LabVIEW Graphical Programming

Fourth Edition

Gary W. Johnson
Richard Jennings

McGraw-Hill
New York Chicago San Francisco Lisbon London
Madrid Mexico City Milan New Delhi San Juan
Seoul Singapore Sydney Toronto
Copyright © 2006 by The McGraw-Hill Companies. All rights reserved. Except as permitted under the United States Copyright
Act of 1976, no part of this publication may be reproduced or distributed in any form or by any means, or stored in a database
or retrieval system, without the prior written permission of the publisher.

ISBN: 978-0-07-150153-8

MHID: 0-07-150153-3

The material in this eBook also appears in the print version of this title: ISBN: 978-0-07-145146-8, MHID: 0-07-145146-3.

All trademarks are trademarks of their respective owners. Rather than put a trademark symbol after every occurrence of a
trademarked name, we use names in an editorial fashion only, and to the benefit of the trademark owner, with no intention of
infringement of the trademark. Where such designations appear in this book, they have been printed with initial caps.

McGraw-Hill eBooks are available at special quantity discounts to use as premiums and sales promotions, or for use in
corporate training programs. To contact a representative please visit the Contact Us page at www.mhprofessional.com.

Information has been obtained by McGraw-Hill from sources believed to be reliable. However, because of the possibility of
human or mechanical error by our sources, McGraw-Hill, or others, McGraw-Hill does not guarantee the accuracy, adequacy, or
completeness of any information and is not responsible for any errors or omissions or the results obtained from the use of such
information.

TERMS OF USE

This is a copyrighted work and The McGraw-Hill Companies, Inc. (“McGraw-Hill”) and its licensors reserve all rights in and to
the work. Use of this work is subject to these terms. Except as permitted under the Copyright Act of 1976 and the right to store
and retrieve one copy of the work, you may not decompile, disassemble, reverse engineer, reproduce, modify, create derivative
works based upon, transmit, distribute, disseminate, sell, publish or sublicense the work or any part of it without McGraw-Hill’s
prior consent. You may use the work for your own noncommercial and personal use; any other use of the work is strictly
prohibited. Your right to use the work may be terminated if you fail to comply with these terms.

THE WORK IS PROVIDED “AS IS.” McGRAW-HILL AND ITS LICENSORS MAKE NO GUARANTEES OR WAR-
RANTIES AS TO THE ACCURACY, ADEQUACY OR COMPLETENESS OF OR RESULTS TO BE OBTAINED FROM
USING THE WORK, INCLUDING ANY INFORMATION THAT CAN BE ACCESSED THROUGH THE WORK VIA
HYPERLINK OR OTHERWISE, AND EXPRESSLY DISCLAIM ANY WARRANTY, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR
PURPOSE. McGraw-Hill and its licensors do not warrant or guarantee that the functions contained in the work will meet your
requirements or that its operation will be uninterrupted or error free. Neither McGraw-Hill nor its licensors shall be liable to you
or anyone else for any inaccuracy, error or omission, regardless of cause, in the work or for any damages resulting therefrom.
McGraw-Hill has no responsibility for the content of any information accessed through the work. Under no circumstances shall
McGraw-Hill and/or its licensors be liable for any indirect, incidental, special, punitive, consequential or similar damages that
result from the use of or inability to use the work, even if any of them has been advised of the possibility of such damages. This
limitation of liability shall apply to any claim or cause whatsoever whether such claim or cause arises in contract, tort or
otherwise.
 Contents

Preface ........................................................ xi
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv

Chapter 1 Roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
LabVIEW and Automation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Virtual instruments: LabVIEW’s foundation . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Why use LabVIEW? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
The Origin of LabVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Introduction ................................................... 9
A vision emerges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
All the world’s an instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
A hard-core UNIX guy won over by the Macintosh . . . . . . . . . . . . . . . . . . . . . 12
Putting it all together with pictures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Favoring the underdog platform for system design . . . . . . . . . . . . . . . . . . . . . 15
Ramping up development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Stretching the limits of tools and machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Facing reality on estimated development times . . . . . . . . . . . . . . . . . . . . . . . . 18
Shipping the first version . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Apple catches up with the potential offered by LabVIEW ................ 19
LabVIEW 2: A first-rate instrument control product becomes
a world-class programming system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
The port to Windows and Sun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
LabVIEW 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
LabVIEW 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
LabVIEW branches to BridgeVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
LabVIEW 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
The LabVIEW RT branch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
LabVIEW 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
LabVIEW 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
LabVIEW 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Crystal Ball Department . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
LabVIEW influences other software products . . . . . . . . . . . . . . . . . . . . . . . . . 32
LabVIEW Handles Big Jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

Chapter 2 Getting Started . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

About the Diagrams in This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Sequencing and Data Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

iii
iv Contents

LabVIEW under the Hood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

The parts of a VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
How VIs are compiled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Multitasking, multithreaded LabVIEW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
The LabVIEW Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Front panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Property nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Block diagram ................................................. 48
SubVIs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Icons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Polymorphic VIs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Typedefs ..................................................... 52
Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
See what the subVIs are up to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Peeking at data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
One step at a time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Execution highlighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Setting breakpoints ............................................. 57
Suspend when called . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Calling Other Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
CINs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Dynamic link libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Programming by Plagiarizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

Chapter 3 Controlling Program Flow . . . . . . . . . . . . . . . . . . . . . . . . . . 61

Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Data Dependency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Adding Common Threads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Looping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
While LOOPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
For Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Shift registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Uninitialized shift registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Globals ........................................................ 71
Global and local variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Built-in global variables—and their hazards . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Local variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Notify and Filter events .......................................... 81
Mechanical actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Dynamic events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Design Patterns ................................................. 87
Initialize and then loop .......................................... 87
Independent parallel loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Client-server .................................................. 90
Client-server (with autonomous VIs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
State machines ................................................ 94
Queued message handler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Event-driven applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
 Contents v

Chapter 4 LabVIEW Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

Numeric Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Building strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Parsing strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Dealing with unprintables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Spreadsheets, strings, and arrays ................................. 110
Arrays ......................................................... 114
Initializing arrays ............................................... 117
Array memory usage and performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Waveforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Data Type Conversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Conversion and coercion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
Intricate conversions and type casting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Flatten To String (. . . Do what?) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Enumerated types (enums) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Get Carried Away Department . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

Chapter 5 Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

Where Do Little Timers Come From? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Using the Built-in Timing Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Timed structures ............................................... 140
Timing sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Execution and priority . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Timing guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Sending timing data to other applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
High-resolution and high-accuracy timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Chapter 6 Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

Polling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
Occurrences .................................................... 155
Notifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Queues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Semaphores .................................................... 161
Me and You, Rendezvous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

Chapter 7 Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

Accessing Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
File Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Writing Text Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
Reading Text Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Formatting to Text Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
Binary Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
Writing binary files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Reading binary files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Writing Datalog Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Reading Datalog Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Datalog file utilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
vi Contents

Chapter 8 Building an Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

Define the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Analyze the user’s needs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Gather specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Draw a block diagram ........................................... 189
Specify the I/O Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Prototype the User Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Panel possibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
First Design and Then Write Your Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Ask a Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Top-down or bottom-up? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Modularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Choose an architecture: Design patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
The VI hierarchy as a design tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Sketching program structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Pseudocoding ................................................. 203
Ranges, coercion, and default values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Handling errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Putting it all together ............................................ 213
Testing and Debugging Your Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Tracing execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Checking performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Final Touches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
VBL epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
Studying for the LabVIEW Certification Exams . . . . . . . . . . . . . . . . . . . . . . . . . . 218
CLAD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
CLD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
Example 1: Traffic light controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Example 2: Car wash controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Example 3: Security system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

Chapter 9 Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

VI Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Control Descriptions .............................................. 232
Custom Online Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Documenting the Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
VI History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
Other Ways to Document .......................................... 235
Printing LabVIEW Panels and Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Putting LabVIEW screen images into other documents . . . . . . . . . . . . . . . . . . 236
Writing Formal Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Document outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Connector pane picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
VI description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Terminal descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
Programming examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Distributing Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

Chapter 10 Instrument Driver Basics . . . . . . . . . . . . . . . . . . . . . . . . . . 243

Finding Instrument Drivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
Driver Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Communication standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Learn about Your Instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Determine Which Functions to Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
 Contents vii

Establish Communications ......................................... 251

Hardware and wiring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
Protocols and basic message passing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256

Chapter 11 Instrument Driver Development Techniques . . . . . . . . . . 257

Plug-and-Play Instrument Drivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
General Driver Architectural Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
Error I/O flow control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Modularity by grouping of functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Project organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
Initialization ................................................... 267
Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
Action and status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
Utility ........................................................ 271
Close . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274

Chapter 12 Inputs and Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275

Origins of Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Transducers and sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
Actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
Categories of signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Grounding and shielding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Why use amplifiers or other signal conditioning? . . . . . . . . . . . . . . . . . . . . . . . 291
Choosing the right I/O subsystem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Network everything! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304

Chapter 13 Sampling Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

Sampling Theorem ............................................... 305
Filtering and Averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
About ADCs, DACs, and Multiplexers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Digital-to-analog converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
Digital codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
Triggering and Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
A Little Noise Can Be a Good Thing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Throughput ..................................................... 319
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321

Chapter 14 Writing a Data Acquisition Program . . . . . . . . . . . . . . . . . 323

Data Analysis and Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Postrun analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
Real-time analysis and display . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
Sampling and Throughput . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Signal bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Oversampling and digital filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
Timing techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
Configuration Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
What to configure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
Configuration editors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
viii Contents

Configuration compilers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362

Saving and recalling configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
A Low-Speed Data Acquisition Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
Medium-Speed Acquisition and Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375

Chapter 15 LabVIEW RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377

Real Time Does Not Mean Real Fast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
RT Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
Designing Software to Meet Real-Time Requirements . . . . . . . . . . . . . . . . . . . . 382
Measuring performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
Shared resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
Multithreading and multitasking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
Organizing VIs for best real-time performance ........................ 391
Context switching adds overhead . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
Scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
Timed structures ............................................... 395
Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399

Chapter 16 LabVIEW FPGA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401

What Is an FPGA? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
LabVIEW for FPGAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
RIO hardware platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
Plug-in cards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
CompactRIO .................................................. 404
Timing and synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Compact Vision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Application Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
Compiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
Synchronous execution and the enable chain . . . . . . . . . . . . . . . . . . . . . . . . . 408
Clocked execution and the single-cycle Timed Loop . . . . . . . . . . . . . . . . . . . . 411
Parallelism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
Pipelining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415

Chapter 17 LabVIEW Embedded . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417

Introduction ................................................... 417
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
LabVIEW Embedded Development Module . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
The technology: What’s happening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Running LabVIEW Embedded on a new target . . . . . . . . . . . . . . . . . . . . . . . . 421
Porting the LabVIEW runtime library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
Incorporating the C toolchain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
The Embedded Project Manager . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
LEP plug-in VIs ................................................ 425
Target_OnSelect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
Other plug-in VIs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
Incorporating I/O srivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
LabVIEW Embedded programming best practices . . . . . . . . . . . . . . . . . . . . . 431
Interrupt driven programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
LabVIEW Embedded targets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
 Contents ix

Chapter 18 Process Control Applications . . . . . . . . . . . . . . . . . . . . . . 437

Process Control Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
Industrial standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
Control = manipulating outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
Process signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
Control system architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
Working with Smart Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
Single-loop controllers (SLCs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
Other smart I/O subsystems ...................................... 463
Man-Machine Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
Display hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
Other interesting display techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
Handling all those front panel items . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
Data Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
Input scanners as servers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
Handling output data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
Display VIs as clients ........................................... 479
Using network connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Real-time process control databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
Simulation for validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
Sequential Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
Interlocking with logic and tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
State machines ................................................ 487
Initialization problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
GrafcetVIEW—a graphical process control package . . . . . . . . . . . . . . . . . . . . 490
Continuous Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
Designing a control strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
Trending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
Real-time trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
Historical trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502
Statistical process control (SPC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
Alarms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 506
Using an alarm handler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 508
Techniques for operator notification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512

Chapter 19 Physics Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513

Special Hardware ................................................ 514
Signal conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
CAMAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 518
Other I/O hardware ............................................. 518
Field and Plasma Diagnostics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
Step-and-measure experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
Plasma potential experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527
Handling Fast Pulses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
Transient digitizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
Digital storage oscilloscopes (DSOs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536
Timing and triggering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
Capturing many pulses .......................................... 539
Recovering signals from synchronous experiments . . . . . . . . . . . . . . . . . . . . . 543
Handling Huge Data Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
Reducing the amount of data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
Optimizing VIs for memory usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
x Contents

Chapter 20 Data Visualization, Imaging, and Sound . . . . . . . . . . . . . . 555

Graphing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 556
Displaying waveform and cartesian data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
Bivariate data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
Multivariate data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
3D Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 570
Intensity Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
Image Acquisition and Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572
System requirements for imaging .................................. 574
Using IMAQ Vision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
IMAQ components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
Sound I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 586
DAQ for sound I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 586
Sound I/O functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587
Sound input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587
Sound output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 588
Sound files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 588
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589

Index .................................................... 591

 Preface

Twenty years have passed since the release of LabVIEW. During this
period, it has become the dominant programming language in the world
of instrumentation, data acquisition, and control. A product of National
Instruments Corporation (Austin, Texas), it is built upon a purely
graphical, general-purpose programming language, G, with extensive
libraries of functions, an integral compiler and debugger, and an appli-
cation builder for stand-alone applications. The LabVIEW development
environment runs on Apple Macintosh computers and IBM PC compat-
ibles with Linux or Microsoft Windows. Programs are portable among
the various development platforms. The concept of virtual instruments
(VIs), pioneered by LabVIEW, permits you to transform a real instru-
ment (such as a voltmeter) into another, software-based instrument
(such as a chart recorder), thus increasing the versatility of available
hardware. Control panels mimic real panels, right down to the switches
and lights. All programming is done via a block diagram, consisting of
icons and wires, that is directly compiled to executable code; there is no
underlying procedural language or menu-driven system.
Working with research instrumentation, we find LabVIEW
indispensable—a flexible, time-saving package without all the frustrat-
ing aspects of ordinary programming languages. The one thing LabVIEW
had been missing all these years was a useful application-oriented book.
The manuals are fine, once you know what you want to accomplish, and
the classes offered by National Instruments are highly recommended
if you are just starting out. But how do you get past that first blank
window? What are the methods for designing an efficient LabVIEW
application? What about interface hardware and real-world signal-
conditioning problems? In this book, we describe practical problem-
solving techniques that aren’t in the manual or in the introductory
classes—methods you learn only by experience. The principles and
techniques discussed in these pages are fundamental to the work of
a LabVIEW programmer. This is by no means a rewrite of the manu-
als or other introductory books, nor is it a substitute for a course in

xi
xii Preface

LabVIEW basics. You are encouraged to consult those sources, as well,

in the process of becoming a skilled LabVIEW developer.
This fourth edition is founded on LabVIEW 8, but we’ve worked
closely with National Instruments to ensure its relevance now and
through future versions of LabVIEW. Chapter 1, “Roots,” starts off
with an entertaining history of the development of LabVIEW. New to
this edition is coverage of material on National Instruments’ certifica-
tion exams. There are three levels of LabVIEW certification: Certified
LabVIEW Associate Developer (CLAD), Certified LabVIEW Developer
(CLD), and Certified LabVIEW Architect (CLA). Each exam builds on
the knowledge required for the previous exam. We have worked closely
with National Instruments to highlight study material in this book
for the first two certification exams. Throughout Chapters 2 through 9
you will find or icons next to sections covered on the cer-
tification exams.
In Chapters 2 through 9, we get down to the principles of program-
ming in G. After a discussion of the principles of dataflow programming,
we discuss programming structures, data types, timing, synchroniza-
tion, and file I/O. Chapter 8, “Building an Application,” shows you how
to design a LabVIEW application. Here we assume that you are not a
formally trained software engineer, but rather a technically skilled per-
son with a job to do (that certainly describes us!). We’ll walk through
the development of a real application that Gary wrote, starting with
selection of hardware, then prototyping, designing, and testing the
program. Chapter 9, “Documentation,” covers this important but oft-
neglected topic. We discuss recommended practice for creating effective
documentation as it pertains to the world of LabVIEW. If you know the
material in Chapters 2 through 9 you should have no problems with
the certification exams. At the end of Chapter 8 we provide three prac-
tice exams for the Certified LabVIEW Developer (CLD) Exam. Use the
knowledge you gained in Chapters 2 through 9 to complete these prac-
tice exams in four hours and you are well on your way to certification.
If you connect your computer to any external instruments, you will
want to read Chapters 10 and 11, “Instrument Driver Basics” and
“Instrument Driver Development Techniques.” We begin with the
basics of communications and I/O hardware (GPIB, serial, and VXI),
then cover recommended driver development techniques and program-
ming practices, especially the virtual instrument standard architecture
(VISA) methods. Instrument drivers can be fairly challenging to write.
Since it’s one of our specialties, we hope to pass along a few tricks.
The basics of interface hardware, signal conditioning, and analog/
digital conversion are discussed in Chapter 12, “Inputs and Outputs,”
and Chapter 13, “Sampling Signals.” Notably, these chapters con-
tain no LabVIEW programming examples whatsoever. The reason is
 Preface xiii

simple: more than half the “LabVIEW” questions that coworkers ask us
turn out to be hardware- and signal-related. Information in this chap-
ter is vital and will be useful no matter what software you may use for
measurement and control. Chapter 14, “Writing a Data Acquisition Pro-
gram,” contains a practical view of data acquisition (DAQ) applications.
Some topics may seem at first to be presented backward—but for good
reasons. The first topic is data analysis. Why not talk about sampling
rates and throughput first? Because the only reason for doing data
acquisition is to collect data for analysis. If you are out of touch with the
data analysis needs, you will probably write the wrong data acquisition
program. Other topics in this chapter are sampling speed, throughput
optimization, and configuration management. We finish with some of
the real applications that you can use right out of the box.
LabVIEW RT brings the ease of graphical programming to the arcane
world of real-time system programming. In Chapter 15, “LabVIEW RT,”
we show you how LabVIEW RT works and how to achieve top per-
formance by paying attention to code optimization, scheduling, and
communications.
When software-timed real-time applications won’t fit the bill,
LabVIEW FPGA is the way to go. LabVIEW FPGA applications are not
constrained by processor or operating system overhead. With LabVIEW
FPGA you can write massively parallel hardware-timed digital con-
trol applications with closed loop rates in the tens of megahertz. Chap-
ter 16, “LabVIEW FPGA,” gives a solid introduction to programming
FPGAs with LabVIEW.
Embedded computer systems are all around us—in our cars, VCRs,
appliances, test equipment, and a thousand other applications. But until
now, LabVIEW has not been a viable development system for those min-
iaturized computers. Chapter 17, “LabVIEW Embedded,” introduces a
new version of LabVIEW capable of targeting any 32-bit microprocessor.
Chapter 18, “Process Control Applications,” covers industrial con-
trol and all types of measurement and control situations. We’ll look at
human-machine interfaces, sequential and continuous control, trend-
ing, alarm handling, and interfacing to industrial controllers, particu-
larly programmable logic controllers (PLCs).We frequently mention a
very useful add-on toolkit that you install on top of LabVIEW, called
the Datalogging and Supervisory Control Module (formerly available
as BridgeVIEW), which adds many important features for industrial
automation.
LabVIEW has a large following in physics research, so we wrote
Chapter 19, “Physics Applications.” Particular situations and solutions
in this chapter are electromagnetic field and plasma diagnostics, mea-
suring fast pulses with transient recorders, and handling very large
data sets. This last topic, in particular, is of interest to almost all users
xiv Preface

because it discusses techniques for optimizing memory usage. (There

are tidbits like this all through the book—by all means, read it cover
to cover!)
Chapter 20, “Data Visualization, Imaging, and Sound,” shows off
some of the data presentation capabilities of LabVIEW. Some third-
party products and toolkits (such as IMAQ for imaging) are featured.
They enable you to acquire video signals, process and display images,
make three-dimensional plots, and record and play sound.
As far as possible, this book is platform-independent, as is LabVIEW
itself. Occasional topics arise where functionality is available on only
one or two of the computer platforms. The LabVIEW user manual con-
tains a portability guide that you can consult when developing applica-
tions that you intend to propagate among various platforms.
Many important resources are available only via the Internet. For
your convenience, Internet addresses are interspersed in the text.
While writing this book, we found that user-supplied example VIs were
hard to obtain, owing to the fact that so many of us work for govern-
ment laboratories and places that just don’t like to give away their soft-
ware. Where it was not possible to obtain the actual code, we attempted
to reconstruct the important aspects of real applications to give you an
idea of how you might solve similar problems.
Third-party LabVIEW products, such as driver and analysis pack-
ages, are described where appropriate. They satisfy important niche
requirements in the user community at reasonable cost, thus expand-
ing the wide applicability of LabVIEW.
If nothing else, we hope that our enthusiasm for LabVIEW rubs off
on you.

Gary W. Johnson and Richard Jennings

 Acknowledgments

We would like to thank the engineers, developers, and managers at

National Instruments who supplied vital information without which this
book would not be possible, particularly Jeff Kodosky, David Gardner,
Joel Sumner, Newton Petersen, P. J. Tanzillo, and Kathy Brown. A spe-
cial thanks goes to Zaki Chasmawala of National Instruments for proof-
reading and highlighting the pertinent parts of Chapters 2 through 9
covered by the certification exams.
Credit also goes to our wives, Katharine Decker Johnson and Patty
Jennings, whose patience during this project cannot be overstated. And
to Elizabeth, Danny, Chris, and David Jennings—thank you for under-
standing.
Finally, thanks to our editor, Wendy Rinaldi.

Gary W. Johnson:
To my wife, Katharine

Richard Jennings:
To my Lord and Savior, Jesus Christ

xv
ABOUT THE AUTHORS

Gary W. Johnson is an instrumentation

engineer at the Lawrence Livermore National
Laboratory. He has a BS degree in electrical
engineering/bioengineering from the
University of Illinois. His professional interests
include measurement and control systems,
electro-optics, communications, transducers,
circuit design, and technical writing. In his
spare time, he enjoys woodworking, bicycling,
and amateur radio. He and his wife, Katharine,
a scientific illustrator, live in Livermore,
California, with their twin Afghan hounds,
Chloe and Derby.
LabVIEW goes aloft. Gary works on a LabVIEW-based
laser wavelength controller for an airborne LIDAR sys-
tem aboard an Air Force C-135.

Richard Jennings is president of Jennings

Embedded Services, LLC in San Antonio,
Texas. His company was founded in 2005 to
serve as a hardware and software resource
for companies engaged in embedded software
and hardware development in emerging
embedded markets such as industrial control,
wireless, and embedded instrumentation. He
is a 15-year veteran hardware and software
engineer. Prior to starting Jennings Embedded
Services, a National Instruments Certified
Alliance partner, Jennings worked as a system
integrator at Sandia National Laboratories and
at Lawrence Livermore National Laboratories
in Livermore, California. He holds an Associate
Do not look into laser with remaining good eye.
Richard wisely wears his laser eye protection while Degree in Laser-ElectroOptics from Texas State
manually aligning laser beams through a piloted jet Technical Institute in Waco, Texas, and is a
burner in the Turbulent Combustion Laboratory. Certified LabVIEW Developer. In 2003 he was
awarded National Instruments’ prestigious
Virtual Instrumentation Professional (VIP)
award. www.jembedded.com
 Chapter

1
Roots

LabVIEW has certainly made life easier for Gary, the engineer. I remem-
ber how much work it was in the early 1980s, writing hideously long pro-
grams to do what appeared to be simple measurement and control tasks.
Scientists and engineers automated their operations only when it was
absolutely necessary, and the casual users and operators wouldn’t dare
to tamper with the software because of its complexity. This computer-
based instrumentation business was definitely more work than fun. But
everything changed when, in mid-1987, I went to a National Instru-
ments (NI) product demonstration. The company was showing off a new
program that ran on a Macintosh. It was supposed to do something with
instrumentation, and that sounded interesting. When I saw what those
programmers had done—and what LabVIEW could do—I was stunned!
Wiring up icons to write a program? Graphical controls? Amazing! I had
to get hold of this thing and try it out for myself.
By the end of the year, I had taken the LabVIEW class and started
on my first project, a simple data acquisition system. It was like watch-
ing the sun rise. There were so many possibilities now with this easy
and fun-to-use programming language. I actually started looking for
things to do with it around the lab (and believe me, I found them).
It was such a complete turnaround from the old days. Within a year,
LabVIEW became an indispensable tool for my work in instrumenta-
tion and control. Now my laboratories are not just computerized, but
also automated. A computerized experiment or process relies heavily
on the human operators—the computer makes things easier by tak-
ing some measurements and doing simple things like that, but it’s far
from being a hands-off process. In an automated experiment, on the
other hand, you set up the equipment, press the Start button on the
LabVIEW screen, and watch while the computer orchestrates a sequence

1
2 Chapter One

of events, takes measurements, and presents the results. That’s how

you want your system to work, and that’s where LabVIEW can save the
day. Let’s start by taking a look at the world of automation.

LabVIEW and Automation

Computers are supposed to make things easier, faster, or more automatic,
that is, less work for the human host. LabVIEW is a unique program-
ming system that makes computer automation a breeze for the scientist
or engineer working in many areas of laboratory research, industrial
control, and data analysis. You have a job to do—someone is probably
paying you to make things happen—and LabVIEW can be a real help in
getting that job done, provided that you apply it properly. But debates
are now raging over this whole business of computers and their influ-
ence on our productivity. For instance, an article I read recently reported
that we now tend to write longer proposals and reports (and certainly
prettier ones) than we used to when only a typewriter was available.
The modern word processor makes it easy to be thorough and commu-
nicate ideas effectively. But does this modern approach always result
in an improvement in productivity or quality? Sometimes we actually
spend more time to do the same old thing. We also become slaves to our
computers, always fussing over the setup, installing new (necessary?)
software upgrades, and generally wasting time.
You must avoid this trap. The key is to analyze your problems and
see where LabVIEW and specialized computer hardware can be used to
their greatest advantage. Then make efficient use of existing LabVIEW
solutions. As you will see, many laboratory automation problems have
already been solved for you, and the programs and equipment are read-
ily available. There are no great mysteries here, just some straightfor-
ward engineering decisions you have to make regarding the advantages
and disadvantages of computer automation. Let’s take a pragmatic
view of the situation. There are many operations that beg for automa-
tion. Among them are
■ Long-term, low-speed operations such as environmental monitoring
and control
■ High-speed operations such as pulsed power diagnostics in which a
great deal of data is collected in a short time
■ Repetitive operations, such as automated testing and calibration,
and experiments that are run many times
■ Remote or hazardous operations where it is impractical, impossible,
or dangerous to have a human operator present
■ High-precision operations that are beyond human capability
■ Complex operations with many inputs and outputs
 Roots 3

In all these cases, please observe that a computer-automated system

makes practical an operation or experiment that you might not other-
wise attempt. An automation may offer additional advantages:
■ It reduces data transcription errors. The old “graphite data acquisi-
tion system” (a pencil) is prone to many error sources not present
in a computer data acquisition system. Indeed, more reliable data
often leads to better quality control of products and new discoveries
in experimental situations.
■ It eliminates operator-induced variations in the process or data col-
lection methods. Repeatability is drastically improved because the
computer never gets tired and always does things the same way.
■ It increases data throughput because you can operate a system at
computer speed rather than human speed.

There are some disadvantages hiding in this process of computer

automation, however:

■ Automation may introduce new sources of error, through improper

use of sensors, signal conditioning, and data conversion, and occa-
sionally through computational (for example, round-off) errors.
■ Misapplication of any hardware or software system is a ticket for
trouble. For instance, attempting to collect data at excessively high
rates results in data recording errors.
■ Reliability is always a question with computer systems. System fail-
ures (crashes) and software bugs plague every high-tech installation
known, and they will plague yours as well.

Always consider the cost-effectiveness of a potential automation solu-

tion. It seems as if everything these days is driven by money. If you can
do it cheaper-better-faster, it is likely to be accepted by the owner, the
shareholders, or whoever pays the bills. But is a computer guaranteed
to save you money or time? If I have a one-time experiment in which
I can adequately record the data on a single strip-chart recorder, an
oscilloscope, or with my pencil, then taking two days to write a special
program makes no sense whatsoever.
One way to automate (or at least computerize) simple, one-time
experiments is to build what I call a LabVIEW crash cart, which is
much like the doctor’s crash cart in an emergency room. When someone
has a short-term measurement problem, I can roll in my portable rack
of equipment. It contains a Macintosh with LabVIEW, analog inter-
face hardware, and some programmable instruments. I can quickly
configure the general-purpose data acquisition program, record data,
and analyze them—all within a few hours. You might want to consider
4 Chapter One

this concept if you work in an area that has a need for versatile data
acquisition. Use whatever spare equipment you may have, recycle some
tried-and-true LabVIEW programs, and pile them on a cart. It doesn’t
even matter what kind of computer you have since LabVIEW runs on
all versions of Windows, Power Macintosh, Sun SPARCstations, Linux,
and HP workstations. The crash cart concept is simple and marvel-
ously effective.
Automation is expensive: The costs of sensors, computers, software,
and the programmer’s effort quickly add up. But in the end, a marvelous
new capability can arise. The researcher is suddenly freed from the labor
of logging and interpreting data. The operator no longer has to orches-
trate so many critical adjustments. And data quality and product quality
rise. If your situation fits the basic requirements for which automation is
appropriate, then by all means consider LabVIEW as a solution.

Virtual instruments: LabVIEW’s foundation

LabVIEW made the concept of virtual instrument (VI) a practical
reality. The objective in virtual instrumentation is to use a general-
purpose computer to mimic real instruments with their dedicated
controls and displays, but with the added versatility that comes with
software. (See Figure 1.1.) Instead of buying a strip-chart recorder, an
oscilloscope, and a spectrum analyzer, you can buy one high-performance

Figure 1.1 This virtual instrument (bottom) is a cus-

tomized version of the real instrument, but having
only the features that you need.
 Roots 5

analog-to-digital converter (ADC) and use a computer running LabVIEW

to simulate all these instruments and more. The VI concept is so funda-
mental to the way that LabVIEW works that the programs you write in
LabVIEW are in fact called VIs. You use simple instruments (subVIs)
to build more complex instruments, just as you use subprograms to
build a more complex main program in a conventional programming
language.

Virtual versus real instrumentation. Virtual instrumentation systems

such as LabVIEW inevitably invite comparison to real physical instru-
mentation. The major drawback in using a personal computer for
implementing virtual instruments is that the computer has only one
central microprocessor. An application that uses multiple instruments
can easily overburden the processor. A stand-alone instrument, how-
ever, may contain any number of processors, each dedicated to specific
processing tasks. In addition, these multiple processors can operate in
parallel, providing a great increase in overall performance. But this
increase in performance results in the expensive price tag accompany-
ing many dedicated instruments, as well as a decrease in flexibility.
Technology from National Instruments is advancing to address
these issues, however. First, a new version of LabVIEW—LabVIEW
RT—allows you to run independent programs on multiple, dedicated,
real-time processors that serve to unload tasks from the desktop host.
Second, LabVIEW permits you to link all kinds of computing platforms
as well as other software applications over a variety of networks, thus
further distributing the workload. Third, plug-in boards now contain
their own processors and are available at very reasonable prices. Digi-
tal signal processors are a good example of special-purpose processors
that find their way onto plug-in boards. Many plug-in data acquisition
boards also have sophisticated direct memory access (DMA), timing,
and triggering capabilities that can span multiple boards, resulting in
improved synchronization and signal coupling between boards. These
developments, along with the development of more capable operating
systems and computer architectures, have brought parallel processing
capabilities to personal computers, making them more sophisticated
platforms for instrumentation and data acquisition applications. How-
ever, sophistication comes at the expense of complexity, because you
must have greater knowledge of these hardware components and their
interconnection than is required to use a stand-alone instrument with
similar capabilities. Virtual instrumentation software is essential for
turning these sophisticated hardware combinations into usable instru-
mentation systems.
Virtual instrumentation offers the greatest benefit over real instru-
ments in the areas of price/performance, flexibility, and customization.
6 Chapter One

For the price of a dedicated high-performance instrument, you can

assemble a personal computer–based system with the fundamental
hardware and software components to design virtual instruments tar-
geted for specific applications. The hardware may be plug-in boards,
external instruments, or a combination of both. In either case, a soft-
ware interface can be as complicated or as simple as needed to serve
the application. You can simplify the operation of a complex stand-
alone instrument with virtual instruments that focus on controlling
only subsets of the instrument’s full capabilities. I, for one, get lost in
the buttons and menus on the panels of many modern instruments and
find a simple VI a welcome relief.
Although LabVIEW has existed since 1986, the virtual instrumen-
tation and block diagram concepts embodied in its design are still at
the leading edge of instrumentation and computer science technology
today. The cost of developing test program software continues to rise
with the increasing complexity of the devices being tested and the
instruments needed to test them. Software modularity, maintainabil-
ity, and reusability, key benefits of LabVIEW’s hierarchical and homo-
geneous structure, are critically important to reducing the burden of
individual software developers. Reusing routines that you have written
and sharing them with others can save a great deal of time and make
programs more reliable.
Virtual instrumentation is becoming increasingly important in the
instrument control world. VMEbus Extensions for Instrumenta-
tion (VXI), a major development in instrumentation, is a standard
that defines physical and electrical parameters, as well as software
protocols, for implementing instrument-on-a-card test systems. A VXI
instrument is a card that plugs into a chassis containing several cards.
Because they are plug-in cards and not stand-alone boxes, individual
VXI instruments do not have front panel user interfaces. Users cannot
interact with a VXI instrument by pressing buttons or reading displays
on a front panel. Instead, VXI systems must be controlled by a com-
puter or other processor-based device, placing new demands on com-
puter software for controlling instrumentation.
VXI instruments are natural candidates for virtual instrumentation
implementations. In the area of user interaction, software front panels
offer a visual means of controlling a faceless VXI instrument. In addi-
tion, the combination of plug-in modules and high-performance timing
and communications capabilities of the VXI bus makes the configura-
tion of a VXI test system much more complex than that of a GPIB test
system. LabVIEW’s method of graphically connecting virtual instru-
ments into a block diagram accelerates the configuring and program-
ming of VXI test systems.
 Roots 7

Why use LabVIEW?

I use LabVIEW because it has significant advantages over conventional
languages and other control and data acquisition packages.

■ My productivity is simply better in LabVIEW than with conven-

tional programming languages. I’ve measured a factor of 5 compared
with C on a small project. Others have reported improvements of 15
times.* Quick prototypes as well as finished systems are now rou-
tinely delivered in what used to be record time.
■ The graphical-user interface is built in, intuitive in operation, simple
to apply, and, as a bonus, nice to look at.
■ LabVIEW’s graphical language—G—is a real programming lan-
guage, not a specialized application. It has few intrinsic limitations.
■ There is only a minimal performance penalty when it is compared
with conventional programming languages, and LabVIEW does some
things better. No other graphical programming system can make this
claim.
■ Programmer frustration is reduced because annoying syntax errors
are eliminated. Have you ever gotten yourself into an argument with
a C compiler over what is considered “legal”? Or made a seemingly
minor error with a pointer and had your machine crash?
■ Rapid prototyping is encouraged by the LabVIEW environment,
leading to quick and effective demonstrations that can be reused or
expanded into complete applications.
■ Sophisticated built-in libraries and add-on toolkits address specific
needs in all areas of science, engineering, and mathematics.
■ Programming in LabVIEW is fun. I would never say that about C
(challenging, yes, but fun, no).

Like any other tool, LabVIEW is useful only if you know how to use
it. And the more skilled you are in the use of that tool, the more you will
use it. After 19 years of practice, I’m really comfortable with LabVIEW.
It’s gotten to the point where it is at least as important as a word pro-
cessor, a multimeter, or a desktop calculator in my daily work as an
engineer.

*“Telemetry Monitoring and Display Using LabVIEW,” by George Wells, Member of

Technical Staff, and Edmund C. Baroth, Manager of the Measurement Technology Cen-
ter, Jet Propulsion Laboratory, California Institute of Technology. Paper given at the
National Instruments User Symposium, March 1993, Austin, Tex.
8 Chapter One

The Origin of LabVIEW

A computer scientist friend of mine relates this pseudobiblical history
of the computer programming world:

In the beginning, there was only machine language, and all was dark-
ness. But soon, assembly language was invented, and there was a glim-
mer of light in the Programming World. Then came Fortran, and the light
went out.

This verse conveys the feeling that traditional computer languages,

even high-level languages, leave much to be desired. You spend a lot
of time learning all kinds of syntactical subtleties, metaphors, com-
piler and linker commands, and so forth, just to say “Hello, world.” And
heaven forbid that you should want to draw a graph or make something
move across the screen or send a message to another computer. We’re
talking about many days or weeks of work here. It’s no wonder that it
took so many years to make the computer a useful servant of the com-
mon person. Indeed, even now it requires at least a moderate education
and plenty of experience to do anything significant in computer pro-
gramming. For the working scientist or engineer, these classical battles
with programming languages have been most counterproductive. All
you wanted to do is make the darned thing display a temperature mea-
surement from your experiment, and what did you get?

(beep) SYNTAX ERROR AT LINE 1326

Thanks a lot, oh mighty compiler. Well, times have changed in a

big way because LabVIEW has arrived. At last, we and other working
troops have a programming language that eliminates that arcane syn-
tax, hides the compiler, and builds the graphical-user interface right
in. There is no fooling around; just wire up some icons and run. And
yet, the thought of actually programming with pictures is so incredible
when contrasted with ordinary computer languages. How did they do
it? Who came up with this idea?
Here is a most enlightening story of the origins of LabVIEW. It’s a
saga of vision; of fear and loathing in the cruel world of computer pro-
gramming; of hard work and long hours; and of breakthroughs, inven-
tion, and ultimate success. The original story, An Instrument That Isn’t
Really, was written by Michael Santori of National Instruments and
has been updated for this book.*

*© 1990 IEEE. Reprinted, with permission, from IEEE Spectrum, vol. 27, no. 8,
pp. 36–39, August 1990.
 Roots 9

Introduction
Prior to the introduction of personal computers in the early 1980s, nearly
all laboratories using programmable instrumentation controlled their
test systems using dedicated instrument controllers. These expensive,
single-purpose controllers had integral communication ports for control-
ling instrumentation using the IEEE-488 bus, also known as the General
Purpose Interface Bus (GPIB).With the arrival of personal computers,
however, engineers and scientists began looking for a way to use these
cost-effective, general-purpose computers to control benchtop instru-
ments. This development fueled the growth of National Instruments,
which by 1983 was the dominant supplier of GPIB hardware interfaces
for personal computers (as well as for minicomputers and other machines
not dedicated solely to controlling instruments).

So, by 1983, GPIB was firmly established as the practical mechanism for
electrically connecting instruments to computers. Except for dealing with
some differing interpretations of the IEEE-488 specification by instru-
ment manufacturers, users had few problems physically configuring their
systems. The software to control the instruments, however, was not in
such a good state. Almost 100 percent of all instrument control programs
developed at this time were written in the BASIC programming language
because BASIC was the dominant language used on the large installed
base of dedicated instrument controllers. Although BASIC had advan-
tages (including a simple and readable command set and interactive
capabilities), it had one fundamental problem: like any other text-based
programming language, it required engineers, scientists, and technicians
who used the instruments to become programmers. These users had to
translate their knowledge of applications and instruments into the lines
of text required to produce a test program. This process, more often than
not, proved to be a cumbersome and tedious chore, especially for those
with little or no prior programming experience.

A vision emerges
National Instruments, which had its own team of programmers struggling
to develop BASIC programs to control instrumentation, was sensitive to
the burden that instrumentation programming placed on engineers and
scientists. A new tool for developing instrumentation software programs
was clearly needed. But what form would it take? Dr. Jim Truchard and
Jeff Kodosky, two of the founders of National Instruments, along with Jack
MacCrisken, who was then a consultant, began the task of inventing this
tool. (See Figure 1.2.) Truchard was in search of a software tool that would
markedly change the way engineers and scientists approached their test
development needs. A model software product that came to mind was
the electronic spreadsheet. The spreadsheet addressed the same general
problem Truchard, Kodosky, and MacCrisken faced—making the com-
puter accessible to nonprogrammer computer users. Whereas the spread-
sheet addressed the needs of financial planners, this entrepreneurial trio
10 Chapter One

Figure 1.2(Left to right): Jack MacCrisken, Jeff Kodosky,

and Jim Truchard, LabVIEW inventors.

wanted to help engineers and scientists. They had their rallying cry—they
would invent a software tool that had the same impact for scientists and
engineers that the spreadsheet had on the financial community.

In 1984, the company, still relatively small in terms of revenue, decided to

embark on a journey that would ultimately take several years. Truchard
committed research and development funding to this phantom product
and named Kodosky as the person to make it materialize. MacCrisken
proved to be the catalyst—an amplifier for innovation on the part of
Kodosky—while Truchard served as the facilitator and primary user.
Dr. T, as he is affectionately known at National Instruments, has a knack
for knowing when the product is right.
Kodosky wanted to move to an office away from the rest of the com-
pany, so he could get away from the day-to-day distractions of the office
and create an environment ripe for inspiration and innovation. He also
wanted a site close to the University of Texas at Austin, so he could access
the many resources available at UT, including libraries for research pur-
poses and, later, student programmers to staff his project. There were
two offices available in the desired vicinity. One office was on the ground
floor with floor-to-ceiling windows overlooking the pool at an apartment
complex. The other office was on the second floor of the building and had
no windows at all. He chose the latter. It would prove to be a fortuitous
decision.
 Roots 11

All the world’s an instrument

The first fundamental concept behind LabVIEW was rooted in a large
test system that Truchard and Kodosky had worked on at the Applied
Research Laboratory in the late 1970s. Shipyard technicians used this
system to test Navy sonar transducers. However, engineers and research-
ers also had access to the system for conducting underwater acoustics
experiments. The system was flexible because Kodosky incorporated
several levels of user interaction into its design. A technician could oper-
ate the system and run specific test procedures with predefined limits
on parameters while an acoustics engineer had access to the lower-level
facilities for actually designing the test procedures. The most flexibility
was given to the researchers, who had access to all the programmable
hardware in the system to configure as they desired (they could also blow
up the equipment if they weren’t careful). Two major drawbacks to the
system were that it was an incredible investment in programming time—
over 18 work-years—and that users had to understand the complicated
mnemonics in menus in order to change anything.

Over several years, Kodosky refined the concept of this test system to
the notion of instrumentation software as a hierarchy of virtual instru-
ments. A virtual instrument (VI) would be composed of lower level virtual
instruments, much like a real instrument was composed of printed cir-
cuit boards and boards composed of integrated circuits (ICs). The bottom-
level VIs represented the most fundamental software building blocks:
computational and input/output (I/O) operations. Kodosky gave particular
emphasis to the interconnection and nesting of multiple software layers.
Specifically, he envisioned VIs as having the same type of construction at
all levels. In the hardware domain, the techniques for assembling ICs into
boards are dramatically different than assembling boards into a chassis.
In the software domain, assembling statements into subroutines differs
from assembling subroutines into programs, and these activities differ
greatly from assembling concurrent programs into systems. The VI model
of homogeneous structure and interface, at all levels, greatly simplifies the
construction of software—a necessary achievement for improving design
productivity. From a practical point of view, it was essential that VIs have
a superset of the properties of the analogous software components they
were replacing. Thus, LabVIEW had to have the computational ability of
a programming language and the parallelism of concurrent programs.

Another major design characteristic of the virtual instrument model was

that each VI had a user interface component. Using traditional program-
ming approaches, even a simple command line user interface for a typ-
ical test program was a complex maze of input and output statements
often added after the core of the program was written. With a VI, the user
interface was an integral part of the software model. An engineer could
interact with any VI at any level in the system simply by opening the
VI’s user interface. The user interface would make it easy to test software
modules incrementally and interactively during system development.
12 Chapter One

In addition, because the user interface was an integral part of every VI, it
was always available for troubleshooting a system when a fault occurred.
(The virtual instrument concept was so central to LabVIEW’s incarnation
that it eventually became embodied in the name of the product. Although
Kodosky’s initial concerns did not extend to the naming of the product,
much thought would ultimately go into the name LabVIEW, which is an
acronym for Laboratory Virtual Instrument Engineering Workbench.)

A hard-core UNIX guy won

over by the Macintosh
The next fundamental concept of LabVIEW was more of a breakthrough
than a slow evolution over time. Kodosky had never been interested in
personal computers because they didn’t have megabytes of memory and
disk storage, and they didn’t run UNIX. About the time Kodosky started
his research on LabVIEW, however, his brother-in-law introduced him to
the new Apple Macintosh personal computer. Kodosky’s recollection of the
incident was that “after playing with MacPaint for over three hours, I
realized it was time to leave and I hadn’t even said hello to my sister.” He
promptly bought his own Macintosh. After playing with the Macintosh,
Kodosky came to the conclusion that the most intuitive user interface for
a VI would be a facsimile of a real instrument front panel. (The Macintosh
was a revelation because DOS and UNIX systems in 1983 did not have
the requisite graphical user interface.) Most engineers learn about an
instrument by studying its front panel and experimenting with it. With its
mouse, menus, scroll bars, and icons, the Macintosh proved that the right
interface would also allow someone to learn software by experimentation.
VIs with graphical front panels that could be operated using the mouse
would be simple to operate. A user could discover how they work, mini-
mizing documentation requirements (although people rarely documented
their BASIC programs anyway).

Putting it all together with pictures

The final conceptual piece of LabVIEW was the programming technique.
A VI with an easy-to-use graphical front panel programmed in BASIC or
C would simplify operation, but would make the development of a VI more
difficult. The code necessary to construct and operate a graphical panel is
considerably more complex than that required to communicate with an
instrument.

To begin addressing the programming problem, Kodosky went back to his

model software product—the spreadsheet. Spreadsheet programs are so
successful because they display data and programs as rows and columns of
numbers and formulas. The presentation is simple and familiar to business-
people. What do engineers do when they design a system? They draw a block
diagram. Block diagrams help an engineer visualize the problem but only
suggest a design. Translation of a block diagram to a schematic or computer
 Roots 13

program, however, requires a great deal of skill. What Kodosky wanted was
a software-diagramming technique that would be easy to use for concep-
tualizing a system, yet flexible and powerful enough to actually serve as a
programming language for developing instrumentation software.

Two visual tools Kodosky considered were flowcharts and state diagrams.
It was obvious that flowcharts could not help. These charts offered a visu-
alization of a process, but to really understand them you have to read the
fine print in the boxes on the chart. Thus, the chart occupies too much
space relative to the fine print yet adds very little information to a well-
formatted program. The other option, a state diagram, is flexible and pow-
erful but the perspective is very different from that of a block diagram.
Representing a system as a collection of state diagrams requires a great
amount of skill. Even after completion, the diagrams must be augmented
with textual descriptions of the transitions and actions before they can be
understood.

Another approach Kodosky considered was dataflow diagrams. Dataflow

diagrams, long recommended as a top-level software design tool, have much
in common with engineering block diagrams. Their one major weakness is
the difficulty involved in making them powerful enough to represent itera-
tive and conditional computations. Special nodes and feedback cycles have
to be introduced into the diagram to represent these computations, making
it extremely difficult to design or even understand a dataflow diagram for
anything but the simplest computations. Kodosky felt strongly, however,
that dataflow had some potential for his new software system.

By the end of 1984, Kodosky had experimented with most of the diagram-
ming techniques, but they were all lacking in some way. Dataflow diagrams
were the easiest to work with up until the point where loops were needed.
Considering a typical test scenario, however, such as “take 10 measure-
ments and average them,” it’s obvious that loops and iteration are at the
heart of most instrumentation applications. In desperation, Kodosky began
to make ad hoc sketches to depict loops specifically for these types of opera-
tions. Loops are basic building blocks of modern structured programming
languages, but it was not clear how or if they could be drawn in a dataflow
concept. The answer that emerged was a box; a box in a dataflow diagram
could represent a loop. From the outside, the box would behave as any
other node in the diagram, but inside it would contain another diagram,
a subdiagram, representing the contents of the loop. All the semantics of
the loop behavior could be encapsulated in the border of the box. In fact,
all the common structures of structured programming languages could be
represented by different types of boxes. His structured dataflow diagrams
were inherently parallel because they were based on dataflow. In 1990, the
first two U.S. patents were issued, covering structured dataflow diagrams
and virtual instrument panels. (See Figure 1.3.)

Kodosky was convinced he had achieved a major breakthrough but was

still troubled by a nagging point. There are times when it is important to
14 Chapter One

For Loop While Loop

Count terminal Shift register

n
n!
Toggling boolean
1
Control (input)

Constant
Indicator RUN
Iteration terminal (output)

Conditional terminal

Figure 1.3 LabVIEW For Loop and While Loop programming structures
with Shift Registers to recirculate data from previous iterations.

force operations to take place sequentially—even when there is no data-

flow requiring it. For example, a signal generator must provide a stimulus
before a voltmeter can measure a response, even though there isn’t an
explicit data dependency between the instruments. A special box to rep-
resent sequential operations, however, would be cumbersome and take up
extra space. During one of their design discussions, Truchard suggested
that steps in a sequence were like frames in a movie. This comment led
to the notion of having several sequential subdiagrams share the same
screen space. It also led to the distinctive graphic style of the Sequence
Structure, as shown in Figure 1.4.

After inventing the fundamentals of LabVIEW block diagramming, it was

a simple matter of using MacPaint to produce pictures of VI panels and
diagrams for several common applications. When Kodosky showed them
to some engineers at NI, the impact was dramatic. The engineers under-
stood the meaning of the diagrams and correctly guessed how to operate

Case Structure Sequence Structure

Diagram identifier Diagram identifier

Which function? Tan 0 [0..1]

GPIB addr
GPIB Write
Tan (x)
Case Selector x f(x) VALUE?
command

Sine 1 [0..1]

Tunnel GPIB Read

Sine (x) Volts

Sequence local
Another frame variable to all Next frame of
of the Case structure subsequent frames the sequence
Figure 1.4LabVIEW Case Structures are used for branching or decision operations, and
Sequence Structures are used for explicit flow control.
 Roots 15

the front panel. Of equal importance, the reviewers expressed great con-
fidence that they would be able to easily construct such diagrams to do
other applications. Now, the only remaining task was to write the soft-
ware, facetiously known as SMOP: small matter of programming.

Favoring the underdog platform

for system design
Although Kodosky felt that the graphics tools on the Macintosh made it
the computer of choice for developing LabVIEW, the clearer marketing
choice was the DOS-based IBM PC. The Macintosh could never be the
final platform for the product because it wasn’t an open machine, and
salespeople would never be able to sell it because the Macintosh was con-
sidered a toy by many scientists and engineers. Politics and marketing
aside, it wasn’t at all clear that you could build a system in which a user
draws a picture and the system runs it. Even if such a system could be
built, would it be fast enough to be useful? Putting marketing concerns
aside momentarily, Kodosky decided to build a prototype on the Macin-
tosh prior to building the real system on a PC.

Kodosky’s affinity for the Macintosh was not for aesthetic reasons alone.
The Macintosh system ROM contains high-performance graphics routines
collectively known as QuickDraw functions. The Macintosh’s most signifi-
cant graphics capability is its ability to manipulate arbitrarily shaped
regions quickly. This capability makes animated, interactive graphics
possible. The graphics region algebra performed by the Macintosh is fast
because of the unique coordinate system built into QuickDraw: the pixels
are between, not on, the gridlines. In addition, the graphics display of the
Macintosh uses square pixels, which simplifies drawing in general and
rotations of bitmaps in particular. This latter capability proves especially
useful for displaying rotating knobs and indicators on a VI front panel.

The operating system of the Macintosh is well integrated. It contains graph-

ics, event management, input/output, memory management, resource and
file management, and more—all tuned to the hardware environment for
fast and efficient operation. Also, the Macintosh uses Motorola’s 68000
family of microprocessors. These processors are an excellent base for large
applications because they have large uniform address space (handling large
arrays of data is easy) and a uniform instruction set (compiler-generated
code is efficient). Remember that this was 1985: the IBM PC compatibles
were still battling to break the 640-kilobyte barrier and had no intrinsic
graphics support. It wasn’t until Microsoft released Windows 3.0 in 1991
that a version of LabVIEW for the PC became feasible.

Ramping up development
Kodosky hired several people just out of school (and some part-time people
still in school) to staff the development team. Without much experience,
16 Chapter One

none of the team members was daunted by the size and complexity of the
software project they were undertaking and instead they jumped into it
with enthusiasm. The team bought 10 Macintoshes equipped with 512
kilobytes of memory and internal hard-disk drives called HyperDrives.
They connected all the computers to a large temperamental disk server.
The team took up residence in the same office near campus used by
Kodosky for his brainstorming session. The choice of location resulted in
11 people crammed into 800 square feet. As it turned out, the working con-
ditions were almost ideal for the project. There were occasional distrac-
tions with that many people in one room but the level of communication
was tremendous. When a discussion erupted between two team members,
it would invariably have some impact on another aspect of the system
they were inventing. The other members working on aspects of the project
affected by the proposed change would enter the discussion and quickly
resolve the issue. The lack of windows and a clock also helped the team
stay focused. (As it turned out, the developers were so impressed with
the productivity of the one-room team concept that the LabVIEW group
is still located in one large room, although it now has windows with a
great view.)

They worked for long hours and couldn’t afford to worry about the time.
All-nighters were the rule rather than the exception and lunch break often
didn’t happen until 3 P.M. There was a refrigerator and a microwave in the
room so the team could eat and work at the same time. The main nutri-
tional staples during development were double-stuff Oreo cookies and pop-
corn, and an occasional mass exodus to Armin’s for Middle Eastern food.

The early development proceeded at an incredible pace. In four months’

time, Kodosky had put together a team and the team had learned how
to program the Macintosh. MacCrisken contributed his project man-
agement skills and devised crucial data structure and software entity
relationship diagrams that served as an overall road map for software
development. They soon produced a proof-of-concept prototype that could
control a GPIB instrument (through a serial port adapter), take multiple
measurements, and display the average of the measurements. In prov-
ing the concept, however, it also became clear that there was a severe
problem with the software speed. It would take two more development
iterations and a year, before the team would produce a viable product.
(See Figure 1.5.)

Stretching the limits of tools and machine

After finishing the first prototype, Kodosky decided to continue working
on the Macintosh because he felt the team still had much to learn before
they were ready to begin the real product. It was at this time that the team
began encountering the realities of developing such a large software sys-
tem. The first problems they encountered were in the development tools.
The software overflowed some of the internal tables, first in the C com-
piler and then in the linker. The team worked with the vendor to remedy
 Roots 17

Figure 1.5 Screen shots from a very early prototype of

LabVIEW. (Thanks to Paul Daley of LLNL who discovered
these old LabVIEW versions deep in his diskette archive.)

the problem. Each time it occurred, the vendor expanded the tables. These
fixes would last for a couple months until the project grew to overflow
them again. The project continued to challenge the capabilities of the
development tools for the duration of the project.

The next obstacle encountered was the Macintosh jump table. The project
made heavy use of object-oriented techniques, which resulted in lots of
functions, causing the jump tables to overflow. The only solution was to
compromise on design principles and work within the limits imposed by
the platform. As it turned out, such compromises would become more com-
monplace in the pursuit of acceptable performance.

The last major obstacle was memory. The project was already getting too
large for the 512-kilobyte capacity of Macintosh and the team still hadn’t
implemented all the required functions, let alone the desirable ones they
had been hoping to include. The prospects looked dim for implementing the
complete system on a DOS-based PC, even with extensive use of overlay-
ing techniques. This situation almost proved fatal to the project. The team
was at a dead end and morale was at an all-time low. It was at this oppor-
tune time that Apple came to the rescue by introducing the Macintosh
Plus in January 1986. The Macintosh Plus was essentially identical to the
existing Macintosh except that it had a memory capacity of one megabyte.
Suddenly, there was enough memory to implement and run the product
with most of the features the team wanted.

Once again, the issue of the marketability of the Macintosh arose. A quick
perusal of the DOS-based PC market showed that the software and hard-
ware technology had not advanced very much. Kodosky decided (with
approval by Dr. Truchard after some persuasion) that, having come this
far on the Macintosh, they would go ahead and build the first version of
LabVIEW on the Macintosh. By the time the first version of LabVIEW
was complete, there would surely, they thought, be a new PC that could
run large programs.
18 Chapter One

Facing reality on estimated

development times
The initial estimates of the remaining development effort were grossly
inaccurate. The April 1986 introduction date passed without a formal soft-
ware release. In May, in anticipation of an imminent shipping date, the
team moved from their campus workroom to the main office, where they
could be close to the application engineers who did customer support. This
event caused much excitement but still no product.

It was at this point that the company became over-anxious and tried to
force the issue by prematurely starting beta testing. The testing was a
fiasco. The software was far from complete. There were many bugs encoun-
tered in doing even the most simple and common operations. Development
nearly ground to a halt as the developers spent their time listening to beta
testers calling in the same problems. (See Figure 1.6.)

As the overall design neared completion, the team began focusing more
on details, especially performance. One of the original design goals was to
match the performance of interpreted BASIC. It was not at all clear how
much invention or redesign it would require to achieve this performance tar-
get, making it impossible to predict when the team would achieve this goal.
On most computational benchmarks, the software was competitive with
BASIC. There was one particular benchmark, the Sieve of Eratosthenes,
that posed, by nature of its algorithm and design, particular problems for
dataflow implementations. The performance numbers the team measured
for the sieve benchmark were particularly horrendous and discouraging—a
fraction of a second for a compiled C program, two minutes for interpreted
BASIC, and over eight hours for LabVIEW.

Figure 1.6 This diagram screen shot is from LabVIEW beta

0.36 in June 1986. The familiar structures (While and For
Loops, and a Case Structure) had appeared by this time.
 Roots 19

Kodosky did his best to predict when the software would be complete,
based on the number of bugs, but was not sure how the major bugs would
be found, much less fixed. Efficiently testing such a complex interactive
program was a vexing and complex problem. The team finally settled on
the bug day approach. They picked a day when the entire team would
stop working on the source code and simply use LabVIEW. They would
try all types of editing operations, build as many and varied VIs as they
could, and write down all the problems they encountered until the white
boards on every wall were full. The first bug day lasted only three hours.
The team sorted the list and for the next five weeks fixed all the fatal
flaws and as many minor flaws as possible. Then they had another bug
day. They repeated this process until they couldn’t generate even one fatal
flaw during an entire day. The product wasn’t perfect, but at least it would
not be an embarrassment.

Shipping the first version

In October 1986, the team figured out how to bypass some of the overhead
in calling a subVI, producing some improvement in performance (for all
but the sieve benchmark, which was better, but still 20 times slower than
BASIC). The decision was made to ship the product. The team personally
duplicated and packaged the first 50 disks and hand-carried them to ship-
ping. Version 1.0 was on the streets.

The reaction to LabVIEW was, in a word, startling. The product received

worldwide acclaim as the first viable, visual, or graphical language. There
were many compliments for a well-designed product, especially from
research and development groups who had had their Macintosh-based
projects canceled by less-adventurous CEOs and marketing departments.
Interestingly enough, the anticipated demand of the targeted BASIC users
did not materialize. These people were apparently content to continue pro-
gramming as they had been doing. Instead, LabVIEW was attracting and
eliciting demands from customers who had never programmed at all but
who were trying to develop systems considered extremely difficult by expe-
rienced programmers in any language. Yet these customers believed they
could successfully accomplish their application goals with LabVIEW.

Apple catches up with the potential

offered by LabVIEW
Shortly after shipment of LabVIEW began, the company received its first
prototype of the Macintosh II. This new version of the Macintosh had many
design features that promised to legitimize the Macintosh in the scientific
and engineering community. The most important of these features was the
open architecture of the new machine. Previous Macintosh versions did
not have the capability to accept plug-in expansion boards. The only mech-
anisms available for I/O were RS-422 serial and SCSI (Small Computer
Systems Interface) ports. National Instruments sold stand-alone interface
20 Chapter One

box products that converted these ports to IEEE-488 control ports, but
performance suffered greatly.

The Macintosh II’s open architecture made it possible to add not only
IEEE-488 support but also other much-needed I/O capabilities, such as
analog-to-digital conversion and digital I/O. The Macintosh II used the
NuBus architecture, an IEEE standard bus that gave the new machine
high-performance 32-bit capabilities for instrumentation and data acqui-
sition that were unmatched by any computer short of a minicomputer (the
PC’s bus was 16 bits).With the flexibility and performance afforded by the
new Macintosh, users now had access to the hardware capabilities needed
to take full advantage of LabVIEW’s virtual instrumentation capabilities.
Audrey Harvey (now a system architect at National Instruments) led the
hardware development team that produced the first Macintosh II NuBus
interface boards, and Lynda Gruggett (now a LabVIEW consultant) wrote
the original LabDriver VIs that supported this new high-performance I/O.
With such impressive new capabilities and little news from the PC world,
National Instruments found itself embarking on another iteration of
LabVIEW development, still on the Macintosh.

Effective memory management turned out to be the key to making this

graphical language competitive with ordinary interpreted languages. The
development team had used a literal interpretation of dataflow program-
ming. Each time data (a wire) leaves a source (a node), the Macintosh Mem-
ory Manager is called to allocate space for the new data, adding tremendous
overhead. Other performance factors involved effective interpretation of
the VI and diagram hierarchy and the scheduling of execution among
nodes on the diagram. It became obvious that memory reuse was vital,
but a suitable algorithm was far from obvious. Kodosky and MacCrisken
spent about four intensive weekend brainstorming sessions juggling the
various factors, eventually arriving at a vague algorithm that appeared to
address everything. The whiteboard was covered with all sorts of instan-
tiation diagrams with “little yellow arrows and blue dots” showing the
prescribed flow of data. Then, one Monday morning, they called in Jeff
Parker (now a LabVIEW consultant) and Steve Rogers (still a LabVIEW
developer at National Instruments) and introduced them to this magnifi-
cent algorithm. The two of them proceeded to implement the algorithm
(to the point that Kodosky and MacCrisken admittedly don’t understand
it anymore!). They kept the famous whiteboard around for about a year,
occasionally referring to it to make sure everything was right. LabVIEW
1.1 included these concepts, known collectively as inplaceness.

While the development team was working with the new Macintosh, they
were also scrambling to meet some of the demands made by customers.
They were simultaneously making incremental improvements in per-
formance, fixing flaws that came to light after shipment began, and try-
ing to plan future developments. As a result of this process, LabVIEW
progressed from version 1.0 to version 1.2 (and the sieve progressed to
23 seconds). LabVIEW 1.2 was a very reliable and robust product. I,
 Roots 21

Figure 1.7This is LabVIEW 1.2. It ran only in black and

white and you couldn’t move an object once it was wired.
Somehow, we early users managed to get a lot accomplished
and enjoyed ourselves, at that.

for one, wrote a lot of useful programs in version 1.2 and can’t recall
crashing. (See Figures 1.7 and 1.8.)

The Macintosh II gave LabVIEW a much-needed and significant boost

in performance. The improvement, however, was short-lived. The internal
architecture of LabVIEW 1.2 was showing signs of distress and the soft-
ware was apparently abusing the Macintosh resource manager as well
as its memory manager. It was becoming clear that the only real way to
enhance capabilities was with a complete redesign. At the least, a rede-
sign could incorporate new diagram analysis algorithms and a fast built-
in compiler that would eliminate performance problems once and for all.
The major objective of the redesign was to achieve execution performance
within a factor of two of compiled C.

Figure 1.8 This is how we got to know the LabVIEW 1 devel-

opment team: the About LabVIEW dialog box had these way-
cool portraits.
22 Chapter One

LabVIEW 2: A first-rate instrument

control product becomes a world-class
programming system
Even as the plans for the next-generation LabVIEW were becoming firm,
customers were fast approaching, and exceeding, the limits of LabVIEW
1.2. Some users were building systems of VIs using up to 8 MB of memory
(then the limit on a Macintosh II). The mega-applications took up to 30 min-
utes to load into memory. Users reeled at the nefarious Too many objects
error message. And members of the development team often shuddered
when they thought of the huge number of allocated structures and the
complexity of their interconnection needed to make such a system work.

The decision to redesign LabVIEW brought with it a new set of pres-

sures. In responding to some of the customer demands, the company had
to admit that version 1.2 was at its design limits. As a result, work began
on a new version, already becoming known as LabVIEW 2. Once the word
was out, the development team was pressured into predicting a release
date. Despite Kodosky’s best intentions, the scope of the redesign resulted
in several missed shipping deadlines. He realized that the design of a
hierarchical dataflow compiler with polymorphic functions and sophisti-
cated memory management was a new science—and it took awhile.

LabVIEW 2 was designed with formalized object-oriented programming

(OOP) techniques, on the insistence of Jeff Parker. OOP has many advan-
tages over common procedural languages, and in fact it was an enabling
technology in this case. Unfortunately, OOP tools for C language develop-
ment were in a rather primitive state in the 1988 time frame, so the team
wrote their own spreadsheet-based development tools to automatically
generate source code files and to keep track of objects and methods. These
tools remain in use because they are more efficient than C++ in terms of
code size and performance, though Kodosky reports that C++ is under
serious consideration now for reasons of future support and portability.

The development team released an alpha version of the software in late

1988. Cosmetically, this version appeared to be in excellent shape. The
compiler was working well and showed great increases in performance.
There were also many enhancements in the editing capabilities of the
product. All these positive signs resulted in an air of excitement and of
imminent release. Unfortunately, the team had a long list of items they
knew had to be fixed to produce a technically sound product. Over a year
elapsed before the team released the final product. In January 1990,
LabVIEW 2 shipped to the first eager customers. I have to say that being
a beta tester was a real thrill: the improvement in speed and flexibility
was astounding.

LabVIEW 2’s compiler is especially notable not only for its performance but
for its integration into the development system. Developing in a standard
programming language normally requires separate compilation and link-
ing steps to produce an executable program. The LabVIEW 2 compiler is
 Roots 23

Figure 1.9 The team that delivered LabVIEW 2 into the hands of
engineers and scientists (clockwise from upper left): Jack Barber,
Karen Austin, Henry Velick, Jeff Kodosky, Tom Chamberlain, Debo-
rah Batto, Paul Austin, Wei Tian, Steve Chall, Meg Fletcher, Rob
Dye, Steve Rogers, and Brian Powell. Not shown: Jeff Parker, Jack
MacCrisken, and Monnie Anderson.

an integral and invisible part of the LabVIEW system, compiling diagrams

in a fraction of the time required by standard compilers. From a user’s
point of view, the compiler is so fast and invisible that LabVIEW 2 is every
bit as interactive as the previous interpreted versions. (See Figure 1.9.)

The port to Windows and Sun

The next major quest in the development of LabVIEW was the portable,
or platform-independent version. Dr. Truchard (and thousands of users)
had always wanted LabVIEW to run on the PC, but until Windows 3.0
came along, there was little hope of doing so because of the lack of 32-bit
addressing support that is vital to the operation of such a large, sophisti-
cated application. UNIX workstations, on the other hand, are well suited to
such development, but the workstation market alone was not big enough
to warrant the effort required. These reasons made Kodosky somewhat
resistant to the whole idea of programming on the PC, but MacCrisken
finally convinced him that portability itself—the isolation of the machine-
dependent layer—is the real challenge. So, the port was on.

Microsoft Windows 3 turned out to be a major kludge with regard to 32-bit

applications (Windows 3, itself, and DOS are 16-bit applications). Only in
Appendix E of the Windows programming guide was there any mention
whatsoever of 32-bit techniques. And most of the information contained
24 Chapter One

in that appendix referred to storage of data and not applications. Finally,

only one C compiler—Watcom C—was suited to LabVIEW development.
But before Watcom C became available, Steve Rogers created a set of glue
routines that translates 32-bit information back and forth to the 16-bit
system function calls (in accordance with Appendix E). He managed to
successfully debug these low-level routines without so much as a symbolic
debugger, living instead with hexadecimal dumps. This gave the develop-
ment team a six-month head start. Rogers summed up the entire situa-
tion: “It’s ugly.” Development on the Sun SPARCstation, in contrast, was a
relative breeze. Like all good workstations, the Sun supports a full range
of professional development tools with few compromises—a programmer’s
dream. However, the X Windows environment that was selected for the
LabVIEW graphical interface was totally different from the Macintosh or
Windows toolbox environments. A great deal of effort was expended on the
low-level graphics routines, but the long-term payoff is in the portability
of X Windows–based programs. Development on the Sun was so conve-
nient, in fact, that when a bug was encountered in the Windows version,
the programmer would often do his or her debugging on the Sun rather
than suffering along on the PC. Kodosky reports, “The Sun port made
the PC port much easier and faster.” LabVIEW 2.5, which was released
in August 1992, required rewriting about 80 percent of LabVIEW 2 to
break out the machine-dependent, or manager, layer. Creating this man-
ager layer required some compromises with regard to the look and feel of
the particular platforms. For instance, creating floating windows (such as
the LabVIEW Help window) is trivial on the Macintosh, difficult on the
PC, and impossible under X Windows. The result is some degree of least-
common-denominator programming, but the situation has continued to
improve in later versions through the use of some additional machine-
dependent programming. (See Figure 1.10.)

LabVIEW 3
The LabVIEW 2.5 development effort established a new and flexible
architecture that made the unification of all three versions in LabVIEW
3 relatively easy. LabVIEW 3, which shipped in July of 1993, included a
number of new features beyond those introduced in version 2.5. Many
of these important features were the suggestions of users accumulated
over several years. Kodosky and his team, after the long and painful port,
finally had the time to do some really creative programming. For instance,
there had long been requests for a method by which the characteristics of
controls and indicators could be changed programmatically. The Attribute
Node addressed this need. Similarly, Local Variables made it possible to
both read from and write to controls and indicators. This is an extension
of strict dataflow programming, but it is a convenient way to solve many
tricky problems. Many subtle compiler improvements were also made that
enhance performance, robustness, and extensibility. Additional U.S. pat-
ents were issued in 1994 covering these extensions to structured dataflow
diagrams such as globals and locals, occurrences, attribute nodes, execu-
tion highlighting, and so forth.
 Roots 25

Figure 1.10 “Some new features are so brilliant that eye protection is recommended when view-
ing.” The team that delivered LabVIEW 2.5 and 3.0 includes (left to right, rear) Steve Rogers, Thad
Engeling, Duncan Hudson, Keving Woram, Greg Richardson, Greg McKaskle; (middle) Dean Luick,
Meg Kay, Deborah Batto-Bryant, Paul Austin, Darshan Shah; (seated) Brian Powell, Bruce Mihura.
Not pictured: Gregg Fowler, Apostolos Karmirantzos, Ron Stuart, Rob Dye, Jeff Kodosky, Jack
MacCrisken, Stepan Riha.

LabVIEW 3 was a grand success, both for National Instruments and

for the user community worldwide. Very large applications and systems
were assembled containing thousands of VIs. The LabVIEW Application
Builder permitted the compilation of true stand-alone applications for dis-
tribution. Important add-on toolkits were offered by National Instruments
and third parties, covering such areas as process control, imaging, and
database access. Hundreds of consultants and corporations have made
LabVIEW their core competency; speaking as one of them, the phones just
keep on ringing.

LabVIEW 4
Like many sophisticated software packages, LabVIEW has both ben-
efited and suffered from feature creep: the designers respond to every
user request, the package bulks up, and pretty soon the beginner is over-
whelmed. April 1996 brought LabVIEW 4 to the masses, and, with it,
some solutions to perceived ease-of-use issues. Controls, functions, and
tools were moved into customizable floating palettes, menus were reor-
ganized, and elaborate online help was added. Tip strips appear when-
ever you point to icons, buttons, or other objects. Debugging became much
more powerful. And even the manuals received an infusion of valuable
new information. I heard significant, positive feedback from new users on
many of these features.
26 Chapter One

LabVIEW 4 enabled the assembly of very large applications consisting

of thousands of VIs, including stand-alone executables, courtesy of the
LabVIEW Application Builder. Most important, National Instruments
realized that LabVIEW was a viable development platform with regard to
creating new features and even extending LabVIEW itself. The first such
extension was the DAQ Channel Wizard, a fairly sophisticated application
that provided a comprehensive user interface for managing channel con-
figurations on plug-in boards. Development time was drastically reduced
compared to conventional C-based coding. Audrey Harvey began the Chan-
nel Wizard development in LabVIEW, showing key features in prototypical
fashion. When it became apparent that her prototype met most of the per-
ceived needs of the user community, it was handed off to Deborah Batto-
Bryant, who perfected it for release with LabVIEW 4.0. This strategy has
since been followed many times. These LabVIEW-based features, when
well designed, are indistinguishable from their C-based counterparts. The
underlying plumbing in LabVIEW 4 was not significantly changed from
LabVIEW 3, though some advanced R&D activities were certainly under
way. The team called the next generation LabVIEW 10, referring to some
distant, future release, and split off a special group dedicated to new fea-
tures such as undo, multithreading, and real-time capability. While some
of these features (including a simple form of undo) almost made it into
LabVIEW 4.1, they had to wait until the next version.

LabVIEW branches to BridgeVIEW

National Instruments decided to target the industrial process control
market—initially with LabVIEW extensions—and ended up launching a
new product, BridgeVIEW. It’s important to make a distinction between
LabVIEW—the development environment—and G, the underlying pro-
gramming language. The two are indeed separable. BridgeVIEW is an
independent development environment that uses the G language and com-
piler, but is optimized for process control applications. Primary differences
between the two products include some changes in the menus, the inclu-
sion of a real-time process control database, extensive system configuration
tools, a new driver concept, and a reduction in the amount of programming
required to create a substantial process control application. The result is
a ready-to-use process control package that you can freely extend with G
language diagrammatic programming. This contrasts sharply with other
process control packages that do what they do, and no more. LabVIEW
users with a control application will find BridgeVIEW very comfortable.

LabVIEW 5
For LabVIEW 5, Jeff Kodosky’s team adopted completely new development
practices based on a formalized, milestone-driven scheme that explicitly
measured software quality. (See Figure 1.11.) Bugs were tracked quan-
titatively during all phases of the development cycle, from code design
to user beta testing. The result, in February 1998, was the most reliable
LabVIEW release ever, in spite of its increased complexity.
 Roots 27

October 2005 LabVIEW 8.0

May 2005 LabVIEW Embedded
June 2003 LabVIEW FPGA
May 2003 LabVIEW PDA
April 2003 LabVIEW 7.0
August 2000 LabVIEW 6.0
May 1999 LabVIEW RT for realtime applications
February 1998 LabVIEW 5.0—multithreading; undo
December 1996 BridgeVIEW for process control
March 1996 LabVIEW 4.0—application builder
March 1995 National Instruments goes public
October 1993 LabVIEW 3.0 for Windows NT
January 1993 LabVIEW 2.5.1 for Sun
August 1992 LabVIEW 2.5 for Windows
January 1990 LabVIEW 2.0—the first graphical compiler

October 1986 LabVIEW 1.0 for Macintosh

April 1983 LabVIEW project begins

Figure 1.11 The LabVIEW development milestones.

The quality improvement process began with a thorough scrubbing of

LabVIEW 4—a laborious but necessary task—prior to adding new features.
Then the fruits of the prior LabVIEW 10 efforts and other advanced concepts
were incorporated. Ultimately, a good deal of the application was rewritten.

Undo was positively the most-requested feature among users, all the way
back to the dawn of the product. But it was also considered one of the
most daunting. Behind the scenes, LabVIEW is an extremely complex col-
lection of objects—orders of magnitude more complicated than any word
processor or drawing program. The simpleminded approach to implement-
ing undo is simply to duplicate the entire VI status at each editing step.
However, this quickly leads to excessive memory usage, making only a
single-level undo feasible. Instead, the team devised an incremental undo
strategy that is fast, memory efficient, and supports multiple levels. It’s
so unique that it’s patented. Of course, it took a while to develop, and it
involved a complete rewrite of the LabVIEW editor. That was step 1.

Step 2 required rewriting the LabVIEW execution system to accommo-

date another advanced feature: multithreaded execution. Applications
that support multiple threads use the scheduling capabilities of the com-
puter’s operating system to divide CPU time between various tasks, or
threads. For instance, the user interface thread can run independently of
the data acquisition thread and the data analysis thread, and all of those
threads can have different priorities. While the potential performance
enhancements are great, multithreading has traditionally been a fairly
difficult programming challenge tackled only by skilled programmers. The
LabVIEW 5 team brought multithreaded development to the masses by
providing a simple, low-risk interface.
28 Chapter One

As any programmer can tell you, the simpler a feature appears, the more
complex is its underlying code—and that certainly describes the conver-
sion of LabVIEW to multithreading. The major task was to evaluate every
one of the thousands of functions in LabVIEW to determine whether it was
thread-safe, reentrant, or needed some form of protection. This evaluation
process was carried out by first writing scripts in PERL (a string manipu-
lation language) that analyzed every line of code in the execution system,
and then having the programmers perform a second (human) evaluation.
As it turned out, about 80 percent of the code required no changes, but
every bit had to be checked. Kodosky said the process was “. . . like doing
a heart and intestine transplant.”

The third and final step in the Great Rewrite involved the compiler. As you
know by now, LabVIEW is unique in the graphical programming world in
that it directly generates executable object code for the target machine.
Indeed, that compiler technology is one of the crown jewels at National
Instruments. At the time, one of the latent shortcomings in the compiler
was the difficulty involved in supporting multiple platforms. Every time a
new one was added, a great deal of duplicated effort was needed to write
another platform-specific compiler process. The solution was to create a
new platform-independent layer that consolidated and unified the code
prior to object code generation.

Now adding a new platform involves only the lowest level of object code
generation, saving time and reducing the chances for new bugs. It also
reduced the code size of the compiler by several thousand lines. It’s
another of those underlying technology improvements that remains invis-
ible to the user except for increased product reliability.

The LabVIEW RT branch

As long as any of us can remember, we’ve been asking for enhanced realtime
performance from LabVIEW (this topic is covered in detail in Chapter 21,
“LabVIEW RT Does Real Time”). It’s also been a hot topic of discussion
for the LabVIEW team as well, and they finally came up with a solution
that was released in the form for LabVIEW RT in May 1999. After years
of discussion and conceptual design that hadn’t completely gelled into a
solution, Jeff Kodosky offered a personal challenge to Darshan Shah: “I
bet you can’t!” Needless to say, Darshan was up to the challenge.

LabVIEW RT required a number of innovations. First, a specialized real-

time operating system (RTOS) was needed to provide faster and more pre-
dictable scheduling response than a desktop OS. RTOSs often lack many
of the services that standard LabVIEW requires, and they must be config-
ured for a specific hardware configuration. Darshan chose PharLap TNT
Embedded because it supported many Windows Win32 API (application
programmer interface) calls, which made it much easier to port LabVIEW
as well as NI-DAQ. To simplify hardware compatibility problems, the first
release of LabVIEW RT ran only on a special DAQ board married to an
 Roots 29

80486 processor, which was chosen for its code compatibility and adequate
performance at low cost. It was a reasonable engineering decision for an
initial product release.

A second challenge was creating a simple and intuitive way of interacting

with the embedded LabVIEW RT processor board. Darshan’s solution was
to provide a simple configuration dialog that tells LabVIEW to compile
and run on either the host machine or the embedded board. This com-
pletely shields the user from the underlying platform change and all the
real-time complexity that one would normally have to face. Another ben-
efit of this architecture was that its development was synergistic with the
other new LabVIEW 5 concepts: the division between the user interface
and execution systems, and multithreading.

LabVIEW 6
June 2000 brought us LabVIEW 6, with a host of new user-interface
features (rendered in 3D, no less), extensive performance and memory
optimization at all levels, and perhaps most important, a very powerful
VI Server. Introduced in LabVIEW 5, the VI Server gives the user exter-
nal hooks into VIs. For instance, you can write programs that load and
run VIs, access VI documentation, and change VI setup information, all
without directly connecting anything to the target VI. Furthermore, most
of these actions can be performed by applications other than LabVIEW,
including applications running elsewhere on a network. This new para-
digm effectively publishes LabVIEW, making it a true member of an
application system rather than exclusively running the show as it had in
the past. You can even export shared libraries (DLLs) in LabVIEW. From
the perspective of professional LabVIEW programmers, we believe that
LabVIEW 6 has truly fulfilled the promise of graphical programming.

At the nuts-and-bolts level, LabVIEW 6 is now completely developed

in C++, in contrast to all the earlier versions, which used C with object
extensions created by the LabVIEW team. They stuck with C for so long
because C++ compiler effectiveness was, in their judgment, not as good
as C. As a point of reference, LabVIEW 6 consists of about 800,000 lines
of code. While that’s a pretty big number, there are plenty of applications
running on your very own computer that are much heftier (including NI-
DAQ, but that’s another story). Also, the growth rate in LabVIEW’s code
size has actually diminished over the years due to improved programming
techniques. It’s always impressive to see continuous improvement in an
application’s performance and efficiency, rather than pure bloat as new
features are added (see Figure 1.12).

LabVIEW 7
LabVIEW 7 Express was released in April of 2003 with new features and
wizards designed to make programming and getting started easier for
30 Chapter One

Figure 1.12 My, how we have grown! This is most of the LabVIEW 6 team.

people new to LabVIEW. Express blocks make configuring instruments,

testing analysis routines, and most common tasks simple. The DAQ Assis-
tant made data acquisition easier for everyone. Now with a few clicks of
a mouse anyone can create a fully functioning data acquisition program.
You’re not limited to simple tasks either; the DAQ Assistant makes com-
plex trigger applications a snap. The LabVIEW team went out of their
way to help the novice with new, simplified ways of doing things without
fatally compromising the power of LabVIEW.

LabVIEW 7 also brings LabVIEW to new platforms far from the desktop.
Targets include a variety of compact computers and operating systems,
including Windows CE, Palm OS, and the ARM processor family. New in
LabVIEW is a compiler for field-programmable gate arrays (FPGAs).
LabVIEW for FPGA continues National Instruments’ push of LabVIEW
Everywhere. This is more than just marketing hype. With LabVIEW
FPGA, engineers are now able to convert LabVIEW VIs into hardware
implementations of G code.
 Roots 31

LabVIEW 8
Released in October 2005, LabVIEW 8 introduced new tools to make
LabVIEW developers more productive and application development and
integration across a wide range of platforms easier. The LabVIEW proj-
ect provides a cohesive environment for application developers. Finally, at
long last, we can simultaneously develop applications for multiple targets.
Developers using LabVIEW FPGA and RT can develop applications on the
host and the target without having to close one environment and switch to
another. The project interface provides a relatively painless way to manage
development and deployment of large applications by many developers.
LabVIEW is the number one programming language in test and measure-
ment and it’s time we had some big time tools to manage our projects.
Look for more object oriented programming as LabVIEW includes built
in OOLV tools. National Instruments continues to take LabVIEW deeper
into education and simulation with the educational version of LabVIEW
for DSPs and the new LabVIEW MathScript. MathScript is an integral
part of LabVIEW that combines dataflow with text-based mathematical
programming. And as shown in Figure 1.13, LabVIEW keeps growing.

Figure 1.13 LabVIEW keeps growing and so does the development team. This is the team circa
2003. Photo courtesy of National Instruments Corporation.
32 Chapter One

Crystal Ball Department

Jeff Kodosky’s predictions in the last edition proved to be amazingly
accurate, so we asked him to share some rumors regarding future
trends and features for LabVIEW. Here’s Jeff:
Probably the most important future goals fall in the area of fleshing out
LabVIEW’s design capabilities. We have always had a contingent of cus-
tomers doing system design using LabVIEW but the strongest area for
us was virtual instrumentation (“acquire-analyze-present”). As LabVIEW
has grown, particularly with the introduction of LabVIEW Real Time and
more recently LabVIEW FPGA, it has become much more suitable for
hard realtime control applications. But control applications typically are
developed in conjunction with modeling or simulating the system under
control, so we have been introducing toolkits for system identification and
control design and simulation, etc., and all these design capabilities fit
superbly well with LabVIEW’s graphical dataflow paradigm. There are
also other “models of computation” that are important for many types of
design and we have several integrated into LabVIEW already (e.g., state
diagrams, text math in the formula box, event structure, network vari-
able) and others that we are continuing to develop (e.g., synchronous data
flow, the next Lego “Robolab,” other specialized models specific to appli-
cation areas). One might even consider a circuit diagram to be a model
of computation, so to that end we are exploring much tighter coupling
between LabVIEW and Electronics WorkBench Multisim.

Our current focus is nicely summarized by our new description: design-

prototype-deploy, where LabVIEW is ideally suited to all aspects of the
process from design, to prototype (in conjunction with platforms like PXI
and cRIO), to deployment (on the same platforms for low volumes or cus-
tom platforms for high volumes).

In the past, LabVIEW was typically the tool for testing the widgets being
produced, but now it is also helping design the widget and may even be
part of the widget.

LabVIEW influences other software products

The concepts found in LabVIEW have influenced the design of many
other instrumentation products, especially in the areas of software front
panels and iconic programming. Such products exist for Macintosh,
DOS, Windows, and UNIX-based computers. Although at first glance
these products are easily confused with LabVIEW, the differences
are many.
As LabVIEW does, most instrumentation software products now offer
some form of front panel–oriented user interface. The role of a Lab-
VIEW front panel is unique, however, in comparison to other products.
A graphical front panel in most instrumentation software products is
simply a display/user-interface mechanism introduced into an otherwise
 Roots 33

conventional, textual language–based programming system. National

Instruments’ LabWindows CVI is such a package. The sole purpose
of the panel in such a system is to serve as a top-level interface to an
instrument or plug-in board. In some packages, users don’t have the
flexibility to customize their own front panel; it’s merely a user-friendly
tool supplied by the vendor. These panels are either not applicable to or
not available as an interface for use within an application program.
In LabVIEW, a front panel is an integral component of every software
module (or VI). A LabVIEW front panel can serve as the interface to
any type of program, whether it is a GPIB instrument driver program,
a data acquisition program, or simply a VI that performs computations
and displays graphs. The methodology of using and building front pan-
els is fundamental to the design and operation of LabVIEW; for every
front panel, there is an associated block diagram program and vice
versa. Front panels can appear at any level in a LabVIEW VI and are
not restricted to the highest level.
Another aspect of LabVIEW that finds its way into other software
products is iconic programming. As with front panels, the iconic pro-
gramming systems found in other products are largely add-on shells
used to hide some of the details of systems based on conventional
programming methodologies. Such products provide a predefined, set
number of operations that can be used in iconic form. While generally
quite simple to use because of their turnkey nature, these systems are
not readily expanded because adding new functions is either impos-
sible or requires a user to revert back to a standard programming lan-
guage to create new icons. These iconic programming systems address
organization of the high-level aspects of a test, but do not have graphi-
cal constructs for specifying programming operations such as execu-
tion order, iteration, or branching. In addition, these systems typically
do not support building a module that can be used in iconic form at
a higher level. Finally, and perhaps most importantly, LabVIEW is a
compiled language, whereas all other iconic packages are interpreted.
A 100-fold increase in performance is typically demonstrated.
LabVIEW’s iconic programming language is not simply an organiza-
tional tool. It is a true programming language complete with multiple
data types, programming structures, and a sophisticated compiler.
LabVIEW is unique for its hierarchy and expandability. VIs have the
same construction at all levels, so the creation of a new high-level VI
is simply a matter of combining several lower-level VIs and putting
a front panel on the new combination. Because LabVIEW’s graphical
programming system offers the functionality and performance of con-
ventional programming languages, users can implement and add new
functions to the system without having to resort to the use of another
programming language.
34 Chapter One

LabVIEW Handles Big Jobs

Some really impressive applications have been written in all versions
of LabVIEW, but in the past few years we’ve seen an absolute explosion
in development that leaves us breathless. Even the LabVIEW team
members have been known to shake their heads in amazement when
shown the size and scope of the latest high-end project from a sophisti-
cated user. Here are a few benchmark projects that we’ve run across.
At the Lawrence Livermore National Laboratory, Gary was
responsible for the LabVIEW software architecture that controlled a
femtosecond laser cutting system. This unique and very complex system
of lasers was developed at a cost of $20 million for a high-value defense
application. Four Windows NT workstations were networked together,
with LabVIEW providing measurement, control, and automatic seq-
uencing for the laser itself, as well as a high-level user interface for
semiskilled operators. Nearly every type of I/O hardware was installed:
DAQ, GPIB, serial, remote SCXI, VME/VXI, and IMAQ imaging. For-
mal software quality assurance procedures were applied because of the
mission-critical nature of the project. A total of about 4 work-years of
software development went into the creation of about 600 VIs.
At Spectrum Astro, Inc., Roger Jellum and Tom Arnold developed
AstroRT, a LabVIEW-based data acquisition and control system for
factory test and on-orbit operations of spacecraft. Now marketed as a
commercial package, AstroRT has been used for many missions, includ-
ing the Air Force Research Lab MightySat II.1 spacecraft, NASA’s
New Millennium Deep Space 1, TDRSS and Swift spacecraft, and the
Naval Research Lab’s Coriolis spacecraft. Running under Windows
NT, AstroRT provides the capability to collect, process, and distribute
frame-synchronized or packetized telemetry received from a spacecraft
or subsystem. The system can also transmit commands to the space-
craft, interface with ground-support equipment (typically via GPIB,
VXI, or Ethernet interfaces), display and archive data, and assist with
mission planning. The whole package encompasses more than 2000 VIs
and requires about 7 work-years to develop.
Perhaps the grandest LabVIEW application of all is MXProLine,
developed by a team lead by Dirk Demol at Honeywell-Measurex
Corporation. Measurex provides process control equipment and sys-
tems for the production of sheet goods, particularly paper and plastic
films. MXProLine is a state-of-the-art distributed process control sys-
tem with LabVIEW representing about 95 percent of the code base (See
Figure 1.14). Dirk’s team worked closely with National Instruments
for many years to recommend enhancements to LabVIEW that would
enable Honeywell-Measurex to implement this very complex software
system with graphical programming. Significant obstacles had to be
 Roots 35

Figure 1.14 MXProLine from Honeywell-Measurex is preconfigured with many user-interface

screens such as this. It’s probably the most elaborate LabVIEW application on the planet.

overcome, particularly in the areas of system configuration, initializa-

tion, and real-time network-transparent data distribution. These chal-
lenges were met by writing an external, real-time database in C++. It
features very low overhead (microsecond access times), a hierarchical
organization that follows the software organization as well as that
of the physical plant, and the ability to share data with applications
outside of LabVIEW. The package is data-driven to the point that the
system can be fully configured without any LabVIEW programming.
International installations can even switch languages on the fly! A typ-
ical MXProLine system can be configured to handle more than 100,000
variables (both physical I/O and computed values), with many operator
workstations and I/O subsystems spread out over a plantwide network.
More than 5000 VIs were written for this project. We know of no other
LabVIEW-based system of this scale, and we’re quite frankly blown
away.
This page intentionally left blank
 Chapter

Getting Started
2
Make no mistake about it: LabVIEW is a programming language, and a
sophisticated one at that. It takes time to learn how to use it effectively.
After you buy the program, you should go through the tutorial man-
ual, and then hopefully you’ll take the basic LabVIEW class, and later
the advanced LabVIEW class; even though it costs money and takes
time, it’s worth it. Ask any of the MBAs in the front office; they know
the value of training in terms of dollars and cents as well as worker
productivity. Remember to keep working with LabVIEW; training goes
stale without practice. There is a great deal of information to absorb.
It’s good to read, experiment, and then read some more. For the pur-
poses of this book, we assume that you’ve been working with LabVIEW
and know the basic concepts.
Another thing you should do is to spend plenty of time looking at the
example VIs that come with the LabVIEW package. Collectively, the
examples contain about 80 percent of the basic concepts that you really
need to do an effective job, plus a lot of ready-to-use drivers and special
functions. The rest you can get right here. Remember that this book is
not a replacement for the user’s manual. Yes, we know, you don’t read
user’s manuals either, but it might be a good idea to crack the manu-
als the next time you get stuck. Most folks think that the manuals are
pretty well written. And the online help fills in details for every pos-
sible function. What we cover in this book are important concepts that
should make your programming more effective.

About the Diagrams in This Book

It’s hard to present LabVIEW through printed media. After all, it’s
an interactive graphical language, and the static nature of paper
can obscure the workings of the underlying structure. In this book,

37
38 Chapter Two

Sequence structures
0 [0..2] 0 [0..2] 1 [0..2] 1 [0..2]

Case structures

True True False

Figure 2.1A standard for printing Sequence and Case structures. This is
but one compromise we must make when displaying LabVIEW programs
on paper.

we’ve used some accepted conventions regarding images of LabVIEW

diagrams and panels. One such convention is shown in Figure 2.1.
Sequence structures and Case structures appear on the screen as single
frames, but may in fact have several hidden frames, or subdiagrams.
You view the subdiagrams by operating the control bar at the top. It’s
too bad you can’t click on these pages. Instead, we’ll just spread out the
subdiagrams in a logical manner that should not cause too much con-
fusion. For details on how you, too, can print and otherwise document
LabVIEW programs, refer to the end of Chapter 9 under “Printing Lab-
VIEW Panels and Diagrams.”

Sequencing and Data Flow

Regular textual programming languages (and most computers) are
based on a concept called control flow, which is geared toward
making things happen one step at a time with explicit ordering. The
LabVIEW language, G, on the other hand, is known formally as a
dataflow language. All it cares about is that each node (an object
that takes inputs and processes them into outputs) have all its inputs
available before executing. This is a new way of programming—one
that you have to get used to before declaring yourself a qualified Lab-
VIEW programmer. Although programming with dataflow is easy, it
can be constraining because you have to operate within the confines of
the dataflow model. This constraint also forces you to write programs
that function well. Whenever you break from the dataflow model,
you risk writing a program that misbehaves. The dataflow concept
should be used to its best advantage at all times. This section dis-
cusses some of the concepts of dataflow programming with which you
should be familiar.
 Getting Started 39

Part of the CPU Instructions in memory

GET INPUT A Program Counter GET A

GET INPUT B
PROCESS A AND B GET B
DISPLAY RESULT PROCESS

DISPLAY

How a procedural language How your computer thinks (if you call this
tells the computer what to do thinking) when programmed with control flow

GET
INPUT A PROCESS DISPLAY
A AND B RESULT
GET
INPUT B

Same problem, this time using dataflow. Notice that

the GET INPUT functions can occur simultaneously,
provided that the computer has such capacity. A
LabVIEW diagram is almost identical to this.

Figure 2.2Compare and contrast control flow programming in a procedural

language with dataflow programming in LabVIEW.

Figure 2.2 compares ordinary procedural programming (such as

Basic or C) with dataflow programming. In all cases, the PROCESS A
AND B step can execute only after GET A and GET B are completed.
In the procedural language, we have forced GET A to happen before
GET B. What if the source of data for B was actually ready to go before
A? Then you would end up wasting time waiting for A. The dataflow
version says nothing about whether GET A or GET B must go first. If
the underlying code (courtesy of the LabVIEW compiler) is intelligent
enough, the two GET tasks will overlap in time as required, enhancing
throughput for this imaginary system.
LabVIEW permits you to have any number of different nodes on a
diagram—all executing in parallel. Furthermore, the LabVIEW envi-
ronment supports parallel execution, or multitasking, between multiple
VIs, regardless of the capability of the operating system or computer.
These capabilities give you great freedom to run various tasks asyn-
chronously with one another without doing any special programming
yourself. On the other hand, you sometimes need to force the sequence
of execution to guarantee that operations occur in the proper order.

LabVIEW under the Hood

Before we go too far, let’s pop open the hood and take a look at how
LabVIEW operates. This is an advanced topic and one not essential for
40 Chapter Two

writing LabVIEW programs, but it helps you write better programs if

you have an idea of what is going on. We’re going to cover just the high-
lights. If you want greater in-depth coverage, there are several excellent
presentations and applications notes on National Instruments’ (NI’s)
Web site: AN 114, Using LabVIEW to Create Multithreaded VIs for
Maximum Performance and Reliability; AN 199, LabVIEW and Hyper-
Threading; and Steve Roger’s “Inside LabVIEW” presentation from
NIWeek 2000. Incidentally, if you want to really understand LabVIEW,
attend NIWeek. It is well worth your time.

The parts of a VI
We see VIs as front panels and block diagrams, but there is a lot that
you don’t see. Each VI is a self-contained piece of LabVIEW software
with four key components (Figure 2.3):

■ Front panel. The front panel code contains all the resources for
everything on the front panel. This includes text objects, decorations,
controls, and indicators. When a VI is used as a subVI, the front panel
code is not loaded into memory unless the front panel is open or the
VI contains a Property node.
■ Block diagram. The block diagram is the dataflow diagram for the
VI. The block diagram is not loaded into memory unless you open the
block diagram or the VI needs to be recompiled.
■ Data. The VI data space includes the default front panel control and
indicator values, block diagram constants, and required memory buf-
fers. The data space for a VI is always loaded into memory. Be careful

Parts of a VI

Data
Always Loaded

Front Block
Code
Panel Diagram

VI code and data are always loaded into

Figure 2.3
memory when the VI is loaded.
 Getting Started 41

when you make current values default on front panel controls or use
large arrays as block diagram constants; all that data is stored on
disk in your VI.
■ Code. This is the compiled machine code for your subVI. This is
always loaded into memory. If you change platforms or versions of
LabVIEW, the block diagram will be reloaded and the code will be
recompiled.

Each VI is stored as a single file with the four components above plus
the linker information about the subVIs, controls, .dll, and external
code resources that the VI needs. When a VI is loaded into memory, it
also loads all its subVIs, and each subVI loads its subVIs, and so on,
until the entire hierarchy is loaded.

How VIs are compiled

There’s an elegant term that the LabVIEW developers use for the
code generated from your block diagram—it’s called a clump. When
you push the Run button on a new VI, LabVIEW translates the block
diagram into clumps of machine code for your platform. The G com-
piler makes several passes through the block diagram to order the
nodes into clumps of nodes. Execution order within each node is fixed
at compile time by dataflow. As the compiler makes its passes, it
determines which data buffers can be reused. Then it compiles the
clumps into machine code. Here’s an example based on Steve Roger’s
presentation: The block diagram in Figure 2.4 gets converted to three
clumps of code. The first clump reads the controls and schedules the
other two clumps for execution; then it goes to sleep. The other two
clumps, one for each For Loop, run to completion, update their indi-
cators, and reschedule the first clump for execution. The first clump
then finishes the diagram, displays the data, and exits the VI. The
traffic cop responsible for running the clumps of code is the LabVIEW
execution engine.

Multitasking, multithreaded LabVIEW

The LabVIEW execution engine runs each clump of code by using coop-
erative multitasking. The execution engine time slices each clump to
make sure all the clumps get a chance to run. Each clump (in Figure
2.4, For Loop 1 or For Loop 2 ) either runs to completion or is switched
out by the execution engine, and another clump runs. The execution
engine proceeds on multitasking each clump of code until either your
program ends or someone pushes the Abort button, which is always
a nasty way to stop a program! Cooperative multitasking has been
42 Chapter Two

Figure 2.4 LabVIEW converts dataflow diagrams

to “clumps” of machine code.

successfully used by LabVIEW since its birth, and it is still the process
being used today. Beginning with LabVIEW 5, we’ve seen its effective-
ness multiplied through multithreading. A thread is an independent
program within a program that is managed by the operating system.
Modern operating systems are designed to run threads efficiently,
and LabVIEW receives a noticeable improvement from multithread-
ing. Each LabVIEW thread contains a copy of the LabVIEW execution
engine and the clumps of code that the engine is responsible for run-
ning. Within each thread the execution engine still multitasks clumps
of code, but now the operating system schedules where and when each
thread runs. On a multiprocessor system, the operating system can
even run threads on different processors in parallel. Today LabVIEW
on all major platforms is multithreaded and multitasking and able to
take advantage of multiprocessor systems. Figure 2.5 illustrates the
multiple threads within LabVIEW’s process. The threads communicate
via a carefully controlled messaging system. Note that one thread is
dedicated to the user interface. Putting the user interface into a sepa-
rate thread has several advantages:

■ Decoupling the user interface from the rest of your program allows
faster loop rates within your code.
 Getting Started 43

LabVIEW Process
Thread
Code
+
Thread
Thread Execution
Code
Engine +
Thread
Messages Execution
Code
User Engine
Thread +
Interface Code
Execution
+
Engine
Execution
Engine

Figure 2.5 Multithreading splits the LabVIEW

process into multiple threads with one thread
dedicated to the user interface. Each thread runs
clumps of code with its own copy of the LabVIEW
execution engine. Messages and other protected
sharing mechanisms are used to communicate
between threads.

■ The user interface doesn’t need to run any faster than the refresh
rate on most monitors—less than 100 Hz.
■ Decoupling the user interface from the rest of the program makes
it possible to run the user interface on a separate computer. This is
what happens in LabVIEW RT.

What does all this mean? It means the two For Loops in Figure 2.4 can
run in parallel and you don’t have to do anything other than draw them
that way. The LabVIEW compiler generates code that runs in parallel,
manages all the resources required by your program, spreads the execu-
tion around smoothly, and does it all transparently. Imagine if you had
to build all this into a test system yourself. National Instruments has
spent more than 20 years making sure LabVIEW is the best program-
ming environment in test and measurement. You don’t need to know
the details of the execution engine or multithreading to program in
LabVIEW, but we think it will make you a better programmer. Stream-
line your block diagram to take advantage of dataflow and parallelism,
and avoid littering your block diagram with user-interface code.

The LabVIEW Environment

We have already covered the difference between the front panel and
the block diagram; but just to reiterate, the block diagram is where you
put the nuts and bolts of your application, and the front panel contains
your user interface. This is not going to be an exhaustive introduction
44 Chapter Two

to LabVIEW—read the manuals and go through the tutorials for that.

The rest of this chapter is an overview of LabVIEW and a jumping off
point to other chapters in the book. Throughout this book we’re going
to focus on good programming practices and aspects that impact how
LabVIEW runs your code.

Front panel
The front panel is the user’s window into your application. Take the
time to do it right. Making a front panel easy to understand while
satisfying complex requirements is one of the keys to successful vir-
tual instrument development. LabVIEW, like other modern graphical
presentation programs, has nearly unlimited flexibility in graphical
user-interface design. But to quote the manager of a large graphic
arts department, “We don’t have a shortage of technology, we have
a shortage of talent!” After years of waiting, we can finally refer you
to a concise and appropriate design guide for user-interface design,
Dave Ritter’s book LabVIEW GUI (2001). Some industries, such as the
nuclear power industry, have formal user-interface guidelines that are
extremely rigid. If you are involved with a project in such an area, be
sure to consult the required documents. Otherwise, you are very much
on your own.
Take the time to look at high-quality computer applications and see
how they manage objects on the screen. Decide which features you
would prefer to emulate or avoid. Try working with someone else’s
LabVIEW application without getting any instructions. Is it easy to
understand and use, or are you bewildered by dozens of illogical, unla-
beled controls, 173 different colors, and lots of blinking lights? Observe
your customer as he or she tries to use your VI. Don’t butt in; just
watch. You’ll learn a lot from the experience, and fast! If the user gets
stuck, you must fix the user interface and/or the programming prob-
lems. Form a picture in your own mind of good and bad user-interface
design. Chapter 8, “Building an Application,” has some more tips on
designing a user interface.
One surefire way to create a bad graphical user interface (GUI) is to
get carried away with colors, fonts, and pictures. When overused, they
quickly become distracting to the operator. (Please don’t emulate the
glitzy screens you see in advertising; they are exactly what you don’t
want.) Instead, stick to some common themes. Pick a few text styles
and assign them to certain purposes. Similarly, use a standard back-
ground color (such as gray or a really light pastel), a standard highlight
color, and a couple of status colors such as bright green and red. Human
factors specialists tell us that consistency and simplicity really
are the keys to designing quality man-machine interfaces (MMIs).
 Getting Started 45

Operators will be less likely to make mistakes if the layouts among

various panels are similar. Avoid overlapping controls; multiple over-
lapping controls containing transparent pieces can really slow down
the user interface.
Group logically related controls in the same area of the screen, per-
haps with a surrounding box with a subtly different background color.
You can see this technique on the panels of high-quality instruments
from companies such as Tektronix and Hewlett-Packard. Speaking of
which, those real instruments are excellent models for your virtual
instruments.

Controls
LabVIEW’s controls have a lot of built-in functionality that you should
use for best effect in your application. And of course this built-in func-
tionality runs in the user-interface thread. Quite a bit of display work,
computational overhead, and user-interface functionality has already
been done for you. All LabVIEW’s graphs support multiple scales on
both the X and Y axes, and all charts support multiple Y scales. Adjust
the chart update modes if you want to see the data go from left to right,
in sweep mode, update all at once in scope mode, or use the default
strip chart mode going from right to left. Control properties are all
easily updateable from the properties page. Figure 2.6 shows the prop-
erty page for an OK button. Changing the button behavior (also called
mechanical action) can simplify how you handle the button in your
application. The default behavior for the OK button is “Latch when
released,” allowing users to press the buttons, change their minds, and
move the mouse off the button before releasing. Your block diagram is
never notified about the button press. On the other hand, latch when
pressed notifies the block diagram immediately. Experiment with the
mechanical behavior; there’s a 1-in-6 chance the default behavior is
right for your application.

Property nodes
Property nodes are an extremely flexible way to manipulate the appear-
ance and behavior of the user interface. You create a Property node by
popping up on a control or indicator and choosing Create . . . Prop-
erty Node. This creates a Property node that is statically linked to
the control. You do not need to wire a reference to it for it to access the
control. The disadvantage is that statically linked Property nodes can
only reference front panel objects on their VI. Figure 2.7 shows both
a statically linked and a dynamically linked Property node. With a
control reference you can access properties of controls that aren’t on
46 Chapter Two

Figure 2.6 Control properties are modified through the Property pages.
Boolean buttons have six different mechanical actions that affect your block
diagram.

the local front panel. This is a handy way to clean up a block diagram
and generate some reusable user-interface code within a subVI. Every
control and every indicator has a long list of properties that you can
read or write. There can be multiple Property nodes for any given front
panel item. A given Property node can be resized by dragging at a cor-
ner to permit access to multiple attributes at one time. Each item in
the list can be either a read or a write, and execution order is sequen-
tial from top to bottom. We’re not going to spend a great deal of time
on Property nodes because their usage is highly dependent upon the
particular control and its application. But you will see them through-
out this book in various examples. Suffice it to say that you can change
almost any visible characteristic of a panel item, and it’s worth your
time to explore the possibilities.
There are some performance issues to be aware of with Property
nodes. Property nodes execute in the user-interface thread. Remember,
in our discussion of dataflow and LabVIEW’s execution system, how
 Getting Started 47

Figure 2.7 Statically linked Property nodes can only reference front panel objects
within the scope of their VIs. Dynamically linked Property nodes use a control
reference to access properties of controls that may be in other VIs.

each node depends on having all its data before it can proceed. Indis-
criminately scattering Property nodes throughout your program will
seriously degrade performance as portions of your block diagram wait
while properties execute in the user interface. A general rule of thumb
is to never place Property nodes within the main processing loop(s).
Use a separate UI loop to interact with the front panel. Look at Chap-
ter 3, “Controlling Program Flow,” to see how you can use Property
nodes within the event structure.
Property nodes can increase the memory requirements of your appli-
cation. Property nodes within a subVI will cause the subVI’s front panel
to be loaded into memory. This may or may not be an issue, depending
on what you have on your front panel, or how many subVIs have Prop-
erty nodes on them. Try to limit the number of Property nodes and
keep them contained with only a few VIs. You could even build a reus-
able suite of user-interface VIs by using Property nodes and references
to front panel items.
One property that you want to avoid is the Value property. This
is such a convenient property to use, but it is a performance hog. In
LabVIEW 6 writing a control’s value with the Value property was
benchmarked at almost 200 times slower than using a local variable.
And a local variable is slower than using the actual control. LabVIEW
is optimized for dataflow. Read and write to the actual controls and
indicators whenever possible. You can do incredible things with Prop-
erty nodes; just have the wisdom to show restraint.
48 Chapter Two

Block diagram
The block diagram is where you draw your program’s dataflow diagram.
For simple projects you might need only a few Express VIs. These are
the “Learn LabVIEW in Three Minutes” VIs that salespeople are so
fond of showing. They are great to experiment with, and you can solve
many simple projects with just a few Express VIs. It is really easy to
take measurements from a DAQ card and log it to disk compared to
LabVIEW 2. However, when time comes to build a complete applica-
tion to control your experiment, you’ll probably find yourself scratching
your head and wondering what to do. Read on. In Chapter 3, “Control-
ling Program Flow,” we’ll talk about design patterns that you can apply
to common problems and build on.
A large part of computer programming consists of breaking a com-
plex problem into smaller and smaller pieces. This divide-and-con-
quer technique works in every imaginable situation from designing
a computer program to planning a party. In LabVIEW the pieces are
subVIs, and we have an advantage that other programming environ-
ments don’t—each subVI is capable of being run and tested on its own.
You don’t have to wait until the application is complete and run against
massive test suites. Each subVI can, and should, be debugged as it is
developed. LabVIEW makes it easy to iteratively develop, debug, and
make all the pieces work together.

SubVIs
Approach each problem as a series of smaller problems. As you break
it down into modular pieces, think of a one sentence statement that
clearly summarizes the purpose. This one sentence is your subVI. For
instance, one might be “This VI loads data from a series of transient
recorders and places the data in an output array.” Write this down
with the labeling tool on the block diagram before you put down a
single function; then build to that statement. If you can’t write a sim-
ple statement like that, you may be creating a catchall subVI. Also
write the statement in the VI description to help with documenta-
tion. Comments in the VI description will also show up in the context
help window. Refer to Chapter 9, “Documentation,” for more tips on
documenting your VIs. Consider the reusability of the subVIs you cre-
ate. Can the function be used in several locations in your program?
If so, you definitely have a reusable module, saving disk space and
memory. If the subVI requirements are almost identical in several
locations, it’s probably worth writing it in such a way that it becomes
a universal solution—perhaps it just needs a mode control. Refer
to Chapter 8, “Building an Application,” for more help on general
application design.
 Getting Started 49

Icons
Each subVI needs an icon. Don’t rest content with the default Lab-
VIEW icon; put some time and effort into capturing the essence of
what the VI does. You can even designate terminals as being manda-
tory on the connector pane. They will show up in bold on the context
help window. Also create an easily recognizable theme for driver VIs.
Figure 2.8 shows the VI hierarchy for the Agilent 34401 Multimeter.

Figure 2.8 VI hierarchy for Agilent 34401 Multimeter. The banner at the top of each icon gives
the VIs a common theme. Each graphic succinctly captures VI function.
50 Chapter Two

Refer to Chapter 11, “Instrument Driver Development Techniques,” for

more information on building instrument drivers.

Polymorphic VIs
Polymorphic VIs, introduced in LabVIEW 6, allow you to handle mul-
tiple data types with a single VI, thus going beyond the limitations
of LabVIEW’s normal polymorphism. A polymorphic VI is like a con-
tainer for a set of subVIs, each of which handles a particular data type.
All those subVIs must have identical connector panes, but otherwise
you’re free to build them any way you like, including totally differ-
ent functionality. To create a new polymorphic VI, visit the File menu,
choose New . . . , and in the dialog box, pick Polymorphic VI. Up pops
a dialog box in which you can select each of your subVIs. You can also
give it an icon, but the connector pane is derived from the subVI. After
you save it, your new polymorphic VI behaves just as any other VI,
except that its inputs and outputs can adapt to the data types that you
wire to them. This is called compile-time polymorphism because
it’s accomplished before you run the VI. LabVIEW doesn’t really sup-
port polymorphism at run time (something that C++ does), although
crafty users sometimes come up with strange type casting solutions
that behave that way. Almost.

Data
As you write your one-sentence description of each VI, consider the
information that has to be passed from one part of the program to
another. Make a list of the inputs and outputs. How do these inputs
and outputs relate to other VIs in the hierarchy that have to access
these items? Do the best you can to think of everything ahead of time.
At least, try not to miss the obvious such as a channel number or error
cluster that needs to be passed just about everywhere.
Think about the number of terminals available on the connector pane
and how you would like the connections to be laid out. Is there enough
room for all your items? If not, use clusters to group related items.
Always leave a few uncommitted terminals on the connector pane in
case you need to add an item later. That way, you don’t have to rewire
everything.

Clusters
A cluster is conceptually the same as a record in Pascal or a struct
in C. Clusters are normally used to group related data elements
that are used in multiple places on a diagram. This reduces wiring
 Getting Started 51

clutter—many items are carried along in a single wire. Clusters also

reduce the number of terminals required on a subVI. When saved as
a custom control with the typedef or strict typedef option (use the
Customize Control feature, formerly known as the Control Edi-
tor, to create typedefs), clusters serve as data type definitions, which
can simplify large LabVIEW applications. Saving your cluster as a
typedef propagates any changes in the data structure to any code
using the typedef cluster. Using clusters is good programming prac-
tice, but it does require a little insight as to when and where they are
best employed. If you’re a novice programmer, look at the LabVIEW
examples and the figures in this book to see how clusters are used in
real life.
An important fact about a cluster is that it can contain only controls
or indicators but not a mixture of both. This precludes the use of a clus-
ter to group a set of controls and indicators on a panel. Use graphical
elements from the Decorations palette to group controls and indicators.
If you really need to read and write values in a cluster, local variables
can certainly do the job. We would not recommend using local variables
to continuously read and write a cluster because the chance for a race
condition is very high. It’s much safer to use a local variable to ini-
tialize the cluster (just once), or perhaps to correct an errant input or
reflect a change of mode. Rule: For highly interactive panels, don’t use a
cluster as an input and output element.
Clusters make sense for other reasons. For instance, passing the
name and calibration information for a signal along with its data in
one cluster makes a nice unit that clarifies the program. Clustered data
minimizes the number of wires on the diagram, too. Data structures
such as clusters and arrays also make handling of data more efficient
when they are closely coupled to the algorithms that you choose. An
algorithm is a stepwise procedure that acts upon some associated data.
Conversely, a well-chosen data structure reflects the organization of
its associated algorithm. Getting them to work in harmony results in
a clear LabVIEW diagram that runs at top speed. And as a bonus, the
computer scientists will think you’re one of them.
A sample of this idea is shown in Figure 2.9. If there are many chan-
nels to process, keep them in an array. Having an array implies using
a For Loop or While Loop to process that array. After the processing
inside a loop, the logical output structure is another array. If the array
needs to carry more information than simple numerics, make it a clus-
ter array, as in the figure. The snippet of code shown here could readily
stand on its own as a subVI, doing a task of data acquisition. Fur-
thermore, it could easily be enhanced by adding items to the clusters,
such as channel calibration information. If you think ahead to relate
algorithms and data structures, that can certainly make your program
52 Chapter Two

MODULES
0
ACTIVE
SLOT
5

DATA OUT
0
CHANNELS
NAME
0 Channel Name
NAME
Channel Name
VALUE
0.26
ACTIVE

For each active module, acquire data

True
For each active channel, append
Unbundle By Name data to output array
MODULES
ACTIVE True
CHANNELS ACTIVE
NAME DATA
Build Array

Empty

Initializes Bundle
shift register

Figure 2.9 The Modules data structure at the upper left implies the algorithm at the bot-
tom because the way that the cluster arrays are nested is related to the way the For Loops
are nested.

easier to understand and modify. The only drawback to complicated

data structures may be performance. All that bundling, unbundling,
and indexing of arrays does take some extra time and memory. An old
rule for writing high-performance programs dictates that you do the
smallest possible amount of data shuffling in high-speed loops and
real-time VIs. For example, it may be worth splitting out the required
data from a highly nested structure of clusters and arrays before it’s
time to enter a For Loop for processing.

Typedefs
An important feature of LabVIEW that can help you manage compli-
cated data structures is the Type Definition, which performs the
same function as a typedef in C and a Type statement in Pascal. You
edit a control by using the LabVIEW Control Editor, in which you
 Getting Started 53

can make detailed changes to the built-in controls. Add all the items
that you need in the data structure. Then you can save your new con-
trol with a custom name. An option in the Control Editor window is
Type Definition. If you select this item, the control can’t be recon-
figured from the panel of any VI in which it appears; it can only be
modified through the Control Editor. The beauty of this scheme is that
the control itself now defines a data type, and changing it in one place
(the Control Editor) automatically changes it in every location where
it is used. If you don’t use a Type Definition and you decide to change
one item in the control, then you must manually edit every VI in which
that control appears. Otherwise, broken wires would result through-
out the hierarchy—a major source of suffering. As you can see, Type
Definitions can save you much work and raise the quality of the final
product.

Arrays
Any time you have a series of numbers or any other data type that
needs to be handled as a unit, the numbers probably belong in an
array. Most arrays are one-dimensional (1D, a column or vector), a few
are 2D (a matrix), and some specialized data sets require 3D or greater.
LabVIEW permits you to create arrays of numerics, strings, clusters,
and any other data type (except for arrays of arrays). The one require-
ment is that all the elements be of the same type. Arrays are often cre-
ated by loops, as shown in Figure 2.10. For Loops are the best because
they preallocate the required memory when they start. While Loops
can’t; LabVIEW has no way of knowing how many times a While Loop
will cycle, so the memory manager will have to be called occasionally,
slowing execution somewhat.

10 Random Number (0-1) Array of DBL numerics Array of DBL numerics

number of
0 0.15877 0.72500 0.73579
elements in
each array To Fractional
Array of strings Array of strings
0 0.15877 0.72500 0.73579

Array of booleans Array of booleans

0
0.5

Creating arrays by using a For Loop. This is an efficient way to build arrays with
Figure 2.10
many elements. A While Loop would do the same thing, but without preallocating memory.
54 Chapter Two

For Loops can also autoindex (execute sequentially for each element)
through arrays. This is a powerful technique in test systems in which
you want to increment through a list of test parameters. Pass the list
as an array, and let your For Loop execute once for each value, or set
of values. Refer to Chapter 4, “Data Types,” for more information on
arrays and clusters.

Debugging
Debugging is a part of life for all programmers (hopefully not too big
a part). As you develop your application, squash each bug as it comes
up. Never let them slip by, because it’s too easy to forget about them
later, or the situation that caused the bug may not come up again. Be
sure to check each subVI with valid and invalid inputs. One thing you
can always count on is that users will enter wrong values. LabVIEW
has some tools and techniques that can speed up the debugging pro-
cess. Read over the chapter of the LabVIEW manual on executing and
debugging VIs. It has a whole list of debugging techniques. Here are a
few of the important ones.

See what the subVIs are up to

The first debugging technique is to open any subVIs that might be of
interest in finding your problem. While a top-level VI executes, open
and observe the lower-level VIs. Their panels will update each time
they are called, allowing you to see any intermediate results. This is
a good way to find errors such as swapped cluster elements or out-of-
range values. If the value on one panel doesn’t agree with the value on
another panel, your wires are crossed somewhere. It’s another reason
for using Type Definitions on all your clusters.
There is one special trick to viewing the panels while running: If a
subVI is set up to be reentrant, then you must double-click that subVI
node on the diagram it is called from while the caller is running. Oth-
erwise, you will just be looking at another copy of the reentrant code.
Reentrant VIs have independent data storage areas allocated for each
instance in which the VI is used.
You will also note that execution slows down when panels are open.
This is caused by extra graphics updates. On a panel that is not dis-
played, the graphics don’t have to be drawn, and that saves lots of
execution time. There is also a saving in memory on VIs that are not
displayed because LabVIEW must duplicate all data that is displayed
in an indicator. If you are running a fast application where timing is
critical, having extra panels open may murder your real-time response,
so this debugging technique may not be feasible. On the other hand,
 Getting Started 55

you can use the slowdown as a debugging tool in driver development.

Sometimes, an instrument doesn’t respond to commands as fast as you
would think, and your VI can get out of sync. Having a lower-level VI’s
panel open can add just enough delay that the system starts work-
ing. Then you have a clue as to where to add some additional delay
or logic.

Peeking at data
It is often helpful to look inside a VI at intermediate values that are
not otherwise displayed. Even if no problems are evident, this can be
valuable—one of those lessons that you can only learn the hard way, by
having a VI that appears to be working, but only for the particular test
cases that you have been using. During the prototyping stage, connect
extra, temporary indicators at various points around the diagram as a
sanity check. Look at such things as the iteration counter in a loop or
the intermediate results of a complicated string search. If the values
you see don’t make sense, find out why. Otherwise, the problem will
come back to haunt you.
While a VI is executing, you can click on any wire with the probe
tool, or pop up on any wire and select Probe from the pop-up menu.
A windoid (little window) appears that displays the value flowing
through that wire. It works for all data types—even clusters and
arrays—and you can even pop up on data in the probe window to
prowl through arrays. You can have as many probe windows as you
like, floating all over the screen, and the probes can be custom con-
trols to display the data just the way you’d like. If the VI executes too
fast, the values may become a blur; you may need to single-step the
VI in that case (see the following section). Probes are great because
you don’t have to create, wire, and then later delete all kinds of tem-
porary indicators. Use probes instead of execution trace when you
want to view data in real time. A Custom Probe is a special subVI
(Figure 2.11) you create that you can dynamically insert into a wire
to see what is going on. Because a custom probe is a subVI you can
put in any functionality you want. Right-click on the wire you want
to probe, and select Custom Probe >> New . . . to create a new
custom probe.

One step at a time

Sometimes you just can’t figure out what’s happening when the VI
runs at a full speed. In that case, try single-stepping the execution
of your program. Single-stepping mode is entered by clicking on the
Pause button . When the VI runs, the diagram will open with the
single-stepping controls displayed in the toolbar, and execution will be
56 Chapter Two

Figure 2.11 Custom probe and the probe’s block diagram. Custom
probes are special VIs you can use to look at live data in a wire
while the VI is executing. This custom probe is for a DBL data
type and lets you conditionally pause the diagram. You can create
custom probes with any functionality you want.

paused with some part of the diagram flashing. You then control execu-
tion with the following buttons:

Step Into allows the next node to execute. If the node is a subVI,
the subVI opens in pause mode, and its diagram is displayed with
part of its diagram flashing. This is a recursive type of debugging,
allowing you to burrow down through the hierarchy.
Step Over lets you execute subVIs without opening them. For
built-in functions, Step Over and Step Into are equivalent. This is my
most used button for stepping through a big diagram.
Step Out Of lets you finish quickly. You can use this button to
complete a loop (even if it has 10,000 iterations to go) or a Sequence
structure without taking the VI out of pause mode. If you hold down
the mouse button, you can control how many levels it steps out of.

If you have trouble remembering what each button does, turn on

Tip Strips (use Tools >> Options Environment). Tip strips tell you in
 Getting Started 57

a clear and context-sensitive manner what will happen with messages

such as Step into subVI “Get Data.” During these single-stepping oper-
ations, you’re free to create probes anywhere in the hierarchy. For that
matter, you can freely open subVIs, view panels, and do just about any-
thing except edit a running VI.

Execution highlighting
Another way to see exactly what is happening in a diagram is to enable
execution highlighting. Click on the execution highlighting button
, and notice that it changes to . Run the VI while viewing the
diagram.
When execution starts, you can watch the flow of data. In the Debug-
ging Preferences, you can enable data bubbles to animate the flow. Each
time the output of a node generates a value, that value is displayed
in a little box if you select the autoprobe feature in the Debugging
Preferences. Execution highlighting is most useful in conjunction with
single-stepping: You can carefully compare your expectations of how
the VI works against what actually occurs. A favorite use for execution
highlighting is to find an infinite loop or a loop that doesn’t terminate
after the expected number of iterations.
While Loops are especially prone to execute forever. All you have to
do is to accidentally invert the boolean condition that terminates the
loop, and you’re stuck. If this happens in a subVI, the calling VI may
be in a nonresponsive state. When you turn on execution highlighting
or single-stepping, any subVI that is running will have a green arrow
embedded in its icon: .
Sometimes a VI tries to run forever even though it contains no loop
structures. One cause may be a subVI calling a Code Interface node
(CIN) that contains an infinite loop. Reading from a serial port using
the older Serial Port Read VI is another way to get stuck because that
operation has no time-out feature. If no characters are present at the
port, the VI will wait forever. (The solution to that problem is to call
Bytes at Serial Port to see how many characters to read, then read
that many. If nothing shows up after a period of time, you can quit with-
out attempting to read. See Chapter 10, “Instrument Driver Basics,” for
more information.) That is why you must always design low-level I/O
routines and drivers with time-outs. In any case, execution highlight-
ing will let you know which node is “hung.”

Setting breakpoints
A breakpoint is a marker you set in a computer program to pause
execution when the program reaches it. Breakpoints can be set on
any subVI, node, or wire in the hierarchy by clicking on them with the
58 Chapter Two

breakpoint tool, . A bold, red outline appears on the selected object.

When the breakpoint is encountered during execution, the VI auto-
matically enters pause mode and highlights the node that the program
will execute next. You can click pause for one of the other single-step-
ping buttons to continue. To clear a breakpoint, point at it with the
breakpoint tool and you’ll notice that the tool changes to . Click, and
the breakpoint is removed.

Suspend when called

Another kind of breakpoint that works only with subVIs is Suspend
When Called. You access this option through the VI Properties dialog
or through the SubVI Node Setup . . . dialog, accessed by popping up
on the icon of a VI on a diagram. If a subVI’s panel is not open when
the breakpoint occurs, it will open automatically. The same thing hap-
pens when you have a numeric control or indicator set to suspend on
a range error—a kind of conditional breakpoint. You might use this
feature when something disastrous would happen if somebody entered
an illegal value. Once the subVI is suspended, it will not proceed until
you click one of the following buttons on the control bar:

is the Run button, which runs the subVI as if it were a top-level

VI being run on its own. You can run it over and over, changing the
input values and observing the results. A really cool thing you can do
while in this state (and no other) is to edit the values of indicators. If
the VI runs for a long time and you need to abort it, you can click the
Abort button , and it will stop the entire hierarchy.
is the Return to caller button. Click this after you have finished
fiddling with any values on the panel. Control returns to the calling
VI, whether or not you have clicked the Run button (remember to do
so if you wish to have the subVI compute new values). The combi-
nation of these two buttons permits you to completely override the
action of a subVI, which is handy when you want to perform a test
with artificial data without modifying any diagrams.

Calling Other Code

CINs
Sometimes, programming in C is simply the best way to solve a prob-
lem. In that case, you can use Code Interface nodes (CINs) which
are written in C and link nicely into LabVIEW. They are ideal when you
want to import an existing code because you can do so almost directly.
CINs are also needed for direct access to your computer’s system
routines, graphics toolbox, and the like. They can be used to speed up
 Getting Started 59

areas where LabVIEW may not be as efficient, that is, conversion of

large arrays of numbers to strings with intricate formatting. In that
case you are forced to chain several string functions together inside a
loop with several string concatenations for each iteration. In that case,
LabVIEW has to call the memory manager on every iteration of the
loop (maybe several times on each iteration), and the resultant process
is very slow. In C, you can compute the final string size, allocate mem-
ory just once, and write to the output string very efficiently. The prob-
lem with CINs is that the average user can’t maintain them because
they require a compiler for, and knowledge of, another language. Fur-
thermore, CINs must be recompiled before they can run on a different
computer. That’s why they’re called sins. They do make a great escape
hatch for situations in which you have no other choice.

Dynamic link libraries

Another way to access external code modules is through the Call Library
node. It looks like a CIN on the diagram, but it’s configured through a
pop-up dialog through which you define the data type and other attri-
butes of each input and output. One of the items you have to configure
is whether LabVIEW can use the library reentrantly. LabVIEW needs
to know if multiple callers can use the piece of code simultaneously or if
access needs to be synchronous. If the Dynamic link library (DLL) func-
tion is not reentrant, then LabVIEW will run it in the user-interface
thread. You can tell from color of the Call library function on the block
diagram how the call is configured. Non-reentrant calls are orange and
reentrant calls are yellow. The inputs and outputs of the Call Library
node have a one-to-one correspondence with the inputs and outputs of
the external module that you are calling. One “gotcha” of DLL program-
ming is that LabVIEW only runs initialization code in a DLL one time
when LabVIEW first loads the DLL. If you’re noticing strange behavior
in your DLL calls, this is something to check.
Windows makes extensive use of Dynamic Link Libraries (DLLs)
to encapsulate drivers and various system functions. Other platforms
use shared libraries for adding greater system functionality. Almost
any compiler can generate a DLL or shared library—and you’re not lim-
ited to C. As long as you know what the parameters are, Call Library
will make the connection.

Programming by Plagiarizing
A secret weapon for LabVIEW programming is the use of available
example VIs as starting points. Many of your problems have already
been solved by someone else, but the trick is to get hold of that code.
60 Chapter Two

The examples and driver libraries that come with LabVIEW are a
gold mine, especially for the beginner. Start prowling these directories,
looking for tidbits that may be useful. Install LabVIEW on your home
computer. Open up every example VI and figure out how it works. If
you have access to a local LabVIEW user group, be sure to make some
contacts and discuss your more difficult problems. The online world is
a similarly valuable resource. We simply could not do our jobs any more
without these resources.
The simplest VIs don’t require any fancy structures or programming
tricks. Instead, you just plop down some built-in functions, wire them
up, and that’s it. This book contains many programming constructs
that you will see over and over. The examples, utilities, and drivers
can often be linked to form a usable application in no time at all. Data
acquisition using plug-in multifunction boards is one area where you
rarely have to write your own code. Using the examples and utilities
pays off for file I/O as well. We mention this because everyone writes
her or his own file handlers for the first application. It’s rarely neces-
sary, because 90 percent of your needs are probably met by the file
utilities that are already on your hard disk: Write to Text File and
Read From Text File. The other 10 percent you can get by modifying
an example or utility to meet your requirements. Little effort on your
part and lots of bang for the buck for the customer. That’s the way it’s
supposed to work.

Bibliography
AN 114, Using LabVIEW to Create Multithreaded VIs for Maximum Performance and
Reliability, www.ni.com, National Instruments Corporation, 11500 N. Mopac Express-
way, Austin, Tex., 2000.
AN 199, LabVIEW and Hyper-Threading, www.ni.com, National Instruments Corpora-
tion, 11500 N. Mopac Expressway, Austin, Tex., 2004.
Brunzie, Ted J.: “Aging Gracefully: Writing Software that Takes Changes in Stride,”
LabVIEW Technical Resource, vol. 3, no. 4, Fall 1995.
Fowler, Gregg: “Interactive Architectures Revisited,” LabVIEW Technical Resource, vol. 4,
no. 2, Spring 1996.
Gruggett, Lynda: “Getting Your Priorities Straight,” LabVIEW Technical Resource, vol. 1,
no. 2, Summer 1993. (Back issues are available from LTR Publishing.)
“Inside LabVIEW,” NIWeek 2000, Advanced Track 1C. National Instruments Corpora-
tion, 11500 N. Mopac Expressway, Austin, Tex., 2000.
Johnson, Gary W. (Ed.): LabVIEW Power Programming, McGraw-Hill, New York, 1998.
Ritter, David: LabVIEW GUI, McGraw-Hill, New York, 2001.
 Chapter

Controlling Program Flow

3
Now we are ready to look at the nuts and bolts of LabVIEW program-
ming, beginning with the fundamentals of how to control LabVIEW
flow. A fundamental theme of LabVIEW programming is dataflow and
executing nodes based on data dependency. A well-written LabVIEW
program flows in a left-to-right and top-to-bottom order, but the execu-
tion order is always governed by dataflow. Without dataflow it is impos-
sible to determine execution order. In fact, if you place independent
items on the block diagram without any data dependency between
them, they will multitask in parallel. Multitasking is a powerful con-
cept made simple in LabVIEW. However, usually you need to order a
program’s execution—random execution in a test and measurement
system is not a good idea. Dataflow can be forced through sequence
frames, or through a common thread flowing through all the block dia-
gram items.

Sequences
The simplest way to force the order of execution is to use a Sequence
structure as in the upper example of Figure 3.1. Data from one frame
is passed to succeeding frames through Sequence local variables,
which you create through a pop-up menu on the border of the Sequence
structure. In a sense, this method avoids the use of (and advantages of )
dataflow programming. You should try to avoid the overuse of Sequence
structures. LabVIEW has a great deal of inherent parallelism, where
GET A and GET B could be processed simultaneously. Using a sequence
guarantees the order of execution but prohibits parallel operations. For
instance, asynchronous tasks that use I/O devices (such as GPIB and

61
62 Chapter Three

0 [0..3] 1 [0..3] 2 [0..3] 3 [0..3]

data A
data B result

result

result

Figure 3.1 In the upper example, Sequence structures force LabVIEW to not
use dataflow. Do this only when you have a good reason. The lower example
shows the preferred method.

serial communications and plug-in boards) can run concurrently with

CPU-bound number-crunching tasks. Your program may actually exe-
cute faster if you can add parallelism by reducing the use of Sequence
structures. Note that Sequence structures add no code or execution
overhead of their own. Perhaps the worst features of Sequence struc-
tures are that they tend to hide parts of the program and that they
interrupt the natural left-to-right visual flow. The lower example in
Figure 3.1 is much easier to understand and permits parallel execution
of GET A and GET B.
Sequencing is mandatory in some problems and can clarify the pro-
gram structure by grouping logically connected operations into neat
frames. Don’t get carried away with the avoidance of Sequence struc-
tures. An alternative to the stacked Sequence structure is the flat
Sequence structure introduced in LabVIEW 8. The flat Sequence struc-
ture executes frames in a nice left-to-right order and does not obscure
dataflow. Good use of dataflow results in a clear, single-page main pro-
gram that is easy to understand.

Data Dependency
A fundamental concept of dataflow programming is data dependency,
which says that a given node can’t execute until all its inputs are avail-
able. So in fact, you can write a program using only dataflow, or not.
Let’s look at an example that is very easy to do with a sequence, but
is better done without. Consider Figure 3.2, which shows four possible
solutions to the problem where you need to open a file, read from it,
and then close it.
Solution A uses a stacked sequence. Almost everybody does it this way
the first time. Its main disadvantage is that you have to flip through
the subdiagrams to see what’s going on. Now, let’s do it with dataflow,
 Controlling Program Flow 63

0 [0..2] 1 [0..2] 2 [0..2] Using the Sequence

Open File Read File data Close File structure rather than
A refnum
dataflow. Effective, but
Path
obscures the meaning
of the program.

Improper use of dataflow. Will the file

refnum
Path be closed before it has been read? You
B data
have no way of knowing. Even if it does
work once, will it always work? Not a
good approach.

Similar to the above, but using artificial

Path
refnum data dependency to guarantee that the
C file is closed after reading. Note that the
extra wire to the Sequence structure
error
doesn't really transfer any useful
Artificial data dependency information, it just ensures ordering.

Optimum use of dataflow programming.

Path
Many VIs have a common thread, in this
D case the file refnum and error I/O cluster,
that can link things together without any
trickery.
Figure 3.2 Four ways to open, read, and close a file, using either Sequence or data-
flow programming methods. Example D is preferred.

where all the functions are linked by wires in some logical order. A
problem arises in solution B that may not be obvious, especially in a
more complicated program: Will the file be read before it is closed?
Which function executes first? Don’t assume top-to-bottom or left-to-
right execution when no data dependency exists! Make sure that the
sequence of events is explicitly defined when necessary. A solution is to
create artificial data dependency between the Read File and the
Close File functions, as shown in example C. We just connected one of
the outputs of the Read File function (any output will do) to the border
of the Sequence structure enclosing the Close File. That single-frame
sequence could also have been a Case structure or a While Loop—no
matter: It’s just a container for the next event. The advantage of this
style of programming is clarity: The entire program is visible at first
glance. Once you get good at it, you can write many of your programs
in this way.

Adding Common Threads

Going further with the idea of data dependency, you can build flow con-
trol right into your subVIs as demonstrated in example D. If you make
a collection of subVIs that are frequently used together, give them all a
common input/output terminal pair so that they can be chained together
64 Chapter Three

without requiring Sequence structures. In the case of LabVIEW’s file

I/O functions, it turns out that the file refnum is duplicated by the
Read File function, and it can be passed along to the Close File func-
tion. Thus, the read operation has to be completed before it permits the
file to be closed. The problem is solved very neatly.
A good common thread is an error code, since just about every opera-
tion that you devise probably has some kind of error checking built into
it, particularly those that do I/O operations. Each VI should test the
incoming error and not execute its function if there is an existing error.
It should then pass that error (or its own error) to the output. You can
assemble this error information into a cluster containing a numeric
error code, a string containing the name of the function that generated
the error, and an error boolean for quick testing. This technique, which
is a universal standard first promoted by Monnie Anderson at National
Instruments, is called Error I/O. Particular examples are the data
acquisition (DAQ) library functions (Figure 3.3) GPIB, VISA, serial,
and file I/O. Error I/O is discussed in detail in Chapter 10, “Instrument
Driver Basics.”

Looping
Most of your VIs will contain one or more of the two available loop
structures, the For Loop and the While Loop. Besides the obvious
use—doing an operation many times—there are many nonobvious

Error out

status code The ubiquitous Error I/O cluster, which contains:

• A Boolean, where true indicates that an error has occurred
ERROR -10243
• An error code number
source • A string indicating the VI which generated the error
AI Group Config

Simple Error
AI Config.vi AI Start.vi DAQ Task ID AI Clear.vi Handler.vi

waveform data Error out

Unbundle By Name status

Error I/O cluster
stop on error

STOP Loop runs until STOP switch

is true or an error occurs

Figure 3.3 A cluster containing error information is passed through all the
important subVIs in this program that use the DAQ library. The While Loop
stops if an error is detected, and the user ultimately sees the source of the
error displayed by the Simple Error Handler VI. Note the clean appearance of
this style of programming.
 Controlling Program Flow 65

ways to use a loop (particularly the While Loop) that are helpful to
know about. We’ll start with some details about looping that are often
overlooked.

While LOOPS
The While Loop is one of the most versatile structures in LabVIEW.
With it, you can iterate an unlimited number of times and then sud-
denly quit when the boolean conditional terminal becomes True. You
can also pop up on the conditional terminal and select Stop If False. If
the condition for stopping is always satisfied (for instance, Stop If False
wired to a False boolean constant), the loop executes exactly one time. If
you put uninitialized shift registers on one of these one-trip loops and
then construct your entire subVI inside it, the shift registers become
a memory element between calls to the VI. We call this type of VI a
functional global. You can retain all sorts of status information in
this way, such as knowing how long it’s been since this subVI was last
called (see the section on shift registers that follows).

Pretest While Loop. What if you need a While Loop that does not exe-
cute at all if some condition is False? We call this a pretest While Loop,
and it’s really easy to do (Figure 3.4). Just use a Case structure to con-
tain the code that you might not want to execute.

Graceful stops. It’s considered bad form to write an infinite loop in

LabVIEW (or any other language, for that matter). Infinite loops run
forever, generally because the programmer told them to stop only when
2 + 2 = 5, or something like that. Gary remembers running Fortran pro-
grams on the old Cyber 175 back at school, where an infinite loop was
rather hard to detect from a timesharing terminal—there was always

Pretest True False

condition
Go here if the
Actual work
Prestest
to be done
condition is
Iteration under normal
false, then
Terminal conditons
quit the loop.

