Attitude Determination and Control System for CubeSat

A Major Qualifying Project

Submitted to the Faculty of
WORCESTER POLYTECHNIC INSTITUTE
In Partial Fulfilment of the Requirements For The
Degree of Bachelor of Science
By
Assaad Farhat

Jighjigh Ivase

Ye Lu

Alan Snapp

Date: 1st March. 2013

Approved by:

Keywords:
1.) CubeSat
2.) ADCS

Prof. Michael A. Demetriou, Major Advisor

"Certain materials are included under the fair use exemption of the U.S. Copyright Law and
have been prepared according to the fair use guidelines and are restricted from further use."

Abstract
This project continued the work on the development and testing of an Attitude Determination and
Control Subsystem (ADCS) for a three-unit Cube Satellite mission led by WPI, the NASA Goddard
Space Flight Center, and the Space Research Centre in Poland. This project focused on hardware
selection in three areas: sensors, actuators, and processors. The attitude maintenance and control scheme
was validated using numerical code written in MATLAB. A secondary goal of the project was to
outline a design for a test-bed where the control policies could be verified experimentally. This project
developed a complete test-bed stand, leaving the construction portion for future ADCS teams, as well as
recommendations for the parts to be used in the experimental section.

Acknowledgements
We would like to primarily thank our advisor:
Professor Michael Demetriou, Ph.D.
Professor, Aerospace Engineering Program
Department of Mechanical Engineering, Worcester Polytechnic Institute
We would also like to extend our thanks to the member of the 2011 ADCS team. In addition, we would
like to thank the advisors and members of the other CubeSat subsystem design teams:
Thermal and Power Subsystem,
Professor John J. Blandino, Ph.D.
Associate Professor, Aerospace Engineering Program
Department of Mechanical Engineering, Worcester Polytechnic Institute
Structural and Mission Analysis Subsystem,
Professor Nikolaos Gatsonis, Ph.D.
Director, Aerospace Engineering Program
Department of Mechanical Engineering, Worcester Polytechnic Institute

ii

Table of Contents
Abstract .......................................................................................................................................................... i
Acknowledgements ....................................................................................................................................... ii
Figures .......................................................................................................................................................... v
Tables .......................................................................................................................................................... vii
Chapter 1 Introduction ............................................................................................................................. 1
1.1

Final Component Selection ......................................................................................................... 1

1.2

Improving Control Algorithm ..................................................................................................... 1

1.3
Testing of Control Policies .......................................................................................................... 2
Chapter 2 Previous Work ......................................................................................................................... 3
Chapter 3 Selecting Algorithm and Simulation Methods ........................................................................ 4
3.1

Proposed Simulation Procedure for ADCS ................................................................................. 4

3.2

Explanation of Iterative Process Diagram ................................................................................... 5

3.3

Example: ADCS System Architecture Based on the ITU PSAT 2.............................................. 6

3.4

Summary Procedure for Software Simulation ............................................................................. 6

3.5

Summary Procedure for Hardware Simulation............................................................................ 7

3.6

B-Dot Stabilizing Controller ....................................................................................................... 8

3.7

Difference Quotient Method ........................................................................................................ 8

3.8
Nyquist Sampling Theorem ....................................................................................................... 11
Chapter 4 Sensors and Actuators Selections .......................................................................................... 14
4.1

Comtech AA Coarse Sun Sensor ............................................................................................... 14

4.2

SSBV CubeSat Sun Sensor........................................................................................................ 16

4.3

EVAL-ADXRS450Z-M Gyroscope .......................................................................................... 18

4.4

Magnetometer ............................................................................................................................ 20

4.5

Magnetorquer Selection ............................................................................................................. 23

4.6

Clyde-Space ADCS Command Board ....................................................................................... 25

4.7
Summary of Final Hardware Selection...................................................................................... 26
Chapter 5 Model and Sensors Readings ................................................................................................ 27
5.1

Geomagnetic Model .................................................................................................................. 27

5.1.1

High Definition Geomagnetic Model (HDGM) ................................................................................... 27

5.1.2

International Geomagnetic Reference Field 11th generation (IGRF) ................................................... 27

5.1.3

World Magnetic Model (WMM) ......................................................................................................... 27

5.2

Sun Sensor Reading ................................................................................................................... 28

5.3

Alternative Determination Methods .......................................................................................... 30

5.3.1

Magnetometer-Only Deterministic Attitude/Rate Determination ........................................................ 30

5.3.2

State Determination with Failed Gyroscope ........................................................................................ 31

iii

Chapter 6

MATLAB Simulation Results............................................................................................... 32

6.1

Performance Improvement of the New Magnetorquer .............................................................. 32

6.2

De-tumbling Optimization ......................................................................................................... 35

6.2.1

Optimized Single Fixed Gain Investigation ......................................................................................... 35

6.2.2

Introduction of multiple fixed gains and adaptive (state dependent) gain ........................................... 37

6.3

Comparison of Three Gain Types ............................................................................................. 38

6.4

Attitude Maintenance ................................................................................................................ 40

6.5
Results ....................................................................................................................................... 40
Chapter 7 ADC Test Bed ....................................................................................................................... 44
7.1

Tabletop Test Bed...................................................................................................................... 44

7.2

Test Bed Design: ....................................................................................................................... 49

7.3

Stress & Displacement Analysis: .............................................................................................. 50

7.3.1

Tabletop Disk (Aluminum):................................................................................................................. 50

7.3.2

Air Bearing Stand (1045 Carbon Steel): .............................................................................................. 51

7.4

Test Bed Material Selection ...................................................................................................... 53

7.4.1

Central Cylinder................................................................................................................................... 53

7.4.2

Support Fins ......................................................................................................................................... 54

7.4.3

Test Bed Baseplate .............................................................................................................................. 55

7.5

Test Bed Configuration ............................................................................................................. 56

7.5.1

Central Processing Unit ....................................................................................................................... 56

7.5.2

Wireless Communication Board .......................................................................................................... 58

7.5.3

On-Board Communication Board ........................................................................................................ 58

7.5.4

I/O Transceiver Board ......................................................................................................................... 59

7.6

Power Considerations ................................................................................................................ 59

7.6.1

Battery ................................................................................................................................................. 60

Chapter 8

Conclusion & Recommendations .......................................................................................... 61

8.1

Test Bed Material Purchasing.................................................................................................... 61

8.2
Attitude Simulation ................................................................................................................... 61
Chapter 9 References ............................................................................................................................. 63
Appendices.................................................................................................................................................. 66
Appendix A: MATLAB Simulation Code input and output guide. ........................................................ 66
Appendix B: Connector Map for ADL LX8PC-AMD Geode LX800 ................................................... 68
Appendix C: Functional Diagram for Accesio 104-COM-8SM ............................................................. 69
Appendix D: Connection Layout Diagram for Diamond-MM-32-AT I/O ............................................. 70

iv

Figures
Figure 1: Iterative ADCS Design with SIL and HIL .................................................................................... 5
Figure 2: ADCS System Architecture ........................................................................................................... 6
Figure 3: Example Software Simulation Setup ............................................................................................. 7
Figure 4: Hardware Simulation Setup ........................................................................................................... 8
Figure 5: Honeywell's HM5883L: .............................................................................................................. 10
Figure 6: Matlab plot of derivatives by both methods. ............................................................................... 10
Figure 7: Dimension of Coarse Sun Sensor ................................................................................................ 14
Figure 8: Fine Sun Sensor ........................................................................................................................... 16
Figure 9: Gyroscope .................................................................................................................................... 18
Figure 10: Gyroscope Pin Connection ........................................................................................................ 19
Figure 11: Magnetometer ............................................................................................................................ 20
Figure 12: Magnetorquer ............................................................................................................................ 23
Figure 13: Magnetorquer Power and Moment Output Comparison ........................................................... 24
Figure 14: Clyde-Space ADCS Board ........................................................................................................ 25
Figure 15: Figure 16: Angular velocity vs. Time from Previous Year ....................................................... 33
Figure 17: Figure 15: Angular velocity vs. Time from Current Result....................................................... 33
Figure 18: Magnetorquer Control Output vs. Time from previous year ..................................................... 34
Figure 19: Magnetorquer Control Output vs. Time from current result ..................................................... 34
Figure 20: De-tumbling Time vs. Gain Value at Time Step of 0.5s ........................................................... 36
Figure 21: De-tumbling Time vs. Gain Value at Time Step of 5s .............................................................. 36
Figure 22: Pointing Accuracy vs time plot ................................................................................................. 41
Figure 23: Controller output during Attitude Maintenance ........................................................................ 42
Figure 24: Visualization of Tabletop Test Bed ........................................................................................... 44
Figure 25: Air-Bearing Shape ..................................................................................................................... 45
Figure 26: Specialty Components SRA250 drawing .................................................................................. 46
Figure 27: Nelson Air Corp Air Bearing drawing ...................................................................................... 47
Figure 28: Air Quality Requirements.......................................................................................................... 48
Figure 29: Test Bed Design ........................................................................................................................ 49
Figure 30: Tabletop Disk Strength Analysis ............................................................................................... 50
Figure 31: Tabletop Disk Displacement ..................................................................................................... 51
Figure 32: Air Bearing Stand Displacement Analysis ................................................................................ 51
Figure 33: Air Bearing Stand Strength Analysis ........................................................................................ 52
Figure 34: The 1045 Carbon High Strength Steel Rod ............................................................................... 53

Figure 35: The 1018 Carbon Steel Sheet .................................................................................................... 54

