Chen - Analog and Digital Control System Design Ocr PDF

Analog (rDzqitai
Control System
Design
Trarufewfurutior ;
State ace, &Algrbri+ic
bfetlwdr
Analog and Digital
Control System Desi~n:
Transfer-Function, State-Space, and
Algebraic Methods
Chi-Tsong Chen
State University of New York at Stony Brook
Saunders College Publishing

Harcourt Brace Jovanovich College Publishers
Fort Worth Philadelphia San Diego New York Orlando Austin
San Antonio Toronto Montreal London Sydney Tokyo

Contents
Chapter 1 Introduction 1
1 .1 Empirical and Analytical Methods 1
1 .2 Control Systems 2
1.2.1 Position Control Systems 2
1 .2.2 Velocity Control Systems 4
1 .2.3 Temperature Control Systems 6
1 .2.4 Trajectory Control and Autopilot 6
1 .2.5 Miscellaneous Examples 7
1 .3 Problem Formulation and Basic Terminology 9
1.4 Scope of the Text 11
Chapter 2 Mathematical Preliminary 14

2.1 Physical Systems and Models 14
2.2 Linear Time-Invariant Lumped Systems 16
2.2.1 Mechanical Systems 17
2.2.2 RLC Networks 20
2.2.3 Industrial Process-Hydraulic tanks 22
2.3 Zero-Input Response and Zero-State Response 24
2.3.1 Zero-Input Response-Characteristic Polynomial 25

xiv CONTENTS
2.4 Zero-State Response-Transfer Function 27

2.4.1 Proper Transfer Functions 30
2.4.2 Poles and Zeros 34
2.5 Block Representation-Complete Characterization 39
2.5.1 The Loading Problem 43
2.6 State-Variable Equations 45
2.7 Solutions of State Equations-Laplace Transform Method 50
2.7.1 Time-Domain Solutions 52
2.7.2 Transfer Function and Characteristic Polynomial 54
2 .8 Discretization of State Equations 57
Problems 60
Chapter 3 Development of Block Diagrams for Control Systems 69
3.1 Introduction 69
3.2 Motors 70
3 .2.1 Field-Controlled DC Motor 70
3 .2.2 Armature-Controlled DC Motor 72
3 .2 .3 Measurement of Motor Transfer Functions 75
3 .3 Gears 78
3.4 Transducers 81
3.5 Operational Amplifiers (Op-Amps) 84
3.6 Block Diagrams of Control Systems 88
3.6.1 Reaction Wheels and Robotic Arms 91
3.7 Manipulation of Block Diagrams 94
3.7.1 Mason's Formula 98
3 .7.2 Open-Loop and Closed-Loop Transfer Functions 101
Problems 102
Chapter 4 Quantitative and Qualitative Analyses of Control Systems 111

4.1 Introduction 111
4.2 First-Order Systems-The Time Constant 111
4.2.1 Effects of Feedback 115
4.3 Second-Order Systems 116
4.4 Time Responses of Poles 123
4.5 Stability 125
CONTENTS XV
4.6 The Routh Test 129

4.6.1 Stability Range 135
4.7 Steady-State Response of Stable Systems-Polynomial Inputs 138
4.7.1 Steady-State Response of Stable Systems-Sinusoidal Inputs 141
4.7 .2 Infinite Time 144
Problems 147
Chapter 5 Computer Simulation and Realization 154

5 .2 Computer Computation of State-Variable Equations 155
5.3 Existing Computer Programs 159
5.4 Basic Block Diagrams and Op-Amp Circuits 162
5.5 Realization Problem 165
5.5.1 Realizations of
5.5 .2 Tandem and Parallel Realizations 172
5 .6 Minimal Realizations 177
5.6.1 Minimal Realization of Vector Transfer Functions 179
Problems 183
Chapter 6 Design Criteria, Constraints, and Feedback 188

6.2 Choice of a Plant 188
6.3 Performance Criteria 189
6.3.1 Steady-State Performance-Accuracy 191
6.3.2 System Types-Unity-Feedback Configuration 194
6.3.3 Transient Performance-Speed of Response 195
6.4 Noise and Disturbances 197
6.5 Proper Compensators and Well-Posedness 198
6.6 Total Stability 202
6.6.1 Imperfect Cancellations 203
6.6 .2 Design Involving Pole-Zero Cancellations 206
6.7 Saturation-Constraint on Actuating Signals 207
6.8 Open-Loop and Closed-Loop Configurations 209
6.9 Two Basic Approaches in Design 216
Problems 217

Xvi CONTENTS
Chapter 7 The Root-Locus Method 223

7.2 Quadratic Systems with a Constant Numerator 223
7.2.1 Desired Pole Region 225
7.2.2 Design using Desired Pole Region 228
7.3 More on Desired Pole Region 231
7.4 The Plot of Root Loci 233
7.4 .1 Properties of Root Loci-Phase Condition 236
7.4.2 Complexities of Root Loci 246
7.4.3 Stability Range from Root Loci-Magnitude Condition 247
7.5 Design using the Root-Locus Method 250
7 .5.1 Discussion 254
7.6 Proportional-Derivative (PD) Controller 255
7.7 Phase-Lead and Phase-Lag Networks 260
7.8 Concluding Remarks 262
Problems 263
Chapter 8 Frequency-Domain Techniques 270

8.2 Frequency-Domain Plots 271
8.3 Plotting Bode Plots 275
8.3.1 Non-Minimum-Phase Transfer Functions 284
8.3.2 Identification 286
8.4 Stability Test in the Frequency Domain 289
8.4.1 Principle of Argument 289
8.4.2 The Nyquist Plot 290
8.4.3 Nyquist Stability Criterion 294
8.4.4 Relative Stability-Gain Margin and Phase Margin 297
8.5 Frequency-Domain Specifications for Overall Systems 300
8.6 Frequency-Domain Specifications for Loop Transfer
Functions-Unity-Feedback Configuration 305
8.6.1 Why Use Bode Plots? 310
8.6.2 Design from Measured Data 311
8.7 Design on Bode Plots 312
8.7.1 Gain Adjustment 314
8.8 Phase-Lag Compensation 315

CONTENTS Xvii
8 .9 Phase-Lead Compensation 321

8 .10 Proportional-Integral (PI) Compensators 327
8 .11 Concluding Remarks 332
Problems 332
Chapter 9 The Inward Approach-Choice of Overall Transfer Functions 339

9.2 Implementable Transfer Functions 340
9.2 .1 Asymptotic Tracking and Permissible Pole-Zero
Cancellation Region 345
9.3 Various Design Criteria 346
9.4 Quadratic Performance Indices 350
9.4.1 Quadratic Optimal Systems 350
9.4.2 Computation of Spectral Factorizations 354
9.4.3 Selection of Weighting Factors 358
9.5 Three More Examples 361
9.5.1 Symmetric Root Loci 365
9.6 ITAE Optimal Systems 367
9.6.1 Applications 371
9.7 Selection Based on Engineering Judgment 375
9.8 Summary and Concluding Remarks 378
Problems 380
Chapter 10 Implementation-Linear Algebraic Method 384
10 .1Introduction 384
10.2 Unity-Feedback Configuration-Model Matching 385
10.3 Unity-Feedback Configuration-Pole Placement by Matching
Coefficients 388
10.3.1 Diophantine Equations 390
10.3 .2 Pole Placement with Robust Tracking 397
10.3 .3 Pole Placement and Model Matching 400
10.4 Two-Parameter Compensators 402
10.4 .1 Two-Parameter Configuration-Model Matching 405
10.5 Effect of Dr(s) on Disturbance Rejection and Robustness 411
10.5 .1 Model Matching and Disturbance Rejection 419

xviii CONTENTS
10.6 Plant Input/Output Feedback Configuration 422

10.7 Summary and Concluding Remarks 425
Problems 428
Chapter 11 State-Space Design 432

11 .1 Introduction 432
11 .2 Controllability and Observability 432
11 .2.1 Pole-Zero Cancellations 438
11 .3 Equivalent State-Variable Equations 440
11 .4 Pole Placement 442
11 .5 Quadratic Optimal Regulator 449
11 .6 State Estimators 453
11 .6.1 Reduced-Dimensional Estimators 456
11 .7 Connection of State Feedback and State Estimators 459
11 .7.1 Comparison with Linear Algebraic Method 461
11.8 Lyapunov Stability Theorem 465
11 .8 .1 Application-A Proof of the Routh Test 467
11 .9 Summary and Concluding Remarks 470
Problems 471
Chapter 12 Discrete-Time System Analysis 475

12.2 Why Digital Compensators? 476
12.3 A/D and D/A Conversions 478
12.4 The z-Transform 481
12.4.1 The Laplace Transform and the z-Transform 484
12.4.2 Inverse z-Transform 487
12.4.3 Time Delay and Time Advance 488
12.5 Solving LTIL Difference Equations 490
12.5 .1 Characteristic Polynomials and Transfer Functions 491
12 .5 .2 Causality and Time Delay 494
12.6 Discrete-Time State Equations 495
12.6 .1 Controllability and Observability 496
12.7 Basic Block Diagrams and Realizations 497
12 .7.1 Realizations of N(z)/D(z) 499
CONTENTS xix
12.8 Stability 500

12.8.1 The Final-Value and Initial-Value Theorems 502
12.9 Steady-State Responses of Stable Systems 503
12.9.1 Frequency Responses of Analog and Digital Systems 506
12.10 Lyapunov Stability Theorem 507
Problems 508
Chapter 13 Discrete-Time System Design 511
13 .1 Introduction 511
13 .2 Digital Implementations of Analog Compensators-Time-Domain
Invariance 512
13 .2 .1 Frequency-Domain Transformations 516
13 .3 An Example 522
13.3.1 Selection of Sampling Periods 524
13 .4 Equivalent Digital Plants 526
13.4.1 Hidden Dynamics and Non-Minimum-Phase Zeros 528
13 .5 Root-Locus Method 534
13 .6 Frequency-Domain Design 540
13 .7 State Feedback, State Estimator and Dead-Beat Design 541
13.8 Model Matching 544
13 .9 Concluding Remarks 547
Problems 548
Chapter 14 PID Controllers 551

14.2 PID Controllers in Industrial Processes 552
14.2.1 Rules of Ziegler and Nichols 558
14.2.2 Rational Approximations of Time Delays 559
14.3 PID Controllers for Linear Time-Invariant Lumped Systems 561
14.4 Digital PID Controllers 563
Appendix A The Laplace Transform 567
A.1 Definition 567

A.2 Inverse Laplace Transform-Partial Fraction Expansion 570
A .3 Some Properties of the Laplace Transform 571

xx CONTENTS
A .4 Solving LTIL Differential Equations 573

A.5 Time Delay 577
Appendix B Linear Algebraic Equations 579
B.! Matrices 579

B .2 Determinant and Inverse 580
B.3 The Rank of Matrices 582
B .4 Linear Algebraic Equations 584
B.5 Elimination and Substitution 586
B .6 Gaussian Elimination with Partial Pivoting 588
References 590
Index 595
Introduction
1 .1 EMPIRICAL AND ANALYTICAL METHODS
The ultimate goal of engineering-in particular, that of control engineering-is to

design and build real physical systems to perform given tasks . For example, an
engineer might be asked to design and install a heat exchanger to control the tem-
perature and humidity of a large building . This has been a longstanding problem in
engineering, and much relevant data has been collected . From the total volume and
geographical location of the building, we can determine the required capacity of the
exchanger and then proceed to install the system . If, after installation, the exchanger
is found to be insufficiently powerful to control the building's environment, it can
be replaced by a more powerful one . This approach, which relies heavily on past
experience and repeated experimentation, is called the empirical method. Although
the empirical method must be carried out by trial and error, it has been used suc-
cessfully to design many physical systems .
The empirical method, however, is inadequate if there is no past experience to
draw from or if experimentation is not feasible because of high cost or risk . For
example, the task of sending astronauts to the moon and bringing them back safely
could not have been carried out by the empirical method . Similarly, the design of
fusion control in nuclear power plants should not be so dealt with . In these cases,
the analytical method becomes indispensable. The analytical method generally con-
sists of four steps : modeling, setting up mathematical equations, analysis, and design .
The first two steps are closely related . If we use simple mathematics, then the model
chosen must be correspondingly simple . If we use sophisticated mathematics, then
1
2 CHAPTER 7 INTRODUCTION
the model can be more complex and realistic . Modeling is the most critical step in
analytical design . If a physical system is incorrectly modeled, subsequent study will
be useless . Once a model is chosen, the rest of the analytical design is essentially a
mathematical problem .
Repeated experimentation is indispensable in the empirical method . It is also
important in the analytical method . In the former, experiments must be carried out
using physical devices, which might be expensive and dangerous . In the latter, how-
ever, experiments can be carried out using models or mathematical equations . Many
computer-aided design packages are available . We may use any of them to simulate
the equations on a digital computer, to carry out design, and to test the result on the
computer. If the result is not satisfactory, we repeat the design . Only after a design
is found to be satisfactory, will we implement it using physical devices .
If the model is adequately chosen, the performance of the implemented system
should resemble the performance predicted by analytical design or computer simu-
lation . However, because of unavoidable inaccuracy in modeling, discrepancies often
exist between the performance of the implemented physical system and that predicted
by the analytical method . Therefore, the performance of physical systems can often
be improved by fine adjustments or tunings . This is why a physical system often
requires lengthy testing after it is implemented, before it is put into actual operation
or mass production . In this sense, experience is also important in the analytical
method .
In the analytical approach, experimentation is needed to set up models, and
experience is needed (due to the inaccuracy of modeling) to improve the performance
of actual physical systems . Thus, experience and experimentation are both used in
the empirical and analytical approaches . The major difference between these two
approaches is that in the latter, we gain, through modeling, understanding and insight
into the structure of systems . The analytical approach also provides systematic pro-
cedures for designing systems and reduces the likelihood of designing flawed or
disastrous systems .
In this text, we study analytical methods in the analysis and design of control
systems.
1 .2 CONTROL SYSTEMS
This text is concerned with the analysis and design of control systems ; therefore, it
is pertinent to discuss first what control systems are . Before giving a formal defini-
tion, we discuss a number of examples .
1 .2 .1 Position Control Systems

The satellite dish in the backyard or on the rooftop of a house has become common
in recent years. It is an antenna aimed at a satellite that is stationary with respect to
the earth and is used to transmit television or other signals . To increase the number
of channels for a television, the dish may be designed to aim at different satellites .

1 .2 CONTROL SYSTEMS 3
A possible arrangement of such a system is shown in Figure 1 .1(a). This system can
indeed be designed using the empirical method . If it is to be designed using the
analytical method, we must first develop a model for the system, as shown in Figure
1 .1(b). The model actually consists of a number of blocks .' Each block represents
a model of a physical device . Using this model, we can then carry out the design .
A large number of systems can be similarly modeled . For example, the system that
aims the antennas shown in Figure 1 .2 at communication satellites and the systems
that control various antennas and solar panels shown in Figure 1 .3 can be similarly
modeled. These types of systems are called position control systems .
There are other types of position control systems . Consider the simplified nu-
clear power plant shown in Figure 1 .4 . The intensity of the reaction inside the reactor
(and, consequently, the amount of heat generated) ; is controlled by the vertical po-
sition of the control rods . The more deeply the control rods are submerged, the more
heat the reactor will generate . There are many other control systems in the nuclear
power plant. Maintenance of the boiler's water level and maintenance or regulation
of the generated voltage at a fixed voltage all call for control systems . Position control
is also needed in the numerical control of machine tools . For example, it is possible
to program a machine tool so that it will automatically drill a number of holes, as
shown in Figure 1 .5.
Control
knob
Motor and load

e el U km 0

ki
~
Potentiometer
0 s(ms+1)
Potentiometer
(b)
Figure 1 .1 Position control system .
'At this point, the reader need not be concerned with the equation inside each block . It will be developed
in Chapter 3 .
Figure 1 .2 Radar antenna. (Courtesy of MIT Lincoln Laboratory.)
1 .2.2 Velocity Control Systems

Driving tapes in video or audio recorders at a constant speed is important in pro-
ducing quality output . The problem is complicated because the load of the driver
varies from a full reel to an empty reel . A possible control system to achieve this is
shown in Figure 1 .6(a). This system can indeed be designed using the empirical
method . If it is to be designed using the analytical method, we must develop a model,
as shown in Figure 1.6(b). Using this model, we can then carry out the design .
Velocity control problems also arise in a large number of industrial applications . To
"grow" an optical fiber with a uniform diameter from melted glass, the speed of
growth must be properly controlled. The speed of the conveyor in a production line
must also be precisely controlled . An error in roller speed of just 0 .1% in a paper
drive system may cause the paper to tear or to pile up on the roller . The velocity of
the rotor of the generator in Figure 1 .4 is kept constant in order to generate a constant
voltage. Velocity control is indeed needed in a wide range of applications .
Dual subreflectors
2 .2-m, 30-GHz
receiving antenna
Housing for signal

regenerator and
baseband switch
Solar array
Figure 1 .3 Communications satellite . (Courtesy of IEEE Spectrum.)
Control F
Steam
Control A1 turbine Generato
rod Voltage
Neutron
detector
0
Pump
I Condenser
Figure 1 .4 Nuclear power plant .
Figure 1 .5 Numerical control system .

5

r dc Motor Tachometer
EWE
Reference signal Differential amplifier Motor and load
I Actua speed
Desired speed I + I U km
kl
I
o-~ k3
ms+l
I
Potentiometer _
k, c
Tachometer
(b)
Figure 1 .6 Velocity control system .
1 .2.3 Temperature Control Systems

We now discuss a different type of example . Consider the temperature control of
the enclosed chamber shown in Figure 1 .7(a). This problem, which arises in the
temperature control of an oven, a refrigerator, an automobile compartment, a house,
or the living quarters of a space shuttle, can certainly be approached by using the
empirical method . If the analytical method is to be used, we must develop a model
as shown in Figure 1 .7(b). We can then use the model to carry out analysis and
design .
Temperature control is also important in chemical processes . The rate of chem-
ical reactions often depends on the temperature . If the temperature is not properly
controlled, the entire product may become useless . In industrial processes, temper-
ature, pressure and flow controls are widely used .
1 .2.4 Trajectory Control and Autopilot

The landing of a space shuttle on a runway is a complicated control problem . The
desired trajectory is first computed as shown in Figure 1 .8 . The task is then to bring
the space shuttle to follow the desired trajectory as closely as possible . The structure
of a shuttle is less stable than that of an aircraft, and its landing speed cannot be
controlled . Thus, the landing of a space shuttle is considerably more complicated
than that of an aircraft . The landing has been successfully accomplished with the
aid of on-board computers and altimeters and of the sensing devices on the ground,
as shown in Figure 1 .8 . In fact, it can even be achieved automatically, without
involving astronauts . This is made possible by the use of an autopilot . The autopilot
is now widely used on aircrafts and ships to maintain desired altitude and/or heading .

1 .2 CONTROL SYSTEMS 7
temperature
Actual temperature
Power supply I
(a)
Control device Switch

e 9
Chamber
Potentiometer
e2
3
Thermocouple and Amplifier

(b)
Figure 1 .7 Temperature control system .
Shuttle
Sensing device
Runway
Figure 1 .8 Desired landing trajectory of space shuttle .
1 .2.5 Miscellaneous Examples

We give two more examples of control systems to conclude this section . Consider
the bathroom toilet tank shown in Figure 1 .9(a). The mechanism is designed to close
the valve automatically whenever the water level reaches a preset height . A sche-
matic diagram of the system is shown in Figure 1 .9(b). The float translates the water
level into valve position. This is a very simple control problem . Once the mechanism

Water
(a)
Water
Preset or
desired
water level ~ I
Controller: I _ I Valve
Actual
water level
floa ~t
H H Water tank
(b)
Figure 1 .9 Bathroom toilet tank.
of controlling the valve is understood, the water level can easily be controlled by
trial and error.
As a final example, a schematic diagram of the control of clothes dryers is shown
in Figure 1 .10 . Presently there are two types of clothes dryers-manual and auto-
matic . In a manual clothes dryer, depending on the amount of clothes and depending
Electricity
Desired Actual
dryness dryness
Experience
(a)
Desired Actual
dryness dryness
Transducer Comparator Switch Dryer
)I I
Sensing devi e
(b)
Figure 1 .10 (a) Manual clothes dryer . (b) Automatic dryer .
1 .3 PROBLEM FORMULATION AND BASIC TERMINOLOGY 9
on experience, we set the timer to, say, 40 minutes . At the end of 40 minutes, this
dryer will automatically turn off even if the clothes are still damp or wet . Its sche-
matic diagram is shown in Figure 1 .10(a) . In an automatic dryer, we select a desired
degree of dryness, and the dryer will automatically turn off when the clothes reach
the desired degree of dryness . If the load is small, it will take less time ; if the load
is large, it will take more time . The amount of time needed is automatically deter-
mined by this type of dryer . Its schematic diagram is shown in Figure 1 :10(b) .
Clearly, the automatic dryer is more convenient to use than a manual one, but it is
more expensive . However, in using a manual dryer, we may overset the timer, and
electricity may be wasted . Therefore, if we include the energy saved, an automatic
dryer may turn out to be more economical .
1 .3 PROBLEM FORMULATION AND BASIC TERMINOLOGY
From the examples in the preceding section, we may deduce that a control system
is an interconnection of components or devices so that the output of the overall
system will follow as closely as possible a desired signal . There are many reasons
to design control systems :
1. Automatic control: The temperature of a house can be automatically main-
tained once we set a desired temperature . This is an automatic control system .
Automatic control systems are used widely and are essential in automation in
industry and manufacturing .
2. Remote control : The quality of reception of a TV channel can be improved by
pointing the antenna toward the emitting station . If the antenna is located at the
rooftop, it is impractical to change its direction by hand . If we install an antenna
rotator, then we can control the direction remotely by turning a knob sitting in
front of the TV . This is much more convenient . The Hubble space telescope,
which is orbiting over three hundred miles above the earth, is controlled from
the earth . This remote control must be done by control systems .
3. Power amplification : The antennas used to receive signals sent by Voyager 2
have diameters over 70 meters and weights over several tons . Clearly, it is
impossible to turn these antennas directly by hand . However, using control sys-
tems, we can control them by turning knobs or by typing in command signals
on computers . The control systems will then generate sufficient power to turn
the antennas . Thus, power amplification is often implicit in many control
systems.
In conclusion, control systems are widely used in practice because they can be
designed to achieve automatic control, remote control, and power amplification .
We now formulate the control problem in the following . Consider the the po-
sition control problem in Figure 1 .1, where the objective is to control the direction
of the antenna. The first step in the design is to choose a motor to drive the antenna .
The motor is called an actuator . The combination of the object to be controlled and
the actuator is called the plant. In a home heating system, the air inside the home is

r
I
I I
I I
I I
r(t) u (t) Object I y(t)
Actuator to be
Reference I Actuating controlled I Plant
signal I signal I output
I
I Plant
I I
L J
Figure 1 .11 Control design problem .
the controlled object and the burner is the actuator . A space shuttle is a controlled
object ; its actuator consists of a number of thrustors . The input of the plant, denoted
by u(t), is called the control signal or actuating signal; the output of the plant,
denoted by y(t), is called the controlled variable or plant output . The problem is to
design an overall system as shown in Figure 1 .11 so that the plant output will follow
as closely as possible a desired or reference signal, denoted by r(t) . Every example
in the preceding section can be so formulated .
There are basically two types of control systems : the open-loop system and the
closed-loop or feedback system . In an open-loop system, the actuating signal is
predetermined by the desired or reference signal ; it does not depend on the actual
plant output. For example, based on experience, we set the timer of the dryer in
Figure 1 .10(a). When the time is up, the dryer will stop even if the clothes are still
damp or wet. This is an open-loop system . The actuating signal of an open-loop
system can be expressed as
u(t) = f(r(t))
where f is some function . If the actuating signal depends on the reference input and
the plant output, or if it can be expressed as
u(t) = h(r(t), y(t))
where h is some function, then the system is a closed-loop or feedback system. All
systems in the preceding section, except the one in Figure 1 .10(a), are feedback
systems. In every feedback system the plant output must be measured and used to
generate the actuating signal . The plant output could be a position, velocity, tem-
perature, or something else . In many applications, it is transformed into a voltage
and compared with the reference signal, as shown in Figure 1 .12. In these transfor-
mations, sensing devices or transducers are needed as shown . The result of the
comparison is then used to drive a compensator or controller . The output of the
controller yields an actuating signal . If the controller is designed properly, the ac-
tuating signal will drive the plant output to follow the desired signal .
In addition to the engineering problems discussed in the preceding section, a
large number of other types of systems can also be considered as control systems .
1 .4 SCOPE OF THE TEXT 11
Figure 1 .12 Feedback control system .
Our body is in fact a very complex feedback control system . Maintaining our body
temperature at 37 Celsius requires perspiration in summer and contraction of blood
vessels in winter . Maintaining an automobile in a lane (plant output) is a feedback
control system: Our eyes sense the road (reference signal), we are the controller, and
the plant is the automobile together with its engine and steering system . An economic
system is a control system . Its health is measured by the gross national product
(GNP), unemployment rate, average hourly wage, and inflation rate . If the inflation
rate is too high or the unemployment rate is not acceptable, economic policy must
be modified . This is achieved by changing interest rates, monetary policy, and gov-
ernment spending . The economic system has a large number of interrelated factors
whose cause-and-effect relationships are not exactly known . Furthermore, there are
many uncertainties, such as consumer spending, labor disputes, or international
crises . Therefore, an economic system is a very complex system . We do not intend
to solve every control problem ; we study in this text only a very limited class of
control problems .
1 .4 SCOPE OF THE TEXT
This text is concerned with the analysis and design of control systems . As it is an
introductory text, we study only a special class of control systems . Every system-
in particular, every control system-is classified dichotomously as linear or nonlin-
ear, time-invariant or time-varying, lumped or distributed, continuous-time or
discrete-time, deterministic or stochastic, and single-variable or multivariable .
Roughly speaking, a system is linear if it satisfies the additivity and homogeneity
properties, time-invariant if its characteristics do not change with time, and lumped
if it has a finite number of state variables or a finite number of initial conditions that
can summarize the effect of past input on future output . A system is continuous-time
if its responses are defined for all time, discrete-time if its responses are defined only
at discrete instants of time . A system is deterministic if its mathematical description
does not involve probability. It is called a single-variable system if it has only one
input and only one output ; otherwise it is called a multivariable system. For a more
detailed discussion of these concepts, see References [15, 18] . In this text, we study
only linear, time-invariant, lumped, deterministic, single-variable systems. Although
this class of control systems is very limited, it is the most important one . Its study
is a prerequisite for studying more general systems . Both continuous-time and
discrete-time systems are studied .
The class of systems studied in this text can be described by ordinary differential
equations with real constant coefficients . This is demonstrated in Chapter 2 by using
examples. We then discuss the zero-input response and the zero-state response . The
transfer function is developed to describe the zero-state response . Since the transfer
function describes only the zero-state response, its use in analysis and design must
be justified . This is done by introducing the concept of complete characterization .
The concepts of properness, poles, and zeros of transfer functions are also intro-
duced. Finally, we introduce the state-variable equation and its discretization . Its
relationship with the transfer function is also established .
In Chapter 3, we introduce some control components, their models, and their
transfer functions . The loading problem is considered in developing the transfer
functions . Electrical, mechanical, and electromechanical systems are discussed . We
then discuss the manipulation of block diagrams and Mason's formula to conclude
the chapter .
The quantitative and qualitative analyses of control systems are studied in Chap-
ter 4 . Quantitative analysis is concerned with the response of systems due to some
specific input, whereas qualitative analysis is concerned with general properties of
systems . In quantitative analysis, we also show by examples the need for using
feedback and tachometer feedback . The concept of the time constant is introduced .
In qualitative analysis, we introduce the concept of stability, its condition, and a
method (the Routh test) of checking it. The problems of pole-zero cancellation and
complete characterization are also discussed.
In Chapter 5, we discuss digital and analog computer simulations . We show
that if the state-variable description of a system is available, then the system can be
readily simulated on a digital computer or built using operational amplifier circuits .
Because it is simpler and more systematic to simulate transfer functions through
state-variable equations, we introduce the realization problem-the problem of ob-
taining state-variable equations from transfer functions . Minimal realizations of vec-
tor transfer functions are discussed . The use of MATLAB, a commercially available
computer-aided design package, is discussed throughout the chapter .
Chapters 2 through 5 are concerned with modeling and analysis problems ; the
remaining chapters are concerned with the design problem . In Chapter 6, we discuss
the choice of plants . We then discuss physical constraints in the design of control
systems . These constraints lead to the concepts of well-posedness and total stability.
The saturation problem is also discussed . Finally, we compare the merits of open-
loop and closed-loop systems and then introduce two basic approaches-namely,
outward and inward-in the design of control systems . In the outward approach,
we first choose a configuration and a compensator with open parameters and then
adjust the parameters so that the resulting overall system will (we hope) meet the
design objective. In the inward approach, we first choose an overall system to meet
the design objective and then compute the required compensators .
Two methods are available in the outward approach : the root-locus method and
the frequency-domain method. They were developed respectively in the 1950s and

1 .4 SCOPE OF THE TEXT 13
1940s . The root-locus method is introduced in Chapter 7 and the frequency-domain

method in Chapter 8 . Both methods are trial-and-error methods .
The inward approach consists of two parts : the search for an overall transfer
function to meet design specifications and the implementation of that overall transfer
function. The first problem is discussed in Chapter 9, where overall systems are
chosen to minimize the quadratic performance index and the ITAE (integral of time
multiplied by absolute error) . It is also shown by examples that good overall transfer
functions can also. b e obtained by computer simulations. The implementation prob-
lem is discussed in Chapter 10, where the difference between model matching and
pole placement is also discussed. We discuss the implementation in the unity-feed-
back configuration, two-parameter configuration, and the plant input/output feed-
back configuration. We also discuss how to increase the degree of compensators to
achieve robust tracking and disturbance rejection . The design methods in Chapter
10 are called the linear algebraic method, because they are all achieved by solving
linear algebraic equations .
The design methods in Chapters 6 through 10 use transfer functions . In Chapter
11, we discuss design methods using state-variable equations . We introduce first the
concepts of controllability and observability and their conditions . Their relationships
with pole-zero cancellations are also discussed . We then use a network to illustrate
the concept of equivalent state-variable equations . Pole placement is then carried
out by using equivalent equations . The same procedure is also used to design full-
dimensional state estimators. Reduced-dimensional estimators are then designed
by solving Lyapunov equations. The connection of state feedback to the output of
state estimators is justified by establishing the separation property . Finally, we
compare the design of state feedback and state estimator with the linear algebraic
method.
Chapters 3 through 11 study continuous-time control systems . The next two
chapters study discrete-time counterparts . Chapter 12 first discusses the reasons for
using digital compensators to control analog plants and then discusses the interfaces
needed to connect analog and digital systems . We introduce the z-transform, differ-
ence equations, state-variable equations, stability, and the Jury test. These are the
discrete-time counterparts of the continuous-time case . The relationship between the
frequency response of analog and digital transfer functions is also discussed . Chapter
13 discusses two approaches in designing digital compensators . The first approach
is to design an analog compensator and then discretize it . Six different discretization
methods are introduced . The second approach is to discretize the analog plant into
an equivalent digital plant and then design digital compensators . All analog methods,
except the frequency-domain method, are directly applicable to design digital com-
pensators without any modification .
If a plant can be modeled as linear, time-invariant, and lumped, then a good
control system can be designed by using one of the methods discussed in this text .
However, many plants, especially industrial processes, cannot be so modeled . PID
controllers may be used to control these plants, as discussed in Chapter 14 . Various
problems in using PID controllers are discussed. The Laplace transform and linear
algebraic equations are reviewed in Appendices A and B : this discussion is not
exhaustive, going only to the extent needed in this text .
Mathematical
Preliminary
2 .1 PHYSICAL SYSTEMS AND MODELS
This text is concerned with analytical study of control systems . Roughly speaking,
it consists of four parts :
1. Modeling
2. Development of mathematical equations
3. Analysis
4. Design
This chapter discusses the first two parts . The distinction between physical systems
and models is fundamental in engineering . In fact, the circuits and control systems
studied in most texts are models of physical systems . For example, a resistor with a
constant resistance is a model ; the power limitation of the resistor is often disre-
garded . An inductor with a constant inductance is also a model ; in reality, the in-
ductance may vary with the amount of current flowing through it . An operational
amplifier is a fairly complicated device ; it can be modeled, however, as shown in
Figure 2 .1 . In mechanical engineering, an automobile suspension system may be
modeled as shown in Figure 2.2. In bioengineering, a human arm may be modeled
as shown in Figure 2.3(b) or, more realistically, as in Figure 2.3(c). Modeling is an
extremely important problem, because the success of a design depends upon whether
or not physical systems are adequately modeled .
Depending on the questions asked and depending on operational ranges, a phys-
ical system may have different models . For example, an electronic amplifier has
14

2 .1 PHYSICAL SYSTEMS AND MODELS 15
vo =A(v2 - v )
Figure 2 .1 Model of operational amplifier .
Figure 2.2 Model of automobile suspension system .
Arm W W W
(a) (b) (c)
Figure 2 .3 Models of arm .
different models at high and low frequencies . A spaceship may be modeled as a

particle in the study of trajectory ; however, it must be modeled as a rigid body in
the study of maneuvering . In order to develop a suitable model for a physical system,
we must understand thoroughly the physical system and its operational range . In this
text, models of physical systems are also called systems. Hence, a physical system
is a device or a collection of devices existing in the real world ; a system is a model
of a physical system . As shown in Figure 2 .4, a system is represented by a unidi-
rectional block with at least one input terminal and one output terminal . We remark
that terminal does not necessarily mean a physical terminal, such as a wire sticking
out of the block, but merely indicates that a signal may be applied or measured from
that point. If an excitation or input signal u(t) is applied to the input terminal of a

16 CHAPTER 2 MATHEMATICAL PRELIMINARY
u y
System
Figure 2.4 System .
system, a unique response or output signal y(t) will be measurable or observable at

the output terminal . This unique relationship between excitation and response, input
and output, or cause and effect is implicit for every physical system and its model .
A system is called a single-variable system if it has only one input terminal and
only one output terminal . Otherwise, it is called a multivariable system . A multi-
variable system has two or more input terminals and/or two or more output terminals .
We study in this text mainly single-variable systems .
2 .2 LINEAR TIME-INVARIANT LUMPED SYSTEMS
The choice of a model for a physical device depends heavily on the mathematics to
be used. It is useless to choose a model that closely resembles the physical device
but cannot be analyzed using existing mathematical methods . It is also useless to
choose a model that can be analyzed easily but does not resemble the physical device .
Therefore, the choice of models is not a simple task. It is often accomplished by a
compromise between ease of analysis and resemblance to real physical systems .
The systems to be used in this text will be limited to those that can be described
by ordinary linear differential equations with constant real coefficients such as
3 d2y(t) + 2 dy(t) du(t) - 3u(t)
dt dt + y(t) = 2 dt
or, more generally,
d o-1 y(t) dy(t)
an dny(t) + an-1 + . . . + a1
dtn dtn-1 dt + a0y(t)
dmu(t) dm-1 u(t) du(t)
= bm + bm 1 dtm - 1 + . . . + bl
dtm dt + b0u(t)
where a; and b; are real constants, and n ? m. Such equations are called nth order
linear time-invariant lumped (LTIL) differential equations . In order to be describable
by such an equation, the system must be linear, time-invariant, and lumped . Roughly
speaking, a system is linear if it meets the additivity property [that is, the response
of u 1(t) + u2(t) equals the sum of the response of u 1 (t) and the response of u 2(t)],
and the homogeneity property [the response of au(t) equals a times the response of
u(t)] . A system is time-invariant if its characteristics-such as mass or moment of
inertia for mechanical systems, or resistance, inductance or capacitance for electrical
systems-do not change with time . A system is lumped if the effect of any past
input u(t), for t < t0 , on future output y(t), for t ? t0, can be summarized by afinite
number of initial conditions at t = t0. For a detailed discussion of these concepts,

2 .2 LINEAR TIME-INVARIANT LUMPED SYSTEMS 17
Friction Spring force

Viscous
Static -
i
k2
000 m h u
I dy/dt A
B'
i
kI ~y
Coulomb
(a) (b) (c)

Figure 2 .5 Mechanical system .
see References [15, 181 . We now discuss how these equations are developed to
describe physical systems .
2.2.1 Mechanical Systems

Consider the system shown in Figure 2.5(a). It consists of a block with mass m
connected to a wall by a spring . The input is the applied force u(t), and the output
is the displacement y(t) measured from the equilibrium position. Before developing
an equation to describe the system, we first discuss the characteristics of the friction
and spring . The friction between the block and the floor is very complex . It generally
consists of three parts-static, Coulomb, and viscous frictions-as shown in Figure
2.5(b) . Note that the coordinates are friction versus velocity . When the mass is
stationary or its velocity is zero, we need a certain amount of force to overcome the
static friction to start its movement . Once the mass is moving, there is a constant
friction, called the Coulomb friction, which is independent of velocity . The viscous
friction is generally modeled as
Viscous friction = kl X Velocity (2.2)
where k l is called the viscous friction coefficient . This is a linear equation . Most
texts on general physics discuss only static and Coulomb frictions . In this text,
however, we consider only viscous friction ; static and Coulomb frictions will be
disregarded . By so doing, we can model the friction as a linear phenomenon .
In general physics, Hooke's law states that the displacement of a spring is pro-
portional to the applied force, that is
Spring force = k2 X Displacement (2.3)
where k2 is called the spring constant. This equation is plotted in Figure 2 .5(c) with
the dotted line. It implies that no matter how large the applied force is, the displace-
ment equals force/k 2. This certainly cannot be true in reality ; if the applied force is
larger than the elastic limit, the spring will break . In general, the characteristic of a
physical spring has the form of the solid line shown in Figure 2.5(c).' We see that
'This is obtained by measurements under the assumption that the mass of the spring is zero and that the
spring has no drifting and no hysteresis . See Reference [18] .

if the applied force is outside the range [A', B'], the characteristic is quite different
from the dotted line . However, if the applied force lies inside the range [A', B'],
called the linear operational range, then the characteristic can very well be repre-
sented by (2.3). We shall use (2.3) as a model for the spring .
We now develop an equation to describe the system by using (2.3) and consid-
ering only the viscous friction in (2.2) . The applied force u(t) must overcome the
friction and the spring force, and the remainder is used to accelerate the mass . Thus
we have
d y(t)
u(t) - k1 dy(t) - kzy(t) = m zdt2
dt
or
ddzt)
m dt + k, dy(t)
dt + k2y(t) = u(t)
This is an ordinary linear differential equation with constant coefficients . It is im-
portant to remember that this equation is obtained by using the linearized relation in
(2.3) and considering only the viscous friction in (2 .2). Therefore, it is applicable
only for a limited operational range .
Consider now the rotational system shown in Figure 2.6(a). The input is the
applied torque T(t) and the output is the angular displacement 0(t) of the load . The
shaft is not rigid and is modeled by a torsional spring . Let J be the moment of inertia
of the load and the shaft . The friction between the shaft and bearing may consist of
static, Coulomb, and viscous frictions . As in Figure 2 .5, we consider only the viscous
friction . Let k1 be the viscous friction coefficient and k2 the torsional spring constant .
Then the torque generated by the friction equals k1dO(t)/dt and the torque generated
by the spring is k20(t) . The applied torque T(t) must overcome the friction and spring
torques; the remainder is used to accelerate the load . Thus we have
20( t )
T(t) - k1 - k20(t) = J
d dtt) ddtz
or
J
d zdtzt)
+ k1 d dtt) + k20(t) = T(t) (2.5a)
This differential equation describes the system in Figure 2.6(a) .
Torque T(t) T(t)

Bearing Bearing
t~
B(t) ) B(t)
Spring /7777 l=
Load
(a) (b)
Figure 2 .6 Rotational mechanical system .

2.2 LINEAR TIME-INVARIANT LUMPED SYSTEMS 19
If we identify the following equivalences:
Translational movement Rotational movement

Linear displacement y -- Angular displacement B
Force u <-- Torque T
Mass m -- Moment of inertia J
then Equation (2 .5a) is identical to (2 .4). The former describes a rotational move-
ment, the latter, a linear or translational movement .
Exercise 2 .2.1
The suspension system of an automobile can be modeled as shown in Figure 2.7.

This model is simpler than the one in Figure 2 .2, because it neglects the wheel mass
and combines the tire stiffness with the spring . The model consists of one spring
with spring constant k2 and one dashpot or shock absorber . The dashpot is a device
that provides viscous frictional force . Let k, be its viscous friction coefficient and
let m be the mass of the car . A vertical force u(t) is applied to the mass when the
wheel hits a pothole . Develop a differential equation to describe the system .
Figure 2 .7 Suspension system of automobile .
[Answer : Same as (2.4).]
Exercise 2 .2.2
Show that the system shown in Figure 2 .6(b) where the shaft is assumed to be rigid
is described by
d2B(t) dO(t)
J + k = T(t) (2 .5b)
dt2 1 dt

+~
v(t) C v( t) L
O
o
I di(t)
v(t)=L dt
Figure 2 .8 Electrical components .
2 .2 .2 RLC Networks
We discuss in this section circuits that are built by interconnecting resistors, capac-
itors, inductors, and current and voltage sources, beginning with the three basic
elements shown in Figure 2 .8. A resistor is generally modeled as v = Ri, where R
is the resistance, v is the applied voltage, and i is current flowing through it with
polarity chosen as shown . In the model v = Ri, nothing is said regarding power
consumption . In reality, if v is larger than a certain value, the resistor will burn out .
Therefore, the model is valid only within the specified power limitation . A capacitor
is generally modeled as Q = Cv, where C is the capacitance, v is the applied voltage,
and Q is the charge stored in the capacitor . The model implies that as the voltage
increases to infinity, so does the stored charge . Physically this is not possible . As v
increases, the stored charge will saturate and cease to increase, as shown in Figure
2.9 . However, for v in a limited range, the model Q = Cv does represent the physical
capacitor satisfactorily . An inductor is generally modeled as 0 = Li where L is the
inductance, 0 is the flux, and i is the current . In reality, the flux generated in an
inductor will saturate as i increases . Therefore the relationship (A = Li is again
applicable only within a limited range of i . Now if R, L, and C change with time,
they are time-varying elements . If R, L, and C are constants, independent of time,
then they are time-invariant elements . Using these linear time-invariant models, we
can express their voltages and currents as
v(t) = Ri(t) (2 .6a)
dv(t)
i(t) = C (2 .6b)
dt
v(t) = L ddt) (2 .6c)
Now we shall use (2 .6) to develop differential equations to describe RLC net-
works . Consider the network shown in Figure 2.10 . The input is a current source
u(t) and the output y(t) is the voltage across the capacitor as shown . The current of
the capacitor, using (2.6b), is
dy(t) _ 2 dy(t)
i (t) = C dt dt

Charge (flux)
Figure 2.9 Characteristics of capacitor and inductor .
This current also passes through the 1-a resistor . Thus the voltage drop across A
and B is
dd(tt) + y(t)
u,B = i,(t) 1 + y(t) = 2
The current i l (t) passing through the 0 .5-fl resistor is
i,(t) dtt) + 2y(t)

05 = 4
Thus we have
u(t) = i 1 (t) + i,(t) = 4 dy(t) + 2y(t) + 2 dy(t)

dt dt
or
ddt) + 2y(t) = u(t) (2.7)

6
This first-order differential equation describes the network in Figure 2.10 .
u(t) 0 .54 2F
B
Figure 2 .10 RC network .

Exercise 2 .2 .3
Find differential equations to describe the networks in Figure 2 .11 . The network in
Figure 2 .11(b) is called a phase-lag network.
R
I14
(a) (b)
Figure 2 .11 Networks .
[Answers : (a) LC d 2 y(t)/dt 2 + RC dy(t)/dt + y(t) = u(t).

(b) C(R 1 + R 2 )dy(t)/dt + y(t) = CR 2 du(t)/dt + u(t).]
2.2 .3 Industrial Process-Hydraulic Tanks

In chemical plants, it is often necessary to maintain the levels of liquids . A simplified
model of such a system is shown in Figure 2 .12, in which
q;, q1, q2 rates of the flow of liquid
A1 , A2 areas of the cross section of tanks
h l , h 2 = liquid levels
R1 , R2 flow resistance, controlled by valves
It is assumed that q1 and q2 are governed by
hl h2
qt = and q2 = (2.8)
R-1 h2 R2
q,
q qZ
Figure 2 .12 Control of liquid levels .

They are proportional to relative liquid levels and inversely proportional to flow
resistances . The changes of liquid levels are governed by
A,dh, = (q; - q,)dt
and
A2dh2 = (q, - q2)dt
which imply
dh, _
(2 .9a)
A1 dt = g` - q,
and
dh2 _ (2 .9b)
A2 dt - q, - q2
These equations are obtained by linearization and approximation . In reality, the flow
of liquid is very complex ; it may involve turbulent flow, which cannot be described
by linear differential equations . To simplify analysis, turbulent flow is disregarded
in developing (2 .9). Let q; and q2 be the input and output of the system . Now we
shall develop a differential equation to describe them . The differentiation of (2 .8)
yields
dh2 = R dqz dht = R dq, + dhz = R dq,
z dt and + R z dq2
dt dt i dt dt ' dt dt
The substitution of these into (2.9) yields
, d12)
q j - q, = A, (R, d~ + R2 (2 .1 Oa)
dq2
q, - q2 - A2 R2 dt (2 .l Ob)
Now we eliminate q, from (2.10a) by using (2 .10b) and its derivative :

z
q j - (q2 + A2R2 dq2 = A IR, ddt2 + A2R2 d 2) + A,R2 dt
(
which can be simplified as
dh2
A,A2R,R2 + (A,(R 1 + R2) + A2R2) q
d z + q2( t) = qi(t) (2 .11)
dt dt
This second-order differential equation describes the input q; and output q2 of the
system in Figure 2 .12 .
To conclude this section, we mention that a large number of physical systems
can be modeled, after simplification and approximation, as linear time-invariant
lumped (LTIL) systems over limited operational ranges . These systems can then be

described by LTIL differential equations . In this text, we study only this class of
systems.
2 .3 ZERO-INPUT RESPONSE AND ZERO-STATE RESPONSE
The response of linear, in particular LTIL, systems can always be decomposed into
the zero-input response and zero-state response . In this section we shall use a simple
example to illustrate this fact and then discuss some general properties of the zero-
input response. The Laplace transform in Appendix A is needed for the following
discussion .
Consider the differential equation
d2y2t) + 3 dy(t) du(t)
+ 2y(t) = 3 - u(t) (2 .12)
dt dt dt
Many methods are available to solve this equation . The simplest method is to use
the Laplace transform . The application of the Laplace transform to (2 .12) yields,
using (A.9),
s2 Y(s) - sy(0 - ) - y(0 - ) + 3[sY(s) - y(0 -)] + 2Y(s)
(2 .13)
= 3[sU(s) - u(0 - )] - U(s)
where y(t) : = dy(t)/dt and capital letters denote the Laplace transforms of the cor-
responding lowercase letters? Equation (2 .13) is an algebraic equation and can be
manipulated using addition, subtraction, multiplication, and division . The grouping
of Y(s) and U(s) in (2 .13) yields
(s2 + 3s + 2)Y(s) = sy(0 - ) + y(0 -) + 3y(0 - ) - 3u(0 - ) + (3s - 1)U(s)
which implies
+ 3)y(0 -) + y(0 - ) - 3u(0 - ) + 3s- 1
Y(s) = (s U(s) (2 .14)
s2 +3s+2 s2 +3s+2
Zero-Input Response Zero-State Response
This equation reveals that the solution of (2 .12) is partly excited by the input u(t),
t ? 0, and partly excited by the initial conditions y(0 - ), y(0 - ), and u(0 -). These
initial conditions will be called the initial state . The initial state is excited by the
input applied before t = 0 . In some sense, the initial state summarizes the effect of
the past input u(t), t < 0, on the future output y(t), for t ? 0. If different past inputs
ut(t), u2(t), . . . , t--- 0, excite the same initial state, then their effects on the future
output will be identical . Therefore, how the differential equation acquires the initial
state at t = 0 is immaterial in studying its solution y(t), for t ? 0 . We mention that
2We use A := B to denote that A, by definition, equals B, and A = : B to denote that B, by definition,
equals A .

2 .3 ZERO-INPUT RESPONSE AND ZERO-STATE RESPONSE 25
the initial time t = 0 is not the absolute time; it is the instant we stari to study the
system.
Consider again (2.14). The response can be decomposed into two parts . The
first part is excited exclusively by the initial state and is called the zero-input re-
sponse. The second part is excited exclusively by the input and is called the zero-
state response . In the study of LTIL systems, it is convenient to study the zero-input
response and the zero-state response separately . We first study the zero-input re-
sponse and then the zero-state response .
2.3.1 Zero-Input Response-Characteristic Polynomial

Consider the differential equation in (2.12). If u(t) = 0, for t ? 0, then (2 .12) reduces
to
d2y(t)
+ 3 dy(t) + 2y(t) = 0
dt2 dt
This is called the homogeneous equation . We now study its response due to a nonzero
initial state. The application of the Laplace transform yields, as in (2 .13),
s2Y(s) - sy(0 -) - y(0 - ) + 3[sY(s) - y(0 - )] + 2Y(s) = 0
which implies
+ 3)y(0 - ) + y(0 - ) - (s + 3)y(0 - ) + y(0-)
Y(s) = (s (2 .15)
s 2 +3s+2 (s+ 1)(s+2)
This can be expanded as
k1
Y(s) = + k2 (2 .16)
s + 1 s + 2
with
(s + 3)y(0 -) + y(0 - )
k1 = 2y(0 - ) + y(0 - )
s+2
and
y(0-)
(s + 3)y(0) +
k_ = -1Y(0 - ) + AID - )]
s + 1 s=-2
Thus the zero-input response is
y(t) = k1e-t + k 2e-2t (2 .17)
No matter what the initial conditions y(0 - ) and y(0 - ) are, the zero-input response
is always a linear combination of the two functions a-t and e -2t. The two functions
et and e- 2t are the inverse Laplace transforms of 1/(s + 1) and 1/(s + 2) . The
two roots -1 and - 2-or, equivalently, the two roots of the denominator of (2 .15)

are called the modes of the system . The modes govern the form of the zero-input
response of the system .
We now extend the preceding discussion to the general case . Consider the nth
order LTIL differential equation
any '"'(t) + an-ty(n-1)(t) + . . . + a t y( ' ) (t) + a oy(t)
= bm u(m)(t) + b.-,u(--')(t) + . . . + b,u ( ' ) (t) + b out) (2 .18)
where
d` d
y (') (t) : = dt` y(t), u(`~(t) : = dt` u(t)
and y(t) : = y1 ' )(t), y(t) : = yt 2) (t) . We define

D(p) := a n p n + a n _,pn - ' + ... + aIp + ao (2 .19a)
and
N(p) := b,,,pm + bm ,pm-' + . . . + b I p + b o (2 .19b)
where the variable p is the differentiator d/dt defined by

d d2 d3
py(t) = d y(t) p 2y(t) = dtz y(t) p3y(t) :=
dt3
y(t) (2.20)
and so forth. Using this notation, (2 .18) can be written as

D(p)y(t) = N(p)u(t) (2 .21)
In the study of the zero-input response, we assume u(t) _- 0. Then (2.21) reduces to
D(p)y(t) = 0 (2 .22)
This is the homogeneous equation . Its solution is excited exclusively by initial con-
ditions . The application of the Laplace transform to (2 .22) yields, as in (2.15),
I(s)
Y(s) =
D(s)
where D(s) is defined in (2 .19a) with p replaced by s and I(s) is a polynomial of s
depending on initial conditions . We call D(s) the characteristic polynomial of (2.21)
because it governs the free, unforced, or natural response of (2.21). The roots of the
polynomial D(s) are called the modes . 3 For example, if
D(s) = (s - 2)(s + 1)2 (s+2-j3)(s+2+j3)
then the modes are 2, - 1, -1 . - 2 + j3, and - 2 - j3 . The root 2 and the complex
roots -2 j3 are simple modes and the root - I is a repeated mode with multi-
3In the literature, they are also called the natural frequencies . However the W, in D(s) = s2 + 2~ws
+ u is also called the natural frequency in this and some other control texts. To avoid possible confusion,
we call the roots of D(s) the modes .

2 .4 ZERO-STATE RESPONSE-TRANSFER FUNCTION 27
plicity 2. Thus for any initial conditions, Y(s) can be expanded as

O _ ki k2 k3 ci c2
Ys
s - 2 + s + 2 - j3 + s + 2 + j3 + s + I + (s + 1) 2
and its zero-input response is, using Table A .1,
y(t) = k,e2` + k2e -(2- j3)t + k3e -(2+33)r + cie -r + -cite-`
This is the general form of the zero-input response and is determined by the modes
of the system .
Exercise 2.3.1
Find the zero-input responses of (2.12) due to y(0 -) = 1, y(0 -) _ -1, and
AV) = YO-) = 1 .
[Answers : y(t) = e y(t) = 3e - r - 2e -2r.]
Exercise 2.3.2
Find the modes and the general form of the zero-input responses of
D(p)y(t) = N(p)u(t) (2.23)
where
D(p) = p2(p - 2)2(p2 + 4p + 8) and N(p) = 3p2 - 10
[Answers : 0, 0, 2, 2, -2 + j2, and -2 - j2; k, + k 2t + k 3e2` + k4te2` +
kse - (2- j2 )t + k6e -(2+ j2)r ]
2 .4 ZERO-STATE RESPONSE-TRANSFER FUNCTION
Consider the differential equation in (2.12) or

d 2y(t) dy(t) du
+ 3 dt + 2y(t) = 3 - u(t) (2 .24)
dt2 d
The response of (2.24) is partly excited by the initial conditions and partly excited
by the input u(t). If all initial conditions equal zero, the response is excited exclu-
sively by the input and is called the zero-state response. In the Laplace transform
domain, the zero-state response of (2.24) is governed by, setting all initial conditions
in (2.14) to zero,
3s - 1
Y(s) = s : G(s)U(s) (2.25)
2 + 3s + 2 U(s) =

where the rational function G(s) = (3s - 1)/(s 2 + 3s + 2) is called the transfer
function. It is the ratio of the Laplace transforms of the output and input when all
initial conditions are zero or
G(s) = Y(s) -T[Output]
(2 .26)
U(s) Initial conditions =0 `[Input] Initial conditions=0
The transfer function describes only the zero-state responses of LTIL systems .
Example 2 .4 .1 (Mechanical system)

Consider the mechanical system studied in Figure 2.5(a). As derived in (2 .4), it is
described by
m d22t) + kI ddt) + k2y(t) = u(t)
The application of the Laplace transform yields, assuming zero initial conditions,
ms2Y(s) + kl sY(s) + k2Y(s) = U(s)
or
(Ms' + kls + k 2)Y(s) = U(s)
Thus the transfer function from u to y of the mechanical system is
- 1
G(s) _ Y(s)
U(s) s + kl s + k2
This example reveals that the transfer function of a system can be readily ob-
tained from its differential-equation description . For example, if a system is de-
scribed by the differential equation
D(p)y(t) = N(p)u(t)
where D(p) and N(p) are defined as in (2 .19), then the transfer function of the system
is
N(s)
G(s) _
D(s)
Exercise 2 .4.1
Find the transfer functions from u to y of the networks shown in Figures 2 .10 and
2.11.
[Answers : 1/(6s + 2), 1/(LCs 2 + RCs + 1), (CR 2s + 1)/(C(R l + R2)s + 1).]

Exercise 2 .4 .2 (Industrial process)
Find the transfer function from q, to q2 of the system in Figure 2 .12 .

[Answer : G(s) = I/(A 1A2 R1R2s2 + (A1 (R1 + R2) + A2R2)s + 1).]
RLC Networks
Although the transfer function of an RLC network can be obtained from its
differential-equation description, it is generally simpler to compute it by using the
concept of the Laplacian impedance or, simply, the impedance . If all initial condi-
tions are zero, the application of the Laplace transforms to (2.6) yields
V(s) = RI(s) (resistor)
V(s) = C1s I(s) (capacitor)
and
V(s) = LsI(s) (inductor)
These relationships can be written as V(s) = Z(s)I(s), and Z(s) is called the
(Laplacian) impedance. Thus the impedances of the resistor, capacitor, and inductor
are respectively R, 1/Cs, and Ls . If we consider I(s) the input and V(s) the output,
then the impedance is a special case of the transfer function defined in (2.26). When-
ever impedances are used, all initial conditions are implicitly assumed to be zero.
The manipulation involving impedances is purely algebraic, identical to the
manipulation of resistances . For example, the resistance of the series connection of
two resistances R, and R2 is R 1 + R2; the resistance of the parallel connection of
R1 and R2 is R1R2/(R 1 + R2) . Similarly, the impedance of the series connection of
two impedances Z 1(s) and Z2(s) is Z1 (s) + Z2(s); the impedance of the parallel
connection of Z1(s) and Z2(s) is Z 1(s)Z2(s)/(Z 1(s) + Z2(s)). The only difference is
that now we are dealing with rational functions, rather than real numbers as in the
resistive case .
Example 2 .4.2
Compute the transfer function from u to i of the network shown in Figure 2.13(a).
Its equivalent network using impedances is shown in Figure 2 .13(b) . The impedance
of the parallel connection of 1/2s and 3s + 2 is
1
(3s + 2)
2s _ 3s + 2
1 6s2 +4s+ 1
+(3s+2)
2s

2t 2
(a) (b) (c)

Figure 2 .13 Network.
as shown in Figure 2 .13(c) . Hence the current I(s) shown in Figure 2 .13 is given by
_ U(s) 6s 2 + 4s + 1
I(s) 3 U(s)
3s + 2 6s2 + 7s +
1 +
6s 2 +4s+ 1
Thus the transfer function from u to i is
I(s) _ 6sz + 4s + 1
G(s) = U(s)
6sz + 7s + 3
Exercise 2.4.3
Find the transfer functions from u to y of the networks in Figures 2 .10 and 2.11
using the concept of impedances .
2.4.1 Proper Transfer Functions

Consider the rational function
G(s) = N(s)
D(s)
where N(s) and D(s) are two polynomials with real coefficients . We use deg to denote
the degree of a polynomial . If
deg N(s) > deg D(s)
G(s) is called an improper rational function. For example, the rational functions
z 1o
s s2 + 1 and
S 1 s9 +ss +s7 - 10
are all improper. If
deg N(s) < deg D(s)

G(s) is called a proper rational function . It is strictly proper if deg N(s) < deg D(s) ;
biproper if deg N(s) = deg D(s). Thus proper rational functions include both strictly
proper and biproper rational functions . If G(s) is biproper, so is G -1(s) = D(s)/N(s) .
This is the reason for calling it biproper .
Exercise 2 .4.4
Classify the following rational functions

1 s2 - 1 s- 1
s2 + 1 2
S + 1 s+ 1 s+ 1
How many of them are proper rational functions?
[Answers : Improper, biproper, strictly proper, improper, biproper ; 3.]
The properness of a rational function G(s) can also be determined from the
value of G(s) at s = o. It is clear that G(s) is improper if G(oo) = 00, proper if
G(oo) is a finite nonzero or zero constant, biproper if G(oo) is finite and nonzero, and
strictly proper if G(00) = 0 .
The transfer functions we will encounter in this text are mostly proper rational
functions . The reason is twofold . First, improper transfer functions are difficult, if
not impossible, to build in practice, as will be discussed in Chapter 5 . Second,
improper transfer functions will amplify high-frequency noise, as will be explained
in the following .
Signals are used to carry information. However, they are often corrupted by
noise during processing, transmission, or transformation . For example, an angular
position can be transformed into an electrical voltage by using the wirewound po-
tentiometer shown in Figure 2 .14 . The potentiometer consists of a finite number of
turns of wiring, hence the contact point moves from turn to turn . Because of brush
(a) (b)
Figure 2.14 Potentiometer and its characteristic .

jumps, wire irregularities, or variations of contact resistance, unwanted spurious

voltage will be generated . Thus the output voltage v(t) of the potentiometer will not
be exactly proportional to the angular displacement 0(t), but rather will be of the
form
v(t) = kO(t) + n(t) (2 .27)
where k is a constant and n(t) is noise . Therefore, in general, every signal is of the
form
v(t) = i(t) + n(t) (2 .28)
where i(t) denotes information and n(t) denotes noise . Clearly in order for v(t) to
be useful, we require
v(t) - i(t)
and for any system to be designed, we require
Response of the system due to v(t)
(2 .29)
response of the system due to i(t)
where - denotes "roughly equal to." If the response of a system excited by v(t) is
drastically different from that excited by i(t), the system is generally useless in
practice . Now we show that if the transfer function of a system is improper and if
the noise is of high frequency, then the system is useless . Rather than discussing the
general case, we study a system with transfer function s and a system with transfer
function 1/s . A system with transfer function s is called a differentiator because it
performs differentiation in the time domain . A system with transfer function 1/s is
called an integrator because it performs integration in the time domain . The former
has an improper transfer function, the latter has a strictly proper transfer function .
We shall show that the differentiator will amplify high-frequency noise ; whereas the
integrator will suppress high-frequency noise. For convenience of discussion, we
assume
i(t) = sin 2t n(t) = 0.01 sin 1000t
and
v(t) = i(t) + n(t) = sin 2t + 0 .01 sin 1000t (2 .30)
The magnitude of the noise is very small, so we have v(t) - i(t). If we apply this
signal to a differentiator, then the output is
dv(t) = 2 cos 2t + 0 .01 x 1000 cos 1000t = 2 cos 2t + 10 cos 1000t

dt
Because the amplitude of the noise term is five times larger than that of the infor-
mation, we do not have dv(t)/dt - di(t)/dt as shown in Figure 2 .15 . Thus a differ-
entiator-and, more generally, systems with improper transfer functions-cannot be
used if a signal contains high-frequency noise .

dv(t)
dt
15
10 I
0
ff
-5
-lo
-15
0 2 4 6 10 12 14
Figure 2 .15 Responses of differentiator.
If we apply (2.30) to an integrator, then the output is

1 0.01
fo v(T)dT = -2 cos 2t - 1000
cos 1000t
The output due to the noise term is practical zero and
j v(T)dT - l(T)dT
o fU
Thus we conclude that an integrator-and, more generally, systems with strictly

proper transfer functions-will suppress high-frequency noise .
In practice, we often encounter high-frequency noise . As was discussed earlier,
wirewound potentiometers will generate unwanted high-frequency noise . Thermal
noise and shot noise, which are of high-frequency compared to control signals, are
always present in electrical systems . Because of splashing and turbulent flow of
incoming liquid, the measured liquid level of a tank will consist of high-frequency
noise . In order not to amplify high-frequency noise, most systems and devices used
in practice have proper transfer functions .
Exercise 2 .4 .5
Consider
v(t) = i(t) + n(t) = cos 2t + 0 .01 cos 0 .001t
Note that the frequency of the noise n(t) is much smaller than that of the information
i(t) . Do we have v(t) - i(t)? Do we have dv(t)/dt - di(t)/dt? Is it true that a
differentiator amplifies any type of noise?
[Answers : Yes, yes, no.]

2.4.2 Poles and Zeros

The zero-state response of a system is governed by its transfer function . Before
computing the response, we introduce the concepts of poles and zeros . Consider a
proper rational transfer function
G(s) = N(s)
D(s)
where N(s) and D(s) are polynomials with real coefficients and deg N(s) < deg D(s) .
o Definition
A finite real or complex number A is a pole of G(s) if IG(A) l where H
denotes the absolute value . It is a zero of G(s) if G(A) = 0 .
Consider the transfer function

N(s) = 2(s 3 + 3s2 - s - 3)
G(s) = (2.31)
D(s) (s - 1)(s + 2)(s + 1)3
We have
N(-2) _ 2[(-2)3 + 3(-2)2 - (- 2) - 3] 6
G(-2) = 00
D(-2) (-3) . 0 . (- 1) - 0 -
Therefore - 2 is a pole of G(s) by definition . Clearly - 2 is a root of D(s) .
Does this imply every root of D(s) is a pole of G(s)? To answer this, we check
s = 1, which is also a root of D(s). We compute G(1):
N(l) 2(1 +3-1 -3) 0
D(1) 0 .3 . 8 0
It is not defined . However 1'Hopital's rule implies
G(1) = N(s) N' (s)
D(s) S-1
D' (s) S-1
2(3s 2 + 6s - 1) - -
16 5x- 00
5s4 + 165 3 + 12s 2 -4s-5 s=1
24
Thus s = 1 is not a pole of G(s) . Therefore not every root of D(s) is a pole of G(s) .
Now we factor N(s) in (2.31) and then cancel the common factors between N(s)
and D(s) to yield
2(s + 3)(s - 1)(s + 1) - 2(s + 3)
G(s) = (2.32)
(s - 1)(s + 2)(s + 1)3 (s + 2)(s + 1) 2
We see immediately that s = 1 is not a pole of G(s) . Clearly G(s) has one zero, - 3,

and three poles, - 2, - 1, and - 1 . The pole - 2 is called a simple pole and the
pole -1 is called a repeated pole with multiplicity 2 .4
From this example, we see that if polynomials N(s) and D(s) have no common
factors,' then all roots of N(s), and all roots of D(s) are, respectively, the zeros and
poles of G(s) = N(s)/D(s) . If N(s) and D(s) have no common factor, they are said
to be coprime and G(s) = N(s)/D(s) is said to be irreducible . Unless stated other-
wise, every transfer function will be assumed to be irreducible .
We now discuss the computation of the zero-state response . The zero-state re-
sponse of a system is governed by Y(s) = G(s)U(s). To compute Y(s), we first
compute the Laplace transform of u(t) . We then multiply G(s) and U(s) to yield Y(s) .
The inverse Laplace transform of Y(s) yields the zero-state response . This is illus-
trated by an example .
Example 2 .4.3
Find the zero-state response of (2 .25) due to u(t) = 1, for t ? 0. This is called the
unit-step response of (2 .25) . The Laplace transform of u(t) is 1/s . Thus we have
3s - 1 1
Y(s) = G(s)U(s) _ (s + 1)(s + 2) s (2.33)
To compute its inverse Laplace transform, we carry out the partial fraction expansion
as
_ 3s - 1 kt k2 k3
Y(s)
(s + 1)(s + 2)s s + 1 + s + 2 + s
where
3s - 1 -4
kt = Y(s) . (s + 1) = 4
(s + 2)s s=-1 (1)(-1)
3s - 1 -7
k2 = Y(s) - (s + 2) = - 3 .5
s= -2 (s + 1)s s=-2
(-1)(-2)
and
_ 3s - 1 -1
k3 = Y(s) . s = -0.5
S=o (s + 1)(s + 2) S=o 2
41f s is very large, (2 .32) reduces to G(s) = I /s 2 and G(x) = 0. Thus - can be considered as a repeated
zero with multiplicity 2 . Unless stated otherwise, we consider only finite poles and zeros .
'Any two polynomials, such as 4s + 2 and 6s + 2, have a constant as a common factor . Such a common
factor, a polynomial of degree 0, is called a trivial common factor . We consider only nontrivial common
factors-that is, common factors of degree 1 or higher .

Using Table A .1, the zero-state response is

y(t) = 4e -t - 3.5e -2` - 0 .5 (2 .34)
Due to the Due to the
Poles of G(s) Poles of U(s)
for t ? 0 . Thus, the use of the Laplace transform to compute the zero-state response
is simple and straightforward .
This example reveals an important fact of the zero-state response . We see from
(2.34) that the response consists of three terms . Two are the inverse Laplace trans-
forms of 1/(s + 2) and 1/(s + 1), which are the poles of the system . The remaining
term is due to the step input . In fact, for any u(t), the response of (2 .33) is generally
of the form
-2t
y(t) = kte - ` + k2e + (terms due to the poles of U(s)) (2.35)
(see Problem 2 .20) . Thus the poles of G(s) determine the basic form of the zero-
state response .
Exercise 2.4 .6
Find the zero-state response of 1/(s + 1) due to e-2`, t ? 0.

[Answer : y(t) = e - ` - e -2a .]
If kt in (2.35) is zero, the corresponding pole -1 is not excited . A similar

remark applies to k2 . Because both kt and k2 in (2.34) are nonzero, both poles of
G(s) in (2.33) are excited by the step input . For other inputs, the two poles may not
always be excited . This is illustrated by an example .
Example 2 .4.4
Consider the system in (2.33). Find a bounded input u(t) so that the pole -1 will
not be excited. If U(s) = s + 1, then
3s - 1
Y(s) = G(s)U(s) (s + 1)
_ (s + 1)(s + 2) .
_ 3s- 1 3(s+2)-7 7
s+ 2 s+ 2 3 s+ 2
which implies
y(t) = 35(t) - 7e -2`

This response does not contain e -t, thus the pole - 1 is not excited. Therefore if
we introduce a zero in U(s) to cancel a pole, then the pole will not be excited by the
input u(t) .
If U(s) is biproper or improper, as is the case for U(s) = s + 1, then its inverse
Laplace transform u(t) will contain an impulse and its derivatives and is not bounded .
In order for u(t) to be bounded, we choose, rather arbitrarily, U(s) = (s + 1)/
s(s + 3), a strictly proper rational function . Its inverse Laplace transform is
u(t) = + 2 e -3t
3
for t ? 0 and is bounded . The application of this input to (2 .33) yields

3s- 1 s+ 1 3s- 1
Y(s)
_ (s + 2)(s + 1) s(s + 3) (s + 2)(s + 3)s
7 10 1
2(s + 2) 3(s + 3) 6s
which implies
2t - - 1
y(t) = 72 e- 10 e- 3t
3 6
for t > 0. The second and third terms are due to the input, the first term is due to
the pole -2 . The term e-i does not appear in y(t), thus the pole - I is not excited
by the input. Similarly, we can show that the input (s + 2)/s(s + 1) or (s + 2)/
(s + 3)2 will not excite the pole -2 and the input (s + 2)(s + 1)/s(s + 3) 2 will
not excite either pole .
From this example, we see that whether or not a pole will be excited depends
on whether u(t) or U(s) has a zero to cancel it. The Laplace transforms of the unit-
step function and sin mot are
1 mo
- and
s s2 + (0 0
They have no zero . Therefore, either input will excite all poles of every LTIL system .
The preceding discussion can be extended to the general case . Consider, for
example,
(s2 10)(s + 2)(s-1)2
Y(s) = G(s)U(s) := 3(S U(s)
s-2)(s+2-j2)(s+2+j2)
The transfer function G(s) has poles at 0, 0, 0, 2, 2, and - 2 j2 . The complex
poles -2 j2 are simple poles, the poles 0 and 2 are repeated poles with multi-
plicities 3 and 2 . If G(s) and U(s) have no pole in common, then the zero-state

response of the system due to U(s) is of the form

y(t) = k1 + k2t + k3t 2 + k4e2` + kste2` + k6e -(2- '2)t + k 7 e -(2+j2)c
(2 .36)
+ (Terms due to the poles of U(s))
(see Problem 2.20.) Thus the poles of G(s) determine the basic form of the response .
Do the zeros of G(s) play any role in the zero-state response? Certainly, they
do. They affect the values of k;. Different sets of k; yield drastically different re-
sponses, as is illustrated by the following example .
Example 2 .4.5
Consider
2
G 1 (s) =
(s+1)(s+1+j)(s+1 - j)
0.2(s + 10)
G2(s) (s + 1)(s + 1 + j)(s + 1 - j)
-0.2(s - 10)
G3(s) (s+1)(s+ 1 +j)(s+ 1
10(5 2 + 0.1s + 0.2)
G4(s)
(s + 1)(s + 1 + j)(s + 1 - j)
The transfer function G1 (s) has no zero, G 2 (s) and G 3 (s) have one zero, and G4(s)
has a pair of complex conjugate zeros at - 0.05 0.444j. They all have the same
2.5
1 .5
0.5
-0 .5
-1 .5
- 2 1
0 1 2 3 4 5 6 7 8 9 10
Figure 2 .16 Unit-step responses of G ;(s) .

2 .5 BLOCK REPRESENTATION-COMPLETE CHARACTERIZATION 39
set of poles, and their unit-step responses are all of the form
y(t) = ki e -t + k2e -(1+ t)t j + k3e -(1-.il)t + k4
with k 3 equal to the complex conjugate of k 2 . Their responses are shown in Figure
2.16, respectively, with the solid line (G 1 (s)), dashed line (G 2 (s)), dotted line (G3 (s)),
and dash-and-dotted line (G 4 (s)). They are quite different. In conclusion, even though
the poles of G(s) determine the basic form of responses, exact responses are deter-
mined by the poles, zeros, and the input. Therefore, the zeros of a transfer function
cannot be completely ignored in the analysis and design of control systems .
2.5 BLOCK REPRESENTATION-COMPLETE CHARACTERIZATION
In the analysis and design of control systems, every device is represented by a block
as shown in Figure 2 .4 or 2 .17(a) . The block is then represented by its transfer
function G(s) . If the input is u(t) and the output is y(t), then they are related by
Y(s) = G(s)U(s) (2.37)
where Y(s) and U(s) are respectively the Laplace transforms of y(t) and u(t) . Note
that we have mixed the time-domain representation u(t) and y(t) and the Laplace
transform representation G(s) in Figure 2 .17(a) . This convention will be used
throughout this text. It is important to know that it is incorrect to write y(t) _
G(s)u(t) . The correct expression is Y(s) = G(s)U(s) .'
Equation (2 .37) is an algebraic equation . The product of the Laplace transform
of the input and the transfer function yields the Laplace transform of the output . The
advantage of using this algebraic representation can be seen from the tandem con-
nection of two systems shown in Figure 2 .17(b) . Suppose the two systems are rep-
resented, respectively, by
Y1(s) = Gl(s)U1(s) Y2(s) = G2(s)U2(s)
In the tandem connection, we have u 2 (t) = y 1 (t) or U2(s) = Y1 (s) and
Y2 (s) = G 2 (s)Y1 (s) = G 2 (s)G 1 (s)U1 (s) (2.38)
U y uI Y, u2 y2 u~ y2

9 G(s) G,(S) G2(S)
(c)
G 2 (s) G,(s)
F
(a) (b)
Figure 2 .17 (a) A system . (b) Tandem connection of two systems . (c) Reduction of (b).
blt can also be expressed as y(t) = fo g(t - T)u(T)dT, where g(t) is the inverse Laplace transform of
G(s) . See Reference [18] . This form is rarely used in the design of control systems.

Thus the tandem connection can be represented by a single block, as shown in Figure
2.17(c), with transfer function G(s) : = G2 (s) G 1 (s), the product of the transfer func-
tions of the two subsystems . If we use differential equations, then the differential
equation description of the tandem connection will be much more complex . Thus
the use of transfer functions can greatly simplify the analysis and design of control
systems.
The transfer function describes only the zero-state response of a system . There-
fore, whenever we use the transfer function in analysis and design, the zero-input
response (the response due to nonzero initial conditions) is completely disregarded .
However, can we really disregard the zero-input response? This question is studied
in this section.
Consider a linear time-invariant lumped (LTIL) system described by the differ-
ential equation
D(p)y(t) = N(p)u(t) (2.39)
where
D(p) = anpn + an -1pn-' + . . . + a lp + a o
N(p) =bm pm+bm _ 1pm - '+ . . .+b1p+bo
and the variable p is the differentiator defined in (2 .19) and (2.20). Then the zero-
input response of the system is described by
D(p)y(t) = 0 (2.40)
and the response is dictated by the roots of D(s), called the modes of the system
(Section 2.3.1). The zero-state response of the system is described by the transfer
function
N(s)
G(s) :
= D(s)
and the basic form of its response is governed by the poles of G(s) (Section 2.4.2).
The poles of G(s) are defined as the roots of D(s) after canceling the common factors
of N(s) and D(s). Thus if D(s) and N(s) have no common factors, then
The set of the poles = The set of the modes (2.41)
In this case, the system is said to be completely characterized by its transfer function .
If D(s) and N(s) have common factors, say R(s), then the roots of R(s) are nodes of
the system but not poles of G(s). In this case the roots of R(s) are called the missing
poles of the transfer function, and the system is said to be not completely charac-
terized by its transfer function . We use examples to illustrate this concept and discuss
its implications .