Conditional
Terminal
Figure 3.4 A pretest While Loop. If the pretest condition
is False, the False frame of the Case structure is executed,
skipping the actual work to be done that resides in the True
case.
66 Chapter Three

an unpredictable delay before the friendly prompt came back, regard-

less of how long your program took to execute. Trouble was, some folks
had to pay real money for computer time, and that computer cost about
$1000 per hour to run! The hazard in LabVIEW is that the only way
to stop an infinite loop is to click the Abort icon in the toolbar. I/O
hardware could be left in an indeterminate state. As a minimum, you
need to put a boolean switch on the front panel, name it STOP, and
wire it to the conditional terminal. The boolean control palette even
has buttons prelabeled STOP. Use whatever is appropriate for your
experiment.
A funny thing happens when you have a whole bunch of stuff going
on in a While Loop: It sometimes takes a long time to stop when you
press that STOP boolean you so thoughtfully included. The problem
(and its solution) is shown in Figure 3.5. What happens in the left
frame is that there is no data dependency between the guts of the loop
and the RUN switch, so the switch may be read before the rest of the
diagram executes. If that happens, the loop will go around one more
time before actually stopping. If one cycle of the loop takes 10 minutes,
the user is going to wonder why the RUN switch doesn’t seem to do
anything. Forcing the RUN boolean to be evaluated as the last thing in
the While Loop cures this extra-cycle problem.

For Loops
Use the For Loop when you definitely know how many times a subdi-
agram needs to be executed. Examples are the building and processing
of arrays and the repetition of an operation a fixed number of times.
Most of the time, you process arrays with a For Loop because LabVIEW
already knows how many elements there are, and the auto-indexing
feature takes care of the iteration count for you automatically: All you
have to do is to wire the array to the loop, and the number of itera-
tions (count) will be equal to the number of elements in the array.

Long time runnin' Long time runnin'

99999
99999 RUN

RUN Data dependency

Figure 3.5 The left example may run an extra cycle after the RUN
switch is turned off. Forcing the switch to be evaluated as the last item
(right) guarantees that the loop will exit promptly.
 Controlling Program Flow 67

Count Terminal
5 elements 999
5 elements
5 elements
10 elements
5 elements
10 elements

Iteration Terminal

Figure 3.6 The rule of For Loop limits: The smallest count always wins,
whether it’s the N terminal or one of several arrays.

But what happens when you hook up more than one array to the For
Loop, each with a different number of elements? What if the count
terminal is also wired, but to yet a different number? Figure 3.6
should help clear up some of these questions. Rule: The smaller count
always wins. If an empty array is hooked up to a For Loop, that loop will
never execute.
Also, note that there is no way to abort a For Loop. In most program-
ming languages, there is a GOTO or an EXIT command that can force
the program to jump out of the loop. Such a mechanism was never
included in LabVIEW because it destroys the dataflow continuity. Sup-
pose you just bailed out of a For Loop from inside a nested Case struc-
ture. To what values should the outputs of the Case structure, let alone
the loop, be set? Every situation would have a different answer; it is
wiser to simply enforce good programming habits. If you need to escape
from a For Loop, use a While Loop instead! By the way, try popping up
(see your tutorial manual for instructions on popping up on your plat-
form) on the border of a For Loop, and use the Replace operation; you
can instantly change it to a While Loop, with no rewiring. The reverse
operation also works. You can even swap between Sequence and Case
structures, and LabVIEW preserves the frame numbers. Loops and
other structures can be removed entirely, leaving their contents wired
in place so far as possible, by the same process (for example, try the
Remove For Loop operation). See the LabVIEW user’s manual for these
and other cool editing tricks that can save you time.

Shift registers
Shift registers are special local variables or memory elements avail-
able in For Loops and While Loops that transfer values from the com-
pletion of one iteration to the beginning of the next. You create them by
popping up on the border of a loop. When an iteration of the loop com-
pletes, a value is written to the shift register—call this the nth value.
68 Chapter Three

Unfiltered Filtered
1.0 1.0

0.8 0.8
RUN
0.5 0.5

0.3 0.3

0.0 0.0
133 183 133 183

Random Number (0-1) Unfiltered

Previous
values
x(i-1) Latest value (X(i)) goes
x(i-2) Filtered into shift register
x(i-3) n
x(i-4) 4.00 Filtered = ∑ xi
i=1
RUN
N
where N = 4
Figure 3.7Using a shift register to perform a simple moving aver-
age on random numbers. Note the difference in the graphs of fil-
tered and unfiltered data. This is a very simple case of a finite
impulse response filter.

On the next iteration, that value is available as a source of data and is

now known as the (n − 1)st value. Also, you can add as many terminals
as you like to the left side, thus returning not only the (n − 1)st value,
but also n − 2, n − 3, and so on. This gives you the ability to do digi-
tal filtering, modeling of discrete systems, and other algorithms that
require a short history of the values of some variable. Any kind of data
can be stored in a shift register—they are polymorphic. Figure 3.7
is an implementation of a simple finite impulse response (FIR) filter,
in this case a moving averager that computes the average of the last
four values. You can see from the strip charts how the random numbers
have been smoothed over time in the filtered case.
Besides purely mathematical applications like this one, there are
many other things you can do with shift registers. Figure 3.8 is a very

Empty string False Array of True

array constant non-empty
strings
0 Build Array

Array of not empty- empty string-

strings add to array throw it away
(the input)

Figure 3.8 Weeding out empty strings with a shift register. A similar
program might be used to select special numeric values. Notice that we
had to use Build Array inside a loop, a relatively slow construct, but also
unavoidable.
 Controlling Program Flow 69

common construct where an array is assembled based on some condi-

tional testing of the incoming data. In this case, it’s weeding out empty
strings from an incoming array. This program is frequently seen in con-
figuration management programs. In that case, a user is filling out a
big cluster array that contains information about the various chan-
nels in the system. Leaving a channel name empty implies that the
channel is unused and should be deleted from the output array. The
Build Array function is located inside a Case structure that checks to
see if the current string array element is empty. To initialize the shift
register, create a one-dimensional (1D) string array constant on the
diagram by popping up on the shift register terminal and selecting
Create Constant.

Uninitialized shift registers

In Figure 3.8, what would happen if that empty string array constant
already had something in it? That data would appear in the output
array ahead of the desired data. That’s why we want an empty array
for initialization. What happens if the empty string array is left out
altogether? The answer is that the first time the program runs after
being loaded, the shift register is in fact empty and works as expected.
But the shift register then retains its previous contents until the next
execution. Every time you ran this modified VI, the output string would
get bigger and bigger (a good way to make LabVIEW run out of mem-
ory, by the way). All shift registers are initialized at compile time as
well: Arrays and strings are empty, numerics are zero, and booleans are
False. There are some important uses for uninitialized shift registers
that you need to know about.
First, you can use uninitialized shift registers to keep track of state
information between calls to a subVI. This is an extremely power-
ful technique and should be studied closely. You will see it dozens of
times in this book. In Figure 3.9, a shift register saves the previous
value of the set point front panel control and compares it to the cur-
rent setting. If a change of value has occurred, the new setting is sent
to a temperature controller via the serial port. By transmitting only
when the value has changed, you can reduce the communications
traffic. This technique, known as change-of-state detection, can
make a VI act more intelligently toward operator inputs. Use a second
(boolean) shift register to force the transmission of the set point the
first time this VI is run. Since the boolean shift register is False when
loaded, you can test for that condition as shown here. Then write True
to the shift register to keep the VI from performing the initialization
operation again.
70 Chapter Three

previous
changed?
setpoint Only initialize
True upon loading
False
init
Write Setpoint has
setpoint not changed.
Setpoint
SP Read Temperature

command PV

Data dependency: read after write Executes just once

Figure 3.9 This VI writes a new set point to a Eurotherm 808 temperature
controller only when the user has changed the value. It then reads the cur-
rent temperature. The upper shift register saves the previous set point value
for change-of-state detection. The lower shift register is used for initialization to
guarantee that a set point will always be written when the VI is first run.

Note that the conditional terminal is wired with a False constant,

which means that the contents of the loop execute only once each
time the VI is called. Such a construct is of little use as a top-level VI.
Instead, its intended use is as a more intelligent subVI—one with state
memory. That is, each time this subVI is called, the contents of the
shift register provide it with logical information regarding conditions
at the end of the previous iteration.
This allows you to program the subVI to take action based not only on
the current inputs, but on previous inputs as well. This concept will be
extrapolated into powerful state machines in subsequent chapters.
The shift register in Figure 3.10 keeps track of the elapsed time
since the subVI was last called. If you compare the elapsed time with

Figure 3.10 This example shows how

to use an uninitialized shift register to
determine how long it’s been since this
subVI was last called. This makes the
subVI time aware and useful for data
logging, watchdog timing, or time-
dependent calculations.
 Controlling Program Flow 71

a desired value, you could use this to trigger a periodic function such
as data logging or watchdog timing. This technique is also used in the
PID (proportional integral derivative) Toolkit control blocks whose
algorithms are time-dependent.

Globals
Global and local variables
Another, very important use for uninitialized shift registers occurs
when you create global variable VIs, also called Functional
Globals. In LabVIEW, a variable is a wire connecting two objects on a
diagram. Since it exists only on one diagram, it is by definition a local
variable. By using an uninitialized shift register in a subVI, two or
more calling VIs can share information by reading from and writing
to the shift register in the subVI. This is a kind of global variable, and
it’s a very powerful notion. Figure 3.11 shows what the basic model
looks like.
If Set Value is True (write mode), the input value is loaded into
the shift register and copied to the output. If Set Value is False (read
mode), the old value is read out and recycled in the shift register. The
Valid indicator tells you that something has been written; it would be
a bad idea to read from this global and get nothing when you expect a
number. Important note: For each global variable VI, you must create a
distinct VI with a unique name.
There are several really nice features of global variables. First, they
can be read or written any time, any place, and all callers will access
the same copy of the data. There is no data duplication. This includes
multiple copies on the same diagram and, of course, communications
between top-level VIs. This permits asynchronous tasks to share infor-
mation. For instance, you could have one top-level VI that just scans

Set value? Valid?

New value Current value Valid?

Set value? (read value:F) Valid?

Set value?
New value
Current
value

New value Current value

0.00 0.00

Figure 3.11 A global variable VI that stores a single numeric value. You could
easily change the input and output data type to an array, a cluster, or any-
thing else.
72 Chapter Three

iteration (1; 0=init)

interval A (sec) event A Get Date/Time
interval B (sec) event B iteration
In Seconds
interval C (min) event C (1;
interval D (ms) event D 1..
Interval Timer interval A (sec) event A

iteration (1; 0 = init) interval A (sec) event A

45 1.00

event B
interval B (sec) event B interval B (sec)

4.00

min 60.00
interval C (min) event C
event C
5.00
interval C (min)

interval D (ms) event D msecs

500.00 now

event D
interval D (ms)

Figure 3.12 A time interval generator based on uninitialized shift registers. Call this
subVI with one of the Event booleans connected to a Case structure that contains a func-
tion you wish to perform periodically. For instance, you could place a strip chart or file
storage function in the True frame of the Case structure.

your input hardware to collect readings at some rate. It writes the data
to a global array. Independently, you could have another top-level VI
that displays those readings from the global at another rate, and yet
another VI that stores the values. This is like having a global database
or a client-server relationship among VIs.
Because these global variables are VIs, they are more than just mem-
ory locations. They can perform processing on the data, such as filtering
or selection, and they can keep track of time as the Interval Timer
VI does in Figure 3.12. Each channel of this interval timer compares
the present time with the time stored in the shift register the last time
when the channel’s Event output was True. When the Iteration input
is equal to zero, all the timers are synchronized and are forced to trigger.
This technique of initialization is another common trick. It’s convenient
because you can wire the Iteration input to the iteration terminal, in
the calling VI, which then initializes the subVI on the first iteration. We
use this timer in many applications throughout this book.
As you can see, functional globals can store multiple elements of dif-
ferent types. Just add more shift registers and input/output terminals.
By using clusters or arrays, the storage capacity is virtually unlimited.
LabVIEW has built-in global variables as well (discussed in the next
section). They are faster and more efficient than functional global vari-
ables and should be used preferentially. However, you can’t embed any
intelligence in them since they have no diagram; they are merely data
storage devices.
 Controlling Program Flow 73

There is one more issue with functional global variables: perfor-

mance. Scalar types including numerics and booleans are very fast and
limited only by LabVIEW’s subVI calling overhead. Arrays or clusters
of numerics and booleans are a little slower, but the number of elements
is what is most important. The real killer comes when your global vari-
able carries clusters or arrays that contain strings. This requires the
services of the memory manager every time it is called, and thus the
performance is relatively poor (almost 10 times slower than a compa-
rably sized collection of numerics).

Built-in global variables—and their hazards

Starting with LabVIEW 3, global variables are built in. To create a
global variable, you select it from the Structures subpalette of the
Function palette, place it on the diagram, and then double-click on it
to open and edit its front panel, exactly as with any subVI. You can
place any number of controls on the panel, name them, and then save
the global variable with a name—again, just as with any subVI. Back
on your main diagram, pop up on the global variable, select one of the
controls that you put in the global, and then choose whether you wish
to read or write data. Finally, wire it to the appropriate source or des-
tination. At any time, you can open the panel of the global variable to
view or change its contents. That can be very helpful during a debug-
ging session.
The difference between built-in globals and the functional globals you
make yourself with a subVI is that the built-in ones are not true VIs
and as such cannot be programmed to do anything besides simple data
storage. However, built-in globals are sometimes faster for most data
types. Another advantage of the built-in globals is that you can have all
the global data for your entire program present in just one global vari-
able but access them separately with no penalty in performance. With
subVI-based globals, you can combine many variables into one global,
but you must read and write them all at once, which increases execu-
tion time and memory management overhead. Thus, built-in globals
are preferred for most applications for performance reasons.
A hazard you need to be aware of when using either type of global
variable is the potential for race conditions. A race condition exists
when two or more events can occur in any order, but you rely on them
to occur in a particular order. While you’re developing your VI, or under
normal conditions, the order may be as expected and all is well. But
under different conditions the order will vary, causing the program
to misbehave. Sequence structures and data dependency prevent
race conditions from being a general problem in LabVIEW, but global
74 Chapter Three

Loop A Loop B
a global control B a global
control A
number (write) number (write)

a global indicator
(read) number

RUN RUN

Figure 3.13 Race conditions are a hazard associated with all global
variables. The global gets written in two places. Which value will it
contain when it’s read?

variables provide a way to violate strict dataflow programming. There-

fore, it’s up to you to understand the pitfalls of race conditions.
In Figure 3.13, two While Loops are executing at the same time (they
don’t necessarily have to be on the same diagram). Both loops write
a value to a global number, and one of the loops reads the value. The
trouble is, which value will the global contain when it’s time to read it?
The value from loop A or the value from loop B? Note that there can be
any number of data writers out there, adding to the uncertainty. You
might have to add an elaborate handshaking, timing, or sequencing
scheme to this simple example to force things to occur in a predictable
fashion.
An all-around safe approach to avoiding race conditions is to write
your overall hierarchy in such a way that a global can only be writ-
ten from one location. This condition would be met, for instance, by
the client-server architecture where there is one data source (a data
acquisition VI) with multiple data readers (display, archive, etc.). You
must also make sure that you never read from a global before it is ini-
tialized.
This is one of the first rules taught for traditional languages: Initial-
ize the variables, then start running the main program. For example,
it would be improper for the data display VI to run before the data
acquisition VI because the global variable is initially empty or contains
garbage. Rule: Enforce the order of execution in all situations that use
global variables.
If your application has a global array, there is a risk of excessive data
duplication. When an array is passed along through wires on a single
diagram, LabVIEW does an admirable job of avoiding array duplica-
tion, thus saving memory. This is particularly important when you
want to access a single element by indexing, adding, or replacing an
element. But if the data comes from a global variable, your program
has to read the data, make a local copy, index or modify the array, and
then write it back. If you do this process at many different locations,
 Controlling Program Flow 75

you end up making many copies of the data. This wastes memory and
adds execution overhead. A solution is to create a subVI that encapsu-
lates the global, providing whatever access the rest of your program
requires. It might have single-element inputs and outputs, addressing
the array by element number. In this way, the subVI has the only direct
access to the global, guaranteeing that there is only one copy of the
data. This implementation is realized automatically when you create
functional globals built from shift registers.
Global variables, while handy, can quickly become a programmer’s
nightmare because they hide the flow of data. For instance, you could
write a LabVIEW program where there are several subVIs sitting on a
diagram with no wires interconnecting them and no flow control struc-
tures. Global variables make this possible: The subVIs all run until
a global boolean is set to False, and all the data is passed among the
subVIs in global variables. The problem is that nobody can understand
what is happening since all data transfers are hidden. Similar things
happen in regular programming languages where most of the data is
passed in global variables rather than being part of the subroutine
calls. This data hiding is not only confusing, but also dangerous. All you
have to do is access the wrong item in a global at the wrong time, and
things will go nuts. How do you troubleshoot your program when you
can’t even figure out where the data is coming from or going to?
The answer is another rule. Rule: Use global variables only where
there is no other dataflow alternative. Using global variables to syn-
chronize or exchange information between parallel loops or top-level
VIs is perfectly reasonable if done in moderation. Using globals to
carry data from one side of a diagram to the other “because the wires
would be too long” is asking for trouble. You are in effect making your
data accessible to the entire LabVIEW hierarchy. One more helpful tip
with any global variable: Use the Visible Items . . . Label pop-up item
(whether it’s a built-in or VI-based global) to display its name. When
there are many global variables in a program, it’s difficult to keep track
of which is which, so labeling helps. Finally, if you need to locate all
instances of a global variable, you can use the Find command from the
Edit menu, or you can pop up on the global variable and choose Find:
Global References.

Local variables
Another way to manipulate data within the scope of a LabVIEW dia-
gram is with local variables. A local variable allows you to read data
from or write data to controls and indicators without directly wiring to
the usual control or indicator terminal. This means you have unlimited
read-write access from multiple locations on the diagram. To create a
76 Chapter Three

local variable, select a local variable from the Structures palette and
drop it on the diagram. Pop up on the local variable node and select the
item you wish to access. The list contains the names of every control
and indicator on the front panel. Then choose whether you want to read
or write data. The local variable behaves exactly the same as the con-
trol or indicator’s terminal, other than the fact that you are free to read
or write. Here are some important facts about local variables.
■ Local variables act only on the controls and indicators that reside on
the same diagram. You can’t use a local variable to access a control
that resides in another VI. Use global variables, or better, regular
wired connections to subVIs to transfer data outside of the current
diagram.
■ You can have as many local variables as you want for each control
or indicator. Note how confusing this can become: Imagine your con-
trols changing state mysteriously because you accidentally selected
the wrong item in one or more local variables. Danger!
■ As with global variables, you should use local variables only when
there is no other reasonable dataflow alternative. They bypass
the explicit flow of data, obscuring the relationships between data
sources (controls) and data sinks (indicators).
■ Each instance of a local variable requires an additional copy of the
associated data. This can be significant for arrays and other data
types that contain large amounts of data. This is another reason that
it’s better to use wires than local variables whenever possible.

There are three basic uses for local variables: control initialization,
control adjustment or interaction, and temporary storage.
Figure 3.14 demonstrates control initialization—a simple, safe, and
common use for a local variable. When you start up a top-level VI, it
is important that the controls be preset to their required states. In
this example, a boolean control opens and closes a valve via a digital

Make sure the front panel

control, Valve, is turned off device 4
at startup. Write to
digital channel 0
Digital
line 0 Line.vi
Local WRITE Valve
CLOSE
the valve variable true = open

Valve
Run

Figure 3.14 Local variables are a convenient way to initialize front

panel controls. The Sequence structure guarantees that the initial-
ization is completed before the While Loop starts.
 Controlling Program Flow 77

output line on a plug-in board. At start-up, a local variable sets the

boolean control to False before starting the While Loop. If you have
many controls to initialize in this manner, you will need just as many
local variables. Note the Sequence structure surrounding the initial-
ization part of the program. This guarantees that the local variable
has done its work before the While Loop starts. Otherwise, a race con-
dition might arise.
The same technique is useful at all levels in a hierarchy, but please
note that you can initialize the controls in a subVI by assigning them
to terminals on the subVI’s icon, then wiring in the desired values from
the calling VI. This is a case when there is a dataflow alternative. How-
ever, if you intend to use a VI at the top level, there is no such alterna-
tive, and you can use local variables with our blessing.
Another problem that is solved by local variables is one we have vis-
ited before: stopping a parallel While Loop. This is a case where the
local variable is a temporary storage device with a scope that extends
throughout the diagram of a single VI. In Figure 3.15, we used a local
variable to stop the slave loop. There is one catch: You can’t use a local
variable with a boolean control that has its mechanical action set to
one of the latch modes. Those are the modes you would normally use
for a stop switch, where the switch resets to its off position after it is
read by the program. Why this limitation? Because there is an ambigu-
ous situation: If the user throws a boolean into its temporary (latched)
position, should it reset after being read directly from its regular ter-
minal, or when it is read by a local variable, or both? There is no uni-
versal answer to this situation, so it is not allowed. Figure 3.15 has a
solution, though. After the two loops stop, another local variable resets
the switch for you.

Master loop
Chart 1

On/Off Switch
Turn off the switch
after stopping loops
Local WRITE variable
Slave loop
On/Off Switch
Chart 2

Local READ variable

NOR function: output
is False when either
On/Off Switch input is True.

Figure 3.15 This example shows how to use local vari-

ables to stop a parallel While Loop. The switch is pro-
grammatically reset because latching modes are not
permitted for boolean controls that are also accessed
by local variables.
78 Chapter Three

Managing controls that interact is another cool use for local variables.
But a word of caution is in order regarding race conditions. It is very
easy to write a program that acts in an unpredictable or undesirable
manner because there is more than one source for the data displayed
in a control or an indicator. What you must do is to explicitly define
the order of execution in such a way that the action of a local variable
cannot interfere with other data sources, whether they are user inputs,
control terminals, or other local variables on the same diagram. It’s dif-
ficult to give you a more precise description of the potential problems
because there are so many situations. Just think carefully before using
local variables, and always test your program thoroughly.
To show you how easy it is to create unnerving activity with local
variables, another apparently simple example of interactive controls is
shown in Figure 3.16. We have two numeric slider controls, for Poten-
tial (volts) and Current (amperes), and we want to limit the product
of the two (Power, in watts) to a value that our power supply can han-
dle. In this solution, the product is compared with a maximum power
limit, and if it’s too high, a local variable forces the Potential control
to an appropriate value. (We could have limited the Current control
as easily.)
This program works, but it has one quirk: If the user sets the Current
way up and then drags and holds the Potential slider too high, the local
variable keeps resetting the Potential slider to a valid value. However,
since the user keeps holding it at the illegal value, the slider oscillates
between the legal and illegal values as fast as the loop can run. In the
section on events we’ll show you an event-driven approach.

Debugging global variables. If your program uses many global variables

or VIs that use uninitialized shift registers for internal state memory,
you have an extra layer of complexity and many more opportunities for

Power limit
Watts = Volts x Amps (False case is empty)
50.00 W Power
Current True
Current A Potential V Power Potential Power too high;
adjust potential
20.0 20.0 50.00 W
Power limit
WRITE local
15.0 15.0 variable

Potential
10.0 10.0
Don't be a CPU hog
5.0 5.0
STOP 10 stop
0.0 0.0

Figure 3.16Limiting the product of two numeric controls by using local variables. But it acts funny
when the user drags and holds the Potential slider to a high value.
 Controlling Program Flow 79

bugs to crop up. Here’s a brief list of common mistakes associated with
global variables and state memory:

■ Accidentally writing data to the wrong global variable. It’s easy to do;
just select the wrong item on one of LabVIEW’s built-in globals, and
you’re in trouble.
■ Forgetting to initialize shift-register-based memory. Most of the
examples in this book that use uninitialized shift registers have a
Case structure inside the loop that writes appropriate data into the
shift register at initialization time. If you expect a shift register to be
empty each time the program starts, you must initialize it as such. It
will automatically be empty when the VI is loaded, but will no longer
be empty after the first run.
■ Calling a subVI with state memory from multiple locations without
making the subVI reentrant when required. You need to use reen-
trancy whenever you want the independent calls not to share data,
for instance, when computing a running average. Reentrant execu-
tion is selected from the VI Properties menu.
■ Conversely, setting up a subVI with state memory as reentrant when
you do want multiple calls to share data. Remember that calls to
reentrant VIs from different locations don’t share data. For instance,
a file management subVI that creates a data file in one location and
then writes data to the file in another location might keep the file
path in an uninitialized shift register. Such a subVI won’t work if it
is made reentrant.
■ Creating race conditions. You must be absolutely certain that every
global variable is accessed in the proper order by its various call-
ing VIs. The most common case occurs when you attempt to write
and read a global variable on the same diagram without forcing the
execution sequence. Race conditions are absolutely the most difficult
bugs to find and are the main reason you should avoid using globals
haphazardly.
■ Try to avoid situations where a global variable is written in more
than one location in the hierarchy. It may be hard to figure out which
location supplied the latest value.

How do you debug these global variables? The first thing you must
do is to locate all the global variable’s callers. Open the global variable’s
panel and select This VI’s Callers from the Browse menu. Note every
location where the global is accessed, then audit the list and see if it
makes sense. Alternatively, you can use the Find command in the Proj-
ect menu. Keep the panel of the global open while you run the main VI.
80 Chapter Three

Observe changes in the displayed data if you can. You may want to
single-step one or more calling VIs to more carefully observe the data.
One trick is to add a new string control called info to the global vari-
able. At every location where the global is called to read or write data,
add another call that writes a message to the info string. It might
say, “SubVI abc writing xyz array from inner Case structure.” You can
also wire the Call Chain function (in the Advanced function palette) to
the info string. (Actually, it has to be a string array in this case.) Call
Chain returns a string array containing an ordered list of the VI call-
ing chain, from top-level VI on down. Either of these methods makes it
abundantly clear who is accessing what.

Debugging local variables and property nodes. Everything

we’ve said regarding global variables is true to a great degree for local
variables. Race conditions, in particular, can be very difficult to diag-
nose. This is reason enough to avoid using local variables except when
absolutely necessary! Execution highlighting and/or single-stepping is
probably your best debugging tool because you can usually see in what
order the local variables are being executed. However, if your VI has a
complicated architecture with parallel loops and timers, the relative
ordering of events is time-dependent and cannot be duplicated in the
slow motion of execution highlighting. In difficult cases, you must first
make sure that you have not inadvertently selected the wrong item
in a local variable. Pop up on the problematic control or indicator, and
select Find >> Local Variables. This will lead you to each local vari-
able associated with that control.
Be sure that each instance is logically correct: Is it read or write
mode, and is it the proper place to access this variable? Your only other
recourse is to patiently walk through the logic of your diagram, just as
you would in any programming language. See? We told you that local
variables are a violation of LabVIEW’s otherwise strict dataflow con-
cepts. And what did they get you? The same old problems we have with
traditional languages.
Property nodes can be a source of bugs just as local variables can.
Controls disappear at the wrong time, scales get adjusted to bizarre val-
ues, and control sliders jiggle up and down. The pop-up Find command
can also locate Property nodes, and you should always go through that
exercise as a first step in debugging. Make sure that you have forced
the execution sequence for each Property node, and that the sequence
makes sense. Complex user interfaces make extensive use of Prop-
erty nodes, and it’s easy to forget to connect an item or two, leaving
a control in an inconsistent condition. After a heavy editing session,
mistakes are bound to crop up. Again, you must patiently walk through
your code.
 Controlling Program Flow 81

Events
Asynchronous event-driven applications are fundamentally different
from data-flow-driven applications. This conflict between data-flow-
driven and event-driven architecture has made writing user-interface
event-driven applications in LabVIEW difficult without deviating from
dataflow. In the past you had to monitor front panel controls using poll-
ing. Figure 3.9 is an example of how polling is used to monitor changes
in a front panel control. The value of the numeric control is saved in a
shift register and compared with the current value on each iteration of
the loop. This was the typical way to poll the user interface. There were
other methods, some more complex than others, to save and compare
control values; but when you used polling, you always risked missing
events. Now with the introduction of the Event structure in Lab-
VIEW 6.1, we finally have a mechanism that allows us to respond to
asynchronous user-interface events.

Notify and Filter events

Let’s start off by defining an event as a notification that “something”
happened. LabVIEW processes event notifications from the operat-
ing system, and now with the Event structure we have a way to get
in on the action and process events on our block diagram. The Event
structure is shown in Figure 3.17. You’ll find it on the Program-
ming >> Structures Function palette. It looks a lot like a Case struc-
ture with multiple cases and a selector on top. You can configure it
to fire on Application events (that is, application exiting, etc.), VI
events (panel close, window move, etc.), Sub Panel Pane events

Figure 3.17 The Event structure looks

a lot like a Case structure, with differ-
ent event cases configured to fire on
specific events. The Event structure
sleeps until an event fires or the time-
out expires.
82 Chapter Three

(mouse down, mouse move, etc.), Control and Indicator events

(mouse down, mouse move, key down, etc.); you can even configure
your own Dynamic user-defined events. Only one case fires for each
event—the structure does not do any looping on its own. The Event
structure has a time-out value in milliseconds. The default is –1, or
never time out. While it’s waiting, the Event structure sleeps and con-
sumes no CPU cycles. Inside each event is a data node that returns
information about the event such as the time the event fired. Need
to log every keystroke and when it occurred? Want to add a macro
capability to your application? Events make it easy. In Figure 3.18
we’re adding an Event case to register for event notifications from the
slider control on the front panel. Once registered for the Value Change
event, this case will fire every time the slider changes value. The
Edit Events dialog shows just a few of the events we can register for

Figure 3.18 Register for the Value Change event from the Slider
control. This event case will fire every time the slider changes
value.
 Controlling Program Flow 83

this one control. The Value Change event is classified as a Notify

event. For the Value Change Notify event, LabVIEW processes the user-
interface action (changing the control value) and then notifies all the
Event structures waiting for the event that the value has changed. We
could have multiple Event structures (not generally a good idea) wait-
ing for notification of this event, and LabVIEW would notify them all
in parallel.
LabVIEW passes Filter events (Filter events end with a “?”) to you
before processing so you can decide how to handle the event. For exam-
ple, the event in Figure 3.19 is configured to filter Mouse Down? events
for two controls: Cluster and OK button. While the Enable button is
False, mouse down actions on these two controls will be discarded. Fil-
ter events let you decide what to do with the users’ action. Note that we
are using a single event case to wait on events from multiple controls.
We can do this with Filter events if the events return identical event
data. Notify events can be merged, but the left data node will return
only the data common to all events. In the section on dynamic events
we’ll see this again.
Notify and Filter events open up a whole new world of user-interface
interaction previously impossible with polling. This newfound power
also requires us to learn some new ways to program. If we’re not careful,
the Event structure has some characteristics that can jump up and bite
us. In Figure 3.18 the Edit Events dialog box has a check box labeled
“Lock front panel until the event case for this event completes.” And that
does exactly what it says: When the Value Change event for the slider
fires, the front panel is locked until the event case completes operation.

Figure 3.19 Filter events allow you to

decide whether to let the event hap-
pen. While the Enable” button is False,
the mouse down events on the Cluster
and OK buttons are discarded. Use Fil-
ter events when you want to intercept
and process events before the event is
handled by LabVIEW.
84 Chapter Three

The locking action gives you a chance to swap out the user interface
or whatever else you need to do to handle the event. The user will not
notice anything as long as you process the event and quickly exit the
event case. However, if you place code in your event case that takes a
while to execute, or if your Event structure is buried in a state machine,
the front panel of your VI will be totally unresponsive. Some practical
guidelines are as follows:
■ Use only one Event structure per block diagram. Let a single Event
structure provide a focal point for all user-interface code.
■ Do only minimal processing inside an Event structure. Lengthy pro-
cessing inside the event case will lock up the user interface.
■ Use a separate While Loop containing only the Event structure. In
the section on design patterns we’ll show you a good parallel archi-
tecture for user-interface-driven applications.

You can combine events, Property nodes, and local variables to create
event-driven applications that are extremely responsive. Because the
Event structure consumes no CPU cycles while waiting for an event,
they make a very low overhead way to enhance your user interface.
Figure 3.20 shows the application from Figure 3.16 rewritten using an

Figure 3.20 Limiting the product of two numeric controls using local variables inside
an Event structure. Interaction is smooth and easy compared to that in Figure 3.16.
 Controlling Program Flow 85

Event structure. We use a single event case to fire on a Value Change

event from any of the three controls. The left data node of the event
case returns a reference to the control that triggered the event from
which we can get the name using a Property node. If the calculated
power exceeds the power limit, we adjust the opposite control’s value.
The result is a responsive piece of user-interface code that works seam-
lessly and consumes almost no CPU cycles, unlike earlier examples
that used polling.

Mechanical actions
The cleanest, easiest, and recommended way to use events and stay
within the dataflow concept is to put controls inside their event case.
This is especially important for boolean controls with a latching
mechanical action. To return to its default state, a boolean latch depends
on being read by the block diagram. If you place the latching boolean
inside its event case, the mechanical action will function properly. But
there’s a funny thing about Value Change events and latched booleans.
What happens when the user selects Edit >> Undo Data Change?
The event fires again, but the control value is not what you expected.
Figure 3.21 illustrates how to handle this unexpected event by using
a Case structure inside the event for any latched boolean. The True
case handles the normal value change, and the False case handles the
unexpected Undo.

Dynamic events
Dynamic events allow you to register and unregister for events while
your application is running. The Filter and Notify events shown so far
were statically linked at edit time to a specific control. As long as the VI

Figure 3.21Value Change events for booleans with a

latched mechanical action require special attention.
86 Chapter Three

Figure 3.22 Dynamic events programmatically register and

unregister controls for events. Note the data from the controls
is returned as a variant. Dynamic events are not limited to
front panel controls, but can be programmatically generated.

is running, changes to the control will fire a Value Change event. Usu-
ally this is the behavior you want. But there are some situations where
dynamic events are a better solution. In Figure 3.22 we used control
references to dynamically register controls for a Value Change event.
The Register For Events property node receives an array of references
of all the controls on every page of a tab control (Figure 3.23). Dynami-
cally registering for this event can save a lot of development time in
high-channel-count applications. Note the data from the control is
returned as a variant. In our case all the controls were clusters with
the same data type, so it is simple to extract the data from the variant.
For clarity the actual controls were not placed in the Value Change

Figure 3.23 A reference to Tab Control is used to register every control on every
page of Tab Control for the Value Change event. This could easily be turned into a
reusable user-interface subVI.
 Controlling Program Flow 87

event case, but good programming practice suggests you should put
them there.
Dynamically registering controls for a Value Change event only
touches the tip of the iceberg in what you can do with dynamic events.
You can create your own user-defined events and use the Event struc-
ture as a sophisticated notification and messaging system. You can also
use the Event structure for “.NET” events. Whatever you use the Event
structure for, try not to stray too far from dataflow programming. In
the section on design patterns we’ll go over some proven ways to use
the Event structure in your applications.

Design Patterns
Your initial objective is to decide on an overall architecture, or pro-
gramming strategy, which determines how you want your application
to work in the broadest sense. Here are some models, also called Lab-
VIEW Design Patterns, that represent the fundamental structure
of common LabVIEW applications. The goals of using design patterns
are that the architecture has already been proved to work, and when
you see someone else’s application that incorporates a design pattern
you’ve used before, you recognize and understand it. The list of design
patterns continues to grow as the years go by, and they are all reliable
approaches. Choosing an appropriate architecture is a mix of analysis,
intuition, and experience. Study these architectures, memorize their
basic attributes, and see if your next application doesn’t go together a
little more smoothly.

Initialize and then loop

The simplest applications initialize and then loop. You perform some
initialization, such as setting up files and starting the hardware, and
then drop into a While Loop that does the main task several times. The
main loop usually consists of four steps that are repeated at a rate that
matches the requirements of your signals:

1. Acquire.
2. Analyze.
3. Display.
4. Store.

In a well-written program, these steps would be encapsulated in subVIs.

After the main loop has terminated, there may also be a shutdown or
cleanup task that closes files or turns off something in the hardware.
88 Chapter Three

0 [0..2] 1 [0..2] 2 [0..2]

Initialization code goes here 0 [0..3] Shutdown code goes here

• Open files • Close files
• Initialize I/O Real work to be
done goes here • Turn off or clear I/O
• Manage configuration info

Interval RUN

Figure 3.24 A Sequence containing frames for starting up, doing the work, and shutting down.
It is a classic for data acquisition programs. The While Loop runs until the operation is finished.
The Sequence inside contains the real work to be accomplished.

This strategy is a classic for the top-level VI in ordinary data acquisi-

tion. The first two design patterns, which follow, work this way.
Figure 3.24 is by far the most commonly used program structure in
all of LabVIEW, primarily because it is an obvious solution. An over-
all Sequence structure forces the order of initialize, main operation,
and shutdown. The main While Loop runs at a regular interval until
the user stops the operation or some other event occurs. Frequently
found inside the main loop is another Sequence structure that contains
the operations to be performed: acquire, analyze, display, and then
store. Unfortunately, this approach obscures dataflow with all those
Sequence structures.
An improvement on the previous example, shown in Figure 3.25,
makes effective use of dataflow programming and can be somewhat
easier to understand. Connections between the three major operations
(initialize, main operation, and shutdown) force the order of execution
in the same way that an overall Sequence structure does, except there
is no obscuring of the overall program. The same methodology can and
should be applied inside the main While Loop by using common threads
between the functions employed there.

Initialization code goes here Main loop does Shutdown code goes here
• Open files the real work • Close files
• Initialize I/O • Turn off or clear I/O
• Manage configuration info graph
config

error error

Interval RUN

Figure 3.25 Dataflow instead of Sequence structure enhances readability. It has the same func-
tionality as Figure 3.24. Note that the connections between tasks (subVIs) are not optional: they
force the order of execution.
 Controlling Program Flow 89

The net result is a clearer program, explainable with just one page. This
concept is used by most advanced LabVIEW users, like you. The example
shown here is in every way a real, working LabVIEW program.

Independent parallel loops

Any time you have two tasks that you need to do at different speeds or
priorities, consider using multiple independent While Loops that oper-
ate in parallel on one diagram (Figure 3.26). You can also use inde-
pendent loops when there is a main task that needs to run regularly
and one or more secondary tasks that run only when the operator
throws a switch. If data needs to be exchanged between the loops, as
it often does, use local or global variables. In this example, a boolean
control with a local variable stops the loops at the appropriate time,
and a smart global variable (operating as a little database) carries a
cluster of settings between the two loops. This is another top-level VI
architecture.
Remember to initialize controls before they are read anywhere else.
If you are using a loop similar to the bottom one shown in this example,
put a time delay inside the loop to keep it from hogging CPU time when
nothing is happening. Normally we’ll use 100 or 200 ms because that’s
about the threshold of human perception.
Global variables tend to hide the flow of data if they are buried
inside subVIs. To make your programs easier to understand, a good
policy is to place all global variables out on the diagram where they are

Real-time operation

Initialize Temperature
Setting Global local var.
chart
500
RUN
Config & Read
Shutdown
Initialize Ramp/Soak
Controller

Shutdown
Manual operation

Manual True False

RUN
Permit manual adjustment
Setting Global Most of the time, the loop is polling
the manual switch, and executes
Write this empty Case structure.
100 Read

Dependency: loops
run after RUN local RUN Slave loop runs until RUN
variable is set to True. local variable is False.

Note that interloop communications Set this subVI to Show

are via global variables. Front Panel When Called.

Figure 3.26 This example uses two independent loops: the upper one for the actual control and the
lower one to handle manual operation (shown simplified here). Thus the main control loop always
runs at a constant interval regardless of the state of the manual loop.
90 Chapter Three

easily spotted. In this example, the Manual Control subVI reads set-
tings information from a global, changes that information, and then
writes it back to the global. We could have hidden the global operations
inside the subVI, but then you wouldn’t have a clue as to how that subVI
acts on the settings information, which is also used by the Ramp/Soak
Controller subVI. It also helps to show the name of the global variable
so that the reader knows where the data is being stored. We also want
to point out what an incredibly powerful, yet simple technique this is.
Go ahead: Implement parallel tasks in C or Pascal.

Client-server
The term client-server comes from the world of distributed comput-
ing where a central host has all the disk drives and shared resources
(the server), and a number of users out on the network access
those resources as required (the clients). Here is how it works in a
LabVIEW application: You write a server VI that is solely responsible
for, say, acquiring data from the hardware. Acquired data is prepared
and then written to one or more global variables that act as a data-
base. Then you design one or more client VIs that read the data in the
global variable(s), as shown in Figure 3.27. This is a very powerful con-
cept, one that you will see throughout this book and in many advanced
examples.

Client VI #1: Reads data from the

global variable and saves it in a file.

data global

Server VI: Acquires data, data

writes it to a global variable.

RUN
data global

data
Client VI #2: Reads data from the
global variable and displays it.
RUN
Unbundle
data global By Name Display
data value

RUN

Figure 3.27 Client-server systems. Client VIs receive data from a server VI
through the use of a global variable. This permits time independence between
several VIs or loops.
 Controlling Program Flow 91

A good use for this might be in process control (Chapter 18, “Process
Control Applications”) where the client tasks are such things as alarm
generation, real-time trending, historical trending to disk, and several
different operator-interface VIs. The beauty of the client-server concept
is that the clients can be

■ Independent loops on one diagram

■ Completely independent top-level VIs in different windows
■ Running at different rates
■ Running on different machines if you use a network

If you want one button to shut down everything, add a RUN boolean
control to a global variable and wire that into all the clients. You can
also set the priority of execution for each VI by two different means.
First, you can put a timer in each loop and make the high-priority
loops run more often. This is simple and foolproof and is generally the
recommended technique. Second, you can adjust the execution prior-
ity through the VI Properties dialog. Typically, the server would have
slightly higher priority to guarantee fresh data and good response to
the I/O systems while the clients would run at lower priority. Use exe-
cution priority with caution and a good dose of understanding about
how LabVIEW schedules VI execution. In a nutshell, high-priority VIs
cut in line ahead of low-priority VIs. If a high-priority VI needs service
too often, it may starve all the other VIs, and your whole hierarchy will
grind to a halt. You can read more about this topic in the LabVIEW
User’s Manual, in the chapter on performance issues. Other references
on this topic are listed in the Bibliography.
Race conditions are a risk with the client-server architecture
because global variables may be accessed without explicit ordering.
The most hazardous situation occurs when you have two or more loca-
tions where data is written to a global variable. Which location wrote
the latest data? Do you care? If so, you must enforce execution order by
some means. Another kind of race condition surrounds the issue of data
aging. For example, in Figure 3.27, the first client VI periodically stores
data in a file. How does that client know that fresh data is available?
Can it store the same data more than once, or can it miss new data?
This may be a problem. The best way to synchronize clients and serv-
ers is by using a global queue. A queue is like people waiting in line.
The server puts new data in the back of the queue (data is enqueued),
and the clients dequeue the data from the front at a later time. If the
queue is empty, the client must wait until there is new data. For a good
example of the queue technique, see the LabVIEW examples in the
directory examples/general/synchexm.llb/QueueExample.VI.
92 Chapter Three

Client-server (with autonomous VIs)

Autonomous client or server VIs can be launched by a master VI to run
independently without being placed in a While Loop on the diagram of
the master VI. Figure 3.28 shows how. Operation of the autonomous
Read Data Server VI is quite similar to that of the previous example
in that it acquires data at an independent rate and transmits it to the
master (client) VI through a global variable or queue. The server VI is
stopped by a global boolean when commanded by the master VI. Note
that the server VI is placed outside the While Loop in the master VI
because the server is called only once and is intended to run until the
master VI terminates. If the server were placed inside the While Loop,
that loop would not even finish its first cycle; it would wait (forever)
until the server VI finished.
There can be multiple autonomous VIs, and their panels may or may
not appear when the VIs are called. For instance, an alarm display VI
would certainly be displayed, while a VI that simply logs data to disk
probably would not. What if an error occurs in the autonomous VI? You
have several options. You can do nothing, as in my simple example, or
maybe return an error cluster when execution is finished. You could

Master VI: Reads data from the global variable and saves it in a file.

data global
Initialization
data
500 data global
Stop Run
data global
Read Data server
Run Stop the
Continuously writes server
Dependency
data to global

error out

Do it data data global

status code error out
no error 0

source data global

Do it
(dummy 200
input) Run

Server VI: Acquires data, writes it to a global variable.

Runs until global Run is False. Front panel is never shown.
Figure 3.28 The upper diagram is a master VI (and client) that launches the lower VI,
which is an autonomous server.
 Controlling Program Flow 93

have the VI generate a dialog box, but what happens after that? A ver-
satile solution is to have the autonomous VI set the Run global boolean
to False and wire the master VI in such a way that it is forced to stop
when the Run global becomes False. That way, any VI in the hierarchy
could potentially stop everything. Be careful. Don’t have the application
come to a screeching halt without telling the user why it just halted.
There is only a slight advantage to the autonomous VI solution com-
pared to placing multiple While Loops on the main diagram. In fact,
it’s an obscure performance benefit that you’ll notice only if you’re try-
ing to get the last ounce of performance out of your system. Here’s the
background: LabVIEW runs sections of code along with a copy of the
execution system in threads. This is called multithreading. The oper-
ating system manages when and where the threads run. Normally a
subVI uses the execution system of its calling VI, but in the VI Proper-
ties dialog (Figure 3.29), you can select a different execution system
and therefore ensure a different thread from the calling VI. Running
a subVI in a separate execution system increases the chances that it
will run in parallel with code in other execution systems on a multi-
processor system. Normally the disadvantage to placing a subVI in a
different execution system is that each call to the subVI will cause a
context switch as the operating system changes from one execution
system (thread) to another. That’s the background; now for the case of
the autonomous VI. If the autonomous VI is placed in a separate execu-
tion system, there will only be one context switch when it is originally
called, and then the autonomous VI will be merrily on its way in its
own execution system. As we said, the advantage is slight, but it’s there
if you need it.

Figure 3.29 There is a slight performance benefit to placing autonomous

VIs in their own execution system.
94 Chapter Three

State machines
There is a very powerful and versatile alternative to the Sequence
structure, called a state machine, as described in the advanced
LabVIEW training course. The general concept of a state machine orig-
inates in the world of digital (boolean) logic design where it is a formal
method of system design. In LabVIEW, a state machine uses a Case
structure wired to a counter that’s maintained in a shift register in a
While Loop. This technique allows you to jump around in the sequence
by manipulating the counter. For instance, any frame can jump directly
to an error-handling frame. This technique is widely used in drivers,
sequencers, and complex user interfaces and is also applicable to situa-
tions that require extensive error checking. Any time you have a chain
of events where one operation depends on the status of a previous oper-
ation, or where there are many modes of operation, a state machine is
a good way to do the job.
Figure 3.30 is an example of the basic structure of a state machine.
The state number is maintained as a numeric value in a shift register,
so any frame can jump to any other frame. Shift registers or local vari-
ables must also be used for any data that needs to be passed between
frames, such as the results of the subVI in this example. One of the con-
figuration tricks you will want to remember is to use frame 0 for errors.
That way, if you add a frame later on, the error frame number doesn’t
change; otherwise, you would have to edit every frame to update the
error frame’s new address.
Each frame that contains an activity may also check for errors
(or some other condition) to see where the program should go next.
In driver VIs, you commonly have to respond to a variety of error

1
Intermediate
results

Bad data 0
Initial state Get Data.vi next
1
state selector state
Good data 2

Keep running

Figure 3.30 A basic state machine. Any frame of the case can jump to
any other by manipulating the state selector shift register. Results are
passed from one frame to another in another shift register. Normal
execution proceeds from frame 1 through frame 2. Frame 0 or frame 2
can stop execution. This is an exceptionally versatile concept, well worth
studying.
 Controlling Program Flow 95

conditions where you may want to retry communications, abort, or do

something else depending on the condition detected. Similarly, in a com-
plex user interface, you may have to handle major changes in display
configuration (using Property nodes) or handle errors arising from user
data entries. The state machine lets you handle these complex situations.
In this example, the program starts in frame 1 where data is acquired. If
the data is no good, the program jumps to frame 0, reports the problem,
and stops the While Loop. If the data is OK, it gets analyzed in frame 2.
Note that an infinite loop could occur here; additional conditions should
be added in a practical program to guarantee stopping after some num-
ber of retries.
Tip: It’s easy to forget to stop the While Loop after the final frame.
The program will execute the last frame over and over, often with very
strange results. Remember to put the little False boolean constant
where it belongs—in the error handler frame and in the last frame of
the sequence.
A state machine can be the diagram behind a flexible user inter-
face, or it can be called as a subVI. In a user interface, the state
machine manages various modes of operation. For instance, the user
may have a series of mode selections when configuring an experi-
ment. A mode selector switch may directly drive the Case structure,
or you may use Property nodes to selectively hide and show valid
mode selection buttons.
Depending upon which button is then pressed, the state machine
jumps to an appropriate frame to handle that selection.
A favorite use for state machines is intelligent subVIs in such appli-
cations as sequencers and profile generators. In those situations, the
state machine subVI is called periodically in the main While Loop.
Each time it’s called, information from the last cycle is available in the
shift register, providing the required history: Where have I been? Data
passed in through the connector pane provides the current information:
Where am I now? And computation within the state machine subVI
determines the future: Where should I go next? Those values may drive
an I/O device through a subVI within the state machine, or they may be
passed back to the calling VI through the connector pane. Figure 3.31 is
an example, and it works as just described. This VI is called every sec-
ond from the main While Loop. Inside, it calls a state machine, Recipe
Control Logic, which keeps track of the state of a complex ramp-and-
soak temperature recipe. On each iteration, the state machine receives
the current process variable (temperature) measurement and some
commands regarding whether or not to continue running the recipe. It
computes a new controller set point, which is returned and then trans-
mitted to the Eurotherm 808 controller.
96 Chapter Three

Settings
Status
Setpoint A/M 1 Heating
Manual
• Recipe Setpoint (°C)
Ramp & Soak
0.0 Now on step Controller.vi
• Manual
3
Restart recipe Controller Addr.
Recipe Group ID Minutes into step
• Run 17.00
0

• Pause
Unit ID
Furnace On Line?
1
• On Line

• Off

Eurotherm 808 Read 3 [0..3] Eurotherm

Run recipe, in AUTO 808 Write
Settings PV
Recipe Control Logic.vi
SL

Restart recipe Now on step

Recipe Minutes into step
Setpoint A/M Status

State machine inside

Figure 3.31 A peek inside the Ramp-and-Soak Controller VI (simplified), which is called
from the main VI. A state machine, Recipe Control Logic, keeps track of complex tempera-
ture profile recipes.

Great confusion can arise when you are developing a state machine.
Imagine how quickly the number of frames can accumulate in a
situation with many logical paths. Then imagine what happens when
one of those numeric constants that points to the next frame has the
wrong number. Or even worse, what if you have about 17 frames and
you need to add one in the middle? You may have to edit dozens of little
numeric constants. And just wait until you have to explain to a novice
user how it all works!
Thankfully, creative LabVIEW developers have come up with an easy
solution to these complex situations: Use enumerated constants
(also called enums) instead of numbers. Take a look at Figure 3.32,
showing an improved version of our previous state machine example.
You can see the enumerated constant, with three possible values (error,
get data, calculate), has replaced the more cryptic numeric constants.
The Case structure now tells you in English which frame you’re on, and
the enums more clearly state where the program will go next. Strings
are also useful as a self-documenting case selector.
 Controlling Program Flow 97

A "get data"
B
Intermediate
results
Bad data "get data"
Initial state error string
Get Data.vi next
get data get data
state
calculate Keep running
Good data

Enum constant with possible states:

{error, get data, calculate}

Figure 3.32 (A) A recommended way to write state machines, using enumerated constants.
(B) Strings are also useful as self-documenting selectors.

Even with enums or strings as selectors, there’s plenty of room for

confusion in designing, debugging, or modifying state machines. Life
is a bit easier when you design them with the aid of a state dia-
gram, an old digital logic designer’s tool shown in Figure 3.33. Each
bubble represents a state where a task is performed and corresponds
to a frame in the Case structure. Paths from one state to another are
annotated to show why the transition is made. Many computer-aided
software engineering (CASE) applications have a state diagram tool

Figure 3.33 LabVIEW State Diagram Editor makes it easy to

design and modify complex state machines. Notice that the only
way out of this state machine is through an error – whoops!
Another transition needs to go from Get Input to Stop.
98 Chapter Three

Figure 3.34 LabVIEW state machine created with the State Diagram Editor Tool-
kit. It is easy to design and prototype, but you cannot modify the state machine
without the editor.

with additional features to simplify state machine design. Figure 3.33

is from the LabVIEW State Diagram Editor Toolkit, and Figure 3.34
shows the block diagram it creates.

Queued message handler

A Queued Message Handler takes a list (queue) of messages and pro-
cesses them in order. Originally called queued state machines, they are
actually fundamentally different from a state machine. A state machine
has predefined transitions built into its architecture. In the state dia-
gram of Figure 3.33 there is a flaw in the logic built into the state
machine, and you cannot get to the stop case without going through
the error case. Because the behavior is built in, you don’t have a way
around it without editing the state machine. State machines are best
when you have a strictly defined behavior (and you code it properly).
That’s often not the case in the lab; the equipment may stay the same,
but the experiment changes from day to day. When the components
of your program stay the same but the order of execution changes, a
queued message handler is a powerful tool.
 Controlling Program Flow 99

Figure 3.35 A queued message handler sequences through a list of messages.

Each case of a queued message handler can jump to any other case,
but unlike for a state machine, the order of the execution is defined
by the list of commands (message queue or array) that you pass in.
In Figure 3.35 the message queue is held in a shift register. With
each iteration of the While Loop, a message at the front of the queue
is pulled off and the appropriate case executes. The loop terminates
when the queue is empty or an Exit command is reached. Note that
we’re also using a type defined cluster as a Data structure. The Data
structure is a defined mechanism providing operating parameters
for subVIs and a way to get data in and out of the queued message
handler. The VI in Figure 3.35 would be used as a subVI in a larger
application.
The queued message handler architecture works best in combination
with other design patterns. It’s tempting to create a queued message
handler with many atomic states, giving you fine-grained control over
your equipment, but this creates problems when one state depends on
another state which depends on another state, etc. Then the calling VI
has to know too much about the inner workings of the queued message
handler. A better solution is to use the queued message handler to call
intelligent state machines. Each intelligent state machine manages all
the specifics of a piece of equipment or of a process. Now the queued
message handler specifies actions, not procedures. Once you master
this, you are well on your way to building robust, reusable applications
that are driven by input parameters, not constrained by predefined
behavior.
100 Chapter Three

Event-driven applications
Event-driven applications are almost trivially simple with the Event
structure. Our guidelines for using the Event structure are as follows:

■ Use only one Event structure per block diagram. Let a single Event
structure provide a focal point for all user-interface code.
■ Do only minimal processing inside an Event structure. Lengthy pro-
cessing inside the event case will lock up the user interface.
■ Use a separate While Loop containing only the Event structure.

The locking nature of the Event structure and the need for immedi-
ate attention to events dictate that you use two parallel loops. You can
use a single loop for both the Event structure and the rest of your code,
but eventually you’ll run up against a timing issue requiring you to
parallelize your application. The Producer/Consumer Design Pattern
template in LabVIEW provides a great place to start.
Figure 3.36 is a modified version of LabVIEW’s Producer/Consumer
Design Pattern. The consumer loop is very similar to the queued mes-
sage handler. Instead of passing in an array we’re using the queue
functions to pass in each item as it happens. A queue is a FIFO (first
in, first out). Messages are enqueued at one end and dequeued at the
other. For more on queues and other synchronization functions, look
at Chapter 6, “Synchronization.” Note that we’re giving our queue a
unique name based on the time of creation. Queues are referenced by
name, and all you need is the name of the queue to put items in it. If
we left the name empty, there’s a real good chance another nameless
queue would put items in ours with disastrous results. We’re also using
a string as the queue element, because they’re easy to use as case selec-
tors. After creation the first thing we do is to enqueue “Init,” telling
our consumer loop to run its initialization case. In the earlier design
patterns, initialization routines were outside the main loop. We still
initialize first, but keep our block diagram clean and simple. We can
also reinitialize without leaving our consumer loop.
The event structure is the heart of the producer loop. The event
case in Figure 3.36 is registered for user menu events. Select “Edit >>
Run-Time Menu…” to create a custom menu for your application
(Figure 3.37). As soon as the user makes a menu selection from your
running application, the item path is enqueued in the event case and
dequeued in the consumer loop. The full item path is a string with a
colon separator between items. Passing menu items as the full item
path eliminates errors we might have if two menus had an item with
the same name. When the user selects New from the File menu, we get
the string “File:New.” The consumer loop uses Match Pattern to split
 Controlling Program Flow 101

Figure 3.36 Producer/Consumer Design Pattern modified for event-driven application. User menu
selections are placed in the queue and handled in the consumer loop. Match Pattern splits the
menu item path at the “:” (colon) separator. Nested cases handle the event actions. (Error handling
is omitted for clarity.)

Figure 3.37Create a custom menu for your application and use the event-
driven Producer/Consumer Design Pattern to handle user selections.
102 Chapter Three

the string at the first colon. We get the two strings File and New. If there
is not a colon in the string, Match Pattern returns the whole string
at its “before substring” terminal. Passing the full item path between
loops also allows us to modularize the consumer loop with a case hier-
archy based partly on menu structure. This event-driven design pat-
tern is powerful and very easy to design.
Design patterns provide a starting point for successful program-
ming. There is no one and only design pattern; otherwise, we would
have shown you that first and omitted the rest. Pick the design pattern
that best fits your skill level and your application. There is no sense in
going for something complicated if all you need is to grab a few mea-
surements and write them to file.

Bibliography
AN 114, Using LabVIEW to Create Multithreaded VIs for Maximum Performance and
Reliability, www.ni.com, National Instruments Corporation, 11500 N. Mopac Express-
way, Austin, Tex., 2000.
AN 199, LabVIEW and Hyper-Threading, www.ni.com, National Instruments Corpora-
tion, 11500 N. Mopac Expressway, Austin, Tex., 2004.
Brunzie, Ted J.: “Aging Gracefully: Writing Software That Takes Changes in Stride,”
LabVIEW Technical Resource, vol. 3, no. 4, Fall 1995.
“Inside LabVIEW,” NIWeek 2000, Advanced Track 1C, National Instruments Corpora-
tion, 11500 N. Mopac Expressway, Austin, Tex., 2000.
 Chapter

4
LabVIEW Data Types

Our next project is to consider all the major data types that LabVIEW
supports. The list consists of scalar (single-element) types such as
numerics and booleans and structured types (containing more than one
element) such as strings, arrays, and clusters. The LabVIEW Controls
palette is roughly organized around data types, gathering similar types
into subpalettes for easy access. There are, in fact, many ways to present
a given data type, and that’s the main reason that there are apparently
so many items in those palettes. For instance, a given numeric type can
be displayed as a simple number, a bar graph, a slider, or an analog
meter, or in a chart. But underneath is a well-defined representation
of the data that you, the programmer, and the machine must mutually
understand and agree upon as part of the programming process.
One thing that demonstrates that LabVIEW is a complete program-
ming language is its support for essentially all data types. Numbers can
be floating point or integer, with various degrees of precision. Booleans,
bytes, strings, and numerics can be combined freely into various struc-
tures, giving you total freedom to make the data type suit the problem.
Polymorphism is an important feature of LabVIEW that simplifies
this potentially complicated world of data types into something that
even the novice can manage without much study. Polymorphism is the
ability to adjust to input data of different types. Most built-in LabVIEW
functions are polymorphic. Ordinary virtual instruments (VIs) that
you write are not truly polymorphic—they can adapt between numeric
types, but not between other data types such as strings to numerics.
Most of the time, you can just wire from source to destination without
much worry since the functions adapt to the kind of data that you sup-
ply. How does LabVIEW know what to do? The key is object-oriented
programming, where polymorphism is but one of the novel concepts

103
104 Chapter Four

that make this new programming technology so desirable. There are, of

course, limits to polymorphism, and we’ll see later how to handle those
special situations.

Numeric Types
Perhaps the most fundamental data type is the numeric, a scalar value
that may generally contain an integer or a real (floating-point) value.
LabVIEW explicitly handles all the possible integer and real repre-
sentations that are available on current 32-bit processors. LabVIEW 8
adds support for 64-bit integers. Figure 4.1 displays the terminals for
each representation, along with the meaning of each and the number
of bytes of memory occupied. LabVIEW floating-point types follow the
IEEE-754 standard, which has thankfully been adopted by all major
CPU and compiler designers.
The keys to choosing an appropriate numeric representation in
most situations are the required range and precision. In general,
the more bytes of memory occupied by a data type, the greater the
possible range of values. This factor is most important with integers;
among floating-point types, even single precision can handle values
up to 1038, and it’s not too often that you need such a large number. An
unsigned integer has an upper range of 2N − 1, where N is the num-
ber of bits. Therefore, an unsigned byte ranges up to 255, an unsigned
integer (2 bytes) up to 65,535, and an unsigned long integer up to
4,294,967,295. Signed integers range up to 2N − 1, or about one-half
of their unsigned counterparts.
If you attempt to enter a value that is too large into an integer con-
trol, it will be coerced to the maximum value. But if an integer math-
ematical process is going on, overflow or underflow can produce
erroneous results. For instance, if you do an unsigned byte addition of
255 + 1, you get 0, not 256—that’s overflow.

Scalar numerics in LabVIEW cover the gamut of integer and real types.
Figure 4.1
The number of bytes of memory required for each type is shown.
 LabVIEW Data Types 105

Floating-point precision can be a very confusing subject, especially

if you’re prone to being anal-retentive. (The info-labview mailgroup
has seen some lively discussions on this subject—to the point of “flame
wars.”) While integers by definition have a precision of exactly 1,
floating-point types are inexact representations. The reason is that
while you may be most comfortable entering and displaying a floating-
point value in decimal (base-10) notation, the computer stores the value
as a binary (base-2) value with a limited number of bits. Some rounding
must occur, so this base-converted storage is therefore inexact.
For example, enter 3.3 into a single-precision, floating-point control,
pop up on the control and set the precision to a large number of digits,
and note that the displayed value is 3.2999999523. . . , which is just a
hair different from what you entered.
The question is, How important is that rounding error to your
application? You should evaluate precision in terms of significant
figures. For instance, the number 123.450 has six significant figures.
According to the IEEE-754 standard, a single-precision float guarantees
7.22 significant figures, a double-precision float gives you 15.95, and an
extended-precision float gives you 19.26. When you configure a numeric
indicator to display an arbitrary number of digits, all digits beyond the
IEEE-specified number of significant figures are insignificant! That is,
you must ignore them for the purposes of computation or display. Double-
precision values are recommended for most analyses and mathematical
operations to reduce cumulative rounding errors. Single-precision values
are sufficient for transporting data from most real-world instruments
(with simple math) since they give you a resolution of 0.1 part per
million, which is better than most real, physical measurements.
IEEE floating-point numbers have a bonus feature: They can tell you
when the value is invalid. If you divide by zero, the value is infinity, and
a LabVIEW numeric indicator will say inf. Another surprisingly useful
value is NaN, or not-a-number. It makes a great flag for invalid data.
For instance, if you’re decoding numbers from a string (see the next sec-
tion) and a valid number can’t be found, you can have the function return
NaN instead of zero or some other number that might be mistaken for
real data. Also, when you are graphing data, NaN does not plot at all.
That’s a handy way to edit out undesired values or to leave a blank
space on a graph. You can test for NaN with the comparison function
Not A Number/Path/Refnum, .

Strings
Every programmer spends a lot of time putting together strings of
characters and taking them apart. Strings are useful for indicators
where you need to say something to the operator, for communications
106 Chapter Four

with a general-purpose interface bus (GPIB) and serial instruments,

and for reading and writing data files that other applications will
access. If you’ve had the displeasure of using old FORTRAN (say, the
1966 vintage), you know what it’s like to deal with strings in a lan-
guage that didn’t even have a string or character type, let alone string
functions. We LabVIEW users are in much better shape, with a nice
set of string-wrangling functions built right in. If you know anything
about the C language, some of these functions will be familiar. Let’s
look at some common string problems and their solutions.

Building strings
Instrument drivers are the classic case study for string building. The
problem is to assemble a command for an instrument (usually GPIB)
based on several control settings. Figure 4.2 shows how string building
in a typical driver works.

1. The Pick Line function is driven by a ring control that supplies a

number (0, 1, or 2) that is mapped into the words OFF, SHORT, or
LONG. These keywords are appended to the initial command string
STPGEN PUL:.
2. The Concatenate Strings function tacks on a substring CUR:. If
you need to concatenate several strings, you can resize Concatenate
Strings (by dragging at any corner) to obtain more inputs.
3. The Format Value function is versatile for appending a single
numeric value. You need to learn about C-style formatting commands

Figure 4.2 String building in a driver VI. This one uses most of the major string-building
functions in a classic diagonal layout common to many drivers.
 LabVIEW Data Types 107

(see the LabVIEW online help for details) to use this function. The
percent sign tells it that the next few characters are a formatting
instruction. What is nice about this function is that not only does
it format the number in a predictable way, but also, it allows you
to tack on other characters before or after the value. This saves
space and gets rid of a lot of Concatenate Strings functions. In this
example, the format string %e translates a floating-point value in
exponential notation and adds a trailing comma.
4. A more powerful function is available to format multiple values:
Format Into String. You can pop up on this function and select
Edit Format String to obtain an interactive dialog box where you
build the otherwise cryptic C-style formatting commands. In this
case, it’s shown formatting an integer, a float, and a string into one
concatenated output with a variety of interspersed characters. Note
that you can also use a control or other string to determine the
format. In that case, there could be run-time errors in the format
string, so the function includes error I/O clusters. It is very handy
and very compact, similar to its C counterpart sprintf().
5. The Append True/False String function uses a boolean control to
pick one of two choices, such as ON or OFF, which is then appended
to the string. This string-building process may continue as needed to
build an elaborate instrument command.

Parsing strings
The other half of the instrument driver world involves interpreting
response messages. The message may contain all sorts of headers,
delimiters, flags, and who knows what, plus a few numbers or important
letters that you actually want. Breaking down such a string is known
as parsing. It’s a classic exercise in computer science and linguistics
as well. Remember how challenging it was to study our own language
back in fifth grade, parsing sentences into nouns, verbs, and all that?
We thought that was tough; then we tackled the reply messages that
some instrument manufacturers come up with! Figure 4.3 comes from
one of the easier instruments, the Tektronix 370A driver. A typical
response message would look like this:

STPGEN NUMBER:18;PULSE:SHORT;OFFSET:-1.37;NVERT:OFF;MULT:ON;

Let’s examine the diagram that parses this message.

1. The top row is based on the versatile Match Pattern function.

Nothing fancy is being done with it here, other than searching for
a desired keyword. The manual fully describes the functions of the
108 Chapter Four

Figure 4.3 Using string functions to parse the response message from a GPIB instrument.
This is typical of many instruments you may actually encounter.

many special characters you can type into the regular expression
input that controls Match Pattern. One other thing we did is to pass
the incoming string through the To Upper Case function. Other-
wise, the pattern keys would have to contain both upper- and lower-
case letters. The output from each pattern match is known as after
substring. It contains the remainder of the original string immedi-
ately following the pattern, assuming that the pattern was found. If
there is any chance that your pattern might not be found, test the
offset past match output; if it’s less than zero, there was no match,
and you can handle things at that point with a Case structure.
2. Each of the after substrings in this example is then passed to another
level of parsing. The first one uses the Decimal String To Number
function, one of several specialized number extractors; others handle
octal, hexadecimal, and fraction/scientific notation. These are very
robust functions. If the incoming string starts with a valid numeric
character, the expected value is returned. The only problem you can
run into occurs in cases where several values are run together such
as [123.E3-.567.89]. Then you need to use the Scan From String
function to break it down, provided that the format is fixed. That is,
you know exactly where to split it up. You could also use Match Pat-
tern again if there are any other known, embedded flags, even if the
flag is only a space character.
3. After locating the keyword PULSE, we expect one of three possible
strings: OFF, SHORT, or LONG. Match First String searches a
string array containing these words and returns the index of the
one that matches (0, 1, or 2). The index is wired to a ring indicator
named Pulse that displays the status.
 LabVIEW Data Types 109

4. Keyword OFFSET is located, and the substring is passed to the

Fract/Exp String To Number function to extract the number,
which in this case is in fractional form.

5. INVERT is located next, followed by one of two possible strings,

ON or OFF. When Gary first wrote this VI, he first used Index and
Strip again, with its search array containing ON and OFF. But a bug
cropped up! It turned out that another keyword farther along in the
string (MULT) also used ON and OFF. Since Match Pattern passes
us the entire remainder of the string, Index and Strip happily went
out and found the wrong ON word—the one that belonged to MULT.
Things were fine if the INVERT state was ON, since we found that
right away. The solution he chose was to use the String Subset func-
tion, split off the first three characters, and test only those. You could
also look for the entire command INVERT:ON or INVERT:OFF.

The moral of the story: Test thoroughly before shipping.

The Scan From String function, like the C scanf() function, can
save much effort when you are parsing strings (Figure 4.4). Like its
complement Format Into String, this function produces one or more
polymorphic outputs (numeric or string), depending upon the contents
of the Format String input. Input terminals allow you to determine the
data format of numeric outputs and to set default values as well.
Other difficult parsing problems arise when you attempt to extract
information from text files. If the person who designed the file format
is kind and thoughtful, all you will have to do is to search for a key-
word, then read a number. There are other situations that border on
the intractable; you need to be a computer science whiz to write a reli-
able parser in the worst cases. These are very challenging problems,
so don’t feel ashamed if it takes you a long time to write a successful
LabVIEW string parsing VI.

Figure 4.4 The Scan From String function is great for parsing simple strings.
110 Chapter Four

Dealing with unprintables

Sometimes you need to create or display a string that contains some
of the unprintable ASCII characters, such as control-x or the escape
character. The trick is to use the pop-up item ‘\’ Codes Display,
which works on front panel control and indicators as well as diagram
string constants. LabVIEW will interpret one- or two-character codes
following the backslash character, as shown in Table 4.1. You can also
enter unprintables by simply typing them into the string. Control
characters, carriage returns, and so on all work fine, except for the
Tab (control-i). LabVIEW uses the Tab key to switch tools, so you
have to type \t. Note that the hexadecimal codes require uppercase
letters. One other disconcerting feature of these escape codes is this:
If you enter a code that can be translated into a printable character
(for instance, \41 = A), the printable character will appear as soon as
you run the VI. To solve that problem, you can pop up on the string
and select Hex Display. Then the control simply displays a series of
two-digit hexadecimal values without ASCII equivalents.
There are many conversion functions in the String palette. Figure 4.5
shows just a few of them in action. These string converters are the
equivalent of functions such as ASC() and CHR() in Basic, where you
convert numbers to and from strings. The Scan From String function
is particularly powerful and not limited to hexadecimal conversion. You
expand it to suit the number of variables in the source string, then enter
an appropriate number of format specifiers in the format string. As with
its complement Format Into String, you can pop up on this function
and select Edit Format String to build formatting commands.

Spreadsheets, strings, and arrays

Frequently you will need to write your data to disk for later use by
spreadsheet programs and other applications that prefer tab-delimited
text. Such a string might look like this:

value <tab> value <tab> value <cr>

TABLE 4.1 Escape Sequences (\ Codes) for Strings

Escape codes Interpreted as
\00-\FF Hexadecimal value of an 8-bit character
\b Backspace (ASCII BS or equivalent to \08)
\f Formfeed (ASCII FF or equivalent to \0C)
\n Newline (ASCII LF or equivalent to \0A)
\r Return (ASCII CR or equivalent to \0D)
\t Tab (ASCII HT or equivalent to \09)
\s Space (equivalent to \20)
\\ Backslash (ASCII\or equivalent to \5C)
 LabVIEW Data Types 111

Figure 4.5 Here are some of the ways you can convert to and from strings that contain unprintable characters.
All these functions are found in the String palette.

Sometimes the delimiter is a comma or other character. No matter,

LabVIEW handles all delimiters. A very important portability note is
in order here. Most platforms are different with respect to their end-
of-line character. A platform-independent constant, End Of Line ( )
is available from the String function palette that automatically gener-
ates the proper character when ported. The proper characters are

Macintosh: carriage return (\r)

PC: carriage return, then line feed (\r\n)
UNIX: line feed (\n) (ul-end)

A really interesting problem crops up with string controls when the

user types a carriage return into the string: LabVIEW always inserts
a line feed. For the examples in this book, we generally use a carriage
return, since we’re Mac users. But remember this portability issue if
you plan to carry your VIs from one machine to another.
To convert arrays of numerics to strings, the easiest technique is to
use the Array To Spreadsheet String function. Just wire your array
into it along with a format specifier, and out pops a tab-delimited string,
ready to write to a file. For a one-dimensional (1D) array (Figure 4.6),
tab characters are inserted between each value, and a carriage return
is inserted at the end. This looks like one horizontal row of numbers
in a spreadsheet. If you don’t want the default tab characters between
values, you can wire a different string to the delimiter input.
You can read the data back in from a file (simulated here) and con-
vert it back into an array by using the Spreadsheet String To Array
function. It has an additional requirement that you supply a type
specifier (such as a 1D array) to give it a hint as to what layout the
data might have. The format specifier doesn’t have to show the exact
112 Chapter Four

Figure 4.6 Converting a 1D array to a spreadsheet-compatible string and back.

field width and decimal precision; plain “%e” or “%d” generally does the
job for floating-point or integer values, respectively.
In Figure 4.7, a two-dimensional (2D) array, also known as a matrix,
is easily converted to and from a spreadsheet string. In this case, you
get a tab between values in a row and a carriage return at the end of
each row. If the rows and columns appear swapped in your spread-
sheet program, insert the array function Transpose 2D Array before
you convert the array to text. Notice that we hooked up Spreadsheet
String To Array to a different type specifier, this time a long integer
(I32). Resolution was lost because integers don’t have a fractional part;
this demonstrates that you need to be careful when mixing data types.
We also obtained that type specifier by a different method—an array

Figure 4.7 Converting an array to a table and back, this time using a 2D array (matrix).
 LabVIEW Data Types 113

constant, which you can find in the array function palette. You can
create constants of any data type on the diagram. Look through the
function palettes, and you’ll see constants everywhere. In this case,
choose an array constant, and then drag a numeric constant into it.
This produces a 1D numeric array. Then pop up on the array and select
Add Dimension to make it 2D.
A more general solution to converting arrays to strings is to use a
For Loop with a shift register containing one of the string conversion
functions and Concatenate Strings. Sometimes this is needed for more
complicated situations where you need to intermingle data from several
arrays, you need many columns, or you need other information within
rows. Figure 4.8 uses these techniques in a situation where you have a
2D data array (several channels and many samples per channel), another
array with time stamps, and a string array with channel names.

Figure 4.8 A realistic example of building a spreadsheet string from a 2D data

array, a timestamp array, and a string array containing channel names for the
file’s header. If this solution is still too slow for your application, you have to imple-
ment it in C code through a Code Interface node.
114 Chapter Four

The names are used to build a header in the upper For Loop, which is
then concatenated to a large string that contains the data. This business
can be a real memory burner (and slow as well) if your arrays are large.
The strings compound the speed and memory efficiency problem.
This is one of the few cases where it can be worth writing a Code
Interface node, which may be the only way to obtain acceptable per-
formance with large arrays and complex formatting requirements. If
you’re doing a data acquisition system, try writing out the data as you
collect it—processing and writing out one row at a time—rather than
saving it all until the end. On the other hand, file I/O operations are
pretty slow, too, so some optimization is in order.

Arrays
Any time you have a series of numbers or any other data type that needs
to be handled as a unit, it probably belongs in an array. Most arrays
are one-dimensional (1D, a column or vector), a few are 2D (a matrix),
and some specialized data sets require 3D or greater. LabVIEW permits
you to create arrays of numerics, strings, clusters, and any other data
type (except for arrays of arrays). Arrays are often created by loops, as
shown in Figure 4.9. For Loops are the best because they preallocate
the required memory when they start. While Loops can’t; LabVIEW
has no way of knowing how many times a While Loop will cycle, so the
memory manager will have to be called occasionally, slowing execution
somewhat.
You can also create an array by using the Build Array function
(Figure 4.10). Notice the versatility of Build Array: It lets you concat-
enate entire arrays to other arrays, or just tack on single elements.
It’s smart enough to know that if one input is an array and the other
is a scalar, then you must be concatenating the scalar onto the array

Figure 4.9Creating arrays using a For Loop. This is an efficient way to build arrays with many elements.
A While Loop would do the same thing, but without preallocating memory.
 LabVIEW Data Types 115

Figure 4.10 Using the Build Array function. Use the pop-up
menu on its input terminals to determine whether the input is
an element or an array. Note the various results.

and the output will be similar to the input array. If all inputs have the
same dimensional size (for instance, all are 1D arrays), you can pop
up on the function and select Concatenate Inputs to concatenate them
into a longer array of the same dimension (this is the default behavior).
Alternatively, you can turn off concatenation to build an array with
n + 1 dimensions. This is useful when you need to promote a 1D array
to 2D for compatibility with a VI that required 2D data.
To handle more than one input, you can resize the function by dragging
at a corner. Note the coercion dots where the SGL (single-precision
floating point) and I16 (2-byte integer) numeric types are wired to the
top Build Array function. This indicates a change of data type because
an array cannot contain a mix of data types. LabVIEW must promote
all types to the one with the greatest numeric range, which in this
case is DBL (double-precision floating point). This also implies that you
can’t build an array of, say, numerics and strings. For such intermixing,
you must turn to clusters, which we’ll look at a bit later.
Figure 4.11 shows how to find out how many elements are in an
array by using the Array Size function. Note that an empty array has
zero elements. You can use Index Array to extract a single element.
Like most LabVIEW functions, Index Array is polymorphic and will
return a scalar of the same type as the array.
If you have a multidimensional array, these same functions still
work, but you have more dimensions to keep track of (Figure 4.12).
Array Size returns an array of values, one per dimension. You have
to supply an indexing value for each dimension. An exception occurs
when you wish to slice the array, extracting a column or row of data,
as in the bottom example of Figure 4.12. In that case, you resize Array
Size, but don’t wire to one of the index inputs. If you’re doing a lot of
work with multidimensional arrays, consider creating special subVIs
116 Chapter Four

Figure 4.11 How to get the size of an array and fetch a single value. Remember
that all array indexing is based on0, not1.

to do the slicing, indexing, and sizing. That way, the inputs to the subVI
have names associated with each dimension, so you are not forced
to keep track of the purpose for each row or column. Another tip for
multidimensional arrays is this: It’s a good idea to place labels next to
each index on an array control to remind you which is which. Labels
such as row and column are useful.

Figure 4.12Sizing, indexing, and slicing 2D arrays are a little different, since you
have two indices to manipulate at all steps.
 LabVIEW Data Types 117

Figure 4.13 Two of the more powerful array editing functions are Delete From Array and Replace Array Subset.
Like all the array functions, they work on arrays of any dimension.

Besides slicing and dicing, you can easily do some other useful array
editing tasks. In Figure 4.13, the Delete From Array function removes
a selected column from a 2D array. The function also works on arrays of
other dimensions, such as all array functions. For instance, you can delete
a selected range of elements from a 1D array. In that case, you wire a
value to the Length input to limit the quantity of elements to delete.
The Replace Array Subset function is very powerful for array
surgery. In this example, it’s shown replacing two particular values in
a 3 × 3 matrix. You can replace individual values or arrays of values
in a single operation. One other handy function not shown here is
Insert Into Array. It does just what you’d expect: Given an input
array, it inserts an element or new subarray at a particular location.
When we didn’t have these functions, such actions required extensive
array indexing and building, which took up valuable diagram space
and added much overhead. You newcomers have it easy.

Initializing arrays
Sometimes you need an array that is initialized when your program
starts, say, for a lookup table. There are many ways to do this, as shown
in Figure 4.14.

■ If all the values are the same, use a For Loop with a constant inside.
Disadvantage: It takes a certain amount of time to create the array.
■ Use the Initialize Array function with the dimension size input
connected to a constant numeric set to the number of elements. This
is equivalent to the previous method, but more compact.
■ Similarly, if the values can be calculated in some straightforward
way, put the formula in a For Loop instead of a constant. For instance,
a special waveform or function could be created in this way.
118 Chapter Four

Figure 4.14 Several ways of programmatically initializing arrays. For

large amounts of data, there are tradeoffs between data storage on disk
and speed of execution with each one.

■ Create a diagram array constant, and manually enter the desired

values. Disadvantage: It’s tedious and uses memory on disk when
the VI is saved.
■ Create a front-panel array control and manually type in the values.
Select Make Current Value Default from the control’s Data
Operations pop-up menu. From now on, that array will always
have those values unless you change them. By popping up on the
control or its terminal, you can select Hide Front Panel Control
or position the control off-screen to keep anyone from modifying the
data. Disadvantage: The data takes up extra space on the disk when
you save the VI.
■ If there is lots of data, you could save it in a file and load it at start-up.

A special case of initialization is that of an empty array. This is not

an array with one or more values set to zero, false, empty string, or the
like! It contains zero elements. In C or Pascal, this corresponds to creat-
ing a new pointer to an array. The most frequent use of an empty array
is to initialize or reset a shift register that is used to hold an array.
Here are some ways to create an empty array:
■ Create a front-panel array control. Select Empty Array from its Data
Operations pop-up, and then Make Current Value Default from
the control’s pop-up menu. This is, by definition, an empty array.
■ Create a For Loop with the count terminal wired to zero. Place
a diagram constant of an appropriate type inside the loop and wire
outside the loop. The loop will execute zero times (i.e., not at all),
but the array that is created at the loop border tunnel will have the
proper type.
 LabVIEW Data Types 119

■ Use the Initialize Array function with the dimension size input
unconnected. This is functionally equivalent to the For Loop with
N = 0.
■ Use a diagram array constant. Select Empty Array from its Data
Operations pop-up menu.

Note that you can’t use the Build Array function. Its output always
contains at least one element.

Array memory usage and performance

Perhaps more than any other structure in LabVIEW, arrays are respon-
sible for a great deal of memory usage. It’s not unusual to collect thou-
sands or even millions of data points from a physics experiment and
then try to analyze or display them all at once. Ultimately, you may see
a little bulldozer cursor and/or a cheerful dialog box informing you that
LabVIEW has run out of memory. There are some things you can do to
prevent this occurrence.
First, read the LabVIEW Technical Note Managing Large Data Sets
in LabVIEW; we’ll summarize it here, but you can find the full article
online at www.ni.com. LabVIEW does its best to conserve memory.
When an array is created, LabVIEW has to allocate a contiguous area
of memory called a data buffer in which to store the array. (By the
way, every data type is subject to the same rules for memory allocation;
arrays and strings, in particular, just take up more space.) Figure 4.15
contains a sampling of functions that do and do not reuse memory in

Functions that can reuse buffers Functions that cannot reuse buffers

Multiply Increment
Build Array Array to Concatenate Strings
Replace Array Spreadsheet
Element String

Code Interface
For loop, indexing Node with input-
Bundle Transpose 2D Array
and reassembling output terminal
an array

Figure 4.15 Here are some of the operations that you can count on for predictable reuse or non-reuse of
memory. If you deal with large arrays or strings, think about these differences.
120 Chapter Four

a predictable way. If you do a simple operation such as multiplying

a scalar by an array, no extra memory management is required. An
array that is indexed on the boundary of a For Loop, processed, and
then rebuilt also requires no memory management. The Build Array
function, on the other hand, always creates a new data buffer. It’s bet-
ter to use Replace Array Element on an existing array, as shown
in Figure 4.16. A quick benchmark of this example showed a perfor-
mance increase of 30 times. This is one of the few cases where you can
explicitly control the allocation of memory in LabVIEW, and it’s highly
advisable when you handle large arrays or when you require maxi-
mum performance.
One confusing issue about multidimensional arrays involves
keeping track of the indices. Which one is the row and which one is
the column?
Which one does a nested loop structure act upon first? You can keep
track of this by looking at the indicator on the panel, such as the one
in Figure 4.16. The top index, called channel, is also the top index
on the Replace Array Element function; this is also true of the Index
Array function.

Figure 4.16 Use Replace Array Element inside a For Loop instead of Build Array
to permit reuse of an existing data buffer. This is much faster. The bottom example
uses five memory buffers compared to two on the top.
 LabVIEW Data Types 121

Figure 4.17 Organization of array indices.

So at least you can keep track of that much. By the way, it is consid-
ered good form to label your array controls and wires on the diagram
as we did in this example. When you access a multidimensional array
with nested loops as in the previous example, the outer loop accesses
the top index and the inner loop accesses the bottom index. Figure 4.17
summarizes this array index information. The Index Array function
even has pop-up tip strips that appear when you hold the wiring tool
over one of the index inputs. They call the index inputs column, row,
and page, just as we did in this example.
All this memory reuse business also adds overhead at execution time
because the memory manager has to be called. Talk about an over-
worked manager! The poor guy has to go searching around in RAM,
looking for whatever size chunk the program happens to need. If a space
can’t be found directly, the manager has to shuffle other blocks around
until a suitable hole opens up. This can take time, especially when mem-
ory is getting tight. This is also the reason your VIs sometimes execute
faster the second time you run them: Most of the allocation phase of
memory management is done on the first iteration or run.
Similarly, when an array is created in a For Loop, LabVIEW can usu-
ally predict how much space is needed and can call the memory manager
just once. This is not so in a While Loop, since there is no way to know in
advance how many times you’re going to loop. It’s also not so when you
are building arrays or concatenating strings inside a loop—two more
situations to avoid when performance is paramount. A new feature in
LabVIEW 7.1 Professional Version shows buffer allocations on the block
diagram. Select Tools >> Advanced >> Show Buffer Allocations to
bring up the Show Buffer Allocations window. Black squares will
appear on the block diagram, showing memory allocations. The effect is
subtle, like small flakes of black pepper; but if you turn them on and off,
122 Chapter Four

the allocations will jump out at you. Figure 4.16 shows what a powerful
tool it can be. The best source of information on memory management
is Application Note 168, LabVIEW Performance and Memory Manage-
ment, found online at www.ni.com.

Clusters
You can gather several different data types into a single, more manage-
able unit, called a cluster. It is conceptually the same as a record in
Pascal or a struct in C. Clusters are normally used to group related data
elements that are used in multiple places on a diagram. This reduces
wiring clutter—many items are carried along in a single wire. Clusters
also reduce the number of terminals required on a subVI. When saved
as a custom control with the typedef or strict typedef option (use the
Customize Control feature, formerly known as the Control Editor,
to create typedefs), clusters serve as data type definitions, which can
simplify large LabVIEW applications. Saving your cluster as a typedef
propagates any changes in the data structure to any code using the
typedef cluster. Using clusters is good programming practice, but it does
require a little insight as to when and where clusters are best employed.
If you’re a novice programmer, look at the LabVIEW examples and the
figures in this book to see how clusters are used in real life.
An important fact about a cluster is that it can contain only con-
trols or indicators, but not a mixture of both. This precludes the use
of a cluster to group a set of controls and indicators on a panel. Use
graphical elements from the Decorations palette to group controls and
indicators. If you really need to read and write values in a cluster, local
variables can certainly do the job. We would not recommend using local
variables to continuously read and write a cluster because the chance
for a race condition is very high. It’s much safer to use a local variable
to initialize the cluster (just once), or perhaps to correct an errant input
or reflect a change of mode. Rule: For highly interactive panels, don’t
use a cluster as an input and output element.
Clusters are assembled on the diagram by using either the Bundle
function (Figure 4.18) or the Bundle By Name function (Figure 4.19).
The data types that you connect to these functions must match the data
types in the destination cluster (numeric types are polymorphic; for
instance, you can safely connect an integer type to a floating-point type).
The Bundle function has one further restriction: The elements must be
connected in the proper order. There is a pop-up menu available on the
cluster border called Reorder Controls in Cluster. . . that you use to
set the ordering of elements. You must carefully watch cluster ordering.
Two otherwise identical clusters with different element orderings can’t
 LabVIEW Data Types 123

Figure 4.18 Using the Bundle and Unbundle functions on clusters.

be connected. An exception to this rule, one which causes bugs that are
difficult to trace, occurs when the misordered elements are of similar
data type (for instance, all are numeric). You can legally connect the
misordered clusters, but element A of one cluster may actually be
passed to element B of the other. This blunder is far too common and is
one of the reasons for using Bundle By Name.
To disassemble a cluster, you can use the Unbundle or the Unbun-
dle By Name function. When you create a cluster control, give each
element a reasonably short name. Then, when you use Bundle By Name
or Unbundle By Name, the name doesn’t take up too much space on the
diagram. There is a pop-up menu on each of these functions (Select
Item) with which you select the items to access. Named access has the

Figure 4.19 Use Bundle By Name and Unbundle By Name in preference to their no-name counterparts.
124 Chapter Four

additional advantage that adding an item to the related cluster con-

trol or indicator doesn’t break any wires as it does with the unnamed
method. Rule: Always use named cluster access except when there is
some compelling reason not to.
When you use Bundle By Name, its middle terminal must be wired.
The functions of the middle terminal on Bundle By Name are (1) to
determine the element ordering and the data types and (2) to set the
item names. Even if you have wired all input elements, you must wire
the middle terminal because the input elements determine only the
data types. The Bundle function does not have this limitation; you need
only wire to its middle terminal when you wish to access a limited set
of a cluster’s elements.
Clusters are often incorporated into arrays (we call them cluster
arrays), as shown in Figure 4.20. It’s a convenient way to package
large collections of data such as I/O configurations where you have
many different pieces of data to describe a channel in a cluster and
many channels (clusters) in an array. Cluster arrays are also used
to define many of the graph and chart types in LabVIEW, a subject
discussed in detail in Chapter 20, “Data Visualization, Imaging, and
Sound.” Figure 4.20 also shows the LabVIEW equivalent of an array
of arrays, which is implemented as an array of clusters of arrays.
Note the difference between this construct and a simple 2D array, or
matrix. Use these arrays of cluster arrays when you want to combine
arrays with different sizes; multidimensional LabVIEW arrays are
always rectangular.

Figure 4.20 Building a cluster array (top) and building an array of clusters that contain an array
(bottom). This is distinctly different from a 2D array.
 LabVIEW Data Types 125

Waveforms
A handy data type, introduced in LabVIEW 6, is the Waveform. It’s a
sensible grouping of information that describes the very common situ-
ation of a one-dimensional time-varying waveform. The waveform type
is similar to a cluster containing a 1D numeric array (the data), tim-
ing information (start time and sample interval), and a variant part
containing user-definable items such as the channel’s name, units, and
error information. Native waveform controls and indicators are avail-
able from the I/O control palette, but other indicators, such as graphs,
will also adapt directly to waveforms. In the Programming palette,
there’s a subpalette called Waveform that includes a variety of basic
operations, such as building, scaling, mathematical operations, and so
forth, plus higher-level functions, such as waveform generation, mea-
surements, and file I/O.
Figure 4.21 shows some of the basic things you might do with a
waveform. Waveforms can come from many sources, among them the
Waveform Generation VIs, such as the Sine Waveform VI that
appears in the figure. The analog data acquisition (DAQ) VIs also
handle waveforms directly. Given a waveform, you can either process it
as a unit or disassemble it for other purposes. In this example, we see
one of the Waveform Measurement utilities, the Basic Averaged
DCRMS VI, which extracts the dc and rms ac values from a signal. You
can explicitly access waveform components with the Get Waveform
Components function, which looks a lot like Unbundle By Name.

Figure 4.21 Here are the basics of the waveform data type. You can access individual components
or use the waveform utility VIs to generate, analyze, or manipulate waveforms.
126 Chapter Four

To assemble a waveform data type from individual components, use

the Build Waveform function. Not all inputs need be supplied; quite
often, the starting time (t0) can be left unwired, in which case it will
be set to zero.
Thanks to polymorphism, waveforms can be wired to many ordinary
functions, but there are some limits and special behavior, as shown in
Figure 4.22. To add a constant offset to each data value in a waveform,
you can use the regular Add node. Another common operation is to sum
the data for two waveforms (for instance, to add uniform noise to a syn-
thetic waveform). In this case, the Add node doesn’t do what you expect.
The result of adding a waveform and a numeric array is counterintui-
tive: You end up with an array of waveforms with the same number of
waveforms as there are elements in the numeric array. Instead, you
must use the Build Waveform function to promote your numeric array
to the waveform type, and then you use the Add node. Another method
is to unbundle the Y array by using Get Waveform Components, add the
two arrays with the Add function, and reassemble with Build Waveform.
LabVIEW 8 correctly handles the situation where you use the Add node
to directly add two waveforms. Both waveforms must have the same dt
values. The resulting waveform retains the t0 and attributes of the
waveform connected to the upper terminal of the Add node.
So what about these waveform attributes? That component is pri-
marily for use with the DAQ functions when you have defined channel
names and units through the DAQ configuration utilities. In that case,
the channel information is passed along with acquired analog data. But
you can also insert your own information into this variant data field,
and read it back as well. Variant data types were originally added to
LabVIEW to handle the complicated data that is sometimes required
by ActiveX objects. Like a cluster, a variant is an arbitrary collection of
items, but a variant lets you change the data type (add or delete items)
while a VI is running.

Figure 4.22 Polymorphism permits standard operations on waveforms.

 LabVIEW Data Types 127

Figure 4.23 The stealth component of waveforms, called attributes, can be accessed when you need them. They
are variant data types, which require some care during access.

In Figure 4.23, we use Set Waveform Attribute to create a new

attribute called My Attribute that is a cluster with a number and a
string inside. This information rides along with that waveform until
it is either modified or deleted. If you use one of the waveform file I/O
utilities, all attributed data will be stored in the file for you. You can
also display it in a waveform indicator, as we have in this figure. Pop
up on the waveform indicator, and under Visible Items you can choose
Attributes. To read the attribute information back into your program,
use the Get Waveform Attribute function. If you supply the name of
an existing attribute, the function returns exactly what you asked for.
If you don’t supply a name, it returns all available attributes. In either
case, the information comes out in the form of a variant. To deal with
that, use the Variant To Data function from the Cluster and Variant
palette. This little fellow converts arbitrary variant data to whatever
type you wish. In this case, the type is set to a cluster of the right
kind, and the information you put in comes back out as desired. What if
you give it the wrong data type? Conversion fails, and Variant To Data
returns an error cluster that tells you what happened.

Data Type Conversions

Thanks to polymorphism, you can usually wire from source to destina-
tion without much worry since the functions adapt to the kind of data
that you supply. For instance, in the top part of Figure 4.24, a constant is
added to each value in an array. That’s a wonderful time saver compared
with having to create a loop to iterate through each element of the array.
There are, of course, limits to polymorphism, as the bottom example in
Figure 4.24 shows. The result of adding a boolean to a string is a little
128 Chapter Four

Figure 4.24 Polymorphism in action. Think how complicated this would be if the functions didn’t adapt
automatically. The bottom example shows the limits of polymorphism. How in the world could you add a
boolean or a string? If there were a way, it would be included in LabVIEW.

hard to define, so the natural polymorphism in LabVIEW doesn’t per-

mit these operations. But what if you actually needed to perform such
an operation? That’s where conversions and type casting come in.

Conversion and coercion

Data in LabVIEW has two components, the data and type descriptor.
You can’t see the type descriptor; it is used internally to give LabVIEW
directions on how to handle the associated data—that’s how polymor-
phic functions know what kind of data is connected. A type descriptor
identifies the type of the data (such as a DBL floating-point array) and
the number of bytes in the data. When data is converted from one type
to another, both the data component and the type descriptor are modi-
fied in some fashion. For example, , where an I32
(signed integer with 32 bits, or 4 bytes) is converted to a DBL (64-bit, or
8-byte, floating point), the value contained in the I32 is changed into a
mantissa and an exponent. The type descriptor is changed accordingly,
and this new data type takes up a bit more memory. For conversion
between scalar numeric types, the process is very simple, generally
requiring only one CPU instruction. But if aggregate types are involved
(strings, clusters, etc.), this conversion process takes some time.
 LabVIEW Data Types 129

First, the value has to be interpreted in some way, requiring that a spe-
cial conversion program be run. Second, the new data type may require
more or less memory, so the system’s memory manager may need to be
called. By now you should be getting the idea that conversion is some-
thing you may want to avoid, if only for performance reasons.
Conversion is explicitly performed by using one of the functions from
the Conversion menu. They are polymorphic, so you can feed them sca-
lars (simple numbers or booleans), arrays, clusters, and so on, as long
as the input makes some sense. There is another place that conversions
occur, sometimes without you being aware. When you make a connec-
tion, sometimes a little gray dot appears at the destination’s terminal:
. This is called a coercion dot, and it performs exactly the
same operation as an explicit conversion function. One other warning
about conversion and coercion: Be wary of lost precision.
A DBL or an EXT floating point can take on values up to 10−237 or
thereabouts.
If you converted such a big number to an unsigned byte (U8), with
a range of only 0 to 255, then clearly the original value could be lost.
It is generally good practice to modify numeric data types to eliminate
coercion because it reduces memory usage and increases speed. Use
the Representation pop-up menu item on controls, indicators, and
diagram constants to adjust the representation.

Intricate conversions and type casting

Besides simple numeric type conversions, there are some more advanced
ones that you might use in special situations. This is an advanced topic,
so we’ve saved it for last.
Figure 4.25 uses an intertype conversion to make a cluster of booleans
into an array. Cluster To Array works on any cluster that contains

Figure 4.25 A cluster of booleans makes a nice user-interface item, but it’s hard to interpret. Here are two
solutions that find the true bit. The bottom solution turned out to be the faster.
130 Chapter Four

only controls of the same type (you can’t arbitrarily mix strings, numer-
ics, etc.). Once you have an array of booleans, the figure shows two ways
to find the element that is True. In the upper solution, the Boolean
Array To Number function returns a value in the range of 0 to 232 – 1
based on the bit pattern in the boolean array. Since the bit that is set
must correspond to a power of 2, you then take the log2 of that number,
which returns a number between 0 and 32 for each button and −1 for
no buttons pressed. Pretty crafty, eh? But the bottom solution turns out
to be faster. Starting with the boolean array, use Search 1D Array to
find the first element that is true. Search 1D Array returns the element
number, which is again a number between 0 and 32, or a large negative
number for no buttons.
This number could then be passed to the selection terminal in a Case
structure to take some action according to which switch was pressed.
Figure 4.26 shows a way to use a nice-looking set of Ring Indica-
tors in a cluster as a status indicator. An array of I32 numerics is con-
verted to a cluster by using the Array To Cluster function. A funny
thing happens with this function. How does LabVIEW know how many
elements belong in the output cluster (the array can have any number
of elements)? For this reason, a pop-up item on Array To Cluster called
Cluster Size was added. You have to set the number to match the
indicator (5, in this case) or you’ll get a broken wire.
One of the most powerful ways to change one data type to another is
type casting. As opposed to conversions, type casting changes only the
type descriptor. The data component is unchanged. The data is in no
way rescaled or rearranged; it is merely interpreted in a different way.
The good news is that this process is very fast, although a new copy of

Figure 4.26 A numeric array is converted to a cluster of Ring Indicators (which are of type I32) by using Array
To Cluster. Remember to use the pop-up item on the conversion function called Cluster Size for the number of
cluster elements. In this case, the size is 5.
 LabVIEW Data Types 131

Figure 4.27 Lifted from the HP 54510 driver, this code segment strips the header off
a data string, removes a trailing linefeed character, then type casts the data to a U16
integer array, which is finally converted to an EXT array.

the incoming data has to be made, requiring a call to the memory man-
ager. The bad news is that you have to know what you’re doing! Type
casting is a specialized operation that you will very rarely need, but
if you do, you’ll find it on the Data Manipulation palette. LabVIEW
has enough polymorphism and conversion functions built in that you
rarely need the Type Cast function.
The most common use for the Type Cast function is shown in
Figure 4.27, where a binary data string returned from an oscilloscope
is type cast to an array of integers. Notice that some header informa-
tion and a trailing character had to be removed from the string before
casting. Failure to do so would leave extra garbage values in the resul-
tant array. Worse yet, what would happen if the incoming string were
off by 1 byte at the beginning? Byte pairs, used to make up I16 integers,
would then be incorrectly paired. Results would be very strange. Note
that the Type Cast function will accept most data types except for clus-
ters that contain arrays or strings.
Warning: The Type Cast function expects a certain byte ordering,
namely big-endian, or most significant byte first to guarantee porta-
bility between platforms running LabVIEW. But there are problems
interpreting this data outside of LabVIEW. Big-endian is the normal
ordering for the Macintosh and Sun, but not so on the PC! This is an
example where your code, or the data you save in a binary file, may not
be machine-independent.
Indeed, there is much trouble in type casting land, and you should try
to use polymorphism and conversion functions whenever possible. Con-
sider Figure 4.28. The first example uses the Type Cast function, while
the second example accomplishes exactly the same operation—writing
a binary image of an array to disk—in a much clearer, more concise
way. Keep looking through the function palettes if the particular data
type compatibility that you need is not apparent. The bottom example
in Figure 4.28 shows how flexible the LabVIEW primitives are. In this
case, the Write File function accommodates any imaginable data type
without putting you through type casting obfuscation.
132 Chapter Four

Figure 4.28“Eschew obfuscation,” the English teacher said. Look for ways to avoid type casting and simplify
your diagrams. Higher-level functions and primitives are extremely flexible.

If you get into serious data conversion and type casting exercises, be
careful, be patient, and prepare to explore the other data conversion
functions in the Data Manipulation palette. Sometimes the bytes are
out of order, as in the PC versus Mac situation. In that case, try Swap
Bytes or Swap Words, and display the data in a numeric indicator
with its format set to hexadecimal, octal, or binary (use the pop-up
menu Format and Precision). Some instruments go so far as to send
the data backward; that is, the first value arrives last. You can use
Reverse Array or Reverse String to cure that nasty situation. Split
Number and Join Number are two other functions that allow you to
directly manipulate the ordering of bytes in a machine-independent
manner. Such are the adventures of writing instrument drivers, a topic
covered in detail in Chapter 10, “Instrument Driver Basics.”

Flatten To String (. . . Do what?)

We mentioned before that the Type Cast function can’t handle certain
complicated data types. That’s because LabVIEW stores strings and
arrays in handle blocks, which are discontiguous segments of memory
organized in a tree structure. When such data types are placed in a
cluster, the data may be physically stored in many areas of memory.
Type Cast expects all the data to be located in a single contiguous area
so that it can perform their simple and fast transformation. You can
read all about LabVIEW data storage details online in Application
Note AN154, LabVIEW Data Storage.
Occasionally you need to transmit arbitrarily complex data types
over a serial link. The link may be a serial port, a network connection
using a protocol such as TCP/IP, or even a binary file on disk, which
is, in fact, a serial storage method. Type Cast can’t do it, but the Flat-
ten To String function can. What this function does is copy all dis-
contiguous data into one contiguous buffer called the data string.
 LabVIEW Data Types 133

Figure 4.29 Use flattened data when you need to transmit complicated data types
over a communications link or store them in a binary file. LabVIEW 8 lets you set
byte order and optionally prepend the data string’s size.

The data string also contains embedded header information for nons-
calar items (strings and arrays), which are useful when you are trying
to reconstruct flattened data. Figure 4.29 shows Flatten To String in
action, along with its counterpart Unflatten From String. The data
string could be transmitted over a network, or stored in a file, and then
read and reconstructed by Unflatten From String. As with all binary
formats, you must describe the underlying data format to the read-
ing program. Unflatten From String requires a data type to properly
reconstruct the original data. In LabVIEW 7.x and previous versions,
the data type descriptor was available as an I16 array. LabVIEW 8’s
data type descriptors are 32-bit, and the type descriptor terminal is no
longer available on Flatten to String. If you still need the type descrip-
tor, select Convert 7.x Data from Flatten to String’s pop-up menu.
The most common uses for flattened data are transmission over a
network or storage to a binary file. A utility VI could use this technique
to store and retrieve clusters on disk as a means of maintaining front-
panel setup information.

Enumerated types (enums)

A special case of an integer numeric is the enum, or enumerated type.
It’s a notion borrowed from C and Pascal where each consecutive value
of an integer, starting with zero, is assigned a name. For instance, red,
green, and blue could correspond to 0, 1, and 2. You create an enum by
selecting an enum control from the Ring and Enum control palette,
and then typing in the desired strings. The handiest use for an enum
is as a selector in a Case structure because the name shows up in
the header of the Case, thus making your program self-documenting
(Figure 4.30). Enums can be compared with other enums, but the com-
parisons are actually based on the integer values behind the scenes,
not the strings that you see. So you don’t need to do any special con-
version to compare two dissimilar enums or to compare an enum with
an integer.
134 Chapter Four

Figure 4.30Enumerated types (enums) are integers with associated text. Here are some useful tips for
comparison and conversion.

Figure 4.30 shows some precautions and tricks for enums. One sur-
prise comes when you attempt to do math with enums. They’re just
integers, right? Yes, but their range is limited to the number of enu-
merated items. In this example, there are three values {Red, Grn, Blu},
so when we attempt to add one to the maximum value (Blu, or 2), the
answer is Red—the integer has rolled over to 0.
You can also do some nonobvious conversions with enums, as shown
in the bottom of Figure 4.30. Given an enum, you can extract the string
value by using Format Into String with a format specifier of %s.
Similarly, you can match an incoming string value with the possible
string equivalents in an enum by using Scan From String. There is
a risk of error here, however: The incoming string has to be an exact
case-sensitive match for one of the possible enum values. Otherwise
Scan From String will return an error, and your output enum will be
set to the value of the input enum. To distinguish this result from a
valid conversion, you could add a value to your enum called Invalid
and handle that value appropriately.

Get Carried Away Department

Here’s a grand finale for conversion and type casting. Say that you
need to fix the length of a string at three characters and add on a null
(zero) terminator. Gary actually had to do this for compatibility with
 LabVIEW Data Types 135

Figure 4.31Making a fixed-length, null-terminated string for compatibility with the C language. The key is the
Array To Cluster conversion that produces a fixed number of elements. This is a really obscure example.

another application that expected a C-style string, which requires the

addition of a null terminator. Let’s use everything we know about the
conversion functions and do it as in Figure 4.31.
This trick (courtesy of Rob Dye, a member of the LabVIEW develop-
ment team) uses the Set Cluster Size pop-up setting in an Array To
Cluster conversion function to force the string length to be four char-
acters, total. The bundler allows parallel access to the whole string, so
it’s easy to change one character, in this case the last one. It turns out
that this program is very fast because all the conversions performed
here are actually type casting operations underneath. The disadvan-
tage of this technique, as compared to using a few string functions, is
that it is hard to understand.
This page intentionally left blank
 Chapter

5
Timing

Software-based measurement and control systems have requirements

for timing and synchronization that are distinctly different from those
of ordinary applications. If you’re writing an application that serves
only a human operator (such as a word processor), chances are that
your only timing requirement is to be fast enough that the user doesn’t
have to sit and wait for the computer to catch up. Exactly when an event
occurs is not so important. But in a control system, there’s a physical
process that demands regular service to keep those boiling kettles from
overflowing. Similarly, a data acquisition system needs to make mea-
surements at highly regulated intervals, lest the analysis algorithms
get confused. That’s why LabVIEW has built-in timing functions to
measure and control time. Depending on your needs, LabVIEW’s tim-
ing functions can be simple and effective or totally inadequate. And the
problems are not all the fault of LabVIEW itself; there are fundamental
limitations on all general-purpose computer systems with regard to
real-time response, whatever that means (1 second? 0.01 second? 1 ns?).
Most applications we’ve seen work comfortably with the available
LabVIEW time measurements that resolve milliseconds, and many
more operate with 1-s resolution. A few applications demand submil-
lisecond resolution and response time, which is problematic owing
primarily to operating system issues. Those cases require special atten-
tion. So, what’s with all this timing stuff, anyway?

Where Do Little Timers Come From?

When LabVIEW was born on the Macintosh, the only timer available
was based on the 60-Hz line frequency. Interrupts to the CPU occurred
every 1/60 s and were counted as long as the Mac was powered on.

137
138 Chapter Five

The system timer on modern computers is based on a crystal-controlled

oscillator on the computer’s motherboard. Special registers count each
tick of the oscillator. The operating system abstracts the tick count for
us as milliseconds and microseconds. It’s funny how we instrumentation
specialists benefited from the consumer-oriented multimedia timing.
Without the need for high resolution time for audio and video, we may
well have been stuck at 60 Hz. One might argue that all those ultrafast
video accelerators and even the speed of CPUs are driven primarily by
the enormous video game market!
Three fine-resolution timers are available within LabVIEW on
all platforms: Wait (ms), Wait Until Next ms Multiple, and Tick
Count (ms). These functions attempt to resolve milliseconds within
the limitations of the operating systems. Two new timing functions,
Timed Loops and Timed Sequence Structure, are deterministic
programming structures similar to While Loops and Sequence struc-
tures. The determinism and resolution of these two structures are
operating system- and hardware platform-dependent. On a PC run-
ning Windows, the Timed Loop has a resolution of 1 ms; but under
a real-time operating system (RTOS) resolution can be 1 µs. Timed
Loops also have other exciting features such as the ability to synchro-
nize to a data acquisition task.
Another kind of timer that is available in LabVIEW is the system
clock/calendar which is maintained by a battery-backed crystal oscil-
lator and counter chip. Most computers have a clock/calendar timer
of this kind. The timing functions that get the system time are Get
Date/Time In Seconds, Get Date/Time String, and Seconds To
Date/Time. These functions are as accurate as your system clock. You
can verify this by observing the drift in the clock/calendar displayed
on your screen. Unless you synchronize to a network time source, it’s
probably within a minute per month or so.
Are you curious about the timing resolution of your system?
Then write a simple timer test VI like the one shown in Figure 5.1.
A For Loop runs at maximum speed, calling the timing function
under test. The initial time is subtracted from the current time,
and each value is appended to an array for plotting. The system in
Figure 5.1 has a resolution of 1 ms and took approximately 1 µs per
iteration.

Using the Built-in Timing Functions

There are two things you probably want to do with timers. First, you
want to make things happen at regular intervals. Second, you want to
record when events occur.
 Timing 139

Figure 5.1 Write a simple timer test VI to check the resolution of your system’s LabVIEW ticker.

Intervals
Traditionally if you want a loop to run at a nice, regular interval, the
function to use is Wait Until Next ms Multiple. Just place it inside the
loop structure, and wire it to a number that’s scaled in milliseconds. It
waits until the tick count in milliseconds becomes an exact multiple
of the value that you supply. If several VIs need to be synchronized,
this function will help there as well. For instance, two independent
VIs can be forced to run with harmonically related periods such as
100 ms and 200 ms, as shown in Figure 5.2. In this case, every 200 ms,
you would find that both VIs are in sync. The effect is exactly like the
action of a metronome and a group of musicians—it’s their heartbeat.
This is not the case if you use the simpler Wait (ms) function; it just
guarantees that a certain amount of time has passed, without regard
to absolute time. Note that both of these timers work by adding
activity to the loop. That is, the loop can’t go on to the next cycle
until everything in the loop has finished, and that includes the timer.

Figure 5.2Wait Until Next ms Multiple (the two left loops)

keeps things in sync. The right loop runs every 100 ms, but
may be out of phase with the others.
140 Chapter Five

Since LabVIEW executes everything inside the loop in parallel, the

timer is presumably started at the same time as everything else. Note
that “everything else” must be completed in less time than the desired
interval. The timer does not have a magic ability to speed up or abort
other functions. Wait Till Next ms Multiple will add on extra delay as
necessary to make the tick count come out even, but not if the other
tasks overrun the desired interval.
Both of these loop timers are asynchronous; that is, they do not
tie up the entire machine while waiting for time to pass. Any time a
call to Wait (ms) is encountered, LabVIEW puts the calling VI into a
scheduler queue until the timer expires. If you want your loop to run
as fast as possible but not hog all the CPU, place a Wait (ms) with
0 ms to wait. The 0-ms wait is a special instruction to the LabVIEW
scheduler to let other waiting tasks run, but to put yours back in the
schedule as soon as possible. This guarantees that other VIs have a
chance to run, but still executes your loop as fast as possible. Rule:
Always put a timer in every loop structure. Otherwise, the loop may
lock out all other VIs, and even interfere with the responsiveness of
the LabVIEW user interface. The exception to this rule occurs when
you know there is some other activity in the loop that will periodi-
cally cause LabVIEW to put that part of the diagram in the scheduler
queue. For instance, file I/O and many DAQ and GPIB operations
have hidden delays or other scheduling behavior that meets this
requirement.

Timed structures
There are three programming structures in LabVIEW 8 with built-in
scheduling behavior: a Timed Loop, a Timed Sequence, and a com-
bination of the two called a Timed Loop with Frames. Timed Loops
are most useful when you have multiple parallel loops you need to exe-
cute in a deterministic order. LabVIEW’s default method of executing
code is multiple threads of time-sliced multitasking execution engines
with a dash of priorities. In other words, your code will appear to run
in parallel, but it is really sharing CPU time with the rest of your
application (and everything else on your computer). Deterministically
controlling the execution order of one loop over another is an incredibly
difficult task—until now.
Timed Loops are as easy to use as a While Loop and ms Timer combi-
nation. The Timed Loop in Figure 5.3 executes every 1 ms and calculates
the elapsed time. You configure the loop rate by adjusting the period, or
dt. You can also specify an offset (t0) for when the loop should start. With
these two settings you can configure multiple loops to run side by side
with the same period but offset in time from each other. How to configure
 Timing 141

Figure 5.3 Timed Loops have a built-in scheduling mechanism. You can configure a Timed Loop’s behavior dynam-
ically at run time through terminals on the Input node, or statically at edit time through a configuration dialog.

the period and offset is all that most of us need to know about Timed
Loops, but there is much more. You can change everything about a Timed
structure’s execution, except the name and the timing source, on the fly
from within the structure. You can use the execution feedback from the
Left Data Node in a process control loop to tell whether the structure
started on time or finished late in the previous execution. Time-critical
processes can dynamically tune the Timed structure’s execution time
and priority according to how timely it executed last time. The left data
node even provides nanosecond-resolution timing information you can
use to benchmark performance.

Timing sources
Timed structures (loops and sequences) run off a configurable timing
source. Figure 5.4 shows a typical configuration dialog. The default
source is a 1-kHz clock (1-ms timer) on a PC. On an RT system with a
Pentium controller, a 1-MHz clock is available, providing a 1-µs timer.
Other configuration options are the Period and Priority. In our dialog
the loop is set to execute every 1000 ms by using the PC’s internal
1-kHz clock. Note that you can also use a timing source connected to
the structure’s terminal. The timing source can be based on your com-
puter’s internal timer or linked through DAQmx to a hardware-timed
DAQ event with DAQmx Create Timing Source.vi (Figure 5.5). You
can even chain multiple timing sources together by using Build Tim-
ing Source Hierarchy.vi. One example in which multiple timing
sources could come in handy is if you wanted a loop to execute based on
time or every time a trigger occurred.
142 Chapter Five

Figure 5.4 Configuration dialog for a Timed Loop with a single frame.

Figure 5.5 Timed structures can be synchronized to external timing sources with DAQmx.
 Timing 143

It’s important to note that once a timing source is started, it will

continue to run even if the timed loop is not running! This can cause
unexpected consequences when your Timed Loop starts running as
fast as possible to try to catch up to its timing source. To avoid this,
use the “<reset at structure start>” feature. This is important if you
use a Timed Loop inside another loop. The loop needs to either reset
its clock when it starts or be set to discard missed periods. Another
method is to clear the timer after exiting the loop, by using Clear
Timing Source.vi. This removes the timer but adds unnecessary
overhead each time the timer has to be recreated.
Timed Loops with Frames and Timed sequence structures have the
unique capability of running off two timing sources. One source trig-
gers the start of the loop, and a second frame timing source governs
each frame. The example in Figure 5.6 shows a multiframe Timed
Loop with dead time inserted between each frame. Figure 5.7 shows
the configuration dialog.

Execution and priority

Each Timed structure runs in its own thread at a priority level just
beneath Time Critical and above High Priority. Timed Loops do not
inherit the priority or execution system of the VI they run in. This
means that a Timed structure will stop any other code from running
until the structure completes. LabVIEW’s Timed structure scheduler
runs behind the scenes and controls execution based on each Timed
structure’s schedule and individual priority. Priorities are set before
the Timed Loop executes, through either the configuration dialog or a
terminal on the Input node. Priority can be adjusted inside the Timed

Figure 5.6 Frames in a multiframe Timed structure can have start times relative to the completion of the previous
frame. The effect is to put dead time between each frame. Frame 2 starts 200 ms after frame 1 completes, frame 3
starts 500 ms after frame 2, and frame 4 starts 200 ms after frame 3.
144 Chapter Five

Figure 5.7 Multiframe structures can have two timing sources.

Loop through the Right Data node. The Timed Loop’s priority is a pos-
itive integer between 1 and 2,147,480,000. But don’t get too excited
since the maximum number of Timed structures you can have in mem-
ory, running or not, is 128.
Timed structures at the same priority do not multitask between
each other. The scheduler is preemptive, but not multitasking. If two
structures have the same priority, then dataflow determines which
one executes first. Whichever structure starts execution first will fin-
ish before the other structure can start. Each Timed Loop runs to
completion unless preempted by a higher-priority Timed Loop, a VI
running at Time Critical priority, or the operating system. It is this
last condition that kills any hope of deterministic operation on a desk-
top operating system.
On a real-time operating system, Timed structures are deterministic,
meaning the timing can be predicted. This is crucial on an RTOS; but
on a non-real-time operating system (such as Windows), Timed struc-
tures are not deterministic. This doesn’t mean that you can’t use Timed
structures on Windows; just be aware that the OS can preempt your
structure at any time to let a lower-priority task run. A real-time oper-
ating system runs code based on priority (higher-priority threads pre-
empt lower-priority threads) while a desktop OS executes code based
on “fairness”; each thread gets an opportunity to run. This is great
 Timing 145

when you want to check e-mail in the background on your desktop, but
terrible for a process controller. If you need deterministic execution,
then use Timed structures on real-time systems such as LabVIEW RT.
They are powerful and easy to use.

Timing guidelines
All LabVIEW’s platforms have built-in multithreading and preemp-
tive multitasking that help to mitigate the effects of errant code con-
suming all the CPU cycles. Even if you forget to put in a timer and
start running a tight little loop, every desktop operating system will
eventually decide to put LabVIEW to sleep and to let other programs
run for a while. But rest assured that the overall responsiveness of
your computer will suffer, so please put timers in all your loops. For
more information on how the LabVIEW execution system works, see
Application Note 114, Using LabVIEW to Create Multithreaded VIs for
Maximum Performance and Reliability.
Here’s when to use each of the timers:
■ Highly regular loop timing—use Wait Until Next ms Multiple or a
Timed Loop.
■ Many parallel loops with regular timing—use Wait Until Next ms
Multiple or Timed Loops.
■ Arbitrary, asynchronous time delay to give other tasks some time to
execute—use Wait (ms).
■ Single-shot delays (as opposed to cyclic operations, such as loops)—
use Wait (ms) or a Timed Sequence.

Absolute timing functions. LabVIEW has a set of timing functions that

deal with absolute time, which generally includes the date in addition
to the time of day. When we are dealing with absolute time, the first
thing we need to do is to establish our calendar’s origin. While every
computer platform uses a slightly different standard origin, the
LabVIEW designers have attempted to give you a common interface.
They set the origin as a time-zone-independent number of seconds that
have elapsed since 12:00 A.M., Friday, January 1, 1904, Coordinated Uni-
versal Time (UTC). For convenience, we call this epoch time. On your
computer, you must configure your time zone (typically through a con-
trol panel setting) to shift the hours to your part of the globe. Internally
LabVIEW calculates time based on UTC, but displays time according
to local time. With LabVIEW 8 you finally have the ability to specify
whether time is UTC or local. Prior to LabVIEW 8, coordinating time
on multiple computers across the Internet was not straightforward.
146 Chapter Five

Now you can schedule an event on multiple computers to UTC without

worrying about time zone offset or daylight savings time.
Epoch time values are available in two formats: as a timestamp and
as a cluster known as a date time rec. Timestamp is a high-resolution
128-bit data type capable of handling dates between 1600 A.D. and
3000 A.D. The upper quad int (64 bits) is the number of seconds since
January 1, 1904, and the lower 64 bits is the fractional second. The
date/time rec cluster contains nine I32 values (second, minute, hour,
day, etc.) and one DBL (fractional second). The seconds value is just
what we’ve discussed: the number of seconds since the beginning of
1904. The date/time rec cluster is useful for extracting individual time
and date values, such as finding the current month.
Here is a quick introduction to the absolute time functions and
their uses:
■ Get Date/Time In Seconds returns a timestamp, which can be
converted into the epoch seconds as a DBL float.
■ Get Date/Time String returns two separate formatted strings
containing time and date. It’s nice for screen displays or printing in
reports and has several formatting options.
■ Format Date/Time String returns a single formatted string con-
taining time and date. You supply a formatting string that allows
you to extract any or all possible components of date and time. For
instance, the format string %H:%M:%S%3u asks for hours, min-
utes, and fractional seconds with three decimal digits with colons
between the values. A typical output string would look like this:
15:38:17.736
■ Seconds To Date/Time returns a date/time rec cluster. You might
use this function when you want to do computations based on a par-
ticular aspect of date or time.
■ Date/Time To Seconds converts the date/time rec back to sec-
onds. It’s useful when you want to combine individual time and
date items, such as letting the user sift through data by the day
of the week.

Figure 5.8 shows how many of these functions can be connected.

Sending timing data to other applications

Many graphing and spreadsheet programs do a lousy job of importing
timestamps of the form 23-Jun-1990 10:03:17. Epoch seconds are OK,
but then the other program has to know how to compute the date and
time, which requires a big algorithm. Most spreadsheet programs can
 Timing 147

Figure 5.8 How to switch between the various formats for system time, particularly getting in and out of the
date/time rec cluster.

handle epoch seconds if you give them a little help. In Microsoft Excel,
for instance, you must divide the epoch seconds value by 86,400, which
is the number of seconds in 1 day. The result is what Excel calls a
Serial Number, where the integer part is the number of days since the
zero year and the fractional part is a fraction of 1 day. Then you just
format the number as date and time. Again, watch out for the numeric
precision problem if you’re importing epoch seconds.

High-resolution and high-accuracy timing

If your application requires higher accuracy or resolution than the built-
in software timing functions can supply, then you will have to use a hard-
ware solution. The first thing to do is to consider your specifications.
Some experiments simply need a long-term stable time base, over
scales of days or even years, to synchronize with other events on other
systems. Or you might be concerned with measuring time intervals
in the subsecond range where short-term accuracy is most important.
These specifications will influence your choice of timing hardware.
148 Chapter Five

For long-term stability, you might get away with setting your com-
puter’s clock with one of the network time references; such options
are built into all modern operating systems. They guarantee NIST-
traceable long-term stability, but there is, of course, unknown latency
in the networking that creates uncertainty at the moment your clock
is updated.
Excellent time accuracy can be obtained by locking to signals received
from the Global Positioning System (GPS) satellites, which provide a
1-Hz clock with an uncertainty of less than 100 ns and essentially no
long-term drift. The key to using GPS timing sources is to obtain a
suitable interface to your computer and perhaps a LabVIEW driver.
One GPS board we like is the TSAT cPCI from KSI. It has a very com-
plete LabVIEW driver with some extra features such as the ability to
programmatically toggle a TTL line when the GPS time matches an
internal register. You can use this programmable Match output to hard-
ware-gate—a data acquisition task. Keep in mind the ultimate limita-
tion for precision timing applications: software latency. It takes a finite
amount of time to call the program that fetches the time measurement
or that triggers the timing hardware. If you can set up the hardware in
such a way that there is no software “in the loop,” very high precision
is feasible. We placed TSAT cPCI cards into multiple remote PXI data
acquisition systems and were able to synchronize data collection with
close to 100 ns of accuracy. This was essential for our application, but a
GPS board is probably overkill for most applications.
Perhaps the easiest way to gain timing resolution is to use a National
Instruments data acquisition (DAQ) board as a timekeeper. All DAQ
boards include a reasonably stable crystal oscillator time base with
microsecond (or better) resolution that is used to time A/D and D/A
conversions and to drive various countertimers. Through the DAQ VI
library, you can use these timers to regulate the cycle time of software
loops, or you can use them as high-resolution clocks for timing short-
term events. This DAQ solution is particularly good with the onboard
counter/timers for applications such as frequency counting and time
interval measurement.
But what if the crystal clock on your DAQ board is not good enough?
Looking in the National Instruments catalog at the specifications
for most of the DAQ boards, we find that the absolute accuracy is
± 0.01 percent, or ± 100 ppm, with no mention of the temperature
coefficient. Is that sufficient for your application? For high-quality
calibration work and precision measurements, probably not. It might
even be a problem for frequency determination using spectral estimation.
In the LabVIEW analysis library, there’s a VI called Extract Single Tone
Information that locates, with very high precision, the fundamental
frequency of a sampled waveform. If the noise level is low, this VI can
 Timing 149

return a frequency estimate that is accurate to better than 1 ppm—but

only if the sample clock is that accurate.
National Instruments does have higher-accuracy boards. In the coun-
ter/timer line, the 6608 counter/timer module has an oven-controlled
crystal oscillator (OCXO) that can be adjusted to match an external
reference, and it has a temperature coefficient of 1 × 10−11/ °C and aging
of 1 × 10−7 per year. With all the National Instruments clock generators,
you typically connect them to your DAQ board via real-time system
integration (RTSI) bus signals or via programmable function input
(PFI) lines. If you’re using an external clock generator, the PFI inputs
are the only way. There are several DAQ example VIs that show you
how to use external clocking for analog and digital I/O operations.
A bargain in the world of stable time bases is the Stanford Research
PRS10 rubidium frequency standard. For about US$1500, you get a
10-MHz oscillator that fits in the palm of your hand with a short-term
stability of 1 × 10−11 over 1 s, aging of 5 × 10−10 per year, and a tempera-
ture coefficient of 5 × 10−11/ °C. It also has the lowest phase noise of any
oscillator at any price, which is important if you are phase lock-multi-
plying the time base to obtain microwave frequencies. It can be locked
to a 1-Hz GPS clock to enhance its long-term stability.
If your facility is set up for it, you can also read the time from time-
base standards such as IRIG (Inter-Range Instrumentation Group)
that use a serial interface with up to 100-ns resolution. Interface
boards for IRIG (and GPS) are available from Bancomm (a division of
Dataum Corporation) and from TrueTime. Both manufacturers offer
plug-in boards that keep your computer’s clock/calendar synchronized,
as well as providing very accurate digital timing outputs. Some
LabVIEW drivers are available, or you can use the shared libraries
(DLLs) or C-callable routines to make your own.

Bibliography
Application Note 200, Using the Timed Loop to Write Multirate Applications in LabVIEW,
www.ni.com, National Instruments Corporation, 11500 N. Mopac Expressway, Austin,
Tex., 2004.
This page intentionally left blank
 Chapter

Synchronization
6
When you build a LabVIEW application, you’ll eventually find the need
to go beyond simple sequencing and looping to handle all the things
that are going on in the world outside your computer. A big step in
programming complexity appears when you start handling events. An
event is usually defined as something external to your program that
says, “Hey, I need service!” and demands that service in a timely fash-
ion. A simple example is a user clicking a button on the panel to start
or stop an activity. Other kinds of events come from hardware, such as
a GPIB service request (SRQ) or a message coming in from a network
connection. Clearly, these random events don’t fall within the simple
boundaries of sequential programming.
There are lots of ways to handle events in LabVIEW, and many of
them involve creating parallel paths of execution where each path is
responsible for a sequential or nonsequential (event-driven) activity.
Since G is intrinsically parallel, you can create an unlimited num-
ber of parallel loops or VIs and have confidence that they’ll run with
some kind of timesharing. But how do you handle those “simultaneous”
events in such a way as to avoid collisions and misordering of execution
when it matters? And what about passing information between loops
or VIs? The complexity and the list of questions just seem to grow and
grow.
LabVIEW has some powerful synchronization features, and there are
standard techniques that can help you deal with the challenges of par-
allel programming and event handling. By the way, this is an advanced
topic, so if some of the material here seems strange, it’s because you
might not need many of these features very often (we’ll start with the

151
152 Chapter Six

easier ones, so even you beginners can keep reading). But once you see
how these features work, you’ll soon become an armed and dangerous
LabVIEW dude.*

Polling
Let’s start off with a straightforward event-handling technique that
does not use any esoteric functions or tricks: polling. When you are
polling, you periodically check the status of something (a flag), look-
ing for a particular value or change of value. Polling is very simple to
program (Figure 6.1), is easy to understand and debug, is very widely
used (even by highly experienced LabVIEW dudes!), and is acceptable
practice for thousands of applications in all programming languages.
The main drawback to polling is that it adds some overhead because
of the repeated testing and looping. (If you do it wrong, it can add lots
of overhead.)
In the left example in Figure 6.1, we’re doing a very simple
computation—the comparison of two values—and then deciding
whether to take action. There’s also a timer in the loop that guaran-
tees that the computation will be done only 10 times per second. Now
that’s pretty low overhead. But what if we left the timer out? Then
LabVIEW would attempt to run the loop as fast as possible—and
that could be millions of times per second—and it might hog the
CPU. Similarly, you might accidentally insert some complex calcula-
tion in the loop where it’s not needed, and each cycle of the polling
loop would then be more costly.

Figure 6.1 A simple polling loop handles a user-interface event. The left example
watches for a change in a numeric control. The right example awaits a True value
from the Boolean button.

*There really is an official LabVIEW Dude. On the registry system at Lawrence Liver-
more National Lab, Gary got tired of getting confused with the other two Gary Johnsons
at the Lab, so he made his official nickname “LabVIEW Dude.” It was peculiar, but effec-
tive. Just like Gary.
 Synchronization 153

The other drawback to polling is that you only check the flag’s state
every so often. If you’re trying to handle many events in rapid suc-
cession, a polling scheme can be overrun and miss some events. An
example might be trying to handle individual bits of data coming in
through a high-speed data port. In such cases, there is often a hard-
ware solution (such as the hardware buffers on your serial port inter-
face) or a lower-level driver solution (such as the serial port handler in
your operating system). With these helpers in the background doing
the fast stuff, your LabVIEW program can often revert to polling for
larger packets of data that don’t come so often.

Events
Wouldn’t it be great if we had a way to handle user-interface events
without polling? We do and it’s called the Event structure. Remem-
ber, an event is a notification that something has happened. The Event
structure bundles handling of user-interface events (or notifications)
into one structure without the need for polling front panel controls, and
with a lot fewer wires! The Event structure sleeps without consuming
CPU cycles until an event occurs, and when it wakes up, it automati-
cally executes the correct Event case. The Event structure doesn’t miss
any events and handles all events in the order in which they happen.
The Event structure looks a lot like a Case structure. You can add,
edit, or delete events through the Event structure’s pop-up menu.
Events are broken down into two types: Notify events and Filter
events. A Notify event lets you know that something happened and
LabVIEW has handled it. Pressing a key on the keyboard can trig-
ger an event. As a notify event, LabVIEW lets you know the key was
pressed so you can use it in your block diagram, but all you can do is
react to the event. Figure 6.2 shows a key down Notify event. A Filter

Figure 6.2 Notify event. Configured to

trigger a notification when any key is
pressed.
154 Chapter Six

Figure 6.3 Filter event. This triggers a

notification when any key is pressed,
but the key press can be modified or
discarded.

event (Figure 6.3) allows you to change the event’s data as it happens
or even discard the event. Filter events have the same name as Notify
events, but end with a question mark (“?”).
The Event structure can save a lot of time and energy, but it can
also drive you crazy if you aren’t alert. The default configuration of
all events is to lock the front panel until the Event case has finished
execution. If you put any time-consuming processing inside the Event
structure, your front panel will lock up until the event completes. Rule:
Handle all but the most basic processing external to the Event structure.
Later in this chapter we’ll show you a powerful design pattern you can
use to pass commands from an event loop to a parallel processing loop
with queues.
Use events to enhance dataflow, not bypass it. This is especially true
with the mechanical action of booleans. A latching boolean does not
reset until the block diagram has read its value. If your latching bool-
ean is outside floating on the block diagram, it will never reset. Be
sure to place any latching booleans inside their Event case so their
mechanical action will be correct. Changing a control’s value through
a local variable will not generate an event – use the property “Value
(Signaling)” to trigger an event.
The advanced topic of Dynamic Events allows you to define and
dynamically register your own unique events and pass them between
subdiagrams. Dynamic events can unnecessarily complicate your
application, so use them only when needed and make sure to docu-
ment what you are doing and why. Figure 6.4 illustrates configuring a
dynamic event to trigger a single Event case for multiple controls. This
eliminates manually configuring an Event case for each control—a
tedious task if you have tens, or even hundreds, of controls. Property
nodes are used to get the control reference to every control on every
 Synchronization 155

Figure 6.4Use control references for dynamic event registration. A single Event case handles multiple Value
Change events.

page of a tab control. The 1D array of references is then used to register

all the controls for a value change event. Whenever any control’s value
is changed, a single Event case will be fired. Inside the Event case we
have a reference to the control and its data value as a variant. Also
note that all the controls are inside the Event case.

Occurrences
An occurrence is a synchronization mechanism that allows paral-
lel parts of a LabVIEW program to notify each other when an event
has occurred. It’s a kind of software trigger. The main reason for using
occurrences is to avoid polling and thus reduce overhead. An occurrence
does not use any CPU cycles while waiting. For general use, you begin
by calling the Generate Occurrence function to create an occurrence
refnum that must be passed to all other occurrence operations.
Then you can either wait for an occurrence to happen by calling the
Wait On Occurrence function or use Set Occurrence to make an
event happen. Any number of Wait On Occurrence nodes can exist
within your LabVIEW environment, and all will be triggered simulta-
neously when the associated Set Occurrence function is called.
Figure 6.5 is a very simple demonstration of occurrences. First, we
generate a new occurrence refnum and pass that to the Wait and Set
functions that we wish to have related to one another. Next, there’s our
old friend the polling loop that waits for the user to click the Do it but-
ton. When that happens, the occurrence is set and, as if by magic, the
event triggers the Wait function, and a friendly dialog box pops up. The
magic here is the transmission of information (an event) through thin
156 Chapter Six

Figure 6.5 A trivial occurrence demon-

stration. The lower loop sets the occur-
rence when the user clicks the Set It
button, causing the Wait On Occurrence
function to trigger the dialog box. Ter-
minating the While Loop without set-
ting the occurrence causes the Wait On
Occurrence function to wait forever.

air from one part of the diagram to another. In fact, the Wait function
didn’t even have to be on the same piece of diagram; it could be out
someplace in another VI. All it has to have is the appropriate occur-
rence refnum.
It appears that, as with global variables, we have a new way to vio-
late dataflow programming, hide relationships among elements in our
programs, and make things occur for no apparent reason in unrelated
parts of our application. Indeed we do, and that’s why you must always
use occurrences with caution and reservation. As soon as the Set and
Wait functions become dislocated, you need to type in some comments
explaining what is going on.
Well, things seem pretty reasonable so far, but there is much more to
this occurrence game. First, there is a ms time-out input on the Wait
On Occurrence function that determines how long it will wait before
it gets bored and gives up. The default time-out value, −1, will cause
it to wait forever until the occurrence is set. If it times out, its output
is set to True. This can be a safety mechanism of a sort, when you’re
not sure if you’ll ever get around to setting the occurrence and you
want a process to proceed after a while. A very useful application for
the time-out is an abortable wait, first introduced by Lynda Gruggett
(LabVIEW Technical Resource, vol. 3, no. 2). Imagine that you have a
While Loop that has to cycle every 100 s, as shown in the left side of
Figure 6.6. If you try to stop the loop by setting the conditional termi-
nal to False, you may have to wait a long time. Instead, try the abort-
able wait (Figure 6.6, right).
To implement the abortable wait, the loop you want to time and abort
is configured with a Wait On Occurrence wired to a time-out value. When-
ever the time-out period has expired, the Wait returns True, and the
While Loop cycles again, only to be suspended once again on the Wait.
 Synchronization 157

An abortable wait (right) can be created from

Figure 6.6
a Wait On Occurrence function by using the time-out
feature.

When the occurrence is set (by the lower loop, in this example), the
Wait is instantly triggered and returns False for the time-out flag, and
its loop immediately stops.
Now for the trickiest feature of the Wait On Occurrence function: the
ignore previous input. It’s basically there to allow you to determine
whether to discard occurrence events from a previous execution of the
VI that it’s in. Actually, it’s more complicated than that, so here is a
painfully detailed explanation in case you desperately want to know
all the gory details.
Each Wait On Occurrence function “remembers” what occurrence it
last waited on and at what time it continued (either because the occur-
rence triggered or because of a time-out). When a VI is loaded, each
Wait On Occurrence is initialized with a nonexisting occurrence.
When a Wait On Occurrence is called and ignore previous is False,
there are four potential outcomes:

1. The occurrence has never been set. In this case Wait On Occurrence
simply waits.
2. The occurrence has been set since this Wait On Occurrence last exe-
cuted. In this case Wait On Occurrence does not wait.
3. The occurrence was last set before this Wait last executed, and last
time this Wait was called it waited on the same occurrence. Wait On
Occurrence will then wait.
4. The occurrence was last set before this Wait last executed, but last
time this Wait was called it waited on a different occurrence. In this
case Wait will not wait!

The first three cases are pretty clear, but the last one may seem a bit
strange. It will arise only if you have a Wait On Occurrence inside a
158 Chapter Six

loop (or inside a reentrant VI in a loop) and it waits on different occur-

rences (out of an array, for example). This can also arise if it is inside a
non-reentrant VI and the VI is called with different occurrences. Fortu-
nately, these obscure cases generally don’t happen.
Wait On Occurrence behaves in this way due to its implementa-
tion. Each occurrence “knows” the last time it was triggered, and each
Wait On Occurrence remembers the occurrence it was last called with
and what time it triggered (or timed out). When Wait On Occurrence
is called and ignore previous is False, it will look at its input. If the
input is the same occurrence as last time, it will look at the time of
the last firing and wait depending on whether the time was later than
the last execution. If the input is not the same as last time, it will
simply look at the time and wait depending on whether it has ever
been fired.
After you’ve tried a few things with occurrences, you’ll get the hang of
it and come up with some practical uses. Some experienced LabVIEW
users have written very elaborate scheduling and timing programs
based on occurrences (far too complex to document here). In essence,
you use sets of timers and various logical conditions to determine when
occurrences are to be triggered. All your tasks that do the real work
reside in loops that wait on those occurrences. This gives you central-
ized control over the execution of a large number of tasks. It’s a great
way to solve a complex process control situation. Just hope you don’t
have to explain it to a beginner.
To sum up our section on occurrences, here is a quote from Stepan
Rhia, a former top-notch LabVIEW developer at National Instruments:
One may think that occurrences are quirky, a pain to use, and that they
should be avoided. One might be right! They are very low level and you
often have to add functionality to them in order to use them effectively.
In other words, they are not for the faint of heart. Anything implemented
with occurrences can also be implemented without them, but maybe not
as efficiently.