Figure 36: The Alloy Steel Hot-Rolled Steel Plate ..................................................................................... 55
Figure 37:ADL LX8PC-AMD Geode LX800 Single Board Computer ..................................................... 56
Figure 38: PC/104 Location for SBC.......................................................................................................... 57
Figure 39: Eurotech COM-1480 Wireless Communication Board ............................................................. 58
Figure 40: Accesio 104-COM-8SM Communication Board ...................................................................... 58
Figure 41: Diamond-MM-32-AT I/O Transceiver Board ........................................................................... 59
Figure 42: Inspired Energy ND2054 Smart Li Ion Battery......................................................................... 60

vi

Tables
Table 1: Data and Specifications of Coarse Sun Sensor ............................................................................. 15
Table 2: Data and Specifications of Fine Sun Sensor ................................................................................. 17
Table 3: Specifications of Gyroscope ......................................................................................................... 19
Table 4: Data and Specifications of Magnetometer .................................................................................... 20
Table 5: Final selections of Hardware ........................................................................................................ 26
Table 6: Method of Sun Sensor Reading .................................................................................................... 29
Table 7: Comparison of Magnetorquers' Data ............................................................................................ 32
Table 8: De-tumbling Time and Energy Consumption with Three Different Gain Types ......................... 38
Table 9: Gain Value for Attitude Maintenance ........................................................................................... 40
Table 10: Sun Pointing Accuracy Comparison ........................................................................................... 41
Table 11: Component Power Information .................................................................................................. 60

vii

Chapter 1 Introduction
For two years, WPI student project teams have been working on a design for a CubeSat system
that will carry the payload of the SphinX-NG X-ray detector to monitor solar fluctuations. These teams
have worked with the NASA Goddard Space Flight Center, the Space Research Center in Poland, and
other polish universities to develop a mission plan and experiment parameters. The previous two years of
projects outlined a simple structure sub-system with mechanical, thermal, and power considerations, as
well as recommended many options for an attitude determination and control system. The focus of our
project was selecting final components to be used in the CubeSat design, continuing to improve the
control algorithms for the processor, and developing a plan to test the control algorithms in the lab.

1.1

Final Component Selection

A major portion of our project was dedicated to the research of satellite components, specifically

components that could be used for the WPI CubeSat mission. This included both actuators, to maneuver
the CubeSat, and sensors, to detect the orientation of the satellite relative to both the sun and the earth.
The actuators were selected based on three key factors: mass, power required while operating, and their
size, as the CubeSat has very specific sizing requirements. Sensors were chosen based upon their mass,
size, and power as well, but more consideration was given to the accuracy of the component as the
CubeSat mission outlined by previous work has stringent pointing requirements for science experiment.

1.2

Improving Control Algorithm

Another goal for this project was modifying the control code to both; accept the new sun sensor

settings outlined by the Structural and Mission Analysis Subsystem team, as well as use the parameters
for the new magnetorquer selected for actuation. Also, new methods of determining the gain values to be
used during the different phases of the mission were researched so that the accuracy of the SphinX-NG
could remain within two degrees of the sun vector for as long as possible.

1.3

Testing of Control Policies

The final portion of our project included research into work where a laboratory set-up on Earth

used control algorithms for small satellites to control idealized versions of the control instruments that
would be used in space. Using information from this research, another goal for this project was to develop
a way that the control policies outlined by previous projects could be tested in the laboratory at WPI.

Chapter 2 Previous Work

Last years CubeSat-ADCs team has already selected the control method and algorithm designs.
For the control of the CubeSat, there are three phases after launch: de-tumbling, initial attitude
determination and recursive attitude determination.
As the most prevalently used algorithms in satellite de-tumbling process, B-Dot controller
controls the satellite by aligning it with the earth magnetic field vectors. Indeed, all small satellite used
some variation of B-Dot as the controller for de-tumbling process [1] [2] [3].
As for initial attitude determination, TRIAD method is one of the earliest and simplest solutions
to the spacecraft attitude determination problem. The method requires two sets of vectors: an observation
vector from magnetometer and sun sensors, and a reference vector for each observation in terms of its
inertial reference frame. Wahbas equation is also known as Error function. It is created to minimize
errors when using TRIAD method. Different weights are applied to the each set of the sensors readings.
The more accurate the sensor is, more weight is given. Immediately afterwards two solutions were
offered to solve the problem: Davenports q-Method and Quaternion Estimator Method (QUEST). Both
of the methods use quaternion as an approach to optimize the result given by Wahbas equation [4].
As for recursive attitude determination, Extended Kalman Filter, [5] [6] [7]aka linear quadratic
estimation, is a two-step process with a discrete time step. The sensors on satellite do not collect data
continuously, so the Kalman filter is choose as it best incorporates the situation. Since most systems are
nonlinear, the optimal estimate of Kalman Filter for linear system models is not the best solution.
Extended Kalman filter (EKF) is used in satellites since it is the nonlinear version of the Kalman filter.
EKF linearizes the system about an estimate of the current mean and covariance and produces an estimate
of the future state.
In addition REQUEST method is a recursive version of Davenports q-method that requires only
one sensor reading. It is good for attitude maintenance during eclipse period since only the magnetic field
measurements are available.

Chapter 3 Selecting Algorithm and Simulation Methods

3.1

Proposed Simulation Procedure for ADCS

The Attitude Determination and Control System (ADCS) comprise hardware such as sensors,

actuators and the ADCS board which, in our case acts as the ADCS computer or processing unit. The
ADCS also includes software that has been coded or programed into the ADCS computer. This software
contains all the control schemes or algorithms of the ADC system. Therefore on a broad scale it can be
observed that there are two major components that make up the ADCS the hardware and software.
Due the unreliable nature of the inexpensive hardware and software that may be used in this
project it is necessary to conduct simulations of hardware and software and indeed the entire ADCS in
order to ensure that the system can cope with likely software and hardware failures. Therefore a possible
software and hardware testing approach will be outlined here. Areas of testing that could be considered in
these tests are situations of flawed sensor readings and actuator failures for instance. Also, the overall
ADCS system operation should eventually be tested using air bearings and a test bed, such as the one
being designed in this project, to better recreate conditions in the actual orbit or flight environment. A
device, such as the Helmholtz coil, for producing uniform magnetic field should be considered to simulate
the earths magnetic field. An IEEE project that has attempted a similar approach shows an iterative
procedure using Hardware-in-Loop and Software-in-Loop simulations where unsatisfactory results are rerun in the loop [8].


Figure 1: Iterative ADCS Design with SIL and HIL 8

3.2

Explanation of Iterative Process Diagram

The ADCS design starts with the design of its control algorithms, for it is the control schemes that

determine the type of software or hardware needed. These algorithms are then tested in an ideal setting
with no induced error. Based on the results of our assessment or simulation, we can either return to the
algorithm design phase to make necessary adjustments to our design or we can proceed. When we are
satisfied with the results from our algorithms in the design phase and simulation, we can then go on to run
the algorithms again, but this time, introducing errors to observe how the system will respond in a crisis
situation [8].
After all software simulation is done, the appropriate hardware can be obtained and tested in
simulation using the software.

3.3

Example: ADCS System Architecture Based on the ITU PSAT 2

Figure 2: ADCS System Architecture 8

Additional Components of ITU PSAT 2 System Architecture Include: Attitude Propagator, Orbit
Propagator, Estimation and Filtering, Actuators [8].

3.4

Summary Procedure for Software Simulation

The ADCS algorithms should be tested on available software platform. Models for the attitude and orbit

propagators can be generated from Satellite Toolkit (STK) since mathematical models generated by individuals
for the both propagators may fail to emulate states or conditions of the mission actual mission. In addition since it is
possible to extract the actual state data from STK we can see how well estimation layer works. ADCS software is
provided either by an interface in STK or by some outside source, software receives state update from STK and
noisy readings, magnetometer calibration errors, gyro biases, etc are applied to state vector to simulate actual sensor
reading during mission [8].


Figure 3: Example Software Simulation Setup 8

3.5

Summary Procedure for Hardware Simulation

Hardware such as sensors, actuators and the on-board- computer also have to be tested in

simulation with experimental platforms which will simulate the in-mission conditions of outer space.
Such a setup is illustrated in the diagram below; where the rack system comprises two computers (PCs), a
monitor to display simulation of satellite and a satellite hardware simulation system providing real
responses to data.


Figure 4: Hardware Simulation Setup 8

3.6

B-Dot Stabilizing Controller

After ejection of the satellite in to orbit, it experiences varying angular velocities. For attitude

determination to take place, the satellite has to be de-tumbled, in other words the angular velocities of the
satellite must be made to approach zero.
The de-tumbling approach from last year was studied as well as the analysis of several types of
de-tumbling algorithms and control policies. After identification of the de-tumbling methods, possible
improvements to the current algorithms and methods were researched.
The stabilization phase is concerned with stopping random motion with no goal for a specific
attitude orientation. A control policy called the B-Dot controller was the most prevalent method found in
research on previous CubeSat missions. There are various policies that stabilize the spacecraft initially but
most of these are variations of the B-dot control, based on Lyapunov functions.