Example 2 .5.1
Consider the system shown in Figure 2 .18 . The input is a current source . The output
y is the voltage across the 2-fl resistor as shown . The system can be described by
the LTIL differential equation
dy(t) - du(t)
0.75y(t) = _ 0.75u(t) (2 .42a)
dt dt
(Problem 2.21) . The equation can also be written, using p = d/dt, as
(p - 0.75)y(t) _ (p - 0.75)u(t) (2 .42b)
The mode of the system is the root of (s - 0.75) or 0.75 . Therefore its zero-input
response is of the form
y(t) = keo .75t (2 .43)
where k depends on the initial voltage of the capacitor in Figure 2 .18 . We see that
if the initial voltage is different from zero, then the response will approach infinity
as t - .
We will now study its zero-state response . The transfer function of the system
is
s - 0.75
G(s) _ = 1 (2 .44)
s - 0.75
Because of the common factor, the transfer function reduces to 1 . Thus the system
has no pole and the zero-state response is y(t) = u(t), for all t. This system is not
completely characterized by its transfer function because the mode 0.75 does not
appear as a pole of G(s) . In other words, the transfer function has missing pole 0.75.
If we use the transfer function to study the system in Figure 2 .18, we would
conclude that the system is acceptable . In reality, the system is not acceptable, be-
cause if, for any reason, the voltage of the capacitor becomes nonzero, the response

will grow without bound and the system will either become saturated or bum out .
Thus the system is of no use in practice .
The existence of a missing pole in Figure 2 .18 can easily be explained from the
structure of the network . Because of the symmetry of the four resistors, if the initial
voltage of the capacitor is zero, its voltage will remain zero no matter what current
source is applied . Therefore, the removal of the capacitor will not affect the zero-
state response of the system . Thus, the system has a superfluous component as far
as the input and output are concerned . These types of systems are not built in practice,
except by mistake .
Example 2 .5 .2
Consider the system described by
(p2 + 2p - 3)y(t) = (p - 2)u(t) (2.45)
The zero-input response of the system is governed by
(p 2 + 2p - 3)y(t) = 0
Its modes are the roots of (s 2 + 2s - 3) = (s - 1)(s + 3) or 1 and -3 . Thus the
zero-input response due to any initial conditions is of the form
yZ1(t) = kiet + k2e -3t
where the subscript zi denotes zero-input . The response of mode 1 approaches in-
finity and the response of mode - 3 approaches zero as t -* oc .
The transfer function of the system is
s-2 - s-2
G(s)- (2 .46)
s2 + 2s - 3 (s - 1)(s + 3)
Thus the zero-state response of the system due to u(t) will generally be of the form
y(t) = kiet + k2e -3t + (Terms due to the poles of U(s))
We see that the two modes appear as the poles of G(s) . Thus the system has no
missing poles and is completely characterized by its transfer function . In this case,
the zero-input response due to any initial conditions will appear in the zero-state
response, so no essential information is lost in using the transfer function to study
the system .
In conclusion, if a system is completely characterized by its transfer function,

then the zero-input response will appear essentially as part of the zero-state response .
It is therefore permissible to use the transfer function in analysis and design without

considering the response due to nonzero initial conditions . If a system is not com-
pletely characterized by its transfer function, care must be exercised in using the
transfer function to study the system . This point is discussed further in Chapter 4 .
Exercise 2 .5.1
Which of the following systems are completely characterized by their transfer func-
tions? If not, find the missing poles .
a. (p2 + 2p + 1)y(t) = (p + 1)p u(t)
b. (p2 - 3p + 2)y(t) = (p - 1) u(t)
c. (p2 - 3p + 2)y(t) = u(t)
[Answers : (a) No, - 1 (b) No, 1 (c) Yes .]
2.5.1 The Loading Problem

Most control systems are built by interconnecting a number of subsystems as shown
in Figures 1 .1(b), 1 .6(b), and 1 .7(b) . In analysis and design, every subsystem will
be represented by a block and each block will be represented by a transfer function .
Thus, most control systems will consist of a number of blocks interconnected with
each other . In connecting two or more blocks, the problem of loading may occur .
For example, consider a 10-volt voltage source modeled as shown in Figure 2 .19(a) .
It has an internal resistance of 10 SL ; its output voltage is 10 volts when nothing is
connected to its terminals . Now consider a device that can be modeled as a 10-fl
resistor . When we connect the device to the voltage source, the output voltage of
the source is not 10 volts, but 10 . 10/(10 + 10) = 5 volts. If we connect a 20-fl
device to the voltage source, the output is 10 . 20/(10 + 20) = 6 .7 volts . We see
that the voltages supplied by the same voltage source will be different when it is
connected to different devices . In other words, for different loads, the voltages sup-
plied by the same voltage source will be different . In this case, the connection is
said to have a loading problem . On the other hand, if the voltage source is designed
so that its internal resistance is negligible or zero, as shown in Figure 2 .19(b), then
10t
b0v t0v
(a) (b)
Figure 2.19 Loading problem.

no matter what device is connected to it, the supplied voltage to the device is always
10 volts. In this case, the connection is said to have no loading problem .
Roughly speaking, if its transfer function changes after a system is connected
to another system, the connection is said to have a loading effect . For example, the
transfer function from u to y in Figure 2 .19(a) is 1 before the system is connected
to any device. It becomes 5/10 = 0 .5 when the system is connected a 10-ft resistor .
Thus, the connection has a loading effect. If the tandem connection of two systems
has a loading effect, then the transfer function of the tandem connection does not
equal the product of the transfer functions of the two subsystems as developed in
(2.38). This is illustrated by an example .
Example 2 .5.3
Consider the networks shown in Figure 2 .20(a) . The transfer function from u, to y,
of network M, is G 1 (s) = s/(s + 1) . The transfer function from u 2 to y2 of network
M2 is G2 (s) = 2/(2 + 3s) . Now we connect them together or set y, = u 2 as shown
in Figure 2 .20(b) and compute the transfer function from u l to y 2 . The impedance
of the parallel connection of the impedance s and the impedance (3s + 2) is
s(3s + 2)/(s + 3s + 2) = s(3s + 2)/(4s + 2). Thus the current I, shown in
Mt M
2
(a) (b)
Isolation amplifier with k = I
(c)
Figure 2 .20 Tandem connection of two systems .

2 .6 STATE-VARIABLE EQUATIONS 45
Figure 2 .20(b) equals

U1(s)
I 1 (s) = (2 .47)
1 + s(3s + 2)
4s + 2
and the current 12 equals
s s
IZ(s) = Il (s) 1 (s) (2 .48)
s + 3s + 2 4s + 2 + s(3s + 2) ' U
Thus, the voltage y2 (t) is given by
2s
Y2 (s) = I2(s) 2 U I (s) (2 .49)
= 3s2 + 6s + 2
and the transfer function from u 1 to y2 of the tandem connection is
Y2(s) _ 2s
(2 .50)
3s2 + 6s + 2
U 1 (s)
This is different from the product of G 1 (s) and G 2(s) or
s 2 2s
G 1(s)G2(s) = s + 1 (2 .51)
3s + 2 = 3s2 + 5s + 2
Thus, Equation (2 .38) cannot be used for this connection .
The loading of the two networks in Figure 2 .20 can be easily explained . The
current 12 in Figure 2 .20(a) is zero before the connection ; it becomes nonzero after
the connection. Thus, the loading occurs . In electrical networks, the loading often
can be eliminated by inserting an isolating amplifier, as shown in Figure 2 .20(c).
The input impedance Z;,, of an ideal isolating amplifier is infinity and the output
impedance Z,,u, is zero . Under these assumptions, 12 in Figure 2 .20(c) remains zero
and the transfer function of the connection is G2 (s)kG 1 (s) with k = 1 .
The loading problem must be considered in developing a block diagram for a
control system. This problem is considered in detail in the next chapter .
2 .6 STATE-VARIABLE EQUATIONS
The transfer function describes only the relationship between the input and output
of LTIL systems and is therefore called the input-output description or external
description . In this section we shall develop a different description, called the state-
variable description or internal description . Strictly speaking, this description is the
same as the differential equations discussed in Section 2 .2. The only difference is
that high-order differential equations are now written as sets of first-order differ-
ential equations. In this way, the study can be simplified.

The state-variable description of LTIL systems is of the form

x 1 (t) = a 11 x1 (t) + a 12x2(t) + a, 3x3 (t) + b1u(t)
X2(t) = a21x1(t) + a22x2(t) + a23x3(t) + b2u(t) (2 .52a)
x3(t) = a31x1(t) + a32x2(t) + a33x3(t) + b3u(t)

y(t) = c1x1 (t) + c 2x2(t) + c3x3(t) + du(t) (2 .52b)
where u and y are the input and output ; xi , i = 1, 2, 3, are called the state variables ;
aid, b i, ci , and d are constants ; and i i (t) : = dxi(t)/dt. These equations are more often
written in matrix form as
x(t) = Ax(t) + bu(t) (State equation) (2 .53a)
y(t) = cx(t) + du(t) (Output equation) (2 .53b)
with
x1 all a12 a13 [b l ]
X = x2 A a21 a22 a23 b = b2 (2 .54a)
x3 ] a31 a32 a33 b3
and
C = [C1 C2 C31 (2 .54b)
The vector x is called the state vector or simply the state. If x has n state variables
or is an n X I vector, then A is an n X n square matrix, b is an n X 1 column
vector, c is a 1 X n row vector, and d is a 1 X 1 scalar . A is sometimes called the
system matrix and d the direct transmission part. Equation (2 .53a) describes the
relationship between the input and state, and is called the state equation . The state
equation in (2.53a) consists of three first-order differential equations and is said to
have dimension 3. The equation in (2 .53b) relates the input, state, and output, and
is called the output equation . It is an algebraic equation ; it does not involve differ-
entiation of x. Thus if x(t) and u(t) are known, the output y(t) can be obtained simply
by multiplication and addition .
Before proceeding, we remark on the notation . Vectors are denoted by boldface
lowercase letters ; matrices by boldface capital letters . Scalars are denoted by regular-
face lowercase letters . In (2.53), A, b, c, and x are boldface because they are either
vectors or matrices ; u, y, and d are regular face because they are scalars .
This text studies mainly single-variable systems, that is, systems with single
input and single output . For multivariable systems, we have two or more inputs
and/or two and more outputs . In this case, u(t) and y(t) will be vectors and the orders
of b and c and d must be modified accordingly . For example, if a system has three
inputs and two outputs, then u will be 3 X 1; y, 2 X 1 ; b, n X 3 ; c; 2 X n; and
d, 2 X 3. Otherwise, the form of state-variable equations remains the same .
The transfer function describes only the zero-state response of systems . Thus,
when we use the transfer function, the initial state or initial conditions of the system

must be assumed to be zero . In using the state-variable equation, no such assumption

is necessary . The equation is applicable even if the initial state is nonzero . The
equation describes not only the output but also the state variables . Because the state
variables reside within a system and are not necessarily accessible from the input
and output terminals, the state-variable equation is also called the internal descrip-
tion . Thus the state-variable description is more general than the transfer function
description . However, if a system is completely characterized by its transfer function
(Section 2.5), the two descriptions are essentially the same . This will be discussed
further in a later section .
In developing state-variable equations for LTIL systems, we must first choose
state variables. State variables are associated with energy . For example, the potential
and kinetic energy of a mass are stored in its position and velocity . Thus, position
and velocity can be chosen as state variables for a mass . For RLC networks, capac-
itors and inductors are energy storage elements because they can store energy in
their electric and magnetic fields . Therefore, all capacitor voltages and inductor
currents are generally chosen as state variables for RLC networks . Resistors will not
store energy; all energy is dissipated as heat . Therefore, resistor voltages or currents
are not state-variables . With this brief discussion, we are ready to develop state-
variable equations to describe LTIL systems .
Example 2.6 .1 (Mechanical system)

Consider the mechanical system shown in Figure 2.5(a) . As discussed in (2.4), its
differential equation description is
my(t) + k,y(t) + k2 y(t) = u(t) (2.55)
where u(t) is the applied force (input), y(t) is the displacement (output), y(t) : _
dy(t)/dt, and y(t) : = d2y(t)/dt2.
The potential energy and kinetic energy of a mass are stored in its position and
velocity ; therefore the position and velocity will be chosen as state variables . Define
x i (t) : = y(t) (2.56a)
and
x2 (t) : = y(t) (2 .56b)
Then we have
x,(t) = y(t) = x 2 (t)
This relation follows from the definition of xl(t) and x2 (t) and is independent of the
system. Taking the derivative of x2 (t) yields
x2(t) = At)
which becomes, after the substitution of (2 .55) and (2.56),

x2 (t) = [ - kzy(t) - kly(t) + u(t)l

m
k,
_ - k2 XI (t) - x2 (t) + 1 u(t)
m m m
These equations can be arranged in matrix form as
, 0 1 0
Cx2(t) '
-k2 _kl + i ] u(t)
[ (2.57a)
J
m m Cx2(t)J m
X I (t)
y(t) = [1 0] (2 .57b)
X2 (t)
This is a state-variable equation of dimension 2 . Note that the direct transmission d
in (2.53) is zero for this example .
We discuss in the following a procedure for developing state-variable equations

for RLC networks that contain a voltage or current source.
Procedure for Developing State-Variable Equations for RLC Networks

Step 1 : Assign all capacitor voltages and inductor currents as state variables . Write
down capacitor currents and inductor voltages as shown in Figure 2 .8. Note
that currents flow from high to low potential .
Step 2: Use Kirchhoff 's voltage and/or current laws to express every resistor's
voltage and current in terms of state variables and, if necessary, the input .
Step 3 : Use Kirchhoff 's voltage and/or current laws to develop a state equation .
This procedure offers only a general guide . For a more detailed and specific
procedure, see Reference [15]. We use an example to illustrate the procedure .
Example 2 .6 .2 (Electrical system)

Consider the RLC network shown in Figure 2 .21 . It consists of one resistor, one
capacitor, and one inductor. The input u(t) is a voltage source and the voltage across
the 3-H inductor is chosen as the output .
Step 1: The capacitor voltage x 1 (t) and the inductor currents x 2 (t) are chosen as
state variables. The capacitor current is 2x 1 (t), and the inductor voltage is
3x2 (t).
Step 2 : The current passing through the 4-fl resistor clearly equals x 2(t). Thus, the
voltage across the resistor is 4x 2(t). The polarity of the voltage must be
specified, otherwise confusion may occur .

452 3H
2F
2x
Step 3 : From Figure 2 .21, we see that the capacitor current 2x, equals x2(t), which
implies
1
x 1 (t) = x2 (t) (2 .58)
2
The voltage across the inductor is, using Kirchhoff 's voltage law,
3x2(t) = u(t) - x,(t) - 4x2 (t)
or
-3 u(t) (2 .59)
x2(t) _ XI(t) -
3 x2(t) +
3
Equations (2 .58) and (2 .59) can be arranged in matrix form as
1
0 0
)jt) 2 XI(t) +
u(t) (2 .60a)
[x2 (t)] -1 -4 [x2(t)] -
[1]
3
3 3
This is the state equation of the network . The output y(t) is
y(t) = 3x2
which is not in the form of cx + du. However the substitution of (2 .59)
yields
y(t) = - x1 (t) - 4x2(1) + u(t)
(2 .oOb)
= [-1 -4-1
[xi(t) 1 + u(t)
J
x2(t)
This is the output equation.

Exercise 2 .6 .1
Find state-variable descriptions of the networks shown in Figure 2 .22 with the state
variables and outputs chosen as shown .
1Q
2F
Figure 2.22 Networks .
[Answers : (a) x = - 0.5x - 0.5u, y = -x (b) x =

C- 1 -2] x + CI
U,
y = [1 0]x.]
2.7 SOLUTIONS OF STATE EQUATIONS-LAPLACE TRANSFORM METHOD
Consider the n-dimensional state-variable equation

i(t) = Ax(t) + bu(t) (2 .61 a)
y(t) = cx(t) + du(t) (2 .61 b)
where A, b, c, and d are respectively n X n, n X 1, 1 X n, and 1 X 1 matrices .

In analysis, we are interested in the output y(t) excited by an input u(t) and some
initial state x(0) . This problem can be solved directly in the time domain or indirectly
by using the Laplace transform . We study the latter in this section. The time-domain
method will be studied in the next subsection .
The application of the Laplace transform to (2 .61a) yields
sX(s) - x(0) = AX(s) + bU(s)
where X(s) = è[x(t) and U(s) _ E[u(t)] . This is an algebraic equation and can be
arranged as
(sI - A)X(s) = x(0) + bU(s) (2 .62)
where I is a unit matrix of the same order as A . Note that without introducing I,
(s - A) is not defined, for s is a scalar and A is an n X n matrix . The premultipli-
cation of (sI - A) -1 to (2.62) yields

2 .7 SOLUTIONS OF STATE EQUATIONS-LAPLACE TRANSFORM METHOD 51
X(s) = (sI - A) - 'x(0) + (sI - A)- 'bU(s) (2.63)

Zero-Input Zero-State
Response Response
This response consists of the zero-input response (the response due to nonzero x(0))
and the zero-state response (the response due to nonzero u(t)). If we substitute X(s)
into the Laplace transform of (2 .61b), then we will obtain
Y(s) = c(sI - A) - 'x(0) + [c(sI - A)- 'b + d]U(s) (2 .64)
Response Response
This is the output in the Laplace transform domain . Its inverse Laplace transform
yields the time response . Thus, the response of state-variable equations can be easily
obtained using the Laplace transform .
Example 2 .7 .1
Consider the state-variable equation
[-6 -3.5l x + [-1 u
x= 4 (2 .65a)
6 J
y = [4 5]x (2 .65b)
Find the output due to a unit-step input and the initial state x(0) = [-2 1 ]', where
the prime denotes the transpose of a matrix or a vector . First we compute
01 _ [-6 -4 .51 - Cs+6 6 s 3 .5 4 ~
sI-A= J
[0 J
Its inverse i s
I
s + 6 3.5
(sI - A) - ' =
-6 s-4
1 1 s - 4 -3 .5
(s+6)(s-4)+6 X 3 .5 6 s+6
Thus we have
1 s - 4 - 3.5 1 21
c(sI-A)-tx(0)=[4 5]sz +2s-3
1 6 s+6 [ - 11
s2+2s-3[4s+ 14 5s+ 16]~
1]
_ -3s-12
s2 +2s-3

and
G(s) : = c(sI - A) - 'b
1 s - 4 -3 .5 1
_ [4 5] s (2 .66)
2 +2s-3 6 s+6 1
_ s + 2
s2 +2s-3
Thus, using (2 .64) and U(s) = 1/s, the output is given by
_ -3s- 12 s + 2 1 _ - 3 s2 - lls+2
Ys
s2 +2s-3 + s 2 +2s-3 s (s- 1)(s+3)s
which can be expanded by partial fraction expansion as
-3 2/3 2/3
Ys =
s- 1 + s+ 3 s
Thus the output is
y(t) _ -3e` + 3 e 3t
3
for t?0.
This example shows that the response of state-variable equations can indeed be
readily obtained by using the Laplace transform .
2.7.1 Time-Domain Solutions

The solution of state-variable equations can also be obtained directly in the time
domain without using the Laplace transform . Following
t2 t"
eat = 1 +ta+-a2+ . . .+-a"+ . . .
2! n!
we define
t2 t"
eAr=I+tA+-A2+ . . .+-A"+ . . . (2 .67)
2! n!
where n! = n . (n - 1) . . . 2 . 1, A2 = AA, A3 = AAA, and so forth . If A is an
n X n matrix, then eAt is also an n X n matrix. If t = 0, then (2.67) reduces to
e = I (2 .68)
As in e(`-7) = ease - a,, we have
e A(i-T) = eAte- AT (2 .69)

2 .7 SOLUTIONS OF STATE EQUATIONS-LAPLACE TRANSFORM METHOD 53
which becomes, if -r = t,
eAre - At =e A'o =e =I (2 .70)
The differentiation of (2 .67) yields

z
deAt =0+A+ 2 A2 + t
3 l A3 + . . .
z z
=A(I+tA+ZI A2 + ) =(I+tA+ZA2+ .)A
which implies
d eAt = Ae At = e AtA (2 .71)
dt
Now using (2 .67) through (2 .71) we can show that the solutions of (2 .61) due to
u(t) and x(O) are given by
x(t) = e Atx(0) + e -AT bu(T)dT (2 .72)

ft
0
Zero-Input
and Response
Zero-State Response
y(t) = ce Atx(0) + ce At fo e -A Tbu(T)dT + du(t) (2 .73)
To show that (2 .72) is the solution of (2 .61 a), we must show that it meets the initial
condition and the state equation . Indeed, at t = 0, (2 .72) reduces to x(0) =
Ix(0) + I 0 x(0) . Thus it meets the initial condition . The differentiation of
(2.72) yields
z(t) _ CeAt x(0) + e fo e - `'Tbu(T)dT]
AeAtx(0) + d (e At) fo e -ATbu(T)dT) + eA` dt (fo e ATbu(T)dT)
A Ce At x(0) + eAt ft e -ATbu(T)dr] + [e _ATbu(T)]

T=t
Ax(t) + e Are -At bu(t) = Ax(t) + bu(t)
This shows that (2 .72) is the solution of (2 .61a) . The substitution of (2 .72) into
(2.61b) yields immediately (2 .73). If u = 0, (2 .61a) reduces to
i(t) = Ax(t)
This is called the homogeneous equation . Its solution due to the initial state x(0) is,
from (2 .72),
x(t) = eAtx(0)

If u(t) = 0 or, equivalently, U(s) = 0, then (2 .63) reduces to

X(s) _ (sI- A) - 'x(0)
A comparison of the preceding two equations yields
E(eA t ) = ( sI - A) - ' or eA` = E - '[(sI - A) - '] (2.74)
Thus eA` equals the inverse Laplace transform of (sI - A) - ' .
To use (2.72), we must first compute eA`. The computation of eA` by using the
infinite power series in (2 .67) is not simple by hand . An alternative method is to use
the Laplace transform in (2.74) . Computer computation of (2.67) and (2 .72) will be
discussed in Section 2 .9.
2 .7 .2 TRANSFER FUNCTION AND CHARACTERISTIC POLYNOMIAL
In this section, we discuss the relationships between state-variable equations and

transfer functions . The transfer function of a system is, by definition, the ratio of the
Laplace transforms of the output and input under the assumption that all initial
conditions are zero . The Laplace transform of the state-variable equation in (2 .61)
was computed in (2.64) . If x(O) = 0, then (2 .64) reduces to
Y(s) = [c(sI - A) - 'b + d]U(s)
Thus the transfer function of the state-variable equation in (2 .61) is
G(s) = Y(s) = c(sI - A) - 'b + d (2 .75)
This formula was used in (2 .66) to compute the transfer function of (2.65) . Now we
discuss its general property . Let det stand for the determinant and adj for the adjoint
of a matrix . See Appendix B . Then (2 .75) can be written as
1
(sI - A)]b + d
G(s) = c det(sI - A) [adj
We call det (sI - A) the characteristic polynomial of A . For example, the charac-
teristic polynomial of the A in (2 .65) is
s + 6 3.5
0(s) : = det (sI - A) = det
-6 s - 4 (2 .76)
=(s+6)(s-4)-(-3 .5x6)=s2+2s-3
We see that if A is n X n, then its characteristic polynomial is of degree n and has
n roots . These n roots are called the eigenvalues of A . The characteristic polynomial
and eigenvalues of A are in fact the same as the characteristic polynomial and modes
discussed in Section 2 .3.1 . They govern the zero-input response of the system . For

2 .8 TRANSFER FUNCTION AND CHARACTERISTIC POLYNOMIAL 55
example, if the characteristic polynomial of A is

L(s) = s 3 (s + 2)(s - 3)
then the eigenvalues are 0, 0, 0, -2, and 3, and every entry of e A` is a linear
combination of the time functions 1, t, t2, e- 2t, and e 3` . Therefore any zero-input
response will be a linear combination of these time functions .
The characteristic polynomial of square matrices often requires tedious com-
putation . However, the characteristic polynomials of the following two matrices
-a, -a 2 -a3 -a4 0 1 0 0
1 0 0 0 0 0 1 0
(2 .77)
0 1 0 0 0 0 0 1
0 0 1 0 -a4 -a 3 -a2 -a,
and their transposes are all
0(s) =
S4 + als 3 + a2 S2 + a3s + a4 (2 .78)
This characteristic polynomial can be read out directly from the entries of the mat-
rices in (2.77). Thus, these matrices are called companion forms of the polynomial
in (2.78).
Now we use an example to discuss the relationship between the eigenvalues of
a state-variable equation and the poles of its transfer function .
Example 2 .8.1
Consider the state-variable equation in (2.65), repeated in the following
1
x= 66 -3.5l x +
C 4
J
1 u (2 .79a)
y = [4 5]x (2 .79b)
Its characteristic polynomial was computed in (2 .76) as
0(s)=s2+2s-3=(s+3)(s- 1)
Thus its eigenvalues are 1 and - 3 . The transfer function of (2 .79) was computed
in (2.66) as
s + 2 s + 2
G(s) =
s2 +2s-3 (s+3)(s- 1)
Its poles are 1 and - 3 . The number of the eigenvalues of (2 .79) equals the number
of the poles of its transfer function .

Example 2 .8.2
.51 x+ [-I]
x= [-6 - 3 u (2.80a)
J
y = [-2 -1]x (2.80b)
Equation (2 .80a) is identical to (2 .79a) ; (2 .80b), however, is different from (2.79b) .
The characteristic polynomial of (2 .80) is
0(s)=s 2 +2s-3
Its eigenvalues are -3 and 1 . The transfer function of (2 .80) is, replacing [4 5] in
(2.66) by [-2 - 1],
1
G(s) = [-2 -1] 2 +
C
2s-3 s 6 4 s + 611 11
_ 1 s+0.5 2s- 1 -s
-1][ .81)
s2+2s-3[-2 s =s2+2s-3 (2
s - 1
(s - 1)(s + 3)
Thus the transfer function of (2 .80) is G(s) = 1/(s + 3) . It has only one pole at
- 3, one less than the number of the eigenvalues of (2.80).
A state-variable equation is called minimal if the number of its eigenvalues or

the dimension of the equation equals the number of the poles of its transfer function .
The state-variable equation in (2 .79) is such an equation . Every eigenvalue of a
minimal state-variable equation will appear as a pole of its transfer function and
there is no essential difference between using the state-variable equation or its trans-
fer function to study the system . Thus, the state-variable equation is said to be
completely characterized by its transfer function . As will be discussed in Chapter
11, every minimal state-variable equation has the properties of controllability and
observability. If a state-variable equation is not minimal, such as the one in (2.80),
then some eigenvalues will not appear as poles of its transfer function and the transfer
function is said to have missing poles. In this case, the state-variable equation is not
completely characterized by its transfer function and we cannot use the transfer
function to carry out analysis and design . This situation is similar to the one in
Section 2 .5 .

2 .9 DISCRETIZATION OF STATE EQUATIONS 57
2 .8 DISCRETIZATION OF STATE EQUATIONS

i(t) = Ax(t) + bu(t) (2 .82a)
y(t) = cx(t) + du(t) (2 .82b)
This equation is defined at every instant of time and is called a continuous-time
equation. If a digital computer is used to generate the input u, then u(t) is (as will
be discussed in Chapter 12) stepwise, as shown in Figure 2 .23(a) . Let
u(t) = u(k) for kT < t < (k + 1)T, k = 1, 2, 3, . . . (2.83)
where T is called the sampling period. This type of signal is completely specified
by the sequence of numbers u(k), k = 0, 1, 2, . . . , as shown in Figure 2 .23(b) . The
signal in Figure 2 .23(a) is called a continuous-time or analog signal because it is
defined for all time ; the signal in Figure 2 .23(b) is called a discrete-time signal
because it is defined only at discrete instants of time . Note that a continuous-time
signal may not be a continuous function of time as shown in the figure.
If we apply a stepwise input to (2.82), the output y(t) is generally not stepwise .
However, if we are interested in only the output y(t) at t = kT, k = 0, 1, 2, . . . , or
y(k) := y(kT), then it is possible to develop an equation simpler than (2 .82) to
describe y(k). We develop such an equation in the following .
The solution of (2.82) was developed in (2 .72) as
x(t) = eA`x(0) + e A` fo e -A Tbu(T)dT (2.84)
This equation holds for any t-in particular, for t at kT. Let x(k) : = x(kT) and
x(k + 1) : = x((k + 1)T) . Then we have
kT
x(k) = eAkTx(0) + e AkT fo e -A 'bu(T)dT (2 .85)
(k+ 1)T
x(k + 1) = eA (k+ 1) Tx(0) + e A(k+1)T jo e - ATbu(T)dT (2.86)
u (t) u (k)
1
3 4 , , )- t 1 3
1 4
1 k
0 1 2 9
- 5 6 0 1 2 5 6
Figure 2 .23 (a) Stepwise continuous-time signal. (b) Discrete-time signal .

We rewrite (2 .86) as
kT
x(k + 1) = eAT Ak
[eTx(0) + eAkT e -A Tbu(T)dT
f0 (2 .87)
(k+ 1)T
+ e A((k+1)T- T) bu(T)dT
J kT
The term inside the brackets equals x(k) . If the input u(t) is stepwise as in (2 .83),
then u(T) in the integrand of the last term of (2.87) equals u(k) and can be moved
outside the integration . Define a = (k + 1)T - T. Then da = -dr and (2 .87)
becomes
x(k + 1) = e ATx(k) - ( J0
T eAada) bu(k)
(2 .88)
= eATx(k) eAada) bu(k)

+ ( fo
This and (2 .82b) can be written as
x(k + 1) = Axk + 6u(k) (2 .89a)
y(k) = cx(k) + du(k) (2 .89b)
with
T
A = e AT b= eA TdT b e= c d = d (2 .89c)
f0
This is called a discrete-time state-variable equation . It consists of a set of first-

order difference equations . We use an example to illustrate how to obtain such an
equation .
Example 2 .9 .1
Consider the continuous-time state-variable equation

_ 1 1 0
x 0- 1 x+ 1 u (2 .90a)
y = [2 1]x (2 .90b)
Discretize the equation with sampling period T = 0 .1 .

2 .9 DISCRETIZATION OF STATE EQUATIONS 59
First we compute
-i
~[eA`] _ (sI - A) - '
[ 0 s + 1
1 1
S + 1 (s + 1)2
(s + 1)2 [ 0 s 1]
0 l
S + 1
Its inverse Laplace transform is, using Table A . 1,
e At = [e - ` to - `]
0 e -`
Using an integration table, we can compute
= 1- e -T I- (I + T)e -T
'4 rdr
0T ef 0 1-e -T
If T = 0.1, then
e 0.1A = [ 0 .9048 0.0905] / 0.0047
0 0.9048
and I fo
\
0'1
eA TdT) b =
1
0.0952
(2 .91)
Thus the discrete-time state-variable equation is
x(k + 1) =
1 0.9048 0 .0905 x(k) +
0.0047
u(k) (2 .92a)
0 0 .9048 0 .0952
y(k) = [2 1]x(k) (2 .92b)
This completes the discretization . This discretized equation is used on digital com-
puters, as will be discussed in Chapter 5, to compute the response of the continuous-
time equation in (2.90).
We have used the Laplace transform to compute eAT in (2.91). If T is small, the
infinite series in (2.67) may converge rapidly . For example, for eAT in (2.91) with
T = 0.1, we have
I + TA + (TA)(TA) _
2 [ 0 0 50 0.9050]
and
1 1 0.9048 0.0905
I + TA + 2 (TA)(TA) + (TA)(TA)(TA) =
6 0 0.9048]

The infinite series converges to the actual values in four terms . Therefore for small
T, the infinite series in (2 .67) is a convenient way of computing eAT. In addition to
the Laplace transform and infinite series, there are many ways (at least 17) of com-
puting eAT. See Reference [47]. In Chapter 5, we will introduce computer software
to solve and to discretize state-variable equations .
PROBLEMS
2 .1 . Consider the pendulum system shown in Figure P2 .1, where the applying force
u is the input and the angular displacement 0 is the output . It is assumed that
there is no friction in the hinge and no air resistance of the mass . Is the system
linear? Is it time-invariant? Find a differential equation to describe it . Is the
differential equation LTIL?
U <
0/
m Figure P2 .1
2 .2 . Consider the pendulum system in Figure P2 .1 . It is assumed that sin 0 and cos
0 can be approximated by 0 and 1, respectively, if 101 < 7r/4 radians . Find an
LTIL differential equation to describe the system .
2 .3 . a . Consider the system shown in Figure P2 .3(a) . The two blocks with mass
ml and m 2 are connected by a rigid shaft . It is assumed that there is no
friction between the floor and wheels . Find a differential equation to de-
scribe the system .
b . If the shaft is long and flexible, then it must be modeled by a spring with
spring constant k2 and a dashpot . A dashpot is a device that consists of a
U
m ~ 7.~~rl7TTTTTi m 2
y y
(a) (b)
Figure P2 .3

PROBLEMS 61
piston and oil-filled cylinder and is used to provide viscous friction . It is

assumed that the force generated by the dashpot equals kl(dy2/dt - dy 1 ldt) .
Verify that the system is described by
d
m2 - k 1 (ddot) - ddtt)) - k2(v2(t) - Y1(t))
dt2t) = u(t)
and
d
m1 dt2 = -ki (~dtt) dtt)) - k2(Yi(t) - Y2(t))
2 .4 . If a robot arm is long, such as the one on a space shuttle, then its employment
must be modeled as a flexible system, as shown in Figure P2 .4. Define
T: Applied torque
k1 : Viscous-friction coefficient
k2 : Torsional spring constant
J,: Moment of inertia
B;: Angular displacement
Find an LTIL differential equation to describe the system .
02 Figure P2 .4
2 .5. Find LTIL differential equations to describe the networks in Figure P2 .5.
(a) (b)
Figure P2 .5
2 .6 . Consider the network shown in Figure P2 .6(a), in which T is a tunnel diode

with characteristics as shown . Show that if v lies between a and b, then the
circuit can be modeled as shown in Figure P2 .6(b). Show also that if v lies
between c and d, then the circuit can be modeled as shown in Figure P2 .6(c),

111111
(a)
L ii
000
(b) (c)
Figure P2 .6
where i,' = i t - i0, v' = v - v 0. These linearized models are often called
linear incremental models, or small-signal models .
2.7. Consider the LTIL differential equation
d2y(t) - 2 dy(t) - 3y(t) = 2 du(t) + u(t)

dt dt dt
a. Find its zero-input response due to y(0 - ) = 1 and y(0 - ) = 2 .
b. What are its characteristic polynomial and modes?
c. Find the set of all initial conditions such that the mode 3 will not be excited .
d. Find the set of all initial conditions such that the mode -1 will not be
excited .
e. Plot (c) and (d) on a two-dimensional plane with y and jy as coordinates . If
a set of initial conditions is picked randomly, what is the probability that
the two modes will be excited?
2.8. Consider the differential equation in Problem 2 .7
a . Find its unit-step response, that is, the zero-state response due to a unit-step
input u(t) .
b. What are its transfer function, its poles and zeros?
c. Find a bounded input that will not excite pole 3 .
d. Find a bounded input that will not excite pole - 1 .
2.9. Consider the differential equation in Problem 2 .7. Find the response due to
y(0 - ) = l, y(0 -) = 2, and u(t) = 2, fort ? 0.

PROBLEMS 63
2 .10. Find the transfer function from u to 0 of the system in Problem 2.2.
2 .11 . Find the transfer functions for the networks in Figure P2.5 . Compute them
from differential equations . Also compute them using the concept of Laplacian
impedances.
2 .12 . Find the transfer functions from u to y, u to y l, and u to y2 of the systems in
Figure P2.3.
2 .13 . Consider a system . If its zero-state response due to a unit-step input is measured
e-2r + sin t, where is the transfer function of the system?
as y(t) = 1 -
2 .14 . Consider a system. If its zero-state response due to u(t) = sin 2t is measured
as y(t) = e -` - 2e -3t + sin 2t + cos 2t, what is the transfer function of the
system?
2 .15 . Consider the LTIL system described by
d2y(t) - 3y(t) = du(t)

+ 2 dy(t) - u(t)
dt dt dt
a. Find the zero-input response due to y(0 - ) = 1 and y(0- ) = 2 . What are
the modes of the system?
b. Find the zero-state response due to u(t) = 1, for t ? 0.
c. Can you detect all modes of the system from the zero-state response?
d. Is the system completely characterized by its transfer function? Will a seri-
ous problem arise if we use the transfer function to study the system?
2 .16. Repeat Problem 2.15 for the differential equation
d2y(t) - 3y(t) = du(t)

+ 2 dy(t) + 3u(t)
dt dt dt
2.17. Consider a transfer function G(s) with poles -1 and - 2 j 1, and zero 3 .
Can you determine G(s) uniquely? If it is also known that G(2) _ -0.1, can
you now determine G(s) uniquely?
2.18. Find the unit-step response of the following systems and plot roughly the
responses :
2
a. G 1 (s) _
(s + 1)(s + 2)
20(s + 0 .1)
b. G 2(s) =
(s + 1)(s + 2)
0.2(s + 10)
C . G3(S) _ (
s + 1)(s + 2)
-20(s - 0.1)
d. G4(s)
(s + 1)(s + 2)
-0.2(s - 10)
e . G 5 (s) =
(s + 1)(s + 2)
Which type of zeros, closer to or farther away from the origin, has a larger
effect on responses?
2.19. Find the poles and zeros of the following transfer functions :
2(s 2 - 9)(s + 1)
a. G(s) = (s +
3)2(s + 2)(s - 1)2
10(s 2 -s+ 1)
b. G(s) _
s +2s3 +s+2
2S2 +8S+8
c. G(s) =
(S + 1)(S2 + 2s + 2)
2.20 . a. Consider a system with transfer function G(s) = (s - 2)/s(s + 1) . Show

that if u(t) = e, t ? 0, then the response of the system can be decomposed
as
Total response = Response due to the poles of G(s)
+ Response due to the poles of U(s)
b. Does the decomposition hold if u(t) = 1, for t ? 0?
c. Does the decomposition hold if u(t) = e 2`, for t ? 0?
d . Show that the decomposition is valid if U(s) and G(s) have no common
pole and if there are no pole-zero cancellations between U(s) and G(s) .
2.21 . Show that the network in Figure 2 .18 is described by the differential equation
in (2.42).
2.22. Consider the simplified model of an aircraft shown in Figure P2 .22 . It is as-
sumed that the aircraft is dynamically equivalent at the pitched angle 0 0, ele-
vator angle u0, altitude ho, and cruising speed v o. It is assumed that small
deviations of 0 and u from 00 and uo generate forces f 1 = k, 0 and f2 = k2 u,
as shown in the figure . Let m be the mass of the aircraft; I, the moment of
PP '
Figure P2.22

PROBLEMS 65
inertia about the center of gravity P ; b9, the aerodynamic damping and h, the
deviation of the altitude from ho. Show that the transfer function from u to h
is, by neglecting the effect of I,
k,k2 12 - k2bs
G(s) =
ms (bs + k,l,)
2.23. Consider a cart with a stick hinged on top of it, as shown in Figure P2 .23 . This
could be a model of a space booster on takeoff . If the angular displacement B
of the stick is small, then the system can be described by
B=B+u
y=0B-u
where f3 is a constant, and u and y are expressed in appropriate units . Find the
transfer functions from u to 0 and from u to y. Is the system completely char-
acterized by the transfer function from u to y? By the one from u to 0? Which
transfer function can be used to study the system?
Figure P2.23
2.24. Show that the two tanks shown in Figure P2 .24(a) can be represented by the
block diagram shown in Figure P2 .24(b) . Is there any loading problem in the
block diagram? The transfer function of the two tanks shown in Figure 2 .12
is computed in Exercise 2 .4.2 . Is it possible to represent the two tanks in Figure
2.12 by two blocks as in Figure P2 .24(b) on page 66? Give your reasons .
2.25. Find state-variable equations to describe the systems in Problems 2 .1 and 2 .2.
2.26. Find state-variable equations to describe the systems in Figure P2 .3 .
2.27. Find state-variable equations to describe the networks in Figure P2 .5 .
2.28. The soft landing phase of a lunar module descending on the moon can be
modeled as shown in Figure P2 .28. It is assumed that the thrust generated is
proportional to rn, where m is the mass of the module . The system can be
described by my = - kth - mg, where g is the gravity constant on the lunar
surface . Define the state variables of the system as x, = y, x2 = y, x3 = m,
and u = m. Find a state-variable equation to describe the system . Is it a time-
invariant equation?

9,
1
9I
h2 R2
AZ ~ ~ q2
(a)
R; 1 q, 1 92
-~ A I R I s+1 A 2 R 2 s+ 1
(b) Figure P2 .24
/"' J / Lunar surface Figure P2 .28
2 .29. Use the Laplace transform to find the response of
[3 -2]
x+ 1] u
C
y = [4 5]x
due to x(O) = [1 2]' and u(t) = 1 for t ? 0.

PROBLEMS 67
2.30 . Use the infinite series in (2 .67) to compute eA` with
A =
[ 0 - 02]
Verify your results by using (2.74).
2.31 . Compute the transfer functions of the following state-variable equations :
a.x= -x+u y = 2x
b. x= 0 - 0 ]x+ u
L [ I]
y = [2 - 2]x
3 1 2
c.x= x+ u
-2 0 4
Y = [l O]x
-3 1 0 [2
d 'x= -2 0 0 + 4 u
1 0 0 1
Y = [1 0 0]x
Do they have the same transfer functions? Which are minimal equations?
2.32. Find the characteristic polynomials for the matrices

-a l 1 0 0 0 -a3
- a2 0 1 1 0 -a 2
-a3 0 0 0 1 -a1
2.33. Find the outputs of the equations in Problem 2 .31 due to a unit-step input and
zero initial state .
2.34. Compute the transfer functions of the state-variable equations in Problems
2.25(b), 2 .26(a), and 2 .27(a). Are they minimal state-variable equations?
2.35. Compute y(k) for k = 0, 1, 2, 3, 4, for the discrete-time state-variable equation
x(k + 1) _ [-2 x(k) + u(k)

10] C4]
y(k) = [1 0]x(k)
due to x(0) = [ 1 - 2]' and u(k) = 1 for all k ? 0.

2.36. Discretize the continuous-time equation with sampling period 0 .1 second
x u
[ -2 0] x + [4]
y = [1 0]x
2.37 . a. The characteristic polynomial of
A=
-6 -3.5]
C 6 4
is computed in (2 .76) as
0(s)=s 2 +2s-3
Verify
0(A) = A 2 + 2A - 31 = 0
This is called the Cayley-Hamilton Theorem . It states that every square
matrix meets its own characteristic polynomial .
b. Show that A2, A3, . . . , can be expressed as a linear combination of I and
A. In general, for a square matrix A of order n, Ak fork ? n can be expressed
as a linear combination of I, A, A 2, . . . , A" -1
2.38. Show that if (2 .67) is used, b in (2 .89c) can be expressed as
T 2 3
b = fo e'TdT b - T (I + 2 A + 31 A2 + 4~ A3 + . . . b
This can be used to compute b on a digital computer if it converges rapidly or

if T is small .
Development of
Block Diagrams
for Control Systems
3.1 INTRODUCTION
The design of control systems can be divided into two distinct parts . One is con-
cerned with the design of individual components, the other with the design of overall
systems by utilizing existing components . The former belongs to the domain of
instrumentation engineers ; the latter, the domain of control engineers . This is a con-
trol text, so we are mainly concerned with utilization of existing components . Con-
sequently, our discussion of control components stresses their functions rather than
their structures .
Control components can be mechanical, electrical, hydraulic, pneumatic, or opti-
cal devices . Depending on whether signals are modulated or not, electrical devices
again are divided into ac (alternating current) or dc (direct current) devices . Even a
cursory introduction of these .devices can easily take up a whole text, so this will
not be attempted . Instead, we select a number of commonly used control compo-
nents, discuss their functions, and develop their transfer functions . The loading prob-
lem will be considered in this development . We then show how these components
are connected to form control systems . Block diagrams of these control systems are
developed . Finally, we discuss manipulation of block diagrams . Mason's formula is
introduced to compute overall transfer functions of block diagrams .
69

70 CHAPTER 3 DEVELOPMENT OF BLOCK DIAGRAMS FOR CONTROL SYSTEMS
3.2MOTORS
Motors are indispensable in many control systems . They are used to turn antennas,
telescopes, and ship rudders ; to close and open valves ; to drive tapes in recorders,
and rollers in steel mills ; and to feed paper in printers . There are many types of
motors : dc, ac, stepper, and hydraulic . The magnetic field of a dc motor can be
excited by a circuit connected in series with the armature circuit ; it can also be excited
by a field circuit that is independent of the armature circuit or by a permanent magnet .
We discuss in this text only separately excited dc motors . DC motors used in control
systems, also called servomotors, are characterized by large torque to rotor-inertia
ratios, small sizes, and better linear characteristics . Certainly, they are more expen-
sive than ordinary dc motors.
3.2.1 Field-Controlled DC Motor

Most of the dc motors used in control systems can be modeled as shown in Figure
3.1 . There are two circuits, called the field circuit and armature circuit . Let i f and is
be the field current and armature current, respectively . Then the torque T generated
by the motor is given by
T(t) = kia(t)if (t) (3.1)
where k is a constant . The generated torque is used to drive a load through a shaft .
The shaft is assumed to be rigid . To simplify analysis, we consider only the viscous
friction between the shaft and bearing . Let J be the total moment of inertia of the
load, the shaft, and the rotor of the motor ; 0, the angular displacement of the load ;
and f, the viscous friction coefficient of the bearing . Then we have, as developed
in (2.5),
z
T(t) = J (3 .2)
d dtzt) + f d d t)
This describes the relationship between the motor torque and load's angular
displacement .
If the armature current is is kept constant and the input voltage u(t) is applied
Armature circuit
Figure 3 .1 DC motor.

3.2 MOTORS 71
to the field circuit, the motor is called a field-controlled dc motor. We now develop
its block diagram .
In the field-controlled dc motor, the armature current ia(t) is constant . Therefore,
(3 .1) can be reduced to
T(t) = ki a(t)if(t) _ : kfif(t) (3.3)
where kf = kia. From the field circuit, we have
dif(t)
Lf dt + Rf if (t) = vf (t) = u(t) (3.4)
The application of the Laplace transform to (3.3) and (3.4) yields, 1 assuming zero
initial conditions,
T(s) - kflf(s)
LfsIf(s) + RfIf(s) = U(s)
which imply
1
If (s) = L f s + R f U(s)
and
k
T(s) = L
fs + R f U(s)
Thus, if the generated torque is considered as the output of the motor, then the
transfer function of the field-controlled dc motor is
T(s) kf
Gm(s) : (3 .5)
U(s) Lfs + Rf
This transfer function remains the same no matter what load the motor drives . Now
we compute the transfer function of the load . The application of the Laplace trans-
form to (3.2) yields, assuming zero initial conditions,
T(s) = Js20(s) + fs0(s)
which can be written as
0(s) 1
G, (s)
' T(s) s(Js + f)
This is the transfer function of the load if we consider the motor torque the input,
and the load's angular displacement the output . Thus, the motor and load in
'Capital letters are used to denote the Laplace transforms of the corresponding lowercase letters . In the
case of T(t) and T(s), we use the arguments to differentiate them .

Motor Load
k T
1 e
-~ f
L s+R . f s(Js+f)
Figure 3.2 Block diagram of field-controlled dc motor .
Figure 3 .1 can be modeled as shown in Figure 3 .2. It consists of two blocks : One
represents the motor; the other, the load . Note that any change of the load will not
affect the motor transfer function, so there is no loading problem in the connection .
The transfer function of the field-controlled dc motor driving a load is the prod-
uct of Gm (s) and G1 (s) or
kflRf.f
G(s) = Gm(S)GZ(S) =S(Lfs + Rf)(JS kf =
+ f) S(TfS + 1)(Tm s + 1)
(3.6)
_ m k
(Tfs + 1)S(Tm S + 1)
where Tf : = L f/R f, Tm : = J/ f and km : = kf/R ff. The constant Tf depends only on

the electric circuit and is therefore called the motor electrical time constant . The
constant Tm depends only on the load and is called the motor mechanical time con-
stant . The physical meaning of the time constant will be discussed in the next chapter .
If the electrical time constant Tf is much smaller than the mechanical time constant
Tm, as is often the case in practice, then G(s) is often approximated by
G(s) (3 ~)
S(T S m + 1)
For a discussion of this type of approximation, see Reference [18]. Note that Tm
depends only on the load.
The field-controlled dc motor is used to drive a load, yet there is no loading
problem in Figure 3 .2. How could this be possible? Different loads will induce
different ia ; even for the same load, is will not be constant. Therefore, the loading
problem is eliminated by keeping is constant. In practice, it is difficult to keep is
constant, so the field-controlled dc motor is rarely used .
3.2 .2 Armature-Controlled DC Motor

Consider the dc motor shown in Figure 3 .1 . If the field current if (t) is kept constant
or the field circuit is replaced by a permanent magnetic field, and if the input voltage
is applied to the armature circuit, then the motor is called an armature-controlled
dc motor. We now develop its transfer function . If if (t) is constant, (3 .1) can be
written as
T(t) = kr i a (t) (3.8)
where kt := ki f (t) is a constant . When the motor is driving a load, a back electro-
motive force (back emf) voltage v, will develop in the armature circuit to resist the

3 .2 MOTORS 73
applied voltage . The voltage v b(t) is linearly proportional to the angular velocity of
the motor shaft :
dO(t)
Vb(t) = kb dt (3.9)
Thus the armature circuit in Figure 3 .1 is described by

dta(t)
Rai a(t) + La + vb(t) = va(t) = u(t) (3 .10)
or
a(t) + k de(t) = u(t)
R i (t) + L di b dt (3 .11)
a a dt
Using (3.2), (3 .8), and (3.11), we now develop a transfer function for the motor . The
substitution of (3 .8) into (3.2) and the application of the Laplace transform to (3 .2)
and (3 .11) yield, assuming zero initial conditions,
kt la(s) = Js 2 0(s) + fs0(s) (3 .12)
Rala(s) + La sla(s) + kbs0(s) = U(s) (3 .13)
The elimination of Ia from these two equations yields

0(s) _ kr
G(s) (3 .14)
= U(s) s[(Js + f)(Ra + Las) + krkb]
This is the transfer function from v a = u to 0 of the armature-controlled dc motor .
Using (3.8), (3 .9), (3 .10), and (3.12), we can draw a block diagram for the
armature-controlled dc motor as shown in Figure 3.3 . Although we can draw a block
for the motor and a block for the load, the two blocks are not independent as in the
case of field-controlled dc motors . A signal from inside the load is fed back to the
input of the motor. This essentially takes care of the loading problem . Because of
the loading, it is simpler to combine the motor and load into a single block with the
transfer function given in (3.14).
Motor Load
I r
I I I
U kr I T I 1
doldt 1 I B
R a +L a s I I
Js+f S
I
I I I I
L - -J
I
kb c
Figure 3 .3 Block diagram of armature-controlled dc motor.

In application, the armature inductance La is often set to zero .2 In this case,

(3.14) reduces to
E )(s) kt km
G(s) = (3 .15)
U(s) s(JRas + kkb + fRa) s(Tm s + 1)
where
: motor gain constant (3 .16)

km : k t kb +t fRa
and
JR a _
TM : =
k t kb: motor time constant (3 .17)
fR a +
The time constant in (3 .17) depends on the load as well as the armature circuit . This
is in contrast to (3.7) where Tm depends only on the load . The electrical and me-
chanical time constants are well defined in (3.6) and (3.7), but this is not possible
in (3 .14) and (3 .15). We often call (3 .14) and (3 .15) motor transfer functions, al-
though they are actually the transfer functions of the motor and load . The transfer
function of the same motor will be different if it drives a different load .
Exercise 3.2 .1
Consider an armature-controlled dc motor with

R a = 20 f1 kt = 1 N.m/A kb = 3 V.s/rad
If the motor is used to drive a load with moment of inertia 2 N.m.s2/rad and with
viscous friction coefficient 0.1 N .m.s/rad, what is its transfer function? If the mo-
ment of inertia of the load and the viscous friction coefficient are doubled, what is
its transfer function? Do the time and gain constants equal half of those of the first
transfer function?
[Answers : 0.2/s(8s + 1), 0.14/s(11 .4s + 1), no .]
The input and output of the transfer function in (3 .15) are the applied electrical
voltage and the angular position of the motor shaft as shown in Figure 3 .4(a). If
motors are used in velocity control systems to drive tapes or conveyers, we are no
longer interested in their angular positions 0 . Instead we are mainly concerned with
2In engineering, it is often said that the electrical time constant is much smaller than the mechanical time
constant, thus we can set Lo = 0. If we write (3.14) as G(s) = k,/s(as + 1)(bs + 1), then a and b are
called time constants . Unlike (3.6), a and b depend on both electrical and mechanical parts of the motor .
Therefore, electrical and mechanical time constants in armature-controlled dc motors are no longer as
well defined as in field-controlled dc motors .

3 . 2 MOTORS 75
Motor and load Motor and load

Input Angular Input Angular
voltage km
displacement voltage km
velocity
S (2n S + 1) mS+l
(a) (b)
Figure 3 .4 Transfer functions of armature-controlled dc motor .
their angular velocities . Let w be the angular velocity of the motor shaft . Then we
have w(t) = d6/dt and
km
W(s) = s0 (S) = S - 1) U(S) U(s)
S(Tm s + TmS + 1 .
The transfer function of the motor from applied voltage u to angular velocity w is
W(s) km
(3.18)
U(S) Tm S + 1
as shown in Figure 3 .4(b) . Equation (3 .18) differs from (3 .15) by the absence of one
pole at s = 0 . Thus the transfer function of motors can be (3 .15) or (3 .18), depending
on what is considered as the output . Therefore it is important to specify the input
and output in using a transfer function . Different specifications lead to different
transfer functions for the same system .
3.2.3 Measurement of Motor Transfer Functions

The transfer function of a motor driving a load is given by (3 .15) . To use the formula,
we must use (3.16) and (3 .17) to compute km and Tm which, in turn, require k1, kn,
Ra, La, J, and f. Although the first four constants may be supplied by the manufac-
turer of the motor, we still need J and f. The moment of inertia J of a load is not
easily computable if it is not of regular shape . Neither can f be obtained analytically .
Therefore, we may have to obtain J and f by measurements. If measurements are to
be carried out, we may as well measure the transfer function directly . We discuss
one way of measuring motor transfer functions in the following .
Let the transfer function of a motor driving a load be of the form shown
in (3.15) . If we apply a voltage and measure its angular velocity, then the trans-
fer function reduces to (3 .18) . Let the applied voltage be a; the velocity can then be
computed as
W(s) = km a _ kma - kmaTm
Tm S + 1 S S Tm S + 1
Its inverse Laplace transform is

w(t) = kma - kmae - `/ T^ (3.19)
for t ? 0. Because Tm is positive, the term e-`/ m approaches zero as t approaches
infinity and the response w(t) is as shown in Figure 3 .5(a) . Thus we have
w(o) = kma (3.20)

W (t) R(t)
>- t
0
w(m)
I
0
-t
(a) (b)
Figure 3.5 (a) Step response of motor . (b) Estimation of T,,, .
This is called the final speed or steady-state speed. If we apply a voltage of known
magnitude a and measure the final speed w(-), then the motor gain constant can
be easily obtained as km = w(oo)/a . To find the motor time constant, we rewrite
(3.19) as
w(t) = kma(1 - e tl m)
which implies, using (3.20),
e -tlm = 1 - w(t) = 1 - w(t)
kma w(-)
Thus we have
- T = In 1 - w(~)) : R(t) (3 .21)
TM
where In stands for the natural logarithm . Now if we measure the speed at any t, say
t = to, then from (3 .21) we have
to
TM = (3.22)
W(to ))
In (1 -
W(- )
Thus the motor time constant can be obtained from the final speed and one additional
measurement at any t.
Example 3 .2 .1
Consider an armature-controlled dc motor driving a load . We apply 5 V to the motor.
The angular speed of the load at to = 2 s is measured as 30 rad/s and the final speed
is measured as 70 rad/s . What are the transfer functions from the applied voltage to
the speed and from the applied voltage to the displacement?
From (3 .20) and (3 .22)~we have
k = w() = 70 = 14
m a 5

3 .2 MOTORS 77
and
-2 -2 -2
TM = 3.57
30 in 0.57 - 0.5596
1 - 70
Thus the transfer functions are
W(s) _ 14 0(s) _ 14
U(s) 3 .57s + 1 U(s) s(3 .57s + 1)
Exercise 3.2.2
Consider an armature-controlled dc motor driving a load . We apply 10 volts to the

motor. The velocity of the load is measured as 20 cycles per second at t = 5 s, and
its steady-state velocity is measured as 30 cycles per second . What is its transfer
function from the applied voltage u to the angular displacement 0(t)? What is its
unit-step response? What is 0(cc)? What does it mean physically?
[Answers: G(s) = 18 .84/s(4 .55s + 1) = 4.14/s(s + 0.22), 0(t) _
85.54e- .22t - 85 .54 + 18 .82t, 0(c) = oc, it means that the load
will continue to turn with steady-state speed 18 .82 rad/s and the
displacement from the original position becomes larger and larger .]
In (3.22), in addition to the final speed, we used only one datum to compute Tm .
A more accurate Tm can be obtained by using more data . We plot R(t) in (3 .21) for
a number of t as shown in Figure 3 .5(b), and draw from the origin a straight line
passing through these points . Then the slope of the straight line equals 1/Tm. This
is a more accurate way of obtaining the motor time constant . This method can also
check whether or not the simplified transfer functions in (3 .18) and (3 .15) can be
used. If all points in the plot are very close to the straight line, then (3 .18) and (3 .15)
are adequate . If not, more accurate transfer functions such as the one in (3 .14) must
be used to describe the motor.
The problem of determining Tm and km in (3.18) from measurements is called
parameter estimation. In this problem, the form of the transfer function is assumed
to be known, but its parameters are not known . We then determine the parameters
from measurements . This is a special case of the identification problem where neither
the form nor the parameters are assumed to be known . If no noise exists in meas-
urements, then the identification problem is not difficult . See Reference [15]. How-
ever, in practice, noise often arises in measurements . This makes the problem dif-
ficult. For a different method of identification, see Section 8 .3.2.

e2
-Ji N.
(a)
Figure 3 .6 Gear train and its characteristics .
3.3 GEARS
Gears are used to convert high speed and small torque into low speed and high
torque or the converse . The most popular type of gear is the spur gear shown in
Figure 3.6(a). Let ri be the radius ; Ni , the number of teeth; 0j, the angular displace-
ment, and Ti, the torque for gear i, i = 1, 2 . Clearly, the number of teeth on a gear
is linearly proportional to its radius ; hence N 1 /r 1 = N2 /r 2 . The linear distance
traveled along the surfaces of both gears are the same, thus 0 1 r 1 = 02 r2 . The linear
forces developed at the contact point of both gears are equal, thus T 1 /r 1 = T2/r 2 .
These equalities can be combined to yield
T1 N 02 (3.23)
T2 = N2 = 01
This is a linear equation as shown in Figure 3 .6(b) and is obtained under idealized
conditions . In reality, backlash exists between coupled gears . So long as the gears
are rotating in one direction, the teeth will remain in contact . However, reversal of
the direction of the driving gear will disengage the teeth and the driven gear will
remain stationary until reengagement .3 Therefore, the relationship between 0 1 and
02 should be as shown in Figure 3 .6(c), rather than in Figure 3.6(b). Keeping the
backlash small will increase the friction between the teeth and wear out the teeth
faster . On the other hand, an excessive amount of backlash will cause what is called
the chattering, hunting, or limited-cycle problem in control systems . To simplify
analysis and design, we use the linear equation in (3 .23).
Consider an armature-controlled dc motor driving a load through a gear train
as shown in Figure 3 .7(a) . The numbers of teeth are N 1 and N2 . Let the total moment
3Here we assume that the mass of the gear is zero, or that the gear has no inertia .
3 .3 GEARS 79
(a)
(b)
Figure 3 .7 Gear train and its equivalence .
of inertia (including rotor, motor shaft, and gear 1) and viscous friction coefficient
on the motor shaft be, respectively, J, and f1 , and those on the load shaft be J 2 and
f2. The torque generated by the motor must drive J 1, overcome f 1, and generate a
torque T1 at gear 1 to drive the second gear. Thus we have
d20,(t) de,(t)
Tmotor - Jl dt 2 + f1 + T, (3 .24)
dt
Torque T1 at gear 1 generates a torque T 2 at gear 2, which in turn drives J 2 and

overcomes f2 . Thus we have
d 2 02 (t) d02 (t)
T2 = J2 dt 2 + f2 dt
which, using 02 = (N1 /N 2)01 , becomes
T2 - J2
N, d 2 6 1 (t) 11
d9 1 (t)
(3.25)
N2 dt2 + f 2 N2 dt

The substitution of this T2 into T, = (N, /N2)T2 and then into (3 .24) yields
d 2 a t (t) d01 (t) N, 2 d 2 0 1 (t)
Tmotor = J, dt2 + f t dt + J2 NZ
dt 2
(') 2
+ f2 Nd9,(t)
(3 .26)
N2 dt
d2 0 1 (t) d0, (t)
J'eq
dt 2 + fteq dt
where
2 N' \ 2
N,
J, eq : = J1 + J2 fieq := fl + f2 (3 .27)
N2 C N2
This process transfers the load and friction on the load shaft into the motor shaft as
shown in Figure 3.7(b) . Using this equivalent diagram, we can now compute the
transfer function and develop a block diagram for the motor and load .
Once the load and friction on the load shaft are transferred to the motor shaft,
we can simply disregard the gear train and the load shaft . If the armature inductance
La is assumed to be zero, then the transfer function of the motor and load from u to
0, is, as in (3.15),
(3 .28)
G(s) OU(S) S(7,,,s + 1)
with (3 .16) and (3.17) modified as

k,
km = (3 .29a)
kr kb + f legRa
and
JtegRa
T,n = (3 .29b)
krkb + f iegRa
U
km 01 N,
S(rr S+ 1) N,
Motor and load Gear

(a)
u 8, N, 82
Motor N2 Load
Gear
(b)
Figure 3 .8 (a) Block diagram . (b) Incorrect block diagram .

3 .4 TRANSDUCERS 81
The transfer function from 0, to 02 is N,/N 2. Thus the block diagram of the motor
driving a load through a gear train is as shown in Figure 3 .8(a). We remark
that because of the loading, it is not possible to draw a block diagram as shown in
Figure 3.8(b).
3 .4 TRANSDUCERS
Transducers are devices that convert signals from one form to another-for example,
from a mechanical shaft position, temperature, or pressure to an electrical voltage .
They are also called sensing devices . There are all types of transducers such as
thermocouples, strain gauges, pressure gauges, and others . We discuss in the follow-
ing only potentiometers and tachometers .
Potentiometers
The potentiometer is a device that can be used to convert a linear or angular dis-
placement into a voltage . Figure 2 .14 shows a wire-wound potentiometer with its
characteristic . The potentiometer converts the angular displacement 9(t) into an elec-
tric voltage v(t) described by
v(t) = kO(t) (3 .30)
where k is a constant and depends on the applied voltage and the type of potentiom-
eter used . The application of the Laplace transform to (3 .30) yields
V(s) = kO(s) (3 .31)
Thus the transfer function of the potentiometer is a constant . Figure 3 .9 shows three
commercially available potentiometers .
Figure 3 .9 Potentiometers .

Tachometers
The tachometer is a device that can convert a velocity into a voltage . It is actually
a generator with its rotor connected to the shaft whose velocity is to be measured .
Therefore a tachometer is also called a tachogenerator . The output v(t) of the tach-
ometer is proportional to the shaft's angular velocity ; that is,
dO(t)
v(t) = k (3.32)
dt
where 0(t) is the angular displacement and k is the sensitivity of the tachometer, in
volts per radian per second . The application of the Laplace transform to (3 .32) yields
V(S)
G(s) = = ks (3.33)
O (S)
Thus the transfer function from 0(t) to v(t) of the tachometer is ks.
As discussed in Section 2.4.1, improper transfer functions will amplify high-
frequency noise and are not used in practice . The transfer function of tachometers
is improper, therefore its employment must be justified . A tachometer is usually
attached to a shaft-for example, the shaft of a motor as shown in Figure 3 .10(a).
Although the transfer function of the tachometer is improper, the transfer function
from u to y t is
km kk n
ks =
S(Tm S + 1) n
T ,S + 1
Motor
u k B
s(z s + 1)
Tachometer
Load Tachometer
(a) (b)
1
Motor Motor
r-
u I km u k W
1 m
'r S+1 S Tn S+1
Tachometer Tachometer
Yi
F- k
(c) (d)
Figure 3.10 Use of tachometer .

3 .4 TRANSDUCERS 83
R Noise n(t)
y
Potentiometer Differentiator
Figure 3 .11 Unacceptable way of generating velocity signal.
as shown in Figure 3 .10(b) . It is strictly proper. Thus electrical noise entered at the
armature circuit will not be amplified . The transfer function from motor torque T(t)
to y t is, from Figure 3 .2,
1 k
ks =
s(Js + f) is + f
which is again strictly proper. Thus, mechanical noise, such as torque generated by
gusts, is smoothed by the moment of inertia of the motor . In conclusion, tachometers
will not amplify electrical and mechanical high-frequency noises and are widely
used in practice . See also Problem 3 .17. Note that the block diagram in Figure 3 .10(b)
can also be plotted as shown in Figure 3 .10(c) . This arrangement is useful in com-
puter computation and operational amplifier circuit realization, as will be discussed
in Chapter 5. We mention that the arrangement shown in Figure 3 .11 cannot be used
to measure the motor angular velocity . Although the potentiometer generates signal
k9(t), it also generates high-frequency noise n(t) due to brush jumps, wire irregu-
larities, and variations of contact resistance . The noise is greatly amplified by the
differentiator and overwhelms the desired signal kdO/dt . Thus the arrangement can-
not be used in practice .
The transfer function of a tachometer is ks only if its input is displacement . If
its input is velocity w(t) = dO(t)/dt, then its transfer function is simply k. In velocity
control systems, the transfer function of motors is km/(Tms + 1) as shown in Figure
3.4(b). In this case, the block diagram of a motor and a tachometer is as shown in
Figure 3 .10(d) . Therefore, it is important to specify what are the input and output
of each block.
Error Detectors
Every error detector has two input terminals and one output terminal. The output
signal is proportional to the difference of the two input signals . An error detector
can be built by connecting two potentiometers as shown in Figure 3 .12(a). The two
potentiometers may be located far apart . For example, one may be located inside a
room and the other, attached to an antenna on the rooftop . The input signals 0, and
0, are mechanical positions, either linear or rotational ; the output v(t) is a voltage
signal . They are related by
v(t) = k[O,(t) - 00(t)] (3.34)
(a)
(b) (c)
Figure 3 .12 Pair of potentiometers and their schematic representations .
or
V(s) = k[O,(s) - O o(s)]
where k is a constant . The pair of potentiometers can be represented schematically
as shown in Figure 3.12(b) or (c). The circle where the two signals enter is often
called the summing point .
3 .5 OPERATIONAL AMPLIFIERS (OP-AMPS)
The operational amplifier is one of the most important circuit elements . It is built in 0
integrated-circuit form . It is small, inexpensive, and versatile . It can be used to build
buffers, amplifiers, error detectors, and compensating networks, and is therefore
widely used in control systems .
The operational amplifier is usually represented as shown in Figure 3 .13(a) and
is modeled as shown in Figure 3 .13(b) . It has two input terminals . The one with a
- " sign is called the inverting terminal and the one with a " +' 9 sign the non-
inverting terminal . The output voltage v, equals A(vz2 - vi,), and A is called the
open-loop gain . The resistor Ri in Figure 3 .13(b) is called the input resistance and

3.5 OPERATIONAL AMPLIFIERS (OP-AMPS) 85
(a) (b)
Figure 3 .13 Operational amplifier .
R0, the output resistance. R; is generally very large, greater than 10 4 fl, and Ro is
very small, less than 50 fl . The open-loop gain A is very large, usually over 10 5,
in low frequencies . Signals in op-amps are limited by supply voltages, commonly
15 V . Because of this limitation, if A is very large or infinity, then we have
Ui1 - Ui2 (3.35)
This equation implies that the two input terminals are virtually short-circuited . Be-
cause Ri is very large, we have
ii = 0 (3 .36)
This equation implies that the two input terminals are virtually open-circuited . Thus
the two input terminals have the two conflicting properties : open-circuit and short-
circuit . Using these two properties, operational amplifier circuits can easily be
analyzed .
Consider the operational amplifier circuit shown in Figure 3 .14(a) . Because of
the direct connection of the output terminal and inverting terminal, we have vi1 =
v,,Thus we have, using the short-circuit property, v o = vie. It means that the output
v tage is identical to the input voltage v f2. One may wonder why we do not connect
vo directly to v f2, rather than through an op-amp . There is an important reason for
doing this . The input resistance of op-amps is very large and the output resistance
is very small, so op-amps can isolate the circuits before and after them, and thus
eliminate the loading problem . Therefore, the circuit is called a voltage follower,
buffer, or isolating amplifier and is widely used in practice .
Consider the circuit shown in Figure 3 .14(b) where Z1 and Zf are two imped-
ances . The open-circuit property i i = 0 implies i 1 = -i0. Thus we have'
V1 - Vil V0 - Vi1
(3 .37)
Z1 Zf
'Because impedances are defined in the Laplace transform domain (see Section 2 .4), all variables must
be in the same domain. We use V,, to denote the Laplace transform of v,.

vi 2
v
D R pm
(a) (b) (c)
C R
(d)
Figure 3 .14 Op-amp circuits.
Because the noninverting terminal is grounded and because of the short-circuit prop-
erty, we have v i , = vie = 0. Thus (3 .37) becomes
Vo Zf
(3.38)
V1 Z1
If Zf = Rf and Z1 = R 1 , then the transfer function from v 1 to v o is -Rf/R 1 and
the circuit can be used as an amplifier with fixed gain -Rf/R 1 . If Rf is replaced by
a potentiometer or an adjustable resistor as shown in Figure 3 .14(c), then the gain
of the amplifier can be easily adjusted .
Exercise 3.5 .1
Show that the transfer function of the circuit in Figure 3 .14(d) equals -1/RCs .
Thus, the circuit can act as an integrator . Show that the transfer function of the
circuit in Figure 33 4(e) equals -RCs . Thus, the circuit can act as a pure differen-
tiator . Integrators and differentiators can be easily built by using operational amplifier
circuits. However, differentiators so built may not be stable and cannot be used in
practice . See Reference [18] . Integrators so built are stable and are widely used in
practice .
Consider the op-amp circuit shown in Figure 3 .15 . The noninverting terminal
is grounded, so v i1 = v ie = 0. Because of ii = 0, we have
if = - (i 1 + + i 3)

3 .5 OPERATIONAL AMPLIFIERS (OP-AMPS) 87
vo = -(av I + bv2 + ev 3 )
Figure 3 .15 Op-amp circuit .
which together with v 1 , = 0 implies

va Vl + V2 + V3
R R/a R/b R/c
Thus we have
vo = - (av, + bv2 + cv 3 ) ( 3 .39)
If a = 1 and b = c = 0, then (3 .39) reduces to v o = -v 1 . It is called an inverting
amplifier as shown in Figure 3.16(a) . The op-amp circuit shown in Figure 3 .16(b)
can serve as an error detector . The output of the first op-amp is
e = - (r - v,)
where v,, is the voltage generated by a tachometer. Note the polarity of the tach-
ometer output . The output of the second op-amp, an inverting amplifier, is u = - e,
thus we have u = r - vu, . This is an error detector . In conclusion, op-amp circuits
are versatile and widely used . Because of their high input resistances and low output
resistances, their connection will not cause the loading problem .
R R
(a) (b)
Figure 3 .16 (a) Inverting amplifier. (b) Error detector .

3.6 BLOCK DIAGRAMS OF CONTROL SYSTEMS
In this section, we show how block diagrams are developed for control systems .
Example 3 .6.1
Consider the control system shown in Figure 3 .17(a) . The load could be a telescope
or an antenna and is driven by an armature-controlled dc motor . The system is
designed so that the actual angular position of the load will follow the reference
signal . The error e between the reference signal r and the controlled signal y is
detected by a pair of potentiometers with sensitivity k1 . The dotted line denotes
mechanical coupling, therefore their signals are identical . The error e is amplified
by a dc amplifier with gain k2 and then drives the motor. The block diagram of this
system is shown in Figure 3 .17(b) . The diagram is self-explanatory .
dc
Amplifier
k2
40 Load
(a)
r +1 ) 1 e U km y
k ~
1
3.
(b)
Figure 3 .17 Position control system .
Example 3 .6.2
In a steel or paper mill, the products are moved by rollers, as shown in Figure 3 .18(a) .
In order to maintain a- prescribed tension, the roller speeds are kept constant and
equal to each other. This can be achieved by using the control system shown in
Figure 3 .18(b) . Each roller is driven by an armature-controlled dc motor . The desired

3 .6 BLOCK DIAGRAMS OF CONTROL SYSTEMS 89
(0 3
(a)
Potentiometer
I is
i
1+ I
E
Isolation Amplifie I I
Tachometer

r amplifier I Load k2
I IJ
k=1 k3 I_ I
V
I I
To other
units
(b)
Motor and load

Potentiometer Amplifier
r + e U km
k1 k3 w
V S+1
Desired speed m
V
k2 c
To other units Tachometer

(c)
Figure 3 .18 Speed control system .
roller speed is transformed by a potentiometer into an electrical voltage . The loading

problem associated with the potentiometer is eliminated by using an isolating am-
plifier. The scale on the potentiometer can be calibrated after the completion of the
design. Therefore, there is no need to consider the potentiometer in the design . The
roller speed is measured using a tachometer . The block diagram of the system is
shown in Figure 3 .18(c) . For this problem, we are interested in the roller's speed,
not its position, so the motor shaft speed is chosen as the output and the transfer
function of the motor and load becomes k .1(7-.s + 1). Similarly, the tachometer's
transfer function is k2 rather than k2 s as shown. We mention that depending on the
wiring, the error signal e can be r - v, r + v, or v - r. For the polarities shown
in Figure 3 .18(b), we have e = r - v.

Example 3 .6.3
Temperature control is essential in many chemical processes . Consider the temper-
ature control system shown in Figure 3 .19(a) . The problem is to control the tem-
perature y inside the chamber . The chamber is heated by steam. The flow q of hot
steam is proportional to the valve opening x ; that is, q = kq x. The valve opening x
is controlled by a solenoid and is assumed to be proportional to the solenoid current
i; that is, x = ki. It is assumed that the chamber temperature y and the steam flow
q are related by
dy
-cy + kq (3 .40)
dt
where c depends on the insulation and the temperature difference between inside
and outside the chamber . To simplify analysis and design, c is assumed to be a
positive constant . This means that if no steam is pumped into the chamber, the
temperature will decrease at the rate of e - ". The application of the Laplace transform
Amplifier
k2
Amplifier Thermocouple
kt
01,
10. 4 0
01
(a)
r u 1 i x q k, y
Pot Ls+R s+c
Desired
temperature
V
(b)
Figure 3 .19 Temperature control system .

to (3.40) yields, assuming zero initial condition,

sY(s) = -cY(s) + kQ(s)
or
Y(s) k,
Q(s) s + c
Thus the transfer function from q to y is kc/(s f c) .
Figure 3 .19(b) shows a block diagram for the heating system . The desired tem-
perature is transformed into an electrical voltage r by the leftmost potentiometer .
The scale on the potentiometer can be calibrated after the completion of the design .
The chamber temperature is measured using a thermocouple and amplified to yield
v = key. From the wiring, we have e = r - v and the wiring is represented by the
summer shown . The error signal e is amplified to yield u . The RL circuit is governed
by
di(t)
u(t) = Ri(t) + L
dt
The application of the Laplace transform yields, assuming zero initial condition,
U(s) = RI(s) + LsI(s)
which implies
(3.41)
U(s) R + Ls
Thus the transfer function from u to i is 1/(Ls + R) as shown. The remainder of
the block diagram is self-explanatory .
Exercise 3 .6.1
Develop the block diagrams in Figures 1 .1, 1 .6, and 1 .7.
3.6 .1 Reaction Wheels and Robotic Arms5

In this subsection, we give two more examples of developing block diagrams for
control systems .
'May be skipped without loss of continuity .

Example 3 .6 .4
In order to point the antenna toward the earth or the solar panels toward the sun, the
altitude or orientation of a satellite or space vehicle must be properly controlled .
This can be achieved using gas jets or reaction wheels . Because of unlimited supply
of electricity through solar panels, reaction wheels are used if the vehicle will travel
for long journeys . Three sets of reaction wheels are needed to control the orientation
in the three-dimensional space ; but they are all identical .
A reaction wheel is actually a flywheel ; it may be simply the rotor of a motor .
It is assumed to be driven by an armature-controlled dc motor as shown in Figure
3 .20(a) and (b) . The case of the motor is rigidly attached to the vehicle . Because of
the conservation of momentum, if the reaction wheel turns in one direction, the
satellite will rotate in the opposite direction . The orientation and its rate of change
can be measured using a gyro and a rate gyro . The block diagram of the control
system is shown in Figure 3 .20(c).
We derive in the following the transfer function G(s) of the space vehicle and
' II s
Reaction
wheel
(a) (b)
Amplifier
r U y
G(s)
IS
Rate gyro
Gyro
(c)
Figure 3 .20 Altitude control of satellite .

motor . Let the angular displacements, with respect to the inertial coordinate, of the
vehicle and the reaction wheel be respectively y and 0 as shown in Figure 3 .20(a) .
They are chosen, for convenience, to be in opposite directions . Clearly, the relative
angular displacement of the reaction wheel (or the rotor of the motor) with respect
to the vehicle (or the stator or case of the motor) is y + 0. Thus the armature circuit
is governed by, as in (3.10),
di,(t)
Raia(t) + La + v b(t) = u(t) (3 .42)
with the back emf voltage vb(t) given by
Vb(t) (3 .43)
= kb ( dt + dt )
Let J and f be the moment of inertia and the viscous friction coefficient on the motor
shaft. Because the friction is generated between the motor shaft and the bearing that
is attached to the satellite, the friction equals f(dO/dt + dy/dt) . Thus the torque
equation is
dzd~t)
T(t) = ktia = J + f (3 .44)
de + dt
Let the moment of inertia of the vehicle be J,, . Then the conservation of angular
momentum implies
Jv t= J (3 .45)
d ae
Using (3 .42) through (3.45), we can show that the transfer function from u to y
equals
t
G(s) .46)
U(s) s[(Las + R a)(JJs + (J + J)f) + krkb(J + J01 (3
This completes the block diagram in Figure 3 .20 .
Ex anhple 3 .6.5
Consider the industrial robot shown in Figure 3 .21(a) . The robot has a number of
joints . It is assumed that each joint is driven by an armature-controlled dc motor
through gears with gear ratio n = Nl/N 2. Figure 3 .21(b) shows a block diagram of
the control of a joint, where the block diagram in Figure 3 .3 is used for the motor .
The J and f in the diagram are the total moment of inertia and viscous friction
coefficient reflected to the motor shaft by using (3 .27). The compensator is a

(a)
r kt T w y Ni
I
PID
L Gs + Ra JS +f s N2
(b)
Figure 3 .21 (a) Industrial robot . (b) Joint control system .
proportional-integral-derivative (PID) compensator, which will be discussed in later

chapters .
3.7 MANIPULATION OF BLOCK DIAGRAMS
Once a block diagram for a control system is obtained, the next step in analysis is
to simplify the block diagram to a single block or, equivalently, to find the overall
transfer function. It is useless to analyze an individual block, because the behavior
of the control system is affected only indirectly by individual transfer function . The
behavior is dictated completely by its overall transfer function .
Two methods are available to compute overall transfer functions : block diagram
manipulation and employment of Mason's formula . We discuss first the former and
than the latter . The block diagram manipulation is based on the equivalent diagrams
shown in Table 3 .1 . The first pair is concerned with summers . A summer must have
two or more inputs and one and only one output . If a summer has three or more
Table 3 . 1 Equivalent Block Diagrams
Y Y1
U Y U
2. G G
Y2 Y3 Y2 Y3
3.
4.
U Y1
5. G
Y
U Y1 Y1
6. G
Y2
L~ G
yz
1 /G l
7.
U
G1 --;I G2
L- ~
G1 G2 H
8.
9.
95

inputs, it can be separated into two summers as shown . A terminal can be branched
out, at branching points, into several signals as shown in the second pair of Table
3.1 with all signals equal to each other . A summer or a branching point can be moved
around a block as shown from the third to the sixth pair of the table . Their equiva-
lenCes can be easily verified . For example, for the fourth pair, we have Y =
GUI U2 for the left-hand-side diagram and Y = G(U, U 2/G) = GU, U 2
for the right-hand-side diagram . They are indeed equivalent .
The last three pairs of Table 3 .1 are called, respectively, the tandem, parallel,
and feedback connections of two blocks . The reduction of the tandem and parallel
connections to single blocks is very simple . We now reduce the feedback connection
to a single block. Note that if W is fed into the summer with a positive sign (that is,
E = U + W), it is called positive feedback . If W is fed into the summer with a
negative sign (that is, E = U - W), it is called negative feedback . We derive only
the negative feedback part. Let the input of G, be denoted by E and the output of
G2 by W as shown in the left-hand-side diagram of the last pair of Table 3 .1 . Then
we have
W=G2 Y E = U - W = U - G2Y Y=G,E (3 .47)
The substitution of the second equation into the last yields
Y = G I(U - G 2 Y) = G I U - G I G2 Y (3 .48)
(1 + GIG2)Y = G I U
Thus we have
Y _ G,
(3.49)
U 1 + GIG2
This is the transfer function from U to Y, for the negative feedback part, as shown
in the right-hand-side diagram of the last pair of Table 3 .1 . In conclusion, the transfer
function of the feedback connection is G,/(1 + G I G2) for negative feedback and
GI/(1 - G IG2) for positive feedback. They are important formulas, and should be
remembered.
Now we use an example to illustrate the use of Table 3 .1 to compute overall
transfer functions of block diagrams .
Example 3.7 .1
Consider the block diagram shown in Figure 3 .22(a) . We first use entry 9 in Table
3 .1 to simplify the inner positive feedback loop as shown in Figure 3 .22(b) . Note
that the direct feedback in Figure 3 .22(a) is the same as feedback with transfer
function 1 shown in Figure 3 .22(b) . Using entries 7 and 9, we have

3 .7 MANIPULATION OF BLOCK DIAGRAMS 97
r + Y r + G
1 - G 2 G3
(a) (b)
r + + y y
G1 G2 1
-G3 /G
1/G F
3
F-
(C) (d)
r + Y
G1G2
G1
1-- F--
G1
(e)
Figure 3 .22 Manipulation of block diagram.
G2
Y _ G' 1 - G2G3 G1 G2
(3.50)
R 1 - G2G3 + G 1 G2
1 + G1 1GG 1
2G3
This is the overall transfer function from R to Y in Figure 3 .22(a) .
A block diagram can be manipulated in many ways . For example, we can move
thIsecond summer in Figure 3 .22(a) to the front of G 1, using entry 4 of Table 3 .1,
to yield the block diagram in Figure 3 .22(c) which can be redrawn, using entry 1 of
Table 3.1, as shown in Figure 3 .22(d) . Note that the positive feedback has been
changed to a negative feedback, but we also have introduced a negative sign into
the feedback block . The two feedback paths are in parallel and can be combined as
shown in Figure 3.22(e) . Thus we have
Y _ G1 G2 _ G1G2
R G3 1 + G l G2 - G2G3
1 + G,G2 -
\1 G1
which is the same as (3.50), as expected .