Notifiers
As Stepan Rhia indicated, adding functionality around occurrences can
make them more useful. And that’s exactly what a notifier does: It
gives you a way to send information along with the occurrence trigger.
Although notifiers still have some of the hazards of occurrences
(obscuring the flow of data), they do a good job of hiding the low-level
trickery.
To use a notifier, follow these general steps on your diagram. First,
use the Create Notifier VI to get a notifier refnum that you pass
to your other notifier VIs. Second, place a Wait On Notification VI
 Synchronization 159

in a loop or wherever you want your program to be suspended until

the notification event arrives along with its data. Third, call the Send
Notification VI to generate the event and send the data. When you’re
all done, call Destroy Notifier to clean up.
Here’s an example that uses notifiers to pass data when an alarm
limit is exceeded. We’ll follow the numbered zones in Figure 6.7 to see
how it works.

1. A new notifier is created. Its refnum is passed to two While Loops.

The two loops don’t have to reside on the same diagram; they just
need to have the refnum passed to them, perhaps through a global
variable.
2. The upper loop reads a numeric value and displays it on a strip
chart every 200 ms. If the value exceeds the Upper Limit (alarm
level), we wish to wake up the other loop and send the value along.
3. Inside the True frame of the Case structure, the value is submitted
to the Send Notification VI. Behind the scenes, that VI generates an
occurrence and queues up the data.
4. The lower While Loop is waiting for the notification to be set. When
it arrives, the Wait On Notification VI returns the data.
5. A message for the user is assembled, and a dialog box pops up.
6. The upper loop stops when the user clicks the Stop button, and the
notifier is destroyed.
7. The lower loop stops when the Wait On Notification returns an error
because the notifier no longer exists.

Figure 6.7 A notifier sends data from the upper loop to the lower one when an
alarm limit is exceeded.
160 Chapter Six

If you actually create and run this VI, you’ll find one misbehavior.
When the notifier is destroyed and the bottom loop terminates, the
Wait On Notification VI returns an empty number, which is passed to
the dialog box to be displayed as a value of 0; it’s a false alarm. To elimi-
nate this bug, all nodes that use the data string should be enclosed in
a Case structure that tests the error cluster coming from the Wait On
Notification VI. In the remainder of this chapter, you’ll see that tech-
nique used as a standard practice.
Wait On Notification includes the time-out. Ignore previous features
of Wait On Occurrence, and they work the same way. However, most of
the time you don’t need to worry about them. The only trick you have
to remember is to include a way to terminate any loops that are wait-
ing. The error test scheme used here is pretty simple. Alternatively, you
could send a notification that contains a message explicitly telling the
recipient to shut down.
One important point to keep in mind about notifiers is that the data
buffer is only one element deep. Although you can use a single notifier
to pass data one way to multiple loops, any loop running slower than
the notification will miss data.

Queues
Queues function a lot like notifiers but store data in a FIFO (first in,
first out) buffer. The important differences between queues and noti-
fiers are as follows:

1. A notifier can pass the same data to multiple loops, but a notifier’s
data buffer is only one element deep. All notifier’s writes are destruc-
tive.
2. A queue can have an infinitely deep data buffer (within the limits
of your machine), but it’s difficult to pass the data to more than one
loop. A queue’s reads are destructive.

Another important difference between queues and notifiers is the

way they behave if no one reads the data. You might think that a queue
with a buffer size of 1 would behave the same as a notifier. But once a
fixed-size queue is filled, the producer loop either can time out and
not insert the data, or can run at the same rate as the consumer.
Figure 6.8 shows the alarm notification application of Figure 6.7
rewritten using queues. Note that the application looks and acts the
same, but with one exception: Because the queue size is infinite, no
alarms will be missed if the user fails to press the acknowledgment
button in time. Every alarm will be stored in the queue until dequeued.
 Synchronization 161

Figure 6.8 Alarm notification using queues. No alarms will be dropped because of user inactivity.

Use a queue when you can’t afford to miss any data passed between
parallel loops.
The asynchronous nature of queues makes for a great way to pass
messages between two loops running at different rates. Figure 6.9
shows an implementation of this using an Event structure in one loop
and a message-driven state machine in the other. The Queued Mes-
sage Handler is one of the most powerful design patterns in LabVIEW
for handling user-interface-driven applications. Notice we’re giving the
queue a unique name, to avoid namespace conflicts. All we need to
insert and remove data from a queue is the queue refnum or its name.
If we have another queue in memory using a string data type, the two
queues will use the same data buffer. This can be hard to troubleshoot!
Rule: Give each queue a unique name at creation.

Semaphores
With parallel execution, sometimes you find a shared resource called
by several VIs, and a whole new crop of problems pop up. A shared
resource might be a global variable, a file, or a hardware driver that
could suffer some kind of harm if it’s accessed simultaneously or in an
undesirable sequence by several VIs. These critical sections of your
program can be guarded in several ways.
162 Chapter Six

Figure 6.9 Queued Message Handler. Each event is placed in queue and handled in order by the con-
sumer loop.

One way is to encapsulate the critical section of code in a subVI. The

LabVIEW execution system says that a given subVI can only be run-
ning under the auspices of one calling VI, unless it is made reentrant
in the VI Properties dialog. For instance, say that you have a scratch
file that several VIs need to read or write. If the open-read/write-close
sequence was independently placed in each of those VIs, it’s possible
that the file could be simultaneously written and/or read. What will
happen then? A simple solution is to put the file access functions in
a subVI. That way, only one caller can have access to the file at any
moment.
 Synchronization 163

Global variables are another kind of shared resource that LabVIEW

programmers have trouble with. A familiar clash occurs when you’re
maintaining a kind of global database. Many VIs may need to access
it, and the most hazardous moment comes during a read-modify-write
cycle, much like the scratch file example. Here, the global variable is
read, the data is modified, and then it is written back to the global
(Figure 6.10A). What happens if several VIs attempt the same opera-
tion? Who will be the last to write the data? This is also known as a
race condition.
While built-in LabVIEW globals have no protection, you can solve
the problem by encapsulating them inside subVIs, which once again
limits access to one caller at a time. Or you can use global variables
that are based on shift registers (Figure 6.10B) and have all the pro-
grammed access there in a single subVI.
Computer science has given us a formal protection mechanism. A
semaphore, also known as a mutex (short for mutually exclusive),
is an object that protects access to a shared resource. To access a criti-
cal section, a task has to wait until the semaphore is available and
then immediately set the semaphore to busy, thus locking out access by
other tasks. It’s also possible for semaphores to permit more than one
task (up to a predefined limit) to be in a critical section.
A new semaphore is created with the Create Semaphore VI. It
has a size input that determines how many different tasks can use the
semaphore at the same time. Each time a task starts using the sema-
phore, the semaphore size is decremented. When the size reaches 0,
any task trying to use the semaphore must wait until the semaphore
is released by another task. If the size is set to 1 (the default), only one

Figure 6.10 Global variables are a classic risk

for race conditions. You can protect such data by
encapsulating it in a subVI or by storing the data
in shift registers.
164 Chapter Six

task can access the critical section. If you’re trying to avoid a race con-
dition, use 1 as the size.
A task indicates that it wants to use a semaphore by calling the Acquire
Semaphore VI. When the size of the semaphore is greater than 0, the
VI immediately returns and the task proceeds. If the semaphore size is
0-, the task waits until the semaphore becomes available. There’s a time-
out available in case you want to proceed even if the semaphore never
becomes available. When a task successfully acquires a semaphore and
is finished with its critical section, it releases the semaphore by calling
the Release Semaphore VI. When a semaphore is no longer needed,
call the Destroy Semaphore VI. If there are any Acquire Semaphore
VIs waiting, they immediately time out and return an error.
Figure 6.11 is an example that uses a semaphore to protect access
to a global variable. Only one loop is shown accessing the critical read-
modify-write section, but there could be any number of them. Or the
critical section could reside in another VI, with the semaphore refnum
passed along in a global variable. This example follows the standard
sequence of events: Create the semaphore, wait for it to become avail-
able, run the critical section, and then release the semaphore. Destroy-
ing the semaphore can tell all other users that it’s time to quit.
Semaphores are widely used in multithreaded operating systems
where many tasks potentially require simultaneous access to a common
resource. Think about your desktop computer’s disk system and what
would happen if there were no coordination between the applications
that needed to read and write data there. Since LabVIEW supports
multithreaded execution of VIs, its execution system uses semaphores
internally to avoid clashes when VIs need to use shared code. Where
will you use semaphores?

Figure 6.11 Semaphores provide a way to protect critical sections of code that can’t tolerate simultaneous
access from parallel calling VIs.
 Synchronization 165

Me and You, Rendezvous

Here’s another interesting situation that can arise with parallel tasks.
Say that you have two independent data acquisition tasks, perhaps
using different hardware, each of which requires an unpredictable
amount of time to complete an acquisition. (In other words, they’re
about as asynchronous as you can get.) But say that you also need to do
some joint processing on each batch of data that the two tasks return;
perhaps you need to combine their data into a single analysis and dis-
play routine. How do you get these things synchronized?
One way to synchronize such tasks is to use the rendezvous VIs.
Each task that reaches the rendezvous point waits until a specified
number of tasks are waiting, at which point all tasks proceed with exe-
cution. It’s like a trigger that requires a unanimous vote from all the
participants; nobody starts until we’re all ready.
To arrange a rendezvous in LabVIEW, begin by creating a rendez-
vous refnum with the Create Rendezvous VI. Pass that refnum to
all the required tasks, as usual. Use the Wait At Rendezvous VI to
cause a loop to pause until everyone is ready to proceed. When you’re
all through, call the Destroy Rendezvous VI.
Figure 6.12 shows a solution to our data acquisition problem using
rendezvous VIs. We’re going to acquire data from two asynchronous

Figure 6.12 Rendezvous VIs can synchronize par-

allel tasks by making them wait for each other.
166 Chapter Six

sources and then analyze it. In this solution, there are four parallel
While Loops that could just as well reside in separate VIs. (Leaving
everything on one diagram indicates a much simpler way to solve this
problem: Just put the two Read Data VIs in a single loop, and wire
them up to the analysis section. Problem solved. Always look for the
easiest solution!)
Start at the top of the diagram, while loop A is responsible for
stopping the program. When that loop terminates, it destroys the
rendezvous, which then stops the other parallel loops. Loops B and
C asynchronously acquire data and store their respective measure-
ments in global variables. Loop D reads the two global values and then
does the analysis and display. The Wait On Rendezvous VIs suspend
execution of all three loops until all are suspended. Note that the Cre-
ate Rendezvous VI has a size parameter of 3 to arrange this condi-
tion. When the rendezvous condition is finally satisfied, loops B and C
begin acquisition while loop D computes the result from the previous
acquisition.
There are always other solutions to a given problem. Perhaps you
could have the analysis loop trigger the acquisition VIs with occur-
rences, and then have the analysis loop wait for the results via a pair
of notifiers. The acquisition loops would fire their respective notifiers
when acquisition was complete, and the analysis loop could be arranged
so that it waited for both notifiers before doing the computation. And
there is no doubt a solution based on queues. Which way you solve
your problem is dependent upon your skill and comfort level with each
technique and the abilities of anyone else who needs to understand and
perhaps modify your program.
 7
Chapter

Files

Sooner, not later, you’re going to be saving data in disk files for future
analysis. Perhaps the files will be read by LabVIEW or another applica-
tion on your computer or on another machine of different manufacture.
In any case, that data is (hopefully) important to someone, so you need
to study the techniques available in LabVIEW for getting the data on
disk reliably and without too much grief.
Before you start shoveling data into files, make sure that you under-
stand the requirements of the application(s) that will be reading your files.
Every application has preferred formats that are described in the appro-
priate manuals. If all else fails, it’s usually a safe bet to write out numbers
as ASCII text files, but even that simple format can cause problems at
import time—things including strange header information, incorrect com-
binations of carriage returns and/or line feeds, unequal column lengths, too
many columns, or wrong numeric formats. Binary files are even worse,
requiring tight specifications for both the writer and the reader. They are
much faster to write or read and more compact than text files though, so
learn to handle them as well. This chapter includes discussions of some
common formats and techniques that you can use to handle them.
Study and understand your computer’s file system. The online
LabVIEW help discusses some important details such as path names
and file reference numbers (refnums). If things really get gritty, you
can also refer to the operating system reference manuals, or one of the
many programming guides you can pick up at the bookstore.

Accessing Files
File operations are a three-step process. First, you create or open a
file. Second, you write data to the file and/or read data from the file.
Third, you close the file. When creating or opening a file, you must

167
168 Chapter Seven

specify its location, or path. Modern computing systems employ a hier-

archical file system, which imposes a directory structure on the storage
medium. You store files inside directories, which can in turn contain
other directories. To locate a file within the file system, LabVIEW uses
a path-naming scheme that works consistently across all operating
systems. On the Macintosh and Windows, you can have multiple drives
attached to your machine, and each drive is explicitly referenced. On
Sun and Linux, the physical implementation is hidden from you. Here
are some examples of absolute path names:

Macintosh
HD80:My Data Folder:Data 123
Windows
C:\JOE\PROGS\DATA\DATA123.DAT
UNIX
/usr/johnny/labview/examples/data_123.dat

LabVIEW’s Path control (from the String and Path palette) automat-
ically checks the format of the path name that you enter and attempts
to coerce it into something valid for your operating system.
Paths can be built and parsed, just as strings can. In fact, you can
convert strings to paths and back by using the conversion functions
String To Path and Path To String. There are also several functions
in the File I/O function palette to assist you in this. Figure 7.1 shows
how the Build Path function can append a string to an existing path.
You might use this if you had a predefined directory where a file should
be created and a file name determined by your program. Given a valid
path, you can also parse off the last item in the path by using the Strip

LabVIEW has functions to build, parse, and compare path names. These are useful when you are
Figure 7.1
programmatically manipulating file names.
 Files 169

Path function. In Figure 7.1, the path is parsed until you get an empty
path. Constants, such as Empty Path and Not A Path, are useful for
evaluating the contents of a path name. The Default Directory con-
stant leads you to a location in the file system specified through the
LabVIEW preferences. You can also obtain a path name from the user
by calling the File Dialog function. This function allows you to prompt
the user and determine whether the chosen file exists.
Once you have selected a valid path name, you can use the Open/
Create/Replace File function either to create a file or to gain access
to an existing one. This function returns a file refnum, a magic num-
ber that LabVIEW uses internally to keep track of the file’s status. This
refnum, rather than the path name, is then passed to the other file I/O
functions. When the file is finally closed, the refnum no longer has any
meaning, and any further attempt at using it will result in an error.
You can navigate and directly manipulate your computer’s file system
from within LabVIEW. In the Advanced File palette are several useful
functions, such as File/Directory Info, List Folder, Copy, Move,
and Delete. This is a pretty comprehensive set of tools, although the
methods by which you combine them and manipulate the data (which
is mostly strings) can be quite complex.

File Types
LabVIEW’s file I/O functions can read and write virtually any file for-
mat. The three most common formats are

■ ASCII text-format byte stream files

■ Binary-format byte stream files
■ LabVIEW datalog-format files

ASCII text files are readable in almost any application and are the
closest thing to a universal interchange format available at this time.
Your data must be formatted into strings before writing. The resulting
file can be viewed with a word processor and printed, so it makes sense
for report generation problems as well. A parsing process, as described
in the strings section, must be used to recover data values after read-
ing a text file. The disadvantages of text files are that all this conver-
sion takes extra time and the files tend to be somewhat bulky when
used to store numeric data.
Binary-format byte stream files typically contain a bit-for-bit
image of the data that resides in your computer’s memory. They cannot
be viewed by word processors, nor can they be read by any program with-
out detailed knowledge of the files’ format. The advantage of a binary
file is that little or no data conversion is required during read and write
170 Chapter Seven

operations, so you get maximum performance. Binary files are usually

much smaller than text files. For instance, a 1000-value record of data
from a data acquisition board may be stored in its raw 16-bit binary
file, occupying 2000 bytes on disk. Converted to text, this record may
take up to 10,000 bytes, or 5 times as much. The disadvantage of binary
files is their lack of portability—always a serious concern.
LabVIEW offers another option, the datalog-format file, which is a
special binary format. This format stores data as a sequence of records of
a single arbitrary data type that you specify when you create the file.
LabVIEW indexes data in a datalog file in terms of these records.
Note that these records can be a complex type, such as a cluster, which
contains many types of data. LabVIEW permits random read access to
datalog files, and you can include timestamps with each record.

Writing Text Files

Here’s the basic procedure for most situations that write data to text
files:

1. Determine the path name.

2. Open or create the file.
3. Convert the data to a string if it’s not already in string format.
4. Write the string to the file.
5. Close the file.

Figure 7.2 uses several of the built-in file I/O functions to perform
these steps. First, the user receives a dialog box requesting the name
and location of a new file. The Open/Create/Replace File function
has an operation mode input that restricts the possible selections to
existing files, nonexisting files, or both. In this example, we wanted to
create a new data file, so we set the mode to create, which forces the
user to choose a new name. The function creates the desired file on
the specified path, and returns a refnum for use by the rest of the file
I/O functions. An output from the Open/Create/Replace File function,

Figure 7.2 The built-in file I/O functions are used to create, open, and write text data to a
file. The Case structure takes care of user cancellations in the file dialog.
 Files 171

canceled, is True if the user clicks the Cancel button. We wired it to

a Case structure—its True case is empty—to skip the file I/O process
if the operation was canceled. Next, the Write To Text File function
writes the already formatted string. After writing, the Close File func-
tion flushes the data to disk and closes the file. Finally, we call Simple
Error Handler (from the Dialog and User Interface function palette)
to notify the user if an error occurs.
Note the clean appearance of this example, thanks to the use of the
flow-through refnum and error I/O parameters. Rule: Always include
an error handler VI to do something sensible when an I/O error occurs.
Each of the file I/O functions returns an error code in its error I/O
cluster. Not handling errors is a risky proposition for file I/O activity.
Lots of things can go wrong. For instance, the disk might be full or a
bad path could be generated if your path selection technique is compli-
cated. Rule: Always use error I/O (error chaining) for file I/O operations,
when available.
Your VIs don’t have to be cluttered with lots of file management func-
tions. A practical example that uses an open refnum to append text
data to an open file is shown in Figure 7.3. Once a file has been opened
and a refnum created, LabVIEW uses the open refnum to access the
file’s Read Mark and Write Mark. Inside the While Loop, a subVI
reads data from some source and returns it in text format. Again, Write
To Text File is called, and the data is appended at the Write Mark (the
end of the file). The loop executes once per second, regulated by Wait
Until Next ms Multiple. The result is a file containing a header fol-
lowed by records of data at 1-s intervals.
This text-based scheme has some performance limitations. If you need
to save data at really high rates (such as thousands of samples per sec-
ond), the first thing to consider is binary-format files, which are very fast.
If you must use text files, avoid closing the file after writing each record.

Figure 7.3 A simple data logger that uses an open refnum for all the file management. In the loop, data is
acquired, formatted into a suitable string, and appended to the file. Simple, eh?
172 Chapter Seven

Figure 7.4 Here’s a simple way to log data to

spreadsheet-compatible files by using a file utility.

The file system maintains a memory-based file buffer, called a disk

cache, which increases disk performance. Closing the file flushes the
file buffer to disk. This forces the disk to move the heads around (called
seeking) and then write the data, a time-consuming process that limits
throughput. If you avoid closing the file until you are done, the disk
cache can keep a moderate amount of data in memory, flushing to disk
only when it becomes full. The disadvantage of using the disk cache is
that if your computer crashes during data collection, data in the disk
cache will be lost. For this reason, LabVIEW has a file I/O function
Flush File that intentionally writes the output buffer to disk. You call
Flush File occasionally to limit the amount of data in the buffer.
The file utility VI Write To Spreadsheet File can simplify your
life when your objective is to write spreadsheet-compatible text. You
often collect data in 1D or 2D numeric arrays, and you need to convert
them to tab-delimited text. As shown in Figure 7.4, you can accumulate
1D arrays by indexing on the border of a loop and then wire it right
into Write To Spreadsheet File. This is really convenient for exporting
data from simulations that create 1D or 2D arrays. The Transpose
option swaps rows and columns; it seems that you almost always need
to transpose. One risk of using this VI as shown here is that LabVIEW
may run out of memory if you let the While Loop run for a sufficiently
long time. Instead, you can put Write To Spreadsheet File inside the
acquisition loop: The VI will also accept 1D arrays, in which case it
writes a single line of data values, separated by tabs and with a trail-
ing end-of-line (EOL) marker. Another risk of using this VI is the time
and additional memory required to convert from numeric to text. A
2-mega-sample array can quickly swell from 8 megabytes as SGL data
to 32 megabytes or larger when converted to ASCII.

Reading Text Files

Reading data from a text file is similar to writing it:
1. Determine the path name.
2. Open the file.
3. Read the string from the file.
 Files 173

Figure 7.5 Use the Read Binary function to read text file from disk.

4. Close the file.

5. Convert the string to a suitable format, if necessary.

Figure 7.5 shows an example that is symmetric with Figure 7.2. The
File Dialog function provides a dialog box that only lets the user pick
an existing file. This is configured from File Dialog’s pop-up configura-
tion menu which is accessible from the block diagram. Open File opens
the file for read-only access and passes the refnum to Read Binary.
The Read Binary function can read any number of bytes wired to the
count input. By default, Read Binary starts reading at the beginning
of the file, and count is –1 (it reads all data). The default data type
returned is string. For the simple case where you wish to read part of
the file, just wire a constant to count and set it to the expected amount
of data. We’re using Read Binary in this example instead of Read Text
because the default behavior of Read Text is to stop at the first EOL.
Figure 7.6 shows several ways to use the Read Text function.
In the Read Text pop-up menu you select whether to read bytes or
lines, and whether or not to convert EOL (Windows = CR/LF, Mac OS =
CR, Linux = LF) characters. The default behavior is to convert EOL
and read lines. Recognizing lines of text is a handy feature, but may not
be what you expect when Read Text returns only the first line of your
file. Figure 7.6A shows Read Text configured not to read lines of text.
Setting count equal to –1 reads an entire file. In Figure 7.6B, Read Text
is configured to read lines and returns the number of lines equal to
count as an array. In Figure 7.6C, Read Text increments through a text
file one line at a time. This can be especially useful when you are incre-
menting through a text file containing an automated test sequence.
Another useful file utility VI is Read From Spreadsheet File,
shown in simplified form in Figure 7.7. This VI loads one or more lines
of text data from a file and interprets it as a 2D array of numbers. You
can call it with number of rows set to 1 to read just one line of data at
a time, in which case you can use the first row output (a 1D array). If
the rows and columns need to be exchanged, set transpose to True. By
default, it assumes that the tab character delimits values on each line.
174 Chapter Seven

Figure 7.6 The Read Text function can read bytes or lines from a text file
depending on its configuration. (A) It reads bytes (−1 = all). (B) It reads lines.
(C) It reads a file one line at a time.

If your format is different, you can enter the appropriate character in

the delimiter string. Like many of the utilities, this one is based on
another utility subVI, Read Lines From File, that understands the
concept of a line of text. The end-of-line character (characters), which
varies (vary) from system to system, is (are) properly handled. This
functionality is now actually built into the Read Text function, as illus-
trated in Figure 7.6. The Read Lines From File goes on to read multiple
lines until the end of the file is encountered.
Words of wisdom: Always remember that you can edit the diagram
of a utility VI such as this to accommodate your special requirements.
It’s a powerful way to program because the groundwork is all done for
you. But also remember to save the modified version with a new name,
and save it someplace besides the vi.lib directory. If you don’t, your next
LabVIEW upgrade may overwrite your changes.

Figure 7.7The file utility VI Read From Spreadsheet File.vi interprets a text file containing
rows and columns of numbers as a matrix (2D array). This is a simplified diagram.
 Files 175

Formatting to Text Files

There’s another way to quickly read and write small amounts of text
in files with control over the formatting. In Figure 7.8A, we used the
Format Into File function from the Strings palette to write a format-
ted ASCII string into a new file. This function works like the Format
Into String function in conjunction with file functions that write the
data to disk; it’s a great space saver. You supply a Format String that
contains all the usual C-style format specifiers, and then you wire up
as many input items as you like. It’s limited to scalars, so you’ll have to
use other functions, such as Array To Spreadsheet String, if you need
to write an array. This example shows a path control to choose the file,
but you can also supply a file refnum if you already have a file open for
writing.
Figure 7.8B shows the complementary function, Scan From File,
doing some on-the-fly parsing of a string as it is loaded from a text file.
In this case, we expect a string that looks like this

12.34V,9

followed by a carriage return. It might represent a measured value

(12.34) followed by a single character for the units (V), then a comma
as a delimiter, and finally a scale factor (9). The Format String was cre-
ated by popping up on the function and selecting Edit Scan String. That
helps to guide you through the process of building the sometimes cryp-
tic collection of format specifiers. The trickiest part here is the match
for any single character in a set comprising all upper- and lowercase
letters: %[A-Z,a-z]. Most of us mere mortals have trouble remember-
ing such details, so we use the dialog. You can keep expanding Scan
From File to handle an arbitrary number of items. Again, it doesn’t
scan arrays; you’ll have to load those as strings and crack them with
Spreadsheet String To Array, or use one of the other file utilities.

Figure 7.8 The (A) Format Into File and (B) Scan
From File functions are great for accessing files con-
taining text items with simple formatting.
176 Chapter Seven

Figure 7.9 Express VIs Read From

Measurement File.vi and Write To
Measurement File.vi take care of a lot
of the work of reading and writing to
files. They have a special LabVIEW
format *.lvm for text and *.tdm for
binary.

A recent addition to LabVIEW includes native measurement file for-

mats for text and binary files. These new formats include the metain-
formation (who, what, when, how) about the measurement in the
measurement file. The Write To Measurement File.vi and Read
From Measurement File.vi convert to and from ASCII text files
(*.lvm) and binary files (*.tdm). The text files include metainformation
in a header within each file while the binary scheme uses an XML file
to describe binary data stored in a separate data file. The express VI
Write To Measurement File.vi includes a lot of functionality, such as
the ability to automatically generate a new file name each time the VI
is run. This makes it easy to acquire sequences of data in a loop and
give each run a new file name. Figure 7.9 shows the Read From Mea-
surement File and Write To Measurement File express VIs.

Binary Files
The main reasons for using binary files as opposed to ASCII text are
that (1) they are faster for both reading and writing operations and (2)
they are generally smaller. They are faster because they are smaller
and because no data conversion needs to be performed (see the section
on conversions). Instead, an image of the data in memory is copied byte
for byte out to the disk and then back in again when it is read. Convert-
ing to ASCII also requires more bytes to maintain numerical precision.
For example, a single-precision floating-point number (SGL; 4 bytes)
has 7 significant figures. Add to that the exponent field and some ±
signs, and you need about 13 characters, plus a delimiter, to represent
 Files 177

it as ASCII. In binary you only need to store the 4 bytes—a savings of

more than 3 to 1. Byte stream binary files (as opposed to datalog files)
are also randomly accessible. That means you can tell the disk drive
to read or write particular areas or individual bytes in the file. Random
access is a bit tricky, however, as we shall see.
One small problem is that the program reading the data must know
every little detail about how the file was written. Data types, byte
counts, headers, and so forth have to be specified. You can’t view it as
an ASCII text string, so there’s little hope of deciphering such a file
without plenty of information. Even LabVIEW datalog files (a special
type of binary file, discussed later) are not immune to this problem.
If you can’t tell the file I/O functions what data type the datalog file
contains, it can’t be accessed. On the other hand, consider the fact that
parsing an arbitrary ASCII text file is also nontrivial, as we discussed
in the section on strings. That’s why there aren’t any universal text file
interpreter VIs.
You should consider using binary files when

■ Real-time performance is crucial. Avoiding the numeric-to-text con-

version and the extra amount of data that accompanies an ASCII
text file will definitely speed things up.
■ You already have a requirement for a certain binary format. For
instance, you may be writing files in the native format of another
application.
■ Random read-write access is required.

To make a binary file readable, you have several options. First, you
can plan to read the file back in LabVIEW because the person who
wrote it should have darned well be able to read it. Within LabVIEW,
the data could then be translated and written to another file format
or just analyzed right there. Second, you can write the file in a for-
mat specified by another application. Third, you can work closely with
the programmer on another application or use an application with the
ability to import arbitrary binary formatted files. Applications such as
Igor, S, IDL, MATLAB, and Diadem do a credible job of importing arbi-
trary binary files. They still require full information about the file for-
mat, however. These are several common data organization techniques
we’ve seen used with binary files:

1. One file, with a header block at the start that contains indexing
information, such as the number of channels, data offsets, and data
lengths. This is used by many commercial graphing and analysis
programs.
178 Chapter Seven

2. One file, organized as a linked list. Each record (say, a channel of

data) is written with its own header that tells the reader both where
this record begins and ends and where to find the next record in the
file (this is called a link). This technique is used by some graphing
and database programs.
3. Two files, one containing only the actual binary data and another
containing an index to the data, perhaps in ASCII format. This is
the format used by LabVIEW in *.tdm files.

In all these formats, the index or header information is likely to

include such things as channel names, calibration information, and
other items you need to make sense of the data. Binary files are much
more than a big bag of bytes.

Writing binary files

LabVIEW’s file I/O functions make it easy to access binary files. The
effort, as with text files, is all in the data formatting. You have to fig-
ure out how to deal with your particular header or index format, and
of course the data itself. Sorry, but that’s your problem, since no two
formats are the same. The best we can do is to show you the basics of
random access files.
Figure 7.10 is a simple driver VI for a binary file format that we just
made up. This format has a header containing one I32 integer (4 bytes)
that is the number of data samples to follow. The data follows imme-
diately afterward and is in SGL floating-point format. Each time this
driver is called, the data array is appended to the file and the header is
updated to reflect the total sample count.

Figure 7.10 This subVI appends an array of SGL floats to a binary file and updates the sample count, an
I32 integer, located at the beginning of the file. (Error handling is omitted for clarity.)
 Files 179

An important concept in file access is the file mark, which is a

pointer to the current location in the file, measured in bytes. It is main-
tained invisibly by the file system. The Set File Position function
permits you to place the mark anywhere you wish. This is how you
achieve random access: Figure out where you are, then move to where
you need to be. If the mark is at the end of the file (EOF) and you call
Write To Binary File, then data is appended. If the mark is in the mid-
dle of the file, then Write To Binary File will overwrite existing data,
starting at the mark.
In this example, the calling VI opens and closes the data file and
passes the file’s refnum via a refnum control. There are no special tricks
to opening or creating a byte stream binary file; LabVIEW does not dif-
ferentiate between binary and text files with regard to file type—it’s
only how you interpret the data that makes a difference. On the first
call to this VI, Initialize should be set to True, which results in the
sample count being set to 0 and written to the file. An uninitialized
shift register stores the sample count for use on subsequent calls.
After initialization, Set File Position sets the write mark to the EOF,
and Write To Binary File appends data to the end of the file. Since
Write To Binary File is polymorphic, we just wired the incoming data
array to the data input. The size of the array is added to the previous
sample count. Then the Write To Binary File is called again, but this
time Set File Position has set the write mark to the start of the file,
which forces it to overwrite at the start of the file. The data in this case
is our updated sample count, an I32 integer, which overwrites the pre-
vious value without bothering data elsewhere in the file.

Reading binary files

Reading the data is even easier, at least for this simple example, as
shown in Figure 7.11. Read File is called with the type input wired to
an I32 constant so that it returns the sample count with the appropri-
ate interpretation. This is another example of polymorphism at work.

Figure 7.11 Reading the simple binary file created

in the previous example. All you have to do is read
the first 4 bytes, which represent the number of
samples to follow, then read that number of data
points. (Error handling is omitted for clarity.)
180 Chapter Seven

Figure 7.12 This VI reads random blocks of data from a binary file written by the LabVIEW DAQ example VI,
Cont Acq to File (scaled). I couldn’t make it much simpler than this.

If you do not wire to type, then the default type, string, is used, and
your data is returned as a string. Next, Read File is called again to read
the data, which is in SGL format. Because count is wired, Read File
returns an array of the specified type. No special positioning of the file
marker was required in this example because everything was in order.
Things are more complex when you have multiple data arrays and/or
headers in one file.
Random access reading is, as we said earlier, a bit tricky because
you have to keep track of the file mark, and that’s a function of the
data in the file. To illustrate, we created a fairly simple VI that allows
you to read random blocks of data from a file created by the LabVIEW
DAQ example VI, Cont Acq to File (scaled), located in the directory
examples/daq/analogin/strmdisk.llb/. The binary file created by that
example contains an I32 header whose value is the number of channels,
followed by scans of SGL-format data. A scan consists of one value for
each channel. The objective is to read and display several scans (we’re
calling this a block of data), starting at an arbitrary location in the file.
Figure 7.12 is a solution (not the best, most feature-packed solution, we
assure you, but it’s simple enough to explain). Here’s how it works.

1. The binary file is opened.

2. A single I32 integer is read by the Read Binary function. Its value is
the number of channels in each scan. For computation of file offset, it
is multiplied by 4, the number of bytes in a value of type SGL. Already,
you can see the dependency on prior knowledge of the file’s contents.
3. The program idles in a While Loop. On each iteration, a new file off-
set is computed. Offset equals scans per block, times the number of
bytes per scan, times the desired block number, plus 4 bytes (which
accounts for the I32 header). Read Binary is set up to read relative
to the start of the file with the computed offset. We used another cool
 Files 181

feature of this file VI: It understands the concept of rows and col-
umns when reading 2D arrays. You can see the row and col values
on the diagram bundled into the count input of Read File.
4. The data is loaded and displayed on a waveform graph. You must
pop up on the graph and select Transpose Array; otherwise, the x
and y axes are swapped.
5. A small While Loop runs until the user clicks the Update button.

When the program is done, the file is closed and errors are checked.
As you can see, even this simple example is rather tricky. That’s the
way it always is with binary files.

Writing Datalog Files

LabVIEW datalog files conveniently store data in binary format one
record at a time. The datalog file I/O functions are polymorphic, and
record can be any LabVIEW data type, although you will usually
use a cluster containing several data types. The datalog functions
are available from the File I/O / Datalog palette. Datalog files require
you to make a connection to the type input of Open/Create/Replace
Datalog function. When the file is opened, the refnum (properly called
a datalog refnum) carries information that describes the type of
data in the file. The Read Datalog and Write Datalog functions
then adapt to the appropriate type. A broken wire will result if you
attempt to wire the data terminal to a source or destination of the
wrong type.
Figure 7.13 is a rework of our simple data logger using LabVIEW
datalog files. At the lower left, the data type is specified as a cluster

Figure 7.13Writing data to a datalog file. The data type in this case is a cluster containing a string and a
numeric array. Records are appended every second. (Error handling is removed for clarity.)
182 Chapter Seven

containing a string and an SGL numeric array. Open/Create/Replace

Datalog responds by creating a file with a datalog refnum that cor-
responds to this type. In the While Loop, Write Datalog appends one
record per iteration, containing a string with the current time and an
array of values. The datalog refnum tells Write Datalog what data type
to expect. Note that datalog files are effectively append only. To edit the
contents of a datalog file, you must open the file and copy the desired
records to another, similar datalog file.
Note the use of a typedef control for determination of data type in
this example. By wiring such a type specifier to all dependent objects
(Open/Create/Replace Datalog, Write Datalog, and Bundle By Name),
you can guarantee that the program will work properly with no bro-
ken wires, even if you change the constituents of the control. This is
extremely important for code maintenance in large projects and is a
good habit to get into. For this simple example, you wouldn’t have to
edit or save the control as a typedef, but at least it’s properly wired and
easy to change if necessary.
You can also create datalog files by using front panel datalogging
functions under the Operate menu. Front panel datalogging appends a
record containing a timestamp and all the front panel data to a file of
your choosing. In the Datalogging submenu, you create a file (Change
Log File Binding . . .) and then manually log the panel at any time
(the Log . . . command) or enable Log At Completion to automati-
cally append a record when execution of the VI ends. The VI need not
have its front panel displayed for Log At Completion to function. This
is a handy way to do datalogging without programming in the usual
sense. Just create a subVI with a set of controls you wish to log (it
does not need any wiring on its diagram). Set it up to Log At Com-
pletion, wire it into the diagram of a calling VI, and every time the
subVI is called, a data snapshot is recorded. The format of this kind
of datalog is a cluster containing two things: a timestamp and a clus-
ter with all front panel controls. The cluster order is the same as the
Tabbing Order, which you set in the Edit >> Set Tabbing Order . . .
menu item.
Front panel datalogging creates standard datalog files, readable by
the methods described in the following section. You can also view the
contents of each record right on the panel by choosing Retrieve . . .
from the Datalogging submenu. This is a convenient way of storing and
manually retrieving VI setups. You can also place the VI from which
you logged data onto a diagram, pop up on it, and select Enable Data-
base Access. A halo that looks like a file cabinet appears around the
icon of the subVI. It has terminals for programmatically accessing
specified records and for reading timestamps and data.
 Files 183

Figure 7.14This VI reads a single datalog of the format written by the previous example.
Note the value of creating a typedef control to set the datalog type. (Error handling is
removed for clarity.)

Reading Datalog Files

You can read one or more records at a time from a datalog file. Figure 7.14
shows the basic method by which you read a single record. The EOF
function reports the number of records in the file, rather than the num-
ber of bytes, as it did with a byte stream text file. Similarly, Read File
responds to its pos offset input by seeking the desired record number
(starting with zero) rather than a byte number. In this example, one
record containing the expected bundle of data is returned. Note how
Read File’s data output is of the same type as the datalog typedef wired
to Open File. The type information is carried in the datalog refnum.
If you want to read more than one record at a time, just connect a
numeric value to the count terminal of Read File, as shown in Fig-
ure 7.15. Its output will then be an array of the data type previously
defined. You can also wire the pos offset input to select which record is
the first to be read. With these two controls, you can read any number
of records (in sequence) from any location in the file.

Datalog file utilities

An easy way to get started with datalog files is to use the utility VIs in
the LabVIEW examples (path: labview/examples/file/datalog.llb). These
VIs are modeled after some of the other file utilities and include func-
tions to open, read, write, and close files containing datalogs with a data
type of your choice. The examples simulate common usage, and there is a
handy example that reads one record at a time when you click a button.

Figure 7.15 This code fragment reads all the records from a
datalog file.
This page intentionally left blank
 Chapter

Building an Application
8
Many applications we’ve looked at over the years seemed as though
they had just happened. Be it Fortran, Pascal, C, or LabVIEW, it was
as if the programmer had no real goals in mind—no sense of mis-
sion or understanding of the big picture. There was no planning. Gary
actually had to use an old data acquisition system that started life
as a hardware test program. One of the technicians sat down to test
some new I/O interface hardware one day, so he wrote a simple pro-
gram to collect measurements. Someone saw what was going on and
asked him if he could store the data on disk. He hammered away for
a few days, and, by golly, it worked. The same scenario was repeated
over and over for several years, culminating in a full-featured . . .
mess. It collected data, but nobody could understand the program,
let alone modify or maintain it. By default, this contrivance became
a standard data acquisition system for a whole bunch of small labs.
It took years to finally replace it, mainly because it was a daunting
task. But LabVIEW made it easy. These days, even though we start by
writing programs in LabVIEW, the same old scenario keeps repeating
itself: Hack something together, then try to fix it up, and document it
sometime later.
Haphazard programming need not be the rule. Computer scientists
have come up with an arsenal of program design, analysis, and quality
management techniques over the years, all based on common sense
and all applicable to LabVIEW. If you happen to be familiar with such
formalisms as structured design, by all means use those methods. But
you don’t need to be a computer scientist to design a quality applica-
tion. Rather, you need to think ahead, analyze the problem, and gener-
ally be methodical. This chapter should help you see the big picture

185
186 Chapter Eight

and design a better application. Here are the steps we use to build a
LabVIEW application:

1. Define and understand the problem.

2. Specify the type of I/O hardware you will need.
3. Prototype the user interface.
4. Design—then write—the program.
5. Test and debug the program.
6. Write the documentation (please!).

To make this whole process clearer, we’re going to go through a real

example: a LabVIEW-based control and datalogging system Gary
developed for Larry’s Vacuum Brazing Lab (call it VBL, for short). At
each step, we outline the general approach and then illustrate the par-
ticular tactics and the problems discovered in the VBL project. At the
end of the chapter we look at the LabVIEW Certification exams.

Define the Problem

If you, the system designer, can’t understand the problem, all is lost!
You will stumble about in the dark, hoping to accidentally solve your
customer’s problems through blind luck or divine intervention.
Sorry, but that’s not going to happen. What you need to do is to spend
a significant amount of time just understanding the problem. At this
point, you don’t need a computer at all, just a pencil, maybe a voltmeter
or an oscilloscope, and the desire to find out what is really going on.

Analyze the user’s needs

Interview the end users in several sessions. Talk to everyone involved,
paying particular attention to the operators—the ones who actually
have to interact with the VIs you write. Even if you are the customer,
you still have to go through this step to make sure that you have all
the important specifications on paper. How else will you know when
you’re done?
Tell your customers about LabVIEW. Show them what it can do in
a live demonstration. That way, you can speak a common language—
you can speak of controls, indicators, files, and subVIs. In return, your
customer should show you the system that needs to be monitored or
controlled. Perhaps there’s an existing measurement and control sys-
tem that has shortcomings—find out what’s wrong with that exist-
ing system. Learn the proper terminology. Maybe you need to read up
on the technology if it’s something you’ve never worked with before.
 Building an Application 187

Vacuum Belljar

Tungsten Heating Element

Materials Being Brazed

Type S
Thermocouple

Eurotherm
Triac Power
Temperature 208 VAC, 30A
Controller
Controller

4-20mA

RS-422 Serial Communications

Figure 8.1 A simplified vacuum brazing furnace.

The temperature controller adjusts power to the
heating element until the measured tempera-
ture matches the desired set point. Not shown
are the vacuum pump and other support equip-
ment. Serial communications make it possible for
LabVIEW to run this furnace.

Gary didn’t know what vacuum brazing was until he spent a couple of
hours with Larry. Here’s what he learned:
The principle of vacuum brazing is simple (Figure 8.1). Brazing involves
joining two materials (usually, but not always, metal) by melting a filler
metal (such as brass) and allowing it to wet the materials to be joined. Sol-
dering (like we do in electronics) is similar. The idea is not to melt the mate-
rials being joined, and that means carefully controlling the temperature.
Another complication is that the base metals sometimes react with the
air or other contaminants at high temperatures, forming an impervious
oxide layer that inhibits joining. Larry’s solution is to put everything in
a vacuum chamber where there is no oxygen, or anything else for that
matter. Thus, we call it vacuum brazing. The VBL has five similar electric
furnaces, each with its own vacuum bell jar. The heater is controlled by
a Eurotherm model 847 digital controller, which measures the tempera-
ture with a thermocouple and adjusts the power applied to the big electric
heater until the measured temperature matches the set point that Larry
has entered.

Gather specifications
As you interview the customers, start making a list of basic functional
requirements and specifications. Watch out for the old standoff
where the user keeps asking what your proposed system can provide.
You don’t even have a proposed system yet! Just concentrate on get-
ting the real needs on paper. You can negotiate practical limitations
later.
188 Chapter Eight

Ask probing questions that take into consideration his or her actual
understanding of the problem, his or her knowledge of instrumenta-
tion, and LabVIEW’s capabilities. If there is an existing computer
system (not necessarily in the same lab, but one with a similar pur-
pose), use that as a starting point. As the project progresses, keep
referring to this original list to make sure that everything has been
addressed. (It’s amazing, but every time I think I’m finished, I check
the requirements document and find out I’ve missed at least one
important feature.)
Review the specifications with the user when you think you under-
stand the problem. In formal software engineering, the requirements
document may adhere to some standards, such as those published by
IEEE, which can be quite involved. In that case, you will have formal
design reviews where the customer and perhaps some outside experts
will review your proposed system design. Although it is beyond
the scope of this chapter, here are some important requirements
to consider:

■ Characterize each signal. Is it digital or analog? Low or high fre-

quency? What kind of I/O hardware is proposed or required for proper
interfacing? Are LabVIEW drivers available?
■ What are the basic operations, steps, event sequences, and proce-
dures that the software needs to perform? This determines much of
the overall structure of the program.
■ Are there critical performance specifications? For example, rapid
response time may be required in a control loop, or very high accu-
racy and noise rejection are desired. Make sure that the hardware,
LabVIEW, and your programming expertise are up to the task.
■ What displays and controls are required? Break down displays by
type, such as charts, graphs, numerics, strings, and booleans. Consider
which screen object controls or displays which signals.
■ Determine the signal analysis needs such as statistics, linearization,
peak detection, frequency transforms, filtering and so on. If unusual
algorithms are required, seek details early in the project. What
needs to be done in real time versus in postrun analysis? This may
drastically affect your choice of computer and the system’s overall
performance.
■ What will be done with the data that will be collected? Compatibil-
ity with other applications (sometimes on other computers) is an
important issue. You must always save the data in a suitable format,
and you should be involved with the development of analysis tools.
Various options for real-time exchange with other programs may be
 Building an Application 189

considered as well (dynamic data exchange; interapplication com-

munication; networking).
■ What other information needs to be recorded besides acquired data?
Often, users like to save run setup information or comments in a
running log format with timestamps.
■ How much of this is really necessary? Separate needs from wants.
Prioritize the tasks. Realize that there are always constraints on
budget and/or time, and you have to work within these constraints.
Speaking of which, get commitment on the budget and schedule early
in the project!

The major requirements for the VBL project I divided into must-haves and
future additions, which prioritized the jobs nicely. Larry was really short
of funding on this job, so the future additions are low priority.

Must-Haves
1. All five systems may operate simultaneously.
2. Procedure: ramp up to a soak setpoint, then go to manual control so
user can tweak temperature up to melting point. Resume ramp on
command.
3. LabVIEW will generate the temperature ramp profiles. User needs a
nice way to enter parameters and maintain recipes.
4. Strip-chart indicators for real-time temperature trends.
5. Notify user (beep and/or dialog) when the soak temp is reached.
6. Alarm (beep and dialog) when temperature exceeds upper limit in
controller.
7. Make post-run trend plots to paste into Microsoft Word report
document.

Future Additions
1. Control vacuum pump-down controller. This will require many RS-232
ports.
2. Trend vacuum measurement. This will require even more RS-232
ports.

Draw a block diagram

Draw a preliminary block diagram of the system. Think in terms of
signal flow all the way from I/O hardware to computer screen and back
out again. What are the inputs and outputs? This will define the sig-
nals, and the signals will define the kind of I/O hardware and data pre-
sentation that you will need. List the characteristics of each signal.
190 Chapter Eight

Some of the items to consider are as follows:

■ Sensor and actuator types, manufacturers, model numbers, and so
forth
■ Number of channels, categorized by signal type
■ Signal characteristics: voltage, current, pulse rate, and so forth
■ Frequency content (determines sampling rates and filtering require-
ments)
■ Isolation and grounding requirements
■ Type of analysis to be performed (influences acquisition and storage
techniques)

Get as much detail as possible right now. This information is vital if

you hope to design a really successful automation package. It would be a
shame to misspecify I/O hardware or write a program that simply didn’t
account for the needs of all the instruments you plan to support.
For VBL, I learned that there would only be two basic signals running
between the computer and each furnace: the measured temperature from
the controller and the setpoint going to the controller (Figure 8.2). In their
raw forms, both are DC analog signals. The Eurotherm controller converts
the signals to and from digital commands that are transmitted over an
RS-422 serial communications line. RS-422 permits multidrop connec-
tions, meaning that multiple controllers can share a single communica-
tions line. Therefore, I knew that no signal conditioning of the usual type
was needed, but a driver for this particular Eurotherm controller would
have to be written. As a bonus, the communications hardware uses dif-
ferential connections to enhance noise rejection—very thoughtful, these
engineers.

Collect details on any special equipment proposed for use in the sys-
tem. Technical manuals are needed to figure out how to hook up the
signal lines and will also give you signal specifications. For all but the
simplest instruments there may be a significant driver development

Trend Alarm RS-422 Measured Temp

LabVIEW
Eurotherm TC
RS-422 Temp Setpoint Temperature
Controller Heater
Control 305.6

Similar for 5 furnaces

Figure 8.2 Draw a block diagram of your project to make sure you know how the pieces fit
together. This one was really simple, with only one input and one output for each of five
furnaces.
 Building an Application 191

effort required (especially for GPIB or serial communications), so you

will need to get a copy of the programming manual as well. Sometimes
one major instrument is the I/O system.
When you have all the information together, write up a system
requirements document—even if it’s only one page. Even if your com-
pany does not require it. Even if it seems like too much work at the
moment. Why? Because it will pay off in the form of a higher-quality
product.

Specify the I/O Hardware

Now that you know which instruments will be used on your project,
follow the steps outlined in Chapter 12, “Inputs and Outputs,” to deter-
mine what kind of I/O hardware to use. The signals your instruments
produce will define the hardware. There are many factors to consider,
so choose wisely.
Since you have to write a LabVIEW program to support the I/O,
having examples of completed drivers available will save you much
time and effort. The LabVIEW instrument driver network (www
.ni.com/idnet) contains hundreds of ready-to-use drivers, all available
for free. Even if your instrument isn’t there, there may be something
very close—a different model by the same manufacturer—that is eas-
ily modified. If you plan to use any of the National Instruments (NI)
multifunction boards, prowl through the DAQ example VIs to find rel-
evant demonstration programs that you can use as starting points. For
simple projects, it’s surprising how quickly the examples can be modi-
fied to become the final application.
Availability and performance of drivers are one more consideration
in your choice of I/O hardware. You may need to do some preliminary
testing with the actual instruments and data acquisition equipment
before you commit to a final list of specifications. Be absolutely sure
that you can meet the user’s needs for overall throughput: the rate
at which data can be acquired, processed, displayed, and stored. Write
some simple test VIs at this point and rest better at night. The Speed
Tester VI in Figure 8.3 is useful when you have a subVI that you want
to exercise at top speed. You can put anything inside the For Loop and
get a fairly accurate measure of its execution time. Make sure that the
speed tester does plenty of iterations so that you get enough timing
resolution.
In most cases you can’t write the application unless you know what
hardware will be used. For instance, in a high-throughput data acquisi-
tion application in which many thousands of samples per second of data
are streamed to disk, you probably can’t do much on-the-fly analysis,
graphics, or changing of windows because of the extreme performance
192 Chapter Eight

Figure 8.3 A general-purpose speed tester VI that is used when you need to
check the throughput or iteration rate of a subVI or segment of code. This
example does 10 floating-point operations.

demands on the CPU. On the other hand, programming of a low-speed

temperature recorder has no such demands and permits a substantial
amount of real-time processing and display.
The Eurotherm controllers use the serial port for communications. Even
though Larry was only asking for one sample per second top speed, I sus-
pected that there could be an upper limit to the total throughput lurking
close by. Serial instruments are notorious for adding lots of header and
trailer information to otherwise simple messages, which compounds the
problem of an already-slow communications link. The bad news was that
there was no driver available, so I had to write one from scratch before
any testing was possible. The good news was that it only took 30 ms to
exchange a message with a controller. That means I could move about
33 messages per second, distributed among the five controllers, or about
6 messages per controller per second. Not a problem, unless he ever wants
to triple the number of furnaces.

Prototype the User Interface

An effective and fun way to start designing your application is to create
prototype front panels that you can show to the users. The panels don’t
have to work at all, or maybe you can just hook up random number
generators to animate the graphs and other indicators. Don’t spend
much time programming; concentrate on designing panels that meet
the specifications. Consider the best types of controls and indicators
for each signal. Make a list of inputs and outputs, grouped in some
logical categories, and pair them up with the most sensible control and
indicator types. If you have to write a formal software requirements
 Building an Application 193

document, you can paste images of your mock front panels into the
document. That makes your intentions clearer.

Panel possibilities
Remember that you have an arsenal of graphical controls available;
don’t just use simple numerics where a slider or ring control would be
more intuitive. Import pictures from a drawing application and paste
them into Pict Ring controls or as states in boolean controls, and so
forth, as shown in Figure 8.4. Color is fully supported (except in this
book). Deal with indicators the same way. Use the Control Editor
to do detailed customization. The Control Editor is accessed by first
selecting a control on the panel and then choosing Customize Con-
trol from the Edit menu. You can first resize and color all the parts of
any control and paste in pictures for any part and then save the control
under a chosen name for later reuse. Be creative, but always try to
choose the appropriate graphic design, not just the one with the most
glitz. After all, the operator needs information, not entertainment.
Booleans with pictures pasted in are especially useful as indicators.
For instance, if the True state contains a warning message and the
False state has both the foreground and background color set to trans-
parent (T, an invisible pen pattern in the color palette when you are
using the coloring tool), then the warning message will appear as if by
magic when the state is set to True. Similarly, an item can appear to

Slider (Text Labels) Pict Ring

Source Mode
Source Mode

Base StepGen, Emitter Com Vcc Vcc Vcc

Base Vcc, Emitter Com o

Base Com, Emitter StepGen Base StepGen, Base Open, Base Short,
Emitter Common Emitter Common Emitter Common

o
o Vcc Vcc
Ring or Enum Vcc

Source Mode Base Common, Base Common,

Emitter Open Emitter StepGen Base Vcc,
Base Com, Emitter StepGen Emitter Common

List Box
Source Mode
Base StepGen, Emitter Com
Base Vcc, Emitter Com
Base Com, Emitter StepGen

Figure 8.4 Controls can be highly customized with various text styles and imported pictures.
Here are four sensible numeric controls that could be used for the same function. Pick your
favorite.
194 Chapter Eight

move, change color, or change size based on its state through the use of
these pasted-in pictures. The built-in LabVIEW controls and indicators
can be made to move or change the size programmatically through the
use of Property nodes.
Real-time trending is best handled by the various charts, which are
designed to update displayed data one point or one buffer at a time.
Arrays of data acquired periodically are best displayed on a graph,
which repaints its entire display when updated. Note that graphs and
charts have many features that you can customize, such as axis labels,
grids, and colors. If you have a special data display problem that is
beyond the capability of an ordinary graph or chart, also consider the
Picture Controls. They allow you to draw just about anything in a
rectangular window placed on the front panel. There are other, more
advanced data display toolkits available from third parties. Chapter 20,
“Data Visualization, Imaging, and Sound,” discusses aspects of graphi-
cal display in detail.
Simple applications may be handled with a single panel, but more
complex situations call for multiple VIs to avoid overcrowding the
screen. Subdivide the controls and indicators according to the function
or mode of operation. See if there is a way to group them such that
the user can press a button that activates a subVI that is set to Show
front panel when called (part of the VI Properties menu, available
by popping up on the icon). The subVI opens and presents the user with
some logically related information and controls, and it stays around
until an Exit button is pressed, after which its window closes—just as
a dialog box does, but it is much more versatile. Look at the Window
options in the VI Properties menu. It contains many useful options
for customizing the look of the VI window. Functions for configuration
management, special control modes, alarm reporting, and alterna-
tive data presentations are all candidates for these dynamic windows.
Examples of dynamic windows and programming tricks illustrating
their use are discussed in Chapter 18, “Process Control Applications.”
For demonstration purposes, these dynamic windows can be quickly
prototyped and made to operate in a realistic manner.
Your next step is to show your mock-up panels to the users and col-
lect their comments. If there are only one or two people who need to
see the demonstration, just gather them around your computer and do
it live. Animation really helps the users visualize the end result. Also,
you can edit the panels while they watch (see Figure 8.5).
If the audience is bigger than your office can handle, try using a
projection screen for your computer down in the conference room. LCD
panels and red-green-blue (RGB) projectors are widely available and
offer reasonable quality. That way, you can still do the live demonstra-
tion. As a backup, take screen shots of the LabVIEW panels that you
 Building an Application 195

Vacuum Brazing Lab—Furnace Programming and Monitors

2.0 SHUTDOWN

1.8 Control Status

1.6 Furnace 1 OFF

1.4 Furnace 2 Heating

Furnace 3 Soaking
1.2
Furnace 4 Cooling
1.0
Furnace 5 Manual
0.8
Temperature Setpoint
0.6
0.68 0. 0.0
0.4 0.56 1. 0.0

0.2 0.39 2. 0.0

0.79 3. 0.0
0.0
23 28 0.93 4. 0.0

Figure 8.5 Can you make that graph a little taller? “No sweat,” I replied, while calmly resizing
the graph.

can edit in a draw/paint application and add descriptive notes. Besides

making a nice handout, hard copies are especially useful for projects
that require quality assurance plans and formal design reviews. You’ve
got to have things in print, you know.
The Vacuum Brazing Lab had only a few things to display during normal
operation: the five furnace temperatures, their setpoints, and their status
(heating, soaking, cooling, etc.). That made one nice front panel with a
big strip chart and a few other indicators. Next, I needed a panel through
which Larry could edit the ramp-and-soak temperature recipe for each
furnace—an ideal application for a dynamic window. Finally, there had
to be a panel for manual control of the furnace. That panel needed such
controls as auto/manual, recipe start/stop, and manual setpoint entry. It,
too, was to be a dynamic window that appeared whenever a button on
the main panel was pressed. I created rough versions of the three panels,
made some dummy connections that animated it all, and showed it to
Larry.

Iterate on your preliminary panel design. Big projects may require

several demonstrations to satisfy the more critical users. This is espe-
cially true if you have collected additional requirements because of
the impact of your presentation. People start seeing alternatives they
previously had not imagined. Take notes during the demonstrations
and add to your requirements list. Discuss the new ideas and require-
ments with your cohorts, and try to come up with likely programming
196 Chapter Eight

solutions to the more difficult problems. Remember to keep the users

in the loop during all phases of the review process. They will be the
ultimate judges of your work, and you certainly don’t want to surprise
them on the day of final delivery.

First Design and Then Write Your Program

Don’t be stymied by the “blank screen syndrome.” The same effect
occurs in many fields when it’s time to create something: the writer
with a blank sheet of paper or the painter with an empty canvas. Pro-
totyping the user interface takes care of the panel, but the diagram
starts as a big collection of unconnected controls and indicators. Where
should you start?
Modern software engineering starts a project in the same way as any
other engineering discipline: with a design—a plan of attack. You must
do the same. The alternative is to sit in front of your computer, hoping
for divine inspiration to somehow guide you toward the optimum pro-
gram. Most likely, you’ll end up with another of those little test pro-
grams that balloon into full-featured, unmaintainable disasters.
Instead, let’s try using some design practices that work well in the
LabVIEW environment. See LabVIEW Power Programming (Johnson
1998) for detailed discussions of software engineering in LabVIEW.
In case you’re wondering why designing and writing the program
are wrapped into this one section instead of being separated, there is
a good reason: Most LabVIEW VIs run nicely on their own, with no
main program. This is an important LabVIEW advantage that is often
overlooked. Once you get past the most general stages of design, you
start writing subVIs that beg to be run, tested, and optimized as you go.
There is no need to write elaborate test programs, no need to wait until
the entire application is assembled, complete with a bunch of buried
bugs.
The ability to do rapid prototyping in LabVIEW is one of the major
reasons for the product’s success. You can quickly throw together a test
program that improves your understanding of the underlying hard-
ware and the ways in which LabVIEW behaves in your particular situ-
ation. These prototype programs are extremely valuable in the overall
development process because they allow you to test important hypoth-
eses and assumptions before committing yourself to a certain approach.
Your first try at solving a problem may be ugly, but functional; it’s the
later iterations of this initial program that you actually deliver. This
concept of iterative design is often the only practical approach to
solving laboratory problems where the requirements change as fast
as you can modify the program. If you can’t seem to arrive at a con-
crete list of requirements, just do the best you can to design a flexible
 Building an Application 197

application that you can update and extend on short notice. Many of
the architectures shown in Chapter 3, “Controlling Program Flow,” are
based on such practical needs and are indeed flexible and extensible.
If you’re an experienced programmer, be sure to use your hard-
learned experience. After all, LabVIEW is a programming language. All
the sophisticated tools of structured analysis can be applied (if you’re
comfortable with that sort of thing), or you can just draw pictures and
block diagrams until the functional requirements seem to match. This
world is too full of overly rigid people and theorists producing more
heat than light; let’s get the job done!

Ask a Wizard
To help you get started building data acquisition applications, National
Instruments added a DAQ Assistant to LabVIEW. The Wizard leads
you through a list of questions regarding your hardware and applica-
tion in the same way you might interview your customer. Then it places
a DAQ solution that is ready to run. From that point, you continue by
customizing the user interface and adding special features. As a design
assistant, it’s particularly valuable for DAQ-based applications, but it
can also suggest basic program designs that are useful for ordinary
applications. As time goes on, LabVIEW Wizards will expand to encom-
pass a broader range of programs.

Top-down or bottom-up?
There are two classic ways to attack a programming problem: from the
top down and from the bottom up. And everyone has an opinion—an
opinion that may well change from one project to the next—on which
way is better.
Top-down structured design begins with the big picture: “I’m going
to control this airplane; I’ll start with the cockpit layout.” When you
have a big project to tackle, top-down works most naturally. LabVIEW
has a big advantage over other languages when it comes to top-down
design: It’s easy to start with the final user interface and then animate
it. A top-down LabVIEW design implies the creation of dummy subVIs,
each with a definite purpose and interrelationship to adjacent subVIs,
callers, and callees, but at first without any programming. You define
the front panel objects, the data types, and the connector pane and its
terminal layout. When the hierarchy is complete, then you start filling
in the code. This method offers great flexibility when you have teams
of programmers because you tend to break the project into parts (VIs)
that you can assign to different people.
Bottom-up structured design begins by solving those difficult low-
level bit manipulation, number-crunching, and timing problems right
198 Chapter Eight

from the start. Writing an instrument driver tends to be this way. You
can’t do anything until you know how to pass messages back and forth
to the instrument, and that implies programming at the lowest level
as step 1. Each of these lower-level subVIs can be written in complete
and final form and tested as a stand-alone program. Your only other
concern is that the right kind of inputs and outputs be available to link
with the calling VIs.
Keeping track of inputs and outputs implies the creation of a data
dictionary, a hierarchical table in which you list important data types
that need to be passed among various subVIs. Data dictionaries are
mandatory when you are using formal software design methods, but
are quite useful on any project that involves more than a few VIs. It
is easy to maintain a data dictionary as a word processing document
that is always open in the background while you are doing the devel-
opment. A spreadsheet or database might be even better; if you use
a computer-aided software engineering (CASE) tool, it will of course
include a data dictionary utility. List the controls and indicators by
name, thus keeping their names consistent throughout the VI hier-
archy. You might even paste in screen shots of the actual front panel
items as a reminder of what they contain. This is also a chance to get
a head start on documentation, which is always easier to write at the
moment you’re involved with the nuts and bolts of the problem. Many
items that make it to the dictionary are saved as typedefs or strict
typedefs that you define in the Control Editor. This saves countless
hours of editing when, for instance, a cluster needs just one more
element.
Don’t be afraid to use both top-down and bottom-up techniques at
once, thus ending up in the middle, if that feels right. Indeed, you need
to know something about the I/O hardware and how the drivers work
before you can possibly link the raw data to the final data display.

Modularity
Break the problem into modular pieces that you can understand. This
is the divide-and-conquer technique. The modules in this case are
subVIs. Each subVI handles a specific task—a function or operation
that needs to be performed. Link all the tasks together, and an appli-
cation is born, as shown in Figure 8.6.
One of the tricks lies in knowing when to create a subVI. Don’t just
lasso a big chunk of a diagram and stick it in a subVI because you ran
out of space; that only proves a lack of forethought. Wouldn’t you know,
LabVIEW makes it easy to do: select part of your diagram, choose Cre-
ate SubVI from the Edit menu, and poof! Instant subVI, with labeled
controls and connector pane, all wired into place. Regardless of how
 Building an Application 199

Task Level Task Type

Highest Application Vacuum Brazing Main VI

Major Task Ramp and Soak Control

Eurotherm Driver-
Driver Write Operation

Serial Port Write

Lowest Built-In Function

Figure 8.6 Using modularity in your program design makes

it much easier to understand. This is one slice through the
VI hierarchy of the VBL project.

easy this may be, you should instead always think in terms of tasks.
Design and develop each task as if it were a stand-alone application.
That makes it easier to test and promotes reuse in other problems.
Each task, in turn, is made up of smaller tasks—the essence of top-
down hierarchical design.
Any properly designed task has a clear purpose. Think of a one-
sentence thesis statement that clearly summarizes the purpose of the
subVI: “This VI loads data from a series of transient recorders and
places the data in an output array.” (A good place to put this thesis
statement is in the VI Documentation dialog.) If you can’t write a simple
statement like that, you may be creating a catchall subVI. Include the
inputs and outputs and the desired behavior in your thesis statement,
and you have a VI specification. This enables a group of programmers
to work on a project while allowing each programmer to write to the
specification using her or his own style. As an example of how power-
ful this is, a recent coding challenge posted to the LabVIEW Developer
Zone (www.ni.com) had more than 100 submissions; although each
submission was different, they all did essentially the same thing. More
than 100 programmers, and all their code was interchangeable because
they all wrote to the same specification. Not only does it make team
development possible, but also maintenance is easier.
Consider the reusability of the subVIs you create. Can the function be
used in several locations in your program? If so, you definitely have a
reusable module, saving disk space and memory. If the subVI require-
ments are almost identical in several locations, it’s probably worth
writing it in such a way that it becomes a universal solution—perhaps
it just needs a mode control. On the other hand, excessive modularity
can lead to inefficiency because each subVI adds calling overhead at
200 Chapter Eight

execution time (a fraction of a microsecond per VI) and takes up more

space on disk.
There are additional advantages to writing a modular program. First,
a simple diagram is easier to understand. Other programmers will be
able to figure out a diagram that has a few subVIs much more read-
ily than a complex diagram with level upon level of nested loops and
Sequence structures. This is like writing a 10,000-line main program in
a procedural language. Second, it is generally easier to modify a modu-
lar program. For instance, you might want to change from one brand
of digital voltmeter to another. By incorporating well-written, modular
drivers, you would just substitute brand X for brand Y and be running
again in minutes. One goal in LabVIEW programming is to make your
diagrams fit on a standard screen (whatever size is most common for
computers like yours). This forces you to design and lay out diagrams
in a thoughtful manner. Another goal is to limit the diagram memory
usage of a subVI to about 500K. Use the VI Properties command from
the File menu to see how big each component of a VI is. These are not
strict rules, however. Main VIs with complex user interfaces, especially
those with many Property nodes, invariably take up much screen space
and memory.

Choose an architecture: Design patterns

Your initial objective is to decide on an overall architecture, or pro-
gramming strategy, which determines how you want your applica-
tion to work in the broadest sense. LabVIEW has a few of the most
common programming models, known as design patterns, built in
as templates accessible from the File >> New . . . dialog. (Figure 8.7)
A design pattern represents the fundamental structure of common
LabVIEW applications. One benefit of using a design pattern is that
the architecture is already proved to work—there is no need to rein-
vent the wheel. Plus, when you see a design pattern you’ve used before
in another application, you immediately recognize and understand it.
The list of design patterns continues to grow as the years go by. Choos-
ing one is a mix of analysis, intuition, and experience. In Chapter 3,
“Controlling Program Flow,” we discussed several design patterns.
Study them, and see if your next application doesn’t go together a little
more smoothly. Although design patterns lay out a common program-
ming structure you can start from, they are not intended to be com-
plete applications. The one aspect missing from almost all the design
patterns is proper error handling. They only show you how to use one
specific error to terminate parallel loops. They don’t even include an
error handler like Simple Error Handler.VI on the diagram! See the
“Handling Errors” section later in this chapter for techniques you can
use to handle errors.
 Building an Application 201

Figure 8.7 LabVIEW design patterns provide a familiar starting point for many applica-
tions. You can access design patterns from the File >> New . . . dialog.

The VI hierarchy as a design tool

A good way to start mapping out your application is to sketch a prelimi-
nary VI hierarchy. To see what this might look like, open a LabVIEW
example VI and select Show VI Hierarchy from the Browse menu.
You don’t need to start off with lots of detail. Just scribble out a map
showing the way you would like things to flow and the likely interac-
tions between various subVIs, as in Figure 8.8. Modular decomposition
in many languages looks like this figure. There’s a main program at
the top, several major tasks in the middle level, and system support or

MAIN
PANEL

MANUAL RAMP & SOAK NOTIFY STORE

INITIALIZE SHUTDOWN
CONTROL CONTROLLER USER DATA

CREATE CHANGE EUROTHERM WRITE

FILE CONFIG. DRIVER FILE

Figure 8.8 A preliminary VI hierarchy for VBL. At this point, I

really didn’t know what was going to happen inside each subVI,
but at least I had an idea about the division of labor.
202 Chapter Eight

I/O drivers at the lowest level. You could also draw this as a nested list,
much as file systems are often diagrammed.
Each node or item in the hierarchy represents a task, which then
becomes a subVI. Use your list of functional requirements as a check-
list to see that each feature has a home somewhere in the hierarchy.
As you add features, you can keep adding nodes to this main sketch, or
make separate sketches for each major task. Modularity again! There
is no need to worry about the programming details inside a low-level
task; just make sure you know what it needs to do in a general way.
Remember that the hierarchy is a design tool that should be referred
to and updated continuously as you write your programs.
The ramp-and-soak controller (one of the major subVIs in VBL) looked
simple enough at first, but turned out to be one of the trickiest routines
I’ve ever written. Without a healthy respect for modularity, I would have
had a much more difficult time. The problems stemmed from interactions
between various modes of operation, such as manual control, ramping, and
soaking. My hierarchy sketch was a real mess by the time I was through,
but it was the only easy way to track those modes and interactions.

Sketching program structure

There is a really powerful, visual tool to design LabVIEW programs.
It’s called LabVIEW sketching, and it requires sophisticated hard-
ware: a pencil and paper. The idea is to “think in LabVIEW” and put
your conceptual program design on paper. You can work at any level
of detail. At first, you’ll be attacking the main VI, trying to make the
major steps happen in the right order. Later on, you will need to figure
out exactly where the data comes from, how it needs to be processed,
and where it needs to go. Representing this information graphically
is a powerful notion. What’s even better is that you can implement
these sketches directly as LabVIEW diagrams without an intermedi-
ate translation step. Figure 8.9 shows Gary’s first sketch of the VBL
application.
You can also do sketching on your computer with LabVIEW. This has
the advantage of permitting you to try out programming techniques
that might simplify the problem. Then your test diagram can be pasted
right into the real VI. I do this a lot when I’m not sure how a built-in
function works. I hook it into a simple test VI and fiddle with it until
I really understand. It’s a much better approach than to bury a mys-
terious function deep inside an important VI, then to figure out why
things aren’t working later on. This scheme works great with top-down
design where you have stub VIs without diagrams. You simply paste
the final results of your experimental development into the diagram
of the stub VI.
 Building an Application 203

Figure 8.9 Sketch for the main VI used in VBL. Not very
detailed at this point, but it sure gets the point across.

Show your sketches to other LabVIEW users. Exchange ideas as

you do the design. It’s a vastly underused resource, this synergy thing.
Everyone has something to contribute—a new way of looking at the
problem, a trick she or he heard about, or sometimes a ready-to-use VI
that fills the bill. To quote from Brad Hedstrom of PE Biosystems, “If
you can’t draw it with a pencil and paper, how can you draw it in G?”

Pseudocoding
If you are an experienced programmer, you may find that it’s easier to
write some parts of your program in a procedural language or pseudo-
code. Of course, you should do this only if it comes naturally. Trans-
lating Pascalese to LabVIEW may or may not be obvious or efficient,
but Gary sometimes uses this technique instead of LabVIEW sketch-
ing when he has a really tough numerical algorithm to hammer out.
Figure 8.10 shows a likely mapping between some example pseudocode
and a LabVIEW data structure, and Figure 8.11 shows the associated
LabVIEW diagram that is a translation of the same code.
There’s something bothersome about this translation process. The
folks who thought up LabVIEW in the first place were trying to free
us from the necessity of writing procedural code with all its unforgiv-
ing syntactical rules. Normally, LabVIEW makes complicated things
simple, but sometimes it also makes simple things complicated. Mak-
ing an oscilloscope into a spectrum analyzer using virtual instruments
is really easy, but making a character-by-character string parser is a
204 Chapter Eight

Pseudo-code LabVIEW Data Structure

MODULES
0
ACTIVE
SLOT
FOR each module defined 5
IF module is active
THEN
FOR each channel
IF channel is active CHANNELS
THEN 0
Read_Data ( module, channel); NAME
Store data in output array; Channel Name
ENDIF
ENDFOR ACTIVE
ENDIF
ENDFOR

Figure 8.10 A mapping between a pseudocode segment and a LabVIEW data

structure.

real mess. In such cases, we choose the best tool for the job at hand and
be happy.
To design the difficult ramp-and-soak control logic in VBL, I found that
a pseudocode approach was easier. I didn’t have to look far to recover the
example that appears in Figure 8.12; it was pasted into a diagram string
constant in the subVI called Recipe Control Logic. I won’t bother to repro-
duce the entire VI diagram here because it has too many nested Case
structures to print in a reasonable space.

Ranges, coercion, and default values

Protect yourself—expect users to supply ridiculous inputs at every pos-
sible opportunity. If your program does not self-destruct under such
abuse, then it is said to be robust. Unfortunately, quick-and-dirty test

Pseudo-code G (LabVIEW) code

FOR each module defined
IF module is active For each active module, acquire data
THEN
True
FOR each channel
IF channel is active For each active channel, append
THEN Unbundle By Name data to output array
MODULES
Read_Data ( module, channel);
ACTIVE True
Store data in output array; ACTIVE
ENDIF CHANNELS
NAME DATA
ENDFOR Build Array
ENDIF
ENDFOR
Empty

Initializes Bundle
shift register

Figure 8.11 The same pseudocode translated into a LabVIEW diagram, which acts on
the data structure from the previous figure. If this is your cup of tea, by all means use
this approach.
 Building an Application 205

Psuedo-code for this VI

if setpoint==manual
freeze ramp timer
do manual updates
else
if recipe==pause
freeze setpoint
else
if step <= Nsteps
if slope > 0 && sp < target

Figure 8.12 If you do decide to use pseudocode to design your

program, include it right on the diagram by pasting it into a
String constant like this.

programs are rarely robust. Commercial applications such as word pro-

cessors are highly robust, lest the publisher go out of business.
You may need to put appreciable effort into bullet-proofing your
LabVIEW application against wayward users. On older versions of
LabVIEW you could use the Data Range . . . pop-up dialog to coerce
values into the desired range (minimum, maximum, and increment).
This was a great way to limit inputs to reasonable values, but there
was no way to tell if a control had Data Range limits set just by looking
at the control. You could get some really strange behavior and spend a
lot of time tracking it down to an invisible range coercion on a control
of a subVI. As a result, setting the Data Range is no longer allowed
for subVIs. Now you will need to use the In Range and Coerce
Function (Figure 8.13) that applies limits programmatically. If the
values are not evenly spaced (such as a 1-2-5 sequence), use a function
similar to the Range Finder VI shown in Figure 8.14.
Many GPIB instruments limit the permissible settings of one control
based upon the settings of another: A voltmeter might permit a range
setting of 2000 V for dc, but only 1000 V for ac. If the affected controls
(e.g., Range and Mode) reside in the same VI, put the interlock logic
there. If one or more of the controls are not readily available, you can
request the present settings from the instrument to make sure that
you don’t ask for an invalid combination.

In Range
upper limit 15.00 and Coerce
coerced value
Numeric
In Range?
lower limit 1.00

Figure 8.13 The In Range and Coerce Function

limits the maximum and minimum values of a
scalar.
206 Chapter Eight

number to find range found

next greater value
numeric array index
Range Finder.vi
numeric array range found Array Max
& Min range found
0 0.0000

Index Array
next greater value
number to find next greater value
numeric array
0.0000 0.0000
index
index
0 number to find Round to
Threshold
1D Array +Infinity

Figure 8.14 A Range Finder utility VI. This one looks through an array of permissible values and
finds the one that is greater than or equal to the value you wish to find. It returns that value and
the index into the number to find the array.

String controls and tables are particularly obnoxious because the user
can (and will) type just about anything into one. Think about the conse-
quences of this for every string in your program. You may need to apply
a filter of some type to remove or replace unacceptable characters. For
instance, other applications may not like a data file in which the name
of a signal has embedded blanks or nonalphanumeric characters.
One nefarious character is the carriage return or end-of-line (EOL)
character for your platform. Users sometimes hit the Return key rather
than the Enter key to complete the entry of text into a LabVIEW string
control, thus appending EOL to the string. Note that training users
will not solve the problem! You have two options. In the LabVIEW
Preferences, under Front Panel, there is an option to treat Return the
same as Enter in string controls. If selected, the two keys act the same
and no EOL character is stored. The other alternative is to write a
subVI to search for EOL and kill it. The Preferences method is nice, but
what happens when you install your VI on another machine? Will that
preference be properly set? And what about all the other preferences
that you so carefully tuned for your application? If a setting is very
important, you must explicitly update the preferences when installing
your VI.
Controls should have reasonable default values. It’s nice to have a VI
that does not fail when run as-opened with default values. After enter-
ing the values you desire, you select Make Current Value Default
from the control’s pop-up or from the Edit menu. Remember to show
the default value and units in parentheses in the control’s label so that
it appears in the Help window. It’s nice to know that the default value is
1000, and whether it’s milliseconds or seconds. But don’t make the data
in graphs and arrays into default values (unless it’s really required);
that just wastes disk space when the VI is saved.
 Building an Application 207

Make intelligent use of default values. In the case of a utility VI

Write Characters to File, the default path is empty, which forces a file
dialog. This can save the use of a boolean switch in many cases. With
numeric controls, you can assign a default value that is unlikely to be
supplied during normal use of the VI and then test for that value. A
large negative integer may make sense; or if your input is a floating-
point type, enter NaN (not a number), and make that the default.
In the Comparison functions, you can use Not A Number/Path/
Refnum to check for NaN. Similarly, empty strings can be tested with
the Empty String/Path function.

Handling errors
Another aspect of robust programs is the way they report run-time
errors. Commercial software contains an incredible amount of code
devoted to error handling, so don’t feel silly when one-half of your dia-
gram is devoted to testing values and generating error clusters or dia-
logs. Most problems occur during the initial setup of a system, but may
rarely occur thereafter. It may seem wasteful to have so much code
lying around that is used only 0.01 percent of the time. But it’s in the
start-up phase that users need the most information about what is
going wrong. Put in the error routines and leave them there. If you want,
add a boolean control to turn off dialog generation. Realize that dialog
boxes tie up the VI from which they are called until the user responds.
This is not a good mode of operation for an unattended system, but you
may still want the option of viewing dialogs during a troubleshooting
session.
There is a distinction between error handling and error recovery.
Error handling provides a graceful method to report an error condition,
and to allow the operator to intervene, without causing more errors. A
simple message “No scope at GPIB address 5” will go a long way toward
getting the problem solved. Error recovery involves programmatically
making decisions about the error and trying to correct it. Serious prob-
lems can occur when error recovery goes awry, so think carefully before
making your application so smart it “fixes” itself. The best applications
will report any and all error messages to guarantee system reliabil-
ity and data quality. Your job during the development phase is not
to ignore errors, but to eliminate the conditions that caused an error
when it crops up.
The simplest and recommended method to implement error handling
in your application is to use an error I/O cluster to connect one VI to
the next. The error I/O cluster works as a common dataflow thread, as
you have already seen in many examples in this book. The error cluster
enforces dataflow and allows VIs downstream to take appropriate action.
208 Chapter Eight

Figure 8.15 Test the error cluster inside each subVI. If there is an input error,
no work is performed by the subVI and the error is passed through without
modification. The Scan From String function does this internally, as do almost
all LabVIEW functions.

It doesn’t make sense to work on bad data, or send commands to an

unresponsive instrument. Figure 8.15 shows how to implement error
handling in a subVI. If there is an input error, the error cluster is
passed unmodified to the next VI in the chain. No work is performed
on any input error. Internally, this is the default behavior of almost all
LabVIEW functions and subVIs. Notable exceptions are functions that
close file I/O or instrument sessions. They always attempt to close the
I/O before passing along the error message.
At the end of every chain of I/O operations you should place an error
handler to alert users of any errors. Some common reasons for failure
in I/O operations are that

■ A system configuration error has been made. For instance, some

equipment has not been properly installed or connected. Also, with
plug-in boards, you must install the correct driver in your operating
system.
■ Improper parameters have been supplied, such as channel numbers
or file path names.
■ A programming error has occurred on the diagram. Typically, you
forget to wire an important parameter.
■ LabVIEW bugs crop up (incredible, but true).
 Building an Application 209

A good error handler tells the user the following:

■ Where the error came from
■ What might have caused the error
■ What actions he or she might take to fix and/or avoid recurrence
■ One or more appropriate options for escape (continue, stop, abort, etc.)
Errors should always be reported in plain language, although an error
code number is usually displayed in addition, in case a system guru
wants to dig deeper. Dialog boxes are a logical choice when the error is
severe enough that user intervention is required. However, avoid bury-
ing dialog boxes deep within the VI hierarchy. Dialogs are best placed
at the end of an error chain higher in the application layer.
Error handler VIs are included with LabVIEW in the Dialog and
User Interface functions. A favorite is called Simple Error Handler,
and it works for just about any purpose. You can feed it an error code
number or a standard error I/O cluster, and it will generate a formatted
message and an optional dialog.
Error I/O is essential in LabVIEW programming because it solves
one of those universal needs, and as such, the file I/O functions, data
acquisition library, and all the newer driver VIs use it. If the error han-
dlers supplied don’t suit your needs, use them as a foundation for your
own customized version. The General Error Handler VI permits you
to enter custom error codes and messages. For complex applications
you can add custom error codes between the range of 5000 and 9999
into LabVIEW with the Error Code File Editor (Tools >> Advanced
>> Edit Error Codes…). Custom error codes are saved in the labview\
user.lib\errors directory. Throughout the application hierarchy, you
can then generate you own error codes, and LabVIEW’s error handlers
will display your plain language message.
If an error condition is severe enough that execution of the VI must
not continue, then shut down the application gracefully. Aborting exe-
cution can be risky if any I/O activity is pending. Some kinds of buff-
ered data acquisition operations keep going, and analog and digital
outputs retain their last values. It may be difficult to clear up these
conditions. Expect the worst and provide a way to programmatically
exit on an error.
There is a fundamental flaw in the LabVIEW design patterns—the
error handling is incomplete. That is so because the design patterns are
illustrations to show how to pass messages and programmatically stop
one loop from another without using globals. Figure 8.16 shows the stock
producer/consumer (events) design pattern shipped with LabVIEW
8.0. Sometimes this example is called a queued message handler.
The top loop produces messages each time the event structure fires.
210 Chapter Eight

Figure 8.16 Error handling in producer/consumer design pattern. Any error terminates a loop. Is
that what you want in a robust application?

The messages are put into a queue (enqueued) by the producer and
removed from the queue (dequeued) by the bottom consumer loop.
Stopping the producer loop releases the queue and stops the consumer
loop when the dequeue function generates an error. If any subVIs in the
consumer loop were included in the error chain between the dequeue
function and the stop condition, it would programmatically alter the
behavior by forcing the loop to terminate on any error. The queued
message handler is a common design pattern for programming user-
interface applications; however, because there is only one-way commu-
nication between the loops, the producer loop will never know if the
consumer loop terminated—it will just go on putting messages into the
queue. Clearly this is not a robust solution!
One flexible error-handling scheme we’ve used, and seen used by
others, is based on the addition of an extra top-level error-handling
loop to which all error reports are passed via a global queue. Any VI
is permitted to deposit an error in the queue at any time, which causes
the error handler to wake up and evaluate that error. Figure 8.17
shows the top-level error-handling loop. The reference to Error Queue
is stored in ErrorQueue.vi (Figure 8.18). Any incoming error placed
into the queue is handled by the Simple Error Handler.vi. Figure 8.19
shows one way to put the code into action.
 Building an Application 211

Error loop handles all errors. The Error Queue reference is stored in
Figure 8.17
ErrorQueue.vi (Figure 8.18).

Figure 8.18 ErrorQueue.vi enqueues incoming errors into

the Error Queue. The queue reference is stored in a shift
register and remains valid until the queue is destroyed.

Figure 8.19 ErrorQueue.vi and the error-handling loop in action. All errors are
reported via the Error Queue.
212 Chapter Eight

Figure 8.20 The event structure responds to front panel events and programmatic events. Dynamic, user-defined
events pass errors from the consumer loop back to the event loop.

Reporting all the errors via a global queue has a lot of advantages,
but we have to add an extra dequeue loop and don’t have a way to
trigger the event loop. Figure 8.20 shows an advanced technique using
dynamic events to pass a user-defined event to the event loop. Using
a single event loop to listen to front panel user events and program-
matic user-defined events enables two-way communications between
the loops. Figure 8.21 shows the Error Event.VI. It is a functional
global that retains the reference to the dynamic event in a shift regis-
ter. Any incoming errors fire the user event and pass the error cluster
to the Simple Error Handler.vi in the Event Loop.
Whichever error-handling method you put into place needs to be
robust and flexible enough to grow with your application. Above all,
the requirements for your application should dictate what level of error
handling or error recovery you need to put in place. The methods we’ve
shown you are by no means the only way to do things. If they don’t
work for you, get creative and invent your own.
 Building an Application 213

Figure 8.21 Incoming errors fire the user-defined event, and the error cluster is
passed to the waiting Event structure.

VBL had to run reliably when unattended. Therefore, I had to avoid any
error dialogs that would lock up the system. When you call one of the
dialog subVIs, the calling VI must wait for the user to dismiss the dialog
before it can continue execution. That’s the reason you should never gen-
erate an error dialog from within a driver VI; do it at the top level only.
My solution was to add a switch to enable dialog generation only for test
purposes. If a communications error occurred during normal operation,
the driver would retry the I/O operation several times. If that failed, the
main program would simply go on as if nothing had happened. This was
a judgment call based on the criticality of the operation. I decided that no
harm would come to any of the furnaces because even the worst-case error
(total communications failure) would only result in a really long constant
temperature soak. This has proven to be the correct choice after several
years of continuous operation.

Putting it all together

Keep hammering away at your application, one VI at a time. Use all
the tools we have discussed for each step of the process. Remember that
each VI can and should be treated as a stand-alone program. Break it
down into pieces you understand (more subVIs) and tackle those one at
a time. Then build up the hierarchy, testing as you go. It’s a reliable path
to success. Never write your program as one huge VI—the LabVIEW
equivalent of a 40,000-line main program—because it will be very dif-
ficult to troubleshoot, let alone understand. Documenting as you go
can also save time in the long run because it’s easier to write about
what you’ve done while you’re doing it. The same is true for quality
214 Chapter Eight

assurance documents such as test procedures. If formal testing is

required, do a dry run of the formal testing as early as possible in the
development process. That means the procedure is complete and any
required test software is written and validated.
One of the LabVIEW inventors, Jack MacCrisken, after watching
many users go through the development process over the years, is a
firm believer in iterative design. His rule of thumb is that every appli-
cation should be written twice. The first time, it’s ugly, but it gets the
job done and establishes a viable hierarchical approach. The second
time, you sweep through and optimize, clean up, and document each
VI, as well as redesign any parts of your program that no longer suit
the overall requirements.
Check your user requirements document from time to time to make
sure that everything has been addressed. Keep the users in the loop
by demonstrating parts of your application as they are completed. It’s
amazing what an operator will say when shown a LabVIEW panel
that mimics real laboratory equipment. You need this feedback at all
stages of development as a form of reality check. In the real world of
science, requirements change. If the experiment succeeds, the require-
ments grow. If it produces unexpected results, requirements change.
And users change, too! You and your programs need to be flexible and
ready to accept change. An excellent article on this subject, written by
Ted Brunzie of JPL, appears in LabVIEW Technical Resource (1995).

Testing and Debugging Your Program

If you have designed your program in a modular fashion, testing will
almost take care of itself: You are able to test each subVI independently,
greatly increasing your chances of success. It’s comforting to know that
each module has been thoroughly thrashed and shows no signs of aber-
rant behavior. Nonmodular programming will come back to haunt you
because a big, complicated diagram is much harder to debug than a
collection of smaller, simpler ones.

Tracing execution
Many procedural languages contain a trace command or mode of oper-
ation, in which you can record a history of values for a given variable.
A string indicator on the panel of a VI will suffice when the message is
more of an informative nature. A handy technique is to use a functional
global with a reference to a queue to pass informative messages. This
is very useful when you’re not sure about the order in which events
occurred or values were computed. An idea Gary got from Dan Snider,
who wrote the Motion Toolbox for Parker Compumotor, is a trace VI
 Building an Application 215

that you can insert in your program to record important events. Val-
ues can be recorded in two ways: to disk or by appending to an array.
The VI can record values each time it’s called, as a simple datalogger
does, or it can watch for changes in a particular value and record only
the changes. This implementation is not perfect, however. If you have
a race condition on a diagram in which you are writing to a global
variable at two locations almost simultaneously, the trace VI may not
be quick enough to separate the two updates. This is where a built-in
LabVIEW trace command would save the day. Perhaps a future version
of LabVIEW will have this feature.
Figure 8.22 shows the connector pane and front panel of the Trace VI.
In this example, trace information may be appended to a file or to a
string displayed on the panel of the VI. Each trace message contains a
timestamp, the value being traced, an optional message string, and the
error code. A single numeric value is traced, but you can replace it with
whatever data you may have. Edit the diagram to change the format-
ting of the string as desired.
The When to log control determines when a trace record will be
logged to the file or string. Log always means a record will be saved
every time the VI is called. Log on change means only record data if

Message ("")
Mode Path in use
When to log Trace summary string
Value to trace error out
error in
Trace.vi

Mode Trace summary string

Append string 8:23:07 PM 2.000000E+0 My notes Err = 0

8:23:10 PM 3.000000E+0 My notes Err = 0
When to log 8:23:32 PM 4.000000E+0 My notes Err = 44

Log on change
Value to trace
4.00

Message ("")
My notes Path in use error out
RAM Disk:a
status code
error in
ERROR 44
status code
source
ERROR 44

source

Figure 8.22The Trace VI can help debug some applications by recording changes
in important variables. You can record to the string indicator on the panel or to a
file.
216 Chapter Eight

Value has changed since the last time the VI was called. The Mode
control determines what operation to perform. You can clear the string,
choose a new file, log to the string, or log to the file. If you are routing
trace information to the string, you may open the VI to display the
trace summary. We also include a mode called nothing, which makes
the VI do exactly that. Why include such a mode? So that you can leave
the Trace VI in place on a diagram but in a disabled condition.
This VI is usually wired into a diagram to monitor a specific variable.
You can also place it in an independent While Loop, perhaps in its own
top-level VI, and have it monitor a global variable for any changes. It’s
important to remember that the Trace VI takes some time to do its job.
Appending to the string starts out by taking a few milliseconds, but it
gradually slows down as the string grows. Logging to a file will also
take some milliseconds, but it’s a constant delay. Displaying the panel
while logging to the string can seriously degrade performance. You may
want to modify the VI for higher performance by stripping out all the
extra features you don’t need. In particular, to trace a single numeric
value, all you have to do is to see if its value has changed, then append
it to an array. That’s much faster than all the string handling, but it’s
not quite as informative.

Checking performance
During development, you should occasionally check the performance of
your application to make sure that it meets any specifications for speed
or response time. You should also check memory usage and make sure
that your customers will have sufficient memory and CPU speed for
the application. LabVIEW has a VI performance profiler that makes it
much easier to verify speed and memory usage for a hierarchy.
Select Profile VIs from the Tools >> Profile >> Performance and
Memory . . . menu to open the Profile Window and start a profiling
session. There are check boxes for selecting timing statistics, timing
details, and memory usage. It’s OK to turn everything on. Click the
Start button to enable the collection of performance data, and then
run your VI and use it in a normal manner. The Profile Window must
stay open in the background to collect data. At any time, you can click
the Snapshot button to view the current statistics. Statistics are pre-
sented in a large table, with one row for each VI. You can save this data
in a tab-delimited text file by clicking the Save button.
Scroll through the table and note the almost overwhelming bounty of
information. (See the LabVIEW user’s manual for details on the mean-
ing of each value.) Click on any column header to sort the information
by that statistic. For instance, you will probably want to know which
VIs use the most memory. If you see something surprising in the list,
 Building an Application 217

such as a subVI using a large amount of memory or taking up too much

time, it’s time to investigate. Note that the memory and timing values
are relative—the profiler itself uses some memory and CPU time—but
the comparisons are accurate.

Final Touches
If you follow the directions discussed here, you will end up with a well-
designed LabVIEW application that meets the specifications deter-
mined early in the process. A few final chores may remain.
Test and verify your program. First, compare your application with
the original specifications one more time. Make sure that all the needs
are fulfilled—controls and indicators are all in place, data files contain
the correct formats, throughput is acceptable, and so forth. Second, you
should abuse your program in a big way. Try pressing buttons in the
wrong order and entering ridiculous values in all controls (your users
will!). When data files are involved, go out and modify, delete, or other-
wise mangle the files without LabVIEW’s knowledge, just to see what
happens. Remember that error handling is the name of the game in
robust programming. If you are operating under the auspices of formal
software quality assurance, you will have to write a detailed valida-
tion and verification plan, followed by a report of the test results. A
phenomenal amount of paperwork will be generated. Formal software
quality assurance methods and software engineering are discussed in
detail in LabVIEW Power Programming (Johnson 1998).
Train your users. Depending on the complexity of the application,
this may require anything from a few minutes to several hours. Gary
once developed a big general-purpose data acquisition package and
had about 15 people to train to use it. The package had lots of features,
and most of the users were unfamiliar with LabVIEW. His strategy
was to personally train two people at a time in front of a live worksta-
tion. The user’s manual he wrote was the only other teaching aid. After
going through the basic operations section of the manual in order dur-
ing the training session, the students could see that the manual was
a good reference when they had questions later. It took about 2 h for
each training session, much of which consisted of demonstration, with
hands-on practice by the users at each major step. Simple reinforce-
ment exercises really help students to retain what you have taught
them. Everyone really appreciated these training sessions because
they helped them get a jump start in the use of the new system.
Last, but not least, make several backup copies of all your VIs and
associated documents. Hopefully, you have been making some kind of
backups for safety’s sake all along. It’s an important habit to get into
because system failures and foul-ups on your part should never become
218 Chapter Eight

major setbacks. A good practice is to keep the working copy on your

hard disk, a daily backup on another (removable) storage device, and
a safety copy on floppies or some other medium, perhaps a file server.
When you make up a deliverable package, you can put a set of master
floppies or CD-R in one of those clear plastic diskette holders in the
back of the user’s manual. It looks really “pro.”

VBL epilogue
It turns out that it took longer to get the equipment fabricated and installed
in Larry’s lab than it did for me to write the application. I did all the test-
ing in my office with a Eurotherm controller next to my Mac, including all
the demonstrations. Since there was plenty of time, I finished off the docu-
mentation and delivered a copy in advance—Larry was pleasantly sur-
prised. Training was spread out over several sessions, giving him plenty of
time to try out all the features. We tested the package thoroughly without
power applied to the furnaces for safety’s sake. Only a few final modifica-
tions and debugging sessions were needed before the package was ready
for real brazing runs. It’s been running constantly since the end of 1992
with excellent results. Larry is another happy LabVIEW customer.

Studying for the LabVIEW

Certification Exams
LabVIEW Certifications are valuable professional credentials that
validate your LabVIEW knowledge and programming skills. Attaining
certification ensures your employers and customers that you actually
know what you are doing. There are three levels of LabVIEW certifica-
tion: the basic Certified LabVIEW Associate Developer (CLAD),
next is the Certified LabVIEW Developer (CLD), followed by the
Certified LabVIEW Architect (CLA). Each level of certification
builds on the knowledge required by the previous level. The CLAD and
CLD examinations focus on LabVIEW programming skills, while the
CLA examination focuses on advanced application development and the
skills required to lead a team of LabVIEW developers. Originally the
CLAD and CLD were one examination with two parts: a 40-question
multiple-choice test for LabVIEW knowledge, followed by a timed
application development examination. Today there are two separate
examinations, and you have to pass the CLAD before you can register
for the CLD.

CLAD
The CLAD is a 1-h, computer-based multiple-choice examination admin-
istered by Pearson Vue. You can register for the CLAD online by following
 Building an Application 219

the links from National Instruments’ Web site (www.ni.com). Study

and apply the material we’ve highlighted in Chapters 2 through 9,
then take the sample examination National Instruments has online,
and you should be fine. You can find links to additional study material
online (www.ni.com). Since the CLAD is a multiple-choice computer-
based examination, you get your score immediately; however, you don’t
get any feedback on what questions you missed. But you shouldn’t have
any trouble if you understand LabVIEW and how the functions work.
Here’s a summary of the CLAD examination topics and some think-
ing questions to lead you down the path of enlightenment. If you look
through Chapters 2 through 9, you’ll see that we’ve covered these in
depth.

■ LabVIEW programming principles. Know how to use dataflow

and polymorphism in your VIs.
■ LabVIEW environment. Obviously if you’re going to be a LabVIEW
programmer, you have to know how to get around and where to find
things on menus and palettes. You also want to know the difference
between a front panel and a block diagram.
■ Data types. A LabVIEW programmer needs to know how LabVIEW
handles data: the different data types, representations, formats,
ranges, scaling, and coercion. What are type definitions and strict
type definitions? Know where they are used and how to create them.
What is a Data structure?
■ Software constructs. Know the LabVIEW programming struc-
tures and how they differ. What is a shift register and what are some
of the ways it can be used? Know which structures index on their
boundary and how they differ in function and performance. What are
tunnels and sequence locals?
■ Graphical-User Interface (GUI) elements. Know the different
charts and graphs, their display properties, and how to format data
for each one. Understand the mechanical actions of booleans. Know
when, where, and how to use Property nodes.
■ Variables and functions. Know the difference between local and
global variables, and understand race conditions. Know the LabVIEW
timing functions, file I/O functions, and how to work with binary and
text files. What is a cluster and how can you bundle or unbundle
a single specific element? Understand how LabVIEW works with
arrays on loop boundaries and how the array functions operate. How
do you pull a single element out of an array? What if it’s a 2D array?
How do you slice out a single row or column? Know the parts of
LabVIEW’s waveform data type and how to access them.
220 Chapter Eight

■ Simple design patterns. Understand LabVIEW’s multitasking

environment and basic design patterns. How does a Case structure
work? What is a state machine?
■ SubVI design. Know about subVI connector panes, icons, and VI
configuration options. How do you make connections required or
optional, and how can you distinguish between them in the Context
Help window?
■ VI design and documentation. Know how to build VIs and appli-
cations according to the LabVIEW Style Guide and the LabVIEW
Development Guidelines. Know how to create VI and control descrip-
tions that show up in the Context Help window. How do you create
tip strips? What are the best practices for documenting VIs?
■ Error handling. Know how error I/O passes error information
between VIs, how a subVI should respond to an input error, and the
proper terminals for error I/O.
■ Debugging tools and techniques. How do LabVIEW’s debugging
tools work and what are some best practices?

That’s a lot of stuff to know, but it’s basic LabVIEW knowledge you
should have if you are going to call yourself a LabVIEW programmer.
The test itself is not too hard. The questions are designed to test your
knowledge, not to confuse or trick you. Here is our own example of one
type of programming problem and question found on the CLAD.
What are the values in Array after the For Loop in Figure 8.23 has
finished?

A) 1, 2, 3, 4, 5
B) 0, 1, 2, 3, 4
C) 1, 3, 6, 10, 15
D) 0, 1, 3, 6, 10

Figure 8.23 This VI represents

one type of program analysis
problem you may encounter on
the CLAD. What are the values
in Array after the For Loop has
finished?
 Building an Application 221

In our example you have to do a little addition to get the answer; but
if you understood For Loops, shift registers, and arrays, it was easy.
The rest of the examination is practical questions on the subjects we
covered in this book.

CLD
The Certified LabVIEW Developer examination is a 4-h test of your
LabVIEW application development skills. You take the examination
at a National Instruments–approved location on a PC with a current
version of LabVIEW installed. You are not allowed to use any toolkits
or outside LabVIEW code other than the templates and examples that
normally come with LabVIEW. All that you are given is a sealed enve-
lope with two things: a blank floppy for your completed application,
and a multipage document with application requirements and func-
tional description(s). And 240 min later you should have a completed
application that
■ Functions as specified. It is always important to have an application
do what the customer wants.
■ Conforms to LabVIEW coding style and documentation standards.
The LabVIEW Development Guidelines is available in the online
help if you need to refer to it.
■ Is created expressly for the examination using VIs and functions
available in LabVIEW.
■ Is hierarchical. All major functions should be performed in subVIs.
■ Uses a state machine. Your state machine should use a type-defined
enumerated control, a queue, or an Event structure for state
management.
■ Is easily scalable to more states and/or features without having to
manually update the hierarchy. Think typedef enum here.
■ Minimizes the use of excessive structures, variables (locals/globals),
and Property nodes.
■ Responds to front panel controls (within 100 ms) and does not use
100 percent of the CPU cycles.
■ Closes all opened references and handles.
■ Is well documented and includes
■ Labels on appropriate wires within the main VI and subVIs
■ Descriptions for each algorithm
■ Documentation in VI Properties >> Documentation for the main
VI and subVIs
222 Chapter Eight

■ Tip strip and descriptions for front panel controls and indicators
■ Labels on all constants

The CLD examination is graded on three criteria:

■ Functionality. Does the application do what was asked? Is the logic
correct? Does the VI respond to front panel events within the time
limit? Does the application produce errors? Is file I/O implemented
correctly?
■ Style. Is the application within professional LabVIEW programming
guidelines? Is it neat? Is it scalable and maintainable? Does it make
excessive use of locals, globals, and Property nodes? Does it use all
the CPU time?
■ Documentation. Place labels on long wires. Algorithms should be
documented with a text label on the block diagram. VIs should have
descriptions. Front panel controls and indicators should have tip
strips and descriptions. VIs should have descriptive icons. All con-
trols, indicators, and constants should have descriptive names.

There is a total of 40 points. Functionality counts 15 points, style

counts 15 points, and documentation counts 10 points. You pass if you
score 75 percent or more.
You have a lot to do in 4 h; if you’re the kind of programmer who likes
to go away and think about a problem before coming up with an elegant
solution, you may find yourself running out of time. Sometimes simple
solutions turn out to be the best. One function that we think you’ll find
handy in many situations is the Elapsed Time Express VI (Figure 8.24)
available from the Programming >> Timing palette. The first time you
call it, you will need to Reset it and give it a Time Target (s). The next
iteration it will return Elapsed Time in seconds and a boolean indi-
cating whether Time Has Elapsed. We recommend spending some
time with this VI and getting to know it well. One idiosyncrasy of this
function is caused by the internal use of the First Call? primitive to
reset itself. Attempting to iteratively develop and debug a subVI built

Figure 8.24 The Elapsed Time Express VI is a handy

function in many situations.
 Building an Application 223

around this function can drive you nuts because Elapsed Time will
reset itself each time you push the subVI’s Run button.
Now that you know what to expect, we’re going to give you three
sample examinations to work through. These are retired CLD exams,
so they are what you can really expect to see when you take the exami-
nation. Remember, you only have 4 h from the time you see the exami-
nation to deliver a documented, scalable, working application that is
data-flow-driven and doesn’t use a ton of locals or Property nodes. Are
you ready? On your mark, get set, Go!

Example 1: Traffic light controller

Task: Your company has received a contract to produce a traffic light

controller for a local city government. The purpose of the traffic light
controller is to maintain traffic at a heavily used 4-way intersection.
You have been assigned to create the software for this system using
LabVIEW. You will create the software for one intersection that includes
north, south, east, and west bound traffic, with each direction having a
left-hand turn lane (Figure 8.25).

Figure 8.25 Front panel for the traffic light controller.

224 Chapter Eight

TABLE 8.1 Traffic Light Pattern with No Left-Turn Sequence

Northbound Southbound Eastbound Westbound
Green Green Red Red
Yellow Yellow Red Red
Red Red Green Green
Red Red Yellow Yellow

General operation. The traffic light controller provides for green, yel-
low, and red lights for each direction. In addition, a second green-
yellow-red light must be provided for the left-hand turn lane.
The periods of time the lights should be on are as follows:

Red: 4 s
Yellow: 2 s
Green: 4 s

The pattern of the lights with no left-turn sequence should be accord-

ing to Table 8.1.
The pattern of lights with a left-turn sequence for an eastbound
vehicle should be according to Table 8.2.
The left-turn sequence should be inserted only if a vehicle is posi-
tioned over the left-turn lane sensor at the time the left-turn sequence
should start.

Example 2: Car wash controller

Task. Your company has received a contract to design a controller

for a touchless car wash. A touchless car wash is one that relies on
water spray only—there are no brushes. The specifications and rules
are listed below. Your job is to design a controller using LabVIEW that
satisfies all specifications.

Instructions. Using a front panel similar to the graphic provided in

Figure 8.26, create a LabVIEW application that implements the actions

TABLE 8.2 Traffic Light Pattern with an Eastbound Turn Sequence

Northbound Southbound Eastbound Westbound

Green Green Red Red
Yellow Yellow Red Red
Red Red Left Green Left Green
Red Red Left Yellow Left Yellow
Red Red Green Green
Red Red Yellow Yellow
 Building an Application 225

Figure 8.26 Front panel for the car wash controller.

of the procedure. For this application, the car wash switches are simu-
lated by switches on the front panel. The car wash starts when a pur-
chase switch has been selected. The execution of a cycle is denoted
by illuminating the appropriate light-emitting diode (LED). For this
application, each cycle’s duration should be set to 5 s unless otherwise
stated in the Operating Rules below.

Car wash specifications

Purchase options
■ $5 Deluxe wash
■ $3 Economy wash
Cycles and associated timing:
1. Underbody wash, 10 s
2. Soap application, 5 s
3. Main high-pressure wash, 5 s
4. Clear water rinse, 5 s
5. Air dry cycle, 10 s

Car wash operating rules

1. Only one purchase selection can be made at a time. The car wash
should not accept a second purchase selection while the car wash is
in operation.
226 Chapter Eight

2. Not all cycles are performed for each wash. The more expensive
wash performs more cycles. The cycles performed include
■ Cycles performed for the deluxe wash: 1 - 2 - 3 - 4 - 5
■ Cycles performed for the economy wash: 2 - 3 - 4

3. Each cycle is initiated by a switch. If the vehicle rolls off of the switch,
the wash immediately pauses and illuminates an indication to the
driver to reposition the vehicle. The amount of time that expires while
the car is out of position should not count against the wash time.
4. Underbody wash cycle: The spray heads for this wash are located
near the entrance of the car wash. The underbody spray heads are
fixed in position, and they require the vehicle to slowly drive over
them to wash the underbody of the vehicle. The underbody wash is
activated under the following conditions:

a. The deluxe wash has been purchased.

b. The car wash is in the underbody wash cycle.
c. Under Body Wash Position Switch is closed (proximity switch).
This cycle of the wash should last for 10 s. Upon completion of
this cycle the controller should signal moving to the next cycle by
activating the Vehicle Out of Position LED.

5. Main wash cycle: Main Wash Position Switch verifies the vehicle is in
the correct location for the wash cycles (cycles 2, 3, and 4) to operate.
Each cycle should last for 5 s. If the vehicle rolls off of the Main Wash
Position Switch, the wash immediately pauses and illuminates an
indication to the driver to reposition the vehicle. The amount of time
that expires while the car is out of position should not count against
the wash time. The wash resumes after the vehicle is properly posi-
tioned. Upon completion of this cycle the controller should signal mov-
ing to the next cycle by activating the Vehicle Out of Position LED.
6. Air dry cycle: The air drier is a set of fixed-position blowers located
near the exit of the car wash. They require the vehicle to drive slowly
out of the car wash through the airstream to dry the vehicle. The air
dry cycle activates on the following conditions:

a. The deluxe wash has been purchased.

b. The car wash has reached the air dry cycle.
c. Air Dry Position Switch is closed (proximity switch).

If the vehicle rolls off of the Air Dry Position Switch, the wash
immediately pauses and illuminates an indication to the driver to
 Building an Application 227

reposition the vehicle. The amount of time that expires while the car
is out of position should not count against the air dry time. The wash
resumes after the vehicle is properly positioned. This cycle of the
wash should last for 10 s. Upon completion of this cycle the controller
should allow the next vehicle in line to select another wash.
7. The car wash must respond to the STOP boolean and Vehicle Posi-
tion Switches within 100 ms. The STOP boolean aborts the opera-
tion of the VI.

Example 3: Security system

Task. Your company has received a contract to produce a security

system for a local manufacturer. The purpose of the security system
is to monitor the perimeter of the manufacturing complex to prevent
unauthorized personnel from entering the grounds. Your job is to
design a security system using LabVIEW that satisfies the provided
specifications.

Instructions. Using a front panel similar to the graphic provided,

create a LabVIEW application that implements the actions of the
security system (Figure 8.27). You will create the software for six

Figure 8.27 Front panel for the security system.

228 Chapter Eight

perimeter zones, with the actual hardware being simulated by boolean

switches on the front panel. The application should be easily scalable to
add more zones without having to manually update the hierarchy.

Definitions
Zone: A perimeter area with its own security status indication.
Alarm: A zone condition indicating an intrusion into a zone.
Bypass: A state in which a zone will not indicate an alarm condition.
Placing a zone in bypass prevents nuisance alarms when mainte-
nance work is performed in the area near the perimeter fence, or
when severe weather may cause false alarms.
Tamper: A condition where the wiring of a zone has been altered
in some way. All zone wiring is supervised. That is, if an individual
attempts to circumvent the system by altering the wiring, that zone
will indicate a tamper condition.

The security system software accepts

Description of controls/indicators.
boolean inputs, via front panel switches (one set for each zone) for the
following functions:
■ Alarm input
■ Tamper input
■ Bypass input

The security system provides for one indicator light for each zone.
The color of the light provides the status information for that zone. The
colors are

■ Green: normal
■ Red: alarm
■ Blue: bypass
■ Orange: tamper

Operation of the security system

System
■ A bypass input should always override an alarm input but should
not turn off an existing alarm condition. An alarm condition should
not be indicated while a zone is in a bypass condition.
■ A tamper input should always override both an alarm input and/
or a bypass input but should not turn off existing alarm and/or
 Building an Application 229

TABLE 8.3 Security System Log Format

Date Time Zone Status
XXXXXXX XXXXXXX XXXXXXX XXXXXXX
XXXXXXX XXXXXXX XXXXXXX XXXXXXX

bypass conditions. Alarms and bypass conditions should not be indi-

cated while a zone is in a tamper condition.
■ An existing condition should be indicated when an overriding input
is removed.

File logging. The security system should produce an ASCII disk file
in a spreadsheet-readable format. The data, when imported into a
spreadsheet, should be in the format shown in Table 8.3, where XXXX
represents logged data.
The Status field should contain a string indicating the condition of
the zone: normal, larm, Bypass, or Tamper.
The log file should be stored in a location relative to the location of
the security application and should be updated only when the status of
a zone changes and closes after every transaction.

Bibliography
Bonal, David: “Mastering the National Instruments Certification Exams,” NIWeek 2005,
National Instruments Corporation, 11500 N. Mopac Expressway, Austin, Tex., 2005.
Brunzie, Ted J.: “Aging Gracefully: Writing Software that Takes Changes in Stride,”
LabVIEW Technical Resource, vol. 3, no. 4, Fall 1995.
Fowler, Gregg: “Interactive Architectures Revisited,” LabVIEW Technical Resource,
vol. 4, no. 2, Spring 1996.
Gruggett, Lynda, “Getting Your Priorities Straight,” LabVIEW Technical Resource, vol. 1,
no. 2, Summer 1993. (Back issues are available from LTR Publishing.)
Johnson, Gary W. (Ed.): LabVIEW Power Programming, McGraw-Hill, New York, 1998.
Ritter, David: LabVIEW GUI, McGraw-Hill, New York, 2001.
Sample Exam, “Certified LabVIEW Developer Examination—Car Wash Controller,”
National Instruments Corporation, 11500 N. Mopac Expressway, Austin, Tex., 2004.
Sample Exam,“Certified LabVIEW Developer Examination—Security System,” National
Instruments Corporation, 11500 N. Mopac Expressway, Austin, Tex., 2004.
Sample Exam, “Certified LabVIEW Developer Examination—Traffic Light Controller,”
National Instruments Corporation, 11500 N. Mopac Expressway, Austin, Tex., 2004.
This page intentionally left blank
 Chapter

Documentation
9
Most programmers hold documentation in the same regard as root
canal surgery. Meanwhile, users hold documentation dear to their
hearts. The only resolution is for you, the LabVIEW programmer, to
make a concerted effort to get that vital documentation on disk and
perhaps on paper. This chapter describes some of the documentation
tools and formats that have been accepted in commercial LabVIEW
software.
The key to good documentation is to generate it as you go. For
instance, when you finish constructing a new subVI, fill in the Get
Info VI description item, described later. Then when you want to put
together a software maintenance document, you can just copy and paste
that information into your word processor. It’s much easier to explain
the function of a VI or control when you have just finished working on
it. Come back in a month, and you won’t remember any of the details.
If you want to write commercial-grade LabVIEW applications, you
will find that documentation is at least as important as having a well-
designed program. As a benchmark, budget 25 percent of your time on
a given contract for entering information into various parts of VIs and
the production of the final document.

VI Descriptions
The VI description in the Documentation dialog box from the VI
Properties is often a user’s only source of information about a VI. Think
about the way that you find out how someone else’s VIs work, or even
how the ones in the LabVIEW libraries work. Isn’t it nice when there’s

231
232 Chapter Nine

online help, so that you don’t have to dig out the manual? Important
items to include in the description are
■ An overview of the VI’s function, followed by as many details about
the operation of the VI as you can supply
■ Instructions for use
■ Description of inputs and outputs
■ List of global variables and files accessed
■ Author’s name and date

Information can be cut and pasted into and out of this window. If
you need to create a formal document, you can copy the contents of
this description into your document. This is how you manage to deliver
a really nice-looking document with each driver that you write with-
out doing lots of extra work. You just write the description as if you
were writing that formal document. You can also use the techniques
described in the following section on formal documents to extract these
comments automatically.
You can also display the VI description by
■ Selecting the Connector Pane and Description option when using the
Print command from the File menu
■ Showing the Context Help window and placing the wiring tool on a
subVI’s icon on the block diagram

Control Descriptions
Every control and indicator should have a description entered through
the Description and Tip pop-up menu. You can display this informa-
tion by showing the Context Help window and placing the cursor on
the control (Figure 9.1). You can also display it by placing the wiring

Figure 9.1 Control descriptions are really

helpful, but only if you remember to type them
in! Also note the availability of a tip strip.
 Documentation 233

tool on the control’s terminal on the block diagram. This is very handy
when you are starting to wire up a very complex VI with many front
panel items. The description should contain the following general infor-
mation, where applicable:

■ Purpose or function of the control or indicator

■ Limits on numeric values (We usually write Range: 0–128.)
■ Default values
■ Special interactions with other control settings

This is the most underused feature of LabVIEW. When confronted

with a new VI, a user typically has no alternative but to guess the func-
tion of each control and indicator. It is a welcome sight to have a brief
message show up in the LabVIEW Help window just by pointing at the
object. Get in the habit of entering a description as soon as you create
the object. Then if you copy the object to other VIs, the description fol-
lows along. It’s much easier than saving the documentation for later.
Remember to tell users about this feature.

Custom Online Help

LabVIEW has an online help system that replaces most of the manu-
als. You can add your own reference documents to this help system as
HTML or as compiled help files (*.chm). The documents could describe
the usage of individual VIs or serve as a user’s manual. To use compiled
help files, you need to create a document (it can contain both graphics
and text) and then use a specialized help compiler for your platform to
generate a source file in Windows Help format.
Once you have a help document compiled, you link it to a particu-
lar VI through the Documentation section of the VI Properties menu.
Particular keywords (Help Tags) in a help file can be linked as well.
That makes it easy for the user to search for a topic or VI name and to
jump directly to the proper part of your spiffy online help document.
Developers are encouraged to place their help files in the Help direc-
tory and to use the <helpdir>: pseudopath in VIs that reference help in
this directory.
When you enter a path in VI Setup’s Help Path, use the current plat-
form’s path conventions. As with other paths in LabVIEW, the path will
be formatted correctly when you move to another platform.
Additionally, you can place VIs and LLBs in the Help directory, and
top-level VIs will be automatically listed in the Help menu. When the
user selects a VI from the menu, it will be loaded (and if you configure
VI Setup to execute upon load, it will start executing immediately).
234 Chapter Nine

You can programmatically access the online help system through the
Control Online Help function from the Programming >> Dialog and
User Interface >> Help function palette. This function lets you open
and close the online help system, display the main Contents menu, or
search for a specific keyword or HTML file. It’s a way to make a Help
button on a LabVIEW panel do something more than display a dialog
box.

Documenting the Diagram

While the overall VI description provides adequate coverage of the pur-
pose and function of a VI, it leaves something to be desired when it
comes to explaining the diagram. There are many helpful items that
you can add to a diagram to make it easier to understand (assuming
that you have laid it out in a neat and orderly fashion). Many of these
techniques appear in the examples in this book.
■ Label every loop and every frame of Case and Sequence structures.
Place a free label (using the text tool) inside each frame that states
the objective of that frame, for example, “Read data from all config-
ured channels.” Sometimes it helps to use a larger font or set the
style to bold to make the label stand out.
■ Label important wires, especially when the wire is very long
or its source is somehow obscured. Make the foreground color
transparent (T) and the background color white; then place the label
right on the wire.
■ Use large, boldface key numbers located near important parts of the
diagram; then use a scrolling diagram string constant (located off
to the side or bottom of the diagram) to explain each key item. This
technique reduces clutter within the diagram but provides lots of
information nearby.
■ Add an index to Sequence structures. If you have a sequence with
more than two or three frames, create a list outside the structure
that helps the user quickly locate the desired frame.

VI History
During development, you can track changes to VIs through the History
Window, which you display by selecting the VI Revision History item
from the Edit menu. In the History Window, you can enter descriptions
of important changes to the VI and then add your comments to the his-
tory log. Revision numbers are kept for each VI, and you can reset the
revision number (which also erases all history entries) by clicking the
Reset button in the History Window.
 Documentation 235

Some kinds of information can be automatically added to this his-

torical record. From the LabVIEW Options dialog, you have access to
a number of settings associated with History logging. For instance,
you can have LabVIEW automatically add an entry every time the VI
is saved or prompt the user for an entry at that time. Settings made
through the Options dialog are the default settings for new VIs. You
can also customize the History settings for individual VIs through the
VI Properties menu.
History information can be printed though the Print command in
the File menu. In the Base and Full editions, this only prints the his-
tory for a single VI. In the Professional Edition, the Print command
extracts history information for all the VIs in a hierarchy or directory.
This capability is included with LabVIEW as a tool for formal soft-
ware development. There is great demand from professional LabVIEW
developers for greater capability in such areas.

Other Ways to Document

One foolproof way of getting important instructions to the user is to
place a block of text right on the front panel, where it can’t be missed.
Like a pilot’s checklist, a concise enumerated list of important steps
can be invaluable. You might even put a suggestion there that says,
“Select Get Info from the File menu for instructions.”
If a front panel has a large string indicator, have it do double duty by
putting instructions in it, then Make Current Value Default. Note that
you can type into an indicator as long as the VI is in edit mode.

Printing LabVIEW Panels and Diagrams

We don’t believe in printing LabVIEW panels and diagrams for the pur-
pose of basic documentation—and we’re not alone in our position. The
interactive, layered, hierarchical nature of LabVIEW is at best poorly
represented on paper. “But my Quality Assurance manager requires
printouts,” you say. Well, ours did, too, but we talked him into accepting
electronic files instead. What use is a big, unreadable stack of paper
anyway? The point is this: The LabVIEW paradigm is completely differ-
ent from procedural languages and should not be forced into that same
old mold. Nobody is going to stop you from printing out your code, but
please think before you kill another tree.
Despite our pleading, you may still have to print a few panels or
diagrams. Perhaps you need to show someone your work where there
is no computer running LabVIEW. Or maybe you’re writing a paper (or
a book!) about your LabVIEW project and you need images of various
screens. In those cases, you can and should print those images. Here
are some of the methods.
236 Chapter Nine

The Print Window command (from the File menu) does just that:
Whichever LabVIEW window is on top gets sent to the printer, using
the current page setup. Limited customization of the printout is avail-
able through Print Options in the VI Properties menu. In particular,
you can request scale-to-fit, which really helps if the panel is large.
When you print a front panel, note that all items on the panel will
be printed, including those you have scrolled out of the usual viewing
area. To avoid printing those items, you should hide them (choose Hide
Control by popping up on the control’s terminal).
The Print command is quite flexible, depending on the edition of
LabVIEW that you have. You can choose which parts of the VI you wish
to print (panel, diagram, connector pane, description, etc.) through the
feature-laden Print dialog box. Printouts are autoscaled (if desired),
which means that reasonably large panels and diagrams will fit on a
single sheet of paper.

Putting LabVIEW screen images

into other documents
There are several ways to electronically place an image of a LabVIEW
screen in another application:

■ Copy and paste items directly.

■ Use a screen-capture utility.
■ Print to rich text format (RTF) files with or without embedded images
(as BMP files).
■ Print to HTML files with PNG, JPEG, or GIF images.
■ Print to encapsulated PostScript (EPS) files.

The big decision is whether you care if the objects are bitmaps or
fully editable vector graphic objects. Bitmaps generally take up more
memory and may have fairly low resolution (say., 72 dots per inch),
but are generally resistant to corruption when printed. Vector graphics
(such as PICT, EPS, and EMF) are more versatile because you can edit
most of the objects in the image, but they sometimes fly apart when
displayed or printed in certain applications.
On the Macintosh and Windows, you can select objects from a
LabVIEW panel or diagram, copy them to the clipboard, and then
paste them into any application that displays PICT (Macintosh) or
EMF (Windows) objects, including word processors. If you paste them
into a drawing application, such as Canvas or CorelDraw, the objects
are accurately colored and fully editable. Diagrams look pretty good,
though you may lose some wire textures, such as the zigzag pattern for
 Documentation 237

strings and the dotted lines for booleans. This is surely the easiest way
to obtain a quality image from LabVIEW.
Screen captures are another quick way to obtain a bitmap image
of anything on your computer’s screen, including LabVIEW panels
and diagrams. Both Macintosh and Windows machines have a built-
in screen-capture capability. On the Macintosh, use command-shift-3,
which saves a bitmap image of the entire screen to a file on the start-up
disk, or command-control-shift-4 to obtain a cursor that you can use to
lasso a portion of the screen and copy it to the clipboard. On Windows,
the Print Screen key copies the complete screen image to the clipboard,
and ALT-Print Screen copies the top window (the one you clicked last)
to the clipboard. Alternatively, you can obtain one of several screen-
capture utilities that can grab a selected area of the screen and save it
to the clipboard or to a file. They’re available as commercial packages
or shareware. The nice thing about many of these utilities is that you
can capture pulled menus and cursors, which is really helpful for user
documentation.
LabVIEW’s Print command can produce a variety of panel and dia-
gram image types, along with all the other documentation text. If you
print to an RTF file, it’s easy to load into word processors. In RTF mode,
you can also ask LabVIEW to save the images external to the main
RTF file (they come out as a set of BMP files). Printing to HTML gives
you a browser-friendly form of documentation. Images are external
since HTML is intrinsically a text-only language, and you can choose
from JPEG, PNG, and GIF for the image files. Keep in mind that you
can use the VI Server features (as we already discussed in the Docu-
ment Directory VI) to automatically generate image files, albeit with
fewer choices in content and file type.
What if you’re creating a really formal document or presentation and
you want the very finest-quality images of your LabVIEW screens?
Here’s Gary’s recipe, using the PostScript printing feature of
LabVIEW. It works on any Macintosh and Windows systems with
certain printer drivers, such as the one for the HP LaserJet 4m; check
your driver to see if it has an EPS file generation feature. You must
have LabVIEW 4.01 or later and Adobe Illustrator 6.0 or later to edit
the resulting images. Here are the steps:
1. Use LabVIEW’s Print command with a custom setup to print what-
ever is desired. Your life will be simpler if you use the scale to fit
option to get everything on one page. Otherwise, objects will be clipped
or lost since EPS records only one page’s worth of information.
2. In the print dialog, print to a file. Macintosh lets you choose whether
the file is pure PostScript or EPS and whether it has a screen pre-
view. Make the file type EPS with a preview.
238 Chapter Nine

3. Open the resulting file with Illustrator 6.0 or later (prior versions
will fail to properly parse the EPS file). The panel and diagram are
fully editable, color PostScript objects. You will need to ungroup and
delete some of the surrounding boxes. Also, many items are masked
and use PostScript patterns. Be careful what you change or delete.
4. Save the file as Illustrator EPS for importing into a page layout pro-
gram. EPS files are absolutely, positively the most robust graphics
files when you’re dealing with complex compositions and you’re nit-
picky about the outcome. Note: If you send the file to a service bureau,
make sure you take along any strange fonts that you might have cho-
sen in LabVIEW, or have Illustrator embed them in the file.

While it takes several steps, the results are gorgeous when printed
at high resolution, such as 1200 dots per inch. This book is an example
of the results.

Writing Formal Documents

If you need to prepare a formal document describing your LabVIEW
project, try to follow the conventions described here. Our objective here
is to help propagate a consistent style of documentation so users can
more quickly understand what you have written. If you work for a big
company, it may have a style guide for technical writers that should
be consulted as well. General document style is beyond the scope of
this book; only the specifics of content and format as they pertain to VI
descriptions will be discussed.

Document outline
For an instrument driver or similar package, a document might consist
of the following basic elements:

1. Cover page
2. Table of contents
3. Introduction—“About This Package”
4. Programming notes—help the user apply the lower-level function
VIs
5. Using the demonstration VI
6. Detailed description of each lower-level function VI

The first five items are general in nature. Use your best judgment
as to what they should contain. The last item, the detailed description
of a particular VI, is the subject of the rest of this section. For major
 Documentation 239

applications, write a user’s manual and a programmer’s manual. In the

user’s manual, begin with a quick start section. Limit this to only a few
pages because most users have the attention span of a 2-year-old. In
that section, put in annotated screen shots that describe the important
controls and explain how to start up and shut down the system. Next,
go into detail about the various LabVIEW panels and modes of opera-
tion. In general, you should model your manuals after those of other
commercial packages that you judge to be effective.

Connector pane picture

Use the methods described under “Putting LabVIEW Screen Images
into Other Documents” to import an image of the connector pane into
your word processor, as shown in Figure 9.2. This gives the reader a
handy key to the VI’s connections, just as you find in the LabVIEW
manuals.
Note the different styles of text in the figure. This is achieved by set-
ting the required connections on the VI’s connector pane. While edit-
ing a VI and showing the connector pane, you pop up on a terminal point
and select This Connection Is. The choices available are Required,
Recommended, or Optional. Required terminals will automatically be
displayed in the Help window (and in printouts) in boldface.
These terminals must be wired when the VI is placed on a diagram,
or the calling VI will be broken. Recommended terminals (the default
setting) are displayed in normal typeface. Optional terminals are those
that the user will rarely have to access, and they are grayed out. In the
Context Help window, the user can select simple diagram help or
detailed diagram help. In the simple mode, optional terminals are
not shown. This simplifies the life of novice users.

VI description
If you have already entered a description of the VI in the Documenta-
tion box with the VI, open it, select all the text, and copy it to the clip-
board to import it into a document in your word processor.

Channel Numbers
DO IT (T) Multiplot Graph
Points in window text time
Scroll Last Values
Time Scale (1=minutes)
CircBuf to Multiplot Graph.vi
Figure 9.2 The connector pane is an important part of the
document for any VI. Different text styles can indicate
the relative importance of each terminal.
240 Chapter Nine

The Print command in the Professional Edition of LabVIEW can

extract all the text from a hierarchy. Just customize it to print only the
documentation, and send the results to an RTF or text file. If you don’t
have the Professional Edition, you can use the Document Directory VI
to do the same thing. Set it to Using Panel, and make the destination
RTF or Text. With either of these methods, you’ll also get all the control
and indicator descriptions. Happy editing. . . .
Names of input and output terminals should be in boldface. If your
word processor can do page layouts, the description can flow around
the connector pane picture. Here is the text that goes with the connec-
tor pane shown in Figure 9.2:
A handy multichannel plotter for use with the Circular Buffer trending VI.
Calls the Circular Buffer driver to return the desired quantity of data for
one or more channels. Displays data as a multiplot graph and returns the
last values for the specified channels. All channels share a common time
base. Supports all the features of the Circular Buffer VI. To see this VI in
action, run the Circular Buffer Example VI, then run this VI. Since the
Circular Buffer is a kind of global variable, its data can be shared between
these independent, top-level VIs.

Terminal descriptions
Every input and output needs a description that includes the following
information.
■ The data type
■ What it does
■ Valid range (for inputs)
■ Default value (for inputs)

This may be the same information that you entered in the control’s
description dialog box. A recommended standard for terminal labeling
makes use of text style to indicate how frequently you need to change
each item. When you create the VI, modify the font for each front panel
item’s label as follows: Use boldface for items that you will have to change
often, use the normal font for items that you plan to use less often, and
put square brackets [label] around the labels of items where the default
value usually suffices. This method maps directly to the required con-
nections behavior previously described. When you write the terminal
descriptions into a document, put them in this prioritized order so the
user will see the most-used items first. We like to start each line with a
picture of the appropriate LabVIEW terminal data type, followed by the
terminal’s name, then its description, range, and defaults:
Peak Voltage sets the maximum peak output voltage. Note
that the peak power setting for 2000 V cannot be greater than 50 W.
 Documentation 241

The possible values are as follows:

0: 16 V (Default)
1: 80 V
2: 400 V
3: 2000 V

Make a complete set of terminal icons that can be kept in a separate

document, open at the same time as your project document. Just copy
and paste the icons. Gary made such a document in Microsoft Word
which he calls “LabVIEW Document Helpers.” It also includes prefor-
matted document sections and styles that speed up and standardize
the work. If you can’t handle all this graphical stuff, at least show the
data type in text format. Here are some examples:

I32 I32 integer

[DBL] Array of DBL numerics
[[DBL]] 2D array of DBL numerics
cluster Cluster
ABC String

Try it out, and feel free to add your favorite documentation tidbits.

Programming examples
If you are producing a commercial package, it’s a good idea to include
some examples using your VIs. Show how the VI is used in a loop or
in conjunction with graphs or an analysis operation. Above all, have a
“try me first” example that is guaranteed to work! This will give the
user greater confidence that the rest of the package is similarly func-
tional. Put completed examples on disk where the interested user can
find them and try them out. Make prominent mention of the existence
of these examples in your documentation, perhaps in the quick start
section.

Distributing Documents
In this modern age, more and more documents are distributed in
electronic form. Except for commercially marketed software packages,
the user is expected to download some kind of document for personal
viewing and/or printing. Text files are of course the least-common-
denominator format, but they leave much to be desired: You can’t include
graphics at all. Word processors and page layout programs can create
nice documents, but they are not generally portable (although Microsoft
242 Chapter Nine

has done a good job of Mac-PC portability with Word, except for graph-
ics items). Besides text and word processor files, there are some good
formats available nowadays, thanks to the World Wide Web.
Portable Document Format (PDF) was created by Adobe Sys-
tems to provide cross-platform portability of elaborate documents. It
based PDF on PostScript technology, which supports embedded fonts
and graphics. You can buy Adobe Acrobat for almost any computer and
use it to translate documents from word processors and page layout
applications into this universal format. Users can obtain a free copy of
Adobe Acrobat Reader from online sources (Adobe Systems has it on
its Web site, www.adobe.com), or you can include it on your distribution
media. Again, Acrobat Reader is available for many platforms. This is
probably the best way to distribute high-quality documentation.
HyperText Markup Language (HTML) is the native format for
many items on the World Wide Web, and all Web browsers can read
files in this format—including local files on disk. The file itself is just
ASCII text, but it contains a complex coding scheme that allows docu-
ments to contain not only text, but also links to other files, including
graphics and sound. You can use Web authoring software, or LabVIEW
automatically generates portable documentation in HTML through the
Print command.
Custom online help is another viable way to distribute documents.
By definition, all LabVIEW users have a way to view such documents.
They can be searched by keyword, and of course you can include graph-
ics. See “Custom Online Help,” or the LabVIEW user’s manual for more
information on creating these files.
 Chapter

Instrument Driver Basics

10
A LabVIEW instrument driver is a collection of VIs that controls a
programmable instrument. Each routine handles a specific operation
such as reading data, writing data, or configuring the instrument. A
well-written driver makes it easy to access an instrument because it
encapsulates the complex, low-level hardware setup and communica-
tions protocols. You are presented with a set of driver VIs that are easy
to understand, modify, and apply to the problem at hand.
The key to writing drivers lies in a simple pronouncement: RTM—
Read The Manual. We’re talking about the programming manual that
goes with the instrument you’re writing the driver for and the online
Application Note Developing LabVIEW Plug and Play Instrument
Drivers. Without these references, you may end up frustrated or, at best,
with a really bizarre driver. The guidelines on writing drivers, com-
plete instrument drivers, as well as many development tools are gath-
ered online at the Instrument Driver Network (ni.com/idnet). National
Instruments (NI) has made it exceptionally simple to create full-
featured plug-and-play instrument drivers with the new Instrument
Driver Project Wizard in LabVIEW 8. We’ll cover communication
basics in this chapter and the Wizard in Chapter 11. Communicating
with an instrument is more than just having a set of VIs.

Finding Instrument Drivers

One reason that LabVIEW has been so widely accepted is because of
the way it incorporates instrument drivers. Unlike in many compet-
ing applications, LabVIEW drivers are written in LabVIEW—complete
with the usual panels and diagrams—instead of arriving as precompiled

243
244 Chapter Ten

black boxes that only the manufacturer can modify. Thus you can start
with an existing driver and adapt it to your needs as you see fit. And
you should adapt it because no off-the-shelf driver is going to match
your application perfectly. Another important asset is the National
Instruments online instrument driver repository at ni.com/idnet. It
contains drivers for hundreds of instruments using a variety of hard-
ware standards such as GPIB, RS-232/422, USB, even ancient VXI, and
CAMAC. Each driver is fully supported by National Instruments. They
are also free. Obtaining drivers online is easy with the Instrument
Driver Finder in LabVIEW 8. From the LabVIEW Help menu select
Help >> Find Instrument Driver…. The Instrument Driver Finder con-
nects you to the instrument driver network at ni.com/idnet and a col-
lection of several thousand instrument drivers. Most of the drivers are
commercial-grade software, meaning that they are thoroughly tested,
fairly robust, documented, and supported. If you can’t find the driver
you need, you might call the manufacturer of the instrument or check
its Web site. Many times the manufacturer will have an in-house devel-
opment version you can use as a starting point. There is no need to
reinvent the wheel!
National Instruments welcomes contributions to the instrument net-
work. If you have written a new driver that others might be interested
in using, consider submitting it. Consultants who are skilled in writing
instrument drivers can join the Alliance Program and become Certi-
fied Instrument Driver Developers. If you have that entrepreneurial
spirit, you could even sell your driver package, providing that there’s
a market. Software Engineering Group (SEG) was probably the first
to do this with its HighwayView package that supports Allen-Bradley
programmable logic controllers (PLCs). It’s been a successful venture.
A few manufacturers of programmable instruments also sell their driv-
ers rather than place them in the library.
Existing instrument drivers are also one of your best resources for
instrument programming examples. Whenever you need to write a new
driver, look at an existing driver that is related to your new project and
see how it was done. Standardization is key to software reuse. A stan-
dard instrument driver model enables engineers to swap instruments
without requiring software changes, resulting in significant savings in
time and money. You should design drivers for similar instruments (for
instance, all oscilloscopes) that have a common interface in the soft-
ware world. This means all the connector panes should be the same.
Rule: Every instrument should be a drop-in replacement for every other
instrument of its genre. This is especially important in the world of
automated test systems requiring interchangeable hardware and soft-
ware components.
 Instrument Driver Basics 245

Driver Basics
Writing a driver can be trivial or traumatic; it depends on your pro-
gramming experience, the complexity of the instrument, and the app-
roach you take. Start by going to the Web site and reading National
Instruments’ Instrument Driver Guidelines, which will give you an
overview of the preferred way to write a driver. Then spend time with
your instrument and the programming manual. Figure out how every-
thing is supposed to work. Next decide what your objectives are. Do
you just need to read a single data value, or do you need to implement
every command? This has implications with respect to the complexity
of the project.

Communication standards
The kinds of instruments you are most likely to use are “smart,” stand-
alone devices that use one of several communication standards, nota-
bly serial, GPIB, USB, and Ethernet. Serial and GPIB are ancient
by computer standards, but still a reliable and common interface on
stand-alone instruments. USB 2.0 adds a high-speed data bus designed
for multimedia but is perfectly suitable for message-based instruments.
Ethernet-based instruments have to share the network bandwidth
with everything else on the network. This can limit their effective-
ness for deterministic acquisition or high-speed transfer, but makes
it great for placing an instrument in a remote location, or sharing an
instrument between multiple controllers. Fortunately VISA provides a
common framework and application programming interface (API) for
controlling instruments over all these busses. Instrument drivers prop-
erly written using VISA API calls can transparently change between
serial, GPIB, USB, or Ethernet without modification.

Serial instruments. Popular serial communication interface stan-

dards include RS-232C, RS-422A, and RS-485A. These standards are
defined by the Electronic Industries Alliance (EIA) and include speci-
fications for cabling, connector pin-outs, signal levels, and timing. Note
that none of these standards say anything about the protocol or mes-
sage formats. This is a popular misconception. Protocols are covered
by independent standards or, more often, by the will and whim of the
instrument manufacturer. The advantages of serial communication are
simplicity and low cost. RS-232, for instance, has been around for a
long time, so there is plenty of inexpensive hardware available. The
disadvantages are that serial systems tend to be slow (you can move
only 1 bit of data at a time) and the protocols are sometimes burden-
some in terms of programming.
246 Chapter Ten

RS-232C is the most common standard, supported by virtually all

computer systems and most serial instruments. It uses the familiar
25-pin or 9-pin D-style connector. The standard defines functions for all
the pins, although most often you will see just a few of the lines used:
transmit data, receive data, and ground. The other pins are used for
handshaking, with such functions as Clear to Send and Ready to Send.
These functions are supported by the LabVIEW VISA serial port func-
tions, if your instrument requires their use. The main limitation of
RS-232 is that it is electrically single-ended which results in poor noise
rejection. The standard also limits the maximum distance to 100 m,
although this is frequently violated in practice. The top data rate is
normally 115,200 Bd (baud, or bits per second), but some systems sup-
port higher speeds. It’s also the most abused and ignored standard
in the industry. Manufacturers take many liberties regarding signal
levels, baud rate, connectors, and so forth. Not all RS-232–compliant
devices are actually compatible. Beware!
RS-422A is similar to RS-232, but uses differential drivers and
receivers, thus requiring two pairs of wires plus a ground as the mini-
mum connection. Because of its superior noise rejection and drive capa-
bility, RS-422 is usable to 1200 m at speeds below 9600 Bd, and up to
1 MBd at shorter distances.
Other standards in this series are RS-432A and RS-485A. RS-432A
is identical to RS-232C, except that the electrical characteristics are
improved to support higher transmission rates over longer distances.
RS-485A is an extension of RS-422A that specifies greater drive cur-
rent and supports multidrop systems (multiple talkers and listeners
on one set of wires). Generally, your computer will need an adapter box
or plug-in board to properly support these standards. National Instru-
ments makes plug-in boards with up to 16 RS-485 ports. You can buy
adapter boxes that provide electrical compatibility among all these
standards from companies such as Black Box.