3.7

Difference Quotient Method

The difference quotient method is the natural method to considering when dealing with

differentiation of discrete signals. This method involves dividing the difference between measurements by
the time interval between them. The value of the magnetic field at a point (k) could subtracted from the

value at a future point (k+1), in which case it is considered as the forward difference. In this case, the
magnetic field value at (k) would have to be stored in order to be used in the differentiation. Alternatively,
the magnetic field value at (k-1) could be subtracted from the value at (k). This is referred at to as the
backward difference. In this method, the first two magnetometer outputs or discrete signals will have to
be stored. Below are the expressions for the above mentioned methods:
Forward Difference:

(1)

Backward Difference:

(2)

Combining the two above approximations we get an improved one known as the central
difference.
Central Difference:

1
2

(3)

where dt = time interval

For the case of the given magnetometer for this project, the difference quotient method found
most appropriate to begin with is the backward difference which makes use of the present as well as the
previous measurements to compute the derivative.
Magnetometer Characteristics


Figure 5: Honeywell's HM5883L:
The magnetometer chosen for the CubeSat is the Honeywells HMC5883L Triple Axis
Magnetometer with minimum and maximum sampling rates 0.75 Hz and 75 HZ respectively. It has a
sampling interval of 0.006 seconds and a magnetic field output of 1.16 gauss along the X and Y-axes,
and 1.08 gauss along the Z-axes.
Based on the minimum and maximum magnetometer output a cosine function was used to model
magnetometer readings as

0.4 cos

1.12. The derivative was then computed analytically and

by using the difference quotient method and both results were compared.
Comparing both derivatives at a high sampling interval to show differences

Figure 6: Matlab plot of derivatives by both methods.

10

In red = dB/dt: using difference quotient method:

= -0.4sin(t)

In blue = dB/dt: found by solving analytically: dB/dt = -0.4sin(t)

The difference quotient method is developed from the Taylor series approximation therefore the
errors that arise due to their use are also based on the same Taylor series approximation and also depend
on the time interval (dt). The errors are as follows:
For the forward difference:

Where B = the 2nd derivative

For the backward difference:

Where B = the 3rd derivative

For the central difference:

3.8

[9]

Nyquist Sampling Theorem

Another method that was considered was differentiation using the Nyquist sampling criterion.

This method is described in the article for differentiation of Multi-dimensional signals by Farid and
Simonelli [10]. The theorem states that a function with a limited upper frequency can be reconstructed
from an infinite sequence of samples spaced at half the sampling rate or with a bandwidth no greater than
the sampling rate. This method is used to differentiate a continuous signal that is interpolated from the
initial discrete signal. This interpolation is done by using the ideal interpolator function

(also called the

sic function). This is based on the assumption that the discrete signal was obtained by sampling an
original continuous function containing frequencies no higher than

of

at a sampling rate

[10].
Using the above assumption the Nyquist sampling theorem implies that the continuous signal

may be reconstructed from the discrete signal as the following samples:

11

(4)

Continuous signal derived from the given discrete signal [10].

Where:

/
/

this is the continuous ideal interpolator of sinc function

f x : the continuous time signal

f k : the discretely sampled counterpart
Assuming that the continuous signal equation above converges, we can then go ahead and
differentiate the continuous function on both sides of the equation.
The derivative operator will only be applied to continuous functions,

and

Let D be the derivative operator [10]

Where:

(5)

the derivative of the sine function

/
/

Arrive at a definition of discrete differentiation by sampling both sides of the above equation on
the original Sampling interval. If the original sampling interval is n and the sampling rate is T we
have:

12

(6)

Where dT is the T-sampled sinc derivative.

Accurate Implementation of this method requires very large filters.
After considering and exploring the appropriate options for differentiation of discrete signals, the
next task is to design appropriate filters for these for the chosen method of differentiation. Kalman filters
can be used to estimate the discrete time sequence governed by the difference equation of the system we
are considering. After desired constraints have been set, derivative filters can then be designed for a given
order and length.

13

Chapter 4 Sensors and Actuators Selections

4.1

Comtech AA Coarse Sun Sensor

Figure 7: Dimension of Coarse Sun Sensor
Sun sensors, although mass-produced according to listed specifications, contain minute
differences in photodiode angles and sensitivities that necessitate calibration for use in space. The goal of
the data from the Coarse Sun Sensor (CSS) which are located along the sides of the satellite is to allow
Attitude Determination & Control (ADC) to determine the Angle of Incidence (AOI) and make attitude
adjustments to bring the view of the Sun. The AOI varies at 15 increments because the CSS requires
only a coarse view of which direction to turn and then the Fine Sun Sensor (FSS) data will allow for finer
adjustments. The CSS data are symmetrical and include voltage outputs that could correspond to either a
positive or negative AOI. The Field of View (FOV) of the CSS at the bottom of the satellite overlaps
with the FOV of those mounted on opposing sides, telling the ADC which AOI the voltage corresponds to
base on whether or not the CSS at the base shows a voltage output.

14

The following table provides the CSS data and specification [11].
Table 1: Data and Specifications of Coarse Sun Sensor
Flight Heritage
Field of View
FoV

Accuracy

ALEXIS, HETE, MOST, ChipSAT and STPSat1

120 fullangle circular field of view

3sigma 5 of 1axis knowledge

Power

None required

Volume

1.27 cm diameter x 0.90 cm height

Mass

10 g w/ 1.27m flying leads

Housing diameter: 1.27 cm

Size

Flange diameter: 2.286 cm

Sensor height: 0.899 cm

Bandwidth

100KHz

Operating
Temperature

40C to 100C

Vibration Test
Levels

14.1g rms protoqual testing procedure

Shock Test Levels

60g protoqual

Radiation TID
Materials

Price

1 Mrads
Epitaxial Silicon help makes the sensor radiation hardened
$2,641.4 US per 1
$13,207 US per 5

15

4.2

SSBV CubeSat Sun Sensor

Figure 8: Fine Sun Sensor
The Fine Sun Sensor (FSS) data allows ADC to make minute attitude adjustments, allowing the
adjacent solar array to maximize solar power generation. The threshold orientation accuracy for the
satellite is 5. ADC required calibration data that would allow for fine adjustment: the data set included
10 increments along the entire FOV and 2 increments surrounding the region normal to the sensor
mounting surface [12].
The FSS has four individual diodes canted towards the sensors quadrants, each receiving voltage
outputs. A potential application of the data is to relate x angle, y angle and voltage by processing
calibration data into a 3D surface. By orienting these surfaces to correspond to sensor setup, ADC could
generate contour projections of AOI based on on-orbit voltages. The intersection of these projections
would be the satellites attitude relative to the Sun.

16

The following table provides the FSS data and specifications:

Table 2: Data and Specifications of Fine Sun Sensor

Flight Heritage

UKube1 and TDS1 satellites in 2012

Mass

5 g

Power

5 mA

Size

33mm x 11mm x 6mm

Operating temperature

25C to 150C

Field of view

120

Update Rate
Bandwidth

10Hz

Accuracy

0.5

Power Supply

3.3 or 5 V

Interface

I/F: 4 analogue channels, 9way NanoD Connector

Material

Aluminum & Titanium

Price

$5,000 US

To capture the necessary calibration data, the sensors were tested using a Calibrated AM0 Sun
Simulator light source at varying angles of incidence, recording voltage output from the photodiodes at
each position. The sensors were also tested across a range of temperatures to see if there is performance
variation within the companys listed temperature range.

17

4.3

EVAL-ADXRS450Z-M Gyroscope

Figure 9: Gyroscope

The EVAL-ADXRS450Z-M is a simple breakout board that enables easy connection of an

ADXRS450 into an existing system. The EVAL-ADXRS450Z-M is a digital-output, single-axis
gyroscope with built-in temperature calibration [13].
The ADXRS450 is available in two packages: an SOIC package for yaw-axis rate sensing, and a
vertical-mount package (VMP) for pitch- or roll-axis sensing. In applications requiring multi-axis sensing,
the innovative VMP greatly simplifies assembly by eliminating the need for additional boards mounted at
90. An advanced, differential, quad sensor design rejects the influence of linear acceleration, enabling
the ADXRS450 to operate in exceedingly harsh environments where shock and vibration are present [14].
The ADXRS450 uses an internal, continuous self-test architecture. The integrity of the
electromechanical system is checked by applying a high frequency electrostatic force to the sense
structure to generate a rate signal that can be differentiated from the baseband rate data and internally
analyzed.
The ADXRS450 is capable of sensing angular rate of up to 300/sec. Angular rate data is
presented as a 16-bit word, as part of a 32-bit SPI message. The ADXRS450 is available in a cavity
plastic 16-lead SOIC (SOIC_CAV) and an SMT-compatible vertical mount package (LCC_V), and is

18

capable of operating across both a wide voltage range (3.3 V to 5 V) and temperature range (40C to
+105C).
The following table presents all the ADXRS450 specifications and pin connections:
Table 3: Specifications of Gyroscope

Operating Temperature

40C to 105C

Sensitivity

80 LSB

Typical Bandwidth

80 Hz

Voltage Supply

3.15V to 5.25V

Board Size

33mm x 33mm x 1.2mm

Price

$59.54 US

Material

Aluminum & Titanium

Connect

Main Computer
SPI digital output with 16bit dataword
Internally temperature compensated
Complete rate gyroscope on a single chip

Figure 10: Gyroscope Pin Connection

19

4.4

Magnetometer

Figure 11: Magnetometer
Table 4: Data and Specifications of Magnetometer

Size

17.78X17.78X0.9mm

Mass

18 mg

Operating Temperature

3085 C

Heading Accuracy

12 Degree

Linearity

0.1%

Measurement Period

6ms

Supply Voltage

2.163.6V

Max Current Draw

6 A

Rate

100 KHz or 400 KHz

Price

14.95$

Materials

NickelIron Permalloy

Last years Design and Analysis project outlined a Magnetometer created by Honeywell, a leader
in the sensor/detector industry. This component satisfied three of the four design guidelines set forth by