Exercise 3 .7 .1
Use block manipulation to find the transfer functions from r to y of the block dia-
grams in Figure 3 .23.
r
-* G4
Figure 3 .23 Block diagrams .
[Answers : G,G2G4/(1 - G2G3 + G 2G4 + G,G2G4), G,/(1 + G 2 + G,G3).]
3.7.1 Mason's Formula

Consider the block diagram shown in Figure 3 .24 . There are two inputs, r and p .
The problem is to find the transfer function from r to y, denoted by Gy and the
transfer function from p to y, denoted by Gyp . Because the system is linear, when
we compute Gyr, we may assume p = 0 or, equivalently, disregard p . When we
compute Gyp , we may disregard r. Once they are computed, the output y due to the
r + + U l y
-0-
Figure 3 .24 Block diagram .

two inputs r and p is given by

Y(s) = Gy,(s)R(s) + Gyp(s)P(s) (3.51)
where Y, R, and P are respectively the Laplace transforms of y, r, and p .
Although Gy, and Gyp can be obtained by block manipulations, the manipula-
tions are not simple . Furthermore, the manipulation in computing Gy, generally can-
not be used to compute Gyp. Therefore, it requires two separate manipulations . In
this subsection, we introduce a formula, called Mason's formula, to compute overall
transfer functions that does not require any block manipulation .
The employment of Mason's formula requires the concepts of loops and forward
paths . Consider the block diagram shown in Figure 3 .24 . Note that every block and
branch is oriented, that is, unidirectional . A loop is any unidirectional path that
originates and terminates at the same point and along which no point is encountered
more than once . A loop gain is the,product of all transfer functions along the loop .
If a loop contains one or more summers, the sign at each summer must be considered
in the loop gain . For example, the loop in Figure 3 .25(a) has two summers . One
branch enters the summer with a negative sign, one with a positive . The one with a
negative sign can be changed to a positive sign by introducing a negative sign into
G2, as is shown in Figure 3 .24(b) . We remark that a branch with no block is the
same as having a block with transfer function 1 . Therefore the loop gain of the loop
in Figure 3 .25(a) or (b) is G,(-G 2) = -G1G2. Similarly, the loop in Figure 3 .25(c)
has loop gain - G, ( - G 2) = G, G 2. Now consider the block diagram in Figure 3 .24 .
It has five loops with loop gains -G1G6, -G4G5, G 3G4G7, - G1G3G4, and G2G4.
Two loops are said to be nontouching if they have no points in common. For
example, the four loops in Figure 3 .24 with loop gains -G4G5, G3G4G7, -G 1 G3G4,
and G2G4 all touch each other because they all pass block G4 or the point denoted
by 1 in the diagram . The two loops with loop gains -G1G6 a id - G4G5 do not
touch each other ; neither do the two loops with loop gains -G,G6 and G2G4. Now
we define
0 = 1 - (I loop gains)
+ ( products of all possible two nontouching loop gains)
(3.52)
- (~ products of all possible three nontouching loop gains)
Although this formula looks very complicated, it is not necessarily so in its appli-
A O 0 O E 0
(a) (b) (c)
Figure 3 .25 Block diagrams with a single loop .

cation . For example, if a block diagram has no loop as in the tandem and parallel
connections in Table 3 .1, then 0 = 1 . If a block diagram has only nontouching
loops, then the formula terminates at the first parentheses . If a block diagram does
not have three or more mutually nontouching loops, then the formula terminates at
the second parentheses . For example, the block diagram in Figure 3 .24 does not have
three or more mutually nontouching loops, and we have
0 = 1 -(-G1G6 - G4G5 + G 3G4G7 - G1G3G4 + G2G4)
(3.53)
+((-G1G6)(-G4G5) + ( - G1G6)(G2G4))
For easy reference, we call 0 the characteristic function of the block diagram . It is
an inherent property of a block diagram and is independent of input and output
terminals .
Now we introduce the concept of forward paths . It is defined for specific input/
output pairs . A forward path from input r to output y is any connection of unidirec-
tional branches and blocks from r to y along which no point is encountered more
than once . A forward path gain is the product of all transfer functions along the path
including signs at summers . For the block diagram in Figure 3 .24, the r-y in-
put/output pair has two forward paths, with gains P 1 = G 1G3G4 and P2 = - G2G4.
The p-y input/output pair has only one forward path, with gain - G4. A loop touches
a forward path if they have at least one point in common .
With these concepts, we are ready to introduce Mason's formula . The formula
states that the overall transfer function from input v to output w of a block diagram
is given by
W(s) Ii PL1
Gwu (3 .54)
V(s) 0
where 0 is defined as in (3 .52),
Pi = the gain of the ith forward path from v to w
Ai = Aset those loop gains to zero if they touch the ith forward path .
and the summation is to be carried out for all forward paths from v to w . If there is
no forward path from v to w, then G,t, = 0. Now we use the formula to compute
Gyr and Gpy for the block diagram in Figure 3 .24. The 0 for the diagram was com-
puted in (3 .53) . There are two forward paths from r to y, with P 1 = G 1G3G4 and
P2 - -G2G4. Because all loops touch the first forward path, we set all loop gains
to zero in (3 .53) to yield
A1 =1
All loops except the one with loop gain -G 1 G6 touch the second forward path,
therefore we have
A2 = 1 - ( - G1G6) = 1 + G1G6

Thus we have
= Y(s) - P101+ P2A2
GYr (3.55)
R(s) A
G1G3G4 ' 1 + ( - G2G4)(1 + G1G6)
1 + G1G6 + G4G5 - G3G4G7 + G1G3G4 - G2G4 + G1G6G4G5 - G1G6G2G4
Now we find the transfer function from p to y in Figure 3 .24 . The A for the
block diagram was computed in (3.53). There is only one forward path from p to y
with gain P 1 = - G4. Because all loops except the loop with loop gain -G 1G6
touch the path, we have
0 1 = 1 - ( - G1 G6) = 1 + G 1G6
Thus the transfer function from p to y in Figure 3 .24 is, using Mason's formula,
PQ
(3.56)
Gyp = P(S) = 1
(-G4)(1 + G,G6 )
1 + G,G6 + G 4 G5 - G3 G4 G7 + G 1G3G4 - G2G4 + G,G6G 4 G5 - G,G 6G2G4
Exercise 3 .7 .2
Use Mason's formula to verify entries 7, 8, and 9 in Table 3 .1 .
Exercise 3.7 .3
Find the transfer functions from r to u and from p to u in Figure 3 .24.

[Answers : Gur = (G 1 (1 + G4G5) - G2G4G7)/A, G up = (- G4G7 + GA)/0,
where A is given in (3.53).]
Exercise 3 .7 .4
Repeat Exercise 3 .7.1 by using Mason's formula . Are the results the same?
3 .7 .2 Open-Loop and Closed-Loop Transfer Functions

Consider the block diagram shown in Figure 3 .26(a) . The transfer function from r
to y, from p1 to y, and from P2 to y are respectively
G1G2 G2 1
Ciyr
yr (3 .57)
1 + G1G2 Gyp1 1 + G1 G2 GYP2 - 1 + G1G2

P2
r + e y r
e _ y
'a- 111+ G1
(a) (b)
Figure 3 .26 Closed-loop and open-loop systems .
These transfer functions are often referred to as closed-loop transfer functions . Now
we open the loop at e as shown, and the resulting diagram becomes the one in Figure
3.26(b) . The transfer functions from r to y, from p l to y, and from P2 to y now
become
Gyr = 0 GYP1 = G 2 GYP2 = I
They are called open-loop transfer functions. We see that they are quite different
from the closed-loop transfer functions in (3.57). This is not surprising, because the
block diagrams in Figure 3 .26 represent two distinctly different systems . Unless
stated otherwise, all transfer functions in this text refer to closed-loop or overall
transfer functions .
PROBLEMS
3.1 . Consider a generator modeled as shown in Figure P3 .1(a). The generator is

driven by a diesel engine with a constant speed (not shown) . The field cirruit-1
is assumed to have resistance R f and inductance L f ; the generator has internal
resistance R g . The generated voltage vg (t) is assumed to be linearly proportional
to the field current if (t)-that is, vg(t) = kgif (t) . Strictly speaking, vg(t) should
appear at the generator terminals A and B . However, this would cause a loading
problem in developing its block diagram . In order to eliminate this problem,
(a)
u kg V
8 RL y
+ Lf S RL + Rg
(b) Figure P3 .1

PROBLEMS 103
we assume that vg(t) appears as shown in Figure P3 .1(a) and combine Rg with
the load resistance RL. Show that the generator can now be represented by the
block diagram in Figure P3 .1(b) and there is no loading problem in the block
diagram . Where does the power of the generator come from? Does the flow of
power appear in the block diagram?
3.2. Consider an armature-controlled dc motor driving a load as shown in Figure
P3.2. Suppose the motor shaft is not rigid and must be modeled as a torsional
spring with constant ks. What is the transfer function from u to 0? What is the
transfer function from u to w = dO/dt?
3.3. The combination of the generator and motor shown in Figure P3 .3 is widely
used in industry . It is called the Ward-Leonard system . Develop a block
diagram for the system . Find the transfer functions from u to v g and from
vg to 0.
Generator
----------------
R f R
111
+ I
1
I
I
u I
I
L
I
a
O
I
l
I
I T II
I f\
c~ 0
I I
I I I I 9
~ I I
L J L J
Ward-Leonard system
Figure P3.3
3.4. a. The system in Figure P3 .4(a) can be used to control the voltage v o(t) of a
generator that is connected to a load with resistance RL. A reference signal
r(t), which can be adjusted by using a potentiometer, is applied through an
amplifier with gain k, to the field circuit of the generator. Draw a block
diagram for the system . The system is called an open-loop voltage regulator .
b. The system in Figure P3 .4(b) can also be used to control the voltage vo(t).
It differs from Figure P3 .4(a) in that part of the output voltage or v, = k v0
is subtracted from r(t) to generate an error voltage e(t) . The error signal
e(t) is applied, through an amplifier with gain k,, to the field circuit of the

generator. Develop a block diagram for the system . The system is called a
closed-loop or feedback voltage regulator .
(a)
(b)
Figure P3 .4
3.5. Consider the armature-controlled dc motor driving a load through a pair of

gears shown in Figure 3 .7(a). The following data are given:
Ra = 50 0 k, = I N m A - ' k,, = 1 V r ad- ' s,
N l /N 2 = 1/2 J2 = 12 N m rad -1 s 2 f2 = 0.2 N m rad- ' s,
Ji = 0.1 Nm rad - ' s 2 , fl = 0.01 N m rad- ' s
Draw a block diagram for the system . Also compute their transfer functions .
3.6. Consider the gear train shown in Figure P3 .6(a). Show that it is equivalent to
the one in Figure P3 .6(b) with
2 2
N~ (N I N3 '\
Jleq = Jl + J2 + J3
N2 N 2 N4
2 2
(Nl/
fleq = fl + f2 N2 + f3 \N2 N4
N'N3 /
3.7 . Find the transfer functions from v ; to v o of the op-amp circuits shown in Figure
P3.7 .

PROBLEMS 105
J,f
U ,1eq ffeq
f2
J3' f 3
N2 N4
(a) (b)
Figure P3 .6
RI C
F,F%6, Vt. o-
F
Figure P3 .7
3 .8 . Consider the operational amplifier circuit shown in Figure P3 .8 . Show that the
output uo equals
R
Uo = R (VB - VA)
This is called a differential amplifier .
Figure P3 .8
3 .9. Develop a block diagram for the system shown in Figure P3 .9. Compute also
the transfer function of each block .
Figure P3.9
3 .10. Consider the temperature control system shown in Figure P3 .10 . It is assumed
that the heat q pumped into the chamber is proportional to u and the temperature
y inside the chamber is related to q by the differential equation dy(t)/dt =
-0.3y(t) + 0.1q(t) . Develop a block diagram for the system and compute the
transfer function of each block.
Temperature
sensor, k,
Figure P3.10
3.11 . In addition to electromechanical transducers, optical transducers are also used

in practice. The transducer shown in Figure P3 .11 consists of a pulse generator
and a pulse counter. By counting the number of pulses in a fixed interval of
time, a signal proportional to the angular speed of the motor shaft can be
generated . This signal is in digital form . Using a digital-to-analog converter
and neglecting the so-called quantization error, the transfer function from 0 to

PROBLEMS 107
v shown in Figure P3 .10 can be approximated by ks. Draw a block diagram

for the system. Indicate also the type of transfer function for each block .
Pulse generator
Control level Light
Transistor
Motor source
pickup
Amplifier 8 ......t0-
Load
k1
Digital-to- Pulse
analog count
converter Digital speed signal Pulse
Figure P3.11
3 .12. Machine tools can be controlled automatically by the instruction recorded on

punched cards or tapes . This kind of control is called the numerical control of
machine tools . Numerical control can be classified as position control and
continuous contour control . A schematic diagram of a position control system
is shown in Figure P3 .12. A feedback loop is introduced in the D/A (digital-
to-analog) converter to obtain a more accurate conversion. Draw a block dia-
gram from ed to 00 for the system. Indicate also the type of transfer function
for each block.
r I
I I
I
Tape
Tape --> reader Pulse -~o D/A converter
generator
. 0 1
I 1
I Digital 1
feedback I
Motor
Amplifier
F
k2
e=k 1 (Bd -B )
Figure P3.12

3.13. In industry, a robot can be designed to replace a human operator to carry out
repeated movements. Schematic diagrams for such a robot are shown in Figure
P3.13 . The desired movement is first applied by an operator to the joystick
shown. The joystick activates the hydraulic motor and the mechanical arm .
The movement of the arm is recorded on a tape, as shown in Figure P3 .13(a).
The tape can then be used to control the mechanical arm, as shown in Figure
P3 .13(b) . It is assumed that the signal x is proportional to u, and the transfer
function from x to y is k,,,/s(Tm s + 1). Draw a block diagram for the system
in Figure P3 .13(b) . Indicate also the type of transfer function for each block .
Transducer
Tape
(a)
Actuator
Transducer
Amplifier
compensating
network
Error Tape F
detector reader Tape
e
(b) Figure P3.13
3.14. Use Table 3 .1 to reduce the block diagrams shown in Figure P3 .14 to single
blocks .

PROBLEMS 109
(a)
r
G1 V
(b)
(c)
Figure P3 .14
3.15. Use Mason's formula to repeat Problem 3 .14 .

3.16 . Use Mason's formula to compute the transfer functions from r l to y and r 2 to
y of the block diagram shown in Figure P3 .16 .
r1 + y
G1 G3
G5
Figure P3.16

3.17. Consider the block diagram shown in Figure P3 .17. It is the connection of the
block diagram of an armature-controlled dc motor and that of a tachometer .
Suppose noises may enter the system as shown . Compute the transfer functions
from n t to w and from n 2 , to w . Are they proper transfer functions?
n2
r-
u I + k, T I I w
1.0 ks
LG s+RQ JS +f I
I
I Tachometer
I
I
Motor and load

Figure P3 .17
Quantitcztive
and Qualitative
Analyses of
Control Systems
4.1 INTRODUCTION
Once block diagrams of control systems are developed, the next step is to carry out
analysis . There are two types of analysis : quantitative and qualitative . In quantitative
analysis, we are interested in the exact response of control systems due to a specific
excitation . In qualitative analysis, we are interested in general properties of control
systems . We discuss first the former and then the latter .
4.2 FIRST-ORDER SYSTEMS-THE TIME CONSTANT
Consider the armature-controlled dc motor driving a load, such as a video tape,

shown in Figure 4.1 . The objective is to drive the tape at a constant linear speed . To
simplify the discussion, we assume that this can be achieved by maintaining the
motor shaft at a constant angular speed .' Using a potentiometer, the desired speed
is transformed into a reference signal r(t) . The reference signal r(t) passes through
an amplifier with gain kl to generate an actuating signal u(t) = k1r(t), which
then drives the motor . The transfer function from u to the motor shaft speed w is
'The radius of the reel changes from a full reel to an empty reel, therefore a constant angular velocity
will not generate a constant linear tape speed . For this reason, the angular velocity of motor shafts that
drive expensive compact disc (CD) players is not constant . The angular velocity increases gradually as
the, pick-up reads from the outer rim toward the inner rim so that the linear speed is constant .
111

112 CHAPTER 4 QUANTITATIVE AND QUALITATIVE ANALYSES OF CONTROL SYSTEMS
Load
Figure 4.1 Open-loop velocity control system .
km /(Tm s + 1), as is shown in Figure 3 .4(b). Thus the transfer function from r to
w is
(s) = klkm
G(s) = W
R(s) Tm S + I
where km and Tm are the motor gain and time constants (see (3 .16) and (3.17)). If
we apply a constant voltage, or r(t) = a, for t ? 0, then R(s) = als, and
klkm a ak,,,k, - akm Tm ki _ akmk, - akm kl
W(s) -
Tm S+ I S S Tm S+ I S S+ 1/T m
which implies
w(t) = akm k i - akm ki e -(1 /m )t (4.2)
for t ? 0 . This response, shown in Figure 4.2(a), is called the step response; it is
called the unit-step response if a = 1 . Because a -t/ m approaches zero as t ~,
we have
w,(t) : = lim w(t) = akm k i
t-oo
This is called the steady-state orfinal speed . If the desired speed is w,., by choosing
a as a = wr/k l km, the motor will eventually reach the desired speed .
In controlling the tape, we are interested in not only the final speed but also the
speed of response ; that is, how fast the tape will reach the final speed. For the first-
order transfer function in (4.1), the speed of response is dictated by Tm , the time
constant of the motor. We compute
et lm
Tm (0 .37)' = 0 .37
2T, (0 .37) 2 = 0 .14
3r, (0 .37) 3 = 0 .05
4Tm (0 .37)' = 0 .02
5Tm (0 .37) 5 = 0.007

4 .2 FIRST-ORDER SYSTEMS-THE TIME CONSTANT 113
w (t) e-
(a) (b)
Figure 4 .2 Time responses .
and plot e - `lm in Figure 4.2(b). We see that if t ? 5T,,,, the value of e - `/m is less
than 1 % of its original value. Therefore, the speed of the motor will reach and stay
within 1% of its final speed in 5 time constants . In engineering, the system is often
considered to have reached the final speed in 5 time constants .
The system in Figure 4 .1 is an open-loop system because the actuating signal
u(t) is predetermined by the reference signal r(t) and is independent of the actual
motor speed . The motor time constant T. of this system depends on the motor and
load (see (3.17)). For a given load, once a motor is chosen, the time constant is
fixed . If the time constant is very large, for example, T,,, = 60 s, then it will take
300 seconds or 5 minutes for the tape to reach the final speed . This speed of response
Amplifier a
e Load Tachometer
ki T r
f
(a)
Desired speed r + e U km w
Pot ~- ki
f
2 c
(b)
Figure 4 .3 Feedback control system .

is much too slow . In this case, the only way to change the time constant is to choose
a larger motor. If a motor is properly chosen, a system with good accuracy and fast
response can, be designed . This type of open-loop system, however, is sensitive to
plant perturbation and disturbance, as is discussed in Chapter 6, so it is used only
in inexpensive or low-quality speed control systems .
We now discuss a different type of speed control system . Consider the system
shown in Figure 4.3(a). A tachometer is connected to the motor shaft and its output
is combined with the reference input to generate an error signal . From the wiring
shown, we have e = r - f. The block diagram is shown in Figure 4.3(b). Because
the actuating signal u depends not only on the reference input but also the actual
plant output, this is a closed-loop or feedback control system . Note that in developing
the transfer function of the motor, if the moment of inertia of the tachometer is
negligible compared to that of the load, it may be simply disregarded. Otherwise, it
must be included in computing the transfer function of the motor and load.
The transfer function from r to w of the feedback control system in Figure 4 .3
is
k1 km
- W(s) - Tm S + 1 = k,km
G0(s)
R(s) I + k1kmk2 Tms + k lk2km + 1
Tm S + I
k1km
klk2 km + 1 k1ko
T's + 1
Tm
(k 1k2k. + 1 s + 1
where
(4 .4a)
T . k1 k2 kmm + 1
d4 .4b)
k . k 1k 2k mm + 1
This transfer function has the same form as (4.1). If r(t) = a, then we have, as in
(4.2),
w(t) = akok1 - akokre -OlTo )t (4.5)
and the steady-state speed is akokl. With a properly chosen, the tape can reach a
desired speed . Furthermore, it will reach the desired speed in 5 X To seconds.
The time constant To of the feedback system in Figure 4 .3 is Tm/(k,k2km + 1).
It now can be controlled by adjusting k 1 or k2 . For example, if Tm = 60 and km =
1, by choosing k 1 = 10 and k2 = 4, we have To = 60/(40 + 1) = 1 .46, and the
tape will reach the final speed in 5 X 1 .46 = 7 .3 seconds . Thus, unlike the open-
loop system in Figure 4 .1, the time constant and, more generally, the speed of re-
sponse of the feedback system in Figure 4 .3 can be easily controlled .

4 .2 FIRST-ORDER SYSTEMS-THE TIME CONSTANT 115
4.2.1 Effects of Feedback

We now use the systems in Figures 4.1 and 4 .3 to discuss some of the effects of
introducing feedback .
1. The time constant of the open-loop system can be changed only by changing
the motor . However, if we introduce feedback, the time constant of the resulting
system can easily be controlled by merely changing the gain of the amplifier .
Thus, a feedback system is more flexible and the choice of a motor is less critical .
2. Although the motor time constant is reduced by a factor of (kik2km + 1) in the
feedback system, as shown in (4 .4a) (this is good), the motor gain constant is
also reduced by the same factor, as shown in (4 .4b) (this is bad) . In order to
compensate for this loss of gain, the applied reference voltage must be increased
by the same factor. This is one of the prices of using feedback .
3. In order to introduce feedback, for example, from mechanical signals to elec-
trical voltages, transducers such as tachometers must be employed . Transducers
are expensive . Furthermore, they may introduce noise and consequently inac-
curacies into systems . Therefore, feedback systems are more complex and re-
quire more components than open-loop systems do .
4. If used properly, feedback may improve the performance of a system . However,
if used improperly, it may have a disastrous effect . For example, consider the
feedback system in Figure 4.3(a). If the wiring at A and B is reversed, then
u(t) = kl(r(t) + f(t)) and the block diagram becomes as shown in Figure 4 .4.
This is a positive feedback system . Its transfer function from r to w is
k,km
W(s) Tms + 1 kl km
G0(s) =
R(s) 1 _ klk2km Tm s + 1 - k,k2km
Tm S + I
If Tm = 1, k ikm = 10 and kik2km = 5, and if r(t) = a, fort ? 0, then

10 a 2.5a 2 .5a
W(s) = .-= -
s- 4 s s- 4 s
which implies
w(t) = 2 .5ae4` - 2.5a
Figure 4.4 Positive feedback system .

We see that if a 0 0, the term 2.5ae4` approaches infinity as t - -. In other

words, the motor shaft speed will increase without bounds and the motor will
bum out or disintegrate . For the simple system in Figure 4.3, this phenomenon
will not happen for negative feedback . However, for more complex systems,
the same phenomenon may happen even for negative feedback . Therefore, care
must be exercised in using feedback . This is the stability problem and will be
discussed later.
Exercise 4 .2 .1
Consider (4.6). If k l k2 km = 1 and r(t) = 10 -2, for t ? 0, what is the speed w(t) of
the motor shaft? What is its final speed?
[Answers : k i km t/l00Tm, infinity .]
4 .3 SECOND-ORDER SYSTEMS
Consider the position control of a load such as an antenna . The antenna is to be

driven by an armature-controlled dc motor . Open-loop control of such a system is
not used in practice because it is difficult to find a reference signal to achieve the
control . (See Problem 4.6). A possible feedback design is shown in Figure 3.17 . Its
transfer function from r to y is
klk2km
Y(s) _ S(Tms + 1) _ k,k 2km
G0(s) =
k,k2 km Tm s 2 + s + k i k2 km
R(s) 1 +
S(Tm S + 1)
klk2km/Tm
2 1 kl k2km
s + -s +
TM TM
If we define con := k,k2 km /T m , and 2~Wn := 1/Tm , then (4.7) becomes
W'2
G ,(s)
o(s) Y(s) __
(4 .8)
R(s) s2 + 2Ccons + Wn
This is called a quadratic transfer function with a constant numerator. In this section
we study its unit-step response .
The transfer function G O (s) has two poles and no zero . Its poles are
-SWn JWn\/I - C2 -Q jWd (4 .9)
where a : _ Ccon and cod : = ton '\/1 - ~2. They are plotted in Figure 4.5. The
constant is called the damping ratio ; (o., the natural frequency ; o-, the damping

4 .3 SECOND-ORDER SYSTEMS 117
Im s
s-plane
Figure 4.5 Poles of quadratic system .
factor; and cod , the damped or actual frequency. Clearly if the damping ratio ~ is 0,
the two poles jwn are pure imaginary . If 0 < ~ < 1, the two poles are complex
conjugate and so forth as listed in the following :
Damping Ratio Poles Remark
~ = 0 Pure imaginary Undamped

0 < < 1 Complex conjugate Underdamped
~ = 1 Repeated real poles Critically damped
~ >1 Two distinct real poles Overdamped
In order to see the physical significance of ~, a, w, and co n , we first compute the

unit-step response of (4.8) for 0 1 . If r(t) = 1, for t ? 0, then R(s) = 1/s
and
wn 1 wn
Y(s)
s2 + 2~ws + (on s (s + o- + jwd)(s + o- - j(od)s
= ki + kz + kz
S S+Q+fwd S+Q-fwd
with
k, = G0 (s)J, . 0 = 1
wz wnz wnz
I k = n
(S + Q - jwd)S (-2j(od)(-o- - Jwd) (2j(od)(cr + jwd)
s=-~-JWd

and
w2
n
k2* -
( -2Jwd)( 0- (
.t Ud)
where k2* is the complex conjugate of k2. Thus the unit-step response of (4.8) is
y(t) = 1 + k2e-(+jwd)` + k 2* e -(0'- i wd)t
which, after some manipulation, becomes
y(t) = 1 - ~n e -` sin (wdt + 0) (4.10)

(od
where
-1 V1-C 2
0 = cos C = tan-' = sin -1 V1 - ~2 (4.11)
C
The angle 0 is shown in Figure 4 .5. We plot sin (wdt + 0) and
e-t
in Figure 4 .6(a)
and (b). The point-by-point product of (a) and (b) yields e - ` sin (wd t + 0). We
see that the frequency of oscillation is determined by wd, the imaginary part of the
poles in (4 .9). Thus, co d is called the actual frequency ; it is the distance of the poles
from the real axis . Note that wn is called the natural frequency ; it is the distance of
the poles from the origin. The envelope of the oscillation is determined by the
damping factor a-, the real part of the poles . Thus, the poles of G0(s) dictate the
response of a - Q` sin (wd t + 0), and consequently of y(t) in (4 .10). The unit-step
response y(t) approaches 1 as t -* cc. Thus the final value of y(t) is 1 .
We now compute the maximum value or the peak of y(t). The differentiation
of y(t) yields
dy(t)
= 0' (0n a - ` sin (codt + 0) - wne - ~` cos (wd t + 8)
dt wd
The peak of y(t) occurs at a solution of dy(t)/dt = 0 or, equivalently, a stationary
point of y(t) . Setting dy(t)/dt to zero yields
sin ((odt + 0) _ wd wnVl - ~2 V1- ~2
= tan (wdt + 0) = = = q (4.12)
COs (wdt + 8 ) / Q wn S
By comparing (4 .11) and (4.12), we conclude that the solutions of (4 .12) are
wd t = k7r k = 0, 1, 2, . . .
Thus the stationary points of y(t) occur at t = kar/w d, k = 0, 1, . . . . We plot y(t)

in Figure 4 .7 for various damping ratios ~. From the plot, we see that the peak occurs
at k = f or
7T IT
tp
wd wn Vl -

4 .3 SECOND-ORDER SYSTEMS 119
sin (C)d t+ 0)
(a)
e r
1/6
(b)
(c)
Figure 4 .6 Response of a - `sin (wdt + 0) .
The substitution of tp into (4 .10) yields

wn wn
ymax : = y(tp) = 1 - a 't" sin (ar + 0) = I + a- ò sin 0
Wd Wd
which, because sin 0 = cod/co,, (see Figure 4.5), reduces to
ymax = 1 + e - `P = 1 + ~" = 1 + (4 .13)
This is the peak of y(t) . It depends only on the damping ratio ~ . If ymax is larger than
the final value y(oc) = 1, the response is said to have an overshoot. From Figure
4.7, we see that the response has an overshoot if ~ < 1 . If ? 1, then there is no
overshoot.
We consider again the unit-step response y(t) in (4.10) . If the damping ratio
is zero, or o- = icon = 0, then a -a` sin ((od t + 0) reduces to a pure sinusoidal
function and y(t) will remain oscillatory for all t. Thus the system in (4.8) is said to

2 3 4 5 6 7 8 9 10
wn t
Figure 4 .7 Responses of quadratic system with various damping ratios.
be undamped. If 0 < ~ < 1, the response y(t) contains oscillation whose envelope
decreases with time as shown in Figure 4 .7 . In this case, the system in (4.8) is said
to be underdamped. If ~ > 1, the two poles of (4.8) are real and distinct and the
unit-step response of (4.8) will contain neither oscillation nor overshoot . In this case,
the system is said to be overdamped. The system is said to be critically damped if
~ = 1 . In this case, the two poles are real and repeated, and the response is on the
verge of having overshoot or oscillation .
The step response of (4.8) is dictated by the natural frequency w, and the damp-
ing ratio ~. Because the horizontal coordinate of Figure 4 .7 is wt, the larger w,,, the
faster the response . The damping ratio governs the overshoot ; the smaller ~, the
larger the overshoot. We see from Figure 4 .7 that, if ~ is in the neighborhood of 0 .7,
then the step response has no appreciable overshoot and has a fairly fast response .
Therefore, we often design a system to have a damping ratio of 0 .7. Because the
response also depends on w,,, we like to control both w and ~ in the design .
Example 4 .3.1
The transfer function of the automobile suspension system shown in Figure 2 .7 is
1 1/m
G(s) =
ms2 +kls+k2
s2 +m
ks + k2
m

4.3 SECOND-ORDER SYSTEMS 121
If the shock absorber is dead (that is, it does not generate any friction), or k, = 0,
then the damping ratio ~ is zero . In this case, the car will remain oscillatory after
hitting a pothole and the car will be difficult to steer . By comparing the transfer
function of the suspension system and (4.8), we have
k2 _ W2
' = 2~con
m " m
If we choose ~ = 0.7 and con = 2, then from Figure 4 .7, we can see that the
automobile will take about 2 seconds to return to the original horizontal position
after hitting a pothole and will hardly oscillate. To have these values, k, and k2 must
be
k2 =22m=4m k, =2 .0.7 .2m=2.8m
Thus, the suspension system of an automobile can be controlled by using suitable
k, and k2 .
Exercise 4 .3.1
Find the damping ratio, damping factor, natural frequency, and actual frequency of
the following systems . Also classify them in terms of dampedness .
9
a. G(s) =
2s 2 + 9
9
b. G(s) =
s2 +3s+9
9
c. G(s)= s2+
12s + 9
[Answers : (a) 0, 0, \/4 .5, \/4 .5, undamped; (b) 0 .5, 1 .5, 3, 2 .6, underdamped ;
(c) ~ = 2, wn = 3, the other two not defined, overdamped .]
With the preceding discussion, we are now ready to study the position control
system in (4.7) . Its block diagram was developed in Figure 3 .17(b) and is repeated
in Figure 4.8(a). In the block diagram, km and T,,, are fixed by the motor and load .
The amplifier gain k2 clearly can be adjusted ; so can the sensitivity of the error
detector (by changing the power supply E) . Although both k, and k2 can be changed,
because
1
to, = \/k,k 2 km /T m and ~ = 2TmWn
only one of to,, and ~ can be arbitrarily assigned . For example, if k, and k2 are chosen
so that wn = 10, we may end up with ~ = 0 .05 . If k, and k2 are chosen so that
r + u km y
S(ms+1)
(a)
u km y
S(Tn S + 1)
(b)
Figure 4.8 Position control systems .
~ = 2, we may end up with w, = 0 .2 . Their step responses are shown in Figure

4.9. The former has too much oscillation ; the latter has no oscillation but is much
too slov~ Thus both responses are not satisfactory . How to choose k 1 and k2 to yield
a satisfactory response is a design problem and will be discussed in later chapters .
Exercise 4.3 .2
(a) Consider the position control system in (4.7). Suppose T1 = 4 and k= 0.25 ;
find k1 and k2 so that w n = 0.25. What is ~? Use Figure 4.7 to sketch roughly its
unit-step response . (b) Can you find k 1 and k2 so that (o,, = 0.25 and = 0 .7?
[Answers : k1 k2 = 1, ~ = 0 .5, no.]
y (t)
4.4 TIME RESPONSES OF POLES 1 23
Exercise 4 .3.3
Suppose the position control system in Figure 4 .8(a) cannot achieve the design ob-
jective . We then introduce' an additional tachometer feedback with sensitivity k3 as
shown in Figure 4.8(b). Show that its transfer function from r to y is
k1 k2km
Y(s) = T m = wn z
G,(s) _
R(s) s2 + 1 + k2k3km 2 + 2~w n s + wn
s + klk2km . s
Tm / Tm
For this system, is it possible to assign ~ and w n arbitrarily by adjusting k 1 and k2?
If Tm = 4, km = 0.25, and k3 = 1, find k1 and k2 so that to, = 0.25 and = 0.7.
[Answers : Yes, k 1 = 1/1 .6, k2 = 1 .6.]
4 .4 TIME RESPONSES OF POLES
From the preceding two sections, we see that poles of overall transfer functions
essentially determine the speed of response of control systems . In this section, we
shall discuss further the time response of poles .
Poles can be real or complex, simple or repeated . It is often convenient to plot
them on the complex plane or s-plane as shown in Figure 4 .10. Their corresponding
responses are also plotted . The s-plane can be divided into three parts : the right half
plane (RHP), the left half plane (LHP) and the pure imaginary axis or jw-axis . To
avoid possible confusion whether the RHP includes the jw-axis or not, we shall use
the following convention: The open RHP is the RHP excluding the j(0-axis and the
closed RHP is the RHP including the jw-axis . If a pole lies inside the open LHP,
then the pole has a negative real part ; its imaginary part can be positive or negative .
If a pole lies inside the closed RHP, then the pole has a positive or zero real part .
Poles and zeros are usually plotted on the s-plane using crosses and circles . Note
that no zeros are plotted in Figure 4 .10. Consider 1/(s + a)' or the pole at - a with
multiplicity n. The pole is a simple pole if n = 1, a repeated pole if n > 1 . To
simplify the discussion, we assume a to be real . Its time response, using Table A .1,
is
1 t o-I e -at (4.14)
n1!
If the pole - a is in the open RHP, or a < 0, then its response increases exponentially
to infinity for n = 1, 2 If the pole is at the origin, or a = 0, and is simple,
then its response is a step function . If it is repeated, with multiplicity n ? 2, then
its response is to -1 /n!, which approaches infinity as t -* oo . If the real pole is
in the open LHP, or a > 0, and is simple, then its response is a -at, which decreases

Im s
e 2.4t te at
1 1 i
0 .37
X Res
X
A4 A3 -a A1
eA3t
t
0 1 0
IA 3
(a)
Im s et cos (t), t
w2
X -
I 2,r
I
I
I Re s
y I 0
e at cosw 2 t
I
I
I (0 1 -
X
I
12-1
1
s-6-jw i + s-6+jw
1 i
=e ot Cosco .t
(b)
Figure 4 .10 Responses of real and complex poles .
exponentially to zero as t If it is repeated with multiplicity 2, then its response

is
at
to

4 .5 STABILITY 125
the product of t, which goes to 00, and e - `, which goes to 0 as t -* oo . Therefore,

it requires some computation to find its value at t -3 oo. We use l'Hopital's rule to
compute
t 1
lim te at = lim - = limt-. ae ` = 0
t__ , eat
Thus, as plotted in Figure 4.10(a), the time response te-` approaches zero as
t - ~ . Similarly, we can show
t"e - at _ 0 as t -- o0
for a > 0, and n = 1, 2, 3, . . . . This is due to the fact that the exponential e - at
with a > 0, approaches zero with a rate much faster than the rate at which t"
approaches infinity . Thus, we conclude that the time response of any simple or
repeated real pole that lies inside the open LHP approaches 0 as t -* 0.
The situation for complex conjugate poles is similar to the case of real poles
with the exception that the responses go to 0 or 00 oscillatorily . Therefore we will
not repeat the discussion . Instead we summarize the preceding discussion in the
following table :
Table 4 .1 Time Responses of Poles as t
Poles Simple (n = 1) Repeated (n ? 2)

Open LHP 0 0
Open RHP Cc x
Origin (s") A constant cc
ja-axis((s 2 + a2)") A sustained oscillation +oo
This table implies the following facts, which will be used later .
1. The time response of a pole, simple or repeated, approaches zero as t -* 00 if
and only if the pole lies inside the open LHP or has a negative real part .
2. The time response of a pole approaches a nonzero constant as t -3 00 if and only
if the pole is simple and located at s = 0.
4 .5 STABILITY
In this section, a qualitative property of control systems-namely, stability-will be

introduced . The concept of stability is very important because every control system
must be stable. If a control system is not stable, it will usually bum out or disinte-
grate . There are three types of stability, bounded-input bounded-output (BIRO) sta-
bility, marginal stability (or stability in the sense of Lyapunov), and asymptotic

stability . In this text, we study only BIBO stability . Therefore, the adjective BIBO
will be dropped .
A function u(t) defined for t ? 0 is said to be bounded if its magnitude does
not approach infinity or, equivalently, there exists a constant M such that
~u(t)J !5 M < -
for allt?0l
o Definition 4 .1
A system is stable if every bounded input excites a bounded output. Otherwise
the system is said to be unstable .
Example 4.5.1
Consider the network shown in Figure 4 .11(a) . The input u is a current source ; the
output y is the voltage across the capacitor . Using the equivalent Laplace transform
circuit in Figure 4 .11(b), we can readily obtain
1
S -
S S
(4.15)
Y(s) = 1 U(s) = s 2 + 1 U(s)
s+-
s
If we apply the bounded input u(t) = 1, for t ? 0, then the output is
_ s 1 1
Y(s)
s2 + 1 s s2 + 1
which implies
y(t) = sin t
It is bounded. If we apply the bounded input u(t) = sin at, for t ? 0, where a is a
positive real constant and a 1, then the output is
s a as[(s2 + a2) - (s2 + 1)]
Y(s) _
S 2 + 1 s2 + as (a2 - 1)(s 2 + 1)(s2 + a2)
a s a s
a2 - 1 s2 + 1 a2 - 1 s2 + a 2
(a) (b)

4 .5 STABILITY 127
which implies
a
y(t) = az _ 1 [cos t - cos at]
It is bounded for any a 0 1 . Thus the outputs due to the bounded inputs u(t) = 1
and sin at with a = 1 are all bounded. Even so, we cannot conclude the stability of
the network because we have not yet checked every possible bounded input . In fact,
the network is not stable, because the application of u(t) = sin t yields
s 1 S
Y(s) = _
s 2 + 1 s2 + 1 (s2 + 1)2
which, using Table A.1, implies
1
y(t) = t sin t
2
This output y(t) approaches positive or negative infinity as t - oc . Thus the bounded
input u(t) = sin t excites an unbounded output, and the network is not stable .
Exercise 4 .5 .1
Consider a system with transfer function 1/s . It is called an integrator. If we apply

the bounded input sin at, will the output be bounded? Can you find a bounded input
that excites an unbounded output? Is the system stable?
[Answers : Yes, step function, no .]
The instability of a system can be deduced from Definition 4 .1 by finding a

single bounded input that excites an unbounded output . However, it is difficult to
deduce stability from the definition because there are infinitely many bounded inputs
to be checked . Fortunately we have the following theorem.
THEOREM 4 .1
A system with proper rational transfer function G(s) is stable if and only if every
pole of G(s) has a negative real part or, equivalently, lies inside the open left
half s-plane .
By open left half s-plane, we mean the left half s-plane excluding the jw-axis .
This theorem implies that a system is unstable if its transfer function has one or
more poles with zero or positive real parts . This theorem can be argued intuitively
by using Table 4 .1 . If a transfer function has one or more open right half plane poles,
then most bounded inputs will excite these poles and their responses will approach

infinity. If the transfer function has a simple pole on the imaginary axis, we may
apply a bounded input whose Laplace transform has the same pole . Then its response
will approach infinity . Thus a stable system cannot have any pole in the closed right
half s-plane . For a proof of the theorem, see Reference [151 or [181 .
We remark that the stability of a system depends only on the poles of its transfer
function G(s) and does not depend on the zeros of G(s). If all poles of G(s) lie inside
the open LHP, the system is stable no matter where the zeros of G(s) are . For
convenience, a pole is called a stable pole if it lies inside the open LHP or has a
negative real part. A pole is called an unstable pole if it lies inside the closed RHP
or has a zero or positive real part . A zero that lies inside the open LHP (closed RHP)
will be called a minimum-phase (nonminimum-phase) zero . The reason for using
such names will be given in Chapter 8 .
Now we shall employ Theorem 4 .1 to study the stability of the network in Figure
4.11 . The transfer function, as developed in (4.15), is
s
G(s) =
s2 + 1
Its poles are j ; they have zero real part and are unstable poles . Thus, the network
is not stable .
Most control systems are built by interconnecting a number of subsystems . In
studying the stability of a control system, there is no need to study the stability of
its subsystems. All we have to do is to compute the overall transfer function and
then apply Theorem 4 .1. We remark that a system can be stable with unstable sub-
systems and vice versa . For example, consider the system in Figure 4 .12(a) . It con-
sists of two subsystems with transfer functions -2 and 1/(s + 1) . Both subsystems
are stable. However, the transfer function of the overall feedback system is
-2
s + 1 _ -2 _ -2
G0(s) =
-2 s+1-2 s-1
1 +
S + 1
which is unstable . The overall system shown in Figure 4 .12(b) is stable because its
transfer function is
2
s - 1 2 2
G0 (s) =
2 s- 1+ 2 s+ 1
1+
s - 1
-2
(a) (b)
Figure 4.12 Stability of overall system and subsystems .

4 .6 THE ROUTH TEST 129
Its subsystem has transfer function 2/(s - 1) and is unstable . Thus the stability of
a system is independent of the stabilities of its subsystems. Note that a system with
transfer function
s2 -2s-3
G(s) = ( 4.16)
(s + 2)(s - 3)(s + 10)
is stable, because 3 is not a pole of G(s). Recall that whenever we encounter rational
functions, we reduce them to irreducible ones . Only then are the roots of the denom-
inator poles . Thus, the poles of G(s) = (s - 3)(s + 1)/(s + 2)(s - 3)(s + 10)
= (s + 1)/(s + 2)(s + 10) are - 2 and - 10. They are both stable poles, and the
transfer function in (4.16) is stable .
Exercise 4 .5.2
Are the following systems stable?

- 1
a.
+ 1
S2 - 1
b. s2
+2s+2
- 1
C.
S2 - 1
d. The network shown in Figure 4 .13(a)

e. The feedback system shown in Figure 4 .13(b)
2H 1H
r + Y
1t 2F 2
(a) (b)
Figure 4 .13 (a) Network. (b) Feedback system.
[Answers : All are stable .]
4 .6 THE ROUTH TEST
Consider a system with transfer function G(s) = N(s)/D(s) . It is assumed that N(s)
and D(s) have no common factor . To determine the stability of G(s) by using Theo-

rem 4 .1, we must first compute the poles of G(s) or, equivalently, the roots of D(s) .
If the degree of D(s) is three or higher, hand computation of the roots is complicated .
Therefore, it is desirable to have a method of determining stability without solving
for the roots . We now introduce such a method, called the Routh test or the Routh-
Hurwitz test.
o Definition 4.2
A polynomial with real coefficients is called a Hurwitz polynomial if all its roots
have negative real parts .
The Routh test is a method of checking whether or not a polynomial is a Hurwitz

polynomial without solving for its roots . The most important application of the test
is to check the stability of control systems . Consider the polynomial
D(s) = a nsn + an _ 1s n-1 + + als + ao an > 0 (4.17)
where a;, i = 0, 1, . . . , n, are real constants . If the leading coefficient an is negative,
we may simply multiply D(s) by - 1 to yield a positive an. Note that D(s) and
-D(s) have the same set of roots ; therefore, an > 0 does not impose any restriction
on D(s) .
Necessary Condition for a Polynomial To Be Hurwitz

We discuss first a necessary condition for D(s) to be Hurwitz . If D(s) in (4.17) is
Hurwitz, then every coefficient of D(s) must be positive. In other words, if D(s) has
a missing term (a zero coefficient) or a negative coefficient, then D(s) is not Hurwitz .
We use an example to establish this condition . We assume that D(s) has two real
roots and a pair of complex conjugate roots and is factored as
D(s) = a4(s + a,)(s + a2)(s + (31 + jy 1)(s + 01 - j'y,)
(4.18)
= a4 (s + ai)(s + a2)(s2 + 2(3,s + /3i + y 2)
The roots of D(s) are - a1, - a2, and - /3 1 jy1 . If D(s) is Hurwitz, then a 1 > 0,
a2 > 0, and (3, > 0. Note that y1 can be positive or negative . Hence, all coefficients
in the factors are positive . It is clear that all coefficients will remain positive after
multiplying out the factors . This shows that if D(s) is Hurwitz, then its coefficients
must be all positive . This condition is also sufficient for polynomials of degree 1 or
2 to be Hurwitz. It is clear that if a1 and ao in a 1 s + ao are positive, then a1s + ao
is Hurwitz . If the three coefficients in
D(s) = a2 S2 + a1 s + a o (4.19)
are all positive, then it is Hurwitz (see Exercise 4.6.2). In conclusion, for a poly-
nomial of degree 1 or 2 with a positive leading coefficient, the condition that all
coefficients are positive is necessary and sufficient for the polynomial to be Hurwitz .
However, for a polynomial of degree 3 or higher, the condition is necessary but not

sufficient . For example, the polynomial

s3 + 2s 2 + 9s + 68 = (s + 4)(s - 1 + 4j)(s - 1 - 4j)
is not Hurwitz although its coefficients are all positive .
Necessary and Sufficient Condition

Now we discuss a necessary and sufficient condition for D(s) to be Hurwitz . For
easy presentation, we consider the polynomial of degree 6 :
D(s) = a 6 s6 + a s s n + a4 S4 + a3 S3 + a2S2 + a 1 s + ao a6 > 0
We form Table 4 .2. 2 The first two rows are formed from the coefficients of D(s) .
The first coefficient (in descending power of s) is put in the (1, 1) position, the
second in the (2, 1) position, the third in the (1, 2) position, the fourth in the (2, 2)
position and so forth. The coefficients are renamed as shown for easy development
of recursive equations . Next we compute k5 = b 61 /b 51 ; it is the ratio of the first
elements of the first two rows . The third row is obtained as follows . We subtract the
product of the second row and k5 from the first row :
b40 = b61 k5 b 51 b41 = b62 - k5b52
= b63 - k5 5 53
b42 b43 - b64 k5 . 0
Note that b40 is always zero. The result is placed at the right hand side of the second
row. We then discard the first element, which is zero, and place the remainder in the
third row as shown in Table 4 .2. The fourth row is obtained in the same manner
from its two previous rows . That is, we compute k4 = b51/b41, the ratio of the first
elements of the second and third rows, and then subtract the product of the third row
and k4 from the second row:
b30 - b51 k4b41 b31
= b52 - k4b42 b32 = b53 - k4 b43
We drop the first element, which is zero, and place the remainder in the fourth row
as shown in Table 4 .2. We repeat the process until the row corresponding to s 0 =
1 is obtained. If the degree of D(s) is n, there should be a total of (n + 1) rows . The
table is called the Routh table .
We remark on the size of the table . If n = deg D(s) is even, the first row has
one more entry than the second row . If n is odd, the first two rows have the same
number of entries . In either case, the number of entries decreases by one at odd
powers of s . For example, the number of entries in the rows of s 5 , s 3 , and s is one
less than that of their preceding rows . We also remark that the rightmost entries of
the rows corresponding to even powers of s are the same . For example, in Table
4.2, we have b64 = b43 = b22 = bo1 = ao .
'The presentation is slightly different from the cross-product method ; it requires less computation and is
easier to program on a digital computer . See Problem 4.16 .

Table 4 .2 The Routh Table
S6 b 61 : = a6 b 62 : = a4 b63 : = a2 b64 : = ao
k5 b61 = (1st row) - k 5 (2nd row)

= S5 b51 . a5 b 52 : = a3 b53 : = a1
b51 = [b40 b41 b42 b431
i
b51 (2nd row) - k4(3rd row)
k4 = S4 b4l b42 b43
b41 = [b30 b3l b 32 b]
~
b41 (3rd row) - k3 (4th row)
k3 = S3 b31 b32
b31 i _ [b2o b21 b221
b31 (4th row) - k2 (5th row)

k2 = S2 b21 b22
b21 = [b1o b111
----------
kl = b21 (5th row) - k l (6th row)

bl , b11
= [boo boll
so bol
THEOREM 4 .2 (The Routh Test)

A polynomial with a positive leading coefficient is a Hurwitz polynomial if and
only if every entry in the Routh table is positive or, equivalently, if and only if
every entry in the first column of the table (namely, b61, b51, b41, b31 , b21, b11,
b 01 ) is positive.
It is clear that if all the entries of the table are positive, so are all the entries in
the first column . It is rather surprising that the converse is also true . In employing
the theorem, either condition can be used . A proof of this theorem is beyond the
scope of this text and can be found in Reference [18] . This theorem implies that if
a zero or a negative number appears in the table, then the polynomial is not Hurwitz .
In this case, it is unnecessary to complete the table .
Example 4 .6 .1
Consider 2s 4 + s 3 + 5s 2 + 3s + 4 . We form
S4 2 5 4 ;
_ 2
S3 1 3 [0 - 1 4] = (1st row) - k3 (2nd row)
k3 1
S2 -1 4 ;
Clearly we have k3 = 2/1, the ratio of the first entries of the first two rows . The
result of subtracting the product of the second row and k3 from the first row is placed
on the right hand side of the s 3 -row. We drop the first element, which is zero, and

4.6 THE ROUTH TEST 133
put the rest in the s 2 -row. A negative number appears in the table, therefore the
polynomial is not Hurwitz.
Example 4 .6 .2
Consider 2s 5 + s4 + 7s 3 + 3s 2 + 4s + 2 . We form
S5 2 7 4
_ 2
S4 1 3 2 ; [0 1 0] = (1st row) - k4(2nd row
k4 1
S3 1 0
A zero appears in the table, thus the polynomial is not Hurwitz . The reader is advised
to complete the table and verify that a negative number appears in the first column .
Example 4 .6 .3
Consider 2s 5 + s4 + 7s 3 + 3s 2 + 4s + 1 .5. We form
S5 2 7 4
2
k4 S4 1 3 1.5 [0 1 1 ] = (1st row) - k4(2nd row)
1 i
1 i
k3 S3 1 1 [0 2 1 .5] = (2nd row) - k3(3rd row)
1
1
k2 S2 2 1 .5 [0 0.25] = (3rd row) - k2(4th row)
2
2
k, Si 0.25 [0 1 .5] = (4th row) - k,(5th row)
0.25
S 1 .5 ,
Every entry in the table is positive, therefore the polynomial is Hurwitz .
Exercise 4 .6.1
Are the following polynomials Hurwitz?

a. 2s4 +2s 3 +3s+2
b. S4 + S 3 + S 2 + S + 1

c . 2S4 + 2s +
d. 2S4 + 5S3 + 5s2 + 2s
e. s5 +3s4 + 10s3 + 12s 2 +7s+3
[Answers : No, no, no, yes, yes .]
Exercise 4 .6 .2
Show that a polynomial of degree 2 is Hurwitz if and only if the three coefficients
of a2s2 + a1s + ao are of the same sign .
Exercise 4 .6 .3
Show that the polynomial

S 3 +a2s2 +a1s+ao
is Hurwitz if and only if
a2 >0 ao >0 and

ao
a
ai - 2 >0
The most important application of the Routh test is to check the stability of
control systems . This is illustrated by an example .
Example 4 .6 .4
Consider the system shown in Figure 4 .14. The transfer function from r toy is, using
Mason's formula,
Y(s)
G0(s) =
R(s)
2s + 1 I
s + 2 s(s2 + 2s + 2)
-2(s- 1) 2s + 1
f (s + 1)(s 2 + 2s + 2) (s + 2)s(s 2 + 2s + 2)]
(2s + 1)(s + 1)
(s + 1)(s + 2)s(s 2 + 2s + 2) + 2(s - 1)s(s + 2) + (2s + 1)(s + 1)

r + 2s+ 1 I
__0__ s+2 s(s 2 +2s+2)
Tachometer
1
0 2s
s+1
Figure 4.14 Feedback system.

(2s + 1)(s + 1)
G0(s)
s5 + 5s4 + 12s 3 + 145 2 + 3s + I
We form the Routh table for the denominator of G0(s):
S5 1 12 3
1
k4 = 5 = 0.2 S4 5 14 1 [0 9.2 2.8]
5
k3 = =0 .54 S3 9.2 2 .8 [0 12 .48 1]
92
k2 = 0.74 s2 12.48 1 [0 2.06]
kl = 6.05 s1 2.06 [0 1 ]
so 1
Because every entry is positive, the denominator of G0(s) is a Hurwitz polynomial .
Thus all poles of GO(s) lie inside the open left half s-plane and the feedback system
is stable .
To conclude this section, we mention that the Routh test can be used to deter-
mine the number of roots of D(s) lying in the open right half s-plane . To be more
specific, if none of the entries in the first column of the Routh table is zero, then the
number of changes of signs in the first column equals the number of open RHP roots
of D(s) . This property will not be used in this text and will not be discussed further .
4.6.1 Stability Range

In the design of a control system, we sometimes want to find the range of a parameter
in which the system is stable . This stability range can be computed by using the
Routh test. This is illustrated by examples .

Example 4 .6.5
Consider the system shown in Figure 4 .15 . If
8
(4.20)
G(s) = (s + 1)(s 2 + 2s + 2)
then the transfer function from r to y is
8k 8k
Go(s) _ (s + 1)(s 2 + 2s + 2) + 8k s 3 + 3s 2 + 4s + (8k + 2)
y
G (s)
Figure 4.15 Unity-feedback system .
We form the Routh table for its denominator:

s3 1 4
k2 = 31 s
2
3 2 + 8k 4 -
2 38k] = [o 1038k]j
10
9 10 - 8k
sI [0 2 + 8k]
k_
' 10 - 8k 3
s 2 + 8k
The conditions for G 0 (s) to be stable are
10 - 8k
> 0 and 2 + 8k > 0
3
These two inequalities imply
1 .25 = > k and k > = - 0.25 (4.21)

8 82
They are plotted in Figure 4.16(a) . From the plot we see that if 1 .25 > k > - 0.25,
then k meets both inequalities, and the system is stable .
I<
i I I I I
-0.25 0 1 .25 -7 .04 -5 0 3 .6 5 .54
(a) (b)
Figure 4 .16 Stability ranges .

I
Example 4 .6 .6
Consider again Figure 4.15 . If

- 1 + j2)(s - 1 - j2) s2 - 2s + 5
Gs = (s (4 .22)
(s- 1)(s + 3 + j3)(s + 3 - j3) s3 +5s 2 + 12s - 18
then the overall transfer function is
s 2 -2s+5
k
kG(s) - s3 + 5s 2 + 12s - 18
G ,(s) =
1 + kG(s) s2 -2s+5
1 +k 2 + 12s - 18
s 3 +5s
k(s 2 - 2s + 5)
s 3 +(5+k)s2 +(12-2k)s+5k- 18
We form the Routh table for its denominator :
S 3 1 12 - 2k
5k - 1k8 =:
5+k 5k-18 0 (12-2k)- [0 x]
5 +
x [0 5k - 181
x
S 5k - 18
The x in the table requires some manipulation :
-2k2 - 3k + 78
(12 - 2k)(5 + k) - (5k - 18)
= -
5 + k 5 + k
-2(k + 7 .04)(k - 5.54)
5 + k
Thus the conditions for G,(s) to be stable are
5+k>0 5k- 18>0
and
x - -2(k + 7.04)(k - 5.54) > 0
5 + k
These three inequalities imply
18
k > -5 k > = 3.6 (4 .23a)
5
and
(k + 7 .04)(k - 5 .54) < 0 (4 .23b)

Note that, due to the multiplication by - 1, the inequality is reversed in (4 .23b) . In

order to meet (4 .23b), k must lie between (-7 .04, 5 .54) . The three conditions in
(4.23) are plotted in Figure 4 .16(b) . In order to meet them simultaneously, k must
lie inside (3.6, 5.54). Thus, the system is stable if and only if
3.6 < k < 5.54
Exercise 4 .6 .4
Find the stability ranges of the systems shown in Figure 4 .17 .
I
y r +
s+a I
k
s+2 s(s+1)
(a) (b)
Figure 4 .17 Feedback systems .
[Answers : (a) 0 < k < cc . (b) 0 < a < 9 .]
4.7 STEADY STATE RESPONSE OF STABLE SYSTEMS-POLYNOMIAL INPUTS
Generally speaking, every control system is designed so that its output y(t) will track
a reference signal r(t). For some problems, the reference signal is simply a step
function, a polynomial of degree 0 . For others, the reference signal may be more
complex . For example, the desired altitude of the landing trajectory of a space shuttle
may be as shown in Figure 4 .18 . Such a reference signal can be approximated by
r(t) =ro +r a t+rz t 2 + . . +r,,,t"'
a polynomial of t of degree m . Clearly, the larger m, the more complex the reference
signal that the system can track . However, the system will also be more complex .
> I
0
Figure 4 .18 Time function.

4 .7 STEADY-STATE RESPONSE OF STABLE SYSTEMS-POLYNOMIAL INPUTS 139
In practice, control systems are designed to track

r(t) = a (step function)
r(t) = at (ramp function)
or
r(t) = at e (acceleration function)
where a is a constant. These are polynomials of degree 0, 1 and 2 . The first two are
used more often . For example, the temperature setting in a thermostat provides a
step reference input, called the set point in industry . To reset the set point means to
change the amplitude of the step reference input .
Although it is desirable to design a system so that its output y(t) will track the
reference input r(t) immediately-that is, y(t) = r(t), for all t ? 0-it is not possible
to achieve this in practice . The best we can hope for is
lim y(t) = r(t)
that is, y(t) will track r(t) as t approaches infinity . This is called asymptotic tracking,
and the response
ys(t) : = lim y(t)
t__
is called the steady-state response . We will now compute the steady-state response
of stable systems due to polynomial inputs .
Consider a system with transfer function
o+ . . . +1 m Sm
G(s) Y(s) I S+
(4 .24)
R(s) ao + a ls + + a"s"
with n ? m. It is assumed that G(s) is stable, or that all the poles of G(s) have
negative real parts . If we apply the reference input r(t) = a, for t ? 0, then the
output is given by
No +f1s + . . + Nmsm a
Y(s) = G (s)R(s)
= a 0+ a s 1 + + a ns" s
k
_ - + (terms due to the poles of G(s))
s
with k given by, using (A.8b),
k = Go
(s) a- . s = G (0)a =
130
a
s S=0 ao
If the system is stable, then the response due to every pole of G(s) will approach
zero as t Thus the steady-state response of the system due to r(t) = a is
ys(t) = lim y(t) = G(0)a = RO a (4.25)

t__ ao

This steady-state response depends only on the coefficients of G0 (s) associated

with so.
Now we consider the ramp reference input . Let r(t) = at. Then
and
a
Y(s) = G"(s) 2
s
k2
+ + (terms due to the poles of G 0(s))
sl
with, using (A.8c) and (A.8d),
a2
k2 = G0(s) . s2 = G O(0)a
s s=0
and
d
k, - G0(s)a
ds s=o
(ao + a,s + . . . + a"s")(,(3, + + rn/3 m sm -1 )
a
(ao + a,s + + an s")2
_ (/o + 0 1 s+ . . . + /msm)(a1 + . . . + na"s " 1)
(ao + a,s + . . . + a" S")2 s=0
a ao1a1-Goal
2
ao
Thus the steady-state response of the system due to r(t) = at equals

ys(t) = Go(0)at + Go(0)a (4.26a)
or
= Ro . at + a aoR1- 130a,
YS (t) (4.26b)
ao ao
This steady-state response depends only on the coefficients of G0(s) associated with
s0 and s .
We discuss now the implications of (4 .25) and (4.26). If Go(0) = 1 or ao = Ro
and if r(t) = a, t ? 0, then
y s(t) = a = r(t)
Thus the output y(t) will track asymptotically any step reference input . If Go(0) _
1 and G,(0) = 0 or ao = / 3 and a, = /3,, and if r(t) = at, t =f 0, then (4.26)
reduces to
ys (t) = at

that is, y(t) will track asymptotically any ramp reference input . Proceeding forward,
if
ao =/3 o a1 =/3 1 and a2 =(32 (4.27)
then the output of G (s) will track asymptotically any acceleration reference input
ate. Note that in the preceding discussion, the stability of G (s) is essential . If G(s)
is not stable, the output of G (s) will not track any r(t).
Exercise 4 .7.1
Find the steady-state responses of

1
a. G (s) = 1 - s2 due to r(t) = a
b. G(s) = s + 1 due to r(t) = a
2 + 3s
c. G(s) = due to r(t) = 2 + t
2+3s+s 2
2
d. G(s) = s + 1 due to r(t) = 3t
68+9s+9s 2
e. G(s) = 68 + 9s + 9s 2 + s3 due to r(t) = a
[Answers : (a) o ; (b) 2a; (c) ys(t) = 2 + t; (d) 6t - 6; (e) cc .]
4.7.1 Steady-State Response of Stable Systems-Sinusoidal Inputs

Consider a system with proper transfer function G(s) = Y(s)/R(s) . It is assumed
that G (s) is stable . Now we shall show that if r(t) = a sin wt, then the output y(t)
will approach a sinusoidal function with the same frequency as t ~.
If r(t) = a sin (qt, then, using Table A . 1,
a (01
R(s) = (4.28)
s2 +w0
Hence, we have
Y(s) = G(s)R(s) = G(s) s2 +w2 = G(s) - (s + j,,,
)(s - jto)
Because G(s) is stable, s = jw are simple poles of Y(s) . Thus Y(s) can be
expanded as, using partial fraction expansion,
k1 kl
Y(s) = + + terms due to the poles of G(s)
s-iw s+jw

with
awo awo a
ki = G 0 (s) = G0(jw) 2 w~ = 2j. Go(jwo)
s + jwo S =jw j o 2
and
* a
ki = 2j. Go( -jwo)
Since all the poles of G 0(s) have negative real parts, their time responses will ap-
proach zero as t - - . Hence, the steady-state response of the system due to r(t) _
a sin wo t is given by
aG(jwo) aGo( -jwo)
YS(t) (4 .29)
2j(s - jwo) 2j(s + jwo)
All coefficients of G 0 (s) are implicitly assumed to be real . Even so, the function
G 0(jwo ) is generally complex . We express it in polar form as
G0(j(oo) = A((o o )eje(-o) (4 .30)
where
A(wo) : = 1G o (j(o o)I = [(Re Go( j(00))2 + (Im Go(jwo))2I (12
and
ImG0(jwo)
0(wo ) :_ < G0(jwo) := tan-' Re
G0(j(oo)
where Im and Re denote, respectively, the imaginary and real parts. A((0o ) is called
the amplitude and 0(wo ), the phase of G0 (s) . If all coefficients of G 0(s) are real,
then A(too) is an even function of wo , and 0(coo ) is an odd function of w0 ; that is,
A(- wo) = A(too) and 0(- wo) = - 0(wo). Consequently we have
Go( - jwo) = A(-wo)eja(-w) = A(wo)e -j e(w) (4 .31)
The substitution of (4 .30) and (4 .31) into (4 .29) yields

aA((o o)ej o(o') ejt - aA((oo )e -j 0( o) e -jt
YS (t) = 2j
2j
e j[) r+B(w )1 - e - jfw t+O(w,)
= aA(wo) (4 .32)
2jj
= aA(wo ) sin (wot + 0(wo ))
This shows that if r(t) = a sin wot, then the output will approach a sinusoidal
function of the same frequency . Its amplitude equals ajG0(jwo )j ; its phase differs
-
from the phase of the input by tan ( [Im G 0 (jwo )/Re Go ( jwo )] . We stress again that
(4 .32) holds only if G 0 (s) is stable.

4.7 STEADY-STATE RESPONSE OF STABLE SYSTEMS-POLYNOMIAL INPUTS 143
Example 4 .7 .1
Consider G0(s) = 3/(s + 0.4) . It is stable . In order to compute its steady-state

response due to r(t) = sin 2t, we compute
Go(j2) = 3 = 3j1 37 - 1 .47e - '1 '37 (4.33)

j2+0.4 2.04e
Thus the steady-state response is
ys(t) = lim y(t) = 1 .47 sin (2t - 1.37) (4.34)
t-x
Note that the phase - 1 .37 is in radians, not in degrees . This computation is very
simple, but it does not reveal how fast the system will approach the steady state .
This problem is discussed in the next subsection .
Exercise 4.7 .2
Find the steady-state response of 2/(s + 1) due to (a) sin 2t (b) I + sin 2t
(c) 2 + 3 sin 2t - sin 3t.
[Answers : (a) 0.89 sin (2t - 1.1). (b) 2 + 0 .89 sin (2t - 1 .1) . (c) 4 + 2 .67
sin (2t - 1 .1) - 0.63 sin (3t - 1 .25).]
The steady-state response of a stable G 0(s) due to sin mot is completely deter-
mined by the value of G0 (s) at s = jwo. Thus G(jco) is called the frequency response
of the system . Its amplitude A(co) is called the amplitude characteristic, and its phase
0(co), the phase characteristic. For example, if G0(s) = 2/(s + 1), then Go(0) =
2, Go(j1) = 2/(j l + 1) = 2/(1 .4ej45') = 1.4e -j45 ', GO(j10) = 2/(j lO + 1) =
0.2e-j 84', and so forth . The amplitude and phase characteristics of G0(s) =
2/(s + 1) can be plotted as shown in Figure 4 .19. From the plot, the steady-state
response due to sin coot, for any wo, can be read out .
Exercise 4 .7 .3
Plot the amplitude and phase characteristics of G0(s) = 2/(s - 1). What is the
steady-state response of the system due to sin 2t?
[Answers : Same as Figure 4 .19 except the sign of the phase is reversed, infinity .
The amplitude and phase characteristics of unstable transfer functions
do not have any physical meaning and, strictly speaking, are not
defined.]

I G,(jw) 4G(jw)
Co
(a) (b)
Figure 4 .19 Amplitude and phase characteristics .
The frequency response of a stable G (s) can be obtained by measurement . We

apply r(t) = sin wot and measure the steady-state response . From the amplitude and
phase of the response, we can obtain A(w(,) and 0(w0). By varying or sweeping to,
G0(jw) over a frequency range can be obtained . Special devices called frequency
analyzers, such as the HP 3562A Dynamic System Analyzer, are available to carry
out this measurement. Some devices will also generate a transfer function from the
measured frequency response .
We introduce the concept of bandwidth to conclude this section . The bandwidth
of a stable G0(s) is defined as the frequency range in which 3
jG,(jw)j ? 0.707 Go(0)j (4 .35)
For example, the bandwidth of G,(s) = 2/(s + 1) can be read from Figure 4 .19 as
1 radian per second. Thus, the amplitude of G0(j(o) at every frequency within the
bandwidth is at least 70 .7% of that at w = 0 . Because the power is proportional to
the square of the amplitude, the power of G0(jw) in the bandwidth is at least
(0.707)2 = 0.5 = 50% of that at to = 0. Thus, the bandwidth is also called the
half-power bandwidth . (It is also called the - 3-dB bandwidth as is discussed in
Chapter 8 .) Note that if G0(s) is not stable, its bandwidth has no physical meaning
and is not defined .
4 .7 .2 Infinite Time
The steady-state response is defined as the response as t - c . Mathematically speak-
ing, we can never reach t = 00 and therefore can never obtain the steady-state
response. In engineering, however, this is not the case. For some systems a response
may be considered to have reached the steady state in 20 seconds . It is a very short
infinity indeed!
'This definition applies only to G(s) with lowpass characteristic as shown in Figure 4 .19 . More generally,
the bandwidth of stable G,(s) is defined as the frequency range in which the amplitude of G,(jw) is at
least 70 .7% of the largest amplitude of G(jn ).

Consider G0(s) = 3/(s + 0.4). If we apply r(t) = sin 2t, then ys(t) = 1 .47
sin (2t - 1 .37) (see [4 .34]). One may wonder how fast y(t) will approach ys(t). In
order to find this out, we shall compute the total response of G 0(s) due to sin 2t . The
Laplace transform of r(t) = sin 2t is 2/(s 2 + 4) . Thus we have
_ 3 2 6
Y(s)
s + 0.4 s2 + 4 (s + 0.4)(s + 2j)(s - 2j) (4 .36)
.37
i11 .44 .47e-i1
1
.37 1 .47e'
s + 0.4 2j(s - 2j) 2j(s + 2j)
which implies
y(t) = 1 .44e -0.41 + 1 .47 sin (2t - 1 .37) (4 .37)
Transient Steady-State
Response Response
The second term on the right hand side of (4 .37) is the same as (4.34) and is the
steady-state response . The first term is called the transient response, because it ap-
pears right after the application of r(t) and will eventually die out . Clearly the faster
the transient response approaches zero, the faster y(t) approaches ys(t). The transient
response in (4.37) is governed by the real pole at - 0.4 whose time constant is
defined as 1/0 .4 = 2 .5. As shown in Figure 4.2(b), the time response of
1/(s + 0 .4) decreases to less than 1% of its original value in five time constants or
5 X 2.5 = 12 .5 seconds . Thus the response in (4 .37) may be considered to have
reached the steady state in five time constants or 12 .5 seconds .
Now we shall define the time constant for general proper transfer functions . The
time constant can be used to indicate the speed at which a response reaches its steady
state. Consider
N(s) N(s)
G(s) = (4 .38)
D(s) (s + a t )(s + a2 )(s + Qt + jc)dl )(s + at - ju dt) . . .
If G(s) is not stable, the response due to its poles will not die out and the time
constant is not defined . If G(s) is stable, then a; > 0 and crt > 0. For each real pole
(s + a,), the time constant is defined as 1/a, . For the pair of complex conjugate
poles (s + o-t jcodt ), the time constant is defined as 1/u 1 ; this definition is
reasonable, because a-t governs the envelope of its time response as shown in Figure
4.6. The time constant of G(s) is then defined as the largest time constant of all poles
of G(s). Equivalently, it is defined as the inverse of the smallest distance of all poles
of G(s) from the imaginary axis . For example, suppose G(s) has poles - 1, -3,
-0.1 j2. The time constants of the poles are 1, 1/3 = 0 .33 and 1/0 .1 = 10 .
Thus, the time constant of G(s) is 10 seconds . In engineering, the response of G(s)
due to a step or sinusoid input will be considered to have reached the steady state
in five time constants . Thus the smaller the time constant or, equivalently, the farther
away the closest pole from the imaginary axis, the faster the response reaches the
steady-state response .

Exercise 4 .7.4
Find the time constants of

S + 10
a.
(s + 2)(s 2 + 8s + 20)
S
b.
(s + 0 .2)(s + 2)(s + 3)
c. A system with the pole-zero pattern shown in Figure 4 .20.
Which system will respond fastest?
Im s
A
-3
X -2
0 -I
Res
-3 -2 -1 0 1 2 3
O --1
X --2 X Pole
- 3 0 Zero
Figure 4 .20 Pole-zero pattern .

[Answers : (a) 0.5; (b) 5 ; (c) 1 ; (a).]
The time constant of a stable transfer function G(s) as defined is open to argu-
ment. It is possible to find a transfer function whose step response will not reach the
steady state in five time constants . This is illustrated by an example .
Example 4 .7.2
Consider G(s) = 1/(s + 1) 3. It has three poles at s = 1 . The time constant of
G(s) is 1 second . The unit-step response of G(s) is
1 1 1 -1 -1 -1
Y(s)
_ (s + 1)3 s s (s + 1)3 + (s + 1)2 + (s + 1)
or
y(t) = I - 0 .5t2e -` - to - ` - e t

PROBLEMS 147
1 2 3 4 5 10
0 i i I i t_i - t
Transient response
Figure 4 .21 Step response .
Its steady-state response is I and its transient response is

-0.5t2e -t - te t - et = -(0.5 t2 + t + 1)e - t
These are plotted in Figure 4 .21 . At five time constants, or t = 5, the value of the
transient response is -0.126 ; it is about 13% of the steady-state response . At
t = 9, the value of the transient response is 0 .007 or 0 .7% of the steady-state
response . For this system, it is more appropriate to claim that the response reaches
the steady state in nine time constants .
This example shows that if a transfer function has repeated poles or, more
generally, a cluster of poles in a small region close to the imaginary axis, then the
rule of five time constants is not applicable . The situation is actually much more
complicated . The zeros of a transfer function also affect the transient response . See
Example 2 .4 .5 and Figure 2 .16 . However, the zeros are not considered in defining
the time constant, so it is extremely difficult to state precisely how many time con-
stants it will take for a response to reach the steady state . The rule of five time
constants is useful in pointing out that infinity in engineering does not necessarily
mean mathematical infinity .
4.1 . Consider the open-loop voltage regulator in Figure P3 .4(a). Its block diagram
is repeated in Figure P4 .1 with numerical values .
a. If RL, = 100 fl and if r(t) is a unit-step function, what is the response vo (t)?
What is its steady-state response? How many seconds will v o(t) take to reach
and stay within 1 % of its steady state?
b. What is the required reference input if the desired output voltage is 20 V?
RL V
10s+3 50 + R L
Figure P4 .1
c. Are the power levels at the reference input and plant output necessarily the
same? If they are the same, is the system necessary?
d If we use the reference signal computed in (b), and decrease RL from 100 i
.
to 50 fl, what is the steady-state output voltage?
4.2. Consider the closed-loop voltage regulator shown in Figure P3 .4(b). Its block
diagram is repeated in Figure P4 .2 with numerical values .
a. If RL = 10012 and if r(t) is a unit-step function, what is the response vo(t)?
What is its steady-state response? How many seconds will v0 (t) take to reach
the steady state?
b. What is the required r(t) if the desired output voltage is 20 V?
c. If we use the r(t) in (b) and decrease RL from 100 ft to 50 c, what is the
steady-state output voltage?
~,
r 2 RL vo
10 -0
10s+3 50 + R L
Figure P4 .2
4.3 . Compare the two systems in Problems 4 .1 and 4 .2 in terms of (a) the time
constants or speeds of response, (b) the magnitudes of the reference signals,
and (c) the deviations of the output voltages from 20 V as RL decreases from
100 c to 50 c. Which system is better?
4.4 . The transfer function of a motor and load can be obtained by measurement.
Let the transfer function from the applied voltage to the angular displacement
be of the form kmls(r,,,s + 1). If we apply an input of 100 V, the speed (not
displacement) is measured as 2 rad/s at 1 .5 seconds . The speed eventually
reaches 3 rad/s . What is the transfer function of the system?
4.5 . Maintaining a liquid level at a fixed height is important in many process-control
systems. Such a system and its block diagram are shown in Figure P4 .5, with
q0 qj
.ir~ v
kI
TS + 1
(a) (b) Figure P4 .5

PROBLEMS 149
T = 10, k, = 10, k2 = 0.95, and k3 = 2. The variables h and q; in the block

diagram are the deviations from the nominal values ho and q0; hence, the
desired or reference signal for this system is zero . This kind of system is called
a regulating system . If h(O) = - 1, what is the response of h(t) for t > 0?
4.6. a. Consider a motor and load with transfer function G(s) = Y(s)/U(s) _
1/s(s + 2), where the unit of the input is volts and that of the output is
radians. Compute the output y(t) due to u(t) = 1, for t ? 0. What is y(t)
as t ---> -?
b. Show that the response of the system due to
a for 0 - t< b
u(t) =
0 fort>b
a pulse of magnitude a and duration b, is given by
ab
y(t) = 2 + a2 e-Zr(1 - e 2b)
for t > b . What is the steady-state response?

c. If a = 1, what is the duration b necessary to move y 30 degrees?
d. If b = 1, what is the amplitude a necessary to move y 30 degrees?
4.7. Consider the position control system shown in Figure 4.8(a) . Let the transfer
function of the motor and load be 1 /s(s + 2) . The error detector is a pair of
potentiometers with sensitivity k1 = 3 . The reference input is to be applied by
turning a knob.
a. If k2 = 1, compute the response due to a unit-step reference input. Plot the
response . Roughly how many seconds will y take to reach and stay within
1 % of its final position?
b . If it is required to turn y 30 degrees, how many degrees should you turn the
control knob?
c. Find a k2 so that the damping ratio equals 0 .7. Can you find a k2 so that the
damping ratio equals 0 .7 and the damping factor equals 3?
4.8. Consider the position control system shown in Figure 4.8(b) . Let the transfer
function of the motor and load be 1/s(s + 2) . The error detector is a pair of
potentiometers with sensitivity k1 = 3 . The reference input is to be applied by
turning a knob. A tachometer with sensitivity k3 is introduced as shown.
a. If k2 = 1 and k3 = 1, compute the response due to a unit-step reference
input. Plot the response . Roughly how many seconds will y take to reach
and stay within 1% of its final position?
b. If it is required to turn y 30 degrees, how many degrees should you turn the
control knob?
c. If k3 = 1, find a k2 so that the damping ratio equals 0.7. If k3 is adjustable,
can you find a k2 and a k3 so that ~ = 0 .7 and ~u) = 3?
d. Compare the system with the one in Problem 4 .7 in terms of the speed of
response .
4.9. Consider a dc motor . It is assumed that its transfer function from the input to
the angular position is 1/s(s + 2). Is the motor stable? If the angular velocity
of the motor shaft, rather than the displacement, is considered as the output,
what is its transfer function? With respect to this input and output, is the system
stable?
4.10 . A system may consist of a number of subsystems . The stability of a system
depends only on the transfer function of the overall system . Study the stability
of the three unity-feedback systems shown in Figure P4 .10 . Is it true that a
system is stable if and only if its subsystems are all stable? Is it true that
negative feedback will always stabilize a system?
Figure P4 .10
4.11 . Which of the following are Hurwitz polynomials?

a. -2s2 - 3s - 5
b.2s2 +3s-5
c.s5 +3s4 +s2 +2s+ 10
d.s4 +3s3 +6s2 +5s+3
e.s 6 +3s5 +7s4 +8s 3 +9s2 +5s+3
4.12 . Check the stability of the following systems :
S3 - I
a.G(s)=
s5 +s 4 +2s3 +2s2 +5s+5
S3 - I
b. G(s) = s
4 + 14s3 + 71s2 + 154s + 120
c. The system shown in Figure P4 .12.
Figure P4.12

PROBLEMS 151
4.13 . Find the ranges of k in which the systems in Figure P4 .13 are stable.
r + s+k Y
(a)
r + 300 Y
+ 184s 3 +760.56 2 + 162s
ks
(b)
Motor
r + 3s+1 1
s+2
Compensating
network
H
Tachometer
(c) Figure P4.13
4.14. Consider a system with transfer function G(s) . Show that if we apply a unit-
step input, the output approaches a constant if and only if G(s) is stable . This
fact can be used to check the stability of a system by measurement .
4.15 . In a modern rapid transit system, a train can be controlled manually or auto-
matically . The block diagram in Figure P4 .15 shows a possible way to control
Amplifier Braking system Train

r + + Train position
1
k 500 >- Y
S2
bs
Tachometer
Figure P4.15

the train automatically . If the tachometer is not used in the feedback (that is,
if b = 0), is it possible for the system to be stable for some k? If b = 0.2,
find the range of k so that the system is stable .
4.16. Let D(s) = a n s n + a- i s' 1 + + a i s + a o , with a n > 0 . Let [j/2] be
the integer part of j/2 . In other words, if j is an even integer, then [j/2] _
j/2 . Ifj is an odd integer, then [j/2] = (j - 1)/2 . Define
bn,j = a,+2-2i i = 1, 2, . . . , [n/2] + 1
ij = an+1-2r
bn - i = 1, 2, . . ., [(n - 1)/2] + 1
For j = n - 2, n - 3, . . ., 2, 1, compute
bj +2,1
kj + 1 -
bj+ 1,t
bj, i = bb+2,i+1 -
kj+1b1+i,i+i i = 1, 2, . . . , [J/2] + I
Verify that the preceding algorithm computes all entries of the Routh table .
4.17 . What are the time constants of the following transfer functions?
s - 2
a s2 + 2s + 1
s - 1
b
. (s + 1)(s 2 + 2s + 2)
s2 +2s-2
C.
(s 2 + 2s + 4)(s 2 + 2s + 10)
s + 10
d.
s + 1
s - 10
e s 2 +2s+2
Do they all have the same time constant?
4.18 . What are the steady-state responses of the system with transfer function
1/(s 2 + 2s + 1) due to the following inputs :
a. ul ( t) = a unit-step function .
b. U2(t) = a ramp function .
C . u3(t) = ui ( t) + u2(t) .
d. u4 (t) = 2 sin 27rt, for t ? 0.
4.19 . Consider the system with input r(t), output y(t), and transfer function
s + 8
G(s) =
(s + 2)(s + 4)

PROBLEMS 153
Find the steady-state response due to r(t) = 2 + 3t. Compute

lim e(t) = lim (r(t) - y(t))
t->oo t__
This is called the steady-state error between r(t) and y(t).