Adding serial ports. If you need more serial ports on your computer,
consider a multiport plug-in board. One important feature to look for
is a 16550-compatible UART. This piece of hardware includes a 16-byte
buffer to prevent data loss (the nefarious overrun error), which other-
wise will surely occur. For any computer with PCI slots, there are sev-
eral sources. Keyspan makes some excellent multiport plug-in boards
for the Macintosh. National Instruments has multiport boards for PCI,
PXI, and PCMCIA busses. You can also use a USB serial port adapter.
Keyspan makes a two-port model that is very popular with LabVIEW
users. For remote systems you might look at National Instruments’
Ethernet Device Server for RS-232 or RS-485. Serial ports on the
 Instrument Driver Basics 247

Ethernet Device Server look like and act as locally attached serial
ports configured with NI VISA.

Troubleshooting serial interfaces. There is great likelihood that you will

encounter problems when setting up RS-232–style serial communica-
tions systems. Here are a few of the more common problems and solu-
tions.

■ Swapped transmit and receive lines. Although the standards clearly

define which is which, manufacturers seem to take liberties. It’s easy
to hook things up backward the first time, and the whole system is
dead. The answer is to use a null modem cable that swaps these criti-
cal lines. You can use a voltmeter to figure out which pins are trans-
mit and receive. Transmit lines will deliver a steady voltage, on the
order of 3 V for RS-232 and ±1.5 V for RS-422 and RS-485. Receive
lines generally stick around 0 V.
■ Failure to properly connect the hardware handshaking lines. For
instance, some instruments won’t transmit until the Clear to Send
(CTS) line is asserted. Study your instrument’s manual and try to
find out which lines need to be connected.
■ Wrong speed, parity, or stop bit settings. Obviously, all parties must
agree on these low-level protocol settings. The Serial Port Init VI
(Serial Instrument I/O function palette) sets these parameters for
most serial interfaces.
■ Serial cable problems. When in doubt, a serial line activity indicator
from Radio Shack is a very useful tool for verifying serial line con-
nections and activity. If you do a lot of serial cable debugging, then a
full RS-232 breakout box from Black Box or the equivalent will allow
for rapid troubleshooting of serial cables.
■ Multiple applications or devices trying to access the serial port. If
the VISA Configure Serial Port function returns an error, make sure
that there are no other programs running on your computer that
might be using the same serial port.

GPIB instruments. Hewlett-Packard Corporation gets credit for invent-

ing this popular communications and control technique back in 1965,
calling it the HP Interface Bus (HP-IB), a name it uses to this day.
The Institute of Electrical and Electronics Engineers (IEEE) formal-
ized it as IEEE 488 in 1978, after which its common name General-
Purpose Interface Bus (GPIB) was adopted. It’s a powerful, flex-
ible, and popular communications standard supported by thousands of
248 Chapter Ten

commercial instruments and computer systems. The most important

characteristics of GPIB are as follows:

■ It is a parallel, or bus-based, standard capable of transferring 1 byte

(8 bits) per cycle.
■ It is fairly fast, transferring up to about 800 Kbytes/s typically and
8 Mbytes/s maximum.
■ Hardware takes care of timing and handshaking.
■ It requires significantly more expensive and complex hardware than
do serial interfaces.
■ Distance is limited to 20 m unless special bus extender units are
added.
■ Up to 15 devices can coexist on one bus; up to 31 can coexist with a
bus expander unit.

There is also a newer version of this standard, IEEE 488.2, estab-

lished in 1987. The original standard didn’t address data formats, sta-
tus reporting, error handling, and so forth. The new standard does. It
standardizes many of these lower-level protocol issues, simplifying
your programming task. National Instruments’ GPIB boards support
IEEE 488.2, and the LabVIEW VISA library is based on it.
Even after the IEEE 488.2 standard was implemented, chaos reigned
in the area of instrument command languages. A consortium of instru-
mentation companies convened in 1990 to define the Standard Com-
mands for Programmable Instruments (SCPI), and programmers’
lives became much easier. SCPI defines
■ Commands required by IEEE 488.2. Example: *IDN? (send instru-
ment identification).
■ Commands required by SCPI. Example: ERR? (send next entry from
instrument’s error queue).
■ Optional commands. Example: DISP (controls selection and presen-
tation of display information).

Commands are structured as a hierarchical command tree. You

specify a function, a subfunction, a sub-subfunction, and so on, sepa-
rated by a colon (:) character. For instance, the SCPI message for an
autoranging voltage measurement would be

SENSE:VOLTAGE:RANGE:AUTO

Programming a SCPI-compliant instrument can be a pleasant sur-

prise because the command set makes sense! Also, the SCPI model
 Instrument Driver Basics 249

Measurement Function

Signal
Routing
INPut SENSe CALCulate FORMat
data
bus

TRIGger MEMory DISPlay

Signal
OUTput SOURce CALCulate FORMat
Routing
data
bus
Signal Generation

Figure 10.1 The SCPI instrument model. Each block, or subsys-

tem, is represented by a command hierarchy.

(Figure 10.1) covers instruments of all kinds with a general model, so

the programs you write tend to be reusable. The LabVIEW Instrument
Driver Wizard creates drivers based on templates that conform to the
SCPI model.

Learn about Your Instrument

You can’t program it if you don’t understand it. When you receive a
new instrument for a driver project, you’re not always familiar with
its purpose or its modes of operation. Expect it to take several hours of
experimenting with the controls and reading the user’s manual before
you’re comfortable with the fundamentals. It’s also a good idea to con-
nect the instrument to some kind of signal source or output monitor.
That way, you can stimulate inputs and observe outputs. It’s good to
have a function generator, a voltmeter, and an oscilloscope, plus test
leads and accessories to accommodate most instruments. Sometimes,
you need to borrow more exotic equipment. Take our advice: Don’t
spend too much time trying to write a driver if you don’t have access
to the actual instrument. Fiddle with the knobs and get acquainted.
You can certainly learn the basics of the command set offline, but it’s
amazing how soon you will have questions that can be answered only
by exchanging some messages.
Obtain and study the programming manual. Some companies write
programming manuals that border on works of art, as you will find
with the newer HP and Tektronix instruments. Other manuals are
little more than lists of commands with terse explanations. (We’ll be
nice guys and let those companies remain anonymous.) Pray that you
don’t end up with one of the latter. By the way, programming manuals
are notorious for errors and omissions. Maybe it’s true because so few
people actually do any low-level programming; they rely on people like
250 Chapter Ten

us to do the dirty chore of writing a driver. If you find mistakes in a

manual, by all means tell the manufacturer so it can fix them in the
next revision.
Even the best manuals can be kind of scary—some complex instru-
ments have really big manuals. What you have to do is to skim through
a couple of times to get a feel for the overall structure of the command
set. Pay attention to the basic communications protocol, which is espe-
cially important for instruments that use the serial port. Also, figure
out how the instrument responds to and reports errors. Take notes,
draw pictures, and make lists of important discoveries.
Interactions between settings need special attention. For instance,
changing from volts to amperes on a digital multimeter (DMM) prob-
ably implies that the range control takes on a different set of legal
values. Sending an illegal range command may cause the meter to
go to the nearest range, do nothing, or simply lock up! (Instruments
are getting better about these sorts of things; in the old days, reboot-
ing an instrument was sometimes required.) Your driver will need to
arbitrate these control interactions, so note them as soon as they’re
discovered.

Determine Which Functions to Program

Nothing is quite so overwhelming as opening the programming manual
and finding out that your instrument has 4378 commands! Obviously,
you’re not going to implement them all; you need to decide which com-
mands are needed for your project.
If you are hired specifically to write an instrument driver, as we con-
sultants are, the customer will probably supply a list of commands that
you are required to support. If the list is long, you simply take more
time, charge more money, and become rich. But if you are writing a
driver for your own specific application, you probably don’t have the
time to develop a comprehensive driver package. Instead, you must
determine the scope of the job and decide exactly which functions you
need to implement. Here are some functions that most simple driver
packages should support:

■ Basic communications. You need the ability to write to and read from
the instrument, with some form of error handling.
■ Sending commands. You need the ability to tell the instrument to
perform a function or change a setting.
■ Transferring data. If the instrument is a measurement device such as
a voltmeter or an oscilloscope, you probably need to fetch data, scale
it, and present it in some useful fashion. If the instrument generates
 Instrument Driver Basics 251

outputs such as a waveform generator, you need to write blocks of

data representing waveforms.
■ Configuration management. You need the ability to load and store the
settings of many of the instrument’s important controls all at once.
Some instruments have internal setup memory, while others let you
read and write long strings of setup commands with ease. Still others
offer no help whatsoever, making this task really difficult.
■ Important controls. There are always a few basic controls that you
simply must support, such as the mode and range on a DMM. More
extensive drivers support more controls.

The basic question to ask is, Will the user do most of the setup manu-
ally through the instrument’s front panel, or will LabVIEW have to do
most of the work? If you’re using LabVIEW on the test bench next to an
oscilloscope, manually adjust the scope until the waveform looks OK,
and then just have LabVIEW grab the data for later analysis. No soft-
ware control functions are required in this case, and all you really want
is a Read Data VI. If the application is in the area of automated test
equipment (ATE), instrument setups have to be controlled to guaran-
tee that they are identical from test to test. In that case, configuration
management will be very important and your driver will have access
to many control functions.
Think about the intended application, look through the programming
manual again, and pick out the important functions. Make a checklist
and get ready to talk to your instrument.

Establish Communications
After you study the problem, the next step is to establish communica-
tions with the instrument. You may need to install communications
hardware (such as a GPIB interface board); then you need to assemble
the proper cables and try a few simple test commands to verify that
the instrument “hears” you. If you are setting up your computer or
interface hardware for the first time, consider borrowing an instru-
ment with an available LabVIEW driver to try out. It’s nice to know
that there are no problems in your development system.
LabVIEW’s Express VI, Instrument I/O Assistant, is an interac-
tive general-purpose instrument controller that you can use to test com-
mands one at a time. Figure 10.2 shows the “*IDN?” query and response
in the I/O Assistant. It’s easy to type in a command and see how the
instrument responds. The parsing window lets you sort out the reply
and assign data types to data transmitted from your instrument. You
can even group bytes together into multibyte data types. In Figure 10.2
252 Chapter Ten

Figure 10.2The Instrument I/O Assistant is an interactive tool for establishing communication with
instruments. The instrument in this example was built with LabVIEW Embedded for the ADI Blackfin
DSP. LabVIEW not only can control instruments, but also can be the instrument firmware.

the assistant made the correct assumptions about the comma-separated

data types returned by our instrument: Manufacturer (string), Model
(string), Hardware Version (DBL), and Software Version (DBL). If the
data types or grouping is wrong, the I/O Assistant makes it easy to
select multiple bytes and group them into other data types. The I/O
Assistant will even plot returned waveform data. Once you have sorted
out the communication details of your instrument, the I/O Assistant
will create an awful VI for you with all the commands you just tested
chained together. This is exactly the kind of VI and code you do not want
to use! It is not a real instrument driver. In Chapter 11 we’ll look at how
to build on these experimental commands and create a modular instru-
ment driver that you can use and reuse as your application grows.

Hardware and wiring

Serial.Problematic because the standards are so often violated (non-
standard standards?), serial hardware can be more challenging to hook
 Instrument Driver Basics 253

1
Serial cable connections for
DSR 6
IBM compatibles using a DB-
2 XMIT 9 connector.
RTS 7
3 RCV If you need a DB-25
CTS 8 connection, use a DB-9 to
4 DTR DB-25 adapter.
RI 9
5 GND

Figure 10.3 For PCs using a DB-9 connector, you can get cables and
connectors like this at your local electronics emporium. DB-9 to DB-25
adapters are also handy.

up than the other types. If you use a lot of serial devices, a breakout box
will quickly pay for itself.
Figure 10.3 shows serial port connections for the very common PC
version. Cables are available with the required DB-9 on one end and a
DB-9 or DB-25 on the free end to mate with most instruments. If you
install a board with extra serial ports or with special interfaces (such
as RS-422 or RS-485), special connectors may be required.
If you must extend the serial cable beyond a few feet, it is important
to use the right kind of cable. Transmit and receive lines should reside
in separate twisted, shielded pairs to prevent crosstalk (Belden 8723
or equivalent). If you don’t use the correct type of cable, the capacitive
coupling will yield puzzling results: Every character you transmit will
appear in the receive buffer.
Your instrument’s manual has to supply information about connector
pin-outs and the requirements for connections to hardware handshaking
lines (such as CTS and RTS). The usual wiring error is to swap transmit
and receive lines on RS-232. The RS-422 adds to the problem by permit-
ting you to swap the positive and negative lines as well. One trick you
can use to sort out the wires is to test each pin with a voltmeter.
The negative lead of the meter should go to the ground pin. Transmit
lines will be driven to a nice, solid voltage (such as +3 V), while receive
lines will be 0 V, or floating randomly. You can also use an oscilloscope
to look for bursts of data if you can force the instrument to go into Talk
mode. A null modem cable or adapter usually does the trick for RS-
232 signal mix-ups. It effectively swaps the transmit and receive lines
and provides jumpers for CTS and RTS. Be sure to have one handy,
along with some gender changers (male-to-male and female-to-female
connectors), when you start plugging together serial equipment.

GPIB. You need some kind of IEEE 488 interface in your computer.
Plug-in boards from National Instruments are the logical choice since
254 Chapter Ten

they are all supported by LabVIEW and they offer high performance.
External interface boxes (SCSI, RS-232, USB, or Ethernet to GPIB)
also use the same driver. Cabling is generally easy with GPIB because
the connectors are all standard. Just make sure that you don’t vio-
late the 20-m maximum length; otherwise, you may see unexplained
errors and/or outright communication failures. Incredibly, people have
trouble hooking up GPIB instruments. Their main problem is failure
to make sure that the connector is pushed all the way in at both ends.
Always start your testing with just one instrument on the bus to avoid
unexpected addressing clashes.

Protocols and basic message passing

GPIB. Figure out how to set the GPIB address of your instrument, or
at least find out what the address is. From the instrument’s manual,
determine what the command terminator is (carriage return, line feed,
EOI asserted, or a combination), and pick out a simple command to try
out. On the first try, you may get a time-out because the GPIB address
is incorrect. If all else fails, go through all the addresses from 1 to 31
(0 is the controller) until you get a response. In really tough cases, you
may have to use a GPIB bus analyzer (HP, Tektronix, National Instru-
ments, and IOTech all make one) to examine the bus traffic. Always
verify the command terminator. If the instrument expects the end or
identify (EOI) signal or a special character and you don’t send it, noth-
ing will happen.
Another classic hang-up with GPIB involves the issue of repeat
addressing. When the controller (your computer) talks to another
device on the bus (an instrument), the first thing the controller has to
do is to address the instrument. If the controller wishes to communi-
cate with the same device again, it may take advantage of the fact that
many (but not all) instruments remember that they are still addressed.
If the instrument doesn’t remember that it is still addressed, then you
need to enable the repeat addressing action in your instrument driver.
You do this by calling a VISA Property node with repeat addressing set
to True.

Serial. Unlike GPIB, serial instruments have little in common with

one another since there are essentially no standards in use. Some
instruments are individually addressed so that they can be used in
multidrop networks such as RS-422A. Others have no addressing and
are designed for use on a dedicated link. Protocols vary from a GPIB-
like ASCII message exchange to pure binary with complex handshak-
ing. Study your instrument’s programming manual and figure out what
protocol it uses. The Instrument I/O Assistant can communicate using
 Instrument Driver Basics 255

ASCII messages or raw binary. You just need to know your instrument
(Read the Manual).
Be sure to set the basic serial parameters: speed (baud rate), bits per
character, stop bits, parity, and XON/XOFF action. LabVIEW’s VISA
Configure Serial Port function sets these parameters and more. If
you use a terminal emulator, you will have to do the same thing in
that program. Check these parameters very carefully. Next to swapped
transmit and receive lines, setting one of these items improperly is the
easiest way to fail to establish communication.
Instruments that use those nasty binary protocols with handshaking
are nearly impossible to test by hand, but the I/O Assistant gives you at
least a fighting chance. For these instruments, try to muddle through
at least one command just to find out if the link is properly connected;
then plunge right in and write the LabVIEW code to handle the proto-
col. Hopefully, you will be 90 percent successful on the first try because
debugging is difficult.
As with GPIB, you can use a serial line analyzer (made by HP, Tek-
tronix, and National Instruments) to eavesdrop on the communications
process. The analyzer stores lots of characters and decodes them to
whatever format you choose (ASCII, hexadecimal, octal, etc.). This really
helps when your driver almost works but still has reliability problems.

Figure 10.4 NI Spy eavesdrops on communications as a troubleshooting aid.

256 Chapter Ten

If you just need to eavesdrop on the communication but don’t need to

monitor each status line, then NI Spy is what you need. NI Spy can log
every VISA call from any application that communicates using VISA.
Figure 10.4 shows a capture file from NI Spy.
In Chapter 11, we’ll start writing drivers by using a modern method-
ical approach.

Bibliography
1999 SCPI Syntax & Style, SCPI Consortium 2515 Camino del Rio South, Suite 340 San
Diego, CA, 1999
 Chapter

Instrument Driver
11
Development Techniques

With every major release of LabVIEW, we’ve watched the general

architecture of instrument drivers change significantly. Back in the
time of LabVIEW 1, about the only I/O functions we had were GPIB
read and write, and serial port read and write. Because the product
was so new, driver architectures weren’t really defined, and there was
a degree of chaos in the blossoming driver library. With LabVIEW 2, we
gained IEEE-488.2 functions and the important application note Writ-
ing a LabVIEW 2 Instrument Driver, which gave developers directions
regarding hierarchical program design. In the time of LabVIEW 3,
those of us involved in driver (and DAQ) development standardized on
error I/O as a common thread. It really cleaned up our diagrams, elim-
inated many Sequence structures, and unified error handling. VME
Extensions for Instrumentation (VXI) instruments appeared, as did
the initial VISA Transition Library (VTL). (VTL was implemented
as a wrapper around the existing GPIB functions, rather than as native
functions.) Many consultants launched their careers by writing drivers
for VXI instruments, some of them using VTL as the foundation. Since
LabVIEW 4, VISA is the primary platform for driver development—in
LabVIEW and in many other instrumentation control languages.
The Virtual Instrument Software Architecture (VISA) provides
a standard communication driver to handle all forms of instrument I/O,
whether the bus is GPIB, VXI, PXI, Serial, Ethernet, or USB. VISA is
an accepted standard and an integral part of LabVIEW. The principle
is simple: There is a consistent interface to all instruments based on
the concept of Open, Read/Write, and Close, much like file operations.
(See Figure 11.1.) When an instrument is opened, a VISA session

257
258 Chapter Eleven

ID? byte count Instr. ID

GPIB0::1::INSTR *IDN? 100

error in (no error) error out

Figure 11.1 VISA Open, Write, Read, Close functions.

identifier (like a file refnum) is created to uniquely identify that

instrument. Data is exchanged in string format for message-based
instruments or as words or bytes for register-level access. Events,
such as triggers and service requests, are a part of the standard, as is
error handling. Today LabVIEW’s GPIB and serial routines use VISA
underneath—whether you knew it or not, if you talk to an instrument
with LabVIEW, you’re using VISA. The major advantage, of course, is
that any driver you write with VISA is immediately portable across
all VISA-supported computer platforms and busses without any addi-
tional programming on your part.
Just as VISA provides a standard communication driver, Standard
Commands for Programmable Instruments (SCPI) defines a
common set of programming commands. SCPI works by describing
manufacturer agreed upon commands and command style across all
instruments. SCPI defines the messages you send to an instrument,
how the instrument replies, and how transmitted data is formatted.
Figure 11.2 shows the SCPI commands used to get voltage (MEAS:
VOLT?) and current (MEAS:CURR?) from a power supply. The reason-
ing behind SCPI is simple: If two instruments have the same controls,
the same measurement capabilities, and the same basic functions, then
it makes sense that you should be able to use the same commands. Of
course, this is the real world, and every manufacturer thinks its unique
instrument requires unique commands. But at least you have a fight-
ing chance if they follow a SCPI format. Learning a completely new
command set and data format for every new instrument is a pain!
Instrument drivers have come a long way since LabVIEW 4; now
VISA and SCPI are the standard for all LabVIEW instrument drivers.
But just because we’re all using the same tools doesn’t mean the drivers
will work well together, or, better yet, be interchangeable. To create

%.;%g
%.;%g

Actual Voltage Level (V)

Actual Current Level (A)
Voltage Measurement Command Current Measurement Command
:MEAS:VOLT? 100 :MEAS:CURR? 100

Figure 11.2 Standard Commands for Programmable Instruments simplifies

instrument commands.
 Instrument Driver Development Techniques 259

instrument drivers that truly work well together, we need to use the
same templates and follow formal guidelines. After all, you must have
standards, recommended practices, and quality assurance mechanisms
in place if you expect to produce reliable, professional-grade software.

Plug-and-Play Instrument Drivers

LabVIEW Plug-and-Play instrument drivers have a common
look and feel, so you can operate the instrument with a minimum of
programming. Additionally, each plug-and-play instrument driver is
interchangeable with drivers of similar instruments. Creating new
plug-and-play instrument drivers sounds like a tall order, and it
indeed would be, were it not for new tools based on the guidelines and
product suggestions of the LabVIEW Instrument Driver Advisory
Board. The board is made up of leading LabVIEW users who help set
standards and recommend improvements to LabVIEW. Many of their
suggestions have been incorporated into the LabVIEW 8 Instrument
Driver Project Wizard. The new Instrument Driver Project Wizard
provides templates, documentation tools, and icon creation tools for the
most common instrument types. This new tool really makes it easy to
create Plug-and-Play instrument drivers. You don’t even have to be a
LabVIEW guru. (See Figure 11.3.)

Figure 11.3 The Instrument Driver Project Wizard has templates for common
instrument types.
260 Chapter Eleven

But don’t rely totally on the wizard; it’s a great framework to start
with, but properly designed instrument drivers need to conform to
the guidelines found at the National Instruments(NI) Web site www
.ni.com/idnet. The two most important documents that you must read
before proceeding with driver programming are as follows:

■ Developing LabVIEW Plug-and-Play Instrument Drivers “describes

the standards and practices for LabVIEW Plug-and Play-drivers,
is intended for driver developers and describes the standards for
structuring VIs, instrument I/O, style, and error reporting. It also
describes instrument driver components and how to integrate these
components. In addition, this document describes a process for devel-
oping useful instrument drivers.”
■ Instrument Driver Guidelines (checklist) “should be used to verify
that your instrument driver complies with the Instrument Driver
Library standards. These guidelines provide the requirements and
recommendations for programming style, error handling, front pan-
els, block diagrams, icons, testing, and documentation, and should
be referenced throughout the driver development process. Use these
guidelines as a checklist to review your driver for instrument driver
certification.”

To help you get started in the right direction, we’ve incorporated NI’s
recommendations and provided a few examples using the Instrument
Driver Project Wizard that represent good driver design techniques.
Even if you’re writing a driver only for yourself, it’s worth following
these guidelines because they will help you structure your drivers and,
in the end, become a better programmer.

General Driver Architectural Concepts

Building an instrument driver is no different from any other LabVIEW
project. You still need to define the problem, do some preliminary test-
ing and prototyping, and then follow good design and programming
practices. Follow up with thorough testing and documentation, and you
will have a versatile product that is easy to use and easy to maintain.
The LabVIEW project created by the Instrument Driver Project Wiz-
ard uses driver template VIs to create a project complete with core
VIs, use case examples, and documentation. Once you’ve created the
project, it’s easy to go through the VIs and modify the commands to
match the format required by your instrument. As you go through the
process, edit the documentation; and when you’re done, you will have a
professional-quality instrument driver.
 Instrument Driver Development Techniques 261

Error I/O flow control

An important characteristic of a robust driver is that it handles error
conditions well. Any time you connect your computer to the outside
world, unexpected conditions are sure to generate errors, and the instru-
ment driver software is the place to trap those errors. Your response
depends on the nature and severity of the error, the state of the overall
system, and options for recovery.
The most straightforward error response is a dialog box, but avoid
placing dialog routines down inside the driver layer. The last thing
an unsuspecting developer using your driver wants is a surprise dia-
log box halting a running system. Dialogs are best placed at a higher
application layer and not in the driver. Assuming that an operator is
always available to respond, the dialog box can present a clear state-
ment about the error: who, what, when, where, and why. Dialogs are
always appropriate for fundamental errors such as basic setup con-
flicts (“No GPIB board in slot 3”) and for catastrophic errors where
the program simply cannot continue without operator intervention
(“The power supply reported a self-test failure and has shut down”).
It’s also helpful to receive informative dialogs when you are setting up
an instrument for the first time. As you develop your driver, make sure
you squash every bug and answer every mysterious error as they pop
up during development. You should never ship a driver that generates
unexplained errors.
Automatic error recovery is used often in serial communications sys-
tems, particularly those that span great distances and/or run at high
speeds. Local-area networks typically have sophisticated means of error
detection and automatic message retransmission. For a serial instru-
ment, you may be able to attempt retransmission of data if a time-out
occurs or if a message is garbled. Retransmission can be added to a
state machine-based driver without too much trouble. If a message is
invalid, or if a time-out occurs, you could jump back to the first frame
of the sequence and send the command again, hoping for success on the
second try. An important feature to include there would be a limit on
the number of retransmissions; otherwise, you would end up in an infi-
nite loop if the instrument never responded. Such automatic actions
should be limited to the lowest levels of a driver package, as in the
example of a serial interface. Serious measurement and control appli-
cations are designed to report any error that is detected, to guarantee
data quality and system reliability. Therefore, you should think care-
fully before making your program so smart that it “fixes” or ignores
potentially dangerous error conditions.
You can use error handling to promote dataflow programming and
simplify applications of a driver package. An error I/O cluster containing
262 Chapter Eleven

error information connects one VI to the next. If an error is generated by

one VI, the next one will be notified of the error and be able to take appro-
priate action. For instance, if a function that initializes the instrument
fails, it probably makes no sense to continue sending commands. The
error I/O chaining scheme can manage this decision at the driver level,
rather than forcing the user to add Case structures to the top-level VI.
VIs respond to an incoming error by skipping all activity and simply
passing the incoming error through. Decisions about how to handle the
error are made by the application developer at a higher level.
A canonical VI with error I/O is shown in Figure 11.4. The error in
and error out clusters always appear in the connector pane as shown.
Doing so permits all users of error I/O to line up their VIs in an even
row. These clusters are available as standard controls in the Array and
Cluster control menu. The cluster contents are always the same:

Item 1: Status (boolean). True means an error occurred. False implies

a warning if the error code number is nonzero.
Item 2: Code (I32). An error code number used for message genera-
tion.
Item 3: Source (string). Contains the name of the VI calling chain
that generated the error.

On the diagram, the error in boolean is tested, and if it’s True, you
may run a special error response program or simply pass the error along
to the error out indicator and do nothing else. If no incoming error
is detected, then it’s up to your program to generate an error cluster

error in (no error) error out

status code status code

0 -1301
error in error out

source source
Canonical Error I/O.vi
Read Data.vi

No Error Error
No incoming error. Run your program. Generate
Incoming error;
error in an error cluster that reports the results
error just pass it along.
(no error)
True=error out

error code
Alternatively, you could take
Current VI's Path special action in response to
the error condition
Strip Path

Figure 11.4 The Canonical error I/O VI. You can encapsulate any program inside the False case.
 Instrument Driver Development Techniques 263

Configure Trigger Measurement

Measurement
Configuration Number (1) Measurement Display
VISA session dup VISA session

error in (no error) error out

Instr. Driver Revision Hold

Instr. Firmware Revision

Figure 11.5 An example of error in/error out, the AOIP Instrumentation OM22 microohmme-
ter, a GPIB instrument. Note how clear and simple this application is, with no need for flow
control structures. Error I/O is an integral part of the underlying VISA driver.

of its own, depending on the outcome of the operation that it performs.

Figure 11.4 shows a couple of tricks in case your subVI doesn’t gen-
erate an error cluster. Generate the Status boolean with the help of
the Not Equal comparison function. Enter the name of the VI into the
Source string automatically by using the Current VI’s Path constant
and the Strip Path function, both from the File I/O function menu.
Figure 11.5 is an example of error I/O in action in the driver for a
GPIB instrument. Error I/O is an integral part of all VISA drivers.
This technique makes it easy to do your error handling in a consistent
fashion all the way from the bottom to the top of the hierarchy. As you
can see from the diagram, applying the AOIP Instrumentation OM22
driver is very easy because of the use of error I/O.
The VISA functions have error I/O built into the function. Each VISA
function checks the error status boolean to see if an error occurred
upstream. If there is an error, the VISA function does nothing and
passes the error message along without modification through the error
out indicator. If there is not an error, the VISA function acts normally.
You can take advantage of this inherent error handling to clean up and
simplify your diagrams. You also want to be alert to any errors and cor-
rect them as soon as they occur.
At some point in your application, usually at the end of a chain of I/O
operations, the error cluster is submitted to an error handler. The error
handler decodes the code number into plain language and appends the
source string. That way, the user will understand what the error is and
where it came from without resorting to an error message table or a lot
of debugging. The message is generally displayed in a dialog box.
Error codes are somewhat standardized. Zero generally means
no error. Positive numbers are warnings, and negative numbers are
errors. Table 11.1 lists some of the error codes that are reserved for use
with generic instrument driver VIs; there are many more codes defined
by the VISA driver as well. If you use one of these codes in your own
264 Chapter Eleven

TABLE 11.1 Predefined VISA Error Codes for Instrument Drivers

Code Description
0 No Error: the call was successful
1073481728 to 1073483775 Warning: Developer defined warnings
–1074003951 Error: Identification query failed
–1074000000 Error: Instrument Defined Error
–1073999873 to –1074001919 Error: Developer defined errors

drivers, the General Error Handler VI that comes with LabVIEW will
be able to interpret them. (Codes from 5000 to 9999 are reserved for
miscellaneous user-defined errors; they are not automatically inter-
preted by the error handlers.)
The General Error Handler utility VI (Figure 11.6) accepts the
standard error cluster or the individual items from an error cluster and
decodes the code numbers into a message that it can display in an error
dialog and in an output string. A useful feature of this handler is its

[user-defined descriptions]
[user-defined codes]
[error code] (0) error?
[error source] (" ") code out
type of dialog (OK msg:1) source out
error in (no error) message
[exception action] (none:0) error out
[exception code]
[exception source]
General Error Handler

message reinitialize to default to display instructions

[error code] (0) error?
Error 1301 occurred at My Bad VI.
0 ERROR

[error source] (" ") Possible reasons:

code out
My special error message
1301

source out
My Bad VI
type of dialog (OK msg:1)
no dialog 0

[user-defined codes] [user-defined descriptions]

error in (no error) error out
0 1301 0 My special error message
status code status code
ERROR 1301 ERROR 1301
source source
My Bad VI [exception action] (none:0) [exception code] [exception source] My Bad VI
no exception 0

Figure 11.6 The General Error Handler utility VI is all you need to decode and display errors.
The arrows indicate how the VI can decode user-defined error codes and messages that you
supply, in addition to the predefined error codes.
 Instrument Driver Development Techniques 265

True

Error Code
0

Error Query Command Bytes VISA resource name out

:SYST:ERR? 256 True
%[^\r\n] %d
VISA resource name If code ever non-zero
then set error out to
error in (no error) be instrument error
0
EDIT: Replace the Error Query Command status error out
status -1074000000
code constant with your instrument's error
query command. Instrument reports:
Timeout error
-1073807339

Error Message

The Error Query VI reads the instrument’s error buffer. Run this VI after each command or set of
Figure 11.7
commands to get fresh error messages.

ability to decode user-defined error codes. You supply two arrays: user-
defined codes and user-defined descriptions. The arrays have a one-to-
one correspondence between code numbers and messages. With this
handler, you can automatically decode both predefined errors (such as
GPIB and VXI) and your own special cases. We would suggest using the
codes in Table 11.1 because they are reserved for instrument-specific
errors.
Error I/O simplifies diagrams by eliminating the need for many flow
control structures. By all means, use it in all your drivers and other
projects.
If you are programming a SCPI-compliant instrument, you can read
errors from the standardized SCPI error queue. A driver support VI,
Error Query, polls the instrument and returns a string containing
any error messages via the regular error out cluster. (See Figure 11.7.)
Add this VI after every command or series of commands. Events are
stored in a queue—oldest events are read out first—so you need to read
events out after every command to keep from filling the queue with old,
uninteresting events. This technique will add additional communica-
tions overhead in the event of an error, but it will catch every error as
it happens.

Modularity by grouping of functions

We’ve seen two extremes of modularity applied to LabVIEW instru-
ment drivers: too much and none. Too much modularity exists when
there is one subVI for every individual command. There are several
problems with this approach. First, the interaction between control
settings is really hard to arbitrate because each function has to query
the instrument to see what state it’s in, rather than just compar-
ing a few control settings on the VI’s panel. Second, the user has to
plow through the directories and libraries, searching for the desired
command. Third, there are just too darned many VIs, which becomes
a nightmare for the poor person who has to maintain the library.
266 Chapter Eleven

Guess what happens when you find a mistake that is common to all
352 VIs? You get to edit every single one of them.
Having no modularity is the other extreme. You may find it in trivial
drivers in which there are only a couple of commands and hence just
one VI. You may also find a lack of modularity in complex drivers writ-
ten by a novice, in which case it’s a disaster. There is nothing quite like
a diagram with sequences inside of sequences inside of sequences that
won’t even fit on a 19-in monitor. In fact, these aren’t really drivers at
all. They are actually dedicated applications that happen to implement
the commands of a particular instrument. This kind of driver is of little
use to anyone else.

Project organization
The correct way to modularize a driver in LabVIEW is to build VIs that
group the various commands by function. In the LabVIEW Plug-and-
Play Instrument Driver model, VIs are organized into six categories:
Initialize, Configuration, Action/Status, Data, Utility, and Close. (See
Figure 11.8.) These intuitive groupings make an instrument easier to
learn and program. The wizard automatically creates stub VIs for you

Figure 11.8 The Instrument Driver Project Wiz-

ard automatically creates a project organized with
core VI stubs, examples, and documentation.
 Instrument Driver Development Techniques 267

Initialize Configuration Action/Status Data Utility Close

Single Waveform

Low Level

Figure 11.9 Default VI Tree for oscilloscopes. Common commands, grouped by function, are easy to find and use.
You create Action/Status VIs as needed.

that implement the basic commands for each instrument. The project
is ordered and grouped according to function, and a VI Tree is created
to make the interrelationship easy to see. The default VI Tree for oscil-
loscopes is shown in Figure 11.9. Once you’ve created the project by
using the wizard, you will need to modify the specific commands inside
each VI to match your instrument. Any additional functions should
be created as new VIs in the project hierarchy and placed on the VI
Tree.VI. The one thing you never want to do is to change the connector
pane or the functionality of the default VIs! Doing so will break their
plug-and-play nature. The templates are designed to expose the com-
mon functionality of each class of instrument in a way that will make
the instruments and the instrument drivers interchangeable in auto-
mated test systems. If you’ve developed a test system before, you can
appreciate the value of replacing hardware without having to rewrite
software. Because the instrument drivers are plug-and-play with the
same connector pane and the same functionality, an application can be
written using VI Server to dynamically call instrument driver plug-ins
that initialize, configure, and take data. Which plug-in VI Server calls
depends on the hardware attached to the computer. Of course, this all
works seamlessly only if each of the VIs has the same connector pane
and generic functionality.

Initialization
The cleanest way to begin any conversation with an instrument is to
create an initialization VI that starts out by calling VISA Open to
establish a connection. The initialization VI should take care of any
special VISA settings, such as serial port speed or time-outs, by call-
ing a VISA Property node, and it should also perform any instru-
ment initialization operations, such as resetting to power up defaults
or clearing memory. Instruments that are SCPI-compliant can be que-
ried with *IDN?, which will return an identification string that you can
check to verify that you are indeed talking to the right instrument.
268 Chapter Eleven

Open instrument Set VISA Attributes

True
Serial Interface C0 30
VISA resource name Instr Instr
error in (no error) Timeout ASRL End In
TermChar
Intf Type ASRL End Out
0 10000 4096
TermChar
Serial Baud
0
Serial Configuration Flow Cntrl
Parity
Baud Rate Data Bits
Flow Control Stop Bits
Parity
Data Bits
Stop Bits

Figure 11.10 Initialize.VI opens the VISA connection and configures port speeds.

An important item to configure in your initialization VI is the termi-

nation character. Many instruments use a special character, such as
a carriage return or line feed, to indicate the end of the message. VISA
can automatically detect a specified character when receiving data (you
must append it yourself when you build a string to send, however). The
Initialize.VI in Figure 11.10 sets the termination character to a line
feed (0x0A) by type casting the line feed constant to an unsigned byte.

Configuration
Configuration VIs are the set of software routines that get the instru-
ment ready to take data. They are not a set of VIs to store and save
instrument configuration—those go under the utility section. A typical
configuration VI, Configure Edge Trigger.VI, is shown in Figure 11.11.

Source Command
Edge Trigger Mode :TRIG:MAIN:EDGE:SOU CH%d;
:TRIG:MAIN:TYP EDGE;

Trigger Source (1: Channel 1)

Slope Command
:TRIG:MAIN:EDGE:SLO RIS;
:TRIG:MAIN:EDGE:SLO FALL;

Trigger Slope (0: Rising) Level Command

%.;:TRIG:MAIN:LEV %g;

Trigger Level (0.0 V)

VISA resource name VISA resource name out

error in (no error) error out
Figure 11.11 Configure Edge Trigger VI. Simple dataflow and logical control names make this block dia-
gram easy to understand. Edit the command string for your instrument.
 Instrument Driver Development Techniques 269

It concisely implements the basic capabilities of the instrument’s trig-

ger section. This VI is organized in its most basic form and is typical
of all good driver VIs. Note some of its important attributes: error I/O
clusters, VISA sessions as input and output common threads, and a
simple dataflow diagram that builds a command string for transmis-
sion to the instrument. If a user needs to add a more advanced com-
mand, there is plenty of panel and diagram space to do so.
An important part of making your driver robust is your expectation
that the user will enter unexpected values into every control. At the
programmatic level, you must add some code on the diagram for range
checking. The In Range And Coerce function and the various com-
parison functions are candidates for range checking. If the values are
not evenly spaced (such as a 1-2-5 sequence), use a Case structure or
a ring control, or select from a list. Changing the representation of
a numeric control or indicator may also limit the range of the input,
particularly for integer types. For instance, a U8 control can only repre-
sent integers between 0 and 255. Anything greater than 255 is coerced
to 255, and anything less than 0 is coerced to 0.
Other difficult situations must be handled programmatically. Many
instruments limit the permissible settings of one control based upon
the settings of another. For example, a voltmeter might permit a range
setting of 2000 V for dc, but only 1000 V for ac. If the affected controls
(e.g., Range and Mode) reside in the same VI, put the interlock logic
there. If one or more of the controls are not readily available, you can
request the present settings from the instrument to make sure that
you don’t ask for an invalid combination. This may seem like a lot of
work, but we’ve seen too many instruments that go stark, raving bon-
kers when a bad value is sent.
String controls don’t have a feature analogous to the Data Range
dialog; all the checking has to be done by your program. LabVIEW
has several string comparison functions that are helpful. You can use
Empty String/Path? to test for empty strings and path names, which
are probably the most common out-of-range string entries. The other
string comparison functions—Decimal Digit?, Hex Digit?, Octal
Digit?, Printable?, White Space?, and Lexical Class—give you
important attributes for the first character in a string. Being polymor-
phic, they also act on the ASCII equivalent of a number. For instance, if
you wired the integer number 0 to the Printable? function, the func-
tion would return False, since the ASCII equivalent of 0 is NUL, which
is not printable. Boundary conditions—values such as zero, infinity
(Inf), not-a-number (NaN), and empty string—cause lots of trouble in
instrument drivers. Use the comparison functions to prevent defective
values from finding their way to the instrument, and always test your
drivers to make sure that ridiculous inputs don’t stop the show.
270 Chapter Eleven

Action and status

VIs that start or stop an instrument’s operation are action VIs. Action
VIs arm a trigger or start the generation or acquisition of a waveform.
Unlike configuration Vis, they do not change settings, but cause the
instrument to act based on its current settings. Status VIs are used by
other VIs to obtain the current status of the instrument.
What happens when your 500-GHz 986ZX computer sends a bunch of
commands in rapid-fire mode to a poor, unsuspecting instrument? Will
the commands be buffered up in high-speed memory for later process-
ing? Or will the instrument simply halt and catch fire? What if there
is a communications error? Instruments are not infinitely fast—they
require some time to respond to commands—and your driver must syn-
chronize itself to the needs of the instrument. Timing and handshaking
techniques, both hardware and software, are available and necessary
to manage these situations.
Sending commands at the wrong time can be a problem. Any com-
mand that initiates a time-consuming operation has the potential for
putting the instrument offline for awhile. For instance, if you tell your
instrument to perform an extensive self-diagnostic test, it may ignore
all GPIB activity until the test is completed, or it might buffer the com-
mands for later execution. Also, the rate at which commands are sent
can be important. Most instruments permit you to send many com-
mands in one string. The question is, Will all the commands in the
string be received before the first one is executed? Most times, the
answer is yes, but there may be dire implications if the instrument
ignores one-half of the commands you just sent. Experience indicates
that testing is the only way to be sure.
There are two ways to solve these timing problems. First, you can
add delays between commands by sending them one at a time in a
timed Sequence structure. The second method is to find out if the pre-
vious operation is complete. If you enable service requests (SRQs)
for your instrument, you can use Wait For RQS, from the VISA >>
VISA Advanced >> Event-handling function palette. This technique is
often used with digital oscilloscopes. Either way you should check your
instrument often for errors, and report them as they happen.

Data
Data VIs should transfer data to or from the instrument with a mini-
mum of user involvement. The VI in Figure 11.12 queries the scope
for scaling information (t0, dt, Y scale, and Y offset) and then builds a
waveform with the data. The VI is clean and simple, VISA and error I/O,
and waveform data. In Figure 11.12 the data is returned as a comma-
separated ASCII string. Sometimes transferring data as ASCII is
 Instrument Driver Development Techniques 271

%f;%f;%f;%f;%f
t0, dt, Y scale, Y offset
Get Waveform preamble Get Waveform t0
WFMPRE:XZE?;XIN?;YZE?;YMU?;YOFF? CURV? dt Waveform Graph
Y
VISA session 80000 VISA session
,
200 ASCII data
%f
error in (no error) error out
Convert comma separated
0 0 ASCII to floating point

Figure 11.12 ASCII data is converted to floating-point waveform.

the easiest, and most instruments support it. But if you have a choice,
binary protocols are faster and more compact.
The VI in Figure 11.13 is the same as that in Figure 11.12 except the
data has been returned as 16-bit signed binary. You can easily change
from a binary string to an array of I16 with the type cast function.
Additional formatting may be required if your binary waveform has
a length header or termination characters. Consult your instrument’s
manual for the proper handling method. LabVIEW provides all the
tools you need to reconstruct your data, including functions to swap
bytes and words. When you are transferring binary data, be sure your
VISA Read function does not terminate on an end-of-line character.
Use a VISA Property node to disable this before the read and reenable
after the read; otherwise, you may not get all your data.

Utility
Sometimes a command doesn’t fit with anything else, and it is nor-
mally used by itself, for instance, utility VIs used for instrument reset,
self-test, calibration, or store/recall configuration. It would be illogical
to throw those VIs in with other unrelated functions.
A good driver package allows you to save and recall collections of
control settings, which are usually called instrument setups. This saves

%f;%f;%f;%f;%f
t0, dt, Y scale, Y offset
Get Waveform preamble Get Waveform t0
WFMPRE:XZE?;XIN?;YZE?;YMU?;YOFF? CURV? dt Waveform Graph
Y
VISA session 80000 VISA session

200 Binary data

error in (no error) error out I16
0
Typecast to signed 16-bit integer array

Figure 11.13 Binary data is converted to floating-point waveform.

272 Chapter Eleven

much time, considering how many controls are present on the front
panels of some instruments. Sophisticated instruments usually have
built-in, nonvolatile storage for several settings, and your driver should
have a feature that accesses those setup memories. If the instrument
has no such memory, then write a program to upload and save setups
on disk for cataloging and later downloading. You may want to provide
this local storage even if the instrument has setup memory: What if the
memory fails or is accidentally overwritten?
Historical note: Digital storage oscilloscopes (DSOs) gradually became so
complex and feature-laden that it was considered a badge of honor for
a technician to obtain a usable trace in less than 5 minutes—all those
buttons and lights and menus and submenus were really boggling! The
solution was to provide an auto-setup button that got something visible on
the display. Manufacturers also added some means by which to store and
recall setups. All the good DSOs have these features, nowadays.

Some instruments have internal setup memory, while others let you
read and write long strings of setup commands with ease. Still others
offer no help whatsoever, making this task really difficult. If it won’t
send you one long string of commands, you may be stuck with the task
of requesting each individual control setting, saving the settings, and
then building valid commands to send the setup back again. This can
be pretty involved; if you’ve done it once, you don’t want to do it again.
Instead, you probably will want to use the instrument’s built-in setup
storage. If it doesn’t even have that, write a letter to the manufacturer
and tell the company what you think of its interface software.

Close
The Close VI is at the end of the error chain. Remember that an impor-
tant characteristic of a robust driver is that it handles error conditions
well. You need to make a final check with your instrument and clear
out any errors that may have occurred. Error I/O is built into the VISA
functions. (See Figure 11.14.) If there is an incoming error, the function
will do nothing and pass the error on unmodified. The VISA Close func-
tion is the exception. VISA Close will always close the session before
passing any error out.

VISA resource name

error in (no error) error out
Figure 11.14 Close VI checks for instrument errors before closing the connection.
 Instrument Driver Development Techniques 273

Documentation
One measure of quality in instrument driver software is the documen-
tation. Drivers tend to have many obscure functions and special appli-
cation requirements, all of which need some explanation. A driver that
implements more than a few commands may even need a function
index so that the user can find out which VI is needed to perform a
desired operation. And establishing communications may not be triv-
ial, especially for serial instruments. Good documentation is the key
to happy users. Review Chapter 9, “Documentation.” It describes most
of the techniques and recommended practices that you should try
to use.
When you start writing the first subVI for a driver, type in control
and indicator descriptions through the Description and Tip pop-up
menus. This information can be displayed by showing the Help window
and is the easiest way for a user to learn about the function of each
control. If you enter this information right at the start, when you’re
writing the VI, you probably know everything there is to know about
the function, so the description will be really easy to write. You can
copy text right out of the programming manual, if it’s appropriate. It
pays to document as you go.
The Documentation item in the VI Properties dialog box from the
File menu is often a user’s only source of information about a VI. The
information you enter here shows up, along with the icon, in the Con-
text Help window. Try to explain the purpose of the VI, how it works,
and how it should be used. This text can also be copied and pasted into
the final document.
Try to include some kind of document on disk with your driver. The
issue of platform portability of documents was mentioned in Chapter 9,
and with drivers, it’s almost a certainty that users of other systems
will want to read your documents. As a minimum, paste all vital infor-
mation into the Get Info box or a string control on a top-level example
VI that is easy to find. Alternatively, create a simple text document. If
you’re energetic and you want illustrations, generate a PDF (Portable
Document Format) document by using Adobe Acrobat, and make sure
it stays with your driver VIs; or create a LabVIEW Help document and
create hyperlinks to appropriate sections in the VI Properties dialog
box. The links will show up in the Context Help window.
If you’re just doing an in-house driver, think about your coworkers
who will one day need to use or modify your driver when you are not
available. Also, remember that the stranger who looks at your pro-
gram 6 months from now may well be you. Include such things as how
to connect the instrument, how to get it to “talk,” problems you have
274 Chapter Eleven

discovered (and their resolutions), and of course a description of each

function VI. Not only is such information useful, but also it may make
you famous rather than infamous.

Bibliography
Developing LabVIEW Plug and Play Instrument Drivers, www.ni.com/idnet, National
Instruments Corporation, 11500 N. Mopac Expressway, Austin, Tex., 2006.
Instrument Communication Handbook, IOTech, Inc., 25971 Cannon Road, Cleveland,
Ohio, 1991.
Instrument Driver Guidelines, www.ni.com/idnet, National Instruments Corporation,
11500 N. Mopac Expressway, Austin, Tex., 2004.
 Chapter

Inputs and Outputs

12
To automate your lab, one of the first things you will have to tackle
is data acquisition—the process of making measurements of physical
phenomena and storing them in some coherent fashion. It’s a vast tech-
nical field with thousands of practitioners, most of whom are hackers
as we are. How do most of us learn about data acquisition? By doing it,
plain and simple. Having a formal background in engineering or some
kind of science is awfully helpful, but schools rarely teach the practical
aspects of sensors and signals and so on. It’s most important that you
get the big picture—learn the pitfalls and common solutions to data
acquisition problems—and then try to see where your situation fits
into the grand scheme.
This chapter should be of some help because the information pre-
sented here is hard-won, practical advice, for the most part. The most
unusual feature of this chapter is that it contains no LabVIEW pro-
gramming information. There is much more to a LabVIEW system
than LabVIEW programming! If you plan to write applications that
support any kind of input/output (I/O) interface hardware, read on.

Origins of Signals
Data acquisition deals with the elements shown in Figure 12.1.
The physical phenomenon may be electrical, optical, mechanical, or
something else that you need to measure. The sensor changes that
phenomenon into a signal that is easier to transmit, record, and
analyze—usually a voltage or current. Signal conditioning amplifies
and filters the raw signal to prepare it for analog-to-digital conversion
(ADC), which transforms the signal into a digital pattern suitable for
use by your computer.

275
276 Chapter Twelve

Physical Sensor Signal ADC

Phenomenon Conditioning

G fl 10110011

Figure 12.1 Elements of the data acquisition process.

Transducers and sensors

A transducer converts one physical phenomenon to another; in our
case, we’re mostly interested in an electric signal as the output. For
instance, a thermocouple produces a voltage that is related to tem-
perature. An example of a transducer with a nonelectrical output is a
liquid-in-glass thermometer. It converts temperature changes to vis-
ible changes in the volume of fluid. An elaboration on a transducer
might be called a sensor. It starts with a transducer as the front end,
but then adds signal conditioning (such as an amplifier), computations
(such as linearization), and a means of transmitting the signal over
some distance without degradation. Some industries call this a trans-
mitter. Regardless of what you call it, the added signal conditioning is
a great advantage in practical terms, because you don’t have to worry
so much about noise pickup when dealing with small signals. Of course,
this added capability costs money and may add weight and bulk.
Figure 12.2 is a general model of all the world’s sensors. If your instru-
ment seems to have only the first couple of blocks, then it’s probably
a transducer. Table 12.1 contains examples of some sensors. The first
one, example A, is a temperature transmitter that uses a thermocouple
with some built-in signal conditioning. Many times you can handle
thermocouples without a transmitter, but as we’ll see later, you always
need some form of signal conditioning. The second example, example
B, is a pressure transmitter, a slightly more complex instrument. We
came across the third example, example C, a magnetic field sensor,

Sensor
Physical Primary Conversion Signal Computations Signal Display
Phenomenon Detector Element Conditioner Transmission

Power Supply and

Support

Figure 12.2 A completely general sensor model. Many times, your sensor is just a trans-
ducer that ends after the conversion element.
 Inputs and Outputs 277

TABLE 12.1 Three Practical Examples of Sensor Systems

Block Example A Example B Example C
Phenomenon Temperature Pressure Magnetic field
Detector — Diaphragm displacement Faraday rotation
Transducer Thermocouple LVDT Laser and photodiode
Signal conditioner Cold junction; amplifier Demodulator ADC
Computations Linearize Linearize; scale Ratio; log; scale
Transmission 0−10 VDC 4–20 mA RS-232 serial
Display Analog meter Analog meter Computer system
Support DC DC (2-wire current loop) DC; cooling

Example A is a temperature transmitter using a thermocouple, with cold-junction compensation, linearization, and
analog output.
Example B is a pressure sensor using a linear variable differential transformer (LVDT) to detect diaphragm displace-
ment, with analog output.
Example C is a magnetic field measurement using an optical technique with direct signal transmission to a computer.
This represents a sophisticated state-of-the-art sensor system (3M Specialty Optical Fibers).

while doing some research. It uses an optical principle called Fara-

day rotation in which the polarization of light is affected by magnetic
fields.
A detailed discussion of sensor technology covering all aspects of the
physics of transducers and the practical matters of selecting the right
sensor is beyond the scope of this book. For the purposes of data acqui-
sition, there are several important things you need to know about each
of your sensors:

■ The nature of the signal it produces—voltage, amplitude range,

frequency response, impedance, accuracy requirement, and so on—
determines what kind of signal conditioning, analog-to-digital con-
verter (ADC), or other hardware you might need.
■ How susceptible is the sensor to noise pickup or loading effects from
data acquisition hardware?
■ How is the sensor calibrated with respect to the physical phenom-
enon? In particular, you need to know if it’s nonlinear or if it has
problems with repeatability, overload, or other aberrant behavior.
■ What kind of power or other utilities it might require? This is often
overlooked and sometimes becomes a show stopper for complex
instruments!
■ What happens if you turn off your data acquisition equipment while
the sensor still has power applied? Will there be damage to any
components?

When you start to set up your system, try to pick sensors and design
the data acquisition system in tandem. They are highly interdependent.
278 Chapter Twelve

When monitored by the wrong ADC, the world’s greatest sensor is of

little value. It is important that you understand the details of how your
sensors work. Try them out under known conditions if you have doubts.
If something doesn’t seem right, investigate. When you call the manu-
facturer for help, it may well turn out that you know more about the
equipment than the designers, at least in your particular application. In
a later section, we’ll look at a holistic approach to signals and systems.
Modern trends are toward smart sensors containing onboard micro-
processors that compensate for many of the errors that plague trans-
ducers, such as nonlinearity and drift. Such instruments are well worth
the extra cost because they tend to be more accurate and remove much
of the burden of error correction from you, the user. For low-frequency
measurements of pressure, temperature, flow, and level, the process
control industry is rapidly moving in this direction. Companies such
as Rosemount, Foxboro, and Honeywell make complete lines of smart
sensors.
More complex instruments can also be considered sensors in a
broader way. For instance, a digital oscilloscope is a sensor of voltages
that vary over time. Your LabVIEW program can interpret this voltage
waveform in many different ways, depending upon what the scope is
connected to. Note that the interface to a sensor like this is probably
GPIB or RS-232 serial communications rather than an analog voltage.
That’s the beauty of using a computer to acquire data: Once you get the
hardware hooked up properly, all that is important is the signal itself
and how you digitally process it.

Actuators
An actuator, which is the opposite of a sensor, converts a signal (per-
haps created by your LabVIEW program) into a physical phenomenon.
Examples include electrically actuated valves, heating elements, power
supplies, and motion control devices such as servomotors. Actuators
are required any time you wish to control something such as tempera-
ture, pressure, or position. It turns out that we spend most of our time
measuring things (the data acquisition phase) rather than controlling
them, at least in the world of research. But control does come up from
time to time, and you need to know how to use those analog and digital
outputs so conveniently available on your interface boards.
Almost invariably, you will see actuators associated with feedback
control loops. The reason is simple. Most actuators produce responses
in the physical system that are more than just a little bit nonlinear and
are sometimes unpredictable. For example, a valve with an electropneu-
matic actuator is often used to control fluid flow. The problem is that the
flow varies in some nonlinear way with respect to the valve’s position.
 Inputs and Outputs 279

Also, most valves have varying degrees of nonrepeatability. They creak

and groan and get stuck—hysteresis and deadband are formal terms
for this behavior. These are real-world problems that simply can’t be
ignored. Putting feedback around such actuators helps the situation
greatly. The principle is simple. Add a sensor that measures the quan-
tity that you need to control. Compare this measurement with the
desired value (the difference is called the error), and adjust the actua-
tor in such a way as to minimize the error. Chapter 18, “Process Control
Applications,” delves more deeply into this subject. The combination
of LabVIEW and external loop controllers makes this whole situation
easy to manage.
An important consideration for actuators is what sort of voltage or
power they require. There are some industrial standards that are fairly
easy to meet, such as 0 to 10 V dc or 4 to 20 mA, which are modest
voltages and currents. But even these simple ranges can have added
requirements, such as isolated grounds, where the signal ground is not
the chassis of your computer. If you want to turn on a big heater, you
may need large relays or contactors to handle the required current; the
same is true for most high-voltage ac loads. Your computer doesn’t have
that kind of output, nor should it. Running lots of high power or high
voltage into the back of your computer is not a pleasant thought! Try to
think about these requirements ahead of time.

Categories of signals
You measure a signal because it contains some type of useful infor-
mation. Therefore, the first questions you should ask are, What infor-
mation does the signal contain, and how is it conveyed? Generally,
information is conveyed by a signal through one or more of the follow-
ing signal parameters: state, rate, level, shape, or frequency content.
These parameters determine what kind of I/O interface equipment and
analysis techniques you will need.
Any signal can generally be classified as analog or digital. A digital,
or binary, signal has only two possible discrete levels of interest—an
active level and an inactive level. They’re typically found in computer
logic circuits and in switching devices. An analog signal, on the other
hand, contains information in the continuous variation of the signal
with respect to time. In general, you can categorize digital signals as
either on/off signals, in which the state (on or off ) is most important,
or pulse train signals, which contain a time series of pulses. On/off sig-
nals are easily acquired with a digital input port, perhaps with some
signal conditioning to match the signal level to that of the port. Pulse
trains are often applied to digital counters to measure frequency,
period, pulse width, or duty cycle. It’s important to keep in mind that
280 Chapter Twelve

digital signals are just special cases of analog signals, which leads to
an important idea:
Tip: You can use analog techniques to measure and generate digital
signals. This is useful when (1) you don’t have any digital I/O hardware
handy, (2) you need to accurately correlate digital signals and analog
signals, or (3) you need to generate a continuous but changing pattern
of bits.
Among analog signal types are the dc signal and the ac signal. Ana-
log dc signals are static or vary slowly with time. The most important
characteristic of the dc signal is that information of interest is con-
veyed in the level, or amplitude, of the signal at a given instant. When
measuring a dc signal, you need an instrument that can detect the
level of the signal. The timing of the measurement is not difficult as
long as the signal varies slowly. Therefore, the fundamental operation
of the dc instrument is an ADC, which converts the analog electric sig-
nal into a digital number that the computer interprets. Common exam-
ples of dc signals include temperature, pressure, battery voltage, strain
gauge outputs, flow rate, and level measurements. In each case, the
instrument monitors the signal and returns a single value indicating
the magnitude of the signal at a given time. Therefore, dc instruments
often report the information through devices such as meters, gauges,
strip charts, and numerical readouts.
Tip: When you analyze your system, map each sensor to an appropri-
ate LabVIEW indicator type.
Analog ac time domain signals are distinguished by the fact that
they convey useful information not only in the level of the signal, but
also in how this level varies with time. When measuring a time domain
signal, often referred to as a waveform, you are interested in some
characteristics of the shape of the waveform, such as slope, locations
and shapes of peaks, and so on. You may also be interested in its fre-
quency content.
To measure the shape of a time domain signal with a digital com-
puter, you must take a precisely timed sequence of individual amplitude
measurements, or samples. These measurements must be taken closely
enough together to adequately reproduce those characteristics of the
waveform shape that you want to measure. Also, the series of measure-
ments should start and stop at the proper times to guarantee that the
useful part of the waveform is acquired. Therefore, the instrument used
to measure time domain signals consists of an ADC, a sample clock, and
a trigger. A sample clock accurately times the occurrence of each ADC.
Figure 12.3 illustrates the timing relationship among an analog
waveform, a sampling clock, and a trigger pulse. To ensure that the
desired portion of the waveform is acquired, you can use a trigger
to start and/or stop the waveform measurement at the proper time
 Inputs and Outputs 281

A Sample

Analog
Waveform

Sampling
Clock

Trigger Sampling starts when this trigger pulse occurs

Time

Figure 12.3An illustration of the relationship between an analog waveform and

the sampling clock and trigger that synchronize an ADC.

according to some external condition. For instance, you may want to

start acquisition when the signal voltage is moving in a positive direc-
tion through 0 V. Plug-in boards and oscilloscopes generally have trig-
ger circuits that respond to such conditions.
Another way that you can look at an analog signal is to convert the
waveform data to the frequency domain. Information extracted from
frequency domain analysis is based on the frequency content of the
signal, as opposed to the shape, or time-based characteristics of the
waveform. Conversion from the time domain to the frequency domain
on a digital computer is carried out through the use of a Fast Fourier
Transform (FFT), a standard function in the LabVIEW digital signal
processing (DSP) function library. The Inverse Fast Fourier Trans-
form (IFFT) converts frequency domain information back to the time
domain. In the frequency domain, you can use DSP functions to observe
the frequencies that make up a signal, the distribution of noise, and
many other useful parameters that are otherwise not apparent in the
time domain waveform. Digital signal processing can be performed by
LabVIEW software routines or by special DSP hardware designed to do
the analysis quickly and efficiently.
There is one more way to look at analog signals, and that is in the
joint time-frequency (JTF) domain (Qian and Chen 1996). This
is a combination of the two preceding techniques. JTF signals have
an interesting frequency spectrum that varies with time. Examples
are speech, sonar, and advanced modulation techniques for communi-
cation systems. The classic display technique for JTF analysis is the
spectrogram, a plot of frequency versus time, of which LabVIEW has
one of the very best, the Gabor spectrogram algorithm (Figure 12.4).
You can order the LabVIEW Signal Processing Toolkit, which includes
282 Chapter Twelve

Figure 12.4These plots illustrate the striking differences between a joint time-
frequency analysis plot of a chirp signal analyzed with the short-time FFT spec-
trogram and the same chirp signal analyzed with the Gabor spectrogram

a compiled application as well as the necessary VIs to do your own

processing with the Gabor spectrogram, short-time FFT, and several
others. It works with live data from a data acquisition board or any
other source.
As Figure 12.5 shows, the signal classifications described in this sec-
tion are not mutually exclusive. A single signal may convey more than

Signal

Digital On-Off Digital Pulse Train

High 5.0
1 2 3 4 5

L ow 0.0

0.00 0.10 0.20 0.30 0.40 0.0 1.0 2.0 3.0 4.0 5.0

Analog DC Analog Time Domain Frequency Domain

overshoot 25
5.0 0
5.50
db
4.83 V
5.00 –50
4.50
0.0 risetime –100
4.00 0.0 2.0 4.0 6.0 8.0 10.0
0.05 0.15 0.25 0.00 0.10 0.20 0.30 0.40 0.50 freq. (Hz)

Figure 12.5 Five views of a signal. A series of pulses can be classified in several
different ways depending on the significance of the signal’s time, amplitude,
and frequency characteristics.
 Inputs and Outputs 283

SIGNAL CLASSIFICATION TYPICAL TYPICAL LABVIEW

INTERFACE PROCESSING DISPLAY

DC or VERY ADC/DAC AVERAGE NUMERIC

LOW FREQUENCY SINGLE SAMPLES STRIP CHART
ANALOG
FILTER,
AC TIME-DOMAIN ADC/DAC THRESHOLD, GRAPH
(WAVEFORM) BUFFERED PULSE STATISTICS

JOINT TIME- ADC INTENSITY

FREQUENCY- BUFFERED SPECTROGRAM
GRAPH
DOMAIN
SIGNAL
(Input or Output)
FILTER,
FREQUENCY- ADC FILTER, TIME
THRESHOLD, GRAPH
DOMAIN BUFFERED WINDOW,
PULSE FFT
STATISTICS

ON-OFF, COMBINATIONAL
TTL I/O PORT LOGIC
BOOLEAN
DIGITAL SINGLE BIT

ON-OFF, TTL I/O PORT, NUMERIC BOOLEAN;

MULTI-BIT MULTIPLE BITS CONVERSION CHART

COUNTER/
PULSE TRAIN ARITHMETIC NUMERIC
TIMER

Figure 12.6 This road map can help you organize your information about each signal to logically
design your LabVIEW system.

one type of information. In fact, the digital on/off, pulse train, and dc
signals are just simpler cases of the analog time domain signals that
allow simpler measuring techniques.
The preceding example demonstrates how one signal can belong to
many classes. The same signal can be measured with different types of
instruments, ranging from a simple digital state detector to a complex
frequency analysis instrument. This greatly affects how you choose sig-
nal conditioning equipment.
For most signals, you can follow the logical road map in Figure 12.6
to determine the signal’s classification, typical interface hardware, pro-
cessing requirements, and display techniques. As we’ll see in coming
sections, you need to characterize each signal to properly determine
the kind of I/O hardware you’ll need. Next, think about the signal attri-
butes you need to measure or generate with the help of numerical meth-
ods. Then there’s the user-interface issue—the part where LabVIEW
controls and indicators come into play. Finally, you may have data stor-
age requirements. Each of these items is directly affected by the signal
characteristics.
284 Chapter Twelve

Connections
Professor John Frisbee is hard at work in his lab, trying to make a
pressure measurement with his brand-new computer:
Let’s see here. . . . This pressure transducer says it has a 0- to 10-V dc out-
put, positive on the red wire, minus on the black. The manual for my data
acquisition board says it can handle 0 to 10 V dc. That’s no problem. Just
hook the input up to channel 1 on terminals 5 and 6. Twist a couple of
wires together, tighten the screws, run the data acquisition demonstra-
tion program, and voila! But what’s this? My signal looks like . . . junk! My
voltmeter says the input is dc, 1.23 V, and LabVIEW seems to be working
OK, but the display is really noisy. What’s going on here?

Poor John. He obviously didn’t read this chapter. If he had, he would

have said, “Aha! I need to put a low-pass filter on this signal and then
make sure that everything is properly grounded and shielded.” What
he needs is signal conditioning, and believe me, so do you. This world
is chock full of noise sources and complex signals—all of them guaran-
teed to corrupt your data if you don’t take care to separate the good
from the bad.
There are several steps in designing the right signal conditioning
approach for your application. First, remember that you need to know
all about your sensors and what kind of signals they are supposed to
produce. Second, you need to consider grounding and shielding. You
may also need amplifiers and filters. Third, you can list your specifica-
tions and go shopping for the right data acquisition hardware.

Grounding and shielding

Noise sources are lurking everywhere, but by following some simple
principles of grounding and shielding, you can eliminate most noise-
related problems right at the source—the wiring. In fact, how you con-
nect a sensor to its associated data acquisition hardware greatly affects
the overall performance of the system (Gunn 1987; Morrison 1986; Ott
1988; White 1986). Another point on the subject: Measurements are
inherently inaccurate. A good system design minimizes the distortion
of the signal that takes place when you are trying to measure it—noise,
nonlinear components, distortion, and so on. The wire between sen-
sor and data acquisition can never improve the signal quality. You can
only minimize the negatives with good technique. Without getting too
involved in electromagnetic theory, we’ll show you some of the recom-
mended practices that instrumentation engineers everywhere use.
Let’s start with some definitions.

Ground. This is absolutely the most overused, misapplied, and mis-

understood term in all electronics. First, there is the most hallowed
 Inputs and Outputs 285

earth ground that is represented by the electrical potential of the soil

underneath your feet. The green wire on every power cord, along with
any metal framework or chassis with 120-V ac main power applied,
is required by the National Electrical Code to be connected through
a low-resistance path to the aforementioned dirt. There is exactly one
reason for its existence: safety. Sadly, somehow we have come to believe
that grounding our equipment will magically siphon away all sources
of noise, as well as evil spirits. Baloney! Electricity flows only in closed
circuits, or loops. You and your equipment sit upon earth ground as
a bird sits upon a high-voltage wire. Does the bird know it’s living at
34,000 V? Of course not; there is no complete circuit. This is not to say
that connections to earth ground—which we will refer to as safety
ground from here on—are unnecessary. You should always make sure
that there is a reliable path from all your equipment to safety ground
as required by code. This prevents accidental connections between
power sources and metallic objects from becoming hazards. Such fault
currents are shunted away by the safety ground system.
What we really need to know about is a reference potential referred to
as signal common, or sometimes as a return path. Every time you see
a signal or measure a voltage, always ask the question, “Voltage—with
respect to what reference?” That reference is the signal common, which
in most situations is not the same as safety ground. A good example is
the negative side of the battery in your car. Everything electrical in the
vehicle has a return connection to this common, which is also connected
to the chassis. Note that there is no connection whatsoever to earth
ground—the car has rubber tires that make fair insulators—and yet,
the electrical system works just fine (except if it’s very old or British).
In our labs, we run heavy copper braid, welding cable, or copper
sheet between all the racks and experimental apparatus according to a
grounding plan (Figure 12.7). A well-designed signal common can even

Control and Data

Rack-mounted Acquisition Rack
instrument or
computer Ground shield at Sensor
this end ONLY
Twisted,
ADC shielded-pair cable +
–
+
–
Experimental
Safety ground, via bonding System
conductors or “green wire”
on power feed

Signal common
(heavy copper braid)

Figure 12.7 Taking the system approach to grounding in a laboratory. Note the
use of a signal common (in the form of heavy copper braid or cable) to tie every-
thing together. All items are connected to this signal common.
286 Chapter Twelve

be effective at higher frequencies at which second-order effects such

as the skin effect and self-inductance become important. See Gunn
(1987), Morrison (1986), Ott (1988), and White (1986) for instructions on
designing a quality grounding system—it’s worth its weight in aspirin.

Electromagnetic fields. Noise may be injected into your measurement

system by electromagnetic fields, which are all around us. Without
delving into Maxwell’s equations, we present a few simple principles of
electromagnetism that you use when connecting your data acquisition
system.
Principle 1. Two conductors that are separated by an insulator
form a capacitor. An electric field exists between the conductors.
If the potential (voltage) of one conductor changes with time, a pro-
portional change in potential will appear in the other. This is called
capacitive coupling and is one way noise is coupled into a circuit.
Moving things apart reduces capacitive coupling.
Principle 2. An electric field cannot enter a closed, conductive sur-
face. That’s why sitting in your car during a thunderstorm is better
than sitting outside. The lightning’s field cannot get inside. This kind of
enclosure is also called a Faraday cage, or electrostatic shield, and is
commonly implemented by a sheet of metal, screen, or braid surround-
ing a sensitive circuit. Electrostatic shields reduce capacitive coupling.
Principle 3. A time-varying magnetic field will induce an electric
current only when a closed, conductive loop is present. Furthermore,
the magnitude of the induced current is proportional to the intensity
of the magnetic field and the area of the loop. This property is called
inductive coupling, or inductance. Open the loop, and the current
goes away.
Principle 4. Sadly, magnetic shielding is not so easy to design for
most situations. This is so because the magnetic fields that we are
most concerned about (low frequency, 50 /60 Hz) are very penetrating
and require very thick shields of iron or even better magnetic mate-
rials, such as expensive mu-metal.

Here are some basic practices you need to follow to block the effects
of these electromagnetic noise sources:
■ Put sensitive, high-impedance circuitry and connections inside a
metallic shield that is connected to the common-mode voltage (usu-
ally the low side, or common) of the signal source. This will block
capacitive coupling to the circuit (principle 1), as well as the entry of
any stray electric fields (principle 2).
■ Avoid closed, conductive loops—intentional or unintentional—often
known as ground loops. Such loops act as pickups for stray magnetic
 Inputs and Outputs 287

fields (principle 3). If a high current is induced in, for instance, the
shield on a piece of coaxial cable, then the resulting voltage drop
along the shield will appear in series with the measured voltage.
■ Avoid placing sensitive circuits near sources of intense magnetic
fields, such as transformers, motors, and power supplies. This will
reduce the likelihood of magnetic pickup that you would otherwise
have trouble shielding against (principle 4).

Another unsuspected source of interference is your lovely color moni-

tor on your PC. It is among the greatest sources of electrical interfer-
ence known. Near the screen itself, there are intense electric fields due
to the high voltages that accelerate the electron beam. Near the back,
there are intense magnetic fields caused by the fly-back transformer
which is used to generate high voltage. Even the interface cards in your
computer, for instance, the video adapter and the computer’s digital
logic circuits, are sources of high-frequency noise. And never trust a
fluorescent light fixture (or a politician).
Radio-frequency interference (RFI) is possible when there is
a moderately intense RF source nearby. Common sources of RFI are
transmitting devices such as walkie-talkies, cellular phones, com-
mercial broadcast transmitters of all kinds, and RF induction heating
equipment. Radio frequencies radiate for great distances through most
nonmetallic structures, and really high (microwave) frequencies can
even sneak in and out through cracks in metal enclosures. You might
not think that a 75-MHz signal would be relevant to a 1-kHz bandwidth
data acquisition system, but there is a phenomenon known as parasitic
detection or demodulation that occurs in many places. Any time that
an RF signal passes through a diode (a rectifying device), it turns into
dc plus any low frequencies that may be riding on (modulating) the
RF carrier. Likely parasitic detectors include all the solid-state devices
in your system, plus any metal-oxide interfaces (such as dirty connec-
tions). As a licensed radio amateur, Gary knows that parasitic detec-
tion can occur in metal rain gutter joints, TV antenna connections, and
stereo amplifier output stages. This results in neighbors screaming at
him about how he cut up their TV or stereo.
When RFI strikes your data acquisition system, it results in unex-
plained noise of varying amplitude and frequency. Sometimes you can
observe stray RF signals by connecting a wide-bandwidth oscilloscope
to signal lines. If you see more than a few millivolts of high-frequency
noise, suspect a problem. Some solutions to RFI are as follows:

■ Shield all the cables into and out of your equipment.

■ Add RF rejection filters on all signal and power leads.
■ Put equipment in well-shielded enclosures and racks.
288 Chapter Twelve

■ Keep known RF sources and cables far away from sensitive

equipment.
■ Keep the person with the walkie-talkie or cellular phone out of
your lab.

Other error sources. Thermojunction voltages are generated any

time two dissimilar metals come in contact with each other in the pres-
ence of a temperature gradient. This principle, known as the Seebeck
effect, is the way in which a thermocouple generates its tiny signal.
Problems occur in data acquisition when you attempt to measure dc
signals that are in the microvolt to millivolt range, such as those from
thermocouples and strain gauges. If you connect your instruments
with wires, connectors, and terminal screws made of different metals
or alloys, then you run the risk of adding uncontrolled thermojunction
voltages to the signals—possibly to the tune of hundreds of microvolts.
The most common case of thermojunction error that we have seen
occurs when operators hook up thermocouples inside an experimental
chamber and bring the wires out through a connector with pins made
of copper or stainless steel. The thermocouple alloy wire meets the con-
nector pin and forms a junction. Then the operators turn on a big heater
inside the chamber, creating a huge temperature gradient across the
connector. Soon afterward, they notice that their temperature readouts
are way off. Those parasitic thermojunction voltages inside the connec-
tor are added to the data in an unpredictable fashion. Here are some
ways to kill the thermojunction bugs:
■ Make all connections with the same metallic alloy as the wires they
connect.
■ Keep all connections at the same temperature.
■ Minimize the number of connections in all low-level signal situations.

Differential and single-ended connections. The means by which you con-

nect a signal to your data acquisition system can make a difference in
system performance, especially in respect to noise rejection. Single-
ended connections are the simplest and most obvious way to connect a
signal source to an amplifier or other measurement device (Figure 12.8).
The significant hazard of this connection is that it is highly susceptible
to noise pickup. Noise induced on any of the input wires, including
the signal common, is added to the desired signal. Shielding the sig-
nal cable and being careful where you make connections to the signal
common can improve the situation. Single-ended connections are most
often used in wide-bandwidth systems such as oscilloscopes and video,
RF, and fast pulse measurements in which low-impedance coaxial
 Inputs and Outputs 289

Vout = G ( Vin )

Single-Ended
Amplifier

V in G

V out

Signal common
Figure 12.8 A single-ended amplifier
has no intrinsic noise rejection prop-
erties. You need to carefully shield
signal cables and make sure that the
signal common is noise-free as well.

cables are the preferred means of transmission. These single-ended

connections are also practical and trouble-free with higher-magnitude
signals (say, 1 V or greater) that only need to travel short distances.
Differential connections depend on a pair of conductors where the
voltage you want to measure, called the normal-mode signal, is the
difference between the voltages on the individual conductors. Differen-
tial connections are used because noise pickup usually occurs equally
on any two conductors that are closely spaced, such as a twisted pair of
wires. So when you take the difference between the two voltages, the
noise cancels but the difference signal remains. An instrumentation
amplifier is optimized for use with differential signals, as shown in
Figure 12.9. The output is equal to the gain of the amplifier times the
difference between the inputs. If you add another voltage in series with
both inputs, called the common-mode signal, it cancels just like the
noise pickup. The optimum signal source is a balanced signal where

Proper differential amplifier application More common application

(Poorer common-mode rejection)
V out = G ((V in + V cm ) – (V cm ))
V out = G ((V in + V cm ) – (V cm ))
= G (Vin )
= G (Vin )
+ Instrumentation Instrumentation
1
V Amplifier Amplifier
2 in + +
–
G V in G
1
+ – –
V
2 in
–
V out V out
Vcm Vcm

Signal common Signal common

Figure 12.9 Instrumentation amplifiers effectively reject common-mode (Vcm) signals, such as noise
pickup, through cancellation. The normal-mode (differential) signal is amplified.
290 Chapter Twelve

the signal voltage swings symmetrically with respect to the common-

mode voltage. Wheatstone bridge circuits and transformers are the
most common balanced sources.
The common-mode voltage can’t be infinitely large, though, since
every amplifier has some kind of maximum input voltage limit, which
is usually less than 10 V—watch out for overloads! Also, beware that
the common-mode rejection ratio (CMRR) of an amplifier generally
decreases with frequency. Don’t count on an ordinary amplifier to reject
very intense RF signals, for instance. Differential inputs are available on
most low-frequency instrumentation such as voltmeters, chart record-
ers, and plug-in boards, for example, those made by National Instru-
ments. Rule: Always use differential connections, except when you can’t.
In Figure 12.10, you can see some typical signal connection schemes.
Shielded cable is always recommended to reduce capacitive coupling

Acceptable use of differential inputs. Strive for this setup.

Data Acquisition Hardware
Shielded Twisted-Pair
in+ Inst. Amp.
+
Ground-Referenced V Shield ADC
d G Vm
Signal Source in – –

Common-Mode Gnd Circuit Common

Voltage + Noise Vcm

Bias current return path, preferably

through system common

Acceptable use of single-ended inputs. The signal cable could also be coax
for volt-level signals. Data Acquisition Hardware
Inst. Amp.
in+
Floating Signal Source, G ADC
Vm
such as battery- Vd Shield
powered equipment
Gnd Circuit Common

Typical misuse of differential inputs. Ground-loop is formed, picks up noise.

Data Acquisition Hardware
in+ Inst. Amp.
+
Ground-Referenced V ADC
d Shield G Vm
Signal Source in – –

Gnd Circuit Common

Ground-loop current flows here

Figure 12.10 Proper use of differential and single-ended signal connections

is shown in the top two parts of the figure. The bottom part is that all-too-
common case in which a ground loop is formed, enhancing the pickup of
noise borne by magnetic fields.
 Inputs and Outputs 291

and electric field pickup. Twisted, shielded pairs are best, but coax-
ial cable will do in situations where you are careful not to create the
dreaded ground loop that appears in the bottom segment of the figure.

Why use amplifiers or other

signal conditioning?
As you can see, the way you hook up your sensors can affect the overall
performance of your data acquisition system. But even if the grounding
and shielding are properly done, you should consider signal condi-
tioning, which includes amplifiers and filters, among other things,
to reduce the noise relative to the signal level.
Amplifiers improve the quality of the input signal in several ways.
First, they boost the amplitude of smaller signals, improving the reso-
lution of the measurements. Second, they offer increased driving power
(lower output impedance), which keeps such things as ADCs from load-
ing the sensor. Third, they provide differential inputs, a technique
known to help reject noise. An improvement on differential inputs, an
isolation amplifier, requires no signal common at the input what-
soever.* Isolation amplifiers are available with common-mode voltage
ranges up to thousands of volts.
Amplifiers are usually essential for microvolt signals such as those
from thermocouples and strain gauges. In general, try to amplify your
low-level signal as close to the physical phenomenon itself as possible.
Doing this will help you increase the signal-to-noise ratio (SNR).
The signal-to-noise ratio is defined as

V 
SNR = 20 log  sig 
Vnoise 
where Vsig is the signal amplitude and Vnoise is the noise amplitude,
both measured in volts rms. The 20 log() operation converts the simple
ratio to decibels (dB), a ratiometric system used in electrical engi-
neering, signal processing, and other fields. Decibels are convenient
units for gain and loss computations. For instance, an SNR or gain of
20 dB is the same as a ratio of 10 to 1; 40 dB is 100 to 1; and so forth.
Note that the zero noise condition results in an infinite SNR. May you
one day achieve this.

*Manufacturers say you don’t need a ground reference, but experience in the field says
it ain’t so. George Wells of JPL reports that Analog Devices 5B modules sometimes cou-
ple large common-mode voltages to adjacent modules. This may be so because of internal
transformer orientation. Other brands may have their own problems. Your best bet is to
provide a proper ground to all signals.
292 Chapter Twelve

Amplifiers are available in stand-alone modular form, such as those

from Action Instruments and Moore Products that mount on DIN rails,
or in chassis-mounted modules, available from Preston and Ectron.
Another choice is a module such as the National Instruments SCXI-
1120, which includes an isolation amplifier and a low-pass filter on each
of eight channels, followed by a multiplexer that routes the selected
channel to a common ADC input. Most plug-in data acquisition boards
include a programmable-gain instrumentation amplifier (PGIA). You
change the gain by sending a command to the board, and the gain can
be changed at any time—even while you are rapidly scanning through
a set of channels. The difference between an onboard amplifier and an
external amplifier is that the onboard amplifier must respond quickly
to large changes in signal level (and perhaps gain setting) between
channels. Errors can arise due to the settling time, which is the time
it takes for the amplifier output to stabilize within a prescribed error
tolerance of the final value. If you scan a set of channels at top speed
with the amplifier at high gain, there is the possibility of lost accu-
racy. National Instruments has gone to great pains to ensure that its
custom-designed PGIAs settle very rapidly, and it guarantees full
accuracy at all gains and scanning speeds. But if you can afford it, the
amplifier-per-channel approach remains the best choice for high-accu-
racy data acquisition.
Filters are needed to reject undesired signals, such as high-frequency
noise, and to provide antialiasing (discussed later in this chapter).
Most of the time, you need low-pass filters which reject high frequen-
cies while passing low frequencies including dc. The simplest filters are
made from resistors and capacitors (and sometimes inductors). Active
filters combine resistors and capacitors with operational amplifiers
to enhance performance. Switched capacitor filters are a modern
alternative to these ordinary analog filters. They use arrays of small
capacitors that are switched rapidly in and out of the circuit, simulat-
ing large resistors on a much smaller (integrated circuit) scale.
A wide variety of external filters are available in modular form for
mounting on printed-circuit boards and DIN rails. They are also avail-
able in plug-in modules, such as the National Instruments SCXI-1141
nine-channel elliptic low-pass filter module, or in rack-mounted enclo-
sures, such as those made by Frequency Devices, Precision Filters, and
TTE. Alternatively, you can use the aforementioned SCXI-1120 with
an amplifier and filter for each channel. (You might have guessed by
now that we like the SCXI-1120 module. We use it a lot because it has
excellent common-mode rejection, filtering, and accuracy.)
Using a signal conditioning system outside of your computer also
enhances safety. If a large overload should occur, the signal conditioner
will take the hit rather than pass it directly to the backplane of your
 Inputs and Outputs 293

Bridge 4-Wire Strain

Excitation Gauge Bridge

Instrumentation Lowpass
+ Amplifier Filter

+ To ADC
–
G fl
–

Signal Common

Figure 12.11 Schematic of a signal conditioner for a strain gauge, a bridge-based

transducer that requires excitation. This signal conditioner has an instrumentation
amplifier with high gain, followed by a low-pass filter to reject noise.

computer (and maybe all the way to the mouse or keyboard!). A robust
amplifier package can easily be protected against severe overloads
through the use of transient absorption components (such as varistors,
zener diodes, and spark gaps), limiting resistors, and fuses. Medical
systems are covered by federal and international regulations regard-
ing isolation from stray currents. Never connect electronic instruments
to a live subject (human or otherwise) without a properly certified isola-
tion amplifier and correct grounding.
Special transducers may require excitation. Examples are resis-
tance temperature detectors (RTDs), thermistors, potentiometers, and
strain gauges, as depicted in Figure 12.11. All these devices produce an
output that is proportional to an excitation voltage or current as well as
the physical phenomenon they are intended to measure. Thus, the exci-
tation source can be a source of noise and drift and must be carefully
designed. Modular signal conditioners, such as the Analog Devices 5B
series, and SCXI hardware from National Instruments include high-
quality voltage or current references for this purpose.
Some signal conditioning equipment may also have built-in multi-
plexing, which is an array of switching elements (relays or solid-state
analog switches) that route many input signals to one common output.
For instance, using SCXI equipment, you can have hundreds of analog
inputs connected to one multiplexer at a location near the experiment.
Then only one cable needs to be run back to a plug-in board in your
computer. This drastically reduces the number and length of cables.
Most plug-in boards include multiplexers.
Multiplexing can cause some interesting problems when not prop-
erly applied, as we mentioned in regard to the amplifier settling time.
Compare the two signal configurations in Figure 12.12. In the top con-
figuration, there is an amplifier and filter per channel, followed by the
294 Chapter Twelve

Amplifier Lowpass Filter Multiplexer

G fl
Signal A
ADC

G fl
Signal B

Multiplexer Amplifier Lowpass Filter

Signal A
G fl ADC

Signal B

Figure 12.12 The preferred amplifier and filter per channel configura-
tion (top); an undesirable, but cheaper, topology (bottom).Watch your
step if your system looks like the latter.

multiplexer which ultimately feeds the ADC. The SCXI-1120, SCXI-

1102, and many packaged data acquisition systems are designed in
this way. It is the preferred arrangement because each filter output
faithfully follows its respective input signal without disturbances from
switching transients. But in the bottom configuration, the amplifier
and filter are located after the multiplexer. This saves money but is
undesirable because each time the multiplexer changes channels, the
very sluggish low-pass filter must settle to the voltage present at the
new channel. As a result, you must scan the channels at a very slow
rate unless you disable the low-pass filter, thus giving up its otherwise
useful properties. The SCXI-1100 32-channel multiplexer amplifier
module is designed in this way and has caught some users unaware.

Practical tips on connecting input signals.Let’s look at a few common

input signal connection situations that you’re likely to come across. We
recommend that you refer to the manual for your signal conditioning or
plug-in board before making any connections. There may be particular
options or requirements that differ from those in this tutorial.
Ground-referenced analog inputs. If one side of your signal has a direct
connection to a reliable ground, use a differential input to avoid ground
loops and provide common-mode noise rejection.
 Inputs and Outputs 295

Floating analog inputs. Battery-powered equipment and instruments

with isolated outputs do not supply a return connection to signal
ground. This is generally good, since a ground loop is easy to avoid.
You can connect such a signal to a referenced single-ended (RSE) input
(Figure 12.13A), an input configuration available on most data acqui-
sition boards, and called RSE in the National Instruments manuals.
Common-mode noise rejection will be acceptable in this case. You can
also use differential inputs, but note that they require the addition of
a pair of leak resistors (Figure 12.13B and C). Every amplifier injects

A. Referenced single-ended connection

Instrumentation
Amplifier
Floating +
source G
–

On-board ground
reference connection

B. Differential connection – DC-coupled source

Instrumentation
Amplifier
Floating +
source G
–

Leak
resistor

C. Differential connection – AC-coupled source

Instrumentation
Amplifier
Floating +
source G
–

Leak
resistors

Figure 12.13 Floating source connections: (A) This is the simplest and most
economical; (B) this is similar but may have noise rejection advantages;
and (C) this is required for ac-coupled sources.
296 Chapter Twelve

a small bias current from its inputs back into the signal connections.
If there is no path to ground, as is the case for a true floating source,
then the input voltage at one or both inputs will float to the amplifier’s
power supply rail voltage. The result is erratic operation because the
amplifier is frequently saturated—operating out of its linear range.
This is a sneaky problem! Sometimes the source is floating and you
don’t even know it. Perhaps the system will function normally for a few
minutes after power is turned on, and then it will misbehave later. Or,
touching the leads together or touching them with your fingers may
discharge the circuit, leading you to believe that there is an intermit-
tent connection. Yikes!
Rule: Use an ohmmeter to verify the presence of a resistive path to
ground on all signal sources.
What is the proper value for a leak resistor? The manual for your
input device may have a recommendation. In general, if the value is too
low, you lose the advantage of a differential input because the input is
tightly coupled to ground through the resistor. You may also overload
the source in the case where leak resistors are connected to both inputs.
If the value is too high, additional dc error voltages may arise due to
input offset current drift (the input bias currents on the two inputs dif-
fer and may vary with temperature). A safe value is generally in the
range of 1 to 100 kΩ. If the source is truly floating, such as battery-
powered equipment, then go ahead and directly ground one side.

Thermocouples. Thermocouples have very low source impedance

and low signal voltage, and they may have either floating or grounded
junctions. For an excellent tutorial on thermocouples, refer to Omega
Engineering’s Temperature Handbook (2001). For general information
on connecting thermocouples to computer I/O, consult National Instru-
ments’ Web site. We generally prefer floating junctions, in which the
welded joint between the thermocouple wires is isolated from the sur-
rounding sheath or other nearby metallic items. This removes another
chance of a ground loop and results in a floating dc-coupled source as
previously described. Always use differential connections to reject noise
on this and other low-level signals, and be sure to use a leak resistor—
perhaps 1 kΩ or so—on one of the inputs. You must also take care to use
extension wire and connectors of the same alloy as that of the thermo-
couple element. Finally, the connections to the signal conditioning must
be isothermal; that is, all connections are at the same temperature.
Thermocouples also require a reference junction measurement to
compensate for the voltage produced at the junction between the ther-
mocouple wires and copper connections. This may be provided by an
additional thermocouple junction immersed in a reference temperature
bath or, more commonly, by measuring the temperature at the reference
temperature block, where the thermocouple alloy transitions to copper,
 Inputs and Outputs 297

typically near the amplifier. All modern signal conditioners include an

isothermal reference temperature feature. The LabVIEW DAQ library
includes VIs for making the corrections and performing linearization.
AC signals. If you want to measure the RMS value of an ac waveform,
you have several choices: Use an external AC to DC converter, digitize
the waveform at high speed and do some signal processing in LabVIEW,
use a plug-in digital multimeter board such as the National Instruments
4030, or use a benchtop DMM with a GPIB connection. For low-frequency
voltage measurements, your best bet is a true RMS (TRMS) converter
module, such as those made by Action Instruments and Ohio Semitron-
ics (which also make single- and three-phase ac power transmitters). The
input can be any ac waveform up to a few kilohertz and with amplitudes
as high as 240 V RMS. For higher frequencies, a more sophisticated con-
verter is required, and you may well have to build it yourself, using an
IC such as the Linear Technology LT1088 RMS to DC converter, which
operates from dc to 100 MHz. To measure low-frequency accurrent, use
a current transformer and an ac converter module as before. If the fre-
quency is very high, special current transformers made by Pearson Elec-
tronics are available and can feed a wideband RMS to DC converter. All
these transformers and ac converters give you signals that are isolated
from the source, making connections to your DAQ board simple. When
the shape of the waveform is of interest, you generally connect the signal
directly to the input of a sufficiently fast ADC or digital oscilloscope, and
then you acquire large buffers of data for analysis.
Digital inputs. Digital on/off signals may require conditioning, depend-
ing upon their source. If the source is an electronic digital device, it prob-
ably has TTL-compatible outputs and can be connected directly to the
digital I/O port of a plug-in board. Check the manuals for the devices
at both ends to ensure that both the levels and current requirements
match. Detecting a contact closure (a switch) requires a pull-up resistor
at the input (Figure 12.14) to provide the high level when the switch is

+5 V on plug-in board
Pull-up resistor TTL digital input
10k on plug-in board

Switch contact

Figure 12.14 Contact closures are sensed by a TTL input with

the help of an external pull-up resistor.
298 Chapter Twelve

open. If your source produces a voltage outside of the 0- to 5-VDC range

of a typical plug-in board, additional signal conditioning is required. You
can buy a board such as one of the National Instruments SSR series
which is compatible with high ac and dc voltages. Even better, the SCXI-
1162HV gives you 32 channels of isolated high-voltage inputs.

Outputs need signal conditioning, too. To drive actuators of one type

or another, you may need signal conditioning. As with input signals,
isolation is one of the major considerations, along with requirements
for extra drive capability that many simple output devices can’t handle
directly.
For digital outputs, most plug-in boards offer simple TTL logic drivers,
which swing from 0 to about +3 V, depending on the load (you need to
consult the specifications for your hardware to make sure). If you want
to drive a solenoid valve, for instance, much more current is needed—
perhaps several amperes at 24 V dc or even 120 V ac. In such cases,
you need a relay of some type. Electromechanical relays are simple and
inexpensive and have good contact ratings, but sometimes they require
more coil current than a TTL output can supply. Sensitive relays, such
as reed relays, are often acceptable. Solid-state relays use silicon con-
trolled rectifiers (SCRs) or triacs to control heavier loads with mini-
mal control current requirements. Their main limitation is that most
units are only usable for ac circuits. Many types of modular I/O systems
have options for these higher voltages and currents, such as the SCXI-
1161 with eight power relays, which you can connect directly to your
plug-in board. If you need to drive other types of logic or require large
output voltages and/or currents at high speed, then a special interface
circuit may have to be custom-designed. Figure 12.15 shows a couple of
simple options for driving light-emitting diodes (LEDs) and heavier dc
loads, such as solenoids, using MOSFETs to handle the power. National
Instruments sells a variety of digital signal conditioners that plug right
into its boards; use them first because they’re convenient.

Driving LEDs. Driving relays or solenoids.

On when output is low.
IN 4002

TTL +5 V TTL MOSFET

Output 300Ω LED Output + Coil
Voltage
Coil

Figure 12.15 Simple circuits you can use with digital outputs on MIO and DIO
series boards. Drive low-current loads like LEDs directly (left). MOSFETs are
available in a wide range of current and voltage ratings for driving heavy dc
loads (right).
 Inputs and Outputs 299

One warning is in order regarding digital outputs. When you turn

on a digital output device, be it a plug-in board or an external module,
which output state will it produce—on, off, or high-impedance (discon-
nected)? Can it produce a brief but rapid pulse train? We’ve seen every-
thing, and you will, too, given time. Connect an oscilloscope to each
output and see what happens when you turn the system on and off.
If the result is unsatisfactory (or unsafe!), you may need a switch or
relay to temporarily connect the load to the proper source. This warn-
ing applies to analog outputs, too.
Most of the analog world gets by with control voltages in the range
of ±10 V or less and control currents of 20 mA or less. These outputs
are commonly available from most I/O boards and modular signal con-
ditioners. If you need more current and/or voltage, you can use a power
amplifier. Audio amplifiers are a quick and easy solution to some of
these problems, although most have a high-pass filter that rolls off the
dc response (that is, they can produce no dc output). Many commercial
dc power supplies have analog inputs and are good for low-frequency
applications. To control big ac heaters, motors, and other heavy loads,
look to triac-based power controllers. They permit an ordinary analog
voltage or current input to control tens or even hundreds of kilowatts
with reasonable cost and efficiency. One thing to watch out for is the
need for isolation when you get near high-power equipment. An isola-
tion amplifier could save your equipment if a fault occurs; it may also
eliminate grounding problems which are so prevalent around high-
power systems.

Choosing the right I/O subsystem

As soon as your project gets underway, make a list of the sensors you
need to monitor along with any actuators you need to drive. Having
exact specifications on sensors and actuators will make your job much
easier. As an instrumentation engineer, Gary spends a large part of
his time on any project just collecting data sheets, calling manufactur-
ers, and peering into little black boxes, trying to elucidate the impor-
tant details needed for interfacing. Understanding the application is
important, too. The fact that a pressure transducer will respond in 1 µs
doesn’t mean that your system will actually have microsecond dynamic
conditions to measure. On the other hand, that superfast transducer
may surprise you by putting out all kinds of fast pulses because there
are some little bubbles whizzing by in the soup. Using the wrong sig-
nal conditioner in either case can result in a phenomenon known as
bogus data.
Since you are a computer expert, fire up your favorite spreadsheet
or database application and make up an instrument list. The process
300 Chapter Twelve

control industry goes so far as to standardize on a format known as

instrument data sheets, which are used to specify, procure, install, and
finally document all aspects of each sensor and actuator. Others work-
ing on your project, particularly the quality assurance staff, will be very
happy to see this kind of documentation. For the purposes of designing
an I/O subsystem, your database might include these items:

■ Instrument name or identifier

■ Location, purpose, and references to other drawings such as wiring
and installation details
■ Calibration information: engineering units (such as psig) and full-
scale range
■ Accuracy, resolution, linearity, and noise, if significant
■ Signal current, voltage, or frequency range
■ Signal bandwidth
■ Isolation requirements
■ Excitation or power requirements: current and voltage

To choose your I/O subsystem, begin by sorting the instrument list

according to the types of signal and other basic requirements. Remem-
ber to add plenty of spare channels! Consider the relative importance of
each instrument. If you have 99 thermocouples and 1 pressure gauge,
your I/O design choice will certainly lean toward accommodating ther-
mocouples. But that pressure signal may be the single most important
measurement in the whole system; don’t try to adapt its 0- to 10-V
output to work with an input channel that is optimized for microvolt
thermocouple signals. For each signal, determine the minimum speci-
fications for its associated signal conditioner. Important specifications
are as follows:

■ Adjustability of zero offset and gain

■ Bandwidth—minimum and maximum frequencies to pass
■ Filtering—usually antialiasing low-pass filters
■ Settling time and phase shift characteristics
■ Accuracy
■ Gain and offset drift with time and temperature (very important)
■ Excitation—built in or external?
■ For thermocouples, cold-junction compensation
■ Linearization—may be better performed in software
 Inputs and Outputs 301

Next, consider the physical requirements. Exposure to weather, high

temperatures, moisture and other contamination, or intense electro-
magnetic interference may damage unprotected equipment. Will the
equipment have to work in a hostile environment? If so, it must be in
a suitable enclosure. If the channel count is very high, having many
channels per module could save both space and money. Convenience
should not be overlooked: Have you ever worked on a tiny module with
itty-bitty terminal screws that are deeply recessed into an overstuffed
terminal box? This is a practical matter; if you have lots of signals to
hook up, talk this over with the people who will do the installation.
If your company already has many installations of a certain type or
I/O, that may be an overriding factor, as long as the specifications are
met. The bottom line is always cost. Using excess or borrowed equip-
ment should always be considered when money is tight. You can do a
cost-per-channel analysis, if that makes sense. For instance, using a
multifunction board in your computer with just a few channels hooked
up through the Analog Devices 5B series signal conditioners is very
cost-effective. But if you need 100 channels, using SCXI would certainly
save money over using the Analog Devices 5Bs. (See Figure 12.16.)

Remote and distributed I/O.Your sensors may not be located close to

your computer system, or they may be in an inaccessible area—a haz-
ardous enclosure or an area associated with a high-voltage source.

Figure 12.16 EG&G Automotive Research (San Antonio, Texas) has the type of industrial
environment that makes effective use of SCXI signal conditioning products. (Photo cour-
tesy of National Instruments and EG&G Automotive Research.)
302 Chapter Twelve

In such cases, remote I/O hardware is appropriate. It makes sense

to locate the acquisition hardware close to groups of sensors—remote
from the computer—because it saves much signal wiring which in turn
reduces cost and reduces the chance of noise pickup. Many remote I/O
systems are commercially available with a variety of communication
interfaces. If more than one I/O subsystem can be connected to a com-
mon communication line, it is often referred to as distributed I/O.
National Instruments carries the Analog Devices 6B series dis-
tributed I/O system. This family includes a wide variety of analog and
digital I/O modules that plug into backplanes that hold 1, 4, or 16 mod-
ules. Backplanes are connected to the host computer via RS-232 or
RS- 485 serial lines, as shown in Figure 12.17. Most computers have
RS-232 built in, and it is adequate for short distances. For RS-485, you
can install an AT-485 RS-485 interface and gain the ability to extend
your reach to 4000 ft. LabVIEW drivers are available for all platforms.
See the National Instruments catalog for details.
A similar distributed I/O system is made by Opto-22. As with the
Analog Devices 6B series, you choose from a variety of analog and
digital I/O modules that are installed in backplanes of various types
and sizes. A brain board (model B1 for digital and B2 for analog I/O)

Host Computer 16-Channel Backplane

(PC, Macintosh, Sun) (6BP16-2)
T
me
ep
tar
eru

lF
wo
C
tno
lor
aP
len
rP
sse
eru
alA
mr
oC
idn
oit
sn

TS
RS-232
PO

RS-485

4-Channel
Backplane
(6BP04-1)

6B50-1

To SSR Series
1-Channel Backplane,
Backplanes SC-2051, or CB-50
(6BP01-1)

Figure 12.17 The Analog Devices 6B series, available through National Instruments,
provides distributed I/O capability with serial communications.
 Inputs and Outputs 303

connects one or more backplanes to an RS-422 serial line running the

Optomux protocol. A free LabVIEW driver is available from National
Instruments. We’ve found Optomux equipment to be a simple, reliable,
low-cost solution for connecting remote sensors and systems around
the lab.
SCXI equipment can be remotely operated over serial links, spe-
cifically RS-232 or RS-485, up to 1.2 km. You can use the SCXI-2000
four-slot chassis with built-in communications, or the SCXI-1124 com-
munications module in any of the standard SCXI chasses, and plug in
all the usual modules. SCXI, of course, maintains transparency with
regard to the NI DAQ driver configurations, regardless of the fact that
it’s connected over a serial link. You may sacrifice some performance,
however.
National Instruments has a new line of modular I/O called Field-
Point. It’s designed primarily for industrial process control applica-
tions, as is the Analog Devices 6B module. Dozens of I/O modules are
available, covering every type of analog and digital signal. It’s ideal
when you need to put the I/O interface close to the process, concentrat-
ing those interfaces in clusters that are linked over a network. Field-
Point features several networking options: RS-232 (including wireless),
RS-485, TCP/IP over Ethernet (10 or 100 Mb/s), and Foundation Field-
bus. There’s also a smart FieldPoint controller module that works as
a LabVIEW RT target over Ethernet. Thus you can create your own
real-time industrial controller—programmed in LabVIEW rather than
ladder logic or C—that snaps on a DIN rail. Now that’s versatile. The
only limitation with FieldPoint at this time is that the driver software
is available only for Windows.

Network everything!
Ethernet connections are just about everywhere these days, including
wireless links. There are many ways that you can take advantage of
these connections when configuring a distributed measurement and
control system, courtesy of National Instruments. Here are a few con-
figurations and technologies to consider.
■ Data can be published via DataSockets, a cross-platform technol-
ogy supported by LabVIEW that lets you send any kind of data to
another LabVIEW-based system or to a Web browser. It’s as easy as
writing data to a file, except it is being spun out on the Web. And it’s
as easy to access as a file, too.
■ Remote data acquisition (RDA) is an extension of the NI DAQ
driver for Windows where a DAQ board plugged into one computer
is accessible in real time over the network on another computer.
304 Chapter Twelve

For instance, you could have a dedicated laboratory system connected

to an experiment, while back in your office you can configure and run
data acquisition operations (from LabVIEW, of course) as if the DAQ
board were plugged into your desktop machine.
■ Ethernet adapters, such as the GPIB-ENET, are available for GPIB
instruments. Also, many new rack-and-stack instruments are featur-
ing Ethernet interfaces. That lets you extend the GPIB bus as far as
you want—perhaps around the world—with nothing more than an
Ethernet jack required on your computer.

Now you have your signals connected to your computer. In Chapter

13, we’ll take a closer look at how those signals are sampled or digitized
for use in LabVIEW.

Bibliography
Beckwith, Thomas G. and R. D. Marangoni, Mechanical Measurements, Addison-Wesley,
Reading, Massachusetts, 1990. (ISBN 0-201-17866-4) Gunn, Ronald, “Designing Sys-
tem Grounds and Signal Returns,” Control Engineering, May, 1987.
Lancaster, Donald, Active Filter Cookbook, Howard W. Sams & Co., Indianapolis, 1975.
(ISBN 0-672-21168-8) Lipshitz, Stanley P., R. A. Wannamaker, and J. Vanderkooy, “Quan-
tization and Diter: A Theoretical Survey,” J. Audio Eng. Soc., 40:(5):355–375 (1992).
Morrison, Ralph, Grounding and Shielding Techniques in Instrumentation, Wiley-
Interscience, New York, 1986.
Norton, Harry R. Electronic Analysis Instruments, Prentice-Hall, Englewood Cliffs,
New Jersey, 1992. (ISBN 0-13-249426-4) Omega Engineering, Inc., Temperature
Handbook, Stamford, Connecticut, 2000. (Available free by calling 203-359-1660 or
800-222-2665.)
Ott, Henry W., Noise Reduction Techniques in Electronic Systems, John Wiley & Sons,
New York, 1988. (ISBN 0-471-85068-3) Pallas-Areny, Ramon and J. G.Webster, Sen-
sors and Signal Conditioning, John Wiley & Sons, New York, 1991. (ISBN 0-471-
54565-1) Qian, Shie and Dapang Chen, Joint Time-Frequency Analysis—Methods and
Applications,
Prentice-Hall, Englewood Cliffs, New Jersey, 1996. (ISBN 0-13-254384-2. Call Prentice-
Hall at 800-947-7700 or 201-767-4990.)
Sheingold, Daniel H., Analog-Digital Conversion Handbook, Prentice-Hall, Englewood
Cliffs, New Jersey, 1986. (ISBN 0-13-032848-0) Steer, Robert W., Jr., “Anti-aliasing
Filters Reduce Errors in ADC Converters,” EDN, March 30, 1989.
3M Specialty Optical Fibers, Fiber Optic Current Sensor Module (product information
and application note),West Haven, Connecticut, (203) 934-7961.
White, Donald R. J. Shielding Design Methodology and Procedures, Interference Control
Technologies, Gainesville, Virginia, 1986. (ISBN 0-932263-26-7)
 Chapter

Sampling Signals
13
Up until now, we’ve been discussing the real (mostly analog) world of
signals. Now it’s time to digitize those signals for use in LabVIEW. By
definition, analog signals are continuous-time, continuous-value
functions. That means they can take on any possible value and are
defined over all possible time resolutions. (By the way, don’t think that
digital pulses are special; they’re just analog signals that happen to be
square waves. If you look closely, they have all kinds of ringing, noise,
and slew rate limits—all the characteristics of analog signals.)
An analog-to-digital converter (ADC) samples your analog signals
on a regular basis and converts the amplitude at each sample time to
a digital value with finite resolution. These are termed discrete-time,
discrete-value functions. Unlike their analog counterparts, discrete
functions are defined only at times specified by the sample interval
and may only have values determined by the resolution of the ADC. In
other words, when you digitize an analog signal, you have to approxi-
mate. How much you can throw out depends on your signal and your
specifications for data analysis. Is 1 percent resolution acceptable? Or
is 0.0001 percent required? And how fine does the temporal resolution
need to be? One second? Or 1 ns? Please be realistic. Additional ampli-
tude and temporal resolution can be expensive. To answer these ques-
tions, we need to look at this business of sampling more closely.

Sampling Theorem
A fundamental rule of sampled data systems is that the input signal
must be sampled at a rate greater than twice the highest-frequency
component in the signal. This is known as the Shannon sampling
theorem, and the critical sampling rate is called the Nyquist rate.

305
306 Chapter Thirteen

Actual 1 kHz Signal

Sampled at 1.2 kHz
Sampled at 5.5 kHz

0 1 2 3 4 5
Time (ms)
Figure 13.1 Graphical display of the effects of sampling rates. When the
original 1-kHz sine wave is sampled at 1.2 kHz (too slow), it is totally unrec-
ognizable in the data samples. Sampling at 5.5 kHz yields a much better rep-
resentation. What would happen if there was a lot of really high-frequency
noise?

Stated as a formula, it says that fs /2 > fa, where fs is the sampling fre-
quency and fa is the maximum frequency of the signal being sampled.
Violating the Nyquist criterion is called undersampling and results in
aliasing. Look at Figure 13.1 which simulates a sampled data system.
I started out with a simple 1-kHz sine wave (dotted lines), and then I
sampled it at two different frequencies, 1.2 and 5.5 kHz. At 5.5 kHz, the
signal is safely below the Nyquist rate, which would be 2.75 kHz, and
the data points look something like the original (with a little bit of infor-
mation thrown out, of course). But the data with a 1.2-kHz sampling
rate is aliased: It looks as if the signal is 200 Hz, not 1 kHz. This effect
is also called frequency foldback: Everything above fs/2 is folded back
into the sub-fs/2 range. If you undersample your signal and get stuck
with aliasing in your data, can you undo the aliasing? In most cases,
no. As a rule, you should not undersample if you hope to make sense
of waveform data. Exceptions do occur in certain controlled situations.
An example is equivalent-time sampling in digital oscilloscopes, in
which a repetitive waveform is sampled at a low rate, but with careful
control of the sampling delay with respect to a precise trigger.
Let’s go a step further and consider a nice 1-kHz sine wave, but this
time we add some high-frequency noise to it. We already know that a
5.5-kHz sample rate will represent the sine wave all right, but any noise
that is beyond fs/2 (2.75 kHz) will alias. Is this a disaster? That depends
on the power spectrum (amplitude squared versus frequency) of the
noise or interfering signal. Say that the noise is very, very small in
amplitude—much less than the resolution of your ADC. In that case
it will be undetectable, even though it violates the Nyquist criterion.
 Sampling Signals 307

10

0
Nyquist Limit (fs /2)
-10
1 KHz Signal
-20
Aliasing occurs
-30
Power (dB)

-40

-50
This area represents aliased energy
-60

-70

-80

-90

-100dB
0 1000 2000 3000 4000 5000 Hz
Frequency

Figure 13.2 Power spectrum of a 1-kHz sine wave with low-pass-filtered noise added. If the ADC
resolves 16 bits, its spectral noise floor is ≈115 dB. Assuming that we sample at 5.5 kHz, any
energy appearing above fs /2 (2.75 kHz) and above ≈115 dB will be aliased.

The real problems are medium-amplitude noise or spurious signals.

In Figure 13.2, I simulated a 1-kHz sine wave with low-pass-filtered
white noise added to it. If we use a 16-bit ADC, the specifications say its
spectral noise floor is about −115 dB below full-scale when displayed as
a power spectrum. (This corresponds to an rms signal-to-noise ratio of
98.08 dB for all frequencies up to the Nyquist limit.) Assuming that the
signal is sampled at 5.5 kHz as before, the Nyquist limit is 2.75 kHz.
Looking at the power spectrum, you can see that some of the noise
power is above the floor for the ADC, and there is also noise present
at frequencies above 2.75 kHz. The shaded triangle represents aliased
energy and gives you a qualitative feel for how much contamination
you can expect. Exactly what the contamination will look like is any-
body’s guess; it depends on the nature of the out-of-band noise. In this
case, you can be pretty sure that none of the aliased energy will be
above −65 dB. The good news is that this is just plain old uncorrelated
noise, so its aliased version will also be plain old uncorrelated noise. It
is not a disaster, just another noise source.

Filtering and Averaging

To get rid of aliasing problems, you need to use a low-pass filter,
known in this case as an antialiasing filter. Analog filters are man-
datory, regardless of the sampling rate, unless you know the signal’s
frequency characteristics and can live with the aliased noise. There
has to be something to limit the bandwidth of the raw signal to fs /2.
The analog filter can be in the transducer, in the signal conditioner,
308 Chapter Thirteen

on the ADC board, or in all three places. One problem with analog fil-
ters is that they can become very complex and expensive. If the desired
signal is fairly close to the Nyquist limit, the filter needs to cut off very
quickly, implying lots of stages (this is more formally known as the
order of the filter’s transfer function). High-performance antialias-
ing filters, such as the SCXI-1141, have such high-order designs and
meet the requirements of most situations.
Digital filters can augment, but not replace, analog filters. Digital
filter VIs are included with the LabVIEW analysis library, and they are
functionally equivalent to analog filters. The simplest type of digital
filter is a moving averager (examples of which are available with
LabVIEW) which has the advantage of being usable in real time on
a sample-by-sample basis. One way to simplify the antialiasing filter
problem is to oversample the input. If your ADC hardware is fast
enough, just turn the sampling rate way up, and then use a digital
filter to eliminate the higher frequencies that are of no interest. This
makes the analog filtering problem much simpler because the Nyquist
frequency has been raised much higher, so the analog filter doesn’t
have to be so sharp. A compromise is always necessary: You need to
sample at a rate high enough to avoid significant aliasing with a mod-
est analog filter; but sampling at too high a rate may not be practical
because the hardware is too expensive and/or the flood of extra data
may overload your poor CPU.
A potential problem with averaging arises when you handle non-
linear data. The process of averaging is defined to be the summation
of several values, divided by the number of values. If your data is, for
instance, exponential, then averaging values (a linear operation) will
tend to bias the data. (Consider the fact that ex + ey is not equal to e(x + y).)
One solution is to linearize the data before averaging. In the case of
exponential data, you should take the logarithm first. You may also
be able to ignore this problem if the values are closely spaced—small
pieces of a curve are effectively linear. It’s vital that you understand
your signals qualitatively and quantitatively before you apply any
numerical processing, no matter how innocuous it may seem.
If your main concern is the rejection of 60-Hz line frequency interfer-
ence, an old trick is to average an array of samples over one line period
(16.66 ms in the United States). For instance, you could acquire data at
600 Hz and average groups of 10, 20, 30, and so on, up to 600 samples.
You should do this for every channel. Using plug-in boards with Lab-
VIEW’s data acquisition drivers permits you to adjust the sampling
interval with high precision, making this a reasonable option. Set up
a simple experiment to acquire and average data from a noisy input.
Vary the sampling period, and see if there isn’t a null in the noise level
at each 16.66-ms multiple.
 Sampling Signals 309

If you are attempting to average recurrent waveforms to reduce noise,

remember that the arrays of data that you acquire must be perfectly in
phase. If a phase shift occurs during acquisition, then your waveforms
will partially cancel one another or cause distortion. Triggered data
acquisition (discussed later) is the normal solution because it helps to
guarantee that each buffer of data is acquired at the same part of the
signal’s cycle.
Some other aspects of filtering that may be important for some of
your applications are impulse response and phase response. For
ordinary datalogging, these factors are generally ignored. But if you
are doing dynamic testing such as in vibration analysis, acoustics, or
seismology, then impulse and phase response can be very important. As
a rule of thumb, when filters become very complex (high-order), they
cut off more sharply, have more radical phase shifts around the cutoff
frequency, and, depending on the filter type, exhibit greater ringing on
transients. Overall, filtering is a rather complex topic that is best left
to the references; the Active Filter Cookbook (Lancaster, 1975) is an old
favorite. Even with an electrical engineering degree and lots of experi-
ence with filters, Gary still uses it to get a quick answer to practical
filtering problems.
The best way to analyze your filtering needs is to use a spectrum ana-
lyzer. Then you know exactly what signals are present and what has
to be filtered out. You can use a dedicated spectrum analyzer instru-
ment (very expensive), a digital oscilloscope with FFT capability (or let
LabVIEW do the power spectrum), or even a multifunction I/O board
running as fast as possible with LabVIEW doing the power spectrum.
Spectrum analyzers are included in the data acquisition examples dis-
tributed with LabVIEW, and they work quite well.

About ADCs, DACs, and Multiplexers

Important characteristics of an ADC or a digital-to-analog converter
(DAC) are resolution, range, speed, and sources of error. (For a detailed
look at all these parameters, consult the Analog-Digital Conversion
Handbook, available from Analog Devices.)
Resolution is the number of bits that the ADC uses to represent the
analog signal. The greater the number of bits, the finer the resolution
of the converter. Figure 13.3 demonstrates the resolution of a hypo-
thetical 3-bit converter, which can resolve 23, or 8, different levels. The
sine wave in this figure is not well represented because of the rather
coarse quantization levels available. Common ADCs have resolutions
of 8, 12, and 16 bits, corresponding to 256, 4096, and 65,536 quanti-
zation levels. Using high-resolution converters is generally desirable,
although they tend to be somewhat slower.
310 Chapter Thirteen

5
ADC Counts

-1
0 10 20 30 40 50 60 70 80 90 100
Time (ms)
Figure 13.3 A sine wave and its representation by a 3-bit ADC sampling every 5 ms.

Range refers to the maximum and minimum voltage levels that the
ADC can quantize. Exceeding the input range results in what is vari-
ously termed clipping, saturation, or overflow/underflow, where the
ADC gets stuck at its largest or smallest output code. The code width
of an ADC is defined as the change of voltage between two adjacent
quantization levels or, as a formula,
range
Code width =
2N
where N is the number of bits and code width and range are mea-
sured in volts. A high-resolution converter (lots of bits) has a small code
width. The intrinsic range, resolution, and code width of an ADC can
be modified by preceding it with an amplifier that adds gain. The code
width expression then becomes
range
Code width =
gain × 2 N
High gain thus narrows the code width and enhances the resolution
while reducing the effective range. For instance, a common 12-bit ADC
with a range of 0 to 10 V has a code width of 2.44 mV. By adding a gain
of 100, the code width becomes 24.4 µV, but the effective range becomes
10/100 = 0.1 V. It is important to note the tradeoff between resolution
and range when you change the gain. High gain means that overflow
may occur at a much lower voltage.
Conversion speed is determined by the technology used in design-
ing the ADC and associated components, particularly the sample-and-
hold (S/H) amplifier that freezes the analog signal just long enough to
do the conversion. Speed is measured in time per conversion or samples
 Sampling Signals 311

dB Digits Bits
8.5
168 28

156 26
7.5 Traditional Instruments, 2000
144 24
Dynamic Range or Resolution

132 6.5 22

120 20 A/D Components, 2000

108 5.5 18

96 16
A/D Components, 1995
4.5
84 14

72 12
3.5

60 10

48 2.5 8

1 10 100 1k 10k 100k 1M 10M 100M 1G 10G 100G 1T

Sampling Rate, samples/second
Figure 13.4The performance of A/D components—including plug-in boards—now closely
approaches or exceeds the performance of rack-and-stack instruments, and usually at
much lower cost. (Courtesy of National Instruments.)

per second. Figure 13.4 compares the typical resolutions and conversion
speeds of plug-in ADC boards and traditional instruments. A very com-
mon tradeoff is resolution versus speed; it simply takes more time or is
more costly to precisely determine the exact voltage. In fact, high-speed
ADCs, such as flash converters that are often used in digital oscilloscopes,
decrease in effective resolution as the conversion rate is increased. Your
application determines what conversion speed is required.
There are many sources of error in ADCs, some of which are a little
hard to quantify; in fact, if you look at the specification sheets for ADCs
from different manufacturers, you may not be able to directly compare
the error magnitudes because of the varying techniques the manufac-
turers may have used.
Simple errors are gain and offset errors. An ideal ADC follows the
equation for a straight line, y = mx + b, where y is the input voltage, x is
the output code, and m is the code width. Gain errors change the slope
m of this equation, which is a change in the code width. Offset errors
change the intercept b, which means that 0-V input doesn’t give you
zero counts at the output. These errors are easily corrected through
calibration. Either you can adjust some trimmer potentiometers so
that zero and full-scale match a calibration standard, or you can make
312 Chapter Thirteen

measurements of calibration voltages and fix the data through a sim-

ple straight-line expression in software. Fancy ADC systems, such as
the E-series plug-in boards from National Instruments, include self-
calibration right on the board.
Linearity is another problem. Ideally, all the code widths are the
same. Real ADCs have some linearity errors that make them deviate
from the ideal. One measure of this error is the differential nonlin-
earity, which tells you the worst-case deviation in code width. Integral
nonlinearity is a measure of the ADC transfer function. It is mea-
sured as the worst-case deviation from a straight line drawn through
the center of the first and last code widths. An ideal converter would
have zero integral nonlinearity. Nonlinearity problems are much more
difficult to calibrate out of your system. To do so would mean taking
a calibration measurement at each and every quantization level and
using that data to correct each value. In practice, you just make sure
that the ADC is tightly specified and that the manufacturer delivers
the goods as promised.
If a multiplexer is used before an ADC to scan many channels, tim-
ing skew will occur (Figure 13.5). Since the ADC is being switched

Multiplexer
(switches every 4 µs)
Sig. Gen. Chan. 1
f = 20 kHz Digital
Chan. 2
ADC
Code
Chan. 3

Chan. 4

1.0
Chan. 4
Chan. 3
Chan. 2
0.5
Chan. 1

0.0

-0.5

-1.0
0 20 40 60 80
Time (µsec)
4 µsec skew
Figure 13.5 Demonstration of skew in an ADC with a multiplexer.
Ideally, there would be zero switching time between channels on the
multiplexer; this one has 4 µs. Since the inputs are all the same sig-
nal, the plotted data shows an apparent phase shift.
 Sampling Signals 313

between several inputs, it is impossible for it to make all the measure-

ments simultaneously. A delay between the conversion of each chan-
nel results, and this is called skew. If your measurements depend on
critical timing (phase matching) between channels, you need to know
exactly how much skew there is in your ADC system.
Multiplexers are used as cost-saving devices, since ADCs tend to be
among the most expensive parts in the system, but the skew problem
can be intolerable in some applications. Then multiple ADCs or sample-
and-hold amplifiers (one per channel) are recommended. This is the
approach taken by the National Instruments SCXI-1140 and 4450-
series dynamic signal acquisition boards, respectively.
You can remove the timing skew in software as long as you know
exactly what the skew is. Most plug-in DAQ boards work as follows: By
default, the multiplexer scans through the channel list at the top speed
of the ADC. For instance, if you have a 100-kHz ADC, there will be a
skew of approximately 10 µs between channels. However, the hardware
and/or the driver software may increase this value somewhat due to all
the extra settling time. For best results, you probably will have to acquire
some test data to be sure. You can also choose to adjust the interchan-
nel delay (i.e., the skew) with the DAQ VIs in LabVIEW. That feature
lets you specify the skew, which makes it easier to correct the data’s time
base. But chances are, even with your adjustments and corrections, the
results won’t be quite as accurate as the multiple-ADC approach, and
that’s why we prefer to use an optimum hardware solution.
One misconception about ADCs is that they always do some averag-
ing of the input signal between conversions. For instance, if you are
sampling every millisecond, you might expect the converted value to
represent the average value over the last millisecond of time. This is
true only for certain classes of ADCs, namely, dual-slope integrators
and voltage-to-frequency converters. Ubiquitous high-speed ADCs, as
found on most multifunction boards, are always preceded by sample-
and-hold amplifiers.
The S/H samples the incoming signal for an extremely brief time
(the aperture time) and stores the voltage on a capacitor for measure-
ment by the ADC. The aperture time depends on the resolution of the
ADC and is in the nanosecond range for 16-bit converters. Therefore,
if a noise spike should occur during the aperture time, you will get an
accurate measurement of the spike, not the real waveform. This is one
more reason for using a low-pass filter.
When a multiplexer changes from one channel to another, there is a
period of time during which the output is in transition, known as the
settling time. Because of the complex nature of the circuitry in these
systems, the transition may not be as clean as you might expect. There
may be an overshoot with some damped sinusoidal ringing. Since your
objective is to obtain an accurate representation of the input signal,
314 Chapter Thirteen

you must wait for these aberrations to decay away. Settling time is the
amount of time required for the output voltage to begin tracking the
input voltage within a specified error band after a change of channels
has occurred. It is clearly specified on all ADC system data sheets. You
should not attempt to acquire data faster than the rate determined by
the settling time plus the ADC conversion time.
Another unexpected source of input error is sometimes referred to
as charge pump-out. When a multiplexer switches from one channel
to the next, the input capacitance of the multiplexer (and the next cir-
cuit element, such as the sample-and-hold) must charge or discharge to
match the voltage of the new input signal. The result is a small glitch
induced on the input signal, either positive or negative, depending
upon the relative magnitude of the voltage on the preceding channel.
If the signal lines are long, you may also see ringing. Charge pump-out
effects add to the settling time in an unpredictable manner, and they
may cause momentary overloading or gross errors in high-impedance
sources. This is another reason to use signal conditioning; an input
amplifier provides a buffering action to reduce the glitches.
Many systems precede the ADC with a programmable-gain instru-
mentation amplifier (PGIA). Under software control, you can change
the gain to suit the amplitude of the signal on each channel. A true
instrumentation amplifier with its inherent differential connections is
the predominant type. You may have options, through software regis-
ters or hardware jumpers, to defeat the differential mode or use various
signal grounding schemes, as on the National Instruments multifunc-
tion boards. Study the available configurations and find the one best
suited to your application. The one downside to having a PGIA in the
signal chain is that it invariably adds some error to the acquisition
process in the form of offset voltage drift, gain inaccuracy, noise, and
bandwidth. These errors are at their worst at high-gain settings, so
study the specifications carefully. An old axiom in analog design is that
high gain and high speed are difficult to obtain simultaneously.

Digital-to-analog converters

Digital-to-analog converters (DACs) perform the reverse action of

ADCs: A digital code is scaled to a proportional analog voltage. We use
DACs to generate analog stimuli, such as test waveforms and actuator
control signals. By and large, they have the same general performance
characteristics as do ADCs. Their main limitations, besides the factors
we’ve already discussed, are the settling time and slew rate. When
you make a change in the digital code, the output is expected to change
instantaneously to the desired voltage. How fast the change actu-
ally occurs (measured in volts per second) is the slew rate. Hopefully,
 Sampling Signals 315

a clean, crisp step will be produced. In actuality, the output may over-
shoot and ring for awhile or may take a more leisurely, underdamped
approach to the final value. This represents the settling time. If you
are generating high-frequency signals (audio or above), you need
faster settling times and slew rates. If you are controlling the current
delivered to a heating element, these specifications probably aren’t
much of a concern.
When a DAC is used for waveform generation, it must be followed
by a low-pass filter, called a reconstruction filter, that performs an
antialiasing function in reverse. Each time the digital-to-analog (D/A)
output is updated (at an interval determined by the time base), a step
in output voltage is produced. This step contains a theoretically infinite
number of harmonic frequencies. For high-quality waveforms, this out-
of-band energy must be filtered out by the reconstruction filter. DACs
for audio and dynamic signal applications, such as National Instru-
ments’ 4450 series dynamic signal I/O boards, include such filters and
have a spectrally pure output. Ordinary data acquisition boards gener-
ally have no such filtering and will produce lots of spurious energy. If
spectral purity and transient fidelity are important in your applica-
tion, be mindful of this fact.

Digital codes
The pattern of bits—the digital word—used to exchange information
with an ADC or DAC may have one of several coding schemes, some
of which aren’t intuitive. If you ever have to deal directly with the I/O
hardware (especially in lower-level driver programs), you will need to
study these schemes. If the converter is set up for unipolar inputs (all-
positive or all-negative analog voltages), the binary coding is straight-
forward, as in Table 13.1. But to represent both polarities of numbers
for a bipolar converter, a sign bit is needed to indicate the signal’s

TABLE 13.1 Straight Binary Coding Scheme for Unipolar, 3-Bit Converter
Decimal fraction of full-scale
Decimal equivalent Positive Negative Straight binary
7 ⁄
78 –7⁄8 111
6 6⁄8 –6⁄8 110
5 5⁄8 –5⁄8 101
4 4⁄8 –4⁄8 100
3 3⁄8 –3⁄8 011
2 2⁄8 –2⁄8 010
1 1⁄8 –1⁄8 001
0 0⁄8 –0⁄8 000
316 Chapter Thirteen

TABLE 13.2 Some Commonly Used Coding Schemes for Bipolar Converters, a 4-Bit Example
Decimal equivalent Fraction of full-scale Sign and magnitude Two’s complement Offset binary
7 ⁄
78 0111 0111 1111
6 6⁄8 0110 0110 1110
5 5⁄8 0101 0101 1101
4 4⁄8 0100 0100 1100
3 3⁄8 0011 0011 1011
2 2⁄8 0010 0010 1010
1 1⁄8 0001 0001 1001
0 0+ 0000 0000 1000
0 0– 1000 0000 1000
–1 –1⁄8 1001 1111 0111
–2 –2⁄8 1010 1110 0110
–3 –3/8 1011 1101 0101
–4 –4⁄8 1100 1100 0100
–5 –5⁄8 1101 1011 0011
–6 –6⁄8 1110 1010 0010
–7 –7⁄8 1111 1001 0001
–8 –8⁄8 Not represented 1000 0000

NOTE: The sign and magnitude scheme has two representations for zero, but can’t represent –8. Also, the only differ-
ence between offset binary and two’s complement is the polarity of the sign bit.

polarity. The bipolar coding schemes shown in Table 13.2 are widely
used. Each has advantages, depending on the application.

Triggering and Timing

Triggering refers to any method by which you synchronize an ADC or
DAC to some event. If there is a regular event that causes each individ-
ual ADC, it’s called the time base, or clock, and it is usually generated
by a crystal-controlled clock oscillator. For this discussion, we’ll define
triggering as an event that starts or stops a series of conversions which
are individually paced by a time base.
When should you bother with triggering? One situation is when you
are waiting for a transient event to occur—a single pulse. It would be
wasteful (or maybe impossible) to run your ADC for a long time, filling
up memory and/or disk space, when all you are interested in is a short
burst of data before and/or after the trigger event. Another use for trig-
gering is to force your data acquisition to be in phase with the signal.
Signal analysis may be simplified if the waveform always starts with
the same polarity and level. Or, you may want to acquire many buffers
of data from a recurrent waveform (such as a sine wave) in order to
average them, thus reducing the noise. Trigger sources come in three
flavors: external, internal, or software-generated.
External triggers are digital pulses, usually produced by special-
ized hardware or a signal coming from the equipment that you are
 Sampling Signals 317

interfacing with. An example is a function generator that has a connec-

tor on it called sync which produces a TTL pulse every time an output
waveform crosses 0 V in the positive direction. Sometimes you have to
build your own trigger generator. When you are dealing with pulsed
light sources (such as some lasers), an optical detector such as a photo-
diode can be used to trigger a short but high-speed burst of data acqui-
sition. Signal conditioning is generally required for external triggers
because most data acquisition hardware demands a clean pulse with a
limited amplitude range. Chapter 19, “Physics Applications,” goes into
detail on external triggering.
Internal triggering is built into many data acquisition devices,
including oscilloscopes, transient recorders, and multifunction boards.
It is basically an analog function in which a device called a compara-
tor or discriminator detects the signal’s crossing of a specified level.
The slope of the signal may also be part of the triggering criteria. Really
sophisticated instruments permit triggering on specified patterns,
which is especially useful in digital logic and communications signal
analysis. The onboard triggering features of National Instruments’
boards are easy to use in LabVIEW, courtesy of the data acquisition
VIs. Newer boards include an advanced system timing controller chip,
the DAQ-STC, with programmable function inputs that solve some of
the more difficult triggering problems you may encounter.
Software-generated triggers require a program that evaluates an
incoming signal or some other status information and decides when to
begin saving data or generating an output. A trivial example is the Run
button in LabVIEW that starts up a simple data acquisition program.
In that case, the operator is the triggering system. On-the-fly signal
analysis is a bit more complex and quickly runs into performance prob-
lems if you need to look at fast signals. For instance, you may want
to save data from a spectrum analyzer only when the process tem-
perature goes above a certain limit. That should be no problem, since
the temperature probably doesn’t change very fast. A difficult problem
would be to evaluate the distortion of an incoming audio signal and
save only the waveforms that are defective. That might require DSP
hardware; it might not be practical at all, at least in real time. The NI
DAQ driver includes basic software trigger functionality, such as level
and slope detection, that works with many plug-in boards and is very
easy to use.

A Little Noise Can Be a Good Thing

Performance of an analog-to-digital (A/D) or D/A system can be enhanced
by adding a small amount of noise and by averaging (Lipshitz, Wanna-
maker, and Vanderkooy 1992). Any time the input voltage is somewhere
318 Chapter Thirteen

between two quantization levels of the A/D system and there is some
dither noise present, the least-significant bit (LSB) tends to toggle
among a few codes. For instance, the duty cycle of this toggling action
is exactly 50 percent if the voltage is exactly between the two quanti-
zation levels. Duty cycle and input voltage track each other in a nice,
proportional manner (except if the converter demonstrates some kind
of nonlinear behavior). All you have to do is to filter out the noise, which
can be accomplished by averaging or other forms of digital filtering.
A source of uncorrelated dither noise, about 1 LSB peak to peak or
greater, is required to make this technique work. Some high-performance
ADC and DAC systems include dither noise generators; digital audio
systems and the National Instruments dynamic signal acquisition
boards are examples. High-resolution converters (16 bits and greater)
generally have enough thermal noise present to supply the necessary
dithering. Incidentally, this resolution enhancement occurs even if you
don’t apply a filter; filtering simply reduces the noise level.
Figure 13.6 demonstrates the effect of dither noise on the quantiza-
tion of a slow, low-amplitude ramp signal. To make this realistic, say
that the total change in voltage is only about 4 LSBs over a period
of 10 s (Figure 13.6A). The vertical axis is scaled in LSBs for clarity.

5 A 5 C
4 4
3 3
2 Actual analog signal, 2
without noise. Actual analog signal,
1 1 with 1.5 LSB peak-peak noise.
0 0
0 2 4 6 8 10s 0 2 4 6 8 10s

5 5
B D
4 4
3 3
2 Digitized signal. Quantization 2
Digitized signal. Quantization
1 steps are objectionable and 1 steps are mixed in noise.
independent of sample rate. Sampling freq = 20 Hz.
0 0
0 2 4 6 8 10s 0 2 4 6 8 10s

5
E
4
3
2
1 Recovered signal, averaged every 0.5 s.
0
0 2 4 6 8 10s

Figure 13.6 Graphs in (A) and (B) represent digitization of an ideal analog ramp of 4-LSB
amplitude which results in objectionable quantization steps. Adding 1-LSB peak-peak
dither noise and low-pass filtering [graphs in (C) through (E)] improves results.
 Sampling Signals 319

The sampling rate is 20 Hz. In Figure 13.6B, you can see the coarse
quantization steps expected from an ideal noise-free ADC. Much imagi-
nation is required to see a smooth ramp in this graph. In Figure 13.6C,
dither noise with an amplitude of 1.5 LSB peak to peak has been added
to the analog ramp. Figure 13.6D is the raw digitized version of this
noisy ramp. Contrast this with Figure 13.6B, the no-noise case with its
coarse steps. To eliminate the random noise, we applied a simple boxcar
filter where every 10 samples (0.5 s worth of data) is averaged into a
single value; more elaborate digital filtering might improve the result.
Figure 13.6E is the recovered signal. Clearly, this is an improvement
over the ideal, noiseless case, and it is very easy to implement in Lab-
VIEW. In Chapter 14, “Writing a Data Acquisition Program,” we’ll show
you how to oversample and average to improve your measurements.
Low-frequency analog signals give you some opportunities to further
improve the quality of your acquired data. At first glance, that ther-
mocouple signal with a sub-1-Hz bandwidth and little noise could be
adequately sampled at 2 or 3 Hz. But by oversampling—sampling at
a rate several times higher than the minimum specified by the Nyquist
rate—you can enhance resolution and noise rejection.
Noise is reduced in proportion to the square root of the number of
samples that are averaged. For example, if you average 100 samples,
the standard deviation of the average value will be reduced by a factor
of 10 when compared to a single measurement. Another way of express-
ing this result is that you get a 20-dB improvement in signal-to-noise
ratio when you average 100 times as many samples. This condition
is true as long as the A/D converter has good linearity and a small
amount of dither noise. This same improvement occurs with repetitive
signals, such as our 1-kHz sine wave with dither noise. If you syn-
chronize your ADC with the waveform by triggering, you can average
several waveforms. Since the noise is not correlated with the signal,
the noise once again averages to zero according to the square root rule.
How you do this in LabVIEW is discussed in Chapter. 14, “Writing a
Data Acquisition Program.”

Throughput
A final consideration in your choice of converters is the throughput
of the system, a yardstick for overall performance, usually measured
in samples per second. Major factors determining throughput are as
follows:
■ A/D or D/A conversion speed
■ Use of multiplexers and amplifiers, which may add delays between
channels
320 Chapter Thirteen

■ Disk system performance, if streaming data to or from disk

■ Use of direct memory access (DMA), which speeds data transfer
■ CPU speed, especially if a lot of data processing is required
■ Operating system overhead

The glossy brochure or data sheet you get with your I/O hardware
rarely addresses these very real system-oriented limitations. Maxi-
mum performance is achieved when the controlling program is written
in assembly language, one channel is being sampled with the amplifier
gain at minimum, and data is being stored in memory with no analy-
sis or display of any kind. Your application will always be somewhat
removed from this particular benchmark.
Practical disk systems have many throughput limitations. The disk
itself takes a while to move the recording heads around and can only
transfer so many bytes per second. The file system and any data conver-
sion you have to do in LabVIEW are added as overhead. If everything
is working well, a stream-to-disk LabVIEW VI will continuously run in
excess of 6 Mbyes/s. For really fast I/O, you simply have to live within
the limits of available memory. But memory is cheap these days, and
the performance is much better than that of any disk system. If you
really need 1 Gbyte of RAM to perform your experiment, then don’t fool
around, just buy it. Tell the purchasing manager that Gary said so.
Double-buffered DMA for acquisition and waveform generation is
built into the LabVIEW support for many I/O boards and offers many
advantages in speed because the hardware does all the real-time work
of transferring data from the I/O to main memory. Your program can per-
form analysis, display, and archiving tasks while the I/O is in progress
(you can’t do too much processing though). Refer to your LabVIEW data
acquisition VI library reference manual for details on this technique.
Performance without DMA is quite limited. Each time you send a
command to a plug-in board, the command is processed by the NI DAQ
driver, which in turn must get permission from the operating system to
perform an I/O operation. There is a great deal of overhead in this pro-
cess, typically on the order of a few hundred microseconds. That means
you can read one sample at a time from an input at a few kilohertz
without DMA. Clearly, this technique is not very efficient and should
be used only for infrequent I/O operations.
If you need to do very much on-the-fly analysis, adding a DSP board
can augment the power of your computer’s CPU by off-loading tasks
such as FFT computations. Using a DSP board as a general-purpose
computer is another story—actually another book! Programming such
a machine to orchestrate data transfers, do control algorithms, and so
on generally requires programming in C or assembly language, using
 Sampling Signals 321

the support tools for your particular DSP board. If you are an experi-
enced programmer, this is a high-performance alternative.
Access to external DSP horsepower without low-level programming
is available from Sheldon Instruments with its LabVIEW add-on
product called QuVIEW. It works with Sheldon Instruments’ PCI-
based DSP boards that feature AT&T’s 32C or TI’s C3x processors
coupled to multifunction I/O modules. With QuVIEW, you use a col-
lection of VIs that automatically download code to the DSP memory
for fast execution independent of the host PC. It’s especially good for
continuous algorithms such as signal generation, filtering, FFTs, and
streaming data to disk. If the extensive library of built-in functions
isn’t enough, you can create your own external code, using C compil-
ers or assemblers from TI or Tartan, or graphical DSP code generators
from MathWorks or Hyperception. In either case, your external code is
neatly encapsulated in familiar VI wrappers for easy integration into
a LabVIEW application.