20

the first Design and Analysis team from 2011 [1]. The Magnetometer selected is the HMC5883L 3-axis
sensor, with a suggested breakout board built by SparkFun. This magnetic sensor had excellent flight
heritage, measured all 3 axis, and operates in the 3.3V range. Also, the component has a very low cost.
Taking the HMC5883L as a baseline, initial research was done to see if there were any new
components made by the same company, or similar parts made by a different company. The Honeywell
Company touted the HMC line as the premiere sensors in the field. The Freescale FXOS8700CQ: 6Axis Xtrinsic Sensor, a component combining an accelerometer and magnetometer, was looked at, but
determined to be much too heavy, with a lesser accuracy for the magnetic measurements. Also, the power
requirements were much greater to power both types of sensors [15]. The HMC5883L was again selected
as the most viable option because of its accuracy, weight, and flight heritage.
Two mounting options were presented by last years Design and Analysis team for the
magnetometer. The first was a static option, where the sensor was on the top +Z face of the design, with
its axis aligned with the body axis of the CubeSat. This option is mechanically simple, cheap, and does
not create any torques or disturbances during normal use. The second option was to use a boom-style
solar panel deployment to get the sensor as far away from the rest of the electric components as possible.
This approach is mechanically complex, and required additional analysis done to acquire inertial
measurements. However, after restrictions from the P-POD document were discussed, it was realized that
a deployable sensor would not be practical for our design as the sensor would have to remain un-deployed
for thirty minutes after the ejection of the CubeSat from the launch vehicle. The magnetometer is
necessary for initial de-tumbling functions, and thus waiting thirty minutes to begin to de-tumble would
vastly increase de-tumbling time. Thus, the top surface mounting option was selected because of the ease
of integration and the simplicity of the design.
The Magnetometer would need to have its data output conditioned by its own breakout board, and
then sent, under last years specifications, to the main computer [16]. From last years design, the ClydeSpace Mission Interface Computer was suggested as the main computer for the WPI mission. This
motherboard is integrated with pluggable processor modules (PPMs), which allow direct connection from

21

the sensor output to the computer. The Clyde-Space component was selected because of its two types of
PPMs, one for high performance data handling (the ARM922T), and one for low power requirements and
redundancies (the MSP430). The HMC5883L would be connected to ARM922T module, as the
magnetometer can output data up to 400 Hz.

22

4.5

Magnetorquer Selection

Figure 12: Magnetorquer
After testing three types of magnetorquers last year, the ADC team determined that the best
possible component for the WPI mission would be the ZARM MTO.2-1 Magnetorquer [1]. This
component had the lowest mass, size, and power requirements of all the types tested, and reached the
magnetic moments required with a low power requirement and much less weight. The Zarm Technik
Corporation was contacted for more information about their MTO.2-1; however they had discontinued the
MTO.2-1 product line. Zarm suggested the possibility of using a MTO.5-1, which was one of the other
products tested by last years team. This piece reaches the same magnetic moments as the MTO.2-1, but
at a much lower power cost. However, the MTO.5-1 is longer, heavier, and more expensive than the
MTO.2-1, but given that there are no other viable options, the MTO.5-1 has been selected as the
magnetorquer of choice for the WPI mission.

23


Figure 13: Magnetorquer Power and Moment Output Comparison

24

4.6

Clyde-Space ADCS Command Board

Figure 14: ClydeSpace ADCS Board
In researching new magnetometers and magnetorquers, the Clyde-Space ADCS board was
discovered. This is a new product made by Clyde-Space, a trusted and prominent company in the CubeSat
industry. This component would serve as a command hub for the systems involved with de-tumbling and
attitude determination/control. With PPM interfaces for multiple sensors, including but not limited to;
coarse and fine sun sensors, GPS systems (an included GPS daughter board is optional), and gyroscopes.
The advanced version of the board comes with its own included magnetometer, but the company has
stated that it is a modular component and could be removed depended on the design constraints specific to
each user. This board could be connected to a PPM on the main computer, and would control the sensors
and actuators connected to it, while sending the necessary information to the main computer. Clyde-Space
has embedded programs in the board, such as de-tumbling and sun tracking, but also states that the board
has the ability to accept custom MatLab coding [17]. All 5 of the CubeSat sun sensors, as well as the
GPS, gyroscope, and magnetometer, will be connected and feeding information to the ADCS board,
which will be directly controlling the actuators using the information being fed to it by the sensors.

25

4.7

Summary of Final Hardware Selection

Table 5: Final selections of Hardware

Sensor/

Model

Manufacture

Number
Needed

Total
Price

Power
Requireme
nt

Sun Sensor

Coarse

Comtech AA

$10,566

None

Sun Sensor

Fine

SSBV Space
and Ground
Systems

$5,000

Magneto
meter

HMC5883L

Honeywell

$14.95

0.25 mW

Gyroscope

ADXRS450

Analog Devices

$59.54

30 mW

Magnetic
Torquer

51 PO

Zarm Technik

Unknow
n

275 mW

Actuator

140 mW

Materials employed included dissipative paints, indium-tin-oxide films, carbon loaded polymers,
thin SiOx coatings on metallic substrates, Nickel-Iron Permalloys, and conductive tapes.

26

Chapter 5 Model and Sensors Readings

The goal of the attitude determination and control system is to develop comprehensive system
and algorithms that the CubeSat can use for its state control, stabilizing and mission required maneuvers.
In order to achieve this goal, acquiring accurate information from geomagnetic model, sun sensor,
magnetometer, and gyroscopes readings is significant for the CubeSat to perform control algorithms.

5.1

Geomagnetic Model
Geomagnetic model is an important piece of information for the CubeSat because it is one of the

reference fields which will be used in the control algorithm. The values from geomagnetic model will be
compared with magnetometer readings in order to determine the CubeSats state which will then be used
to control he magnetorquer output. Three geomagnetic field models are available for use.
5.1.1

High Definition Geomagnetic Model (HDGM)

It is provided by US National Geophysical Data Center for high accuracy geomagnetic

referencing. This model includes the main geomagnetic field, crustal field to degree 720 and also secular
variation. This model significantly reduces geomagnetic referencing errors. It is updated each year. But
the model is priced at $19,997 per year [18].

5.1.2

International Geomagnetic Reference Field 11th generation (IGRF)

As selected by last years ADC team, this model holds adequate accuracy as indicated by the

mission requirements. This model is also used by Satellite Tool Kit (STK) for geomagnetic field reference
[19]. It is updated every five years. The detailed descriptions and calculations of magnetic field vector can
be found in CUBESA-ADC12.
5.1.3

World Magnetic Model (WMM)

This is an alternative geomagnetic model. It is not as accurate as IGRF. As our mission does not

require high accuracy of the attitude, WMM is optional as the computational time is less than that of the

27

IGRF. The strength of the magnetic field changes over time. According to the data from WMM 20102015, the greatest change in magnetic field strength is about 90 nT/yr [20]. Compare the error produce by
the models with the magnetic strength shift, it is well within the tolerance. WMM models the
geomagnetic field in the following way:
,

The geomagnetic field varies in space and time.

accounts for 95% of the total field strength.

is the dominating part of the field. It

is much smaller when compared with

. This field

cannot be modeled with low degree spherical harmonic models therefore is excluded in WMM.
models the field from magnetized crustal rocks, so its not useful in determining the magnetic field in the
arises from varies resources such as ionosphere and magnetosphere. It varies with

space.

time and space [20].

The advantage of WMM is that it is implemented in the latest version of MATLAB (MATLAB
2012b). If we have the processor that can run MATLAB script, its very convenient to use WMM instead
of IGRF.

5.2

Sun Sensor Reading

The selections of both fine and coarse sun sensors were finalized. The top unit in the 3U-CubeSat

is occupied by the mission payload and it is not accessible for any purposes, therefore, the original design
of 5 plus 1 sun sensor setting came down to only 4 plus 1, which means only 4 coarse sun sensors will be
used in addition to one fine sun sensor. The selected coarse sun sensors provide only one-axis information
about the Suns position; in other words, the Suns position cannot be determined accurately with the oneaxis sun sensors. As a result, a rough estimation of the Suns position is required to effectively control the
CubeSat. The following table indicates the detailed methods of estimating the Suns position.

28

Table 6: Method of Sun Sensor Reading

Available face s of
sun sensor reading

Phase angle degrees

Sun Position vector

in Body frame

90

180

90

90

90

0, 1, 0

No sensor reading

270

90

0, 1, 0

180

0, 0, 1

X, Y

45

90

2/2, 2/2, 0

X,Z

135

2/2, 0, 2/2

X, Y

135

90

2/2, 2/2, 0

X,Z

180

135

2/2, 0, 2/2

Y,Z

90

135

0, 2/2, 2/2

X, Y,Z

45

135

0.5, 0.5, 2/2

X, Y,Z

135

135

1, 0, 0
1, 0, 0

0.5, 0.5, 2/2

The +Z face is pointing to the sun. Whenever the reading from +Z face is available, which is the
reading from the fine sun sensor, it will be taken as a prior value and any readings from other coarse
sensors will be ignored. The method of estimation primarily assumes that if the sun sensor on only +X
face has readings, the Suns position is considered to be aligned with the +X axis and if sun sensors on
both +X and +Y faces have readings, then the Suns position is considered to be aligned with 45 degrees
from both +X and +Y direction and so the same for all others. Notice that there is no Y face because
only 4 coarse sun sensors are used. No sensor reading indicates that none of the 5 sun sensors on board
has sight of the Sun, so the Sun is assumed to be at the Y phase. Since the mission design group will be
providing detailed orbital information, then the information on whether or not the CubeSat is during
eclipse is known. As a result, its adequate to assume if the CubeSat is known to be in lighting period,

29

then the no sensor reading indicates the Sun is in the Y face. With the list of all possible phase angles
the Suns position vector in the body frame then can be calculated and thus be used in the control
algorithms.