4.20. Derive (4.25) by using the final-value theorem (see Appendix A) . Can you use
the theorem to derive (4.26)?
4.21 . Consider a system with transfer function G(s) . It is assumed that G(s) has no
poles in the closed right half s-plane except a simple pole at the origin . Show
that if the input u(t) = a sin wot is applied, the steady-state response excluding
the dc part is given by Equation (4.32).
4.22 . What is the time constant of the system in Problem 4 .19? Is it true that the
response reaches the steady state in roughly five time constants?
4.23 . Consider a system with transfer function
s + 1
G(s) =
(s + 2)(s + 4)
Compute its unit-step response . Is it true that the response reaches the steady
state in roughly five time constants?
4.24. What are the bandwidths of the transfer function 1/(s + 3) and the transfer
functions in Problems 4 .19 and 4 .23?
Computer
Simulation
and Realization
6
5.1 INTRODUCTION
In recent years computers have become indispensable in the analysis and design of
control systems . They can be used to collect data, to carry out complicated com-
putations, and to simulate mathematical equations . They can also be used as control
components, and as such are used in space vehicles, industrial processes, autopilots
in airplanes, numerical controls in machine tools, and so on . Thus, a study of the
use of computers is important in control engineering .
Computers can be divided into two classes : analog and digital . An intercon-
nection of a digital and an analog computer is called a hybrid computer. Signals on
analog computers are defined at every instant of time, whereas signals on digital
computers are defined only at discrete instants of time . Thus, a digital computer can
accept only sequences of numbers, and its outputs again consist only,Qf sequences
of numbers. Because digital computers yield more accurate results and are more
flexible and versatile than analog computers, the use of general-purpose analog
computers has been very limited in recent years . Therefore, general-purpose analog
computers are not discussed in this text ; instead, we discuss simulations using op-
erational amplifier (op-amp) circuits, which are essentially special-purpose or cus-
tom-built analog computers .
We first discuss digital computer computation of state-variable equations . We
use the Euler forward algorithm and show its simplicities in programming and com-
putation . We then introduce some commercially available programs . Op-amp circuit
implementations of state-variable equations are then discussed . We discuss the rea-
154

5 .2 COMPUTER COMPUTATION OF STATE-VARIABLE EQUATIONS 155
sons for not computing transfer functions directly on digital computers and then
introduce the realization problem . After discussing,the problem, we show how trans-
fer functions can be simulated on digital computers or built using op-amp circuits
through state-variable equations .
5 .2COMPUTER COMPUTATION OF STATE-VARIABLE EQUATIONS

i(t) = Ax(t) + bu(t) (5.1)
y(t) = cx(t) + du(t) (5.2)
where u(t) is the input ; y(t) the output, and x(t) the state . If x(t) has n components
or n state variables, then A is an n X n matrix, b is an n X 1 vector, c is a I X n
vector, and d is a 1 X 1 scalar. Equation (5 .2) is an algebraic equation . Once x(t)
and u(t) are available, y(t) can easily be obtained by multiplication and addition .
Therefore we discuss only the computation of (5.1) by using digital computers .
Equation (5.1) is a continuous-time equation ; it is defined at every instant of
time . Because every time interval has infinitely many points and because no digital
computer can compute them all, we must discretize the equation before computation .
By definition, we have
x(to) _ x(to + a) - x(to)
x(to) lim
dt a-+o a
The substitution of this into (5.1) at t = to yields
x(to + a) - x(to) = [Ax(to) + bu(to)]a
or
x(to + a) = x(to ) + aAx(to) + abu(to)
= Ix(to) + aAx(to) + abu(to)
where I is a unit matrix with the same order as A. Note that x + aAx =
(1 + aA)x is not well defined (why?). After introducing the unit matrix, the equation
becomes
x(to + a) = (I + aA)x(to) + bu(to)a (5.3)
This is a discrete-time equation, and a is called the integration step size . Now, if
x(to) and u(to ) are known, then x(to + a) can be computed algebraically from (5 .3).
Using this equation repeatedly or recursively, the solution of (5.1) due to any x(O)
and any u(t), t ? 0, can be computed . For example, from the given x(0) and u(0),
we can compute
x(a) = (I + aA)x(0) + bu(0)a
We then use this x(a) and u(a) to compute
x(2a) = (I + aA)x(a) + bu(a)a

156 CHAPTER 5 COMPUTER SIMULATION AND REALIZATION
Proceeding forward, x(ka), k = 0, 1, 2, . . . , can be obtained . This procedure can

be expressed in a programmatic format as
DO 10 k = 0, N
10 x((k + 1)a) = (I + aA)x(ka) + bu(ka)a
where N is the number of points to be computed . Clearly this can easily be pro-
grammed on a personal computer . Equation (5.3) is called the Euler forward algo-
rithm . It is the simplest but the least accurate method of computing (5 .1).
In using (5.3), we will encounter the problem of choosing the integration step
size a. It is clear that the smaller a, the more accurate the result. However, the
smaller a, the larger the number of points to be computed for the same time interval .
For example, to compute x(t) from t = 0 to t = 10, we need to compute 10 points
if a = 1, and 1000 points if a = 0 .01 . Therefore, the choice of a must be a
compromise between accuracy and amount of computation . In actual programming,
a may be chosen as follows . We first choose an arbitrary a o and compute the re-
sponse. We then repeat the computation by using a, = ao/2 . If the result of using
a, is close to the result of using a o, we stop, and the result obtained by using a, is
probably very close to the actual solution . If the result of using a, is quite different
from the result of using a o, ao is not small enough and cannot be used . Whether or
not a, is small enough cannot be answered at this point . Next we choose a 2 = a,/2
and repeat the computation . If the result of using a 2 is close to the result of using
a,, we may conclude that a, is small enough and stop the computation . Otherwise
we continue the process until the results of two consecutive computations are suf-
ficiently close.
Example 5 .2.1
Compute the output y(t), from t = 0 to t = 10 seconds, of
x(t)
y(t) = [1
[ _ 0.5 - I .5]
- 1]x(t)
X(t) +
I
I0
u(t)
(5.4)
due to the initial condition x(0) = [2 -1]' and the input u(t) = 1, for t ? 0,
where the prime denotes the transpose .
For this equation, (5 .3) becomes
+ a a
Cx2(to + a)] - ([0 01] 1 -0.5 -1 .5 1 / 1x2(t0 )] + 1I1 1
a [xi(to)1 + I i 1 a
[- 0 .5a 1 - 1 .5a I x2(to) J 1

5 .2 COMPUTER COMPUTATION OF STATE-VARIABLE EQUATIONS 157
which implies
x, ((k + 1)a) = x, (ka) + ax2 (ka) (5 .5a)
x2 ((k -- 1)a) _ -0 .5ax,(ka) + (1 - 1 .Sa)x2(ka) + a (5 .5b)

The output equation is
y(ka) = x,(ka) - x 2(ka) (5 .5c)
Arbitrarily, we choose a o = 1 . We compute (5 .5) from k = 0 to k = 10. A

FORTRAN program for this computation is as follows :
REAL X1(0 :1500), X2(0 :1500), Y(0 :1500), A
INTEGER K
A=1 .0
X1(0) = 2 .0
X2(0) = -1 .0
DO 10 K=O, 10
X1 (K + 1) = X1 (K) + A*X2(K)
X2(K+1)=-0 .5*A*X1 (K) + (1 .0 - 1 .5*A)*X2(K) + A
Y(K) = X1(K) - X2(K)
PRINT*, 'T=',K, 'Y=', Y(K)
10 CONTINUE
END
where A stands for a. The result is printed in Table 5 .1 and plotted in Figure 5 .1
using + . We then repeat the computation by using a, = a 0/2 = 0 .5 and compute
21 points . The result is plotted in Figure 5 .1 using 0, but we print only 11 points in
Table 5 .1. The two results are quite different, therefore we repeat the process for
Table 5 .1 Computation of (5.4) Using Four Different a
a= 1.0 a=0.5 a = 0.25 a=0.125 Exact

T = 0 .0 Y = 3.0000 Y = 3.0000 Y = 3 .0000 Y = 3.0000 Y = 3.0000
T = 1.0 Y = 0 .5000 Y = 1 .3125 Y = 1 .6259 Y = 1.6468 Y = 1 .6519
T = 2.0 Y = 1 .2500 Y = 1 .3008 Y = 1 .4243 Y = 1.4350 Y = 1 .4377
T = 3.0 Y = 1 .6250 Y = 1 .5286 Y = 1 .5275 Y = 1 .5292 Y = 1.5297
T = 4.0 Y = 1 .8125 Y = 1 .7153 Y = 1.6702 Y = 1 .6677 Y = 1 .6673
T = 5.0 Y = 1 .9063 Y = 1 .8350 Y = 1.7851 Y = 1 .7814 Y = 1.7807
T = 6.0 Y = 1 .9531 Y = 1.9059 Y = 1.8647 Y = 1 .8612 Y = 1.8605
T = 7.0 Y = 1 .9766 Y = 1.9468 Y = 1.9164 Y = 1 .9135 Y = 1 .9130
T = 8.0 Y = 1 .9883 Y = 1 .9700 Y = 1.9488 Y = 1 .9467 Y = 1 .9463
T = 9.0 Y = 1 .9941 Y = 1.9831 Y = 1.9689 Y = 1 .9673 Y = 1 .9672
T = 10.0 Y = 1 .9971 Y = 1.9905 Y = 1.9811 Y = 1 .9800 Y = 1 .9799
y (t)
Figure 5 .1 Results of computer simulation.
a = 0.25 and then for a = 0.125 . The last two results are very close, therefore we
stop the computation .
The exact solution of (5 .4) can be computed, using the procedure in Section
2.7, as
y(t) = 2 + 4e -' - 3e -o.5 t
The computed result is very close to the exact solution.
Exercise 5 .2 .1
Discretize and then compute the solution of x(t) = - 0.5x(1) + u(t) due to x(0) _
1 and u(t) = 1, for t ? 0 . Compare your result with the exact solution .
The purpose of introducing the preceding algorithm is to show that state-variable

equations can easily be programmed on a digital computer . The introduced method
is the simplest but yields the least accurate result for the chosen integration step size .
There are many other methods . For example, we may discretize a continuous-time
5.3 EXISTING COMPUTER PROGRAMS 159
equation into a discrete-time state-variable equation as shown in (2 .89) and then

carry out the computation . This is discussed in the next section .
5.3 EXISTING COMPUTER PROGRAMS
Before the age of personal computers, control systems were mostly simulated on
mainframe computers . Many computer programs are available at computing centers .
Two of the most widely available are
IBM Scientific Routine Package
LSMA (Library of Statistics and Mathematics Association)
In these packages, there are subroutines for solving differential equations, linear
algebraic equations, and roots of polynomials . These subroutines have been thor-
oughly tested and can be employed with confidence . Most computing centers also
have UNPACK and Eispack, which were developed under the sponsorship of the
National Science Foundation and are considered to be the best for solving linear
algebraic problems . These programs are often used as a basis for developing other
computer programs .
Many specialized digital computer programs have been available for simulating
continuous-time systems in mainframe computers : CSMP (Continuous System Mod-
eling Program), MIDAS (Modified Integration Digital Analog Simulator), MIMIC
(an improved version of MIDAS), and others . These programs were major simulation
tools for control systems only a few years ago . Now they are probably completely
replaced by the programs to be introduced in the following .
Personal computers are now widely available . A large amount of computer
software has been developed in universities and industries . About 90 programs
are listed in Reference [30] . We list in the following some commercially available
packages:
CTRL-C (Systems Control Technology)
EASY5 (Boeing Computer Services Company)
MATLAB (Math Works Inc.)
MATRIX,, (Integrated Systems Inc .)
Program CC (Systems Technology)
Simnon (SSPA Systems)
For illustration, we discuss only the use of MATLAB .* The author is familiar with
version 3 .1 and the Student Edition of MATLAB, which is a simplified edition of
version 3 .5. Where no version is mentioned, the discussion is applicable to either
*The author has experience only with PC-MATLAB and has no knowledge of the relative strength or
weakness of the programs listed . MATLAB is now available in many universities . The Student Edition
of MATLAB" is now available from Prentice Hall .

version. In MATLAB, a matrix is represented row by row separated by semicolons ;

entries of each row are separated by spaces or commas . For example, the following
matrices
-1 .5 - 2 -0 .5 1
A= 1 0 0 b = 0 c=[1 .5 0 0 .5] d = 0 (5 .6)
0 1 0 0
are represented, respectively, as
a=[-1 .5 -2 -0 .5 ;1 0 0 ;0 1 0] ; b=[1 ;0 ;0] ; c=[1 .5 0 0 .5] ; d=[0] ;
After each statement, if we type the ENTER key, the statement will be executed,
stored, and displayed on the monitor. If we type a semicolon, the statement will be
executed and stored but not displayed . Because b has three rows, entries of b are
separated by semicolons . The row vector c can also be represented as C = [1 .5,0,0 .5]
and the scalar d as d = 0 without brackets .
The state-variable equation
x = Ax + bu (5.7a)
y = cx + du (5.7b)
is represented as (a, b, c, d, iu), where iu denotes the ith input . In our case, we have
only one input, thus we have iu = 1 .
Suppose we wish to compute the unit-step response of (5.7). In using version
3.1 of MATLAB, we must specify the initial time t o, the final time tf , and the time
interval a at which the output will be printed or plotted . For example, if we choose
to = 0, tf = 20, and a = 1, then we type
t = 0 :1 :20 ;
The three numbers are separated by colons . The first number denotes the initial time,
the last number denotes the final time, and the middle number denotes the time
interval at which the output will appear . Now the following commands
y=step(a,b,c,d,1,t) ;
plot(t,y)
will produce the solid line shown in Figure 5 .2. It plots the output at t = 0, 1, 2,
. . . . 10; they are connected by straight lines . The following commands
t = 0 :0 .05 :20 ;
plot(t,step(a,b,c,d,1,t))
will generate the dotted line shown in Figure 5 .2, where the output is plotted every
0.05 second . In using version 3 .5 or the Student Edition of MATLAB, if we type
step(a,b,c,d,1)
then the response will appear on the screen . There is no need to specify t o, tf, and
a. They are chosen automatically by the computer .
5 .3 EXISTING COMPUTER PROGRAMS 161
Figure 5 .2 Step responses .
We discuss how MATLAB computes the step response of (5.7). It first trans-
forms the continuous-time state-variable equation into a discrete-time equation as in
(2.89). This can be achieved by using the command c2d which stands for continuous
to discrete . Note that the discretization involves only A and b; c and d remain
unchanged . If the sampling period is chosen as 1, then the command
[da,db] = c2d(a,b,1)
will yield
-0.1375 -0.8502 -0.1783 0.3566
da = 0.3566 0.3974 -0.1371 db = 0 .2742
0.2742 0.7678 0.9457 0.1086
Thus, the discretized equation of (5.6) is
-0.1375 - 0 .8502 - 0 .1783 0.3566
x(k + 1) = 0.3566 0.3974 -0.1371 x(k) + 0.2742 u(k) (5.8a)
0.2742 0.7678 0.9457 0.1086
y(k) = [1 .5 0 0.5]x(k) (5.8b)
This equation can be easily computed recursively .
If the input u(t) is stepwise, as shown in Figure 2 .19(a), the discretized equation
in (5.8) will give the exact response of (5.7) at sampling time . Clearly, a step function
is stepwise, so the two responses in Figure 5 .2 are the same at the sampling instants
even though one sample period is 1 second and the other is 0 .05 second . We mention

that several methods are listed in MATLAB for computing e". One of them is to
use the infinite series in (2.67). In using the series, there is no need to compute the
eigenvalues of A .
5 .4 BASIC BLOCK DIAGRAMS AND OP-AMP CIRCUITS
A basic block diagram is any diagram that is obtained by interconnecting the three
types of elements shown in Figure 5 .3 . The three types of elements are multipliers,
adders, and integrators . The gain k of a multiplier can be positive or negative, larger
or smaller than 1 . An adder or a summer must have two or more inputs and one and
only one output . The output is simply the sum of all inputs . If the input of an
integrator is x(t), then its output equals fo x(T)dT. This choice of variable is not as
convenient as assigning the output of the integrator as x(t) . Then the input of the
integrator is x(t) : = dx/dt as shown in Figure 5 .3. These three elements can be easily
built using operational amplifier (op-amp) circuits . For example, a multiplier with
gain k can be built as shown in Figure 5 .4(a) or (b) depending on whether k is
positive or negative . The adder can be built as shown in Figure 5.4(c). Figure 5 .4(d)
shows an implementation of the integrator with R = 1 kfl = 1000 fl, C = 10 -3 F
orR = 1 Mfg. = 106 fl, C = 1 F = 10 -6 F. For simplicity, the grounded inverting
terminals are not plotted in Figure 5 .4(b through f).
Now we show that every state-variable equation can be represented by a basic
block diagram . The procedure is simple and straightforward . If an equation has n
state variables or, equivalently, has dimension n, we need n integrators . The output
of each integrator is assigned as a state variable, say, x,; then its input is x, . If it is
assigned as -x,, then its input is -x,. Finally, we use multipliers and adders to
build up the state-variable equation . This is illustrated by an example . Consider
C 2(t)
1(t) I
= 1
C2
-0.3 xl(t)1
-8 I C X2(t) J +
[-2]x
0
u(t) (5.9a)
y(t) _ [ - 2 3] Cx2(t) + 5u(t) (5.9b)

I
It has dimension 2 and needs two integrators . The outputs of the two integrators are
assigned as x1 and x2 as shown in Figure 5 .5. Their inputs are xl and i 2 . The first
equation of (5.9a) is xl = 2x 1 - 0.3x2 - 2u . It is generated in Figure 5 .5 using
the solid line . The second equation of (5 .9a) is x2 = x1 - 8x2 and is generated using
the dashed line . The output equation in (5 .9b) is generated using the dashed-and-
x
s
Figure 5 .3 Three basic elements .

5 .4 BASIC BLOCK DIAGRAMS AND OP-AMP CIRCUITS 1 63
MFh6'
k>0
(a) (b)
(c) (d)
e
Rib
R/c R/C
(e) (f)
Figure 5 .4 Implementation of basic elements .
dotted line . It is indeed simple and straightforward to develop a basic block diagram
for any state-variable equation . Conversely, given a basic block diagram, after as-
signing the output of each integrator as a state variable, a state-variable equation can
be easily obtained . See Problem 5 .5.
Every basic element can be built using an op-amp circuit as shown in Figure
5 .4; therefore every state-variable equation can be so built through its basic block
Figure 5 .5 Basic block diagram of (5 .9).

diagram . For example, Figure 5 .6(a) shows an op-amp circuit implementation

of (5 .9) .
The implementation in Figure 5 .6(a), however, is not desirable, because it uses
unnecessarily large numbers of components . In practice, op-amp circuits are built to
perform several functions simultaneously . For example, the circuit in Figure 5 .4(e)
can act as an adder and multipliers . If its inputs are x1, x,, and x3, then its output is
-u
R/2
R/2
R
-VW-
x
R/8
(a)
(b)
Figure 5 .6 Implementations of (5 .9) .

5 .5 REALIZATION PROBLEM 165
- (ax, + bx 2 + cx3). The circuit in Figure 5 .4(f) can act as an integrator, an adder,
and multipliers . If we assign its output as x, then we have -x = av, + bv 2 + cv3 .
If we assign the output as -x, then we have X = av1 + bv2 + cv3. It is important
to mention that we can assign the output either as x or -x, but we cannot alter its
relationship with the inputs.
Figure 5 .6(b) shows an op-amp circuit implementation of (5.9) by using the
elements in Figure 5 .4(e) and (f). It has two integrators. The output of one integrator
is assigned as x,; therefore, its input equals -x, = - 2x, + 0.3x2 + 2u as shown .
The output of the second integrator is assigned as -x2; therefore its input should
equal x2 = x1 - 8x2 as shown. The rest is self-explanatory . Although the numbers
of capacitors used in Figure 5 .6(a) and (b) are the same, the numbers of operational
amplifiers and resistors in Figure 5 .6(b) are considerably smaller.
In actual operational amplifier circuits, the range of signals is limited by the
supplied voltages, for example 15 volts . Therefore, a state-variable equation may
have to be scaled before implementation. Otherwise, saturation may occur . This and
other technical details are outside the scope of this text, and the interested reader is
referred to References [24, 50].
5 .5 REALIZATION PROBLEM
Consider a transfer function G(s) . To compute the response of G(s) due to an input,
we may find the Laplace transform of the input and then expand G(s)U(s) into partial
fraction expansion . From the expansion, we can obtain the response . This procedure
requires the computation of all poles of G(s)U(s), or all roots of a polynomial. This
can be easily done by using MATLAB . A polynomial in MATLAB is represented
by a row vector with coefficients ordered in descending powers. For example, the
polynomial (s + 1) 3(s + 1 .001) = s4 + 4.001s3 + 6.003s2 + 4.003s + 1 .001
is represented as
p = [1,4 .001,6 .003,4 .003,1 .001 ]
or
p = [1 4 .001 6 .003 4 .003 1 .001]
The entries are separated by commas or spaces . The command

roots(p)
will generate
- 1 .001
- 1 .0001
- 1 + 0 .0001 i
- 1 - 0.0001 i
We see that the results differ slightly from the exact roots - 1, - 1, - 1, and - 1 .001 .
If we perturb the polynomial to s 4 + 4.002s 3 + 0.6002s2 + 4.002s + 1, then the

command
roots([1 4 .002 6 .002 4 .002 1])
will generate
-1 .2379
- 9 .9781 + 0 .2081
- 9 .9781 - 0 .2081
-0 .8078
The roots are quite different from those of the original polynomial, even though the
two polynomials differ by less than 0 .03% . Thus, roots of polynomials are very
sensitive to their coefficients . One may argue that the roots of the two polynomials
are repeated or clustered together ; therefore, the roots are sensitive to coefficients .
In fact, even if the roots are well spread, as in the polynomial
p(s) = (s + 1)(s + 2)(s + 3) . . . (s + 19)(s + 20)
the roots are still very sensitive to the coefficients . See Reference [15, p . 219] .
Furthermore, to develop a computer program to carry out partial fraction expansion
is not simple . On the other hand, the response of state-variable equations is easy to
program, as is shown in Section 5 .2. Its computation does not require the compu-
tation of roots or eigenvalues, therefore it is less sensitive to parameter variations .
For these reasons, it is desirable to compute the response of G(s) through state-
variable equations .
Consider a transfer function G(s) . If we can find a state-variable equation
x(t) = Ax(t) + bu(t) (5 .1Oa)
y(t) = cx(t) + du(t) (5 .10b)
such that the transfer function from u to y in (5 .10) equals G(s) or, from (2.75),
G(s) = c(sI - A) - 'b + d
then G(s) is said to be realizable and (5 .10) is called a realization of G(s) . The term
realization is well justified, for G(s) can then be built or implemented using op-amp
circuits through the state-variable equation . It turns out that G(s) is realizable if and
only if G(s) is a proper rational function . If G(s) is an improper rational function,
its realization will assume the form
x(t) = Ax(t) + bu(t) (5 .11 a)
y(t) = cx(t) + du(t) + d 16(t) + dzii(t) + ( 5 .11 b)

where u(t) = du(t)/dt and ii(t) = d 2U(t)/dt z. In this case, differentiators are needed
to generate u(t) and fi(t) . Note that differentiators are not included in Figure 5 .3 and
are difficult to build in practice . See Reference [18, p. 456] . Therefore, the state-
variable equation in (5 .11) is not used in practice . Thus, we study only the realization
of proper rational transfer functions in the remainder of this chapter .

5 .5 .1 Realizations of N(s)/D(s)
Instead of discussing the general -case, we use a transfer function of degree 4 to
illustrate the realization procedure . Consider
3 + b2s 2 + b,s + b o - N(s)
G(s) = Y(s) = b4s4 + b,SS3 (5.12)
U(s) a4s 4 + a3 + a2S2 + a ts + a o D(s)
where a ; and b, are real constants and a4 = 0. If b4 =A 0, the transfer function is

biproper; if b4 = 0, it is strictly proper. Before realization, we must carry out two
preliminary steps : (i) decompose G(s) into the sum of a constant and a strictly proper
rational function, and (ii) carry out a normalization . This is illustrated by an example .
Example 5 .5.1
Consider
S 4 + 2s3 - s2 + 4s + 12
G(s) = 2S (5.13)
4 + 105 3 + 20S 2 + 205 + 8
Clearly we have G(cc) = 1 /2 = 0 .5 ; it is the ratio of the coefficients associated

with s 4. We compute
(S 4 +2s 3 -S 2 +4s+ 12)-
0.5(254 + 105 3 + 205 2 + 20s + 8)
Gs(s) : = G(s) - G(-)
2S4 + 105 3 + 2052 + 20s + 8 (5.14)
- 3S 3 - 11s 2 - 6s + 8
2S4 + 1053 + 205 2 + 20s + 8
It is strictly proper . Next we divide its numerator and denominator by 2 to yield
-1 .5s3 - 5.5s2 - 3s + 4 N(s)
GS(s) = s
4 + 55 3 + 105 2 + 10s + 4 . D(S)
This step normalizes the leading coefficient of D(s) to 1 . Thus (5 .13) can be written
as
s4 + 2s3 - s2 + 4s + 12
G(s) = 2S
4 + iOs 3 + 2052 + 20s + 8 (5 .15)
-1 .5s 3 - 5 .5s2 - 3s + 4
= 0.5 +
s4 + 5 s3 + 10s2 + 10 s+4
This completes the preliminary steps of realization . We mention that (5 .14) can also
be obtained by direct division as
0.5
2s4 + 10s 3 + 205 2 + 20s + 8)s 4 + 2s 3 - s 2 + 4s + 12
S 4 + 5S 3 + 105 2 +10s + 4
- 3S3 - lls 2 - 6s + 8

Using the preceding procedures, we can write G(s) in (5 .12) as

Y(s) b S3 + b2 S2 + b1S + bo N(s)
G(s) _ = G(oc) + a 3 3 2 - =: d + (5 .16)
U(s) s + a3 s + a2 s + ai s + ao D(s)
Now we claim that the following state-variable equation
-a2 -a 1 -aa 1
1 0 0 0 L I
x(t) = x(t) + u(t) (5.17a)
0 1 0 0
0 0 1 0
y(t) = [b3 b2 bt bo]x(t) + du(t) (5 .17b)
with d = G(oo), is a realization of (5.12) or (5.16). The dimension of this equation
equals 4, the degree of the denominator of G(s) . The value of G(oo) yields the direct
transmission part d. The first row of A consists of the coefficients, except the leading
coefficients, of D(s) with sign changed. The pattern of the remainder of A is fixed ;
if we delete the first row and the last column, then it reduces to a unit matrix . The
pattern of b is again fixed. It is the same for all G(s); the_first entry is 1 and the rest
zero . The row vector c is formed from the coefficients of N(s) . Thus the state-variable
equation in (5 .17) can be easily formed from the coefficients in (5.16). Given a G(s),
there are generally many (in fact, infinitely many) realizations . For convenience, we
call (5 .17) the controllable form realization for reasons to be discussed in Chapter
12. See also Problem 5 .18 .
Now we use Mason's formula to show that the transfer function of (5 .17) from
u to y equals (5.16). Figure 5 .7 shows a basic block diagram of (5 .17). It has 4
integrators with state variables chosen as shown . The first equation of (5.17a) yields
the lower part of Figure 5 .7. The second equation of (5 .17a) is x2 = x 1 ; thus, we
simply connect the output of the first integrator, counting from the left, to the input
of the second integrator . The third equation of (5 .17a) is i3 = x2; thus, we connect
0
0
Figure 5 .7 Basic block diagram of (5 .17) .

the output of the second integrator to the input of the third integrator . A similar
remark applies to the fourth equation of (5 .17a). From (5 .17b), we can readily draw
the upper part of Figure 5 .7. The basic block diagram has four loops with loop gains
-a 3 /s,-a 2 /s 2, - a,/s 3 and -a
0 /s 4. They touch each other. Thus, we have
a3+ -a2+ -a l - ao
O= 1 _ = 1 +a3 + a22 +a3 + a
S S S S ) S S S S
There are five forward paths from u to y with path gains :
b0 b~ b2 b3
P' P4
_ S4 P2 S3 P3 S2 S
and
P5 = d
Note that the first four paths touch all the loops ; whereas P 5 , the direct transmission
path, does not touch any loop . Thus we have
L 1 = A2 = 03 = A4 = 1
and
A5 = A
Therefore the transfer function from u to y is
5
tEP;O;
Pi +P2 +P3 +P4 +d0
G(s) = - Q
0
b2 + b3
b3
S + s s + bsa
= d +
1+a+az+a3+a
S S S S
b3s3 + b2s2 + b ' S + b0
= d +
s4 + a3s 3 + a2 s 2 + as + ao
This is the same as (5.16). Thus (5.17) is a realization of (5 .12) or (5.16). This can
also be shown by computing c(sI - A) - 'b + d in (5.17). This is more tedious and
is skipped .
Because G(s) is scalar-that is, G(s) = G'(s)-we have
G(s) = c(sI - A) - 'b + d = b'(sI - A') - 'c' + d
Thus, the following state-variable equation
- a3 1 0 0 b3
X(t) = X(t) + b2 u(t) (5.18a)

-a,a2 00 01 10 b1
- ao 0 0 0 b0

y(t) = [1 0 0 0]x(t) + du(t) (5 .18b)

is also a realization of (5.16). It is obtained from (5 .17) by taking the transpose of
A and interchanging b and c . This is called the observable form realization of (5 .16)
for reasons to be discussed in Chapter 12 . Although we use the same x, the x in
(5.17) and the x in (5 .18) denote entirely different variables .
Example 5 .5 .2
Consider the G(s) in (5.13). It can be written as, as discussed in (5.15),
s4 +2s3 -s 2 + 4s + 12
G(s) =
2s4 + 10s3 + 20s2 + 20s + 8 (5 .19)
-1.5s3 - 5.5s2 - 3s + 4
=0.5+
s4 +5s3 + 10s2 + 10s+4
The denominator of G(s) has degree 4, therefore the realization of G(s) has dimension
4. Its controllable-form realization is
-5 -10 -10 -4 1
_ 1 0 0 0 0
x x+ 0 U (5 .20a)
0 1 0 0
0 0 1 0 -0 -
y = [ -1 .5 - 5 .5 - 3 4] x + 0 .5u (5 .20b)
Its observable-form realization is
-5 1 0 0 -1 .5
-10 0 1 0 -5.5
7C = X + u
-10 0 0 1 -3
-4 0 0 0 4
y=[1 0 0 0]x+0 .5u
Example 5 .5.3
Consider
1 0.50 s 2 +0 s +0.5
G(s) _ = _
2s 3 s3 +0 s 2 +0 s +0
s3
The denominator of G(s) has degree 3, therefore the realization of G(s) has dimension
3. Because G(s) is strictly proper or G(c) = 0, the realization has no direct trans-

mission part . The controllable-form realization of G(s) is

0 0 0 1
x= 1 0 0 x+ 0 u
0 1 0 0
y = [0 0 0.5]x
Exercise 5 .5 .1
Find controllable- and observable-form realizations of

2s + 10
a.
s + 2
3s2 -s+2
b
S 3 + 2S 2 + 1
4s 3 +2s+ 1
C.
2s3 + 3s 2 + 2
Exercise 5 .5 .2
Find realizations of
5
a. 2s2 + 4s + 3
2
b.
S 3 +S - 1
3
C.
2s5 + 1
To conclude this section, we mention that realizations can also be generated

using MATLAB . The command tf2ss stands for transfer function to state space or
state-variable equation . The G(s) in (5 .19) can be represented by
num = [1 2 -1 4 12] ; den= [2 10 20 20 8] ;
where num stands for numerator and den, denominator . Then the command
[a,b,c,d] =tf2ss(num,den)

will generate the controllable-form realization in (5.20). To find the step response
of G(s), we type
step(num,den)
if we use version 3 .5 or the Student Edition of MATLAB . Then the response will
appear on the monitor . If we use version 3 .1 of MATLAB to find the step response
of G(s) from 0 to 20 seconds and with print-out interval 0 .1, we type
t=0 :0.1 :20 ;
y=step(num,den,t) ;
plot(t,y)
Then the response will appear on the monitor . Inside MATLAB, the transfer function
is first realized as a state-variable equation by calling tf2ss, and then discretized by
calling c2d . The response is then computed and plotted.
5.5 .2 Tandem and Parallel Realizations

The realizations in the preceding section are easy to obtain, but they are sensitive to
parameter variations in computer computation and op-amp circuit implementations .
In this subsection, we discuss two different types of realizations . They are to
be obtained from block diagrams . We use examples to illustrate the realization
procedures .
Example 5 .5 .4
20 _ 2 1 10
(5.21)
G(s)=(s+2)2(2s+3) s+2 s+2 2s+3
with grouping shown; the grouping is quite arbitrary . It is plotted in Figure 5.8(a).
Now we assign the output of each block as a state variable as shown . From the first
u 2 1 x2 10 x3 =Y
s+2 s+2 2s+3
(a)
U 4s+2 s +4s+8 1
s+3 s 2 +2s+2 s+I
(b)
Figure 5.8 Tandem connections .

block from the left, we have

2
X, (s) = s + 2 U(s) or (s + 2)X, (s) = 2U(s)
which becomes, in the time domain,

z, + 2x, = 2u or zl = -2x, + 2u
From the third block, we have
10
X3(s) 2(s) or (s + 1 .5)X3 (s) = 5X 2 (s)
_ 2s + 3 X
which becomes
.z 3
+ 1.5X3 = 5x2 or x3 = - 1 .5x3 + 5x2
Similarly, the second block implies x2 = - 2x2 + x, . These equations and y = x 3
can be arranged as
X, - 2 0 0][x,] 2
x2 = 1 - 2 0 x2 + 0 u (5.22a)
x3 0 5 - 1.5 X3 0
y = [0 0 1 ] x (5.22b)
This is called a tandem or cascade realization because it is obtained from the tandem
connection of blocks .
Example 5 .5.5
Consider
(4s + 2)(s2 + 4s + 8)
G(s) = (5 .23)
(s + 3)(s 2 + 2s + 2)(s + 1)
It has pairs of complex conjugate poles and zeros . If complex numbers are not
permitted in realization, then these poles and zeros cannot be further factored into
first-order factors . We group (5 .23) and then plot it in Figure 5.8(b). The first block
has the biproper transfer function (4s + 2)/(s + 3) . If we assign its output as a
state variable as x,, then we have
4s + 2
XI (s) = s + 3 U(s) or (s + 3)X,(s) = (4s + 2)U(s)
which implies
z, + 3x, = 4u + 2u or z, _ - 3x, + 2u + 4u

This equation contains the derivative of u, thus we cannot assign the output of the
first block as a state variable . Now we assign the outputs of the first and second
blocks as w, and w 2 as shown in Figure 5.8(b). Then we have
0
W1 (s) = 4s +
3
U(s)
= C
4 + s + / U(s)
and
s2 +4s+8 2s+6
W2(S) = s2 + 2s + 2 W, (s) = 1 + s
2 + 2s + 2
W, (S)
The first transfer function is realized in the observable form as

x, = - 3x, - lOu w, = x, + 4u (5 .24)
The second transfer function is realized in the controllable form as
W,
[x3] [ 1 0] [x3 ] + [0] (5 .25)
W2 = [2 6]
The substitution of w1 in (5.24) into (5 .25) yields

.z2 = - 2x2 - 2x3 + x1 + 4u X3 = x2 (5 .26a)
w2 - 2x2 + 6x3 + x1 + 4u (5 .26b)
If we assign the output of the third block in Figure 5 .8(b) as x4, then we have, using
(5.26b),
x4 = x4 +w2 = -x4 +2x2 +6x3 +x1 +4u (5 .27)
These equations and y = x4 can be arranged as

.z, -3 0 0 0 x1 -10
x2 1 -2 -2 0 x2 4
+ u (5 .28a)
x3 0 1 0 0 x3 0
J4 1 2 6- 1 x4 4
y = [0 0 0 1]x (5 .28b)
This is a tandem realization of the transfer function in (5 .23).
If a transfer function is broken into a product of transfer functions of degree 1

or 2, then we can obtain a tandem realization of the transfer function . The realization
is not unique . Different grouping and different ordering yield different tandem real-

izations. The outputs of blocks with transfer function b/(s + a) can be assigned as
state variables (see also Problem 5.9). The outputs of blocks with transfer function
(s + b)/(s + a) or of degree 2 cannot be so assigned .
In the following, we discuss a different type of realization, called parallel
realization .
Example 5 .5.6
Consider the transfer function in (5 .21). We use partial fraction expansion to expand
it as
20 - 40 - 20 40
G(s) _ = + + (5.29)
(s + 2)2(2s + 3) s + 2 (s + 2)2 s + 1.5
(a)
(b)
Figure 5.9 Parallel connections .
Figure 5 .9 shows two different plots of (5.29). If we assign the output of each block
as a state variable as shown, then from Figure 5 .9(a), we can readily obtain
Xl = -2x 1 + x2 x2 = - 2x2 + u
and
X3 = - 1 .5X3 + 40u y = - 20x1 - 40x2 + x3

They can be expressed in matrix form as

x, -2 1 0 xl 0
X2
X = 0 - 2 0 x2 + 1 u (5 .30a)
3 0 0 - 1 .5 x3 40
y = [-20 -40 1]x (5 .30b)
This is a parallel realization of G(s) in (5.29). The matrix A in (5 .30a) is said to be
in a Jordan form ; it is a generalization of diagonal matrices .
Exercise 5 .5.3
Show that the block diagram in Figure 5 .9(b) with state variables chosen as shown
can be described by
z, -2 0 0 x [-20
X2 = 1 - 2 0 x2 + - 40 a (5 .31 a)
X3 0 0 - 1 .5 x3]
' 1
y = [0 1 40] x (5.31 b)
Example 5 .5.7
Consider the transfer function in (5 .23). Using partial fraction expansion, we expand
it as
(4s + 2)(s2 + 4s + 8)
G(s) _
(s + 3)(s2 + 2s + 2)(s + 1) (5 .32)
-5 5 4s + 12
s+1 + s+3 + s2 +2s+2
It is plotted in Figure 5 .10. With the variables chosen as shown, we have
zl = -x, - 5u X2- 3x2 +Su
-5 x I
s+1
u 5 x 2 y
s+3
4s 12
s2 +2s+2
Figure 5.10 Parallel connections .

5 .6 MINIMAL REALIZATIONS 177
and
x3 ] 1u
x4 1 0] [x4 + C0
w = [4 12]
CJ
They can be combined to yield
xl -1 0 0 0 ^ xl -5
x2 0 -3 0 0 x2 5
u (5 .33a)
x3 0 0 -2 -2 x3 1
x4 0 0 1 0 x4 0
y = [1 1 4 12] x (5 .33b)
This is a parallel realization of the transfer function in (5.32).
Tandem and parallel realizations are less sensitive to parameter variations than
the controllable- and observable-form realizations are . For a comparison, see Ref-
erence [13].
5 .6 MINIMAL REALIZATIONS

s3 + 4s2 + 7s + 6
G(s) = s (5 .34)
4 + 5s3 + lOs2 + lls + 3
It is strictly proper and has 1 as the leading coefficient of its denominator . Therefore,
its realization can be read from its coefficients as
-5 -10 -11 -3 1
1 0 0 0 0
x = x + u (5 .35a)
0 1 0 0 0
0 0 1 0 0
y = [l 4 7 6]x (5 .35b)
This is the controllable form realization . Its observable form realization is

-5 1 0 0 1
-10 0 1 0 4
x 7 u (5 .36a)
-11 0 0 1 x+
-3 0 0 0 6

y = [1 0 0 0]x (5 .36b)
Both realizations have dimension 4 . If we use either realization to simulate or im-
plement G(s), then we need 4 integrators . It happens that the numerator and denom-
inator of G(s) have the common factors' + 2s + 3 . If we cancel this common
factor, then G(s) in (5 .34) can be reduced to
s + 2
G(s) = (5 .37)
s2 + 3s + 1
This can be realized as
x=
C 1
0] X +
I Iu
(5 .38a)
y = [1 2]x (5 .38b)
or
x
= C -1 0 ] x + C 2 ~ a (5 .39a)
y = [1 0]x (5 .39b)
These realizations have dimension 2 . They are called minimal realizations of G(s)
in (5 .34) or (5.37), because they have the smallest number of state variables among
all possible realizations of G(s) . The realizations in (5 .35) and (5 .36) are called
nonminimal realizations. We mention that minimal realizations are minimal equa-
tions discussed in Section 2.8. Nonminimal realizations are not minimal equations .
If D(s) and N(s) have no common factor, then the controllable-form and
observable-form realizations of G(s) = N(s)/D(s) are minimal realizations . On the
other hand, if we introduce a common factor into N(s) and D(s) such as
s + 2 (s + 2)P(s)
G(s) _
s2 + 3s + 1 (s2 + 3s + 1)P(s)
then depending on the polynomial P(s), we can find many nonminimal realizations .
For example, if the degree of P(s) is 5, then we can find a 7-dimensional realization .
Such a nonminimal realization uses unnecessarily large numbers of components and
is not desirable in practice.
Exercise 5.6 .1
Find two minimal realizations for

s2 - 1
G(s) =
s3 +s2 +s-3

5.6 MINIMAL REALIZATIONS 179
G 1 (s)
e2
G 2 (s)
Figure 5 .11 Two-input, one-output system .
Exercise 5 .6.2
Find three different nonminimal realizations of dimension 3 for the transfer function
G(s) = 1/(s + 1) .
5 .6 .1 Minimal Realization of Vector Transfer Functions'

Consider the connection of two transfer functions shown in Figure 5 .11 . It has two
inputs r and y and one output e, and is called a two-input, one-output system . Using
vector notation, the output and inputs can be expressed as
R(s)
E(s) = G 1 (s)R(s) + G2(s)Y(s) = [G,(s) G2(S)1 Y(s) (5.40)
where GI(s) and G2 (s) are proper rational functions . We assume that both GI (s) and
G2(s) are irreducible-that is, the numerator and denominator of each transfer func-
tion have no common factor . Certainly, we can find a minimal realization for G,(s)
and that for G 2(s). By connecting these two realizations, we can obtain a realization
for the two-input, one-output system . This realization, however, may not be a min-
imal realization.
We now discuss a method of finding a minimal realization for the two-input,
one-output system in Figure 5 .11 . First, the system is considered to have the follow-
ing 1 X 2 vector transfer function
G(s) = [G, (s) G2(S)1 (5.41)
We expand G(s) as
G(s) _ [d, d2] + [Gs,(s) Gs2(s)] = : [d, d2] + (5.42)

ID1(s)
D2(S)1
w here d, = G (cc) and G s,(s) are strictly proper, that is, deg N(s) < deg D (s) . Let
D(s) be the least common multiplier of D, (s) and D2(s) and have 1 as its leading
'The material in this section is used only in Chapter 10 and its study may be postponed .

coefficient. We then write (5 .42) as
G(s) = [d, d2] + [N1(s) N2(s)]

D(s)
Note that deg Ni (s) < deg D(s), for i = 1, 2. For convenience of discussion, we
assume
D(s) = s 4 + a3S 1 + a2S 2 + a 1 s + a0
(5 .43)
N1 (s)=b 31 s3 +b21 s2 +b ia s+b0 ,
and
N2 (s) - b 32s3 + b22S2 + b 12 s + b 02
Then a minimal realization of (5.42) is
1 0 0 b 31 b32
- a20 1 0 b 22 r(t)1
X(t) _ x(t) + b21 (5 .44a)
-a, 0 0 1 b T12 C y(t)
-a0 0 0 0 _b01 b02
r(t)
e(t) = [1 0 0 O]x(t) + [d 1 d2] l (5 .44b)
y(t) J
This is essentially a combination of the two observable-form realizations of G 1 (s)
and G2 (s).
Example 5 .6.1
Consider G 1 (s) = (s + 3)/(s + 1)(s + 2) and G 2(s) = (-s + 2)/2(s + 1)
connected as shown in Figure 5 .11 . Let the input of G 1 (s) be r and the output be e 1 .
We expand G 1 (s) as
E1 (s) - s+3 - s+3
GI(s) =
R(s) (s+ 1)(s+2) s2 +3s+2
From its coefficients, we can obtain the following minimal realization :
x i (t) I = C-3 -21[x1(t)1
+ 0 r(t) (t)
C 1 0 x2(t) ['1
e1(t) _ [1 3] x, (t)
x2(t) ]
To realize G 2 (s) with input y and output e2, we expand it as
E2(s) (-s + 2) 1 .5
G2(S)
z(s)
Y(s) 2(s + 1) - -0 .5 + s + 1

5 .6 MINIMAL REALIZATIONS 181
Figure 5 .12 Nonminimum realization .
Then the following one-dimensional state-variable equation

x3(t) = - X3(t) + y(t)
e2(t) = 1 .5x3 (t) - 0.5y(t)
is a minimal realization of G 2(s). The addition of el(t) and e2(t) yields e(t). The
basic block diagram of the connection is shown in Figure 5 .12 and the combination
of the two realizations yields
z 1 (t) -3 -2 0 xi(t) 1 0
r(t)
x2 (t) = 1 0 0 x2(t) + 0 0
x3(t) 0 0 -1 x3 (t) 0 1 y(t)
X, (t)
r(t)
e(t) = e l (t) + e2(t) = [1 3 1.5] x2(t) + [0 -0.5]
Y(01
X3(t)
This is a three-dimensional realization .
Now we develop a minimal realization . The transfer functions G 1 (s) and G 2(s)
are combined to form a vector as
s+3 -s+2
G(s) = (s + 1)(s + 2) 2(s + 1) (5 .45)
s + 3 3
_ [0 -0.5] + (s + 1)(s + 2) 2(s + 1)
The least common multiplier, with leading coefficient 1, of D 1 and D 2 is

D(s)=(s+ 1)(s+2)=s2 +3s+2

0 0
Figure 5 .13 Minimal realization.
We then write (5 .45) as

1
G(s) = [0 -0 .5]+s2 [s + 3 1 .5(s + 2)]
+ 3s + 2
Thus a minimal realization is
5] [
x(t)
=C _ 2 1]
x(t) +
[
I 3 r(t)]
Y(t)
(5 .46a)
e (t) = [1 0]x(t) + [0 r(t)

-0 .5] (5 .46b)
l y(t)
Its basic block diagram is shown in Figure 5 .13. It has two integrators, one less than
the block diagram in Figure 5 .12. Thus, the realization in Figure 5 .12 is not a minimal
realization . Note that the summer in Figure 5 .11 is easily identifiable with the right-
most summer in Figure 5 .12 . This is not possible in Figure 5 .13. The summer in
Figure 5.11 is imbedded and cannot be identified with any summer in Figure 5 .13.
Exercise 5 .6 .3
Find minimal realizations of the following 1 X 2 proper rational functions with

inputs r and y, and output e :
a. Cs-1 s+ 1]
s(s + 1)
2s2 + 1 -sz + s - 2
b.
s2 +3s+ 1 s2 +3s+ 1

PROBLEMS 183
PROBLEMS
5.1 . Write a program to compute the output y of the following state-variable equa-
tions due to a unit-step input . The initial conditions are assumed to be zero .
What integration step sizes will you choose?
-1 -2 1 .5
a. z =
81 -0.9 x + 1 .1 u
y = [0 .7 2.1]x
9 1 1 .1
b. x= 1 .9 -0.1
0 4.5 x+ 0 J u
1 2 5 2
y = [2.5 1 1 .2]x
5.2 . Repeat Problem 5 .1 using a commercially available software . Compare the

results with those obtained in Problem 5 .1 .
5.3 . Draw basic block diagrams for the equations in Problem 5 .1 .
5.4 . Draw operational amplifier circuits for the two state-variable equations in Prob-
lem 5 .1 . Use the elements in Figure 5 .4(e) and (f ) .
5 .5 . Develop a state-variable equation for the basic block diagram shown in Figure
P5.5 .
Figure P5.5

5.6. Develop a state-variable equation for the op-amp circuit shown in Figure P5 .6.
R/c I R
R/c 2
INM oy
C R/c 3
--A (- -
R W--R
RC = 1
R/a 3
R R/a 0
Figure P5 .6
5.7 . Draw a basic block diagram for the observable-form equation in (5 .18) and
then use Mason's formula to compute its transfer function .
5.8. Find realizations for the following transfer functions and draw their basic block
diagrams.
s 2 +2
a. G 1 (s) _
4s 3
3s 4 + 1
b. G2 (s) _
2s 4 + 3s 3 + 4s 2 + s + 5
(s + 3) 2
c. G3O
G3(S) __
(S + 1)2 (S + 2)
5.9. Show the equivalence of the blocks in Figure P5 .9.
b
U -31' . s-a
Figure P5.9

PROBLEMS 185
5 .10. Use Problem 5 .9 to change every block of Figure P5 .10 into a basic block
diagram and then develop state-variable equations to describe the two systems .
1 I
3
s+2 s+2
U V
y
x2
(a)
U I 1 x2 1 3 y
s+2 s+I s+2
(b)
Figure P5 .10
5.11. Find tandem and parallel realizations for the following transfer functions :
(s + 3)2
a
. (s + 1)2(s + 2)
(s + 3) 2
b
. (s + 2)2(s2 + 4s + 6)
5.12. a . Find a minimum realization of
S2 -4
G(s) _ 2
(s - 2)(2s + 3s + 4)
b. Find realizations of dimension 3 and 4 for the G(s) in (a).
5.13. a. Consider the armature-controlled dc motor in Figure 3 .1 . Show that if the
state variables are chosen as x l = 0, x2 = B and x 3 = i, then its state-
variable description is given by
xl 0 1 0 x, 0
z2 = 0 - f /J k,/J x, + 0
X3 0 - k,,/L, - R,/L, x3 1 L/ Q U
0 = [1 0 0]x

b. The transfer function of the motor is, as computed in (3.14),

_ t
G(s) = U(
U(s) s[(Js + f)(Ra + Las) + k~kb]
Find a realization of G(s) . [Although the realization is different from the
state-variable equation in (a), they are equivalent . See Section 11 .3.]
c. Every state variable in (a) is associated with a physical quantity . Can you
associate every state variable in (b) with a physical quantity?
5.14 . Consider the block diagram shown in Figure P5 .14. (a) Compute its overall
transfer function, realize it, and then draw a basic block diagram . (b) Draw a
basic block diagram for each block and then connect them to yield the overall
system . (c) If k is to be varied over a range, which basic block diagram, (a) or
(b), is more convenient?
r + 2s 2 +s-1 y
+s 2 +3s+2
Figure P5 .14
5.15 . Consider the block diagram shown in Figure P5 .15 . It has a tachometer feed-
back with transfer function 2s . Differentiators are generally not built using
operational amplifier circuits . Therefore, the diagram cannot be directly simu-
lated using operational amplifier circuits . Can you modify the diagram so that
it can be so simulated?
r + 1 y
s s+2)
Figure P5 .15
5 .16 . Find minimal realizations for the following 1 X 2 vector transfer functions :
a. s + 2 s + 11
(s+ 1) 2 s + 2
b. s - 2 s + 1
1(s +1)(s+2) s + 2
C. s 2 +S+ 1 5 3- 1
[ S 3 + 2S 2 + I S 3 + 2S 2 + l
5 .17 . Consider the block diagram shown in Figure 5 .11 . Suppose G 1 (s) and G2(S)
are given as in Problem 5 .16(c) . Draw a basic block diagram . Can you identify
the summer in Figure 5 .11 with a summer in your diagram?

PROBLEMS 187
5.18. a. Consider the block diagram in Figure 5 .7. Show that if the state variables
x 1 , x2 , x3 , and x4 in Figure 5 .7 are renamed as x4 , x3 , x2 , and x 1 , then the
block diagram can be described by
0 1 0 0
0 0 1 0
x(t) + u(t)
0 0 0 1
-ai -a2 - "3 -
y(t) = I& b, b2 b3]x(t) + du(t)
This is also called a controllable-form realization of (5.16). See References
[15, 16] . This is an alternative controllable form of (5 .17) .
b. Show that an alternative observable form of (5 .18) is
0 0 - ao
1 0 0
x(t) _ x(t) +
1 0 - a2
0 1 b3
y(t) = [0 0 0 1]x(t) + du(t)

M mw- Y.P?l M ,
Design Criteria,
Constraints,
and Feedback
6.1 INTRODUCTION
With the background introduced in the preceding chapters, we are now ready to
study the design of control systems . Before introducing specific design techniques,
it is important to obtain a total picture of the design problem . In this chapter, we
first discuss the choice of plants and design criteria and then discuss noise and
disturbance problems encountered in practice . These problems impose constraints
on control systems, which lead to the concepts of well-posedness and total stability .
We also discuss the reason for imposing constraints on actuating signals . Feedback
systems are then shown to be less sensitive to plant perturbations and external dis-
turbances then are open loop systems . Finally, we discuss two general approaches
in the design of control systems .
6.2 CHOICE OF A PLANT
We use an example to illustrate the formulation of the design problem . Suppose we

are given an antenna . The antenna's specifications, such as weight and moment of
inertia, and its operational range, such as average and maximum speed and accel-
eration, are also given. We are asked to design a control system to drive the antenna .
The first step in the design is to choose a motor . Before choosing the type of motor,
we must estimate the required horsepower . Clearly if a motor is too small, it will
not be powerful enough to drive the antenna; if it is too large, the cost will be
188

6 .3 PERFORMANCE CRITERIA 189
unnecessarily high. The torque required to drive the antenna is

Torque = J6(t)
where J is the moment of inertia of the antenna and 0(t) is its angular displacement .
The power needed to drive the antenna is
Power = Torque X Velocity = J6(t)6(t)
This equation shows that the larger the acceleration and the velocity, the larger the
power needed . Let bmax and 6max be the maximum acceleration and velocity . Then
the required power is
Power - J(emax)( 6max) (6.1)
In this computation, the moments of inertia of motor and gear trains, which are not
yet determined, are not included . Also, we consider neither the power required to
overcome static, Coulomb, and viscous frictions nor disturbances due to gusting .
Therefore, the horsepower of the motor should be larger than the one computed in
(6.1). After the size of a motor is determined, we must select the type of motor : dc,
ac, or hydraulic . The choice may depend on availability at the time of design, cost,
reliability, and other considerations . Past experience may also be used in this choice .
For convenience of discussion, we choose an armature-controlled dc motor to drive
the antenna . A dc generator is also chosen as a power amplifier, as shown in Figure
6.1 . This collection of devices, including the load, is called the plant of the control
system . We see from the foregoing discussion that the choice of a plant is not unique .
1 0
Cm
T -
V h = kh6 Bm Antenna
JL
(a)
--- ----
I
I I
I vs Nm I
U I G, (s) G2 (S) 0
I NL
I I
I G (s) I
L I
(b)
Figure 6 .1 (a) Plant. (b) Its block diagram .

190 CHAPTER 6 DESIGN CRITERIA, CONSTRAINTS, AND FEEDBACK
Different designers often choose different plants ; even the same designer may choose
different plants at different times .
Once a plant is chosen, the design problem is concerned with how to make the
best use of the plant . Generally, the plant alone cannot meet the design objective .
We must introduce compensators and transducers and hope that the resulting overall
system will meet the design objective . If after trying all available design methods,
we are still unable to design a good control system, then either the design objective
is too stringent or the plant is not properly chosen . If the design objective can be
relaxed, then we can complete the design . If not, then we must choose a new plant
and repeat the design . Thus a change of the plant occurs only as a last resort . Other-
wise, the plant remains fixed throughout the design .
6 .3 PERFORMANCE CRITERIA
Once a plant with transfer function G(s) is chosen, the design problem is to design
an overall system, as shown in Figure 6 .2, to meet design specifications . Different
applications require different specifications . For example, in designing a control
system to aim an astronomical telescope at a distant star, perfect aiming or accuracy
is most important; how fast the system achieves the aiming is not critical . On the
other hand, for a control system that drives guided missiles to aim at incoming enemy
missiles, speed of response is as critical as accuracy . In general, the performance of
control systems is divided into two parts : steady-state performance, which specifies
accuracy, and transient performance, which specifies the speed of response . The
steady-state performance may be defined for a step, ramp, or acceleration reference
input . The transient response, however, is defined only for a step reference input .
Before proceeding, it is important to point out that the behavior of a control
system depends only on its overall transfer function G0 (s) from the reference input
to the plant output y . It does not depend explicitly on the plant transfer function
G(s) . Thus, the design problem is essentially the search for a G 0 (s) to meet design
specifications . Let G0(s) be of the form
B(s) - (30 + 13s + 32s+ . . . + RmS n
G,(s) - (6.2)
A(s) a 0 + a l s + a2s2+ + as
with n ? m. If G0(s) is not stable, the output y(t) will approach infinity for almost
any reference input and the system will break down or burn out . Thus, every G 0(s)
r -1
I
I I
r u Y
G(s) T
I I
I I
I GO (S) I
L
Figure 6.2 The design problem .

to be designed must be stable . The design problem then becomes the search for a
stable G0(s) to meet design specifications .
6.3.1 Steady-State Performance-Accuracy

The steady-state performance is concerned with the response of y(t) as t approaches
infinity . It is defined for step, ramp, and acceleration inputs .
Step Reference Inputs

If the reference signal r(t) is a step function with amplitude a, that is, r(t) = a, for
t ? 0, then the steady-state output ys(t) of the overall system, as derived in (4 .25)
or by using the final-value theorem, is
ys(t) = lim y(t) = lim sY(s) = lim sG0(s) a = aGo(0) = a 00

t__ S-0 s-*0 s a0
In this derivation, the stability condition of G0(s) is essential . The percentage steady-
state error is then defined as
a - y(t)
Position error : = ep (t) : = lim
a
a-a-go
ao ao -/3 o
= 11 - Go (0)I
a ao
Because step functions correspond to positions, this error is called the position error .
Clearly if G0(0) = 1 or 130 = ao then the position error is zero . If we require the
position error to be smaller than y or 100y percent, then
ao-00
< y
a0
which, using a0 > 0 because of the stability assumption of G0(s), implies
-a0y < Ro - a0 < a0 y

or
(1 - y)ao < /3 0 < (1 + y)a o (6.4)
Thus, the specification on the position error can easily be translated into the constant
coefficients a 0 and 0 0 of G0(s). Note that the position error is independent of a, and
f3j, for i ? 1 .

Exercise 6 .3.1
Find the range of 60 so that the position error of the following transfer function is
smaller than 5% .
G(s) = go
s2 +2s+2
[Answer : 1.9 < go < 2.1 .]
Ramp Reference Inputs

If the reference signal is a ramp function or r(t) = at, for t ? 0 and a > 0, then the
steady-state output y,(t) of the overall system, as computed in (4.26), is
yr(t) = G0(0) at + Go(0) a = aoao at + a0R1 z aoa1 a

ao
(6.5)
where G ,(0) = dG0 (s)/dsI s _ 0. The percentage steady-state error due to a ramp
function is defined as
r(t) - y(t) at - y,(t)
Velocity error : = e (t) : = lim a
t__ a
_ I(1 - Go(0))t - Go(0)j (6.6)
ao -ao 1 a0f31- 0oai
a0 J a z0
This error will be called the velocity error, because ramp functions correspond to
velocities . We see that if G0(0) = ,130/x 0 1, then r(t) and ys (t) have different
slopes, as shown in Figure 6.3(a), and their difference approaches infinity as t -.
Thus the velocity error is infinity . Therefore, in order to have a finite velocity error,
we must have G 0 (0) = 1 or a0 = a 0 . In this case, r(t) and ys (t) have the same
0 0
(a) (b)
Figure 6.3 Velocity errors .

slope, as is shown in Figure 6.3(b), and the velocity error becomes finite and equals
ao/3, - g oal
Velocity error = e(t) =
a o0
Thus the conditions for having a zero velocity error are a o = /3o and a, = /3,, or
G0(0) = 1 and Go(0) = 0 . They are independent of a ; and /3 i, for i ? 2.
The preceding analysis can be extended to acceleration reference inputs or any
inputs that are polynomials oft . We will not do so here . We mention that, in addition
to the steady-state performances defined for step, ramp, and acceleration functions,
there is another type of steady-state performance, defined for sinusoidal functions .
This specification is used in frequency-domain design and will be discussed in
Chapter 8 .
The plant output y(t) is said to track asymptotically the reference input r(t) if
lim ly(t) -r(t)l = 0

t__
If the position error is zero, then the plant output will track asymptotically any step
reference input . For easy reference, we recapitulate the preceding discussion in the
following. Consider the design problem in Figure 6 .2. No matter how the system is
designed or what configuration is used, if its overall transfer function GO(s) in (6.2)
is stable, then the overall system has the following properties :
1. If Go(0) = 1 or a o = /3 o, then the position error is zero, and the plant output
will track asymptotically any step reference input.
2. If G o(0) = 1 and Go(0) = 0, or a o = /3o and a, = /3,, then the velocity error
is zero, and the plant output will track asymptotically any ramp reference input .
3. If G 0(0) = 1, Go(0) = 0, and Go(0) 0, or a o = /3o, a, = /3,, and a 2 = /3 z,
then the acceleration error is zero, and the plant output will track asymptotically
any acceleration input.
Thus, the specifications on the steady-state performance can be directly translated
into the coefficients of G 0(s) and can be easily incorporated into the design .
The problem of designing a system to track asymptotically a nonzero reference
signal is called the tracking problem . The more complex the reference signal, the
more complex the overall system . For example, tracking ramp reference inputs im-
poses conditions on a o, a,, /3o, and /3, ; tracking step reference inputs imposes con-
ditions only on a o and /3o . If the reference signal is zero, then the problem is called
the regulating problem . The response of regulating systems is excited by nonzero
initial conditions or disturbances, and the objective of the design is to bring such
nonzero responses to zero . This objective can be achieved if the system is designed
to be stable ; no other condition such as G o(0) = 1 is needed . Thus, if a system is
designed to track a step reference input or any nonzero reference input, then the
system can also achieve regulation . In this sense, the regulating problem is a special
case of the tracking problem.

6 .3 .2 System Types-Unity-Feedback Configuration

The conditions for the steady-state performance in the preceding section are stated
for overall closed-loop transfer functions . Therefore, they are applicable to any con-
trol configuration once its overall transfer function is computed . In this subsection,
we discuss a special case in which the conditions can be stated in terms of open-
loop transfer functions . Consider the unity-feedback configuration shown in Figure
6.4(a), where G(s) is the plant transfer function and C(s) is a compensator. We define
G,(s) = G(s)C(s)
and call it the loop transfer function . A transfer function is called a type i transfer
function if it has i poles at s = 0. Thus, if G,(s) is of type i, then it can be expressed
as
N1(s)
G I (s)
= s'D,(s)
with N,(0) ~ 0 and D,(0) :~ 4- 0. Now we claim that if G,(s) is of type 1, and if the
unity-feedback system is stable, then the position error of the unity-feedback system
is 0 . Indeed, if G,(s) is of type 1, then the overall transfer function is
N,(s)
(s) _ sD,(s) - N1(s)
Go
N,(s) sD,(s) + N,(s)
1 +
sD, (s)
Therefore we have
N1(0) = N,(0)
(0) =
G O(O) = 1
X D,(0) + N,(0) N1 (0)
which implies that the position error is zero . Thus, the plant output will track asymp-
totically any step reference input . Furthermore, even if there are variations of the
parameters of N,(s) and D,(s), the plant output will still track any step reference input
so long as the overall system remains stable . Therefore, the tracking property is said
to be robust. Using the same argument, we can show that if the loop transfer function
is of type 2 and if the unity-feedback system is stable, then the plant output will
track asymptotically and robustly any ramp reference input (Problem 6.7).
r
0 00
(a) (b) (c)
Figure 6 .4 (a) Unity-feedback system . (b) Unity-feedback system with a forward gain .
(c) Nonunity-feedback system .

If G,(s) is of type 0, that is, G,(s) = N,(s)/D,(s) with N(0) 34- 0 and D,(0) ~ 0,
then we have
N1(s) N(0)
G ,(s) and Go O
0 =
DI(s) + N,(s) D1 (0) + N,(0) 0 1
and the position error is different from zero . Thus, if the loop transfer function in
Figure 6 .4(a) is of type 0, then the plant output will not track asymptotically any
step reference input . This problem can be resolved, however, by introducing the
forward gain
D,(0) + N1(0)
k =
N,(0)
as shown_in Figure 6.4(b) . Then the transfer function G0 (s) from r to y has the
property Go(0) = 1, and the plant output will track asymptotically any step reference
input r. In practice, there is no need to implement gain k. By a proper calibration or
setting of r, it is possible for the plant output to approach asymptotically any desired
value .
There is one problem with this design, however . If the parameters of GI (s) _
G(s)C(s) change, then we must recalibrate or reset the reference input r . Therefore,
this design is not robust . On the other hand, if G,(s) is of type 1, then the tracking
property of Figure 6 .4(a) is robust, and there is no need to reset the reference input .
Therefore, in the design, it is often desirable to have type I loop transfer functions .
We mention that the preceding discussion holds only for unity-feedback sys-
tems. If a configuration is not unity feedback, such as the one shown in Figure 6.4(c),
even if the plant is of type 1 and the feedback system is stable, the position error is
not necessarily zero . For example, the transfer function of Figure 6 .4(c) is
1
S 1
G0(s) _
2 s + 2
1 + -
S,
Its position error, using (6 .3), is

2 - 1
ep = = 0.5 = 50%
2
The position error is not zero even though the plant is of type 1 . Therefore, system
types are useful in determining position or velocity errors in the unity-feedback
configuration, but not necessarily useful in other configurations .
6.3.3 Transient Performance-Speed of Response

Transient performance is concerned with the speed of response or the speed at which
the system reaches the steady state . Although the steady-state performance is defined
for step, ramp, or acceleration reference inputs, the transient performance is defined

Y Y
Overshoot
Ys
0 .9 Ys
0 .04 y,
tr ts t t
(a) (b)
Figure 6 .5 Transient performance .
only for step reference inputs . Consider the outputs due to a unit-step reference input
shown in Figure 6 .5, in which ys denotes the steady state of the output . The transient
response is generally specified in terms of the rise time, settling time, and overshoot .
The rise time can be defined in many ways . We define it as the time required for
the response to rise from 0 to 90% of its steady-state value, as shown in Figure 6 .5.
In other words, it is the smallest tr such that
At) = 0.9ys
The time denoted by is in Figure 6 .5 is called the settling time. It is the time for the
response to reach and remain inside 2% of its steady-state value, or it is the
smallest is such that
ly(t) - Ysl 0.02y s for all t - is
Let ymax be the maximum value of l y (t)l, for t ? 0, or
Ymax : = max ly(t) I
Then the overshoot is defined as
Ymax -Ys
Overshoot : = X 100%
Ys
For the response in Figure 6.5(a), if ymax = 1.3ys , then the overshoot is 30% . For
the response in Figure 6.5(b), because ymax = ys, the overshoot is zero or there is
no overshoot.
Control systems are inherently time-domain systems, so the introduced speci-
fications are natural and have simple physical interpretations . For example, in point-
ing a telescope at a star, the steady-state performance (accuracy) is the main concern ;
the specifications on the rise time, overshoot, and settling time are not critical . How-
ever, in aiming missiles at an aircraft, both accuracy and speed of response are
important . In the design of an aircraft, the specification is often given as shown in
Figure 6 .6. It is required that the step response of the system be confined to the
region shown . This region is obtained by a compromise between the comfort or

6 .4 NOISE AND DISTURBANCES 1 97
Figure 6 .6 Allowable step response.
physical limitations of the pilot and the maneuverability of the aircraft . In the design
of an elevator, any appreciable overshoot is undesirable . Different applications have
different specifications .
A system is said to be sluggish if its rise time and settling time are large. If a
system is designed for a fast response, or to have a small rise time and a small
settling time, then the system may exhibit a large overshoot, as can be seen from
Figure 4 .7 . Thus, the requirements on the rise time and overshoot are often conflict-
ing and must be reached by compromise .
The steady-state response of G0(s) depends only on a number of coefficients of
G0(s) ; thus the steady-state performance can easily be incorporated into the design .
The transient response of G0(s) depends on both its poles and zeros . Except for some
special cases, no simple relationship exists between the specifications and pole-zero
locations . Therefore, designing a control system to meet transient specifications is
not as simple as designing one to meet steady-state specifications .
6.4 NOISE AND DISTURBANCES
Noise and disturbances often arise in control systems . For example, if a potentiom-
eter is used as a transducer, noise will be generated (because of brush jumps, wire
irregularity, or variations of contact resistance) . Motors and generators also generate
noise because of irregularity of contact between carbon brushes and commutators .
Shot noise and thermal noise are always present in electronic circuits . Therefore,
noise, usually high-frequency noise, exists everywhere in control systems .
Most control systems will also encounter external disturbances . A cruising air-
craft may encounter air turbulence or air pockets . A huge antenna may encounter
strong or gusting winds . Fluctuations in power supply, mechanical vibrations, and
hydraulic or pneumatic pressure will also disturb control systems .
Variation of load is also common in control systems . For example, consider a
motor driving an audio or video tape . At the beginning and end, the amounts of tape
on the reel are quite different ; consequently, the moments of inertia of the load are
not the same. As a result, the transfer function of the plant, as can be seen from
(3.17), is not the same at all times . One way to deal with this problem is to choose
(a) (b)
Figure 6 .7 Systems with noise or disturbance entering at every block .
the average moment of inertia or the largest moment of inertia (the worst case),
compute the transfer function, and use it in the design . This transfer function is
called the nominal transfer function. The actual transfer function may differ from
the nominal one . This is called plant perturbation . Aging may also change plant
transfer functions . Plant perturbations are indeed inevitable in practice .
One way to deal with plant perturbation is to use the nominal transfer function
in the design . The difference between actual transfer function and nominal transfer
function is then considered as an external disturbance . Thus, disturbances may arise
from external sources or internal load variations . To simplify discussion, we assume
that noise and/or disturbance will enter at the input and output terminals of every
block, as shown in Figure 6 .7 . These inputs also generate some responses at the
plant output . These outputs are undesirable and should be suppressed or, if possible,
eliminated . Therefore, a good control system should be able to track reference inputs
and to reject the effects of noise and disturbances .
6 .5 PROPER COMPENSATORS AND WELL-POSEDNESS
In this and the following sections we discuss some physical constraints in the design
of control systems . Without these constraints, design would become purely a math-
ematical exercise and would have no relation to reality . The first constraint is that
compensators used in the design must have proper transfer functions . As discussed
in the preceding chapter, every proper transfer function can be realized as a state-
variable equation and then built using operational amplifier circuits . If the transfer
function of a compensator is improper, then its construction requires the use of pure
differentiators. Pure differentiators built by using operational amplifiers may be un-
stable. See Reference [18]. Thus compensators with improper transfer functions
cannot easily be built in practice . For this reason, all compensators used in the design
will be required to have proper transfer functions .
In industry, proportional-integral-derivative (PID) controllers or compensators
are widely used . The transfer functions of proportional and integral controllers are
kp and k;/s ; they are proper transfer functions . The transfer function of derivative
controllers is kd s, which is improper . However, in practice, derivative controllers

6 .5 PROPER COMPENSATORS AND WELL-POSEDNESS 199
are realized as
k, s
1 + Nkd s
for some constant N. This is a proper transfer function, and therefore does not violate
the requirement that all compensators have proper transfer functions . In the remain-
der of this chapter, we assume that every component of a control system has a proper
transfer function. If we encounter a tachometer with improper transfer function ks,
we shall remodel it as shown in Figure 3 .10(c) . Therefore, the assumption remains
valid.
Even though all components have proper transfer functions, a control system so
built may not have a proper transfer function . This is illustrated by an example .
Example 6.5.1
Consider the system shown in Figure 6.7(a). The transfer functions of the plant and
the compensator are all proper . Now we compute the transfer function Gyr(s) from
r to y . Because the system is linear, in computing Gyr(S), all other inputs shown
(n;, i = 1, 2, and 3) can be assumed zero or disregarded . Clearly we have
-(s + 1) s -s
s+2 s+ 1 s+2 -s
Gyr(S) = _ - 0.5s
-(S + 1) S I- S s+ 2 - s
1+
s+ 2 s+ 1 s+ 2
It is improper! Thus the properness of all component transfer functions does not
guarantee the properness of an overall transfer function .
Now we discuss the implication of improper overall transfer functions . As dis-

cussed in the preceding section, in addition to the reference input, noise and dis-
turbance may enter a control system as shown in Figure 6.7(a). Suppose r(t) = sin
t and n1 (t) = 0 .01 sin 10,000t where r(t) denotes a desired signal and n l (t) denotes
a high-frequency noise . Because the transfer function Gyr (S) is -0.5s, the plant
output is simply the derivative of r(t) and nl(t), scaled by -0.5. Therefore, we have
d
y(t) = (-0.5) (sin t + 0.01 sin 10,000t)
dt
-0.5 cos t - 0.5(0.01) X 10,000 X cos 10,000t
- 0.5 cos t - 50 cos 10,000t
Although the magnitude of the noise is one-hundredth of that of the desired signal
at the input, it is one hundred times larger at the plant output . Therefore, the plant

200 CHAPTER 6 DESIGN CRITERIA, CONSTRAINTS. AND FEEDBACK
output is completely dominated by the noise and the system cannot be used in
practice.
In conclusion, if a control system has an improper closed-loop transfer function,
then high-frequency noise will be greatly amplified and the system cannot be used .
Thus, a workable control system should not contain any improper closed-loop trans-
fer function . This motivates the following definition .
o Definition 6 .1
A system is said to be well-posed or closed-loop proper if the closed-loop
transfer function of every possible input/output pair of the system is proper .
We have assumed that noise and disturbance may enter a control system at the
input and output terminals of each block . Therefore, we shall consider not only the
transfer function from r to y, but also transfer functions from those inputs to all
variables . Let Gpq denote the transfer function from input q to output p. Then the
system in Figure 6 .7(a) or (b) is well posed if the transfer functions Gel., Gvr, Gur,
Gyr, Gen,, Gun, , Gyn ,, Gene, Gun2, Gyn2, Gen3, Gun3, Gyn3, are all proper . These transfer
functions are clearly all closed-loop transfer functions and, strictly speaking, the
adjective closed-loop is redundant . It is kept in Definition 6 .1 to stress their difference
from open-loop transfer functions .
The number of possible input/output pairs is quite large even for the simple
systems in Figure 6 .7. Therefore, it appears to be difficult to check the well-posed-
ness of systems. Fortunately, this is not the case. In fact, the condition for a feedback
system to be well posed is very simple . A system that is built with blocks with proper
transfer functions is well posed if and only if
A(co) =A 0 (6 .8)
where 0 is the characteristic function defined in (3.37). For the feedback systems in
Figure 6.7, the condition becomes
A(oc) = 1 + C(oc)G(cc) 0 0 (6 .9)
For the system in Figure 6.7(a), we have C(s) = - (s + 1)/(s + 2) and G(s) =
s/(s + 1) which imply C(cc) = - 1, G(co) = 1 and 1 + C(oo)G(oo) = 0 . Thus the
system is not well posed . For the system in Figure 6 .7(b), we have
+ 1) 2s
1 + C(s)G(s)I, = I + -(s = I + (-1)(2) -1 71~ 0
s + 2 s + 1 S= .
Thus the system is well posed . As a check we compute the closed-loop transfer
functions from n2 to u, y, e, and v in Figure 6.7(b). In this computation, all other
inputs are assumed zero . The application of Mason's formula yields
_ 1 1 1 s+ 2
G,,2(s)
2s-(s+ 1) 1 - 2s s + 2 - 2s -s + 2
s + I (s + 2) s + 2 s + 2

6. 5 PROPER COMPENSATORS AND WELL-POSEDNESS 201
2s
(S + 1) 2s(s + 2)
Gyn2 (s) _
1 + 2s . - (s + 1) (s + I)( - s + 2)
S + I s+2
- 2s(s + 2)
G en2 (S) _ - G1'2 (S) -
(S + I)( - S + 2)
and
2s -(S + 1) 2s
s+ 1 s+2 s+2 2s
G,'2(S) = 2s -(s + 1) s + 2 - 2s -s + 2 (6 .10)
1 +
S + 1 s+ 2 s+ 2
They are indeed all proper . Because the condition is (6.8) can easily be met, a control
system can easily be designed to be well posed . We remark that if a plant transfer
function G(s) is strictly proper and if C(s) is proper, then the condition in (6 .9) is
automatically satisfied . Note that the conditions in (6 .8) and (6.9) hold only if the
transfer function of every block is proper . If any one of them is improper, then the
conditions cannot be used .
To conclude this section, we discuss the relationship between well-posedness
and properness of compensators . Properness of compensators is concerned with
open-loop properness, whereas well-posedness is concerned with closed-loop prop-
erness . Open-loop properness does not imply closed-loop properness, as is demon-
strated in the system in Figure 6.7(a). It can be verified, by computing all possible
closed-loop transfer functions, that the system in Figure 6 .8 is well posed . However,
the system contains one improper compensator . Thus, well-posedness does not imply
properness of compensators . In conclusion, open-loop properness and closed-loop
properness are two independent concepts . They are also introduced for different
reasons. The former is needed to avoid the use of differentiators in realizing com-
pensators; the latter is needed to avoid amplification of high-frequency noise in
overall systems .
ni n
s+1
s+
4s+3
s+2
Figure 6.8 Well-posed system with an improper compensator .