Bibliography
3M Specialty Optical Fibers: Fiber Optic Current Sensor Module (product information
and application note), West Haven, Conn., 1996
Beckwith, Thomas G., and R. D. Marangoni: Mechanical Measurements, Addison-Wesley,
Reading, Mass., 1990.
Gunn, Ronald: “Designing System Grounds and Signal Returns,” Control Engineering,
May 1987.
Lancaster, Donald: Active Filter Cookbook, Howard W. Sams & Co., Indianapolis, Ind.,
1975.
Lipshitz, Stanley P., R. A. Wannamaker, and J. Vanderkooy: “Quantization and Dither: A
Theoretical Survey,” J. Audio Eng. Soc., vol. 40, no. 5, 1992, pp. 355–375.
Morrison, Ralph: Grounding and Shielding Techniques in Instrumentation, Wiley-
Interscience, New York, 1986.
Norton, Harry R.: Electronic Analysis Instruments, Prentice-Hall, Englewood Cliffs, N.J.,
1992.
Omega Engineering, Inc.: Temperature Handbook, Stamford, Conn., 2000. (Available free
by calling 203-359-1660 or 800-222-2665.)
Ott, Henry W.: Noise Reduction Techniques in Electronic Systems, Wiley, New York,
1988.
Pallas-Areny, Ramon, and J. G. Webster: Sensors and Signal Conditioning, Wiley,
New York, 1991.
Qian, Shie, and Dapang Chen: Joint Time-Frequency Analysis—Methods and Applica-
tions, Prentice-Hall, Englewood Cliffs, N.J., 1996. (Call Prentice-Hall at 800-947-7700
or 201-767-4990.)
Sheingold, Daniel H.: Analog-Digital Conversion Handbook, Prentice-Hall, Englewood
Cliffs, N.J., 1986.
Steer, Robert W., Jr.: “Anti-aliasing Filters Reduce Errors in ADC Converters,” Electronic
Design News (EDN), March 30, 1989.
White, Donald R. J.: Shielding Design Methodology and Procedures, Interference Control
Technologies, Gainesville, Va., 1986.
This page intentionally left blank
 Chapter

Writing a Data Acquisition Program

14
We’re going to bet that your first LabVIEW application was (or will be)
some kind of data acquisition system. We say that because data acqui-
sition is by far the most common LabVIEW application. Every experi-
ment or process has signals to be measured, monitored, analyzed, and
logged, and each signal has its own special requirements. Although it’s
impossible to design a universal data acquisition system to fit every
situation, there are plenty of common architectures that you can use,
each containing elements that you can incorporate into your own
problem-solving toolkit.
As we mentioned in the Introduction, what you’ll learn in this chap-
ter are all the other things that the LabVIEW manuals and examples
don’t cover. The act of fetching data from an input device is the easy
part, and it’s a subject that is already well discussed. On the other
hand, how do you keep track of channel assignments and other configu-
ration information? And how does data analysis affect program design?
These topics are important, yet rarely mentioned. It’s time to change
all that.
Your data acquisition application might include some control (out-
put) functionality as well. Most experiments have some things that
need to be manipulated—some valves, a power supply set point, or
maybe a motor. That should be no problem as long as you spend some
time designing your program with the expectation that you will need
some control features. If your situation requires a great deal of control
functionality, read Chapter 18, “Process Control Applications,” to see
some methods that may work better than simply adding on to a data
acquisition design.
Plan your application as you would any other, going through the
recommended steps. We’ve added a few items to the procedure that

323
324 Chapter Fourteen

emphasize special considerations for data acquisition: data analysis,

throughput, and configuration management requirements. Here
are the basic steps.

1. Define and understand the problem; define the signals and deter-
mine what the data analysis needs are.
2. Specify the type of I/O hardware you will need, and then determine
sample rates and total throughput.
3. Prototype the user interface and decide how to manage
configurations.
4. Design and then write the program.

If your system requires extra versatility, such as the ability to quickly

change channel assignments or types of I/O hardware, then you will
need to include features to manage the system’s configuration. Users
should be able to access a few simple controls rather than having to
edit the diagram when a configuration change is needed.
Data throughput, the aggregate sampling rate measured in sam-
ples per second, plays a dominant role in determining the architecture
of your program. High sampling rates can severely limit your ability
to do real-time analysis and graphics. Even low-speed systems can be
problematic when you require accurate timing. Fortunately, there are
plenty of hardware and software solutions available in the LabVIEW
world.
The reason for assembling a data acquisition system is to acquire
data, and the reason for acquiring data is to analyze it. Surprisingly,
these facts are often overlooked. Planning to include analysis features
and appropriate file formats will save you (and the recipients of your
data) a lot of grief.
The canonical LabVIEW VI for a simple, yet complete, data acquisi-
tion program is shown in Figure 14.1. It begins with a VI that handles
I/O configuration and the opening of any data files. The rest of the
program resides in a While Loop that cycles at a rate determined by
the sample interval control. The Read Data VI communicates with
hardware and returns the raw data, which is then analyzed for dis-
play and stored in files. All subVIs that can produce an error condition
are linked by an error I/O cluster, and an error handler tells the user
what went wrong. Simple as this diagram is, it could actually work,
and do some rather sophisticated processing at that. Try simplifying
your next data acquisition problem to this level. Add functionality
as required, but keep things modular so that it’s easy to understand
and modify. This chapter is devoted to the building blocks shown in
Figure 14.1.
 Writing a Data Acquisition Program 325

Initialize
hardware; Graph
Open files
config Handle
data errors
error
path
error

Sample interval RUN

Figure 14.1 A generic data acquisition program includes the functions shown
here.

Data Analysis and Storage

Data analysis has a different meaning in every application. It depends
on the kind of signals you are faced with. For many applications, it
means calculating simple statistics (minimum, maximum, mean, stan-
dard deviation, etc.) over some period of time. In spectroscopy and chro-
matography, it means peak detection, curve fitting, and integration. In
acoustics and vibration studies, it means Fourier transforms, filtering,
and correlation. Each type of analysis affects your LabVIEW program
design in some way. For instance, doing a fast Fourier transform (FFT)
on a large array in real time requires lots of processing power—your
system could become so burdened that data collection might be dis-
rupted. Such analysis drives the performance requirements of your
VIs. And everyone worries about timing information, both for real-time
analysis and for reading data from files. It’s obvious that your program
has to measure and store time markers reliably and in a format that is
useful to the analysis programs.
What you need to avoid is analysis by accident. Time and again we’ve
seen LabVIEW programs that grab data from the hardware and stuff
it into a file with no thought about compatibility with the analysis pro-
gram. Then the poor analyst has to grind along, parsing the file into
readable pieces, trying to reconstitute important features of the data
set. Sometimes, the important information isn’t available on disk at all,
and you hope that it has been written down somewhere. Disaster! Gas-
tric distress also occurs when a new real-time analysis need crops up
and your program is so inflexible that the new features can’t be added
without major surgery.
We recommend a preemptive strike. When someone proposes a new
data acquisition system, make it a point to force that person to describe,
326 Chapter Fourteen

in detail, how the data will be analyzed. Make sure he or she under-
stands the implications of storing the megabytes or gigabytes of data
that an automated data acquisition system may collect. If there is a
collective shrug of shoulders, ask them point-blank, “. . . Then why are
we collecting data at all?” Do not write your data acquisition program
until you understand the analysis requirements.
Finally, you can get started. Divide the analysis job into real-time
and postrun tasks, and determine how each aspect will affect your
program.

Postrun analysis
You can analyze data with LabVIEW, another application, or a custom
program written in some other language. Sometimes, more than one
analysis program will have to read the same data file. In all cases, you
need to decide on a suitable data file format that your data acquisi-
tion program has to write. The file type (typically text or binary), the
structure of the data, and the inclusion of timing and configuration
information are all important. If other people are involved in the analy-
sis process, get them involved early in the design process. Write down
clear file format specifications. Plan to generate sample data files and
do plenty of testing so as to assure everyone that the real data will
transfer without problems.
It’s a good idea to structure your program so that a single data
saver VI is responsible for writing a given data file type. Raw data and
configuration information go in, and data files go out. You can easily
test this data saver module as a stand-alone VI or call it from a test
program before your final application is completed. The result is a mod-
ule with a clear purpose that is reliable and reusable. LabVIEW can
read and write any file format (refer to Chapter 7, “Files,” for a general
discussion of file I/O). Which data format you use depends on the pro-
gram that has to read it.

Datalog file format. If you plan to analyze data only in LabVIEW, the
easiest and most compact format is the datalog file, discussed in
detail in Chapter 7. A datalog file contains a sequence of binary data
records. All records in a given file are of the same type, but a record
can be a complex data structure, for instance, a cluster containing
strings and arrays. The record type is determined when you create the
file. You can read records one at a time in a random-access fashion or
read several at once, in which case they are returned as an array. This
gives your analysis program the ability to use the data file as a simple
database, searching for desired records based on one or more key fields
in each record, such as a timestamp.
 Writing a Data Acquisition Program 327

The disadvantage of datalog format files is that they can only be read
by LabVIEW or by a custom-written program. However, you can eas-
ily write a translator in LabVIEW that reads your datalog format and
writes out files with another format. Another hazard (common to all
binary file formats) is that you must know the data type used when the
file was written; otherwise, you may never be able to decipher the file.
You might be able to use the automatic front panel datalogging fea-
tures of LabVIEW. They are very easy to use. All you have to do is to
turn on the datalogging by using the Datalogging submenu in the
Operate menu for a subVI that displays the data you wish to save.
Every time that the subVI finishes executing (even if its front panel
is not displayed), the front panel data are appended to the current log
file. The first time the subVI is called, you will receive a dialog ask-
ing for a new datalog file. You can also open the subVI and change log
files through the Datalogging menu. To access logged data, you can
use the file I/O functions, or you can place the subVI of interest in a
new diagram and choose Enable Database Access from its pop-up
menu. You can then read the datalog records one at a time. All the front
panel controls are available—they are in a cluster that is conveniently
accessed by Unbundle By Name. But for maximum file performance,
there’s nothing better than wiring the file I/O functions directly into
your diagram. Open the datalog file at the start of the experiment, and
don’t close it until you’re done.

Datalog File Handler VI makes it easy. It seems as if your diagrams

always end up with a large number of file I/O functions scattered all
over the place, with all the attendant wiring. One way to clean up the
situation is to encapsulate all file I/O operations into a single inte-
grated VI. This approach works for many other applications as well,
such as instrument drivers. The principle is simple: Create a VI with a
Mode control wired to a Case structure containing the various opera-
tions you need to perform. For files, the modes might be Create, Write,
and Read. Gary did exactly that for the Datalog File Handler VI, and
it really makes datalog files easy to use (Johnson 1994).
On the panel in Figure 14.2, you can see the Mode control and its
four possible choices. You begin by creating a new file or opening an
existing one. The file’s path is saved in an uninitialized shift register,
so you don’t have to wire to the path input for other I/O operations.
Append mode appends data in the Cluster in control to the file. Read
returns a selected range of data as a cluster array. You can choose the
starting record number and the number of records to read. The last
mode, Read Timestamps, returns an array of timestamp values for the
entire file. This tells you the number of records in the file (equal to the
number of elements in the Timestamps array). It’s also a convenient
328 Chapter Fourteen

Path (""; dialog) Path in use

Cluster in Cluster array out
Records to read (1) Timestamps
Start record (0) error out
error in (no error)
Mode (0=create/open)
Datalog File Handler

Mode (0=create/open) Path (""; dialog)

Ram Disk:datalog.dat

Cluster array out

5 Data cluster
Cluster in
Records to read (1) Timestamp
Timestamp
6 2924459674.000
0.000
Data
Data Start record (0)
99 -1.44E-2
99 0
1.23E+2

Timestamps
Cluster in and out are
Typedefs; edit with 5 0.00
Control Editor

Figure 14.2 The Datalog File Handler VI encapsulates datalog file I/O operations into
a single, integrated VI.

way to locate a particular record of data based on time. The Cluster

in and Cluster array out controls are typedefs. Modify them with
the Control Editor to match your data type. Note: The Timestamp is a
required item in the cluster.

ASCII text format. Good old ASCII text files are your best bet for por-
table data files. Almost every application can load data from a text
file that has simple formatting. The ubiquitous tab-delimited text
format is a likely choice. Format your data values as strings with a tab
character between each value, and place a carriage return at the end of
the line; then write it out. The only other thing you need to determine
is the type of header information. Simple graphing and spreadsheet
applications are happy with column names as the first line in the file:

Date Time Channel_1 Channel_2

5-13-82 01:17:34 5.678 -13.43
5-13-82 01:17:44 5.665 -13.58

A spreadsheet could be programmed to interpret all kinds of header

information, if you need to include more. Other applications are less
versatile with respect to headers, so be sure you know who’s going to
 Writing a Data Acquisition Program 329

Channel names Channel Pressure1 3

Flow
0.0
Temperture
Level
Pressure1 -5.0
Pressure2
-10.0
STOP 3 33

Write header
This is my header.
Write To
Spreadsheet File.vi

Channel names Write Characters append

To File.vi Data source
Graph
500
Strings []
String list *Channel*
converter.vi Attribute node Channel stop

Figure 14.3This simple datalogger example writes a text header and then tab-delimited data.
The String List Converter VI initializes the Channels control with all the channel names.

be reading your file. Figure 14.3 shows a simple datalogger that writes
a text header, followed by tab-delimited text data. It uses the easy-
level file VIs for simplicity. We wrote a little utility VI, String List
Converter, to make it easier for the user to enter channel names. The
names are typed into a string control, separated by carriage returns.
String List Converter translates the list into a set of tab-delimited
names for the file header and into a string array that you can wire to
the Strings[] attribute for a ring control, as shown.
The disadvantages of ASCII text files are that they are bulkier than
binary files, and they take much longer to read and write (often sev-
eral hundred times longer) because each value has to be converted to
and from strings of characters. For high-speed data-recording applica-
tions, text files are out of the question. You might be able to store a few
thousand samples per second as text on a fast computer, but be sure to
benchmark carefully before committing yourself to text files.

Custom binary formats. LabVIEW can write files with arbitrary binary
formats to suit other applications. If you can handle the requisite pro-
gramming, binary files are really worthwhile—high on performance and
very compact. It’s also nice to open a binary file with an analysis pro-
gram and have it load without any special translation. Keep in mind the
fact that LabVIEW datalog files are also binary format (and fast, too),
but are significantly easier to use, at least within LabVIEW. If you don’t
really need a custom binary format, stick with datalogs for simplicity.
330 Chapter Fourteen

Binary file handlers require significant programming experience.

Even if you have all the formatting information, be prepared to spend
time working out the programming details. Software manufacturers
will generally supply you with a description of their application’s native
binary file format if you ask the right person. Usually, you will get
some kind of program listing that was lifted from the manufacturer’s
file I/O routines. If the company is interested in making its file format
public, it will supply an application note and sometimes even machine-
readable code. The folks at Wavemetrics supply all this information
for Igor (their analysis and graphing package for Macintosh and Win-
dows) with the application. Because the information was available, the
VIs to read and write Igor binary were easy to create and are now
public domain. Figure 14.4 shows the panel for the VI that writes a

Path (default: dialog)

Wave Note ("")
WaveName
Wave Data (SGL) Path chosen
Y & X-axis Units ("")
X scale & offset (1,0)
Y scale min & max
Y scale is valid (F)

SGL to Igor Binary File

WaveName Path (default: dialog)

wave0 Gary:An Igor Wave

Wave Data (SGL) Path chosen

0 123.00

Y & X-axis Units ("")

V Y

s X
Wave Note ("")

X scale & offset (1,0) This note will ride along with the data, right
into Igor.
1.00 hsA

0.00 hsB

Y scale min & max

0.00 max
Y scale is valid (F)
0.00 min

Figure 14.4 This VI creates a file suitable for direct loading with Wavemetrics’
Igor.
 Writing a Data Acquisition Program 331

single-precision floating-point array to an Igor binary file. The diagram

is very complex, so it’s not shown. You can obtain these Igor support
VIs from Wavemetrics. By the way, Igor can also read arbitrary binary
formats directly. With that capability, you can write your LabVIEW
data as simple binary arrays and then load them right into Igor with
no special programming whatsoever.
Some binary formats such as the previously described Igor format
are unsuitable for continuous data acquisition—where you would like
to append records one at a time as the experiment proceeds. They are
designed for single-shot experiments in which the entire file is writ-
ten at once. This is fine for single buffers of data from an oscilloscope,
but less useful for a simple datalogger. Buffering data in memory (in
arrays) is one way to solve this problem, but if your computer should
crash, that data might be lost. You could rewrite the entire file occa-
sionally to avoid such a catastrophe, if you are forced to use one of these
single-shot formats. On the other hand, if you are not stuck with such
a predefined format, it is easy and efficient to write continuously to
a binary file, as you can see by examining the techniques used in the
DAQ disk streaming example VIs.

Direct links to other applications. Several other important analysis

applications are available with LabVIEW links. The ability to write
data directly to another application is very appealing since it makes
background analysis without user intervention a reality; it also elimi-
nates the intermediate file conversion hassle.
Starting with LabVIEW 8.0, you can use a MathScript Script
node to directly execute MathScript scripts from the LabVIEW
diagram. It’s located in the Programming >> Structures palette and
looks and behaves much as a Formula node. You type in any valid
MathScript script and assign variables to input and output terminals
that you create.
MATLAB is another popular multiplatform mathematics and simu-
lation application, with a variety of toolboxes for specialties in science
and engineering. While it will certainly read data files (text or binary),
you can also use a MATLAB Script node, which is very similar to
the MathScript Script node, and, of course, only runs under Windows.
Remember that you do need a legal copy of MATLAB installed on your
machine, but not for the MathScript scripts.
Microsoft Excel is very popular for data analysis, with its large
installed base on both Windows and Macintosh. Perhaps its major
limitations are a lack of a binary file import capability and a maxi-
mum of 65,536 rows, making it unsuitable for very large data sets.
But there is excellent support for direct interaction with Excel from
LabVIEW.
332 Chapter Fourteen

Macintosh users can exchange data via AppleEvents; there are exam-
ple VIs that read and write data. Windows users can implement DDE
or Active X connections. See the LabVIEW examples. Using ActiveX,
you can make Excel do just about anything, from creating and saving
workbooks to executing macros and reading or writing data. Again, the
LabVIEW examples will get you started.

Timestamps. Most data that we collect is a function of time. There-

fore, timing information needs to be stored along with the data in your
files. The precision or resolution of these timestamps will be driven by
the requirements of your experiment and the limitations of LabVIEW’s
timing functions. Requirements for formatting of the timestamps will
be determined by the application that reads the data file.
Consider an ordinary datalogging application where a few dozen
channels are stored to disk as ASCII text every few seconds. A resolu-
tion of 1 s is probably adequate, and a source for this timing informa-
tion is Get Date/Time In Seconds, or one of the other built-in timing
functions. You can then format the returned value as seconds relative
to 1 Jan 1904, as seconds relative to the start of your experiment, or
divide it by 60 to obtain minutes, and so forth. Saving a simple numeric
timestamp has the advantage that it is easy to interpret. If you use a
spreadsheet application that can manipulate time and date formats
(such as mm/dd/yy hh:mm:ss), then Get Date/Time String may be
appropriate.
You can get similar results by using the simple Tick Count (ms)
function, which returns a relative number of milliseconds. The uncer-
tainty of these higher-resolution timers depends on your machine and
on LabVIEW’s workload. You may want to do some testing if you need
really accurate software-derived timestamps.
Data that is acquired at a constant rate (periodic sampling) needs a
timestamp only at the beginning of the collection period because the
time at any sample is easily calculated. This is certainly true of data
acquired with hardware-timed I/O using the DAQ library. There is no
reason to store a timestamp with every sample if you already know the
sample interval with acceptable precision; it would just take up extra
disk space. Only data that is taken aperiodically requires such detailed
timestamp information.
Sampling rates higher than about 10 Hz usually are handled by data
acquisition hardware (or smart I/O subsystems) because there is too
much timing uncertainty in fast LabVIEW loops on general-purpose
computers. Using hardware can simplify your timestamp requirements.
Because you know that the hardware is sampling at a steady rate, there
is no need to store a timestamp for every sample period. Rather, you
need only save an initial time and a value for the sampling interval.
 Writing a Data Acquisition Program 333

The data analysis program should then be able to reconstruct the time
base from this simple scheme.
A technique used in really fast diagnostics is to add a timing fidu-
cial pulse to one or more data channels. Also known as a fid, this pulse
occurs at some critical time during the experiment and is recorded on
all systems (and maybe on all channels as well). It’s much the same
as the room full of soldiers and the commander says, “Synchronize
watches.” For example, when you are testing explosives, a fiducial pulse
is distributed to all the diagnostic systems just before detonation. For
analog data channels, the fiducial pulse can be coupled to each chan-
nel through a small capacitor, creating a small glitch in the data at the
critical moment. You can even synchronize nonelectronic systems by
generating a suitable stimulus, such as flashing a strobe in front of a
movie or video camera. Fiducial pulses are worth considering any time
you need absolute synchronization among disparate systems.
If you need accurate time-of-day information, be sure to reset the
computer clock before the experiment begins. Personal computer clocks
are notorious for their long-term drift. If you are connected to the Inter-
net, you can install a utility that will automatically reset your system
clock to a standard time server. For Windows, there are several public
domain utilities, such as WNSTIME (available from sunsite.unc.edu),
that you can install. For Macintosh, you can use the Network Time
Server option in the Date and Time Control Panel. These utilities even
take into consideration the time zone and the daylight savings time
settings for your machine. The only other trick is to find a suitable
Network Time Protocol server. One we’ve had good luck with is NASA’s
norad.arc.nasa.gov server.
In Chapter 5, “Timing,” we list a host of precision timing devices that
you might consider, such as GPS and IRIG standards. Such hardware
can provide absolute timing standards with accuracy as good as a few
hundred nanoseconds. That’s about as good as it gets.

Passing along configuration information. Your analysis program may

need information about the configuration of the experiment or soft-
ware that generated the data. In many cases, channel names are all
that is needed, and you can pass them along as the column titles in a
spreadsheet file. Beyond that, you have two basic choices: Use a sepa-
rate configuration file or add a file header to each data file. Both
methods are highly dependent upon the ability of the analysis program
to read and interpret the information.
If you’re using the Measurement and Automation Explorer
(MAX), it has a useful feature called Create Report (File >> Create
Report). Create Report allows you to save the contents of your DAQmx
tasks in HTML format. The information includes channel names
334 Chapter Fourteen

and descriptions, scale factors, hardware device information, and sen-

sor information. This is great information to include with your docu-
mentation. It even includes connection diagrams showing you how to
connect the signals to your DAQ equipment.
Binary files almost always have headers because the program that
reads them needs information about the type and location of the data
within. Here is a list of the kind of information contained in a binary
file header:

Experiment identification
Channel name
Creation time and date
Data type (single- or double-precision)
Data file version (to avoid incompatibility with future versions)
The y-axis and x-axis units number of data points
The y-axis and x-axis scale factors
Flag to indicate whether user comments follow the data segment

As always, the format of this binary file header will be highly speci-
fied on a byte-by-byte basis. You need to make sure that each item is of
the proper data type and length before writing the header out to the file.
An effective way to do this is to assemble all the items into a cluster,
Type Cast it (see Chapter 4, “LabVIEW Data Types”) to a string, and
then write it out, using byte stream mode with the Write File function.
Alternatively, the header can be text with flag characters or end-of-line
characters to help the reading application parse the information.
For text files, generating a header is as simple as writing out a series
of strings that have been formatted to contain the desired information.
Reading and decoding a text-format header, on the other hand, can be
quite challenging for any program. If you simply want an experimental
record or free-form notepad header for the purposes of documentation,
that’s no problem. But parsing information out of the header for pro-
grammatic use requires careful design of the header’s format. Many
graphing and analysis programs can do little more than read blocks of
text into a long string for display purposes; they have little or no capac-
ity for parsing the string. Spreadsheets (and of course programming
languages) can search for patterns, extract numbers from strings, and
so forth, but not if the format is poorly defined. Therefore, you need to
work on both ends of the data analysis problem—reading as well as
writing—to make sure that things will play together.
Another solution to this header problem is to use an index file that is
separate from the data file. The index file contains all the information nec-
essary to successfully load the data including pointers into the data file.
 Writing a Data Acquisition Program 335

It can also contain configuration information. The data file can be binary
or ASCII format, containing only the data values. We’ve used this tech-
nique on several projects, and it adds some versatility. If the index file is
ASCII text, then you can print it out to see what’s in the data file. Also,
the data file may be more easily loaded into programs that would other-
wise choke on header information. You still have the problem of loading
the configuration, but at least the data can be loaded and the configura-
tion is safely stored on disk. One caution: Don’t lose one of the files!
The configuration file can be formatted for direct import into a
spreadsheet and used as a printable record of the experiment. This
turns out to be quite useful. Try to produce a complete description of
the hardware setup used for a given test, including module types, chan-
nel assignments, gain settings, and so forth. Here’s what a simple con-
figuration file might look like:

Source expt RUN306 08-SEP-1992 21:18:51.00 HD:Data:RUN306

File Module ID Slot Crate Signal Name Chan Units
TEST0.T Kinetic 3525 1 1 TC1 1 Deg C
TEST0.T Kinetic 3525 1 1 TC2 2 Deg C
TEST0.T Kinetic 3525 1 1 Upper src temp 4 Deg C
TEST0.T Kinetic 3525 1 1 Lower src temp 5 Deg C
END

When the experimenter has a question regarding the signal connec-

tions, you can refer to this list, which is usually clipped into the labora-
tory logbook. We’ll discuss some methods for generating configuration
files later in this chapter.

Using a real database. If your data management needs are more com-
plex than the usual single-file and single-spreadsheet scheme can han-
dle, consider using a commercial database application for management
of configurations, experimental data, and other important information.
The advantages of a database are the abilities to index, search, and sort
data with concurrent access from several locations but with explicitly
regulated access to the data, thus enhancing security and reliability.
The Enterprise Connectivity Toolkit (formerly the SQL Toolkit)
from National Instruments enables LabVIEW to directly communicate
with any Open Database Connectivity (ODBC)–compliant database
application using Structured Query Language (SQL) commands.
The toolkit is available for all versions of Windows as well as Macintosh
and is compatible with nearly all major databases. It was originally cre-
ated by Ellipsis Products and marketed as DatabaseVIEW.
The SQL Toolkit can directly access a database file on the local disk
using SQL commands to read or write information, or the database
can exist on a network—perhaps residing on a mainframe computer
or workstation. You begin by establishing a session, or connection to
336 Chapter Fourteen

Figure 14.5 The Enterprise Connectivity Toolkit connects LabVIEW to commercial databases.
This example inserts simulated weather data into a climate database. (Example courtesy of
Ellipsis Products, Inc.)

the target database(s), and then build SQL commands using LabVIEW
string functions. The commands are then executed by the SQL Toolkit
if you are connecting directly to a file or by a database application that
serves as a database engine. Multiple SQL transactions may be active
concurrently—an important feature, since it may take some time to
obtain results when you are using a complex SQL command to access
a very large database.
Figure 14.5 shows an example that simulates a year of weather in
Boston, Massachusetts, and inserts each day’s statistics into a dBASE
database. The dataflow is pretty clear. First, a connection is made to
the database by using the Connect VI for the data source DBV DEMO
DBASE. The first two frames of the sequence (not shown) create the
empty climate table in the database. Second, frame 2 begins by creat-
ing the year of simulated data. A dynamic SQL INSERT statement is
also created by using wildcard characters (?) to designate the values
that will be inserted later. In the For Loop, each field value is bound
to its associated parameter in the wildcard list, and then the Execute
Prepared SQL VI inserts data into the climate table. The dynamic
SQL method can speed inserts up to 4 times over the standard con-
vert, parse, and execute method. Third, the SQL statement reference is
discarded in the End SQL VI. When the loop is complete, the last frame
of the Sequence structure (not shown) disconnects from the database.
 Writing a Data Acquisition Program 337

Real-time analysis and display

LabVIEW’s extensive library of built-in analysis functions makes it easy
to process and display your newly acquired data in real time. You are
limited by only two things: your system’s performance and your imagi-
nation. Analysis and presentation are the things that make this whole
business of virtual instrumentation useful. You can turn a voltmeter
into a strip chart recorder, an oscilloscope into a spectrum analyzer, and
a multifunction plug-in board into . . . just about anything. Here, we’ll
look at the general problems and approaches to real-time analysis and
display. In later chapters we will discuss particular applications.
Once again, we’ve got the old battle over the precise definition of real
time. It is wholly dependent upon your application. If 1-min updates
for analysis and display are enough, then 1 min is real time. If you need
millisecond updates, then that’s real time, too. What matters is that you
understand the fundamental limitations of your computer, I/O hard-
ware, and LabVIEW with regard to performance and response time.
The kind of analysis you need to perform is determined by the nature
of your signals and the information you want to extract (Table 14.1).
Assuming that you purchase the full version of LabVIEW, there are
about 400 analysis functions available. Other functions (and useful
combinations of the regular ones) are available from the examples and
from others who support LabVIEW through the Alliance Program. If you
ever need an analysis function that seems obvious or generally useful,

TABLE 14.1 Signal Types and Analysis Examples

Signal type Typical analysis
Analog—DC Scaling
Statistics
Curve fitting
Analog—Time domain Scaling
Statistics
Filtering
Peak detection and counting
Pulse parameters
Analog—Frequency domain Filtering
Windowing
FFT/power spectrum
Convolution/deconvolution
Joint time frequency analysis
Digital on-off Logic
Digital pulse train Counting
Statistics
Time measurement
Frequency measurement
338 Chapter Fourteen

be sure to contact National Instruments (NI) to find out if it’s already

available. NI also takes suggestions—user input is really what makes
this palette grow.

Continuous versus single-shot data analysis. Data acquisition may

involve either continuous data or single-shot data. Continuous data
generally arrives one sample at a time, like readings from a voltmeter.
It is usually displayed on something such as a strip chart, and probably
would be stored to disk as a time-dependent history. Single-shot data
arrives as a big buffer or block of samples, like a waveform from an
oscilloscope. It is usually displayed on a graph, and each shot would be
stored as a complete unit, possibly in its own file. Analysis techniques
for these two data types may have some significant differences.
There is a special form of continuous data that we call block-mode
continuous data where you continuously acquire measurements, but
only load them into your LabVIEW program as a block or buffer when
some quantity of measurements has accumulated. Multiple buffer-
ing or circular buffering can be carried out by any smart instrument,
including DAQmx. The advantage of block-mode buffered operation is
the reduced I/O overhead: You need to fetch data only when a buffer is
half full, rather than fetch each individual sample. The disadvantage
is the added latency between acquisition of the oldest data in the buffer
and the transfer of that data to the program for processing. For analy-
sis purposes, you may treat this data as either continuous or single-
shot since it has some properties of both.
Here is an example of the difference between processing continuous
and single-shot data. Say that your main interest lies in finding the
mean and standard deviation of a time-variant analog signal. This is
really easy to do, you notice, because LabVIEW just happens to have a
statistical function called Standard Deviation and Variance which
also computes the mean. So far, so good.

Single-shot data. A single buffer of data from the desired channel is

acquired from a plug-in board. DAQmx returns a numeric array con-
taining a sequence of samples, or waveform, taken at a specified sam-
ple rate. To compute the statistics, wire the waveform to the Standard
Deviation and Variance function and display the results (Figure 14.6).

Figure 14.6 Statistics are easy to calculate by using the built-in Stan-
dard Deviation and Variance function when data is acquired as a single
shot (or buffer), in this case using the DAQmx Assistant.
 Writing a Data Acquisition Program 339

Compute statistics
Compute statistics after the run
during the run
200 Mean

sample Std Dev

device Mean Population Weighting

channel Std Dev
This array could
Running Mean
RUN be quite large.
reset stats on & Sigma.vi
first iteration

Figure 14.7 Gary had to write a special function, Running Mean & Sigma, that
accumulated and calculated statistics during execution of a continuous data
acquisition process. Building an array for postrun calculations consumes much
memory.

You might note that there is a conversion function before the input ter-
minal to Standard Deviation. That’s because DAQmx uses the dynamic
data type, while most of the analysis library uses double-precision. The
result is extra memory management and some loss of efficiency and
speed, but it is currently unavoidable.

Continuous data. You can collect one sample per cycle of the While
Loop by calling AI Single Scan, as shown in Figure 14.7. If you want
to use the built-in Standard Deviation function, you have to put all
the samples into an array and wait until the While Loop finishes
running—not exactly a real-time computation. Or, you could build the
array one sample at a time in a shift register and call the Standard
Deviation function each time. That may seem OK, but the array grows
without limit until the loop stops—a waste of memory at best, or you
may cause LabVIEW to run out of memory altogether. The best solu-
tion is to create a different version of the mean and standard deviation
algorithm, one that uses an incremental calculation.
Gary wrote a function called Running Mean & Sigma that recom-
putes the statistics each time it is called by maintaining intermediate
computations in uninitialized shift registers (Figure 14.8). It is fast and
efficient, storing just three numbers in the shift registers. A Reset switch
sets the intermediate values to zero to clear the function’s memory.
The idea came right out of the user’s manual for his HP-45 calcula-
tor, proving that inspiration is wherever you find it. Algorithms for this
and hundreds of other problems are available in many textbooks and
in the popular Numerical Recipes series (Press 1990). You can use the
concept shown here for other continuous data analysis problems.
National Instruments introduced another option for single-shot data
analysis called the Point-by-Point library. It contains nearly all the
analysis functions including signal generation, time and frequency
domain, probability and statistics, filters, windows, array operations, and
340 Chapter Fourteen

x mean
0.00 0.00 x mean
iteration (1; 0=init) std dev
iteration (1; 0=init) std dev
Running Mean & Sigma
1 0.00

True x

∑x
0.0

∑x^2

sx m = sx / N; mean
iteration (1; 0=init) s = sqrt( 1/(N-1) * m
N
x2 (x2-(sx^2)/N) ); s
std dev

Figure 14.8Running Mean & Sigma calculates statistics on an incremental basis by stor-
ing intermediate computations in uninitialized shift registers. The Reset switch clears the
registers.

linear algebra. Internally, each function uses the same basic techniques
we’ve just discussed for the Running Mean and Sigma VI. But there’s
some real numerical sophistication in there, too. Consider what it means
to do an FFT calculation on a point-by-point basis. A simple solution
would be to just save up a circular buffer of data and then call the regular
FFT function on each iteration. Instead, NI implements the mathemati-
cal definition of the FFT, which is the integral of a complex exponential.
This is, in fact, much faster, but makes sense only when performed on a
sliding buffer. As a benchmark, you’ll find that a 1024-point FFT runs
about 50 percent faster in the point-by-point mode. It’s an even larger dif-
ference when you’re processing a buffer that’s not a power of 2.

Faster analysis and display.Real-time analysis may involve significant

amounts of mathematical computation. Digital signal processing (DSP)
functions, such as the FFT and image processing functions, operate on
large arrays of data and may require many seconds, even on the fastest
computers. If execution time becomes a problem, you can:
■ Make sure that you are using the most efficient computation tech-
niques. Try to simplify mathematical expressions and processes and
seek alternative algorithms.
■ Avoid excessive array manipulation and duplication.
 Writing a Data Acquisition Program 341

■ Figure out ways to reduce the amount of data used in the calcula-
tions. Decimation is a possibility.
■ Do the analysis postrun instead of in real time.
■ Get a faster computer.
■ Use a DSP coprocessor board.

Some of the options you may reject immediately, such as postprocess-

ing, which is of little value when you are trying to do feedback control.
On the other hand, if you really need that 8192-point power spectrum
displayed at 5 kHz, then you had better be using something faster than
the average PC. Always be sure that the analysis and display activities
don’t interfere with acquisition and storage of data.

Reducing the volume of data. Execution time for most algorithms is

roughly proportional to the size of the data arrays. See if you can do
something to reduce the size of your arrays, especially when they are to
be processed by one of the slower functions. Here are some ideas:
Sample at the minimum rate consistent with the Nyquist criteria
and input filtering for your signals. Many times, your data acquisition
system will have to sample several channels that have widely differ-
ent signal bandwidths. You may be able to rewrite the acquisition part
of your program so that the low-bandwidth signals are sampled at a
slower rate than the high-bandwidth signals. This may be more com-
plex than using a single I/O function that reads all channels at once,
but the reduction in array size may be worthwhile.
Process only the meaningful part of the array. Try to develop a tech-
nique to locate the interesting part of a long data record, and extract
only that part by using the Split Array or Array Subset functions.
Perhaps there is some timing information that points to the start of the
important event. Or, you might be able to search for a critical level by
using Search 1D Array, Peak Detector, or a Histogram function.
These techniques are particularly useful for sparse data, such as that
received from a seismometer. In seismology, 99.9 percent of the data
is just a noisy baseline containing no useful information. But every
so often an interesting event is detected, extracted, and subjected to
extensive analysis. This implies a kind of triggering operation. DAQmx
has a software triggering feature whereby data is transferred to the
LabVIEW data space only if it passes some triggering criteria, includ-
ing slope and level. This feature works for plug-in boards that don’t
have similar hardware triggering functionality.
Data decimation is another possible technique. Decimation is a pro-
cess whereby the elements of an array are divided up into output arrays,
much as a dealer distributes cards. The Decimate 1D Array function
342 Chapter Fourteen

can be sized to produce any number of output arrays. Or, you could
write a program that averages every n incoming values into a smaller
output array. Naturally, there is a performance price to pay with these
techniques; they involve some amount of computation or memory man-
agement. Because the output array(s) is (are) not the same size as the
input array, new memory buffers must be allocated, and that takes
time. But the payoff comes when you finally pass a smaller data array
to those very time-consuming analysis VIs.

Improving display performance. All types of data displays—especially

graphs and images—tend to bog down your system. Consider using
smaller graphs and images, fewer displayed data points, and less fre-
quent updates when performance becomes a problem.
Figure 14.9 shows two ways of reducing the update rates of graph-
ics, or anything else, be it an indicator or a computation. Figure 14.9A
fixes the update rate of a graph in terms of a time interval by using the
Interval Timer VI, described in Chapter 5.
Figure 14.9B permits a graph to update every n cycles of a data
acquisition loop by testing the remainder output of the Quotient and
Remainder function. Yet another way to reduce display overhead is to
add an update display button connected to the Case structure in lieu of
these automatic update techniques.
The displays could also appear in independent top-level VIs that
receive data from the main acquisition loop via global variables. This
is the client-server concept from Chapter 8, “Building an Application,”
and it works well here. You write the data (probably an array) to a
global variable in the data acquisition loop (the server). Then another
VI containing graphs or other displays (the client) reads data from the
global variable asynchronously. Analysis can be performed in either

A B
True True

Waveform Waveform
Data Source Graph Data Source Graph

Update
graph (sec)

Interval Timer.vi RUN RUN

Update every N times

Figure 14.9 Limit the updating rates of graphics (or anything else) by using (A) the Interval
Timer VI or (B) modulo arithmetic with the Quotient and Remainder function.
 Writing a Data Acquisition Program 343

VI or both VIs. The advantage is that the client can be set to execute
at lower priority than the server, as well as execute at a slower rate.
Also, an arbitrary number of clients can run simultaneously. You could
really get carried away and use another global that indicates that the
display VI wants an update, thus creating a handshaking arrangement
that avoids writing to the global on every acquisition. The disadvan-
tage of the client-server scheme is that even more copies of the data are
required, trading memory usage for speed.
Graphics accelerators are now common. All new PCs and Macs
include graphics coprocessors that significantly reduce the overhead
associated with updating the display (by as much as a factor of 50 in
some cases). LabVIEW still has to figure out where the text, lines, and
boxes need to go, and that takes some main CPU time. But the graphics
board will do most of the low-level pixel manipulation, which is cer-
tainly an improvement. The LabVIEW Options item Smooth Updates
makes a difference in display performance and appearance as well.
Smooth updates are created through a technique called off-screen bit-
maps where graphics are drawn to a separate memory buffer and then
quickly copied to the graphics display memory. The intent is to enhance
performance while removing some jumpiness in graphics, but smooth
updates may actually cause the update time to increase, at least on
some systems. Experiment with this option, and see for yourself.

Sampling and Throughput

How much data do you need to acquire, analyze, display, and store in
how much time? The answer to this question is a measure of system
throughput. Every component of your data acquisition system—
hardware and software—affects throughput. We’ve already looked at
some analysis and display considerations. Next, we’ll consider the input
sampling requirements that determine the basic data generation rate.
Modern instrumentation can generate a veritable flood of data.
There are digitizers that can sample at gigahertz rates, filling multi-
megabyte buffers in a fraction of a second. Even the ubiquitous, low-
cost, plug-in data acquisition boards can saturate your computer’s bus
and disk drives when given a chance. But is that flood of data really
useful? Sometimes; it depends on the signals you are sampling.

Signal bandwidth
As we saw in Chapter 13, “Sampling Signals,” every signal has a mini-
mum bandwidth and must be sampled at a rate at least 2 times this
bandwidth, and preferably more, to avoid aliasing. Remember to
include significant out-of-band signals in your determination of the
sampling rate.
344 Chapter Fourteen

If you can’t adequately filter out high-frequency components of the

signal or interference, then you will have to sample faster. A higher
sampling rate may have an impact on throughput because of the larger
amount of raw data that is collected. Evaluate every input to your sys-
tem and determine what sampling rate is really needed to guarantee
high signal fidelity.
Sometimes, you find yourself faced with an overwhelming aggregate
sampling rate, such as 50 channels at 180 kHz. Then it’s time to start
asking simple questions such as, Is all this data really useful or neces-
sary? Quite often, there are channels that can be eliminated because of
low priority or redundancy. Or, you may be able to significantly reduce
the sampling rate for some channels by lowering the cutoff frequency
of the analog low-pass filter in the signal conditioner. The fact that
some channels need to go fast doesn’t mean that they all do.

Oversampling and digital filtering

Low-frequency analog signals give you some opportunities to further
improve the quality of your acquired data. At first glance, that thermo-
couple signal with a sub-1-Hz bandwidth and little noise could be ade-
quately sampled at 2 or 3 Hz. But by oversampling—sampling at a rate
several times higher than the Nyquist frequency—you can enhance res-
olution and noise rejection. Noise is reduced in proportion to the square
root of the number of samples that are averaged. For example, if you
average 100 samples, the standard deviation of the average value will
be reduced by a factor of 10 when compared to a single measurement.
This topic is discussed in detail in “A Little Noise Can Be a Good Thing”
in Chapter 13, “Sampling Signals.” We use oversampling for nearly all
our low-speed DAQ applications because it’s easy to do, requires little
extra execution time, and is very effective for noise reduction.
Once you have oversampled the incoming data, you can apply a
digital low-pass filter to the raw data to remove high-frequency noise.
There are a number of ways to do digital filtering in LabVIEW: by using
the Filter functions in the analysis library or by writing something of
your own. Digital filter design is beyond the scope of this book, but at
least we can look at a few ordinary examples that might be useful in
a data acquisition system. For more information, see LabVIEW Signal
Processing (Chugani et al. 1998).
An excellent resource for filter design is National Instruments’ Dig-
ital Filter Design Toolkit, which is part of the Signal Processing
Toolkit for LabVIEW and LabWindows/CVI. It is a stand-alone appli-
cation for designing and testing all types of digital filters. You typically
begin with a filter specification—requirements for the filter response
in terms of attenuation versus frequency—and use the filter designer
to compute the required coefficients that the LabVIEW VIs can use at
 Writing a Data Acquisition Program 345

run time. The toolkit makes it easy by graphically displaying all results
and allowing you to save specifications and coefficients in files for com-
parison and reuse. Displays include magnitude, phase, impulse, and
step response, a z plane plot, and the z transform of the designed filter.
Once you have designed a filter and saved its coefficients, an included
LabVIEW VI can load those coefficients for use in real time.

FIR filters. If you are handling single-shot data, the built-in Finite
Impulse Response (FIR) filter functions are ideal. The concept behind
an FIR filter is a convolution (multiplication in the frequency domain) of
a set of weighting coefficients with the incoming signal. In fact, if you look
at the diagram of one of the FIR filters, you will usually see two VIs: one
that calculates the filter coefficients based on your filter specifications,
and the Convolution VI that does the actual computation. The response
of an FIR filter depends only on the coefficients and the input signal, and
as a result the output quickly dies out when the signal is removed. That’s
why they call it finite impulse response. FIR filters also require no initial-
ization since there is no memory involved in the response.
For most filters, you just supply the sampling frequency (for calibra-
tion) and the desired filter characteristics, and the input array will be
accurately filtered. You can also use high-pass, bandpass, and bandstop
filters, in addition to the usual low-pass, if you know the bandwidth of
interest. Figure 14.10 shows a DAQ-based single-shot application where

FIR filter specifications optional FIR filter specification

Noisy raw data
Topology PB Gain Waveform
Filtered data
graph
Equi-ripple FIR 3.00
6.0
Type #Taps SB Gain
4.0
Lowpass 30 60.00 2.0
Lower PB Upper PB Scale 0.0
-2.0
1.00 0.00 dB
-4.0
Lower SB Upper SB Window
-6.0
100.00 0.00 Hanning 0.0 100.0m 200.0m 300.0m

AI Acquire Add Digital FIR Waveform

Build
Waveform.vi Waveforms.vi Filter.vi graph
noisy Array
device 1
filtered
n samples 300

1000.00 optional FIR filter

sample rate FIR filter specification
amplitude
specifications
0.1
Periodic Random
Noise Waveform.vi

Figure 14.10 A practical application of FIR low-pass filtering applied to an array of single-shot
data, a noisy sine wave. Bandpass filtering could also be used in this case, since we know the
exact frequency of interest.
346 Chapter Fourteen

a noisy sine wave is the signal. We applied a sinusoidal input signal,

and then we added Periodic Random Noise to exercise our filter.
The Digital FIR Filter VI is set up to perform a low-pass filter
operation. The sampling frequency (fs) for the filter VI is the same
one used by the DAQ VI and is measured in hertz. This calibrates the
low cutoff frequency control in hertz as well. Since the input signal
was 10 Hz, we selected a cutoff of 50 Hz, which is sufficiently high to
avoid throwing away any significant part of the signal. Topology is set
for the FIR filter by specification, where you supply only the passband
and stopband frequencies. Examine the graph and note that the fil-
tered waveform is delayed with respect to the raw data. This is where
the Digital Filter Design Toolkit comes in handy. You can choose the
optimum cutoff frequency to yield the desired results. Filter design is
always a compromise, so don’t be disappointed if the final results are
in some way less than perfect.
You can compensate for the delay in an FIR filter by routing the
filtered signal through the Array Subset function and removing the
required number of samples from the start of the waveform. That num-
ber is approximately equal to the number of taps; being the nonmath-
ematical types, we usually test it out and see what the magic number
is. Since the delay is invariant once you set the number of taps, FIR
filters can effectively supply zero delay. (Technically, this is because
they have linear phase distortion.) This is important when you need
to maintain an absolute timing reference with respect to a trigger or
another signal.
A time window generally should be applied to all signals prior to
filtering or other frequency domain processing. Time windows reduce
spectral leakage and other finite-sequence length artifacts in
spectral estimation calculations such as FFTs and power spectra. Spec-
tral leakage occurs because an FFT has discrete frequency bins. If the
signal does not happen to precisely land in one of these bins, the FFT
smears the signal energy into adjacent bands. The other problem, due
to the finite data record length, shows up as extra energy spread out
all over the spectrum. If you think about it, an arbitrary buffer of data
probably does not start and end at zero. The instantaneous step from
zero to the initial value of the data represents a transient, and tran-
sients have energy at all frequencies.
The job of the time window is to gradually force the start and end of
your data to zero. It does so by multiplying the data array by some func-
tion, typically a cosine raised to some power, which by definition is zero
at both ends. This cleans up much of the spectral leakage and finite-
sequence length problems, or at least makes them more predictable.
One side effect is a change in absolute amplitude in the results. If you
use the Scaled Window VI from the Signal Processing >> Waveform
 Writing a Data Acquisition Program 347

Conditioning palette, the gain for each type of window is properly com-
pensated. The Digital FIR Filter VI is also properly compensated. There
are quite a few window functions available, and each was designed
to meet certain needs. You should definitely experiment with various
time windows and observe the effects on actual data.

IIR filters. Infinite Impulse Response (IIR) filters are generally a

closer approximation to their analog kin than are FIR filters. The output
of an IIR filter depends not only on the input signal and a set of forward
coefficients, but also on the output of the filter and an additional set
of reverse coefficients—a kind of feedback. The resulting response
takes (theoretically) an infinite amount of time to arrive at its final
value—an asymptotic response just like the exponential behavior of
analog filters. The advantage of an IIR filter, compared to an FIR filter, is
that the feedback permits more rapid cutoff transitions with fewer coef-
ficients and hence fewer computations. However, IIR filters have nonlin-
ear phase distortion, making them unsuitable for some phase-sensitive
applications. Also, you can choose a set of coefficients that make an IIR
filter unstable, an unavoidable fact with any feedback system.
IIR filters are best suited to continuous data, but with attention to
initialization they are also usable with single-shot data. In the Fil-
ter palette, you will find most of the classic analog filter responses in
ready-to-use VIs: Butterworth, Chebyshev, Inverse Chebyshev, ellipti-
cal, and Bessel. If you are at all familiar with RC or active analog fil-
ters, you’ll be right at home with these implementations. In addition,
you can specify arbitrary responses not available using analog tech-
niques. Of course, this gets a bit tricky, and you will probably want the
Digital Filter Design Toolkit or some other mathematical tool to obtain
valid coefficients.
Figure 14.11 shows the old NI DAQ function AI Continuous Scan
VI acquiring seamless buffers of data from a single channel containing
a 5-Hz sine wave. For interest, we added noise, just as in the previous
FIR example. The Digital IIR Filter behaves in this application simi-
larly to its analog counterpart, but we didn’t need a soldering iron to
build it.
For classical filters such as this Butterworth filter, controls are quite
similar to the design parameters for corresponding analog filters. In
particular, you choose the filter type (high-pass, low-pass, etc.), and
then you choose the order. Order, in the analog world, translates to
the number of inductors and capacitors in the filter implementation.
Higher order results in a sharper response, but as you might expect,
the implementation requires additional computations. Sampling fre-
quency and high- and low-cutoff frequencies are calibrated in hertz, as
with the FIR filters.
348 Chapter Fourteen

IIR filter specifications

Noisy raw data
Topology Waveform
Filtered data
Butterworth graph
4.0
Type Order
Lowpass 4 2.0

Lower Fc Upper Fc 0.0

20.00 200.00
-2.0
PB Ripple SB Attenuation
-4.0
1.00 60.00 0.0 100.0m 200.0m 300.0m STOP

AI Continuous Index Add Digital IIR Build Waveform

Scan.vi Array Waveforms.vi Filter.vi Array graph
noisy
device 1 filtered
n samples 300

1000.00
noise IIR filter specifications
scan rate
amplitude 0.1
Periodic Random
Noise Waveform.vi

0=init 50
stop

Figure 14.11 A continuous acquisition process with a Butterworth IIR filter. Once ini-
tialized, this digital filter works just as an analog filter does.

Median filters. If your data contains outliers—also known as spikes or

fliers—consider the Median Filter VI. The median filter is based on a
statistical, or nonlinear, algorithm. Its only tuning parameter is called
rank, which determines the number of values in the incoming data
that the filter acts upon at one time. For each location i in the incom-
ing array, the filter sorts the values in the range (i − rank) to (i + rank)
and then selects the median (middle) value, which then becomes the
output value. The algorithm slides along through the entire data set in
this manner. The beauty of the median filter is that it neatly removes
outliers while adding no phase distortion. In contrast, regular IIR and
FIR filters are nowhere near as effective at removing outliers, even
with very high orders. The price you pay is speed: The median filter
algorithm tends to be quite slow, especially for higher rank. To see how
the median filter works, try out the Median Filtering example VI.

Moving averagers. Moving Averagers are another appropriate filter

for continuous data. The one demonstrated in Figure 14.12, Moving
Avg Array, operates on data from a typical multichannel data acqui-
sition system. An array containing samples from each channel is the
 Writing a Data Acquisition Program 349

Number of Channels Raw signal Display chan

5 Averaged 1

Strip chart
Samples to avg 2.00

6 1.40

Average 0.80

ON 0.20

-0.40
STOP -1.00

Pseudo-data acquisition system 1 sample per chan

Strip
chart
Moving Average
Number of
Array.vi
Channels

5 Samples
Display chan
to avg

Average 250
0=initialize
Stop

Figure 14.12 Demonstration of the Moving Avg Array function, which applies a low-pass filter
to an array of independent channels. The For Loop creates several channels of data, like a data
acquisition system. Each channel is filtered, and then the selected channel is displayed in both
raw and filtered forms.

input, and the same array, but low-pass-filtered, is the output. Any
number of samples can be included in the moving average, and the
averaging can be turned on and off while running. This is a particular
kind of FIR filter where all the coefficients are equal to 1; it is also
known as a boxcar filter.
Other moving averagers are primarily for use with block-mode con-
tinuous data. They use local memory to maintain continuity between
adjacent data buffers to faithfully process block-mode data as if it were
a true, continuous stream.
After low-pass filtering, you can safely decimate data arrays to reduce
the total amount of data. Decimation is in effect a resampling of the
data at a lower frequency. Therefore, the resultant sampling rate must
be at least twice the cutoff frequency of your digital low-pass filter, or
else aliasing will occur. For instance, say that you have applied a 1-kHz
low-pass filter to your data. To avoid aliasing, the time interval for each
sample must be shorter than 0.5 ms. If the original data was sampled
at 100 kHz (0.01 ms per sample), you could safely decimate it by a
factor as large as 0.5/0.01, or 50 to 1. Whatever you do, don’t decimate
350 Chapter Fourteen

without knowledge of the power spectrum of your incoming signal. You

can end up with exactly the same result as sampling too slowly in the
first place.

Timing techniques
Using software to control the sampling rate for a data acquisition
system can be a bit tricky. Because you are running LabVIEW on a
general-purpose computer with lots of graphics, plus all that operating
system activity going on in the background, there is bound to be some
uncertainty in the timing of events, just as we discussed with regard
to timestamps. Somewhere between 1 and 1000 Hz, your system will
become an unreliable interval timer. For slower applications, however,
a While Loop with a Wait Until Next ms Multiple function inside
works just fine for timing a data acquisition operation.
The best way to pace any sampling operation is with a hardware timer.
Most plug-in boards, scanning voltmeters, digitizers, oscilloscopes, and
many other instruments have sampling clocks with excellent stability.
Use them whenever possible. Your data acquisition program will be
simpler and your timing more robust.
Your worst timing nightmare occurs when you have to sample one
channel at a time—and fast—from a “dumb” I/O system that has no abil-
ity to scan multiple channels, no local memory, and no sampling clock.
Gary ran into this problem with some old CAMAC A/D modules. They
are very fast, but very dumb. Without a smart controller built into the
CAMAC crate, it is simply impossible to get decent throughput to a
LabVIEW system. Remember that a single I/O call to the NI DAQ or NI
GPIB driver typically takes 1 ms to execute, thus limiting loop cycle time
to perhaps 1000 Hz. Other drivers, typically those based on peek and
poke or other low-level operations including some DLL or CIN calls, can
be much faster, on the order of microseconds per call. In that case, you at
least have a fighting chance of making reliable millisecond loops.
If the aggregate sampling rate (total channels per second) is pressing
your system’s reliable timing limit, be sure to do plenty of testing and/
or try to back off on the rate. Otherwise, you may end up with unevenly
sampled signals that can be difficult or impossible to analyze. It’s bet-
ter to choose the right I/O system in the first place—one that solves the
fast sampling problem for you.

Configuration Management
You could write a data acquisition program with no configuration man-
agement features, but do so only after considering the consequences.
The only thing that’s constant in a laboratory environment is change,
so it’s silly to have to edit your LabVIEW diagram each time someone
 Writing a Data Acquisition Program 351

wants to change a channel assignment or scale factor. Make your pro-

gram more flexible by including configuration management VIs that
accommodate such routine changes with a convenient user interface.

What to configure
Even the simplest data acquisition systems have channel names that
need to be associated with their respective physical I/O channels. As
the I/O hardware becomes more complex, additional setup information
is required. Also, information about the experiment itself may need to
be inseparably tied to the acquired data. For instance, you certainly
need to know some fundamental parameters such as channel names
and sample intervals before you can possibly analyze the data. On the
other hand, knowing the serial number of a transducer, while useful, is
not mandatory for basic analysis. Table 14.2 is a list of configuration-
related items you might want to consider for your system.
In a sense, all this information comprises a configuration data-
base, and any technique that applies to a database could be applied
here: inserting and deleting records, sorting, searching, and of course
storing and fetching database images from disk. These are tasks for
a configuration editor. Additionally, you need a kind of compiler or
translator program that reads this user-supplied information, validates
it, and then transmits it in suitable form to the I/O hardware and the
acquisition program. (See Figure 14.13.)
The level of sophistication of such an editor or compiler is limited only
by your skill as a programmer and your creative use of other applications

TABLE 14.2 Items to Include in Configuration

I/O hardware-related
Port number, such as GPIB board or serial port selection
Device address
Slot number, where multiple I/O modules or boards are used
Module or board type
Channel number
Channel name
Channel gain, scale factor, or linearization formula
Sampling rate; may be on a per-module or per-channel basis
Filtering specifications
Triggering parameters—slope, level, AC/DC coupling
Experiment-related
Experiment identifier; short ID numbers make good foundations for data file names
Purpose or description of the experiment
Operator’s name
Start date and time
Data file path(s)
Transducer type or description for each channel
Transducer calibration information and serial number
352 Chapter Fourteen

Di
Da sk
Ed tab
ito as
rV eV
Is Is

Co
mp
ile
rV
Is
Re
a
Ac l-T
I/O qu ime
isi
Ha tio Da
rd n V ta
wa Is
re

SCX
I-10
01

SCXI

1140

SCXI

1140

SCXI

1140

SCXI

1140

SCXI

1140

SCXI

1140

SCXI

1140

SCXI

1140

SCX SCXI

MAINF
I 1140
RAME

SCXI

1140

SCXI

1140

Figure 14.13 Process flow diagram for configuration management.

on your computer. The simplest editor is just a cluster array into which
you type values. The most complex editor we’ve heard of uses a com-
mercial database program that writes out a configuration file for Lab-
VIEW to read and process. You can use the SQL Toolkit (available
from National Instruments) to load the information into LabVIEW by
issuing SQL commands to the database file. When the experiment is
done, you might be able to pipe the data—or at least a summary of the
results—back to the database. Do you feel adventurous?
Assuming that you’re using DAQmx, much of this configuration
information is addressed by the Measurement and Automation
Explorer (MAX). MAX is a convenient and feature-rich graphical-
user interface that you use to maintain and test DAQmx tasks. From
within LabVIEW, about all you need to do is to let the user pick from a
list of predefined tasks. All the other information about each channel—
scale factors, hardware assignments, and so forth—is already defined
in the file, so your LabVIEW program doesn’t need to keep track of it.
Supplementary information, such as experimental descriptions, will of
course be maintained by programs that you write.

Configuration editors
By all means, use MAX if you can. But it can’t address every imagin-
able need, so you should know something about the alternatives. Aside
from that rather elaborate application of a commercial database, there
 Writing a Data Acquisition Program 353

are some reasonable ways for you to program LabVIEW as a configura-

tion editor. There are two basic editor types: interactive editors and
static editors. An interactive editor is a LabVIEW VI that runs while
you enter configuration information, supplying you with immediate
feedback as to the validity of your entries. A static editor is generally
simpler, consisting of a VI that you run after entering information into
all its control fields. If an error occurs, you receive a dialog box and
try again. With either editor, once the entries are validated, the infor-
mation is passed on to a configuration compiler (which may actually
be part of the editor program). If the configuration is loaded from an
external file, it may pass through an editor for user updates or directly
to the compiler for immediate use.

Static editors for starters.Ordinary LabVIEW controls are just fine for
entering configuration information, be it numeric, string, or boolean
format. Since a typical data acquisition system has many channels, it
makes sense to create a configuration entry device that is a cluster
array. The cluster contains all the items that define a channel. Mak-
ing an array of these clusters provides a compact way of defining an
arbitrary number of channels (Figure 14.14).
One problem with this simple method is that it’s a bit inconvenient
to insert or delete items in an array control by using the data selection

Channels Configuration out

1 0
Channel Name
Channel Name
Channel Two
Channel Two
Channel Number
Channel Number
44
44
On Scan? ON

True
Empty cluster Configuration
array control
Channel is on scan-- add to array out

(hidden)

True
Channels Channel Name Empty name
Channel
Channel Number
On Scan?

needs a name.

Figure 14.14A static configuration editor using a cluster array. It checks to see that a channel
name has been entered and appends on scan channels to the output configuration array.
354 Chapter Fourteen

items in the array control pop-up. A solution is to include an On Scan

switch, as I did in this example. Only when a channel is on scan is
its configuration information passed along to the I/O driver. When a
channel is off scan, the analog-to-digital converter (ADC) or digitizer
may safely ignore that channel, thus saving some I/O operations. Make
the switch turn red when a channel is off scan to ensure that the user
doesn’t accidentally leave a channel turned off.
The diagram for this editor checks to see that the channel name is
not empty (you could also check for invalid characters, name too long,
etc.). If the name is empty, a dialog pops up, telling the user, “Chan-
nel 17 needs a name.” If the name is OK and the channel is on scan,
then the name and channel number are appended to the output cluster
array for use later on. You can also use the empty channel name as a
switch to take the channel off scan.
Obviously, you can add as many items as you need to the chan-
nel description cluster, but the amount of error checking and cross-
verification you could do will become extensive. Error checking is part
of making your program robust. Even this simple example needs more
checking: What if the user assigns the same channel number to two or
more channels? What if two channels have the same name?
For more complex I/O systems, a nested structure of cluster arrays
inside of cluster arrays may be required. Figure 14.15 configures

Voltage Inputs
0
0
Tag Name PI-203
Position
(0...15) Descriptive Name
Tank Pressure

Module Range
Mux No. 0 to 5 VDC (AD6)
0
Slope Intercept Units
1.00 0.00 PSIG
High Alarm Low Alarm
350.00 0.00

Alarm Print File Dialog Beep

Responses

Figure 14.15Configuration for a more complicated I/O system

(Opto-22 Optomux) requires nested cluster arrays with lots of
controls.
 Writing a Data Acquisition Program 355

Opto-22 analog input modules. The outer array selects a multiplexer

(MUX) number. Each multiplexer has up to 16 channels that are con-
figured by the inner cluster array. Since there were four possible alarm
responses, we grouped them inside yet another cluster, way down
inside. As you can see, the data structure becomes quite complex and
might be unmanageable for a casual user. We’ve found that array con-
trols are hopelessly confusing to some operators. This is a clear case of
making the user interface convenient for the programmer rather than
for the user. Such a structure might, however, be used as the output of
a more user-friendly interactive editor.
The LabVIEW Table control can also be used for configuration
entries, and it’s more likely that your users will understand it. As in
a spreadsheet, columns can be assigned to each setup parameter with
one row per channel. Tables have an advantage over cluster arrays
because tables have high information density; many channel defini-
tions will fit in a small screen area. But tables have one glaring disad-
vantage: Data validation is mandatory on each and every cell. Because
a table is just a 2D string array, the user is free to type any characters
into any cell. What happens when you are hoping for a numeric entry,
but receive something nonnumeric?
Any time you write a loop to process a Table control, you will encoun-
ter the nefarious empty row problem. It occurs when the user types
something into a cell and then deletes everything. Even though the
cell is empty, the 2D array now has another element that contains an
empty string. If you pass this 2D array to a loop for processing, it will
attempt to process the empty strings. You can test for empty strings
later or use the VI in Figure 14.16, Remove Empty Table Rows.
It removes any number of trailing empty rows. Size the Table control
so that it displays only the proper number of columns; then hide the
horizontal scroll bars to keep the user from entering too many columns
of data.
Figure 14.17 is a static editor that interprets the data in a Table con-
trol. It does no error checking, but that could be added right after each
of the Index Array functions. As a minimum, you have to verify that all
numbers are in range and that the channel name is acceptable. Cross-
checking for more complex data inconsistency gets really interesting.
For instance, you might have to verify that the right number of chan-
nels is assigned according to the multiplexer type. Perhaps a future
version of LabVIEW will include a more powerful Table control that
has many of these data filter features, such as numeric validation.

Interactive configuration editors are more ver-

Interactive editors.
satile and more user-friendly than static editors because they pro-
vide instant feedback to the user. You can add pop-up windows, add
356 Chapter Fourteen

Table with empty

Table trailers removed
0 1 0 1
10 100 10 100
20 100 20 100
30 60 30 60
40 20 40 20
50 0 50 0

The cells in this row

are empty, but LabVIEW
thinks this table has
sizes in sizes out
seven rows anyway
0 7 0 6

2 2

For each row, see if all columns are empty.

Table with empty

Empty table False
trailers removed
0
0
Table
Some data in row.
Keep it.

sizes out
sizes in True

Entire row is empty.

Ignore it.

Figure 14.16This VI gets rid of empty table rows. Erasing a cell completely
does not remove that row of the 2D array. But this VI does.

Config Table Config Out

Channel Name Mux Slot Chan 0 Name
Pressure 1 1 1 7
Pressure 2 1 2 4 Pressure 1

Supply Volts 1 2 9 Mux Slot Chan

1 1 7

For each row...

Config Table Bundle Config Out

name 0
mux 1
slot 2
chan 3

Figure 14.17 This is a simple configuration editor that inter-

prets the contents of a Table control and converts the data into
a cluster array.
 Writing a Data Acquisition Program 357

status displays, or even turn LabVIEW into a menu-driven system.

Plan to do some extra programming if you want these extra features.
Elaborate interactive editor projects are among the most challenging
programming tasks we’ve ever tackled.

Popup editors. Say you have several different kinds of I/O modules,
and they all have significantly different configuration needs. If you try
to accommodate them all with one big cluster array, various controls
would be invalid depending on which module was selected. You really
need separate input panels for each module type if you want to keep
the program simple.
One solution is to put several buttons on the panel of the main
configuration VI that open customized configuration editor subVIs
(Figure 14.18). Each button has its mechanical action set to Latch
When Released, and each editor subVI is set to Show front panel
when called. The editor subVIs do the real configuration work. Note
that they don’t really have to be any fancier than the static editors
we already discussed. When you create one of these pop-up subVIs,
remember to disable Allow user to close window in the Window
Options of the VI Setup dialog. Otherwise, the user may accidentally
close the window of the VI while it’s running, and then the calling VI
will not be able to continue.

Figure 14.18 This configuration editor model

uses subVIs that appear when an appropri-
ate button is pressed. The subVI is the actual
user interface (in this case, a configuration
editor) and may do other processing as well.
Each subVI is set to show the front panel when
called.
358 Chapter Fourteen

Information may be returned by each editor subVI for further pro-

cessing, or each editor can function as a stand-alone program, doing all
the necessary I/O initialization, writing of global variables, and so forth.
A worthy goal is for each editor to return a standard data structure
(such as a cluster array), regardless of the type of I/O module that it
supports. This may simplify the data acquisition and analysis processes
(see the section entitled “Configuration Compilers” which follows).
Another way you can accommodate modules with different config-
uration needs is to use Property nodes to selectively hide controls
when they don’t apply. When the user selects module type A, the spe-
cial controls for module B are hidden, and so forth. Note that you can
selectively hide controls that are part of a cluster: Either pop up on the
particular control in the cluster you wish to hide, and select Create >>
Property Node, or get the array of control references to all the con-
trols in the cluster and sort out which ones you want. There’s another
way to tackle this problem: Place the clusters for each module on a tab
control. This is the simplest and often the best solution.
Here is a model for an interactive editor hierarchy that Gary origi-
nally built to support SCXI analog input modules. It turned out to be
quite versatile, and we’ve used it for many other kinds of instruments.
Figure 14.19 shows the window that appears when the user clicks a
button on the top-level configuration manager VI. The figure also shows
a simplified version of the diagram. This program relies on a global
variable, Inst Config Globals, to pass the configuration cluster array
along to other parts of the data acquisition system. The configuration
data is loaded from the global at start-up and then circulates in a shift
register until the user clicks the Exit button. Two subVIs act on the
configuration data: Read/Write SCXI Config and Edit SCXI Con-
fig. The Read/Write SCXI Config VI can either load or store an image
of the cluster array to a binary file. The file path is chosen by another
subVI, Change Config File, which opens as a dialog box when called.
It permits the user to pick an existing file or create a new one. The real
work of editing the configuration occurs inside Edit SCXI Config.
Indeed, the tough part of this interactive editor is the Edit SCXI Con-
fig VI. Figure 14.20 shows the front panel, which was designed purely
for user comfort and efficiency with no thought whatsoever for the com-
plexity of the underlying program. So, where’s the diagram? It’s way
too complicated to show here. It’s based on a state machine that looks
at the present control settings and then modifies the display accord-
ingly. For instance, if the user changes the Module # control, all the
other controls are updated to reflect the present setting of channel 1
of that module.
This is accomplished by reading the settings from the configuration
cluster array and loading them into the controls with local variables.
 Writing a Data Acquisition Program 359

Present Config
Save config to file
0
Module Type Slot #
Reload config from file SCXI-1120 1
8-channel isolated input
Module Gain
Switch config file 1
Channels CJC IC Sensor
0
Edit channel config Name

First channel
Signal Type
EXIT Type J thermocouple

Gain Scale
Path
100 1.000000 Scan?
Maxima:SCXI_Config.CFG
Offset Engr Units ON

0.000000 DegC

Switch files Read config Write config Edit config Path

from file to file
Inst Config Inst Config
Globals True True True True Globals
SCXI mods SCXI mods
Read Write

Present Config
Change file From file To file Edit

Don't hog the machine 200

Exit

Figure 14.19 The SCXI Analog Config Dialog VI helps the user manage SCXI analog input modules. The
configuration is stored in a global variable for use elsewhere in the data acquisition program, and it can be
loaded from and saved to files. (False frames of all the Case structures simply pass each input to its respec-
tive output.)

Also, the Module Gain and channel Gain controls are selectively dis-
abled (dimmed) by Attribute nodes. For instance, if the user picks an
SCXI-1120 module, each channel has its own gain setting, so the Gain
control is enabled while the Module Gain control is disabled.
This is the fanciest configuration editor that Gary could design in
LabVIEW, and it uses just about every feature of the language. This
editor is particularly nice because, unlike in the static editor with its
cluster array, all the controls update instantly without any clumsy array
navigation problems. However, the programming is very involved—as
it would be in any language. So we suggest that you stick to the sim-
pler static editors and use some of the ideas described here until you’re
really confident about your programming skills. Then design a nice user
interface, or start with my example VI, and have at it. If you design the
360 Chapter Fourteen

This button opens a subVI

that permits changes to the
table, but with error checking.

Edit module list OK

Available modules
This button returns to the
Module # Module type Slot # calling VI with no changes.
CANCEL
1 SCXI-1120 8-channel isolated input 1
Changing the Module 2 SCXI-1120 8-channel isolated input 2 Module Gain is disabled
number causes all the 3 SCXI-1100 32-channel mux 3 (using an Attribute Node)
controls below to update when the selected module
to the settings of Channel 1 does not use this feature.
of that module.
Module # 2 Module Gain 1 CJC Thermistor
Changing the Channel
number causes all of its
associated controls to Channel calibration All of these controls
update to their current Channel # 9 ENTER
accept user changes.
settings. Scale Offset Local Variables update
Name their contents when the
1.00 0.00
Module or Channel
My channel name
Result = volts * scale + offset number is changed.

Signal Type Gain Engr Units

Type E thermocouple 100 DegC

Status display informs

user of any errors.
Channel number out of range for this module type.

Figure 14.20 Panel of the SCXI configuration editor, showing some of the features that you can implement if
you spend lots of time programming.

overall configuration program in a modular fashion, you can replace

the simple editor with the fancy editor when it’s complete.
Only you can decide if great expenditures of effort are necessary. If
you have only one or two users, or if they only have to deal with con-
figuration management once a month, then it’s probably not worth the
effort. On the other hand, you may be able to develop an elaborate edi-
tor VI that can serve as a template for other purposes, thus leveraging
your development efforts. When Gary wrote the SCXI configuration
manager, he spent some time designing it with the intent that much of
the code would be reused for other types of I/O hardware in his applica-
tion. That’s why the VI in Figure 14.19 has such a clean design.

Status displays. Since an interactive editor is a program that runs all

the time, it can do such things as entry validation even as the user
changes settings. You can give the user feedback by adding a status
display that describes any errors or inconsistencies among the various
settings. In Figure 14.21, for instance, you have a choice of several I/O
modules, each having a certain limited range of channels. Any disagree-
ment between module type and channel quantity needs to be rectified.
A ring indicator contains predefined status messages. Item 0 is OK to
 Writing a Data Acquisition Program 361

Status (ring indicator)

Only 1 to 4 channels permitted with 1

Not So Fancy module.

Module Type Number of channels

Not So Fancy EXIT
6

Module Type Not So Fancy

too many
Status

4
Number of ok
channels max chans

exit

Figure 14.21 How to use a ring indicator as a status display. The While
Loop runs all the time, so the user’s settings are always being evaluated.
The ring indicator contains predefined messages—in this case, item 0 is
OK to continue. Any real work for this configuration editor would be done
inside the While Loop.

continue, while item 1 is the error message shown. The status has to be
zero to permit the Exit button to terminate the While Loop.
You can also use a string indicator to display status messages, feed-
ing it various strings contained in constants on the diagram. The string
indicator can also display error messages returned from I/O operations
or from an error handler VI. You would probably use local variables to
write status messages from various places on your diagram, or keep
the status message in a shift register if you base your solution on a
state machine. That’s what we did with the SCXI configuration editor.
Some configuration situations require constant interaction with the
I/O hardware to confirm the validity of the setup. For instance, if you
want to configure a set of VXI modules, you might need to verify that
the chosen module is installed. If the module is not found, it’s nice to
receive an informative message telling you about the problem right
away. Such I/O checks would be placed inside the overall While Loop.
A good status message is intended to convey information, not admonish
the user. You should report not only what is wrong, but also how to cor-
rect the problem. “ERROR IN SETUP” is definitely not helpful, although
that is exactly what you get from many commercial software packages.
Dialogs can also be used for status messages, but you should reserve
them for really important events, such as confirming the overwriting of a
file. It’s annoying to have dialog boxes popping up all the time.
362 Chapter Fourteen

Menu-driven systems. When PCs only ran DOS, and all the world was
in darkness, menu-driven systems were the standard. They really are
the easiest user interfaces to write when you have minimal graphics
support. The classic menu interface looks like this:

Choose a function:
1: Initialize hardware
2: Set up files
3: Collect data
Enter a number >_

In turn, the user’s choice will generate yet another menu of selections.
The good thing about these menu-driven prompting systems is that the
user can be a total idiot and still run your system. On the other hand,
an experienced user gets frustrated by the inability to navigate through
the various submenus in an expedient manner. Also, it’s hard to figure
out where you are in the hierarchy of menus. Therefore, we introduce
menus as a LabVIEW technique with some reluctance. It’s up to you to
decide when this concept is appropriate and how far to carry it.
The keys to a successful menu-driven system are aids to navigation
and the ability to back up a step (or bail out completely, returning to
step 1) at any time. Using subVIs that open when called allows you to
use any kind of controls, prompting, status, and data entry devices that
might be required.
A LabVIEW menu could be made from buttons, ring controls, or slid-
ers. If you use anything besides buttons, there would also have to be a
Do it button. Programming would be much like the first pop-up editor
example, earlier in this section. To make nested menus, each subVI
that opens when called would in turn offer selections that would open
yet another set of subVIs.
If you lay out the windows carefully, the hierarchy can be visible
on the screen. The highest-level menu VI would be located toward the
upper-left corner of the screen. Lower-level menus would appear offset
a bit lower and to the right, as shown in Figure 14.22. This helps the
user navigate through nested menus. LabVIEW remembers the exact
size and location of a window when you save the VI. Don’t forget that
other people who use your VIs may have smaller screens.

Configuration compilers
A configuration compiler translates the user’s settings obtained
from a configuration editor into data that is used to set up or access
hardware. The compiler may also be responsible for storing and recall-
ing old configuration records for reuse. The compiler program may or
may not be an integral part of an editor VI.
 Writing a Data Acquisition Program 363

Figure 14.22Layout for hierarchical menu subVIs that helps

users navigate. The Exit button returns you to the previous
menu. You might use the VI Setup dialog to customize the
window presentation; the Run arrow is not needed.

The control layout on a configuration editor’s panel should be optimized

for efficient user interaction. But those controls may not be very efficient
when it comes time to send information to a real-time data acquisition
driver or file storage VI. It would be very inefficient, for example, to have
the driver interpret and sort arrays of strings that describe channel
assignments every time the driver is called. Rather, you should write a
compiler that interprets the strings at configuration time and creates a
numeric array that lists the channel numbers to be scanned. Your objec-
tive is to write a program that does as little as possible during data
acquisition and analysis phases, thus enhancing throughput.
Figures 14.23 through 14.25 illustrate a simple compiler that supports
some digitizer modules. Each module has four channels with independent
setups. Figure 14.23 is a cluster array into which the user has entered
the desired setup information, perhaps by using an interactive editor VI.
Channel setup information is in the form of four clusters, which makes
it easy for the user to see all the setups at once. However, this is not an
efficient way to package the information for use by an acquisition VI.
Figure 14.24 shows a more desirable structure, where each channel is an
element of an array. The compiler must make this translation.
Here are the steps to be taken by this configuration compiler, whose
diagram appears in Figure 14.25:
1. Convert the individual channel clusters into an array.
2. If the module is online, build the output configuration cluster and
append it to the output array.
364 Chapter Fourteen

0
Model GPIB Addr Slot # (N) This digitizer is:
User
Config LC 6810 7 11 1 2 4
# Active Chans
ON-LINE
Channel Name Coupling Sensitivity Offset

Chan. 1: Channel_One Non-Inverting 4.096 V Bipolar

DC Coupled

Chan. 2: Channel_Two Differential 10.24 V Bipolar

DC Coupled

Chan. 3: Channel_Three Differential 4.096 V Unipolar +

AC Coupled

Chan. 4: Channel_Four Inverting 10.24 V Bipolar

AC Coupled

Figure 14.23 The input to this configuration compiler for digitizers is a cluster array that a
user has filled in. The four clusters (channel names, coupling, sensitivity, and offset) need to be
checked for consistency and converted to arrays for use by the acquisition VIs.

3. Check the settings for consistency (for example, there may be limita-
tions on sensitivity for certain coupling modes).
4. Initialize each module and download the settings.
5. Write the output configuration cluster array to a global variable
which will be read by the acquisition VI.

We left out the gory details of how the settings are validated and
downloaded since that involves particular knowledge of the modules
and their driver VIs. Even so, much of the diagram is taken up by

0
Model GPIB Addr Slot # (N)
Config
Out 1 LC 8210 1 1
1 2 4
# Active Chans

0
Channel Name Coupling
Channel
No. Channel_One Non-Inverting
DC Coupled

Sensitivity Offset
4.096 V Bipolar

The compiler’s output is a cluster array. Note that it carries all the
Figure 14.24
same information as the input cluster array, but in a more compact form.
 Writing a Data Acquisition Program 365

For each digitizer defined...

empty config True Inst Config

cluster array This digitizer is ON-LINE. Process its configuration. Globals

CAMAC

0 [0..1]
Config Out
Do setups for 6810 modules

Model model
GPIB Addr Various driver calls would
Slot # (N) be made at this point to
User Config # Active Chans download settings to the
Channel Name module and verify that the
Coupling settings are consistent.
Sensitivity (Omitted for clarity)
Offset
On-Line?
Convert 4
clusters into
1 cluster

Figure 14.25 Diagram for the simple compiler. Data types are converted by the inner For Loop. Set-
tings are checked for consistency, and digitizers are initialized by the inner Case structure. If a digi-
tizer is online, its configuration is added to the output array, which is passed to the acquisition VIs by
a global variable.

bundlers and unbundlers and other conversion functions that are

required to reformat the incoming data. This is typical of these com-
pilers, and it sometimes gets rather messy because of the number of
items and special conditions that you have to deal with. Where pos-
sible, encapsulate related parts of the compiler in subVIs to make the
diagram easier to understand and more compact. You would probably
put all the driver functions for a given type of digitizer module in a
subVI because they are logically related.
Somewhere in the edit or compile phase you should try to communi-
cate with each I/O device to see if it responds, and report an error if it
doesn’t. It would be uncouth to permit the user to get all the way to the
data acquisition phase before announcing that an important instru-
ment is dead on arrival.

Saving and recalling configurations

Another useful function that the compiler can perform is the creation
of a permanent record of the configuration. The record might be a
spreadsheet-format text file suitable for printing, a special format for
use by a data analysis program, or a LabVIEW datalog or binary file
that you can read back into the configuration program for later reuse.
It makes sense to put this functionality in with the editors and compil-
ers because all the necessary information is readily available there.

Many systems we’ve worked on

Printable records (spreadsheet format).
needed a hard-copy record of the experiment’s configuration for the
366 Chapter Fourteen

Create strings for each module defined

string under construction
header
String to
Model\tGPIB\sAddres
0 [0..1] spreadsheet
s\tSlot\sNo.\r
file
LC6810\t

Config Model
GPIB Addr
Slot
Channels Name\tCoupling\tSen
sitivity\tOffset\r

For each channel %s\t%d\t%d\t%d\r

Channel Name
Coupling
Sensitivity
Offset Format Into String

Figure 14.26 This is a routine that converts the configuration cluster of Figure 14.24 into a
tab-delimited text file for documentation purposes. As you can see, there’s lots of string build-
ing to do.

lab notebook. The direct approach is to generate a text file that you can
load and print with a spreadsheet or word processor. Add tab characters
between fields and carriage returns where appropriate to clean up the
layout. Titles are a big help, too. Figure 14.26 is a diagram that inter-
prets the configuration cluster array of Figure 14.24 into tab-delimited
text. There is a great deal of string building to do, so don’t be surprised
if the diagram gets a little ugly.
Here is what this text converter VI produces, as interpreted by a
word processor with suitable tab stops. As you can see, most of the set-
tings are written in numeric form. You could add Case structures for
Coupling, Sensitivity, and Offset to decode the numbers into something
more descriptive.

Model GPIB Address Slot No.

LC6810 1 7
Name Coupling Sensitivity Offset
Channel One 7 2 0
Channel Two 1 4 2
Channel Three 0 3 1
Channel Four 4 5 1

Another way to print a configuration record is to lay out a suitable

LabVIEW front panel and print that. You can use all the usual indica-
tors as well as formatted strings displayed in string indicators, if that
helps you get more information crammed onto one screen. The obvious
 Writing a Data Acquisition Program 367

Current Open VI Close LV Object

VI's Path Reference Invoke Node Reference
VI
Print Panel To Printer
Entire Panel?
F: Print only the
visible part
Figure 14.27 The VI Server can do all sorts of run-time magic. Here, it’s generating a
printout of a panel—but only the part that’s visible.

way to print the panel is to use the Print command in the File menu,
but you can also print programmatically. Here’s how. First, create
a subVI with the displays you want to print. If there are extra inputs
that you don’t want to print, pop up on those items on the diagram
and choose Hide Control. Second, turn on the Print At Completion
option in the Operate menu to make the VI print automatically each
time it finishes execution. All the print options in the VI Setup dialog
will apply, as will the Page Setup settings. When you call this carefully
formatted subVI, its panel need not be displayed in order to print.
Another way to generate a printout automatically is through the VI
Server. This method is quite versatile since you can have the VI Server
print any part of any VI, any time, any place. For instance, you can print
the visible part of a VI’s panel to the printer (Figure 14.27). Or you could
route the front-panel image to an HTML or RTF file for archiving.

Manual save and recall. Your configuration manager should be able to

recall previous setups for reuse. Otherwise, the user would have to type
in all that information each time. You can use the Datalogging func-
tions in the Operate menu (see “Writing Datalog Files” in Chapter 7).
The Datalogging functions make it easy to retrieve setups at a later
time. Each record has a timestamp and date stamp so you can see when
the setup was created. Thus, your operator can manually save the cur-
rent front panel setup and sift through old setups at any time.
The crude way to save setups is by entering values into controls
and then choosing Make Current Values Default from the Operate
menu. However, this option is not available when the VI is running, so
it’s more of a system manager configuration item than something your
operators would use. But these default values are appropriate for items
that the operator doesn’t need to change, and this is the easiest and
most reliable way of saving setup data.

Automatic save and recall.We prefer to build the configuration editor

with binary file support as previously described for the Read/Write
368 Chapter Fourteen

SCXI Config VI. This solution is very convenient for the user. Previous
settings can be automatically loaded from a standard file each time
the configuration editor is called, and then the file is updated when an
editing session is complete.
Gary wrote the Setup File Handler VI to make this easy (Johnson
1995). It’s also a handy way to manage front panel setups where you
want to save user entries on various controls between LabVIEW ses-
sions. For example, it’s nice to have the last settings on run-time con-
trols return to their previous states when you reload the VI. The Setup
File Handler is an integrated subVI which stores a cluster containing
all the setup data in a shift register for quick access in real time and in
a binary file for nonvolatile storage. Multiple setups can be managed
by this VI. For instance, you might have several identical instruments
running simultaneously, each with a private collection of settings.
As you can see from the panel in Figure 14.28, the setup data is
a cluster. Internally, setup clusters are maintained in an array. The
element of the array to be accessed is determined by the Setup Index
control. The Setup cluster is a typedef that you should edit with the
Control Editor, and probably rename, for your application. You usually
end up with a very large outer cluster containing several smaller clus-
ters, one for each subVI that needs setup management.

Setup file path (""; dialog) Path in use

Setup Index (0) Duplicate Index
Setup Setup
Mode (1=read file) Error out
Error in (no error)
Setup File Handler.vi

Setup Index (0) Mode (1=read file) Duplicate Index

0 0

Setup file path (""; dialog) Path in use

RAM Disk:MyConfig.CFG

Setup Setup

Numeric Boolean Numeric Boolean

444.00 ON 444.00 ON

String String

a data string a data string

Figure 14.28 The Setup File Handler VI stores and retrieves setup clusters
from disk. It’s very useful for configuration management and for recalling
run-time control settings at a later time.
 Writing a Data Acquisition Program 369

Initialize controls: Read setup Your main program, in which Update the setup database
database, then use Local Variables the user updates the controls. and write it out to disk.

A number
Numeric A number Numeric
A Boolean Boolean
Boolean A Boolean
Setup Index A string String Duplicate Index
String A string
index
the setup
error
Error in
Read mem Write both Error out
200
Read from memory (faster) Exit
unless you really want the
disk image.

Figure 14.29 Using the Setup File Handler in a typical application.

To see how this all works, look at the Setup File Handler example
in Figure 14.29. It begins by calling the Setup File Handler VI with
Mode = Init to create or open a setup file. If the file doesn’t exist, the
settings in a default setup cluster are used for all setups. In most appli-
cations, all subsequent calls to Setup File Handler would be performed
in subVIs. In the example subVI shown in Figure 14.29, the Setup File
Handler is the source for control initialization data. Local variables
initialize controls for the user interface, or the data might be passed
to a configuration subVI. After the main part of the program finishes
execution, the new control settings or configuration data are bundled
back into the setup cluster, and the Setup File Handler updates the
memory and disk images. Bundle and Unbundle by Name are very
handy functions, as you can see. Try to keep the item names short in
the setup cluster to avoid huge Bundlers.
Where should the setup file be saved? You can use the LabVIEW file
constant This VI’s Path to supply a guaranteed-valid location. This
technique avoids a familiar configuration problem where you have a
hand-entered master path name that’s wrong every time you move the VI
to a new computer. In your top-level VI, insert the little piece of code
shown in Figure 14.30 to create a setup file path. If the current VI is in a
LabVIEW library, call the Strip Path function twice as shown. If the VI
is in a regular directory, call Strip Path once. This technique even
works for compiled, stand-alone executables created by the LabVIEW
Application Builder.
Another way to obtain a guaranteed-valid (and locatable) path is
with the Get System Directory VIs written by Jeff Parker and pub-
lished in LabVIEW Technical Resource (Parker 1994). He used CINs to
locate standard directories, such as System: Preferences on the Macin-
tosh or c:\windows on the PC. (You could simply assume that the drive
and directory names are always the same, but sooner or later someone
will rename them behind your back!)
370 Chapter Fourteen

Current VI's Path

Setup file path

MySetup.SET
Figure 14.30 Use the path manipulation functions
Strip Path and Build Path to create a guaranteed-
valid path for your setup files. If the current VI is in
a LabVIEW library, call Strip Path twice, as shown.
Otherwise, call it only once.

A Low-Speed Data Acquisition Example

Life is unfair. You spend all this time reading about high-powered
methods of building a data acquisition system, only to find out you
need a system up and running now—not next week, not next month.
What should you do? As we’ve stated before, your best bet is to program
by plagiarizing. Sift through the examples that come with LabVIEW
and see if there’s something that might do at least 50 percent of the job.
Or, search online and tap into the works of others. Eventually, you will
build a personal library of VIs that you can call on to quickly assemble
working applications. Following is an example that qualifies as a real
solution to everyday data acquisition needs.
Here is a low-speed data acquisition VI that uses the DAQ Assistant
and Write to Measurement File Express VIs. It’s flexible enough to
be the foundation of a somewhat larger application. It is intended for
situations in which you need to acquire, display, and store data from a
few analog channels at speeds up to perhaps 10 samples per second—a
very common set of requirements. The diagram in Figure 14.31 shows
the basic VI.
Express VIs provide a simple starting point when you have to get
going fast. The DAQ Assistant walks you through configuring your
DAQ measurement in just a few simple steps. Figure 14.32 shows the
DAQ Assistant’s configuration dialog configured to measure a single
point from each of eight voltage channels. You can interactively test the
configuration and change the input signal range if needed. The DAQ
Assistant will even show you how to connect the signals for each chan-
nel. This example measures simple voltages, but the DAQ Assistant
is fully capable of handling thermocouples, RTD’s, strain gauges, and
just about any measurement you might typically come across in the
lab or the field. Once you’re done, click OK and the DAQ Assistant will
generate the LabVIEW code for your application. You can leave it as
an Express VI or choose to Open Front Panel and convert the Express
 Writing a Data Acquisition Program 371

Figure 14.31 Diagram of a simple low-speed data acquisition VI using Express

VIs. It’s useful for a few channels at sampling rates up to about 10 Hz. This is a
simple, but useful, example to which you can add features as required.

Figure 14.32 The DAQ Assistant Configuration Wizard

walks you through your measurement configuration. You
can interactively test and modify until it’s right.
372 Chapter Fourteen

block to a standard VI. Although once it’s converted to a VI, you can no
longer use the Wizard. In our example we’ve left it as an Express VI.
For timing we’re using the Timed Loop set to run once every 1000 ms.
We covered the Timed Loop in detail in Chapter 5, “Timing.” The Timed
Loop gives us a high-priority, fairly deterministic software timing
mechanism for our data acquisition. Once the Timed Loop is started, it
will interrupt any other process in order to run at its scheduled time.
We are writing the data to a text-based LabVIEW measurement file
each iteration of the loop. The system overhead is fairly low, we only have
eight channels at 1 Hz, and so we can afford to save them as text data.
If we needed something faster or if file size were an issue, we could use
the binary TDM format. Chapter 7, “Files,” has some more information
on these and other types of file I/O in LabVIEW. Figure 14.33 shows the
configuration dialog for the Express VI Write to Measurement File. It
creates a text-based LabVIEW measurement file with an informational
header the first time it is called. Each subsequent time it appends the
timestamped data to the file. This VI takes only a few minutes to put
together, yet it can solve many problems. We hope you find it useful.

Figure 14.33 Dialog for the Express VI Write to Measurement File. It creates a
text-based LabVIEW measurement file (lvm) and appends timestamped data
on each iteration.
 Writing a Data Acquisition Program 373

Medium-Speed Acquisition and Processing

Many applications demand that you sample one or more analog chan-
nels at moderate speeds, on the order of tens to thousands of samples
per second, which we’ll call medium-speed acquisition. At these speeds,
you must have the I/O hardware do more of the time-critical work.
The key to any moderate- and high-speed acquisition is hardware-
timed, circular buffered I/O using direct memory access (DMA).
Although it sounds complicated, DAQmx sets this up for you transpar-
ently. Here’s how the technology works:

Timed buffered I/O. For DAQ applications running at speeds greater

than a few tens of hertz, you must turn to hardware-timed buffered I/O.
Every DAQ board has a piece of hardware called a first-in, first-out
(FIFO) memory buffer that is located on the plug-in board. As data is
read in, samples from the ADC are temporarily stored in the FIFO pend-
ing a bus transfer to your computer’s main memory, giving the software
extra time to take care of other business. An important piece of informa-
tion is the size of the available FIFO buffer, which varies among models
of plug-in boards. For instance, an E-series board has at least 512 words
(a word is 16 bits, or one sample). If you are collecting data from a single
channel at 10,000 samples per second, the FIFO buffer will fill up in
51.2 ms. If there is any overhead at all, the driver will not be able to
get around to uploading the data before it is overwritten. If your Lab-
VIEW program always had to generate or read each individual sample,
there would be little hope of sampling at accurate rates much beyond 10
Hz. With buffered I/O, precision hardware timers control the real-time
flow of data between memory buffers and the ADCs and DACs. Buffered
operations can be triggered by hardware or software events to start
the process, thus synchronizing the process with the physical world.
LabVIEW transfers data between the plug-in board and a buffer when-
ever the buffer is full (for inputs) or empty (for outputs). These opera-
tions make maximal use of hardware: the onboard counter/timers, and
any DMA hardware that your system may have.

DMA. The technique used to get memory off the board and in to mem-
ory is called direct memory access. DMA is highly desirable because
it improves the rate of data transfer from the plug-in board to your
computer’s main memory. In fact, high-speed operations are often impos-
sible without it. The alternative is interrupt-driven transfers, where
the CPU has to laboriously move each and every word of data. Most
PCs can handle several thousand interrupts per second, and no more.
A DMA controller is built into the motherboard of every PC, and all
but the lowest-cost plug-in boards include a DMA interface. The latest
374 Chapter Fourteen

boards have multiple DMA channels and bus mastering capability.

When a board is allowed to take over as bus master (in place of the
host CPU), throughput can increase even more because less work has
to be done at interrupt time. With DMA, you can (theoretically, at least)
acquire data and generate waveforms nearly as fast as your computer’s
bus can move the data.

Circular buffered I/O. Circular buffered I/O is a powerful technique

because the buffer has an apparently unlimited size, and the input or
output operation proceeds in the background with no software over-
head. A circular buffer is an area of memory that is reused sequentially.
Data is streamed from the DAQ hardware to the circular buffer by the
DMA controller. When the end of the memory buffer is reached, the
buffer wraps around and starts storing at the beginning again. Hope-
fully you’ve had time to read the data before it has been overwritten.
The only burden on your program is that it must load or unload data
from the buffer before it is overwritten. We really like circular buff-
ered I/O because it makes the most effective use of available hardware
and allows the software to process data in parallel with the acquisition
activity.
Figure 14.34 shows the DAQmx example VI Cont Acq&Graph
Voltage To File (Binary).vi (examples >> DAQmx >> Analog In >>
Measure Voltage.llb). This VI follows our canonical DAQ model where
we initialize outside the loop, acquire and process (or save) inside the
loop, and close out our references after the loop. It’s all tied nicely
together by chaining the error I/O to enforce dataflow. When we config-
ure our sample clock, DAQmx configures the board for hardware-timed
I/O. By enabling continuous samples, DAQmx automatically sets up a
circular buffer. Data is streamed to disk inside the loop as raw binary
data. This example can easily stream data from multiple channels at
10,000 samples per second until your hard drive fills up. For more DAQ

Figure 14.34 The Cont Acq&Graph Voltage To File (Binary).vi. This medium-speed example can acquire and
save binary data to disk at 10,000 samples per second until you run out of storage.
 Writing a Data Acquisition Program 375

examples you can modify to fit your application needs, look at the
DAQmx examples shipping with LabVIEW.

Bibliography
Chugani, Mahesh, et al.: LabVIEW Signal Processing, Prentice-Hall, Englewood Cliffs,
N.J., 1998.
Johnson, Gary W.: “LabVIEW Datalog Files,” LabVIEW Technical Resource, Vol. 2, No. 3,
Summer 1994. (Back issues are available from LTR Publishing.)
Johnson, Gary W.: “Managing Front-Panel Setup Data,” LabVIEW Technical Resource,
Vol. 3, No. 1, Winter 1995. (Back issues are available from LTR Publishing.)
Parker, Jeff: “Put config Files in Their Place!” LabVIEW Technical Resource, Vol. 2, No. 3,
Summer 1994. (Back issues are available from LTR Publishing.)
Press, William H., et al.: Numerical Recipes in C, Cambridge Press, New York, 1990.
This page intentionally left blank
 15
Chapter

LabVIEW RT

In just a few years, LabVIEW RT has grown from a niche product to

become a staple of automation. LabVIEW RT on one of National Instru-
ments’ many industrial computers conveniently fills the gap between
a Programmable Logic Controller (PLC) and a PC. The result is
a whole new class of industrial controller called the Programmable
Automation Controller (PAC). PACs fill a need for high-loop-rate,
closed-loop control systems requiring advanced analysis and integra-
tion with the corporate network. And PACs do it for a reasonable cost.
Whether you’re programming a PAC or just need a reliable DAQ sys-
tem in the lab, LabVIEW RT has the hardware and software to fill most
needs. Applications written in LabVIEW RT are deployed on an embed-
ded target (one with no keyboard, mouse, or display) and run under
a commercial real-time operating system (RTOS). You build your
application under Windows on a regular PC, and then you download
the code via Ethernet to run on the real-time target system. All the
functionality of regular LabVIEW is available. LabVIEW RT and PACs
let us attack a whole new class of applications while maintaining the
familiar ease of development and use for which LabVIEW is famous.
In this chapter, we’ll start with some background information about
real-time systems and LabVIEW RT, and then we’ll cover the important
issues that crop up when you develop real-time applications, including
timing, multithreading, task scheduling, and task priorities. Let’s start
off by defining what we mean by real time.

Real Time Does Not Mean Real Fast

There are several characteristics that define a real-time application.
Too often we hear the term real time misused as meaning real fast.
LabVIEW G code will not run any faster under a real-time operating

377
378 Chapter Fifteen

system than it will under any other operating system! LabVIEW is

compiled into machine code and runs at the speed of the processor.
Only a faster CPU will give you faster code. What the RTOS gives you
that your desktop system can’t is determinism, which is perhaps the
most important factor in defining real time.
Determinism means that you can accurately predict or determine
when a section of your program will execute. If you are operating a
closed-loop controller on the factory floor, you may have a calculation
and a set point update that you need to make after each analog input
reading. It’s just as important to your process that the update be made
within a critical time window (meet the deadline) as it is that your
program make the correct calculation. How late or early the code is
in actually running with respect to the desired time is known as its
latency or worst-case response time. If your process will fail if you’re
1 s late, then 1 s is your real-time latency tolerance. As a benchmark,
with LabVIEW RT and software timing alone we can easily achieve
1-kHz closed-loop control with latencies of less than 50 µs. With hard-
ware timing, latency can drop to nanoseconds.
To look at this determinism issue in another way, let’s consider the
requirements of an embedded controller for a robot on a factory assem-
bly line. As product comes down the line, the robot needs to put its arm
in the right place at exactly the right time to do its job. The correctness
of the robot’s program depends as much on the robot being on time as
it does on making the right calculation to be in place.
A real-time system can be classified as hard real time or soft real
time based partly on the consequences of missing a timing deadline.
A soft real-time system allows for some deadlines to be missed with
only a slight degradation in performance but not a complete failure.
In a soft real-time system, the assembly line may be designed so that
the product will stay in position until the robot can come do its job. The
assembly line may have to slow down to compensate for any delays
that result from the robot being a little late, but the end result is that
no harm is done other than a little less product comes off the line.
In a hard real-time system, the robot would have only a very small
time window when the product was there. If the robot did not do its job
at the right time, the product would be gone or worse. When a system
has tasks that have to be run on time or else the system will fail, it is
a hard real-time system. In some cases, this failure to meet a deadline
could result in the loss of life and/or property.
An RTOS is an operating system (OS) that has been designed to do
one thing: run your code in a predictable manner with minimal latency.
That means it meets the requirements for hard real time. In contrast,
all modern desktop operating systems operate on a fairness basis. Each
application or process gets a little bit of time on the CPU regardless of
 LabVIEW RT 379

its priority. LabVIEW on a desktop computer is subject to the timing

uncertainties caused by the desktop OS, and we cannot be sure that the
OS will not suspend the execution of our most important, time-critical
VI while it goes out to perform some routine maintenance. We have all
seen this happen when a background event briefly interrupts a high-
priority acquisition and control loop. Clearly, this kind of soft real time
cannot be used to generate or respond to time-critical tasks in which the
latency needs to be better than perhaps a few hundred milliseconds.*
Because LabVIEW RT runs on an RTOS, we can write software that
will meet a guaranteed deadline. But this predictability comes with a
price: We have to pay close attention to threading, priorities, memory
management, and other tricky technical issues that we’ll cover later in
this chapter. Most important, we need to have realistic expectations.
Designing software for a real-time system requires different goals
from designing software for use with a desktop OS. Rapid prototyp-
ing is one of the areas in which LabVIEW excels, but it can lead to
some bad habits. Too often we get in the habit of just throwing things
together until we get the right answer, but without paying much atten-
tion to how efficiently our code executes. Because performance is just
as important in a real-time system as having the right answer, we need
to take a different approach to software design.
Sometimes you just need a headless (no GUI) version of an existing
LabVIEW program. In that case, you can simply use LabVIEW RT to
recompile your existing code onto an RT-compatible target, and you are
done. LabVIEW RT is fully cross-platform-compatible with all other
versions of LabVIEW and makes building headless embedded systems
easy. On the other hand, if you are interested in building control sys-
tems that can meet time-critical demands, then read on. We’ll try to
show you the basic information that you’ll need to build a system to
meet your real-time demands.

RT Hardware
The LabVIEW RT development environment runs on a Windows host.
The host serves as the local storage for the RT applications during
development as well as provides the user interface. When the Run but-
ton is pressed, the LabVIEW VIs are downloaded to the embedded tar-
get, where they are compiled and run on their own dedicated processor

*There’s nothing wrong with using a desktop system in a soft real-time application;
there are perhaps millions of such systems in operation today. You only have to prove
that the system meets your requirements for determinism under all circumstances. Ded-
icating a computer to a control task, deleting unneeded applications, and keeping users
away leave it lightly loaded, thus reducing latency.
380 Chapter Fifteen

Figure 15.1 The first platform for LabVIEW RT was the 7030-
series RT DAQ card with an onboard 486 processor and analog
and digital I/O. Today the PCI 7041/6040E uses a 700-MHz PIII
with 32-Mbyte RAM and 32-Mbyte Flash. (Photo courtesy of
National Instruments.)

under a real-time OS. The user interface runs only on the host, not
on the embedded target. Normal LabVIEW user-interface interaction
takes place over a TCP/IP (Ethernet) communications link using Front
Panel Protocol (FPP) routines that National Instruments provides
in LabVIEW RT.
LabVIEW RT runs on the PCI 7041/6040E series RT DAQ boards,
on selected PXI crate controllers, Compact Fieldpoint controllers, Com-
pact Vision systems, and even off-the-shelf dedicated desktop PC hard-
ware (Figures 15.1 and 15.2). The 7041 intelligent DAQ board plugs

Figure 15.2Programmable automation controllers (PACs) combine PLC

ruggedness with PC-like functionality. From left to right: CompactRIO,
Compact Vision, PXI, Compact FieldPoint. (Photo courtesy of National
Instruments.)
 LabVIEW RT 381

into an existing computer’s PCI slot. Each intelligent RT DAQ card

consists of a main processor card and a daughter card for data acquisi-
tion tasks. The processor card has a PIII/700-MHz CPU with related
support circuitry and 32 Mbytes of RAM for user programs, and
32 Mbytes of permanent storage. The daughter card is National Instru-
ments’ standard 6040E 12-bit MIO card that has been adapted to fit
onto the processor card. It’s a handy way to upgrade an existing system
to enhance time-critical performance. It’s also a way to add parallel
processing power to a single PC, since you can run multiple processor
cards simultaneously.
LabVIEW RT on a rugged CompactRIO, Compact Vision, or Compact
FieldPoint provides an industrial-strength hardware platform with a
reliable OS and flexible programming environment. LabVIEW RT on
one of National Instruments’ PXI controllers offers the greatest speed
and flexibility. All the PXI DAQ modules that work with DAQmx and
most modular instruments will work in real time under LabVIEW RT
with the PXI controller. If you already have a PXI system, you should
be able to upgrade it to real-time performance without a major rework
of any hardware. You can also use a desktop PC as an RT controller
and use your existing PCI DAQ cards. Any Pentium 4–based computer
system with the same PCI and Ethernet chip set as the PXI board
should be capable of running LabVIEW RT applications. Check out
NI’s website (www.ni.com) for complete and up-to-date information.
The sole purpose of a real-time system is to perform time-critical or
real-time tasks. In LabVIEW RT the user interface has been stripped
out to allow all the processor cycles to be used for the real-time task.
All user interaction with the RT card and communication between the
host and the plug-in RT card are done through shared memory on the
7041 and over Ethernet on the modular platforms. We can use this
shared memory for high-speed communications when necessary, but
for most of our routine tasks we can just use LabVIEW’s built-in TCP/
IP functions and the high-level VI Server functions.
The only way that you can interact with the RT system on the PXI
crate is over Ethernet. LabVIEW RT has all of LabVIEW’s communi-
cation routines, including the VI Server, TCP/IP, and Web server. Just
make sure that your communication routines run in a separate loop at
a lower priority than your real-time tasks. You don’t want your user
interface to get in the way of your time-critical tasks. That’s why we
moved to a real-time system in the first place.
Since the RT card does not have an Ethernet port, the LabVIEW
team has implemented TCP/IP and VI Server functionality through
shared memory. This is just one of the many ways LabVIEW takes
care of the details for us so that we can concentrate on our applica-
tion and don’t worry about the implementation. Some of the VI Server
382 Chapter Fifteen

functions, such as Make Current Values Default, don’t make any sense
on a card without any local storage, and so they have been imple-
mented only on the PXI system. These minor details are covered in the
LabVIEW RT manuals.
One of the first things you will notice after you have booted up your
PXI crate into the real-time environment is that the graphical display is
gone. You are left with a black, DOS-like screen and a cursor that doesn’t
respond to keyboard input. In the real-time system, all user-interface
code is stripped out, and only the compiled block diagram code is left.
On the PXI system there is a printf function that will print a line to the
monitor for debugging during development, but other than that there is
no resident user-interface at all! All user-interface activity is handled by
the Windows-based development system over the Ethernet link.

Designing Software to Meet

Real-Time Requirements
Ugly code is ugly code! Don’t download some poorly designed spaghetti
code onto a LabVIEW RT system and expect to have it magically work
better. Chances are it will be worse. Programming an application that
will meet a hard real-time deadline requires you to write code that will
execute efficiently. Fortunately, programming in LabVIEW RT is not
any different from programming in regular LabVIEW, and we have
the entire suite of LabVIEW tools to work with. All you have to do is to
learn some new guidelines, many of which will probably improve the
performance of your non-RT applications as well.
A good model when you are designing software for deterministic
behavior is to look at the techniques used in hardware design. In a
very real sense, what we are trying to do with our software is to emu-
late a hardware task. Many real-time tasks started out as hardware
tasks but needed the flexibility that a software design can bring. The
art is to implement the design in software without paying too much of
a performance penalty.
When you need to design a circuit, you need to know the inputs and
outputs just as in software. You can predict how each device you put
into your circuit will behave because you know the physical character-
istics of each device and you can verify the performance with hardware
probes such as oscilloscopes and logic analyzers.
This same modularization and performance verification is exactly
how you must go about designing your software for a real-time system.
Since time is a critical element of a real-time system specification, you
need to know exactly how much time each subVI requires to execute.
When you design a real-time system, one of the obvious questions
you need to ask yourself is, What is our real-time task? Suppose we
have a process controller that needs to run at 1 kHz and we also have
 LabVIEW RT 383

a safety loop that needs to reset a watchdog timer once per second to
keep the process from being shut down. The safety loop is the real-
time task, and the 1-kHz process will have to be preempted while the
safety loop executes. To build a successful real-time system, you need
to balance a limited number of real-time tasks with very close atten-
tion paid to how those tasks might interact or clash with one another.
The remainder of your code should be tolerant of preemption; that is, it
should tolerate much greater latency.
The next thing you need to know is the timing requirements that
you have to meet. LabVIEW’s built-in timing function uses the PC’s
real-time clock, which is based on milliseconds. To achieve timing loops
faster than 1 kHz, you’ll need to use hardware-timed acquisition, where
a DAQ board provides the clock. This is not a problem, but you need to
know in advance what your restrictions are so you can design for suc-
cess. Be sure to leave enough spare time in your real-time task so that
background tasks can execute. For example, a 10-kHz PID loop probably
isn’t going to do you much good if you can’t send it a new set point.
Once you have your process timing requirements, you need to ana-
lyze your code or VIs and to determine how long each VI or collection of
VIs takes to execute. After all your code is optimized and verified to fit
within your timing requirements, you construct your program so that
each time-critical piece of code can execute on time, every time.

Measuring performance
LabVIEW’s built-in VI profiler provides a good first indicator of VI
execution time. But to help you get down to the real nitty-gritty of
how many processor cycles a VI will take, National Instruments has
also given us some tools to read the Pentium’s built-in timestamp func-
tion. This allows you to get nanosecond-scale readings on execution
time. The RDTSC Timing library includes a low-level VI (RDTSC.vi)
that calls a CIN to read the CPU clock, and a utility to convert the result-
ing tick count to seconds (RDTSC Diff To Time.vi). There are also some
examples that measure and display timing jitter. All you have to do is to
insert your code to be tested. Figure 15.3 shows a very simple example.

0 [0..2] Processor
1 [0..2] 2 [0..2]
Speed (MHz)
Initial timestamp Final timestamp
733.00
Time (uS) Code to be
tested
goes here rdtsc.vi
rdtsc.vi
RDTSC Diff To
Time.vi

Figure 15.3 The RDTSC library includes a VI to read the Pentium timestamp
and convert it to microseconds. Use it for direct measurement of jitter and
other small delays in your code.
384 Chapter Fifteen

Have you ever wondered what the real effect of that coercion dot was?
This will tell you. You can find the RDTSC VIs on the Web at www
.ni.com.
An oscilloscope is an essential tool on the hardware test bench, and
it should be one of the first tools you turn to when analyzing real-time
embedded performance as well. You can bring out many of the internal
timing signals of your data acquisition card to an external terminal
block like the SCB-68. A logic analyzer is another handy tool when
most of the signals you’re concerned about are digital. Looking at your
timing signals is a great way to monitor the performance of your soft-
ware and hardware.
We once had a data acquisition problem where we had to sample
a photodetector immediately before and after a laser pulse (see the
timing diagram in Figure 15.4). The laser fired at 30 Hz, and the first
sample needed to be taken 250 µs before the laser pulse and the second
sample 500 µs after the laser pulse. To provide this uneven timing, we
programmed our DAQ card to sample the same channel twice each
time an external scan clock pulsed. Then we set the interchannel delay
on the DAQ card to 750 µs. This meant that whenever the DAQ board
received a trigger it would take the first sample and then wait 750 µs
before taking the second sample. By monitoring our board’s timing sig-
nals we were able to verify exactly when each scan was beginning and
when each sample was being taken.
Monitoring our DAQ card’s timing signals also allowed us to see a
“gotcha” of NI DAQ: When you use an external scan clock, the first
sample is not taken until one interchannel delay after the input trig-
ger. This meant that we were totally missing the event we were trying
to see. But once we saw where our timing problem was, we were able to
adjust our pretrigger input to the external scan clock so that the tim-
ing was just as specified. After the triggering was set correctly and
the board was configured to store the data into a circular buffer, our

750 µs 250 µs 500 µs

D Setup Time B

A
C
Figure 15.4 Monitor AI CONVERT* to see when each A/D conversion is taking
place. We needed to sample a photodiode (A) 250 µs before a laser pulse (B) and
500 µs after the pulse (C). The DAQ card was configured to use an external scan
clock trigger (D) with an interchannel delay of 750 µs. An extra delay equal to the
interchannel delay (750 µs) was required by NI DAQ to set up the acquisition.
 LabVIEW RT 385

DAQ processing application only had to monitor the buffer for new
data and act on it as it came in. You should use hardware triggering
any time you can, even in LabVIEW RT. It will give you the most stable
and repeatable system. It’s also fully verifiable in that you can directly
observe the signals externally.
To view your board’s internal timing signals, you need to use the Route
Signal VI (available in the DAQ Miscellaneous function palette) to bring
them out to a PFI pin on the connector where you can connect to them.
You can wire this VI into your DAQ configuration routines and leave it
there during development. It doesn’t add any overhead to the data acqui-
sition tasks, and you will find yourself relying on these signals whenever
you need to verify timing. Figure 15.5 shows an example where the AI
Read One Scan VI has been modified to include signal routing. Some of
the key pin-outs to monitor on the E series cards are as follows:

AI STARTSCAN generates a 50-ns-wide, positive-going pulse at the

beginning of each analog input scan. Use the Route Signal VI to con-
nect the AI STARTSCAN signal to PFI 7.
AI CONVERT* goes low for 50 ns at each analog/digital conversion.
Use the Route Signal VI to connect AI CONVERT* to PFI 2. This
digital signal is useful to see when each analog/digital conversion is
taking place in a scan.
AO UPDATE* goes low for 50 ns when analog output circuitry is
being updated. Route it to PFI 5.

Look at the documentation that comes with your DAQ card to verify
your particular pin-outs.

iteration (init:0)
..0
signal name
coupling &
input config PFI 7
scaled data
input limits
Route Signal.vi
task id
device (1)

channels (0)
binary data
buffer size 0:
0
don't allocate AI scan start
[number of signal source error out
error in AMUX boards]

Figure 15.5 AI Read One Scan modified to bring out the data acquisition card’s internal
timing signals. The configuration case has been altered to include the Route Signal VI.
The signal being routed is AI STARTSCAN to PFI 7. Each time AI Read One Scan
executes, PFI 7 will produce a pulse.
386 Chapter Fifteen

Almost any decent bench oscilloscope will let you see these brief tran-
sitions. If you’re lucky and have a digital oscilloscope with a variable-
persistence display, you can probably get a decent measurement of the
software latencies, which will show up as jitter. But we’ll show you an
even better way to get an accurate timestamp of each and every transi-
tion by using something you’re just as likely to have around: an extra
DAQ card with a counter (Figure 15.6). There is an example VI called
Count Buffered Edges. This VI uses one of the DAQ-STC counters
on an E-series DAQ card to measure the period of a digital signal at the
gate of one of your counters.
The counter measures the period of your signal by counting a signal
connected to the counter’s source and storing that count in a buffer each
time it receives a signal at its gate. On the example VI, Count Buffered
Edges, the counter’s source is connected to the internal 20-MHz clock.
This means the count increments every 50 ns. One of the features of
buffered period measurement is that with each new period the counter
is rezeroed, and you don’t have to worry about the counter rolling over.
You can use the Count Buffered Edges VI to measure the perfor-
mance of your real-time data acquisition tasks. Use the Route Signal
VI when you configure your analog input or output to send the appro-
priate timing signal out to the connector. Then connect the output of
that PFI pin to the gate of a counter on your extra DAQ board.
Now you can use this simple tool to “peek under the hood” of your real-
time system and analyze your data acquisition tasks. Count Buffered

SOURCE0
(20 MHz Clock)
5 6 1 2 3 4 5 1 2

GATE0
(External signal)
Count during
LOW interval
Read count,
stored in buffer n=6 n=5

Counter0
Clock pulses per interval

SOURCE0
Clock
OUT0
GATE0

DIG. GND.
Figure 15.6 A DAQ counter counts the 20-MHz internal clock, which
is routed to the SOURCE input. Each time the counter’s GATE is
triggered, the total count is stored in a buffer. The counter is then
rezeroed and counting continues.
 LabVIEW RT 387

Edges lets you see just how deterministic your loops are. The histogram
plot keeps a running tally of latencies that you can use for long-term
profiling to verify that your code is executing on time, every time.
Sometimes your measurements may seem very erratic, and this can
be caused by noise on the digital lines. If you have a fast oscilloscope,
you might be able to see noise or glitches that are triggering the coun-
ter. One way to cure the problem is by conditioning your signal with a
one-shot, which is a pulse generator with a programmable pulse width.
Set the pulse width to a value that’s a bit less than the expected inter-
val you’re measuring, and the extra pulses in between will be ignored.
There is a simple VI in vi.lib (vi.lib\DAQ\1EASYIO.llb\Generate
Delayed Pulse.vi) that will let you use one of your counters as a delay
generator. Set the delayed pulse width to be several tens of microsec-
onds, connect the signal you want to buffer to the counter’s gate, and
monitor your cleaned-up signal at the counter’s source.
If, for example, you want to look at AO UPDATE* to see when the
analog output circuitry on your real-time card is being updated, you
connect PFI 5 to counter 0’s GATE. The slightly delayed buffered pulse
will be at counter 0’s OUT. Buffering your signals in this way will make
them easier to see, and it will also eliminate some of the noise that
might be giving you false readings. No extra software overhead is added
to your program other than the time required to initialize the counters.
Figure 15.7 shows the results obtained when a digital line was toggled

Counter Configuration
Latency
device counter 50
1 0 40 Last Buffer

mean
task type
20
buffered semiperiod measurement 100068.78
standard deviation
Buffer Size counts to read 0
2.55
10000 100
-20
Gate Specification

-40
Gate Selection
-50
PFI Line default 0 10 20 30 40 50 60 70 80 90 100
STOP
RTSI Line RTSI 0 Histogram
500000
Misc low 100000

Gate Polarity
10000
positive (low to high) # of data points outside
1000 histogram bounds

above 0
Histogram Configuration 100
below 0
Center Period (usec) + - Range # bins
10
100068.35 150.00 100
1
-100.0 -50.0 0.0 50.0 100.0
Microseconds

Figure 15.7 Screen capture from the Count Buffered Edges VI after toggling a digital line every
100 ms for 20 h. The histogram of latencies shows that there were no latencies greater than 50 µs.
388 Chapter Fifteen

every 100 ms by writing to a digital port in a LabVIEW RT For Loop.

The peak jitter was only 50 microseconds after 20 h of operation.
Once you’ve streamlined each section of your code and verified that
it can meet its deadlines, it’s time to start putting everything together
to make a complete application. This is when we need to know how
the LabVIEW execution system works and how your operating system
affects execution.

Shared resources
If unexplained timing glitches occur when you run your code, then you
may have the problem of two sections of code competing for a shared
resource. The quickest way to lose deterministic execution of your code
is to have multiple tasks in competition.
A shared resource is a software object that cannot be accessed
simultaneously by multiple threads. Examples of shared software
resources are single-threaded device drivers such as NI DAQ, global
variables, shared subVIs that are not reentrant, synchronization code
(semaphores, queues, occurrences, rendezvous), networking protocols,
file I/O, and LabVIEW’s memory manager.
To protect a shared resource such as NI DAQ calls, LabVIEW places
a mutex around it. A mutex, or mutual exclusion, is a software mecha-
nism to prevent two threads from accessing a shared resource at the
same time. You can think of a mutex as a lock on a door that has only
one key. When your program uses NI DAQ to access a data acquisi-
tion board, it must first check to see if NI DAQ is in use. If NI DAQ is
available, your program gets the “key,” unlocks the door, and enters the
protected area, and the door is locked behind it. No other program can
use the resource until the door is unlocked and the key is returned.
For example, you can write a program in regular LabVIEW that
uses two data acquisition loops running in parallel; maybe one is ana-
log input and the other is digital I/O. Because you’re not concerned
about or monitoring the real-time behavior of these DAQ loops, every-
thing appears to be fine—only the average time of the loops is impor-
tant. But if you do this in a real-time execution system where you
are paying close attention to deterministic timing, you will see how
each DAQ task is interfering with the other as they compete for the
shared resource.
If two tasks competing for a shared resource are at different pri-
orities, you can get a condition known as priority inversion, where
a high-priority task has to wait for a low-priority task to finish. If a
low-priority VI has the key to the shared resource, that VI cannot be
preempted by a higher-priority VI. The high-priority task will have
to wait for the low-priority task to finish. One method used by the
 LabVIEW RT 389

RTOS to resolve a priority inversion, known as priority boosting, is to

temporarily raise the priority of the task using the shared resource to
time critical until the shared resource is released. Then the low-priority
task is reduced back to its original priority and the higher-priority task
is free to run.
A shared resource that you don’t usually think about is LabVIEW’s
memory manager. LabVIEW manages memory dynamically behind
the scenes, and it is so transparent that you usually never even have
to think about it. But each time LabVIEW manipulates a region of
memory, it uses a mutex to protect the region so that other parts of
your application don’t interfere halfway through the process and mess
things up. For example, this means that a low-priority subVI that is
building an array in a different execution system from our time-critical
loop can have a dramatic effect on deterministic behavior. A solution to
this particular problem is to preallocate all your arrays at the begin-
ning of the program and avoid resizing them. While you can’t totally
avoid memory management, you should at least follow some of the gen-
eral guidelines in Application Note 168, LabVIEW Performance and
Memory Management.

Multithreading and multitasking

Ever since version 1.0, LabVIEW has given us the ability to execute
pieces of our program in parallel. LabVIEW’s dataflow paradigm makes
it easy for us: All we have to do is to wire up two parallel While Loops
and presto! It works! LabVIEW performs this magic by multitasking,
whereby each section of code is given a small slice of time to run on the
processor before it is pulled off and another section is allowed to run.
The LabVIEW execution engine controls the multitasking in a coop-
erative manner. To a great degree, the multitasking that takes place
is transparent to us; our code apparently executes in parallel. In the
latest versions of LabVIEW, we still have the cooperative multitasking
of the LabVIEW execution engine, but it has been enhanced by the
addition of multithreading.
Beginning with LabVIEW 5.0, when you write a LabVIEW applica-
tion on a multithreaded OS and push the Run button, your compiled
program is split up into multiple copies of the LabVIEW execution sys-
tem. Now, in addition to LabVIEW executing portions of your diagram
in parallel, you have the OS running multiple copies of the LabVIEW
execution system in parallel. These multiple systems running in parallel
are a powerful programming concept that is made easy by LabVIEW.
There are at least two references you should consult with regard
to multithreading and execution of VIs in LabVIEW. Application
Note 114, Using LabVIEW to Create Multithreaded VIs for Maximum
390 Chapter Fifteen

Performance and Reliability, is available from the LabVIEW help sys-

tem (under printed manuals). If you have a copy of LabVIEW 5, the G
Programming Reference Manual has a helpful chapter entitled “Under-
standing the G Execution System.”
The default configuration of LabVIEW has one copy of the execution
system for each of four priorities: time critical is the highest, then high,
above normal, and normal. The default also calls for one copy for each
of five categories: standard, instrument I/O, DAQ, other1, and other2.
(The category names are arbitrary. That is, the DAQ category can be
assigned to any task, not just DAQ operations.) An additional prior-
ity, background, is not enabled by default, but you can enable it and
change the number of threads in each execution system by using the
Threadconfig VI located in /vi.lib/utilities/sysinfo.llb (Figure 15.8).
Each of the 20 default copies of the LabVIEW execution system runs
a set of VIs that you assign to it as its thread runs on the processor.
The LabVIEW threads and priorities that you assign to your VIs are
mapped directly to priorities within the OS. The OS assumes complete
responsibility for which thread gets the processor.

Figure 15.8 You can set how many threads are assigned to each LabVIEW execu-
tion system with the Thread Configuration VI. Find it in labview/vi.lib/utility/
sysinfo.llb.
 LabVIEW RT 391

A thread can be in one of three basic states:

Running. The processor is running the instructions that make up

this thread.
Ready. Some other thread is running, but this thread could run if it
were given access to the processor.
Blocked. The thread is waiting for an external event such as an
interrupt or a timer. Even if the processor were available, this thread
could not run.

The RTOS used by LabVIEW RT uses a combination of round-

robin scheduling and preemptive scheduling. In round-robin sched-
uling, each thread that is ready to run gets a slice of time to execute.
In preemptive scheduling, the operating system takes control of the
processor at any instant. That’s a way for higher-priority tasks to get
preferential access to the processor, thus improving the response time
of specified tasks. The RTOS will give the processor to the ready thread
with the highest priority, but if there are multiple threads with the
same priority, the RTOS will give each thread a small slice of time.
Each thread will run until it blocks (waiting for a timer or interrupt), a
higher-priority thread unblocks, or the RTOS pulls it off to run another
thread at the same priority. Any work that the thread did not complete
will have to wait until the next time it can run.
The one exception to the time-slicing feature of each execution sys-
tem occurs when a VI is set to run at subroutine priority. A subrou-
tine VI will always run to completion within its execution system. A
thread running a subroutine VI can still be preempted by the RTOS,
but no other VI running in the same execution system as a subroutine
VI will be able to run until the subroutine VI is finished.

Organizing VIs for best

real-time performance
To make best use of the parallel execution systems, consider which VIs
are logically grouped together and keep them within the same execu-
tion system. If you have a communications loop that you use to pass
data between the RT system and the host computer, then place that
loop and all related VIs into one thread or execution system. Set the
thread to a low priority, such as background or normal, and let the
RTOS take responsibility for running this thread when higher-priority
threads are sleeping. By letting the RTOS take care of the hard work
for us, we can maximize the efficiency of our application.
Place your real-time task in a time-critical thread, and it will run
before any other thread. When the time-critical priority thread is ready
392 Chapter Fifteen

to run, then the scheduler will preempt or remove whatever thread is

running from the processor and the time-critical priority thread will
run until it blocks. Be sure that you never have more than one time-
critical thread, or else the RTOS will implement round-robin scheduling
between the time-critical threads and you will lose your deterministic
performance.
There are several things that you should watch out for when assign-
ing priorities to your VIs in LabVIEW RT that are different from
normal LabVIEW. A VI set to time-critical priority must have a sleep
function somewhere inside its main loop. The only time low-priority
threads get a chance to run is when the time-critical thread blocks or
sleeps. A time-critical thread that never sleeps will starve out all other
threads.
There are at least two ways to create a sleep function. First, you can use
the Wait For Next ms Multiple or the Wait ms function inside its main
loop. Either of these timers can wake up the loop once every integral
number of milliseconds. While the loop is sleeping, the RTOS will sched-
ule lower-priority threads until it is time to wake up the loop again.
Second, you can use the DAQ function AI Single Scan. You can con-
figure a hardware-timed acquisition external to your time-critical
loop and call AI Single Scan to read each scan as it becomes available
(Figure 15.9). This is counterintuitive because normally calling a DAQ
function in this way causes the loop to use all the processor time until
the function returns. NI DAQ has been modified for LabVIEW RT so
that this one VI provides its own sleep function.
When a loop in a time-critical priority VI is put to sleep, the entire
thread is put to sleep. No other loops or VIs in that thread will run
until the thread wakes up. This means that you cannot have two par-
allel loops running at different rates inside the time-critical thread. If
you need to have multiple tasks operate in the time-critical loop, then
you need to implement a scheduler to control execution.
In addition to the various execution systems that manage the
sub-VIs, another execution system is used for the user interface in

continuous
device Acquire and wait for next scan
0

scaled data
channels

scan rate

Figure 15.9 AI Single Scan provides its own sleep function. The VI
shown here would typically be set to time-critical priority in LabVIEW
RT. This is not standard practice in regular LabVIEW.
 LabVIEW RT 393

regular LabVIEW. But in LabVIEW RT this user-interface thread

becomes a TCP/IP communication thread between the RT system and
the real user interface running on the host development system. That’s
how we remove the burden of a user interface from the RT processor.
Having a large number of execution systems allows us to make more
efficient use of the processor but at the expense of speed and determin-
ism. The more execution systems you use, the more likely it is that you
will need to share data among two or more execution systems. Lab-
VIEW takes care of the details of this for you, but there is still some
overhead involved in passing data back and forth.
There are several options for sharing data between a real-time loop
and a lower-priority loop. If you don’t need to collect every data point,
then you can use a LabVIEW 2–style shift register global set to sub-
routine priority with skip if busy enabled. When you drop a VI set to
subroutine priority on the block diagram, an extra menu item shows
up when you command click on the subVI: “Skip Subroutine Call If
Busy.” The real-time loop will skip the VI if the VI is in use by a lower-
priority thread. You may miss a few data points, but you can build
whatever functionality you need into your global. To avoid missing
data points, you can build a queue or buffer to transfer data to the
lower-priority loop.
A good application for all that we’ve discussed so far is point-by-point
control loop I/O. In these situations, you need to make a measurement,
do a computation (such as PID), and then generate an output with
minimal delay or timing jitter. You would build such a VI in the manner
shown in Figure 15.9 and set the VI’s priority to time critical. It should
run as a separate top-level VI, and it should be the only time-critical VI
in the system. You can exchange data such as the setpoint, measure-
ment, and status information via global variables or queues. We’ll see
such an example later.

Context switching adds overhead

It’s important to choose carefully which thread your VIs are assigned
to. The actual names of the execution systems are not important; that
is, DAQ VIs do not need to run in the execution system named DAQ.
What is important is that you avoid placing subVIs that are called
repeatedly in separate execution systems from their caller. You assign
a VI to run in a specific thread using the Execution options from the VI
Properties dialog (Figure 15.10).
A context switch occurs each time the OS pulls one thread off the
CPU and puts another thread on. When the OS swaps threads, all the
stack information and all the instructions in the CPU cache need to be
swapped out to main memory. Then all the new thread’s information
394 Chapter Fifteen

Figure 15.10 You can assign a VI to go in a specific execution system by using

the Execution options from the VI Properties dialog. The default setup for
each VI is normal priority in the same execution system as the caller. It’s the
setting you almost always want for subVIs.

is loaded into the CPU so that the thread is in the same state it was
in before it was removed. This can take a significant amount of time.
Any time you force the operating system to perform a context switch by
calling a subVI in another execution system, you cause delays that will
affect how efficiently your code executes. If the subVI needs to be in a
separate execution system, then consider making it into a top-level VI,
just as you would for a time-critical VI.
Consider what happens when you have a top-level VI set to run in
the standard execution system and a subVI set to run in the DAQ exe-
cution system. Each time that subVI is called, the thread containing
the top-level VI will be removed from the processor and the thread con-
taining the subVI will run. When the subVI is finished, another context
switch occurs as the OS switches back to the top-level VI. If the subVI
is in a loop that runs very frequently, the extra overhead will be easy
to observe. This shows why you should not arbitrarily assign execution
systems to subVIs.
When a subVI is set to execute in the same execution system as the
caller, it will inherit the caller’s priority and run in whatever execu-
tion system it was called from. That is, if you have a top-level VI set
to time-critical priority in the standard execution system and it calls
a subVI set to normal priority and the same execution system as the
caller, there will not be a context switch from time-critical to normal
priority. The subVI will have its priority elevated to the same priority
as the calling VI.
 LabVIEW RT 395

Scheduling
Timed structures simplify the way you schedule real-time execution
order. Efficient scheduling and the mechanisms used to enforce it used
to be a complex art form driven by which VI has the highest priority and
the earliest deadline. Now you just place down a Timed Loop, give it a
priority and a period, and off you go. If you need a parallel task at a dif-
ferent loop rate and a higher priority, that’s no problem! Once again Lab-
VIEW makes something incredibly easy that was terribly complicated.

Timed structures
There are three structures with built-in scheduling behavior: Timed
Loops, Timed Loop with Frames, and Timed Sequences. All three
provide a way to schedule real-time program execution. Each Timed
structure runs in its own thread at a priority level just beneath time
critical and above high priority. Timed Loops do not inherit the prior-
ity or execution system of the VI in which they run. This means that a
Timed structure will stop any other code from running until the struc-
ture completes. LabVIEW’s Timed Structure Scheduler runs behind the
scenes and controls execution based on each Timed structure’s schedule
and individual priority. Each Timed Loop’s priority is a positive inte-
ger between 1 and 2,147,480,000. Priorities are set before the Timed
Loop executes, through either the configuration dialog or a terminal
on the Input node. Figure 15.11 shows the configuration dialog for

Configuration dialog for a Timed Loop. Timed Loops on RT systems

Figure 15.11
have a 1-MHz timing source available.
396 Chapter Fifteen

a Timed Loop. Priority can be adjusted inside the Timed Loop through
the right data node, so you can dynamically adjust performance if
needed. Timed structures at the same priority do not multitask between
each other. The scheduler is preemptive, but not multitasking. If two
structures have the same priority, then dataflow determines which one
executes first. Whichever structure starts execution first will finish
before the other structure can start. Each Timed Loop runs to comple-
tion unless it is preempted by a higher-priority Timed Loop, a VI run-
ning at time-critical priority, or the operating system.
Figure 15.12 illustrates a round robin schedule implemented with
3 timed loops. All 3 loops use the 1 kHz system clock with a period
(dt) of 30 ms. Loop B’s start (t0) is offset by 10 ms from Loop A and
Loop C is offset from Loop A by 20 ms. The timeline in Figure 15.13
shows each loop’s time slice provided each task finishes in its allotted
10 ms. If one of the tasks took longer than 10 ms to complete the Timed
Structure Scheduler would adjust execution based on priority, and the
other loops’ configuration options for late iterations. Any loop with a
higher priority will preempt a lower priority loop. Once the higher pri-
ority loop has finished the preempted loop will get a chance to finish.

Figure 15.12 Three Timed Loops with the same

period, but offset in phase (t0). They will run one
after the other as shown in Figure 15.13.
 LabVIEW RT 397

Loop A
Loop B
Loop C
Time
Figure 15.13 Time line for the three Timed Loops in Figure 15.12.
All loops have the same period, but are offset in phase.

Late starts by timed loops are governed by the selected actions on late
iterations as shown in Figure 15.11. The default selections will cause a
timed loop with a late start to forgo execution until its next scheduled
timeslot. Deselecting “Discard missed periods” causes the timed loop to
try to make up any missed iterations. This can have unexpected conse-
quences, so use with caution and always verify that your application is
doing what you expect.
Timing sources for Timed structures can be based on your computer’s
internal timer or linked through DAQmx (Figure 15.14) to a hardware-
timed DAQ event with DAQmx Create Timing Source.vi. You can
even chain multiple timing sources together with Build Timing

Figure 15.14 Hardware-timed acquisition and control. Timed Loops can be synchronized to DAQmx events.
398 Chapter Fifteen

Figure 15.15 Shared variable is used to communicate between the

RT process and the host PC (Figure 15.16).

Source Hierarchy.vi. This could allow you to execute based on time

or, with the appropriate hardware, on a change on any configured digi-
tal line.

Communications
LabVIEW has several technologies to make exchanging data transpar-
ent between distributed real-time targets and a host PC. The easiest to
use is the Network-Published Shared Variable. Figures 15.15 and
15.16 show all the code you need to communicate between a real-time
process and a host PC. Shared variables communicate using National
Instruments’ Publish-Subscribe Protocol (NI-PSP) with a Shared
Variable Engine (SVE) that hosts the shared variables on the net-
work. When you write to a shared variable, LabVIEW sends the data
to the SVE which then publishes (sends) the data to all subscribers.
You can directly link front panel control and indicators to shared vari-
ables (Figure 15.17). This is really easy, but totally violates dataflow.
Shared variables may not be right for every application, especially
if you need to communicate with programming languages outside of
LabVIEW. For those applications consider the Real-Time Communica-
tion Wizard (Tools >> Real-Time Module >> Communication Wizard).

Figure 15.16 Host PC reads data from the

shared variable.
 LabVIEW RT 399

Figure 15.17 Data binding allows you to bind front panel objects to
shared variables without programming.

The Communication Wizard converts an existing real-time VI into a

client-server architecture. Controls and indicators in the time-criti-
cal code are replaced with real-time FIFOs (First-In, First-Out)
that transfer data to a parallel lower-priority communication loop. An
additional communication VI is created that your host uses to talk to the
time-critical process. Programmatically it’s considerably more involved
than the shared variable approach, but it’s there if you need it.

Bibliography
Application Note 200, Using the Timed Loop to Write Multirate Applications in
LabVIEW, www.ni.com, National Instruments Corporation, 11500 N. Mopac Express-
way, Austin, Tex., 2004.
Hays, Joe: “Advanced Real-time Programming Techniques,” NIWeek 2000, Advanced
Track Session 2D, available from www.natinst.com/niweek.
Li, Yau-Tsun Steven, and Sharad Malik: Performance Analysis of Real-Time Embedded
Software, Kluwer Academic Publishers, Boston, 1998.
Simon, David E.: An Embedded Software Primer, Addison-Wesley Longman, Reading,
Pa., 1999.
This page intentionally left blank
 Chapter

LabVIEW FPGA
16
LabVIEW FPGA takes real-time programming to the hardware level.
Although LabVIEW RT is a deterministic real-time environment, it is
still software running on a microprocessor, and any microprocessor has
only a limited number of clock cycles available to share among all the
running processes. LabVIEW RT’s deterministic behavior means that
you can accurately predict when your code will run; but as your appli-
cation grows and you start running more processes, you have to begin
making tradeoffs that affect determinism.
LabVIEW FPGA applications are not constrained by processor or oper-
ating system overhead. Code is written in parallel, is mapped directly
into parallel hardware registers, and runs in parallel. With LabVIEW
FPGA you can write massively parallel hardware-timed digital control
applications with tight closed-loop rates in the tens of megahertz.