5.3

Alternative Determination Methods

Provided by the Hubble Space telescope, four of its six world class gyros failed. And there are

several known cases where the gyro failed during mission. In order to secure our mission, back up
determination algorithms should be kept on board in case of parts failure.

5.3.1

Magnetometer-Only Deterministic Attitude/Rate Determination

This method was previously used on Solar, Anomalous, and Magnetospheric Particles Explorer

(SAMPEX) which is a small explorer spacecraft. It has been found that the algorithm is effective of
yielding accuracy of 1.5 degree in attitude and 0.01 degree/second in rates. This Method can be used as a
back-up algorithm in case the gyroscope failed during the mission [21]. This algorithm allows us to
determine the angular rate of the CubeSat.
The deterministic scheme constructs second vector measurement from the 1st derivative of B
resolved in the reference and body frames. The transformation equations are given below:

AB

AB

(7)

Where A is the attitude matrix, is the angular velocity vector, and R and A represents reference
and body frame respectively [21].

30

Along with the deterministic attitude determination from magnetometer-only data, Real-Time
Sequential Filter (RTSF) is used for accurate determination and faster convergence.

5.3.2

State Determination with Failed Gyroscope

The Far ultraviolet Spectroscopic Explorer (FUSE) had failed gyro during mission [22]. The

study after the failure of FUSE uses hybrid integrated-rate parameters (IRP) Euler filter approach.

(8)

where D t is the spacecraft attitude matrix. t

velocity and t

, with t the spacecraft angular

defined as follows:

(9)

The discrete-time version of the attitude matrix is:

31

1 ,

1 ,

(10)

(11)

Chapter 6 MATLAB Simulation Results

The previous ADC team had achieved quite satisfying results with MATLAB simulation for both
stabilization and attitude determination and maintenance phases. This year, the results were further
improved.

6.1

Performance Improvement of the New Magnetorquer

Compare with last year, new magnetorquer was selected this year. Although more weight is

added to the CubeSat and the structure group had to modify the layout of each component to fit them into
the CubeSat, they have shown much better performance during stabilization phase.
Table 7: Comparison of Magnetorquers' Data
Model
Maximum moment
Max power

MTO.2-1
0.2 Am2
0.2W

MTO.5-1
0.5 Am2
0.3W

Using the same setting for the simulation from last years ADC group, e.g. 5 degrees for yaw,
pitch and roll angle and 5 degrees per second for initial angular velocity and the updated magnetorquer
information, comparisons of the velocity profile and control output profile are shown below.

32


Figure 15: Figure 16: Angular velocity vs. Time from Previous Year

Figure 17: Figure 15: Angular velocity vs. Time from Current Result

33


Figure 18: Magnetorquer Control Output vs. Time from previous year

Figure 19: Magnetorquer Control Output vs. Time from current result

34

When the angular velocity for each components of the CubeSat is less than 0.1 degree per second,
then its considered that the satellite is stabilized. Using this criterion, the optimized stabilizing time from
the last year was given to be around 3300 seconds whereas with the latest magnetorquer, the stabilizing
time was much less and was estimated at about 1100 seconds. By analyzing the controller output, the
reason why the new magnetorquer has such a huge improvement became obvious. The maximum moment
produced by the new magnetorquer is 0.5 Am2 which is twice more than that of 0.2 Am2 of the old one.
Also the plot from previous year showed that the old magnetorquers peak at the first 1000 seconds. As
with the new magnetorquer, the output moment only went up to 0.4 Am2 for the very first 100 seconds
and went only slightly above 0.2 Am2 after 100 seconds. The extra moment produced by the new
magnetorquer had indeed improved the stabilizing time significantly.

6.2

De-tumbling Optimization
To further improve the performance of the control algorithm during stabilization period, the team

decided to carry on with more investigations into the gain determination.

6.2.1

Optimized Single Fixed Gain Investigation

With the updated specifications for the new magnetorquer, updated de-tumbling time was

acquired using the MATLAB simulation. Using 5 degrees for yaw, pitch and roll angle and 5 degrees per
second for initial angular velocity, the summary of the de-tumbling time with different noise filters is
shown below in the plot. EKF denotes extended Kalman filter, LPF denotes low pass filter and triad
denotes the raw data without any filtering.

35


Figure 20: Detumbling Time vs. Gain Value at Time Step of 0.5s

Figure 21: Detumbling Time vs. Gain Value at Time Step of 5s
At time interval of 0.5 seconds, the optimized de-tumbling time of around 1100 seconds is
observed at gain value of approximately -120000 to -125000. In the case of time interval of 5 seconds, the
optimized de-tumbling time of around 1250 seconds is observed to be at gain value of approximately 115000 to -120000. The different results for different time interval revealed that if the sampling
frequencies of the gyroscope and magnetorquer are high, better de-tumbling time was expected to be
achieved and the magnetorquer can produce higher torque to control the satellite. If the sampling
frequencies were low, the controller gain was relatively small because large gain will tend to overshoot

36

the system. In any cases, the gain cannot exceed certain values depending on the sampling frequency and
further investigations of the critical value are needed for ensuring the system stability.
6.2.2

Introduction of multiple fixed gains and adaptive (state dependent) gain

Recall the equations for magnetorquer output and torque applied:

where

(12)

is the controller output, B is the magnetic field vector,

is the angular velocity vector

and C is the controller gain.

The previous simulation was based merely on a single fixed value as controller gain. After further
investigating, it is possible to improve the performance of the magnetorquer by implementing multiple
fixed gain or adaptive gain. The cross product gave a vector and C was set to be a 3-by-3 matrix so to
match the dimension of , the controller output as a vector.
As a multiple fixed gain, the controller gain C has the following form:

C
0
0

0
0

(13)

C1, C2 and C3 have values on the order of 105 which are on the same order as the gain value for
single fixed gain. For the simplicity of investigating the effects of implementing multiple fixed gains, C
was taken to be a diagonal matrix. Instead of having single gain for all three axes, now each axis has its
own gain value which means that output of each magnetorquer can be modified.
As for adaptive gain, the controller gain C has the following form:

37

,
0
0

0
,
0

0
0
,

, ,

(14)

denotes that it is a function of all components of the magnetic field and angular

velocity. Similarly, for the simplicity, C was also taken to be a diagonal matrix. The possible
improvement by implementing adaptive gains was that at different strengths of magnetic field and angular
velocities, the optimal controller gains would be different; therefore having adaptive gains can possibly
improve the performance of the controller more.

6.3

Comparison of Three Gain Types

With the latest inertia matrix, along with the specifications of the new magnetorquer, MATLAB

simulation results using single fixed gain, multiple fixed gains, and adaptive gain were obtained. However,
the plots of angular velocity and controller output vs time are very similar and its rather difficult to judge
which gain mode is better. In order to study the differences among the three methods, the squares of the
controller output are integrated over the same time period of 2500 seconds (note that controller output is
directly proportional to the power consumption, its reasonable to assume that the more controller outputs,
the more energy it consumes) and below is the results of relative energy consumption by magnetorquer
for each type of gains.
Table 8: Detumbling Time and Energy Consumption with Three Different Gain Types
Single fixed gain

Multiple fixed gains

Adaptive gain

9.1998

11.6167

6.7102

y-axis [A.m ]

15.9138

11.4179

13.2193

z-axis [A.m2]

10.4868

11.8592

13.7142

sum [A.m ]

35.6003

34.8938

33.6437

De-tumble time [s]

1480

1340

1250

x-axis [A.m ]

38

The total relative energy consumption for single fixed gain is the highest amongst all three, and
three magnetorquers have uneven output, meaning the magnetorquer in y-axis does more work and
leaving the other two in x- and z-axes do less work. In addition, the de-tumbling time using single fixed
gain is the longest among all. Note that the de-tumbling time presented here is higher than the ones
presented in the previous section of determination of optimal controller gain; the reason is inertia matrix
has been updated since then and was used for the simulation results in this section.
The advantage of using multiple fixed gains is clearly seen that the total energy consumption is
less, the de-tumbling time is less and more importantly each magnetorquer is able to provide similar
amount of output so to achieve an even wear for the magnetorquers and reduce the possibility of
breakdown during mission.
Lastly for the adaptive gain, both the energy consumption and the de-tumbling time are further
reduced. The state depending function used for this simulation is:

19500 5

B
0.00005

B
0.00005

(15)

The denominators in each term in function above were chosen relative to the maximum values of
each component readings. B was on the order of maximum of 10-5 [Tesla] and was set initially to 5
degrees/s. This reasonable expression for the adaptive gain produces satisfying results. In order to achieve
the even magnetorquer output, better de-tumbling time, and better energy consumption, further
investigations are needed to find the best adaptive expressions for each entry. As in the simulation,
controller gains are simplified to have only diagonal entries, replacing zeroes in the gain matrix could
possibly improve the performance of the controller even more.