6 .6 TOTAL STABILITY
In the design of control systems, the first requirement is always the stability of
transfer functions, G0(s), from the reference input r to the plant output y . However,
this may not guarantee that systems will work properly . This is illustrated by an
example .
Example 6.6 .1
Consider the system shown in Figure 6 .9 . The transfer function from r to y is

s - 1 1 1
S + 1 s- 1 s+ 1 _ 2
G0(s) = 2 X (6 .11)
s- 1 1 = 2 X 1 s+ 2
1 + 1 +
S + 1 s- 1 s+ 1
It is stable . Because G0(0) = 1, the position error is zero . The time constant of the
system is 1/2 = 0 .5. Therefore, the plant output will track any step reference input
in about 5 X 0.5 = 2 .5 seconds . Thus the system appears to be a good control
system.
A close examination of the system in Figure 6 .9 reveals that there is a pole-zero
cancellation between C(s) and G(s) . Will this cause any problem? As was discussed
earlier, noise or disturbance may enter a control system . We compute the transfer
function from n to y in Figure 6 .9:
1 1
s 1 s 1 s+ 1
G,, (s) = .12)
1 + s - 1 1 1 (S - 1)(s + 2) (6
1 +
S + 1 s- 1 s+ 1
r I y
2
S-1
C(s) G(s)
Figure 6 .9 Feedback system with pole-zero cancellation .
It is unstable! Thus any nonzero noise, no matter how small, will excite an un-
bounded plant output and the system will bum out . Therefore the system cannot be
used in practice, even though its transfer function from r to y is stable . This motivates
the following definition .
6 .6 TOTAL STABILITY 203
o Definition 6.2
A system is said to be totally stable if the closed-loop transfer function of every
possible input-output pair of the system is stable .
Because of the presence of noise, every control system should be required to

be totally stable . Otherwise, noise will drive some variable of the system to infinity
and the system will disintegrate or burn out . From Example 6.6 .1, we see that if a
plant has an unstable pole, it is useless to eliminate it by direct cancellation . Although
the canceled pole does not appear in the transfer function G0(s) from r to y, it appears
in the transfer function from n to y. We call the pole a missing pole or a hidden pole
from G0(s). Any unstable pole-zero cancellation will not actually eliminate the un-
stable pole, only make it hidden from some closed-loop transfer functions .
We now discuss the condition for a system to be totally stable . Consider a system
that consists of a number of subsystems . Every subsystem is assumed to be com-
pletely characterized by its proper transfer function . See Section 2 .5. Let G0(s) be
the overall transfer function . If the number of poles of G 0(s) equals the total number
of poles of all subsystems, then the system is completely characterized by G 0(s). If
not, the system is not completely characterized by G0(s) and G 0(s) is said to have
missing poles . Missing poles arise from pole-zero cancellation' and their number
equals
Number of poles of G0(s) - [Total number of poles of all subsystems]
With this preliminary, we are ready to state the condition for a system to be
totally stable. The condition can be stated for any closed-loop transfer function and
its missing poles . A system is totally stable if and only if the poles of any overall
transfer function and its missing poles are all stable poles . For example, consider
the system shown in Figure 6 .9. Its overall transfer function GO (s) from r to y is
computed in (6.11). It has 1 pole, which is less than the total number of poles of
G(s) and C(s) . Therefore, there is one missing pole . The pole of G 0(s) is stable, but
the missing pole s - 1 is unstable . Therefore, the system is not totally stable . The
same conclusion can also be reached by using the transfer function GY,, from n to y
computed in (6 .12). The number of poles of Gy equals the total number of poles of
G(s) and C(s) . Therefore, Gy has no missing pole . The system in Figure 6 .9 is totally
stable if and only if all poles of Gy are stable poles . This is not the case . Therefore,
the system is not totally stable .
6.6.1 Imperfect Cancellations

As discussed in the preceding section, a system cannot be totally stable if there is
any unstable pole-zero cancellation . In fact, it is impossible to achieve exact pole-
'In addition to pole-zero cancellations, missing poles may also arise from parallel connection and
other situations . See Reference [15, pp . 436-437] . In this text, it suffices to consider only pole-zero
cancellations .

zero cancellation in practice . For example, suppose we need a 10-kil resistor to

realize C(s) = (s - 1)/(s + 1) in Figure 6 .9 . However, the resistor we use may
not have a resistance exactly equal to 10 W . . Furthermore, because of change in
temperature or aging, the resistance may also change . Thus, the compensator we
realize may become C'(s) = (s - 0.9)/(s + 1.1), rather than the intended
(s - 1)/(s + 1), as shown in Figure 6 .10, and the unstable pole (s - 1) will not
be canceled. Even if we can realize C(s) = (s - 1)/(s + 1) exactly, the plant
transfer function 1/(s - 1) may change due to load variations or aging . Again, exact
pole-zero cancellation cannot be achieved. In conclusion, because of inexact imple-
mentation of compensators and plant perturbations due to load variations and aging,
all we can achieve is imperfect pole-zero cancellation . In this section, we study the
effect of imperfect cancellations . The transfer function from r to y in Figure 6 .10 is
s - 0.9 1
s + 1 .1 s - 1 - 2 2(s - 0.9)
G'(s) = 2 X
S - 0.9 1 s + 0 .1s - 1 .1 + s - 0.9
1 +
S + 1 .1 s- 1 (6.13)
2(s - 0.9) 2(s - 0.9)
s2 + l .ls - 2 (s + 2.0674)(s - 0.9674)
It is unstable! Thus, its step response y(t) will approach infinity and is entirely
different from the one in Figure 6 .9. In conclusion, unstable pole-zero cancellations
are permitted neither in theory nor in practice .
Stable pole-zero cancellations, however, are an entirely different matter . Con-
sider the system shown in Figure 6 .11(a) . The plant transfer function is G(s) =
3/(s 2 + O.ls + 100), and the compensator transfer function is C(s) =
(s2 + O.ls + 100)/s(s + 2) . Note that C(s) is of type 1, therefore the position error
of the unity feedback system is zero . The overall transfer function from r to y is
s2 +Ols+ 100 3
_ s(s + 2) s 2 +O.ls+ 100
G(s)
S 2 + O.ls + 100 3
1 +
s(s + 2) s2 + O .ls + 100 (6.14)
3
_ s(s + 2) 3
3 s2 +2s+3
1 +
s(s+2)
The number of the poles of G 0 (s) is 2, which is 2 less than the total number of poles
of G(s) and C(s) . Thus, G0(s) has two missing poles ; they are the roots of s 2 + O.ls
+ 100. Because the poles of G 0 (s) and the two missing poles are stable, the system
is totally stable. The unit-step response of G0 (s) is computed, using MATLAB, and
plotted in Figure 6.11(b) with the solid line .
Now we study the effect of imperfect pole-zero cancellations . Suppose the com-
pensator becomes

6 .6 TOTAL STABILITY 205
I- t
r
S-0.9 - 4
S+ 1 .1
Figure 6 .10 Feedback system with an imperfect cancellation .
s2+0 .09s+99
C'(s)=
s(s + 2)
due to aging or inexact realization . With this C(s), the transfer function of Figure
6.11 (a) becomes
s2 +0.09s+99 3
s(s + 2) s 2 +0 .1s+ 100
G' (s) _ s 2 +0.09s+99 3
1 +
s(s + 2) s2 +0.ls+ 100
_ 3(s 2 + 0.09s + 99)
(6.15)
s(s + 2)(s 2 + 0.1s + 100) + 3(s 2 + 0.09s + 99)
_ 3s 2 + 0.27s + 297
S 4 + 2.1s 3 + 103.2s2 + 200.27s + 297
P
r + s2 +0 .15+100 3 y
-*0-3.
s(s+2) s +0 .15+100
(a)
1 .2
1
0.8
0.6
0.4
0.2
0
1 2 3 4 5 6 8
(b)
Figure 6.11 (a) Feedback system. (b) Its step response.

Its unit-step response is plotted in Figure 6 .11(b) with the dotted line . We see that
the two responses in Figure 6 .11(b) are hardly distinguishable . Therefore, unlike
unstable pole-zero cancellation, imperfect stable pole-zero cancellations may not
cause any serious problem in control systems .
6.6.2 Design Involving Pole-Zero Cancellations

Unstable pole-zero cancellations are not permitted in the design of control systems .
How about stable pole-zero cancellations? As was discussed earlier, pole-zero can-
cellations do not really cancel the poles . The poles become hidden in some closed-
loop transfer functions, but may appear in other closed-loop transfer functions . For
example, the transfer function from disturbance p to the plant output y in Figure
6.11 (a) is given by
3
s +0.ls+ 100
2
Gyp =
s +Ols+ 100
2 3
1 +
s(s + 2) + O.ls + 100
s2
(6,16)
3
s +0.1s+ 100
2 3s(s + 2)
3 (s + 2s + 3)(s2 + O.ls + 100)
2
1 +
s(s + 2)
We see that the complex-conjugate poles s + O.ls + 100 = (s + 0 .05 + j9 .999)
2
(s + 0 .05 - j9.999), which do not appear in G0(s), do appear in Gyp. Because the
poles have a very small real part and large imaginary parts, they will introduce high-
frequency oscillation and slow decay . Indeed, if the disturbance is modeled as a unit-
step function, then the excited output is as shown in Figure 6 .12. Even though the
response eventually approaches zero, it introduces too much oscillation and takes
too long to decay to zero . Therefore, the system is not good, and the stable pole-
zero cancellation should be avoided .
Can pole-zero cancellations be used in the design? Unstable pole-zero cancel-
lations are not permitted, because the resulting system can never be totally stable .
0 .06
0 .05
0 .04
0 .03
0 .02 r r r 41
0 .01
0
-0.01
-0.02, J
-0 .03
-0.04 .
0 1 2 3 5 6 7 8 9 10
Figure 6.12 Effect of step disturbance .

6 .7 SATURATION-CONSTRAINT ON ACTUATING SIGNALS 207
Figure 6.13 Permissible pole cancellation region .
A system can be totally stable with stable pole-zero cancellations . However, if can-
celed stable poles are close to the imaginary axis or have large imaginary parts, then
disturbance or noise may excite a plant output that is oscillatory and slow decaying .
If poles lie inside the region C shown in Figure 6 .13, then their cancellations will
not cause any problems . Therefore, perfect or imperfect cancellation of poles lying
inside the region C is permitted in theory and in practice . The exact boundary of the
region C depends on each control system and performance specifications, and will
be discussed in the next chapter .
There are two reasons for using pole-zero cancellation in design . One is to
simplify design, as will be seen in the next chapter . The other reason is due to
necessity . In model matching, we may have to introduce pole-zero cancellations to
insure that the required compensators are proper . This is discussed in Chapter 10 .
6.7 SATURATION-CONSTRAINT ON ACTUATING SIGNALS
In the preceding sections, we introduced the requirements of well-posedness and

total stability in the design of control systems . In this section, we introduce one more
constraint. This constraint arises from modeling and from devices used, and is prob-
ably the most difficult to meet in design .
Strictly speaking, most physical systems are nonlinear . However, they can often
be modeled as linear systems within some limited operational ranges . If signals
remain inside the ranges, the linear models can be used . If not, the linear models do
not hold, and the results of linear analyses may not be applicable . This is demon-
strated by an example .
Example 6.7 .1
Consider the system shown in Figure 6 .14 . The element A is an amplifier with gain
2. The overall transfer function is

A
e n
-> s+2
2
(a) (b)
Figure 6 .14 Amplifier with saturation .
s + 2
2
s(s - 1) 2(s + 2) 2(s + 2)
G0 (s) = (6.17)
s+2 s2 -s+2s+4 s2 +s+4
1+2-
s(s - 1)
Because G0(0) = 4/4 = 1, the system has zero position error and the plant output
will track any step reference input without an error . The plant outputs excited by
r = 0.3, 1.1, and 1 .15 are shown in Figure 6 .15 with the solid lines .
In reality, the amplifier may have the characteristic shown in Figure 6 .14(b) .
For ease of simulation, the saturation is approximated by the dashed lines shown .
The responses of the system due to r = 0 .3, 1 .1, and 1 .15 are shown in Figure 6 .15
with the dashed lines . These responses are obtained by computer simulations . If
r = 0.3, the amplifier will not saturate and the response is identical to the one
obtained by using the linear model . If r = 1 .1, the amplifier saturates and the re-
sponse differs from the one obtained by using the linear model . If r = 1 .15, the
Figure 6 .15 Effect of saturation .

6 .8 OPEN-LOOP AND CLOSED-LOOP CONFIGURATIONS 209
response approaches infinity oscillatorily and the system is not stable, although the
linear model is always stable for any r . This example shows that linear analysis
cannot be used if signals run outside the linear range .
This example is intentionally chosen to dramatize the effect of saturation . In

most cases, saturation will make systems only more sluggish, rather than unstable .
In any case, if a control system saturates, then the system will not function as de-
signed . Therefore, in design, we often require
u(t)I < M (6 .18)
for all t ? 0, where u(t) is the actuating signal and M is a constant . This constraint
arises naturally if valves are used to generate actuating signals . The actuating signals
reach their maximum values when the valves are fully open . In a ship steering
system, the constraint exists because the rudder can turn only a finite number of
degrees. In electric motors, because of saturation of the magnetic field, the constraint
also exists . In hydraulic motors, the movement of pistons in the pilot cylinder is
limited. Thus, the constraint in (6 .18) exists in most plants . Strictly speaking, similar
constraints should also be imposed upon compensators . If we were to include all
these constraints, the design would become very complicated . Besides, compared
with plants, compensators are rather inexpensive, and hence, if saturated, can be
replaced by ones with larger linear ranges . Therefore, the saturation constraint is
generally imposed only on the plant .
Actuating signals depend on reference input signals . If the amplitude of a ref-
erence signal is doubled, so is that of the actuating signal . Therefore, in checking
whether or not the constraint in (6 .18) is met, we shall use the largest reference input
signal . However, for convenience, the constraint M in (6 .18) will be normalized to
correspond to unit-step reference inputs . Therefore, in design, we often require the
actuating signal due to a unit-step reference input to have a magnitude less than a
certain value .
To keep a plant from saturating is not a simple problem, because in the process
of design, we don't know what the exact response of the resulting system will be .
Hence, the saturation problem can be checked only after the completion of the de-
sign . If saturation does occur, the system may have to be redesigned to improve its
performance.
6 .8 OPEN-LOOP AND CLOSED-LOOP CONFIGURATIONS
In this section we discuss the configuration of control systems . Given a control

problem, it is generally possible to use many different configurations to achieve the
design . It is therefore natural to compare the relative merits of various configurations .
To simplify the discussion, we compare only the two configurations shown in Figure
6.16. Figure 6 .16(a) is an open-loop configuration, Figure 6 .16(b) a closed-loop or
feedback configuration .

I I
L__ G(s)__ J
I I
LG(G I
(a) (b)
Figure 6.16 Open- and closed-loop systems .
In order to compare the two configurations, we must consider noise and dis-
turbances as discussed in Section 6 .4. If there were no noise and disturbances in
control systems, then there would be no difference between open-loop and closed-
loop configurations . In fact, the open-loop configuration may sometimes be prefer-
able because it is simpler and less expensive . Unfortunately, noise and disturbances
are unavoidable in control systems .
The major difference between the open-loop and feedback configurations is that
the actuating signal of the former does not depend on the plant output . It is pre-
determined, and will not change even if the actual plant output is quite different
from the desired value . The actuating signal of a feedback system depends on the
reference signal and the plant output . Therefore, if the plant output deviates from
the desired value due to noise, external disturbance, or plant perturbations, the de-
viation will reflect on the actuating signal . Thus, a properly designed feedback sys-
tem should perform better than an open-loop system . This will be substantiated by
examples .
Example 6 .8.1
Consider the two amplifiers shown in Figure 6 .17 . Figure 6 .17(a) is an open-loop
amplifier . The amplifier in Figure 6 .17(b) is built by connecting three identical open-
loop amplifiers and then introducing a feedback from the output to the input as
shown, and is called a feedback amplifier. Their block diagrams are also shown in
Figure 6 .17 . The gain of the open-loop amplifier is A = - 1OR/R = -10 . From
Figure 6.17(b), we have
e= -(1R r+ RR y) = -10 r+R y) =A r+R y/

f f fs
Thus the constant (3 in the feedback loop equals R/R f . Note that the feedback is
positive feedback . The transfer function or the gain of the feedback amplifier is
A3
A0 = 1 - 13A3 (6 .19)

Rf= 10 .101 R
A/V~__
1OR 1OR IOR IOR
-MM- -AA&- -Wv-
r R R R
0-NW'- -NA- JWb- Y
0
Y r + e
~A=-10H I A I A A
I1I
(a) (b)
Figure 6 .17 Open- and closed-loop amplifiers .
In order to make a fair comparison, we shall require both amplifiers to have the same
gain-that is, A = A = - 10 . To achieve this, /3 can readily be computed as /3 =
0.099, which implies Rf = 10.101R .
The feedback amplifier needs three times more hardware and still has the same
.gain as the open-loop amplifier . Therefore, there seems no reason to use the former .
Indeed, this is the case if there is no perturbation in gain A .
Now suppose gain A decreases by 10% each year due to aging . In other words,
the gain becomes - 9 in the second year, - 8.1 in the third year, and so forth . We
compute
( - 9)3
A 1 - 0 .099( - 9)3 - 9.96
(-8 .1)3
A = -9.91
1 - 0.099(-8 .1)3
and so forth . The results are listed in the following :
1 2 3 4 5 6 7 8 9 10
-A 10 9.0 8.1 7.29 6.56 5 .9 5.3 4.78 4.3 3.87
-A0 10 9.96 9.92 9.84 9.75 9.63 9.46 9.25 8.96 8.6
We see that although the open-loop gain A decreases by 10% each year, the closed-
loop gain decreases by from 0 .4% in the first year to 4 .1 % in the tenth year . Thus
the feedback system is much less sensitive to plant perturbation .

If an amplifier is to be taken out of service when its gain falls below 9, then the
open-loop amplifier can serve only one year, whereas the closed-loop amplifier can
last almost nine years . Therefore, even though the feedback amplifier uses three
times more hardware, it is actually three times more economical than the open-loop
amplifier . Furthermore, the labor cost of yearly replacement of open-loop amplifiers
can be saved.
Example 6 .8.2
Consider a speed control problem in an aluminum factory . Heated aluminum ingots
are pressed into sheets through rollers as shown in Figure 6 .18(a) . The rollers are
driven by armature-controlled dc motors ; their speeds are to be kept constant in order
to maintain uniform thickness of aluminum sheets . Let the transfer function of the
motor and rollers be
10
G(s) = 5s + 1 (6 .20)
Its time constant is 5 ; therefore it will take 25 seconds (5 X time constant) for the
rollers to reach the final speed . This is very slow, and we decide to design an overall
system with transfer function
2
G(s) = (6 .21)
s + 2
Its time constant is 1/2 = 0 .5 ; therefore the speed of response of this system is
much faster . Because G 0(0) = 2/2 = 1, the plant output of G0(s) will track any
step reference input without an error .
Now we shall implement G0(s) in the open-loop and closed-loop configurations
shown in Figure 6.18(b) and (c) . For the open-loop configuration, we require
10
5s+ 1
C2 (s) 10 Y
5s+ 1
(a)
(c)
Figure 6 .18 Speed control of rollers .

G0(s) = G(s)C,(s). Thus the open-loop compensator C,(s) is given by

G0(s) 2 5s + 1 _ 5s + 1
C , (s) = (6 .22)
G(s) s + 2 10 5(s + 2)
For the closed-loop configuration, we require
C2(s)G(s)
G,(s) - (6 .23)
1 + C2(s)G(s)
or
G0(s) + G o(s)C2(s)G(s) = C2(s)G(s)
Thus the closed-loop compensator C 2(s) is given by
G0(s) 2/(s + 2) _ 2/(s + 2)
CZ
(s) = G(s)(1 - G0(s)) 10 2 . 10 s
5s+ 1 C1 _ s+2) 5s+ 1 s+2 (6 .24)
5s + 1 1
_ 5s = 1 + -
5s
It consists of a proportional compensator with gain 1 and an integral compensator
with transfer function 115s ; therefore it is called a PI compensator or controller. We
see that it is a type I transfer function .
If there are no disturbances and plant perturbation, the open-loop and closed-
loop systems should behave identically, because they have the same overall transfer
function. Because G0(0) = 1, they both track asymptotically any step reference input
without an error. In practice, the load of the rollers is not constant . From Figure
6.18(a), we see that before and after the engagement of an ingot with the rollers,
the load is quite different. Even after the engagement, the load varies because of
nonuniform thickness of ingots . We study the effect of this load variation in the
following .
Plant Perturbation
The transfer function of the motor and rollers is assumed as G(s) = 10/(5s + 1)
in (6.20). Because the transfer function depends on the load, if the load changes, so
does the transfer function . Now we assume that, after the design, the transfer function
changes to
9
G(s) = (6 .25)
(4.5s + 1)
This is called plant perturbation . We now study its effect on the open-loop and
closed-loop systems .
After plant perturbation, the open-loop overall transfer function becomes
5s + 1 9
G00(s) = C,(s)G(s) (6 .26)
= 5(s + 2) (4.5s + 1)

Because Goo(0) = 9/10 0 1, this perturbed system will not track asymptotically
any step reference input . Thus the tracking property of the open-loop system is lost
after plant perturbation .
Now we compute the overall transfer function of the closed-loop system with
perturbed G(s) in (6.25). Clearly, we have
5s + 1 9
C2(s)G(s) 5s 4.5s + 1
1 + C2(s)G(s) 5s+I 9
1+
5s 4.5s + 1 (6,27)
9(5s + 1) 45s + 9
5s(4 .5s + 1) + 9(5s + 1) 22.5s2 + 50s + 9
Because G 0( .(0) = 1, this perturbed overall closed-loop system still track any step
reference input without an error . In fact, because the compensator is of type 1, no
matter how large the plant perturbation is, the system will always track asymptoti-
cally any step reference input, so long as the overall system remains stable . This is
called robust tracking . In conclusion, the tracking property is destroyed by plant
perturbation in the open-loop system but is preserved in the closed-loop system .
Disturbance Rejection
One way to study the effect of load variations is to introduce plant perturbations as
in the preceding paragraphs . Another way is to introduce a disturbance p(t) into the
plant input as shown in Figure 6 .18(b) and (c). Now we study the effect of this
disturbance in the open-loop and closed-loop systems . From Figure 6 .18(b), we see
that the transfer function from p to y is not affected by the open-loop compensator
C1(s) . If the disturbance is modeled as a step function of magnitude a, then it will
excite the following plant output
10 a
Yp (s) _ s (6.28)
Ss + I
Its steady-state output, using the final-value theorem, is
10 a = lOa
yp (oo) = lim yp(t) = lim sYp(s) = lim s
t-. S-o S-0 5s + 1 ' s
In other words, in the open-loop configuration, the step disturbance will excite a
nonzero plant output ; therefore, the speed of the rollers will differ from the desired
speed . For example, if the disturbance is as shown in Figure 6 .19(a), then the speed
will be as shown in Figure 6 .19(b) . This differs from the desired speed and will
cause unevenness in thickness of aluminum sheets . Thus, the open-loop system is
not satisfactory .

P(t) Yo Yc
r L r r~y~
75100
I I ` ' '_ t ' 1 ' 1 ' > t ' ' ' ' ' t
0 25 50 1 0 25 50 75 100 0 25 50 75 100
(a) (b) (c)

Figure 6.19 Effects of disturbance.
Now we study the closed-loop system . The transfer function from p to y is,
using Mason's formula,
G(s) _ 10/(5s + 1) _ lOs
G
Gyp (6.29)
(s) = 1 + G(s)C 2(s) 1 + 10 5s + 1 (5s + 1)(s + 2)
5s + 1 5s
Now if the disturbance is P(s) = a/s, the steady-state output y p due to the disturbance
is
yp(oo) = lim y (t) = lim sYP (s) = lim sG YP(S) Sa

t__ P S-0 S-0
lOsa
=lim = 0
S-0 (5s + 1)(s + 2)
This means that the effect of the disturbance on the plant output eventually vanishes .
Thus, the speed of the rollers is completely controlled by the reference input, and
thus, in the feedback configuration, even if there are disturbances, the speed will
return, after the transient dies out, to the desired speed, as shown in Figure 6.19(c) .
Consequently, evenness in the thickness of aluminum sheets can be better
maintained .
We remark that in the closed-loop system in Figure 6 .18(c), there is a pole-zero
cancellation . The canceled pole is - 1/5, which is stable but quite close to the
jw-axis . Although this pole does not appear in G0(s) in (6.21), it appears in G,d(s)
in (6.29). Because of this pole (its time constant is 5 seconds), it will take roughly
25 seconds (5 X time constant) for the effect of disturbances to vanish, as is shown
in Figure 6 .19(c) . It is possible to use different feedback configurations to avoid this
pole-zero cancellation . This is discussed in Chapter 10 . See also Problem 6 .14.
From the preceding two examples, we conclude that the closed-loop or feedback
configuration is less sensitive to plant perturbation and disturbances than the open-
loop configuration. Therefore, in the remainder of this text, we use only closed-loop
configurations in design .

6.9TWO BASIC APPROACHES IN DESIGN
With the preceding discussion, the design of control systems can now be stated as
follows : Given a plant, design an overall system to meet a given set of specifications .
We use only feedback configurations because they are less sensitive to disturbances
and plant perturbation than open-loop configurations are . Because improper com-
pensators cannot easily be built in practice, we use only compensators with proper
transfer functions . The resulting system is required to be well posed so that high-
frequency noise will not be unduly amplified . The design cannot have unstable pole-
zero cancellation, otherwise the resulting system cannot be totally stable . Because
of the limitation of linear models and devices used, a constraint must generally be
imposed on the magnitude of actuating signals . The following two approaches are
available to carry out this design:
1. We first choose a feedback configuration and a compensator with open param-
eters . We then adjust the parameters so that the resulting feedback system will
hopefully meet the specifications .
2. We first search for an overall transfer function G0(s) to meet the specifications .
We then choose a feedback configuration and compute the required
compensator.
These two approaches are quite different in philosophy . The first approach starts
from internal compensators and works toward external overall transfer functions .
Thus, it is called the outward approach . This approach is basically a trial-and-error
method . The root-locus and frequency-domain methods discussed in Chapters 7 and
8 take this approach . The second approach starts from external overall transfer func-
tions and then computes internal compensators, and is called the inward approach.
This approach is studied in Chapters 9 and 10 . These two approaches are independent
and can be studied in either order. In other words, we may study Chapters 7 and 8,
and then 9 and 10, or study first Chapters 9 and 10, and then Chapters 7 and 8 .
To conclude this chapter, we mention a very important fact of feedback . Con-
sider a plant with transfer function G(s) = N(s)/D(s) and consider the feedback
configuration shown in Figure 6 .20. Suppose the transfer function of the compensator
is C(s) = B(s)/A(s) . Then the overall transfer function is given by
B(s) N(s)
C(s)G(s) _ A(s) D(s) _ B(s)N(s)
G,(s) =
I + C(s)G(s) 1 + B(s) N(s) A(s)D(s) + B(s)N(s)
A(s) D(s)
r + u Y
C(s) G (s)
Figure 6 .20 Feedback system .

PROBLEMS 217
The zeros of G(s) and C(s) are the roots of N(s) and B(s) ; they remain to be the zeros
of G0(s). In other words, feedback does not affect the zeros of G(s) and C(s) . The
poles of G(s) and C(s) are the roots of D(s) and A(s) ; after feedback, the poles of
G0(s) become the roots of A(s)D(s) + B(s)N(s) . The total numbers of poles before
feedback and after are the same, but their positions have now been shifted from D(s)
and A(s) to A(s)D(s) + B(s)N(s) . Therefore, feedback affects the poles but not the
zeros of the plant transfer function . The given plant can be stable or unstable, but
we can always introduce feedback and compensators to shift the poles of G(s) to
desired position . Therefore feedback can make a good overall system out of a bad
plant . In the outward approach, we choose a C(s) and hope that G 0(s) will be a good
overall transfer function . In the inward approach, we choose a good G 0 (s) and then
compute C(s) .
PROBLEMS
6.1 . Find the ranges of f3i so that the following transfer functions have position
errors smaller than 10% .
a.. + 00
s 2 +2s+2
2 + f31 s + f3o
b. s3a2s 2
+3s +2s+3
c. s 3 I32s 2 +p1s+I3
+2s 2 +9s+68
G_I
6.2. Find the ranges of f3 ; so that the transfer functions in Problem 5 .1 have velocity
errors smaller than 10% .
6.3. Consider the three systems shown in Figure P6 .3. Find the ranges of k so that
the systems are stable and have position errors smaller than 10% .
6.4. Repeat Problem 6 .3 so that the systems have velocity errors smaller than 10% .
6.5. a. Find the range of k o such that the system in Figure P6 .5 is stable . Find the
value of k0 such that the system has a zero position error or, equivalently,
such that y will track asymptotically a step reference input .
b. If the plant transfer function in Figure P6 .5 becomes 5 .1/(s - 0.9) due to
aging, will the output still track asymptotically any step reference input? If
not, such a tracking is said to be not robust.
6 .6. Consider the unity feedback system shown in Figure 6 .4(a). We showed there
that if the loop transfer function G1(s) = C(s)G(s) is of type I or, equivalently,
can be expressed as
G,(S)
sD I(s)
where N1(0) 0 0 and D,(0) =A 0, and if the feedback system is stable, then the
1 Y
k -0
(s + 2)(s + 1)
(a)
r + 2 Y
(D----v- k
s(s + 1)(s + 2)
(b)
s+1 Y
s(s+2)
k 2s
(c) Figure P6 .3
r
ko 5 Y
s-I
Figure P6 .5
plant output will track asymptotically any step reference input . Now show that
the tracking is robust in the sense that, even if there are perturbations in N(s)
and D(s), the position error is still zero as long as the system remains stable .
6.7. a. Consider the unity feedback system shown in Figure 6.4(a). Show that if
G,(s) is of type 2 or, equivalently, can be expressed as
N,(s)
G,(s)
= s2D1(s)
with N,(0) =A 0 and D,(0) =A 0, and if the unity feedback system is stable,
then its velocity error is zero . In other words, the plant output will track
asymptotically any ramp reference input .
b. Show that the tracking of a_ramp reference input is robust even if there are
perturbations in N,(s) and D,(s) as long as the system remains stable . Note
that GI(s) contains 1/s 2, which is the Laplace transform of the ramp refer-
ence input. This is a special case of the internal model principle, which
states that if G1(s) contains R(s), then y(t) will track r(t) asymptotically and
the tracking is robust. See Reference [15] .

PROBLEMS 219
6.8. Consider the system shown in Figure P6 .8. Show that the system is stable . The
plant transfer function G(s) is of type 1, is the position error of the feedback
system zero? In the unity feedback system, the position and velocity error can
be determined by system type. Is this true in nonunity feedback or other con-
figurations?
G (s)
Figure P6.8
6.9. Show that if a system is designed to track t2, then the system will track any
reference input of the form ro + r t t + r2 t2.
6.10 . The movement of a recorder's pen can be controlled as shown in Figure
P6.10(a) . Its block diagram is shown in Figure P6 .10(b). Find the range of k
such that the position error is smaller than I% .
Input Summing Modulator ac

network amplifier
ac
motor
r = 0.02 m
(a)
1 Bm
k s(s + 2) 0 .02 Y
(b)
Figure P6 .10

6 .11 . Consider the systems shown in Figure P6 .11. Which of the systems are not
well posed? If not, find the input-output pair that has an improper closed-loop
transfer function.
r + V r + Y
1 s+I
(a) (b)
mom
(c) (d)
Figure P6 .11
6 .12 . Discuss the total stability of the systems shown in Figure P6 .11 .
6 .13 . Consider the speed control of rollers discussed in Figure 6 .18. We now model
the plant transfer function as 10/(Ts + 1), with r ranging between 4 and 6 .
Use the compensators computed in Figure 6.18 to compute the steady-state
outputs of the open-loop and feedback systems due to a unit-step reference
input for the following three cases : (a) T equals the nominal value 5, (b) T =
4, and (c) r = 6. Which system, open-loop or feedback system, is less sensitive
to parameter variations?
6.14 . a. Consider the plant transfer function shown in Figure 6 .18 . Find a k in Figure
P6 .14, if it exists, such that the overall transfer function in Figure P6 .14
equals 2/(s + 2) .
b. If the plant has a disturbance as shown in Figure 6 .18, find the steady-state
output of the overall system in (a) due to a unit-step disturbance input .
c. Which feedback system, Figure 6 .18(c) or Figure P6 .14, is less sensitive to
plant perturbations? The loop transfer function in Figure 6 .18(c) is of type
1 . Is the loop transfer function in Figure P6 .14 of type 1?
r + y Y
1o
O 5s + 1
Figure P6.14

PROBLEMS 221
6.15 . Consider the system shown in Figure P6 .15. The noise generated by the am-
plifier is represented by n . If r = sin t and n = 0 .1 sin 10t, what are the steady-
state outputs due to r and n? What is the ratio of the amplitudes of the outputs
excited by r and n?
n
Amplifier
5 1
2 10
Figure P6 .15
6.16 . Consider the systems shown in Figure P6 .16 . (a) If the plant, denoted by P,
has the following nominal transfer function
1
s(s2 + 2s + 3)
show that the two systems have the same steady-state output due to r(t) = sin
0.1t. (b) If, due to aging, the plant transfer function becomes
1
s(s 2 + 2.1s + 3 .06)
what are the steady-state outputs of the two systems due to the same r?
(c) Which system has a steady-state output closer to the one in (a)?
3 +2s+3) 3(s 2 + 2s + 3)
P -
s3 + 3 .5S 2 +Ss+3 ff r + s 2 + 3 .5s + 5
(a) (b)
Figure P6 .16
6.17 . The comparison in Problem 6 .16 between the open-loop and closed-loop sys-
tems does not consider the noise due to the transducer (which is used to intro-
duce feedback) . Now the noise is modeled as shown in Figure P6 .17 .
a. Compute the steady-state y, due to n(t) = 0.1 sin lOt .
b. What is the steady-state y, due to r(t) = sin O .lt and n(t) = 0 .1 sin lOt?
c. Compare the steady-state error in the open-loop system in Figure P6.16(a)
with the one in the closed-loop system in Figure P6 .17 . Is the reduction in
the steady-state error due to the feedback large enough to offset the increase
of the steady-state error due to the noise of the transducer?
r + 3(s2 +2s+3) 1 v
9
sZ +3 .5s+5 s(s` + 2 . Is + 3 .06)
Figure P6 .17
6.18. a. Consider the feedback system shown in Figure P6 .18. The nominal values
of all k, are assumed to be 1 . What is its position error?
b. Compute the position error if k, = 2 and k2 = k3 = 1 . Compute the position
error if k2 = 2 and k, = k3 = 1 . Compute the position error if k3 = 2 and
k, = k2 = 1.
r +
50
A
Figure P6 .18

The Root-Locus
Method
7 .1 INTRODUCTION
As was discussed in the preceding chapter, inward and outward approaches are
available for designing control systems . There are two methods in the outward ap-
proach: the root-locus method and the frequency-domain method . In this chapter,
we study the root-locus method .
In the root-locus method, we first choose a configuration, usually the unity-
feedback configuration, and a gain, a compensator of degree 0 . We then search the
gain and hope that a good control system can be obtained . If not, we then choose a
different configuration and/or a more complex compensator and repeat the design .
Because the method can handle only one parameter at a time, the form of compen-
sators must be restricted . This is basically a trial-and-error method . We first use an
example to illustrate the basic idea.
7 .2QUADRATIC TRANSFER FUNCTIONS WITH A CONSTANT NUMERATOR
Consider a plant with transfer function
(~ ~)
G(s) s(s + 2)
223

224 CHAPTER 7 THE ROOT-LOCUS METHOD
It could be the transfer function of a motor driving a load . The problem is to design
an overall system to meet the following specifications :
1. Position error = 0
2. Overshoot < 5%
3. Settling time < 9 seconds
4. Rise time as small as possible .
Before carrying out the design, we must first choose a configuration and a
compensator with one open parameter. The simplest possible feedback configuration
and compensator are shown in Figure 7 .1 . They can be implemented using a pair of
potentiometers and an amplifier . The overall transfer function is
1
k
+ 2) _ k
G 0(s)
,(s) = (7.2)
1 s2 + 2s + k
1+k
s(s + 2)
The first requirement in the design is the stability of G0(s). Clearly, G0 (s) is stable
if and only if k > 0. Because G0(0) = k/k = 1, the system has zero position error
for every k > 0. Thus the design reduces to the search for a positive k to meet
requirements (2) through (4) . Arbitrarily, we choose k = 0.36 . Then G0 (s) becomes
0.36
G ,(s)
(s) 2
+ 2s + 0.36 (s + 0.2)(s + 1 .8) (73)
One way to find out whether or not G0(s) will meet (2) and (3) is to compute
analytically the unit-step response of (7.3). A simpler method is to carry out computer
simulation . If the system does not meet (2) or (3), then k = 0.36 is not acceptable .
If the system meets (2) and (3), then k = 0.36 is a possible candidate . We then
choose a different k and repeat the process . Finally, we choose from those k meeting
(2) and (3) the one that has the smallest rise time . This completes the design.
From the preceding discussion, we see that the design procedure is very tedious
and must rely heavily on computer simulation . The major difficulty arises from the
fact that the specifications are given in the time domain, whereas the design is carried
out using transfer functions, or in the s-plane . Therefore, if we can translate the time-
domain specifications into the s-domain, the design can be considerably simplified .
This is possible for a special class of transfer functions and will be discussed in the
next subsection .
r Y
I
s(s +2)
Figure 7 .1 Unity-feedback system .

7 .2 QUADRATIC TRANSFER FUNCTIONS WITH A CONSTANT NUMERATOR 225
7 .2.1 Desired Pole Region

Consider a control system with transfer function
(0 2
Go(s) (7 .4)
2 + 2C(o,,s + con
where C is the damping ratio and Wn the natural frequency . It is a quadratic transfer
function with a constant numerator . This system was studied in Section 4 .3. The
poles of G0 (s) are
-b ft)n &)"'\/6 2 - 1
If ~ < 1, the poles are complex conjugate as shown in Figure 7 .2(a) . For ~ < 1, the
unit-step response of G0(s), as derived in (4.10), is
t Wn 1
Y(t) _ ~- (,) 2
IS 2 + 2,Ycon s + S J
I - ~" e - " sin ((oa t + 0)
(
od
where cod = con(1 - .2)1 / 2, 0 = SWn, and 0 = cosThe steady-state response

of y(t) is y s = 1, and the maximum value, as computed in (4.13), is
- 1/2
= max ly(t)l = 1 + e--6'('
ymax
2)
Thus the overshoot, as defined in Section 6 .3 .3, is

e _ 7r~r(h - ~ 2)- U2
Overshoot =
Im s
0)d i
~
C = 1 o . 60
E X~rI > Re s
n 40
~> 1 Kv I ~=cosO 0
20
Pole
0 0.2 0.4 0.6 0.8 1 .0 1 .2
s-plane Damping ratio 4
(a) (b)
Figure 7 .2 Damping ratio and overshoot .

We see that the overshoot depends only on the damping ratio . The relationship is
plotted in Figure 7 .2(b). From the plot, the range of ~ for a given overshoot can be
obtained . For example, if the overshoot is required to be less than 20%, then the
damping ratio must be larger than 0 .45, as can be read from Figure 7.2(b). Now we
translate this relationship into a pole region . Because
~ = cos 0
where 6 is defined in Figure 7.2(a), and because cos 0 is a decreasing function of 0
in 0 < 0 < 90, if ~ ? ~0, then 0 < 00 = cos -' ~0. Using this fact, the specification
on overshoot can easily be translated into a pole region in the s-plane . For example,
we have
Overshoot :!s~ 10% - ~ ? 0.6 - 0 < cos - '0.6 = 53
and
Overshoot !!~ ; 5% - ~ ? 0.7 -> 0 < cos - '0.7 = 45 0
In other words, for the system in (7.4), if the overshoot is required to be less than
5%, then the poles of G0(s) must lie inside the sector bounded by 45, as is shown
in Figure 7 .3. This translates the specification on overshoots into a desired pole
region .
Next we consider the settling time . As defined in Section 6 .3 .3, the settling time
is the time needed for the step response to reach and stay within 2% of its steady-
state value . The difference between y(t) in (7 .5) and its steady state y s = I is
con e - Ut (on e _ f _ e at
D : _ I y(t) - I I = sin (coa t + 8) (7 .6)
UI - ~ 2
Wd Cod
for ~ < 1 . Note that a- = own is the magnitude of the real part of the poles (see
Figure 7 .2) ; it is the distance of the complex-conjugate poles from the imaginary
10 percent overshoot Im s
5 percent
overshoot
Re s
Figure 7 .3 Overshoot and pole region .

axis . Thus the settling time t, is the smallest t such that

e -Ut
0.02 for t > t, (7 .7)
~1- ~2
Clearly, for given ~ and w,,, the settling time can be computed from (7.7). It is,
however, desirable to develop a formula that is easier to employ. If ~ < 0 .8, then
(7.7) becomes
e - Ut
D < ~1 <_ 1 .7e - t
~a
This is smaller than 0 .02 if t ? 4.5/a-. Hence, given a settling time t, if 0' ? 4.5/t,
or, equivalently,
4.5
-(Real parts of the poles) (7 .8)
? t s
then the specification on settling time can be met . Although (7 .8) is developed for
complex poles with damping ratio smaller than 0 .8, it also holds for real poles with
~ > 1 .05 . Thus, in general, if both poles of G 0(s) in (7 .4) lie on the left-hand side
of the vertical line shown in Figure 7 .4, then the specification on settling time can
be met. The condition in (7.8) is consistent with the statement that the step response
reaches and remains within 1 % of its steady-state value in five time constants . The
settling time is defined for 2% and equals 4 .5 X time constant .
Now we can combine the specifications on overshoot and settling time . The
poles of (7.4) must lie inside the section as shown in Figure 7 .3 to meet the overshoot
specification and must lie on the left-hand side of the vertical line in Figure 7 .4 to
meet the settling time specification . Therefore, to meet both specifications, the poles
must lie inside the region denoted by C in Figure 7 .4. The exact boundary of C can
be obtained from the specifications on overshoot and settling time .
Im S
C 45
Re s
0
~I -
Overshoot J
Settling time
Figure 7 .4 Desired pole region .

Exercise 7 .2.1
Find the desired pole regions for the following specifications :

a. Overshoot < 20%, and settling time < 3 seconds .
b. Overshoot < 10%, and settling time < 10 seconds .
By definition, the rise time is the time required for the step response of G0(s) to
rise from 0 to 90 percent of its steady-state value . The translation of the rise time
into a pole region cannot be done quantitatively, as in the case of overshoot and
settling time . All we can say is that, generally, the farther away the closest pole s
from the origin of the s-plane, the smaller the rise time . Strictly speaking, this state-
ment is not correct, as can be seen from Figure 4 .7. The rise times of the responses
in Figure 4 .7 are all different, even though the distances of the corresponding com-
plex poles from the origin all equal w,, . On the other hand, because the time scale
of Figure 4.7 is wt, as the distance w increases, the rise time decreases . Thus, the
assertion holds for a fixed ~ . Because there is no other better guideline, the assertion
that the farther away the closest pole from the origin, the smaller the rise time will
be used in the design .
We recapitulate the preceding discussion as follows :
Overshoot Sector with 0 = cos ~ and ~ is determined from Fig. 7 .2(b) .
Settling time 4 .5/(shortest distance of poles from the jw-axis) .
Rise time Proportional to 1 /(shortest distance of poles from the origin) .
These simple rules, although not necessarily exact, are very convenient to use in
design .
7 .2 .2 Design Using Desired Pole Region

Now we return to the design problem in Figure 7 .1 . As discussed earlier, if k is
positive, then the system is stable and has zero position error . Hence, in the follow-
ing, we shall find a positive k to meet transient specifications. The specification on
overshoot requires that the poles of G0(s) lie inside the sector bounded by 0 = 45,
as shown in Figure 7 .5 . The settling time requires the poles to be located on the left-
'The system has two poles . If they are complex, the distances of the two complex-conjugate poles from
the origin are the same. If they are real and distinct, then one pole is closer to the origin then the other .
We consider only the distance to the closer pole . If a system has three or more poles, then we consider
the pole closest to the origin .

Im s
i
k=5 12-
\_ i
k=21
\.\ i I -
45'
k=0 .36 k= 1 :\ k=0
-3
I X)( )( J( x>< I >- Res
-2 0 1
k=0 k=0.75 1
\ k=0.36
~ I -1 -
T k=2 I
7 I
7 k=5 -2-
Figure 7 .5 Poles of G0 (s) .
hand side of the vertical line passing through -4 .5/t s = - 0.5 . Hence, if all poles
of G0 (s) lie inside the shaded region in Figure 7 .5, the overall system will meet the
specifications on overshoot and settling time .
The poles of G0(s) in (7 .2) for k = 0 .36, 0.75, 1, 2, and 5 are computed in the
following:
Gain Poles Comments

k = 0 .36 -0.2, -1.8 meet (2) but not (3)
k = 0.75 -0.5, -1.5 meet both (2) and (3)
k=1 -1, -1 meet both (2) and (3)
k = 2 -1 j meet both (2) and (3)
k = 5 - 1 j2 meet (3) but not (2)
They are plotted in Figure 7 .5. Note that there are two poles for each k. For k =
0.36, although one pole lies inside the region, the other is on the right-hand side of
the vertical line . Hence if we choose k = 0.36, the system will meet the specification
on overshoot but not that on settling time . If we choose k = 5, then the system will
meet the specification on settling time but not that on overshoot . However for k =
0 .75, 1, and 2, all the poles are within the allowable region, and the system meets

the specifications on overshoot and settling time . Now we discuss how to choose a
k from 0 .75, 1, and 2, so that the rise time will be the smallest . The poles corre-
sponding to k = 0.75 are -0 .5 and - 1 .5 ; therefore, the distance of the closer pole
from the origin is 0 .5 . The poles corresponding to k = 1 are -1 and - 1 . Their
distance from the origin is I and is larger than 0.5 . Therefore, the system with
k = 1 has a smaller rise time than the one with k = 0.75 . The poles corresponding
to k = 2 are -1 j I . Their distance from the origin is V 1 + 1 = 1 .4, which is
the largest among k = 0.75, 1, and 2 . Therefore the system with k = 2 has the
smallest rise time or, equivalently, responds fastest . The unit-step responses of the
system are shown in Figure 7 .6 . They bear out the preceding discussion .
For this example we are able to find a gain k to meet all the specifications . If
some of the specifications are more stringent, then no k may exist. For example, if
the settling time is required to be less than 2 seconds, then all poles of G0(s) must
lie on the left-hand side of the vertical line passing through - 4.5/2 = -2 .25. From
Figure 7 .5, we see that no poles meet the requirement . Therefore, no k in Figure 7 .1
can yield a system with settling time less than 2 seconds . In this case, we must
choose a different configuration and/or a more complicated compensator and repeat
the design .
Exercise 7.2.2
Consider a plant with transfer function 2/s(s + 4) . (a) Find the range of k in Figure
7 .1 such that the resulting system has overshoot less than 5% . (b) Find the range of
k such that the system has settling time smaller than 4 .5 seconds . (c) Find the range
of k to meet both (a) and (b). (d) Find a value of k from (c) such that the system has
the smallest rise time .
[Answers : (a) 0 < k < 4 . (b) 1 .5 < k < o . (c) 1 .5 < k < 4. (d) k = 4.]
1 .2
k=2
k=1
0 .8
/ k = 0 .75
0 .6
04
0 .2
0 1 2 3 5 6 7 9 10
Figure 7.6 Unit-step responses .

7 .3 MORE ON DESIRED POLE REGION 23 1
7 .3 MORE ON DESIRED POLE REGION
The example in the preceding sections illustrates the essential idea of the design
method to be introduced in this chapter . The method consists of two major
components :
1. The translation of the transient performance into a desired pole region . We then
try to place the poles of the overall system inside the region by choosing a
parameter .
2. In order to facilitate the choice of the parameter, the poles of the overall system
as a function of the parameter will be plotted graphically . The method of plotting
is called the root-locus method.
In this section we discuss further the desired pole region . The root-locus method is
discussed in the next section .
The desired pole region in Figure 7 .4 is developed from a quadratic transfer
function with a constant numerator . We shall check whether it is applicable to other
types of transfer functions . Consider
G 0(s) = 1 (7 .9)
s)
(s2 + 1 .2s + 1) (1 +
a
In addition to the complex-conjugate poles -0 .6 jO .8, G0(s) has one plot at - a.

If a = -, (7 .9) reduces to (7.4) and its unit-step response, as shown in Figure 7 .7,
has an overshoot of 10% and a settling time of about 7 seconds, as predicted by the
desired pole region developed in Figure 7 .4. Consider now a = 4 . The pole s =
- 4 is located quite far away from the pair of complex-conjugate poles . Because the
response e-4t of the pole approaches zero rapidly, the unit-step response of G(s)
with a = 4 is quite close to the one with a = -, as shown in Fig . 7 .7. Thus the
unit-step response of GO (s) is essentiaffy determined by the pair of complex-conju-
gate poles -0 .6 j0 .8 . In this case the pair is called the dominant poles and G 0(s)
Y (t)
a=- a=4 a=1 a = 0 .6 a = 0 .2
t
2 4 6
Figure 7 .7 Unit-step responses of (7 .9) .

essentially reduces to a quadratic transfer function with a constant numerator . Thus

the desired pole region in Figure 7 .4 can still be employed .
The unit-step responses of G0(s) for a = 1, 0.6, and 0 .2 are also shown in Figure
7 .7. We see that as the pole moves closer to the origin, the overshoot becomes smaller
(eventually becoming zero) and the settling time becomes larger . In these cases,
the quantitative specification for the overshoot in Figure 7.2 no longer holds . How-
ever, the specifications regarding the settling time and rise time appear to be still
applicable .
Next we examine the effect of introducing a zero by considering
S
1 + -
a
G0(s) _ (7.10)
s2 + 1 .2s + 1
Figure 7 .8(a) shows the unit-step responses of G 0(s) for a = 4, 1, 0 .6, and 0 .2, and
Figure 7 .8(b) shows the unit-step responses of G 0(s) for a = -4, - 1, - 0.6, and
-0.2. The responses for a = 4 and -4 are quite similar to the one for a = -. In
other words, if the zero is far away (either in the right half plane or in the left half
plane) from the complex conjugate poles, the concept of dominant poles is still
applicable . As the left-half-plane zero moves closer to the origin, the overshoot and
settling time become larger . However, the rise time becomes smaller. If the zero is
in the right half plane, or a < 0, the unit-step response will become negative and
then positive. This is called undershoot.2 For a = - 4, the undershoot is hardly
detectable . However as the right-half-plane zero moves closer to the origin, the
undershoot becomes larger. The overshoot, settling time, and rise time also become
larger. Thus, the quantitative specifications developed in Figure 7 .4 are no longer
applicable . This is not surprising, because the response of a system depends on its
poles and zeros, whereas zeros are not considered in the development of the desired
pole region .
Even for the simple systems in (7 .9) and (7.10), the relationships between the
specifications for the transient performance and the pole region are no longer as
precise as for quadratic transfer functions with a constant numerator . However, be-
cause there is no other simple design guideline, the desired pole region developed
in Figure 7 .4 will be used for all overall transfer functions . Therefore, if an overall
transfer function is not quadratic as in (7 .4) and cannot be approximated by (7 .4),
then there is no guarantee that the resulting system will meet the transient specifi-
cations by placing all poles inside the desired pole region . It is therefore important
to simulate the resulting system on a computer, to check whether or not it really
meets the specifications . If it does not, the system must be redesigned .
ZIt was shown by Norimatsu and Ito [49] that if G0 (s) has an odd number of open right-half-plane real
zeros, then undershoots always occur in step responses of G,(s) .
7 .4 PLOT OF ROOT LOCI 233
a=4 \
a = 0.6
I
a=~
t
2 4 6
(a)
(b)
Figure 7 .8 Unit-step responses of (7.10) .
7 .4 PLOT OF ROOT LOCI
From the example in Section 7 .2, we see that the design requires the computation
of the poles of G0(s) as a function of k. In this section, we shall discuss this problem .
Consider the unity-feedback system shown in Figure 7 .9, where G(s) is a proper
rational function and k is a real constant . Let G(s) = N(s)/D(s) . Then the overall

r + U Y
> k -* G (s)
Figure 7.9 Unity-feedback system .
transfer function is
N(s)
kG(s) _ k D(s) _ kN(s)
G0 (s)
1 + kG(s) N(s) D(s) + kN(s)
1 + k
D(s)
The poles of G0(s) are the zeros of the rational function
1 + kG(s) = 1 + k N(s) (7 .11)

D(s)
or the roots of the polynomial
D(s) + kN(s) (7 .12)
or the solutions of the equation

1 + kG(s) = 0 (7 .13)
The roots of D(s) + k/N(s) or the poles of G0(s) as a function of a real k are called
the root loci . Many software programs are available for computing root loci . For
example, for G(s) = 1/s(s + 2) = 1/(s 2 + 2s + 0), the following commands in
version 3 .1 of MATLAB
num=[1 ] ;den =[1 2 0] ;
k= 0 :0.5:10 ;
r= rlocus(num,den,k) ;
plot(r,'x')
will plot 21 sets of the poles of kG(s)/(1 + kG(s)) fork = 0, 0.5, 1, 1 .5, . . ., 9 .5,
and 10. If we use version 3 .5 or the Student Edition of MATLAB, the command
rlocus(num,den)
will plot the complete root loci on the screen . Therefore, to use an existing computer
program to compute root loci is very simple . Even so, it is useful to understand the
general properties of root loci . From the properties, we can often obtain a rough plot
of root loci even without any computation or measurement . This can then be used
to check the correctness of computer printout .
To simplify discussion, we assume
G(s) = q(s + z 1 )(s + z2) (7 .14)
(s + P1)(s + P2)(s + P3)

where -z, and -p, denote, respectively, zeros and poles, and q is a real constant,
positive or negative . Because G(s) is assumed to have real coefficients, complex-
conjugate poles and zeros must appear in pairs . Now we shall write 1 + kG(s) _
0 as
q(s + z 1 )(s + z2 ) 1
G(s) _ _ - k (7 .15)
(s + p1)(s + p2)(s + p3)
Then the roots of D(s) + kN(s) are those s, real or complex, which satisfy (7 .15)
for some real k. Note that for each s, say s 1 , each factor on the left-hand side of
(7:15) is a vector emitting from a pole or zero to s 1 , as shown in Figure 7 .10. The
magnitude I I is the length of the vector . The phase < is the angle measured from
the direction of positive real axis ; it is positive if measured counterclockwise, nega-
tive if measured clockwise . The substitution of
s l + z; = Is, + z,l ei'I s s +z,> = Is, + z,l eje
and
S1 + p, = Is, + p,l ei<(s ' + P;) = Is, + p,l ei0i
into (7 .15) yields
Iql Is, + z1lei(4 +e,+e2)
Is1 + z2l
1
i,+02+03> (7.16)
Is, + P1l ISI + P21 is, + P31 e k
This equation actually consists of two parts : the magnitude condition
IqI Is, + zil ISI + z2I 1
(7.17)
Is1 + PIl ISI + P21 ISI + P31 k
IM S
Re s
Figure 7.10 Vectors in s-plane .

and the phase condition

,~, 1
< q + 0 1 + 02 - ( 0 1 + `f'2 + 03) _ - (7.18)
k
Note that < q equals 0 if q > 0, 7r or - 7r if q < 0, 0; and 4; can be positive (if
measured counterclockwise) or negative (if measured clockwise) . In the remainder
of this and the next sections, we discuss only the phase condition . The magnitude
condition will not arise until Section 7 .4.3 .
Because k is real, we have
+ it + 37T + 5 7r, . . . if k > 0
k 0, 21T, 41r, . . . if k < 0
Two angles will be considered the same if they differ by 21T radians or 360 or
their multiples. Using this convention, the phase condition in (7.18) becomes
, ~, ~rr ifk>0
To tal phase q + 01 + 02 - (4 + 4'2 a'3)to .19)
0 if k < 0 (7
+ =
We see that the constant k does not appear explicitly in (7.19). Thus the search
for the root loci becomes the search for all s 1 at which the total phase of G(s 1) equals
0 or in If s1 satisfies (7 .19), then there exists a real k 1 such that D(s 1 ) + k,N(s 1 ) _
0. This k1 can be computed from (7.17).
We recapitulate the preceding discussion in the following . The poles of G0(s)
or, equivalently, the roots of D(s) + kN(s) for some real k are those s 1 such that the
total phase of G(s 1) equals 0 or IT. The way to search for those s 1 is as follows . First
we choose an arbitrary s, and draw vectors from the poles and zeros of G(s) to s 1
as shown in Figure 7 .10 . We then use a protractor to measure the phase of each
vector . If the total phase is 0 or 7r, then s 1 is a point on the root loci . If the total
phase is neither 0 nor IT, then s 1 is not on the root loci . We then try a different point
and repeat the process . This is a trial-and-error method and appears to be hopelessly
complicated . However, using the properties to be discussed in the next subsection,
we can often obtain a rough sketch of root loci without any measurement .
7.4.1 Properties of Root Loci-Phase Condition

Consider a transfer function with real coefficients expressed as
+ z1)(s + z2) . . . (s + z,y,) _ , N(s)
G(s) : - q(s (7 .20)
(s + P1)(s + P2) . . . (s + P) D(s)
This step can be achieved easily by using
tf2zp
in MATLAB . The transfer function has n poles and m zeros, with n ? m. In dis-
cussing the root loci, it is important to express every term in the form of (s + a) .
Note that (s + a) is a vector extending from - a to s . If we use the form I + rs,
then its meaning is not apparent . The root loci will be developed graphically on the

complex s-plane . We first plot all poles and zeros on the s-plane and then measure
the angle from every pole and zero to a chosen s . Note that the constant q does not
appear on the plot, but it still contributes a phase to (7.19) . To simplify discussion,
we assume in this section that q > 0. Then the phase of q is zero and the total phase
of G(s) will be contributed by the poles and zeros only . We discuss now the general
properties of the roots of the polynomial
D(s) + kN(s) (7 .21)
or the zeros of the rational function
N(s)
1 + k = 1 + kG(s)
D(s)
or the solutions of the equation
N(s) q(s + zl)(s + z2) . (s +Z m) - 1
-k (7 .22)
D(s) (s + Pi)(s + P 2) . . . (s + p)
as a function of real k. To simplify discussion, we consider only k ? 0. In this case,
the root loci consist of all s at which G(s) has a total phase of ar radians or 180.
This is the phase condition in (7 .19); the magnitude condition in (7 .17) will not be
used in this section .
PROPERTY 1
The root loci consist of n continuous trajectories as k varies continuously from 0
to cc . The trajectories are symmetric with respect to the real axis .
The polynomial in (7 .21) has degree n. Thus for each real k, there are n roots.
Because the roots of a polynomial are continuous functions of its coefficients, the n
roots form n continuous trajectories as k varies from 0 to oc . Because the coefficients
of G(s) are real by assumption, complex-conjugate roots must appear in pairs . There-
fore the trajectories are symmetric with respect to the real axis .
PROPERTY 2
Every section of the real axis with an odd number of real poles and zeros
(counting together) on its right side is a part of the root loci for k ? 0. 3
If k ? 0, the root loci consist of those s with total phases equal to 180 . Recall
that we have assumed q > 0, thus the total phase of G(s) is contributed by poles
and zeros only. We use examples to establish this property . Consider
(s + 4)
Gt(s) = (7.23)
(s - 1)(s + 2)
3 More generally, if q > 0 and k > 0 or q < 0 and k < 0, then every section of the real axis whose right-
hand side has an odd number of real poles and real zeros is part of the root loci . If q < 0 and k > 0 or
q > 0 and k < 0, then every section of the real axis whose right-hand side has an even number of real
poles and real zeros is part of the root loci .

Their poles and zeros are plotted in Figure 7 .11(a) . If we choose s 1 = 2 .5 in Figure
7.11(a) and draw vectors from poles 1 and -2 to s 1 and from zero -4 to s1, then
the phase of every vector is zero . Therefore, the total phase of G 1(s 1) is zero. Thus
s, - 2.5 is not a zero of 1 + kG1(s) = 0 for any positive real k. If we choose
s2 = 0 and draw vectors as shown in Figure 7 .11(a), then the total phase is
0 - 0 - 7r = - 7r
which equals 7r after the addition of 27T. Thus s2 = 0 is on the root loci . In fact,
every point in [ - 2, 1 ] has a total phase of ar, thus the entire section between [ - 2,
1 ] is part of the root loci . The total phase of every point between
[-4, -2] can be shown to be 2rr, therefore the section is not on the root loci . The
total phase of every point in (, -4] is -rr, thus it is part of the root loci . The two
sections (-, -4] and [-2, 1] have odd numbers of real poles and zeros on their
right-hand sides.
The transfer function in (7 .23) has only real poles and zeros . Now we consider
2(s + 2)
G2(s) = (s + (7 .24)
3)2(s + 1 + j4)(s + 1 - j4)
which has a pair of complex-conjugate poles . The net phase due to the pair to any
point on the real axis equals 0 or 2rr as shown in Figure 7 .11(b) . Therefore, in
applying property 2, complex-conjugate poles and zeros can be disregarded . Thus
for k > 0, the sections (--, -3] and [-3, -2] are part of the root loci .
Exercise 7 .4.1

s + 4
G3(S) = (7.25)
(s - 1)(s + 1) 2
Find the root loci on the real axis for k ? 0.
Exercise 7 .4.2

(s+2+j2)(s+2-j2)
G4(s)
4(S) (7.26)
(s - 1)(s + 2)(s + 3)
Find the root loci on the real axis for positive k.
PROPERTY 3
The n trajectories migrate from the poles of G(s) to the zeros of G(s) as k
increases from 0 to oc .

(a) (b)
Figure 7 .11 Root loci on real axis.
The roots of (7 .21) are simply the roots of D(s) if k = 0. The roots of D(s)
+ kN(s) = 0 are the same as the roots of
D(s)
+ N(s) = 0
k
Thus its roots approach those of N(s) as k ---> - . Therefore, as k increases from 0 to
cc, the root loci exit from the poles of G(s) and enter the zeros of G(s) . There is one
problem, however. The number of poles and the number of zeros may not be the
same. If n (the number of poles) > m (the number of zeros), then m trajectories will
enter the m zeros . The remaining (n - m) trajectories will approach (n - m) asymp-
totes, as will be discussed in the next property .
PROPERTY 4
For large s, the root loci will approach (n - m) number of straight lines, called
asymptotes, emitting from
I Poles - I Zeros
(7 .27a)
(No. of poles - No. of zeros' 0
called the centroid.4 These (n - m) asymptotes have angles
Ir 3 1r 5Tr
(7 .27b)
n - m n - m n - m
4If G(s) has no zeros, then the centroid equals the center of gravity of all poles .

These formulas will give only (n - m) distinct angles. We list some of the
angles in the following table .
n - m Angles of asymptotes
1 180
2 901
3 60, 180-
4 45, 135
5 36, 108, 180

We justify the property by using the pole-zero pattern shown in Figure 7 .12(a).
For s l very large, the poles and zeros can be considered to cluster at the same point-
say, a-as shown in Figure 7 .12(b) . Note the units of the scales in Figure 7 .12(a)
and (b). Consequently the transfer function in (7 .22) can be approximated by
q(s + zl ) . . . (s + zm) q for s very large (7.28)
(S + p i ) . . . (S + pn) (S - a)n-m
In other words, all m zeros are canceled by poles, and only (n - m) poles are left
at a. Now we compute the relationship among zi, pi, and a. After canceling q, we
turn (7 .28) upside down and then expand it as
(s+pl) . . .(s+pn)= S n +(y pi)Sn-1 + . . . y-m (S - a
(S + Zl) . . . (S + Zm) s m + (I'Zi ) S m-1 + . . .
Im s Im s
A
(a) (b)
Figure 7.12 Asymptotes .

which implies, by direct division and expansion,

sn-n + [(Ipi) - ( YZi)]Sn-m-1 + . . . - sn-m - (n - m)as"-m-1 + . . .
Equating the coefficients of sn -m-t yields
(n - m)a = - [(1pi) - (Y zi)] = Y,(-p i ) - (I - z, )
or
1(-pt) - (Y- - zi ) (I Poles) - (I zeros)
a = _
n - m (No. of poles) - (No. of zeros)
This establishes (7 .27a) .
With all (n - m) number of real poles located at a, it becomes simple to find
all s, with a total phase of IT, or, more generally, 7r, 31r, 5 7r, . . . . Thus
each pole must contribute err/(n - m), 31r/(n - m), 51r/(n - m) This
establishes (7 .27b) . We mention that (n - m) asymptotes divide 360 equally and
are symmetric with respect to the real axis .
Now we shall use this property to find the asymptotes for G,(s) in (7 .23) and
G2(s) in (7.24). The difference between the numbers of poles and zeros of G,(s) is
1 ; therefore, there is only one asymptote in the root loci of G,(s) . Its degree is
7r/1 = 180 ; it coincides with the negative real axis . In this case, it is unnecessary
to compute the centroid . For the transfer function G 2(s) in (7.24), the difference
between the numbers of poles and zeros is 3 ; therefore, there are three asymptotes
in the root loci of G 2(s). Using (7 .27a), the centroid is
-3-3- 1 - j4 - 1 +j4-(-2) - -6
=-_ -2
3 3
Thus the three asymptotes emit from (- 2, 0). Their angles are 60 and 180 . Note
that the asymptotes are developed for large s, thus the root loci will approach them
for large s or large k.
Now we shall combine Properties (3) and (4) as follows : If G(s) has n poles
and m zeros, as k increases from 0 to -, n trajectories will emit from the n poles .
Among the n trajectories, m of them will approach the m zeros ; the remaining
(n - m) trajectories will approach the (n - m) asymptotes. 5
Exercise 7 .4.3
Find the centroids and asymptotes for G3(s) in (7.25) and G4 (s) in (7.26).
[Answers : (1.5, 0), 90; no need to compute centroid, 180 .]
5 The G(s) in (7 .20) can be, for s very large, approximated by q/s" - ' . Because it equals zero at s = -,
G(s) can be considered to have n - m number of zeros at s = - . These zeros are located at the end of
the (n - m) asymptotes. If these infinite zeros are included, then the number of zeros equals the number
of poles, and the n trajectories will emit from the n poles and approach the n finite and infinite zeros .

PROPERTY 5
Breakaway points-Solutions of D(s)N'(s) - D'(s)N(s) = 0.
Consider the transfer function G(s) = N(s)/D(s) = (s + 4)/(s - 1)(s + 2) .

Part of the root loci of 1 + kG(s) is shown in Figure 7.11(a), repeated in Figure
7.13 . Ask increases, the two roots of D(s) + kN(s) move away from poles 1 and
-2 and move toward each other inside the section [-2, 1] . As k continues to
increase, the two roots will eventually collide and split or break away . Such a point
is called a breakaway point. Similarly, as k approaches infinity, one root will ap-
proach zero - 4 and another will approach - oo along the asymptote that coincides
with the negative real axis . Because the root loci are continuous, the two roots must
come in or break in somewhere in the section (--, - 4] as shown in Figure 7 .13 .
Such a point is also called a breakaway point. Breakaway points can be computed
analytically .
A breakaway point is where two roots collide and break away ; therefore, there
are at least two roots at every breakaway point . Let s o be a breakaway point of
D(s) + kN(s) . Then it is a repeated root of D(s) + kN(s) . Consequently, we have
D(so) + kN(so) = 0 (7.29a)
and
[D(s) + kN(s)] = D'(so) + kN'(so) = 0 (7.29b)

ds S=S'
where the prime denotes differentiation with respect to s. The elimination of k from
(7.29) yields
D' (so)
D(so) - N(so) = 0
N'(so)
Im S
S2
sI
i 90
Res
-6 -4 -2 A
-0.8
Figure 7 .13 Root loci of G,(s).

which implies
D(so)N'(so) - D'(so)N(so) = 0 (7 .30)
Thus a breakaway point so must satisfy (7 .30) and can be obtained by solving the
equation . For example, if G(s) = (s + 4)/(s - 1)(s + 2), then
D(s) =s 2 +s-2 D'(s) = 2s + 1
N(s) = s + 4 N' (s) = 1
and
D(s)N'(s) - D'(s)N(s) = s2 + s - 2 -(2s 2 + 9s + 4) = 0 (7 .31)
or
s 2 +Ss+6=0
Its roots are - 0.8 and -7.2 . Thus the root loci have two breakaway points at A =
- 0.8 and B = - 7 .2 as shown in Figure 7 .13 . For this example, the two solutions
yield two breakaway points . In general, not every solution of (7 .30) is necessarily a
breakaway point for k ? 0. Although breakaway points occur mostly on the real
axis, they may appear elsewhere, as shown in Figure 7 .14(a). If two loci break away
from a breakaway point as shown in Figure 7 .13 and Figure 7 .14(a), then their
tangents will be 180 apart. If four loci break away from a breakaway point (it has
four repeated roots) as shown in Figure 7 .14(b), then their tangents will equally
divide 360.
With the preceding properties, we are ready to complete the root loci in Figure
7.13 or, equivalently, the solutions of
s + 4 1
(s - 1)(s + 2) k
Ims
Breakaway
point
(a) (b)
Figure 7 .14 Breakaway points.

for k ? 0 . As discussed earlier, the sections (- -, -4] and [-2, 1] are parts of the
root loci . There is one asymptote that coincides with the negative real part . There
are two breakaway points as shown. Because the root loci are continuous, the root
loci must assume the form indicated by the dotted line shown in Figure 7 .13 . The
exact loci, however, must be obtained by measurement . Arbitrarily we choose an s 1
and draw vectors from zero - 4 and poles - 2 and 1 to s 1 as shown in Figure 7 .13 .
The phase of each vector is measured using a protractor. The total phase is
90 - 135 - 158 = -203
which is different from 180 . Thus s 1 is not on the root loci . We then try s2, and
the total phase is measured as -190 . It is not on the root loci . We then try s 3, and
the total phase roughly equals -180 . Thus s3 is on the root loci . From the fact that
they break away at point A, pass through s3 , and come in at point B, we can obtain
the root loci as shown . Clearly the more points we find on the root loci, the more
accurate the plot. The root loci in Figure 7 .13 happens to be a circle with radius 3 .2
and centered at -4 . This completes the plot of the root loci of G1 (s) in (7.23).
Exercise 7 .4.4
Find the breakaway points for G3(s) in (7 .25) and G4(s) in (7 .26). Also complete the
root loci of G3(S)-
PROPERTY 6
Angle of departure or arrival .
Every trajectory will depart from a pole . If the pole is real and distinct, the
direction of the departure is usually 0 or 180. If the pole is complex, then the
direction of the departure may assume any degree between 0 and 360 . Fortunately
this angle can be measured in one step . Similarly the angle for a trajectory to arrive
at a zero can also be measured in one step . We now discuss their measurement .
Consider the transfer function G 2(s) in (7.24). Its partial root loci are obtained
in Figure 7 .11(b) and repeated in Figure 7 .15(a) . There are four poles, so there are
four trajectories . One departs from the pole at - 3 and enters the zero at - 2. One
departs from another pole at - 3 and moves along the asymptote on the negative
real axis . The last two trajectories will depart from the complex-conjugate poles and
move toward the asymptotes with angles 60 . To find the angle of departure, we
draw a small circle around pole -1 + j4 as shown in Figure 7 .15(b) . We then find
a point s 1 on the circle with a total phase equal to in Let s 1 be an arbitrary point on
the circle and let the phase from pole -1 + j4 to s 1 be denoted by 01. If the radius
of the circle is very small, then the vectors drawn from the zero and all other poles
to s 1 are the same as those drawn to the pole at - 1 + j4 . Their angles can be
measured, using a protractor, as 76, 63 , and 90 . Therefore, the total phase of G 2(s)
40
X
63 76
Re s
(a) (b)
Figure 7 .15 Root loci of G 2(s).
at s, is
76 - (63 + 63 + 90 + 9,) = - 140 - 0,
Note that there are two poles at - 3, therefore there are two 63 in the phase equation .
In order for s, to be on the root loci, the total phase must be 180 . Thus we have
01 = 40 . This is the angle of departure .
Once we have the asymptote and the angle of departure, we can draw a rough
trajectory as shown in Figure 7 .15. Certainly, if we find a point, say A = j5 shown
in the figure, with total phase 180, then the plot will be more accurate . In conclusion,
using the properties discussed in this section, we can often obtain a rough sketch of
root loci with a minimum amount of measurement .
Exercise 7 .4.5
Compute the angle of arrival for G 4(s) in (7 .26) and then complete its root loci .

a+jf3 a+j/3> a+j/3

a+ V30 a<a+~_3f3 a>a+~3p
c x < X c > < x
a-j a-jP~\ a-j/3 I

(a) (b) (c)
Figure 7 .16 Root loci of (7.32) .
7 .4 .2 Complexities of Root Loci
14 Using the properties discussed in the preceding subsection, we can often obtain a
rough sketch of root loci . However, because the roots of polynomials are very sen-
sitive to coefficients, small changes in coefficients may yield entirely different root
loci . For example, consider the transfer function
1
G(s) = - a - j/3)(s - a + jf3)(s- a) 732
(s
It has three poles . If a = a + V"3-f3 or, equivalently, if the three poles form an
equilateral triangle, then the root loci are all straight lines as shown in Figure 7 .16(a).
If a < a + \f3, then the root loci are as shown in Figure 7 .16(b) . If a >
a + ~/3, then the root loci have two breakaway points as shown in Figure 7 .16(c).
Although the relative positions of the three poles are the same for the three cases,
their root loci have entirely different patterns .
As an another example, consider
s + 1
G(s) = (7 .33)
S 2(s + a)
Its approximate root loci for a = 3, 7, 9, and 11 are shown in Figure 7 .17 . It has
two asymptotes with degrees 90 0, emitting respectively from
0+0+(-a)-(-1)-1-a
-1,-3,-4,-5
3 - 1 2
As a moves away from the origin, the pattern of root loci changes drastically . There-
fore to obtain exact root loci from the properties is not necessarily simple . On the
other hand, none of the properties is violated in these plots . Therefore the properties
can be used to check the correctness of root loci by a computer .
'This example was provided by Dr . Byunghak Seo.

X
-11
Figure 7.17 Root loci of (7.33) .
7 .4 .3 Stability Range from Root Loci-Magnitude Condition

The plot of root loci up to this point used only the phase condition in (7.19). Now
we discuss the use of the magnitude equation in (7 .17). We use it to find the stability
range of systems . Consider the unity-feedback system shown in Figure 7 .18, where
the plant transfer function is given by
N(s) _ s2 -2s+5
G (s)
D(s) s3 + 5s2 + 12s - 18 (7.34)
(s - 1 + j2)(s - 1 - j2)
(s - 1)(s + 3 + j3)(s + 3 - j3)
The system was studied in (4.22) and the stability range of k was computed, using
the Routh test, as
3 .6 < k < 5 .54
Now we shall recompute it using the root-locus method . First we plot the root loci
of
(s - 1 + j2)(s - 1 - j2) _ -1
(7.35)
(s - 1)(s + 3 + j3)(s + 3 - j3) k
for k > 0. The section (--, 1], plotted with the heavy line in Figure 7 .19, is part
of the root loci because its right-hand side has one real pole . The difference between
the numbers of poles and zeros is 1 ; therefore, there is one asymptote with degree
180, which coincides with the negative real axis . The angle of departure at pole
-3 + j3 is measured as 242 ; the angle of arrival at zero 1 + j2 is measured as
y
--4 G (s)

Im s
A
242
x 3-
2-
k2\ -143
-3.7 -0.27 1 kt
I
I I ' I II 1 A / X I I I - Re s
-6 -5 -4 -2 -1 ` O 1 2 3 4
~1
\
-2- 0
--3
Figure 7 .19 Stability range from root loci.
-143. We compute
D(s)N'(s) - D'(s)N(s) = (s3 + 5s 2 + 12s - 18)(2s - 2)
-(3S 2 + 10s + 12)(s 2 - 2s + 5) = _S4 + 4s3 + 7s 2 - 86s - 24
Its roots are computed, using MATLAB, as 3 .9870 j2 .789, - 3.7, and - 0.274 .
Clearly, - 3.7 and - 0.274 are breakaway points, but not the complex-conjugate
roots . Using the breakaway points and departure angles, we can readily obtain a
rough sketch of root loci as shown in Figure 7.19 with heavy lines . As k increases
from 0 to -, the three roots of D(s) + kN(s) or, equivalently, the three poles of the
unity-feedback system in Figure 7.18 will move along the trajectories as indicated
by arrows .
The rates of emigration of the three closed-loop poles are not necessarily the
same . To see this, we list the poles for k = 0, 1 and 2:
k=0: 1,-3j3
k = 1 : 0 .83,-3 .4j2
k = 2 : 0 .62,-5 .1,-2 .9
We see that, at k = 1, the complex-conjugate poles have moved quite far away from
-3 j3, but the real pole is still very close to 1 . As k continues to increase, the
complex-conjugate poles collide at s = - 3.7, and then one moves to the left and
the other to the right on the real axis . The pole moving to the left approaches --
as k approaches infinity . The one moving to the right collides with the real pole
emitting from 1 at s = - 0.274 . They split and then enter the complex-conjugate
zeros of G(s) with angles 143 . They cross the imaginary axis roughly at s =
jl .