What Is an FPGA?
FPGA is the acronym for field-programmable gate array. The tech-
nology has been around for a while, but only recently have the costs
come down and the tools improved enough that FPGAs have become
commonplace. An FPGA is essentially a 2D array of logic gates with
software programmable interconnections. FPGAs range in size from
devices with tens of thousands of gates to devices with millions of gates
and multiple embedded microprocessors. The number of gates is really
just a marketing number. Each FPGA vendor combines those gates to
create its own unique architecture. The basic programming block of
the Xilinx parts used by National Instruments is defined as “slices.”
Each slice contains two storage elements, arithmetic logic gates, large

401
402 Chapter Sixteen

Figure 16.1 PXI 7831R is National Instruments’ first reconfigu-

rable intelligent DAQ card. User code is downloaded into the 1
million gate LabVIEW programmable FPGA. (Photo courtesy of
National Instruments.)

multiplexers, a fast carry look-ahead chain, a horizontal cascade chain,

and two function generators. The function generators are configurable
as 4-input lookup tables, 16-bit shift registers, or 16-bit memory. As you
develop and compile your applications, look at the compilation sum-
mary and keep an eye on how many slices your application uses. The
1 million gate XCV1000 FPGA used on the 7831R (Figure 16.1) con-
tains 5120 slices. Try out different approaches to a problem and see
how each approach affects the number of slices used. Programming
with LabVIEW FPGA doesn’t require knowledge of a hardware defini-
tion language such as VHSIC hardware description language (VHDL)
or the underlying hardware; but understanding what’s underneath the
hood and following a few basic guidelines we cover in this chapter will
help you write cleaner, faster code.
An FPGA is a blank slate you can use to create almost any logic or
control circuit you want, and once the programmable interconnections
between the gates are made, your software becomes hardware. They can
be configured to precisely emulate discrete logic systems, or even replace
obsolete components. You can find FPGAs serving as the timing and
triggering logic on many plug-in cards, including National Instruments’
DAQ cards. Unfortunately the FPGA on the standard E-series and
M-series DAQ cards is not large enough to allow user-programmable
 LabVIEW FPGA 403

code, but hopefully in the future all NI’s DAQ cards will have user-
programmable logic. A presentation at NIWeek04 pointed out the major
shift in design philosophy between the LabVIEW FPGA programmable
DAQ cards and traditional DAQ cards and drivers (NI DAQ or DAQmx).
System designers of traditional DAQ cards design a flexible hard-
ware personality into the card and engineer a full-featured driver to
expose the hardware as the designer intended. It’s that “as the designer
intended” part that causes trouble. The designer couldn’t anticipate
your application or its exotic analog and digital triggering conditions.
LabVIEW FPGA and the RIO cards give you, the end-user/programmer,
a wide-open piece of hardware and a lightweight driver to handle com-
munications between the card and the PC. Whatever you want to do on
the card is in your hands. This is a revolutionary change.

LabVIEW for FPGAs

It’s easy to visualize and draw out parallel operations in LabVIEW, and
this maps really well into logic on an FPGA. LabVIEW FPGA gives you
the ability to quickly prototype right down at the hardware level and
to easily build low-level hardware interfaces with LabVIEW. This is a
huge advantage for digital design and development teams. Richard’s
group at Sandia builds custom hardware for telemetry. One of the big-
gest challenges has been to find a way to interface custom digital hard-
ware to a desktop PC for test and development. Now adding intuitive,
graphical interfaces to low-level communication protocols such as SPI,
I2C, and custom serial and parallel busses is easy.

RIO hardware platforms

National Instruments’ RIO (Reconfigurable I/O) LabVIEW program-
mable hardware (Figure 16.2) wraps a high-level interface, and a lot
of I/O goodies, around a Xilinx FPGA. Programming the FPGA is easy.
You have almost infinite control over the asynchronous I/O. By asyn-
chronous I/O we mean the input/output lines can be used indepen-
dently of each other; however, you can synchronize them if you want.
You can use the digital-to-analog converters (DACs) for arbitrary pulse
and waveform generation. Complex triggering with digital or analog
triggers is no problem. The DIO lines can be used for counters, timers,
and frequency dividers; just about anything you want to do with stan-
dard logic is possible.

Plug-in cards
National Instruments currently has seven R-series intelligent DAQ
cards varying in form factor, size of FPGA, and number and type of
404 Chapter Sixteen

Figure 16.2 LabVIEW FPGA reconfigurable hardware includes

intelligent DAQ cards, timing and synchronization modules,
Compact Visions systems, and the CompactRIO Programmable
Automation Controller. (Photo courtesy of National Instruments.)

I/O lines. All the cards are designed to let you create your own process-
ing algorithms that can run at rates up to 40 MHz. You can configure
digital lines as inputs, outputs, counter/timers, pulse-width modula-
tor (PWM) encoder inputs, or custom communication busses. On cards
with analog capabilities you can have up to eight simultaneous 16-bit
analog inputs at rates up to 200 kHz, and up to eight simultaneous
16-bit analog outputs at rates up to 1 MHz. The more we use LabVIEW
FPGA, the more we’re convinced that all DAQ cards should be user-
programmable. All the trigger lines and PXI local bus lines are exposed
on the RIO hardware, and you may use any triggering scheme, analog
or digital, you desire. Because the input, processing, and control out-
put are all embedded into the FPGA fabric, you will find that the RIO
plug-in cards have exceptional performance for single-point I/O control
applications.

CompactRIO
CompactRIO or cRIO combines a LabVIEW RT controller for advanced
floating-point analysis and control with an intelligent FPGA backplane
for hard real-time timing, synchronization, and control. A cRIO system
 LabVIEW FPGA 405

Figure 16.3 CompactRIO is a compact 3.5-in × 3.5-in × 7-in indus-

trial controller with a LabVIEW RT processor and LabVIEW
FPGA programmable backplane. (Photo courtesy of National
Instruments.)

consists of an embedded real-time controller, a four- or eight-slot chas-

sis with a LabVIEW FPGA programmable backplane, and a mix of
industrial I/O modules (Figure 16.3).

Timing and synchronization

Advanced timing and synchronization applications can use National
Instruments’ PXI 6652 and 6653 control modules to control timing
across multiple PXI chassis for almost unlimited customization. In
addition to synchronizing multiple chassis, each controller can drive
the STAR trigger line and the TTL triggers on the PXI backplane.
Typical uses for these LabVIEW FPGA programmable cards include
advanced timing, triggering, and user-defined delay applications.

Compact Vision
National Instruments’ Compact Vision system is a rugged industrial vision
system that uses a LabVIEW programmable FPGA for complex timing
and synchronization. Each system has three FireWire camera ports, a
VGA port, an Ethernet port, 15 digital inputs, and 14 digital outputs.
You can program the digital lines with LabVIEW FPGA for quadrature
encoders, generating strobe pulses for the vision system, synchronizing to
external stimuli, or controlling relays and other actuators.
406 Chapter Sixteen

Application Development
There is no single definitive programming method for LabVIEW
FPGA, but there are some significant differences between LabVIEW
on a desktop and LabVIEW code running on an FPGA. Here is a short
list, which we’ll expand on throughout the chapter:

■ Local variables. Some things that are bad on the desktop, such as
local variables, are good on FPGA. A local variable in desktop Lab-
VIEW is inefficient and results in extra copies of the data everywhere
it is used. On FPGA a local variable is a hardware register that can
be efficiently accessed from multiple locations. Race conditions can
still occur if you are not careful, so be sure to have one writer and
many readers.
■ SubVIs. SubVIs are efficient chunks of reusable code on the desktop,
but they can have unexpected consequences on the FPGA. Unless a
subVI is marked reentrant, there will be only one copy of the subVI
mapped into hardware. A single copy of a VI in hardware may be
what you want if you need to Read–Modify–Write a shared piece
of data in multiple loops, but it will also mean that the loops have
a hardware dependency on one another. This can cause unexpected
jitter and may even destroy an expected real-time response.
■ Parallelism. A common pitfall of LabVIEW programmers new to
LabVIEW FPGA is to use the same logic flow they used on the desk-
top and to ignore the physical parallelism exposed by the FPGA.
LabVIEW’s parallel dataflow paradigm maps really well into FPGA
programming because any node that has all its inputs satisfied will
execute. This lets you easily write parallel control loops that actu-
ally run in parallel. And because your code is mapped into hardware
registers, the number of deterministic parallel tasks is only limited
by the size of the FPGA and the efficiency of your code.
■ Flat hierarchy. Even though your application may have a hier-
archical structure to it, it is compiled and mapped onto a “flat”
two-dimensional array of logic without any hierarchy at all. Avoid
excessively complicated logic with many nested Case structures
because this maps very poorly onto the FPGA and consumes too many
resources. Minimizing resource usage is a good thing to practice
because no matter how big the FPGA is, sooner or later you’ll write
a program that is bigger.

Compiling
Compiling for the FPGA can take some time and is the only painful
part of programming with LabVIEW FPGA. When you compile for your
FPGA target, LabVIEW converts your block diagram into VHDL and
 LabVIEW FPGA 407

Figure 16.4 Successful Compile Report from LabVIEW FPGA. The number to
watch is the number of slices used. In this case 196 slices out of 5120 were used.

then uses the Windows command line to call Xilinix’s tools. The end
product is a netlist prescribing how the hardware is interconnected.
The compile process can take a long time, many tens of minutes on a
PC with limited RAM. On a dual-Xeon machine with 2-Gbyte RAM an
average compile is 3 min, and the longest has been 20 min. If you’re
developing on a PXI controller, set up a fast PC with a lot of RAM as a
compile server. You’ll save a lot of time.
Figure 16.4 shows a compile report that is returned after the project
has successfully compiled. Here’s a quick definition of terms:
■ BUFGMUX. BUFG is a buffer for a global clock and MUX is a multi-
plexer. As you add derivative clocks you will consume BUFGMUXs.
■ DCM. Digital Clock Manager is used for clock multiplication. In
this project we derived an 80-MHz clock from the 40-MHz onboard
clock.
■ External IOBs. IOB is an input/output block. The XCV1000 on the
7831R has 324 IOBs, 216 of which are used.
■ LOCed IOBs. These are located IOBs, or constrained to a fixed
location.
■ Slices. This is the most important number to watch as you optimize
your code. The fewer the slices used, the more efficient your code is
and the more you can do in a fixed FPGA space.
■ Base clock. This is the fundamental onboard clock. You can derive
faster clocks by using the 40-MHz reference. In Figure 16.4 we com-
piled our application using an 80-MHz clock.
408 Chapter Sixteen

■ Compiled rate. You can compile your code to use clocks of 40, 80,
120, 160, and even 200 MHz, but the faster you go, the more jitter
you introduce. Additionally, the digital I/O of the XCV1000 is slew-
rate-limited at 20 MHz. The safest practice is to use 40 MHz as your
base clock unless you really need some section of code to run faster.

As you develop applications for the FPGA, take time during develop-
ment to compile and notice how many slices your application consumes.
You can even cut and paste the compile report onto the block diagram
or into the history report to keep a running record of the application
size. Excessively convoluted logic consumes valuable resources, can
cause unnecessary latency, and may even cause your code not to com-
pile. To get maximum performance and the best use out of the FPGA,
keep the code on the FPGA simple and clean.

Debugging
LabVIEW has some great debugging tools, but unfortunately they
aren’t available for compiled code on the FPGA. However, you can use
LabVIEW’s graphical debugging tools, including probes and execution
highlighting, when you use the Emulation mode. To use the emulator,
right-click on the FPGA Target in the project folder and select Target
Properties. . . . I/O in Emulation mode can use randomly generated
data, or you can use the actual I/O hardware of your FPGA target.
This is a great way to do some rapid prototyping and debugging since
VIs in Emulation mode do not have to go through the long compilation
process. Of course there is no real-time response in Emulation mode
since your VI is battling every other process on the host computer for
CPU time. To get hard real-time response, you have to compile and run
down on the hardware where it is much harder to peak into a running
process. See Figure 16.5.
Place digital outputs at strategic points in your code, and toggle
the digital line each time that section of code executes. Each loop is a
potential point you may want to look at. Toggling a digital output once
each loop iteration is a great way to see what is happening. Once your
VI is compiled and running, you can connect an oscilloscope to these
digital watch points to get a real-time look into your process. You might
be surprised at what you find out.

Synchronous execution
and the enable chain
LabVIEW is a data-flow-driven programming language, and LabVIEW
FPGA takes special care to enforce dataflow in the generated code.
This is done by transparently adding synchronization logic to code
 LabVIEW FPGA 409

Figure 16.5 This FPGA project uses Emulation

mode. All LabVIEW’s debugging tools are avail-
able for use on the FPGA in Emulation mode.
You can use the actual I/O hardware of the FPGA
target as I/O in emulation, which allows you to
verify algorithm functionality.

on your block diagram. Figure 16.6A shows a few simple logic func-
tions chained together. What you might expect, if you are familiar with
VHDL programming, is for all the logic functions to map to combinato-
rial logic and execute in a single clock cycle (as shown in Figure 16.6B).
But LabVIEW treats each function as a separate entity requiring syn-
chronization at its inputs and outputs, and LabVIEW does not opti-
mize the code across synchronization boundaries. Instead of getting
the code in Figure 16.6B, you get code that functions like that in Figure
16.6C. This is a huge difference in performance; if each function takes
one clock tick, we’ve gone from one tick to four ticks of the clock to exe-
cute the same code. Figure 16.6C doesn’t tell the whole story though,

Figure 16.6(A) What you drew. (B) What you wanted. (C) What you
got. LabVIEW enforces dataflow on the VHDL code it generates.
410 Chapter Sixteen

because transparently added to each function is an extra input called

the enable chain. The combination of synchronous execution and the
enable chain enforces dataflow but also uses FPGA real estate and
slows down execution.
A good way to illustrate the effect of the synchronization logic and
enable chain on execution is to see how fast you can toggle a digital line
on a RIO card. There are two loops in Figure 16.7. Loop A is a normal
While Loop, and loop B is a single-cycle Timed Loop. Each loop runs as
fast as possible and toggles a digital line. We can benchmark loop per-
formance by monitoring the digital lines on an oscilloscope. With the
VI compiled to use a 40-MHz clock, you would expect to see a 20-MHz
square wave output from both loops. But the dataflow nature means
each piece of the block diagram adds overhead and loop A takes three
ticks to execute, resulting in a 6.67-MHz square wave. Loop B, however,
generates a 20-MHz square wave because it is a single-cycle Timed
Loop. A single-cycle Timed Loop uses clocked execution and executes
all the code inside the loop within a single clock cycle. You can maxi-
mize performance and avoid the enable chain with a single-cycle loop.

Figure 16.7 Two loops toggle digital lines as fast as possible. (A)
Loop overhead and the enable chain cause loop A to execute in
3 clock ticks. (B) Single-cycle Timed Loops execute completely
with each clock tick.
 LabVIEW FPGA 411

Clocked execution and the

single-cycle Timed Loop
Clocked execution was introduced in LabVIEW 7.1 with the single-
cycle Timed Loop. Code placed within a single-cycle Timed Loop is spe-
cially optimized to execute on a single clock tick. Logic functions in the
loop map directly to combinatorial FPGA logic. LabVIEW shift regis-
ters map directly to FPGA flip-flops. The flip-flops hold state from one
clock cycle to the next. This leads to an efficient FPGA implementation,
but unfortunately not all LabVIEW functions are able to map directly
to logic or are capable of executing in a single clock tick. Only combina-
tions of logic capable of executing in a single clock tick can be placed
in a single-cycle loop. Look in the online help for a list of unsupported
functions for the single-cycle Timed Loop.
One workaround you can use to fit more logic into a single clock tick
is to slow down the clock. In the project in Figure 16.8 we have our
40-MHz clock and two derived clocks, one at 20 MHz and the other at
80 MHz. You can have multiple clocks in an FPGA project using the
onboard clock to create derived clocks. To create a new derived clock,
right-click on the 40-MHz onboard clock in the project window and
select New FPGA Derived Clock. Figure 16.9 shows how to configure
your new clock. Once it is configured, you can use it in your project. The
7831R target we are using normally compiles with a 40-MHz clock, but

Figure 16.8 You can have multiple clocks in an FPGA

project. On this target they are all derived from the
40-MHz onboard clock.
412 Chapter Sixteen

Figure 16.9New FPGA Derived Clock dialog. Configure new derivative clocks based on
the onboard clock for your target.

we could change the top-level clock to use our 80-MHz derived clock if
we needed things to run a little faster. Using a faster top-level clock
allows the code to execute faster, but it also introduces jitter into the
system. If you set a clock rate that is too fast, the VI will not compile and
you will have to select a lower clock and compile again. Figure 16.10

Figure 16.10 Single-cycle Timed Loops can use derived clocks as their tim-
ing source.
 LabVIEW FPGA 413

Figure 16.11Timing diagram for an SPI communication. Counting clock

edges provides an easy way to determine states.

shows a single-cycle Timed Loop with a 20-MHz derived clock as its

time base. Clocked execution with the single-cycle Timed Loop and
derived clocks provides a lot of flexibility.

Parallelism
Programming for an FPGA is distinctively different from that for Lab-
VIEW on a desktop. An FPGA executes parallel instructions in par-
allel whereas a microprocessor executes instructions sequentially. It’s
important to take advantage of this parallelism when you develop your
program. A parallel state machine turns out to be an easy way to build
a serial bus controller. Figure 16.11 shows a timing diagram for a sim-
ple serial peripheral interface (SPI) communication sequence. SPI is a
common bus in chip-to-chip communications between microprocessors
and peripheral components. One of the strengths of LabVIEW FPGA
is the ability to interface with low-level devices using communications
busses such as SPI. This lets you build test systems that interface
directly with board-level components. It is fairly easy to design an SPI
bus controller by counting clock edges on a timing diagram. Each com-
ponent will have a diagram in its data sheet, and it’s fairly simple to get
out a ruler and mark out the transitions. Each clock edge represents a
different state. The Chip Select line is set low in state 0 and high in 17.
Data is latched out on the MOSI (master out slave in) line on each fall-
ing edge. The MISO (master in slave out) latches in data on each rising
edge. Figure 16.12 illustrates how parallel execution with a separate
case for each line results in a clean, easy-to-understand diagram.

Pipelining
A powerful extension of parallelism is pipelining. Pipelining breaks a
series of sequential tasks into parallel steps for faster loop rates. This
parallel execution can end up increasing the throughput of a sequential
task. The total time for the code to execute is still the sum of the time
required for each individual task, but the loop rate is faster because
414 Chapter Sixteen

Figure 16.12 SPI controller as a parallel state machine. CLK

toggles each iteration. CS is asserted low at 0 and high at 17.
MOSI clocks out data 1 bit at a time at states 1, 3, 5, 7, 9, 11, 13,
15. MISO reads in data at 2, 4, 6, 8, 10, 12, 14, 16. The other cases
are empty.

each piece of code executes in parallel. The diagram in Figure 16.13

show two identical tasks performing the same function. The sequence
frame waits for a rising edge on DIO0 before reading the bit on DIO1.
Data is shifted in and compared for a match with a synch word. When a
match is found, DIO2 is toggled. Figure 16.13A shows the nonpipelined
approach. Each action happens sequentially. Figure 16.13B shows the
pipelined version. There are still three distinct sections of code, but
now they happen in parallel. Pipelining this process does not reduce
the total time between the rising edge on DIO0 and the toggle of DIO2,
but it does allow the VI to work with a faster clock on DIO0.

Conclusions
The great thing about programming with LabVIEW FPGA is that it’s
just LabVIEW. You don’t have to learn another language or develop-
ment environment. The hardware interface tool LabVIEW FPGA puts
in your toolbox is powerful. Once you understand the subtleties of
the hardware target, you’ll wonder how you ever did without it. Just
remember to keep your code clean and simple, take advantage of the
 LabVIEW FPGA 415

Figure 16.13 Pipelined versus nonpipelined approaches to a problem. The pipe-

lined version is more efficient and can work with a faster clock on DIO0.

inherent parallelism in the FPGA, and verify through each step of

development. Also be aware of the impact of I/O arbitration on loop
speed. By default, LabVIEW places extra flip-flops as arbiters, but they
add execution overhead. Incremental development and test is the best
way to verify the impact of each function on your code.

Bibliography
“Thinking Inside the Chip,” NIWeek04, National Instruments Corporation, 11500 N.
Mopac Expressway, Austin, Tex., 2004.
“Virtex-II Platform FPGAs: Complete Data Sheet,” DS031 (v, 3.3), Xilinx, Inc., 2100 Logic
Drive, San Jose, Calif., June 24, 2004.
This page intentionally left blank
 Chapter

LabVIEW Embedded
17
In the last edition of this book we showed you how to build an embed-
ded system using LabVIEW for Linux and a PC104 computer. It was
big and power-hungry, but at the time it was the only way to build an
embedded system with LabVIEW. National Instruments has been hard
at work on its own version of LabVIEW for embedded systems, and it’s
so much better than anything you or I could do. LabVIEW can now
target any 32-bit microporcessor (Figure 17.1) To get an inside look at
LabVIEW Embedded, we asked P. J. Tanzillo, National Instruments’
product support engineer for the LabVIEW Embedded platform, to
give us a hand and tell us what this new technology is all about. So,
without further ado, here’s P. J.:

Introduction
LabVIEW is much more than a tool for instrument control and data acqui-
sition. With its extensive built in signal processing and analysis libraries,
LabVIEW has developed into an ideal environment for algorithm develop-
ment and system design. Furthermore, with the addition of the LabVIEW
modules like the LabVIEW Real-Time Module and the LabVIEW FPGA
Module, it has become possible to run these applications not just on a
Windows PC, but on other hardware platforms as well. In this chapter,
we will examine the newest such module called the LabVIEW Embed-
ded Development Module. This latest addition to the LabVIEW family
allows you to develop applications for any 32-bit microprocessor.

History
In the late 1990s, it became clear that Personal Digital Assistants (PDAs)
were gaining mainstream acceptance as a viable computing platform.
National Instruments saw this platform as a potential compliment to their

417
418 Chapter Seventeen

Figure 17.1 LabVIEW targets devices from the desktop to the palmtop, and
beyond. LabVIEW Embedded can target any 32-bit microprocessor. (Photo cour-
tesy of National Instruments)

industrial control hardware, and so investigation began into how some-

one might use LabVIEW to program PDAs. The initial concept proposal
included a potential use case: A factory foreman monitors and controls
his/her headless automation controller with a simple LabVIEW Graphical
User Interface on a handheld device.

At the time, however, PDAs were running on several processor architec-

tures with various operating systems, and no single technology was emerg-
ing as a standard. Since LabVIEW is a compiled language, it would have
been necessary to port the entire LabVIEW compiler and run-time engine
in order to support a new architecture. Porting LabVIEW to a single new
architecture would have been a significant effort, and this approach would
have required several such efforts to address any significant number of
PDA users.

Furthermore, the application memory space on PDAs was signifi-

cantly smaller than any other platform that LabVIEW then supported.
 LabVIEW Embedded 419

Therefore, even more effort would have to be made to reduce the code foot-
print of LabVIEW built executables. It became clear that a new approach
would be needed to port LabVIEW to these small, inconsistent targets.

Therefore, investigation began into creating a LabVIEW C code gen-

erator. This approach was chosen because if LabVIEW were to generate
ANSI C, then it could use existing C cross-compilers to build the applica-
tion for any PDA. Furthermore, the code footprint would not be an issue
since LabVIEW could then take advantage of the aggressive code dead-
stripping of C compilers.
After a significant development effort, this technology was released for
the first time in 2003 as the LabVIEW PDA Module. As development con-
tinued and the technology matured, it became evident that this approach
of C Code generation could be applied more generally toward embedded
microprocessors. Since C cross-compilers already existed for almost any
embedded platform, it was simply an issue of decoupling the LabVIEW C
Code from the PDA and adding optimizations so that LabVIEW generates
efficient C Code. Once that effort was completed, and once a framework
and documentation was developed to help a user port LabVIEW to any
target, the LabVIEW Embedded Development Module was released in
May 2005.

LabVIEW Embedded Development Module

From a high level, the goals of the LabVIEW Embedded Development
Module were simple. National Instruments wanted to make program-
ming of embedded systems easier by allowing the LabVIEW program-
ming experience to be extended to arbitrary hardware. In many cases, a
system’s specifications can be met by a commercial off the shelf hardware
platform like NI CompactRIO. Thus, the LabVIEW Real-Time Module is
sufficient for such situations. However, there are many other cases where
constraints such as size, power consumption, and production volume will
make it necessary for custom hardware to be created. With the intro-
duction of the LabVIEW Embedded technology, this second case is now
achievable with LabVIEW as well.

Unfortunately, since the embedded development market is so fragmented

with seemingly infinite combinations of processor architecture, I/O periph-
erals, and operating systems, it again proved impossible to identify the
“correct” platform to meet the needs of any significant number of embed-
ded programmers. Therefore, it became necessary to allow for LabVIEW
generated C Code to be integrated with any operating system and embed-
ded toolchain (Figure 17.2).

The technology: What’s happening

In order to get from a LabVIEW VI to a cross-compiled embedded applica-
tion, several steps must be completed.
420 Chapter Seventeen

Figure 17.2 LabVIEW generates C code.

First, LabVIEW must traverse the block diagram and generate ANSI C
code for the top-level VI as well as for every subVI in the VI hierarchy. The
LabVIEW C Code Generator generates simple C primitives wherever
possible. For example, while loops convert to while() statements, and the
add function generates a simple ‘+’. More complex functions, however, can-
not provide such a direct mapping. Therefore, an additional component
of the LabVIEW generated C Code is the LabVIEW C Runtime Library.
This is analogous to the LabVIEW run-time engine on the desktop, and
we will discuss the LabVIEW runtime library in greater detail later in
this chapter.

Next, all of the generated code must be passed through a cross-compiler

that is specific to the desired target. In addition to the generated code, you
can also add preexisting C source code to your LabVIEW Embedded Proj-
ect. These files are passed through the cross-compiler with the generated
code and linked into the executable. Once the compiling and linking is
complete, you will have a cross-compiled executable that is specific to your
target in your build folder along with all generated source and object files.
This executable needs to be loaded into the correct location in memory
on the device, and the appropriate execute command needs to be sent so that
the application begins running. The typical communication mechanisms
for downloading code and communicating with an embedded device are an
RS-232 serial connection, an Ethernet connection, or a JTAG emulator.

Finally, in order to have live front panel controls and indicators, the target
and the host need to establish communications. This can again be done
through any protocol that can be understood by both the host and the
target. This can be implemented in one of two possible ways.
 LabVIEW Embedded 421

The simpler method of the two is called instrumented debugging. In

this approach, additional function calls are added to the generated C Code
that handle sending updates back to LabVIEW. Since details of the com-
munication protocol are built into the implementation of instrumented
debugging, you only need to provide LabVIEW with implementations for
open, read, write, and close. Common communications used for instru-
mented debugging are serial and TCP/IP, but other successful implemen-
tations have been seen using shared memory or CAN. The benefits of this
approach lie mostly in the ease of implementation. In addition, there is
no need for external JTAG emulators which can be very expensive and
cumbersome. This approach, however, will expand the code footprint by as
much as 40%, and since the additional functions are added to the gener-
ated code, there is a non-negligible overhead that will be incurred by the
communication routines.

The other debug method is called on-chip debugging. This method does
not call for any additional functions to be added to the generated code,
so debug and release builds are identical. Rather than adding additional
function calls, here the LabVIEW C code generator adds only additional
comments. Each wire, control, and indicator on the block diagram results
in a separate generated comment. At compile time, these comments are
mapped to physical memory locations on the chip, and this information is
stored in a debug database. When LabVIEW needs a value of a specific
wire, indicator, or probe, it looks to the debug database to see what seg-
ment of physical memory that it needs to read. Then, it requests the value
from the JTAG emulator/debugger using that debugger’s provided API.
This approach is considerably more complex, and thus, it requires more
implementation effort. For example, the means of mapping lines of C code
to physical memory locations is compiler dependent and non-trivial. In
addition, if the debugging interface requires lengthy processor halts to
read memory, this method can be rather intrusive to the program’s exe-
cution. However, high quality probes that can efficiently read and write
memory can result in almost completely non-intrusive debugging. In addi-
tion, full debugging information can be included without increasing the
final code footprint.

Running LabVIEW Embedded on a new target

Before you begin porting LabVIEW to your target, you must first be able
to build, download, and run a simple application in C. This expertise is
necessary to ensure that the Board Support Package (BSP) is correctly
installed and configured. In addition, this will ensure that you have the
necessary toolchain on the host so that you can compile and link the
LabVIEW generated C code to create embedded applications. In addition,
you will need to select hardware that is compatible. The following are
some required and recommended specifications for a LabVIEW Embed-
ded target.
422 Chapter Seventeen

Hardware requirements:
■ 32-bit processor architecture
■ 256K of application memory (plus whatever is required for your embed-
ded OS)

Hardware recommendations (but not requirements):

■ Operating System—This will allow you to take advantage of parallel-
ism and multithreading in LabVIEW Embedded.
■ Hardware Floating Point Unit—Since the LabVIEW Analysis Library
is built on floating point math, there will be a significant performance
gain if you are able to complete that math in hardware rather than
software emulation.

Once your hardware, toolchain, and BSP are in place, you can begin the
process of porting LabVIEW to your target. This consists of four main
steps, each of which we will discuss in detail. They are:

Porting the LabVIEW Runtime Library

Incorporating the C Toolchain
Incorporating I/O Drivers
Customizing the LabVIEW Environment

The rest of this chapter will be spent discussing these four steps in detail.

Porting the LabVIEW runtime library

The LabVIEW C code generator will produce the same C code regardless
of the target platform. Each LabVIEW function that does not map directly
to a C primitive is implemented using a function call. These generic func-
tions are implemented in the LabVIEW C Runtime Library. These
functions are themselves implemented atop an even lower layer of generic
functions that handle memory movement, string manipulation, simple
timing, as well as a few other miscellaneous functions.

One of the major components of porting the LabVIEW C Runtime Library

is providing a mapping of these generic low level function calls to func-
tions that are specifically supported by your runtime environment. This
will ensure that the LabVIEW generated C Code can be compiled for your
target platform. This mapping needs to be completed for about 50 specific
functions in one header file called LVDefs_plat.h. Here, a series of #define
statements map the generic function calls like StrCaselessCompare( x, y )
to Operating System (OS) specific function calls like strcasecmp( x, y ) (for
Linux and VxWorks) or _stricmp( x, y ) (for Windows). In addition to pro-
viding this mapping, LVDefs_plat.h also contains some important feature
support flags such as defining whether or not a file system is present or
whether TCP/IP or UDP communication is supported. Since most operat-
ing systems have migrated to similar APIs, there is no reason to begin
 LabVIEW Embedded 423

creating LVDefs_plat.h from scratch. In most cases, it is wise to simply

copy an existing implementation that is for an OS most like yours, and
then begin to modify it where necessary.

In order for your target to support some of the more advanced features in
LabVIEW, you will also need to implement some OS specific functions. For
instance, in order for the timing primitives to be supported, you will need
to implement the LVGetTicks() function. Similar functions are defined in
the porting guide that are necessary for further LabVIEW synchroniza-
tion and timing features like Timed Loops and Notifiers. Source files for
the OS specific pieces of the LabVIEW C runtime library can be found in
an operating system specific folder in the C Generation codebase.

Finally, the main entry point for all applications that LabVIEW gener-
ates can be found in LVEmbeddedMain.c. This file contains functions that
perform set up and tear down of common pieces such as occurrences and
FIFOs as well as any hardware specific setup and tear down routines
that are necessary for a given target. The main function in LVEmbed-
dedMain.c initializes all global variables and then calls the top-level VI
in the project. After the top-level VI has completed, a shutdown routine
is completed. These initialization and shutdown routines are defined by
two macros in LVEmbeddedMain.c called LV_PLATFORM_INIT and
LV_PLATFORM_FINI. These macros are defined per target rather than
per OS because different routines may be required for separate hard-
ware platforms that share the same OS. For instance, although they both
run eCos, a PowerPC for engine control and an ARM processor in the
Nintendo Gameboy Advance would require very different hardware ini-
tialization routines.

Incorporating the C toolchain

Once the LabVIEW Runtime Library has been ported, you should be able
to manually compile and run LabVIEW generated C Code on your target.
The next section will discuss the architecture of LabVIEW Embedded, the
infrastructure that exists for incorporating your toolchain so that build-
ing, downloading, and debugging is an automated process.

Let’s first review the process of cross-compiling a simple C program from

a command line. It is important to understand this well, as this is the pro-
cess that LabVIEW will need to invoke for a complete and ported target. A
typical compiler command will consist of the following components:

compiler –XXX(C Flags) sourcefile.c –I Include files

The output of such a command will be an object file (*.o) that will have
the same file name as the C source file. A similar compile command will
need to be executed for every C source file in your project (as well as the
LabVIEW C runtime library). Once the compile step is completed, you will
need to link all the compiled object files and pre-existing libraries into a
single stand-alone executable file. This is done by running the compiler
424 Chapter Seventeen

command again, this time with appropriate flags set so that linking is
performed. This command will typically look like this:

compiler –XXX(L Flags) "destinationfile" object files

Once the application has been successfully linked, you will have a cross
compiled executable for your target platform. This is what we will be
referring to as “building” your application. Though the syntax and order
of components of the compiler for your target may vary, these key compo-
nents will need to be in place for the compiler and the linker to succeed.

The Embedded Project Manager

In order to generate C Code and build that generated code into an embed-
ded application, you will need a LabVIEW Embedded Project (*.lep) in
addition to the VIs in your hierarchy. The .lep file contains information
about the C Code generation options as well as any external source files
that you wish to build along with the generated code. The LabVIEW
Embedded Project Manager (Figure 17.3) is the common user interface
that allows you to modify all of your projects for all of your targets. It pro-
vides a single source for all project configuration data (such as source files

The Embedded Project Manager controls

Figure 17.3
embedded builds.
 LabVIEW Embedded 425

and additional compiler flags) as well as target specific settings (such as

IP address or COM port settings).

The LabVIEW Embedded Project Manager is also the place where you
interact with the target and the project. For example, there is a “Build”
button that will cause LabVIEW to generate the C Code for all of the VIs
in the project (as well as for the entire VI hierarchy for each of these VIs)
and compile and link the generated code. The implementation of this and
other actions within the LabVIEW Embedded Project Manager vary from
target to target, and will therefore need to be modified when adding sup-
port for a new target.

LEP plug-in VIs

Each LabVIEW Target has a series of implementation VIs called Plug-In
VIs. These VIs govern the interaction between LabVIEW and the target
specific toolchain. For example, when the user hits the “Build” button in the
LabVIEW Embedded Project Manager, there is a plug-in VI executed that
is called “XXX_Build.vi” where XXX is a target specific prefix. This Build
VI is responsible for pulling the build options and target information from
the .lep file, translating these options into C code generation options and
C compiler flags. Then, this VI executes all of the appropriate compiler
commands such that at the end of its execution, a finished, cross-compiled
executable is present in the project folder (Figure 17.4). Each target will
have a different Build VI, and each will invoke a different compiler and
linker. In order to keep a clear distinction between plug-in VIs for different
targets, a naming convention and folder hierarchy has been introduced.

Once you have installed the LabVIEW Embedded Development Module,

you will find a folder for each of the operating systems of the shipping
example targets. Within each of these folders, you will find a folder called
OS_LEP_TargetPlugin where OS is the name of the operating system.

Figure 17.4 Build.vi turns your project into an executable. Only one case is shown for clarity.
426 Chapter Seventeen

Within the OS folders that support multiple hardware platforms, you

will find an additional folder for each supported hardware platform. For
instance in the vxworks folder, there are two additional folders called
ixdp425 and cmd565. Each of these sub targets will have their own plug-in
folder called OS_HW_LEP_TargetPlugin where OS is the name of the
operating system and HW is the name of the hardware platform. Each of
the child targets will inherit much of their functionality from the parent
while overriding only that which must be changed to support the specific
hardware platform. Just as is the case with porting the LabVIEW C run-
time library to your target, much of the code that exists in the plug-in VIs
can be used generically for any target platform. Therefore, it is advisable
to begin your implementation of a new target by copying an existing tar-
get and modifying it. An Embedded Target Creation Utility has been
added to the LabVIEW Embedded Development Module to help automate
the process of copying, renaming, and relinking an existing target. If none
of the existing example targets meet the needs of your target platform,
there is also a blank template target included with only the Code present
that is common to all embedded targets.

Target_OnSelect
When exploring the target plug-in VIs for the first time, it makes the most
sense to begin with the Target_OnSelect.vi. This VI consists of a series
of cases in a case structure, each of which corresponds to a target menu
selection from the Embedded Project Manager. In the future, this VI will
be replaced by a target specific xml file that will provide the mapping from
the LabVIEW project to specific implementation VIs.

Common target menu selections like Build, Run, Debug, and Build Options
will likely be present for most targets, but you have the option to add cus-
tom target menu options for your target by modifying the Target_Menu.
vi and the Target_OnSelect.vi. For example, the LabVIEW Embedded
Module for ADI Blackfin Processors includes an option to reset the Black-
fin board. In the future, this information will also be defined in the target
specific xml file.

Other plug-in VIs

The Build plug-in VI is responsible for converting a VI to a cross compiled
executable that can be run on the target platform. This task is primarily
accomplished in the ScriptBt.vi (Bt is short for Build Tools), and it can
be broken down into four primary steps:

Code generation
1. The CGen plug-in VI is responsible for traversing the block diagram of
the Top-Level VI and generating ANSI C code for all of the applicable
VIs in the project. This capability is built into LabVIEW Embedded, and
 LabVIEW Embedded 427

it is exposed through a VI server call. Once this VI is complete, the proj-

ect folder should have a *.c file for every VI in the VI hierarchy that is
not part of the native LabVIEW function set. The input to this VI server
call is a cluster of configuration options that will determine the specifics
of the C code that is generated. For example, one of the elements of the
configuration cluster is a Boolean control called GenerateDebugInfo.
If this value is true, then additional function calls will be added to the
generated code so that the embedded application will communicate back
to LabVIEW in order to update front panel indicators and acquire val-
ues from front panel controls (this is called Instrumented Debugging).
If the GenerateDebugInfo control is set to false (the default value), then
this code will not be included in the generated C code. This option (and
most of the other options that are discussed in this section) are exposed
to the user through the Build Options dialog that can be found for all
example targets in Target >> Build Options…. The following are some
of the other most commonly used C generation options:
■ GenerateGuardCode. Guard code prevents a user from making
common mistakes that would typically cause an application to crash.
For example, guard code can prevent dividing by zero and indexing
out of range in an array.
■ GenerateSerialOnly. The default generated C code has the same
execution behavior as LabVIEW on the desktop. For example, par-
allel While Loops run in parallel. However, the LabVIEW C Code
Generator generates code that uses cooperative multitasking in a
single thread. Additional threads are used only by Timed Loops.
The generated C code is difficult to read and does not resemble code
that a human would write. Set this attribute to TRUE if you do not
want to generate the cooperative multitasking code. The code is eas-
ier to read and usually runs much faster without the cooperative
multitasking code. However, you lose the parallel execution behav-
ior. Parallel While Loops execute in sequence with one While Loop
executing only after the other While Loop completely finishes. This
execution behavior most closely resembles subroutine priority VIs in
LabVIEW for Windows.
■ DestinationFolder. DestinationFolder indicates where you want the
LabVIEW C Code Generator to save the generated C files. Unless the
Incremental Build attribute is TRUE, any C files in the destination
folder that have the same name as VIs in the build are overwritten.
■ GenerateIntegerOnly. If a block diagram does not contain any
floating-point data types, the GenerateIntegerOnly attribute generates
C code that does not have any floating-point declarations or operations.
You can link the generated code with a run-time library compiled with
the _Integer_Only flag set to produce applications that run without
hardware or software floating-point support. Generate-IntegerOnly
does not support fixed-point operations in the Embedded Development
Module.
428 Chapter Seventeen

■ GenerateCFunctionCalls. Determines the calling interface to

subVIs only if you set GenerateSerialOnly to TRUE. The calling
interface to a subVI usually is generated in such a way that if an
input to or output from the subVI is unwired, the default data is
used. This increases the overall amount of code and generated data
for a VI relative to what you might use in a normal C program. If
all the inputs and outputs are wired, default data is not needed and
a more efficient interface could be generated. Set this attribute to
TRUE to generate the interface to all VIs as C style function calls
without any default data initialization, which can reduce the code
size by as much as 50% for a small VI. An error occurs if any input or
output to any VI is unwired when the C code is generated.
■ OCDIComments. Extra comments are placed in the generated
C code to help figure out where certain items such as wires and nodes,
are in the generated code for use during non-intrusive debugging.
These comments make the generated code harder to read. Use this
option to turn off these comments when they are not needed. The Lab-
VIEW Embedded Development Module ships with an example target
called Code Generation Only. This target only generates C Code for a
VI when the build button is pressed and does not attempt to compile
the generated C Code. Furthermore, most of the C code generation
options that have been discussed in this section are exposed as build
options in the Target >> Build Options…dialog. Feel free to experi-
ment with the build options using this target and examine the gener-
ated C Code to observe the differences.
2. The Prepare.vi is responsible for readying the project folder for the
compilation processes as well as querying the *.lep file and building up
a list of source and include files based in the current project information.
For the most part, the prepare VI is relatively consistent in implemen-
tation from target to target. In most cases, the Prepare.vi accepts the
project configuration information (like the build settings and the proj-
ect folder) as an input, and produces an array of source files, include
files, and precompiled libraries to be used by the ScriptCompiler and
ScriptLinker VIs as inputs. The LabVIEW C Code runtime library
can either be pre-built and linked in as a library or the source files can
be built each time the project is built. The advantage to re-building the
C code runtime library each time is that the C Compiler can apply much
more aggressive deadstripping to the object files and therefore make the
code footprint of the final application considerably smaller. In order to
achieve such an end, each of the files in the LabVIEW C runtime library
should be added to the array of source files in the build VI. If code size is
not a major issue, however, it is wise to pre-build the runtime library and
add it to the array of libraries in the Prepare.vi. This will reduce the time
that it takes to build a LabVIEW Embedded application considerably.
3. The ScriptCompiler.vi is responsible for scripting the compiler for
each of the files in the source file array that is produced by the
Prepare.vi. This typically consists of a list of the generated C files (one
for each VI in the VI hierarchy of the project) as well as any external C
 LabVIEW Embedded 429

files that the user has included in the project. In essence, this VI is
responsible for building up the correct compiler command in a large
string and executing this from the command line using the SysExec.
vi from the LabVIEW functions palette. The inputs to this VI are typi-
cally the configuration options for the project, an array of source files,
and an array of file paths that need to be included. The result of a suc-
cessful execution of this VI is a project folder full of object files that will
be used by the linker to produce the executable. To observe the build
process of any target, you will only need to have the ScriptCompiler.vi
open while you are building a project.
4. The ScriptLinker.vi works in much the same way as the Script-
Compiler.vi, only now, the command that is executed is the command
necessary to link all of the object files to produce one stand alone
executable. Just as you monitored the progress of the ScriptCompiler.vi
in the previous section, you can also monitor the execution of the
ScriptLinker.vi in a similar way.
5. The Debug.vi is responsible for initiating a connection between Lab-
VIEW and the running application so that you can have an interactive
LabVIEW debugging experience including probes, breakpoints, and live
front panels. This can be completed in one of two ways—instrumented
debugging (TCP, Serial, CAN, etc.) and on-chip debugging (JTAG). In
either case, the Debug plug-in VI relies on the proper and complete
implementation of the debugging plug-in VIs. Once one or both of these
mechanisms are in place, you only need to begin the debugging session
from the Debug.vi and LabVIEW will handle the background commu-
nications to the application. For instrumented debugging via TCP/IP
four things need to happen:
■ The generated code has debug information included.
■ The compiler directive of UseTCPDebugging=1 is set.
■ The application is run with the correct host IP address as a param-
eter.
■ The niTargetStartTCPDebug is called after the application is running.

Incorporating I/O srivers

In traditional embedded C programming, analog and digital I/O has typi-
cally been done only on a register level. Even C level driver APIs are a
rarity, and most data acquisition has to be done by direct memory access.
LabVIEW Embedded introduces a means of abstracting the low level
implementation of such tasks in a unified API called Elemental I/O
(Figure 17.5). Although the register level access still needs to be included
in the implementation of the Elemental I/O nodes when porting to a new
target it must only be done once per target platform. The result is a simple
access point to the analog and digital peripherals that are available in
your system. Furthermore, the Elemental I/O implementation changes
based on the current target, making LabVIEW code that was developed
on one platform completely portable to another supported platform.
430 Chapter Seventeen

Figure 17.5 Elemental I/O provides simple access to analog and digital peripherals.

In order to enable easy register level programming in the LabVIEW envi-

ronment, a new structure was added to LabVIEW Embedded called the
Inline C Node (Figure 17.6). This new structure is used much in the same
way that the Formula Node is used in core LabVIEW. However, the Inline
C Node allows a user to add their own C and assembly code to the gener-
ated C code to facilitate easy register access and small code optimizations.

I/O driver development can differ greatly depending on the type and com-
plexity of your device. For example, digital input and output is typically
done by accessing the general purpose I/O pins on the device, and devel-
opment of a driver can be relatively straightforward. However, a buffered
DMA analog acquisition through an external A/D converter can be much
more challenging to develop. In all cases, however, as with any I/O, there
will typically be some initialization code that must be executed before
the acquisition can be completed. In the traditional LabVIEW I/O pro-
gramming model of Open -> Read/Write -> Close, the initialization code

Figure 17.6Inline C node allows you to add your own C and assembly code directly
on the LabVIEW block diagram.
 LabVIEW Embedded 431

obviously resides in the open VI. However, since an elemental I/O node
consists of a single block, the implementation of this becomes slightly less
apparent. To accomplish initialization within the Elemental I/O single
node framework, it is best to declare a static integer in your Inline C node
called “initialized.” This can act as a flag that should be set and reset
accordingly as the acquisition is initialized and completed.

LabVIEW Embedded programming best practices

When developing an embedded application, system constraints such as
memory limitations and time-critical code requirements can play a cru-
cial role in your programming approach. This document outlines good
programming practices that can be used to help optimize your embed-
ded application with LabVIEW 7.1 Embedded Edition. The techniques
described are applicable for LabVIEW on the desktop as well, but the
benefit will be most evident on an embedded target.

Avoid dynamic memory allocation. Dynamic memory allocation is very

expensive in time-critical code. In LabVIEW Embedded, dynamic memory
allocation can occur when using the Build Array and Concatenate String
functions. Alternatively, the Build Array primitive can be replaced with
a Replace Array Subset function in order to replace elements in a pre-
allocated array. The preallocated array should be created outside of the
loop by using an array constant or with the Initialize Array function. Lab-
VIEW code is shown below (Figure 17.7) to contrast the different imple-
mentations.

Avoid placing large constants inside loops. When a large constant is

placed inside a loop, memory is allocated and the array is initialized at the
beginning of each iteration of the loop. This can be an expensive operation
in time-critical code. A better way to access the data is to place the array
outside of the loop and wire it through a loop tunnel, or to use a global
variable. Examples of the two recommended methods are shown below
(Figure 17.8).

Use global variables instead of local variables. Every time a local vari-
able is accessed, extra code is executed to synchronize it with the front
panel. Code performance can be improved, in many cases, by using a
global variable instead of a local. The global has no extra front panel syn-
chronization code and so executes slightly faster than a local.

Use shift registers instead of loop tunnels for large arrays. When pass-
ing a large array through a loop tunnel, the original value must be copied
into the array location at the beginning of each iteration, which can be
expensive. The shift register does not perform this copy operation, but
make sure to wire in the left shift register to the right if you don’t want
the data values to change (Figure 17.8A).
432 Chapter Seventeen

Figure 17.7 Avoid dynamic memory allocation. (A) Array is pre-initialized

and filled in inside the loop. (B) Build Array function allocates new memory
each time it is called.

Use integer operations instead of floating point operations. If your pro-

cessor does not have a floating point unit, converting to floating point to
perform an operation and then converting back to an integer data type can
be very expensive (Figure 17.9). In the examples below, using a Quotient &
Remainder function is faster than a normal Divide function, and using a
Logical Shift function is faster than a Scale by a Power of 2 function.

Avoid automatic numeric conversions. Another technique for improv-

ing code performance is to remove all implicit type conversions (coercion
dots) (Figure 17.10). Use the Conversion functions to explicitly convert
data types as this avoids a copy operation and a data type determination.

Avoid Case structures for simple decision making. For simple decision
making in LabVIEW, it is often faster to use the Select function rather
than a Case structure. Since each case in a Case structure can contain its
own block diagram there is significantly more overhead associated with
this structure when compared with a Select function. However, it is some-
times more optimal to use a case structure if one case executes a large
amount of code and the other cases execute very little code. The decision
to use a Select function versus a Case structure should be made on a case
by case basis.
 LabVIEW Embedded 433

Figure 17.8 (A) Memory for look up table is allocated once outside
the loop. (B) A single copy of look up table is stored in global vari-
able. (C) Memory for look up table is allocated each loop iteration.

Figure 17.9 Integer opera-

tions are faster than float-
ing point if your processor
does not have a floating
point unit.
434 Chapter Seventeen

Figure 17.10 (A) Explicitly type convert inputs

to subVIs. (B) Type coercion involves a data copy
and data type determination.

Avoid In Range and Coerce in time-critical code. The In Range and

Coerce function has significant overhead associated with it due to the spe-
cial user configurable features and extra data type determination opera-
tions. This function should be re-implemented with comparison and Select
functions if it is used in time-critical code.

Avoid clusters in time-critical code. Clusters passed into subVIs will

result in unnecessary data type information being passed as well. In order
to speed up your code, try not to use clusters in time-critical areas of your
block diagram.

Often the best results can be obtained by using a hybrid of LabVIEW and
C code. The Inline C Node and Call Library Node allow the use of C code
directly within your LabVIEW block diagram. See the Embedded Devel-
opment Module documentation for more information on the use of the
Inline C and Call Library Nodes. By following good embedded program-
ming practices, you can better optimize your code to meet the constraints
of your embedded application. Implementing one or two of these tech-
niques may noticeably improve the performance of your application, but
the best approach is to incorporate a combination of all these techniques.

Interrupt driven programming

Many of you may be familiar with event driven programming on a PC.
In this programming model, code is not executed sequentially. Rather, a
user interacts with a user interface, and this interaction triggers some
subset of code to be executed. If there is no user interaction, the program
sits in an idle state waiting for the next event. The analogous programming
model for embedded systems is called interrupt driven programming.
In this model, hardware events are detected by the processor and desig-
nated code is executed in response; the specialized segments of code are
called Interrupt Service Routines (ISRs). These routines are typically
 LabVIEW Embedded 435

Figure 17.11 Interrupt driven Timed Loop executes once per interrupt.

executed in a separate interrupt thread that is very high priority. There-

fore, it is typical practice to spend as little time in the ISR as possible.
Usually, such routines are only responsible for re-enabling the interrupt
and setting some flag that can then be handled by the main program.

In LabVIEW Embedded, you can designate a VI as an ISR VI. Thus, you

can develop G code that is triggered by some asynchronous hardware
event, and this code will steal the execution from your top level VI at any
point. However, since the ISR VIs do run in interrupt time, it is desirable
to do as little in the ISR VI as possible.

The timed loop is a specialized version of the while loop that was added to
LabVIEW 7.1. This structure allows you not only to control the loop rate,
but also set independent priorities for the execution thread of that loop. In
LabVIEW Embedded, this structure simply spawns an OS thread, and the
embedded OS’s scheduler handles scheduling the different threads. This
is relevant to interrupt driven programming because you also have the
ability to assign an external timing source to a timed loop. Furthermore,
you can then programmatically fire a timed loop from an ISR VI. This
allows for a block diagram with multiple parallel timed loops with each
loop’s timing source associated with some interrupt (Figure 17.11).

LabVIEW Embedded targets

The LabVIEW Embedded Development Module comes with several exam-
ple targets to serve as a model when porting to your custom hardware. For
a complete list of supported hardware, visit ni.com/embedded.
This page intentionally left blank
 Chapter

Process Control Applications

18
Industrial process control has its roots in the big process industries,
sometimes called the Four P’s: paper, petrochemicals, paint, and phar-
maceuticals. These plants are characterized by having thousands of
instruments measuring such variables as pressure, temperature, flow,
and level, plus hundreds of control elements such as valves, pumps, and
heaters. They use a great deal of automatic control, including feedback,
sequencing, interlocking, and recipe-driven schemes. Modern control
systems for these plants are, as you might imagine, very complex and
very expensive. Most large process control systems are designed and
installed through cooperative efforts between manufacturers, system
integrators, and the customer’s control engineering staff. These are the
Big Guns of process control.
Chances are that you are probably faced with a smaller production
system, laboratory-scale system, or pilot plant that needs to be mon-
itored and controlled. Also, your system may need to be much more
flexible if it’s experimental in nature. Even though your needs are dif-
ferent, many of the concepts and control principles you will use are the
same as those used in the largest plants, making it worth your while to
study their techniques.
Large process control systems generally rely on networked minicom-
puters with a variety of smart I/O subsystems, all using proprietary
software. Until recently, most software for these larger systems was not
open, meaning that the end user could not add custom I/O interfaces
or special software routines nor interconnect the system with other
computers. Even the smaller process control packages—most of which
run on PCs—have some of these lingering limitations. You, however,
have an advantage—the power and flexibility of LabVIEW. It’s not just
a process control package. You can begin with a clean slate and few
fundamental limitations.

437
438 Chapter Eighteen

To help LabVIEW users address the special needs of process control

applications, National Instruments created an add-on toolkit that you
install on top of LabVIEW. It’s called the Datalogging and Supervi-
sory Control (DSC) module. Take everything that we talk about in
this chapter, do all the programming, add device servers (drivers) for
popular industrial I/O, and you’ve basically got DSC. As always, if the
capability of DSC doesn’t meet your needs, you’re still developing in
LabVIEW, so you can do regular G programming and customize to suit.
We liked DSC when it came out because it solved most of our process
control challenges right out of the box. Whether you use DSC, or not,
the rest of this chapter will make the whole picture clearer.

Process Control Basics

In this chapter, we’ll cover the important concepts of process control
and then design the pieces of a small process control system, using
good LabVIEW practices. First, a plug for the ISA: The U.S.-based
Instrumentation, Systems, and Automation Society (ISA) sets many
of the standards and practices for industrial process control. Gary
joined the ISA some years ago while working on the control system
for a large experimental facility, and he found his membership to
be quite helpful. It offers a catalog of books, publications, standards
documents, and practical training courses that can make your plant
and your work practices safer and more efficient. The concepts and
terminology presented in this chapter are straight out of the ISA pub-
lications, in an effort to keep us all speaking a common language.
For information about membership or publications, contact the ISA
at www.isa.org.

Industrial standards
Standards set by the ISA and other organizations address both the
physical plant—including instruments, tanks, valves, piping, wiring,
and so forth—as well as the man-machine interface (MMI) and any
associated software and documentation. The classic MMI was a silk-
screened control panel filled with controllers and sequencers, digital
and analog display devices, chart recorders, and plenty of knobs and
switches. Nowadays, we can use software-based virtual instruments
to mimic these classic MMI functions. And what better way is there to
create virtual instruments than LabVIEW?
Engineers communicate primarily through drawings, a simple but
often overlooked fact. In process control, drawing standards have been in
place long enough that it is possible to design and build a large process
plant with just a few basic types of drawings, all of which are thoroughly
 Process Control Applications 439

specified by ISA standards. By the way, ISA standards are also regis-
tered as ANSI (American National Standards Institute) standards.
Some of these drawings are of particular interest to you, the con-
trol engineer. (See? We’ve already promoted you to a new position, and
you’re only on the third page. Now read on, or you’re fired.)

Piping and instrument diagrams and symbols. The single most impor-
tant drawing for your plant is the piping and instrument diagram
(P&ID). It shows the interconnections of all the vessels, pipes, valves,
pumps, transducers, transmitters, and control loops. A simple P&ID
is shown in Figure 18.1. From such a drawing, you should be able to
understand all the fluid flows and the purpose of every major item
in the plant. Furthermore, the identifiers, or tag names, of every
instrument are shown and are consistent throughout all drawings and
specifications for the plant. The P&ID is the key to a coherent plant
design, and it’s a place to start when you create graphical displays in
LabVIEW.

MAX ETHYLENE
1 (MAX) (100PPM)
TOTAL C2'S
TO 2ND STAGE PCV
FT FI AC (0.4% WT) AC
COMPRESSOR 1329
1313 1313
PT
1328
PC
1328
2 PC
FILTER FILTER
1329
C2 C2=
PCV PT C3= C3=
1328 1329
LCV ETHANE C2
1319 (NEGLECT
ETHYLENE)
1 PROPYLENE ETHYLENE C2=
FT TI TE C3= AT
NEW 1314 1344 1344
FROM E-A3-63 10
K.O. FC V/L
TI TE
DRUM 1314 TE TC
SET 1345 1345 20
F-A3–201 21 1346 1346
LAG L
LC LT LT LC
X V
1318 1318 1319 1319
ORGANIC F- A3-200 NEW
TE TI FC FT
1347 1347 1317 1317
WATER STEAM
LCV
1318
50 PSIG
4
WATER LE LT
1320 1320 E - A3-200A E - A3-200B

F -A3-202 LT LC
1321 1321

TO DEETHANIZER
LCV BOTTOMS
1321

Loop Types:
FCV
1. Simple Indication Loop FC FT 1315
2. Simple Control Loop 1316 1315

3. Cascade Control Loop

4. Complex Loop
FC
1315
3

Figure 18.1 A piping and instrument diagram is the basis for a good process plant
design.
440 Chapter Eighteen

TABLE 18.1 Abbreviated List for the Generation of ISA Standard Instrument Tags
First letter Second letter
measured or readout or Succeeding letters
initiating variable output function (if required)
A Analysis Alarm Alarm
B Burner, combustion User’s choice User’s choice
C Conductivity Controller Controller
D Density/damper Differential
E Voltage (elect) Primary element
F Flow Ratio/bias Ratio/bias
G Gauging (dimensional) Glass (viewing device)
H Hand (manual) High
I Current (electrical) Indicate Indicate
J Power Scanner
K Time Control station
L Level Light Low
M Moisture/mass Middle/intermediate
N User’s choice User’s choice User’s choice
O User’s choice Orifice, restriction
P Pressure Point (test) connection
Q Quantity Totalize/quantity
R Radiation Record Record
S Speed/frequency Safety/switch Switch
T Temperature Transmitter Transmitter
U Multipoint/variable Multifunction Multifunction
V Vibration Valve, damper, louver Valve, damper, louver
W Weight Well
X Special Special Special
Y Interlock or state Relay/compute Relay/compute
Z Position, dimension Damper or louver drive

SOURCE: From S5.1, Instrumentation Symbols and Identification. Copyright 1984 by Instrument Society of
America. Reprinted by permission.

Tag names should follow ISA standard S5.1, a summary of which

is shown in Table 18.1. The beauty of this naming system is that it is
both meaningful and concise. With just a few characters, you can deter-
mine what the instrument controls or measures (for example, P for
pressure, T for temperature) as well as its function (for example, C for
controller, V for valve, T for transmitter). A numeric suffix is appended
as a hierarchical serial number. Most plants use a scheme where the
first digit is the zone or area, the second is the major vessel, and so
forth. Tag names appear everywhere—on the P&ID, instrument data
sheets, loop diagram, in your control system database, and, of course,
on a little metal tag riveted to the instrument. Since this is the only
national standard of its type, most process engineers are familiar with
this naming scheme—a good reason to consider its use.
When you write a LabVIEW program for process control, you should
use tag names everywhere. On the diagram, you can label wires,
 Process Control Applications 441

cluster elements, and frames of Case and Sequence structures that are
dedicated to processing a certain channel. On the panel, tag names make
a convenient, traceable, and unambiguous way to name various objects.
Here are some examples of common tag names with explanations:

PI-203 Pressure Indicator—a mechanical pressure gauge

LT-203 Level Transmitter—electronic instrument with 4–20-mA
output
TIC-203 Temperature Indicating Controller—a temperature control-
ler with readout
PSV-203 Pressure Safety Valve—a relief valve
FCV-203 Flow Control Valve—valve to adjust flow rate
ZSL-203 Position Switch, Low-Level—switch that closes when a
valve is closed

The little balloons all over the P&ID contain tag names for each
instrument. Some balloons have lines through or boxes around them
that convey information about the instrument’s location and the
method by which a readout may be obtained. Figure 18.2 shows some
of the more common symbols. The intent is to differentiate between a
field-mounted instrument, such as a valve or mechanical gauge, and
various remotely mounted electronic or computer displays.
Every pipe, connection, valve, actuator, transducer, and function in
the plant has an appropriate symbol, and these are also covered by ISA
standard S5.1. Figure 18.3 shows some examples that you would be
likely to see on the P&ID for any major industrial plant. Your system
may use specialized instruments that are not explicitly covered by the
standard. In that case, you are free to improvise while keeping with

HV
Discrete instrument, field-mounted
552

TIC Discrete instrument, in a location

312 normally accessible by operator

PIR Shared display or control, in a location

126 normally accessible by operator

I Computer function, in a location normally

175 accessible by operator

Figure 18.2 Function symbols, or balloons, used

to identify instruments on the P&ID.
442 Chapter Eighteen

Electrical control
signal
PY
552 HS
FIR
552
531
Instrument S 3-way solenoid valve
air supply Temperature
Pneumatic line gauge

Diaphragm- FT
IA operated valve
531 TI
530
VENT

Fails with this

path open Turbine flow
meter
FV FE
552 531

Figure 18.3 Some instrument symbols, showing valves, transmitters, and associated
connections. These are right out of the standards documents; your situation may
require some improvising.

the spirit of the standard. None of this is law, you know; it’s just there
to help. You can also use replicas of these symbols on some of your
LabVIEW screens to make it more understandable to the technicians
who build and operate the facility.

Other drawing and design standards. Another of our favorite ISA stan-
dards, S5.4, addresses instrument loop diagrams, which are the
control engineer’s electrical wiring diagram. An example appears in
Figure 18.4. The reasons for the format of a loop diagram become clear
once you have worked in a large plant environment where a signal may
pass through several junction boxes or terminal panels before finally
arriving at the computer or controller input. When a field technician
must install or troubleshoot such a system, having one (or only a few)
channels per page in a consistent format is most appreciated.
Notice that the tag name appears prominently in the title strip,
among other places. This is how the drawings are indexed, because the
tag name is the universal identifier. The loop diagram also tells you
where each item is located, which cables the signal runs through, instru-
ment specifications, calibration values (electrical as well as engineering
units), and computer database or controller setting information.
We’ve found this concise drawing format to be useful in many labora-
tory installations as well. It is easy to follow and easy to maintain.
There are commercial instrument database programs available that
contain built-in forms, drawing tools, and cross-referencing capability
 Process Control Applications 443

Field Process Area Building A Control Panel CP-4

Panel Rear Panel Front
FROM PUMP P-370
SW: MODEL NO.
SET POINT
TRANSMITTER: MODEL NO.
SPEC NO.
2'5

REV FSH 301-1

C 1-1-C
FSH 120VAC FY 301A-1
301 NO FSH 301-2
1-1-3 FY 2-3-C
120VAC
2 FAH
301A
I 2-2-C 2-3-2 3 301
112 2-2-5 FY 301A-2

FY 301B-2 (W)
FY 301B-1 (B)
ELEV. PLC: ANNUNCIATOR:

FSH 301(B)

120VAC
300'-0" MODEL NO. MODEL NO.
AS 20 PSIG SPEC NO. POINT POS.
FE 0-100"WC CABLE-30 28-1
301 JB
FE S 3-15 PSIG JB 30 200
301 O 3 7
FT-301 (B) TUBE BUNDLE 28 3-15 PSIG
4 8
FV-301 (B)
1
L2
28-2 S 3
L1
2
ELEV. AS 100 PSIG CONTROLLER:
SOLENOID: FY MODEL NO.
300'-0" 301B
MODEL NO. SPEC NO.
FV SPEC NO.
1
301 I S 12 21
VALVE: MODEL NO. O FIC
301
SPEC NO. AS 20 PSIG S REV

TO UNIT NO.3

NOTES:

1. FV-301 fully open at 3 psig and fully closed at 15 psig.

No. Date Revisions By Apr
2. FE-301 consist of 20 diameters upstream and 5 diameters downstream.
FRESH FEED FLOW CONTROL
REFERENCE DRAWINGS: TO UNIT NO. 3
WITH HIGH FLOW LIMITING.
P&ID XXXX Sheet 1 LOOP DIAGRAM
Instrument installation details XXX, YYY
Control board specification XXX JOB No. DRAWING No. REV.

Figure 18.4 An instrument loop diagram, the control engineer’s guide to wiring. (Reprinted by
permission. Copyright ©1991 by the Instrument Society of America. From S5.4,Instrument Loop
Diagrams.)

that make management of large process control projects. much easier.

Ask ISA for a catalog of process control products, if you’re interested.
Here are a few other ISA standards that can be valuable to you:

S5.5, Graphic Symbols for Process Displays

S51.1, Process Instrumentation Terminology
S50.1, Compatibility of Analog Signals for Electronic Industrial Pro-
cess Equipment

If you are involved with the design of a major facility, many other
national standards will come into play, such as the National Electri-
cal Code and the Uniform Mechanical Code. That’s why plants are
designed by multidisciplinary teams with many engineers and design-
ers who are well versed in their respective areas of expertise. With
these standards in hand and a few process control reference books, you
might well move beyond the level of the mere LabVIEW hacker and
into the realm of the registered professional control engineer.
444 Chapter Eighteen

Control = manipulating outputs

Controlling a process implies a need for some kind of output signal
from the control system. (At last! Something that uses all those ana-
log and digital output modules that somebody sold you!) The simplest
control mode, manual control, relies on the operator to turn a knob
or throw a switch to manipulate the process through some actuator.
Programming a manual control system is about as simple as things
get. The other control mode, automatic control, requires a hardware
or software machine to make the adjustments. What both modes have
in common (besides output hardware) is the use of feedback. In man-
ual control, the operators see that the temperature is too high, and
they turn down the heater. In automatic control, the controller makes
a measurement (the process variable), compares it with the desired
value, or setpoint, and adjusts the output, or manipulated variable,
accordingly.
The output of the controller may be either digital or analog in nature.
Digital-output controllers are also known as on/off controllers or
bang-bang controllers. An example is the thermostat on your home’s
furnace. On/off controllers are generally the simplest and cheapest con-
trol technique, but they may be a bit imprecise—the process variable
tends to cycle above and below the setpoint. Analog controllers, on the
other hand, are proportional in nature, adjusting the output in infinitesi-
mally small steps to minimize the error. They are powerful and versatile,
but generally more complex and expensive than on/off controllers.
The most common types of automatic controllers in use today rely on
the well-known proportional-integral-derivative (PID) family of
algorithms. The simplest form of the algorithm is a pure proportional
(P) controller (Figure 18.5A). An error signal is derived by subtract-
ing the process variable from the setpoint. Error is then multiplied by
a proportional gain factor, and the result drives the output, or manip-
ulated variable. If the process reacts by “going in the right direc-
tion,” the process variable will tend toward the setpoint—the result of
negative feedback. If the process goes the wrong direction, positive
feedback results, followed by undesirable oscillation or saturation. By
raising the proportional gain, the manipulated variable is more strongly
driven, thus reducing the error more quickly. There is a practical limit
to gain, however, due to delay (lag) in the process that will eventually
cause a destabilizing phase shift in the control loop—the loop ends up
responding to old information. Therefore, some residual error is always
present with a proportional controller, but the error magnitude may
be sufficiently small for your requirements that a simple proportional
controller is adequate.
One way to eliminate this error is to mathematically integrate or
reset the error over time. Combining these two control techniques
 Process Control Applications 445

A) Proportional controller Proportional

+
Setpoint Σ Kp Controller
output
–

Process variable

1 Integral
B) Proportional-Integral
(PI) controller Ti s
Proportional
+
+ error +
Setpoint Σ Σ Kp Controller
output
–

Process variable

C) Proportional-Integral- 1 Integral
Derivative (PID) controller with Ti s
derivative on measurement Proportional
+ + +
error Controller
Setpoint Σ Σ Kp
output
– –
Td s
Derivative
Process variable
Figure 18.5 Signal flow diagrams for proportional, integral, and derivative algorithms
are the basis for much of today’s practical feedback control. These are just a few exam-
ples of P/PI/PID configuration; there are many more in actual use.

results in the proportional-integral (PI) controller (Figure 18.5B).

With the PI algorithm, the integral reduces the error by spreading the
process reaction over time, thus giving it time to respond to controller
commands. Like any control algorithm, too much of a good thing may
result in substandard or unstable performance. Also, the integral can
wind up (where the value of the integral grows to a very large value)
when the process is sufficiently out of control for a long period of time
or when the controlled device does not respond as expected. This can
result in very long recovery times, so most controllers include an anti-
reset windup feature to limit integral action. PI controllers are the
most widely used in general applications.
When a process is subject to sudden upsets or large changes in set-
point, the controller may not respond quickly enough. In such cases,
a derivative, or rate, term may be added (Figure 18.5C). Taking the
derivative of the error (a common technique) effectively increases the
controller’s gain during periods when the process variable is chang-
ing, forcing a quick correcting response. Unfortunately, the derivative
is a kind of high-pass filter that emphasizes the controller’s response
to noise. Therefore, the full PID algorithm can only be used when the
446 Chapter Eighteen

signal has little noise or where suitable filtering or limiting has been
applied.
National Instruments offers a set of PID algorithms available in the
PID Control Toolkit. You can use them to build all kinds of control
schemes; usage will be discussed later in this chapter. PID control can
also be accomplished through the use of external “smart” controllers
and modules. Greg Shinskey’s excellent book Process Control Systems
(1988) discusses the application, design, and tuning of industrial con-
trollers from a practical point of view. He’s our kinda guy.
There are many alternatives to the common PID algorithm so often
used in industrial control. For instance, there are algorithms based on
state variable analysis which rely on a fairly accurate model of the pro-
cess to obtain an optimal control algorithm. Adaptive controllers, which
may or may not be based on a PID algorithm, modify the actions of the
controller in response to changes in the characteristics of the process.
Predictive controllers attempt to predict the trajectory of a process to
minimize overshoot in controller response. Modern fuzzy logic control-
lers are also available in LabVIEW in the PID Toolkit. That package
includes a membership function editor, a rule-base editor, and an infer-
ence engine that you can incorporate into your system. If you have
experience in control theory, there are few limitations to what you can
accomplish in the graphical programming environment. We encourage
you to develop advanced control VIs and make them available to the
rest of us. You might even make some money!
So far, we have been looking at continuous control concepts that
apply to steady-state processes where feedback is applicable. There are
other situations. Sequential control applies where discrete, ordered
events occur over a period of time. Valves that open in a certain order,
parts pickers, robots, and conveyors are processes that are sequential
in nature. Batch processes may be sequential at start-up and shut-
down, but operate in steady state throughout the middle of an opera-
tion. The difficulty with batch operations is that the continuous control
algorithms need to be modified or compromised in some way to handle
the transient conditions during start-up, shutdown, and process upsets.
A special form of batch process, called a recipe operation, uses some
form of specification entry to determine the sequence of events and
steady-state setpoints for each batch. Typical recipe processes are
paint and fuel formulation (and making cookies!), where special blends
of ingredients and processing conditions are required, depending on
the final product.
Early sequential control systems used relay logic, where electrome-
chanical switching devices were combined in such a way as to imple-
ment boolean logic circuits. Other elements such as timers and stepper
switches were added to facilitate time-dependent operations. System
 Process Control Applications 447

inputs were switch contacts (manual or machine-actuated), and out-

puts would drive a variety of power control devices such as contactors.
Most modern sequential control systems are based on programma-
ble logic controllers (PLCs) which are specialized industrial com-
puters with versatile and nearly bulletproof I/O interface hardware.
Millions of PLCs are in service in all industries, worldwide, and for
good reasons: They are compact, cost-effective, reliable, and easy to pro-
gram. It’s much easier to modify a PLC’s program than it is to rip out
dozens of hardwired relays mounted in a rack. PLCs can be networked,
and LabVIEW turns out to be a fine man-machine interface.

Process signals
The signals you will encounter in most process control situations are
low-frequency or dc analog signals and digital on/off signals, both
inputs and outputs. Table 18.2 lists some of the more common ones.
In a laboratory situation, this list would be augmented with lots of
special analytical instruments, making your control system heavy on
data acquisition needs. Actually, most control systems end up that way
because it takes lots of information to accurately control a process.
Industry likes to differentiate between transducers and trans-
mitters. In process control jargon, the simple transducer (like a ther-
mocouple) is called a primary element. The signal conditioner that
connects to a primary element is called a transmitter.
In the United States, the most common analog transmitter and con-
troller signals are 4–20-mA current loops, followed by a variety of volt-
age signals including 1–5, 0–5, and 0–10 V. Current loops are preferred
because they are resistant to ground referencing problems and voltage
drops. Most transmitters have a maximum bandwidth of a few hertz,
and some offer an adjustable time constant which you can use to opti-
mize high-frequency noise rejection. To interface 4–20-mA signals to an
ordinary voltage-sensing input, you will generally add a 250-Ω preci-
sion resistor in parallel with the analog input. (If you’re using National
Instruments’ signal conditioning—particularly SCXI—order its 250-Ω
terminating resistor kits.) The resulting voltage is then 1–5 V. When

TABLE 18.2 Typical Process Control Signals and Their Usage

Analog inputs Digital inputs Analog outputs Digital outputs

Pressure Pressure switch Valve positioner Valve open/close
Temperature Temperature switch Motor speed Motor on/off
Flow rate Flow switch Controller setpoint Indicator lamps
Level Level switch Heater power
Power or energy Position switch
448 Chapter Eighteen

you write your data acquisition program, remember to subtract out the
1-V offset before scaling to engineering units.
On/off signals are generally 24 V dc, 24 V ac, or 120 V ac. We pre-
fer to use low-voltage signals because they are safer for personnel. In
areas where potentially flammable dust or vapors may be present, the
National Electrical Code requires you to eliminate sources of ignition.
Low-voltage signals can help you meet these requirements as well.
The world of process control is going through a major change with
emerging digital field bus standards. These busses are designed to
replace the 4–20-mA analog signal connections that have historically
connected the field device to the distributed control systems with
a digital bus that interconnects several field devices. In addition to
using multidrop digital communications networks, these busses use
very intelligent field devices. They are designed for device interoper-
ability, which means that any device can understand data supplied by
any other device. Control applications are distributed across devices,
each with function blocks that execute a given control algorithm. For
instance, one control algorithm can orchestrate a pressure transmitter,
a dedicated feedback controller, and a valve actuator in a field-based
control loop.
The first of the digital communication protocols, HART, was created
by Rosemount and supported by hundreds of manufacturers. It adds
a high-frequency carrier (such as that used with 1200-baud modems)
which rides on top of the usual 4–20-mA current loop signal. Up to
16 transmitters and/or controllers can reside on one HART bus, which
is simply a twisted pair of wires. HART allows you to exchange mes-
sages with smart field devices, including data such as calibration
information and the values of multiple outputs—things that are quite
impossible with ordinary analog signals.
Perhaps the most important new standard is Foundation Field-
bus (from the Fieldbus Foundation, naturally), which has interna-
tional industry backing. The ISA is also developing a new standard,
SP-50, which will ultimately align with Foundation Fieldbus and other
standards. Because the field devices on these field bus networks are so
intelligent, and because the variables shared by devices are completely
defined, the systems integration effort is greatly reduced. It is thus
much more practical to implement a supervisory system with LabVIEW
and one or more physical connections to the field bus network. You
no longer need drivers for each specific field device. Instead, you can
use drivers for classes of field devices, such as pressure transmitters
or valves. National Instruments is an active player in the Foundation
Fieldbus development process, offering several plug-in boards and NI-
FBUS host software. At this writing, the standard is well established
 Process Control Applications 449

and is in wide use. As development continues, look for extensive

LabVIEW support for Foundation Fieldbus—it’s in the National Instru-
ments catalog.

Control system architectures

Hopefully, you will be designing your LabVIEW process control appli-
cation along with the plant you wish to control. Choices of transducers
and actuators, as well as the overall process topology, drastically affect
the possibilities for adequate control. The process control engineer is
responsible for evaluating the chemistry and physics of the plant with
an eye toward controllability. The best software in the world can’t over-
come the actions of improperly specified valves, can’t correct unusable
data from the wrong type of flowmeter, nor can it control wildly unsta-
ble chemical reactions. You and your design team need to be on top of
these issues during all phases of the project.
You will need to choose an appropriate control system architecture.
There are many variables and many personal and corporate biases
that point in different directions. Industrial control journals are filled
with case histories and design methodologies that represent the cumu-
lative knowledge of thousands of plant designs. For the novice, these
journals are a source of inspiration, as are consultants and, of course,
the major engineering firms. At the end of this section, we’ll list some
possible design parameters you might use to develop a checklist that
leads to the right system.

Distributed control system (DCS).For the largest installations, distrib-

uted control systems have held the market for decades, though that
position is being eroded by the improved capability of smaller, more
open systems including PCs. Important characteristics of a DCS are as
follows (see Figure 18.6):
■ A hierarchical set of intelligent nodes, or controllers, ranging from
smart I/O scanners and controllers to man-machine interface sta-
tions, on up to corporate-level management information systems
■ Loop controllers physically close to the process being controlled
■ Heavy reliance on local-area networks (usually redundant, for
reliability)
■ A globally shared, real-time database scheme where all nodes can
freely share information
■ Multiple levels of access security through passwords and communi-
cations restrictions
450 Chapter Eighteen

Data
Archive
1.00
√
0.80 X

0.60
0.40
0.20
0.00

-4
00
-2
00
0
20
0
40
0

Trend Process Report

Displays Mimic Display Generator
CS
IX
CCSS 11 CS
IX
IX IX
04
111 11
00441 04

MMI MMI Communications

Controller Controller Module

Redundant Network

CS CS
IX IX
CS 11 11
IX 04 04
11
04

I/O Communications Communications

Controller Module Module

SCXI- SCXI
1001 -1001

SCXI SCXI

1140 1140

SCXI SCXI

1140 1140

SCXI SCXI

Proprietary PLC Corporate

1140 1140

SCXI SCXI

1140 1140

SCXI SCXI

1140 1140

SCXI SCXI

1140 1140

SCXI SCXI

1140 1140

SCXI SCXI

1140 1140

SCXI SCXI SCXI SCXI

MAINFRAM MAINFRAM
1140 1140

I/O Interface
E

Information
SCXI SCXI

1140 1140

SCXI SCXI

1140 1140

System

Figure 18.6 A distributed control system, or DCS, encompasses many nodes communicat-
ing over networks and many I/O points.

■ Integration of several different kinds of computers and I/O hardware

■ Dozens, and sometimes hundreds, of nodes
■ Many thousands of I/O points
■ Very high cost, often in the millions of dollars

Older DCSs used proprietary operating systems and custom hard-

ware, but new designs are moving more toward open, or public, stan-
dards in I/O hardware and operating systems such as Windows and
Linux. All DCS vendors work closely with the customer because the
system specification, configuration, and start-up phases are quite com-
plex. Turnkey systems (where the system vendor does the lion’s share
of the work) are the rule rather than the exception.
 Process Control Applications 451

A DCS may incorporate many kinds of computers and I/O subsys-

tems, integrated into one (hopefully) harmonious package. In principle,
a DCS could be designed with multiple systems running LabVIEW, but
that may be stretching things a bit. At this time, it just isn’t very well
suited to this kind of massive application. Some major control system
manufacturers are using LabVIEW at this time, but only in certain
subsystems as part of an overall proprietary DCS. But given a few
years, LabVIEW—particularly with DSC—may well penetrate these
high-end applications.

SCADA in the middle. The middle ground in process control is han-

dled by a broad range of systems collectively known as Supervisory
Control and Data Acquisition (SCADA) systems (Boyer 1993). The
name pretty well describes what they do. At the bottom of the architec-
ture are the data acquisition features—measurements for the purpose
of feedback and operator information. Next up the line are the control
features, including manual commands and continuous and sequential
automatic algorithms. At the top level are the supervisory aspects—the
user interface, trending, report generation, and so forth. A key feature
of SCADA is distributed I/O, including remote sensing and a heavy reli-
ance on communications or networking with remote terminal units.
In contrast to a large DCS, SCADA systems are chosen for simpler
systems where the total number of I/O points is unlikely to exceed a
thousand, but with wide distribution. An example would be an oil or
gas pipeline. You probably would not find control engineers at a large
pulp plant who claim that their facility is run by a SCADA system. A
small municipal waterworks is a more likely location. But the defining
bounds of SCADA are becoming ever more fuzzy, thanks to the power
and popularity of the personal computer and modern software.

Enter the personal computer. At the other end of the scale from a mega-
DCS is the personal computer, which has made an incredible impact on
plant automation. The DCS world has provided us with software and
hardware integration techniques that have successfully migrated to
desktop machines. A host of manufacturers now offer ready-to-run PC-
based process control and SCADA packages with most of the features
of their larger cousins, but with a much lower price tag and a level of
complexity that’s almost . . . human in scale. LabVIEW with DSC fits
into this category.
A wide range of control problems can be solved by these cost-effective
small systems. In the simplest cases, all you need is a PXI system
running LabVIEW with plug-in I/O boards or maybe some outboard
I/O interface hardware. You can implement all the classical control
schemes— continuous, sequential, batch, and recipe—with the functions
452 Chapter Eighteen

built into LabVIEW and have a good user interface on top of it all. Such
a system is easy to maintain because there is only one programming
language, and only rarely would you need the services of a consultant
or systems house to complete your project. The information in this
chapter can lead you to a realistic solution for these applications.
There are some limitations with any stand-alone PC-based pro-
cess control system. Because one machine is responsible for servicing
real-time control algorithms as well as the user interface with all its
graphics, there can be problems with real-time response. If you need
millisecond response, consider using outboard smart controllers (see
the following section) or LabVIEW RT. The I/O point count is another
factor to consider. Piling on 3000 analog channels is likely to bring
your machine to its knees; you must consider some kind of distributed
processing scheme.
LabVIEW, like the specialized process control packages, permits you
to connect several PCs in a network, much like a DCS. You can config-
ure your system in such a way that the real-time tasks are assumed
by dedicated PCs that serve as I/O control processors, while other PCs
serve as the man-machine interfaces, data recorders, and so forth. All
the machines run LabVIEW and communicate via a local-area network
using supported protocols such TCP/IP. The result is an expandable
system with the distributed power of a DCS, but at a scale that you
(and perhaps a small staff) can create and manage yourself. The prob-
lem is programming.

PCs with smart controllers. Another way to accommodate medium- and

small-scale plants is to use your PCs with smart external controllers,
such as PLCs and single-loop PID controllers, to off-load critical real-time
tasks. This approach has some advantages. Dedicated, special-purpose
controllers are famous for their reliability, more so than the general-
purpose computer on which you run LabVIEW. They use embedded
microprocessors with small, efficient executive programs (like miniature
operating systems) that run your control programs at very high speeds
with no interference from file systems, graphics, and other overhead.
PLCs can be programmed to perform almost any kind of continuous
or sequential control, and larger models can be configured to handle
thousands of I/O points and control loops, especially when networked.
However, the PLC lacks a user interface and a file system, among
other amenities—just the kinds of things that your LabVIEW pro-
gram can provide. LabVIEW makes an ideal man-machine interface,
data recorder, alarm annunciator, and so forth. It’s a really synergistic
match. About the only disadvantage of this approach is that you have
two programming environments to worry about: LabVIEW plus the
PLC’s programming package.
 Process Control Applications 453

LabVIEW RT enhances the real-time performance of LabVIEW

for time-critical applications such as feedback control. The big advan-
tage here is that you avoid having to learn any additional languages,
such as those required for programming a PLC. You configure one or
more LabVIEW RT systems that reside on a TCP/IP network and then
exchange process information between those RT systems (known as
peer-to-peer communications) or with the MMI machine. See Chap-
ter 15, “LabVIEW RT,” for more information on LabVIEW RT.
Single-loop controllers (SLCs) do an admirable job of PID, on/off,
or fuzzy logic control for industrial applications. They are generally
easy to configure, very reliable, and many offer advanced features such
as recipe generation and automatic tuning. Like PLCs, however, they
lack the ability to record data, and it may not be feasible to locate the
SLC close to the operator’s preferred working location. By using a com-
munications link (usually RS-232), LabVIEW can integrate SLCs into
a distributed process control system.

Choose your system. In summary, here are some important control

system design parameters that might lead you to a first-cut decision
as to whether LabVIEW (probably with DSC) is a practical choice for
your application.

Number of I/O points and number of network nodes. What is the

overall scale of your system? Up to a few hundred I/O points, even
a stand-alone PC running LabVIEW has a good chance of doing the
job. With somewhat higher point counts, high-speed requirements, or
a need to have the I/O distributed over a wider area, PLCs offer lots
of advantages, and LabVIEW makes a nice MMI. If you are tackling
an entire plant, then LabVIEW might be suited to some local control
applications, but a DCS or a large, integrated SCADA package may
be a better choice. LabVIEW with DSC, for example, is probably suit-
able for installations with several thousand points.
Planned expansion. Continuing with the first topic, remember to
include any plans for future expansion. A lab-scale pilot plant is ideal
for LabVIEW, but its plant-scale implementation may be a whole
’nother story. A poorly chosen control system may result from under-
estimation of the overall needs. Beware.
Integration with other systems. Needs for corporate-level computing
will influence your choice of control computers and networks. For
instance, if you live in a UNIX/X Window world, then a SPARC sta-
tion or Linux box running LabVIEW might be a good platform.
Requirements for handling unique, nonstandard instruments.
LabVIEW has tremendous advantages over commercial process
454 Chapter Eighteen

control software products when it comes to handling unusual instru-

ment interfaces. Because you can easily write your own drivers in G,
these special situations are a nonproblem. Similarly, DSC permits
you to add ordinary LabVIEW drivers, running in parallel with the
native Process Control server-based drivers.
Ease of configuration and modification. What plans do you have
for modifying your process control system? Some system architec-
tures and software products are rather inflexible, requiring many
hours of reconfiguration for seemingly simple changes. Adding a new
instrument interface to a large DCS may seemingly require an act
of Congress, as many users have discovered. Meanwhile, LabVIEW
remains a flexible solution, so long as you design your application
with some forethought.
Overall reliability. It’s hard to compare computers, operating sys-
tems, and software when it comes to reliability. Suffice it to say that
simple is better, and that’s why many process engineers prefer PLCs
and SLCs: They are simple and robust. The MMI software is, in a
sense, of secondary importance in these systems because the process
can continue to operate (if desired) even while the MMI is offline.
We’ve found that the system architecture is often more important
than the brand of computer or software. Using redundant hard-
ware is the ultimate solution where reliability is paramount. Many
PLC and DCS vendors offer redundancy. Additionally, you should
consider adding fail-safe mechanisms, such as watchdog timers, to
insure safety and prevent major hazards and losses when things go
awry. Software bugs, hardware failures, and system crashes can be
very costly in process control.*
Existing company standards and personal biases. Everyone from
the CEO to the janitor will have opinions regarding your choice of
hardware and software. For the small, experimental systems that
we work on, LabVIEW has many advantages over other products. As
such, it has become very popular—a definite positive bias—and with
good reason. However, we would not use it to control a new hundred
million-dollar facility, no matter how biased we might be.

*A friend of ours at a major process control system manufacturer reports that Bill
Gates and company have effectively lowered our expectations. Prior to the great rise in
the popularity of Windows in process control, systems were based on VAX/VMS, UNIX,
and very expensive custom software. If a system crashed in a plant operation, our friend’s
company would get a nasty phone call saying, “This had better never happen again!”
Now, he finds that Windows and the complex application software come crashing down
so often that plant operators hardly bat an eye; they just reboot. It’s a sad fact, and we
wonder where this is all going.
 Process Control Applications 455

Cost. The bottom line is always cost. One cost factor is the need for
system integration services and consultants, which is minimal with
LabVIEW, but mandatory with a DCS. Another factor is the num-
ber of computers or smart controllers that you might need: A stand-
alone PC is probably cheaper than a PC plus a PLC. Also, consider
the long-term costs such as operator training, software upgrades and
modifications, and maintenance. Plan to evaluate several process
control packages— as well as LabVIEW with DSC—before commit-
ting your resources.

Working with Smart Controllers

Adding smart controllers to your process control system is smart engi-
neering. PLCs, SLCs, and other programmable devices are practical
solutions to your real-world control problems. They are completely opti-
mized to perform real-time continuous or sequential control with the
utmost in reliability while managing to be flexible enough to adapt to
many different situations. Take advantage of them whenever possible.

Programmable logic controllers (PLCs). A programmable logic control-

ler (PLC) is an industrial computer with a CPU, nonvolatile program
memory, and dedicated I/O interface hardware (Bryan and Bryan
1988). Conceived by General Motors Corporation in 1968 as a replace-
ment for hardwired relay logic, PLCs have grown dramatically in their
software and hardware capabilities and now offer incredible power and
flexibility to the control engineer. Major features of a modern PLC are
as follows:
■ Modular construction
■ Very high reliability (redundancy available for many models)
■ Versatile and robust I/O interface hardware
■ Wide range of control features included, such as logic, timing, and
analog control algorithms
■ Fast, deterministic execution of programs—a true real-time system
■ Communications supported with host computers and other PLCs
■ Many programming packages available
■ Low cost

You can buy PLCs with a wide range of I/O capacity, program storage
capacity, and CPU speed to suit your particular project. The simplest
PLCs, also called microcontrollers, replace a modest number of relays
and timers, do only discrete (on/off, or boolean) logic operations, and
456 Chapter Eighteen

HS-203 LAH-203 ZSL-301 FV-301

HS-303 Normally open Normally closed

contact contact

FV-301 FV-302
TIMER1
15 SEC
Coil (output)

Figure 18.7 Ladder logic is a language for PLCs.

Each item in the diagram corresponds to an emu-
lated piece of hardware—contacts (switches), timers,
relay coils, and so forth.

support up to about 32 I/O points. Midrange PLCs add analog I/O and
communications features and support up to about 1024 I/O points. The
largest PLCs support several thousand I/O points and use the latest
microprocessors for very high performance (with a proportionally high
cost, as you might expect).
You usually program a PLC with a PC running a special-purpose
application. Ladder logic (Figure 18.7) is the most common language
for programming in the United States, though a different concept,
called GRAFCET, is more commonly used in Europe. Some program-
ming applications also support BASIC, Pascal, or C, either intermixed
with ladder logic or as a pure high-level language just as you would use
with any computer system.
There are about 100 PLC manufacturers in the world today. Some of
the major brands are currently supported by OLE for Process Control
(OPC) and LabVIEW drivers. Check the National Instruments Web
site for new releases.
The ISA offers some PLC-related training you may be interested in.
Course number T420, Fundamentals of Programmable Controllers,
gives you a good overview of the concepts required to use PLCs. The
textbook for the class, Programmable Controllers: Theory and Imple-
mentation (Bryan and Bryan 1988), is very good even if you can’t attend
the class. Another course, T425, Programmable Controller Applica-
tions, gets into practical aspects of system design, hardware selection,
and programming in real situations.

Advantages of PLCs. Programmable controllers have become a favorite

tool in the control industry because of their simplicity, low cost, rug-
ged hardware, and reliable real-time performance. These are clear-cut
advantages that will make them candidates for your project as well.
 Process Control Applications 457

Interfacing couldn’t be easier. Input and output modules are avail-

able for direct connection to 240-V ac and lower ac and dc voltages and
a wide range of analog signals. Compare this with most interface sys-
tems which require extra isolation devices or adapters, especially for
high voltages and currents.
Reliability may be the main reason for choosing a PLC over other
solutions. Both the hardware and the operating software in a PLC are
simple in nature, making it easier to prove correctness of design and
simultaneously lowering the number of possible failure points. Con-
trast this with a general-purpose, desktop computer. How often does
your computer hang up or crash? If your process demands predictable
performance, day in and day out, better consider using a PLC. See
Figure 18.8.

PLC communications and register access. Manufacturers of PLCs have

devised many communications techniques and pseudostandard proto-
cols. For instance, Modicon has its proprietary Modbus protocol, while
Allen-Bradley has its Data Highway Plus—and the two are completely
incompatible. The good news is, through an adapter module, any com-
puter with RS-232 or Ethernet capability can communicate with these
various data highways, assuming that you have the right driver soft-
ware. All midrange and high-end PLCs have the ability to communicate

Figure 18.8 PLCs like this Siemens model are popular in process control. SinecVIEW
is a LabVIEW driver package that communicates with Siemens PLCs via a serial
interface. (Photo courtesy of National Instruments and CITvzw Belgium.)
458 Chapter Eighteen

peer to peer, that is, between individual PLCs without host interven-
tion. A variety of schemes exist for host computers as well, serving as
masters or slaves on the network.
Once a PLC has been programmed, your LabVIEW program is free
to read data from and write data to the PLC’s registers. A register may
represent a boolean (either zero or one), a set of booleans (perhaps 8 or
16 in one register), an ASCII character, or an integer or floating-point
number. Register access is performed at a surprisingly low level on
most PLCs. Instead of sending a message like you would with a GPIB
instrument (“start sequence 1”), you poke a value into a register that
your ladder logic program interprets as a command. For instance, the
ladder logic might be written such that a 1 in register number 10035
is interpreted as a closed contact in an interlock string that triggers a
sequential operation.
There are a few things to watch out for when you program your PLC
and team it with LabVIEW or any other host computer. First, watch
out for conflicts when writing to registers. If the ladder logic and your
LabVIEW program both write to a register, you have an obvious con-
flict. Instead, all registers should be one-way; that is, either the PLC
writes to them or your LabVIEW program does. Second, you may want
to implement a watchdog program on the PLC that responds to fail-
ures of the host computer. This is especially important when a host
failure might leave the process in a dangerous condition. A watchdog is
a timer that the host must hit, or reset, periodically. If the timer runs to
its limit, the PLC takes a preprogrammed action. For instance, the PLC
may close important valves or reset the host computer in an attempt to
reawaken it from a locked-up state.
PLC communications is another area where the Datalogging and
Supervisory Control (DSC) module driver model has advantages over
the much simpler LabVIEW driver model. With DSC, National Instru-
ments supplies a large set of OPC drivers that connect to a wide variety
of industrial I/O devices, including PLCs. Also, many hardware manu-
facturers supply OPC-compliant drivers that you can install and use
with DSC. OPC is a layered, plug-and-play communications interface
standard specified by Microsoft for use with all versions of Windows.
Essentially, it provides a high-performance client-server relationship
between field devices and user applications, including LabVIEW, Excel,
Visual Basic, and many other packages. There’s even an organization—
the OPC Foundation—that promotes the OPC standards. Visit them at
www.opcfoundation.org.

An OPC-based PLC driver example. When properly designed, OPC drivers

are easy to configure and use, and add only modest overhead to the com-
munications process. Included with DSC is the Industrial Automation
 Process Control Applications 459

Servers CD with hundreds of OPC drivers that you install into your
operating system and configure with a utility program that is external
to LabVIEW. This basic configuration step tells the driver what com-
munications link you’re using (serial, Ethernet, etc.) and how many
and what kind of I/O modules you have installed. Once this low-level
step is performed, you work entirely within the features of the DSC
tool set to configure and use individual tags (I/O points). The Tag Con-
figuration Editor is the place where you create and manage tags.
For a given tag, you assign a tag name, a PLC hardware address, scale
factors, alarm limits, and trending specifications. For very large data-
bases, you may find it easier to manage the information in a spread-
sheet or a more sophisticated database program. In that case, the Tag
Configuration Editor allows you to export and import the entire tag
database as ASCII text. Once the tag database is loaded, the editor
does a consistency check to verify that the chosen OPC server recog-
nizes the hardware addresses that you’ve chosen.
When you install DSC, LabVIEW gains a large number of control and
function subpalettes. Among the functions are VIs to read, write, and
access properties of tags—and it’s really easy to do. Figure 18.9 shows
a few of the basic VIs and how they’re used for simple read/write activi-
ties. In most cases, you just supply a tag name and wire up the input
or output value. The database and OPC drivers take care of the rest.
You can also use G Wizards, which create blocks of G code on the dia-
gram. A pop-up item available on every control and indicator brings
up a dialog through which you assign a tag name and an action. For
instance, if you want an indicator to display the value of an analog tag,
a small While Loop with the Read Tag VI will appear on the diagram,
already wired to the indicator. A large number of options are available
such as alarm actions (color changes and blinking) as well as the fre-
quency of update.

Figure 18.9 Here are a few of the VIs included with

DSC that allow you to read and write tag values that
reside in the database
460 Chapter Eighteen

The OPC-based database has several features that improve perfor-

mance. When you define a tag, you assign a scan interval. That’s the
interval at which the OPC driver goes out and interrogates the hard-
ware for an input tag. The rest of the time, there is no I/O activity,
thus reducing overhead for both your host computer and the PLC or
other I/O hardware. VIs that read tag values, such as Read Tag, can
be configured to sleep until a new value is available. For output tags,
you can configure tags to send data to the hardware only when there’s
a change in value. Again, this drastically reduces I/O message activity
and is a key to the excellent performance that we’ve seen in systems
using DSC.

LabVIEW PLC driver example: HighwayVIEW. Of the PLC driver pack-

ages currently available, HighwayVIEW, from SEG, was the first. It
supports Allen-Bradley models PLC-2, PLC-3, and PLC-5 and accepts
all data types for both inputs and outputs. Communications are han-
dled by a program that is installed in your computer’s operating sys-
tem (a Control Panel device on the Macintosh; a DLL under Windows).
This communications handler is particularly robust, offering automatic
retries if an error occurs. Figure 18.10 shows a simple example of High-
wayVIEW in action as a simple on/off limit controller application. An

Figure 18.10 A simple example using HighwayVIEW. An

array of floating-point values is read from the PLC, and
one is compared with a setpoint. If the tank level is too
high, a pump is turned on by setting a bit within a PLC
register to 1.
 Process Control Applications 461

array of floating-point values is read by the PLC-5 Read Float VI. If the
tank level is greater than the setpoint, a drain pump is turned on by
setting a bit with the PLC-5 Write VI.

Single-loop controllers (SLCs)

Single-loop controllers are an old favorite of process engineers. Years
ago, they were pneumatically operated, using springs, levers, bellows,
and orifices to implement an accurate PID algorithm. To this day,
there are still many installations that are entirely pneumatic, mostly
where an intrinsically safe control system is required. Modern SLCs
are compact, microprocessor-based units that mount in a control panel.
Some models can control more than one loop. They are compatible with
common input signals including thermocouples, RTDs, currents, and
voltages, with various ranges. The output may be an analog voltage or
current or an on/off digital signal controlled by a relay or triac. Other
features are alarms with contact closure outputs and ramp-and-soak
recipe operation. (See Figure 18.11.)
The front panel of a typical SLC consists of an array of buttons and
one or more digital displays. They are fairly easy to understand and
use, though some operators are frustrated by the VCR programming
syndrome where there are too many obscure functions and too many
menu choices. This is where LabVIEW can enhance the usefulness
of an SLC by simplifying the presentation of controls and indicators.
Low-level setup information can be hidden from the user while being
properly maintained by your program.

Your Measuring
Sensor or Transmitter Monitroller I
Power Supply Board
Thermocouple
Front Panel
System Power,
Display Board Terminal
Block 120 VAC
RTD
Connections

100Ω Signal Conversion and

Processing Board Control Output Card 1
Strain, Pressure Relay, Triac
+ Signalizer™ Card or Current-Loop
P
– Valve,
Rate �eat/Cool
Control Output Card 2 Element,
Analog to On-Off Pump
Signal Relay, Triac
Current Digital or PID
Conditioning Conversion or Current-Loop
Algorithms
4 to 20 mA
Voltage Display Serial I/O Card
Formatting Serial RS232C
V 1 to 5 V
or Isolated
AC Voltage/Current Analog System Processing, 20 mA Loop
Timing, Control, Printer,
Output
and Status Signals Computer,
BCD Output Card Terminal,
AC RMS includes EAROM
Isolated, Data Logger
3-State BCD
DC Voltage/Current
Control Signal
+
– Analog Output Recorder,
Remote Display

Figure 18.11 Block diagram of a typical single-loop controller. (Courtesy Analogic.)

462 Chapter Eighteen

All of the more sophisticated controllers offer communications

options. Most use RS-232 or RS-422/485 serial, but at least one fea-
tures GPIB (Research, Incorporated, with its Micristar controller). The
communications protocols all vary widely, generally following no stan-
dard at all. One exception is a line of controllers from Anafaze that use
the Allen-Bradley Data Highway protocol, the same one used by Allen-
Bradley PLCs. Anafaze controllers are supported by AnaVIEW from
SEG. As a bonus, some Anafaze models can handle up to 16 PID loops in
one compact controller; they are more appropriately named multiple-
loop controllers. A few other SLCs have LabVIEW drivers, such as
the Eurotherm 808/847. (See Figure 18.12.) With a data highway, GPIB,
or RS-422 party line, a multidrop network of controllers is feasible,
permitting many stations to communicate with your LabVIEW system
while consuming only one communications port.
The easiest way to control an SLC is with a good driver. A few control-
lers are available on the web. (e.g., Eurotherm), and a few are supported
by commercial software vendors, as in the case of Anafaze. Consider
the availability of drivers when choosing a controller; otherwise, you
will have to write your own. A good driver package will permit you to
read and write most any function in the controller.

Setpoint Present Temp 0.00 °C

Serial port Group ID
100 100.0
0 0
80 80.0

baud rate Unit ID 60.0

60
9600 0
40 40.0

20 20.0

STOP HIGH ALARM LOW ALARM 0 0.0

Group & Unit Read temp Update setpoint Read alarm status

Serial port Present Temp LOW ALARM

True
Low Alarm
High Alarm
PV HIGH ALARM

Setpoint SL STOP
baud rate

millisecond multiple 500 status

Figure 18.12 This example uses the Eurotherm 808/847 single-loop controller driver to update the
setpoint and read the process variable and alarm status.
 Process Control Applications 463

We really recommend SLCs when you have a modest number of control

loops. Their high reliability and simplicity, combined with the power and
flexibility of a LabVIEW MMI, makes a fine process control system.

Other smart I/O subsystems

There are many other ways to distribute the intelligence in your pro-
cess control system besides PLCs and SLCs. All you really need is a
remote or distributed I/O subsystem with local intelligence and com-
munications that can execute either sequential or continuous control
algorithms.
Multifunction scanners are more often used as simple data acquisi-
tion units, but they contain a surprising amount of control function-
ality as well. You can download programs to do scanning and units
conversion, limit checking, and some kinds of feedback control. You
have many options with the many types of interface modules available
for these scanners. Check for availability of LabVIEW drivers before
deciding on a particular model.
An often overlooked option is to use LabVIEW-based slave systems
with little or no MMI functionality. The slave nodes communicate with
the main MMI node, exchanging data in some LabVIEW formats that
you decide upon. Slaves handle all the real-time data acquisition and
control activity without the performance penalties of a busy graphical
display or lots of user intervention. This technique is advantageous
when you need sophisticated algorithms that a PLC or other simple
processor can’t handle. Also, since everything is written in LabVIEW,
it’s easier to maintain. Networking techniques are discussed later in
this chapter.
One downside to using LabVIEW in the loop may be in the area of reli-
ability, depending upon how critical your tasks are. Where I work, it is a
policy never to use a general-purpose computer for safety interlocks.
PLCs, on the other hand, are considered acceptable. The differences
are in the ruggedness of the hardware and in the simplicity of the PLC
software, which simply doesn’t crash and which also has a very deter-
ministic response time. The same cannot be said of a typical LabVIEW
system running on top of a complex operating system. You will have to
be the judge on this matter of safety and reliability.

Man-Machine Interfaces
When it comes to graphical man-machine interfaces, LabVIEW is
among the very best products you can choose. The library of stan-
dard and customizable controls is extensive; but more than that,
464 Chapter Eighteen

Process Mimic Alarm Summary

9/25/96 8:52:17 PM *** TIC-103
Upper vessel inlet STATE ALARM
9/25/96 8:52:27 PM *** TIC-103
Upper vessel inlet STATE ALARM
...CLEARED
9/25/96 8:53:49 PM *** TIC-104
Lower vessel inlet STATE ALARM

ACKNOWLEDGE ALARM DETAIL

Controller Faceplate Real-time Trend

TIC-203 MAN 10.0
RED TIC-103 BLUE TIC-104

SP PV OUT 8.0
53.1 66.3 8.2 6.0

100 100 4.0

2.0
80 80
0.0
60 60
-2.0
40 40
-4.0
20 20
-6.0
0 0 22:26:36 22:27:01 22:27:26 22:27:51 22:28:16
09/25/1996 09/25/1996 09/25/1996 09/25/1996 09/25/1996
HIGH

Figure 18.13 Some of the basic process displays that you can make in LabVIEW.

the function of each control is determined by you, the crafty program-

mer. In the process control world, there are several commonly used dis-
plays that you will probably want to implement in one form or another
(Figure 18.13):

■ Process mimic displays are based on simplified P&ID diagrams.

Various elements of the display are animated, tying measurements
and status information directly to physical elements in the plant.
Operators really like this kind of display. You can draw a representa-
tion of the system and import the picture into LabVIEW, then over-
lay it with various controls and indicators. Valves, heaters, and alarm
indicators are usually booleans, while various numeric controls and
indicators are placed near their respective instruments. DSC pro-
vides a library of process control symbols that speed the creation of
your displays.
■ Trending displays (graphs or strip charts), of which there are two
types: historical trends and real-time trends. A real-time trend dis-
plays up-to-the-minute data, but may not go very far back in time.
 Process Control Applications 465

It relies primarily on data stored in memory. Historical trends usu-

ally read and display data from large disk files. Trending is discussed
later in this chapter.
■ Controller faceplate displays look and act much like the front
panels of SLCs. They permit the operator to observe and control all
the settings of PID and other types of analog controllers. Digital I/O
or alarm status can also be presented in the form of boolean indica-
tors, like the example in Figure 18.14.
■ Alarm summary displays give the operator a concise overview of
important status information, particularly I/O points that are out of
specification in some way. You can use scrolling string indicators or
tables to list recent messages or boolean indicators as a kind of sta-
tus panel.

As an example, we decided to create a nice controller faceplate dis-

play that uses some of LabVIEW’s more advanced features. This one
mimics some of the commercial SLC front panels and is typical of the
faceplates we’ve seen displayed on commercial DCS screens. (On a
DCS, they usually put 8 or 10 faceplates on one screen, and the face-
plate is a standard indicator element.) Figure 18.14 shows the panel of
the controller subVI. You would normally size the VI window such that

Group & Unit

Tagname
Units Final A/M
Previous A/M Final SP/hi/lo Size VI window so that only
Previous SP/hi/lo
the faceplate shows
Tagname string indicator
AUTO switch
Group & Unit Previous SP/hi/lo Final SP/hi/lo
TIC-203 MAN
Group ID SP 50.00 SP 42.90
SP PV OUT
2 hi hi
80.00 42.9 82.65 9.2 74.70

Unit ID lo DegC lo
20.00 100 100 20.00
1
80 80
Previous A/M
60 Final A/M
Tagname MAN
MAN
40 40
TIC-203
60

20 20
Units
0 0
DegC
HIGH EXIT

Customized vertical slider control Alarm indicators (HIGH and LOW)

Figure 18.14 Here’s the panel of a subVI that mimics a single-loop controller
faceplate. You can call it from other VIs to operate many controllers since every-
thing is programmable.
466 Chapter Eighteen

only the faceplate part of the panel is visible; the other items are just
parameters for the calling VI. This subVI is programmable in the sense
that the tag name, units, and other parameters are passed from the
caller. This fact permits you to use one subVI to operate many control
loops. The VI must be set to Show front panel when called.
The SP (setpoint) slider control is customized via the various pop-up
options and the Control Editor. We first added two extra sliders by
popping up on the control and selecting Add Slider. They serve as high
and low alarm limits. We set the Fill Options for each of these alarm
sliders to Fill to Maximum (upper one) and Fill to Minimum (lower
one). In the Control Editor, we effectively erased the digital indicators
for the two alarm limits by hiding them behind the control’s main digi-
tal display.
The diagram in Figure 18.15 is fairly complex, so we’ll go through
it step by step. It illustrates the use of Local variables for control ini-
tialization. The I/O hardware in this example is, again, the Eurotherm
808 SLC. This VI relies on the controller to perform the PID algorithm,
though you could write a set of subVIs that perform the same task

Group
& Unit Read temperature True

PV 3 AUTO: read output

from controller
PV
OUT
HIGH
OU Local var
Units Units
SP
hi
LOW
6 Final SP/
hi/lo
Tagname Tagname lo

Setpoint

1 True
Setpoint changed;
False cases are empty
4 update controller

2 SP

Set controls to AUTO

previous values Final A/M
Previous A/M True = AUTO
True
AUTO Auto/man state changed;
1000
Local var
5 update controller

Previous SP/hi/lo 8
EXIT
SP
Local var

False
Dependency MANUAL: write from OUT
control to controller
SO
7
OUT

Figure 18.15 Diagram for the faceplate controller.

 Process Control Applications 467

with the PID Control Toolkit in LabVIEW, in conjunction with any type
of I/O.

1. Tag name and Units are copied from incoming parameters to indi-
cators on the faceplate. This gives the faceplate a custom look—as if
it was written just for the particular channel in use.
2. Previous settings for Auto? (the auto/manual switch) and SP (the
setpoint slider) are written to the user controls on the faceplate by
using local variables. This step, like step 1, must be completed before
the While Loop begins, so they are contained in a Sequence struc-
ture with a wire from one of the items to the border of the While
Loop.
3. The process variable (current temperature) is read from the control-
ler, checked against the alarm limits, and displayed.
4. The current value for the setpoint is compared against the previous
value in a Shift Register. If the value has changed, the setpoint is
sent to the controller. This saves time by avoiding retransmission of
the same value over and over.
5. In a similar fashion, the Auto? switch is checked for change of state.
If it has changed, the new setting is sent to the controller.
6. If the mode is automatic, the Out (output) control is updated by
using a Local variable. The value is read from the controller.
7. If the mode is manual, Out supplies a value that is sent to the con-
troller. These two steps illustrate an acceptable use of read-write
controls that avoids race conditions or other aberrant behavior.
8. The loop runs every second until the user clicks Exit, after which
the final values for Auto? and SP/hi/lo are returned to the calling
VI for use next time this VI is called.

Figure 18.16 shows how this subVI might be used in an applica-

tion to support multiple controllers. The panel contains a configura-
tion array that identifies each controller, a set of buttons to call the
desired controller, and an initialization array. When one of the buttons
is pressed, our old friend, the Which Button VI, returns a nonzero
value and Case 1 is executed (Case 0 just recirculates data in the Shift
Register, waiting for a button to be pressed).
Inside the Case, the Configuration cluster array is indexed to
extract the setup information for the selected controller, which is then
passed to the Controller Faceplate subVI, which opens when called.
Similarly, the controller data array, circulating in the Shift Regis-
ter, is indexed and unbundled. When the Controller Faceplate subVI
finishes, the data it returns is bundled and the current element in the
468 Chapter Eighteen

Init controller data Configuration

Controller
0 SP/hi/lo 0 to operate:
Group & Unit Tagname

SP 0.00 TIC-203 TIC-203

Group ID
hi 0.00 2 Units TIC-204
DegC
lo 0.00 Unit ID
TIC-321
1
A/M MAN STOP

Which Button.vi
TIC-203
TIC-204
= 0 if no buttons are pressed
TIC-321

1 [0..1]
Call Controller Faceplate with previous settings
Configuration Controller faceplate

Replace
Init Array
controller Element
data data for all ctlrs

A/M A/M
SP/hi/lo SP/hi/lo

data for this controller

array index

Case 0 just recirculates the

stop
controller data in the shift register

Figure 18.16 An application that uses the Faceplate Controller subVI. The Shift Register acts
as a database for previous controller values.

data array is replaced. The next time this controller is called, data from
the previous call will be available from the Shift Register. Note the use
of Bundle by Name and Unbundle by Name. These functions show
the signal names so you can keep them straight.
This overall procedure of indexing, unbundling, bundling, and replac-
ing an array element is a versatile database management concept that
you can use in configuration management. Sometimes the data struc-
tures are very complex (arrays of clusters of arrays, etc.), but the proce-
dure is the same, and the diagram is very symmetrical when properly
laid out. Little memory management is required, so the operations are
reasonably fast. One final note regarding the use of global memory of
this type. All database updates must be controlled by one VI to serial-
ize the operations. If you access a global variable from many locations,
sooner or later you will encounter a race condition where two callers
attempt to write data at the same time. There is no way of knowing
 Process Control Applications 469

who will get there first. Each caller got an original copy of the data at
the same time, but the last one to write the data wins. This is true for
both the built-in LabVIEW globals and the one based on Shift Regis-
ters that you build yourself.
If you want to save yourself lots of effort, check out the DSC MMI
G Wizard. It lets you interactively create your MMI from a library of
standard display elements such as trends, faceplates, and alarms.
What’s amazing is that you don’t have to do any G programming at
all, if you don’t want to.

Display hierarchy
Naturally, you can mix elements from each of the various types of dis-
plays freely because LabVIEW has no particular restrictions. Many
commercial software packages have preformatted displays that make
setups easy, but somewhat less versatile. You could put several of these
fundamental types of displays on one panel; or better, you may want to
segregate them into individual panels that include only one type of dis-
play per panel. A possible hierarchy of displays is shown in Figure 18.17.
Such a hierarchy relies heavily on global variables for access to infor-
mation about every I/O point and on pop-up windows (VIs that open
when called).
Your displays need to be organized in a manner that is easily under-
stood by the operator and one that is easy to navigate. A problem that
is common to large, integrated control systems is that it takes too long
to find the desired information, especially when an emergency arises.
Therefore, you must work with the operators so that the display hierar-
chy and navigation process make sense. The organization in Figure 18.17
is closely aligned with that of the plant that it controls. Each subdis-
play gives a greater level of detail about, or a different view of, a par-
ticular subsystem. This method is generally accepted by operators and
is a good starting point.

PLANT Plant Overview Mimic

PROCESSING UNIT Unit 1 Mimic Unit 2 Mimic Unit 3 Mimic

SUB-UNIT Alarm Summary Tank A Mimic Tank B Mimic

DETAIL Trend 1 Trend 2 Faceplates Detail Mimic

Figure 18.17 Hierarchical process displays allow the operator to

see any level of detail without having overcrowded screens.
470 Chapter Eighteen

The VI hierarchy can be mapped one for one with the display hierar-
chy by using pop-up windows (subVIs set to Show front panel when
called). Each VI contains a dispatcher loop similar to the one shown
later in this chapter (in Figure 18.29). Buttons on the panel select which
display to bring up. A Case structure contains a separate display subVI
in each frame. The utility VI, Which Button, multiplexes all the but-
tons into a number that selects the appropriate case. The dispatcher
loop runs in parallel with other loops in the calling VI.
Each subVI in turn is structured the same way as the top-level VI,
with a dispatcher loop calling other subVIs as desired. When it’s time
to exit a subVI and return to a higher level, an exit button is pressed,
terminating the subVI execution, closing its panel, and returning con-
trol to the caller. An extension of this exit action is to have more than
one button with which to exit the subVI. A value is returned by the
subVI indicating where the user wants to go next (Figure 18.18). The
caller’s dispatcher is then responsible for figuring out which subVI to
call next.
Making the actual buttons is the fun part. The simplest ones are the
built-in labeled buttons Or, you can paste in a picture that
is a little more descriptive. Another method is to use a transparent
boolean control on top of a graphic item on a mimic display, such as
a tank or reactor. Then all the operator has to do is click on the tank,
and a predefined display pops up. This is a good way to jump quickly

Level 1 Level 2 Level 3

MAIN

UNIT1
MIMIC
TANKA
MIMIC

UNIT1 Exit to UNIT1

MIMIC MIMIC

Exit and go to UNIT1 TRENDS

MAIN

UNIT1
TRENDS

Figure 18.18 Traversing the display hierarchy.

When you click an exit button on a subdisplay VI,
it returns a value that tells the calling VI where to
go next.
 Process Control Applications 471

to an associated trend, alarm, or other detail display. Remember that

the operator has to know that the buttons are there because he or she
can’t see them. Therefore, you must be consistent in your use of such
buttons, or the operators will get confused. To make a transparent bool-
ean, select a simple boolean control such as the Flat Square Button,
and use the Coloring tool to make both the foreground and background
transparent (T).

Power Windows: Use the VI Server. You can do some really interesting
display tricks (and a lot more) with the VI Server functions, which you
find in the Application Control function menu. They provide ways to
manipulate VIs that go way beyond the regular VI Setup items, such as
Show front panel when called.
The VI Server lets you act on VIs in three basic ways:

■ Property Nodes change or interrogate attributes of a VI, such as

the appearance of the front panel and VI information. Note: This is
not the same kind of Property Node (formerly known as Attribute
Nodes) that you use in conjunction with controls and indicators.
■ Invoke Nodes cause a VI to take some kind of action, such as run-
ning, saving, or printing. This is also called invoking a method (one
of those object-oriented programming terms).
■ Call By Reference Nodes allow you to run a VI—and pass param-
eters to and from it—without the actual VI being wired into the dia-
gram. This enables you to dynamically choose and run VIs while the
application is running.

The VI Server uses a call-by-name approach, where you supply a

string or path containing the exact name of the VI you wish to access
to the Open VI Reference node. Once the VI is loaded into memory,
you can start manipulating properties and methods.
Let’s start with the situation where you need to change the mode of
operation of your application in a significant way and you would prefer
not to have all the required VIs in memory at all times. For instance,
when the user clicks the Configure button, you can load the configura-
tion manager from disk, rather than having it memory-resident at all
times. Figure 18.19 shows a simple solution. We begin by opening a
VI reference, which locates the Config VI and returns a VI refnum.
Next, a Property Node opens the VI’s front panel (but doesn’t run it),
sets the window size and position, and changes the title. An Invoke
Node then runs the VI, and finally the reference is closed. This leaves
the Config VI running. When it terminates, you can force its window
to close and throw the Config VI out of memory by adding a bit of VI
472 Chapter Eighteen

True
Open the VI's panel; Run the VI. Close ref and
Size to 640 by 480 pixels; Don't wait check errors.
Set the title string. until it stops.
Open VI Close LV Object
vi path Reference Property Node Invoke Node Reference
HD:Proj: VI VI
Config.vi
Configure FP.Open Run VI
FP.PanelBounds Wait until done
Front Panel Window:
FP.Title
Panel Bounds
left right This is the window title
0 640 Front Panel Window:Title
top bottom
0 480

Figure 18.19 The VI Server opens a VI, sets the appearance of the panel, runs the VI, and
leaves it running. It’s a kind of remote control, and it works even over a network.

Server code to its diagram. All you need to do is have it open a VI refer-
ence (to itself), then have a Property Node set the Front Panel Open
property to False.
The Call By Reference node operates in a manner similar to wiring
the designated subVI directly into the diagram, right down to pass-
ing parameters to controls and receiving results back from indicators
on the subVI. The important difference is that the subVI is dynami-
cally loaded and unloaded, and which subVI is loaded is determined by
the path name that you supply. In order to make it possible for you to
dynamically substitute “any” VI into the Call By Reference node, you
must make the connector pane identical in all of the candidate VIs.
Next, you have to tell the VI Server what your standard connector
pane looks like. Beginning with an Open VI Reference function, pop
up on the type specifier VI input and choose Create Control. A
VI Refnum control will appear on the panel. Drag the icon of one of your
standard VIs onto that VI Refnum control, and the connector pane will
appear there. You have now created a class of VIs for the VI Server.
Drop a Call By Reference Node on the diagram, connect it to the
Open VI Reference function, and you’ll see your connector pane appear.
Then all you have to do is wire to the terminals as if the standard VI
were actually placed on the diagram. Figure 18.20 shows the complete
picture. In this example, the selected subVI runs in the background
without its panel open unless it’s configured to Show Front Panel When
Called. You can also add Property Nodes to customize the appearance,
as before.
The VI Server offers additional opportunities for navigation among
display VIs. To speed the apparent response time of your user interface
when changing from one MMI window to another, consider leaving sev-
eral of the VIs running at all times, but have only the currently active
 Process Control Applications 473

type specifier VI error out

Refnum (for type only)
status code
0

source

N
12.00

type specifier VI Refnum Run the VI. Close ref and

(for type only) check errors.

Close LV Object
vi path Call By Reference Node Reference
HD:Proj:
Config.vi
HD:a path N
Open VI
Reference
error out

Figure 18.20 The Call By Reference node loads a VI from disk and
exchanges parameters with it as if the subVI were wired directly
into the diagram. This can save memory since the subVI is loaded
only when required.

one visible. When the user clicks a button to activate another window,
use an Invoke Node to open the desired panel. Its window will pop to
the front as fast as your computer can redraw the screen. A similar
trick can prevent users from getting lost in a sea of windows. If your
display hierarchy is complex and you allow more than one active win-
dow at a time, it’s possible for the user to accidentally hide one behind
the other. It’s very confusing and requires excessive training to keep
the user out of trouble. Instead, try to simplify your display navigation
scheme, or use Invoke Nodes to assist the user.

Other interesting display techniques

You can customize controls and indicators by replacing the built-in pic-
tures with custom ones from a drawing package. A useful example,
submitted by Corrie Karlsen of LLNL, is shown in Figure 18.21. The
system being controlled has a carousel with 10 sample bottles and a
liquid sampling valve. When the operator moves the valve icon, a step-
per motor moves the carousel to the appropriate bottle. The operator
presses a fill button (not shown), and the valve is cycled open for a
predetermined period of time and then closed. This LabVIEW indicator
then responds by updating the full/empty status of the sample bottles,
which are boolean indicators.
You can create boolean controls that look like valves with different
open and closed states by pasting in a pair of pictures. If the operator
prefers feedback regarding the status of a valve that has limit switches,
474 Chapter Eighteen

Figure 18.21 The sample valve is a horizontal slider control with the
valve picture pasted in as the handle. The control has no axis labels,
and the housing is colored transparent. The bottles are boolean indi-
cators, and the pipes are static pictures.

you can place a Pict Ring indicator for status on top of a transparent
boolean control for manual actuation. For a valve with high and low
limit switches, the ring indicator would show open, transit, closed, and
an illegal value where both limit switches are activated.
You can animate pipes, pumps, heaters, and a host of other process
equipment by using picture booleans or Pict Rings. To simulate motion,
create a picture such as a paddle wheel in a drawing program, then
duplicate it and modify each duplicate in some way, such as rotating
the wheel. Then, paste the pictures into the Pict Ring in sequence. Con-
nect the ring indicator to a number that cycles through the appropriate
range of values on a periodic basis. This is the LabVIEW equivalent of
those novelty movies that are drawn in the margins of books.
Perhaps the most impressive and useful display is a process mimic
based on a pictorial representation of your process. A simple line draw-
ing such as a P&ID showing important valves and instruments is easy
to create and remarkably effective. If your plant is better represented
as a photograph or other artwork, by all means use that. You can import
CAD drawings, scanned images, or images from a video frame grabber
to enhance a process display.
Property nodes allow you to dynamically change the size, posi-
tion, color, and visibility of controls and indicators. This adds a kind of
animation capability to LabVIEW. The Blinking attribute is especially
useful for alarm indicators. You can choose the on/off state colors and
the blink rate through the LabVIEW Preferences.

Handling all those front panel items

Process control panels tend to be loaded with many controls and indi-
cators. This can make your diagram very busy and hard to under-
stand. What you need to do is bundle related controls into arrays or
 Process Control Applications 475

1 [0..2] 2 [0..2]
Unbundle Outputs for indicators
FV-305 Build array of controls.
NOTE: Order-dependent! FCV-305 ON?
FV-204
FCV-204 ON?
FV-304
FCV-304 ON?
FV-401
FCV-401 ON?
FV-402
Unbundle FCV-402 ON?
TIC-101 DO 0..7
TIC-101 ON?
TIC-102
TIC-102 ON?
TIC-201
TIC-201 ON?

Figure 18.22 Clean up your diagram by combining controls and indicators into arrays or clus-
ters. Other frames in this sequence are the sources and sinks for data. The unbundler subVI
on the right extracts boolean values from an array or cluster. You could also use Unbundle By
Name.

clusters and pass them to subVIs for processing. Similarly, indica-

tors are unbundled for display after being returned from source VIs
(Figure 18.22). A Sequence structure helps to conserve screen space. It
doesn’t impair the visual flow of the program too much because only a
single item is passed from one frame to the next.
One hazard in this array-based scheme is that all the controls and
indicators are order-dependent; that is, element three of the control
array is always PV-401. Any wiring or indexing errors anywhere in
your program will send the wrong value. An alternative is to use clus-
ters and the Bundle by Name and Unbundle by Name functions. There
is still a limitation with named clusters, however; you can’t arbitrarily
configure channels (insert or delete them) during execution with a con-
figuration manager. Channels that you access must always exist, and
editing of one or more diagrams will always be necessary to change
the overall configuration. The way out of this quandary is a real-time
database, a topic we cover later in this chapter.
Experts in human factors tell us that having too much information
on one screen leads to information overload. Operators can’t find what
they’re looking for, and the important values get lost in a sea of colored
lights. If you find that your MMI display VI has a ridiculous number of
controls and/or indicators, it’s time to break it into multiple panels. It
will then be easier for operators to use, and easier for the programmer
to design and maintain. This is just one more reason for using hierar-
chical displays.

Data Distribution
If you think about MMI display hierarchies, one thing you may won-
der about is how the many subVIs exchange current data. This is a
data distribution problem, and it can be a big one. A complex process
476 Chapter Eighteen

Control VIs Output Output I/O Input Data global Display VIs
queue VI handler VI Hardware handler VI variable
Output
Output
Queue
Queue Output Input
Handler Handler

Figure 18.23 Data distribution in a simple LabVIEW process control system. Control VIs write
new settings to an output queue, from which values are written to the hardware by an output
handler VI. An input handler VI reads data and stores it in a global variable for use by multiple
display VIs.

control system may have many I/O subsystems of different types, some
of which are accessed over a network. If you don’t use a coherent plan
of attack, performance (and perhaps reliability) is sure to suffer. In
homemade LabVIEW applications, global variables generally solve the
problem, if used with some caution.
Most commercial process control systems, including DSC, use a real-
time database to make data globally accessible in real time. You can
emulate this in your LabVIEW program at any level of sophistication.
The general concept is to use centralized, asynchronous I/O handler
tasks to perform the actual I/O operations (translation: stand-alone
VIs that talk to the I/O hardware through the use of LabVIEW driv-
ers). Data is exchanged with the I/O handlers via one or more global
variables or queues (Figure 18.23). User-interface VIs and control VIs
all operate in parallel with the I/O handlers.

Input scanners as servers

The client-server model is very effective for the input data in process
control systems. An input handler task, a server VI which I like to call
a scanner, periodically polls all the defined input channels, applies
scale factors, and writes the results to global variables. The scanner
is also a logical place to check any alarm limits. Depending on your
application’s complexity, the scanner VI may also be responsible for
some aspects of trending or data archiving.
Figure 18.24 is a simplified scanner VI. It is called by a top-level VI
and runs periodically (in parallel with the top-level VI) until a boolean
global variable called RUN is set to False. Two input handler VIs—one
for analog inputs and the other for digital—acquire raw data, filter it,
and scale it to engineering units. Handler outputs are cluster arrays
containing each channel’s value, its name, and any other information
that client tasks might need. This information is written to global vari-
ables and passed to alarm generator VIs that test each channel for
alarm limit violations. Alarm flags or messages are stored in a queue
(see the section entitled “Using an Alarm Handler,” which follows).
 Process Control Applications 477

Input Input Alarm

handlers globals generators

Analog in

Digital in

Scan interval
RUN

Figure 18.24 A scanner VI that handles analog and digital

inputs. The I/O configuration subVI, at left, passes informa-
tion to the input handlers. Data from the input handlers
is written to other global variables for use by client VIs.
Alarm limits are also checked, and the results are written
to alarm message queues.

As you might expect, process control has many of the same configu-
ration needs as a data acquisition system. Each input handler requires
information about its associated input hardware, channel assign-
ments, and so forth. This information comes from an I/O configuration
VI, which supplies the configurations in cluster arrays. If configura-
tion information is needed elsewhere in the hierarchy, it can be passed
directly in global variables, or it can be included in the data cluster
arrays produced by the input handlers.
Here’s an important tip regarding performance. Where possible, avoid
frequent access of string data, particularly in global variables. Strings
require extra memory management, and the associated overhead yields
relatively poor performance when compared to all other data types. Not
that you should never use strings, but try to use them sparingly and only
for infrequent access. For instance, when you open an operator display
panel, read channel names and units from the configuration database
and display them just once, not every cycle of a loop.

Handling output data

For efficiency, it may be better to update an output only when its value
changes because some devices have a fair amount of overhead associ-
ated with communications. Like the input model we’ve been discuss-
ing, you can use an output handler and an output global variable
that contains the values for each output channel. The global variable
carries an array with each element of the array corresponding to an
478 Chapter Eighteen

Output Output
globals handlers

Analog out

Digital out

Scan interval
RUN

Figure 18.25 An output scanner, which is quite

similar to an input scanner.

output channel. Any control VI can update values in the output global.
The output handler checks for changes in any of the output channels and
updates the specific channels that have changed. You can combine the
handler and its associated global variable into an output scanner, as in
Figure 18.25,which works like an input scanner in reverse.
To test for changes of state on an array of output values, use a pro-
gram similar to the one in Figure 18.26. A Shift Register contains
the values of all channels from the last time the handler was called.
Each channel’s new value is compared with its previous value, and if
a change has occurred, the output is written. Another Shift Register is
used to force all outputs to update when the VI is loaded. You could also
add a force update or initialize boolean control to do the same thing.
Depending upon the complexity of your system, channel configuration

Analog out

For each channel...

True
False
previous values Change detected: No change;
write the value Skip this channel
False = initialize

Figure 18.26 An output handler that only writes to an output channel when a change of value
is detected. The upper Shift Register stores the previous values for comparison, while the lower
Shift Register is used for initialization. Its value is False when the VI is loaded, forcing all out-
puts to update. Additional initialization features may be required if the program is stopped then
restarted.
 Process Control Applications 479

information may be required to perform the output write operation. If

so, you can read the configuration global variable and pass that infor-
mation along to the subVI that writes the data.
Because the Analog Output global variable contains an array of val-
ues, control VIs have to update that array, not overwrite it. Updates
are easy to do. Just read the array from the global, use Replace Array
Element to surgically change the value of a single channel, then write
the array back to the global. Please note that there is again a great
probability of a race condition arising. If the global variable is accessed
by multiple callers, they will clash over who writes last. The solution is
to encapsulate all write operations for a global variable inside a subVI,
effectively serializing access. There is, of course, no limit to the number
of VIs that can simultaneously read the global data.
Another way to avoid race conditions is to use an output queue. A
queue is a first-in, first-out, or FIFO, construct that works like people
standing in line. When a high-level VI needs to update an output, it
calls an intermediate VI that adds the request to an output queue.
Later, the output handler VI reads and empties the queue and updates
each specified output.
The Datalogging and Supervisory Control (DSC) module includes a
very capable process control database,*—one of the main reasons that
you should consider using this package for your more complex appli-
cations. It’s actually written in G (though they don’t supply the dia-
grams for you to fiddle with) and the tag access VIs make it very easy
to access information by tag name. All of the difficult synchronization
problems, especially with outputs, are taken care of for you. The data-
base interfaces with drivers based on the OPC server model, giving
you a consistent and transparent connection with all of your I/O points.
Inputs are scanned at rates that you choose, alarms are checked, and
outputs are updated when a change is detected. It works even with
other networked PCs running DSC, exchanging input and output data
with remote I/O as if it was on the local machine.

Display VIs as clients

By now you should be getting the idea that global variables are key
elements when you are designing a versatile process control system.
VIs that display data are very easy to write because global variables

* Historical note: In early 1991, Gary wrote a letter to the LabVIEW developers describ-
ing the elements and behaviors of a process control database, hoping that they would build
such a beast into LabVIEW. As National Instruments got more involved with the process
control industry, they realized how important such a capability really was. Audrey Har-
vey actually implemented the first prototype LabVIEW SCADA package in LabVIEW 3
sometime in 1993. It eventually became the basis for DSC, now available as DSC.
480 Chapter Eighteen

DISPLAY SELECT
Goto Screen
CHARTS MIMIC METERS MAIN
0

Expanded Outside Diameter (in.)

Target: Target 0.000 Max Above 0.000 Max Below 0.000

Statistical: UCL Setpoint 0.000 LCL Setpoint 0.000 Range UCL 0.000

Pre-Control: Upper Limit 0.000 Lower Limit 0.000 Range Upper Limit 0.000

Longitudinal Change (%)

Target: Desired L.C. 0.000 L.C. Setpoint 0.000

Statistical: UCL 0.000 LCL 0.000 Range UCL 0.000

Pre-Control: Upper Limit 0.000 Lower Limit 0.000 Range Upper Limit 0.000

Dips per Kft

Statistical: Process UCL 0.000 Process LCL 0.000 Spec UCL 0.000

Pre-Control: Process U.L. 0.000 Process L.L. 0.000 Spec U.L. 0.000

Figure 18.27Front panel of a display and control VI that is used to

monitor and set operating specifications.

solve the data distribution problem. The input scanner supplies up-to-
date values for all inputs, while the configuration part of your system
supplies such things as channel names, units, and other general infor-
mation, all available through global variables. Figures 18.27 and 18.28
show how we wrote a typical display/control VI by using the methods

True
1 [0..2] False case
Display present values Read controls is empty

L.C. Setpoint UCL

PLC data
2
Range Upper Limit
LCL
25
Upper Limit
Range UCL
Mean 17
Lower Limit Flags
+3sigma Desired L.C.
18
-3sigma

MAIN Which
seconds 2.00 Button.vi Goto Screen
200 MIMIC
CHARTS
Interval Timer.vi
METERS

Figure 18.28 Diagram for the display/control VI. The Case structure and interval
timer limit how often the displays and controls are updated. A Sequence logically
groups the controls and indicators, eliminating clutter. The VI loops until the user
clicks a button to jump to another display.
 Process Control Applications 481

we’ve discussed. Since this display example (Figure 18.27) is part of a

hierarchical suite, there are buttons marked Display Select near the
top. Clicking one of them causes this VI to terminate and returns an
appropriate value in Goto Screen (which would normally be posi-
tioned off-screen so the user doesn’t see it). The rest of the panel is a
collection of numeric controls and indicators, grouped by function, with
the help of some decorations. A fancy process mimic graphic could also
be placed underneath the controls.
In the associated diagram in Figure 18.28, you can see that the Which
Button VI determines whether the VI will continue to run, based on the
state of all the Display Select buttons. The While Loop runs at a fairly
fast rate, on the order of every 200 ms or so, to provide rapid response
to the buttons. It would be wasteful, however, to update all the display
items at that rate, since the data is only updated by the scanners every
couple of seconds. The Interval Timer VI controls a Case structure
that contains all the real activity. Every two seconds, the Case is True;
otherwise, the False case, which contains nothing, executes.
Input data comes from several global variables containing data from
a PLC and some computed statistics which are written by the input
scanner VI. Output data from controls on the panel are built into an
array and written to another global variable called Flags, which is used
elsewhere to indicate system status. This is the only VI in the hierar-
chy that writes to Flags global variable, so there is no reason to create a
queue or other protective device for that global variable. For clarity, we
used a Sequence structure to logically group the front panel items.
If the display needs to call lower-level displays, add a second, inde-
pendent While Loop as a dispatcher, as shown in Figure 18.29,
just like the top-level VIs have. Some buttons will then activate the

DISPATCHER
0 [0..4] 1 [0..4]
MAIN (we're already here!) Call SPECS display. When SPECS is
next finished, it returns a value for the
MAIN next screen. Each frame of the
screen
SPECS dispatcher Case is similar.
Case zero is MAIN.
CHARTS
SPECS display VI
METERS
Which
MIMIC Goto screen
button.vi

Global to stop
QUIT RUN
main loop

Figure 18.29 The dispatcher loop in a higher-level VI that calls the display VI from the preceding
figures.
482 Chapter Eighteen

lower-level displays. Other buttons that cause a return to higher-level

displays are wired as shown in Figure 18.28. This dispatcher concept is
flexible and easy to program.

Initializing controls. All controls present on your display panel have

to be initialized when the VI is called. Otherwise, they will revert to
their default values, which will then be accepted as if those were user
inputs—probably not a great idea. The simplest way to initialize con-
trols is to supply initial values from the calling VI through the con-
nector pane. Every control will have to be wired to the connector pane,
and that may be a problem: You have only 20 terminals, one of which is
probably the Goto Display output. If you have more than 19 controls to
initialize, some will have to be colocated in clusters. Think about your
panel layout and figure out which controls might be logically grouped
in this way. You can effectively hide the border of a cluster by coloring it
transparent so that it disappears behind the other graphical elements
of your display.
You can also initialize controls with local variables from within the
display VI. For the present example, initial values for the manual flag
controls are easy to obtain because each control’s previous value was
written to the Flags global variable. After indexing each value out of
the Flags array, use local variables to set each control’s value, then
start the main While Loop.