39

6.4

Attitude Maintenance
As desired by mission payload, the purpose of the CubeSat is to point to the sun as accurate as

possible during lighting period. Last year has already achieved some results, but with the updated
magnetorquer and inertia matrix information, additional investigation was needed. The algorithm which
was already implemented in MATLAB simulation is quaternion feedback controller [1].

(16)
Proportional gain

, corrects the pointing error whereas derivative gain

corrects the angular

velocity which tends to stabilize the system by applying torque opposite to the direction of rotation. So
when the angular velocity is small, the term of proportional gain leads the magnetorquer output and when
the angular velocity is high, the derivative gain will keep the magnetorquer output more for stabilization.

6.5

Results
After hundreds of hours of simulation run, the optimal gain was determined as follows:
Table 9: Gain Value for Attitude Maintenance
Gain Type

Value

Proportional gain

0.0003

Derivative gain

0.0004

Maximum derivative gain

0.006

Minimum derivative gain

0.0003

Derivative gain in eclipse

0.02

Because of the random errors produced in the MATLAB simulation, 10 simulation results were
obtained and average of the sun pointing accuracy over time was taken and below is the plot:

40


Figure 22: Pointing Accuracy vs time plot
Then the percentage of the sun pointing accuracy under certain degrees can be found. Note that
the percentage was calculated out of the time during the sunny period which means the time spent by the
CubeSat in eclipse period was omitted in the percentage calculation. A comparison of the sun pointing
accuracy was also shown below.
Table 10: Sun Pointing Accuracy Comparison
Accuracy / degree

New Percentage

Percentage from last year

0.25

3.72%

5.21%

0.50

36.53%

15.14%

0.75

58.52%

30.95%

1.00

67.56%

44.47%

1.50

78.06%

62.70%

2.00

84.45%

72.09%

2.50

90.21%

3.00

93.10%

41

As seen from the table, the CubeSat now spends 12.3% more time within the 2 degree accuracy.
And up to 93% within 3 degree accuracy. Although less time was spent within the 0.25 degree accuracy,
the extra 20-30% more time within 1 degree is of the teams interest. The improvement of the pointing
accuracy could be resulted from few changes that were made: the change of the inertia matrix, new
magnetorquer and simpler sun sensor algorithm.

Figure 23: Controller output during Attitude Maintenance

From the figure of magnetorquer output, it can be seen that the controller peaks at its limit, but
its only for the first few hundreds seconds. The new magnetorquer produced extra power to control the
CubeSat, and reduced the time to achieve its desired pointing angle more quickly. But when it came to
minor maneuverer for pointing corrections, the new magnetorquers had no advantages over the old and
less powerful ones. What the new magnetorquer can achieve was to shorten the time for the CubeSat to

42

turn so that the fine sun sensor can obtain steady readings and achieve a small gain in the point
percentages for one sunny period. However as for the whole mission, the extra time spent in sight of the
Sun would be amplified in hours and gave the payload instrument more accurate readings.

43

Chapter 7 ADC Test Bed

7.1

Tabletop Test Bed

Figure 24: Visualization of Tabletop Test Bed
The purpose of the test bed is to prove that our Attitude Determination and control system
simulation works using magnetorquers, gyroscope, and on board computer. The rotational air bearing can
rotate up to fifty degrees off nominal in both the pitch and roll axes and support up to eighty kilograms.
The air bearing has three orthogonal reaction wheels for attitude control and three orthogonal torque coils
for reaction wheel desaturation or attitude control. The testbed uses inertial measurement units and built
in-house sun and Earth sensors for attitude estimation. The testbed communicates wirelessly with a
nearby ground station computer for reprogramming and data storage. The air bearing has manual as well
as automatic center of mass adjustment devices used to reduce torque due to gravity on the system
The primary objective of the air bearing structure is to provide a physical platform capable of
interfacing with the hemispherical base as well as provide attachment points and stability for the
components necessary to perform ADCS testing [23]. The foundation of the structure is the hemisphere
that oats on the cushion of compressed air. The hemisphere is the only portion of the vehicle that is in
contact with the fixed inertial air bearing support column.
The hemisphere, however, is not actually touching the support column, but rather slightly above
the column supported completely by a layer of compressed air. This support method is what gives the air
bearing the ability to rotate in all three axes with significantly reduced friction while also being
constrained to zero translational motion.

44


Figure 25: AirBearing Shape
The air bearing that supports the platform is located on top of a pedestal structure (12 in high) and
it allows the platform to move without friction 45 deg about the two horizontal axes (x and y) and 360deg
about the vertical (z) axis. The bearing is the SRA250 spherical air bearing designed and manufactured by
Specialty Components Inc. The bearing itself is made of 600 aluminum and can hold up to 748 lbf of load
when operating at 80 psi air pressure. The GIT platform bearing is operated at 30 psi which corresponds
to approximately 300 lbf of vertical load.
The aluminum platform provides a mounting surface for the several simulator subsystems. The
location of the center of mass (desired to be at the center of rotation of the simulator in this case the
center of the bearing rotor) can be changed by positioning different counterweights in various slots and
holes located at several places on the platform. Care has been taken to position all major components of
the simulator, such as momentum wheels, batteries, amplifiers, etc. in a symmetric fashion. This makes it
easy to balance the platform as well as locate the principal axes [24].
Specialty Components were able to provide us a drawing of the SRA250 spherical air bearing in
order to make sure that our testbed design will be able to carry the SRA250 and match with our
dimensions.

45


Figure 26: Specialty Components SRA250 drawing

As a backup plan in case we wont be able to by the SRA250 from Specialty Components, we
contacted another air bearing company located in NH called Nelson Air Corp and requested the drawing
for their spherical air bearing. Then, we compared both dimensions and re-dimensioned the testbed on
SolidWorks to make sure whichever air bearing the next year team decides to go with it will be able to be
mounted on this specific testbed.

46


Figure 27: Nelson Air Corp Air Bearing drawing

Due to the small flying heights typical of air bearings, external gas pressure sources must be free
of contaminants in order to allow proper operation.
Incoming air/gas should be conditioned to the following:

Absolute filtration of particles to 1 microns

Relative humidity less than 85% (non-condensing)

No oils/ resins/ wax (coalescing / charcoal filter)

System designs should include provisions to prevent accidental motion of the bearing without adequate
pressure. Pressure Interlocks and/or locking mechanisms with hard stops are prudent additions to
mechanisms which use air bearings.

47


Figure 28: Air Quality Requirements

48

7.2

Test Bed Design:

Figure 29: Test Bed Design
Cylindrical Stand:

Side Support:

Bottom Disk (Stand)

5 in diameter

5 in horizontal distance

15 in diameter

12 in height

7.5 in height

1 in thickness

1045 Carbon Steel

0.5 in thickness

Hot-Rolled Alloy Steel

1018 Carbon Steel

49

7.3

Stress & Displacement Analysis:

Stress Analysis Characteristics:

M = 6 [Kg]
Ftotal = 60 [N]
F / Peace = 15 [N]

7.3.1

Tabletop Disk (Aluminum):

Yield Strength = 27,574,200 [N/m^2]

Figure 30: Tabletop Disk Strength Analysis

50

Max Displacement = 8.784e-004 [mm]

Figure 31: Tabletop Disk Displacement
7.3.2

Air Bearing Stand (1045 Carbon Steel):

Max Displacement = 2.517e-006 [mm]

Figure 32: Air Bearing Stand Displacement Analysis

51

Yield Strength = 620,422,016 [N/m^2]

Figure 33: Air Bearing Stand Strength Analysis

52

7.4

Test Bed Material Selection

The updated test bed design for this year required materials to be selected for three key features

of the test bed stand. The parts of the test bed design that were selected to be constructed this year were
the baseplate, the support fins, and the central cylinder. Once these pieces are put together, a future team
could outline a test bed experimental section and table-top design that fits to the support stand that will be
constructed this year. Leaving the open-ended design seems to be the best idea, as it gives future teams a
starting point when they begin to design an experimental set up for a test bed. After urging from the
mechanical department, MSC Industrial Direct products were selected as the best options due to the array
of products available, pricing, and quality of the products, as well as discounts the university may be able
to arrange with the company itself.

7.4.1

Central Cylinder

Figure 34: The 1045 Carbon High Strength Steel Rod


Possibly the most important part of the test bed design is the central cylinder that supports the air
bearing and the rest of the tabletop test bed. This part needs to be able to withstand the stresses and
displacements outlined above. The highest quality steel available in the required dimensions is the 1045
High Strength Cold Finished Steel, made conveniently in the 5 diameter by 12 size that our test bed

53

design outlines. The machining of this product will be the most time consuming and delicate. The higher
quality of the steel requires greater care and more practiced machining techniques. This piece will have
slats machined into the side at the 4 places where the support fins attach to the side. The slats will be 0.5
to have the fins slide in before they are welded to the cylinder. The bottom of the cylinder will have screw
holes machined into it to allow screws to attach from the optical table, through the baseplate, to the
cylinder. The top of the cylinder will have holes machined to allow the future attachment of a spherical
air bearing, upon which the experiment portion of the test bed will sit.
7.4.2

Support Fins

Figure 35: The 1018 Carbon Steel Sheet
The fins used to support the main cylinder are necessary to reduce the stresses on the cylinder and
make the overall test bed much more stable. The quality of the material used in the construction of these
fins does not, however, need to be as high as the material used in the cylinder. The material selected is
1018 Cold Finished Steel. This product is a highly used product in industry, and holds all of the
characteristics that the test bed design outlines. This product will be purchased in 1 sheet, with the
required thickness of 0.5. That sheet will then be cut into the four different 5 by 7.5 triangular
support fins. This product has the least amount of machining required, as it is simply being cut. These fins

54

will be inserted into the slats of the cylinder and baseplate, and welded as well to the cylinder and
baseplate.
7.4.3

Test Bed Baseplate

Figure 36: The Alloy Steel HotRolled Steel Plate
The baseplate of the test bed design will need to be solid enough to support all of the normal
stresses caused by the test bed in motion, as well as need to be able to be machined. The most important
characteristic for the baseplate will be its ability to be machined to the required dimensions and still
provide the support that the test bed calls for. The baseplate will have many holes machined into it, for
fixing the cylinder to the baseplate, as well as fixing the baseplate to the optical table on which the whole
test bed is mounted. The baseplate will also have four slats machined into its top surface to allow the four
support fins to slide into before they are welded together. The slats will be 0.5 wide, and 5 in length
from the outer diameter of the baseplate towards the center of the test bed.