7.4 PLOT OF ROOT LOCI 249
From the preceding discussion, we can now determine the stability range of k.
At k = 0, the unity-feedback system has one unstable pole at s = 1 and one pair
of stable complex-conjugate poles at - 3 j3 . As k increases, the unstable closed-
loop pole moves from 1 into the left half plane . It is on the imaginary axis at
k = k1 , and then becomes stable fork > k l. Note that the complex-conjugate closed-
loop poles remain inside the open left half plane and are stable as k increases from
0 to k 1. Therefore, the unity-feedback system is stable if k > k1 . The three closed-
loop poles remain inside the open left half plane until k = k 2, where the root loci
intersect with the imaginary axis . Then the closed-loop complex-conjugate poles
move into the right half plane and become unstable . Therefore, the unity-feedback
system is stable if k 1 < k < k2.
To compute k 1 and k2, we must use the magnitude equation . The magnitude of
(7.35) is
i s- 1 + j21 Is -
1 -j21 -1 1
(7 .36)
is fIIs+3+j3IIs+3-131 = k k
where we have used the fact that k > 0 . Note that k1 is the gain of the root loci at
s = 0 and k2 the gain at s = j 1 . To compute k1, we sets = 0 in (7 .36) and compute
1 I - 1+ j21 - 1- 12l V V 1
(7.37)
kl 13 + j31 13 - j31 \ V 3 .6
which implies k1 = 3.6 . This step can also be carried out by measurement . We draw
vectors from all the poles and zeros to s = 0 and then measure their magnitudes .
Certainly, excluding possible measurement errors, the result should be the same as
(7 .37) . To compute k2, we draw vectors from all the poles and zeros to s = j 1 and
measure their magnitudes to yield
1 .4 X 3 .2 1
1 .4 x 3.6 X 5 k2
which implies
k2 = 5.6
Thus we conclude that the overall system is stable in the range
3.6 = k1 < k < k2 = 5.6
This result is the same as the one obtained by using the Routh test .
Exercise 7 .4.6
Consider the G2(s) in (7 .24) with its root loci plotted in Figure 7 .15 . Find the range
of positive k in which the system is stable .
[Answer : 0 < k < 38 .]

7 .5 DESIGN USING THE ROOT LOCUS METHOD
In this section we discuss the design using the root-locus method . We use the ex-
ample in (7 .1) to develop a design procedure . The procedure, however, is applicable
to the general case . It consists of the following steps :
Step 1 : Choose a configuration and a compensator with one open parameter k such
as the one in Figure 7 .1 .
Step 2 : Compute the overall transfer function and then find the range of k for the
system to be stable and to meet steady-state specifications . If no such k
exists, go back to Step 1 .
Step 3: Plot root loci that yield the poles of the overall system as a function of the
parameter.
Step 4: Find the desired pole region from the' specifications on overshoot and set-
tling time as shown in Figure 7 .4 .
Step 5: Find the range of k in which the root loci lie inside the desired pole region .
If no such k exists, go to Step 1 and choose a more complicated compen-
sator or a different configuration .
Step 6 : Find the range of k that meets 2 and 5 . If no such k exists, go to Step 1 .
Step 7: From the range of k in Step 6, find a k to meet the remaining specifications,
such as the rise time or the constraint on the actuating signal . This step
may require computer simulation of the system .
We remark that in Step 2, the check of stability may be skipped because the stability
of the system is automatically met in Step 5 when all poles lie inside the desired
pole region . Therefore, in Step 2, we may simply find the range of k to meet the
specifications on steady-state performance .
Example 7 .5.1
We use an example to illustrate the design procedure . Consider a plant with transfer
function
s + 4
G(s) = (7.38)
(s + 2)(s - 1)
This plant has two poles and one zero . Design an overall system to meet the following
specifications :
1. Position error < 10%
2. Overshoot < 5%
3. Settling time < 4 .5 seconds
Step 1 : We try the unity-feedback configuration shown in Figure 7 .20.

7 .5 DESIGN USING THE ROOT-LOCUS METHOD 251
r + V
)'0 s+4
(s +2)(s- 1
Step 2 : The overall transfer function is

s + 4
k
(s + 2)(s - 1) _ k(s + 4)
G0(s) = (7 .39)
s+4 s 2 +(k+ 1)s+4k-2
l+k
(s + 2)(s - 1)
The conditions for G0(s) to be stable are 4k - 2 > 0 and k + 1 > 0,
which imply
2
k > - = 0.5
4
Thus the system is stable for k > 0 .5. Next we find the range of k to have
position error less than 10% . The specification requires, using (6.3),
4k - 2 - 4k -2 1
ep(t) = 0.1 (7.40)
4k - 2 4k - 2 2k - 1
where we have used the fact that k > 0.5, otherwise the absolute value
sign cannot be removed . The inequality in (7.40) implies
10<2k- 1
or
11
k >_ = 5.5 (7 .41)
2
Thus, if k ? 5 .5, then the system in Figure 7.20 is stable and meets spec-
ification (1) . The larger k is, the smaller the position error .
Steps 3 and 4: Using the procedure in Section 7.4.1, we plot the root loci of
1 + kG(s) = 0 in Figure 7.21 . For convenience of discussion, the poles
corresponding to k = 0.5, 0.7, 1, 5, . . . are also indicated . They are actually
obtained by using MATLAB. Note that for each k, there are two poles, but
only one is indicated . The specification on overshoot requires all poles to
lie inside the sector bounded by 45 . The specification on settling time
requires all poles to lie on the left-hand side of the vertical line passing
through -4 .5/t s = - 1 . The sector and the vertical line are also plotted
in Figure 7.21 .
Step 5 : Now we shall find the ranges of k to meet the specifications on overshoot
and settling time. From Figure 7.21, we see that if 0.5 < k < 1, the two
Im s
Re s
S
Figure 7 .21 Root loci of (7 .38) .
poles lie inside the sector bounded by 45 . If 1 < k < 5, the two poles
move outside the sector . They again move inside the sector fork > 5 . Thus
if 0 .5 < k < 1 or 5 < k, the overall system meets the specification on
overshoot . If k < 1, although one pole of G0(s) is on the left-hand side of
the vertical line passing through - 1, one pole is on the right-hand side .
If k > 1, then both poles are on the left-hand side . Thus if k > 1, the
system meets the specification on settling time .
Step 6 : The preceding discussion is summarized in the following :
k > 0 .5: stable
k > 5.5: meets specification (1) . The larger k is, the smaller the
position error .
k > 5 or 1 > k > 0.5: meets specification (2)
k > 1 : meets specification (3) .
Clearly in order to meet (1), (2), and (3), k must be larger than 5 .5.
Step 7: The last step of the design is to find a k in k > 5.5 such that the system
has the smallest rise time . To achieve this, we choose a k such that the
closest pole is farthest away from the origin . From the plot we see that as
k increases, the two complex-conjugate poles of G0(s) move away from
the origin . At k = 13 .3, the two complex poles become repeated poles at
s = - 7.2. At k = 15, the poles are -10 .4 and - 6 .4; one pole moves
away from the origin, but the other moves closer to the origin . Thus, at
k = 13.3, the poles of G0(s) are farthest away from the origin and the
system has the smallest rise time . This completes the design.

7 .5 DESIGN USING THE ROOT-LOCUS METHOD

253
It is important to stress once again that the desired pole region in Figure 7 .4 is
developed for quadratic transfer functions with a constant numerator. The G0(s) in
(7 .39) is not such a transfer function . Therefore, it is advisable to simulate the re-
sulting system. Figure 7 .22 shows the unit-step responses of the system in (7 .39) for
k = 13.3 (dashed line) and k = 5 .5 (solid line) . The system with k = 13.3 is better
than the one with k = 5 .5. Its position error, settling time, and overshoot are roughly
4%, 1 .5 seconds, and 10% . The system meets the specifications on position error
and settling time, but not on overshoot . This system will be acceptable if the re-
quirement on overshoot can be relaxed . Otherwise, we must redesign the system .
The root loci in Figure 7 .21 are obtained by using a personal computer ; there-
fore, the gain k is also available on the plot . If the root loci are obtained by hand,
then the value of k is not available on the plot . In this case, we must use the magnitude
equation
(s + 4) -1
(s + 2)(s - 1) k
to compute k. For example, to find the value of kl shown in Figure 7 .21, we draw
vectors from all poles and zeros to s 1 and then measure their magnitudes to yield
(s + 4) _ 3 .2 _ -1
(s + 2)(s - 1) S=Sj
3 .2 x 5 k,
which implies k1 = 5. To compute k2, we draw vectors from all poles and zeros to
s2 and measure their magnitudes to yield
(s + 4) 3 .2 -1
(s + 2)(s - 1) S=S2 5.2 x 8.2 k2
1 .4
1 .2
0 .8
06
0 .4
0 .2
00 1 2 3 4 5 6 7 9 10

which implies k2 = 13 .3. Thus, the gain can be obtained from the magnitude
equation .
7.5.1 Discussion
1. Although we studied only the unity-feedback configuration in the preceding
section, the root-locus method is actually applicable to any configuration as long
as its overall transfer function can be expressed as
k)
G0 (s) = N(s, (7 .42)
p(s) + kq(s)
where p(s) and q(s) are polynomials, independent of k, and k is a real parameter
to be adjusted . Since the root-locus method is concerned only with the poles of
G0(s), we plot the roots of
p(s) + kq(s) (7 .43a)
or the solutions of
q(s) 1
(7 .43b)
p(s) k
as a function of real k. We see that (7.43a) and (7.43b) are the same as (7 .12)
and (7.13), thus all discussion in the preceding sections is directly applicable to
(7.42). For example, consider the system shown in Figure 7 .23 . Its overall trans-
fer function is
s+k2 10
_ k1 + 2 s(s2 + 2s + 2)
G ,(s)
(s)
s + k2 10
1 + k1
s + 2 s(s2 +2s+2)
10k1(s + k2)
As + 2)(s2 + 2s + 2) + 10k 1(s + k2)
It has two parameters, k1 and k2. If we use a digital computer to plot the root
loci, it makes no difference whether the equation has one, two, or more param-
eters. Once the root loci are obtained, the design procedure is identical to the
one discussed in the preceding sections . If the root loci are to be plotted by
hand, we are able to handle only one parameter at a time . Arbitrarily, we choose
r + s + k2 u 10 Y
ki
(s +2) s(s 2 +2s+2)
Figure 7 .23 System with two parameters .

7 .6 PROPORTIONAL-DERIVATIVE (PD) CONTROLLER 255
k i = 5 . Then G0 (s) becomes

50(s + k2)
GG(s) =
[s(s + 2)(s 2 + 2s + 2) + 50s] + k2 50
This is in the form of (7.42) . Thus the root-locus method is applicable . In this
case, the root loci are a function of k2.
2. The root-locus method considers only the poles . The zeros are not considered,
as can be seen from (7.42) . Thus the method is essentially a pole-placement
problem . The poles, however, cannot be arbitrarily assigned ; they can be as-
signed only along the root loci .
3. The desired pole region in Figure 7.4 is developed for quadratic transfer func-
tions with a constant numerator. When it is used to design other types of transfer
functions, it is advisable to simulate resulting systems to check whether they
really meet the given specifications .
7 .6 PROPORTIONAL-DERIVATIVE (PD) CONTROLLER
In this section we give an example that uses a proportional-derivative (PD) controller .

2
(7 .44)
G(s) = s(s + 1)(s + 5)
Design an overall system to meet the specifications :
1. Velocity error as small as possible
2. Overshoot < 5%
3. Settling time < 5 seconds
As a first try, we choose the unity-feedback system shown in Figure 7 .24 . The
overall transfer function is
2k 2k
G(s) _
s(s + 1)(s + 5) + 2k s3 + 6s 2 + 5s + 2k
A necessary condition for G0 (s) to be stable is k > 0. Thus we plot the root loci of
2 _ 1
s(s + 1)(s + 5) k
Y
2 y
s(s + 1) (s + 5)

Im s
for k > 0 . The root loci are shown in Figure 7 .25 . There are three asymptotes with
centroid at
0 - 1 - 5
_ -2
3
and with angles 60 and 180 . The breakaway point can also be computed ana-
lytically by solving
D(s)N'(s) - D'(s)N(s) _ -(3s2 + 12s + 5) _ -3(s + 0 .47)(s + 3 .5)
Its solutions are - 0.47 and - 3.5 . Clearly - 0.47 is a breakaway point, but - 3 .5
is not.'
In order for the resulting system to have settling time less than 5 seconds, all
the poles of G0(s) must lie on the left-hand side of the vertical line passing through
the point - 4.5/t s = - 0.9. From the root loci in Figure 7 .25 we see that this is not
possible for any k > 0 . Therefore, the configuration in Figure 7 .24 cannot meet the
specifications .
As a next try, we introduce an additional tachometer feedback as shown in
Figure 7 .26 . Now the compensator consists of a proportional compensator with gain
71t is a breakaway point of the root loci for k < 0 .

7 .6 PROPORTIONAL-DERIVATIVE (PD) CONTROLLER 257
2 Y
s(s + 1) (s + 5)
Figure 7 .26 PD controller.
k and a derivative compensator with transfer function k,s, thus it is called a PD

compensator or controller. 8 It has two parameters, k and k, . Because we can handle
only one parameter at a time, we shall choose a value for k. First we choose k = 1
and carry out the design . It is found that the design is not possible for any k 1 . Next
we choose k = 5. Then the overall transfer function of Figure 7.26 becomes
2k
s(s + 1)(s + 5)
G0 (s) =
2k 2k1s
1 + s(s+ 1)(s+5)+s(s+ 1)(s + 5)
_ 2k (7 .45)
s(s + 1)(s + 5) + 2k + 2k 1 s
10
2
s3 +6s +5s+2k,s+ 10
The root loci of (s 3 + 6s 2 + 5s + 10) + k, (2s) or of
1 _ 2s
k, s 3 + 6s 2 + 5s + 10 (7 .46)
2s
(s + 5 .42)(s + 0.29 + j 1.33)(s + 0 .29 - j l .33)
are plotted in Figure 7.27. There are three trajectories . One moves from pole - 5.4
to the zero at s = 0 along the negative real axis ; the other two are complex conjugates
and approach the two asymptotes with centroid at
(- 5.42 - 0.29 + j 1.33 - 0.29 - j1 .33) - (0) - - 3
a = 3 - 1
and angles 90. Some of k, are also indicated on the plot .
sA different arrangement of PD controllers is U(s) = ( k + k 1 s)E(s) . See Chapter 11 . The arrangement

in Figure 7 .26, that is, U(s) = kE(s) + k1 sY(s), is preferable, because it differentiates y(t) rather than
e(t), which often contains discontinuity at t = 0 . Therefore, the chance for the actuating signal in Figure
7 .26 to become saturated is less .

IM s
\ I I
\I k1= 7 1 -3
I I
I 6
I \ 1 -2
\ Ik =1
I 4\ 3 1 X
I -I
k 1 =1 k 1 =3 \
!k1=4 k 1 =5
\ I \
Re s
-5 .4 1 2
I
/ I
/
/
/
/
Figure 7 .27 Root loci of (7.46).
Because G 0(0) = 1, the system in (7.45) has zero position error and its velocity
error, using (6.7), is
5+2k, - 0 2k, + 5
e(t) =
10 10
Thus the smaller k,, the smaller the error . To meet the specification on overshoot,
all poles must lie in the sector bounded by 45, as shown in Figure 7.27 . The real
pole lies inside the sector for all k, > 0. The complex poles move into the section
at about k, = 3 and move out at about k, = 6.5. Therefore, if 3 < k, < 6.5, then
all three closed-loop poles lie inside the sector and the system meets the specification
on overshoot. To meet the specification on settling time, all poles must lie on the
left-hand side of the vertical line passing through - 4.5/5 = - 0.9. The real pole
moves into the right-hand side at about k, = 5 ; the complex poles move into the
left-hand side at about k, = 2 .5 . Therefore if 2.5 < k, < 5, then all poles lie on the
left-hand side of the vertical line and the system meets the specification on settling
time . Combining the preceding two conditions, we conclude that if 3 < k, < 5, then
the system meets the specifications on overshoot and settling time .
The condition for the system to have the smallest rise time is that the closest
pole be as far away as possible from the origin . Note that for each k,, G0(s) in (7.45)
has one real pole and one pair of complex-conjugate poles . We list in the following

7.6 PROPORTIONAL-DERIVATIVE (PD) CONTROLLER 259
the poles and their shortest distance from the origin for k t = 3, 4, and 5 :
Because the system corresponding to k t = 4 has the largest shortest distance, it has
the smallest rise time among k t = 3, 4, and 5 . Recall that the velocity error is smaller
if k t is smaller. Therefore, if the requirement on velocity error is more important,
then we choose k t = 3. If the requirement on rise time is more important, than we
choose kt = 4. This completes the design .
The overall transfer function in (7 .45) is not quadratic ; therefore, the preceding
design may not meet the design specifications . Figure 7 .28 shows the unit-step re-
sponses of (7.45) for kt = 4 (solid line) and 3 (dashed line) . The overshoot, settling,
and rise times of the system with k t = 4 are, respectively, 0, 3 .1 and 2 .2 seconds .
The system meets all design specifications . The overshoot, settling, and rise times
of the system with k t = 3 are, respectively, 4 .8%, 6 .1, and 1 .9 seconds . The system
does not meet the specification on settling time but meets the specification on over-
shoot . Note that the system with kt = 3 has a smaller rise time than the system with
k t = 4, although the distance of its closest poles from the origin for kt = 3 is
1 .2
0 .8 Rise Settling -
Overshoot time, time,
percent seconds seconds
0.6
Figure 7 .26 (k = 5, k 1 = 4) 0 2 .2 3 .1
0 .4 - - - Figure 7 .26 (k = 5, k I = 3) 4.8 1 .9 6 .1 -
Figure 7 .29 (k = 5, a = 1, a = 2 .52) 3 .6 2 .2 4 .5

0 .2
0 1 2 3 4 6 7 8 9 10

smaller than that for k, = 4. Therefore the rule that the farther away the closest pole
from the origin, the smaller the rise time, is not applicable for this system . In con-
clusion, the system in Figure 7.26 with k = 5 and k, = 4 meets all design require-
ments and the design is completed .
7 .7 PHASE-LEAD AND PHASE-LAG NETWORKS
Consider again the design problem studied in Section 7.6. As shown there, the design
cannot be achieved by using the configuration in Figure 7.24 . However, if we intro-
duce an additional tachometer feedback or, equivalently, if we use a PD controller,
then the design is possible . The use of tachometer feedback, however, is not the only
way to achieve the design . In this section, we discuss a different design by using a
compensating network as shown in Figure 7.29 . The transfer function of the com-
pensating network is chosen as
s + a
C(s) : (7 .47)
_ s + aa
It is called a phase-lead network, if a > 1 ; a phase-lag network, if a < 1 . The
reason for calling it phase-lead or phase-lag will be given in the next chapter . See
also Problem 7 .9.
The transfer function of the system in Figure 7.29 is
s + a 2
k
- s + aa s(s + 1)(s + 5)
G(s) s + a 2
1 + k
s + as s(s + 1)(s + 5) (7 . 48)
_ 2k(s + a)
s(s + 1)(s + 5)(s + aa) + 2k(s + a)
Its denominator has degree 4 and the design using (7.48) will be comparatively
complex . To simplify design, we shall introduce a stable pole-zero cancellation .
Because both - 1 and - - 5 lie inside the desired pole region, either one can be
canceled. Arbitrarily, we choose to cancel the pole at - 1 . Thus we choose a = 1
in (7.47) and the overall transfer function in (7.48) reduces to
G(s) (7.49)
As + 5)(s 2+ a) + 2k
r + s+a 2 Y
)'0 k
s+aa s(s + 1) (s + 5)
Figure 7 .29 Unity-feedback system with compensating network .

7 .7 PHASE-LEAD AND PHASE-LAG NETWORKS 261
If k is chosen as k = 5, then (7.49) becomes

10
G0(s) =
s3 + (5 + a)s 2 + 5as + 10 (7 .50)
10
(S 3 +5S2 + 10)+as(s+5)
The root loci of (s 3 + 5s2 + 10) + as(s + 5) or, equivalently, of
-1 _ s(s + 5)
a s3 + 5s2 + 10 (7 .51)
s(s + 5)
(s + 5.35)(s - 0.18 + jl .36)(s - 0.18 - j1.36)
as a function of a are plotted in Figure 7 .30.

Now the specification on overshoot requires that the roots be inside the sector
bounded by 45 as shown. The settling time requires that the roots be on the left-
hand side of the vertical line passing through point (-0 .9, 0). We see from Figure
7.30 that if a, < a < a3, then these two specifications are satisfied . In order to have
the rise time be as small as possible, the pole closest to the origin should be as far
away as possible from the origin . Again from Figure 7.30 we rule out the range from
a2 to a3. The roots corresponding to any a in (a,, a2) are roughly the same distance
from the origin, thus we may pick an a from this range . The last specification is that
the velocity error be as small as possible . The velocity error of G0(s) in (7.50) is,
IM S
Re s
Figure 7.30 Root loci of (7.51).

from (6.7),
5a - 0 a
e,(t) = X 100%
10 2
Hence in order to have the smallest possible velocity, error, we choose a to be a, .
The parameter a, can be obtained from (7 .51) by measurement as
-1 s(s + 5)
al (s + 5 .35)(s - 0.18 + j 1 .36)(s - 0.18 - j l .36) S=sj
_ 1 .4 X 4.1 1
4 .45 X 1 .2X2.7 2.52
which implies a, = 2 .52 . Hence by choosing k = 5, a = 1, and a = 2.52, the
system in Figure 7 .29 may meet all the design specifications, and the design is
completed . The total compensator is
k(s + a) - 5(s + 1)
(7.52)
s + aa s + 2.52
It is a phase-lead network .
The unit-step response of the system in Figure 7 .29 with (7 .52) as its compen-
sator is plotted in Figure 7 .28 with the dotted line . Its overshoot is about 3 .6%; its
settling time is 4.5 seconds . It also responds very fast . Thus the design is satisfactory.
7 .8 CONCLUDING REMARKS
We give a number of remarks to conclude this chapter .

1. The root-locus method is basically a graphical method of plotting the roots of
a polynomial as a function of a real parameter . In the plotting, we use only the
phase condition in (7.19). From the properties of roots of polynomials, we can
often obtain a rough sketch of root loci without any measurement or computa-
tion. Plotting exact root loci is best done on a personal computer.
2. The root-locus design method tries to choose a parameter such that the poles
are in a nice location . To guide the choice, we develop a desired pole region
from quadratic transfer functions with a constant numerator. If an overall transfer
function is not quadratic, there is no guarantee that the resulting system will
meet the given specifications .
3. The method is a trial-and-error method . Therefore, it may take us several trials
before we succeed in designing an acceptable system .
4. In the root-locus design method, the constraint on actuating signals is not con-
sidered. The constraint can be checked only after the completion of the design .
If the constraint is not met, then we may have to redesign the system.
5. The design method considers only the poles of overall transfer functions . The
zeros are not considered . Therefore, the method is a special case of the pole-

PROBLEMS 263
placement problem in Chapter 10 . The pole-placement problem in Chapter 10

can assign poles in any location ; the root-locus method can assign poles only
along the root loci.
PROBLEMS
7.1 . Sketch the root loci for the unity-feedback system shown in Figure 7 .1 with
s + 4
a. G(s) = s
2(s + 1)
b .G(s)=(s+4)(s+6)
(s - 1)(s + 1)
s2 +2s+2
c. G (s) =
(s + 1) 2(s2 + 4s + 6)
7.2. Sketch the root loci of the polynomials
a. s3 +252 +3s+ks+2k
b. s3 (l + 0.001s)(1 + 0 .002s) + k(1 + 0.1s)(1 + 0.25s)
7.3. Use the root-locus method to show that
a. The polynomials3 + s2 + s + 2 has one real root in (-2, - 1) and a pair
of complex-conjugate roots with real part in (0, 1) . [Hint: Write the poly-
nomial as s2(s + 1) + k(s + 2) with k = 1.1
b. The polynomial
s5 + 2s4 - 15s3 + s2 - 2s - 15
has three real roots and a pair of complex-conjugate roots . Also show that
the three real roots lie in (5, 3), (0, - 3), and (- 5, - cc) .
7.4. The root loci of the system shown in Figure P7 .4(a) are given in Figure P7.4(b) .
Find the following directly from measurement on the graph .
a. The stability range of k.
b. The real pole that has the same value of k as the pair of pure imaginary
poles .
c. The k that meets (i) overshoot < 20%, (ii) settling time <_ 10 seconds, and
(iii) smallest possible position error .
7.5. Consider the feedback system shown in Figure P7 .5 . Sketch root loci, as a
function of positive real k, for the following :
1 _ 4(s + 2)
a . G(s) = s(s + 1)' H(s) s + 4
s2 +4s+4 _ s+5
b. G(s) = H(s)
s(s - 1) ' s2 + 2s + 2

r + 2
V
41
(s+3) (s 2 +2s+2)
(a)
IM s
k->oo
3
' ~ I I 1
~ 2
k=0
X
k --3 oo k=0 k--oo
Re s
-4 -3 -2 -1 0 1
k=0
)<
1
t
i 2
(b) Figure P7 .4
Figure P7 .5
7.6. Consider the unity-feedback system shown in Figure 7.1. Let the plant transfer
function be
s + 1
G(s) _
(s-0.2+j2)(s-0.2-j2)
Find the ranges of k to meet the following
a. Position error < 10%
b. a. and overshoot < 15%
c. a., b., and settling time < 4 .5 seconds
d. a., b., c ., and the smallest possible rise time .

PROBLEMS 265
7 .7 . A machine tool can be automatically controlled by a punched tape, as shown

in Figure P7.7(a). This type of control is called numerical control . By neglect-
ing the quantization problem, the system can be modeled as shown in Figure
P7.7(b). Find a gain k such that the system has zero position error and zero
overshoot. Numerical control cannot have overshoot, otherwise it will overcut
or the tool will break . No constraints are imposed on the settling and rise times.
Digital measurement
Hydraulic motor
Amplifier
-l Machine
D/A tool
Tape Comparator converter
reader 1 S2
I
.000, END
L=2H
(a)
1 -0 3
k 1+2s s(O.3s+ 1)
(b)
Figure P7 .7
7 .8 . The depth below sea level of a submarine can be maintained by the control
system shown in Figure P7 .8 . The transfer function from the stem plane angle
0 to the actual depth y of the submarine can be modeled as
10(s + 2)2
G(s) _ (s + 10)(s 2 + 0.1)
The depth of the submarine is measured by a pressure transducer, which is
assumed to have a transfer function of 1 . Find the smallest k such that the
Actuator and
submarine dynamics
Y
Figure P7 .8

position error is less than 5%, the settling time is less than 10 seconds, and the
overshoot is less than 2% .
7 .9. a. Consider C(s) = (s + 2)/(s + 1). Compute its phase at s = jl . Is it
positive or negative?
b. Consider C(s) = (s + a)/(s + b) . Show that the phase of G(j(o) for every
u> > 0 is positive for 0 < a < b and negative for 0 < b < a . (Thus, the
transfer function is called a phase-lead network if b > a and a phase-lag
network if a > b .)
7 .10. Consider the unity-feedback system shown in Figure P7 .10 . Use the Routh test
to find the range of real a for the system to be stable . Verify the result by using
the root-locus method . Find the a such that the system has the smallest settling
time and overshoot. Is it a phase-lead or phase-lag network?
)'0 s+a I
s+3
Figure P7 .10
7 .11 . The speed of a motor shaft can be controlled accurately using a phase-locked
loop [39] . The schematic diagram of such a system and its block diagram are
shown in Figure P7 .11 . The desired speed is transformed into a pulse sequence
with a fixed frequency . The encoder at the motor shaft generates a pulse stream
whose frequency is proportional to the motor speed . The phase comparator
generates a voltage proportional to the difference in phase and frequency .
Sketch the root loci of the system . Does there exist a k such that the settling
time of the system is smaller than I second and the overshoot is smaller than
10 percent?
Desired Motor
speed speed
Phase Compensation Amplifier
Encoder detector network Motor
k
1lN1fL ncode
(a)
0.2 0.4s+ 1 20
k
S 4s 0 .ls+ 1
(b)
Figure P7 .11

PROBLEMS 267
7 .12 . The transfer function from the thrust deflection angle u to the pitch angle 0 of
a guided missile is found to be
4(s + 0.05)
G(s) =
s(s + 2)(s - 1 .2)
The configuration of the compensator is chosen as shown in Figure P7 .12. The
transfer function of the actuator is G 1(s) = 1/(s + 6.1). If kt = 2k2, find a
kt, if it exists, such that the position error is less than 10%, the overshoot is
less than 15%, and the settling time is less than 10 seconds .
Actuator Missile
u 0
G I (s) G (s) 1
Figure P7 .12
7 .13 . Consider the control system shown in Figure P7 .13 . Such a system may be
used to drive potentiometers, dials, and other devices . Find kt and k2 such that
the position error is zero, the settling time is less than 1 second, and the over-
shoot is less than 5% . Can you achieve the design without plotting root loci?
Amplifier Motor Gear train
r U 10 1
4 i s(s+1) 5
Tachometer
k s
Figure P7 .13
7 .14 . One way to stabilize an ocean liner, for passengers' comfort, is to use a pair
of fins as shown in Figure P7 .14(a) . The fins are controlled by an actuator,
which is itself a feedback system consisting of a hydraulic motor . The transfer
function of the actuator, compared with the dynamics of the liner, may be
simplified as a constant k. The equation governing the roll motion of the
liner is
JO(t) + 716(t) + aO(t) = ku(t)
where 0 is the roll angle, and ku(t) is the roll moment generated by the fins .
The block diagram of the linear and actuator is shown in Figure P7 .14(b) . It
is assumed that a/J = 0.3, r,/2V'aJ = 0 .1, and k/a = 0.05. A possible
configuration is shown in Figure P7 .14(c) . If kt = 5, find a k2, if it exists, such

that (1) position error < 15%, (2) overshoot <_ 5%, and (3) settling time < 30
seconds. If no such k2 exists, choose a different k, and repeat the design .
r I
I
u k I B
Js 2 +ris+a I
I
I
L
(a) (b)
0 .015
s 2 +0 .11s+0.3
(c) Figure P7 .14
7 .15 . A highly simplified model for controlling the yaw of an aircraft is shown in
Figure P7 .15(a), where 0 is the yaw error and 0 is the rudder deflection . The
rudder is controlled by an actuator whose transfer function can be approximated
as a constant k. Let J be the moment of inertia of the aircraft with respect to
the yaw axis . For simplicity, it is assumed that the restoring torque is propor-
tional to the rudder deflection (t) ; that is,
JO(t) = -kcp(t)
The configurations of compensators are chosen as shown in Figure P7 .15(b),
(c), and (d), where G(s) = - k/Js 2 = - 2/s 2. We are required to design an
overall system such that (1) velocity error < 10%, (2) overshoot < 10%,
(3) settling time < 5 seconds, and (4) rise time is as small as possible . Is it
possible to achieve the design using configuration (b)? How about (c) and (d)?
In using (c) and (d), do you have to plot the root loci? In this problem, we
assume that the saturation of the actuating signal will not occur .
7 .16. Consider the plant discussed in Section 6 .2 and shown in Figure 6.1 . Its transfer
function is computed as
300
G(s) =
s(s3 + 184s2 + 760.5s + 162)
PROBLEMS 269
G (s)
(b)
(c)
k S+/3 G(s)
s s+a
(d)
Figure P7 .15
Design an overall system such that (1) position error < 10%, (2) settling time
~ 5 seconds, and (3) overshoot is as small as possible .
7 .17 .Consider the system shown in Figure 7 .26 . Let k = 10 . Use the root-locus
method to find a k, so that the system meets the specifications listed in
Section 7.6.
7.18 . Consider the system shown in Figure 7 .26 . Let k, = 4. Use the root-locus
method to find a k so that the system meets the specifications listed in Section
7.6.
7.19 . In Figure 7 .29, if we choose a = 5, then the system involves a stable pole-
zero cancellation at s = - 5 . Is it possible to find k and a in Figure 7 .29 so
that the system meets the specifications listed in Section 7 .6? Compare your
design with the one in Section 7 .7 which has a stable pole-zero cancellation at
Frequency-Domain
Techniques
8.1 INTRODUCTION
In this chapter we introduce a design method that, like the root-locus method, takes
the outward approach. In this approach, we first choose a configuration, then search
a compensator and hope that the resulting overall system will meet design specifi-
cations . The method is mainly limited to the unity-feedback configuration shown in
Figure 8 .1, however . Because of this, it is possible to translate the design specifi-
cations for the overall system into specifications for the plant transfer function G(s) .
If G(s) does not meet the specifications, we then search for a compensator C(s) so
that C(s)G(s) will meet the specifications and hope that the resulting unity-feedback
configuration in Figure 8 .1 will perform satisfactorily . Thus, in this method we work
directly on G(s) and C(s) . However, the objective is still the overall system
G0(s) = G(s)C(s)/(1 + G(s)C(s)) . This feature is not shared by any other design
method.
The method has another important feature ; it uses only the information of G(s)
along the positive imaginary axis, that is, G(jto) for all co ? 0. Thus the method is
called the frequency-domain method. As discussed in Chapter 4, G(jco) can be ob-
tained by direct measurement. Once G(jw) is measured, we may proceed directly
to the design without computing the transfer function G(s). On the other hand, if we
are given a transfer function G(s), we must first compute G(jw) before carrying out
the design . Thus we discuss first the plotting of G(jw) .
270

8 .2 FREQUENCY-DOMAIN PLOTS 271
r U y
C(s) G (s)
8 .2 FREQUENCY DOMAIN PLOTS
We use a simple example to illustrate the basic concept . Consider

1
G (s) _ s + 0 (8 .1)
.5
or
1
G(j(o) = jw + 0
.5
We discuss the plot of G(jw) as a function of real w ? 0. Although w is real, G(j(O)
is, in general, complex . If w = 0, then G(0) = 2. If w = 0 .2, then
1 _ 1 _
G (j0.2 = = L9e -j22
0.5 + j0 .2 V0.29eta-'(o .2/0 .5) _ 0 .53 1ej22
Similarly, we can compute
- 87
G(jO.5) = 1 .4e -j45 G(j2) = 0.5e -j76 G(j lO) = 0 .1e j
Using these data we can plot a number of G(jw) by using different sets of coordi-
nates . The plot in Figure 8 .2(a) is called the polar plot . Its horizontal and vertical
axes are, respectively, Re G(jw) and Im G(jw), where Re and Im stand for the real
part and imaginary part . Thus a point in the plane is a vector with magnitude or gain
IG(j(o)l and phase 5C G(jw) . For example, if w = 0, it is the vector or point A
shown in Figure 8.2(a); it has magnitude 2 and phase 0 . Point B, corresponding to
w = 0 .5, has magnitude 1 .4 and phase -45 . The plot of G(s) in (8.1) happens to
be a semicircle as shown .
The plot in Figure 8 .2(b) is called the log magnitude-phase plot. It is a plot of
gain versus phase on rectangular coordinates as shown . The gain on the vertical
coordinate is expressed in decibels (dB), defined as
dB = 20 log IG(jw)j
For example, we have
20 log 2 = 6 dB
20 log 1 .9 = 5 .6 dB
20 log 1 .4 = 2.9 dB
20 log 0.5 = - 6 dB

272 CHAPTER 8 FREQUENCY-DOMAIN TECHNIQUES
Im G(jw) I G(j(O) dB I
-10
FA
B
-45 1 2 Re G(jw) 0 .01 0 .1 1 10 100 1000 w
klowk,
2,"
ff
-2
I I
-1 / 0
w = 0 .5
-10-
I
1 2
I I
3
>-
log w
(a)
-20-
dB
4 G (jw)
FA 0 .5
2 -45 w
0.1 1 10 100
- -45
I _901 -
(b) (c)
Figure 8 .2 Frequency of plots of G(s) .
and
20 log 0 .1 = - 20 dB
Note that the decibel gain is positive if I G(jw)I 1, and negative if IGjwI < 1 . The
>
phase, on the horizontal coordinate, is in linear angles . The log magnitude-phase

plot of G(s) in (8.1) is shown in Figure 8 .2(b). For example, point A, corresponding
to w = 0, has magnitude 6 dB and phase 0 ; point B, corresponding to w = 0 .5,
has magnitude 2 .9 dB and phase -45 and so forth . The plot is quite different from
the one in Figure 8.2(a).
The plot in Figure 8 .2(c) is called the Bode It actually consists of two plots:
plot.
gain versus frequency, and phase versus frequency . The gain is expressed in decibels,
the phase in degrees . The frequency on the horizontal coordinate is expressed in
logarithmic units as shown . Thus w = 1 corresponds to 0 ; w = 10 corresponds to
1 ; w = 100 corresponds to 2, and so forth . Note that w = 0 appears at -oo. Thus
point A should appear at -- and has 6 dB and zero degree . The complete Bode
plot of G(s) in (8.1) is plotted in Figure 8.2(c). We remark that w appears as a
variable on the plots in Figure 8 .2(a) and (b) whereas it appears as coordinates in

8 .2 FREQUENCY-DOMAIN PLOTS 273
Figure 8.2(c). Although the three plots in Figure 8 .2 look entirely different, they are
plots of the same G(jw) . It is clear that if any plot is available, the other two plots
can be obtained by change of coordinates .
With digital computers, the computation and plotting of G(jw) become very
simple . Even so, it is useful to be able to estimate a rough sketch of G(j(O) . This is
illustrated by an example .
Example 8.2 .1
Plot the polar plot of
2
(8.2)
G(s) = s(s + 1)(s + 2)
Before computing G(jw) for any w, we shall estimate first the values of G(j(O) as
w ---> 0 and w ---> - . Clearly we have
1
s - 0 or w - 0: G(s) - 2 or G(jc)) ~ =* IG(jco)l < G(jw) _ -90
2s
and
s or w -- ~: G(s) - 3 ~ IG(j(o)I -> 0, ~ G(jw) _ -270 0

S
They imply that for w very small, the phase is - 90 and the amplitude is very large .
Thus the plot will start somewhere in the region denoted by A shown in Figure
8.3(a). As w increases to infinity, the plot will approach zero or the origin with phase
Im G(jco) Im G(jc))
w=~ 1 1
Re G(jw) I i Re G(jw)
0 0
0 .6
162 B
A
A
(a) (b)
Figure 8 .3 Polar plot of (8.2).

- 270 or + 90 as shown in Figure 8.3(a). Recall that a phase is positive if measured

clockwise, negative if measured counterclockwise . Now we compute G(jw) at
w= 1 :
2 _ j162 2
G(~1) = j45' = 0.6e -
jl(jl + 1)(jl + 2) ej90 1 .4e 2.2e 127
It is plotted in Figure 8 .3(a) as point B . In other words, as w increases from 0 to ~,
the plot will start from region A, pass through point B, and approach the origin along
the positive imaginary axis . Thus the plot may assume the form shown in Figure
8.3(b). Thus a rough polar plot of G(s) in (8.2) can be easily obtained as shown .
Clearly the more points of G(jw) we compute, the more accurate the plot .
We stress once again that in plotting G(jw), it is useful to estimate first the
values of G(s) at w = 0 and w = oo. These values can be used to check the cor-
rectness of a digital computer computation .
Exercise 8 .2 .1
Plot the polar, log magnitude-phase, and Bode plots of G(s) = 1/(s + 2) .
To conclude this section, we discuss the plot of G(s) = 1/(s + 0 .5) using
MATLAB . We first list the commands for version 3 .1 of MATLAB:
n=[1] ;d=[1 0 .5] ;
w = logspace( -1,2) or w = logspace( -1,2,200) ;
[re,im] = nyquist(n,d,w) ;
plot(re,im),title('Polar plot')
[mag,pha]=bode(n,d,w) ;
db=20*log10(mag) ;
plot(pha,db),title('Log magnitude-phase plot')
semilogx(w,db),title('Bode gain plot')
semilogx(w,pha),title('Bode phase plot')
The numerator and denominator of G(s) are represented by the row vectors n and
d, with coefficients arranged in descending power of s, separated by spaces or com-
mas. Command logspace( -1,2,200) generates 200 equally spaced frequencies in
logarithmic scale between 10 -1 = 0 .1 and 102 = 100 radians per second. If 200
is not typed, the default is 50 . Thus logspace( -1,2) generates 50 equally spaced
frequencies between 0.1 and 100. Command nyquist(n,d,w) 1 computes the real part
and imaginary part of G(jw) at w. Thus, plot(re,im) generates a polar plot. Com-
mand bode(n,d,w) computes the magnitude and phase of G(jw) at w . The magni-
'The name Nyquist will be introduced in a later section . The Nyquist plot of G(s) is defined as G(jw)
for w ? 0 and w < 0, whereas the polar plot is defined for only w ? 0 . Because command nyquist(n,d,w)
in version 3 .1 of MATLAB computes only positive w, a better name would be polar(n,d,w) .

8 .3 PLOTTING BODE PLOTS 275
tude is converted into decibels by db=20*Iog10(mag) . Command plot(pha,db)

plots the phase on the x- or horizontal axis with linear scale and the gain in decibels
on the vertical axis . Thus the plot is a log magnitude-phase plot . Command semi-
logx(w,db) plots w on the horizontal axis with logarithmic scale and the gain in dB
on the vertical axis . Thus the plot is a Bode gain plot . Similarly, semilogx(w,pha)
generates a Bode phase plot . For version 3 .5 or the Student Edition of MATLAB,
the commands
n=[1] ;d=[1 0 .5];
bode(n,d)
will plot the Bode gain and phase plots on the screen . The command
nyquist(n,d)
will plot the polar plot (G(jw), w ? 0) with the solid line and its mirror image
(G(jw), (o < 0) with the dashed line on the screen . Thus, the use of MATLAB to
generate frequency plots is very simple .
8.3 PLOTTING BODE PLOTS
In this section, we discuss the plot of Bode plots by hand . One may wonder why we
bother to study this when the plot can be easily obtained on a personal computer .
Indeed, one can argue strongly for not studying this section . But in the study, we
can learn the following : the reason for using logarithmic scales for frequency and
magnitude, the mechanism for identifying a system from its Bode plot, and the reason
for using the Bode plot, rather than the polar or log magnitude-phase plot, in the
design. Besides, the plot of Bode plots by hand is quite simple ; it does not require
much computation .
We use an example to discuss the basic procedure of plotting Bode plots .
Consider
5s + 50 _ 5(s + 10)
G(s) (8 3)
- s 2 + 99 .8s - 20 (s - 0.2)(s + 100)
First we write it as
-5X101 +1O s)
G(s) =
0.2 X 1001 - 012 s 1 + 1
10 s
o
-2.5(l + t s)
I1
0 .2s)( 1 + 100 s)

It is important to express every term in the form of 1 + Ts . The gain of G(s) in

decibels is 20 log IG(jw)l or
20 log IG(s)I = 20 log 1 -2.51 + 20 log

(8 .5)
- 20 log 1 - 1
0 2 s - 20 log 1 + 100
s
and the phase of G(s) is
-G(s)= <(-2 .5)+<C1+0s)
-
C 02
1- s) -<
(100
1+
s)
We see that the decibel gain of G(s) is simply the algebraic sum of the decibel gain
of each term of G(s). Adding all gains of the terms in the numerator and then
subtracting those in the denominator yields the gain of G(s) . Similar remarks apply
to the phase of G (s) . Other than the constant term, (8.4) consists only of linear factors
of form (1 + Ts) . Therefore we discuss these linear factors first .
Gain of Linear Factors

Consider 20 log 11 Tjw l, for T > 0 and w ? 0. We use the following
approximations :
to very small or :
.1/T 20 log i1 jwT l - 20 log 1 = 0 dB
(o<0
w very large or w ? 10/T : 20 log I1 jWTI .- 20 log wT
In other words, for w very small, 20 log 11 jwrl can be approximated by a
horizontal line passing through 0 dB . For w very large, 20 log I1 jwr( can be
approximated by the straight line 20 log WT. If w = 1/T, then 20 log wT = 20 log
1 = 0, thus the straight line 20 log wT intersects the 0 dB line at w = 1/7 . The
gain of 20 log wT increases 20 dB whenever w increases 10 times, thus the straight
line has a slope of 20 dB/decade as shown in Figure 8 .4. The point 1/T where the
straight line intersects with the 0-dB line is called the corner frequency. Once the
corner frequency is determined, the two straight lines can be readily drawn . The two
straight lines are called the asymptotes .
Now the question is : How good is the approximation? For w :5 0.1/T, the
difference between 20 log 11 j(o7- and the 0-dB line is less than 0 .04 dB . Similarly,
1
if w j 10/T, the difference between 20 log 1 jwTI and 20 log wT is again less
I
than 0.04 dB . Thus, for w < 0 .1/T and w ? 10/T, or for w less than one-tenth of the
corner frequency and larger than ten times the corner frequency, the Bode gain plot
almost coincides with the asymptotes. For w in the frequency interval
(0.1/T, 10/T), the exact plot differs from the asymptotes . The largest difference
occurs at corner frequency 1/T, and equals
20 log 11 jwr 1 = 20 log 11 j II = 20 log \ = 3 dB

dB
20
0 01 1 N I 10
T T ~~ I T
Poles ' I
-20 'S
-20 dB/decade
-20 log I I jro)I
Figure 8 .4 Bode gain plot of 20 log I 1 jTwl with T > 0.
If we pick the 3-dB point and draw a smooth curve as shown in Figure 8 .4, then the
curve will be a very good approximation of the Bode gain plot of (1 Ts) . Note
that the Bode gain plot of (1 + Ts) is identical to that of (1 - -rs) . If they appear
in the numerator, their asymptotes go up with slope 20 dB/decade . If they appear
in the denominator, their asymptotes go down with slope -20 dB/decade . In other
words, the asymptotes of 20 log 11 jrwl go up with slope 20 dB/decade ; the
asymptotes of -20 log Il jTtol go down with slope -20 dB/decade .
Now we plot the Bode gain plot of (8.4) or (8.5). The corner frequency of zero
(1 + s/10) is 10, and its asymptote goes up with slope 20 dB/decade as shown in
Figure 8 .5 with the dashed lines . The corner frequency of pole (I - s/0 .2) is 0 .2,
and its asymptote goes down with slope - 20 dB/decade as shown with the dotted
dB
40-
i
20
20-
8 dB
0 .01 0 .1 0 2 " 10 / 100 -\ 1000

-20 dB/ decade '\
-20 W1 \ \ - 20
-40-
Figure 8 .5 Bode gain plot of (8 .4) .

lines. Note that the negative sign makes no difference in the gain plot . The corner
frequency of pole (1 + s/100) is 100, and its asymptote goes down with slope
- 20 dB/decade as shown with the dashed-and-dotted lines . Thus, there are three
pairs of asymptotes . Now we consider gain - 2.5 in (8.4). Its decibel gain is 20 log
- 2 .51 = 8 dB; it is a horizontal line, independent of w, as shown . The sum of the
horizontal line and the three pairs of asymptotes is shown by the heavy solid line .
It is obtained by adding them point by point . Because the plot consists of only straight
lines, we need to compute the sums only at w = 0 .2, 10, 100, and at a point larger
than w = 100. The sum can also be obtained as follows . From w = 0 to w = 0.2,
the sum of an 8-dB line and three 0-dB lines clearly equals 8 dB . Between [0.2, 101,
there is only one asymptote with slope -20 dB/decade . Thus we draw from w =
0.2 a line with slope -20 dB/decade up to w = 10 as shown . Between [10, 100],
there is one asymptote with slope 20 dB/decade and one with slope - 20 dB/decade,
thus the net is 0 dB/decade . Therefore, we draw a horizontal line from w = 10 to
w = 100 as shown . For w ? 100, two asymptotes have slope -20 dB/decade
and one has slope 20 dB/decade . Thus, the net is a straight line with slope
-20 dB/decade . There are three corner frequencies, at w = 0.2, 10, and 100 .
Because they are far apart, the effects of their Bode gain plots on each other are
small . Thus the difference between the Bode plot and the asymptotes at every corner
frequency roughly equals 3 dB . Using this fact, the Bode gain plot can then be
obtained by drawing a smooth curve as shown . This completes the plotting of the
Bode gain plot of (8.4).
Phase of Linear Factors

Consider the phase of (1 -rs) or, more precisely, of (1 jwr) with T > 0 and
w ? 0 . We shall use the following approximations :
w very small or w < 0 .1/T : 1 (1 j(oT) ~ (1) = 0

and
< (+jwT) = 90
w very large or w ? 10/T : <1 (1 jwr) --
4 (- jwT) = -90'
In other words, for w in ( - co, 0.1/T), the phase of (1 jwT) can be approximated
by 0; for w in (10/T, oc), the phase of (1 + j(oT) can be approximated by 90 and
the phase of (1 - jwr), by -90 as shown in Figure 8 .6 . We then connect the end
points by a dashed straight line as shown . The exact phase of (1 + j (0 r) is also
plotted in Figure 8 .6 using a solid line . We see that at w = 0.1/T and w = 10/r,
the differences are 5 .7. There is no difference at w = 1/r . The differences at the
midpoints between 0 . 11,r and 1 / T and between 1/7 - and 10/Tare 3 .4. The differences
are fairly small between the dashed straight line and the exact phase plot . Thus, the
phase plot of (1 + jwr) can be approximated by the straight lines . Note that the
phase of (1 - j (o r) equals the reflection of the phase of (1 + jwr) to negative
angles . Thus the phase of (1 - jwr) can also be approximated by the straight lines .

5 .7
901 - 3 .4
3 .4
45'
5 .7'
w
1 101
T TI
I
(1 -jTcu) (zero)
II =- (I +jzc)) (Pole)
_901 -
Figure 8.6 Bode phase plots of 1 jcoT with T > 0.
For the gain, we have

I1 + jcoTI = I1 - jwTl
For the phase, we have
(I + jtoT) = - 4 (I - jwT) (1 - jcoT) _ - (1 + j(0 T)
Thus, the Bode gain plots of (1 + Ts) and (1 - Ts) are the same . If they appear as
zeros, then their plots are as shown in the upper part of Figure 8 .4; if they appear
as poles, then their plots are as shown in the lower part of Figure 8 .4. The situation
in phase plots is different . If (1 + Ts) appears as a zero, or (1 - Ts) appears as a
pole, then their phases are as shown in the upper part of Figure 8 .6. If (1 + Ts)
appears as a pole, or (1 - Ts) appears as a zero, then their phases are as shown in
the lower part of Figure 8 .6.
With the preceding discussion, we are ready to plot the Bode phase plot of (8.3)
or (8.6). The phase of gain -2 .5 is 180 . It is a horizontal line passing through 180
as shown in Figure 8 .7. The corner frequency of zero (1 + s/10) is 10 ; its phase
for w smaller than one-tenth of 10 is 0 ; its phase for w larger than ten times 10 is
+90 . The phase for w in (1, 100) is approximated by the dashed line shown in
Figure 8 .7. The phase for pole (1 - s/0 .2) is - 4 (1 - jw/0 .2). Its phase is 0,
for w < 0.02, and - (- 90) = + 90 , for w > 2 . The approximated phase is plotted
with dotted lines. Similarly the phase of pole (1 + s/100) or - < (1 + j(01 1 00) is
plotted with dashed-and-dotted lines . Their sum is denoted by the solid line ; it is
obtained using the procedures discussed for the gain plot . By smoothing the straight
lines, the Bode phase plot can then be obtained (not shown) .
Once a Bode phase plot is completed, it is always advisable to check the plot
for w -* 0 and w - 0c. If w -~ 0, (8 .3) reduces to - 2.5 and its phase is 180 . If w
--> oc, (8 .3) reduces to 5s/s 2 = 5/s and its phase is - 90 which also equals 270 .
Thus the plot in Figure 8 .7 checks with the two extreme cases . Note that phases are
considered the same if they differ by 360 or its multiples . For example, if the phase

I
270 _ I
I I I I
I
180 I
I I I
I I 1 I
90- ---1
1 ` --- - - - -I - -
co
0 .01 0.02 0.1 0 .2 1 2 10 - 100 1000
_90-
Figure 8 .7 Bode phase plot of (8.3).
of -2 .5 is plotted as -180 , then the phase plot in Figure 8 .7 will be shifted down
by 360 , which is the one generated by calling bode in MATLAB .
Exercise 8 .3.1
Plot the Bode plots of

10
a. G(s) =
(s + 5)(s + 100)
2(s - 5)
b. G(s) =
(s + 5)2(s + 10)
Up to this point, we have considered transfer functions that contain only linear
factors . Now we discuss transfer functions that also contain quadratic factors and
poles or zeros at the origin .
Poles or Zeros at the Origin

The repeated pole 1 /s` has a logarithmic magnitude of - 20 log joi _ - 20i log
I
w dB and a phase of - i 90. If i = 1, the gain plot is a straight line passing

x
through the 0-dB line at co = 1 and with slope -20 dB/decade, as shown in Figure
8.8 . Its phase is a horizontal line passing through - 90. In Figure 8 .8, we also plot
the Bode plot of 1/s` for i = 2, - 1, and -2. The plots are very simple .

Gain (dB) Phase
180 i=-
90 Zeros
0.1 10 100
90 i =+1
Poles
180 i=+2
-40 dB /decade
Figure 8.8 Bode plot of 1/s` .
Quadratic Factors
Consider the complex-conjugate poles 2
w2 1
n (8 .7)
E(s) : _ 2
s2 + 2~ws + wn
1+ S+
CO CWn/
with 0 < ~ < 1 . The logarithmic magnitude for s = j(o is
-
20 log E(jw)I _ - 20 log 1 +
I (jw) -
((on ) 2
As for linear factors, we use the following approximations :

w very small or w < 0-1(on : 20 log E(jw)I 20 log III = 0 dB
-
w very large or w ? 10w,,
: 20 log JE(jw)I -20 log (w/ (0n)2 I I
-40 log (o/(on

They are two straight lines, called asymptotes. One asymptote has slope zero, the
other has slope -40 dB/decade . They intersect at corner frequency w = w n as
shown in Figure 8.9. The Bode gain plot for w in (0 .1 (on , 10w,,) depends highly on
damping ratio r and differs greatly from the asymptotes as shown in Figure 8 .9. If
~ = 0 .5, the gain plot is very close to the asymptotes . If ~ = 0 .7, the difference at
the corner frequency is 3 dB . The smaller ~, the larger the overshoot . Thus in plotting
the Bode gain plot of the quadratic factor, we first compute the corner frequency
and draw the two asymptotes, one with slope zero, the other with slope
2The Bode gain plot of 1/(1 - (2~/w,, )s + (s/w, )2 ) equals that of (8 .7), and its Bode phase plot equals
that of (8 .7) reflected to positive angles . To simplify discussion, we study only (8 .7) .

20
15
10
0.2
b 0.3
c 5
c7 0.5
0
-5
-10
0
~1..l MV1
01
b 90
1 \
a 0.3
0.5 i%P`
180
0.7
0 .1 0.2 0.4 0 .6 0 8 1 2 4 6 8 10
0)/O)
n
Figure 8 .9 Bode plot of quadratic term .
-40 dB/decade . We then compute the damping ratio and use Figure 8 .9 to draw
an approximate Bode gain plot . Note that if the quadratic factor appears as zeros,
then one asymptote will go up with slope + 40 dB/decade, and the plots in Figure
8.9 are reversed.
The phase plot of the quadratic factor in (8.7) can be approximated by straight
lines as follows :
(o very small or w < 0.1 w,,: 4 E(jw) - 1 = 0
w very large or w ? 10w0: E(jw) - ~ (jw/(O) 2 = - 180
The exact phases for w in (0.1 w,,, 10w) are plotted in Figure 8 .9 for various ~. For
small ~, they are quite different from the dashed straight line shown . Thus in plotting

the Bode phase plot of (8.7), we must compute the corner frequency w n as well as
the damping ratio ~ and then use Figure 8 .9.
Example 8 .3.1
Plot the Bode plot of the transfer function
50(s + 2)
G(s) =
s(s2 + 4s + 100)
To compute w and ~ of the quadratic term, we equate
wn = 100 2~w = 4
which imply w = 10 and ~ = 0 .2. We then express every term of G(s) in the form
Gain (dB)
0)
0)
Figure 8 .10 Bode plot.

of 1 + () as in (8 .4) or (8.7):
50 X2(i+Zs
G (s) =
z z
100s( 1+ s l+2 (10) s+ (10)]
100 s+ 100) 1
The asymptotes of the zero, the pole at the origin, and the quadratic term are plotted
in Figure 8 .10 using, respectively, the dashed, dotted, and dashed-and-dotted lines .
The sums of these asymptotes are denoted by the thin solid lines . Because the damp-
ing ratio is 0 .2, the Bode plot is quite different from the asymptotes at co = co n =
10. Using the plot in Figure 8 .9, we can obtain the Bode gain and phase plots of
G(s) as shown in Figure 8 .10 with the heavy solid lines .
Exercise 8 .3 .2

1
a. G(s) =
sz(s + 1)
100
b. G(s) = s(s
z + 4s + 100)
8 .3 .1 Non-Minimum-Phase Transfer Functions

Consider the following two transfer functions
s - z s + z
G,(s) _ and G2(s) =
S + p s + p
with their poles and zeros plotted in Figure 8 .11 . The transfer function G I(s) has one
zero s = z in the open right half plane . If we reflect the zero into the left half plane,
Im s Im s
Figure 8.11 Non-minimum-phase transfer function .

we obtain G 2(s). Now we compare the gains and phases of G 1(s) and G2(s) on the
positive jw-axis . From Figure 8 .11, we can see that the vectors from zero z . to any
point on the jw-axis and from zero - z to the same point have the same length ;
therefore, we have
IG1(jw)I = IGZ(j(0 )I
for all w ? 0 . Actually, this fact has been used constantly in developing Bode gain
plots. Although G1(s) and G2(s) have the same gain plots, their phases are quite
different . From Figure 8 .11, we have
<G1(jw)= 01 <G2(j ('J)=02-4)
-0
At w = 0, 0 1 = 180 > 02 = 0. At w - 00, 01 = 02 = 90 . In general, 01 ? 02
for all w > 0 . Thus we have
< G1(j(o) ? < G2(j(0)
for all w ? 0. Thus, if a transfer function has right-half-plane zeros, reflecting these
zeros into the left half plane gives a transfer function with the same amplitude but
a smaller phase than the original transfer function at every w ? 0. This motivates
the following definition :
o Definition 8 .1
A proper rational transfer function is called a minimum-phase transfer function if
all its zeros lie inside the open left half s-plane . It is called a non-minimum-phase
transfer function if it has zeros in the closed right half plane . Zeros in the closed
right half plane are called non-minimum-phase zeros. Zeros in the open left half
plane are called minimum-phase zeros .
We mention that there are two different definitions of minimum-phase transfer

functions in the literature . One requires both poles and zeros to lie inside the open
left half plane ; the other requires only zeros . Both definitions are ambiguous about
zeros on the jw-axis . Our definition of non-minimum-phase zeros includes zeros on
the jw-axis .
Exercise 8.3 .3

(s-2) (s + 2)
and
s(s - 1) s(s - 1)
Do they have the same amplitude plots? How about their phase plots? Phases are
considered the same if they differ by 360 or their multiples . If their phases are
set equal at w = 00, which transfer function has a smaller phase at every w ? 0?

Exercise 8 .3.4
Which of the following are non-minimum-phase?

S - 1 s+ 1 s(s+2)
(s + 2)2 (s - 2)2 (s + 1)2
[Answers : 1 and 3.]
8.3.2 Identification
Determination of the transfer function G(s) of a system from measured data is an
identification problem . As discussed in Section 4 .7.1, if a system is stable, then
IG(jco)I and o G(jw) can be obtained by measurement . This is also possible if G(s)
has only one unstable pole at s = 0 (see Problem 4 .21). From IG(jw)I and ~ G(jw),
we can readily obtain the Bode plot of G(s). Now we discuss how to obtain G(s)
from its Bode plot . This is illustrated by examples .
Example 8.3 .2
Find the transfer function of the Bode plot shown in Figure 8 .12(a) . First we ap-
proximate the gain plot by the three straight dashed lines shown . They intersect at
w = 1 and w = 10. We begin with the leftmost part of the gain plot . There is a
dB dB
20
4
A
-90,
0 .1 0.2 1 5 10 20 100
-I Nw Co
0.1 1 --90 10 100
90
-]go, - --180
-270' -
(a) (b)
Figure 8 .12 Bode plots.

straight line with slope - 20 dB/decade, therefore the transfer function has one pole
at s = 0 . At w = 1, the slope becomes -40 dB/decade, a decrease of 20 dB/decade,
thus there is a pole with corner frequency w = 1 . At w = 10, the slope becomes
- 20 dB/decade, an increase of 20 dB/decade, therefore there is a zero with corner
frequency w = 10 . Thus, the transfer function must be of the form
G(s) =
This form is very easy to obtain . Wherever there is a decrease of 20 dB/decade in

slope, there must be a pole . Wherever there is an increase of 20 dB/decade, there
must be a zero. The constant k can be determined from an arbitrary w. For example,
the gain is 37 dB at w = 1 or s = j 1 . Thus we have
k\/1 + 0 .01
37 = 20 X log = 20 X log
1XU1+1
j IX (I J1)
k X 1.005
20 X log
1.414
which implies
1 .414
Ik~ = 1037/20 = 1 .4 x 101-115 = 1 .4 . 70.79 = 99 .6
1 .005
Now we use the phase plot to determine the signs of each term . The phase of
G(s) for w very small is determined by k/s . If k is negative, the phase of k/s is
+ 90; if k is positive, the phase is - 90. From the phase plot in Figure 8 .12(a), we
conclude that k is positive and equals 99 .6. If the sign of pole 1 s is negative, the
pole will introduce positive phase into G(s) or, equivalently, the phase of G(s) will
increase as w passes through the corner frequency 1 . This is not the case as shown
in Figure 8 .12(a), therefore we have 1 + s . If the sign of zero 1 0 .ls is positive,
the zero will introduce positive phase into G(s) or, equivalently, the phase of G(s)
will increase as w passes through the corner frequency at 10 . This is not the case as
shown in Figure 8 .12(a), thus we have 1 - 0 .1s and the transfer function of the
Bode plot is
99.6(1 - 0. 1s) -9.96(s - 10)
G(s) =
s(1 + s) s(s + 1)

Example 8.3.3
Find the transfer function of the Bode plot in Figure 8.12(b) . The gain plot is first
approximated by the straight lines shown . There are three corner frequencies : 0.2,
5, and 20. At w = 0 .2, the slope becomes - 20 dB/decade; therefore, there is one
pole at 0 .2 or (1 s/0 .2) . At w = 5, the slope changes from -20 dB/decade to
0; therefore, there is a zero at 5 or (1 s/5) . At w = 20, the slope changes from
0 to -40 dB/decade ; therefore, there is a repeated pole or a pair of complex-
conjugate poles with corner frequency 20 . Because of the small bump, it is a quad-
ratic term . The bump is roughly 10 dB high, and we use Figure 8 .9 to estimate its
damping ratio ~ as 0 .15. Therefore, the transfer function of the Bode plot is of the form
k
C1 5)
G(s) _ 2
s 2X0 .15 s
0.21 20 s + 20
The gain of G(s) at w -> 0 or s = 0 is 40 dB. Thus we have
20 X log JG(0)J = 20 X log Jkl = 40
which implies Jkl = 100 or k = 100. Using the identical argument as in the
preceding example we can conclude from the phase plot that we must take the
positive sign in all . Thus the transfer function of the Bode plot is
100 1 +
( 5) 1600(s + 5)
G(s) = )2)
s 0.3 s (s + 0.2)(s2 + 6s + 400)
1+ 1+ 20 s+ 20))
C 0.2)( v
From these examples, we see that if a Bode plot can be nicely approximated by
straight lines with slope 20 dB/decade or its multiples, then its transfer function
can be readily obtained . The Bode gain plot determines the form of the transfer
function . Wherever the slope decreases by 20 dB/decade, there is a pole ; wherever
it increases by 20 dB/decade, there is a zero . Signs of poles or zeros are then
determined from the Bode phase plot . If the Bode plot is obtained by measurement,
then, except for a possible pole at s = 0, the system must be stable . Therefore, we
can simply assign positive sign to the poles without checking the phase plot . If the
transfer function is known to be of minimum phase, then we can assign positive sign
to the zeros without checking the phase plot . In fact, if a transfer function is stable
and of minimum phase, then there is a unique relationship between the gain plot and
phase plot, and we can determine the transfer function from the gain plot alone . To
conclude this section, we mention that devices, such as the HP3562A Dynamic
System Analyzer, are available to measure Bode plots and then generate transfer
functions . This facilitates considerably the identification of transfer functions .

8 .4 STABILITY TEST IN THE FREQUENCY DOMAIN 289
8 .4 STABILITY TEST IN THE FREQUENCY DOMAIN
Consider the unity-feedback system shown in Figure 8 .1 . We discuss in this section

a method of checking the stability of the feedback system from its open-loop transfer
function G(s)C(s). The transfer function of the unity-feedback system is
G,(s)
G ,(s)
(s)
1 + G(s)C(s) ' 1 + G,(s) (8.8)
where G,(s) : = G(s)C(s) is called the loop transfer function. The stability of G0 (s)
is determined by the poles of G0(s) or the zeros of the rational function
1 + G,(s)
Recall that we have introduced two methods of checking whether or not all zeros of
(1 + G,(s)) have negative real parts . If we write G,(s) = N,(s)/D,(s), then the zeros
of (1 + G,(s)) are the roots of the polynomial DI (s) + N,(s) and we may apply the
Routh test . Another method is to plot the root loci of G,(s) _ - 1/k with k = 1,
as was discussed in Section 7 .4.3. In this section, we shall introduce yet another
method of checking whether or not all zeros of (1 + G I (s)) lie inside the open left
half plane . The method uses only the frequency plot of G,(s) and is based on the
principle of argument in the theory of complex variables .
8 .4 .1 Principle of Argument
Consider a rational function F(s) . The substitution of a value s ;, a point in the
s-plane as shown in Figure 8 .13, into F(s) yields a value F(s, ), a point in the
F-plane. In mathematical terminology, s, in the s-plane is said to be mapped into
F(s l ) in the F-plane. In this sense, the polar plot in Figure 8 .2(a) is the mapping of
rttaP0A
IM s Im s
Re s %%'4,4
*
o,wA4h.,0, MZN% Re F(s)
s-plane F-plane
(a) (b)
Figure 8 .13 Mapping of C, .

the positive jcw-axis of the s-plane by G(s) in (8.1) into the G-plane . A simple closed
curve is defined as a curve that starts and ends at the same point without going
through any point twice .
Principle of Argument
Let C 1 be a simple closed curve in the s-plane . Let F(s) be a rational function of
s that has neither pole nor zero on C 1. Let Z and P be the numbers of zeros and
poles of F(s) (counting the multiplicities) encircled by C 1 . Let C 2 be the mapping
of C1 by F(s) into the F-plane . Then C2 will encircle the origin of the F-plane
(Z - P) times in the same direction as C 1 .
Because F(s) has no pole on C 1, the value of F(s) at every s on C 1 is well

defined. Because F(s) is a continuous function of s, the mapping of C 1 by F(s) is a
continuous closed curve, denoted by C2, in the F-plane as shown in Figure 8 .13.
The closed curve, however, is not necessarily simple-that is, it may intersect with
itself. Furthermore C 2 will not pass through the origin, because F(s) has no zero on
C1 . Now if Z - P is positive-that is, F(s) has more zeros than poles encircled by
C1-then C2 will travel in the same direction as C 1 . If Z - P is negative, then C2
will travel in the opposite direction. In counting the encirclement, we count the net
encirclement . For example, if C2 encircles the origin once in the clockwise direction
and three times in the counterclockwise direction, then C2 is considered to encircle
the origin two times in the counterclockwise direction . Consider the mapping shown
in Figure 8 .13 . The contour C 1 is chosen to travel in the clockwise direction and
encircles four poles and two zeros of F(s) . Thus we have Z - P = - 2. Now if we
plot F(s) along C 1 , the resulting contour C2 must encircle the origin of the F-plane
twice in the direction opposite to C 1 or in the counterclockwise direction . Indeed,
the contour C2 encircles the origin three times in the counterclockwise direction,
once in the clockwise direction . Therefore the net encirclement is twice in the coun-
terclockwise direction . One way to count the encirclements is to draw a straight line
from the origin as shown in Figure 8 .13 ; then the number of encirclements equals
the number of crossings of the straight line by C 2 minus the number of reverse
crossings by C2-
8.4 .2 The Nyquist Plot

Consider
G(s) 1 GI(Gs)1(s) Ps) (8.9)

where F(s) = I + GI(s) . The condition for G0(s) to be stable is that the numerator
of F(s) is a Hurwitz polynomial or, equivalently, F(s) has no zero in the closed right
half s-plane . In checking stability, the contour C 1 will be chosen to enclose the entire
closed right half plane as shown in Figure 8 .14(a), in which the radius R of the
semicircle should be very large or infinity . The direction of C 1 is arbitrarily chosen
to be clockwise . Now the mapping of C 1 by F(s) into the F-plane is called the Nyquist

Im s Im s
hIF Re s
IF
R
)
Re s
(a) (b)
Figure 8.14 Contour of C, .
plot of F(s) . Similarly, we may define the Nyquist plot of GI(s) as the mapping of
C1 by G,(s) . We first use an example to illustrate the plotting of the Nyquist plots
of G,(s) and F(s) .
Example 8 .4.1
Consider
8s
G,(s) (8 .10)
_ (s - 1)(s - 2)
We compute G,(j l) = 2 .6ej 161 ' G,(j2) = 2.6ei"and G,(j 10) = 0.8e-''07' .
Using these, we can plot the polar plot of GI(s) or the plot of G,(jw) for w ? 0 as
shown in Figure 8 .15 with the solid curve . It happens to be a circle . Because all
coefficients of G,(s) are real, we have
G,( -j(o) = G i (j(o)
where the asterisk denotes the complex conjugate . Thus the mapping of G,(jw), for
w < 0, is simply the reflection, with respect to the real axis, of the mapping of
G,(jw), for w ? 0, as shown with the dashed curve . Because G,(s) is strictly proper,
every point of the semicircle in Figure 8 .14(a) with R --* is mapped by GI(s) into
0. Thus, the complete Nyquist plot of G,(s) consists of the solid and dashed circles
in Figure 8 .15 . As w travels clockwise in Figure 8 .14(a), the Nyquist plot of G,(s)
travels counterclockwise as shown .
If GI(s) is proper, then the mapping of the infinite semicircle in Figure C 1 by
G1(s) is simply a point in the G,-plane . Thus the Nyquist plot of G,(s) consists mainly
of G,(jw) for all w. The polar plot of G,(s), however, is defined as G,(jw) for w
0. Therefore the Nyquist plot of G,(s) consists of the polar plot of G,(s) and its
reflection with respect to the real axis . Therefore, the Nyquist plot can be readily
obtained from the polar plot .

Re G1
Re F
Figure 8 .15 Nyquist plot of (8 .10).
Because F(s) = 1 + G,(s), the Nyquist plot of F(s) is simply the Nyquist plot
of GI(s) shifted to the right by one unit . This can be more easily achieved by choosing
the coordinates of the F-plane as shown in Figure 8 .15. In other words, the origin
of the F-plane is the point (- 1,0) in the G,-plane . Therefore, once the Nyquist plot
of G,(s) is obtained, the Nyquist plot of F(s) is already there .
Exercise 8 .4 .1
Plot the Nyquist plots of 1/(s + 1) and 2/(s - 1).
The transfer function in (8 .10) has one zero and no pole on the imaginary axis,
and its Nyquist plot can easily be obtained . Now if a transfer function G,(s) contains
poles on the imaginary axis, then GI(s) is not defined at every point of C, in Figure
8.14(a) and its Nyquist plot cannot be completed. For this reason, if G,(s) contains
poles on the imaginary axis as shown in Figure 8 .14(b), then the contour C 1 must
be modified as shown . That is, wherever there is a pole on the imaginary axis, the
contour is indented by a very small semicircle with radius r. Ideally, the radius r
should approach zero. With this modification, the Nyquist plot of G,(s) can then be
completed . This is illustrated by an example . Before proceeding, we mention that
the command nyquist in MATLAB will yield an incorrect or incomplete Nyquist
plot if GI(s) has poles on the imaginary axis .
Example 8 .4.2
Consider
S + 1
G,(s) = s2(s - 2) (8.11)
Its poles and zero are plotted in Figure 8 .16(a). Because GI(s) has poles at the origin,
the contour C, at the origin is replaced by the semicircle re' , where 0 varies from

G (C) Re G 1
-2 - G,(A) 1 -]I ~ ReF
I /
R I
I
(a) (b)
Figure 8 .16 Nyquist plot of (8.11).
- 90 to 90 and r is very small . To compute the mapping of the small semicircle

by G,(s), we use, for s very small,
G,(s) = (8.12)
s 2(s - 2) s2( -2) 2s 2
Now consider point A or s = rej 0 in Figure 8 .16(a) . It is mapped by (8 .11) or (8 .12)
into
e-180
GI(A)
- I - = 1 ej180-
2r 2ej0 2r 2 2r 2
Its phase is 180 and its amplitude is very large because r is very small . Similarly,
we have the following
ej180- 1 e j90 -
B: s = re j45- GI(B) = 2ej90 =
2r 2r 2
ej 180 1
C: s = re j90 G,(C)
= 2r 2ej 180 2r 2
We then compute G,(jO.1) = 50ej9 , G,(jl) = 0.6ej710, and G,(jlO) =
0 .0lej 162. From these, we can plot G,(jc)) for w > 0 (including section ABC) as
shown in Figure 8 .16(b) with the solid lines . The plot of G,(jw) for w < 0 is the
reflection, with respect to the real axis, of the plot for w > 0 and is shown in Figure
8.15(b) with the dashed lines . Every point on the large semicircle with R ---> - is
mapped by G,(s) into the origin of the G,plane . Thus the plot in Figure 8 .16(b) is
the conplete Nyquist plot of G,(s) in (8.11). If the Nyquist plot of F(s) = 1 + GI(s)
is desired, we may simply add a set of coordinates as shown in Figure 8.16(b) .

Exercise 8 .4 .2
Plot the Nyquist plots of

1
G,(s) =
and F(s) = 1 + G,(s) .
8.4 .3 Nyquist Stability Criterion
Consider the unity-feedback system shown in Figure 8 .17(a). Its overall transfer
function is
G,(s) _ : G,(s)
GG(s) = (8.13)
3)
+ G,(s) F(s)
THEOREM 8 .1 (Nyquist Stability Criterion)
The G0(s) in (8 .13) is stable if and only if the Nyquist plot of G,(s) does not pass
through critical point (- 1, 0) and the number of counterclockwise encirclements
of (- 1, 0) equals the number of open right-half-plane poles of GI(s).
To prove this theorem, we first show that G0(s) is stable if and only if the Nyquist
plot of F(s) does not pass through the origin of the F-plane and the number of
counterclockwise encirclements of the origin equals the number of open right-half-
plane poles of G,(s) . Clearly G0(s) is stable if and only if F(s) has no closed right-
half-plane zeros . If the Nyquist plot of F(s) passes through the origin of the F-plane,
then F(s) has zeros on the imaginary axis and G9(s) is not stable . We assume in the
following that F(s) has no zeros on the imaginary axis . Let Z and P be, respectively,
the numbers of open right-half-plane zeros and poles of F(s) or, equivalently, the
numbers of zeros and poles of F(s) encircled by C1 . Because F(s) and G 1(s) have
the same denominator, P also equals the number of open right-half-plane poles of
G,(s) . Now the principle of argument states that
N = Z - P
Clearly F(s) has no open right-half-plane zeros, or G O(s) is stable if and only if
Z = 0 or N = - P . Because C 1 is chosen to travel in the clockwise direction, the
stability condition requires the encirclements to be in the counterclockwise direction .
This establishes the assertion.
(a) (b)
Figure 8.17 Unity-feedback systems .

From the discussion in the preceding subsection, the encirclement of the Nyquist
plot of F(s) around the origin of the F-plane is the same as the encirclement of the
Nyquist plot of GI(s) around point (- 1, 0) on the GIplane . This establishes the
theorem. We note that if the Nyquist plot of G,(s) passes through (-1,0), then G0(s)
has at least one pole on the imaginary axis and G0(s) is not stable.
We discuss now the application of the theorem .
Example 8.4.3
Consider the unity-feedback system in Figure 8 .17(a) with G,(s) given in (8.10).
G,(s) has two poles in the open right half plane . The Nyquist plot of G,(s) is shown
in Figure 8 .15. It encircles (- 1, 0) twice in the counterclockwise direction . Thus,
the unity-feedback system is stable . Certainly, this can also be checked by computing
8s
- 1)(s - 2)
G,(s) = G 1 (s) (s
1 + G,(s) 1 + 8s
(s - 1)(s - 2)
8s 8s
(s- 1)(s-2)+8s s 2 +5s+2
which is stable .
Example 8 .4.4
Consider the unity-feedback system in Figure 8 .17(a) with GI(s) given in (8.11) .
GI(s) has one open right-half-plane pole . Its Nyquist plot is shown in Figure 8 .16;
it encircles (- 1, 0) once in the clockwise direction . Although the number of encir-
clements is right, the direction is wrong . Thus the unity-feedback system is not stable .
This can also be checked by computing
G,(s) - s + 1
G ,(s) =
I + GI (s) s3 - 2S2 + S + I
which is clearly not stable .
In application, we may encounter the problem of finding the range of k for the
system in Figure 8 .17(b) or
kG,(s)
G,(s) = (8.14)
1 + kG,(s)
to be stable . Although Theorem 8 .1 can be directly applied to solve the problem, it
is more convenient to modify the Nyquist stability criterion as follows :

THEOREM 8 .2
The G0(s) in (8 .14) is stable if and only if the Nyquist plot of G1(s) does not pass
through the critical point at (- 1/k, 0) and the number of counterclockwise
encirclements of (-1/k, 0) equals the number of open right-half-plane poles
of GI(s).
This theorem reduces to Theorem 8 .1 if k = 1 . The establishment of this theo-

rem is similar to that of Theorem 8 .1 and will not be repeated . We will now discuss
its application.
Example 8 .4.5
Consider the unity-feedback system shown in Figure 8 .17(b) with
_ 8
G, (s) (8 .15)
(s + 1)(s2 + 2s + 2)
Find the stability range of k.