Using network connections

To create a distributed control system (even a very small one), you will
need to use a network. The most versatile network connections use
Ethernet and TCP/IP, which is directly supported by LabVIEW through
VIs in the Networking function palette. IP (Internet Protocol) per-
forms the low-level packaging of data into datagrams and determines a
network path by which messages can reach their destination. However,
it offers no handshaking and doesn’t even guarantee delivery. TCP
(Transmission Control Protocol) adds handshaking and guaran-
tees that messages arrive in the right order, making it quite reliable.
UDP (Universal Datagram Protocol) is similar to TCP, but does no
handshaking, making it somewhat faster and also offering the possibil-
ity of broadcasting a message to many recipients. You establish sessions
between each LabVIEW system on the network, then exchange binary
data in formats of your choosing. There are several examples of TCP/IP
clients and servers included in the networking examples. These example
VIs could be modified to run in the background with local communica-
tions via global variables. A Data Server VI would then act as a tunnel
 Process Control Applications 483

for data that reappears in a Data Client elsewhere on the network. You
can have as many connections, or ports, open simultaneously as you
need. Several TCP/IP examples are included with LabVIEW, and you
can refer to other books that cover the subject in depth (Johnson 1998;
Travis 2000).
The VI Server includes transparent, multiplatform network access.
LabVIEW itself can be the target, or particular VIs can be manipulated
across the network in the same ways that we have already described.
This makes it possible, for instance, to use a Call By Reference node
on a Windows machine to execute a VI on a remote Macintosh run-
ning LabVIEW. An example VI shows how to do exactly that. Remem-
ber that you have to set up the VI Server access permissions in the
LabVIEW Options dialog.
DataSockets are a great cross-platform way to transfer many kinds
of data. A DataSocket connection consists of a client and a server that
may or may not reside on the same physical machine. You specify the
source or destination of the data through the familiar URL designator,
just like you’d use in your Web browser. The native protocol for Data-
Socket connections is DataSocket Transport Protocol (dstp).
To use this protocol, you must run a DataSocket server, an applica-
tion external to LabVIEW that is only available on the Windows plat-
form as of this writing. You can also directly access files, File Transfer
Protocol (ftp) servers, and OPC servers. Here are a few example URLs
that represent valid DataSocket connections:
■ dstp://servername.com/numericdata where numericdata is the
named tag
■ opc:\\machine\National Instruments.OPCModbus\Modbus Demo Box.4:0
■ ftp://ftp.natinst.com/datasocket/ping.wav
■ file:\\machine\mydata\ping.wav

Figure 18.30 shows a very simple example where data is written in

one While Loop and read in another. The DataSocket Write function
includes inputs for the URL and polymorphic data. In this case we’re
writing a numeric array to a URL that specifies the local machine and
a tag called MyGraphData. The DataSocket Read function is simi-
lar, requiring a URL and a data type specifier. Since the URLs match,
the two loops in this example represent a server (writer) and a client
(reader). The LabVIEW examples include a nice remote-access demon-
stration that reads weather data from a machine running LabVIEW
at National Instruments headquarters in Austin, Texas. As you can
see, DataSockets are a flexible data tunnel that requires no special
programming on your part.
484 Chapter Eighteen

Open a datasocket connection for writing data

DataSocket Write
dstp://localhost/MyGraphData

360.00
200 stop

Open a datasocket connection for reading data

DataSocket Read
dstp://localhost/MyGraphData

data type 0 Graph

ms timeout 500

stop

Figure 18.30 Two While Loops share data via a DataSocket

connection. The loops could just as well reside on different
computers.

Files as mailboxes. An old, reliable way to exchange data between

VIs over the network is to use files as mailboxes that any node can
remotely fetch and read or write. For one-way transmission, files are
easy to use. Just write data to a file in some predetermined format on a
periodic basis, then the remote node can fetch it asynchronously, read,
and decode the information. Things get tricky when you need synchro-
nized, bidirectional communications because you must create a system
of file access permission limits using flags or some other technique.

Real-time process control databases

If you buy a commercial DCS or PC-based industrial process con-
trol package, you will find that the entire software environment is
database-driven. For user interaction, there is invariably a program
called the Database Builder that edits and compiles all the configura-
tion information. The run-time programs are then centered around this
database: I/O handlers pass values in and out to hardware, alarm han-
dlers compare measurements with limit values, and trenders display
values on the screen. LabVIEW, however, is not in itself a database-
driven package. In fact, it would be much less flexible if it were cen-
tered around a predefined database structure.
A particular requirement for a database used in process control is
that it must be a real-time database, which implies that it is memory-
resident and available for thousands of transactions per second. For
added versatility, the database may also be distributed among vari-
ous nodes in the system. This permits any node to efficiently exchange
 Process Control Applications 485

information with any other node on the network through a consis-

tent message-passing scheme. A global, real-time, database requires
a sophisticated client-server or peer-to-peer relationship among all
nodes on the network. Another attribute of a good database is access by
name. That is, you supply the tag name and attribute (such as Setpoint
or Alarm High) that you wish to access, and the database looks up the
name and accesses the correct items. This eliminates the headaches
we’ve all experienced trying to keep track of array index values when
data is carried in arrays. Such programming is possible in LabVIEW,
but is beyond the scope of this book and beyond the abilities of all but
the most experienced programmers. This is one of the areas where you
may want to turn to DSC or another SCADA product.

Simulation for validation

An effective way of testing or validating your process control system

is by simulating real I/O. The obvious way is to replace each physical
input with an appropriate control or calculated data source and replace
each physical output with an appropriate indicator. If you’ve designed
your application with the client-server model discussed here, simula-
tion turns out to be quite easy. When we wrote our first well-designed
client-server-based system, we didn’t have any hardware for testing, so
it was imperative that the software have a built-in test mode. Since the
inputs all came from a PLC, the core of the input scanner VI consisted
of a series of PLC input operations, and all values were collected in a
single array for use by the clients. We created a boolean control that
selected either normal or simulation mode and used that to control a
Case structure. The normal frame of the Case structure contained the
PLC programming, while the simulation frame contained a set of math-
ematical expressions in a big Formula Node. There was one expression
for each signal, and all results were collected into an array, just like
the real data. To keep things interesting, the formulas contained sine
functions and random number generators.
The status of almost all output signals in a process control system
are probably displayed by indicators on various panels. For simulation,
then, all you must do is disable the VIs that talk to the hardware. If
your outputs are handled by output servers, encapsulate the hardware
output VIs in a Case structure tied to a mode control, just like we did
for those simulated inputs.
If your system has closed-loop controls, where inputs respond to out-
puts, consider writing output data to global variables that can be read
by the input simulation code. Then add some scale factors or perhaps
some time-dependent response to this feedback data. The PID Control
Toolkit contains a simple plant simulator that does such a trick.
486 Chapter Eighteen

There are great payoffs to built-in simulation. Most importantly, you

can do a thorough job of testing your application offline. That means
you can work anywhere you like, and you don’t have to wait for time on
the real equipment. There is a much lower risk of sending the real pro-
cess out of control, losing data, or encountering some other unforeseen
condition or bug. As a bonus, the simulation mode is handy for demon-
stration purposes, like showing the boss how crafty you are.

Sequential Control
Every process has some need for sequential control in the form of inter-
locking or time-ordered events. We usually include manual inputs in
this area—virtual switches and buttons that open valves, start motors,
and the like. This is the great bastion of PLCs, but you can do an admi-
rable job in LabVIEW without too much work. We’ve already discussed
methods by which you read, write, and distribute data. The examples
that follow fit between the input and output handlers.

Interlocking with logic and tables

G’s boolean logic functions make ordinary interlocking simple
(Figure 18.31). Boolean conditions can be derived from a variety of
sources such as front panel switches, digital inputs, or comparison func-
tions acting on analog signals. Ladder logic, as shown in Figure 18.31,
is the most common language for PLC programming. It comes from the
days of electromechanical switching and remains a useful documenta-
tion technique for logical systems.
One way to add versatility to your logical operations is to use a bool-
ean table, as shown in Figure 18.32. A table is a two-dimensional array
where the columns are the mode or step number and the elements
in each row represent some state of the system, such as a permissive
interlock. The Mode control selects the column through the use of a 2D
Index Array function. The output of the Index Array function is a 1D
array of booleans—a column slice from the incoming table. Another set
of Index Array functions selects the individual interlock permissives.

Ladder Logic LabVIEW Equivalent

A C D OUT
C D
OUT

B A
B

Figure 18.31 Simple interlock logic, comparing a ladder logic network with its LabVIEW
equivalent.
 Process Control Applications 487

Input A
Mode Startup 0
Output A
Boolean Table
row 0 Input B
col 0
Output B

Input A

Output A
Boolean Table row 0

Output B
row 1
Mode
Input B

Figure 18.32 An interlock table. The 2D boolean array (a table) is

sliced by an array indexer to deliver an interlock permissive vector,
which then is ANDed with a series of inputs.

If one of the input switches is True and its corresponding permissive

(from the table) is True, then the output is True. You can add whatever
kind of logic is required and expand the table as necessary.
The Mode control could be a Ring control that selects between sys-
tem modes such as Startup, Run, and Maintenance. This is an effective
way to change the action of a large number of interlock chains without
a great deal of wiring. Another use for the Mode control is to tie it to
an internally generated value in your program. For instance, the mode
might change in response to the time of day or the temperature in a
reactor.
Tables are an efficient way to store interlock information, and they
are very fast. The only bad thing about tables is that they can be hard
to debug. When a table is very large and your program changes modes
very often, it’s not always obvious what is going on. Directly wired
logic is generally easier to understand and debug, and should be used
preferentially.

State machines
The state machine architecture is about the most powerful LabVIEW
solution for sequential control problems. A State Machine consists of
a Case structure inside of a While Loop with the Case selector carried
in a Shift Register. Each frame of the state machine’s Case structure
has the ability to transfer control to any other frame on the next itera-
tion or to cause immediate termination of the While Loop. This allows
you to perform operations in any order depending on any number of
conditions—the very essence of sequential control.
488 Chapter Eighteen

Start Fill
Fill stay here
Start Fill next frame
Timeout

time now Duration

Set up some output Upper Limit
conditions here, like Start time
opening the valve that
start fills a tank.
time
Upper limit False = upper
Open Valve limit reached
continue tank level continue

Figure 18.33 This state machine implements an operation that terminates when the tank is full
or when a time limit has passed. The Timeout of the Case (not shown) takes action if a time-out
occurs. The Upper Limit frame (also not shown) is activated when the upper limit for tank level
is exceeded.

Figure 18.33 is an example of a tank-filling operation. The objective

is to fill the tank, stopping when an upper level is reached or when
a certain time has passed. In the first frame, Start, you open a valve
(presumably by writing to an output device) that starts filling the tank.
The starting time is saved in a shift register. The program then pro-
ceeds to the second frame, Fill, where the elapsed time and present
tank level are checked. The tank level is obtained from an analog input
or from a global variable, which in turn is written by an analog input
handler. If the elapsed time exceeds the specified duration, the pro-
gram goes to the Timeout frame. If the upper limit is exceeded, jump to
the Upper Limit frame. Otherwise, keep looping on the Fill frame. The
Timeout and Upper Limit frames may then take appropriate action to
close the valve, or whatever; other operations may follow.
More activity can be managed in this same structure by looping over
several frames rather than just one. The sequence could be 1-2-3-1-2-3,
and so forth, with the termination conditions causing a jump to the
fourth or fifth frame. Very complex looping is possible, although, like
any scheme, it can be hard to debug if you’re not careful. We usually put
an indicator on the panel that shows which state it’s in at the moment.
Single-stepping then shows the execution order. If the process has to
run fast, you can accumulate a history of all the states by building an
array on the boundary of the While Loop for later review.
Multiple state machines can run in parallel with each managing a
different task. You could have several on one diagram with a global
boolean that requests all the loops to stop. This would permit you to
break your control problem down into logical, related pieces that are
easier to design and debug.
 Process Control Applications 489

Initialization problems
Initialization is important in all control situations and particularly so
in batch processes that spend much of their time in the start-up phase.
When you first load your LabVIEW program, or when you restart for
some reason, the program needs to know the state of each input and
output to prevent a jarring process upset. For instance, the default val-
ues for all your front panel controls may or may not be the right values
to send to the output devices. A good deal of thought is necessary with
regard to this initialization problem. The DSC database includes user-
defined initialization for all I/O points, drastically reducing the amount
of programming you’ll need to do.
When your program starts, a predictable start-up sequence is neces-
sary to avoid output transients. Begin by scanning all the inputs. It’s
certainly a safe operation, and you probably need some input data in
order to set any outputs. Then compute and initialize any output val-
ues. If the values are stored in global variables, the problem is some-
what easier because a special initialization VI may be able to write
the desired settings without going through all the control algorithms.
Finally, call the output handler VI(s) to transfer the settings to the
output devices. Also remember to initialize front panel controls, as dis-
cussed earlier.
Control algorithms may need initialization as well. If you use any
uninitialized Shift Registers to store state information, add initializa-
tion. The method in Figure 18.34 qualifies as another Canonical VI.
The technique relies on the fact that an uninitialized boolean Shift
Register contains False when the VI is freshly loaded or compiled.
(Remember that once the VI has been run, this is no longer the case.)
The Shift Register is tested, and if it’s False, some initialization logic
inside the Case structure is executed. A boolean control called Initial-
ize is included to permit programmatic initialization at any time, such
as during restarts.
You could also send flags to your control algorithms via dedicated
controls or global variables. Such a flag might cause the clearing of

False
Shift register to True
be initialized real work
123 goes here Normal
Initialize operation
here
This shift register contains
False when first loaded,
Initialize Don’t force init after
forcing initialization
first time through
Figure 18.34 The upper Shift Register, perhaps used for a control algorithm, is initialized by
the lower Shift Register at startup or by setting the Initialize control to True.
490 Chapter Eighteen

intermediate calculations, generation of empty arrays for data stor-

age, or cause the outputs to go to a desired state. All of these items
are worth considering. Be sure to test your control system thoroughly
by stopping and then restarting in various states. Start-up is the time
where most process disasters occur. Of course, bad things never happen
on our projects.

GrafcetVIEW—a graphical
process control package
There is an international standard for process control programming
called GRAFCET. It’s more popular in Europe, having originated
in France, but is found occasionally in the United States and other
countries. You usually need a special GRAFCET programming pack-
age, much like you need a ladder logic package to program a target
machine, which is usually a PLC. The language is the forerunner of
sequential function charts (SFCs) (IEC standard 848), which are
similar to flowcharts, but optimized for process control and PLCs in
particular (Figure 18.35). Emmanuel Geveaux and Francis Cottet at
LISI/ENSMA in conjunction with Saphir (all in France) have devel-
oped a LabVIEW package called GrafcetVIEW that allows you to do
GRAFCET programming on LabVIEW diagrams. It’s a natural match
because of the graphical nature of both languages.
One enhancement to LabVIEW that the authors had to make was
additional synchronization through the use of semaphores, a classic
software handshaking technique. In many control schemes, you may

1
Graphical representation of a step.

2 5 7
15
Graphical representation of an initial step.

9
Graphical representation of a transition.

Graphical representation of the relation 11 20 15

between steps and relations.

When several steps are related to the same transition,

they are joined with this symbol.
Figure 18.35The GRAFCET language is a kind of sequential function chart. Once you under-
stand the symbols, it’s a reasonably clear way to describe a sequential operation. (Courtesy
Emmanuel Geveaux, LISI/ENSMA.)
 Process Control Applications 491

GrafcetVIEW Panel GrafcetVIEW Diagram

Boolean indicators
Initial Step VI

S0 S1 Upstream
Synchronization VI

Transition VI
E0

Step VI
S2

E1
Transition VI

Downstream
Picture, pasted into front panel Synchronization VI

Figure 18.36 This is one of the GrafcetVIEW demonstration VIs. The panel graphic was pasted
in, and boolean controls and indicators were placed on top. Compare the LabVIEW diagram
with actual GRAFCET programming in Figure 18.35.

have parallel tasks that execute continuously and independently, but

which occasionally need to force one another to pause to guarantee
sequential timing during certain operations. The GrafcetVIEW solu-
tion is a set of subVIs that manages semaphores. It may also be pos-
sible to implement the solution via Occurrences, but their semaphore
trick seems to be immune to some of the difficult starting and ending
occurrence generation situations.
GrafcetVIEW requires only a few subVIs to mimic the GRAFCET
language, and they are wired together in a manner that matches the
drawings. Figure 18.36 shows the panel and diagram of one of the
GrafcetVIEW example programs. The most remarkable feature of this
diagram is that it apparently implements cycles, or closed loops, in
LabVIEW. As you may know, cycles of any type are illegal; just try wir-
ing the output of a function or VI to its input. That’s why G is prop-
erly called an acyclic dataflow programming language. So how did
they do it? If you inspect the GrafcetVIEW subVI terminals, the wiring
really does flow from inputs to outputs in the usual, legal manner, but
the routing makes it seem cyclic. Amazing!
Once your GRAFCET logic diagram has been built into a subVI,
you combine it with some run-time VIs that initialize the digital I/O
hardware and others that actually execute the logic with appropriate
synchronization. GrafcetVIEW uses a series of global variables to pass
information about the logic to the run-time engine VIs. The final dia-
grams are very concise. The package also makes it easy to simulate the
492 Chapter Eighteen

behavior of an application by replacing actual I/O points with boolean

controls and indicators. Such a validation technique is valuable in any
control implementation.
Even if you’re not planning to run your system with GrafcetVIEW,
it makes a good learning and demonstration tool for the language. For
more information on GRAFCET, visit www.tecatlant.fr.

Continuous Control
Continuous control generally implies that a steady-state condition
is reached in a process and that feedback stabilizes the operation over
some prolonged period of time. Single-loop controllers, PLCs, and other
programmable devices are well suited to continuous control tasks, or
you can program LabVIEW to perform the low-level feedback algo-
rithms and have it orchestrate the overall control scheme. LabVIEW
has some advantages, particularly in experimental systems, because it’s
so easy to reconfigure. Also, you can handle tricky linearizations and
complex situations that are quite difficult with dedicated controllers.
Not to mention the free user interface.
Most processes use some form of the PID algorithm as the basis for
feedback control. Gary wrote the original PID VIs in the LabVIEW
PID Control Toolkit with the goal that they should be easy to apply
and easy to modify. Every control engineer has personal preferences as
to which flavor of PID algorithm should be used in any particular situ-
ation. You can easily rewrite the supplied PID functions to incorporate
your favorite algorithm. Just because he programmed this particular
set (which he personally trusts) doesn’t mean it’s always the best for
every application. The three algorithms in the package are
■ PID—an interacting positional PID algorithm with derivative action
on the process variable only
■ PID (Gain Schedule)—similar to the PID, but adds a tuning sched-
ule that helps to optimize response in nonlinear processes
■ PID with Autotuning—adds an autotuning wizard feature that
guides you through the loop tuning process

There is also a Lead/Lag VI that is useful for more advanced control

strategies, such as feedforward control. Lead refers to the phase shift
associated with a single-pole high-pass filter or differentiator, while lag
refers to a low-pass filter or integrator. The Lead/Lag VI is also useful
in simulations where you need to emulate the time-domain response
of a first-order system. We won’t elaborate on the contents of the whole
package here. Just order the toolkit from National Instruments and
read the manual.
 Process Control Applications 493

Setpoint (%) Output

10.00 Direction (F:reverse)

Setpoint (%) Output
Process Variable (%) Override Process Variable (%)
Hysteresis (0 %)
5.00 Normal Override
Hysteresis On-off Controller.vi
Hysteresis (0 %) Direction (F:reverse)

1.00 Reverse

Setpoint (%) Process Variable (%)

Hysteresis (0 %) Override
Normal

Output

Direction (F:reverse)

Figure 18.37 An on/off controller with hysteresis, which functions much like a thermostat. Add
it to your process control library.

You can use the math and logic functions in LabVIEW to implement
almost any other continuous control technique. For instance, on/off or
bang-bang control operates much like the thermostat in your refrig-
erator. To program a simple on/off control scheme, all you need to do
is compare the setpoint with the process variable and drive the output
accordingly. A comparison function does the trick, although you might
want to add hysteresis to prevent short-cycling of the output, as we did
in the Hysteresis On/Off Controller VI in Figure 18.37. This control
algorithm is similar to the ones in the PID Control Toolkit in that it
uses an uninitialized shift register to remember the previous state of
the output.

Designing a control strategy

The first step in designing a continuous control strategy is to sketch a
flowchart of your process, showing control elements (i.e., valves) and
measurements. Then add feedback controllers and any other required
computations, as textbooks on the subject of process control recommend.
The finished product will probably look like some kind of a P&ID dia-
gram. Your goal is to translate this diagram into a working LabVIEW
system. The question is whether to have LabVIEW do the real-time feed-
back control loop calculations or to have an external smart controller
494 Chapter Eighteen

Cascade/feedforward surge tank level control LabVIEW equivalent using PID control VIs
Tld
FT1

Lead-Lag.vi

SP Valve
LC + PID.vi
FT-1
+
∑
SURGE
SP
TANK
LC

LT1
LT-1 FT-2 RUN

FT2

With the PID functions, you can map a control strategy from a textbook diagram to
Figure 18.38
a LabVIEW diagram.

do the job. As we’ve already discussed, the choice depends on perfor-

mance and reliability requirements as well as personal preference.
If you decide to have LabVIEW do the processing, translate the flow-
chart into a LabVIEW block diagram using the PID control VIs with
the math and logic functions of LabVIEW. An example of a translated
control scheme is shown in Figure 18.38. The only elements missing
from this simplified program are the loop tuning parameters and auto/
manual switching.
If you use an external controller, the program is even simpler. All you
will have to do is write settings such as setpoints and tuning parame-
ters and read the status of various measurements from the controller.
You can supplement the functionality of the external controllers with
math and logic in your LabVIEW program. In all cases, there will be
the usual requirements for interlocking and initialization.
Your choice of control topology must complement the process, or the
resulting performance may suffer from instability, overshoot, or long-
term error, and it may be very difficult to tune or optimize. This is
where training in process control strategies, books, and practical expe-
rience come into play. It’s better to choose the right approach, if possi-
ble, rather than force-fitting a generic controller into a difficult process
situation. Here is one very common example, taken from an exchange
Gary had with someone on the info-labview mailgroup.
Question: I am using the PID algorithm to control temperature. I would
like to make a step change in controlled temperature as sharp as possible
without overshoot and ripples.

Seems like a reasonable request, doesn’t it? Well, it turns out that
control engineers have been struggling with this exact problem for the
last 60 years with mixed results and a very wide range of possible solu-
tions, depending upon the exact character of the process, the available
 Process Control Applications 495

control equipment, and the sophistication of the engineer. We can list a

few solutions that you might want to look into.

1. Use feedforward control techniques. Basically, this is where you

characterize your process (temperature versus controller output,
both static and dynamic response) and force the output to move in
the right direction with the proper dynamics to compensate for the
response of the process. Feedforward is quite safe as far as instabil-
ity is concerned. The Lead/Lag VI was included in the PID Control
Toolkit for such control schemes. You can read about it in several
books (Corripio 1990; McMillan 1995; Shinskey 1988). We would use
this technique as a first choice. Note that it’s not available in single-
loop controllers, but LabVIEW can do it.
2. Create a PID tuning schedule. Your process gain and temporal
response vary with absolute temperature, which implies that your
PID tuning must be different at every temperature. Process gain
is the temperature change divided by the controller output change.
Typically, the process gain is much lower at high temperatures, lead-
ing to sluggish response. Ideally, you create a mathematical model of
your process either theoretically or by gathering empirical data. The
model is then loaded into a simulation package (such as MATLAB’s
SimuLink), and you derive optimal control parameters from that. For
simpletons like us, we generally perform an open-loop step response
measurement at several temperatures, as described in the toolkit
manual, and create a table of tuning parameters. Then we use the PID
(Gain Schedule) VI, fed by a cluster containing those PID param-
eters. Figure 18.39 shows what it looks like. The tuning schedule

Gain Scheduling

Scheduling input
Setpoint
Measure PV Update output
Gain schedule (breakpoints) Setpoint PID (Gain AO Update
0 Schedule).vi Channel.vi
100.00

PID parameters
0
Kc 4.8518 Gain Scheduling
init
Ti (min) 0.0211
ms 250 Run
Td (min) 0.0053

Run Setpoint
ON 2.50

Figure 18.39 This is one way to implement a PID tuning schedule. Properly
tuned, it yields good control performance over a wide range of operating
conditions.
496 Chapter Eighteen

can evaluate break points based on the process variable, setpoint,

controller output, or an extra numeric input called the gain schedul-
ing variable. That pretty much covers any situation you’ll encounter.
In some processes, you can get away with constant values for the inte-
gral and derivative terms while varying only the proportional gain.
3. Another technique related to tuning schedules is sometimes called
approach control gain adjustment or start-up management and
is included with some single-loop controllers. Basically, you reduce
the controller’s gain as the error (setpoint minus process variable)
decreases. This can reduce overshoots (sometimes). It generally fails
with processes that have a long dead time.
4. Ramp the setpoint. Many controllers will nicely track a ramped
setpoint without the severe upset normally seen with large step
changes in setpoint. You can add a profile to the ramp to further
improve the response. Typically, the ramp’s rate of change can be
larger at the outset, decreasing as it nears the final value.
5. If you want to use a simple PID algorithm, use as much proportional
gain as the process can stand without oscillation. The proportional
term is your friend. Also be sure to use derivative, which is a sta-
bilizing feedback term. In fact, you can make a PD controller, thus
avoiding the reset (integral) term which encourages overshoot. The
trouble is, there will be some long-term error, which you can cancel
by adding a dc bias to the controller output. The bias is determined
experimentally. For one experiment, we generated lots of small steps
in temperature, recording the process response, then determined a
bias schedule, which can be a piecewise linear fit. Resulting error
is pretty low for an invariant process. An alternative is a control
scheme that switches modes, from PD at start-up to PID at steady
state. Again, a tuning schedule can perform this trick. Look up
batch controllers in Shinskey (1988). Batch systems cannot toler-
ate overshoot.
6. Consider fuzzy logic, predictive, or model-based controllers.
These are advanced control schemes beyond the scope of this book,
but exceptional performance is possible, with enough effort.

Automatic parameter tuning is available in the PID Toolkit and

from VI Engineering. Its Automatic PID Tuner can help you select opti-
mum tuning parameters using a variety of well-known analysis tech-
niques. Note that these tools are not for continuous, online use, so they
won’t automatically solve your dynamic problems.

Scaling input and output values. All the functions in the PID Con-
trol Toolkit have a cluster input that allows you to assign setpoint,
 Process Control Applications 497

process variable, and output ranges. You supply a low and a high value,
and the controller algorithm will normalize that to a zero to 100 per-
cent range or span, which is customarily used for PID calculations.
Here is an example. Assume that you use a multifunction board ana-
log input to acquire a temperature signal with a range of interest span-
ning 0 to 250°C.You should set the sp low value to zero, and the sp high
value to 250 in the PID option cluster. That way, both the setpoint and
process variable will cover the same range. On the output side, perhaps
you want to cover a range of 0 to 10 V. You should assign out low to zero
and out high to 10. That will keep the analog output values within a
sensible range.
Once everything is scaled, the PID tuning parameters will begin to
make sense. For instance, say that you have a temperature transmitter
scaled from −100 to +1200°C. Its span is 1300°C. If the controller has a
proportional band of 10 percent, it means that a measurement error of
130°C relative to the setpoint is just enough to drive the controller out-
put to saturation, if you have scaled your setpoint in a similar manner.
In the PID tuning parameters, they call for a proportional gain value,
which is equal to 100 divided by the proportional band, in percent. So
for this example, the proportional gain would be 10.

PID tricks and tips. All of the PID VIs contain a four-sample FIR low-pass
filter on the process variable. It’s there to reduce high-frequency noise,
which is generally a good thing. However, it also induces some additional
phase shift. This extra lag must be included in any process simulation.
Gary sometimes goes into the VI and rips out the filter subVI, and does
his own filtering as part of the data acquisition process.
Nonlinear integral action is used to help reduce integrator windup
and the associated overshoot in many processes. The error term that is
applied to the integrator is scaled according to the following expression:
error
Error =
1 + 10( error 2/span2)

where error is the setpoint minus the process variable and span is
the controller’s setpoint span. Thus, when the deviation or error is very
large, such as during process start-up, the integral action is reduced to
about 10 percent of its normal value.
Because derivative action is effectively a high-pass filter, high-
frequency noise is emphasized, leading many users to abandon deriva-
tive action altogether. To combat such problems, derivative limiting is
sometimes used. It’s basically a low-pass filter that begins acting at
high frequencies to cancel the more extreme effects of the derivative.
In the PID Toolkit, there is no implicit limit on the derivative action.
498 Chapter Eighteen

However, if you leave the process variable filter in place, there’s a good
chance that it will serve as a nice alternative.

Timing and performance limitations. Always be wary of timing limita-

tions and system performance when you are doing continuous control.
DCS manufacturers rate their I/O controllers in terms of loops per sec-
ond, referring to how many PID calculations can be performed in one
second under average conditions. If the system software is well written,
adding more loops will cause a gradual degradation of performance
rather than an outright failure of any kind. It’s much better to have a
loop running 20 percent slow than not at all. Also, there should be no
way to accidentally upset the steady operation of a continuous control
algorithm. Real processes have plenty of unpredictable features with-
out help from cantankerous software controls.
According to control theory, a sampled control system needs to run
about 10 times faster than the fastest time constant in the plant under
control. For instance, a temperature control loop is probably quite
slow—a time constant of 60 s is common in a small system. In this case,
a cycle time of about 6 s is sufficient. Faster cycling offers little or no
improvement in performance. In fact, running all your control VIs too
fast degrades the overall response time of your LabVIEW application.
If you use the timing functions available in LabVIEW to regulate exe-
cution of a feedback algorithm, be aware of the actual precision avail-
able on your computer—typically 1 ms. Therefore, the fastest practical
loop cycle times are on the order of 10 ms (100 Hz) for most LabVIEW
systems. To go faster, you must obtain better timing information (from
a hardware timer) or run the algorithm on a LabVIEW RT system
or use another external device with suitable performance. By the way,
most industrial single-loop controllers cycle no faster than about
5 Hz (200 ms).
Here is an example of how timing accuracy can affect a control algo-
rithm. A PID algorithm has two time-dependent terms, the integral and
derivative responses. When the algorithm is called, the amount of time
since the last execution, .t, is used in the calculation. If .t is in error, then
the response of the algorithm may also be in error. The error magnitude
depends on the tuning parameters of the PID as well as the state of the
process. If the process is in steady state, then the time-dependent terms
are zero anyway, and the timing error does not matter. But during a
process upset, the response to timing errors can be very hard to predict.
For best results, you had better make sure that your control loops run
with a steady rhythm and at a sufficiently high speed.
The PID Control Toolkit supports either internal or external tim-
ing. Internal timing uses LabVIEW timing functions with the reso-
lution limits previously mentioned. The advantage of this method is
 Process Control Applications 499

that the PID functions keep track of the elapsed time between each
execution. External timing requires you to supply the actual cycle time
(in seconds) to the PID function VI. If you are using the DAQ library,
the actual scan period for an acquisition operation is returned by the
Waveform Scan VI, for instance, and the value is very precise. Each
PID VI has an optional input called dt. If dt is set to a value less than
or equal to zero seconds (the default), internal timing is used. Positive
values are taken as gospel by the PID algorithm.
The DAQ analog I/O example Analog IO Control Loop (hw timed)
is an example of a data acquisition operation where you can place a
PID loop on one of the input channels driving an analog output. The
trick is to wire actual scan rate (in scans per second) from the DAQ
VI, AI Start, through a reciprocal function to the PID VI. This provides
the PID algorithm with an accurate time interval calibration. The data
acquisition board precisely times each data scan so you can be assured
that the While Loop runs at the specified rate. This example should run
reliably and accurately at nearly 1 kHz, including the display.

Trending
The process control industry calls graphs and charts trend displays.
They are further broken down into real-time trends and historical
trends, depending on the timeliness of the data displayed. Exactly
where the transition occurs, nobody agrees. Typically, a historical trend
displays data quite a long time into the past for a process that runs
continuously. A real-time trend is updated frequently and only displays
a fairly recent time history. Naturally, you can blur this distinction
to any degree through crafty programming. Historical trending also
implies archival storage of data on disk for later review while real-time
trending may not use disk files.

Real-time trends
The obvious way to display a real-time trend is to use a Waveform
Chart indicator. The first problem you will encounter with chart indi-
cators is that historical data is displayed only if the panel containing
the chart is showing at all times. As soon as the panel is closed, the old
data is gone. If the chart is updating slowly, it could take quite a long
time before the operator sees a reasonable historical record. A solution
is to write a program that stores the historical data in arrays and then
write the historical data to the chart with the Chart History item in
a Property node. You should take advantage of strip charts whenever
possible because they are simple and efficient and require no program-
ming on your part.
500 Chapter Eighteen

Consider using an XY Graph as a real-time trend display for extra

versatility. You maintain arrays that contain some number of recent
measurements and another array containing accurate timestamps.
Then it’s a simple matter of graphing the data versus the timestamps
whenever a real-time trend needs to be displayed. If the arrays are
stored in global variables, any VI can read and display data from any
channel at any time.
An important concern with this technique is memory management.
If you just keep appending samples to your trending arrays, you will
ultimately run out of memory. Also, the graphs will gradually become
sluggish because of the large numbers of plot points. Some kind of
length-limited array or a data compression technique is required for
long-term operation.
A sophisticated method of trend data management uses a circular
buffer. A circular buffer is an array that you program to act as if it is a
continuous ring of data in memory rather than a straight line of finite
length. Because the array never has to change size or be relocated,
no real-time memory management is required and the performance is
very good. The amount of data you can store is limited only by avail-
able memory. Unfortunately, the programming is rather complex: You
must keep track of where the oldest and newest data are in the buffer
and figure out a way to map the (physical) linear array into a (virtual)
circular one.
The Circular Buffer VI has some interesting features that make
it more useful than the regular strip charts. First, you preallocate all
the required memory by setting the desired number of channels and
samples per channel. Second, you periodically call the VI, supplying
an array that contains the measurements for each channel. These are
written to the next location in the circular buffer. At the same time,
timestamps are recorded in a separate buffer. The timestamps are in
epoch seconds, where zero is 1-Jan-1904. They are internally managed
so you don’t even have to supply a time measurement.
The interesting part comes when you want to read the data. The VI
returns an array of data for any one channel and an array of timestamps
suitable for xy plotting. The number of samples is adjustable, as is the
start time of the returned array. That way, you can scroll back through
time, looking at a window of data rather than trying to zoom in by using
the graph controls. It’s much faster, too, because the graph only has to
display a limited number of points. If the buffer is sized for 10,000 sam-
ples, you might be hesitant to attempt to display them all at once.
Figure 18.40 shows the panel and diagram of an example VI that
uses the Circular Buffer VI for real-time trending. Two independent
While Loops store and retrieve data. Because the Circular Buffer VI is
a kind of global variable, you can call it in read or write mode any place,
 Process Control Applications 501

Plotted Data
2.0

1.5

1.0

0.5

0.0
09:51:55 PM 09:52:20 PM 09:52:44 PM

Data Scroll

Channel to Display
STOP 1

Points in
Write Interval (sec) Read Interval (sec) Graph Window
1 2 50

Acquire and write data to buffer

True = initialize

10 Make sinewaves for data

Channels Write

Circular
1000 buffer.vi Reset stop button
Samples/Chan 1000 stop
stop
Write Interval (sec)

Read data from buffer

True
Read
Plotted Data

Data Scroll X Minimum

Plotted Data
X Maximum
Text Times.vi
Points in
Graph Window
1000
stop
Channel to Display Read Interval (sec)

Figure 18.40 An example of the Circular Buffer real-time trending subVI in action.
Parallel While Loops are used, one to write sinusoidal data for 10 channels, and the
other to read a specified range from a channel for display. Note that the graphed data
can be scrolled back in time.

any time. The data written here is a set of sine waves, one for each
of the 10 channels defined. The buffer is initialized to contain 1000
samples per channel in this example. Every two seconds, a fresh set of
values for all 10 channels is written.
The Data Scroll slider allows you to display a range of data from
the past without disturbing the data recording operation. The Points
502 Chapter Eighteen

in Graph Window control lets you display a selected number of data

points on the graph. To properly display time, we used the graph’s X
Scale Formatting menu to set the scale to time/date. On the diagram,
a Property node sets the x axis minimum and maximum values. This
technique guarantees correct interpretation of timing regardless of
actual sampling intervals because a timestamp is stored with each
sample, and data is always plotted versus timestamps.
Memory usage and performance are very good. There is a noticeably
long delay when you call Circular Buffer with its initialization bool-
ean set to True. At that time, it allocates memory for the entire buffer,
which can be quite large. The actual amount of memory used for the
buffer itself is calculated from
Bytes = ( N + 1) × M × 8
where N is the number of channels and M is the number of samples
per buffer. For instance, 10 channels and 10,000 samples per channel
requires 880,000 bytes of memory. (The 1 in the formula accounts for
the hidden timestamp buffer.) Only one duplication of the data array
is required, and that is also static memory usage, so the actual mem-
ory usage is twice the result from the formula. When you write data,
the Replace Array Element function is used, so no reallocation of
memory is required. When reading, the amount of memory required
is proportional to the number of data points to be displayed, a number
significantly smaller than the size of the buffer for practical usage.
Thanks go to Marty Vasey for his original work on this memory-based
circular buffer concept. He designed it out of necessity: A customer was
threatening bodily harm if they couldn’t scroll back through their real-
time data and do it fast. This program does the trick.

Historical trends
Memory-resident data is fine for real-time trending where speed is
your major objective. But long-term historical data needs to reside on
disk both because you want a permanent record and because disks
generally have more room. You should begin by making an estimate
of the space required for historical data in your application. Consider
the number of channels, recording rates, and how far back in time the
records need to extend. Also, the data format and content will make a
big difference in volume. Finally, you need to decide what means will be
used to access the data. Let’s look at some of your options.
All of the basic file formats discussed in Chapter 7, “Files,”—datalogs,
ASCII text, and proprietary binary formats—are generally applicable
to historical trending. ASCII text has a distinct speed disadvantage,
however, and is probably not suited to high-performance trending
 Process Control Applications 503

where you want to read large blocks of data from files for periodic redis-
play. Surprisingly, many commercial PC-based process control applica-
tions do exactly that, and their plotting speed suffers accordingly. Do
you want to wait several minutes to read a few thousand data points?
Then don’t use text files. On the other hand, it’s nice being able to
directly open a historical trending file with your favorite spreadsheet,
so text files are certainly worth considering when performance is not
too demanding.
LabVIEW datalog files are a better choice for random access histori-
cal trending applications because they are fast, compact, and easy to
program. However, you must remember that only LabVIEW can read
such a format, unless you write custom code for the foreign application
or a LabVIEW translator program that writes a more common file for-
mat for export purposes.

The HIST package. A custom binary file format is the optimum solu-
tion for historical trending. By using a more sophisticated storage algo-
rithm such as a circular buffer or linked list on disk, you can directly
access data from any single channel over any time range. Gary wrote
such a package, called HIST, which he used to sell commercially.
The HIST package actually includes two versions: Fast HIST and
Standard HIST. Fast HIST is based on the real-time circular buffer
that we’ve just discussed, but adds the ability to record data to disk in
either binary or tab-delimited text format. This approach works well
for many data acquisition and process control packages. Based on a
single, integrated subVI, Fast HIST has very low overhead and is capa-
ble of recording 100 channels at about 80 Hz to both the memory-based
circular buffer and to disk. Utility VIs are included for reading data,
which is particularly important for the higher-performance binary for-
mat files.
Standard HIST is based on a suite of VIs that set up the files, store
data, and read data using circular buffers on disk rather than in
memory. At start-up time, you determine how many channels are to
be trended and how many samples are to be saved in the circular buf-
fers on disk. Sampling rates are variable on a per channel basis. The
use of circular buffers permits infinite record lengths without worry
of overflowing your disk. However, this also means that old data will
eventually be overwritten. Through judicious choice of buffer length
and sampling parameters and periodic dumping of data to other files,
you can effectively trend forever. (See Figure 18.41.)
Data compression is another novel feature of Standard HIST. There
are two compression parameters—fencepost and deadband—for
each channel (Figure 18.42). Fencepost is a guaranteed maximum
update time. Deadband is the amount by which a channel’s value
504 Chapter Eighteen

Start of circular buffer

Latest value & timestamp

Pointer to next location;
contains oldest data

End of circular buffer

Figure 18.41 Representation of a circular buffer on disc.
This buffer contains the values and timestamps from a
single channel. Other channels would be located before and
after this one. The program must keep track of pointers to
the start, end, and next locations in the file.

must change before it is updated on disk. The combination of these

two parameters is very flexible. If deadband is zero, then the chan-
nel will be updated each time the Store HIST Data VI is called
(as fast as once per second). If deadband is a very large number, then
the channel will be updated every time the fencepost period passes. If
you choose a moderate deadband value, then transients will be stored
with high fidelity while steady-state conditions will be trended with a
minimum frequency to preserve buffer space. Each time you store data,
you update each channel’s fencepost and deadband (data compression
parameters), giving real-time control over which channels are being

These four samples were

all outside the deadband

Previous value,
± Deadband

Fencepost
(4 time units)
Figure 18.42 Illustration of the action of deadband-fencepost data
compression. Values are stored at guaranteed intervals deter-
mined by the fencepost setting, which is four time units in this
example. Also, a value is stored when it deviates from the last
stored value by greater than +/-deadband (measured in engineer-
ing units).
 Process Control Applications 505

trended at what rates. Similar data compression techniques are found

in DSC and in many other process control packages.
Standard HIST includes all the necessary file services, such as file
creation; in addition, it allows you to restart historical trending with an
existing set of files, picking up right where you left off. This is possible
because the indexing information in the circular buffer data file is kept
in a separate file. Also, you can run up to four separate HIST file sets
simultaneously. That permits you to break up your data as you see fit.
The data files are in a special binary format which makes reading the
data in other programs impossible without writing special programs.
For this reason, a VI called HIST Data to Text File is included to do a
translation to tab-delimited text within LabVIEW. You could also write
your own translation VIs since the data from every channel is readily
available in ordinary LabVIEW arrays.

Statistical process control (SPC)

What do you do with tons of data you’ve collected from monitoring your
process or test system? In manufacturing situations, statistical pro-
cess control (SPC), also known as statistical quality control (SQC),
techniques are commonly used to emphasize the range over which the
process is in control and what the process is capable of producing. Once
a process is in control, it can be improved. Statistical techniques are
used to monitor the mean and variability of a process. There is always
some random variability in a process, but there may also be other non-
random causes (i.e., systematic errors) present which must be identi-
fied and corrected.
SPC techniques aid in identifying whether or not special causes are
present so that the process is corrected only when necessary. A pro-
cess parameter is usually plotted against control limits. If the process
parameter exceeds these control limits, there is a very high probabil-
ity that the process is not in control, which means that there is some
special cause present in the process. This special cause could be a new
piece of equipment, a new untrained operator, or sundry other prob-
lems. Once a process is in control, SPC techniques can predict the yield
or defect rate (in parts per million, for example) of the process. There
are also techniques for designing experiments in order to improve the
process capability. SPC techniques are pretty popular these days with
everybody trying to meet ISO 9000 quality criteria. Some good refer-
ences for applying SPC techniques are listed at the end of this chapter
(Montgomery 1992; Wheeler and Chambers 1992).
The Statistical Process Control Tools, available from National
Instruments as part of the Enterprise Connectivity Toolset, makes a
great starting point for any SPC application (Figure 18.43). It comes
506 Chapter Eighteen

X-bar Chart Point exceeded control limits X-bar Histogram # bins 12

74.03 UCL 600

74.02 CL
500
74.01 LCL
400
74.00 X-bar Chart Limits
300
73.99 UCLx 74.015

CLx 74.002 200

73.98
LCLx 73.989
Sample Number 314 100

Range Chart Point exceeded control limits Range 0

73.95 74.00 74.05
0.08 UCL

CL min 73.95 max 74.05

0.06
LCL
0.04 Upper Specification Limit
Range Chart Limits 74.05
0.02
UCLr 0.048 Lower Specification Limit

0.00 CLr 0.023 73.95 # Samples 100

LCLr 0.000 COMPUTE

Process Shift detected at Sample # 276 in X-bar Chart
PROCESS CAPABILITY

Calc Mode CALCULATE VIEW CONTROL 20

For more info use
MANUAL CONTROL LIMITS LIMIT CALCULATION # Samples for Limits STOP
the LV Help Window

Figure 18.43 The SPC Toolkit demo VI displays several of the built-in chart types and statistical
analysis techniques included in the package.

with a comprehensive set of VIs and examples and a thorough manual.

Common presentation techniques for SPC are control charts, Pareto
analysis, histograms, and other statistical calculations. To make your
life easier, the toolkit includes a set of graph and table controls that
are customized for these standard presentations. Special VIs compute
process statistics and prepare data for display. Most of the VIs expect
arrays of data for processing, so you can either load historical data from
disk or maintain recent data in memory, perhaps in a circular buffer.
Application is straightforward, but be warned that the world of SPC
can be daunting to the newcomer. Don’t expect magical improvements
in your process unless you study the references or get clear instruc-
tions from an expert.

Alarms
A vital function of any process control system is to alert the opera-
tor when important parameters have deviated outside specified lim-
its. Any signal can be a source of an alarm condition, whether it’s a
measurement from a transducer, a calculated value such as an SQC
control limit, or the status of an output. Alarms are generally classified
 Process Control Applications 507

High Limit High alarm

In Range and Coerce

Analog signal Out of range alarm

Low Limit Low alarm

Pressure switch
WARNING
high limit

Automatic mode Caution

Normal

Figure 18.44 Using comparison functions to detect alarm

conditions on analog signals (top). Digital signals are com-
bined with boolean logic to generate alarm states (bottom).

by severity, such as informative, caution, or warning, and there are a

variety of audiovisual alarm presentation methods available.
You should start by listing the signals and conditions in your pro-
cess that require some form of alarm response and determining the
appropriate responses. For analog signals and computed values, use
the comparison functions (Figure 18.44, top) to generate boolean
alarm flags. Digital signals, such as contact closure inputs, are com-
bined with the logic functions, such as AND, OR, and NOT, to generate
flags (Figure 18.44, bottom). Naturally, you can combine these tech-
niques as required.
Change-of-state detection is another important function that you
may need for alarm generation. This enables you to react to a signal that
is generally constant but occasionally undergoes a sudden deviation.
To detect a change of state, the VI must remember the previous state
of the signal, which implies the use of a Shift Register as shown in
Figure 18.45.

0=init False Input True

RUN INIT
Change of state

On to Off

Off to On

Figure 18.45 A change of state can be detected by comparing a boolean’s present value with its
previous value stored in a Shift Register. Three different comparisons are shown here. This would
make a reasonable subVI for general use.
508 Chapter Eighteen

Input

last state
Off to On
detected

Latched
Off -> On
Alarm
Acknowledge

Figure 18.46 This example latches an alarm state

until the Alarm Acknowledge input is set to True.

Whenever the input changes state, the appropriate boolean indica-

tors will be set to True. Because the logic functions are polymorphic,
you can change all the boolean controls and indicators to arrays of bool-
eans without rewiring these last two VIs (Figs. 18.44 and 18.45). This
allows you to process several channels at once. If you do modify this
example to use arrays, the initial array inside the Case must contain
the desired number of elements.
An extension of this technique employs latching, used when you
need the boolean indicator to remain in the alarm state until a sepa-
rate reset signal clears the alarm. The reset function is called alarm
acknowledge in process control terminology. In Figure 18.46, the
upper Shift Register detects an off-to-on change of state, as in the pre-
vious example. The lower Shift Register is set to True when a change
of state occurs, and stays that way until Alarm Acknowledge is True.
Again, the inputs and outputs could be arrays.

Using an alarm handler

Detecting alarms is the easy part. But there is a potential data distri-
bution problem just like we encountered with input and output signals.
If your system is fairly complex, alarms may be generated by several
VIs with a need for display elsewhere. An alarm handler adds a great
deal of versatility to your process control system, much as the I/O han-
dlers do for analog and digital signals. It acts as a centralized clearing-
house for alarm messages and status information.

Global Alarm Queue VI. One way to handle distributed alarm generation
is to store alarm messages in a global queue. The approach is similar
to the one we used with an output handler VI. Multiple VIs can deposit
alarm messages in the queue for later retrieval by the alarm handler.
 Process Control Applications 509

Enqueue? valid False

Alarm Message In Alarm Message Out Dequeue

ALARM:Alarm Message Queue.vi the queue

1 valid
Enqueue? valid
•enqueue Enqueue?
0
•dequeue Alarm
Message Out
Alarm Message In Alarm Message Out

The Message The Message True

Enqueue
11:37:44 - TIC-101 11:37:44 - TIC-101
HIGH ALARM HIGH ALARM
Destination Code Destination Code
7 7

Alarm Message In

Figure 18.47 A global queue stores alarm messages. A message consists of a cluster containing a
string and a numeric. In Enqueue mode, elements are appended to the array carried in the Shift
Register. In Dequeue mode, elements are removed one at a time.

A queue guarantees that the oldest messages are handled first and
that there is no chance that a message will be missed because of a tim-
ing error. The alarm message queue in Figure 18.47 was lifted directly
from the Global Queue utility VI that was supplied with older versions
of LabVIEW. All we did is change the original numeric inputs and out-
puts to clusters. The clusters contain a message string and a numeric
that tells the alarm handler what to do with the message. You could
add other items as necessary. This is an unbounded queue which grows
without limits.
Alarms can be generated anywhere in your VI hierarchy, but the I/O
handlers may be the best places to do so because they have full access
to most of the signals that you would want to alarm. You can combine
the output of an alarm detector with information from your configura-
tion database to produce a suitable alarm message based on the alarm
condition. For instance, one channel may only need a high-limit alarm
while another needs both high- and low-limit alarms. And the contents
of the message will probably be different for each case. All of these
dependencies can be carried along in the configuration. Once you have
formatted the message, deposit it in the global alarm queue.

Alarm Handler VI. Once the alarm messages are queued up, an alarm
handler can dequeue them asynchronously and then report or distribute
them as required. The alarm handler in Figure 18.48 performs two such
actions: (1) It reports each alarm by one of four means as determined
by the bits set in the Destination Code number that’s part of the mes-
sage cluster; and (2) it appends message strings to a string indicator,
510 Chapter Eighteen

No. of Messages to Keep Alarms noted

5 1
No. of Messages
Alarm Messages to Keep Alarms noted
11:37:44 - TIC-101 HIGH ALARM ACKNOWLEDGE Alarm Messages
Alarm Handler.vi

ACKNOWLEDGE

Dequeue True
messages
Valid Queue -- Process 3 [0..3]
until all gone

Dialog
dequeue Message
0
Dest Code True
Alarm Message Queue
1
Alarms
noted 2
If "ACK" button
is pressed, set 3
messages to
empty string. Alarm
No. of Messages to Keep Keep N Lines Messages

Current alarm messages

ACKNOWLEDGE

Figure 18.48 An alarm handler VI. It reads messages from the global alarm queue and reports them
according to the bits set in the Destination code. Messages are appended to the Current Messages
string. The subVI, Keep N Lines, keeps several of the latest messages and throws away the older ones
to conserve space in the indicator for current messages.

Current Messages, for direct display. This alarm handler would be

called periodically by a high-level VI that contains the alarm message
display.
There are other operations that an alarm handler might perform. If
many other VIs need to display alarm messages or status information,
you could have the alarm handler copy messages to a global variable
for reading and display elsewhere. The handler could also call subVIs
or set special global flags that cause some control operations to occur.
Another way to manage alarms is to use a global database as a repos-
itory for all your data. Each I/O point is represented by a cluster of
information, and part of that information is a set of flags that repre-
sents alarm limits and alarm status flags. Alarm generators (probably
part of the I/O handlers) set the alarm status flags based on the current
value and the desired limits. Since the database is globally accessible,
any VI can read the status of any alarm and take appropriate action.
 Process Control Applications 511

All of this is limited by the real-time performance of the database, so

you should approach with caution. If you use DSC, all aspects of alarm
handling described here will be taken care of automatically.

Techniques for operator notification

The fun part of this whole alarm business is notifying the operator. You
can use all kinds of LabVIEW indicators, log messages to files, make
sounds, or use external annunciator hardware. Human factors special-
ists report that a consistent and well-thought-out approach to alarm
presentation is vital to the safe operation of modern control systems.
Everything from the choice of color to the wording of messages to the
physical location of the alarm readouts deserves your attention early
in the design phase. When automatic controls fail to respond properly,
it’s up to the operator to take over, and he or she needs to be alerted in
a reliable fashion.
Boolean indicators are simple and effective alarm annunciators.
Besides the built-in versions, you can paste in graphics for the True
and/or False cases for any of the boolean indicators. Attention-getting
colors or shapes, icons, and descriptive text are all valuable ideas for
alarm presentation.
You can log alarm messages to a file to provide a permanent record.
Each time an alarm is generated, the alarm handler can call a subVI
that adds the time and date to the message then appends that string
to a preexisting text file. Useful information to include in the message
includes the tag name, the present value, the nature of the alarm, and
whether the alarm has just occurred or has been cleared. This same file
can also be used to log other operational data and important system
events, such as cycle start/stop times, mode changes, and so forth. DSC
stores alarms and other events in an event log file. SubVIs are avail-
able to access and display information stored in the event log.
The same information that goes to a file can also be sent directly to
a printer. This is a really good use for all those old dot-matrix serial
printers you have lying around. Since the printer is just a serial instru-
ment, it’s a simple matter to use the Serial Port Write VI to send it
an ASCII string. If you want to get fancy, look in your printer’s manual
and find out about the special escape codes that control the style of the
output. You could write a driver VI that formats each line to emphasize
certain parts of the message. As a bonus, dot-matrix printers also serve
as an audible alarm annunciator if located near the operator’s station.
When the control room printer starts making lots of noise, you know
you’re in for some excitement.
Audible alarms can be helpful or an outright nuisance. Traditional
control rooms and DCSs had a snotty-sounding buzzer for some alarms,
512 Chapter Eighteen

and maybe a big bell or Klaxon for real emergencies. If the system
engineer programs too many alarms to trigger the buzzer, it quickly
becomes a sore point with the operators. However, sound does have its
place, especially in situations where the operator can’t see the display.
You could hook up a buzzer or something to a digital output device or
make use of the more advanced sound recording and playback capabili-
ties of your computer (see Chapter 20, “Data Visualization, Imaging,
and Sound”). LabVIEW has a set of sound VIs that work on all plat-
forms and serve as annunciators.
Commercial alarm annunciator panels are popular in industry
because they are easy to understand and use and are modestly priced.
You can configure these units with a variety of colored indicators that
include highly visible labels. They are rugged and meant for use on the
factory floor. Hathaway/Beta Corporation makes several models, rang-
ing from a simple collection of lamps to digitally programmed units.
That covers sight and sound; what about our other senses? Use your
imagination. LabVIEW has the ability to control most any actuator.
Maybe you could spray some odoriferous compound into the air or drib-
ble something interesting into the supervisor’s coffee.

Bibliography
Boyer, S. A.: SCADA: Supervisory Control and Data Acquisition, ISA Press, Raleigh, N.C.,
1993.
Bryan, Luis A., and E. A. Bryan: Programmable Controllers: Theory and Implementation,
C Industrial Text Company, Atlanta, GA, 1988.
Corripio, Armondo B.: Tuning of Industrial Control System,. ISA Press, Raleigh, N.C.,
1990.
Hughes, Thomas A.: Measurement and Control Basics, ISA Press, Raleigh, N.C., 1988.
Johnson, Gary (Ed.): LabVIEW Power Programming, McGraw-Hill, New York, 1998.
McMillan, Gregory K.: Advanced Temperature Control, ISA Press, Raleigh, N.C., 1995.
Montgomery, Douglas C.: Introduction to Statistical Quality Control, Wiley, New York,
1992.
Shinskey, F. G.: Process Control Systems, McGraw-Hill, New York, 1988.
Travis, Jeffrey: Internet Applications in LabVIEW, Prentice-Hall, Upper Saddle River,
N.J., 2000.
Wheeler, Donald J., and D. S. Chambers: Understanding Statistical Process Control,
2d ed., SPC Press, Knoxville, Tenn., 1992.
 Chapter

Physics Applications
19
Physics is Phun, they told us in Physics 101, and, by golly, they were
right! Once we got started at Lawrence Livermore National Laboratory
(LLNL), where there are plenty of physics experiments going on,
we found out how interesting the business of instrumenting such
an experiment can be. One problem we discovered is just how little
material is available in the way of instructional guides for the budding
diagnostic engineer. Unfortunately, there isn’t enough space to do a
complete brain dump in this chapter. What we will pass along are a
few references: Bologna and Vincelli (1983); and Mass and Brueckner
(1965). Particularly, we’ve gotten a lot of good tips and application
notes from the makers of specialized instruments (LeCroy Corporation
2000). Like National Instruments, they’re all in the business of selling
equipment, and the more they educate their customers, the more
equipment they are likely to sell. So, start by collecting catalogs and
look for goodies like sample applications inside. Then, get to know your
local sales representatives, and ask them how to use their products.
Having an experienced experimental physicist or engineer on your
project is a big help, too.
We’re going to treat the subject of physics in its broadest sense for the
purpose of discussing LabVIEW programming techniques. The common
threads among these unusual applications are that they use uncon-
ventional sensors, signal conditioning, and data acquisition equipment,
and often involve very large data sets. Even if you’re not involved in
physics research, you are sure to find some interesting ideas in this
chapter.
Remember that the whole reason for investing in automated data
acquisition and control is to improve the quality of the experiment. You
can do this by improving the quality and accuracy of the recorded data

513
514 Chapter Nineteen

and by improving the operating conditions of the experiment. Calibrat-

ing transducers, instruments, and recording devices as well as moni-
toring and stabilizing the physical parameters of the experiment all
lead to better results. Having a flexible tool such as LabVIEW makes
these goals much easier to achieve than in the past. Twenty years ago,
we had computers, but they were so cumbersome to configure that the
researcher needed a large staff just to support simple data acquisition.
That meant less money and time available for improving the experi-
ment, examining the data, and doing physics.

Special Hardware
In stark contrast to ordinary industrial situations, physics experiments
are, by their nature, involved with exotic measurement techniques and
apparatus. We feel lucky to come across a plain, old low-frequency pres-
sure transducer or thermocouple. More often, we’re asked to measure
microamps of current at high frequencies, riding on a 35-kV dc potential.
Needless to say, some fairly exotic signal conditioning is required. Also,
some specialized data acquisition hardware, much of which is rarely
seen outside of the physics lab, must become part of the researcher’s
repertoire. Thankfully, LabVIEW is flexible enough to accommodate
these unusual instrumentation needs.

Signal conditioning
High-voltage, high-current, and high-frequency measurements require
specialized signal conditioning and acquisition equipment. The sources
of the signals, though wide and varied, are important only as far as their
electrical characteristics are concerned. Interfacing the instrument to
the data acquisition equipment is a critical design step. Special ampli-
fiers, attenuators, delay generators, matching networks, and overload
protection are important parts of the physics diagnostician’s arsenal.

High-voltage measurements. To measure high-voltage signals, you will

need to reduce the magnitude of the signal to something that your analog
to digital converter (ADC) can safely and accurately accommodate. For
most signals, resistive voltage dividers are the easiest to use. You can
buy commercial high-voltage probes of several types. Those intended
for routine ac and dc voltmeters, such as those made by Fluke-Phillips,
are sufficient for frequencies up to about 60 Hz. Special high-voltage
oscilloscope probes such as the Tektronix P6015 are available with
divider ratios of 10, 100, and 1000 to 1 and are useful up to 20 kV dc, at
frequencies up to 75 MHz, and with input impedances of 10 to 100 MΩ.
All you have to do is match the output impedance (and capacitance, for
 Physics Applications 515

high frequencies) to that of your data acquisition equipment, perhaps

by using a suitable amplifier. A ’scope probe with a 1-MΩ resistor for
termination works fine with a plug-in data acquisition board.
For measurements up to hundreds of kilovolts, you can buy high-
voltage probes from Ross Engineering and others. These probes are
compact (considering their voltage ratings) and are intended for per-
manent installation. As with ’scope probes, you have to match the out-
put impedance for highest accuracy (Figure 19.1).
Pulsed high voltages can be challenging because of their extra high-
frequency content. Resistive probes incorporate frequency compensation
capacitors to flatten frequency response (as always, proper matching at
the output is paramount). Capacitive dividers and custom-made pulse
transformers may be optimum in certain cases.
If the measurement you are making is part of a floating system
(that is, it is not referenced to ground), you will also need an isola-
tion amplifier. Most commercial isolation amplifiers operate only at
lower voltages, so they must be inserted after a voltage divider. Some-
times, you can float an entire measurement system at high voltage.
This requires an isolated power supply for the electronics and typically
incorporates a fiber-optic link for communications. Many accelerator
experiments have been constructed around such systems.
Safety is your number one concern in high-voltage system design.
First, protect the personnel, then protect the equipment, then protect the
data. Floating systems are probably the most dangerous because they
tempt you to reach in and turn a knob—a sure way to an early grave.
Cover all high-voltage conductors with adequate insulation, and mark all
enclosures with warning labels. Enclosures should be interlocked with

V in Part of high-voltage probe

C1
+ R1
Input protection
High-voltage
source
C2
R2 V out Amplifier
–
To ADC

System low side of power supply

ground

R2 Compensated when: R1 C1 = R2 C2
V out = V in C1 represents stray capacitance
R1 + R2
Figure 19.1 General application of a high-voltage probe. Frequency compensation is
only needed for ac measurements. Always make sure that your amplifier doesn’t load
down the voltage divider.
516 Chapter Nineteen

safety devices that shut down and crowbar (short-circuit) the source of
high voltage. Cables emerging from high-voltage equipment must be
properly grounded to prevent accidental energization. At the input to
your signal conditioning equipment, add overvoltage protection devices
such as current-limiting resistors, zener diodes, Transorbs, metal-oxide
varistors (MOVs), and so forth. Protect all inputs, because improper
connections and transients have a way of creeping in and zapping your
expensive equipment . . . and it always happens two minutes before a
scheduled experiment.

Current measurements. It seems that we spend more time measur-

ing electrical currents than almost anything else in the physics lab.
Whether it’s high or low current, ac or dc, the measurement is always
more complex than expected.
AC and pulse currents are often the easiest to measure because
you can use a current transformer. Simple robust devices, current
transformers rely on a dynamic magnetic field to couple a fraction
of the measured current in the primary winding into the secondary
winding. Pearson Electronics makes a wide variety of solenoidal current
transformers, usable up to tens of thousands of amps and 20 MHz.
The output has an impedance of 50 Ω, so it matches well to wideband
amplifiers and digitizer inputs. There are two limitations to current
transformers. First, DC current tends to prematurely saturate the
core, resulting in distortion and other amplitude errors. Second, these
transformers are naturally ac-coupled, so you have to be aware of the
low-frequency response limit of your transformer. Specifically, dc and
low-frequency ac information will be lost. On the other hand, there is
no concern about contact with high dc voltages unless the transformer
insulation breaks down. Note that the output of a current transformer
is an image of the current waveform, not an equivalent rms or average
value. Therefore, you must record waveforms, not low-speed dc signals.
Hall Effect devices are used to make clamp-on dc-coupled current
probes. These solid-state devices respond to the instantaneous magnetic
field intensity surrounding a conductor, producing a proportional out-
put. One commercial Hall Effect current probe is the Tektronix AM503S.
With it, you can measure AC or DC currents from milliamps to 500 A
at frequencies up to 50 MHz, and to do this you merely clamp the probe
around the conductor. Though expensive, they are real problem solvers.
Current shunts, also known as current viewing resistors, are the
simplest, and often cheapest, current transducers, consisting of a resis-
tor that is placed in series with the source of current. The resulting volt-
age drop follows Ohm’s Law, where the voltage is equal to the current
multiplied by the resistance. You can make your own shunt for low cur-
rents by inserting a low-value resistor in the circuit. For high currents,
 Physics Applications 517

System
+ under test
Input protection
Power
supply
Current Current Amplifier
– Shunt
To ADC

System low side of power supply

ground
Figure 19.2 Signal conditioning for a current shunt installed on the preferred side
of the system under test.

up to thousands of amps, there are commercially made shunts avail-

able. Standard outputs are usually 50 or 100 mV at full current, so
you may need an amplifier if your ADC system can’t handle low-level
signals. Try to connect the shunt on the low side, or ground return, of
the power source to eliminate the common-mode voltage (Figure 19.2).
Note that this method is only feasible if the system under test or the
power supply can be isolated from ground. If you must wire the shunt
in series with the high side of the power source, you need an isolation
amplifier (Figure 19.3). Unfortunately, isolation amplifiers are expen-
sive, and they generally don’t have very wide bandwidths.

High-frequency measurements. Many experiments involve the detec-

tion of pulsed phenomena, such as nuclear particle interactions, lasers,

high side of power supply

+
Current Current
+ Shunt
To ADC
Power –
supply
System Isolation
– under test amplifier

System
ground
Figure 19.3 This example shows an isolation amplifier with a high side
current shunt. The common-mode voltage is that which appears across
the system under test, and may exceed the capability of nonisolated
amplifiers.
518 Chapter Nineteen

and explosions. The detector may sense ion currents or some electri-
cal event in a direct manner, or it may use a multistep process, such
as the conversion of particle energy to light, then light to an electri-
cal signal through the use of a photomultiplier or photodiode. Making
quantitative measurements on fast, dynamic phenomena is quite chal-
lenging, even with the latest equipment, because there are so many
second-order effects that you must consider. High-frequency losses due
to stray capacitance, cable dispersion, and reflections from improperly
matched transmission lines can severely distort a critical waveform.
It’s a rather complex subject. If you’re not well versed in the area of
pulsed diagnostics, your best bet is to find someone who is.
There are a number of instruments and ancillary devices that you
will often see in high-frequency systems. For data acquisition, you have
a choice of acquiring the entire waveform or just measuring some par-
ticular characteristic in real time. Waveform acquisition implies the
use of a fast transient digitizer or digitizing oscilloscope, for which
you can no doubt find a LabVIEW driver to upload the data. Later in
this chapter, we’ll look at waveform acquisition in detail. Some experi-
ments, such as those involving particle drift chambers, depend only on
time interval or pulse coincidence measurements. Then you can use
a time interval meter, or time-to-digital converter (TDC), a device
that directly measures the time between events. Another specialized
instrument is the boxcar averager, or gated integrator. This is an
analog instrument that averages the signal over a short gate interval
and then optionally averages the measurements from many gates. For
periodic signals, you can use a boxcar averager with a low-speed ADC
to reconstruct very high frequency waveforms.

CAMAC
CAMAC (Computer Automated Measurement and Control) is an old,
reliable, and somewhat outdated standard for data acquisition in the
world of high-energy physics research. It remains a player in research
labs all over the world. LabVIEW has instrument drivers available for
use with a variety of CAMAC instruments and controllers, particularly
those from Kinetic Systems and LeCroy. Because other hardware plat-
forms have replaced CAMAC, we’re no longer going to say much about
it. If you want to read more, you can probably find a previous edition of
this book in which there’s a long CAMAC section.

Other I/O hardware

There are some other I/O and signal-conditioning standards that you
may run into in the world of physics research. And, of course, there really
is no limit on what you can use; it’s just that there are some specialized
 Physics Applications 519

functions that have been implemented over the years that aren’t always
commonly available.

FASTBUS. FASTBUS (ANSI/IEEE 960-1986) represents the fastest,

highest-density data acquisition hardware available today. It is a modu-
lar standard featuring a 32-bit address and data bus, an asynchronous
ECL backplane capable of 80 MHz, and the ability to use multiple mas-
ters or controllers. It’s in fairly wide use in high-energy physics at such
institutions as CERN in Switzerland. It’s a sophisticated I/O system,
but it’s not for the casual user or those with miniature budgets. But if
you get involved with the right laboratory, you may well see FASTBUS
equipment. Currently, you can interface it to a LabVIEW system by
using CAMAC as an intermediary or with a PCI bus interface, either of
which is perfectly acceptable.

VME. The VME bus is a high-performance, general-purpose computing

platform. It’s an open standard with modules available from hundreds
of manufacturers offering state-of-the-art CPUs and peripherals as
well as a wide variety of I/O hardware. Interfaces for everything from
analog I/O to image processing hardware to exotic military test equip-
ment is available in VME format. You can use LabVIEW on VME-based
CPUs that run a suitable operating system, such as Windows. Or, you
can use an MXI interface to connect a desktop machine to a VME back-
plane. An example is the National Instruments VMEPCI8000, which
connects a PCI-bus computer to a B-size VME system, making your
computer appear as if it were plugged into the VME backplane. Your
computer can directly access the VME address space, and VME bus
masters can directly access the computer’s memory and resources. Up to
eight MXI-based devices, such as VXI crates, can be daisy-chained into
such a system. To access devices on the VME bus, you use NI-VXI/VISA
drivers. VISA and VXI drivers are discussed in detail in Chapter 11,
“Instrument Driver Development Techniques.”

VXI. VXI is taking over the title of workhorse data acquisition interface
for physics applications, particularly when high-performance transient
digitizers are required. It’s well planned and supported by dozens of
major manufacturers, and offers many of the basic functions you need.
And, of course, there is excellent LabVIEW support. Drivers are avail-
able for many instruments and PCs can be installed right in the crate.

NIM. Another older standard still in use in the physics community

is the NIM (Nuclear Instrumentation Manufacturers) module
format, originally established in 1964. It’s not an I/O subsystem like
CAMAC or VXI; instead, it’s a modular signal conditioning standard.
520 Chapter Nineteen

Modules are either full- or half-height and plug into a powered NIM
bin with limited backplane interconnections. Because the CAMAC for-
mat was derived from the earlier NIM standard, NIM modules can
plug into a CAMAC crate with the use of a simple adapter.
Many modern instruments are still based on NIM modules, partic-
ularly nuclear particle detectors, pulse height analyzers, and boxcar
averagers such as the Stanford Research SR250. We still use them in
the lab because there are so many nice functions available in this com-
pact format, such as amplifiers, trigger discriminators, trigger fan-outs,
and clock generators.

Field and Plasma Diagnostics

Among the many low-speed applications of LabVIEW in physics are
those that deal with field mapping and DC plasma diagnostics.
These applications combine data acquisition with motion control for
field mapping and the generation of ramped potentials for plasma
diagnostics. There are many variations on these experiments, so we’ll
just cover a few simple applications that might give you some insight
as to how LabVIEW might help out in your lab.

Step-and-measure experiments
When you have a system that generates a static field or a steady-state
beam, you probably will want to map its intensity in one, two, or three
dimensions and maybe over time as well. We call these step-and-
measure experiments because they generally involve a cyclic procedure
that moves a sensor to a known position, makes a measurement, moves,
measures, and so forth, until the region of interest is entirely mapped.
Some type of motion control hardware is required, along with a
suitable probe or sensor to detect the phenomenon of interest.

Motion control systems. There are many actuators that you can use to
move things around under computer control—the actuators that make
robotics and numerically controlled machines possible.
For simple two-position operations, a pneumatic cylinder can be con-
trolled by a solenoid valve that you turn on and off from a digital out-
put port. Its piston moves at a velocity determined by the driving air
pressure and the load. Electromagnetic solenoids can also move small
objects back and forth through a limited range.
But more important for your step-and-measure experiments is the
ability to move from one location to another with high precision. Two
kinds of motors are commonly used in these applications, stepper
motors and servo motors. Think of a stepper motor as a kind of
 Physics Applications 521

digital motor. It uses an array of permanent magnets and coils arranged

in such a way that the armature (the part that moves) tends to snap from
one location to the other in response to changes in the polarity of the DC
current that is applied to each coil. Stepper motors are available with
anywhere from 4 to 200 steps per revolution. Linear stepper motors are
also available. Steppers have the advantage of simplicity and reasonable
precision. Their main disadvantages are that you can’t position them
with infinite resolution and their speed is somewhat limited, typically
1000 to 2000 steps per second. There is one option for increasing
resolution, called microstepping. With microstepping, the effective
steps per revolution can be increased 10 times, and sometimes more.
This requires special control hardware and sometimes compromises the
performance of the motor with regard to torque.
Servomotors are similar to ordinary motors except that they are
intended for variable-speed operation, down to and including stalled,
for indefinite periods. Both AC and DC servo motors are in wide use.
A servo system also includes a positional or velocity feedback device,
such as a shaft encoder or tachometer, to precisely control the motion of
the shaft. (Such feedback may also be used with stepper motors.) With
the use of feedback, servo motor systems offer extremely high precision
and the ability to run at high speed. Their disadvantage is the require-
ment for somewhat complex control circuitry.
The next item you need to select is a motion controller (Figure 19.4).
It may take the form of a packaged system with RS-232 GPIB; or Ether-
net communications from companies such as Gallil, Parker Compumo-
tor, or Klinger; or as plug-in boards for the PC from companies such as
National Instruments and Parker Compumotor. Motor controllers are

Amplifier or
Motion
Controller Power Supply

Motion
Commands
IB Fe
GP ed
or ba
al Encoder
Seri ck
or Tachometer
n
tio s
Mo and Po
m m we
C o r

Motor

Plug-in Motion
Control Board
Figure 19.4 A typical motion control system consists of a computer, a controller (internal or external), an
amplifier or power supply, and a motor, perhaps with an encoder or tachometer for feedback.
522 Chapter Nineteen

also available in various modular formats such as CAMAC, VME, and

VXI. LabVIEW drivers are available for many of these devices; we’ll
discuss some of them later. With any motor, you also need a power sup-
ply—also known as an amplifier—which is a power electronics package
to drive the coils of the motor in an appropriate fashion. The amplifier
supplies current to the motor, provides signal conditioning for encod-
ers or tachometers, and isolates the sensitive I/O ports of the motion
controller from the harsh external world.
About the easiest way to integrate motion control into a LabVIEW
application is to use the plug-in motion control boards and drivers
from National Instruments. Several years ago, National Instruments
acquired nuLogic, a company famous for its Macintosh- and Windows-
compatible motion boards and LabVIEW drivers. Boards are available
that support up to four axes of steppers and servos, in PCI, Compact-
PCI, and PXI formats and a stand-alone controller with a FireWire
(IEEE-1394) interface. The company also offers a line of stepper and
servo power supplies, or you can wire the boards up to another man-
ufacturer’s power supply. Most of the boards also include general-
purpose analog and digital I/O.
The high-performance motion boards (FlexMotion 7344 series) are
highly programmable, with onboard real-time processors that allow you
to write sophisticated motion routines and have them execute without
intervention from the host computer. Figure 19.5 shows an example of
a FlexMotion application. As with any programmable motion controller,
you can load a sequence of commands into a buffer on the FlexMotion
board and then have it execute that sequence autonomously. One of the
first VIs in the sequence, Begin Program Storage, tells the board that
all subsequent commands are to be stored, not executed immediately.
Several trajectory commands are sent next, including acceleration,
deceleration, velocity, and so forth. Finally the End Program Storage VI
tells the board to terminate command storage. The sequence can now be
initiated through software by calling the Run Program VI, or you can
configure the board to run the program when a hardware trigger occurs.

Axis
Load
Initialize Begin Program Load Target Acceleration/ Load Start End Program
Controller.flx Storage.flx Position.flx Deceleration.flx Velocity.flx Motion.flx Storage.flx

Board ID

Accel/
Target Pos 10000 5000 Vel 100000
decel

Figure 19.5 This example stores a motion sequence for later execution on a FlexMotion board. The
stored program can be quite complex, and it can run without intervention from LabVIEW.
 Physics Applications 523

A few other motion control systems are also supported by LabVIEW

drivers. In the driver library, you’ll find VIs for several models from
Aerotech Unidex and Newport Corporation. For other products,
check with your favorite manufacturer.

Reading encoders. A recurring theme in the world of motion control is

reading an encoder. Encoders can be of the incremental or absolute
type. Incremental encoders produce a continuous train of pulses when
moved while absolute encoders have a parallel output that directly
tells you the absolute position of a shaft or linear actuator. Incremental
encoders typically have two outputs that are 90 degrees out of phase,
in which case they are called quadrature encoders. The phase shift per-
mits determination of direction, assuming that you have the proper
readout logic.
To read a quadrature encoder with LabVIEW, you have a few choices.
First, you can use any of the motion control products described previ-
ously. The driver VIs always have a way to read the encoder position,
and they also typically include direct measurement of velocity. Second,
you can try to use a special-purpose integrated circuit that converts
quadrature pulses to a format suited to a regular DAQ board with
counter-timers. Application note AN084, Using Quadrature Encod-
ers with E Series DAQ Boards, shows how it’s done. If that sounds
troublesome, consider buying the BNC-2120 accessory for an E-series
DAQ board. It includes the necessary circuitry, and there’s an example
VI that does the measurement.
Here’s a better solution: Use one of the timing I/O boards from
National Instruments. The 660x family of counter-timer boards has an
exceptionally versatile chip onboard, the NI-TIO, that has the ability to
directly interpret quadrature encoder signals, among other things. On
National Instruments’ Web site you’ll find the handy Position Meas
Quad Encoder VI. All you have to do is connect the two quadrature
signals from your encoder to the source and up/down pins of one of the
counters on your TIO board. Many encoders also have a Z index pulse
output that is asserted only once per revolution; it’s useful for obtain-
ing an absolute orientation on a rotational measurement. If you have
a Z index, wire it to the gate input of the counter. That’s all there is for
signal connections.
The VI takes advantage of some advanced features of the TIO boards.
First, you can activate low-pass filtering on the signal input lines
to reduce noise susceptibility. Second, you can activate resolution
multiplication, where the hardware interpolates the encoder pulses
by a factor of 2 or 4. The output of the VI is the measured position in
counts, which you will probably want to scale to degrees or some other
useful unit.
524 Chapter Nineteen

Motion application: An ion beam intensity mapper. One lab that Gary
worked in had a commercially made ion beam gun that we used to
test the sputter rate of various materials. We had evidence that the
beam had a nonuniform intensity cross section (the xy plane), and that
the beam diverged (along the z axis) in some unpredictable fashion. In
order to obtain quality data from our specimens, we needed to charac-
terize the beam intensity in x, y, and z. One way to do this is to place
sample coupons (thin sheets of metal or glass) at various locations in
the beam and weigh them before and after a timed exposure to obtain
relative beam intensity values. However, this is a tedious and time-
consuming process that yields only low-resolution spatial data. Pre-
liminary tests indicated that the beam was steady and repeatable, so
high-speed motion and data acquisition was not a requirement.
Our solution was to use a plate covered with 61 electrically isolated
metal targets (Figure 19.6). The plate moves along the z axis, pow-
ered by a stepper motor that drives a fine-pitch lead screw. The stepper
motor is powered by a Kinetic Systems 3361 Stepper Motor Controller
CAMAC module. Since the targets are arranged in an xy grid, we can

Stepper Motor

Kinetic Systems 3361

Stepper Motor Controller

61 Isolated
Targets
Push Rod

Ion Gun
100 Ω
Current Sampling
Resistors AM
UX
(61) 64
TM
ult
iple
xe
r +

Target – G
+ –
System Ground Bias
Supply MI
O-16
Inp
uts

Figure 19.6 The ion beam intensity mapper hardware. The plate holding 61 isolated
probes moves along the axis of the ion beam. Current at each probe is measured at
various positions to reconstruct a 3D picture of the beam intensity.
 Physics Applications 525

obtain the desired xyz mapping by moving the plate and taking data
at each location. Each target is connected to a negative bias supply
through a 100-Ω current-sensing resistor, so the voltage across each
resistor is 100 mV/mA of beam current. Current flows because the tar-
gets are at a potential that is negative with respect to the (positive) ion
gun, which is grounded. A National Instruments AMUX-64T samples
and multiplexes the resulting voltages into an NB-MIO-16 multifunc-
tion board. The bias voltage was limited to less than 10 V because that’s
the common-mode voltage limit of the board. We would have liked to
increase that to perhaps 30–50 V to collect more ions, but that would
mean buying 61 isolation amplifiers.
The front panel for this experiment is shown in Figure 19.7. Two
modes of scanning are supported: unidirectional or bidirectional (out
and back), selectable by a boolean control (Scan Mode) with pictures
pasted in that represent these motions. A horizontal fill indicator shows
the probe position as the scan progresses to cover the desired limits.
Examine the diagram of the main VI in Figure 19.8. An overall While
Loop keeps track of the probe location in a shift register, which starts
at zero, increases to a limit set by Total Range, then steps back to
zero if a bidirectional scan is selected. The VI stops when the scan is
finished then returns the probe to its starting location. Arithmetic in
the upper half of the loop generates this stepped ramp.

I/O Setup Scan Setup

DAQ device 0 Step size (in.) 0.50

N (KS-3361) 4 samples/ch 100

GPIB 1 scan rate 1000

Calibration Status
Motor steps/inch 1600 Motor position 3200

Total range (in.) 11.00 Path RAM Disk:abc

Scan Mode
2.00

Position 0.0 2.0 4.0 6.0 8.0 10.0 12.0

Chan. Names 0 X1Y1

Figure 19.7 Front panel of the ion beam intensity scan VI. The
user runs the VI after setting all the parameters of the scan. A
horizontal fill indicator shows the probe position.
526 Chapter Nineteen

Scan Mode

Keep track of position Motor position

Location (steps)
Location 0.00
Step size (in.)
Go the other True
Motor steps/inch
way?
Sign (L/R) 1 +=left, -=right

End
Total range (in.) Position
Position

Steps to move (+ or -) Move back

GPIB 0 [0..2] to zero

Create data file Tell controller to move

KS 3361
Path N (KS-3361) Write Count Stepper
Write Count

Chan. Wait for it to

start moving 1000
Names

1 [0..2]
Poll controller til move is done
True
CCW
CW

DONE
Read Count WHOOPS ! ! ! ! !
YOU HIT A
300
LIMIT SWITCH

2 [0..2]
Acquire and average data
position
DAQ device
samples/ch currents
Write To
Spreadsheet
scan rate File.vi

append

Figure 19.8 Diagram of the ion beam intensity scan VI. The While Loop executes once for each step of the
positional ramp. The sequence moves the probe and acquires data.

The Sequence structure inside the main loop has three frames.
Frame zero commands the motor controller to move to the new posi-
tion. Frame one polls the motor controller, awaiting the done flag, which
indicates that the move is complete. If the VI encounters a limit switch,
it tells the user with a dialog box and execution is aborted. Frame two
acquires data by calling a subVI that scans the 64 input channels at
high speed and averages the number of scans determined by a front
panel control. Finally, the position and the data are appended to a
data file in tab-delimited text format for later analysis. As always, the
 Physics Applications 527

format of the data was determined beforehand to assure compatibility

with the analysis software.

Plasma potential experiments

The plasma we’re talking about here doesn’t flow in your veins. This
plasma is the so-called fourth state of matter, where most of all the
atoms are ionized. Plasma is formed by depositing sufficient energy
into a substance (usually a gas) to rip away one or more electrons from
the atoms, leaving them positively charged. The ionizing energy may
come from an electrical discharge (an electron bombardment) or some
form of ionizing radiation, such as light, gamma rays, or nuclear par-
ticles. The ions and electrons in a plasma love to recombine and react
with other materials, so most plasma experiments are performed in
a vacuum chamber with a controlled atmosphere containing only the
desired gaseous species.
Some parameters that we like to determine in experiments are the
plasma space potential, electron temperature, floating potential, and
ion and electron densities. A very simple plasma diagnostic technique,
the Langmuir probe, makes most of these measurements simple. The
method involves the measurement of ion and/or electron current flow-
ing in a small metal probe that is positioned in contact with the plasma.
By varying the voltage (potential) applied to the probe and measuring
the current, a curve called the probe characteristic is acquired. All you
need is a variable voltage source, a sensitive current monitor, and a
recording mechanism (anyone for LabVIEW?). Figure 19.9 shows how
a simple Langmuir probe experiment might be connected.

Langmuir probe

+ +
DAC output
or function Plasma
generator – –
Probe
current Vacuum chamber
Isolation
amplifier or Input protection
power supply +
Current Amplifier Current
Shunt
– measurement
to ADC

Figure 19.9 Basic electrical connections for a Langmuir probe experiment. The probe
voltage is supplied by an isolation amplifier, programmable power supply, or function
generator with a large output voltage swing.
528 Chapter Nineteen

Electrically, the most important piece of equipment is the voltage

source that drives the probe. Depending on the experiment, you may
need a source that can produce anywhere from ±10 V at 10 mA up to
±50 V at 100 mA, or greater. Ordinary DACs, such as those built into
a typical plug-in board, can only reach ±10 V and a few mA, which
may be sufficient for some laboratory experiments. For low-speed DC
experiments, we like to use a laboratory power supply with an external
analog input driven by a DAC. The output of such a supply is isolated
from ground and is nearly bulletproof. For experiments in which you
need a high-speed ramp waveform or pulse, you may need a function
generator and a high-performance amplifier. As for the current-sensing
function, pick a shunt resistor that only drops a fraction of a volt at the
expected probe current (R = V/I ), and follow it with the usual differen-
tial amplifier on your DAQ board or signal conditioner. Input protection
is a good idea because the plasma source generally has a high-voltage
source. Accidental contact with any high-voltage elements (or arcing)
will likely destroy the amplifier.
To obtain the Langmuir probe characteristic, your LabVIEW pro-
gram will have to generate a ramp waveform. Steady-state plasma
experiments can be performed at low speeds over a period of seconds
or minutes, so it’s practical for your program to calculate the desired
voltage and drive an output device. Figure 19.10 shows a simple V-I
scan experiment that uses a precalculated array of values represent-
ing a ramp waveform that is sent to a DAC repeatedly. You can use any
conceivable waveform to initialize the data array. Nested For Loops
step through the values. After each step, a measurement is taken from
the system (for instance, the current from a Langmuir probe). Inside
the inner For Loop the new voltage is written to the output, then the
resulting probe current is measured. Error I/O links the DAC Out and
Read Probe subVIs so that they execute in the proper order. If extra
settling time is required before the measurement, add a suitable delay
between the DAC Out and Read Probe VIs.
The XY Graph (current versus voltage) acts like a real-time xy
recorder. Since LabVIEW doesn’t have a built-in xy strip chart, you can
simulate one by updating the graph each time the inner For Loop exe-
cutes. A shift register carries the accumulated array of clusters, where
the cluster contains x (voltage) and y (current) data. An empty graph
constant initializes the shift register. If you don’t care to see the graph
update for every sample point, put it outside the inner For Loop and
let the cluster array build itself on the border of the loop. This is much
faster and more efficient, too.
The Read Probe subVI acquires and scales the current measurement.
If you want to store the data on disk, that subVI would be a logical place
 Physics Applications 529

Steps V start V end Ramp time Cycles Cycle #

20 -10.00 10.00 5.00 s 2 2

Voltage to probe Current vs Voltage

10.0 10.0
8.0
5.0 6.0
mA 4.0
0.0
2.0
-5.0 0.0
-2.0
-10.0 -10.0 -5.0 0.0 5.0 10.0 V

Cycles Do the V-I scan several times

empty graph
Update output volts, measure input current
Ramp time 0

Voltage to
probe
1000 Current vs
Steps Voltage

V end

V start
Ramp Pattern.vi Cycle #

Figure 19.10 A simple V-I scan experiment. This is another way of generating low-speed,
recurrent ramps, using a precalculated array of output voltages. Values are sent to a digi-
tal to analog converter (DAC) repeatedly, and the response of the system under test is
measured and recorded after each output update. The current versus voltage graph acts
like a real-time xy chart recorder.

to write the data to a file. Outside the For Loops, you would open a suit-
able file and probably write some kind of header information. Remem-
ber to close the file when the VI is through.
In faster experiments, LabVIEW may not have enough time to
directly control the stimulus and response sequence. In that case, you
can use a couple of different approaches, based on plug-in data acquisi-
tion boards or external, programmable equipment.
The data acquisition library supports buffered waveform generation
as well as buffered waveform acquisition on plug-in boards with direct
memory access (DMA). Essentially, you create an array of data repre-
senting the waveform, then tell the DMA controller to write the array to
the DAC, one sample at a time, at a predetermined rate. Meanwhile, you
can run a data acquisition operation also using DMA that is synchro-
nized with the waveform generation. A set of examples is available in
530 Chapter Nineteen

the directory labview\examples\DAQ\anlog-io. Synchronization of the

input and output operations can be determined three possible ways.
■ The DAC update clock is externally wired to an input (SCANCLK)
which triggers a scan of the ADC. Each time a new step value in
the ramp is generated, the ADC collects one value from each defined
input channel (i.e., one scan).
■ The ADC scans continuously, and its clock output triggers the DAC
updates.
■ An external input is supplied to both the ADC and DAC clocks. The
signal can be derived from an onboard counter-timer set up as a pulse
generator or from an external pulse generator.

The VI in Figure 19.11 uses the first method where DAC updates trig-
ger ADC scans. The controls allow you to specify the ramp waveform’s
start and stop voltages, number of points, and the DAC update rate.

# points update rate

per cycle # cycles (points/sec) V Start V Stop

STOP 100 6 1000.00 -2.00 2.00

X-Y graph
device
2.0
1 M
e 1.0
AO channel a
s 0.0
0
u
r -1.0
AI channels
e
0,1 -2.0
d
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
Ramp Voltage

Ramp Pattern.vi

V Stop
V Start update rate (points/sec)
# points
per cycle AO channel
device
AO tas�I �
# cycles 1st ch
0 X-Y graph
AI channels AI taskID
ac� done
scan clock source
# scans error
I/O Connector
status

For an AT-MIO-16F-5, hook OUT5 (pin 49, DAC update clock) to OUT2 stop
(pin 46, which clocks each scan when doing interval scanning).

Figure 19.11 Using the Data Acquisition library and an AT-MIO-16F-5 board, this VI simultane-
ously generates a ramp waveform and acquires data from one or more analog inputs. Only one
external jumper connection is required; note that the pin number and signal name vary slightly
between models of DAQ boards.
 Physics Applications 531

Multiple cycles can be generated, and each cycle is plotted indepen-

dently. For the analog inputs, you can acquire data from one or more
channels with adjustable scan rate and adjustable number of points to
acquire. Only one channel is plotted in this example, but data from all
scanned channels is available on the diagram.
This example makes effective use of the intermediate-level data
acquisition functions. Two tasks are defined, one for input and one
for output. Error I/O controls the execution sequence and, of course,
handles errors. Synchronization of the input and output operations is
determined by the DAC update clock which is externally wired to an
input which clocks each scan of the ADC. Each time a new step value
in the ramp is generated, the ADC collects one value from each defined
input channel (i.e., one scan). Referring to the diagram, here are the
details of the program’s steps:

1. The Ramp Pattern VI (from the Signal Generation VIs in the

Analysis library) creates an array of values corresponding to a ramp
waveform. You could use other functions if you needed a different
waveform.
2. AI Config sets up the analog input hardware on the board for the
desired number of channels and allocates a data acquisition buffer
that’s sized for the number of scans to be acquired.
3. AI Start begins the buffered acquisition process that was defined
by AI Config. The Trigger Type is set to 2 (I/O connector) which tells
the board to use an external scan clock.
4. AO Config sets up the analog output hardware on the board and
allocates a data buffer sized for the number of points to generate.
5. AO Write loads the ramp waveform into the output buffer in prepa-
ration for waveform generation.
6. AO Start starts the buffer waveform generation process that was
defined by AO Config. The number of buffer iterations is set by the
Number of Cycles control, meaning that the waveform will be pro-
duced several times.
7. AI Read fetches the acquired data from the buffer and makes it
available to the XY Graph, where the measured voltage is plotted
versus the calculated scan ramp. AI Read makes an estimate (based
on the number of points to acquire and the sample rate) of how long
to wait for data before timing out, so your program won’t get stuck if
there is a hardware failure.
8. The While Loop will continue executing the AI Read until an error
occurs or the user clicks the Stop button.
532 Chapter Nineteen

9. Always clean up your mess. AI Clear and AO Clear terminate any

I/O in progress and release their respective memory buffers. No
error checking is done here because these functions can do no harm,
even if no AI or AO operations are in progress. The Simple Error
Handler checks for and reports any data acquisition errors.

In this example, where the analog output is updated first, there is a

very short time delay before the ADC acquires a sample. Your experi-
ment may not have time to settle, resulting in an unpredictable error.
An alternative is to acquire a new reading from the ADC and then
update the DAC output. This maximizes settling time and is probably
a better overall solution.
For the E-series boards, a multitude of internal soft connections are
possible, thanks to some elaborate onboard switching. This makes it
easier to configure an E-series board for flexible scanning, like our syn-
chronization problem. The Synchronize AI AO for E-Series Boards
VI obtains the ADC scan clock from the DAC update clock in a manner
similar to the example in Figure 19.11. Multiple AI and AO channels
are supported.
Another way to solve this problem is to use a function generator
for the stimulus waveform and a plug-in board or external digitiz-
ing instrument for the measurement. A great way to synchronize the
stimulus and measurement operations is to use a time-base clock. If
you can get your hands on a modern arbitrary waveform gener-
ator, you will find that it has a clock output. Each time the instru-
ment updates its DAC output, a pulse is delivered to the clock output.
This clock can then be applied to the external clock input of your data
acquisition board or digitizer. If the signal generator is programmed
to produce 1024 clocks in one cycle of the waveform, your digitizer is
guaranteed to acquire exactly 1024 ADC samples. Nice and neat, and
always in phase. To see how to program your DAQ board, look at the
DAQ example VIs Acquire N Scans-ExtChanClk and Acquire N
Scans-ExtScanClk.
If you can’t use a clock, then there may be some way to use a trigger
signal. Even simple analog function generators have a trigger output
that occurs when the waveform starts. Use that to start a data acquisi-
tion operation or trigger a digital oscilloscope (see the section on trig-
gering which follows). All you have to do is make sure that the period
of the waveform is a little longer than the DAQ operation. Also, as
with any external signal generation, you may want to dedicate a sec-
ond ADC channel to monitoring the stimulus waveform, just to be sure
that the amplitude is what you think it is.
 Physics Applications 533

Handling Fast Pulses

Pulse and transient phenomena abound in the world of physics research.
As we mentioned before, there is a host of specialized instrumentation
to accompany the various detectors that produce transient waveforms.
In this section, we’ll look at some of those instruments, the LabVIEW
programs that support them, and some of the programming tricks that
may be of general use.

Transient digitizers
A whole class of instruments, called transient digitizers, have been
developed over the years to accommodate pulsed signals. They are
essentially high-speed ADCs with memory and triggering subsystems,
and are often sold in modular form, such as CAMAC and VXI and more
recently as plug-in boards. Transient digitizers are like digital oscil-
loscopes without the display, saving you money when you need more
than just a few channels. In fact, the digital oscilloscope as we know it
today is somewhat of a latecomer. There were modular digitizers, sup-
ported by computers and with analog CRTs for displays, back in the
1960s. Before that, we used analog oscilloscopes with Polaroid cameras,
and digitizing tablets to convert the image to ones and zeros. Anyone
pine for the “good ol’ days”? The fact is, digitizers make lots of sense
today because we have virtual instruments, courtesy of LabVIEW. The
digitizer is just so much hardware, but VIs make it into an oscilloscope,
or a spectrum analyzer, or the world’s fastest strip-chart recorder.
If you can’t find a transient recorder that meets your requirements,
check out the digital oscilloscopes available from the major manufac-
turers. They’ve got amazing specifications, capability, and excellent
value these days. You might also consider plug-in oscilloscope boards
which have the basic functionality of a ’scope but without the user
interface. National Instruments offers a line of plug-in digital oscil-
loscope boards in PCI, PXI, and PCMCIA formats. At this writing,
boards are available with sampling rates up to 100 MHz at 8-bit reso-
lution and up to 16 Msamples of onboard memory, featuring analog
as well as digital triggering. Some boards are programmed through
NI-DAQ, like any other plug-in board, while others are handled by
the NI-SCOPE driver, which is IVI-based. In either case, example VIs
make it easy to get started and integrate these faster boards into your
application.
There are several other major manufacturers of oscilloscope boards that
supply LabVIEW drivers. Gage Applied Sciences offers a very wide
range of models, in ISA and PCI bus format for Windows machines.
534 Chapter Nineteen

Input characteristics. High-speed digitizers, including fast oscilloscopes,

generally have a 50-Ω input impedance. Low-speed models may be high-
impedance, and many offer switchable input impedance. AC or DC cou-
pling is also available on most models, but you can always add your own
coupling capacitor if the input is DC-coupled only. All modern digitizers
have input amplifiers and/or attenuators to adjust the scale factor. Old
CAMAC digitizers were fixed at one range, such as 2 V. You have to sup-
ply your own wideband amplifiers and attenuators with those.

Triggering. Most digitizers rely on a TTL or ECL-level external trigger

to start or stop acquisition. Most models have some form of internal,
level- or edge-sensitive triggering, much like an oscilloscope. Through
software commands, you begin acquisition by arming the digitizer after
which you either poll the digitizer to see if it has received a trigger
event, or, in the case of most GPIB instruments, wait for an SRQ telling
you that it’s time to read the data.
Like digital oscilloscopes, you can store pretrigger and/or posttrig-
ger data. Once the digitizer is armed, its ADC samples at the desired
rate and stores data in high-speed memory, which is arranged as a
circular buffer. When the trigger occurs, the current memory location is
saved, and sampling continues until the desired number of posttrigger
samples are acquired. Because data was being stored before the trig-
ger, you can retrieve pretrigger data as well. This is very useful because
there may well be important information that arrives before the trig-
ger event in many experiments. The DAQ library does a similar trick
with plug-in boards of all types.

Data storage and sampling. An important choice you have to make when
picking a transient recorder is how much memory you will need and
how that memory should be organized. Single-shot transient events
with a well-characterized waveshape are pretty easy to handle. Just
multiply the expected recording time by the required number of sam-
ples per second. With any luck, someone makes a digitizer with enough
memory to do the job. Another option is streaming data to your com-
puter’s memory, which is possible with plug-in oscilloscope boards. For
instance, a good PCI bus board can theoretically move about 60 million
1-byte samples per second to memory, assuming uninterrupted DMA
transfers. Similarly, VXI systems can be configured with gigabytes of
shared memory tightly coupled to digitizer modules. It appears that
the possibilities with these systems are limited only by your budget.
Far more interesting than simple transient events is the observation
of a long tail pulse, such as atomic fluorescence decay. In this case, the
intensity appears very suddenly, perhaps in tens of nanoseconds, decays
rapidly for a few microseconds, then settles into a quasi-exponential
 Physics Applications 535

decay that tails out for the best part of a second. If you want to see the
detail of the initial pulse, you need to sample at many megahertz. But
maintaining this rate over a full second implies the storage of millions
of samples, with much greater temporal resolution than is really neces-
sary. For this reason, hardware designers have come up with digitiz-
ers that offer multiple time bases. While taking data, the digitizer
can vary the sampling rate according to a programmed schedule. For
instance, you could use a LeCroy 6810 (which has dual time bases)
to sample at 5 MHz for 100 υs, then have it switch to 50 kHz for the
remainder of the data record. This conserves memory and disk space
and speeds analysis.
Another common experiment involves the recording of many rapid-fire
pulses. For instance, the shot-to-shot variation in the output of a pulsed
laser might tell you something significant about the stability of the
laser’s power supplies, flash lamps, and so forth. If the pulse rate is too
high, you probably won’t have time to upload the data to your computer
and recycle the digitizer between pulses. Instead, you can use a digitizer
that has segmented memory. The Tektronix RTD720A is such a beast.
It can store up to 1024 events separated by as little as 5 υs, limited only
by the amount of memory installed. After this rapid-fire sequence, you
can upload the data at your leisure via GPIB. Other instruments with
segmented memory capability are various HP, Tektronix, and LeCroy
digital oscilloscopes, and the LeCroy 6810. Figure 19.12 shows how it’s
done with a National Instruments NI-5112 digital oscilloscope board,
which supports multiple-record mode, whereby a series of waveforms
is stored in onboard memory. The board takes 500 ns to rearm between
acquisitions, so you won’t lose too much data. This advanced capability
is only available through the NI-SCOPE driver.

Calibration. As a practical precaution, you should never trust a digi-

tizer when it comes to amplitude calibration. While the time bases are

Store each record as a separate

dimension in the waveform array.
number of records
niScope Fetch Multi
input channel Waveform.vi
instrument handle waveform array
error in
error out
niScope Actual
Record Length.vi
record no.

Figure 19.12 With the right oscilloscope board and the NI-SCOPE driver, you can
rapidly acquire multiple waveform records in onboard memory. This bit of code
reads out the data into a 2D array.
536 Chapter Nineteen

generally quite accurate, the same cannot be said for high-speed ADC
scale factors. We like to warm up the rack of digitizers, then apply
known calibration signals, such as a series of precision dc voltages,
and store the measured values. For greater accuracy at high frequen-
cies, you should apply a pulse waveform of known amplitude that
simulates the actual signal as closely as possible. If you want to get
fancy, add an input switching system that directs all the input chan-
nels to a calibration source. Then LabVIEW can do an automatic cali-
bration before and after each experiment. You can either store the
calibration data and apply it to the data later or apply the pre-run
calibration corrections on the fly. Some plug-in digitizers, like the
NI-5112, have onboard calibrators that simplify this task. One pre-
caution: If you have several digitizers with 50-Ω inputs, make sure
they don’t overload the output of your calibrator. It’s an embarrass-
ing, but common, mistake.

Digital storage oscilloscopes (DSOs)

As far as the physics researcher is concerned, the digital storage oscil-
loscope (DSO) has several uses—for troubleshooting and initial setup
of experimental apparatus, for convenient display of live waveforms
during an experiment, and as a transient digitizer when combined
with a LabVIEW system. Most DSOs have a GPIB interface and a Lab-
VIEW driver. Therefore, you can use them to make permanent records
of transient events while simultaneously gaining a convenient real-
time waveform display. The only limitation we’ve seen is in some really
old models that were incredibly slow at accepting commands and trans-
ferring data.
As a utility instrument, the oscilloscope (either analog or digital) is
unsurpassed. As much as we would like to think that a digitizer and
a LabVIEW display can duplicate an oscilloscope, it never really pans
out in the practical world. The reasons are partly aesthetic, partly por-
tability, but mostly that some of us like real knobs. Some of the digital
oscilloscopes aren’t very good in this respect either. All covered with
buttons and nary a knob in sight. For general-purpose tests and mea-
surements, we’ll take a real ’scope any day.
For real-time waveform display, LabVIEW sometimes approaches
the common DSO in terms of display update speed. If you have a mod-
est record length, such as 512 samples, and a fast computer, a GPIB
or VXI digitizer can do a credible job of mimicking a real-time display.
You can also throw lots of traces on one screen, which is something
that most DSOs can’t do. Nevertheless, every lab we’ve ever worked
in has had every available oscilloscope fired up, displaying the live
waveforms.
 Physics Applications 537

Timing and triggering

The key to your success in pulsed experiments is setting up timing and
triggering of all the diagnostic instruments and recording equipment.
Just having a rack full of digitizers that all go off at the same time may,
or may not, solve your problems. All of your pulsed diagnostics must
trigger at the proper time, and you have to know exactly when that
trigger occurred.

What’s all this triggering stuff, anyhow? The experiment itself may be
the source of the main trigger event, or the trigger may be externally
generated. As an example of an externally triggered experiment, you
might trigger a high-voltage pulse generator to produce a plasma.
Experiments that run on an internally generated time base and experi-
ments that produce random events, such as nuclear decay, generally
are the source of the trigger event. The instruments must be armed
and ready to acquire the data when the next event occurs. All of these
situations generate what we might classify as a first-level, or primary,
trigger (Bologna and Vincelli 1983).
First-level triggers are distributed to all of your pulsed diagnostic
instruments to promote accurate synchronization. Some utility hard-
ware has been devised over the years to make the job easier. First, you
probably need a trigger fan-out, which has a single input and multiple
outputs to distribute a single trigger pulse to several instruments, such
as a bank of digitizers. A fanout module may offer selectable gating
(enabling) and inverting of each output. A commercial trigger fanout
CAMAC module is the LeCroy 4418 with 16 channels. Second, you
often need trigger delays. Many times, you need to generate a rapid
sequence of events to accomplish an experiment. Modern digital delays
are very easy to use. Just supply a TTL or ECL-level trigger pulse, and
after a programmed delay, an output pulse of programmable width and
polarity occurs. Examples of commercial trigger delays are the Stan-
ford Research Systems DG535 and the LeCroy 4222. Third, you may
need a trigger discriminator to clean up the raw trigger pulse from
your experimental apparatus. It works much like the trigger controls
on an oscilloscope, permitting you to select a desired slope and level on
the incoming waveform.
A fine point regarding triggering is timing uncertainty, variously
known as trigger skew or jitter. Sometimes you need to be certain
as to the exact timing relationship between the trigger event and the
sampling clock of the ADC. A situation where this is important is wave-
form averaging. If the captured waveforms shift excessively in time,
then the averaged result will not accurately represent the true wave-
shape. Most transient digitizers have a sampling clock that runs all
the time. When an external trigger occurs, the current sample number
538 Chapter Nineteen

is remembered by the digitizer’s hardware. Note that the trigger event

could fall somewhere in between the actual sampling events that are
driven by the clock. One way to force synchronization is to use a master
clock for your experiment that drives all the digitizers. Then derive the
trigger signal from the clock so that triggering always occurs at the
same time relative to the clock. Unfortunately, this trick can’t be used
when the trigger event is random or uncontrollable in nature. In those
cases, you must make sure that the trigger generator is stable and
repeatable and that the sampling rate is high enough that a timing
offset of plus or minus one sample will do no harm.
Figure 19.13 illustrates an application using several of these trigger
processing devices with a pulsed laser as the source of the primary
trigger event. In this case, the laser excites some plasma phenomenon
in a target chamber. A photodiode samples the laser beam to provide
a primary trigger pulse. Then a high-voltage pulse generator is trig-
gered to collect ions in the plasma. Finally, the digitizers are triggered
to observe the ion current. The delay helps adjust the zero time for
the digitizers to center the recorded data with respect to the physi-
cal event. Note that a delay generator cannot produce an output that
occurs before the input. (We can’t count the number of times we needed
such a device!) If you want a physical event to occur before the trigger,
you have to delay everything else in the experiment.
Second-level triggers are another useful tool in physics diagnostics.
Some experiments require a bit more intelligence to decide when to
trigger the data acquisition hardware. For instance, the coincidence of
two pulses or digital words (a logical operation) is commonly used to
detect high-energy particles. Coincidence detection helps to filter out
undesired, anomalous events. In such cases, you can use commercial
triggering logic systems, such as those made by LeCroy. Or, if it’s a simple
problem, just put together some analog and/or digital logic yourself.
Third-level triggers represent an even higher level of sophistication
in trigger event detection. In many experiments, there may be enough
noise or other false events that the first-level and even the second-level

Trigger Trigger
fanout delay
Pulsed YAG laser Trigger
Amplifier ∆T to
Flashtube
Photo- Trigger digitizers
diode discriminator
Trigger to
pulsed
HV generator
TTL

Figure 19.13 Typical application of a trigger discriminator, fanout, and delays used in
a pulsed laser experiment.
 Physics Applications 539

triggers cause you to record many nonevents. If this occurs very often,
you end up with a disk full of empty baselines and just a few records
of good data. That’s okay, except for experiments where each event con-
sists of 85 megabytes of data. The poor person analyzing the data (you?)
is stuck plowing through all this useless information.
The solution is to use a hardware- or software-based, second-level
trigger that examines some part of the data before storage. For instance,
if you have 150 digitizers, it might make sense to write a subVI that
loads the data from one or two of them and checks for the presence of
some critical feature—perhaps a critical voltage level or pulse shape
can be detected. Then you can decide whether to load and store all the
data or just clear the digitizer memories and rearm.
If there is sufficient time between events, the operator might be able
to accept or reject the data through visual inspection. By all means, try
to automate the process. However, often there is barely time to do any
checking at all. The evaluation plus recycle time may cause you to miss
the actual event. If there is insufficient time for a software pattern rec-
ognition program to work, you can either store all the data (good and
bad) or come up with a hardware solution or perhaps a solution that
uses fast, dedicated microprocessors to do the pattern recognition.
Many modern digital oscilloscopes have a trigger exclusion feature
that may be of value in physics experiments. You define the shape of a
good pulse in terms of amplitude and time, typically with some tolerance
on each parameter. After arming, the ’scope samples incoming wave-
forms and stores any out-of-tolerance pulses. This is good for experi-
ments where most of the data is highly repetitive and nearly identical,
but an occasional different event of some significance occurs.

Capturing many pulses

Many pulsed experiments produce rapid streams of pulses that you
need to grab without missing any events. Assuming that you have all
the right detectors and wideband amplifiers, you need to select the
right acquisition hardware.
There are two ways to treat multipulse data: Capture each pulse indi-
vidually, or average the pulses as they occur. Averaging has the distinct
–
advantage of improving the signal-to-noise ratio on the order of √ n , where
n is the number of pulses averaged. Of course, if you are able to capture
and store all of the pulses, you can always average them afterward.
Figure 19.14 is a VI that uses a digitizer with segmented memories
to acquire many waveforms. Note that the digitizer must be accurately
triggered and in phase with every pulse to avoid statistical errors.
Additionally, all the pulses must be similar. The diagram shows some
of the array indexing tricks required to average the waveforms.
540 Chapter Nineteen

Data Waveform Graph

Segment 31 -0.04 9.0

Sample 49
5.0

0.0
-3.0
0 10 20 30 40 49

Array size returns an array of values

A waveform for multi-dimensional arrays

segments Array Size Index Array Average data for each sample
32 samples
Index Add Array
seg# Array
1 Elements Waveform
data Graph

6810 1-Chan
Read.vi Data
segments
0

Disable indexing for Channels; Sum array elements and

note that the top index on Index Array divide by the number of
is the same item as the top index on elements to obtain average
the array indicator on this VI's panel

Figure 19.14 Averaging waveforms acquired from a digitizer that has segmented memories.
Be really careful with array indexing; this example shows that LabVIEW has a consistent
pattern with regards to indexes. The first index is always on top, both on the panel and on
the diagram.

This VI is memory-intensive because it assembles multiple wave-

forms (1D arrays) into a large 2D array. At first, it may seem that there
is no other solution because you need to have all of the waveforms
available before averaging can be performed. There is another solution:
Use an ensemble averager subVI like the one in Figure 19.15. Each
time this VI is called, it sums the incoming data with the previous data
on a sample-by-sample basis. When called with Average set to True,
the sum is divided by the number of samples. The Iteration control
initializes the two uninitialized shift registers. At initialization time,
the incoming waveform determines the length of the array.
Now you can rewrite the data-averaging VI in a way that saves mem-
ory. The improved version is shown in Figure 19.16, and it produces
the same result but with much less memory usage. To further improve
performance, averaging is turned off until the last iteration of the For
Loop. This step saves many floating-point division operations. If you
want to see the averaging operation evolve in real time, you could move
the Waveform Graph inside the For Loop and turn averaging on all
the time. Each time a new segment of data is read, the graph will be
updated with a newly computed average.
 Physics Applications 541

X[] Averaged X[]

0 0.00 0 0.00 X[]
Average (T) Averaged X[]
Iteration
Iteration (0:initialize) Average (T) (0:initialize)

0 Average Ensemble Averager.vi

1 [0..1]
X[] False case is empty 0 [0..1]
Accumulate
True Initialize
sum

number of
records Averaged X[]
1.00

Iteration (0:initialize)

Average (T)

Figure 19.15 This shows an ensemble averager VI. It accumulates (sums) the data from mul-
tiple waveforms, then calculates the average on command. It uses uninitialized shift registers
to store the data between calls.

Individual pulses need to be stored when the information contained in

each pulse is significant. The big issues (as far as LabVIEW is concerned)
are how often the pulses occur and how fast your data transfer rate
and throughput are. If the pulses only occur every second or so, there
is no problem. Any simple acquisition loop that you write with a good
external digitizer will probably work. But sometimes you will want to
capture pulses at much higher rates, perhaps hundreds or thousands of
pulses per seconds. That calls for digitizers with segmented memories,
as previously mentioned. For intermediate rates, it’s possible to do the
acquisition in LabVIEW, but you had better do some testing to verify
performance. Remember to optimize your program (described later in
this chapter), primarily by minimizing graphics and memory usage.

segments
32 Average only on the last iteration Waveform Graph

avg? 9.0
Ensemble
seg# Waveform
Averager 5.0
Graph

0.0
init
-3.0
6810 1-Chan
Disable 0 10 20 30 40 49
Read.vi
indexing

Figure 19.16 Using the ensemble averager drastically simplifies the diagram and cuts memory
usage while yielding the same results as before.
542 Chapter Nineteen

If the running average value of many pulses is interesting, then you

can use some other instruments. Most DSOs offer built-in signal aver-
aging. Assuming that you can trigger the DSO reliably, this is a really
easy way to obtain averaged waveforms over GPIB. For a more modular
approach, DSP Technology makes a line of averaging memories for use
with its CAMAC digitizers. It uses high-speed memory and an adder
to accumulate data from synchronized waveforms. After the desired
number of events (N), you read out the memory and divide the data
by N. Some advanced VXI modules may support this kind of waveform
acquisition and processing as well.
You can also use a boxcar averager to acquire averaged pulse
waveforms. The basic timing diagram for a boxcar averager is shown in
Figure 19.17. The gate interval is a short period of time during which
the analog signal voltage is sampled and its amplitude stored as charge
on a capacitor. The output of the boxcar is the average of many such
gate events over a selected period of time. If the gate is fixed at one
location in time, relative to the recurrent analog waveform, then the
boxcar output represents the average value of the waveform at that
time—just like putting cursors on the graph of the waveform. Now
we can vary the gate delay with an analog voltage, thus scanning the
sampling location over the entire period of the pulse. By scanning
slowly, the boxcar output accurately reconstructs the analog waveform.
With this technique, a simple xy chart recorder or slow data acquisition
system can record the shape of nanosecond waveforms.

Trigger

Gate delay

Boxcar gate (variable)

Analog waveform
(sync'd to trigger)

Average value during gate:

periodically acquired by ADC
Figure 19.17 Basic timing diagram for a boxcar
averager. The output of the boxcar is the average of
many measurements that are acquired during many
gate events. By slowly varying the gate delay, the
boxcar output reconstructs the shape of the analog
waveform.
 Physics Applications 543

One boxcar system, the Stanford Research Systems SR250, offers

a GPIB interface module (for which we have a LabVIEW driver) that
directly supports gate scanning and also measures the boxcar output.
Even without this special support, you can use a ramp generator VI like
the one described in the section on Plasma Potential Experiments to
scan the gate and make the output measurements. On the SR250, the
gate delay signal is a 0–10-V input (drive it directly from your MIO-16
output), and the gate delay ranges from 1 ns to 10 ms, so you can cover
a very wide range of pulse widths. It also offers a high-speed sampling
module that shrinks the gate time to about 10 ps. The boxcar output
ranges up to ±2 V and is well filtered, making accurate measurements
a snap.
Equivalent time sampling (ETS) is a trick used by high-speed
digitizers, DSOs, and the E-series DAQ and oscilloscope boards from
National Instruments. Realize that very fast ADCs are expensive
and often lack high resolution. But sample-and-hold amplifiers can
be designed with very short aperture (sampling) times—on the order
of picoseconds—at a more reasonable cost. This allows the hardware
designer to use a scheme similar to that of the boxcar averager. They
delay the triggering of the sample-and-hold and ADC by a variable
amount, just like the boxcar’s gate, thus scanning the sampling point
across the period of a recurrent waveform. Once the sample-and-hold
has acquired a voltage, a low-speed ADC can read that voltage at
a relatively leisurely rate and with good resolution. The trigger delay
can be chosen randomly or it may proceed in a monotonic fashion like
the boxcar. Thus, the waveform is reconstructed in memory for readout
and display.
ETS, like the boxcar, is a specialized hardware solution that solves
a particular problem: acquiring fast, recurrent waveforms. Physics
diagnostic problems often require esoteric hardware like that, and you
should try to learn about special instruments to enhance your arsenal of
solutions. LabVIEW alone can’t replace every kind of system, but it can
sure make the job easier, once you’ve got the right front-end hardware.

Recovering signals from

synchronous experiments
Say that you have a beam of light chopped at 10 kHz that is partially
absorbed in an experiment. The remaining light strikes a photodiode,
producing a small-amplitude square wave, and you want to measure
the amplitude. To complicate things, assume the likely scenario where
there is a great deal of 60-Hz noise and some nasty impulse noise from
other apparatus. If you digitize this signal and display it, there’s a good
chance you won’t even find the 10-kHz waveform, let alone measure its
amplitude. You must do some signal processing first.
544 Chapter Nineteen

One solution is to apply a narrow-bandwidth digital bandpass fil-

ter, tuned to the desired center frequency. The concept is also known
as a frequency-selective voltmeter. It’s simple and often adequate
because much out-of-band noise is rejected. All it takes is a DAQ wave-
form acquisition VI, followed by one of the IIR filter VIs. To determine
the amplitude, you might use the AC & DC Estimator VI from the
Measurement function palette in the Analysis library. Its ac amplitude
estimate is obtained from the mean and variance of the time domain
signal after passing the signal through a Hann time window.
In the lock-in library you’ll find a VI called Signal Recovery By
BP Filter. It acquires data from one channel of a DAQ board, applies
a low-pass filter, and displays the ac estimate (Figure 19.18).
There are some shortcomings to this filter-based technique. Impulses
and other transient signals often contain energy that falls within the
passband of the filter, making that polluting energy inseparable from
your signal. Also, filters with narrow passbands (high Q) are prone
to ringing when excited by impulse or step waveforms. If the incom-
ing signal is sufficiently noisy, you may not be able to recover your

device Bandpass filter Amplitude

4 Frequency 2.0
channels
100.00 1.5
0 0
Bandwidth
1.0
scan rate 10.00
1000.00 0.5

N scans to read 0.0

500 STOP 0 36

Acquire data Bandpass filter Measure amplitude

filter type
device AI Continuous Butterworth AC & DC
Scan.vi Bandpass Filter.vi Estimator.vi
channels
N scans Fs
0 Amplitude
scan rate

Bandwidth
Fhi continue
3
2 order
Flo 10

Frequency

stop status

Figure 19.18 The Signal Recovery by BP Filter VI demonstrates a way of measuring ampli-
tude in a selected frequency band. For this screen shot, I swept the input frequency around the
100-Hz region while running the VI.
 Physics Applications 545

signal reliably. To see how bad the problem can be, try the Pulse Demo
VI from the Analysis examples. Turn the noise way up, and watch the
pulse disappear from view. Note that the addition of some signal con-
ditioning filters and amplifiers, plus proper grounding and shielding,
may remove some of the interference from your signal and is always a
worthwhile investment.
In experiments that operate with a constant carrier or chopping fre-
quency, the signal amplitude can be recovered with lock-in amplifiers,
which are also called synchronous amplifiers or phase-sensitive
detectors (PSDs). The basic principle revolves around the fact that
you need only observe (demodulate) the range of frequencies within
a narrow band around the carrier frequency. To do this, the lock-in
requires a reference signal—a continuous sine or square wave that is
perfectly synchronized with the excitation frequency for the experi-
ment. The lock-in mixes (multiplies) the signal with the reference,
resulting in an output containing the sum and difference frequen-
cies. Since the two frequencies are equal, the difference is zero, or dc.
By low-pass-filtering the mixer output, you can reject all out-of-band
signals. The bandwidth, and hence the response time, of the lock-in
is determined by the bandwidth of the low-pass filter. For instance,
a 1-Hz lowpass filter applied to our example with the 10-kHz signal
results in (approximately) a 1-Hz bandpass centered at 10 kHz. That’s
a very sharp filter, indeed, and the resulting signal-to-noise ratio can
be shown to be superior to any filter (Meade 1983).
Lock-ins can be implemented by analog or digital (DSP) techniques.
It’s easiest to use a commercial lock-in with a LabVIEW driver, avail-
able from Stanford Research or EG&G Princeton Applied Research.
If you want to see a fairly complete implementation of a lock-in in
LabVIEW, check out the example in the lock-in library. For a simple
no-hardware demonstration, we created a two-phase lock-in with
an internal sinusoidal reference generator. This example VI, Two-
phase Lock-in Demo, simulates a noisy 8-kHz sine wave, and you
can change the noise level to see the effect on the resulting signal-to-
noise ratio.
The DAQ 2-Phase Lock-in VI is a real, working, two-phase DSP
lock-in where the signal and reference are acquired by a plug-in board.
The reference signal can be a sine or square wave, and the amplitude
is unimportant. SubVIs clean up the reference signal and return a
quadrature pair of constant-amplitude sine waves. To generate these
quadrature sine waves, we had to implement a phase-locked loop (PLL)
in LabVIEW. We used a calibrated IIR low-pass filter as a phase shift
element. When the filter’s cutoff frequency is properly adjusted, the
input and output of the filter are exactly 90 degrees out of phase. This
condition is monitored by a phase comparator, which feeds an error
signal back to the low-pass filter to keep the loop in lock.
546 Chapter Nineteen

We tested this application on a Macintosh G3/400 with a PCI-MIO-

16E1 board. It’s possible to obtain stable amplitude and phase measure-
ments with sampling rates up to at least 30 kHz, processing buffers as
short as 500 samples. That yields a maximum demodulated (output)
bandwidth of up to 30 Hz. A 1-kHz signal can be accurately measured
when its rms amplitude is below 1 LSB on the ADC if the time constant
is long enough. To make this application run faster, you can eliminate
graphical displays or get a faster computer.

Handling Huge Data Sets

Experiments have a way of getting out of hand with respect to the
quantity of data generated. At the NASA Ames Research Center’s
Hypervelocity Free-Flight Aerodynamic Facility (HFFAF) (Bogdanoff
et al. 1992), they use about 124 channels of transient digitizers with
anywhere from 8 to 256K samples per channel—about 16 megabytes
per shot, typical. (The good news is, they can only do a couple of shots
per week.) In another common situation, seismology, they need to digi-
tize and record data at hundreds of samples per second from many
channels—forever. Forever implies a really big disk drive. Obviously,
there must be some practical solutions to memory and disk limitations,
and that’s what we’ll look at next.

Reducing the amount of data

The concept of third-level triggers can certainly reduce the amount of
data to be handled. Or, you can do some kind of preliminary data reduc-
tion to compact the raw data before storing it on disk.
In seismology, an algorithm called a P-picker continuously processes
the incoming data stream, searching for features that indicate the pres-
ence of a P wave (a high-frequency, perpendicular shock wave), which
indicates that an interesting seismic event has occurred. If a P wave is
detected, then a specified range of recent and future data is streamed to
disk for later, more detailed analysis. Saving all of the raw data would
be impractical, though they sometimes set up large circular buffer
schemes just in case a really important event (“The Big One”) occurs.
That permits a moderate time history to remain in memory and/or on
disk for immediate access. But in the long run, the P-picker is key.
If your data is of a contiguous but sparse nature, like that of a seismic
data recording system, consider writing a continuous algorithm to detect
and keep interesting events. Because the detection algorithm may require
significant execution time, it probably needs to run asynchronously with
the data acquisition task. If you are using a plug-in board, you can use
buffered DMA. Some kinds of smart data acquisition units also support
 Physics Applications 547

buffering of data, such as the HP3852. Either way, the hardware continues
collecting and storing data to a new buffer even while the computer is
busy analyzing the previous buffer.
If you can’t afford to throw away any of your data, then data reduc-
tion on the fly might help reduce the amount of disk space required.
It will also save time in the post-experiment data reduction phase. Is
your data of a simple, statistical nature? Then maybe all you really
need to save are those few statistical values, rather than the entire
data arrays. Maybe there are other characteristics—pulse parameters,
amplitude at a specific frequency, or whatever—that you can save in
lieu of raw data. Can many buffers of data be averaged? That reduces
the stored volume in proportion to the number averaged. If nothing
else, you might get away with storing the reduced data for every wave-
form, plus the raw data for every Nth waveform. Decimation is another
alternative if you can pick a decimation algorithm that keeps the criti-
cal information, such as the min/max values over a time period. As you
can see, it’s worth putting some thought into real-time analysis. Again,
there may be performance limitations that put some constrains on how
much real-time work you can do. Be sure to test and benchmark before
committing yourself to a certain processing scheme.

Optimizing VIs for memory usage

Large data sets tend to consume vast quantities of memory. Some
experiments require the use of digitizers that can internally store
millions of samples because there simply isn’t time during the run to
stream the data to your computer. When the run is over, it’s time to
upload, analyze, and save all those samples. If your computer has only,
say, 32 MB of RAM, then it’s unlikely that you will be able to load more
than a few hundred thousand samples of data in one chunk. You have
two choices: Buy more memory or write a smarter program.
Expanding the memory capacity of your machine is certainly worth-
while. LabVIEW is a big memory user, no doubt about it. And modern
operating systems aren’t getting any smaller. Even virtual memory
(VM), available on all of LabVIEW’s platforms, may not save you
because there are significant real-time performance penalties due to
the disk swapping that VM causes. For best performance and simplicity
in programming, nothing beats having extra memory. Of course it costs
money, but memory is cheap compared to the aggravation of poor sys-
tem performance.
So, you find yourself a little short on memory, a little short on cash,
and a little down in the mouth because LabVIEW dies every time
you try to load those megasamples of data. It’s time to optimize, and
perhaps rewrite, your program. By the way, we guarantee that you
548 Chapter Nineteen

will see an increase in the performance of your program when you

optimize memory usage. Anything you do to reduce the activity of our
old friend the memory manager will speed things up considerably.
Here are the basic memory optimization techniques that everyone
should know.
Begin by “RTFMing” (reading the fine manual), this time concentrat-
ing on the chapter in the LabVIEW user manual called “Performance
Issues.” The LabVIEW team is fully aware of the importance of explain-
ing how LabVIEW allocates memory, and they continue to improve and
expand the available documentation. Also check the current technical
notes. LabVIEW Application Note AN168, LabVIEW Performance
and Memory Management, is relevant to the current problem: han-
dling large quantities of data.

Tip 1: Use cyclic processing rather than trying to load all channels at
once. Once you see this concept, it becomes obvious. As an example,
here is a trivial example of the way you can restructure a data acquisi-
tion program. In Figure 19.19A, 10 waveforms are acquired in a For
Loop, which builds a 10-by-N 2D numeric array. After collection, the
2D array is passed to a subVI for storage on disk. The 2D array is
likely to be a memory burner. Figure 19.19B shows a simple, but effec-
tive change. By putting the data storage subVI in the loop, only one
1D array of data needs to be created, and its data space in memory
is reused by each waveform. It’s an instant factor-of-10 reduction in
memory usage.

A. All the data is loaded at once into a huge 2D array.

Not memory efficient!
Load all waveforms Save all the
waveforms
10
channels

BIG 2D array

B. One waveform at a time is loaded and written

to a file. Much more memory-efficient.

10 Load a waveform Save it

channels

Figure 19.19 Use cyclic processing of data (b) to

avoid storing all of your data in memory at the
same time, as in a.
 Physics Applications 549

Of course, you can’t always use cyclic processing. If you need to per-
form an operation on two or more of the waveforms, they have to be
concurrently memory-resident. But keep trying: If you only need to
have two waveforms in memory, then just keep two, and not the whole
bunch. Perhaps you can use shift registers for temporary storage and
some logic to decide which waveforms to keep temporarily.

Tip 2: Break the data into smaller chunks. Just because the data source
contains a zillion bytes of data, you don’t necessarily have to load it
all at once. Many instruments permit you to read selected portions of
memory, so you can sequence through it in smaller chunks that are
easier to handle. The LeCroy 6810 CAMAC digitizer and most digital
oscilloscopes have such a feature, which is supported by the LabVIEW
driver. If you’re using a plug-in board, the DAQ VI AI Read permits
you to read any number of samples from any location in the acquisition
buffer. Regardless of the data source, it’s relatively simple to write a
looping program that loads these subdivided buffers and appends them
to a binary file one at a time. Similarly, you can stream data to disk and
then read it back in for analysis in manageable chunks.

Tip 3: Use the smallest practical numeric types.Remember that LabVIEW

supports many numeric types, from a single-byte integer on up to a
16-byte extended-precision (EXT) float. The minimum amount of mem-
ory that your data will consume is the number of samples multiplied
by the size of the numeric type, in bytes. Lots of digitizers have 8-bit
resolution, which fits perfectly into a single-byte integer, either signed
(I8) or unsigned (U8). That’s a case of maximum efficiency. Many times,
however, you will want to scale the raw binary data to engineering
units which require a floating-point representation. For most cases, try
to use the single-precision float (SGL), which is 4 bytes long and offers
seven significant figures of precision. In the rare case that greater pre-
cision is required, use double-precision (DBL) which occupies 8 bytes.
The latter type is also required by all VIs in the Analysis library. If you
use any of those VIs, you may as well keep all of your data in DBL for-
mat. Why? That’s the next tip.

Tip 4: Avoid coercion and conversion when handling arrays. Minimizing

data space requires constant vigilance. Begin by looking at your data
sources, such as driver VIs. Make sure that the data is born in the small-
est practical type. Drivers that were created way back in version 1.2
of LabVIEW are notorious for supplying all data in EXT format (the
only numeric type then available). Figure 19.20 shows one such exam-
ple, an old CAMAC driver VI. You might even create your own versions
of driver VIs, optimized for your memory-intensive application.
550 Chapter Nineteen

GPIB Address

True Change the representation of

this indicator to match the
DATA
conversion function

Change this conversion

0 function to something
DATA i16 smaller than EXT
FORMAT
Figure 19.20 This old driver (from the Kinetic Systems 3988 library) converts
16-bit binary data to EXT floating point, which wastes much memory. Change the
indicated conversion function to something smaller, such as SGL. Save big! And
get the Ginsu knife!

Most of the analysis functions in the LabVIEW library are partly

polymorphic and handle only DBL-format floating-point or waveform
data. Using one of these functions on your more compact SGL data
results in a doubling of memory usage. There are several things you
can do if you need to conserve memory. First, call National Instruments
and complain that you want totally polymorphic analysis VIs. Second, if
it’s not too complex, try writing your own equivalent function subVI in
LabVIEW, one which uses only your chosen (smaller) data type. Realize
that your routine may run a bit slower than the optimized, CIN-based
analysis VI. The G Math Toolkit contains a plethora of analysis VIs
that use no CINs, and that can be the basis for your custom, memory-
optimized analysis. Third, you can perform the analysis at a time when
less data is resident in memory—perhaps after the data acquisition
phase is complete.
Keep a sharp eye open for coercion dots at terminals that handle
large arrays. Anywhere you see one, the data type is being changed,
and that means that reallocation of memory is required. This takes
time and, especially if the new data type is larger, results in the con-
sumption of even more memory. If the coercion dot appears on a subVI,
open the subVI and try to change all of the related controls, indicators,
and functions to the same data type as your source. Eliminate as many
coercion dots as you can. Sometimes this is impossible, as with the
Analysis library, or with any VI that contains a CIN. By the way, never
change the data type that feeds data into a CIN! A crash will surely
ensue because the CIN expects a certain data type; it has to be recom-
piled to change the data type. Maybe one day we’ll have polymorphic
CINs, too. Finally, don’t convert between data types unnecessarily. For
instance, changing a numeric array to a string for display purposes is
costly in terms of memory and performance. If you must display data,
try to use an indicator in the native format of that data.
 Physics Applications 551

Tip 5: Avoid data duplication. LabVIEW Application Note AN168,

LabVIEW Performance and Memory Management, goes into some
detail on the subject of efficient use of memory. In general, the G com-
piler tries very hard to avoid data duplication by reusing memory buf-
fers. You can help by choosing the best combination of functions and
structures to promote reuse. Here are some of the most important rules
to keep in mind.
■ An output of a function reuses an input buffer if the output and
the input have the same data type, representation, and—for arrays,
strings, and clusters—the same structure and number of elements.
This category includes all of the Trig & Log functions; most of the
Arithmetic functions, including those that operate on boolean data;
a few string functions such as To Upper Case; the Transpose Array
function; and the Bundle function to the extent that the cluster input
and output can use the same buffer. Most other functions do not
meet this criteria; among them are the Build Array and Concatenate
String functions.
■ In general, only one destination can use the buffer of the data source.
More specifically, where the source of a data path branches to two or
more destinations that change the input data, the source buffer can
be used by only one. A way to conserve memory is to do processing in
a serial fashion—one subVI after another—rather than wiring the
subVIs in parallel.
■ If there are multiple destinations that read data but do not change it,
they may use the same source buffer. For example, Array Size doesn’t
duplicate the buffer since it doesn’t change it. Also, under certain
circumstances, when a node that alters data and one or more nodes
that merely read data have the same source, LabVIEW can deter-
mine that the readers execute before the node that changes the data
so that the source buffer can be reused. (For this to occur, all nodes
involved must be synchronous.)
■ If the source is a terminal of a control that is not assigned to a con-
nector terminal or receives no data from the calling VI, the source
data is stored in a separate buffer. Without an external source to sup-
ply data when the subVI executes, the data is locally supplied. This
means you can save memory by not trying to display it. You can also
use the VI Properties dialog to change the priority to Subroutine.
Subroutine priority eliminates all extra buffers associated with indi-
cators and slightly reduces subVI calling overhead. Note that it also
prevents you from running the VI interactively!
■ If a buffer can be reused by an input and output of a structure or subVI,
the buffer can be used by other nodes wired to that node. Thus, in a
552 Chapter Nineteen

hierarchical program, a buffer can be reused from the top-level VI to

the lowest subVI and back up to the top. Don’t be afraid to use subVIs,
so long as they contain no additional array duplications internally.
■ Data allocated to a top-level VI cannot be deallocated during execution.
On the other hand, data allocated in a subVI can be deallocated.

You have some control over the LabVIEW memory manger through
the Options dialog, in the Performance and Disk items. The first
option is Deallocate memory as soon as possible. When a subVI finishes
execution, this option determines whether or not its local buffers will
be deallocated immediately. If they are deallocated, you gain some free
memory, but performance may suffer because the memory will have to
be reallocated the next time the subVI is called. On the Macintosh, there
is an option called Compact memory during execution. When enabled,
LabVIEW tells the Macintosh memory manager to defragment memory
every 30 seconds. Defragmented memory enhances speed when allocat-
ing new buffers. However, this compacting operation usually induces
a lull in performance every 30 seconds which may be objectionable in
real-time applications.
If your application uses much more memory than you think it should,
it’s probably because of array duplication. Try to sift through these
rules (study the application note for more) and figure out where you
might modify your program to conserve memory.

Tip 6: Use memory diagnostics. There are some ways to find out how
much memory your VIs are actually using. Please note that these tech-
niques are not absolutely accurate. The reasons for the inaccuracy are
generally related to the intrusiveness of the diagnostic code and the
complexity of the memory management system. But as a relative mea-
sure, they are quite valuable.
Begin by using the VI Properties dialog in the File menu to find out
how much memory is allocated to your VI’s panel, diagram, code, and
data segments. Of primary interest is the data segment, because that’s
where your buffers appear. Immediately after compiling a VI, the data
segment is at its minimum. Run your VI, and open the VI Properties dia-
log again to see how much data is now allocated. You can try some simple
tests to see how this works. Using an array control, empty the array and
check the data size. Next, set the array index to a large number (such as
10,000), enter a value, and check the data size again. You should see it
grow by 10,000 times the number of bytes in the data type.
The most extensive memory statistics are available through the
Profiling feature of LabVIEW. Select Profile VIs from the Tools >>
Advanced menu, and start profiling with the Profile Memory Usage
and Memory Usage items selected (Figure 19.21). Run your VIs, take
 Physics Applications 553

Figure 19.21 The Profiling window gives you timing and memory usage statistics on your application. The first
VI in the list is the biggest memory user; that’s the place to concentrate your efforts.

a snapshot at some critical time, and examine the memory-related col-

umns. If you click on a column header, the entire spreadsheet will be
sorted in descending order by that column. Important statistics include
minimum, maximum, and average bytes used, in addition to timing
information. Find out which VIs use the most memory and try to opti-
mize them first. The Profiler is a quantitative tool for this effort. Some-
times, the least-expected VI is the biggest memory hog!

Bibliography
Accelerator Technology Division AT-5: Low-Level RF LabVIEW Control Software User’s
Manual, Los Alamos National Laboratory, LA-12409-M, 1992. (Available from National
Technical Information Service, Springfield, Va.)
Bogdanoff, D. W., et al.: “Reactivation and Upgrade of the NASA Ames 16 Inch Shock
Tunnel: Status Report,” American Institute. of Aeronautics and Astronautics 92-0327.
Bologna, G., and M. I. Vincelli (Eds.): “Data Acquisition in High-Energy Physics,” Proc.
International School of Physics, North-Holland, Amsterdam, 1983.
Mass, H. S. W., and Keith A. Brueckner (Eds.): Plasma Diagnostic Techniques, Academic
Press, New York, 1965.
Meade, M. L.: Lock-in Amplifiers: Principles and Applications, Peter Peregrinus, Ltd.,
England, 1983. (Out of print.)
2000 LeCroy Research Instrumentation Catalog, LeCroy Corporation, New York, 2000.
(www.lecroy.com.)
This page intentionally left blank
 Chapter

Data Visualization, Imaging, and Sound

20
Today’s sophisticated computer systems make it possible for you to col-
lect and observe information in ways that used to be impractical or
impossible. Once the bastion of the High Priests of Graphic Arts, the
ability to produce multidimensional color plots, acquire and manipu-
late images, and even make movies of your data is directly available
(live!) through LabVIEW. All you need to do is configure your system
appropriately and, for some features, buy some third-party LabVIEW
applications that offer these advanced capabilities. This chapter tells
you how.
We humans have a marvelous ability to absorb visual information,
but only when it is presented in optimum form. A well-designed graph or
image conveys a great deal of information and gives us unexpected insight
into hidden relationships buried in our data. You have to invest some pro-
gramming effort, time, money, and thought to create the right visualiza-
tion instrument, but the results are both valuable and impressive.
Keep in mind the fact that imaging is just another data acquisition
and analysis technique. That’s a fact that is overlooked when you haven’t
seen it in action. Images can be born of cameras and live scenes—the
usual pictures—or they can arise from other forms of numerical data.
Images are more than pretty pictures; they can be quantitative as well,
spatially and in intensity. Further, you can make those images into
movies and even combine them with sound by using the new concepts
embodied in multimedia. LabVIEW does all of these things with the
help of some add-on products.
There are dozens, perhaps hundreds, of data visualization (graphing,
imaging, and presentation) programs available on various computer
platforms these days. But LabVIEW has one big advantage over most

555
556 Chapter Twenty

of them: It can tie in data visualization with other operations in your

real-world laboratory. For example, you can direct a robot arm, mea-
sure temperatures, control an electron beam welder, acquire images of
the weld, process and display those images, store data on disk . . . and
do it all in real time, right there in LabVIEW. This is an exceptional
capability! Only a few years ago, you had to use several different pro-
grams (probably not simultaneously, either) or write a zillion lines of
custom code to perform all of these tasks. Sometimes, the processing
you wish to perform isn’t practical in LabVIEW. In that case, you can
still exchange data files with other specialized programs on various
computers.
This chapter covers graphing of data, image acquisition and process-
ing, and the recording and production of sound. LabVIEW has built-in
graphs that are adequate for both real-time and posttest data presen-
tation of ordinary two-dimensional data. It also has the capability to
display images in color or gray scale. With the addition of add-on prod-
ucts discussed here, you can also acquire and analyze video images,
make multidimensional plots, QuickTime movies, and other advanced
displays.

Graphing
The simplest way to display lots of data is in the form of a graph.
We’ve been drawing graphs for about 200 years with pens and paper,
but LabVIEW makes graphs faster and more accurate. Most impor-
tant, they become an integral part of your data acquisition and analy-
sis system.
Part of your responsibility as a programmer and data analyst is
remembering that a well-designed graph is intuitive in its meaning
and concise in its presentation. In his book The Visual Display of Quan-
titative Information (1983), Edward Tufte explains the fundamentals
of graphical excellence and integrity and preaches the avoidance
of graphical excess that tends to hide the data. We’ve been appalled
by the way modern presentation packages push glitzy graphics for
their own sake. Have you ever looked at one of those 256-color three-
dimensional vibrating charts and tried to actually see the information?
Here are your major objectives in plotting data in graphical format:
■ Induce the viewer to think about the substance rather than the
methodology, graphic design, the technology of graphic production,
or something else.
■ Avoid distorting what the data has to say (for instance, by truncating
the scales, or plotting linear data on logarithmic scales).
 Data Visualization, Imaging, and Sound 557

■ Try to present many numbers in a small space, as opposed to diffus-

ing just a few data points over a vast area. This makes trends in the
data stand out.
■ Make large data sets more understandable by preprocessing.
■ Encourage the eye to compare different pieces of data.
■ Reveal the data at several levels of detail, from a broad overview
to the fine structure. You can do this by making more than one plot
with different scales or by allowing the user to rescale a graph—a
standard LabVIEW feature.
■ Design the graph in such a way that it serves a clear purpose with
respect to the data—for description, exploration, or tabulation.
■ The graph should be closely integrated with other descriptions of
the data. For example, a graph should be synergistic with numeric
displays of statistics derived from the same data.
■ Note that some data is better displayed in a format other than a
graph, such as a table.
■ Above all, make sure that the graph shows the data.