55

7.5

Test Bed Configuration

The Attitude Determination and Control project from last year suggested a less robust, and thus

cheaper, design for the components and actuators of the test bed. The aluminum board affixed at the top
of the test bed design will be the surface to which all test bed components are attached. These components
include control computer processors, data handling transmitter/receiver boards, sensors for acquisition,
and actuators for motion control.
The configuration and alignment of these items is crucial as the main goal of the test bed is to
balance its own center of rotation with its center of balance. Last years team researched designs of
successful test beds and suggested taking a similar approach, with a few cutbacks taken where they could
be to reduce the overall cost of creating this test bed. All of the components suggested below have been
used by either the Naval Post-Graduate Test Bed, or by the Journal of Applied Research and Technology
schematic posted in the IEEE publication.
7.5.1

Central Processing Unit

Figure 37:ADL LX8PCAMD Geode LX800 Single Board Computer
The overall goal of the test bed is to produce stable results using the type of actuators that are
going to be used on the CubeSat while controlling them with the algorithms designed to control the
CubeSat.
The test bed itself will be attempting to directly mimic the control policies that are going to be
implemented within the CubeSat. The main processor of the test bed will need to be able to command the

56

data from all of the sensors, motors, and other processors and control the system to a stable plane of
motion.
The Advanced Digital Logic LX8PC-AMD Geode LX800 Single Board Computer has been
selected to run the test bed for three main characteristics. It has the capability to run at the speeds that the
CubeSats ADCS board runs at, allowing us to prove that control of a system similar to our design for the
CubeSat can be handled through the ADCS board sub-computer, which is being imitated by this
processor in the test bed design. It is stackable in the PC/104 format, allowing easier integration into the
design of the test bed, and the part is scalable in price, allowing, if possible, a less robust design, reducing
the budget [17].
This component will also function as the power management system for the test bed. The power
supply will be connected with a serial +5V direct current connection, as well as with all power pins of the
PC/104 configuration for power management. The PC/104 format connector is shown below.

Figure 38: PC/104 Location for SBC

57

7.5.2

Wireless Communication Board

Figure 39: Eurotech COM1480 Wireless Communication Board
The design of the test bed calls for a balance of the components on the spherical air bearing. To
avoid the complicity of the wiring to all of the computer components from our data acquisition system,
wireless communication will be used. A single Eurotech COM-1480 is needed to carry out this task. The
board is PC/104 compatible, allowing the board to be easily integrated into the system with the Single
Board Computer and the power supply. This is ideal because the weight of the board will be distributed
evenly with the rest of the components, making it easier to balance the entire system. An outline of the
system is shown below. Further analysis will need to be done on this product to find a good, and yet
cheap, receiver for this system [25].

7.5.3

On-Board Communication Board

Figure 40: Accesio 104COM8SM Communication Board

58

The board selected for on board communication will need to be able to do two things. First, this
component needed to connect easily both electrically and physically to the computer systems selected for
processing. This is achieved by the 104-COM-8SMs PC/104 Bus connectivity. Second, the serial
connections of this board needed to be compatible with the systems sending it information. These systems
would be the motors controlling the balance of the system. The motors for the test bed have not been
selected due to the unknown strength of the magnetorquers, which the motors need to be on par with [26].

7.5.4

I/O Transceiver Board

Figure 41: DiamondMM32AT I/O Transceiver Board
The actuators of the test bed (the magnetorquers, yet to be selected) will need to be interfaced
with the control program through an Input/output controller. This board allows the computer system to
easily control the actuators. Connection from this board to the computer is carried out through the PC/104
Bus. The Diamond-MM-32-AT is compatible with many different types of analog and digital inputs,
allowing it to be a good choice for the control of the yet unknown actuators [27].

7.6

Power Considerations
The table below shows the power required for each component highlighted. As it can be seen, the

components all connect to the power source with a +5V direct current connection (DC).

59

Table 11: Component Power Information

Type

Power mA

LX8PCAMD Geode LX800

1480

5 DC

Eurotech COM1480

900

5 DC

Accesio 104COM8SM

80

5 DC

DiamondMM32AT I/O

410

5 DC

7.6.1

Voltage V

Battery

Figure 42: Inspired Energy ND2054 Smart Li Ion Battery
The main consideration for the battery was its ability to supply the power required evenly to the
systems. The ND2054 series of batteries from Inspired Energy provide a scalable option for power
supply. The ND2054 is rated to function at levels acceptable for our theorized test bed specifications, and
has been used by the Naval Post-Graduate Test Bed design. Another benefit of the ND2054 series is the
mass. It weighs in at just less than 0.25 kg, which keeps the overall mass of the components well below
the cap of 6 kg. The possibility of needing more than one battery exists, as the power requirements for the
motors and magnetorquers could be greater than that provided by one battery. This is not necessarily a
drawback, as the use of multiple batteries could allow easier balancing of the overall system, simplifying
the computations necessary.

60

Chapter 8 Conclusion & Recommendations

This project increased the effectiveness of the MATLAB control code, as the total time within the
required pointing increased by about 12.5% [1]. The use of a stronger magnetic torquer allowed the
control algorithm to better de-tumble the CubeSat, as the stronger magnetic moment provided by the
bigger torquer turns the spacecraft body much faster. The updated inertial matrix provided by the
Structural and Mission Analysis team was not as idealized as the previous estimate, but nonetheless still
allowed full attitude control using our specified policies. Also, a theorized set up for a test bed has been
developed, and materials selected for the planned construction of the test bed stand.

8.1

Test Bed Material Purchasing

With materials outlined as above, the decision was made for this project to suggest specific parts

and have a succinct set of parts to order available for future WPI CubeSat teams. Originally parts were
selected from the online McMaster-Carr order guide. However, after urging from the mechanical
department, MSC Industrial Direct parts should be purchased for use in the test bed design.

8.2

Attitude Simulation
The new 4+1 sun sensor setting and the nature of one-axis readings from the coarse sun sensors

have limited the readings from the sensors. As in simulation section, algorithm for sun sensor was tested
and approved to be effective in the overall control system; however, the current method of obtaining sun
sensor readings can be improved using complicated algorithm integrate the sun sensors and gyroscope
and achieve more accurate sun position estimation. The advantage of acquiring more accurate sun
position is avoiding controller overshoot the system too much which could lead to save the time for initial
attitude determination. But associated disadvantage by using complicated algorithm is that it consumes
more computing power. This is one aspect of what the future work would be.

61

Also as of future interest, the expressions for the gain selection in stabilization phase can be
investigated further to find improve the control algorithms. From previous section, system with adaptive
gains produced better results. As for the computing power consumed by calculating the adaptive gains,
depending on the selection of the expressions, for the current selection the power used is minor compare
with other calculations. So for the improvement of the stabilization phase, adaptive gains should be
investigated and determined.

62

Chapter 9 References

[1]
[2]
[3]
[4]
[5]

[6]
[7]
[8]
[9]
[10]
[11]

[12]
[13]

[14]
[15]
[16]
[17]
[18]

N.N.&.D.V.ElizabethDawson,"AttittudeDeterminationandControlSystemDesignFor
ACubeSatMission,"WPI,Worcester,2012.
D.V.Guerrant,"DesingandAnalysisofFullyMagneticControlforPicosatellite
Stabilization,"SanLuisObispo,2005.
P.W.Steyn,"MagneticAttitudeDeterminationandControlforLowEarthOrbitingSmall
Satellites,"Matieland.
Y.Yang,"QuaternionBasedModelforMomentumBiasedNadirPointingSpacecraft,"
AerospaceScienceandTechnology,pp.199202,2010.
M.B.A.B.Y.H.M.S.A.M.SiMohammed,"Simulationofmicrosatelliteattitudeusing
Kalmanfilteringinorbitresults,"SimulationModellingPracticeandTheory,pp.257277,
2008.
G.W.a.G.Bishop,"AnIntroductiontotheKalmanFilter,"UniversityofNorthCarolinaat
ChapelHill,ChapelHill,NC,2006.
M.I.Ribeiro,"KalmanandExtendedKalmanFilters:Concept,DerivationandProperties,"
2004.
N.K.Y.B.K.a.G.I.Ure,"TheDevelopmentofaSoftwareandHardwareintheLoopTest
SystemforITUPSATIINanoSatelliteADCS,"IEEE,pp.115,2011.
K.A.KOPECKY,"NumericalDifferentiation,"2007.
H.F.&.E.P.Simoncelli,"Differentiationofdiscretemultidimensionalsignal,"IEEE
Transactionsonimageprocessing.
"CoarseSunsensordatasheet,"Aeosatro,[Online].Available:
http://www.aeroastro.com/uploads/data%20sheets/Coarse%20Sun%20Sensor%20Data
%20Sheet.pdf.
"CubeSatSunSensor,"SSBV,[Online].Available:
http://www.ssbv.com/ProductDatasheets/page39/page29/index.html.
"ADXRS450ZEvaluationBoard,"AnalogDevices,[Online].Available:
http://www.analog.com/en/memssensors/memsgyroscopes/adxrs450/products/EVAL
ADXRS450Z/eb.html.
"Gyroscope,"Gyroscope.org,[Online].Available:http://www.gyroscopes.org/index.asp.
T.Bak,"SpacecraftAttitudeDeterminationaMagnetometerApproach,"Aalborg,1999.
"TripleAxisMagnetometerBreakout,"SparfunElectronics,[Online].Available:
https://www.sparkfun.com/products/10530.
"SingleBoardComputer,"EmbeddedSolution,[Online].Available:http://www.adl
usa.com/products/cpu/datapage.php?pid=ADLLX8PC.
"HighDefinitionGeomagneticModel,"NOAA,[Online].Available:
63