The Nyquist plot of GI(s) is plotted in Figure 8 .18 . G,(s) has no open RHP pole .
Thus the feedback system is stable if and only if the Nyquist plot does not encircle
(- 1/k, 0) . If -1/k lies inside [0, 41, the Nyquist plot encircles it once ; if -1/k
lies inside [-0.8, 01, the Nyquist plot encircles it twice . Thus if -1/k lies inside
[0, 4] or [-0.8, 0], the feedback system is not stable . On the other hand, if
1
- oo < - < - 0.8 (8 .16a)
k
or
1
4 < -k 00 (8 .16b)
1mG(jco)
Re G(jw)
Figure 8 .18 Nyquist plot of (8 .15) .

then the Nyquist plot does not encircle (-1/k, 0) and the feedback system in
Figure 8 .17(b) is stable. Now (8 .16a) implies 0 < k < 5/4 and (8 .16b) implies
0 ? k > - 1/4 . Thus the stability range of the system is
1 5
-4<k<4 (8 .17)
This result is the same as (4 .21) which is obtained by using the Routh test . We
mention that the stability range can also be obtained by using the root-locus method .
This will not be discussed .
We give a special case of Theorem 8 .1 in which loop transfer functions have

no open right-half-plane poles . Poles on the imaginary axis, however, are permitted.
COROLLARY 8 .1
If the loop transfer function G,(s) in Figure 8 .17(a) has no open right-half-plane
poles, then G0 (s) in (8 .13) is stable if and only if the Nyquist plot of G,(s) does
not encircle critical point (- 1, 0) nor passes through it .
Before proceeding, we mention that the Nyquist criterion is applicable to non-

unity-feedback configurations such as the one shown in Figure 8 .19 . Its transfer
function from r to y is
C1(s)G(s)
G ,(s)
0(s) (8 .18)
1 + C,(s)C2(s)G(s)
If we define G,(s) : = C(s)C2(s)G(s), then G0(s) in (8 .18) is stable if and only if the
Nyquist plot of G,(s) does not pass through (- 1, 0) and the number of counter-
clockwise encirclements of ( - 1, 0) equals the number of open right-half-plane poles
of G,(s) .
8 .4 .4 Relative Stability-Gain Margin and Phase Margin

The Routh test is probably the easiest method of checking stability . Therefore, the
purpose of introducing the Nyquist criterion is not so much for checking stability,
but because it will reveal the degree of stability and then provide a method of de-
signing control systems . If the Nyquist plot of G,(s) passes through critical point
G (s)
C2 (s)
Figure 8 .19 Non-unity-feedback system .

(-1, 0) at, say, s= jwo, then the numerator of (1 + GI(s)) equals zero at s = jcoo.
In other words, the numerator has a zero on the imaginary axis and is not Hurwitz .
Consequently, G0 (s) is not stable . Thus, the distance between the Nyquist plot of
GI (s) and critical point (- 1, 0) can be used as a measure of stability of G0(s) .
Generally, the larger the distance, the more stable the system . The distance can be
found by drawing a circle touching the Nyquist plot as shown in Figure 8 .20(a).
Such a distance, however, is not easy to measure and is not convenient for design .
Therefore, it will be replaced by phase margin and gain margin .
Consider the Nyquist plot for w ? 0 or, equivalently, the polar plot shown in
Figure 8 .20 . Let cop > 0 be the frequency at which l G,(jwp) = 180 . This is called
the phase crossover frequency ; it is the frequency at which the polar plot of G,(s)
passes through the negative real axis . The distance between -1 and G,(jcop) as
shown in Figure 8 .20(b) is called the gain margin. The distance, however, is not
measured on linear scale; it is measured in decibels defined as
Gain margin : = 20 log I - 11 - 20 log I G1(jwp)I
(8.19)
-20 log IG,(jwp) I
For example, if G,(jwp) _ -0.5, then the gain margin is +6 dB . If G,(jwp) = 0,
then the gain margin is ~ . Note that if G,(jw ) lies between -1 and 0, then
jGr(jcop) j < 1 and the gain margin is positive . If ~G,(jwp) I > 1, or the polar plot of
GI (jwp) intersects the real axis on the left hand side of -1 as shown in Figure
8.20(c), then the gain margin is negative .
Let cog > 0 be the frequency such that IG I (j(og)I = 1 . It is the frequency at
which the polar plot of G,(s) intersects with the unit circle as shown in Figure 8 .20(b)
and is called the gain crossover frequency . If we draw a straight line from the origin
to GI(jwg), then the angle between the straight line and negative real axis is called
the phase margin . To be more precise, the phase margin is defined as
Phase margin : = I< (-1)I - I < GI(jwg)I = 180 0 - I 1 G,(jwg )I (8.20)
where the phase of GI(jwg) must be measured in the clockwise direction . Note that
if the intersection with the unit circle occurs in the third quadrant as shown in Figure
/OL
(a) (b) (c)

Figure 8 .20 Relative stability .

8.20(b), then the phase margin is positive. If the intersection occurs in the second
quadrant as shown in Figure 8 .20(c), then the phase margin is negative .
The phase and gain margins can be much more easily obtained from the Bode
plot. In fact, the definitions in (8 .19) and (8 .20) are developed from the Bode plot .
Recall that the Bode plot and polar plot differ only in the coordinates and either one
can be obtained from the other . Suppose the polar plots in Figure 8 .20(b) and
(c) are translated into the Bode plots shown in Figure 8 .21 (a) and (b). Then the gain
crossover frequency co g is the frequency at which the gain plot crosses the 20 log
1 = 0-dB line . We then draw a vertical line downward to the phase plot . The distance
in degrees between the - 180 line and the phase plot is the phase margin . If the
phase of G,(jc) g lies above the - 180 line, the phase margin is positive, as shown
)
in Figure 8 .21(a). If it lies below, the phase margin is negative, as shown in Figure
8.21(b) . The phase crossover frequency p is the frequency at which the phase plot
to
crosses the -180 line . We then draw a vertical line upward to the gain plot . The
distance in decibels between the 0-dB line and the gain plot at ce p is the gain margin .
If the gain at (op lies below the 0-dB line, as shown in Figure 8.21(a), the gain margin
is positive. Otherwise it is negative, as shown in Figure 8 .21(b) . Thus, the phase and
gain margins can readily be obtained from the Bode plot .
Although the phase and gain margins can be more easily obtained from the
Bode plot, their physical meaning can be more easily visualized on the Nyquist plot .
For example, if G,(s) has no pole in the open right half plane and has a polar plot
roughly of the form shown in Figure 8 .20, then its Nyquist plot (the polar plot plus
its reflection) will not encircle (- 1, 0) if both the phase and gain margins are
positive . Thus, the overall system G 0 (s) = G,(s)/(1 + G,(s)) is stable . In conclusion,
I G' (iw) I I G,(jw) I
dB dB
Gain Gain
margin crossover
20 20
0 frequency
w/
0.1 x
-180
Phase margin > 0
(a) (b)
Figure 8 .21 Phase margin and gain margin .
if G,(s) has no open right-half-plane pole, then generally the overall system
G0(s) = G,(s)/(1 + G,(s)) is stable if the phase margin and the gain margin of GI(s)
are both positive . Furthermore, the larger the gain and phase margins, the more stable
the system . In design we may require, for example, that the gain margin be larger
than 6 dB and that the phase margin be larger than 30 . If either the phase margin
or gain margin equals 0 or is negative, then the system G 0(s) is generally not stable .
If a loop transfer function G,(s) has open right-half-plane poles, in order for
G0(s) to be stable, the Nyquist plot must encircle the critical point . In this case, the
polar plot may have a number of phase-crossover frequencies and a number of gain-
crossover frequencies as shown in Figure 8 .22(a), thus phase margins and gain mar-
gins are not unique . Furthermore, some phase margins must be positive and some
negative in order for the system to be stable . Thus, the use of phase margins and
gain margins becomes complicated. For this reason, if loop transfer functions have
open right-half-plane poles, the concepts of phase and gain margins are less useful .
Even if loop transfer functions have no pole in the open right half plane, care
must still be exercised in using the phase and gain margins . First, the polar plots of
such transfer functions may not be of the form shown in Figure 8 .20 . They could
have more than one phase margin and/or more than one gain margin as shown in
Figure 8 .22(b) . Moreover, if a polar plot is as shown in Figure 8 .22(c), even though
G,(s) has a large phase margin and a large gain margin, the closed loop system
G0(s) = G,(s)/(1 + G,(s)) has a poor degree of stability. Therefore, the relationship
between the degree of stability and phase and gain margins is not necessarily exact .
Exercise 8 .4.3
Find the gain and phase margins of the following transfer functions :
2
a.
s(s + 1)
5
b.
s(s + 1)(s + 2)
1
C.
s 2 (s + 1)
8 .5 FREQUENCY DOMAIN SPECIFICATIONS FOR OVERALL SYSTEMS
With the preceding background, we are ready to discuss the design problem . The
problem is : given a plant with transfer function G(s), design a feedback system with
transfer function G0(s) to meet a set of specifications . The specifications are generally
stated in terms of position error, rise time, settling time, and overshoot . Because they

8 .5 FREQUENCY-DOMAIN SPECIFICATIONS FOR OVERALL SYSTEMS 301
(a) (b) (c)
Figure 8.22 Applicability of phase and gain margins .
are defined for the time responses of G0(s), they are called time-domain specifica-
tions. Now, if the design is to be carried out using frequency plots, we must translate
the time-domain specifications into the frequency domain . This will be carried out
in this section . Recall that the specifications are stated for the overall transfer function
GO (s), not for the plant transfer function G(s).
Steady-State Performance
Let G0(s) be written as
G0(jw) = IG0(jw)Ie' e(~)

where 6(w) = tan - ' [Im G0(j (o)/Re G0(jw)] . The plot of I G0(j(o)j with respect to
(o is called the amplitude plot and the plot of 6(w) with respect to w is called the
phase plot of G0(s) . Typical amplitude and phase plots of control systems are shown
in Figure 8 .23. From the final-value theorem, we know that the steady-state per-
formance (in the time domain) is determined by G0(s) as s ---> 0 or G0(jw) as
w -> 0 (in the frequency domain) . Indeed, the position error or the steady-state error
due to a step-reference input is, as derived in (6.3),
Position error = ep(t) = 11 - Go(0)I X 100% (8 .21)
Thus from Go(0), the position error can immediately be determined . For example, if
G0(0) = 1, then ep = 0; if Go(0) = 0.95, then the position error is 5% . The velocity
error or the steady-state error due to a ramp-reference input is, as derived in (6 .6),
Velocity error = e (t) = I(1 - Go(0))t - Go(0)1 X 100% (8 .22a)
In order to have finite velocity error, we require G0(0) = 1 . In this case, (8 .22a)
reduces to
Velocity error = vp(t) = IGo(0)I X 100% (8 .22b)

which implies that the velocity error depends only on the slope of G0(jw) at w = 0 .
If the slope is zero, velocity error is zero . Thus the steady-state performance can be
easily translated into the values of G0(jw) and its derivatives at w = 0 .
Transient Performance
The specifications on the transient performance consist of overshoot, settling time,
and rise time. These specifications are closely related to MP and bandwidth shown
in Figure 8 .23. The constant MP, called the peak resonance, is defined as
MP : = max I G0(j(o)I (8 .23)
for (o ? 0. It is the largest magnitude of G0(jw) in positive frequencies . The band-
width is defined as the frequency range in which the magnitude of GO(jw) is equal
to or larger than 0 .707jGo(0)I. The frequency w, with the property IG0(j(qw )j =
0.7071Go(0)I is called the cutoff frequency . If G0(0) = 1 or 0 dB, the amplitude of
G,,(s) at w, is
20 log 0 .707 = - 3 dB
(0
(b)
Figure 8 .23 Typical frequency plot of control systems .

8 .5 FREQUENCY-DOMAIN SPECIFICATIONS FOR OVERALL SYSTEMS 303
and, because the power is proportional to IGo (jw)1 2, the power of G0(s) at ww is
(0.707) 2 = 0.5
thus w, is also called the - 3 dB or half-power point. In conclusion, the cut-off
frequency of G0(s) is defined as the frequency at which its amplitude is 70% or
3 dB below the level at w = 0, or at which the power is half of that at w = 0 .
Now we discuss the relationship between the specifications on transient per-
formance, and the peak resonance and bandwidth . Similar to the development of the
desired pole region in Chapter 7, we consider the following quadratic transfer
function
w2
n (8.24)
G0(s) : _
S 2 + 2,'wns + wn
Clearly we have
w2
n
G0(j (o) =
- w2 + 2gwnw + wn
and
wn
G (jw)12 - (w2 - w2)2 + (2~wn w)2
By algebraic manipulation, it can be shown that
1
for 1 < 0707
.
M,=max I G0(jw)l = 2C'\11 - ~2 (8.25)
1 for ? 0.707
Thus the peak resonance MP depends only on the damping ratio. The same situation
holds for the overshoot as shown in Figure 7.2(b). Now we plot (8 .25) on Figure
7.2(b) to yield Figure 8 .24 . From Figure 8 .24, the relationship between overshoot
and peak resonance can easily be obtained . For example, if we require the overshoot
to be less than 20%, then from Figure 8 .24, we see that Mn must be less than 1 .25.
Similarly, if we require the overshoot to be less than 10%, then we require M P <-
1 .05. Thus, the peak resonance dictates the overshoot .
The bandwidth of G0(s) is related to the speed of response . The larger the band-
width, the faster the response. This rule of thumb is widely accepted in engineering,
even though the exact relationship is not known and the statement may not be true
for every control system . The speed of response here may mean the rise time or
settling time . We give a plausible argument of the statement by using the quadratic
transfer function in (8.24). From the horizontal coordinate wnt in Figure 4 .7, we
argued in Section 7 .2 .1 that the rise time is inversely proportional to wn. The Bode
plot of (8.24) is shown in Figure 8 .9 . The intersection of the - 3-dB horizontal line
with the gain plot yields the cutoff frequency and the bandwidth . Because the hor-

100%
M
L L'
i n
8 2
a
0
a 1
C)
o.
a
Overshoot -
0 I 0
0 .2 0.4 0 .6 0 .8 1 .0
Damping ratio
Figure 8.24 Peak resonance and overshoot .
izontal coordinate is w/o), the bandwidth is proportional to (0,, . Thus we conclude

that the larger the bandwidth, the smaller the rise time or the faster the response .
In addition to bandwidth, we often impose constraint on the amplitude of the
gain plot at high frequencies . For example, we may require the magnitude of G0(jw)
to be less than, say, e for all w > to, as shown in Figure 8 .25. For the same bandwidth,
if wd is closer to the cutoff frequency, then the cutoff rate of the gain plot will be
larger as shown . The implication of this high-frequency specification on the time
response is not clear . However, it has a very simple implication on noise rejection .
Noise or unwanted signals with frequency spectra lying in the range w > w d will be
greatly suppressed or eliminated .
The peak resonance and bandwidth are defined on the magnitude plot of G,(s) .
The phase plot of G,(s) is not used . If the phase plot is not a linear function of w,
then distortion will occur . Distortion of signals in control systems is not as critical
as in signal processing and filter design ; hence, in the design of control systems,
generally no specification is imposed on the phase plot of G 0(s).
We summarize the preceding discussion in Table 8 .1 . The second column lists
the time-domain specifications, the third column lists the frequency-domain speci-
fications . Both are specified for overall systems . The last column of the table will
be developed in the next section .
IG(io)I
w
Figure 8 .25 Two frequency plots with same bandwidth .

8 .6 FREQUENCY-DOMAIN SPECIFICATIONS FOR LOOP TRANSFER FUNCTIONS 305
Table 8.1 Specifications of Control Systems
Time Domain Frequency Domain

Overall System Overall System Loop Transfer Function
Steady-State Position error 11 - Go (0)J I1/(1 + K,)!
Performance Velocity error IG,(0)I if G,(0) = 1 1l/K
Transient Overshoot Peak resonance Gain and phase margins
Performance Rise time Bandwidth Gain crossover
Settling time frequency
High-frequency gain High-frequency gain
8 .6 FREQUENCY DOMAIN SPECIFICATIONS FOR LOOP TRANSFER

FUNCTIONS-UNITY-FEEDBACK CONFIGURATION
With the frequency-domain specifications discussed in the preceding section, the

design problem now becomes : Given a plant with frequency-domain plot G(j(O),
design a system such that the overall transfer function G0 (jw) will meet the speci-
fications . There are two possible approaches to carry out the design . The first ap-
proach is to compute G0(jw) from G(jw) for a chosen configuration . If G0 (j(O) does
not meet the specifications, we then introduce a compensator and recompute G0(jw) .
We repeat the process until G0 (j(o) meets the specifications . The second approach
is to translate the specifications for G0(jw) into a set of specifications for G(jw) . If
G(j(o) does not meet the specifications, we introduce a compensator C(s) so that
C(jw)G(jw) will meet the specifications . Because the first approach is much more
complicated than the second approach, it is rarely used in practice . The method
introduced in the remainder of this chapter takes the second approach .
In order to take this approach, the specifications for the overall system must be
translated into those for G(jw) . We shall do so for the unity-feedback configuration
shown in Figure 8 .1 . Define G,(s) = C(s)G(s) . Then the transfer function of the
unity-feedback system in Figure 8 .1 is
C(s)G(s) G,(s)
(s)
G ,(s) (8 .26)
1 + C(s)G(s) 1 + G,(s)
We call G,(s) the loop transfer function.
Steady-State Performance
Consider the loop transfer function G,(s) . We define
K,, : = lim G,(s) = G,(0) (8 .27a)
S-0
K, : = lim sG,(s) (8 .27b)
s-o

Now we compute steady-state errors in terms of Kp and K,,. The error e = r - y of

the unity-feedback system in Figure 8 .1 can be computed in the Laplace-transform
domain as
E(s) = R(s) - Y(s) = R(s) - G0(s)R(s) (8 .28)
G1(s) 1
_ (1 - GI(s)R(s) =
1 + 1 + G,(s) R(s)
If r(t) = 1 or R(s) = 1/s, then the steady-state error in (8 .28) is the position error
defined in (6 .3).3 Using the final-value theorem, we have
1 1
ep (t) = lim Ie(t)I = lim IsE(s)I = lim s -
t-oo S-o S-0 1 + GI(s))s (8.29)
I 1
x 100%
1 + G1(0) 1 + Kp
We see that the position error depends only on KP, thus KP is called the position-
error constant .
If r(t) = t or R(s) = 1/s 2 , then the steady-state error in (8 .28) is the velocity
error defined in (6 .6) 4 Using the final-value theorem, we have
s 1 1
e(t) = lim Ie(t)j = lim IsE(s)l = lim 1 + GI(s)/ s 2
t__ S-o S-o (8.30)
1 1
= lim x 100%
S-o s + sG,(0) K
Here we have implicitly assumed that e (t) approaches a constant, otherwise the
final-value theorem cannot be applied . We see that the velocity error depends only
on K,,, thus K is called the velocity-error constant . Once the position or velocity
error is specified, we can use (8 .29) or (8 .30) to find the range of Kp or K,,. This
translates the steady-state specifications for overall systems into specifications for
loop transfer functions .
As discussed in Section 6.3.2, the position or velocity error of unity-feedback
systems can be determined from the system type of G,(s) . The loop transfer function
G,(s) is of type i if it has i poles at the origin . For example, G,(s) = 2/(s - 1) and
(s + 2)/(s 2 + s + 10) are of type 0, and their position-error constants are - 2 and
2/10 = 0 .2. Their velocity-error constants are zero . The loop transfer function
G,(s) = 1/s and 1/s(s + 2) are of type 1 . Their position-error constants are infinity,
and their velocity-error constants are 1 and 1/2 = 0 .5 . These are summarized in
Table 8.2 in which k denotes some nonzero constant and Ka is defined as lim s 2 G,(s)
and is called the acceleration-error constant . S-o
'If r(t) = a, then the error must be divided by a. Because a = 1, this normalization is not needed .
4If r(t) = at, then the error must be divided by a. Because a = 1, this normalization is not needed .

8 .6 FREQUENCY-DOMAIN SPECIFICATIONS FOR LOOP TRANSFER FUNCTIONS 307
Table 8.2 Error Constants
KP K K,
Type 0 k 0 0
1
Type 1 00 k 0 eP =
1 + KP
1
Type 2 00 00 k e =
K
Now if G,(s) is of type 1, then KP = oo and ee = 0. Thus the unity-feedback

system in Figure 8 .1 will track any step-reference input without an error . If G, (s) is
of type 2, then K = - and e = 0. Thus the system will track any ramp-reference
input without an error . These are consistent with the conclusions in Section 6 .3.2.
To conclude this part, we mention that (8 .29) and (8 .30) are established for unity-
feedback systems . They are not necessarily applicable to other configurations .
Transient Performances
The transient performance of G 0(s) is specified in terms of the peak resonance MP,
bandwidth, and high-frequency gain . Now we shall translate these into a set of
specifications for G,(s) . To do so, we must first establish the relationship between
G0(jco) and G,(jw) = G(jco)C( jw) . Let the polar plot of GI(s) be as shown in Figure
8 .26(a). Consider the vector G,(j(o t) . Then the vector drawn from (-1, 0) to G1(j(o1)
equals 1 + G,(jwt). Their ratio
G`(jw1)
I + GI(jot) = Go(jw1)
yields G0(jw t ) . Therefore it is possible to translate G,(jto) graphically into G0(jco) .
To facilitate the translation, we first compute the loci on the G,-plane that have
constant IG0(jw)I . Let x + jy be a point of G,(jw) on the G,-plane . Clearly we have
1/2
G,(jco) + jy _ x 2 + y2
G o (J w)j = = x (8 .31
1 + G,(j(o) 1 + x + jy - ((1 + x) 2 + y2
Let IG0(jw)l = M. Then (8.31) implies

x2 + y 2
m2 - y2
(1 +x)2 +
or
(1 + x)2M 2 + y 2M 2 = x2 + y2
'This subsection establishes the last column in Table 8 .1 . It may be skipped without loss of continuity .
Im G i I G0(jw) I
Re G 1
(a) (b)
Figure 8 .26 From G,(j(o) to G,(jco).
which, after some simple algebraic manipulation, becomes

MM2)2 MM2) 2
- 1 + y2 - M
lx C1 /
It is the equation of a circle with radius M/(1 - 2) and centered at
(M 2 /(1 - M 2 ), 0). A family of such circles for various M are plotted in Figure 8 .27.
They are called the constant M-loci. 6 If the constant M-loci are superposed on the
polar plot of G,(s) as shown, then G0 (jw) can be read out directly as shown in Figure
8.26(b) . The circle with the largest M which the polar plot of G l (s) touches tangen-
tially yields the peak resonance MI.
In order to better see the relationship between the peak resonance and phase
and gain margins, we expand part of the plot of Figure 8 .27 in Figure 8 .28 . From
the intersections of the circle of M = 1 .2 with the real axis and the unit circle, we
may conclude that if the gain margin of G,(s) is smaller than 5 .3 dB or the phase
margin of G l (s) is smaller than 50, then the peak resonance of G 0(s) must be at least
1 .2. Conversely, if a polar plot is roughly of the form shown in Figure 8 .28, then
we may have the following :
Gain margin ? 10 dB, Phase margin ? 45 =* Mp -= 1 .3
and
Gain margin ? 12 dB, Phase margin ? 60 = Mp - 1 .0
This establishes a relationship between the peak resonance of G0 (s) and the gain and
phase margins of G I (s).
Using Figure 8 .28, we can also read the bandwidth of G0(s) from the polar plot
of G,(s) . If G0(0) = 1, cutoff frequency w, is the frequency at which the polar plot
intersects with the M-circle of 0 .7. In the example in Figure 8 .28, as w increases,
the polar plot first passes through the unit circle centered at the origin and then
61t is also possible to plot the loci of constant phases of Ge (s) on the G,-plane . The plot consists of a
family of circles called constant N-loci. The plot of constant M- and N-loci on the log magnitude-phase
plot is called the Nichols chart . They are not used in this text and will not be discussed .
Im G(jc))
Re G1 (jco)
Figure 8 .27 Constant M-loci .
Re G1 (jo))
Figure 8.28 Constant M-loci and phase and gain margins .
309

through the M-circle of 0 .7, thus we have

wg < w, (8 .32)
where wg is the gain-crossover frequency of G,(s). Note that w, is defined for G0 (s),
whereas cog is defined for G,(s) . Although (8 .32) is developed for the polar plot in
Figure 8 .28, it is true in general, and is often used in frequency-domain design . For
example, if an overall system is required to respond as fast as possible or, equiva-
lently, to have a bandwidth as large as possible, then we may search for a compen-
sator C(s) so that the loop transfer function G,(s) = C(s)G(s) has a gain-crossover
frequency as large as possible .
The high-frequency specification on G0 (s) can also be translated into that
of G,(s) = C(s)G(s) . If G(s) is strictly proper and if C(s) is proper, then
G 1 (jcI)J << 1, for large w. Thus we have
Gl(jw)
IG0(jw)I = IG,(j(o)I
1 + G,(jw)
for large w. Hence the specification of G0 (jw) < e, for w ? cod, can be translated
to IG,(jw)l < e, for w ? wd.
The preceding discussion is tabulated in the last column of Table 8 .1 . We see
that there are three sets of specifications on accuracy (steady-state performance) and
speed of response (transient performance) . The first set is given in the time domain,
the other two are given in the frequency domain . The first two sets are specified for
overall systems, the last one is specified for loop transfer functions in the unity
feedback configuration . It is important to mention that even though specifications
are stated for loop transfer functions, the objective is still to design a good overall
system.
8.6.1 Why Use Bode Plots?

With the preceding discussion, the design problem becomes as follows : Given a
plant with proper transfer function G(s), find a compensator C(s) in the unity-feed-
back configuration in Figure 8 .1 such that the loop transfer function GI (s) =
C(s)G(s) will meet the specifications on Kp, or K, the phase margin, gain margin,
and gain-crossover frequency . The design is to be carried out using only frequency
plots. This chapter introduced three frequency plots-namely, the polar plot, log
magnitude-phase plot, and Bode plot . These three plots are all equivalent, and the-
oretically any one can be used in the design . In practice, however, the Bode plot is
used almost exclusively, for the following two reasons . First, the Bode plot can be
drawn from its asymptotes ; thus the Bode plot is the easiest to draw among the three
plots. Second and more important, we have
Bode plot of C(s)G(s) = Bode plot of G(s) + Bode plot of C(s)
This applies to the magnitude plot as well as to the phase plot, that is,
20 log IC(jw)G(jw)I = 20 log IC(jw)I + 20 log IG(jw)I

8.6 FREQUENCY-DOMAIN SPECIFICATIONS FOR LOOP TRANSFER FUNCTIONS 311
and
< C(jco)G(j(o) = -~', C(jco) + < G(jw)
Thus in the design, if the Bode plot of G(s) does not meet the specifications, we
simply add the Bode plot of C(s) to it until the sum meets the specifications . On the
other hand, if we use the polar plot of G(s) to carry out the design, the polar plot of
G(s) is of no use in subsequent design because the polar plot of C(s)G(s) cannot
easily be drawn from the polar plot of G(s). Therefore the polar plot is less often
used. In the log magnitude-phase plot, the frequency appears as a parameter on the
plot, thus the summation of C(j(o) and G(jw) involves summations of vectors and
is not as convenient as in the Bode plot. Thus, the Bode plot is most often used in
the frequency-domain design .
8.6.2 Design from Measured Data

If the transfer function of a plant is known, we can compute its Kp or K, plot its
Bode plot, and then proceed to the design . Even if the transfer function is not known,
we can still carry out the design from measured data . This is a unique feature of the
frequency-domain method. Instruments are available to measure the frequency plot,
in particular, the Bode plot of plants . Once a Bode plot is obtained by measurement,
we can then read out its phase margin and gain margin . Its position- or velocity-
error constants can also be read out from the plot . At very low frequencies, if the
slope of the gain plot is zero, as shown in Figure 8 .29(a), the transfer function is of
type 0 ; if the slope is - 20 dB/decade, it is of type 1 ; if the slope is - 40 dB/decade,
it is of type 2; and so forth . For a type 0 transfer function, we have, at very low
frequencies or at w = 0,
20 log IG(j(o)I = 20 log IKPJ
Therefore, if we extend the horizontal line to intersect with the vertical coordinate
at, say, a dB, then we have
20 log IKj = a
dB dB
-20 db/decade
20 log I Kv I = a
a
~ w i =gv
I >' a)
0 .1 1 10 wl
-40 db/decade
(a) (b)
Figure 8.29 Bode plots .

which implies KP = 10'20. Thus the position-error constant can be easily obtained
from the plot.
Every type 1 transfer function can be expressed as
k(1 + bls)(1 + b2s)
G (s)
s(1 + ais)(1 + a2s)(1 + a3s) . . .
Clearly, its position-error constant is infinity . At very low frequencies, the Bode gain
plot is governed by
k k K
20 log IG(j(o)l - 20 log = 20 log = 20 log (8 .33)
Jw w w
This is a straight line with slope -20 dB/decade . We extend the straight line to
intersect with the vertical axis at, say, a dB and intersect with the horizontal axis at,
say, w l radians per second . The vertical axis passes through w = 1, thus (8 .33)
becomes
K
20 log a
1
which implies K = 10 / 20. The gain of (8.33) is 0 dB at w = w l , thus we have
K
0 = 20 log
w1
which implies K = w 1. Thus, the velocity-error constant of type 1 transfer functions
can also be easily obtained from the Bode gain plot . In conclusion, from the leftmost
asymptote of Bode gain plots, the constants KP and K can be easily obtained . Once
KP, K, the phase margin, and the gain margin are read out from the measured data,
we can then proceed to the design.
8 .7 DESIGN ON BODE PLOTS
Before discussing specific design techniques, we review the problem once again .
Given a plant with transfer function G(s), the objective is to design an overall system
to meet a set of specifications in the time domain . Because the design will be carried
out by using frequency plots, the specifications are translated into the frequency
domain for the overall transfer function G0(s), as shown in Table 8 .1 . For the unity-
feedback configuration shown in Figure 8 .1, the specifications for G0(jw) can be
further translated into those for the loop transfer function G1(s) = G(s)C(s) as shown
in Table 8 .1 . Therefore the design problem now becomes : Given a plant with Bode
plot G(jw), find a compensator C(s) in Figure 8 .1 such that the Bode plot of C(s)G(s)
will meet the specifications on position- or velocity-error constant, phase margin,
gain margin, gain-crossover frequency, and high-frequency gain . If this
is successful, all we can hope for is that the resulting overall system G0 (s) =
G1(s)/(1 + G,(s)) would be a good control system . Recall that the translations of
8. 7 DESIGN ON BODE PLOTS 313
the specifications in Table 8 .1 are developed mainly from quadratic transfer func-
tions; they may not hold in general . Therefore, it is always advisable to simulate the
resulting overall system to check whether it really meets the design specifications .
The search for C(s) is essentially a trial-and-error process . Therefore we always
start from a simple compensator and, if we are not successful, move to a more
complicated one . The compensators used in this design are mainly of the following
four types : (a) gain adjustment (amplification or attenuation), (b) phase-lag compen-
sation, (c) phase-lead compensation, and (d) lag-lead compensation. Before pro-
ceeding, we mention a useful property . Consider the Bode plots shown in Figure
8.30. It is assumed that the plant has no open right-half-plane poles nor open right-
half-plane zeros . Under this assumption, the phase can be estimated from the slopes
of the asymptotes of the gain plot . If a slope is -20 dB/decade, the phase will
approach -90. If a slope is -40 dB/decade, the phase will approach - 180 . If a
slope is - 60 dB/decade, the phase will approach - 270 . Because of this property,
if the slope of the asymptote at the gain-crossover frequency is - 60 dB/decade, as
shown in Figure 8 .30(a), then the phase will approach - 270 and the phase margin
will be negative . Consequently the feedback system will be unstable . On the other
hand, if the slope of the asymptote at the gain-crossover frequency is -20
dB/decade as shown in Figure 8 .30(b), then the phase margin is positive . If the slope
of the asymptote at the gain-crossover frequency is -40 dB/decade, the phase
margin can be positive or negative . For this reason, if it is possible, the asymptote
at the gain-crossover frequency is designed to have slope - 20 dB/decade . This is
the case in almost every design in the remainder of this chapter .
dB dB
a) w
-270
(a) (b)
Figure 8 .30 Bode plots .

8.7 .1 Gain Adjustment

The simplest possible compensator C(s) is a gain k. The Bode gain plot of
C(s)G(s) = kG(s) is 20 log Jki + 20 log JG(jro)J . Thus, the introduction of gain
k will simply shift up the Bode gain plot of G(s) if JkJ > 1, and shift it down if
JkJ < 1 . If gain k is positive, its introduction will not affect the phase plot of G(s).
For some problems, it is possible to achieve a design by simply shifting a Bode gain
plot up or down .
Example 8 .7.1
Consider the unity-feedback system shown in Figure 8 .31 . Let the plant transfer
function be G(s) = 1/s(s + 2) and let the compensator C(s) be simply a constant
k. Find a gain k such that the loop transfer function kG(s) will meet the following :
1. Position error <_ 10% .
2. Phase margin > 60, gain margin ? 12 dB .
3. Gain-crossover frequency as large as possible .
The plant is of type 1, thus its position-error constant KP is infinity and its
position error 11/(1 + KP)I is zero for any k. The Bode plot of G(s) = l/s(s + 2)
or
1 0.5
G(s) _ _
2s1 +~sl s(1 + I
C S)
is shown in Figure 8.32 with the solid lines . The gain plot crosses the 0-dB line
roughly at 0.5; thus the gain-crossover frequency is cog = 0 .5 rad/s . The phase
margin is then measured from the plot as 76 . To find the gain margin, we must first
find the phase-crossover frequency . Because the phase approaches the -180 line
asymptotically as shown, it intersects the line at w = -, and the phase-crossover
frequency w, is infinity . The gain plot goes down to -c dB with slope
-40 dB/decade as co - - . Thus, the gain margin at top = oo is infinity. Thus, the
Bode plot of G(s) meets the specifications in (1) and (2) . If we do not require the
specification in (3), then there is no need to introduce any compensator and the
design is completed. It is important to point out that no compensator means C(s) _
k = 1 and that the unity feedback is still needed as shown in Figure 8 .31.
Compensator Plant
C(s) 1
s(s+2)

8 .8 PHASE-LAG COMPENSATION 315
10 20 50 100
o .l o .z I
I nun
I II Compensated 1
-40 db/decade
Uncompensated
I 1
0
I I
I
I
0 .5 2 10 20 50 100
0
I I I II I I I I I I I
I1I I I I I I I II I
111)-w
I ~
I
I
Q = 1 .15
-90 I
76
-180 1 60
Figure 8 .32 Bode plot of 1/s(s + 2) .
Now we shall adjust k to make the gain-crossover frequency as large as possible .

If we increase k from 1, the gain plot will move upward and the gain-crossover
frequency will increase (shift to the right) . This will cause the phase margin to
decrease, as can be seen from Figure 8 .32 . In other words, as the gain-crossover
frequency increases, the phase margin will decrease . To find the largest permissible
k, we draw a horizontal line with 60 phase margin . Its intersection with the phase
plot yields the largest permissible gain-crossover frequency, which is read from the
plot as cog = 1 .15 . If we draw a vertical line upward from W,,' = 1 .15, then we can
read from the gain plot that if we add 8 dB to the gain plot or shift the gain plot up
8 dB, then the new gain-crossover frequency will be shifted to cog = 1 .15 . Thus the
required gain k is
20 log k = 8
which implies k = 2.5. Note that the gain margin remains infinity for k = 2.5 . Thus
the design is completed by introducing k = 2 .5.
This problem was also designed using the root-locus technique in Section 7 .2.
The result was k = 2 . For this problem, the root-locus and frequency-domain meth-
ods yield comparable results .
8 .8 PHASE-LAG COMPENSATION
First we use the example in Figure 8 .31 to show that adjustment of a gain alone
sometimes cannot achieve a design .

Example 8 .8.1
Consider a plant with transfer function G(s) = 1/s(s + 2) . Find a compensator C(s)
in Figure 8 .1 such that C(s)G(s) will meet the following : (1) velocity error < 10%,
(2) phase margin ? 60, and (3) gain margin ? 12 dB .
First we choose C(s) = k and see whether or not the design is possible . The
loop transfer function is
k
G,(s) = kG(s) (8.34)
= s(s + 2)
It is of type 1 and its velocity-error constant is
k
K = lim sG,(s) _ - (8.35)
S-o 2
In order to meet (1), we require
1 2
:5:: 0.1 (8.36)
K~, k
dB
Figure 8 .33 Bode plot of 20/s(s + 2) .

8.8 PHASE-LAG COMPENSATION 317
which implies k ? 20. We choose k = 20 and plot the Bode plot of

1 10
kG(s) = 20 (8 .37)
- + 2) = 1
s 1 + 2s
in Figure 8 .33 . We plot only the asymptotes . We see that the gain-crossover fre-
quency roughly equals wg = 4.2 rad/s, and the phase margin roughly equals 26 .
The phase-crossover frequency is wp = cc, and the gain margin is infinity . Although
the gain margin meets the specifications, the phase margin does not . If we restrict
C(s) to be k, the only way to increase the phase margin is to decrease k. This,
however, will violate (1). Thus for this problem, it is not possible to achieve the
design by adjusting k alone .
Now we introduce a compensating network that is more complex than a gain .

Consider the network shown in Figure 8 .34(a). Its transfer function is
1
R2 + -
_ E2(s) Cs _ 1 + R2Cs 1 + aT1s
C, (s) (8.38)
E, 1 + (R, + R2)CS I + T, s
(s) R 1 + R2
+ Cs
where
R
0< a := < 1 (8.39)
R 1 + R2
and
T, : = (R, + R 2 )C (8.40)
The pole and zero of C 1 (s) are -1/T, and -1/aT 1. Because a < 1, the zero is
farther away from the origin than the pole, as shown in Figure 8 .34(b) . For any
w ? 0, the phase of CI(s) equals 0 - (P. Because 4i ? 0, the phase of C 1(s) is
negative for all w ? 0. Thus the network is called a phase-lag network .
Im s
(a) (b)
Figure 8.34 Phase-lag network.

dB
1 I ,'
T1 aT1 ,'
log w
0 -20 dB d/ ecade
90 - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
10
log w
1\ 5 .7
-90
Figure 8.35 Bode plot of (8.38).
Now we plot the Bode plot of C,(s). The corner frequency of the pole is 1/T, .
We draw two asymptotes from 1/T,, one with slope -20 dB/decade . The corner
frequency of the zero is 1/aT, . We draw two asymptotes from 1/aT,, one with slope
20 dB/decade . Note that the corner frequency 1/aT, is on the right-hand side of the
corner frequency 1/T,, because a < 1 . The summation of these yields the Bode gain
plot of Cl(s) as shown in Figure 8 .35. For w very small, the Bode gain plot is a
horizontal line with gain I or 0 dB . For w very large, C,(s) can be approximated by
(1 + aT,s)/(1 + Ts) .- aTis/T,s = a; thus its Bode gain plot is a horizontal line
with gain a or 20 log a dB. Because a < 1, 20 log a is a negative number . Thus the
phase-lag network introduces an attenuation of 20 Ilog al . The Bode phase plot of
C, (s) can be similarly obtained as shown in Figure 8 .35 . We see that the phase is
negative for all w as expected .
A phase-lag network has two parameters a and T,- The amount of attenuation
introduced by the network is determined by a . In employing a phase-lag network,
we use mainly its attenuation property. The phase characteristic will not be used
except the phase at 10/aT,, ten times the right-hand-side corner frequency . From
Figure 8 .6, we see that the phase at 10/aT, is at most 5 .7. Now we shall redesign
the problem in Example 8.8.1.
Example 8 .8.1 (continued)

Consider a plant with transfer function G(s) = 1/s (s + 2) . Find a compensator C(s)
in Figure 8 .1 such that C(s)G(s) will meet (1) velocity error <_ 10%,(2) phase margin
60, and (3) gain margin ~ 12 dB .

8 .8 PHASE-LAG COMPENSATION 319
Now we choose C(s) as kC,(s), where C,(s) is given in (8.38) . Because

C, (0) = 1, the velocity-error constant K of C(s)G(s) is
k
K = lim sG,(s) = lim skC,(s)G(s) _ -
s-o s-o 2
Thus, in order to meet (1), we require, as in (8 .36), k ? 20. The Bode plot of
1
20
s(s + 2)
is plotted in Figure 8 .33 with the solid lines . Its gain-crossover frequency is cog =
4 .5 rad/s and its phase margin is 25 . The phase margin, however, is required to be
at least 60 . This will be achieved by introducing a phase-lag network . First we
search for a new gain-crossover frequency that has a phase margin of 60 plus 6
(this will be explained later), or 66 . This can be found by drawing a horizontal line
with 66 phase margin . Its intersection with the phase plot yields the new gain-
crossover frequency. It is read from Figure 8 .33 as cog' = 0 .9. We then draw a vertical
line upward to the gain plot. We see that if the gain is attenuated by 20 dB at
to << cog', then the new gain plot will pass through cog' = 0.9 at the 0-dB line . A
phase-lag network will introduce an attenuation of 20 Ilog al, thus we set
20 log a = - 20
which implies a = 0 .1 .
The remaining problem is to determine T, in (8.38) . If the right-hand-side corner
frequency 1/aT, is close to cog, the phase margin at cog' will be greatly reduced by
the phase of the network . If the corner frequency is very far away from cog (on the
left-hand side), then .the phase margin will be hardly affected by the phase of the
network . Therefore it seems that we should choose T, so that 1 /aT, is as small as
possible. However, this will cause a different problem. If 1/aT, is very small, then
C(s) has a pole very close to the origin, as shown in Figure 8 .34(b) . If we plot the
root locus of C(s)G(s), then we can see that the closed-loop system will have a pole
very close to the origin and the corresponding time constant will be very large .
Therefore, if 1 /aT, is very small, the speed of response of the resulting unity-feed-
back system may become slow . For this reason, we do not want to place 1 /aT, too
far away from cog . In practice, 1/aT, is often placed at one decade below wg ; that
is,
1 wg1
aTl 10
In this case, the phase-lag network has at most a phase lag of 5.7 at cog, as shown
in Figure 8 .35 . Thus the phase of G(s) at cog' will be reduced by roughly 6 after
introducing C(s) . This is the reason for adding 6 to the required phase margin in

determining cog.' For this problem, we have

10 10 _
aTt=a) ,= 0 -11 .1
a .9
and
11 .1 11.1 _
T` a 0 .1 = 111
Thus the compensator is
1+aT ts 1+11 .ls
C(s) = kC1(s) = 20 . (8.41)
1 + T ts = 20 1 + Ills
With this compensator, the Bode plot of C(s)G(s) will meet the specifications
on velocity error and phase margin . Now we check whether it will meet the speci-
fication on gain margin . From Figure 8 .33, we see that the phase-crossover frequency
is cop = oo and the gain margin is infinity. Thus, C(s)G(s) meets all specifications,
and the design is completed .
We summarize the procedure of designing phase-lag networks in the following .

Step 1 : Compute the position- or velocity-error constant from the specification on
steady-state error .
Step 2 : Plot the Bode plot of the plant with the required error constant . Measure
the phase margin and gain margin from the plot .
Step 3 : If we decide to use a phase-lag compensator, determine the frequency at
which the Bode plot in Step 2 has the required phase margin plus 6 . This
frequency is designated as the new gain-crossover frequency cog .
Step 4: Measure the attenuation needed to bring the gain plot down to cog. This
attenuation will be provided by a phase-lag network . Let this attenuation
be a. Compute a from a = - 20 log a or a = 10 - "~ 20
Step 5: Compute Tt from
1 wg
(8 .42)
aTt - 10
that is, the corner frequency 1/aT 1 is placed one decade below the new
gain-crossover frequency wg.
Step 6: If the resulting system satisfies all other specifications, then the design is
completed . The compensating network can be realized as shown in Figure
8 .34(a).
7The design on Bode plots is carried out mostly by measurements . It is difficult to differentiate between
5 .7 and 6 on a plot . Therefore we need not be concerned too much about accuracy in this method.

8 .9 PHASE-LEAD COMPENSATION 321
To conclude this section, we remark that the use of a phase-lag network will
reduce the gain-crossover frequency . Consequently the bandwidth of the unity-feed-
back system may be reduced . For this reason, this type of compensation will make
a system more sluggish .
8 .9 PHASE-LEAD COMPENSATION
In this section we introduce a network that has a positive phase for every w > 0 .
Consider the network shown in Figure 8 .36(a) . It is built by using two resistors, one
capacitor, and an amplifier . The impedance of the parallel connection of R 1 and the
capacitor with impedance 1/Cs is
1
R1
Cs R1
R1Cs + 1
R + 1
1 Cs
Thus the transfer function from e1 to e2 in Figure 8 .36(a) is
E2(s) Rl + R2 R2
C2(s) : _ _
El(s) R2
R + R1
2 R 1 Cs + I (8 .43)
_ R1 + R2 R2+ R 1R2Cs 1 + bT2s
R2 R1 + R2 + R1R2Cs 1 + T2s
where
b= R1+R2 >1 (8 .44)

R2
and
+ R2 . C
T2 = R1R1R2 (8 .45)
Im s
(a) (b)
Figure 8 .36 Phase-lead network .

The pole and zero of C 2(s) are plotted in Figure 8 .36(b) . The phase of C 2 (s) equals
0 - 0 as shown. Because 0 > 0 for all positive w, the phase of C 2 (s) is positive.
Thus it is called a phase-lead network. Note that C 2(0) = 1 .
The corner frequency of the zero is 1/bT 2 ; the corner frequency of the pole is
1/T 2 . Because b > 1, l/bT 2 < 1/T 2 and the corner frequency of the zero is on the
left-hand side of that of the pole . Thus the Bode gain plot of C 2 (s) is as shown
in Figure 8 .37. The gain at low frequencies is 1 or 0 dB . For large w, we have
C2(s) .- bT2 s/T2 s = b, thus the gain is b or 20 log b dB as shown . Unlike the phase-
lag network, the phase of the phase-lead network is essential in the design, therefore
we must compute its phase . The phase of (1 + bT2s)/(1 + T 2s) at s = jw equals
bT2 w - T2 w
P(w) = tan -i bT2w - tan -i T2 w = tan-i
t
1 + bTZw2
Thus we have
bT2 w - T2w
tan COO = (8.46)
1 + bTZw2
Since the phase plot is symmetric with respect to the midpoint of 11bT2 and 1/T2, as
shown in Figure 8.37, the maximum phase occurs at the midpoint . Because of the
logarithmic scale, the midpoint w,,, is given by
log to,, log

2 2 2 2
or
1
(8.47)
V bT2
dB
20 log b /
10 log b i
log w
1 (t)
m 1
bT2 T2
90 0
00 log w
1
m
Vb T 2
-900 ----- - - - - - - --- -- - - -- --- - - - - - -
Figure 8 .37 Bode plot of (8.43).

Thus the maximum phase at am equals, substituting (8 .47) into (8.46),
(b - 1)
tan
(b - 1)TZ wm V _ b - 1
`Ym
= 1 + bTzwm 1 + 1 2V
which implies
,~, b- I l+sin4
sin Ì'm and b (8 .48)
= b + 1 = 1 - sin
We see that the larger the constant b, the larger the maximum phase 0m . However,
the network in Figure 8 .36 also requires a larger amplification . In practice, constant
b is seldom chosen to begreater than 15 . We mention that the gain equals 10 log b
at w = wm , as shown in Figure 8 .37 .
The philosophy of using a phase-lead network is entirely different from that of
using a phase-lag network . A phase-lag network is placed far away from the new
gain-crossover frequency so that its phase will not affect seriously the phase margin.
A phase-lead network, on the other hand, must be placed so that its maximum phase
will contribute wholely to the phase margin . Therefore, com should be placed at the
new gain-crossover frequency . To achieve this, however, is not as simple as in the
design of phase-lag networks . The procedure of designing phase-lead networks is
explained in the following :
Step 1 : Compute the position-error or velocity-error constant from the specifica-
tion on steady-state error .
Step 2: Plot the Bode plot of kG(s), the plant with the required position- or veloc-
ity-error constant. Determine the gain-crossover frequency co g and phase-
crossover frequency w,,. Measure the phase margin 0, and gain margin
from the plot .
Step 3 : If we decide to use a phase-lead compensator, calculate qf = (required
phase margin) - cbl . The introduction of a phase-lead compensator will
shift the gain-crossover frequency to the right and, consequently, decrease
the phase margin . To compensate for this reduction, we add 0, say 5 , to
qi. Compute 0m = i/r + 0.
Step 4: Compute constant b from (8.48), which yields phase 4m .
Step S. If we place this maximum phase at wg or, equivalently, set wm = wg,
because the network has positive gain, the gain-crossover frequency of
C2,(s)G(s) will be shifted to the right and the maximum phase will not
appear at the new gain-crossover frequency. For this reason, we must com-
pute first the new gain-crossover frequency before placing w m . We draw a
horizontal line with gain - 10 log b. Its intersection with the Bode gain
plot of kG(s) yields the new gain-crossover frequency, denoted by w9' .
Measure the phase margin 42 of kG(s) at this frequency. If 0, - 02 > 0,
choose a larger 0 in Step 3 and repeat Steps 4 and 5 . If 0, - 42 < 0, go
to the next step.

Step 6: Set to. = to' and compute T 2 from (8.47). If the resulting system satisfies
all other specifications, the design is completed. The network can then be
realized as shown in Figure 8 .36(a).
Example 8 .9.1
We shall redesign the system discussed in the preceding section by using a phase-
lead network. Consider a plant with transfer function G(s) = 1/s(s + 2) . Find a
compensator C(s) in Figure 8 .1 such that C(s)G(s) will meet (1) velocity error <_
10%, (2) phase margin ? 60, and (3) gain margin ? 12 dB.
Now we shall choose C(s) as kC2(s), where C 2 (s) is given in (8 .43). Because
C2(0) = 1, the velocity-error constant K0 of C(s)G(s) is
k
K = lim sG,(s) = lim skC2(s)G(s) _
s-o s-o 2
Thus we require k ? 20 in order to meet the specification in (1). The Bode plot of
kG(s) = 20/s(s + 2) is plotted in Figure 8.38 with the solid lines . The phase margin
tp, is 26. The required phase margin is 60. Thus we have = 60 - 26 = 34.
If we introduce a phase-lead network, the gain-crossover frequency will increase
and the corresponding phase margin will decrease . In order to compensate for this
reduction, we choose arbitrarily 0 = 5 . Then we have tp~ = 34 + 5 = 39. This
is the total phase needed from a phase-lead network.
dB
co'x = 6 .8
WR = 6.3 Phase-lead compensator
I
10 ! 20 50 100
Figure 8 .38 Phase-lead compensation .

Next we use (8 .48) to compute b as

+ sin 39 _ 1.63
1
.4
b - 1 - sin 390 0.37 - 4
We then draw a horizontal line with - 10 log b = - 10 log 4.4 = - 6.44 dB as
shown in Figure 8 .38. Its intersection with the gain plot yields the new gain-crossover
frequency as cog' = 6.3. The corresponding phase margin is then read from the plot
as 02 = 18. Because Y'1 - Y'2 = 26 - 18 = 8, which is larger than 0 = 5,
the choice of 0 = 5 is not satisfactory . As a next try, we choose 0 = 9 . Then we
have 471 = 34 + 9 = 43 and the corresponding b is computed from (8.48) as
5.9. Next we draw a horizontal line with - 10 log b = -10 log 5 .9 = - 7 .7 dB .
Its intersection with the gain plot yields ii = 6.8 rad/s . The phase margin at ceg is
17 . The reduction of the phase margin is 26 - 17 = 9 which equals 0. Thus the
choice of 0 = 9 is satisfactory and the corresponding b = 5 .9 can be used in the
design . Now we set wm = (sg' or, using (8 .47),
con, = Cog 6.8

V bT2
which implies
1 1 1
2 = .06
~Wg 5.9x6 .8 _ 16 .5 = 0
Thus the required phase-lead compensator is
+ bT2 s _ 1 + 0.36s
CZ(s)
2(S) = I + T2s 1 + 0.06s
and the total compensator is
00.36(s
.36(s + 2.8) + 2.8
C(s) = kC2(s) = 20 . .9) = 120 . s (8.49)
0.06( + 16 S + 16.7
This completes the design .
The introduction of phase-lead compensators will increase the gain-crossover

frequency and, consequently, the bandwidth of the overall system . This, in turn, will
increase the speed of response of the resulting system . Before proceeding, we com-
pute the root loci and step responses of the preceding three designs in Figure 8 .39.
The design using the gain adjustment alone (shown with the solid line) has a very
large overshoot, although its phase margin is 60 and gain margin is infinity . This
is not entirely surprising, because the relationships among the overshoot, damping
factor, peak resonance, phase margin, and gain margin are only approximate . The
design using a phase-lag network (shown with the dashed line) is rather sluggish ;
the settling time of the system is very large . The design using a phase-lead network

Gain Phase-lag Phase-lag
C(s)=20 . 1+11 .1s C(s)=20 . I 1+0 .36s

C(s) = 2.5 I + Ills +0 .06s
Pole of -1 j1 .2 -0.98 -3 .2
G0 (s) -0.96 j O .96 -7 .7 j6.6
Root loci Im S Im s Im s
0 .09
-0.009 I
-0 .961 -7 .9 1 -2 .78
X - Re s )( I 0 )- Re s )( I \o) - Re s
-1 0 -2 1 -16 .67 1 -2
Step 1 .6
response
1 .4 Phase lead
Phase lag PI compensator
1 .2
0.8
0.6
0.4
0.2
0~
0 4 4 6 8 10 12 14 16 18 20
Figure 8 .39 Various designs.
(shown with the dotted line) has the smallest overshoot among the three designs ; its
rise time and settling time are also smallest . Therefore it is the best design among
the three .
The phase-lead compensation does not always yield a good design . For example,
if the phase in the neighborhood of the gain-crossover frequency decreases rapidly,
then the reduction of the phase margin due to a phase-lead compensator will offset
the phase introduced by the compensator . In this case, the specification on phase
margin cannot be met . Thus, the phase-lead compensation is not effective if the
phase at the gain-crossover frequency decreases rapidly . To conclude this section,
we compare phase-lag and phase-lead networks in Table 8 .3.

8 .10 PROPORTIONAL-INTEGRAL(PI) COMPENSATORS 327
Table 8.3 Comparisons of Phase-Lead and Phase-Lag Networks
Phase-Lag Network Phase-Lead Network

1 . The pole is closer to the origin than the The zero is closer to the origin than the
zero . Its phase is negative for every pole . Its phase is positive for every
positive w. positive w .
2. Shifts down the gain-crossover Shifts up the gain-crossover frequency ;
frequency ; consequently, decreases the consequently, increases the bandwidth
bandwidth and the speed of response . and the speed of response .
3. Placed one decade below the new gain- Placed over the new gain-crossover
crossover frequency to reduce the effect frequency to add the maximum phase
of the network on the phase margin. on the phase margin.
4. No additional gain is needed . Additional gain is needed .
5. Design can be achieved in one step . Design may require trial and error .
In addition to the phase-lag and phase-lead networks, we may also use a network
with transfer function
1 + aT1s 1 + bT2s
C3(s) _ I + T,s I + T 2 s
in the design . The transfer function is the product of the transfer function of a phase-
lag network and that of a phase-lead network . Thus it is called a lag-lead network .
In the design, we use the attenuation property of the phase-lag part and the positive
phase of the phase-lead part . The basic idea is identical to those discussed in the
preceding two sections and will not be repeated .
8.10 PROPORTIONAL-INTEGRAL (PI) COMPENSATORS
A compensator that consists of a gain (also called a proportional compensator) and

an integral compensator is called a PI compensator or controller . The transfer func-
tion of an integral compensator with gain k. is ki/s . Thus, the transfer function of PI
compensators is
ki ki + ks ki(l + cs)
C3(s) = k + - _ _ (8.50)
s s s
where c = k/k i. This is a special case of the phase-lag network shown in Figure
8 .34(b) with the pole located at the origin . The phase of C3(s) is clearly negative for
all positive to. We shall use this compensator to redesign the problem in the preceding
two sections . Although PI controllers are a special case of phase-lag networks, the
procedure for designing phase-lag networks cannot be used here . Instead we will
use the idea in designing phase-lead networks in this problem .

Example 8 .10.1
Consider a plant with transfer function G(s) = 1/s(s + 2) . Find a PI compensator
C3 (s) in Figure 8.1 such that C 3 (s)G(s) will meet (1) velocity error <_ 10%, (2) phase
margin ? 60, and (3) gain margin ? 12 dB.
The transfer function of C3 (s)G(s) is
k (l + cs) 1 _ k (1 + cs)
C3(s)G(s) = (8 .51)
s s(s + 2) 2s2(l + 0.5s)
which has two poles at s = 0 . Thus it is of type 2, and the velocity error is zero for
any ki ( see Section 6.3.2). The same conclusion can also be reached by using Table
8.2. For a type 2 transfer function, the velocity-error constant K is infinity . Thus,
its velocity error 1/KI is zero for any ki.
We first assume c = 0 and plot in Figure 8.40 the Bode plot of
k,
(8.52)
+ 2 s~
2s2
(1
with k;/2 = 1 . The gain-crossover frequency is co g = 1 and the phase margin can
be read from the plot as -27. Because the phase plot approaches the - 180 line
at co = 0, the phase-crossover frequency is cop = 0 and the gain margin is negative
infinity . Changing ki will shift the gain plot up or down and, simultaneously, shift
the gain-crossover frequency to the right or left . However, the phase margin is always
0) A
P
U)
0 .1 0 .2 1 2 10 20
__90 1
180
-27 --270o
Figure 8 .40 Bode plot of 1/s 2(1 + 0 .5s).

8 .10 PROPORTIONAL-INTEGRAL (PI) COMPENSATORS 329
negative and the gain margin remains negative infinity . Thus, if c = 0, the unity-
feedback system is unstable for any k ; and the design is not possible .
If k, > 0 and c > 0, the Bode phase plot of C3(s)G(s) is given by
+ Cs
(C3(s)G(s)) = (s2) + (11 + 0.5s) (8 .53)
(1 +Cs
-180 +
1 + 0.5s )
If c < 0 .5, then (1 + cs)/(1 + 0 .5s) is a phase-lag network and its phase is negative
for all positive w . In this case, the phase in (8 .53) is always smaller than -180.
Thus, the phase margin is negative, and the design is not possible . If c > 0 .5, then
(1 + cs)/(1 + 0 .5s) is a phase-lead network, and it will introduce positive phases
into (8.53). Thus the design is possible. We write
l+cs 1+cX0.5s
(8 .54)
1 + 0.5s 1 + 0.5s
with c > I and compare it with (8.43), then c = b and (8 .48) can be used to compute
c. Now the phase margin of C3(s)G(s) is required to be at least 60, therefore we
have
- - I + sin 60 1 + 0 .866
= 13 .9
1 - sin 60 1 - 0.866
Substituting this c into (8 .54) yields
l+cs 1+13.9X0 .5s 1+6.95s
1 + 0.5s 1 +0.5s 1 + 0.5s
and C3(s)G(s) in (8 .51) becomes
k,(21 + 6.95s)
C3 (s)G(s) = 2s (1 + 0 .5s) (8 .55)
The corner frequencies of the pole and zero of the phase-lead network are, respec-
tively, 1/0 .5 = 2 and 1/6 .95 = 0 .14 . Thus the maximum phase of the network
occurs at
w,,,=V0 .14X2=10 .28=0.53 (8 .56)
and equals 60. Figure 8 .41 shows the Bode plot of (8 .55) with k,/2 = 1 . We plot
only the asymptotes for the gain plot . Because the phase of 1/s 2 is -180, the phase
of (8 .55) is simply the phase of (1 + 6 .95s)/(1 + 0 .5s) shifted down by 180 . From
the plot, we see that the gain-crossover frequency is roughly 3 radians per second .
If we draw a vertical line down to the phase plot, the phase margin can be read out
as 25. This is less than the required 60 . Thus if k,12 = 1 or k, = 2, C 3 (s)G(s) in
(8.55) is not acceptable .

If we increase k the gain plot will move up, the gain-crossover frequency will
shift to the right, and the corresponding phase margin will decrease . On the other
hand, decreasing ki will move down the gain plot and shift the gain-crossover fre-
quency to the left . Thus the phase margin will increase . Now c in the phase-lead
network (1 + cs)/(1 + 0 .5s) is chosen so that the phase is 60 at Wm = 0.53 . Thus,
we shall choose ki so that the gain-crossover frequency equals The gain at W,
is roughly 22 dB, so we set
20 log 2 = - 22
(
which implies
= 10 -22/20 = 0.08
k`
2
a If we choose ki = 2 X 0.08 = 0.16, then C3(s)G(s) in (8 .55) will meet the speci-
fication on phase margin .
The phase plot approaches the -180 line at to = 0 and W = oc, as shown in
Figure 8 .41 . Thus, there are two phase-crossover frequencies . Their corresponding
gain margins are, respectively, -- and co . It is difficult to see the physical meaning
of these gain margins from Figure 8 .41, so we shall plot the Nyquist plot of
C3(s)G(s), where the phase and gain margins are originally defined . Figure 8 .42
0)
0 .1 0 .14 1 1 2 3 10
I I
I --20 1
I
I I
I
I 1
I I
I 4 I
0 .53 I ^ 1
U)
0.1 I 1 1 10
I 1
1 --90 I
y I
y 25
60
-180 t
Figure 8 .41 Bode plot of (1 + 6 .95s)/s 2 (l + 0.5s).
8 .10 PROPORTIONAL-INTEGRAL (PI) COMPENSATORS 331
Im Im
s-plane
Figure 8 .42 Nyquist plot of C 3 (s)G(s) in (8 .57) .
shows the Nyquist plot of (8.55) with k i = 0.16 or

0 .16(1 + 6.95s)
C3 (s)G(s) = (8.57)
2s (1 + 0.5s)
We see that the Nyquist plot encircles the critical point (- 1, 0) once in the coun-
terclockwise direction and once in the clockwise direction . Therefore, the number
of net encirclements is zero . The loop transfer function in (8 .57) has no open right-
half-plane pole, so we conclude from the Nyquist stability criterion that the unity-
feedback system in Figure 8 .1 is stable . From the Nyquist plot, we see that the phase
margin is 60. There are two gain-crossover frequencies with gain margins oo and
- -. If there are two or more phase-crossover frequencies, we shall consider the one
that is closest to (- 1, 0) . In this case, we consider the phase-crossover frequency
at w = -; its gain margin is -. Thus C3(s)G(s) also meets the requirement on gain
margin and the design is completed. In conclusion, if we introduce the following PI
compensator
C3(S) = 0.16(1 + 6.95s)

S
the unity-feedback system in Figure 8 .1 will meet all specifications . The response
of this system is shown in Figure 8 .39 with the dashed-and-dotted lines . Its response
is slower than the one using the phase-lag compensator and much slower than the
one using the phase-lead network . Therefore, there is no reason to restrict compen-
sators to PI form for this problem .

We give a number of remarks to conclude this chapter .

1. An important feature of the Bode-plot design method is that it can be carried
out from measured data without knowing transfer functions . All other methods
discussed in this text require the knowledge of transfer functions to carry out
design .
2. The design is carried out on loop transfer functions . In order to do so, specifi-
cations for overall systems are translated into those for loop transfer functions
in the unity-feedback configuration . Thus the method is applicable only to this
configuration.
3. The relationships between the specifications such as rise time and overshoot for
overall systems and the specifications such as phase and gain margins for loop
transfer functions are not exact . Therefore, it is advisable to simulate the re-
sulting systems after the design .
4. If a plant transfer function has open right-half-plane poles, then its Bode plot
may have two or more phase and gain margins . In this case, the use of phase
and gain margins becomes complex . For this reason, the Bode-plot design
method is usually limited to plants without open right-half-plane poles .
5. In this chapter, we often use asymptotes of Bode gain plots to carry out the
design . This is done purposely because the reader can see better the plots and
the design ideas . In actual design, one should use more accurate plots . On the
other hand, because the relationships between phase and gain margins and time
responses are not exact, design results using asymptotes may not differ very
much from those using accurate plots .
6. The method is a trial-and-error method . Therefore, a number of trials may be
needed to design an acceptable system .
7. In the Bode-plot design method, the constraint on actuating signals is not con-
sidered. The constraint can be checked only after the completion of the design .
If the constraint is not met, we may have to redesign the system .
8. The method is essentially developed from the Nyquist plot, which checks
whether or not a polynomial is Hurwitz. In this sense, the design method con-
siders only poles of overall systems . Zeros are not considered .
PROBLEMS
8.1 . Plot the polar plots, log magnitude-phase plots, and Bode plots of
10 20
and
(s - 2) s(s + 1)
8.2. Plot the Bode plots of the following transfer functions :
s + 5
a. G(s) = s(s +
1)(s + 10)

PROBLEMS 333
s-5
b.G(s)=
s(s + 1)(s + 10)
10
c. G (s) =
(s + 2)(s 2 + 8s + 25)
8.3. a. Consider the Bode gain plot shown in Figure P8 .3. Find all transfer func-
tions that have the gain plot .
CO
Figure P8.3
b. If the transfer functions are known to be stable, find all transfer functions
that have the gain plot .
c. If the transfer functions are known to be minimum phase, find all transfer
functions that have the gain plot .
d. If the transfer functions are known to be stable and of minimum phase, find
all transfer functions that have the gain plot . Is this transfer function unique?
8.4. Consider the three Bode plots shown in Figure P8 .4. What are their transfer
functions?
dB dB
Phase Phase
A
0 10 100 1000
CO
(a) (b)
Figure P8 .4

dB
-180,
(c)
Figure P8 .4 (Continued)
8 .5 . Consider
k(1 + 0.5s)
G(s) = s(bs + 1)(s + 10)
Its Bode plot is plotted in Figure P8 .5 . What are k and b?
dB
(0
90 1 -
0 .1 0.2 1
(0
10 20 100
Figure P8 .5

PROBLEMS 335
8.6. A typical frequency response of a Fairchild A741 operational amplifier is

shown in Figure P8 .6. What is its transfer function?
IA(jw)I A(jc))
0
1 10 10 6 Hz
Figure P8 .6
8.7. Use the Nyquist criterion to determine the stability of the system shown in
Figure 8 .17(a) with
2s + 10
a. G 1(s) =
(s + 1)(s - 1)
2s + 1
b. G 1 (s) = s(s
- 1)
100(s + 1)
c. G i (s) = s(s
- 1)(s + 10)
8.8. Consider the unity-feedback system shown in Figure 8 .17(b) . If the polar plot
of G1(s) is of the form shown in Figure P8 .8, find the stability range for the
following cases :
a. G 1(s) has no open right-half-plane (RHP) pole and zero .
Figure P8 .8
b . G I (s) has one open RHP zero, no open RHP pole .

c. G,(s) has one open RHP pole .
d. G I(s) has two open RHP poles, one open RHP zero .
e. G,(s) has three open RHP poles.
8.9 . Find the stability range of the system in Figure 8 .17(b) with
s + 2
G,(s) -_
(s2 + 3s + 6 .25)(s - 1)
using (a) the Routh test, (b) the root-locus method and (c) the Nyquist stability
criterion .
8.10. What are the gain-crossover frequency, phase-crossover frequency, gain mar-
gin, phase margin, position-error constant, and velocity-error constant for each
of the transfer functions in Problem 8 .2?
8.11 . Repeat Problem 8 .10 for the Bode plots shown in Fig . P8 .4.
8.12 . Consider the unity-feedback system shown in Figure 8.1 . The Bode plot of the
plant is shown in Figure P8 .12. Let the compensator C(s) be a gain k. (a) Find
the largest k such that the phase margin is 45 degrees . (b) Find a k such that
the gain margin is 20 dB .
-90
Figure P8.12
8.13 . Consider the system shown in Figure 8 .1 . The Bode plot of the plant is shown
in Figure P8 .13. The compensator C(s) is chosen as a gain k. Find k to meet
(1) phase margin ?60, (2) gain margin ? 10 dB, and (3) position error S25% .

PROBLEMS 337
dB
40
20
0
w
10 100
8
0.1
0
--90
-180 -
--270
Figure P8 .13
8.14. The Bode plot of the plant in Figure 8 .1 is shown in Figure P8 .14 . Find
a phase-lag network to meet (1) phase margin ?60 and (2) gain margin
?10 dB .
0 .1 1 10 100
w
0
-90_
Figure P8 .14
8.15. Consider the control of the depth of a submarine discussed in Problem 7 .8.
Design an overall system to meet (1) position error :10%, (2) phase margin
?60, and (3) gain margin >_ 10 dB . Compare the design with the one in Prob-
lem 7 .8 .
8.16 . Consider the ship stabilization problem discussed in Problem 7 .14 . Design an
overall system to meet (1) position error X15%, (2) phase margin ?60,
(3) gain margin ? 10 dB, and (4) gain-crossover frequency ? 10 rad/s .
8.17 . Consider the problem of controlling the yaw of the airplane discussed in Prob-
lem 7 .15. Design an overall system to meet (1) velocity error <_10%, (2) phase
margin j30%, (3) gain margin ?6 dB, and (4) gain-crossover frequency as
large as possible .
8.18. Consider the plant transfer function given in Problem 7 .16, which is factored
as
300
G(s) =
s(s + 0.225)(s + 3 .997)(s + 179 .8)
Design an overall system to meet (1) position error = 0, (2) phase margin
?55, (3) gain margin >6 dB, and (4) gain-crossover frequency is not smaller
than that of the uncompensated plant . [Hint: Use a lag-lead network .]
8.19 . a. Plot the Bode plot of
s - 1
G(s) = s
2(s + 10)
What are its phase margin and gain margin?

b. Compute Ge(s) = G(s)/(1 + G(s)) . Is G0(s) stable?
c. Is it always true that the unity-feedback system is stable if the phase and
gain margins of its loop transfer function are both positive?
The Inward
Approach-Choice
of Overall Transfer
Functions
9.1 INTRODUCTION
In the design of control systems using the root-locus method or the frequency-domain
method, we first choose a configuration and a compensator with open parameters .
We then search for parameters such that the resulting overall system will meet design
specifications . This approach is essentially a trial-and-error method ; therefore,
we usually choose the simplest possible feedback configuration (namely, a unity-
feedback configuration) and start from the simplest possible compensator-namely,
a gain (a compensator of degree 0) . If the design objective cannot be met by searching
the gain, we then choose a different configuration or a compensator of degree 1
(phase-lead or phase-lag network) and repeat the search. This approach starts from
internal compensators and then designs an overall system to meet design specifica-
tions; therefore, it may be called the outward approach.
In this and the following chapters we shall introduce a different approach, called
the inward approach. In this approach, we first search for an overall transfer function
to meet design specifications, and then choose a configuration and compute the
required compensators . Choice of overall transfer functions will be discussed in this
chapter. The implementation problem-namely, choosing a configuration and com-
puting the required compensators-will be discussed in the next chapter .
Consider a plant with proper transfer function G(s) = N(s)/D(s) as shown in
Figure 9 .1 . In the inward approach, the first step is to choose an overall transfer
function G0(s) from the reference input r to the plant output y to meet a set of
339

340 CHAPTER 9 THE INWARD APPROACH-CHOICE OF OVERALL TRANSFER FUNCTIONS
_
-1
i
r I Y
G (s) T- ~
i __L~
i i
i G o (s) i
L---_ J
Figure 9.1 Design of control systems .
specifications . We claim that

G0(s) = 1
is the best possible system we can design . Indeed if G0(s) = 1, then y(t) = r(t), for
t ? 0 . Thus the position and velocity errors are zero ; the rise time, settling time, and
overshoot are all zero . Thus no other G 0(s) can perform better than G 0 (s) = 1 . Note
that although r(t) = y(t), the power levels at the reference input and plant output
are different . The reference signal may be provided by turning a knob by hand ; the
plant output y(t) may be the angular position of an antenna with weight over several
tons .
Although G0(s) = 1 is the best system, we may not be able to implement it in
practice . Recall from Chapter 6 that practical constraints, such as proper compen-
sators, well-posedness, and total stability, do exist in the design of control systems .
These constraints impose some limitations in choosing G0(s). We first discuss this
problem .
9.2 IMPLEMENTABLE TRANSFER FUNCTIONS

N(s)
G(s)
D(s)
where N(s) and D(s) are two polynomials and are assumed to have no common
factors . We assume n = deg D(s) ? deg N(s), that is, G(s) is proper and has degree
n. An overall transfer function G0(s) is said to be implementable if there exists a
configuration such that the transfer function from the reference input r to the plant
output y in Figure 9 .1 equals G 0(s) and the design meets the following four
constraints :
1. All compensators used have proper rational transfer functions .
2. The resulting system is well-posed .
3. The resulting system is totally stable .
4. There is no plant leakage in the sense that all forward paths from r to y pass
through the plant.
The first constraint is needed, as discussed in Section 5 .4, for building compen-
sators using operational amplifier circuits . If a compensator has an improper transfer
function, then it cannot be easily built in practice. The second and third constraints

9 .2 IMPLEMENTABLE TRANSFER FUNCTIONS 341
are needed, as discussed in Chapter 6, to avoid amplification of high-frequency noise

and to avoid unstable pole-zero cancellations . The fourth constraint implies that all
power must pass through the plant and that no compensator be introduced in parallel
with the plant. This constraint appears to be reasonable and seems to be met by every
configuration in the literature. This constraint is called "no plant leakage" by
Horowitz [35] .
If an overall transfer function G0(s) is not implementable, then no matter what
configuration is used to implement it, the design will violate at least one of the
preceding four constraints . Therefore, in the inward approach, the G0(s) we choose
must be implementable .
The question then is how to tell whether or not a G0(s) is implementable . It turns
out that the answer is very simple .
THEOREM 9.1
Consider a plant with proper transfer function G(s) = N(s)/D(s) . Then G0 (s) is
implementable if and only if G0(s) and
G0(s)
T(s) :
= G (s)
are proper and stable .
We discuss first the necessity of the theorem . Consider, for example, the con-
figuration shown in Figure 9 .2 . Noise, which may enter into the intput and output
terminals of each block, is not shown . If the closed-loop transfer function from r to
y is G0(s) and if there is no plant leakage, then the closed-loop transfer function from
r to u is T(s). Well-posedness requires every closed-loop transfer function to be
proper, thus T(s) and G0(s) must be proper. Total stability requires every closed-loop
transfer function to be stable, thus G0(s) and T(s) must be stable. This establishes
the necessity of the theorem . The sufficiency of the theorem will be established
constructively in the next chapter .
C3 C4
u
CI C2 G (s)
Cs C7 C6
Figure 9 .2 Feedback system without plant leakage.

We discuss now the implication of Theorem 9 .1. Let us write

G(s) = N(s) G,(s) = N,(s) T(s) = Nr(s)
D(s) D0(s) Dr(s)
We assume that the numerator and denominator of each transfer function have no
common factors . The equality G0(s) = G(s)T(s) or
N0(s) _ N(s) N,(s)
D0(s) D(s) Dr(s)
implies
deg D0 (s) - deg N0 (s) = deg D(s) - deg N(s) + (deg Dr (s) - deg Ne(s))
Thus if T(s) is proper, that is, deg Dr(s) ? deg Ne(s), then we have
deg D0 (s) - deg N0(s) ? deg D(s) - deg N(s) (9.1)
Conversely, if (9.1) holds, then deg Dr (s) >- deg Ne (s), and T(s) is proper.
Stability of G0(s) and T(s) requires both D0 (s) and Dr(s) to be Hurwitz . From
= G,(s) = N,(s) D(s)
T(s) = Ne(s)
Dr(s) G(s) D0(s) N(s)
we see that if N(s) has closed right-half-plane (RHP) roots, and if these roots are not
canceled by N0(s), then Dr(s) cannot be Hurwitz . Therefore, in order for T(s) to be
stable, all the closed RHP roots of N(s) must be contained in N0 (s) . This establishes
the following corollary .
COROLLARY 9 .1
Consider a plant with proper transfer function G(s) = N(s)/D(s) . Then G0(s) _
N0(s)/D 0 (s) is implementable if and only if
(a) deg D0(s) - deg N0(s) ? deg D(s) - deg N(s) (pole-zero excess inequality) .
(b) All closed RHP zeros of N(s) are retained in N0(s) (retainment of non-
minimum-phase zeros) .
(c) D0(s) is Hurwitz .
As was defined in Section 8 .3.1, zeros in the closed RHP are called non-mini-
mum-phase zeros . Zeros in the open left half plane are called minimum-phase zeros .
Poles in the closed RHP are called unstable poles . We see that the non-minimum-
phase zeros of G(s) impose constraints on implementable G 0(s) but the unstable
poles of G(s) do not . This can be easily explained from the unity-feedback config-
uration shown in Figure 9 .3. Let
G (s) = N(s) C(s) = Nc(s)
D(s) Dr (s)
be respectively the plant transfer function and compensator transfer_ function . Let

r + e U Y
C(s) G (s)
Figure 9.3 Unity-feedback configuration .
G0 (s) = N0 (s)/D 0(s) be the overall transfer function from the reference input r to
the plant output y . Then we have
N0(s) C(s)G(s) N(s)Nc (s)
G ,(s)
o(s)
D0(s) 1 + C(s)G(s) D(s)D,(s) + N(s)Nc(s) I921
We see that N(s) appears directly as a factor of N0(s). If a root of N(s) does not
appear in N0(s), the only way to achieve this is to introduce the same root in D(s)Dc(s)
+ N(s)NN,(s) to cancel it . This cancellation is an unstable pole-zero cancellation if
the root of N(s) is in the closed right half s-plane . In this case, the system cannot be
totally stable and the cancellation is not permitted . Therefore all non-minimum-phase
zeros of G(s) must appear in N0(s). The poles of G(s) or the roots of D(s) are shifted
to D(s)D,(s) + N(s)N N(s) by feedback, and it is immaterial whether D(s) is Hurwitz
or not . Therefore, unstable poles of G(s) do not impose any constraint on G0(s), but
non-minimum-phase zeros of G (s) do. Although the preceding assertion is developed
for the unity-feedback system shown in Figure 9 .3, it is generally true that, in any
feedback configuration without plant leakage, feedback will shift the poles of the
plant transfer function to new locations but will not affect its zeros . Therefore the
non-minimum-phase zeros of G(s) impose constraints on G0(s) but the unstable poles
of G(s) do not .
Example 9 .2.1
Consider
+ 2)(s - 1)
G(s) - (s
s(s - 2s + 2)
Then we have
G0(s) = 1 Not implementable, because it violates (a) and (b) in Corollary 9 .1 .
s + 2
G 1) Not implementable, meets (a) and (c) but violates (b).
(s) = (s + 3)(s +
s - 1
G0 Not implementable, meets (a) and (b), violates (c).
(s) = s(s + 2)
s - 1
G0(s) = Implementable .
(s + 3)(s + 1)

S - 1
G0 Implementable.
(s) = (s + 3)(s + 1)2
(2s - 3)(s -1)
Qs) = Implementable .
(s+2)
- 3)(s - 1)(s + 1)
o(s) =
G ,(s) ( s + 2)5 Implementable .
Exercise 9.2 .1
Given G(s) = (s - 2)/(s - 3)2, are the following implementable?