The worst thing you can do is over decorate a graph with what Tufte
calls chartjunk. Chartjunk is graphical elements that may catch the
eye but tend to hide the data. You don’t want to end up with a graphical
puzzle. Simply stated, less is more (Figure 20.1). Though this is more of
a problem in presentation applications, there are some things to avoid
when you set up LabVIEW graphs:
■ High-density grid lines—they cause excessive clutter.
■ Oversized symbols for data points, particularly when the symbols
tend to overlap.
■ Colors for lines and symbols that contrast poorly with the back-
ground.
■ Color combinations that a color-blind user can’t interpret (know thy
users!).
■ Too many curves on one graph.
■ Insufficient numeric resolution on axis scales. Sometimes scientific
or engineering notation helps (but sometimes it hinders—the num-
ber 10 is easier to read than 1.0E1).
■ Remove or hide extra bounding boxes, borders, and outrageous col-
ors. Use the coloring tool, the transparent (T) color, and the control
editor to simplify your graphs.
558 Chapter Twenty

Default graph Remove unneeded elements

1.0 Plot 0 1.0

0.5 0.5

0.0 0.0

-0.5 -0.5
0 50 100 150 0 50 100 150

...but maybe all you really need is the data itself. Axis values are only required for quantitative graphing.
Controls and legends should not be displayed unless needed.

Figure 20.1 Sometimes, less is more when presenting data in graphs. Use the coloring tool, the
control editor, and the various pop-up menus to customize your displays.

Displaying waveform and cartesian data

Most data that you’ll acquire is a function of time or some other vari-
able and is described by y = f(x), where x is the independent variable
(e.g., time) and y is the dependent variable (e.g., voltage). This is gener-
alized cartesian (xy) data. The x values are most often based on time.
Time-series, or waveform, data is effectively displayed as an ordinary
graph or, in the case of real-time data, as a strip chart. Graphs and
charts are standard LabVIEW indicators with many programmable
features. Read through the graphing chapter of the LabVIEW user
manual to see how all those features are used. In particular, Property
Nodes are worthy of much study because graphs have a large number
of varying features.
Figure 20.2 shows the simplest ways to use the Waveform Graph.
Like many LabVIEW indicators, it’s polymorphic, so you can wire sev-
eral data types directly to it. Numeric arrays (any numeric represen-
tation) can be wired directly and are displayed with the x axis scaled
with values beginning with zero (x0 = 0) and increasing by one (∆x = 1)
for each data point in the array. You can also control x0 and ∆x by
bundling your data into a cluster as shown in Figure 20.2. Finally, the
waveform data type can be wired directly to a Waveform Graph, and
all the scaling information is carried along. Any time your data has
evenly spaced x values, use the Waveform Graph. Compared with an
XY Graph, it takes less memory, is faster, and is certainly the easiest
to set up.
 Data Visualization, Imaging, and Sound 559

Waveform Graph 1

1.0
1D array
Sine Pattern.vi Waveform Graph 1 0.0
32 1.00
-1.0
1D array 0 50 127

Xo Cluster of Xo, ∆X, [Y] Waveform Graph 2

5.0 5.00 1.0

∆X Xo
Waveform Graph 2
10.0 ∆X 10.00 0.0

-1.0
Bundle Cluster of Xo, ∆X, [Y] 32 1.00
5 1275

Waveform type Waveform Graph 3

t0 1.0
Y 0
Sine Waveform.vi Waveform Graph 3
10:28:16 PM
Sun, Feb 5, 2040 0.00 0.0

Waveform type 0.06

dt -1.0
0.13
0.00 0.0 500.0m 1.0

Figure 20.2 The Waveform Graph is polymorphic. This example shows some of the basic data types that
it accepts for the making of single-variable plots.

Polymorphism makes the Waveform Graph more versatile because

the graph directly accepts several data types. In the previous example,
you saw that a single variable could be plotted with or without x-axis
scaling. Figure 20.3 shows how to make a multiplot graph where each
signal has the same number of points. You scale the x axis in a manner
consistent with the previous example.
What if the signals don’t have the same number of points? Two dimen-
sional arrays require that each row of data has the same number of
points, so that approach won’t work. Your first guess might be to make
an array of arrays. Sorry, can’t do that in LabVIEW. Instead, make
an array that contains a cluster with an array inside it (Figure 20.4).
There are two ways to build this data structure. One uses a Bundle
function for each array followed by Build Array. The other way uses
the Build Cluster Array function, which saves some wiring. The
index of the outer array is the plot number, and the index of the inner
array is the sample number.
A slightly different approach, shown in Figure 20.5, is required if
you want to make multiple plots with different numbers of points and
you want to scale the x axis. For this example, we plotted a sine wave
and a line representing its mean value (which might be useful for a
control chart in statistical process control). The mean value is used as
both elements of a two-element array, which is the y-axis data for the
560 Chapter Twenty

Waveform Graph 1
Waveform 1.0
Sine Pattern.vi Graph 1 2D array
128 0.0
0 0.00
samples -1.0
2D array 0
0 50 127
Build
Array Xo Cluster of Xo, ∆X, [[Y]]
5.0 Waveform Waveform Graph 2
∆X
Graph 2 Xo 5.00
1.0
10.0
∆X 10.00 0.0
Cluster of Xo,
0 0.00 -1.0
∆X, [[Y]]
0 5 1275

frequency Sine Waveform.vi Array of waveforms Waveform Graph 3

Waveform
5.00 Graph 3 0 1.0
t0
Y 0
10:28:16 PM
frequency Triangle Waveform.vi Sun, Feb 5, 2040 0.00 0.0
Array of
1.00 Build waveforms 0.03
dt
Array -1.0
0.001000 0.06
0.0 500.0m 1.0

Figure 20.3 Three ways of making multiple plots with a Waveform Graph when the number of data points in
each plot is equal. Note that the Build Array function can have any number of inputs. Also, the 2D array could
be created by nested loops.

second plot. The trick is to use some math to force those two data points
to come out at the left and right edges of the plot.
So far, we’ve been looking at graphs with simple, linear x axes which
are nice for most time-based applications, but not acceptable for para-
metric plots where one variable is plotted as a function of another.
For that, use the XY Graph. Figure 20.6 shows a way of building one of
the data structures that defines a multiplot xy graph. A plot is defined

Figure 20.4 Making multiple plots where each plot has a different number of samples. Build Cluster Array
performs the same function as the combination in the dashed rectangle. Data could consist of waveforms
instead of the 1D arrays used here.
 Data Visualization, Imaging, and Sound 561

Array of cluster of Xo, ∆X, [Y] Waveform Graph

0 1.0 Data
Xo 17.00
Mean
∆X 10.00 0.0
-0.6 Mean
0 0.00
17 500 1297 0.48

Sine Pattern.vi ∆X 10.0 Xo 17.0

Build
Array Waveform Graph
0.6
Bundle

Array of cluster
Array Size Build of Xo, ∆X, [Y]
Array

Mean.vi
Mean

Figure 20.5Yet another data structure for multiple plots accepted by the
Waveform Graph. In this example, the sine wave is plotted with a line
representing the mean value. This second plot has only two data points.

XY Graph Plot Legend Array of cluster of array of cluster of (x,y)

60.0 Spiral Graph Palette
0
Array of cluster of (x,y)
40.0 One Circle
0
X 0.00
20.0 Scale Legend
Y 0.00
Y

0.0 X

-20.0 Two Y
Use the Show Name pop-up
-40.0 Cursor Legend item to display cursor name

-60.0 One 35.87 34.84

-50.0 0.0 50.0
X Two -26.54 -12.66

# of points
150 Spiral: X = i*sin(i), Y = i*cos(i)

X bundle bundle

5.0
Y Build
Array XY Graph

# of points
30 Circle: X = k*sin(i), Y = k*cos(i)

X bundle bundle Array of cluster of

array of cluster of
5.0 (x,y)
Y
k 50.0

Figure 20.6One of the data structures for an XY Graph is shown here. It probably makes sense to use this struc-
ture when you gather (x,y) data-point pairs one at a time. This is the way it had to be done in LabVIEW 2.
562 Chapter Twenty

XY Graph Array of cluster of [X] and [Y]

60 0
[X] 0 0.00
40

20 [Y] 0 0.00
0

-20

-40

-60
-60 -40 -20 0 20 40 60

# of points
150 Spiral: X = i*sin(i), Y = i*cos(i)

X bundle

5.0 Y
Build
Array XY Graph

# of points
30 Circle: X = k*sin(i), Y = k*cos(i)

bundle Array of cluster

X
of [X] and [Y]
5.0 Y
k 50.0

Figure 20.7 Similar to the last XY Graph example (Figure 10.6), but
using the other available data structure. This one is convenient when
you have complete arrays of x and y data.

as an array of clusters that contains x and y data-point pairs. In this

example, we turned on all the display options on the graph: the Plot
Legend, Graph Legend, Graph Palette, and Cursor Legend. These are
also available on the Waveform Graph.
Figure 20.7 shows how to use another available data structure for an
xy graph. Here, a plot is defined as a cluster of x and y data arrays. Why
would you use this format in place of the previous one? That depends
on how your data is acquired or created. If you collect single data-point
pairs, for instance, with a low-speed data recorder, then the previous
format is probably best. For data that arrives in complete arrays for
each plot, use the format in Figure 20.7.
A real-time xy strip chart can be constructed as shown in
Figure 20.8. The trick is to add a new xy data pair to an ever-growing
array that resides in a Shift Register. For each iteration of the While
Loop, the acquired data is stored in the array, and all data is displayed
in an xy graph. The biggest limitation is that performance decreases as
time goes on because the amount of data to be displayed keeps grow-
ing. You also risk running out of memory. As an alternative, consider
 Data Visualization, Imaging, and Sound 563

XY Graph
device
1 2.0

channels

Chan 1
0,1 0.0

STOP
-3.0
Interval -4.0 -2.0 0.0 2.0 4.0
500 ms Chan 0

0 Build Array

device
Constant to
initialize XY Graph
shift register
0
channels
Bundle
Interval 1
stop

Figure 20.8 An XY Graph makes a handy real-time xy chart

recorder. Keep appending data to an array stored in a shift
register. Caution: You may run out of memory after extensive
operation.

using the Circular Buffer VI, described in Chapter 18, “Process Con-
trol Applications.”

Bivariate data
Beyond the ordinary cartesian data we are so familiar with, there is
also bivariate data, which is described by a function of two variables,
such as z = f(x,y). Here are some examples of bivariate data that you
might encounter:

■ A photographic image is an example of bivariate data because the

intensity (z) changes as you move about the (x,y) area of the picture.
■ Topographic data of a surface, such as the Earth, contains elevations
(z) that are a function of latitude and longitude (x,y).
■ A series of waveforms can be treated as bivariate data. For instance,
the z axis can be amplitude, the x axis time, and y the sequence num-
ber. This is also called a waterfall plot.

Bivariate data may be displayed in many formats. Surface plots

have a three-dimensional (3D) look to them and are the most compu-
tationally intensive displays to generate. Surface plots may be solid
564 Chapter Twenty

models (sometimes including shading) or wire frames with or without

hidden lines removed. These plots can be very dramatic and give the
reader insight into the general nature of the data, if properly displayed.
The disadvantage of surface plots is that it is often difficult to use them
quantitatively. Even with calibrated axes and the application of color,
it’s still hard to read out the value of a particular point as you would do
on a simple xy graph. We’ll look at a practical way of obtaining three-
dimensional plots in LabVIEW in the next section.
Adding color to a surface plot effectively adds another dimension to
the display, although it can be more confusing than helpful unless the
data is well behaved. An example of effective use of color would be to
display the topography of a mountain range and color the surface
according to the temperature at each location. In general, temperature
decreases with elevation, so a smooth progression of colors (or grays)
will result. On the other hand, if you used color to encode wind speed
at each location, you would probably end up with every pixel being a
different color because of the rather uncorrelated and chaotic nature of
wind-speed variations over large areas. Don’t be afraid to try coloring
your plots—just be sure that it adds information, not just glitz.
Bivariate data can also be displayed as an image. The Intensity
Graph is a LabVIEW indicator that displays the value at each (x,y)
location as an appropriate color. You choose the mapping of values to
colors through a Color Array control, which is part of the Intensity
Graph. This type of display is very effective for data such as tempera-
tures measured over an area and, of course, for actual images. Use of
the Intensity Graph is covered later in this chapter.
Yet another way to display bivariate data is to use a contour plot.
A contour plot consists of a series of isometric lines that represent
constant values in the data set. Familiar examples are topographical
maps, maps of the ocean’s depth, and isobars on a weather chart show-
ing constant barometric pressure. LabVIEW offers limited contour
plotting capability as part of the 3D Graph tools that we’ll cover later.
Many other data analysis and graphing packages already offer contour
plotting, so you might plan on exporting your data. We might mention
that contour plots are not as easy to generate as it might seem. The
big problem is interpolation of the raw data. For instance, even though
your measurements are acquired on a nice xy grid, how should the
program interpolate between actual data points? Is it linear, logarith-
mic, or something else? Is it the same in all directions? Worse yet, how
do you handle missing data points or measurement locations that are
not on an even grid? Gary watched an analyst friend of his struggle
for years on such a problem in atmospheric dispersion. Every change
to the interpolation algorithm produced drastically different displays.
 Data Visualization, Imaging, and Sound 565

STOP RESET Update interval 1.0 sec

Vertical Gain

0.0
max
0.2

0.4

0.6

0.8

1.0 min

Z-angle

0.0 5.0 10.0

Number of Traces
10

Figure 20.9A Waterfall Plot is a simple and efficient way to

display waveforms that change with time.

You generally need some other display or analysis techniques to vali-

date any contour plot.
Waterfall plots are useful for observing a series of waveforms that
change with time. For instance, you may want to observe the behavior
of a power spectrum display while changing the system under observa-
tion. Figure 20.9 is a simple and efficient waterfall graphing VI that
originally shipped with the LabVIEW 2 examples. It accepts a 1D array
of data and does hidden-line removal.

Multivariate data
Beyond bivariate data is the world of multivariate data. Statisticians
often work in a veritable sea of data that is challenging to present because
they can rarely separate just one or two variables for display. For instance,
a simple demographic study may have several variables: location, age,
marital status, and alcohol usage. All of the variables are important, yet
it is very difficult to display them all at once in an effective manner.
The first thing you should try to do with multivariate data is analyze
your data before attempting a formal graphical presentation. In the
demographic study example, maybe the location variable turns out to
be irrelevant; it can be discarded without loss of information (although
you would certainly want to tell the reader that location was evalu-
ated). Ordinary xy graphs can help you in the analysis process, perhaps
by making quick plots of one or two variables at a time to find the
important relationships.
Some interesting multivariate display techniques have been devised
(Wang 1978). One method that you see all the time is a map. Maps can
be colored, shaded, distorted, and annotated to show the distribution of
almost anything over an area. Can you draw a map in LabVIEW? If you
have a map digitized as xy coordinates, then the map is just an xy plot.
566 Chapter Twenty

XY California Graph City locations

100.0 2
City
90.0 Sacramento

80.0 Horiz 26.00

70.0 Vert 63.00

Sacramento
60.0
San Francisco
50.0
Cursor Display
40.0
0 San Francisco 18.50 53.50
30.0
Los Angeles 57.00 16.00
20.0
Los Angeles Sacramento 26.00 63.00
10.0
0.00 0.00
0.0
0.0 20.0 40.0 60.0 80.0 100.0

City locations XY California Graph

Horiz
Vert
City XY California Graph
Graph setup notes:
Active Cursor Background set to T (transparent), X
Cursor Plot and Y axes set to fixed scales, major
0 grids on. Point style set to none, since
Cursor Index
the cursors provide markers.
2 Cursor Locked Interpolation off so that no lines are
Cursor Name Visible drawn.
Cursor Name
Color box constant California map pasted in from a drawing
Cursor Color program where it was colored gray.
0 Cursor Style Choose Send To Back from Edit menu
so that the graph overlays the map.
2 Cursor Point Style

Figure 20.10Mapping in LabVIEW. We pasted in a picture (the California state map)

from a drawing program, then overlaid it with an xy graph to display city locations.
Cursor names annotate each city location.

Or, you can paste a predrawn map into your LabVIEW panel and over-
lay it with a graph, as we did in Figure 20.10. In this example, we cre-
ated a cluster array containing the names and locations of three cities
in California with coordinates that corresponded to the map. When the
VI runs, the horizontal and vertical coordinates create a scatter plot.
This might be useful for indicating some specific activity in a city (like
an earthquake). Or, you could add cursors to read out locations, then
write a program the searches through the array to find the nearest
city. Note that a map need not represent the Earth; it could just as well
represent the layout of a sensor array or the surface of a specimen.
The hardest part of this example was aligning the graph with the map.
 Data Visualization, Imaging, and Sound 567

The map that we pasted into LabVIEW was already scaled to a reason-
able size. We placed the xy graph on top of it and carefully sized the
graph so that the axes were nicely aligned with the edges of the map.
Through trial and error, we typed horizontal and vertical coordinates
into the cluster, running the VI after each change to observe the actual
city location.
City names are displayed on the graph with cursor names. On the dia-
gram, you can see a Property Node for the graph with many elements—all
cursor-related—displayed. You can programmatically set the cursor
names, as well as setting the cursor positions, marker styles, colors, and
so forth. This trick is very useful for displaying strings on graphs.

Picture Controls. The Picture controls add a new indicator, called a

Picture, into which you can programmatically draw all types of geo-
metric figures, bitmaps, and text. The toolkit includes many examples,
most of them for the creation of specialized graphs with features that
exceed those of the standard LabVIEW types. Figure 20.11 shows a few
samples. The basic concept behind the Picture controls is that you start
with an empty picture, then chain the various drawing VIs together,

Smith plot Polar plot

90
2.00
1.75
1.50
1.25
1.00
0.75
0.50
0.25
180 0.00

270

Histogram XY Plot
10.0 1.0

0.9
9.0
0.8
8.0
0.7
7.0 0.6

6.0 0.5

0.4
5.0
0.3
4.0
0.2
3.0 0.1

2.0 -0.0

-0.1
1.0

0.0 -0.2
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4

Figure 20.11Here are some of the specialized plots that you

can draw with the Picture controls. These are included in the
examples.
568 Chapter Twenty

adding objects at each step. Each object requires specifications, such

as position, size, and color. Let’s take a look at an interesting example
using this toolkit.
The most interesting multivariate data display technique we’ve heard
of is the Chernoff Face (Wang 1978). The technique was invented by
Herman Chernoff at Stanford University in 1971 for the purpose of
simultaneously displaying up to 18 variables in the form of a simplified
human face. His idea works because we are quite sensitive to subtle
changes in facial expressions. Some applications where the Chernoff
face has been successful are showing relative living quality in various
metropolitan areas, displaying psychological profiles, and, my favorite,
showing the state of Soviet foreign policy over the years. The trick is to
map your variables into the proper elements of the face with appropri-
ate sensitivity, and of course, to draw the face in a controllable manner.
I spent some time with the VIs in the Picture controls, and managed to
build a crude face (Figure 20.12) with three variables: face width, eye
width, and mouth shape.

Picture 200 by 200 pixel area

Face width Eye width Mouth angle ±deg
63 8 25.71

100 20 30.0

80 20.0
15
60
10 0.0
40

20 5
-20.0
0 0 -30.0

Hor center Vert center

100 100

Hor center 15 Eye H offset

Face ht Centered Oval Picture

80
Mouth
Face width
Mouth angle
9
Eye ht Eye width

40
Vert center Eye V offset

Figure 20.12 A simplified Chernoff Face VI, with only three variables: face width,
eye width, and mouth angle. This VI calls two subVIs that we wrote to draw some
simple objects using the Picture Control Toolkit.
 Data Visualization, Imaging, and Sound 569

Figure 20.13 This Draw Centered Oval VI does some simple cal-
culations to locate an oval on a desired center point, then calls
the Draw Oval Picture VI. Note the consistent use of Picture in–
Picture out to maintain dataflow.

Because the Picture functions are all very low-level operations, you will
generally have to write your own subVIs that perform a practical opera-
tion. In this example, we wrote one that draws a centered oval (for the face
outline and the eyes) and another that draws a simple mouth (see Fig-
ure 20.13). All the Picture functions use Picture in–Picture out to support
dataflow programming, so the diagram doesn’t need any sequence struc-
tures. Each VI draws additional items into the incoming picture—much
like string concatenation. Don’t be surprised if your higher-level subVIs
become very complicated. Drawing arbitrary objects is not simple. In fact,
this exercise took us back to the bad old days when computer graphics
(on a mainframe) were performed at this level, using Fortran, no less.
You could proceed with this train of development and implement a
good portion of the actual Chernoff Face specification, or you could take
a different tack. Using the flexibility of the Picture VIs, you could cre-
ate some other objects, such as polygons, and vary their morphologies,
positions, colors, or quantities in response to your multivariate data
parameters. Be creative, and have some fun.
570 Chapter Twenty

3D Graphs
Occasionally, data sets come in a form that is best displayed in three
dimensions, typically as wire-frame or surface plots. When you
encounter such data, the first thing you should do is try to figure out
how to plot it in the conventional 2D manner. That will save you a good
deal of effort and will generally result in a more readable graphical
display. You should be aware that 3D graphing is much slower than 2D
because of the extra computations required (especially for large data
sets), so it’s generally limited to non-real-time applications. Many data
analysis applications have extremely flexible 3D graphing capability.
We prefer to analyze our more complicated data with MATLAB, HiQ,
IDL, or Igor.
If you need to display 3D data in LabVIEW, you are limited to the
Windows version, which includes 3D Graph ActiveX objects that you
drop on the front panel. A set of 3D Graph VIs manipulates that object
through Invoke nodes and Property nodes. As with any 3D plotting
problem, the real challenge is formatting your data in a manner accept-
able to one of the available plot types.
Figure 20.14 shows the 3D Surface Graph indicator displaying
a waterfall plot and a contour plot. Data in this example consists of
an array of sine waves with varying amplitude. In the real world, you
might be acquiring waveforms from DAQ hardware or an oscilloscope.
There are many possible plot styles, and a great deal of extra custom-
ization is possible through the use of the various 3D graph property
VIs. A typical surface plot (plot style cwSurface) is shown with 10 per-
cent transparency selected to allow the axes to peek through. You can
also have the graph show projections to either of the three planes. With
the plot style set to cwSurfaceContour and xy projection enabled, a
shaded contour plot is produced.
Other available 3D display types are the 3D Parametric Graph
and 3D Curve Graph. A parametric graph is most useful for geomet-
ric solid objects where 2D arrays (matrices) are supplied for x, y, and z
data. If the data you are displaying looks more like a trajectory through
space, then the 3D Curve Graph is appropriate. It accepts 1D arrays
(vectors) for each axis.
There’s a great deal of versatility in the 3D Graph, and you’ll have to
explore and experiment with the examples to see how your data can be
massaged into a suitable form. Hopefully, the LabVIEW team will come
up with a cross-platform solution to 3D graphing. Even when they do,
the challenge of data formatting won’t go away.
Historical note: The original 3D graphing package for LabVIEW
was SurfaceVIEW from Metric Systems, a company started by
Jeff Parker, who was one of the original LabVIEW team members.
 Data Visualization, Imaging, and Sound 571

Basic Projection
3D Graph 3D Surface.vi Properties.vi Properties.vi

20

plot style xy projection

Z matrix

transparency show projections only

Sinewaves with
200
random amplitudes stop

cwSurface, with cwSurfaceContour, with

transparency=10% xy projection on and
show projections only on

Figure 20.14 Example of a 3D Surface Graph being used as a waterfall display. A rendered surface and a con-
tour plot are shown.

Jeff informs us that the product is no longer available, but it certainly

represents one of the first successful add-on packages that helped to
make LabVIEW what it is today.

Intensity Chart
Another interesting indicator, the Intensity Chart, adds a third dimen-
sion to the ordinary strip chart. It accepts a 2D array of numbers, where
the first dimension represents points along the y axis and the second
dimension represents points along the x axis, like a block-mode strip
chart. Each value in the array is mapped to a color or intensity level.
The biggest trick is setting the color or gray-scale range to accommodate
your data. Fortunately, there is an option in the Intensity Chart pop-
up for the z-scale called AutoScale Z. That gets you into the ballpark,
572 Chapter Twenty

Spectrum Plot Intensity Chart

20.0 -100
128

0.0
100
F
-20.0 80
r
e 60
-40.0 q
40
-60.0 20
0
-80.0
779 Time 894
0 2000 4000 6000 8000 10000 -20

device channel samples sample rate Window Mapping Display Unit

4 0 256 20000 Blackman-Harris dB Vrms STOP

device
Spectrum Plot
channel Intensity
build
samples Chart
array
sample rate
spectrum
Window
Simple Spectrum
Display Unit Analyzer
stop
Mapping (from analysis
examples)

Figure 20.15 The Intensity Chart in action displaying a power spectrum. White regions on
the chart represent high amplitudes. Frequency of a triangle wave was swept from 4 to 2 kHz
while this VI was running. The chart scrolls to the left in real time.

after which you can set the overall range and the breakpoints along
the scale by editing the values on the color ramp. The colors can also
be determined by an attribute node. This is discussed in the LabVIEW
user’s manual.
In Figure 20.15, the Intensity Chart gives you another way to look at
a rather conventional display, a power spectrum. In this example, you
can see how the power spectrum has changed over a period of time. The
horizontal axis is time, the vertical axis is frequency (corresponding to
the same frequency range as the Spectrum Plot), and the gray-scale
value corresponds to the amplitude of the signal. We adjusted the gray-
scale values to correspond to the interesting −20- to −100-dB range of
the spectrum.

Image Acquisition and Processing

Your LabVIEW system can be a powerful imaging tool if you add the
right combination of hardware and software products. You will likely
need hardware, in the form of a frame grabber and perhaps some video
output devices, and software, in the form of image acquisition and pro-
cessing VIs. The important thing to remember is that images are just
 Data Visualization, Imaging, and Sound 573

another kind of data, and once you’re working in LabVIEW, that data
is going to be easy to handle. In fact, it’s no harder to acquire, analyze,
and display images than it is to acquire, analyze, and display analog
waveforms. Good news, indeed.
The basic data format for an image is a two-dimensional (2D) matrix,
where the value of each element (a pixel) is the intensity at a physi-
cal location. Image acquisition involves the use of a frame grabber
for live video signals or the use of mathematical transforms applied to
other forms of data to convert it into a suitable 2D format. Standard
image processing algorithms operate only on this normally formatted
image data.
Image processing algorithms can be broadly grouped into five areas:

■ Image enhancement: Processing an image so that the result is

of greater value than the original image for some specific purpose.
Examples: adjusting contrast and brightness, removing noise, and
applying false color.
■ Image restoration: A process in which you attempt to reconstruct or
recover a corrupted image (which is obviously easier when you know
what messed up the image in the first place). Examples: deblurring,
removal of interference patterns, and correcting geometric distor-
tion. The Hubbell Space Telescope benefited from these techniques
during its initial operation.
■ Image segmentation: Dividing an image into relevant parts,
objects, or particles. These algorithms allow a machine to extract
important features from an image for further processing in order to
recognize or otherwise measure those features.
■ Object representation and description: Once an image has been
segmented, the objects need to be described in some coherent and
standardized fashion. Examples of these descriptive parameters are
size, coordinates, shape factors, shape equivalents, and intensity
distribution.
■ Image encoding: Processing of images to reduce the amount of
data needed to represent them, in other words, data compression
and decompression algorithms. These methods are very important
when images must be transmitted or stored.

Like signal processing, image processing is at once enticing and fore-

boding: There are so many nifty algorithms to choose from, but which
ones are really needed for your particular application? When Gary first
decided to tackle an imaging problem, he ran across the hall to visit a
fellow EE and asked for a book. What he came back with was Digital
Image Processing by Gonzales and Wintz (1987), which is the favorite
574 Chapter Twenty

of many short courses in image processing. What you really need to do

is study a textbook such as that one and get to know the world of image
processing—the theory and applications—before you stumble blindly
into a VI library of algorithms. Also, having example programs, and
perhaps a mentor, are of great value to the beginner.
The most important imaging software for LabVIEW is IMAQ Vision
from National Instruments. Years ago, Graftek developed Concept
V.i, a highly modular package of CIN-based VIs that supports image
acquisition and processing with interactive displays, file I/O, and a
generally free exchange of data with the rest of LabVIEW. In 1996,
National Instruments decided that imaging technology had reached
the point where images could be acquired and manipulated just like
data from a DAQ board. It reached an agreement with Graftek, pur-
chased Concept V.i, and renamed it IMAQ Vision. You can now develop
a flexible imaging application with LabVIEW and IMAQ as easily as
you would write a data acquisition program with the DAQ library.
Hence the name, IMaging ACquisition.
IMAQ has two family members offering different levels of image pro-
cessing capability. The base package includes fundamental functions
for image acquisition and file management, display and sizing utilities,
histogram computation, and image-to-array conversions. The advanced
package includes image processing and analysis functions such as
lookup table transformations and spatial filters to allow for contrast
and brightness adjustment, extraction of continuous and directional
edges, noise removal, detail attenuation, texture enhancement, and so
forth. It also includes frequency processing operations, arithmetic and
logic operators, and object manipulations such as thresholding, ero-
sion, dilation, opening, closing, labeling, hole filling, particle detection,
and so forth.
Graftek will continue to support IMAQ Vision (as well as older ver-
sions of Concept V.i). Its specialty will be drivers for third-party frame
grabbers, advanced and specialized algorithms, and custom applica-
tion development. If you need special assistance with a custom imaging
application, contact Graftek.

System requirements for imaging

Basic graphing and simple image processing tasks are handled eas-
ily by any of the common LabVIEW-compatible computers, even the
older, lower-performance Macs and PCs. But serious, real-time dis-
play and analysis call for more CPU power, more memory, and more
disk space. Audio and video I/O adds requirements for specialized
interface hardware that may not be included with your average com-
puter, though the newer multimedia-ready machines sometimes have
 Data Visualization, Imaging, and Sound 575

decent configurations. Let’s see what kinds of hardware you might con-
sider above and beyond your generic data acquisition system.

Computer configuration. If it’s images you’re dealing with, then it’s

time to get serious about hardware. Image processing and display
are positively the most CPU- and memory-intensive tasks you can
run on your computer. Images contain enormous quantities of data.
Many algorithms perform millions of calculations, even for apparently
simple operations. And storing high-resolution images requires a disk
drive with endless capacity. You can never afford the ideal computer
that’s really fast enough, or has enough memory, or big enough disks.
But practical cost-performance trade-offs can lead you to a reasonable
choice of hardware.
The fastest general-purpose computers available today come close
to handling full-speed video, but be really cautious of manufacturers’
claims. If you run custom-written applications with direct access to the
I/O hardware, real-time video is possible on these faster machines. But
if you expect LabVIEW to do lots of other things at the same time, such
as running an experiment, there will be performance problems. As with
high-speed data acquisition, buy and use specialized hardware to off-
load these burdensome processing tasks, and you will be much happier
with the performance of your system.

Video I/O devices. Unless your computer already has built-in video
support (and LabVIEW support for that), you will need to install a
frame grabber in order to acquire images from cameras and other
video sources. A frame grabber is a high-speed ADC with memory that
is synchronized to an incoming video signal. Most frame grabbers have
8-bit resolution, which is adequate for most sources. It turns out that
cameras aren’t always the quietest signal sources, so any extra bits of
ADC resolution may not be worth the extra cost. Preceding the ADC,
you will generally find an amplifier with programmable gain (contrast)
and offset (brightness). These adjustments help match the available
dynamic range to that of the signal. For the North American NTSC
standard (RS-170) video signal, 640 by 480 pixels of resolution is com-
mon. For the European PAL standard, up to 768 by 512 pixels may be
available, if you buy the right grabber. Keep in mind that the resolu-
tion of the acquisition system (the frame grabber) may exceed that of
the video source, especially if the source is videotape. Both color and
monochrome frame grabbers are available. Color images are usually
stored in separate red-green-blue (RGB) image planes, thus tripling
the amount of memory required.
Frame grabbers have been around for many years, previously as
large, expensive, outboard video digitizers, and more recently in the
576 Chapter Twenty

form of plug-in boards for desktop computers. With the birth of mul-
timedia, frame grabbers have become cheaper and even more capable
and are sometimes built into off-the-shelf computer bundles. The trick
is to choose a board that LabVIEW can control through the use of suit-
able driver software or DLLs. Alternatively, you can use a separate
software package to acquire images, write the data to a file, and then
import the data into LabVIEW, but that defeats the purpose of building
a real-time system.
In general, there are three classes of frame grabbers. First, there are
the low-cost multimedia types, sometimes implemented as an external
camera with a digitizer, such as the Connectix Quickcam. Also in this
class are cheap plug-in boards and the built-in video capability in some
Macintosh models. The primary limitations on these low-cost systems
are that they include no hardware triggering, so you can’t synchro-
nize image acquisition with an experiment, and they often contain an
autogain feature. Like automatic gain control in a radio, this feature
attempts to normalize the contrast, whether you like it or not. This can
be a problem for quantitative applications, so beware!
The second class includes midrange plug-in boards, such as the
BitFlow Video Raptor, the IMAQ PCI-1408, and Scion LG-3. These boards
have triggering, hardware gain and offset control, and various synchro-
nization options. They offer good value for scientific applications.
High-performance boards make up the third class. In addition to the
features of the midrange boards, you can install large frame buffers on
these boards to capture bursts of full-speed images for several seconds,
regardless of your system’s main memory availability or performance.
Special video sources are also accommodated, such as line scan cam-
eras, variable-rate scanning cameras, and digital cameras. Some DSP
functionality is also available, including built-in frame averaging, differ-
encing, and other mathematical operations, in addition to full program-
mability if you’re a C or assembly language hacker. Examples are the
BitFlow Data Raptor, Imaging Technology IC-PCI, and the Scion AG5.
Now that you’ve got an image, how are you going to show it to some-
one? For live display you can use your computer’s monitor from within
LabVIEW or another application; or you might be able to use an exter-
nal monitor. Hard-copy or desktop-published documents are always in
demand.
One catch in the business of video output from your computer is the
fact that high-resolution computer displays use a signal format (RGB
component video) that is much different than the regular baseband
NTSC or PAL standard format required by VCRs. There are several
solutions (again, multimedia demands easy and cheap solutions, so
expect to see some better ways of doing this). First, you can buy spe-
cial converter boxes that accept various RGB video signals and convert
 Data Visualization, Imaging, and Sound 577

them to NTSC. An example is TelevEyes from Digital Vision Corpora-

tion, which converts the signals from many 640-by-480 pixel resolution
RGB color displays to NTSC or S-video.
Another solution is to use a frame grabber board with video output
capability (most boards have this feature). You write processed images
to the output buffer, and the data passes through a DAC to create a
standard video signal. You can also use computer display boards that
generate NTSC outputs.

Using IMAQ Vision

IMAQ Vision is very well designed, and the included example pro-
grams make it pretty clear how you put together an image acquisi-
tion and processing application. Also, the manual is quite thorough.
You can also buy IMAQ Vision Builder, an IMAQ starter application
developed by National Instruments. It includes a comprehensive set of
image processing functions and makes it easy to try out various fea-
tures of IMAQ. What we’ll show you here are a few basic concepts and
a real world application.

IMAQ components
A complete IMAQ application really consists of two major parts. First is
the image acquisition driver, called NI-IMAQ, that talks to the frame
grabber. This driver, including its image acquisition boards, is free from
National Instruments. If you want to use a non-NI board, you can often
obtain driver VIs that are compatible with NI-IMAQ from Graftek. The
NI-IMAQ driver is quite similar to the NI-DAQ driver, including the
use of refnums and error I/O as common threads. You call the IMAQ
Init VI to establish a new session with a channel on your board and
then with various acquisition VIs that are organized in the familiar
easy and advanced levels. If you’ve used DAQ, this is a piece of cake.
The other part of IMAQ is the Vision package, which is a very
comprehensive image processing toolkit that costs about as much as
LabVIEW itself. Without Vision, you’re limited to the basic acquisi-
tion operations and some file I/O. With it, you’re limited only by your
knowledge of image processing. First-time users find that a bit daunt-
ing, because there are literally hundreds of functions in Vision, many
of which have optional parameters that seem to require a Ph.D. to
comprehend (what’s a Waddel Disk Diameter, anyway?). Your best
resources are the example VIs, the IMAQ manuals, and image process-
ing textbooks. Even so, we’ve been pleasantly surprised at how easy it
is to grab an image and extract key features. Let’s look further into how
you use IMAQ.
578 Chapter Twenty

Image handles. IMAQ carries information about an image (name, size,

resolution, data, and so forth) in a data structure that is accessible
by each function in the IMAQ library but hidden from LabVIEW. To
indicate which image you are working with, you use a cluster called an
image handle which is a kind of refnum. Image handles are normally
created by IMAQ Create and marked for destruction by IMAQ Dis-
pose. You must never fiddle with the value of an image handle, or the
program will cease to function (or worse). A new image handle must
be created for every independent image. Like all data in LabVIEW,
images (and their handles) are no longer accessible once the calling VI
finishes execution.
One advantage to this image handle scheme is that you have explicit
control of image creation and destruction, unlike the situation with
LabVIEW’s data types. If you create an image handle, load an image,
then submit the handle to 10 different image processing functions from
the IMAQ library, rest assured that there will be no duplication of the
data array (image). In contrast, a LabVIEW array, if passed in parallel
fashion to several different DSP subVIs, may require many duplicates.
You might be able to chain some functions in series, if you’re careful,
thus eliminating much duplication.

A simple acquisition and display application. Figure 20.16 shows a basic

IMAQ application that acquires and displays single frames with soft-
ware timing. You begin by calling IMAQ Init with the desired interface
name, which implies a board and channel number as defined in Mea-
surement Explorer. That allocates resources in the NI-IMAQ driver
and returns an IMAQ Session identifier that you pass to any other
acquisition VIs. Next, you need to allocate memory for an image with
the IMAQ Create VI, which returns an image handle.
At this point, you can begin an arbitrarily complex sequence of acqui-
sition and processing steps. For instance, you might want to do a con-
tinuous, double-buffered acquisition based on the high-level IMAQ

Initialize IMAQ Allocate image Acquire and display Close

board
IMAQ session
IMAQ Init.vi IMAQ Snap.vi IMAQ Close.vi
Image Name IMAQ Create
img0 image image handle
Interface
error
Name
8 bits IMAQ WindDraw
Image Type
stop
100

Figure 20.16 A simple IMAQ application that repeatedly grabs an image from a plug-in frame grab-
ber and displays it.
 Data Visualization, Imaging, and Sound 579

Grab Acquire VI or its lower-level components. In our case, we’ll use

the simpler IMAQ Snap VI to acquire frames one at a time. With it
placed in a While Loop, we get a frame every time the millisecond timer
fires, or as fast as the frame grabber can go.
To display an image, you have several choices. The fastest and easi-
est way is to use the built-in IMAQ image windows. And the simplest
way to do that is to call the IMAQ WindDraw VI. A whole family of
display VIs let you manipulate the window’s size, position, and color
palette. Display windows can even be interactive (more on that later).
Each time you call IMAQ WindDraw, the display is updated.
When the application is finished, you should call IMAQ Close to release
the hardware resources. You can also call IMAQ Dispose to release the
image memory, though this step actually is done for you behind the scenes.
Display windows remain visible unless you call IMAQ WindClose or
another display VI that explicitly makes the window disappear (the user
can also manually close the window).

Interactive windows. Many applications require user interaction with

the image. In IMAQ, users can draw arbitrary shapes, including text,
or use a variety of tools to highlight a certain area of the image. For
instance, you may want them to select a desired area for analysis,
known as a region of interest (ROI). IMAQ Vision includes a set of
tools that permit the user to draw shapes and text on the image and
then lets your program find out what they drew. There’s even a tool
palette that you can customize and show/hide programmatically.
Figure 20.17 shows a simple example of ROI selection. An image is
read from a file and then displayed. The IMAQ WindToolsSelect VI
chooses the tool type, in this case a rectangle. At any time, the user can
click and drag in the image window. When you call the IMAQ Wind-
GetROI VI, it returns the coordinates and type of ROI object that was
drawn. Your real-world applications may become quite complex. If you
need to manage a great deal of user interaction, consider writing a state
machine that handles the various modes of operation and updates the
image at the appropriate times.

Image files. IMAQ Vision has a set of VIs that make it easy to read
and write files in standard image formats, including TIFF, JPEG, BMP,
AIPD, and PNG. No data conversion steps are required. All you need
to do is wire your image handle to IMAQ Write File or IMAQ Read
File.

An image processing example. As we’ve said, image processing is a

fairly complicated field of study. But there’s at least one very famil-
iar sequence of steps that we’ve seen so often that it qualifies as a
580 Chapter Twenty

Accept ROI ROI Descriptor

Global Rectangle Contours

Accept
0 23 x1Left 0
ID
47 y1Top External 0
145 x2Right Type

203 Rectangle 3
y2Bottom
Coordinates
0 23

IMAQ IMAQ IMAQ Imaq

Create ReadFile WindDraw Wait for user to
WindGetROI
select ROI area
My Image

ROI
Descriptor
IMAQ
WindToolsSelect 100

Tool Rectangle

Accept ROI

Figure 20.17 The simplest example of reading the description of a user-selected

region of interest.

canonical image processing example. What we’re going to do is locate

one or more important particles in the image. A particle is generally
defined as any closed path. One way to do this is to use the sophisti-
cated pattern matching VIs in IMAQ. In that case, you define a tem-
plate by selecting a typical object in an image, and then the matching
routine attempts to locate similar objects, regardless of their position
or rotation. It’s very powerful, and it is certainly worth considering. But
for now we’re going to concentrate on a more conventional stepwise
approach that’s very flexible and easy to optimize.
The first step is to turn an acquired gray-scale image into a sim-
pler bitmap image. This step is known as thresholding, and there
are many algorithms available to do it. Try using the IMAQ AutoB-
Threshold VI, which features five different algorithms that you can
test on your data. What’s nice about auto-thresholding is that it does
a good job of normalizing the image, thus reducing problems related
to overall contrast and brightness. Figure 20.18 is a complete work-
ing example based on thresholding. It came from a project that Gary
worked on involving the alignment of a telescope, where the objective
was to locate the centroids of two concentric circles. Let’s see how it
works.
Starting with an acquired image (read from a file in this case), we
apply IMAQ AutoBThreshold to locate the 50 percent threshold value.
 Data Visualization, Imaging, and Sound 581

0 Red 255 Window Num 1

Bitmap Green 255
Palette Edge Detected
Blue 255
IMAQ
IMAQ IMAQ
AutoB
IMAQ IMAQ Multi Imaq WindDraw
Threshold
Create ReadFile Threshold IMAQ Morphology EdgeDetection

Raw
Coefficients
(2D)
IMAQ IMAQ
path clustering Close Open Roberts Complex Complex
Method Operation Operation Method Particle Measure

Processed Image X center of mass 8 Parameters to

IMAQ measure
Create Y center of mass 9

Figure 20.18 An image processing example that uses thresholding, morphology, and spatial filter-
ing to locate concentric circles. It’s deceptively simple.

Next, IMAQ MultiThreshold uses that threshold value to convert the

grayscale image into a binary one. Instead of overwriting the original
image, we chose to create a new one for the binary image that will be
further processed. In the images in Figure 20.18, you can see how the
original noisy pattern has been transformed into a fairly clear binary
pattern. By default, the IMAQ WindDraw VI displays an image as
grayscale. But if you want to display a binary image, you need to set the
color palette as shown in the figure; otherwise, all you’ll see is black.
Here’s a trick for troubleshooting your image processing sequence.
You can wire IMAQ WindDraw to any image on the diagram to see what
has happened at that step. But something unexpected may appear: You
always see the final result of the processing sequence no matter where
you connect IMAQ WindDraw. Why is that? On a fast computer, a race
condition may occur where the whole chain of VIs executes before IMAQ
WindDraw actually copies image data to the window. Remember that
the image is not duplicated when it’s passed along through a chain of
IMAQ VIs. If you single-step your program, you will see the expected
image at the step where you’ve made the connection. A solution to this
problem is to break the chain between VIs and wire the input to the
following IMAQ VI to an empty image constant. Pop up on the VI’s
Image Source input and select Create Constant. That will keep the
downstream chain happy without overwriting your other image.
582 Chapter Twenty

Our thresholded image isn’t perfect, as you can see, because it con-
tains some holes in the dark area due to noise in the original image.
There are many ways to filter out noise-related pixels and other small
aberrations, but the techniques that work best in this case are dila-
tion, eroding, opening, and closing:

■ Erosion eliminates isolated background pixels (white specks on a

black background).
■ Dilation is the opposite of erosion (i.e., it fills in tiny holes).
■ Opening is an erosion followed by a dilation.
■ Closing is a dilation followed by an erosion.

You really have to see these morphological transformations in action

to understand how they work, and you need to try various sequences
of transformations on your image to see what works best. IMAQ Mor-
phology is the VI that does all of these operations and more. In our
case, we needed a close followed by an open to clean up the image.
Since our objective is to extract the concentric circles, the next step
is edge detection, a kind of spatial filtering that locates contours in
an image. The IMAQ EdgeDetection VI has several methods, and again
you should experiment. The final image in the figure shows the result-
ing clean, concentric circles.
Our final step is measuring the locations of the circles. IMAQ Com-
plexParticle locates all the particles in a binary image and returns a
set of statistics, called a complex report, for each one. Included in the
output cluster array are such items as area, perimeter, a global rect-
angle, and statistical summations that are useful in other calculations.
To make sense of this raw data, the complex report is passed to IMAQ
ComplexMeasure. The image processing community has devised a
large number of standardized ways to quantify various attributes of a
particle, and this VI apparently does them all. You supply an array of
parameters (attributes) that you wish to measure, and the VI returns
a 2D array containing a value for each parameter on each particle. In
this case, we asked for the X and Y centers of mass, or centroid, so the
VI returns a 2 by 2 array. All you have to do is look at the values in the
array to determine where the circles are in relation to each other.
As you can see, this apparently simple task of locating a couple of
circles took quite a few steps. Also, you can imagine that no image pro-
cessing routine is 100 percent bulletproof. Given a noisy or distorted
image, it may fail. Do plenty of testing with real images, perhaps using
Vision Builder, to optimize your program. Also consider assigning alarm
limits to some of those complex measurements to validate the results of
your algorithm if it’s used in real time.
 Data Visualization, Imaging, and Sound 583

X Resolution 320 Picture in Image in picture

Image Y Resolution 240 Draw 8-bit Pixmap.vi

(a Picture function)

Bi-Linear Index Array

Interpolation Type Image in picture
IMAQ IMAQ Bundle
Resample ImageToArray Array DrawAreaSize
Size

Figure 20.19 IMAQ images can be converted to LabVIEW Pictures for on-panel display.

Other ways to display images. You don’t have to display your images in
a floating IMAQ window, especially if you don’t need any user inter-
action. A simple way to place the image on a LabVIEW panel is to
use a Picture indicator. Figure 20.19 shows a practical example that
includes resampling of the image to adjust its size without losing too
much information. The IMAQ Resample VI lets you choose the X and
Y resolution and one of four methods for interpolation. This step is not
required if your IMAQ image is exactly the size that you want it to be
in the Picture indicator. The next step is to transfer the image data
from the special IMAQ format into a LabVIEW 2D array, which is done
by the IMAQ ImageToArray VI. That’s a useful function any time
you need to manipulate image data in a manner not already included
in IMAQ. Its complement is IMAQ ArrayToImage. Finally, you use
one of the Picture functions such as the Draw 8-bit Pixmap VI to
convert raw 2D data into a picture data type for display.
We added an extra feature to this example. The Picture indicator is
automatically sized to fit the image. You get the array dimensions (cor-
responding to image width and height), bundle them up, and pass them
to a Property Node for the Picture. The DrawSizeArea property is the
one you need to set.

LabVIEW has another means

Intensity Graph for built-in image display.
by which you can display images right on the front panel: the Inten-
sity Graph. The Intensity Graph accepts several data types, among
them a 2D array corresponding to an image. We wrote a simple VI
(Figure 20.20) to load a raw binary image from a file and format the
data for display in an Intensity Graph. Because a raw binary file con-
tains no header to describe the dimensions of the image, the user has
to supply that information. The control called Rows supplies this key
information. If you enter the wrong number, the image is hopelessly
scrambled.
Looking at the diagram, the file is read as a string which is then con-
verted to an array of U8 using the String to Byte Array conversion
function. A For Loop breaks the long array of values into rows, which
are then placed in a 2D array for display. An Attribute Node for the
584 Chapter Twenty

Intensity Graph
0 255

20 225
40
200
60
175
80
150
100
125
120
100
140 Rows
75 195
160
50
180
195 25 Bytes in file
0 25 50 75 100 125 160 31200
0

Split off one row at a time

Read Characters Rows from the raw image array
From File.vi String To Intensity
Byte Array For each row...
Graph

Array
Subset
Intensity Graph
Bytes
Property Node
in file
columns X Scale.Range:Maximum
X Scale.Range:Minimum
0.00
Y Scale.Range:Maximum
rows Y Scale.Range:Minimum
Plot Area:Size:Width
Plot Area:Size:Height

Figure 20.20 The Intensity Graph displaying an image from a binary file. The
user has to know the size of the image—particularly the number of rows. The
raw data string is converted to an array, which is then transformed into a 2D
array suitable for display.

Intensity Graph sets the y- and x-axis scales to fit the actual number of
rows and columns in the image. Note that the y axis has zero at the top,
rather than its usual place at the bottom. That makes the image come
out right-side up. You also have to select Transpose Array from the
pop-up menu on the Intensity Graph to swap the x- and y-axis data.
An important tweak that you can make is to match the number of
displayed pixels to the data. Ideally, one array element maps to one
pixel, or at least there is an integral ratio between the two. This reduces
aliasing in the image. In this example, we made this adjustment by set-
ting the Plot Area Size (x and y values) with the Property Node.
The Ramp control (part of the Intensity Graph) sets the data range
to 0–200, corresponding to a pleasing gray scale ranging from black to
white. You can adjust the gray scale or color gradations by editing the
numeric markers on the Ramp. Pop up on the Ramp control and set
 Data Visualization, Imaging, and Sound 585

the marker spacing to Arbitrary. Pop up again to add or delete mark-

ers. You can then change the value of each marker and slide it up and
down the ramp to create breakpoints in the (normally linear) mapping
of intensity values to display brightness or color. For fun, you can pick
various colors or put in palette breakpoints by adding elements to the
Color Table, another item available through the Attribute Node. This is
a means of generating a kind of color lookup table (CLUT).
Colors are stored as unsigned long integers (U32) with the following
representation. The least-significant byte is the value of blue, the next
byte is green, the next is red, and the most-significant byte is unused,
but is normally set to zero. (In some applications, this extra byte is
called the alpha channel and represents transparency.) For instance,
100 percent red (255 in the red byte) plus 50 percent green produces
orange. This is also known as 32-bit color. To select a color, you have
several options. On the diagram, you can place a Color Box Constant
(from the Additional Numeric Constants palette), then pop up on it
to choose a color from those available in LabVIEW. A corresponding
control on the panel is the Color Box constant from the Numerics pal-
ette. A related control, the Color Ramp, allows you to map an arbitrary
number to an arbitrary color. All of these controls and constants use a
visual, rather than numeric, representation of color. Remember that
the actual color on your monitor is limited to the palette permitted by
LabVIEW, which in turn is limited by your monitor setting.
For arbitrary colors, you must generate an appropriate U32 number.
Figure 20.21 shows two of the possible ways to combine values for red,
green, and blue into a composite RGB value. The data conversion func-
tions (Type Cast and Join Numbers) come from the Advanced >>
Data Manipulation palette. You might use this technique for combining
separate RGB color image planes into a single, displayable color image.
The complementary operation is also feasible, using the Split Num-
bers function to break a color into its RGB components.

Figure 20.21Colors can be generated programmatically, but it requires some data

manipulation tricks.
586 Chapter Twenty

Sound I/O
Recording and reproducing sound can be useful for scientific analysis,
operator notification, or as a novelty. The whole idea is to convert acous-
tic vibrations into an electrical signal (probably with a microphone),
digitize that signal, and then reverse the process. In most cases, we use
the human audible range of 20 Hz to 20 kHz to define sound, but the
spectrum may be extended in both directions for applications such as
sonar and ultrasonic work. Depending upon the critical specifications
of your application, you may use your computer’s built-in sound hard-
ware, a plug-in board (DAQ or something more specialized), or external
ADC or DAC hardware. As usual, a key element is having a LabVIEW
driver or other means to exchange data with the hardware.
The simplest sound function that’s included with LabVIEW is the
Beep VI, found in the Sound function palette. On all platforms, it plays
the system alert sound through the built-in sound hardware.

DAQ for sound I/O

An ordinary DAQ board is quite capable of recording and reproducing
sound up to its Nyquist sampling limit. The only technical shortcom-
ings of such a board may be in the areas of harmonic and intermodu-
lation distortion, the lack of built-in antialiasing filters, and perhaps
a dynamic range limitation with 12-bit boards. For those reasons
National Instruments developed its dynamic signal acquisition boards,
featuring antialiased 16-bit inputs and outputs and very low distor-
tion. In any case, all you need is the standard DAQ library to do the
recording and playback in LabVIEW.
The maximum frequency that your system can process continuously
is probably limited by CPU performance. The keyword is process: Do
you want to do real-time filtering, FFTs, and display of signals with a
40-kHz bandwidth? You probably won’t make it, at least with today’s
general-purpose CPUs and LabVIEW. Perhaps a DSP board can take
care of the numerical processing; check out the DSP products from sev-
eral vendors, but remember that extra programming—usually in C—is
required. At lower frequencies, though, you can certainly build a usable
real-time system. Be sure to run some benchmarks before committing
to a real-time DAQ approach.
The good news about DAQ for sound applications is that it’s easy
to use. Review the analog I/O part of this book for details on single-
shot and continuous waveform acquisition and generation. For wide-
bandwidth signals, disk streaming is usually the answer; and again,
the solutions are built into the DAQ and file I/O libraries.
A recurring issue with sound I/O is that of file formats and stan-
dards. On each platform, there are certain file formats that the
 Data Visualization, Imaging, and Sound 587

operating system can conveniently record and play back through stan-
dard hardware. Therefore, it’s desirable to use those formats for your
own, DAQ-based sound I/O. In the sections that follow, we’ll look at
some options for handling those formats.

Sound I/O functions

Once upon a time, there were no sound I/O functions built into
LabVIEW, so we relied on a variety of third-party and donated VIs.
Now we have a full array of VIs that provide multiplatform support
for sound input, output, and various sound file formats. The sound I/O
VIs support standard kinds of sound hardware that you’ll find on any
Macintosh or PC, such as the SoundBlaster. As long as your hardware
can be accessed by the ordinary operating system interfaces, LabVIEW
should be able to access it as well. Signal types include monaural and
stereo in 8- or 16-bit formats using offset binary coding. Several sam-
pling rates are available, including 11,025, 22,050, and 44,100 samples
per second.
National Instruments made things easy for us once again by design-
ing the sound I/O VIs much like the DAQ library. You configure a port,
read or write data, and start/stop these buffered operations in the obvi-
ous way. Examples included with LabVIEW cover most of the features
you’ll be interested in using. Let’s take a quick look at some input and
output applications.

Sound input
It’s very easy to set up an input data streaming application, as shown
in Figure 20.22. The SI Config VI allocates an input buffer and defines
the type of data and sampling rate, much like its DAQ cousin, AI
Config. Then you call SI Start to begin a continuous buffered acquisi-
tion. There’s typically no triggering hardware available on consumer-
grade sound boards, so the SI library does not support triggering.
(Perhaps National Instruments could add software triggering like the
NI-DAQ driver has.) Once the acquisition is started, you put the SI
Read VI in a loop to read buffers as they become available. If you don’t
read often enough, you’ll get a buffer overflow error. Adding a timer in
the loop prevents the SI Read operation from taking up all the CPU
time while it waits for the next available buffer. Data is returned as I16
or I8 arrays; stereo data is 2D. When you’re done, call SI Stop and then
SI Clear to free up memory.
If all you want to do is a single-shot acquisition of sound, there’s a
simpler VI in the sound palette, called Snd Read Waveform. It com-
bines all the steps of Figure 20.22, but without the While Loop, and
returns all possible data types (mono, stereo, 8- and 16-bit).
588 Chapter Twenty

Sound format Mono 16-bit data

sound quality 200

mono
100
rate

Amplitude
44100 0

bits per sample

-100
16 bit
-200
Buffer size

11025 -300
766662 771071
STOP Time

Mono 16-bit data

Sound
format SI Config.vi SI Start.vi SI Stop.vi SI Clear.vi
taskID

error
SI Read.vi
Buffer size
100

stop

Figure 20.22 A chain of sound input VIs performs a hardware-timed, buffered acquisi-
tion from your computer’s built-in sound hardware.

Sound output
Sound output functions are similar to those for input, with the addi-
tion of a few extra features that help control the flow of data. In
Figure 20.23, a waveform is computed and then written to the output
buffer by the SO Write VI. In contrast with sound input behavior, this
buffer size is variable and depends only on the amount of data that
you pass to SO Write. Once the buffer is filled, you call SO Start to
generate the signal. If the buffer is fairly long and you need to know
when it’s done, you can call the SO Wait VI, which returns only when
the current sound generation is complete. It’s also possible to call the
SO Pause VI to temporarily stop generation; call SO Start again to
restart it.
For single-shot waveform generation, you can use the Snd Write
Waveform VI, which performs the configure, write, start, wait, and
clear steps.

Sound files
Included with the sound VIs are utilities to read and write Windows
standard .wav files with all available data types and sample rates.
They’re called Snd Read Wave File and Snd Write Wave File, and
 Data Visualization, Imaging, and Sound 589

Sound format samples Waveform Graph

sound quality 10000 40000

mono amplitude

rate 30000 Play Sound

11025 f1 PLAY
bits per sample 0.08

16 bit f2
-40000
0.15 STOP 1 10 100 1000 10000

samples Chirp Pattern.vi

amplitude
f1 Waveform Graph

f2
True
Sound Play Sound
format SO Config.vi SO Write.vi SO Start.vi SO Wait.vi SO Clear.vi
taskID

error

100
stop

Figure 20.23 Sound output is as easy as sound input. In this example we generate a cool sound-
ing chirp each time the Play button is pressed.

they work on all platforms. The interface to these VIs meshes with the
sound-in and sound-out VIs by including the sound format cluster and
the four data array types.
For Macintosh users who need access to AIFF format files, Dave
Ritter of BetterVIEW Consulting has created a CIN-based VI to play
them. It’s available at www.bettervi.com.

Bibliography
Cleveland, William S.: The Elements of Graphing Data, Wadsworth, Monterey, Calif.,
1985.
Gonzalez, Rafael C., and Paul Wintz: Digital Image Processing, Addison-Wesley, Reading,
Mass., 1987.
Tufte, Edward R.: The Visual Display of Quantitative Information, Graphics Press,
Cheshire, Conn., 1983.
Wang, Peter C. C.: Graphical Representation of Multivariate Data, Academic Press,
New York, 1978.
This page intentionally left blank
 Index

“\” Codes Display, 110 AN084 (Using Quadrature problem definition, 186–191
Encoders with E Series process for, 196–197
DAQ Boards), 523 pseudocoding, 203–204
A AN114 (Using LabVIEW to Create specifications definition, 187–189
Abortable wait, 156 Multithreaded VIs for user interface prototyping for,
Aborting, of For Loops, 67 Maximum Performance and 192–196
Absolute encoders, 523 Reliability), 145, 389–390 user needs analysis for, 186–187
Absolute timing functions, 139, AN154 (LabVIEW Data Storage), with VI hierarchy, 201–202
145–146 132 Application development
AC & DC Estimator, 544 AN168 (LabVIEW Performance (for LabVIEW FPGA),
AC signals, 297 and Memory Management), 406–414
Ac time domain, 280 122, 548, 551 clocked execution, 411–413
Accessibility, random, 177 Anafaze controllers, 462 compiling, 406–408
Acquire Semaphore VI, 164 Analog design, 314 debugging, 408
Action Instruments, 292 Analog Devices 6B Series, 302, 303 parallelism, 413, 414
Active filters, 292 Analog filters, 308 pipelining, 413–415
ActiveX objects, 126 Analog IO Control Loop synchronization, 408–411
Actual scan rate, 499 (hw timed), 499 Application events, 81
Actuators, 278–279 Analog outputs, 299 Application Note 084 (Using
Acyclic dataflow, 491 Analog signals, 305 Quadrature Encoders with E
A/D (analog-to-digital) system, 317 Analog transmitters, 447 Series DAQ Boards), 523
Adaptive controllers, 446 Analog voltage, 278 Application Note 114 (Using
ADCs (see Analog-to-digital Analog-to-digital converters LabVIEW to Create
converters) (ADCs), 309–314 Multithreaded VIs for
Addressing, 254 characteristics of, 309 Maximum Performance and
Adobe Illustrator, 237 coding schemes for, 315–316 Reliability), 145, 389–390
Aerotech Unidex, 523 errors in, 311 Application Note 154 (LabVIEW
AI Clear, 532 multiplexing vs., 313 Data Storage), 132
AI Config, 531 signals and, 275 Application Note 168 (LabVIEW
AI CONVERT, 385 use of, 305 Performance and Memory
AI Read, 531, 549 Analog-to-digital (A/D) system, 317 Management), 122, 548, 551
AI Start, 531 AnaVIEW, 462 Application program interface
AI STARTSCAN, 385, 499 Anderson, Monnie, 64 (API), 28, 245
Air Force Research Lab Antialiasing, 292 Applications:
MightySat II.1, 34 Antialiasing filter, 307 event-driven, 100–102
Alarm acknowledge, 508 Anti-reset window, 445 mega-applications, 22
Alarm annunciator panels, 512 AO Clear, 532 Applied Research Laboratory, 11
Alarm Handler VI, 509–511 AO Config, 531 Approach control gain
Alarm handlers, 508–511 AO Start, 531 adjustment, 496
Alarm summary displays, 465 AO UPDATE, 385 Arbitrary waveform generator, 532
Alarms, 506–512 AO Write, 531 Architecture, 87, 200
alarm handler for, 508–511 Aperture time, 313 Arnold, Tom, 34
for operator notification, 511–512 API (see Application program Array constant, 112, 113
Algorithms, 51 interface) Array Size function, 115
Aliasing, 306 Append True/False String Array to Cluster function, 130, 135
Allen-Bradley: function, 107 Array To Spreadsheet String, 111
Data Highway protocol, 457, 462 Apple Computer, 17 Arrays, 53–54, 114–122
PLCs, 244 Application Builder (LabVIEW), clusters and, 124
Alliance Program, 244 25, 26 coercion/conversion when
Alpha channels, 585 Application design, 185–218 handling, 549
Ames Research Center’s block diagrams for, 189–191 defined, 114
HFFAF, 546 with DAQ Assistant, 197 empty, 69
Amperes (Current) controls, 78 debugging, 204–207, 214–218 FPGA (see Field-programmable
Amplifier(s), 291, 522 design patterns for, 200–201 gate arrays)
for instrumentation, 289 for error handling, 207–213 initialization of, 117–119
isolation, 291, 515 I/O hardware specification for, memory usage/performance
lock-in, 545 191–192 of, 119–122
PGIA, 314 with LabVIEW sketching, one-dimensional, 69
Amplitude measurements, 280 202–203 shift registers and, 431
AMUX-64T, 525 for modularity, 198–200 two-dimensional, 112

591
592 Index

Arrays of arrays, 124 Blocked (thread), 391 CAMAC (Computer Automated

ArrayToImage (IMAQ), 583 BMP (embedded image) files, 236 Measurement and
Artificial data dependency, 63 BNC-2120 accessory, 523 Control), 518
ASCII text files, 3, 167, 176 Board Support Package (BSP), 421 Cancellation, 171
(See also Text files) Boards: Canvas (program), 236
ASCII text format, postrun DAQ, 148 Capacitive coupling, 286
analysis in, 328–329 interface, 149 Capacitor, 286
AstroRT, 34 plug-in, 253–254 Cartesian data, on graphs, 558–563
ATE (automated test Bogus data, 299 CASE (see Computer-aided
equipment), 251 Boolean Array To Number software engineering
Attribute Node, 24, 583–584 function, 130 applications)
Attributes, 126, 127 Boolean control, 478 Case structures, 432
AutoBThreshold (IMAQ), 580 Boolean logic functions, 486–487 Caution (see Alarms)
Auto-indexing, 66 Boolean strings, 262–263 C-callable routines, 149
Automated test equipment Boolean table, 486 CERN, 519
(ATE), 251 Boosting, priority, 389 Certification exams, 218–229
Automatic control, 444 Bottom-up structured design, (See also specific types,
Automatic error recovery, 261 197–198 e.g.: Certified LabVIEW
Automatic parameter tuning, 496 Boxcar averager, 518, 542 Developer)
Automatic save and recall, 367–370 Brain board, 302 Certified Instrument Driver
Automation, 2–4 Brazing, 187 Developers, 244
Autonomous VIs, 92–94 Breakpoints: Certified LabVIEW Architect
AutoScale Z, 571 for debugging, 57–58 (CLA), 217
Auto-setup, 272 defined, 57 Certified LabVIEW Associate
Autotuning, PID with, 492 setting of, 57–58 Developer (CLAD), 218–221
Average, 540 Suspend When Called, 58 Certified LabVIEW Developer
Averaging (signal sampling), BridgeVIEW, 26 (CLD), 218, 221–229
308–309 Brightness, 575 example questions for, 223–229
Avoidance, of sequence Brunzie, Ted, 214 grading tenets of, 222
structures, 62 BSP (Board Support Package), 421 Change Log File Binding, 182
Buffered I/O, circular, 374–375 Change-of-state detection, 69, 507
Buffering, 119, 386–387 Channel (top index), 120
B BUFG, 407 Channel Wizard (DAQ), 26
BUFGMUX, 407 Charge pump-out, 314
Backups, 217 Bug day approach, 19 Chart History, 499
Balanced signals, 51, 290 Build Array, 114, 559 Chartjunk, 557
Bancomm, 149 Build Cluster Array, 559 Charts:
Bandwidth, signal, 343–344 Build Path, 168 control, 506
Base clock, 407 Build Timing Source Hierarchy VI, flowcharts, 13
Basic Averaged DCRMS VI, 125 141, 397, 398 intensity, 571–572
BASIC programming, 9, 18 Build VI, 425 SFCs, 490
Batch controllers, 496 Build Waveform function, 126 strip, 464–465
Batch processes, 446 Built-in global variables, 73–75 Chernoff face, 568, 569
Beep, 586 Bundle By Name, 122, 468, 475 CINs (see Code Interface Nodes)
Benchmark products (LabVIEW), Bundle function, 122, 559 Circular buffer, 500, 563
34–36 Byte stream files, binary-format Circular buffered I/O, 374–375
Best practices (LabVIEW (see Binary-format byte CLA (see Certified LabVIEW
Embedded), 431–434 stream files) Architect)
BetterVIEW Consulting, 589 Bytes at Serial Port, 57 CLAD (see Certified LabVIEW
Bias current, 296 Associate Developer)
Big-bang controllers, 444 Classes, of field devices, 448
Big-endian (byte ordering), 131
Binary files, 167, 169–170, 176–181
C CLD (see Certified LabVIEW
Developer)
ASCII files vs., 176 C codes: Clear Timing Source VI, 143
reading, 179–181 for digital timing, 149 Clear to Send (CTS) line, 247
writing, 178–179 generator and generation Clients, 90
Binary formats, custom, 329–331 of, 419, 420, 427–428 Client-server, 90–94
Binary-format byte stream files, in LabVIEW Embedded, 422–424 Client-server model, 476
169–170 (See also C Runtime Library, 422–423 Clipping, 310
Binary files) Cabling, 254 Clock(s):
Bipolar converters, 315 Calibration, of transient digitizers, base, 407
Bivariate data, on graphs, 563–565 535–536 crystal, 148
Block diagrams, 12–13, 48, Call by reference nodes, 471 time base, 316
189–191 Call Libraries, 59, 434 (See also Timing)
 Index 593

Clocked execution, 411–413 Computer Automated PIC VIs and, 497–498

Close (IMAQ), 579 Measurement and Control timing/performance limitations
Close File, 63, 64, 171 (CAMAC), 518 on, 498–499
Close VIs, 272 Computer languages, 8 Continuous data:
Closing, 582 Computer-aided software real time analysis with, 339–340
Clump, 41 engineering (CASE) single-shot data vs., 338
Cluster Size, 130 applications, 97, 98 Continuous time functions, 305
Cluster To Array, 129, 130 Concatenate Strings, 106 Continuous-value functions, 305
Clusters, 50–52, 122–125 Concept V.i program, 574 Contour plot, 564
arrays and, 124 Conditional terminal, 65 Contrast, 575
contents of, 262 Configuration cluster array, 467 Control(s):
Customize Control and, Configuration management, automatic, 444
51–53, 122 350–370 continuous, 446, 492–499
defined, 122 editors for, 352–362 control strategy for, 493–496
numeric types and, 115 interactive editors, 355–357 I/O values in, 496–497
saved as strict typedef, 51 for menu-driven systems, PIC VIs and, 497–498
strict typedef and, 122 360–361 timing/performance
in time-critical code, 434 pop-up editors, 357–360 limitations on, 498–499
typedefs and, 122 static editors, 353–355 description documentation of,
CMRR (common-mode rejection status displays for, 360–361 232–233
ration), 290 recording in, 365–370 feedback, 453
Code boolean, 262 automatic save and recall, initializing, 482
Code Interface Nodes (CINs), 367–370 Iteration, 540
58–59 manual save and recall, 367 manual, 444
Code width, 310 Configuration VIs, 268–269 process (see Process control
Codes: Conflicts, in namespace, 161 applications)
insertion of other, into LabVIEW, Connections, 284–304 sequential, 446, 486–492
58–59 amplifier use for, 291–294 boolean logic functions in,
user-defined, 265 differential, 289, 290 486–487
(See also specific types, e.g.: differential/single-ended, GrafcetVIEW and, 490–492
C code) 288–291 initialization problems in,
Coding schemes: grounding/shielding of, 284–291 489–490
of Sequence structures, 62 of differential/single-ended state machine architecture
for signal sampling, 315–316 connections, 288–291 for, 487–488
Coercion: electromagnetic fields and, Control charts, 506
arrays and, 549 286–288 Control Editor (see Customize
conversion and, 128–129 error sources in, 288 Control)
Coercion dots, 115, 432, 550 ground, 284–286 Control events, 82
Col values, 181 of input signals, 294–298 Control flow, 38
Color, 585 AC signals, 297 Control limits, 505
Columns, 116, 121 digital inputs, 297–298 Control Online Help function, 233
Command rate, 270 floating analog inputs, Control reference, 45
Common threads, 63–64 294–295 Control system architectures,
Common-mode rejection ration ground-referenced analog 449–455
(CMRR), 290 inputs, 294 distributed control system,
Common-mode signal, 289 thermocouples, 295–296 449–451
Communication(s): with networking, 303–304 with personal computers,
with instrument drivers, of output signals, 298–299 451–452
251–256 required, 239 selection of, 453–455
LabVIEW RT and, 398–399 single-ended, 288 Supervisory Control and Data
of PLCs, 457–458 Connector pane picture, 239 Acquisition, 451
standards for, 245–249 Consistency, 44 Controller(s):
CompactRio (cRIO), 404–405 Constants: adaptive, 446
CompactVision system, 405 array, 112, 113 batch, 496
Comparator, 317 looping and, 431 big-bang, 444
Comparison functions, 507 Consumer events, 160, 210 desired, 467
Compiled help files, 232 dequeuing by, 210 Hysteresis On/Off Controller, 493
Compiled rate, 408 design pattern with, 209 on/off, 444
Compile-time polymorphism, 50 Cont Acq to File (scaled), 180 PI, 445
Compiling, 406–408, 423–424 Context switching, 377, 393–394 PID, 444
Complex reports, 582 Continuous control, 446, predictive, 446, 496
ComplexMeasure (IMAQ), 582 492–499 proportional, 444
ComplexParticle (IMAQ), 582 control strategy for, 493–496 smart (see Smart controllers)
Compression, data, 503 I/O values in, 496–497 Controller faceplate displays, 465
594 Index

Controller signals, 447

Controls (LabVIEW), 45, 46
D in custom binary formats,
329–331
Conversion(s), 127–135 D/A (digital-to-analog) system, 317 database for, 335–336
arrays and, 549 DACs (see Digital-to-analog in datalog file format, 326–327
coercion and, 128–129 converters) Datalog File Handler VI for,
of data from one type to DAQ (see Data Acquisition 327–328
another, 128 Program) direct links in, 331–332
by Flatten to String, 132–133 DAQ Assistant, 197 timestamps for, 332–333
type casting and, 134–135 DAQ boards, 148 real time, 337–343
type setting and, 129–132 DAQ cards, R-series, 403–404 with continuous data, 339–340
Conversion speed, 310 DAQ Channel Wizard, 26 display performance for,
Convert 7.x Data, 133 DAQ Library, 297 342–343
Converters: DAQ VIs, 125 with single-shot data,
bipolar, 315 DAQmx Create Timing Source VI, 338–339
time-to-digital, 518 141, 397 speed of, 340–341
Cooperative multitasking, 41–42 Data, 50–54, 103–135 volume of data for, 341–342
Copy, 169 arrays, 53–54, 114–122 Data buffer, 119
CorelDraw, 236 bogus, 299 Data dependency, 62–63
Coriolis spacecraft (Naval Cartesian, 558 Data descriptor, 128
Research Lab), 34 checking of, 55, 56 Data dictionary, 198
Cost, 455 clusters, 50–52, 122–125 Data distribution (process control
Count, 180, 183 compression of, 503 applications), 475–486
Count Buffered Edges, 386–387 continuous, 338–340 with display VIs, 479–482
Count input, 173 conversion of, 128 with input scanners, 476–477
Coupling, 286 conversions of, 127–135 of output data, 477–479
Crash cart, 3 Convert 7.x Data, 133 real-time process control
Crashing (computer), 454 enumerated types, 133–134 databases for, 484–485
Create... Property Node, 45 multivariate, 565 using network connections,
Create (IMAQ), 578 numeric, 104–105 482–484
Create Control, 472 polymorphic, 68 validation and, 485–486
Create Notifier, 158 for real time analysis, 341–342 Data fields, variant, 126
Create Rendezvous VI, 165 single-shot, 338–339 Data Highway Plus (Allen-Bradley),
Create Semaphore VI, 163 strings, 105–114 457, 462
Create SubVI, 198 typedefs, 52–53 Data input, 179
Creep, feature, 25 volume of, 341–342 Data Manipulation palette, 131
cRIO (CompactRio), 404–405 waveform, 558 Data quantity:
Critical selections, 161 waveforms, 125–127 memory optimization for,
Crowbar (short-circuit), 516 Data Acquisition Program (DAQ), 547–553
Crystal clocks, 148 323–375 in physics applications,
CTS (Clear to Send) line, 247 boards for, 148 546–553
Current (amperes) controls, 78 Channel Wizard in, 26 reduction of, 546–547
Current measurements, 516–517 configuration management Data Range, 205
Current Messages, 510 features of, 350–370 Data Scroll, 501
Current shunts (current viewing compilers for, 362–365 Data Server VI, 482
resistors), 516 editors for, 352–362 Data sheets, 300
Current transformer, 516 recording in, 365–370 Data Storage (Application Note
Current viewing resistors data analysis for, 325–343 AN154), 132
(current shunts), 516 postrun analysis for, 326–336 Data storage/sampling, in transient
Currents, 296 real time analysis, 337–343 digitizers, 534–535
Cursor names, 567 low-speed example of, 370–372 Data strings, 132
Curve Graph, 3D, 570 medium-speed example of, Data structures, 51
Custom binary formats, for postrun 373–375 Data terminal, 181
analysis, 329–331 sampling in, 343–350 Data types:
Custom online help, 233–234, 242 digital filtering for, 344–350 in CLAD, 219
for document distribution, 242 oversampling, 344 determination of, 182
as documentation, 233–234, 242 signal bandwidth and, 343–344 Data VIs, 270–271
Custom Probe, 55 for sound I/O, 586–587 Data visualization (see Graphs)
Customize Control (Control VIs for, 125 Database(s):
Editor), 193 (See also under DAQ) for debugging, 421
clusters and, 51–53, 122 Data analysis, 325–343 for postrun analysis, 335–336
sliders in, 466 postrun, 326–336 for process control, 484–485
use of, 193 in ASCII text format, 328–329 real-time, 475, 476, 484
CVI (LabWindows), 33 configuration information for, Database-driven software
Cycles (closed loops), 491 333–335 environment, 484
 Index 595

Dataflow, 38 Delays: Digital Clock Manager (DCM), 407

acyclic, 491 interchannel, 313 Digital filtering, 344–350
programming with, 38–39 lag, 444, 492 analog filters and, 308
structured, 13 trigger, 537 FIR filters, 345–347
Dataflow diagrams, 13 Delete, 169 IIR filters, 347–348
Datalog File Handler VI, 327–328 Delete From Array, 117 median filters, 348
Datalog files, 170, 181–183 Delimiter input, 111, 174 moving averages, 348–350
postrun analysis in, 326–327 Demodulation, 287 timing for, 350
reading, 182–183 Dequeued, 100, 210 Digital Image Processing (Gonzalez
utilities for, 183 Derivative (rate), 445 and Wint), 573–574
writing, 181–182 Derivative limiting, 497 Digital inputs, 297–298
Datalog refnum, 181 Derivative responses, 498 Digital multimeter, 250
Datalog-format file, 170 (See also Description pop-up menu, 232, 273 Digital storage oscilloscopes (DSOs):
Datalog files) Descriptions, user-defined, 265 flash converters used in, 311
Datalogging and Supervisory Design, 196 history of, 272
Control (DSC) module, 438, analog, 314 for pulse and transient
458, 479 bottom-up structured, 197–198 phenomena, 536–539
DataSocket Read function, 483 in CLAD, 220 uses of, 536
DataSocket Transport Protocol of high-voltage systems, 515–516 Digital timing, 149
(dstp), 483 iterative, 196 Digital Vision Corporation, 577
DataSocket Write function, 483 of man-machine interfaces, 44 Digital-to-analog converters
DataSockets, 303, 483 of SubVI, 220 (DACs), 314–315
Date time rec, 146 top-down structured, 197 characteristics of, 309
Date/Time, Seconds To, 138, 146 (See also Application design) coding schemes for, 315–316
Date/Time In Seconds, 138 Design patterns, 87–102, 200–201 for pulse and waveform
dB (decibels), 291 in CLAD, 220 generation, 403
DBL (double-precision floating client-server, 90–94 Digital-to-analog (D/A) system, 317
point) numeric type, 115 event-driven applications, Digitizers, transient (see Transient
DC Estimator, 544 100–102 digitizers)
DC plasma diagnostics, 520 independent parallel loops, 89–90 Dilation, 582
DCM (Digital Clock Manager), 407 initialize and then loop, 87–89 Dimension size, 117, 119
DCRMS VI, 125 with producer/consumer Direct memory access (DMA), 5,
DCS (distributed control system), events, 209 320, 373–374
449–451 Queued Message Handler, 98–99 Discrete-time functions, 305
Deadband, 279, 503–504 site machines, 94–98 Discrete-value functions, 305
Debug database, 421 Design standards, 442–443 Discriminator, 317, 537
Debugging, 54–58 Desired controller, 467 Dispatcher, 481
during application design, Desktop PCs: Dispatcher loop, 470
204–207, 214–218 flat hierarchy in LabVIEW use Display hierarchy (MMIs), 469–473
breakpoint setting for, 57–58 of, 406 Display VIs (as clients), 479–482
in CLAD, 220 local variables in LabVIEW use Displays:
data checking for, 55, 56 of, 406 in IMAQ Vision, 583
with execution highlighting, 57 parallelism in LabVIEW use performance of, 342–343
generating info for, 427 of, 406 status, 360–361
of globals, 78–80 for real time, 379, 381 Dispose (IMAQ), 578, 579
instrumented, 421 subVIs in LabVIEW use of, 406 Distributed control system (DCS),
in LabVIEW FPGA, 408 Destination Code, 509 449–451
of local variables, 80 Destroy Notifier VI, 159 Distributed I/O, 302
on-chip, 421 Destroy Rendezvous VI, 165 Distribution:
with performance checking, Destroy Semaphore VI, 164 of documentation, 241–242
216–217 Detailed diagram help, 239 with HTML, 242
of property nodes, 80 Determinism, 378 with PDF, 242
with single-stepping, 55–57 Development (LabVIEW), 15–18 using custom online help, 242
with subVIs, 54–55 Diagrams: Dithering, 318
with Suspend When Called block, 12–13, 48, 189–191 DLLs (see Dynamic Link Libraries)
breakpoints, 58 dataflow, 13 DMA (see Direct memory access)
with trace commands, documentation of, 234 Documentation, 231–242
214–216 help with, 239 in CLAD, 220
of VIs, 54 instrument loop, 442 CLD graded on, 222
Decibels (dB), 291 for piping instruments, 439–442 connector pane picture in, 239
Decimal String To Number, 108 state, 13, 97 of control descriptions, 232–233
Decoupling, 42 Dialog boxes, 261 custom online help as,
Default Directory constant, 169 Differential connections, 288–291 233–234, 242
Default string, 180 Differential linearity, 312 of diagrams, 234
596 Index

Documentation (cont.) Empty Array (function), 118, 119 ETS (Equivalent time
distribution of, 241–242 Empty arrays, 69, 118 sampling), 543
formal, 238–241 Empty Path, 169 Event driven programming, 434
of instrument drivers, 273–274 Empty String/Path, 269 Event handling, functional globals
outline of, 238–239 Emulation mode, 408 and, 212
printing, 235–238 Enable chain, 410 Event output, 72
programming examples in, 241 Enable Database Access, 182 Event structure, 81
screen images, 236–237 Encapsulated PostScript (EPS) Event-driven applications,
of terminal descriptions, 240–241 files, 236 100–102
VI descriptions and, 231–232, Encoders: Events, 81–87
239, 240 absolute, 523 controls for, 85
of VI history, 234–235 incremental, 523 defining, 81, 151, 153
writing, 238–241 PWM, 404 dynamic, 85–87
DOS systems, 12 quadrature, 523 dynamic user-defined, 82
Double-buffered DMA, 320 reading of, 523 filter, 83–85
Double-precision floating point Encoding images, 573 Notify, 153
(DBL) numeric type, 115 End or identify (EOI) signal, 254 notify, 81–83
Dr. T (see Truchard, Jim) End-of-line (EOL), 206 Sub Panel Pane, 81–82
Draw 8-bit Pixmap, 583 Enhancement, image, 573 synchronization of, 153–155
Drawing standards, 442–443 Enqueued, 91, 100, 210 user-defined, 212
Drivers, 191 (See also Instrument Ensemble averager, 540 Value Change, 86–87
drivers) Enterprise Connectivity Toolset, VI, 81
DSC (see Datalogging and 505–506 Excel (Microsoft), epoch
Supervisory Control module) Enumerated constraints time in, 147
DSOs (see Digital storage (enums), 96 Excitation, 293
oscilloscopes) Enumerated types (enums), Exclusion, trigger, 539
DSP board, 320 133–134 Executable code, 420, 423
dstp (DataSocket Transport Enums, 133–134 Execution:
Protocol), 483 EOF function, 183 of timing, 143–145
Dynamic events, 85–87, 154, 212 EOI (end or identify) signal, 254 of VIs, 54
Dynamic Link Libraries (DLLs), EOL (end-of-line), 206 Execution highlighting, 57
29, 59, 149 Epoch time, 145–147 Execution overhead, 62
Dynamic links, 45 EPS (encapsulated PostScript) EXIT command, 67
Dynamic memory allocation, 431 files, 236 Exp String To Number
Dynamic user-defined events, 82 Equivalent time sampling function, 109
(ETS), 543 Expansion, 453
Equivalent time-sampling, 306 Exponent, 128
E Erosion, 582 Exporting timing data, 146–147
Error(s): Extended precision (EXT)
Ectron, 292 in actuators, 279 float, 549
Edit Format String, 110 in ADCs, 311 External IOBs (input/output
Editors, 352–362 automatic recovery of, 261 blocks), 407
interactive, 355–357 gain, 311 External timing, 498, 499
for menu-driven systems, in grounding/shielding External triggers, 316–317
360–361 connections, 288 Extract Single Tone Information
pop-up, 357–360 offset, 311 VI, 148
static, 353–355 Error Code File Editor, 209
status displays for, 360–361 Error Event VI, 212
EG&G Automotive Research, 301
Elapsed Time, 222
Error handler, 208 F
Error handling:
Electromagnetic fields, 286–288 application design for, 207–213 Faceplate displays, 465
Electronics WorkBench Multisim, 32 in CLAD, 220 False String function, 107
Electrostatic shield, 286 Error in, 262, 263 Familiarization, 249–250
Elemental I/O, 429 Error I/O, 64, 257 Fan-out, trigger, 537
Embedded Development Module flow control drivers, 261–265 Faraday cage, 286
(LabVIEW), 419 in VISA functions, 272 Fast Fourier Transform (FFT), 281
Embedded image (BMP) files, 236 Error out, 262, 263 FASTBUS, 519
Embedded LabVIEW (see Error Query, 265 Feature creep, 25
LabVIEW Embedded) Error Queue, 210 Feedback, 444
Embedded Project Manager, E-Series DAQ boards, 523 Feedback control, 453
424–425 Ethernet adapters, 304 Feedforward, 495
Embedded Target Creation Ethernet Device Server, Femtosecond laser cutting
Utility, 426 246–247 system, 34
EMF objects, 236 Ethernet devices, 245, 457 Fencepost, 503
 Index 597

FFT (Fast Fourier Transform), 281

Field bus, 448
Flatten to String, 132–133
Floating analog inputs, 294–295
G
Field diagnostics, 520–527 Floating point operations, 432 G language, 26
ion beam intensity mapper Floating windows, 24 G Math Toolkit, 550
application, 524–527 Floating-point (real) value, 104 G Programming Reference Manual
motion control systems for, Flow control, 61–102 (LabVIEW 5), 390
520–523 common threads for, 63–64 G Wizards, 459
reading encoders and, 523 data dependency and, 62–63 Gain errors, 311
Field mapping, 520 design patterns for (see Design Gain scheduling variable, 496
Field-Point, 303 patterns) Gape Applied Sciences, 533
Field-programmable gate arrays error I/O, 261–265 Gate input, 523
(FPGA), 30, 406 (See also events and, 81–87 Gated integrator, 518
LabVIEW FPGA) controls for, 85 Gates, Bill, 454
FIFO (First In, First Out), 399 dynamic, 85–87 General Error Handler VI, 209, 264
File Dialog function, 169, 173 filter, 83–85 General Purpose Interface Bus
File mark, 179 notify, 81–83 (GPIB), 9, 245, 247, 248
File refnum, 169 globals and, 71–80 hardware/wiring issues of,
File transfer protocols (FTP), 483 debugging of, 78–80 253–254
File/Directory Info, 169 global variables, 71–75 history of, 9
Files, 167–183 local variables, 75–78, 80 protocols/basic message passing
accessing, 167–169 with looping, 64–71 of, 254
ASCII text, 167 For Loops, 66, 67 voltage and, 278
binary, 169–170, 176–181 pretest While Loops, 65 Generate Occurrence, 155
reading, 179–181 shift registers for, 67–71 GenerateCFunctionCalls, 428
writing, 178–179 stops for, 65, 66 GenerateDestinationFolder, 427
datalog, 170, 181–183 While Loops, 65–66 GenerateGuardCode, 427
help, 232 with sequences, 61–62 GenerateIntegerOnly, 427
in IMAQ Vision, 579 Flowcharts, 13 GenerateSerialOnly, 427
as mailboxes, 484 Fluke-Phillips, 514 Generator, arbitrary waveform, 532
object, 423 Flush File, 172 Get Date/Time In Seconds, 146
for sound I/O, 588–589 For Loops, 64, 66, 67 Get Date/Time String, 138, 146
text, 169–176 Force update, 478 Get Waveform Attributes, 127
formatting to, 175–176 Formal documentation, Get Waveform Components
reading, 172–174 238–241 function, 125
writing, 170–172 Format Date/Time String, 146 Global Alarm Queue VI, 508–509
types of, 169–170 Format Into File, 175 Global Positioning System (GPS),
Fill button, 473 Format Into String, 107, 110, 134 148, 149
Fill frame, 488 Format menu, 132 Global queue, 210, 508
Filter events, 83–85, 153 Format String input, 109, 175 Global References, 75
Filtering: Format Value, 106 Global variables, 71–75
in signal sampling, 307–309 Formatting, 175–176 built-in, 73–75
spatial, 582 Foundation Fieldbus, 303, 448 local variables vs., 431
Filters, 291 Four P’s, 437 VI for, 71
active, 292 FPGA (see Field-programmable gate Globals, 71–80
analog, 308 arrays; LabVIEW FPGA) debugging of, 78–80
for antialiasing, 307 FPP (Front Panel Protocol), 380 functional, 65, 71, 212
digital, 308, 344–350 Fract/Exp String To Number global variables, 71–75
FIR filters, 68, 345–347 function, 109 local variables, 75–78, 80
IIR filters, 347–348 Frame grabbers, 575–576 GOTO command, 67
median filters, 348 Frequency domain, 281 GPIB (see General Purpose
moving averages, 348–350 Frequency foldback, 306 Interface Bus)
timing for, 350 Frequency-selective voltmeter, 544 GPIB-ENET, 304
low-pass, 307 Front Panel Protocol (FPP), 380 GPS (see Global Positioning
reconstruction, 315 Front-panel, 44–45, 474–475 System)
Find command, 75, 80 FTP (file transfer protocols), 483 GPS interface boards, 149
Finite impulse response (FIR) Functional globals, 65, 71, 212 Grab Acquire (IMAQ), 578, 579
filters, 68, 345–347 Functional requirements and GRAFCET, 456
First Call, 222 specifications, 187 GrafcetVIEW, 490–492
First In, First Out (FIFO), 399 Functionality (CLD), 222 Graftek, 574
First row output, 173 Functions (CLAD), 219 Graphical user interface (GUI),
First-level triggers, 537 Future predictions, for 12, 219
Flags, 152 LabVIEW, 32 Graphs, 464–465, 556–572
Flash converters, 311 Fuzzy logic, 446, 496 with bivariate data, 563–565
Flat hierarchy, 406 intensity charts as, 571–572
598 Index

Graphs (cont.) High-accuracy timing, 147–149 IMaging ACquisition, 574 (See also
with multivariate data, 565–569 High-frequency measurements, IMAQ Vision)
3D, 570–571 517–518 IMAQ tools (see specific types,
with waveform and Cartesian Highlighting, execution, 57 e.g.: Morphology)
data, 558–563 High-resolution timing, 147–149 IMAQ Vision, 577–585
Ground, 284–286 High-side, of power source, 517 display in, 583
Ground loops, 286–287 High-voltage systems, 514–516 example, 579–582
Ground return (low side), 517 HighwayView, 244, 460–461 files in, 579
Grounding, 284–291 HIST Data to Text File, 505 image handles of, 578–579
of differential/single-ended HIST package, 503–505 intensity graphs and, 583–585
connections, 288–291 Histograms, 506 interactive windows in, 579
electromagnetic fields and, Historical trends, 499, 502–503 Importing timing data, 146–147
286–288 Hit (reset), 458 Impulse response, 309
error sources in, 288 Honeywell-Measurex In Range and Coerce Function,
ground, 284–286 Corporation, 35 205, 269, 434
Ground-referenced analog HP Interface Bus (HP-IB), 247 Incremental encoders, 523
inputs, 294 HP-IB (HP Interface Bus), 247 Independent parallel loops, 89–90
Gruggett, Lynda, 156 HTML (see Hypertext Markup Index Array, 115, 486–487
GUI (see Graphical user interface) Language) Indicator events, 82
HyperDrives, 16 Inductance, 286
Hypertext Markup Language Inductive coupling, 286
H (HTML), 232 Industrial standards (process
for document distribution, 242 control applications),
Hall Effect devices, 516 printing to, 236 438–443
Handle blocks, 132 Hypervelocity Free-Flight for drawing and design, 442–443
Hard real time, 378 Aerodynamic Facility for piping and instrument
Hardware: (HFFAF), 546 diagrams/symbols, 439–442
GPIB issues with, 253–254 Hysteresis, 279 Infinite loops, 65, 105
instrument driver issues with, Hysteresis On/Off Controller, 493 Info string, 80
252, 254 Informative alarms (see Alarms)
I/O, 191–192 Init (IMAQ), 577, 578
for LabVIEW Embedded, 422
for LabVIEW RT, 379–382
I Initialization, 179
of arrays, 117–119
for physics applications, 514–520 I16 (2-byte integer) numeric of boolean control, 478
CAMAC, 518 types, 115 sequential control problems
FASTBUS, 519 Iconic programming, 33 with, 489–490
NIM module, 519–520 Icons, 49–50 Initialization VIs, 267–268
for signal conditioning, IEEE (Institute of Electrical and Initialize and then loop, 87–89
514–518 Electronics Engineers), 247 Initialize Array function, 117, 119
VME bus, 519 IEEE 488, 247, 253 Inline C Node, 430
VXI, 519 IEEE 488.2, 248 Inplaceness, 20
remote I/O, 302 IEEE-488 bus, 9 Input:
serial instruments issues with, IFFT (Inverse Fast Fourier characteristics of, 534
252–253 Transform), 281 count, 173
for signal conditioning, 514–518 Ignore previous input, 157 data, 179
for current measurements, IIR filters, 347–348 delimiter, 111, 174
516–517 Illustrator (Adobe), 237 ignoring, 157
for high-frequency Image encoding, 573 for length, 117
measurements, 517–518 Image handles, 578–579 previous, 157
for high-voltage Image Source, 581 sound, 587–588
measurements, 514–516 Images, 573 type, 179, 181
Hardware platforms: ImageToArray (IMAQ), 583 Input offset current, 296
for LabVIEW FPGA, 379, 380 Imaging, 555, 572–586 Input scanners, 476–477
RIO, 379, 380 computer configuration for, 575 Input signals, 294–298
HART, 448 with IMAQ Vision, 577–585 AC signals, 297
Help: display in, 583 digital inputs, 297–298
compiled files for, 232 example, 579–582 floating analog inputs, 294–295
custom online, 233–234, 242 files in, 579 ground-referenced analog
Help Tags, 232 image handles of, 578–579 inputs, 294
Hex Display, 110 intensity graphs and, 583–585 thermocouples, 295–296
HFFAF (Hypervelocity Free-Flight interactive windows in, 579 Input/output blocks (IOBs),
Aerodynamic Facility), 546 system requirements for, external, 407
Hide Control, 236 574–577 input/output (I/O) signals:
Hide Front Panel Control, 118 video I/O devices for, 575–577 circular buffered, 374–375
 Index 599

continuous control values for,

496–497
Integral nonlinearity, 312
Integral responses, 498
L
drivers for, 429–431 Integration, 444 LabDriver VIs, 20
Error, 64 Intensity charts, 571–572 Laboratory Virtual Instrument
error flow control, 261–265 Intensity graphs, 564, 583–585 Engineering Workbench
handler, 476 Interactive editors, 355–357 (see under LabVIEW)
hardware specifications for, Interactive windows (IMAQ LabVIEW (Laboratory Virtual
191–192 Vision), 579 Instrument Engineering
imaging devices, 575–577 Interchannel delay, 313 Workbench) (general):
in LabVIEW Embedded, 429–431 Interface boards, 149 benchmark products created
subsystems for, 299–303, 463 Interface names, 578 with, 34–36
Insert Into Array, 117 Interfacing, 457 challenges during development
Institute of Electrical and Internal timing, 498–499 of, 16–18
Electronics Engineers Internal triggering, 317 desktop vs. FPGA use of,
(IEEE), 247 Internet Protocol (IP), 482 406–414
Instrument data sheets, 300 Interrupt Service Routines (ISRs), development of, 15–18
Instrument Driver Advisory Board 434–435 future of, 32
(LabVIEW), 259 Interrupt-driven programming on Macintosh computers, 12, 15,
Instrument Driver Finder, 244 (LabVIEW Embedded), 19–21
Instrument Driver Guidelines 434–435 multitasking in, 41–42
(NI), 245 Interval Timer VI, 72, 481 multithreading in, 42–43
Instrument Driver Project Wizard, Intervals, timing, 139–140 origin of, 8–9
243, 259 Invalid enums, 134 for PC, 15
Instrument drivers, 243–274 Inverse Fast Fourier Transform portability of, 23–24
close VIs and, 272 (IFFT), 281 programming technique of,
communication standards for, Inversion, priority, 388 12–15, 39–40
245–249 Invoke nodes, 471 release of, 19
communication with, 251–256 “Invoking a method,” 471 screen images of, 236–237
configuration VIs and, 268–269 I/O (see Input/output) software products influenced
Data VIs and, 270–271 IOBs (input/output blocks), 407 by, 32–34
defined, 243 Ion beam intensity mapper use of, 7–8, 37
development of, 257–259 application, 524–527 virtual instrument model and,
documentation of, 273–274 IP (Internet Protocol), 482 11–12
for error I/O flow control, 261–265 IRIG interface boards, 149 vision of, 9–10
familiarization with, 249–250 ISA (Instrumentation, Systems, (See also specific applications,
functions of, 250–251 and Automation e.g.: LabVIEW Embedded)
GPIB instruments, 247–249 Society), 438 LabVIEW 1.2, 20–21
hardware/wiring issues of, Isolation amplifiers, 291, 515 LabVIEW 2, 22–24
252, 254 Isothermal, 296 LabVIEW 3, 24, 25, 479
HighwayView and, 244 ISRs (Interrupt Service Routines), LabVIEW 4, 25–26
initialization VIs and, 267–268 434–435 LabVIEW 5, 26–28, 390
modularity achieved with, Iteration control, 540 LabVIEW 6, 29, 30
265–266 Iteration input, 72 LabVIEW 7, 29, 30
obtaining, 243–245 Iterative design, 196 LabVIEW 8, 31
organizing with, 266–267 Instrument Driver Project
plug-and-play, 259–260 Wizard in, 243
protocols/basic message passing J Sequence structures in, 62
of, 254–256 time functions in, 145–146
serial instruments, 245–247 Jellum, Roger, 34 LabVIEW Application Builder,
utility VIs and, 271–272 Jitter, 537 25, 26
Instrument I/O Assistant, 251–252 Join Number, 132, 585 LabVIEW C code generator,
Instrument loop diagrams, 442 Joint time-frequency (JTF) 419, 420
Instrument setups, 271, 272 domain, 281 LabVIEW C Runtime Library,
“An Instrument That Isn’t Really” JTAG emulator/debugger, 421 422–423
(Michael Santori), 8 JTF (joint time-frequency) LabVIEW Certifications (see
Instrumentation: domain, 281 specific types, e.g.: Certified
amplifier for, 289 LabVIEW Developer)
real, 5–6
virtual, 4–6 K LabVIEW DAQ Library, 297
LabVIEW Data Storage (Application
Instrumentation, Systems, Karlsen, Corrie, 473 Note AN154), 132
and Automation Society Keywords, 232 LabVIEW Development
(ISA), 438 Kinetic Systems, 524 Guidelines, 221
Instrumented debugging, 421 Kodosky, Jeff, 9–17, 19, 20, 22–24, LabVIEW Embedded, 417–435
Integers, 104, 115 28, 32 best practices for, 431–434
600 Index

LabVIEW Embedded (cont.) LabVIEW Signal Processing infinite, 65, 105

C code and, 422–424 Toolkit, 281, 282 initialize and then loop, 87–89
Embedded Project Manager LabVIEW sketching, 202–203 For Loops, 64, 66–67
for, 424–425 LabVIEW Technical Notes, 119 pretest While Loops, 65
hardware requirements/ (See also specific types, e.g.: shift registers for, 67–71
recommendations for, 422 Managing Large Data Sets stops for, 65, 66
history of, 417–419 in LabVIEW) Timed Loop with Frames, 140
interrupt-driven programming LabWindows CVI, 33 Timed Loops, 138, 140, 395
in, 434–435 Ladder logic, 456, 486 While Loops, 64–66
I/O drivers incorporated into, Lag (delay), 444, 492 Loops, 13
429–431 Langmuir probe, 527 Loops per second, 498
LabVIEW Embedded Latch modes, 77 Low side (ground return), 517
Development Module, 419 Latching, 508 Low-pass filter, 307
LabVIEW runtime library and, Latency, 378, 379 Low-speed DAQ, 370–372
422–423 Lawrence Livermore National LT1088 RMS to DC converter, 297
plug-in VIs in, 425–429 Laboratory, 34
steps in, 419–421
targets and, 421, 435
LCD panels, 194
Lead, 492 M
LabVIEW Embedded Development Lead/Lag VI, 492 MacCrisken, Jack, 9, 16, 20, 23, 214
Module, 419 Leak resistors, 295 Machine-dependent (manager)
LabVIEW FPGA LEDs (light-emitting diodes), 298 layer, 24
(field-programmable gate Length input, 117 Macintosh computers, 15–17
array), 377–391 Libraries: LabVIEW development on, 12, 15
application development for, C Runtime Library, 422–423 sound files on, 589
406–414 Call Libraries, 59, 434 timers and, 137
clocked execution and, DAQ Library, 297 Macintosh II, LabVIEW
411–413 for digital timing, 149 development on, 19–21
CompactRio and, 404–405 DLLs, 29, 59, 149 Macintosh Plus, 17
CompactVision system LabVIEW runtime library, MacPaint, 12, 14
and, 405 422–423 Magnetic shielding, 286
compiling of, 406–408 RDTSC library, 383 Mailboxes, files as, 484
dubugging in, 408 shared, 59 Make Current Value Default,
parallelism in, 413, 414 VTL, 257 118, 206
pipelining in, 413–415 Light-emitting diodes (LEDs), 298 Manager (machine-dependent)
plug-in cards of, 403, 404 Limiting, derivative, 497 layer, 24
RIO hardware platforms for, Line feed, 111 Managing Large Data Sets in
379, 380 Linear Technology, 297 LabVIEW (LabVIEW
synchronization and, 405, Linearity, 312 Technical Notes), 119
408–411 Link list, 178 Manipulated variable, 444
LabVIEW GUI (John Ritter), 44 Linked code, 420 Man-machine interfaces (MMI),
LabVIEW Instrument Driver Linking, 423 463–475
Advisory Board, 259 Links, 45, 178 design of, 44
LabVIEW MathScript, 31 List, of links, 178 display hierarchy in, 469–473
LabVIEW PDA Module, 419 List Folder, 169 front-panel item handling in,
LabVIEW Performance and Local variables, 24, 75–78, 80 474–475
Memory Management debugging of, 80 standards for, 438
(see Application Note 168 in desktop and FPGA LabVIEW techniques for, 473–474
(LabVIEW Performance and use, 406 Mantissa, 128
Memory Management)) finding, 80 Manual control, 444
LabVIEW RT, 28–29, 377–399 global variables vs., 431 Manual save and recall, 367
communications and, 398–399 sequence, 61 Maps, 565
defining, 377–379 Lock-in amplifiers, 545 Match First String, 108
hardware for, 379–382 Locking, 83, 84, 250 Match Pattern, 100–101, 107, 108
scheduling and, 395–398 Log always, 215 MathScript (LabVIEW), 31
software design for, 382–394 Log At Completion, 182 Matrix (2D array), 112
context switching in, 393–394 Log command, 182 Median filters, 348
for measuring performance, Log on change, 215 Medium-speed DAQ, 373–375
383–388 Logic functions, boolean, 486–487 Mega-applications, 22
multithreading/multitasking Loop tunnels, 431 Memory:
in, 389–391 Looping, 64–71 allocation of, 431
shared resources for, 388–389 aborting, 67 diagnostics for, 552
VI organization in, 391–393 constants and, 431 optimization of, 547–553
LabVIEW runtime library, 422–423 dispatcher loop, 470 problems with, during LabVIEW
LabVIEW SCADA, 479 independent parallel loops, 89–90 development, 17, 20
 Index 601

segmented, 535 Mutliple time bases, 535 Number of Cycles control, 531
state, 70 Mutually exclusive (mutex), Numbers and numbering:
Memory manager, 121 163, 388 incorrect, 96
Memory Usage, 119–122, 552 MXProLine, 35 of rows, 173
Menu-driven systems, editors for, MyGraphData, 483 state, 94
360–361 Numeric types, 104–105
Message passing: clusters and, 115
of GPIB instruments, 254 N DBL, 115
of instrument drivers, 254–256 I16, 115
of serial instruments, 254–256 Namespace conflicts, 161 SGL, 115
Metainformation, 176 Naming: Nyquist rate/criterion,
Metal-oxide varistors (MOVs), 516 of cursors, 567 305–307
Metric Systems, 570 of tags, 439–441, 467
Microcontrollers, 455 unique, 71
Microsoft Excel, epoch time in, 147 NaN (see Not-a-number)
Nanosecond waveforms, 542
O
Microsoft Windows 3.0, 15, 23
Microstepping, 521 NASA: Object files, 423
MightySat II.1 (Air Force Ames Research Center’s Object oriented programming
Research Lab), 34 HFFAF, 546 (OOP), 22
Milliseconds (ms) timing, New Millennium Deep Space 1, 34 Occurrences, 155–158
138, 139 National Electrical Code, 285, 443 OCDIComments, 428
MMI (see Man-machine interfaces) Naval Research Lab Coriolis OCX (oven-controlled crystal
MMI G Wizard, 469 spacecraft, 34 oscillator), 149
Modbus protocol, 457 Needs analysis, user, 186–187 Offset, 180, 183
Mode control, 216, 486 Negative feedback, 444 Offset errors, 311
Model-based controllers, 496 Network connections, 482–484 Ohm’s Law, 516
Modem, null, 247, 253 Network-Published Shared OLE for Process Control
Modicon, 457 Variable, 398 (OPC), 456
Modularity, 198–200, 265–266 New Millennium Deep Space 1 Omega Engineering Temperature
Moore Products, 292 (NASA), 34 Handbook, 296
Morphological transformations, 582 Newport Corporation, 523 On-chip debugging, 421
Morpohology (IMAQ), 582 NI Spy, 256 One-dimensional (1D) string array,
Motion control systems, 520–523 NI-FBUS host software, 448 69, 111
Motion Toolbox, 214 NI-IMAQ, 577 Online help, custom (see Custom
Motorola 68000 microprocessors, 15 NIM (Nuclear Instrumentation online help)
Move, 169 Manufacturers) module, On/off controllers, 444
Moving averager, 308 519–520 OOP (object oriented
Moving averages, 348–350 NI-PSP (Publish-Subscribe programming), 22
MOVs (metal-oxide varistors), 516 Protocol), 398 OPC (OLE for Process Control), 456
ms time out, 156 NI-SCOPE driver, 533 OPC-based PLC (example),
Ms (milliseconds) timing, 138, 139 Nodes, 38, 202 458–460
Multidrop systems, 246 call by reference, 471 Open standards, 450
Multimedia, 555 CINs, 58–59 Open VI Reference node, 471
Multimeter, digital, 250 property, 45–47, 80, 194, 471, 474 Open/Create/Replace File, 169,
Multiple state machines, 488 Noise: 170, 182
Multiple-loop controllers, 462 dithering of, 318 Opening, 582
Multiple-record mode, 535 elimination of, 284 Operating systems,
Multiplexing: in signal sampling, 317–319 multithreaded, 164
ADCs vs., 313 Nonlinearity, integral, 312 Operation mode, 170
in signal conditioning, 293 Normal-mode signal, 289, 485 Operators, notification alarms
timing skew and, 312–313 Not A Number/Path/Refnum, for, 511–512
Multipulse data, 539–543 105, 207 Opto-22, 302
Multitasking, 41–42, 389 Not-a-number (NaN), 105, 207 Output (out) control, 467
cooperative, 41–42 Notification VI, 159 Output data:
preemptive, 145 Notifiers, for synchronization, analog, 299
in RT software design, 389–391 158–160 data distribution of, 477–479
Multithreaded operating Notify events, 83, 153 manipulation of, 444–447
systems, 164 NuBus architecture, 20 in process control applications,
Multithreading, 27, 42–43, 145, Nuclear Instrumentation 444–447
389–391 Manufacturers (NIM) for sound, 588
MultiThreshold (IMAQ), 581 module, 519–520 Oven-controlled crystal oscillator
Multivariate data, on graphs, Null modem, 247, 253 (OCX), 149
565–569 Null (zero) terminators, 134 Overflow, 104, 310
Mutex (Mutually exclusive), 163 NuLogic, 522 Oversampling, 308, 319, 344
602 Index

P motion control systems for,

520–523
Polymorphism, 103, 179, 559
Pop-up editors, 357–360
P (proportional) controllers, 444 reading encoders and, 523 Pop-up menus, 232, 273
PACs (programmable automation hardware for, 514–520 Portability, 23–24
controllers), 377 CAMAC, 518 Portable Document Format
Page, 121 FASTBUS, 519 (PDF), 242
Parallel loops, independent, 89–90 NIM module, 519–520 porting), 23
Parallelism: for signal conditioning, 514–518 Ports:
in desktop and FPGA LabVIEW VME bus, 519 addition of, 246–247
use, 406 VXI, 519 for serial instruments, 246–247
in LabVIEW FPGA, 413, 414 for plasma diagnostics, 527–532 Pos offset, 183
Parametric Graph, 3D, 570 pulse and transient phenomena Position Meas Quad
Parametric plots, 560 in, 533–546 Encoder VI, 523
Parasitic detection, 287 DSOs for, 536–539 Positive feedback, 444
Pareto analysis, 506 multipulse data, 539–543 Postrun analysis for, 326–336
Parker, Jeff, 20, 570, 571 signal recovery for, 543–546 in ASCII text format, 328–329
Parker Compumotor, 214 transient digitizers for, configuration information for,
Parsing, 107–109 533–536 333–335
Path control, 168 PI (proportional-integral) in custom binary formats,
Path To String, 168 controllers, 445 329–331
Pattern matching (IMAQ), 580 PIC VIs, 497–498 database for, 335–336
Pause button, 55 Pick Line, 106 in datalog file format, 326–327
PCI 7041/6040E, 380, 381 PICT objects, 236 Datalog File Handler VI for,
PCI bus interface, 519 Pict Ring, 193, 474 327–328
PCs (see Personal computers) Picture Controls, 194, 567, 583 direct links in, 331–332
PDA Module, 419 Picture in--Picture out, 569 timestamps for, 332–333
PDA Module (LabVIEW), 419 Pictures, 567 PostScript printing, 237
PDAs (Personal Digital Assistants), PID (algorithm), 492 Posttrigger data, 534
417–419 PID (Gain Schedule), 492, 495 Potential (volts) controls, 78
Peak voltage, 240 P&ID (piping instrument diagrams Power (watts) controls, 78
Peer-to-peer communications, 453 and symbols), 439–442 Power spectrum, 306
Performance: PID Control Toolkit, 446, 485, P-picker, 546
continuous control limitations 492, 497 Precision, 104
on, 498–499 PID (proportional-integral- Precision menu, 132
RT software design derivative) controllers, 444 Predictive controllers, 446, 496
measurements of, 383–388 PID tuning schedule, 495–496 Preemptive multitasking, 145
strings affecting, 476 PID with Autotuning, 492 Preemptive scheduling, 391
Performance and Disk items, 552 Pipelining (LabVIEW FPGA), Prepare VI, 428
Performance and Memory 413–415 Preston (company), 292
Management (Application Piping instrument diagrams and Pretest While Loops, 65
Note 168), 122 symbols (P&ID), 439–442 Pretrigger data, 534
Performance checking, 216–217 Pixels, 573 Primary elements, 447
PERL language, 28 Plagiarizing, 59–60 Print command, 235, 236
Personal computers (PCs), 15 Plasma diagnostics, 527–532 Print Options, 236
control system architectures PLCs (see Programmable logic Print Window, 236
with, 451–452 controllers) Printing:
desktop, 379, 381 Plot(s): of documentation, 235–238
Personal Digital Assistants (PDAs), Contour, 564 functions for, 269
417–419 defined, 560, 562 to HTML, 236
PGIA (programmable-gain parametric, 560 with PostScript, 237
instrumentation amplifier), surface, 563, 570 Priority, 91, 391
314 waterfall, 563, 565 Priority boosting, 389
PharLap TNT Embedded, 28 wire-frame, 570 Priority inversion, 388
Phase response, 309 Plug-and-play instrument drivers, Probe characteristic, 527
Phase-sensitive detectors 259–260, 266 Probe tools, 55
(PSDs), 545 Plug-in boards, 253–254 Problem definition, 186–191
Physics applications, 513–553 Plug-in cards, 403, 404 Process control applications,
data quantity in, 546–553 Plug-in VIs, 425–429 437–512
memory optimization for, code generation for, 426–429 alarms and, 506–512
547–553 Target_OnSelect, 426 alarm handler for, 508–511
reduction of, 546–547 Points in Graph Window, 501–502 for operator notification,
for field diagnostics, 520–527 Polling, 152–153 511–512
ion beam intensity mapper Polymorphic data, 68 for continuous control, 492–499
application, 524–527 Polymorphic VIs, 50 control strategy for, 493–496
 Index 603

I/O values in, 496–497 Programmable Controllers: Theory Quantization, 309

PIC VIs and, 497–498 and Implementation Queue(s), 91, 160–161, 479
timing/performance (ISA), 456 defined, 160
limitations on, 498–499 Programmable logic controllers for synchronization, 160–161
control system architectures for, (PLCs), 377, 455–461 use of, 161
449–455 advantages of, 456–457 Queued Message Handler, 98–99,
DCS, 449–451 from Allen-Bradley, 244 161, 209
with personal computers, communications techniques of, Quick start, 239, 241
451–452 457–458 QuickDraw, 15
SCADA, 451 HighwayView, 460–461 QuVIEW, 321
selection of, 453–455 OPC-based example of, 458–460
data distribution in, 475–486 sequential control systems
with display VIs, 479–482 based on, 447 R
with input scanners, 476–477 Programmable-gain
of output data, 477–479 instrumentation amplifier Race conditions, 73, 91, 163
real-time process control (PGIA), 314 Radio-frequency interference
databases for, 484–485 Programming: (RFI), 287
using network connections, in CLAD, 219 Ramp Pattern VI, 531
482–484 with dataflow, 38–39 Random accessibility, 177
validation and, 485–486 iconic, 33 Range, 104, 310
industrial standards for, 438–443 of LabVIEW, 12–15, 39–40 Range Finder VI, 205
for drawing and design, by plagiarizing, 59–60 Rapid prototyping, 196
442–443 Programs, robust, 204 Rate, of commands, 270
for piping and instrument Projectors, RGB, 194 RDA (see Remote data acquisition)
diagrams/symbols, 439–442 Property nodes, 45–47, 194, RDTSC, 383
man-machine interfaces and, 471, 474 Read Datalog, 181
463–475 debugging of, 80 Read File, 63, 64
display hierarchy in, 469–473 defined, 45 Read From Measurement
front-panel item handling in, Proportional band, 497 File VI, 176
474–475 Proportional (P) controllers, 444 Read From Spreadsheet File, 173
techniques for, 473–474 Proportional-integral (PI) Read From Text File, 60
output manipulation in, 444–447 controllers, 445 Read function (DataSocket), 483
process signals in, 447–449 Proportional-integral-derivative Read Lines From File, 174
for sequential control, 486–492 (PID) controllers, 444 Read Mark, 171
boolean logic functions in, Protocols: Read Text function, 173
486–487 of GPIB instruments, 254 Read the fine manual (RTFM), 548
GrafcetVIEW and, 490–492 of instrument drivers, 254–256 Read The Manual (RTM), 243
initialization problems in, of serial instruments, 254–256 Reading:
489–490 Prototyping: binary files, 179–181
state machine architecture for, rapid, 196 datalog files, 182–183
487–488 of user interface, 192–196 text files, 172–174
with smart controllers, 452, PSDs (phase-sensitive Ready (thread), 391
455–463 detectors), 545 Real instrumentation, 5–6
I/O subsystems, 463 Pseudocoding, 203–204 Real knobs, 536
PLCs, 455–461 Publishing, 398 Real time (RT), 377–379 (See also
trending and, 499–506 Publish-Subscribe Protocol LabVIEW RT)
HIST package and, 503–505 (NI-PSP), 398 Real time analysis, 337–343
historical trends, 502–503 Pull-up resistors, 297 with continuous data, 339–340
real-time trends, 499–502 Pulse, 108, 403 display performance for, 342–343
SPC and, 505–506 Pulse and transient phenomena, with single-shot data, 338–339
Process Control Systems 533–546 speed of, 340–341
(Greg Shinsky), 446 DSOs for, 536–539 volume of data for, 341–342
Process mimic displays, 464 multipulse data, 539–543 Real (floating-point) value, 104
Process signals, 447–449 signal recovery for, 543–546 Real-time databases, 475, 476,
Process variable, 444 transient digitizers for, 533–536 484–485
Processing, image, 573 Pulse-width modulator (PWM) Real-time operating system
Producer, 160, 210 encoder, 404 (RTOS), 28, 377, 378
Producer events, 209 PXI system, 381–382, 405 Real-time trends, 499–502
Profile Memory Usage, 552 Real-time xy strip chart, 562
Profile VIs, 216 Rebooting, 250, 454
Profiling, 552 Q Recipe operation, 446
Reconfigurable I/O (RIO), 403
Program architecture, 87
Programmable automation Quad int, 146 Reconstruction filters, 315
controllers (PACs), 377 Quadrature encoders, 523 Record (Pascal), 50
604 Index

Recording, 365–370 Robust programs, 204 Screen captures, 237

automatic save and recall, Rogers, Steve, 20, 24 Screen images, 236–237
367–370 ROI (region of interest), 579 ScriptBt VI, 426
manual save and recall, 367 Rosemount, 448 ScriptCompiler, 428–429
Recovery, error, 261 Round-robin scheduling, 391 ScriptLinker, 428, 429
Red File (IMAQ), 579 Route Signal VI, 385 SCSI (Small Computer Systems
Red-green-blue (RGB) Rows, 116, 121 Interface) ports, 19
projectors, 194 first, 173 SCXI equipment, 292, 294, 303
Redundancy, 454 in IMAQ, 583 Second-level triggers, 538
Reentrant, 28, 54, 162 number of, 173 Seconds To Date/Time, 138, 146
Reference junction, 296 output of, 173 Seeback effect, 288
Reference numbers (refnums), 167 values of, 181 SEG (Software Engineering
datalog, 181 RS-232C, 246, 278, 303 Group), 244
file, 169 RS-422A, 246 Segmentation, images, 573
Referenced single-ended (RSE) RS-432A, 246 Segmented memory, 535
input, 295 RS-485A, 246 Select function, 432
Refnums (see Reference numbers) RSE (referenced single-ended) Select Item, 123
Region of interest (ROI), 579 input, 295 Self-inductance, 286
Register For Events, 86 R-series DAQ cards, 403–404 Semaphores, 163
Release Semaphore VI, 164 RT (real time) (see LabVIEW RT) defined, 490
Remote data acquisition (RDA), RTDs (resistance temperature synchronization with, 161–165
303–304 detectors), 293 used in multithreaded operating
Remote I/O hardware, 302 RTF (see Rich text format files) systems, 164
Remote sensors, 302 RTFM (read the fine manual), 548 Sensor systems, 278
Remote terminal units, 451 RTM (Read The Manual), 243 Sensors, 277–278, 302
Rendezvous VIs, 165–166 RTOS (see Real-time operating Sequence local variables, 61
Reorder Controls in Cluster, 122 system) Sequence structures and
Repeat addressing, 254 Run button, 58 sequencing, 14, 61–62
Replace Array Element, 120, Running (thread), 391 avoidance of, 62
479, 502 coding schemes of, 62
Replace Array Subset, 117 in LabVIEW 8, 62
Representation, 129 S Sequential control, 446, 486–492
Required connections, 239 boolean logic functions in,
Resample (IMAQ), 583 Safety: 486–487
Resampling, 583 in high-voltage system design, GrafcetVIEW and, 490–492
Reset Structure at Start 515–516 initialization problems in,
feature, 143 signal conditioning for, 292–293 489–490
Reset windup, 445 Safety ground, 285 state machine architecture for,
Resetting, 234, 444, 458 Sample-and-hold (S/H) 487–488
Resistance temperature detectors amplifier, 310 Sequential function charts (SFCs),
(RTDs), 293 Samples, 280 490
Resistors, pull-up, 297 Sampling theorem, 305–307 Serial communications (see specific
Resolution (bits), 309 Santori, Michael, 8 types, e.g.: RS-232C)
Resolution multiplication, 523 Saturation, 310 Serial devices, 245
Restoration, image, 573 Save and recall: parameters of, 255
Return paths, 285 automatic, 367–370 problems with, 247
Return to Caller button, 58 manual, 367 Serial instruments, 245–247
Reverse Array, 132 Save button, 216 hardware/wiring issues of,
Reverse String, 132 SCADA (see Supervisory Control 252–253
Revision history, 234–235 and Data Acquisition) ports for, 246–247
RFI (radio-frequency Scaler (single-element) types, 103 protocols/basic message passing
interference), 287 Scan From File, 175 of, 254–256
RGB (red-green-blue) Scan From String, 108–110 RS-series of, 246
projectors, 194 Scan interval, 460 troubleshooting of, 247
Rhia, Stephen, 158 Scan Mode, 525 Serial Number, 147
Rich text format (RTF) files, Scanner, 476 Serial Port functions, 255, 511
236, 237 Scheduler queue, 140 Serial Port Init VI, 247
Ring and Enum control Scheduling: Server, 90
palette, 133 LabVIEW RT and, 395–398 Service requests (SRQs), 270
Ring Indicators, 130 preemptive, 391 Servo motors, 520
RIO (reconfigurable I/O), 403 round-robin, 391 Set Cluster Size, 135
RIO hardware platforms SCPI (Standard Commands Set File Position Function, 179
(LabVIEW FPGA), 379, 380 for Programmable Set Occurrence, 155
Ritter, Dave, 44, 589 Instruments), 248, 258 Set Value, 71
 Index 605

Set Waveform Attribute, 127 Signal conditioning, 284, 291 communications techniques of,
Setpoint (SP), 466, 467 hardware for, 514–518 457–458
Settling time, 292, 313–315 for current measurements, HighwayView, 460–461
SFCs (sequential function charts), 516–517 OPC-based example of, 458–460
490 for high-frequency single-loop controllers, 461–463
SGL (single-precision floating measurements, 517–518 Snapshot button, 216
point) numeric types, 115 for high-voltage Snd Read Wave File, 588–589
S/H (sample-and-hold) measurements, 514–516 Snd Read Waveform, 587
amplifier, 310 multiplexing in, 293 Snd Write Wave File, 588–589
Shah, Darshan, 28, 29 for safety, 292–293 Snd Write Waveform, 588
Shannon sampling theorem, Signal Processing Toolkit, 281, 282 Snider, Dan, 214
305–307 Signal recovery, 543–546 SNR (signal-to-noise ration), 291
Shared libraries, 59, 149 Signal Recovery By BP Filter, 544 SO Pause, 588
Shared resources: Signal sampling, 305–321 SO Wait, 588
defined, 388 analog-to-digital converters for, SO Write, 588
for RT software design, 388–389 309–314 Soft real time, 378, 379
synchronization of, 161–165 averaging in, 308–309 Software constructs, in CLAD, 219
Shared Variable Engine (SVE), 398 coding schemes for, 315–316 Software design (RT), 382–394
Sheldon Instruments, 321 digital-to-analog converters for, context switching in, 393–394
Shielding, 286 (See also Grounding) 314–315 for measuring performance,
Shift registers: filtering in, 307–309 383–388
for large arrays, 431 noise in, 317–319 multithreading/multitasking in,
for looping, 67–71 sampling theorem, 305–307 389–391
uninitialized, 69–71 throughput and, 319–321 shared resources for, 388–389
Shinskey, Greg, 446 triggering and, 316–317 VI organization in, 391–393
Short-circuit (crowbar), 516 Signal-to-noise ration (SNR), 291 Software Engineering Group
Show Buffer Allocations, 121 Signed integers, 104 (SEG), 244
Show VI Hierarchy, 201 Significant figures, 105 Software-generated triggers, 317
Shunting, 516–517 Simple diagram help, 239 Sound files, 588–589
SI Clear, 587 Simple Error Handler, 171, 209, Sound I/O, 586–589
SI Config, 587 210, 212, 532 DAQ for, 586–587
SI Read, 587 Simplicity, 44 files for, 588–589
SI Stop, 587 Simulation mode, 485 functions of, 587
Sign bit, 315, 316 Sine Waveform VIs, 125 input, 587–588
Signal(s), 275–304 Single-element (scaler) types, 103 output, 588
actuators, 278–279 Single-ended connections, 288–291 SoundBlaster, 587
ADCs and, 275 Single-loop controllers (SLCs), 453 Source boolean (string), 262–263
analog, 305 Single-precision floating point Source pins, 523
balanced, 51, 290 (SGL) numeric types, Sources (timing), 141–143
categories of, 279–283 115, 549 SP (see Setpoint)
common-mode, 289 Single-shot data: Sp low, 497
connections between, 284–304 continuous data vs., 338 SP-50, 448
amplifier use for, 291–294 real time analysis with, 338–339 Span (range), 497
grounding/shielding of, Single-stepping, 55–57 Spatial filtering, 582
284–291 Sins (see Code Interface Nodes) SPC (statistical process control),
of input signals, 294–298 Site machines, 94–98 505–506
with networking, 303–304 6B Series (see Analog Devices Specifications definition,
of output signals, 298–299 6B Series) 187–189
controller, 447 16550-compatible UART, 246 Spectrogram, 281
EOI, 254 sketching (LabVIEW), 202–203 Spectrum Astro, Inc., 34
input, 294–298 Skew, 537 Speed, of real time analysis,
AC signals, 297 Skin effect, 286 340–341
digital inputs, 297–298 “Skip if busy,” 393 Speed Tester, 191
floating analog inputs, 294–295 SLCs (single-loop controllers), 453 Spike, 313
ground-referenced analog Slew rate, 314–315 Split Number, 132, 585
inputs, 294 Slices, 401, 407 Spreadsheet programs, 12
thermocouples, 295–296 Slider controls, 78, 466 Spreadsheet String To Array
I/O subsystem for, 299–303 Small Computer Systems Interface specifier, 111
normal-mode, 289 (SCSI) ports, 19 Spreadsheets, strings in, 110–114
origins of, 275–283 Smart controllers, 452, 455–463 SQC (see Statistical process control)
sensors, 277–278 I/O subsystems, 463 SRQs (service requests), 270
transducers, 276–277 programmable logic controllers, Standard Commands for
Signal bandwidth, 343–344 455–461 Programmable Instruments
Signal common, 285–286 advantages of, 456–457 (SCPI), 248, 258
606 Index

Start button, 216 SubVIs, 48 reading, 172–174

Start frame, 488 debugging with, 54–55 writing, 170–172
Start-up management, 496 defining, 5, 48, 198 Thermocouples, 295–296
State diagrams, 13, 97 design of, 220 Third-level triggers, 538
State machines, 70, 487–488 in desktop and FPGA LabVIEW 32-bit color, 585
State memory, 70 use, 406 Threadconfig, 390
State number, 94 development of, 19 Threads, 42
State variable analysis, 446 Sun SPARCstation, 24 Thread-safe, 28
Static editors, 353–355 Supervisory Control and Data 3D Curve Graph, 570
Static links, 45 Acquisition (SCADA), 3D graphs, 570–571
Statistical process control (SPC), 451, 479 3D Parametric Graph, 570
505–506 Surface plots, 563, 570 Thresholding, 580
Statistical Process Control SurfaceVIEW, 570 Throughput, 191, 319–321
Tools, 505 Suspend When Called Tick Count (ms), 138
Statistical quality control (see breakpoints, 58 Time base (clock), 316
Statistical process control) SVE (Shared Variable Engine), 398 Time buffered I/O, 373
Status boolean, 262 Swap Bytes (Swap Words), 132 Time critical, 389
Status displays, for editors, Swap Words (Swap Bytes), 132 Time domain, ac, 280
360–361 Switched capacitor filters, 292 Time Has Elapsed, 222
Step Into, 56 Switching: Time out (ms), 156
Step Out Of, 56 context, 377 Time Target(s), 222
Step Over, 56 stop, 77 Time-critical code, 434
Step-and-measure experiments, Sync, 317 Timed Loop with Frames, 140, 395
520–527 Synchronization, 151–166 Timed Loops, 138, 140, 395
ion beam intensity mapper of events, 153–155 Timed Sequence Structure, 138,
application, 524–527 in LabVIEW FPGA, 405, 140, 395
motion control systems for, 408–411 Timed structures, 385, 395–398
520–523 notifiers for, 158–160 Timeout frame, 488
reading encoders and, 523 of occurrences, 155–158 Timers, 137–138
Stepper motors, 520 with polling, 152–153 Time-sampling, equivalent, 306
Stop button, 531 queues for, 160–161 Timestamps, 146, 332–333
Stop switch, 77 with rendezvous VIs, 165–166 Time-to-digital converter
Stops (looping), 65, 66 with semaphores, 161–165 (TDC), 518
Store HIST Data VIA, 504 of shared resources, 161–165 Timing, 137–149 (See also
Strict typedef, 51, 122, 198 (See also Timing) Synchronization)
String(s), 105–114 Synchronous amplifiers, 545 absolute timing functions, 139
Array To Spreadsheet String, 111 continuous control limitations
building, 106–107 on, 498–499
comparison functions for, 269 T digital, 149
data, 132 for digital filtering, 350
as default, 180 T (transparent) color, 557 epoch, 145, 146
info, 80 Tag Configuration Editor, 459 execution/priority of, 143–145
parsing, 107–109 Tag names, 439–441, 467 external, 498, 499
performance affected by, 476 Tags (I/O points), 459 guidelines for, 145–146
in spreadsheets, 110–114 Tanzillo, P. J., 417 high-resolution/high-accuracy,
unprintables, 110, 111 Target Properties, 408 147–149
String Subset, 109 Targets, 421, 435 importing/exporting data for,
String to Byte Array, 583 Tasks, 198 146–147
String To Path, 168 TCP (Transmission Control internal, 498–499
Strip charts, 464–465 Protocol), 482 intervals for, 139–140
Strip Path function, 168–169 TDC (time-to-digital converter), 518 in LabVIEW 8, 145–146
Struct (C), 50 Technical Notes (LabVIEW) (see in milliseconds, 138, 139
Structured dataflow, 13 LabVIEW Technical Notes) sources for, 141–143
Structured design: Tektronix, 514 structures for, 140–141
bottom-up, 197–198 TelevEyes, 577 timers for, 137–138
top-down, 197 Temperature Handbook (Omega Timing skew, 312–313
Structured types, 103 Engineering), 296 Tip pop-up menu, 232, 273
Style (CLD), 222 Terminal descriptions, 240–241 Tip strips, 25, 56–57
Sub Panel Pane events, 81–82 Termination character, 268 To Upper Case function, 108
Subdiagrams, 13 Terminators, null, 134 Toolchain, 421
Subroutine, 551 Test and verify, 217 Top index (channel), 120
Subroutine priority, 391 Text files, 169–176 Top-down structured design, 197
Subscribers, 398 ASCII, 167 Total Range, 525
SubVI Node Setup, 58 formatting to, 175–176 Trace commands, 214–216
 Index 607

Trace VI, 214–217

Training, of users, 217
Type casting, 130, 134–135
Type Definition (typedefs),
V
Transducers, 276–277, 447 52–53, 198 Vacuum brazing, 187
Transfer function, 308 clusters saved as, 51, 122 Valid indicator, 71
Transformations, morphological, 582 Control Editor and, 53 Validation, 485–486
Transformers, current, 516 for data type determination, 182 Value, 216
Transient digitizers, 518, 533–536 defined, 52 Value Change event, 86–87
calibration of, 535–536 strict, 51 Variables, 219 (See also specific
data storage/sampling in, Type descriptor, 128 types, e.g.: Global variables)
534–535 Type input, 179, 181 Variant data fields, 126
input characteristics of, 534 Type setting, 129–132 Variant To Data, 127
for pulse and transient Type specifier, 111, 472 VHDL (VHSIC hardware
phenomena, 533–536 Typedef, strict (see Strict typedef) description language), 402
triggering in, 534 Typedefs (see Type Definition) VHSIC hardware description
Transient phenomena (see Pulse language (VHDL), 402
and transient phenomena) VI descriptions:
Transmission Control Protocol U in documentation, 231
(TCP), 482 documentation of, 231–232,
Transmit lines, 247 UART, 16550-compatible, 246 239, 240
Transmitters, 447 UDP (Universal Datagram VI events, 81
Transparent (T) color, 557 Protocol), 482 VI hierarchy, 201–202
Transpose Array, 181, 584 Unbundle, 123 VI history:
Transpose option, 172 Unbundle By Name, 123, 468, 475 documentation of, 234–235
Transposing, 173 Underflow, 104, 310 of revisions, 234
Trends and Trending, 499–506 Undersampling, 306 VI Properties, 200, 273, 552
displays of, 464–465, 499 Undo, 27 VI Revision History, 234
HIST package and, 503–505 Unflatten From String, 133 VI Server, 471–473
historical trends, 502–503 Uniform Mechanical Code, 443 VI Tree, 267
real-time, 499 Uninitialized shift registers, 69–71 Video I/O devices, 575–577
real-time trends, 499–502 Unipolar inputs, 315 Videotape, 575
statistical process control and, Unique naming, 71 Virtual instrument model, 11–12
505–506 Units, 467 Virtual Instrument Software
Trigger delays, 537 Universal Datagram Protocol Architecture (VISA), 257,
Trigger discriminator, 537 (UDP), 482 272 (See also under VISA)
Trigger exclusion, 539 University of Texas at Virtual instruments (VIs), 4, 40–41
Trigger fan-out, 537 Austin (UT), 10 autonomous, 92–94
Triggering, 309 UNIX systems, 12, 23 block diagrams in, 40
complex, 403 Unprintables, 110, 111 categories of, 266
defined, 316 Unsigned integers, 104 code in, 41
external, 316–317 Up/down pins, 523 compiling, 41
first-level, 537 Upper Limit frame, 488 configuration, 268–269
internal, 317 USB devices, 245 data space in, 40–41
second-level, 538 User interface prototyping, debugging of, 54
signal sampling and, 316–317 192–196 defining, 5
third-level, 538 User needs analysis, 186–187 description documentation of,
in transient digitizers, 534 User training, 217 231–232, 239, 240
TRMS (true RMS) converter User-defined codes, 265 executing of, 54
module, 297 User-defined descriptions, 265 front panel of, 40
Troubleshooting serial User-defined events, 82, 212 hierarchy of, 201–202
instruments, 247 Using LabVIEW to Create organization of, 391–393
Truchard, Jim, 9–11, 17, 23 Multithreaded VIs for parts of, 40–41
True RMS (TRMS) converter Maximum Performance and of plug-and-play instrument
module, 297 Reliability (see Application drivers, 266
TrueTime, 149 Note 114 (Using LabVIEW plug-in, 425–429
Tufte, Edward, 556, 557 to Create Multithreaded VIs real instrumentation vs., 4–6
Turnkey systems, 450 for Maximum Performance revision history of, 234–235
2-byte integer (I16) numeric and Reliability)) in RT software design, 391–393
types, 115 Using Quadrature Encoders with (See also specific types, e.g.:
2D (two-dimensional) array E Series DAQ Boards Autonomous VIs)
(matrix), 112 (AN084), 523 Virtual memory (VM), 547
Two-dimensional (2D) array UT (University of Texas at VIs (see Virtual instruments)
(matrix), 112 Austin), 10 VISA API, 245
Type, wiring to, 180 Utilities, for datalog files, 183 VISA Configure Serial Port
Type Cast, 585 Utility VIs, 271–272 function, 255
608 Index

VISA Open, 267 Wait At Rendezvous VI, 165 Windows (OS), 454
VISA Property node, 267 Wait For RQS, 270 WindToolsSelect (IMAQ), 579
VISA serial port functions, 246 Wait on Notification, 158–159 Wire-frame plots, 570
VISA session identifier, 257, 258 Wait On Occurrence, 155 Wiring, 180
VISA Transition Library (VTL), 257 Wait Until Next ms Multiple, Wiring issues:
Visible Items, 127 138, 139 of GPIB instruments, 253–254
Vision (IMAQ), 574 Warnings (see Alarms) of serial instruments,
Vision Builder (IMAQ), 577 Watchdog program, 458 252–253
Vision package (IMAQ), 577 Watcom C, 24 Write Datalog, 181
The Visual Display of Quantitative Waterfall plots, 563, 565 Write File (IMAQ), 579
Information (Edward Watts (Power) controls, 78 Write function (DataSocket), 483
Tufte), 556 Waveform Chart, 499 Write Mark, 171
VM (virtual memory), 547 Waveform Generation VIs, 125 Write To Measurement
VME bus, 519 Waveform Graph, 558 File VI, 176
VMEbus Extensions for Waveform Measurement Write to Spreadsheet File, 172
Instrumentation (VXI), 6, utilities, 125 Write to Text File, 60, 171
311, 519 Waveform Scan VI, 499 Writing:
Voltage: Waveforms, 125–127, 558 to binary files, 178–179
analog, 278 attributes of, 126 to datalog files, 181–182
GPIB and, 278 generation of, 403, 529 to text files, 170–172
peak, 240 graphs using, 558–563 Wrong numbering, 96
waveform data of, 278 nanosecond, 542
Volts (Potential) controls, 78 recording of, 516
VTL (VISA Transition Library), 257 of voltage, 278 X
VXI (see VMEbus Extensions for Wells, George, 291
Instrumentation) Which Button, 467, 470, 481 XY Graph, 500, 528, 560
VXI instrument, 6 While Loops, 64–66
Wind up, 445
WindClose (IMAQ), 579 Z
W WindDraw (IMAQ), 579 Z index pulse, 523
Wait, abortable, 156 WindGetROI (IMAQ), 579 Zero (null) terminators, 134
Wait (ms), 138, 140 Windoid, 55 Z-scale, 571