[19]
[20]
[21]

[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]

[35]
[36]
[37]

http://www.ngdc.noaa.gov/geomag/hdgm.shtml.
"InternationalGeomagneticReferenceField:theeleventhgeneration,"USDepartment
ofCommerce,2010.
e.StefanMaus,"TheUS/UKWorldMagneticModelfor20102015,"NOAA,Boulder,
2010.
J.D.a.D.F.B.G.A.NatansonandM.S.Challa,"MagnetometerOnlyAttitudeandRate
DeterminationforaGyroLessSpacecraft,"NationalAeronauticsandSpace
Administration,Greenbelt,Maryland.
J.T.a.Y.O.RickHarman,"GYROLESSATTITUDEANDRATEESTIMATIONGYROLESS
ATTITUDEANDRATEESTIMATION".
D.Meissner,"Athreedegreeoffreedomtestbedfornanosatelliteandcubesatattitude
dynamics,determination,andcontrol,"NavalPostgraduateSchool,2009.
P.D.W.M.CoreyWhitcombCrowell,"DevelopmentandAnalysisofaSmallSatellite
AttitudeDeterminationandControlSystemTestbed,"Boston,2011.
"COM1480,"Eurotech,[Online].Available:
http://www.eurotech.com/en/products/COM1480.
"104COM8SMFamily,"[Online].Available:http://accesio.com/go.cgi?p=../104/104
com8sm.html.
"DiamondMM32AT,"DiamondSystems,[Online].Available:
http://www.diamondsystems.com/products/diamondmm32at.
J.Prado,"Threeaxisairbearingbasedplatformforsmallsatelliteattitudedetermination
andcontrolsimulations,"IEEE,2003.
HewlettPackardCorporation,"AdvancedConfigurationandPowerInterface
Specification,Revision2.0,"2000.
"TNOcoarsesunsensor,"ESA,[Online].Available:
http://telecom.esa.int/telecom/www/object/index.cfm?fobjectid=28404.
"WilkinsonMicrowaveAnistropyProbe:Explanatorysupplement,"NASA,2003.
"Freescalesemiconductor,"[Online].Available:
http://cache.freescale.com/files/sensors/doc/data_sheet/FXOS8700CQ.pdf?fpsp=1.
"CubeSatAttitudeDeterminationandControlSystems,"ClydeSpace,[Online].Available:
http://www.clydespace.com/cubesat_shop/adcs.
"CubeSatSunSensor,"CubeSatshop,[Online].Available:
http://www.cubesatshop.com/index.php?page=shop.product_details&product_id=104&
flypage=flypage.tpl&pop=0&option=com_virtuemart&Itemid=65&vmcchk=1&Itemid=65.
"NumericalDifferentiation,"[Online].Available:
http://elm.eeng.dcu.ie/~ee317/Course_Notes/handout1.pdf.
"3AxisDigitalCompassIC,"Honeywell,[Online].Available:
http://dlnmh9ip6v2uc.cloudfront.net/datasheets/Sensors/Magneto/HMC5883LFDS.pdf.
T.H.K.Z.H.TianXiang,"DesignandOnorbitPerformanceoftheAttitude
DeterminationandControlSystemfortheZDPS1APicoSatellite,"ActaAstronautica,pp.
64

[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]

182196,2011.
W.M.A.R.F.B.ThomasW.Flatley,"ABDotAcquisitionControllerfortheRADARSAT
Spacecraft,"GoddardSpaceFlightCenter,Greenbelt,Maryland.
D.P.V.PaulTisa,"PerformanceAnalysisofControlAlgorithmsforFalconSat3,"
AmericanAstronauticalSociety,p.18,2006.
M.L.P.a.Y.Oshman,"SpacecraftAttitudeRateEstimationFromGeomagneticField
Measurements".
F.L.Markley,"AttitudeDeterminationUsingVectorObservationsandThesingularValue
Decomposition,"TheJournaloftheAstronauticalSciences,pp.245258,1988.
J.M.E.M.AwaisAbbas,"ATTITUDEDETERMINATIONANDCONTROLSUBSYSTEM
SATELLITECONTROLLER,"IEEE,p.6,2011.
L.S.LiDa,"RealTimeSequentialKalmanFilterSensorRegistrationAlgorithm,"IEEE,
2006.
E.G.L.JohnL.Crassidis,"ATTITUDEDETERMINATIONUSINGCOMBINEDGPSAND
THREEAXISMAGNETOMETERDATA".
M.I.G.A..W.B.Group,"Annex1NetFDIInflows,20002008,"MIGA,Washington,
2009.
B.J.a.B.Carpenter,"OptimalPlacementofSpacecraftSunSensorsUsingStochastic
Optimization,"inIEEEAerospace,2004.
J.M.G.AageSkullestad,"HControlofaGravityGradientStabilizedSatelite,"Control
EngineeringPractice,pp.975983,2000.
"SensorDescription,"[Online].Available:
http://www.ufa.cas.cz/html/upperatm/DSS_Description_&_protokol.pdf.

65

Appendices
Appendix A: MATLAB Simulation Code input and output guide.
1) Control
a) PD.m
i) input: time, desired quaternion and current state vector
ii) output: 3x1 magnetorquer moment vector
b) Stabilizer.m
i) Input: Current state vector and geomagnetic vector
ii) Output: 3x1 magnetorquer moment vector
2) Determination
a) EKF.m
i) Input: time and time step size
ii) Output: 7x1 state vector
b) intEKF.m
c) LowPass.m
i) Input: count number and old estimated state vector
ii) Output: new estimated state vector
d) TRIAD_est.m
i) Input: time
ii) Output: estimated state vector
3) Math
a) angle_diff.m
i) input: two column vectors x1 and x2
ii) output: angle difference in radian between the two input vectors
b) angular_error.m
i) input: sun position vector, angular error type and error angle
ii) output: point error angle
c) dcm_from_q.m
i) input: quaternion (either 4x1 or 1x4)
ii) output: 3x3 directional cosine matrix
d) delta_q.m
i) input: two column quaternions: desired and actual in form of {k,j,i,1}
ii) output: a unit vector given the difference between the quaternions
e) q_from_dcm.m
i) input: direction cosine matrix
ii) output: quaternion
f) q_to_ypr.m
i) input: a row quaternion matrix
ii) output: a row matrix containing yaw pitch and roll angles in radian
g) QXx_ypr.m
i) input: yaw pitch and roll angles in radian
ii) output: directional cosine matrix
h) Wx_to_wypr.m
i) input: angular velocity measurement matrix and 3x1 vector of yaw pitch and roll angles
ii) output: body frame velocity
i) ypr_to_q.m
i) input: 3x1 vector of yaw pitch and roll angles
ii) output: 4x1 quaternion

66

4) Models (it contains similar algorithm as in sensors, but it produces the model for each parameter)
a) model_desq.m
i) input: time
ii) output: desired quaternion of the satellite
b) model_magnet.m
i) input: time and mode (real for measurement with noise)
ii) output: real strength of magnet field vector
c) model_nadir.m
i) input: time, mode (real for measurement with noise), frame (body for satellite body frame)
and form (Quat for converted quaternion output)
ii) output: quaternion of the satellites motion relative to the nadir
d) model_sun.m
i) input: time and mode (real for measurement with noise)
ii) output: sun vector model
5) Sensors
a) sensor_gyro.m
i) input: mode (ideal for original system, real for noise corrupted system)
ii) output: real or ideal angular velocity vector
b) sensor_magnet.m
i) input: time, mode (real for measurement with noise) and frame (body for satellite body
frame)
ii) output: magnetometer measurement from pre-generated data from STK
c) sensor_sun.m
i) input: time, mode (real for measurement with noise) and frame (body for satellite body
frame)
ii) output: angular error of the sun sensor
iii) note: when the angle is out of the field of view of the fine sun sensor, the angular error
becomes larger
6) Main folder
a) inertia.m
i) input: CubeSat unit and x, y, z positions of the cg point
ii) Output: inertia tensor
b) jacobian.m
i) input: none
ii) output: Jacobian matrix

67

Appendix B: Connector Map for ADL LX8PC-AMD Geode LX800

68

Appendix C: Functional Diagram for Accesio 104-COM-8SM

69

Appendix D: Connection Layout Diagram for Diamond-MM-32-AT I/O

70