1 s-2 s - 2 (s - 2)(s - 3)
a b c. d.
. s+ 1 .s+1 (s+1)2 (s+ 1)3
[Answers : No, no, yes, yes .]
Exercise 9 .2.2
Given G(s) = (s + 1)/(s - 3)2, are the following implementable?

1 s - 2 s - 2 (s-2)s 4
a b c d
. s + 1 . s + 1 . (s + 1)2 . (s + 2)6
[Answers : Yes, no, yes, yes .]
From the preceding examples, we see that if the pole-zero excess inequality is
met, then all poles and all minimum-phase zeros of G0(s) can be arbitrarily assigned .
To be precise, all poles of G0(s) can be assigned anywhere inside the open left half
s-plane (to insure stability) . Other than retaining all non-minimum-phase zeros of
G(s), all minimum-phase zeros of G0(s) can be assigned anywhere in the entire
s-plane . In the assignment, if a complex number is assigned as a zero or pole, its
complex conjugate must also be assigned . Otherwise, the coefficients of G 0 (s) will
be complex, and G0(s) cannot be realized in the real world . Therefore, roughly speak-
ing, if G0(s) meets the pole-zero excess inequality, its poles and zeros can be arbi-
trarily assigned .
Consider a plant with transfer function G(s). The problem of designing a system
so that its overall transfer function equals a given model with transfer function G m(s)
is called the model-matching problem . Now if Gm(s) is not implementable, no matter
how hard we try, it is not possible to match Gm(s) without violating the four con-
straints . On the other hand, if Gm(s) is implementable, it is possible, as will be shown
in the next chapter, to match Gm(s). Therefore, the model-matching problem is the
same as our implementability problem . In conclusion, in model matching, we can

arbitrarily assign poles as well as minimum-phase zeros so long as they meet the
pole-zero excess inequality.
To conclude this section, we mention that if G o is implementable, it does not
mean that it can be implemented using any configuration. For example, G0(s) =
1/(s + 1) 2 is implementable for the plant G(s) = 1/s(s - 1). This G0(s), however,
cannot be implemented in the unity-feedback configuration shown in Figure 9 .3 ; it
can be implemented using some other configurations, as will be discussed in the
next chapter . In conclusion, for any G(s) and any implementable G0(s), there exists
at least one configuration in which Gs(s) can be implemented under the preceding
four constraints .
9.2.1 Asymptotic Tracking and Permissible Pole-Zero

Cancellation Region
A control system with overall transfer function
/ i t s + 02s22 + . . .+
G0(s) = 00 + R mS (9 .3)
ao + ats + a2 5 + + as"
with an > 0 and n ? m, is said to achieve asymptotic tracking if the plant output
y(t) tracks eventually the reference input r(t) without an error, that is,
lim ly(t) - r(t)I = 0
t__
Clearly if G0(s) is not stable, it cannot track any reference signal . Therefore, we
require G0(s) to be stable, which in turn requires a ; > 0 for all i t . Thus, the denom-
inator of G0(s) cannot have any missing term or a term with a negative coefficient .
Now the condition for GO (s) to achieve asymptotic tracking depends on the type of
r(t) to be tracked . The more complicated r(t), the more complicated G 0(s). From
Section 6 .3.1, we conclude that if r(t) is a step function, the conditions for G0(s) to
achieve tracking are G0(s) stable and ao = /3 0. If r(t) is a ramp function, the con-
ditions are G0(s) stable, a0 = (30; and a t = (3t . Ifr(t) = ate , an acceleration function,
then the conditions are G0(s) stable, a0 = 60 , at = at, and a 2 = N2 If r(t) = 0,
the only condition for y(t) to track r(t) is G0(s) stable. In this case, the output may
be excited by nonzero initial conditions, which in turn may be excited by noise or
disturbance. To bring y(t) to zero is called the regulating problem . In conclusion,
the conditions for G0(s) to achieve asymptotic tracking are simple and can be easily
met in the design.
Asymptotic tracking is a property of G 0(s) as t or a steady-state property
of G0(s). It is not concerned with the manner or the speed at which y(t) approaches
r(t) . This is the transient performance of G0(s). The transient performance depends
on the location of the poles and zeros of G0(s). How to choose poles and zeros to
meet the specification on transient performance, however, is not a simple problem .
'Also, they can all be negative . For convenience, we consider only the positive case .

In choosing an implementable overall transfer function, if a zero of G(s) is not

retained in G0 (s), we must introduce a pole to cancel it in implementation . If the
zero is a non-minimum-phase zero, the pole that is introduced to cancel it is not
stable and the resulting system will not be totally stable . If the zero is minimum
phase but has a large imaginary part or is very close to the imaginary axis, then, as
was discussed in Section 6 .6 .2, the pole may excite a response that is very oscilla-
tory or takes a very long time to vanish . Therefore, in practice, not only the non-
minimum-phase zeros of G(s) but also those minimum-phase zeros that are close to
the imaginary axis should be retained in G0(s), or the zeros of G(s) lying outside the
region C shown in Figures 6 .13 or 7 .4 should be retained in G 0(s). How to determine
such a region, however, is not necessarily simple . See the discussion in Chapter 7 .
Exercise 9.2 .3
What types of reference signals can the following systems track without an error?
s + 5
a
. s3 +2s2 +8s+5
8s + 5
b. s3
+2s2 +8s+5
2s2 +9s+68
C.
s3 +2s2 +9s+68
[Answers : (a) Step functions . (b) Ramp functions . (c) None, because it is not
stable . ]
9 .3 VARIOUS DESIGN CRITERIA
The performance of a control system is generally specified in terms of the rise time,
settling time, overshoot, and steady-state error . Suppose we have designed two sys-
tems, one with a better transient performance but a poorer steady-state -performance,
the other with a poorer transient performance but a better steady-stage performance .
The question is : Which system should we use? This difficulty arises from the fact
that the criteria consist of more than one factor. In order to make comparisons, the
criteria may be modified as
J : = kl X (Rise time) + k2 X (Settling time)
(9 .4)
+ k 3 X (Overshoot) + k4 X (Steady-state error)
where the k, are weighting factors and are chosen according to the relative importance
of the rise time, settling time, and so forth. The system that has the smallest J is
called the optimal system with respect to the criterion J. Although the criterion is

9.3 VARIOUS DESIGN CRITERIA 347
reasonable, it is not easy to track analytically . Therefore more trackable criteria are
used in engineering .
We define
e(t) : = r(t) - y(t)
It is the error between the reference input and the plant output at time t as shown
in Figure 9 .4. Because an error exists at every t, we must consider the total error in
[0, oc) . One way to define the total error is
J l e(t)dt
:= (9.5)
fo
This is not a useful criterion, however, because of possible cancellations between
positive and negative errors . Thus a small Jl may not imply a small e(t) for all t. A
better definition of the total error is
J2 (9 .6)
:=
fo e(t)I dt
This is called the integral of absolute error (IAE) . In this case, a small J2 will imply
a small e(t) . Other possible definitions are
J3 = fo le(t)I2 dt (9 .7)
Figure 9 .4 Errors .

and
J4 : = f0 tl e(t)I dt (9 .8)
The former is called the integral of square error (ISE) or quadratic error, and the
latter the integral of time multiplied by absolute error (ITAE) . The ISE penalizes
large errors more heavily than small errors, as is shown in Figure 9 .4. Because of
the unavoidable large errors at small t due to transient responses, it is reasonable not
to put too much weight on those errors . This is achieved by multiplying t with I e(t)I .
Thus the ITAE puts less weight on e(t) for t small and more weight on e(t) for t
large . The total errors defined in J 2 , J3, and J4 are all reasonable and can be used in
design .
Although these criteria are reasonable, they should not be used without consid-
ering physical constraints . To illustrate this point, we consider a plant with transfer
function G(s) = (s + 2)/s(s + 3). Because G(s) has no non-minimum-phase zero
and has a pole-zero excess of 1, G0 (s) = a/(s + a) is implementable for any positive
a. We plot in Figure 9 .5(a) the responses of G 0(s) due to a unit-step reference input
for a = I (solid line), a = 10 (dashed line), and a = 100 (dotted line) . We see that
the larger a is, the smaller J2, J3 , and J4 are . In fact, as a approaches infinity, J2, J3 ,
and J4 all approach zero . Therefore an optimal implementable G 0 (s) is a/(s + a)
with a = -.
As discussed in Section 6 .7, the actuating signal of the plant is usually limited
by
~u(t)I < M for all t ? 0 (9 .9)
This arises from limited operational ranges of linear models or the physical con-
straints of devices such as the opening of valves or the rotation of rudders . Clearly,
the larger the reference input, the larger the actuating signal . For convenience, the
u(t) in (9.9) will be assumed to be excited by a unit-step reference input and the
constant M is proportionally scaled . Now we shall check whether this constraint will
be met for all a. No matter how G0(s) is implemented, if there is no plant leakage,
the closed-loop transfer function from the reference input r to the actuating signal u
is given by
T(s) G (s)) (9.10)
= G
If r is a step function, then the actuating signal u equals
0(s) 1 = a(s + 3)
U(s) = T(s)R(s) = G (9.11)
G(s) s (s + 2)(s + a)
This response is plotted in Figure 9 .5(b) for a = 1, 10, and 100. This can be obtained
by analysis or by digital computer simulations . For this example, it happens that
Iu(t)I max = u(0) = a . For a = 100, u(0) is outside the range of the plot . We see

9 .3 VARIOUS DESIGN CRITERIA 349
1 .2
a = 100
0 .8
0 -t
0 .5 1 1 .5 2 2 .5 3 3 .5 4
(a)
d(t)
10
9
2
a = 1
0
0.5 1 1 .5 2 2 .5 3 3 .5 4
(b)
Figure 9 .5 (a) Step responses . (b) Actuating signals .
that the larger a is, the larger the magnitude of the actuating signal . Therefore if a
is very large, the constraint in (9.9) will be violated.
In conclusion, in using the performance indices in (9.6) to (9.8), we must include
the constraint in (9.9). Otherwise we can make these indices as small as desired and
the system will always be saturated . Another possible constraint is to limit the band-
width of resulting overall systems . The reason for limiting the bandwidth is to avoid
amplification of high-frequency noise . It is believed that both constraints will lead
to comparable results . In this chapter we discuss only the constraint on actuating
signals .

9.4 QUADRATIC PERFORMANCE INDICES
In this section we discuss the design of an overall system to minimize the quadratic
performance index
jo [y(t) - r(t)] 2 dt (9 .12a)
subject to the constraint

u(t)I < M (9 .12b)
for all t ? 0, and for some constant M. Unfortunately, no simple analytical method
is available to design such a system . Furthermore, the resulting optimal system may
not be linear and time-invariant . If we limit our design to linear time-invariant sys-
tems, then (9 .12) must be replaced by the following quadratic performance index
J o [q(y(t) - r(t))2 + u2(t)] dt (9 .13)

= f
where q is a weighting factor and is required to be positive . If q is a large positive
number, more weight is placed on the error . As q approaches infinity, the contribution
of u in (9 .13) becomes less significant, and at the extreme, (9 .13) reduces to (9.7).
In this case, since no penalty is imposed on the actuating signal, the magnitude of
the actuating signal may become very large or infinity ; hence, the constraint in
(9 .12b) will be violated . If q = 0, then (9 .13) reduces to
-
f u2 (t)dt
0
and the optimal system that minimizes the criterion is the one with u =- 0 . From
these two extreme cases, we conclude that if q in (9.13) is adequately chosen, then
the constraint in (9 .12b) will be satisfied . Hence, although we are forced to use the
quadratic performance index in (9 .13) for mathematical convenience, if q is properly
chosen, (9 .13) is an acceptable substitution for (9.12).
9.4 .1 Quadratic Optimal Systems

G( s ) = N(s) (9 .14)
D(s)
It is assumed that N(s) and D(s) have no common factors and deg N(s) < deg D(s)
= n. The design problem is to find an overall transfer function to minimize the
quadratic performance index
J = fo [q(y(t) - r(t)) 2 + u2(t)]dt (9 .15)

9 .4 QUADRATIC PERFORMANCE INDICES 351
where q is a positive constant, r is the reference signal, y is the output, and u is the
actuating signal . Before proceeding, we first discuss the spectral factorization .
Consider the polynomial
Q(s) : = D(s)D( - s) + qN(s)N( - s) (9 .16)
It is formed from the denominator and numerator of the plant transfer function and
the weighting factor q. It is clear that Q(s) = Q(-s) . Hence, if s, is a root of Q(s),
so is -s 1 . Since all the coefficients of Q(s) are real by assumption, if s i is a root of
Q(s), so is its complex conjugate s* . Consequently all the roots of Q(s) are symmetric
with respect to the real axis, the imaginary axis, and the origin of the s-plane, as
shown in Figure 9 .6 . We now show that Q(s) has no root on the imaginary axis .
Consider
Q(jw) = D(jw)D( - jw) + gN(jw)N( - jw) (9.17)
= ID( j (0)1 2 + ql N(j(0 )1 2
The assumption that D(s) and N(s) have no common factors implies that there exists
no coo such that D(jcoo) = 0 and N(jwo) = 0 . Otherwise s2 + wo would be a
common factor of D(s) and N(s) . Thus if q 0 0, Q(jw) in (9.17) cannot be zero for
any w . Consequently, Q(s) has no root on the imaginary axis . Now we shall divide
the roots of Q(s) into two groups, those in the open left half plane and those in the
open right half plane . If all the open left-half-plane roots are denoted by D0(s), then,
because of the symmetry property, all the open right-half-plane roots can be denoted
by D,(- s). Thus, we can always factor Q(s) as
Q(s) = D(s)D(- s) + gN(s)N(- s) = D0(s)D0(- s) (9.18)
where D0(s) is a Hurwitz polynomial . The factorization in (9 .18) is called the spectral
factorization .
With the spectral factorization, we are ready to discuss the optimal overall trans-
fer function . The optimal overall transfer function depends on the reference signal
r(t) . The more complicated r(t), the more complicated the optimal overall transfer
function . We discuss in the following only the case where r(t) is a step function .
Im s
X d X
0
I X X >< >< I >- Re s
-c -b -a a b c
X -d X
Figure 9 .6 Distribution of the roots of Q(s) in (9.16).

Problem Consider a plant with transfer function G(s) = N(s)/D(s), as shown in

Figure 9.1, where N(s) and D(s) have no common factors and deg N(s) < deg D(s)
= n. Find an implementable overall transfer function G0(s) to minimize the quadratic
performance index
J = fo [q(y(t) - r(t))2 + U2 (t)] dt
where q > 0, and r(t) = 1 for t ? 0, that is, r(t) is a step-reference signal .
Solution First we compute the spectral factorization :
Q(s) : = D(s)D( - s) + qN(s)N( - s) = D0(s)D,(-s)
where D0(s) is a Hurwitz polynomial . Then the optimal overall transfer function is
given by
G0(s) = qN(0) . N(s) (9 .19)

Do(0) D0(s)
The proof of (9.19) is beyond the scope of this text ; its employment, however,
is very simple. This is illustrated by the following example .
Example 9 .4 .1
G(s) _ N(s) 1
(9 .20)
D(s) s(s + 2)
Find G0 (s) to minimize
T = fo [9(y(t) - 1)2 + u 2 (t)] dt (9 .21)
Clearly we have q = 9,
D(s) = s(s + 2) D(-s) = -s(-s + 2)
and
N(s) = 1 N(- s) = 1
We compute
Q(s) : = D(s)D( - s) + qN(s)N( - s)
= s(s + 2)( - s)( - s + 2) + 9 . 1 . 1 (9 .22)
_ -s 2(-s 2 +4)+9=s 4 -4s 2 +9

It is an even function of s . If terms with odd powers of s appear in Q(s), an error

must have been committed in the computation . Using the formula for computing the
roots of quadratic equations, we have
4V16-4 . 9 = 4j 20
s2 = 2 2 =2j =3e + j e
with
B = tan-' = 48
2
-)
Thus the four roots of Q(s) are
u 3ej0/2 = V 3ej 24 _\ej24 = V 3ej(180+24) _ \ ej204-
\e-j24 -N/3e j24 _ N/-3ej(180-24) _ \ ej156
as shown in Figure 9.7. The two roots in the left column are in the open right half
s-plane; the two roots in the right column are in the open left half s-plane . Using the
two left-half-plane roots, we form
D(s) _ (s + \ej24 )(s + \e - j24 )
S2 + '\13(e j24' + e-j240)s + 3 (9 .23)
=s 2 +2 N/3(cos24 )s+3=s 2 +3.2s+3
This completes the spectral factorization . Because q = 9, N(O) = 1, and D(0) _
3, the optimal system is, using (9.19),
1 3
G (s)
,(s) (9 .24)
3 s 2 + 3 .2s + 3 s 2 + 3 .2s + 3
This G (s) is clearly implementable . Because G(0) = 1, the optimal system has a
zero position error . The implementation of this optimal system will be discussed in
the next chapter .
Im s
-tee -j24 ~3e j24

X
> Res
- 0
_ -ej 24 ~3e -j 24
Figure 9 .7 Roots of (9.22).

Exercise 9 .4.1
Given G(s) = (s - 1)/s(s + 1), find an implementable overall transfer function to

minimize
J = fo [9(y(t) - 1)2 + U2 (t)] dt (9.25)
[Answer : G0 (s) = - 3(s - 1)/(s + 3)(s + 1).1
9 .4 .2 Computation of Spectral Factorizations

The design of quadratic optimal systems requires the computation of spectral fac-
torizations . One way to carry out the factorization is to compute all the roots of Q(s)
and then group all the left-half-plane roots, as we did in (9.23). This method can be
easily carried out if software for solving roots of polynomials is available . For ex-
ample, if we use PC-MATLAB to carry out the spectral factorization of Q(s) in
(9.22), then the commands
q=[1 0 -4 0 9] ;
r = roots(q)
yield the following four roots :
r = -1 .5811 + 0 .7071 i
-1 .5811 - 0 .7071 i
1 .5811 + 0 .7071 i
1 .5811 - 0 .7071 i
The first and second roots are in the open left half plane and will be used to form
D0(s). The command
poly([r(1) r(2)])
yields a polynomial of degree 2 with coefficients
1 .0000 3 .1623 3 .0000
This is D0(s). Thus the use of a digital computer to carry out spectral factorizations
is very simple .
We now introduce a method of carrying out spectral factorizations without solv-
ing for roots . Consider the Q(s) in (9.22). It is a polynomial of degree 4 . In the
spectral factorization of
Q(s) = s4 - 4s 2 + 9 = D0(s)D0(- s) (9.26)
the degrees of polynomials D0 (s) and D,(- s) are the same . Therefore, the degree of
D0 (s) is half of that of Q(s), or two for this example . Let

D0(s) = b o + b1s + b 2 s 2 (9 .27)
where b; are required to be all positive .2 If any one of them is zero or negative, then
D0(s) is not Hurwitz. Clearly, we have
D,(-s) = bo + b,(-s) + b2 (- s) 2 = bo - b t s + b2s 2 (9 .28)
The multiplication of D0(s) and Do( - s) yields
D0(s)D,(-s) = (bo + bt s + b 2s2)(bo - bts + b2s2)
= bo + (2b0b2 - b2 )s 2 + b22s 4
It is an even function of s. In order to meet (9 .26), we equate
b0 = 9
2b0b2 - b2 = - 4
and
b2 = I
Thus we have b0 = 3, b 2 = 1 and
bi =2b0b2 +4=2 .3 .1 + 4 = 10
which implies b t = N/10 . Note that we require all b; to be positive ; therefore, we
have taken only the positive part of the square roots . Thus the spectral factorization
of (9.26) is
D0 (s)=3+V10s+s 2 =3+3 .2s+s 2
We see that this procedure is quite simple and can be used if a digital computer and
the required software are not available . The preceding result can be stated more
generally as follows : If
Q(s) = a 0 + a2s2 + a4 s4 (9 .29)
and if
D0(s) = b0 + bts + b2s2 (9 .30)
then
b0 = V b2 = bt = -\I(-a2 + 2b0b 2) (9 .31)
Note that before computing b t , we must compute first b0 and b2 .
Now we shall extend the preceding procedure to a more general case . Consider
Q(s) = a0 + a2s2 + a4s4 + a6 S6 (9 .32)
It is an even polynomial of degree 6. Let

D0(s) = b0 + bt s + b 2s 2 + b3S 3 (9 .33)
2Also, they can all be negative . For convenience, we consider only the positive case .

Then
D0(-s) = b - b,s + b2s 2 - b3S3
and
D0 (s)D 0(-s) = bo + (2b b 2 - b,)s2 + (b2 - 2b1b3 )s4 - b3s6 (9 .34)
Equating (9 .32) and (9 .34) yields

b2 = a (9 .35a)
2bb2 - bi = a 2 (9 .35b)
b2 - 2b, b3 = a4 (9 .35c)
and
b3 = - ab (9 .35d)
From (9 .35a) and (9 .35d), we can readily compute b = \ and b3 = 6 In \7-a .

other words, the leading and constant coefficients of D0 (s) are simply the square
roots of the magnitudes of the leading and constant coefficients of Q(s) . Once b
and b3 are computed, there are only two unknowns, b1 and b2 , in the two equations
in (9.35b) and (9.35c) . These two equations are not linear and can be solved itera-
tively as follows. We rewrite them as
b 1 = U2b b, - a2 (9.36a)
b 2 = Va4 + 2b,b3 (9 .36b)
First we choose an arbitrary b2-say, b2)-and use this b2) to compute b 1 as

bi' ) = V2bbz) - a2
We then use this b(') to compute b2 as
bz' ) = Aa4 + 2b11) b3
If bZ') happens to equal b2) , then the chosen b2) is the solution of (9 .36). Of course,
the possibility of having b2') = b2(o) is extremely small . We then use b2') to compute
a new b1 as
b;2) = V2bbz') - a 2
and then a new b 2 as
bz) = 'V/a4 + 2bi2)b 3
If b22) is still quite different from b2'), we repeat the process . It can be shown that
the process will converge to the true solutions .3 This is an iterative method of car-
rying out the spectral factorization . In application, we may stop the iteration when
the difference between two subsequent b2') and b2`+ 1) is smaller than, say, 5% . This
is illustrated by an example .
31f we compute b2 = ( a2 + b ;)/2b o and b, = (b22 - a4)/2b 3 iteratively, the process will diverge .

Example 9 .4.2
Compute the spectral factorization of
Q(s) = 25 - 41s 2 + 20s4 - 4s6 (9.37)
Let
D0(s) = b + b 1 s + b2 S 2 + b3s3
Its constant term and leading coefficient are simply the square roots of the corre-
sponding coefficients of Q(s) :
bo = 25=5 b3 =VI -41 =N/ 4 = 2
The substitution of these into (9 .36) yields
b, = V 10b 2 + 41
b2 = 120 + 4b,
Now we shall solve these equations iteratively . Arbitrarily, we choose b2 as b
0 and compute
(0) (1) (2) (3) (4) (5)

b, 6.4 10.42 10.93 10.99 10 .999
b2 0 6.75 7.85 7.98 7.998 7.9998
We see that they converge rapidly to the solutions b, = 11 and b2 = 8 . To verify

the convergence, we now choose b2 as b2) = 100 and compute
(U) (1) (2) (3) (4) (J)
b, 32.26 12.77 11.19 11.02 11 .002

b2 100 12.21 8.43 8.05 8.005 8.0006
They also converge rapidly to the solutions b, = 11 and b2 = 8 .

The preceding iterative procedure can be extended to the general case . The basic
idea is the same and will not be repeated .

Exercise 9.4.2
Carry out spectral factorizations for

a. 4s4 - 9s2 + 16
b. -4s6 + lOs4 - 20s2 + 16
[Answers : 2s2 + 5s + 4, 2s 3 + 6.65s2 + 8.56s + 4.]
9.4 .3 Selection of Weighting Factors

In this subsection we discuss the problem of selecting a weighting factor in the
quadratic performance index to meet the constraint ~u(t)j < M for all t ? 0. It is
generally true that a larger q yields a larger actuating signal and a faster response.
Conversely, a smaller q yields a smaller actuating signal and a slower response .
Therefore, by choosing q properly, the constraint on the actuating signal can be met .
We use the example in (9 .20) to illustrate the procedure.
Consider a plant with transfer function G(s) = 1/s(s + 2) . Design an overall
system to minimize
J = fo [q(y(t) - 1)2 + u2(t)]dt
It is also required that the actuating signal due to a unit-step reference input meet
the constraint ~u(t)j < 3, for all t ? 0. Arbitrarily, we choose q = 100 and compute
Q(s) = s(s + 2)(-s)(-s + 2) + 100 .1 .1 = s4 - 4s2 + 100
Its spectral factorization can be computed as, using (9 .31),
D0 (s)=s2 + 24s+ 10=s 2 +4.9s+ 10
Thus the quadratic optimal transfer function is
qN(0) N(s) _ 100 . 1 1
0(s) - R(s) -
G ,(s) Do(0) D0 (s) 10 s2 + 49s + 10
10
s 2 +49s+ 10
The unit-step response of this system is simulated and plotted in Figure 9.8(a). Its
rise time, settling time, and overshoot are 0 .92 s, 1 .70 s, and 2 .13%, respectively.
Although the response is quite good, we must check whether or not its actuating
signal meets the constraint. No matter what configuration is used to implement G 0(s),
if there is no plant leakage, the transfer function from the reference signal r to the
actuating signal u is
G0(s) _ 10 s(s + 2) _ 10s(s + 2)
T(s) =
G(s) s2 +4.9s+ 10 1 s 2 +4.9s+ 10

YO y(r~
I-
,- t yt 0 I I-t
0
5 5 10 u(t) 5
u (t)
rat 0 i >t
5
(b) (c)
q = 100 0 .64 9
Rise time = 0.92 6 .21 2 .01
Settling time = 1 .70 10 .15 2 .80
Overshoot = 2 .13% 0% 0 .09%
U (0+) = 10 0 .8 3
Figure 9 .8 Responses of quadratic optimal systems .
The unit-step response of T(s) is simulated and also plotted in Figure 9.8(a). We see
that u(0 + ) = 10 and the constraint J u(t) l < 3 is violated . Because the largest mag-
nitude of u(t) occurs at t = 0 + , it can also be computed by using the initial-value
theorem (see Appendix A) . The response u(t) due to r(t) = 1 is
10
U(s) = T(s)R(s) =
s 2 + 4.9s + 10 s
The application of the initial-value theorem yields
u(0+) = lim sU(s) = lim sT(s)R(s) = lim T(s) = 10
s-,oo
Thus the constraint Ju(t) l < 3 is not met and the selection of q = 100 is not
acceptable 4
Next we choose q = 0 .64 and repeat the design . The optimal transfer function
is found as
0.64 x 1 1 0.8
G0 (s) =
0.8 S2 + 6 s + 0 .8 s 2 + 2.4s + 0.8
4 1t is shown by B . Seo [511 that if a plant transfer function is of the form (b 1 s + bo )/s(s + a), with
bo = 0, then the maximum magnitude of the actuating signal of quadratic optimal systems occurs at
t = 0 + and ~u(t)j < u(0 + ) = V.

Its unit-step response and the actuating signal are plotted in Figure 9.8(b). The re-
sponse is fairly slow . Because Ju(t)j < u(0 + ) = 0.8 is much smaller than 3, the
system can be designed to respond faster . Next we try q = 9, and compute
Q(s) = s(s + 2)(-s)(-s + 2) + 9 . 1 . 1
Its spectral factorization, using (9.31), is
DO (s) =s 2 +V10 s+3=s 2 +3.2s+3
Thus the optimal transfer function is
qN(0) N(s) = 9 . 1 1
GO(s) = = 3 (9.38)
D0(0) DO(s) 3 s2 + 3 .2s + 3 s2 + 3.2s + 3
and the transfer function from r to u is
G0(s) 3s(s + 2)
T(s)
S)
=
s 2 +3 .2s+3
Their unit-step responses are plotted in Figure 9 .8(c). The rise time of y(t) is 2 .01
seconds, the settling time is 2 .80 seconds, and the overshoot is 0 .09%. We also have
Iu(t)l < u(0 + ) = T(-) = 3, for all t. Thus this overall system has the fastest response
under the constraint J u(t) l < 3.
From this example, we see that the weighting factor q is to be chosen by trial
and error . We choose an arbitrary q, say q = q 0, and carry out the design . After the
completion of the design, we then simulate the resulting overall system . If the re-
sponse is slow or sluggish, we may increase q and repeat the design. In this case,
the response will become faster . However, the actuating signal may also become
larger and the plant may be saturated . Thus the choice of q is generally reached by
a compromise between the speed of response and the constraint on the actuating
signal .
Optimality is a fancy word because it means "the best ." However, without
introducing a performance index, it is meaningless to talk about optimality . Even if
a performance index is introduced, if it is not properly chosen, the resulting system
may not be satisfactory in practice . For example, the second system in Figure 9 .8 is
optimal with q = 0.64, but it is very slow . Therefore, the choice of a suitable
performance index is not necessarily simple.
Exercise 9 .4.3
Given a plant with transfer function G(s) = (s + 2)/s(s - 2), find a quadratic
optimal system under the constraint that the magnitude of the actuating signal due
to a unit step reference input is less than 5 .
[Answer : G0(s) = 5(s + 2)/(s 2 + 7s + 10) .]

9 .5 THREE MORE EXAMPLES 361
9.5 THREE MORE EXAMPLES
In this section we shall discuss three more examples . Every one of them will be
redesigned in latter sections and be compared with quadratic optimal design .
Example 9 .5.1
Consider a plant with transfer function [19, 34]
2
s = 9
s(s 2 + 0.25s + 6 .25)
Design an overall system to minimize
J = f
[q(y(t) - 1)2 + u2(t)]dt (9 .40)
The weighting factor q is to be chosen so that the actuating signal u(t) due to a unit-
step reference input meets Ju(t)l < 10 for t ? 0. First we choose q = 9 and compute
Q(s) = D(s)D( - s) + qN(s)N( - s)
= s(s 2 + 0 .25s + 6 .25) . (-s)(s 2 - 0.25s + 6 .25) + 9 . 2 . 2 (9.41)
= -s 6 - 12.4375s4 - 39 .0625s2 + 36
The spectral factorization of (9.41) can be carried out iteratively as discussed in
Section 9.4.2 or by solving its roots . As a review, we use both methods in this
example . We first use the former method . Let
D0(s) = b + b,s + b2s2 + b3S 3
Its constant term and leading coefficient are simply the square roots of the corre-
sponding coefficients of Q(s) :
b = N/36 = 6 b3 = '\/ _11 _ Vi = 11
The substitution of these into (9.36) yields

b, = V12b 2 + 39 .0625
b2 = U2b, - 12.4375
Now we shall solve these equations iteratively . Arbitrarily, we choose b2 as b2(o)
0 and compute
b, 6.25 6.49 6.91 7.30 7.53 7.65 7.70 7.73 7.75 7.75
b2 0 0.25 0.73 1 .18 1 .47 1 .62 1.69 1 .72 1 .74 1 .74 1 .75

We see that they converge to the solution b, = 7 .75 and b2 = 1 .75 . Thus we have
Q(s) = D0(s)D,,(-s) with
D0 (s) = s3 + 1 .755 2 + 7 .75s + 6
Thus the optimal overall transfer function that minimizes (9.40) with q = 9 is
_ qN(0) N(s) 9 .2 2
G(s) D0(0) D0(s) 6 s3 + 1 .755 2 + 7 .75s + 6 (9 .42)
6
s 3 + 1 .755 2 + 7 .75s + 6
For this overall transfer function, it is found by computer simulation that ju(t)j < 3,
for t ? 0. Thus we may choose a larger q . We choose q = 100 and compute
Q(s) = D(s)D(-s) + 100N(s)N( - s)
=
_S6 - 12 .4375s4 - 39 .0625s 2 + 400
Now we use the second method to carry out the spectral factorization . We use
PC-MATLAB to compute its roots . The command
r =roots([ -1 0 -12 .4375 0 - 39.0625 0 400])
y (f)
t
0 1 2 3 4 5 6 7 8
Figure 9.9 Responses of various designs of (9.39).

yields
r= -0 .9917+3 .0249i
- 0 .9917 - 3 .02491
0 .9917 + 3 .02491
0 .9917 - 3 .0249i
1 .9737
-1 .9737
The first, second, and last roots are in the open left half plane . The command
poly([r(1) r(2) r(6)])
0
yields [1 .000 3 .9571 14.0480 20.0000] . Thus we have D (s) = s 3 + 3 .957s 2
+ 14.048s + 20 and the quadratic optimal overall transfer function is
20
G(s) = (9 .43)
s3 + 3.957s 2 + 14.048s + 20
For this transfer function, the maximum amplitude of the actuating signal due to a
unit-step reference input is 10 . Thus we cannot choose a larger q . The unit-step
response of G0(s) in (9 .43) is plotted in Figure 9 .9 with the solid line. The response
appears to be quite satisfactory .
Exaryiple 9 .5.2
G(s) =
s( + 1) (9 .44)
Find an overall transfer function to minimize the quadratic performance index
J = fo [100(y(t) - 1)2 + u2 (t)]dt (9 .45)
where the weighting factor has been chosen as 100 . We first compute
Q(s) = D(s)D( - s) + qN(s)N( - s)
= s(s - 1)(-s)(-s - 1) + 100(s + 3)(-s + 3)
=S4 - lOls2 +900
= (s + 9 .5459)(s - 9.5459)(s + 3 .1427)(s - 3 .1427)

where we have used PC-MATLAB to compute the roots of Q(s) . Thus we have
Q(s) = D 0(s)D0( - s) with

D (s) = (s + 9 .5459)(s + 3 .1427) = s 2 + 12.7s + 30

and the quadratic optimal system is given by
qN(0) N(s) 10(s + 3)
G ,(s)
(s) D (0) D (s) s2 + 12.7s + 30 (9 .46)
Its response due to a unit-step reference input is shown in Figure 9 .10(a) with the
solid line . The actuating signal due to a unit-step reference input is shown in Figure
9 .10(b) with the solid line ; it has the property I u(t)I < 10 for t ? 0.
Y (t) u(t)
2
,- ITAE in (9.59)
L Quadratic optimal
0.8
ITAE in (9 .57)
0.6
Computer simulation
0.4
02
0.1 0.2 0 .3 0 .4 0 .5
(a) (b)
Figure 9 .10 Responses of various designs of (9.44).
Example 9.5.3
S - 1
G(s) = (9.47)
s(s - 2)
It has a non-minimum-phase zero . To find the optimal system to minimize the quad-
ratic performance index in (9.45), we compute
Q(s) = s(s - 2)(-s)(-s - 2) + 100(s - 1)(-s - 1) = s 4 - 104s 2 + 100
= (s + 10.1503)(s - 10.1503)(s + 0 .9852)(s - 0.9852)
Thus we have D(s) = s2 + 11 .14s + 10 and
-10(s - 1)
G ,(s)
(s) (9.48)
s2 + 11 .14s + 10

Its unit-step response is shown in Figure 9 .11 with the solid line. By computer
simulation we also find Ju(t)l !5 10 for t ? 0 if the reference input is a unit-step
function .
Y (t)
t
5 6
Figure 9 .11 Responses of various designs of (9 .47) .
9.5.1 Symmetric Root Loci -5

We discuss in this subsection the poles of G 0(s) as a function of q by using the root-
locus method. Consider the polynomial
D(s)D( - s) + qN(s)N( - s) (9.49)
The roots of (9 .49) are the zeros of the rational function
1 + qG(s)G(-s)
or the solution of the equation
= G (s)G ( - s) = s) (9.50)
41 D(s)D(-
These equations are similar to (7 .11) through (7.13), thus the root-locus method can
be directly applied . The root loci of (9 .50) for G(s) = 1/s(s + 2) are plotted in
5 This section may be skipped without loss of continuity .

Im s
q=9
1 ) I >-Res
q= 1
Figure 9.12 Root loci of (9.50) .
Figure 9.12. The roots for q = 0 .64, 4, 9, and 100 are indicated as shown . We see
that the root loci are symmetric with respect to the imaginary axis as well as the real
axis . Furthermore the root loci will not cross the imaginary axis for q > 0 . Although
the root loci reveal the migration of the poles of the quadratic optimal system, they
do not tell us how to pick a specific set of poles to meet the constraint on the actuating
signal .
We discuss now the poles of G0(s) as q It is assumed that G(s) has n poles
and m zeros and has no non-minimum-phase zeros . Then, as q - -, 2m root
loci of G(s)G ( - s) will approach the 2m roots of N(s)N( - s) and the remaining
(2n - 2m) root loci will approach the (2n - 2m) asymptotes with angles
(2k + 1) 7r
k = 0, 1,2, . . .,2n - 2m - 1
2n - 2m
Ims Ims
/ / 3
*
>)K Re s -H
1 1
\ / x
n-m=3 / n-m=4
Figure 9 .13 Distribution of optimal poles as q --> - .

9 .6 ITAE OPTIMAL SYSTEMS (34) 367
(see Section 7.4, in particular, (7 .27)). Thus as q approaches infinity, m poles of

G(s) will cancel m zeros of G(s) and the remaining (n - m) poles of G(s) will
distribute as shown in Figure 9 .13, where we have assumed that the centroid defined
in (7 .27a) is at the origin . The pole pattern is identical to that of the Butterworth
filter [13].
9 .6 ITAE OPTIMAL SYSTEMS [331
In this section we discuss the design of control systems to minimize the integral of
time multiplied by absolute error (ITAE) in'(9.8). For the quadratic overall system
1
G(s)
_ s2 + 2~s + 1
the ITAE, the integral of absolute error (IAE) in (9.6), and the integral of square
error (ISE) in (9.7) as a function of the damping ratio ~ are plotted in Figure 9 .14.
The ITAE has largest changes as C varies, and therefore has the best selectivity . The
ITAE also yields a system with a faster response than other criteria, therefore Graham
and Lathrop [33] chose it as their design criterion . The system that has the smallest
ITAE is called the optimal system in the sense of ITAE or the ITAE optimal system.
Consider the -overall transfer function
a
G (s) = (9.51)
s" + an i s" -1 + + a2 s2 + a 1 s + a o
This transfer function contains no zeros . Because G(0) = 1, if G(s) is stable, then
the position error is zero, or the plant output will track asymptotically any step-
reference input . By analog computer simulation, the denominators of ITAE optimal
systems were found to assume the forms listed in Table 9 .1 . Their poles and unit-
step responses, for w o = 1, are plotted in Figures 9 .15 and 9.16 . We see that the
optimal poles are distributed evenly around the neighborhood of the unit circle . We
also see that the overshoots of the unit-step responses are fairly large for large n.
These systems are called the ITAE zero-position-error optimal systems .
J4 (ITAE)
0 0.4 0.8 1 .2
Figure 9.14 Comparison of various design criteria.

CHAPTER 9 THE INWARD APPROACH-CHOICE OF OVERALL TRANSFER FUNCTIONS

368
Table 9.1 ITAE Zero-Position-Error Optimal Systems

s+wo
s 2 + 1 .4wos + w.
s3 + 1 .75wos 2 + 2 .15wos + wo
s4 + 2 .1wos 3 + 3 .4wos 2 + 2.7wos + (01
s 5 + 2.8wos4 + 5 .0wps 3 + 5 .5wos 2 + 3 .4wos + w0
s 6 + 3 .25 wos 5 + 6 .60wos4 + 8 .60wos 3 + 7 .45 wos 2 + 3 .95 wos + wo 6OS
+ wo
s7 + 4 .475wos 6 + 10.42wos 5 + 15 .08wps4 + 15 .54wos 3 + 10 .64was 2 + 4 .58(0 .15wos + w8
s8 + 5 .20w0s' + 12 .80wos 6 + 21.60wos 5 + 25.75wos 4 + 22.20wos 3 + 13.30wos 2 + 5 0
We now discuss the optimization of

als + ao (9.52)
+ . . . + a2s 2 + a t s + a o
G(s) s + ants"-1
with respect to the ITAE criterion . The transfer function has one zero ; their coeffi-
cients, however, are constrained so that G (s) has zero position error and zero ve-
. By
locity error. This system will track asymptotically any ramp-reference input
analog computer simulation, the optimal step responses of G (s) in (9 .52) are found
as shown in Figure 9 .17 . The optimal denominators of G (s) in (9 .52) are listed in
.
Table 9 .2 . The systems are called the ITAE zero-velocity-error optimal systems
3rd 4th 0 5th 0

2nd order +j +j
+j 0 _ +j
i
A
i/ 0
o
i i
0 0 10 0
0
N
_j -J -j
0
0
6th 0 7th 8th

+j 0 -Q- +j +j
q
~~0
i
i ~o
I 0 I 0 0
1 0 -1
0
0
N o
1 0 _j _j
0 0
0
Figure 9 .15 Optimal pole locations .

9.6 ITAE OPTIMAL SYSTEMS (34) 369
5 10 15
Normalized time
Figure 9.16 Step responses of ITAE optimal systems with zero position error .
1 .5
IN
~k!@VWIMRI
0 .5
0
0 5 10 15
Normalized time
Figure 9.17 Step responses of ITAE optimal systems with zero velocity error .

Table 9 .2 ITAE Zero-Velocity-Error Optimal Systems
s2 +3.2wos+wo
s3 + 1 .75wos2 + 3.25wos + wo
s4 + 2.41wos3 + 4.93wos 2 + 5.14wps + wo
S' + 2.19wos4 + 6.SOwos 3 + 6.30wos 2 + 5.24
OS + 0)5
s6 + 6.12woss + 13 .42wos4 + 17.16wos3 + 14.14wos 2 + 6 .76wos + w60
Similarly, for the following overall transfer function

a2s2 + as + ao
G ,(s) _ (9.53)
sn + an _1S n-1 + . . . + a2S 2 + a1 s + ao
the optimal step responses are shown in Figure 9 .18 and the optimal denominators
are listed in Table 9 .3 . They are called the ITAE zero-acceleration-error optimal
systems . We see from Figures 9 .16, 9.17, and 9 .18 that the optimal step responses
for G0(s) with and without zeros are quite different . It appears that if a system is
required to track a more complicated reference input, then the transient performance
will be poorer . For example, the system in Figure 9 .18 tracks acceleration reference
inputs, but its transient response is much worse than the one for the system in Figure
9 .16, which can track only step-reference inputs . Therefore a price must be paid if
we design a more complex system .
Figure 9.18 Step responses of ITAE optimal systems with zero acceleration error .

Table 9.3 ITAE Zero-Acceleration-Error Optimal Systems
s3 + 2 .97coo s 2 + 4 .94wos + coo

s4 + 3 .71cuo s 3 + 7 .88was 2 + 5 .93wos + coo
s5 + 3 .81 w0 s4 + 9 .94wos3 + 13 .44wos 2 + 7.36coos + wo
s6 + 3 .93wo s 5 + 11 .68wos4 + 18 .56 J3 s3 + 19.3wos 2 + 8 .06wos + wo
9.6.1 Applications
In this subsection we discuss how to use Tables 9 .1 through 9 .3 to design ITAE
optimal systems . These tables were developed without considering plant trans-
fer functions. For example, for two different plant transfer functions such as
I/s(s + 2) and 1/s(s - 10), the optimal transfer function G (s) can be chosen as
2
s2 + 1 .4w0s + wo
The actuating signals for both systems, however, will be different . Therefore too in
both systems should be different . We shall use the constraint on the actuating signal
as a criterion in choosing w o. This will be illustrated in the following examples .
Example 9 .6.1
Consider the plant transfer function in (9 .20) or
1
G(s)=s(s+2)
Find a zero-position-error system to minimize ITAE . It is also required that the
actuating signal due to a unit-step reference input satisfy the constraint
Iu(t)I - 3
for all t.
The ITAE optimal overall transfer function is chosen from Table 9 .1 as
tv2
G(s) = 0
s2 + 1 .4,0 s + cwo
It is implementable . Clearly the larger the to o, the faster the response . However, the
actuating signal will also be larger . Now we shall choose to o to meet ~u(t)j <_ 3 . The
transfer function from r to u is
U(s) = G (s) cwos(s + 2)
T(s)
:= R(s) G(s) s2 + 1 .4ce0s + wo

Y (t) u(t)
A
1 .4 3
1 .2 2.5 1
1
2 1
I
1 .5 1
0.8 1 1
1 ITAE zero-velocity-error
1
06
0.5
04 0
ITAE zero-position-error
02 -0.5 /
0 t
1 2 3 4 5 6 7 8 9 10 1 0 1 2 3 4 5 6 7 8 10 t
(a) (b)
Figure 9 .19 Step responses of (9 .54) (with solid lines) and (9 .55) (with dashed lines) .
i Consequently, we have
wos(s + 2) 1
U(s) = T(s)R(s) =
s 2 + 1.4w0 s + w z0 S
By computer simulation, we find that the largest magnitude of u(t) occurs at t =
0+ .6 Thus the largest magnitude of u(t) can be computed by using the initial-value
theorem as
toos(s + 2) 2
Umax = u(0 + ) = I'M sU(s) = I' M 2 2 = WO
S + 1.4wo s + (0o
S__ S_ .
In order to meet the constraint Ju(t)l < 3, we set wo = 3 . Thus the ITAE optimal
system is
3 3
G0(s) = 2 (9 .54)
s + 1 .4x V-3s+3 s z +2.4s+3
This differs from the quadratic optimal system in (9 .38) only in one coefficient.
Because they minimize different criteria, there is no reason that they have the same
overall transfer function . The unit-step responses of G0(s) and T(s) are shown in
Figure 9 .19 with solid lines . They appear to be satisfactory .
Exercise 9 .6.1
Consider a plant with transfer function 2/s 2 . Find an optimal system to minimize
the ITAE criterion under the constraint J u(t)I !5 3 .
[Answer : 6/(s z + 3.4s + 6).]
61f the largest magnitude of u(t) does not occur at t = 0, then its analytical computation will be com-
plicated. It is easier to find it by computer simulations .

Example 9 .6.2
Consider the problem in Example 9 .6.1 with the additional requirement that the
velocity error be zero . A possible overall transfer function is, from Table 9 .2,
3 .2wos + coo
G(s) =
s2 + 3.2WOs + coo
However, this is not implementable because it violates the pole-zero excess inequal-
ity. Now we choose from Table 9 .2 the transfer function of degree 3 :
3.25&)02 s + coo
G(s) _ s
3 + 1 .75cwO S 2 + 3.25 (002S + ceo
This is implementable and has zero velocity error . Now we choose co o so that the
actuating signal due to a unit-step reference input meets Ju(t)l < 3. The transfer
function from r to u is
(s) _ (3.25 (oos + coo)s(s + 2)
T(s) - G
G(s) s3 + 1 .75wos2 + 3.25wos + wo
Its unit-step response is shown in Figure 9 .19(b) with the dashed line . We see that
the largest magnitude of u(t) does not occur at t = 0 + . Therefore, the procedure in
Example 9 .6.1 cannot be used to choose wO for this problem . By computer simula-
tion, we find that if wO = 0.928, then fa(t)f < 3. For this wo, G (s) becomes
2.799s + 0.799
G(s) = (9 .55)
s3 1 .624s2 + 2.799s + 0.799
+
This is the ITAE zero-velocity-error optimal system . Its unit-step response is plotted
in Figure 9 .19(a) with the dashed line . It is much more oscillatory than that of the
ITAE zero-position-error optimal system . The corresponding actuating signal is plot-
ted in Figure 9.19(b).
Example 9 .6.3
Consider the plant transfer function in (9.39), that is,
2
G(s)
= s(s2 + 0.25s + 6.25)
Find an ITAE zero-position-error optimal system . It is also required that the actuating
signal u(t) due to a unit-step reference input meet the constraint J u(t)I < 10, for
t ? 0. We choose from Table 9.1
w3
G (s) =
s3 + 1.75wos2 + 2.15wos + wo

By computer simulation, we find that if to o = 2.7144, then I u(t)I <_ u(0) = 10 for
all t > 0 . Thus the ITAE optimal system is
20
G(s) = (9 .56)
s3 + 4.75s 2 + 15.84s + 20
Its unit-step response is plotted in Figure 9 .9 with the dashed line . Compared with
the quadratic optimal design, the ITAE design has a faster response and a smaller
overshoot . Thus for this problem, the ITAE optimal system is more desirable .
Example 9.6.4
s + 3
G(s) = s(s
- 1)
signal u(t) due to a unit-step reference input meet the constraint Iu(t) l <_ 10, for
t ? 0. The pole-zero excess of G(s) is 1 and G(s) has no non-minimim-phase zero;
therefore, the ITAE optimal transfer function
G(s) = ~o (9 .57)
s+too
is implementable . We find by computer simulation that if W o = 10, then G(s) meets
the design specifications . Its step response and actuating signal are plotted in Figure
9.10 with the dashed line . They are almost indistinguishable from those of the quad-
ratic optimal system . Because G (s) does not contain plant zero (s + 3), its imple-
mentation will involve the pole-zero cancellation of (s + 3), as will be discussed in
the next chapter . Because it is a stable pole and has a fairly small time constant, its
cancellation will not affect seriously the behavior of the overall system, as will be
demonstrated in the next chapter .
In addition to (9.57), we may also choose the following ITAE optimal transfer
function
W2
G, (s) __ o (9 .58)
s2 + 1.4wos + t002
It has pole-zero excess larger than that of G(s) and is implementable . We find by
computer simulation that if too = 24.5, then the G(s) in (9 .58) or
_ 600.25
G(s) (9 .59)
s2 + 34 .3s + 600 .25
meets the design specifications . Its step response and actuating signal are plotted in
Figure 9 .10 with the dotted lines . The step response is much faster than the ones of
(9.57) and the quadratic optimal system . However, it has an overshoot of about 4.6%.

9.7 SELECTION BASED ON ENGINEERING JUDGMENT 375
Example 9 .6 .5
S - I
G(s)
s (s - 2)
signal u(t) due to a unit-step reference input meet the constraint ju(t)~ < 10, for
t ? 0. This plant transfer function has a non-minimum-phase zero and no ITAE
standard form is available to carry out the design . However, we can employ the idea
in [34] and use computer simulation to find its ITAE optimal transfer function as [54]
-10(s - 1)
G (s) _ (9 .60)
sz + 5.1s + 10
under the constraint l u(t)l < 10 . We mention that the non-minimum-phase zero
(s - 1) of G(s) must be retained in G(s), otherwise G(s) is not implementable . Its
step response is plotted in Figure 9 .11 with the dashed line . It has a faster response
than the one of the quadratic optimal system in (9.48) ; however, it has a larger
undershoot and a larger overshoot . Therefore it is difficult to say which system is
better.
9 .7 SELECTION BASED ON ENGINEERING JUDGMENT
In the preceding sections, we introduced two criteria for choosing overall transfer
functions . The first criterion is the minimization of the quadratic performance index .
The main reason for choosing this criterion is that it renders a simple and straight-
forward procedure to compute the overall transfer function . The second criterion is
the minimization of the integral of time multiplied by absolute error (ITAE) . It was
chosen in [33] because it has the best selectivity . This criterion, however, does not
render an analytical method to find the overall transfer function ; it is obtained by
trial and error and by computer simulation . In this section, we forego the concept of
minimization or optimization and select overall transfer functions based on engi-
neering judgment. We require the system to have a zero position error and a good
transient performance . By a good transient performance, we mean that the rise and
settling times are small and the overshoot is also small . Without comparisons, it is
not possible to say what is small . Fortunately, we have quadratic and ITAE optimal
systems for comparisons. Therefore, we shall try to find an overall system that has
a comparable or better transient performance than the quadratic or ITAE optimal
system. Whether the transient performance is comparable or better is based on en-
gineering judgment ; no mathematical criterion will be used . Consequently, the se-
lection will be subjective and the procedure of selection is purely trial and error .

Example 9.7.1
Consider the plant transfer function in (9.39), or
2
G(s)=
s(s 2 +0.25s+6 .25)
We use computer simulation to select the following two overall transfer functions
20 20
G' (s) = (s (9.61)
+ 2)(s2 + 2.5s + 10) s3 + 4.5s2 + 15s + 20
20 20
Go2(s) =
(s + 10)(s2 + 2s + 2) s3 + 12s 2 + 22s + 20 (9.62)
The actuating signals of both systems due to a unit-step reference input meet the
4 constraint Ju(t) l < 10 for t ? 0 . Their step responses are plotted in Figure 9 .9 with,
respectively, the dotted line and the dashed-and-dotted line . The step response of
Gol(s) lies somewhere between those of the quadratic optimal system and the ITAE
optimal system . Therefore, G0,(s) is a viable alternative of the quadratic or ITAE
optimal system.
The concept of dominant poles can also be used to select G0(s). Consider the
overall transfer function in G o2(s). It has a pair of complex-conjugate poles at
-1 j 1 and a real pole at - 10. Because the response due to the real pole dies
out much faster than does the response due to the complex-conjugate poles, the
response of G02(s) is essentially determined or dominated by the complex-conjugate
poles . The complex-conjugate poles have the damping ratio 0 .707, and consequently
the step response has an overshoot of about 5% (see Section 7.2.1), as can be seen
from Figure 9.9 . However, because the product of the three poles must equal 20 in
order to meet the constraint on the actuating signal, if we choose the nondominant
pole far away from the imaginary axis, then the complex-conjugate poles cannot be
too far away from the origin of the s-plane . Consequently, the time constant of G02(s)
is larger than that of Gol (s) and its step response is slower, as is shown in Figure
9.9. Therefore for this problem, the use of dominant poles does not yield a satisfac-
tory system. We mention that if the complex-conjugate poles are chosen at
-2 j2, then the system will respond faster . However, the real pole must be chosen
as 20/8 = 2 .5 in order to meet the constraint on the actuating signal . In this case,
the complex-conjugate poles no longer dominate over the real pole, and the concept
of dominant poles cannot be used.

9 .7 SELECTION BASED ON ENGINEERING JUDGMENT 377
Example 9 .7 .2
Consider the plant transfer function in (9.44), or
G (s) = s (s + 1) (9.63)
We have designed a quadratic optimal system in (9 .46) and two ITAE optimal sys-
tems in (9 .57) with wo = 10 and (9.59). Their step responses are shown in Figure
9.10. Now using computer simulation, we find that the following
- 784
G0(s) (9.64)
s 2 + 50 .4s + 784
has the response shown with the dashed-and-dotted line in Figure 9 .10 . It is com-
parable with that of the ITAE optimal system in (9 .59) under the same constraint on
the actuating signal . Thus, the overall transfer function in (9 .64) can also be used,
although it is not optimal in any sense .
Example 9 .7 .3
S - 1
G(s) =
s(s - 2)
We have designed a quadratic optimal system in (9 .48) and an ITAE optimal system
in (9.60) . Their step responses are shown in Figure 9 .11 . Now we find, by using
computer simulation, that the response of
10(s - 1)
G0(s) = - (9.65)
(s + N/_10 )2
lies somewhere between those of (9 .48) and (9 .60) under the same constraint on the
actuating signal . Therefore, (9.65) can also be chosen as an overall transfer function .
In this section, we have shown by examples that it is possible to use computer

simulation to select an overall transfer function whose performance is comparable
to that of the quadratic or ITAE optimal system . The method, however, is a trial-
and-error method . In the search, we vary the coefficients of the quadratic or ITAE
system and see whether or not the performance could be improved . If we do not
have the quadratic or ITAE optimal system as a starting point, it would be difficult
to find a good system . Therefore, the computer simulation method cannot replace
the quadratic design method, nor the standard forms of the ITAE optimal systems .
It can be used to complement the two optimal methods .
9 .8SUMMARY AND CONCLUDING REMARKS
This chapter introduced the inward approach to design control systems . In this ap-
proach, we first find an overall transfer function to meet design specifications and
then implement it . In this chapter, we discussed only the problem of choosing an
overall transfer function. The implementation problem is discussed in the next
chapter.
The choice of an overall transfer function is not entirely arbitrary ; otherwise we
may simply choose the overall transfer function as 1 . Given a plant transfer function
G(s) = N(s)/D(s), an overall transfer function G0(s) = N0(s)/D 0(s) is said to be
implementable if there exists a configuration with no plant leakage such that G0(s)
can be built using only proper compensators. Furthermore, the resulting system is
required to be well posed and totally stable-that is, the closed-loop transfer function
of every possible input-output pair of the system is proper and stable . The necessary
and sufficient conditions for G 0(s) to be implementable are that (1) G 0(s) is stable,
(2) G0 (s) contains the non-minimum-phase zeros of G(s), and (3) the pole-zero ex-
cess of G0(s) is equal to or larger than that of G (s). These constraints are not stringent ;
poles of G0(s) can be arbitrarily assigned so long as they all lie in the open left half
s-plane; other than retaining all zeros outside the region C in Figures 6 .13 or 7 .4,
all other zeros of G0(s) can be arbitrarily assigned in the entire s-plane .
In this chapter, we discussed how to choose an implementable overall system
to minimize the quadratic and ITAE performance indices . In using these performance
indices, a constraint on the actuating signal or on the bandwidth of resulting systems
must be imposed; otherwise, it is possible to design an overall system to have a
performance index as small as desirable and the corresponding actuating signal will
approach infinity . The procedure of finding quadratic optimal systems is simple and
straightforward; after computing a spectral factorization, the optimal system can be
readily obtained from (9.19). Spectral factorizations can be carried out by iteration
without computing any roots, or computing all the roots of (9 .16) and then grouping
the open left half s-plane roots . ITAE optimal systems are obtainable from Tables
9.1 through 9.3 . Because the tables are not exhaustive, for some plant transfer func-
tions (for example, those with non-minimum-phase zeros), no standard forms are
available to find ITAE optimal systems . In this case, we may resort to computer
simulation to find an ITAE optimal system .
In this chapter, we also showed by examples that overall transfer functions that
have comparable performance as quadratic or ITAE optimal systems can be obtained
by computer simulation without minimizing any mathematical performance index .
It is therefore suggested that after obtaining quadratic or ITAE optimal systems, we
may change the parameters of the optimal systems to see whether a more desirable
system can be obtained . In conclusion, we should make full use of computers to
carry out the design.
We give some remarks concerning the quadratic optimal design to conclude this
chapter.
1. The quadratic optimal system in (9 .19) is reduced from a general formula in
Reference [10]. The requirement of implementability is included in (9.19) . If no

9 .8 SUMMARY AND CONCLUDING REMARKS 379
such requirement is included, the optimal transfer function that minimizes (9 .15)
with r(t) = 1 is
qN(O)
Go(s) = N+(s) (9 .66)
Do(0) D0 (s)
where N + (s) is N(s) with all its right-half-plane roots reflected into the left half
plane . In this case, the resulting overall transfer function may not be imple-
mentable. For example, if G(s) = (s - 1)/s(s + 1), then the optimal system
that minimizes
J = fo [q(y(t) - 1) 2 + u2 (t)]dt (9 .67)
with q = 9 is
3(ss + 1)
z
S +4s+3
which does not retain the non-minimum-phase zero and is not implementable .
For this optimal system, J can be computed as J = 3 . See Chapter 11 of Ref-
erence [12] for a discussion of computing J . The implementable optimal system
that minimizes J in (9 .67) is
- 3(s - 1)
G(s) =
s2 + 4s + 3
For this implementable G0 (s), J can be computed as J = 21 . It is considerably
larger than the J for G 0 (s). Although G"(s) has a smaller performance index, it
cannot be implemented.
2. If r(t) in (9.15) is a ramp function, that is r(t) = at, t ? 0, then the optimal
system that minimizes (9 .15) is
N(s) -s qN(O) N(s)
GG (s) = q(k 1 + k2 s) = 1 + (9 .68)
D0 (s) k1 D0 (0) D0 (s)
where
N(0) d FN(-s)]
k1 = and
Do (0) ds D"(-s) s=o
or, if N(s) = No + N1 s + + Nm sm and D0(s) = D o + D 1 s + + D,, s",

NOD, - D O N,
k = DNoO and k =
D zO
The optimal system in (9 .68) is not implementable because it violates the pole-
zero excess inequality . However, if we modify (9 .68) as
k(s) _ \ 1 + k2s) qN(0) N(s)"+ 1

ki l D0(0) DD (s) + Es
(9 .69)

where n : = deg D0(s) and c is a very small positive number, then G0(s) will be
implementable. Furthermore, for a sufficiently small E, D0 (s) + Es"' is Hur-
witz, and the frequency response of G0(s) is very close to that of G 0(s) in (9.68).
Thus (9 .69) is a simple and reasonable modification of (9.68).'
3. The quadratic optimal design can be carried out using transfer functions or using
state-variable equations . In using state-variable equations, the concepts of con-
trollability and observability are needed . The optimal design requires solving an
algebraic Riccati equation and designing a state estimator (see Chapter 11). For
the single-variable systems studied in this text, the transfer function approach
is simpler and intuitively more transparent . The state-variable approach, how-
ever, can be more easily extended to multivariable systems .
PROBLEMS
9.1 . Given G(s) = (s + 2)/(s - 1), is G0(s) = 1 implementable? Given G(s) _

(s - 1)/(s + 2), is G0(s) = 1 implementable?
9.2. Given G(s) = (s + 3)(s - 2)/s(s + 2)(s - 3), which of the following G0(s)
are implementable?
s- 2 s+ 3 - 2
s(s + 2) (s + 2)(s - 3) (s + 2) 2
(s+4)(s-2) s-2
s4 +4s2 +3s+6 s3 +4s+2
9.3. Consider a plant with transfer function G(s) = (s + 3)/s(s - 2).
a. Find an implementable overall transfer function that has all poles at - 2
and has a zero position error .
b. Find an implementable overall transfer function that has all poles at - 2
and has a zero velocity error . Is the choice unique? Do you have to retain
s + 3 in G0(s)? Find two sets of solutions : One retains s + 3 and the other
does not.
9.4. Consider a plant with transfer function G(s) = (s - 3)/s(s - 2).
a. Find an implementable overall transfer function that has all poles at - 2
and has a zero position error.
b. Find an implementable overall transfer function that has all poles at - 2
and has a zero velocity error. Is the choice unique if we require the degree
of G0(s) to be as small as possible?
7 This modification was suggested by Professor Jong-Lick Lin of Cheng Kung University, Taiwan .

PROBLEMS 381
9.5. What types of reference signals will the following G 0(s) track without an error?
-5s-2
a. G0(s)
-s2 -Ss-2
4s2 +s+ 3
b.G0(s)=
s5 +35 4 +45 2 +s+3
- 2S 2 + 154s + 120
c. G0(s) = s
4 + 14s 3 + 715 2 + 154s + 120
9.6. Consider two systems . One has a settling time of 10 seconds and an overshoot
of 5%, the other has a settling time of 7 seconds and an overshoot of 10% . Is
it possible to state which system is better? Now we introduce a performance
index as
J = k, . (Settling time) + k2 (Percentage overshoot)
If k, = k2 = 0 .5, which system is better? If k, = 0 .8 and k2 = 0.2, which
system is better?
9.7. Is the function
J = fo [q(y(t) - r(t)) + u(t)]dt
with q > 0 a good performance criterion?

9.8. Consider the design problem in Problem 7 .15 or a plant with transfer function
G(s) = -2/s 2. Design an overall system to minimize the quadratic per-
formance index in (9 .15) with q = 4 . What are its position error and velocity
error?
9.9. In Problem 9.8, design a quadratic optimal system that is as fast as possible
under the constraint that the actuating signal due to a step-reference input must
have a magnitude less than 5 .
9.10 . Plot the poles of G0(s) as a function of q in Problem 9 .9 .
9.11 . Consider the design problem in Problem 7 .14 or a plant with transfer function
0.015
G(s)= 52
+0.11s+0 .3
Design an overall system to minimize the quadratic performance index in
(9.15) with q = 9 . Is the position error of the optimal system zero? Is the
index of the optimal system finite?
9.12 . Consider the design problem in Problem 7 .12 or a plant with transfer function
4(s + 0.05)
G(s) =
s(s + 2)(s - 1 .2)
Find a quadratic optimal system with q = 100. Carry out the spectral factor-
ization by using the iterative method discussed in Section 9 .4.2.

9 .13 . Let Q(s) = D0 (s)D0 (- s) with

Q(s) = a o + a 2s 2 + a4S4 + . . . + a2i s 2"
and
D0(s)=bo +b1s+b 2 s 2 + . . . + bs"
Show
ao = bo2
a 2 = 2b0b2 - bi
a4 = 2b0b4 - 2b1b3 + b22
a 2r = 2bob2n - 2b1b2- 1 + 2b2b2 , -2 + (_ 1)"bn

where b, = 0, for i > n .
9.14. The depth of a submarine can be maintained automatically by a control system,
as discussed in Problem 7 .8. The transfer function of the submarine from the
stern angle 0 to the actual depth y can be approximated as
10(s + 2)2
G(s) =
(s + 10)(s2 + 0.1)
Find an overall system to minimize the performance index
J = fo [(y(t) - 1)2 + 02I dt

9.15 . Consider a plant with transfer function s/(s 2 - 1) . Design an overall system
to minimize the quadratic performance index in (9 .15) with q = 1 . Does the
optimal system have zero position error? If not, modify the overall system to
yield a zero position error .
9.16 . Consider a plant with transfer function G(s) = 1/s(s + 1) . Find an imple-
mentable transfer function to minimize the ITAE criterion and to have zero
position error. It is also required that the actuating signal due to a unit-step
reference input have a magnitude less than 10 .
9.17 . Repeat Problem 9.16 with the exception that the overall system is required to
have a zero velocity error .
9.18 . Repeat Problem 9.16 for G(s) = 1/s(s - 1) .
9.19 . Repeat Problem 9.17 for G(s) = 1/s(s - 1).
9.20 . Find an ITAE zero-position-error optimal system for the plant given in Problem
9.8. The magnitude of the actuating signal is required to be no larger than the
one in Problem 9 .8.

PROBLEMS 383
9.21 . Find an ITAE zero-position-error optimal system for the plant in Problem 9 .11 .
The real part of the poles of the optimal system is required to equal that in
Problem 9 .11 .
9.22 . Is it possible to obtain an ITAE optimal system for the plant in Problem 9 .12
from Table 9.1 or 9 .2? If yes, what will happen to the plant zero?
9.23 . Repeat Problem 9 .22 for the plant in Problem 9 .14.
9.24. a. Consider a plant with transfer function G(s) = (s + 4)/s(s + 1) . Design
an ITAE zero-position-error optimal system of degree 1 . It is required that
the actuating signal due to a unit-step reference input have a magnitude less
than 10.
b. Consider a plant with transfer function G(s) = (s + 4)/s(s + 1) . Design
an ITAE zero-position-error optimal system of degree 2 . It is required that
the actuating signal due to a unit-step reference input have a magnitude less
than 10.
c. Compare their unit-step responses.
9.25. Consider the generator-motor set in Figure 6 .1 . Its transfer function is assumed
to be
300
G(s) = s
4 + 184s3 + 760s 2 + 162s
It is a type I transfer function . Design a quadratic optimal system with q =
25 . Design an ITAE optimal system with u(0 +) = 5. Plot their poles . Are
there many differences?
9.26. Consider a plant with transfer function 1/s 2. Find an optimal system with zero
velocity error to minimize the ITAE criterion under the constraint Ju(t) l <_ 6 .
[Answer : (6s + 2.5)/(s3 + 2.38s 2 + 6s + 2 .5) .]
9.27 . If software for computing step responses is available, adjust the coefficients of
the quadratic optimal system in Problem 9 .8, 9 .11, 9 .12, 9.14, or 9 .15 to see
whether a comparable or better transient performance can be obtained .
Implementation
Linear Algebraic
Method
4
10.1 INTRODUCTION
The first step in the design of control systems using the inward approach is to find
an overall transfer function to meet design specifications . This step was discussed
in Chapter 9 . Now we discuss the second step-namely, implementation of the
chosen overall transfer function . In other words, given a plant transfer function G(s)
and an implementable G0(s), we shall find a feedback configuration without plant
leakage and compute compensators so that the transfer function of the resulting
system equals G0(s). The compensators used must be proper and the resulting system
must be well posed and totally stable .
The preceding problem can also be stated as follows : Given a plant G(s) and
given a model G 0 (s), design an overall system so that the overall transfer function
equals or matches G0(s). Thus the problem can also be called the model-matching
problem. In the model-matching problem, we match not only poles but also zeros ;
therefore, it can also be called the pole-and-zero placement problem . There is a
closely related problem, called the pole-placement problem . In the pole-placement
problem, we match or control only poles of resulting overall systems ; zeros are not
specified . In this chapter, we study both the model-matching and pole-placement
problems.
This chapter introduces three control configurations . They are the unity-feed-
back, two-parameter, and plant input/output feedback configurations . The unity-
384

10 .2 UNITY-FEEDBACK CONFIGURATION-MODEL MATCHING 385
feedback configuration can be used to achieve any pole placement but not any model
matching. The other two configurations, however, can be used to achieve any model
matching. In addition to model matching and pole placement, this chapter also stud-
ies robust tracking and disturbance rejection .
The idea used in this chapter is very simple. The design is carried out by match-
ing coefficients of compensators with desired polynomials. If the denominator D(s)
and numerator N(s) of a plant transfer function have common factors, then it is not
possible to achieve anypoleplacement or any model matching . Therefore, we require
D(s) and N(s) to have no common factors or to be coprime . Under this assumption,
the conditions of achieving matching depend on the degree of compensators . The
larger the degree, the more parameters we have for matching . If the degree of com-
pensators is large enough, matching is always possible . The design procedures in
this chapter are essentially developed from these concepts and conditions .
10.2 UNITY FEEDBACK CONFIGURATION-MODEL MATCHING
We discuss in this section the implementation of an implementable G 0(s) by using

the unity-feedback configuration shown in Figure 10 .1 . Let G(s) and C(s) be re-
spectively the transfer function of the plant and compensator . If the overall transfer
function from r to y is G0(s), then we have
G a,(s) 10 .1
(s) = 1 + C(s)G(s)
which implies
G0 (s) + C(s)G(s)G0(s) = C(s)G(s)
and
C(s)G(s)(1 - G0(s)) = G0 (s)
Thus, the compensator can be computed from
C(s) G(s) (10 .2)

G(s)(1 - G0(s))
We use examples to illustrate its computation and discuss the issues that arise in its
implementation .
r y
C(s) G (s)

386 CHAPTER 10 IMPLEMENTATION-LINEAR ALGEBRAIC METHOD
Example 10 .2.1
Consider a plant with transfer function 1/s(s + 2) . The optimal system that mini-
mizes the ITAE criterion and meets the constraint ~u(t)I < 3 was computed in (9 .54)
as 3/(s 2 + 2.4s + 3) . If we implement this G0(s) in Figure 10 .1, then the
compensator is
3
s 2 +2.4s+3
C(s) =
1 1 - 3
s(s + 2) s2+2.4s+3 (10.3)
3
s2 +2.4s+3 _ 3(s+2)
1 s2 + 2.4s s + 2.4
s(s + 2) s 2 + 2.4s + 3
It is a proper compensator. This implementation has a pole-zero cancellation. Be-
cause the cancelled pole (s + 2) is a stable pole, the system is totally stable . The
condition for the unity-feedback configuration in Figure 10 .1 to be well posed is
1 + G(oc)C(oo) =A 0. This is the case for this implementation ; therefore, the system
is well posed. In fact, if G(s) is strictly proper and if C(s) is proper, then the unity-
feedback configuration is always well posed .
This implementation has the cancellation of the stable pole (s + 2) . This can-
celed pole is dictated by the plant, and the designer has no control over it . If this
cancellation is acceptable, then the design is completed . In a latter section, we shall
discuss a different implementation where the designer has freedom in choosing can-
celed poles .
Example 10 .2.2
Consider a plant with transfer function 2/(s + 1)(s - 1). If G0(s) = 2/(s 2 + 2s
+ 2), then
2
s2 +2s+2
C(s) =
2 2
(s+ 1)(s- 1) 1 s2 +2s+2
2
s2 +2s+2 (s+ 1)(s- 1)
2 s2 + 2s s(s + 2)
(s + 1)(s - 1) s2 + 2s + 2

10 .2 UNITY-FEEDBACK CONFIGURATION-MODEL MATCHING 387
P
r + e (s+ 1)(S- 1) U 2 V
s (s + 2) (s +1) (s-1
C(s) G (s)
The compensator is proper . The implementation is plotted in Figure 10 .2. The im-
plementation has a stable pole-zero cancellation and an unstable pole-zero cancel-
lation between C(s) and G(s) .
As discussed in Chapter 6, noise and/or disturbance may enter a control system
at every terminal . Therefore we require every control system to be totally stable . We
compute the closed-loop transfer function from p to y shown in Figure 10 .2:
2
(S + 1)(S - 1)
Gyp (S) : = Y(S) =
P(S) 1 + (S + 1)(S - 1) 2
s(s + 2) (S + 1)(S - 1)
2
_ (S + 1)(S - 1) 2s(s + 2)
S(s + 2) + 2 (S - 1)(S + 1)(S 2 + 2s + 2)
s(s + 2)
It is unstable . Thus the output will approach infinity if there is any nonzero, no
matter how small, disturbance . Consequently, the system is not totally stable, and
the implementation is not acceptable .
Exercise 10 .2 .1
Consider a plant with transfer function 1/s 2. Implement G0(s) = 6/(s 2 + 3.4s
+ 6) in the unity-feedback configuration . Is the implementation acceptable?
[Answer : C(s) = 6s/(s + 3 .4), unacceptable .]
These examples show that the implementation of G0(s) in the unity-feedback

configuration will generally involve pole-zero cancellations . The canceled poles are
determined by the plant transfer function, and we have no control over them . In
general, if G(s) has open right-half-plane poles or two or more poles at s = 0, then
the unity-feedback configuration cannot be used to implement any G 0(s) . Thus, the
unity-feedback configuration cannot, in general, be used in model matching .

10 .3 UNITY FEEDBACK CONFIGURATION-POLE PLACEMENT

BY MATCHING COEFFICIENTS
Although the unity-feedback configuration cannot be used in every model matching,

it can be used to achieve arbitrary pole placement . In pole placement, we assign only
poles and leave zeros unspecified . We first use an example to illustrate the basic idea
and to discuss the issues involved .
Example 10 .3 .1
1
G (s) =
s(s + 2)
and consider the unity-feedback configuration shown in Figure 10 .1 . If the compen-
sator C(s) is a gain of k (a transfer function of degree 0), then the overall transfer
function can be computed as
kG(s) = 2 k
GG (s) -
1 + kG (s) s + 2s + k
This G0 (s) has two poles . These two poles cannot be arbitrarily assigned by choosing
a value for k. For example, if we assign the two poles at - 2 and - 3, then the
denominator of G0(s) must equal
s2 +2s+k=(s+2)(s+3)=s 2 +5s+6
Clearly, there is no k to meet the equation . Therefore, if the compensator is of degree
0, it is not possible to achieve arbitrary pole placement .'
Next let the compensator be proper and of degree 1 or
B0 + B1s
C(s) =
A0 + A 1s
with A 1 =A 0 . Then the overall transfer function can be computed as
B0 + B 1 s
G ,(s)
o(s)
1 + C(s)G(s) s(s + 2)(A 1s + A 0) + Bs + B 0
B1s + B0
A1 s3 + (2A 1 + A o )s2 + (2A0 + B 1 )s + B0
This G0 (s) has three poles . We show that all these three poles can be arbitrarily
assigned by choosing a suitable C(s) . Let the denominator of G0 (s) be
D0(s) = s 3 + F2S 2 + F1s + F0
'The root loci of this problem are plotted in Figure 7 .5 . If C(s) = k, we can assign the two poles only
along the root loci .

10 .3 UNITY-FEEDBACK CONFIGURATION-POLE PLACEMENT BY MATCHING COEFFICIENTS 389
where Fi are entirely arbitrary . Now we equate the denominator of G0 (s) with D0(s)
or
A1s3 + (2A 1 + A0)s2 + (2Ao + B 1)s + B0 = s3 + F2s2 + Fls + F0
Matching the coefficients of like power of s yields
A1 = 1 2A 1 +A 0 =F2 2Ao +B1 =F1 B0 =F0
which imply
A1 = 1 A 0 =F2 -2A 1 B 1 =F1 -2F2 +4A1 B0 =F0
For example, if we assign the three poles G0(s) as - 2 and - 2 2j, then D0(s)
becomes
D0(s) = s3 + F2S 2 + Fs + F0 = (s + 2)(s + 2 + 2j)(s + 2 - 2j)
=s3 +6s2 + 16s + 16
We mention that if a complex pole is assigned in D0(s), its complex conjugate must
also be assigned . Otherwise, D0(s) will have complex coefficients . For this set of
poles, we have
A1 = 1 A 0 =6-2=4 B1 = 16-2 .6+4=8 B0 = 16
and the compensator is
8s + 16
C(s) = s + 4
This compensator will place the poles of G0(s) at - 2 and - 2 2j. To verify this,
we compute
8s + 16 1
_ C(s)G(s) s + 4 s(s + 2)
G0(s)
1 + C(s)G(s) 1 + 8s + 16 1
s+4 s(s+2)
8s+ 16 8s+ 16
s(s + 2)(s + 4) + 8s + 16 s3 +6s2 + 16s+ 16
Indeed G0 (s) has poles at - 2 and - 2 2j. Note that the compensator also intro-
duces zero (8s + 16) into G0(s). The zero is obtained from solving a set of equations,
and we have no control over it . Thus, pole placement is different from pole-and-
zero placement or model matching .
This example shows the basic idea of pole placement in the unity-feedback
configuration. It is achieved by matching coefficients . In the following, we shall
extend the procedure to the general case and also establish the condition for achieving

pole placement. Consider the unity-feedback system shown in Figure 10.1 . Let
N(s) B(s)
G(s) = G ,(s)
D(s) C(s) A(s) (s) D(s)
and deg N(s) < deg D(s) = n. The substitution of these into (10 .1) yields
B(s) N(s)
C(s)G(s) A(s) D(s)
Q ,(s)
(s)
1 + C(s)G(s) + B(s) N(s)
1-~
A(s) D(s)
which becomes
N(s) = B(s)N(s)
G(s) = (10.4)
D(s) A(s)D(s) + B(s)N(s)
Given G(s), if there exists a proper compensator C(s) = B(s)/A(s) so that all
poles of G(s) can be arbitrarily assigned, the design is said to achieve arbitrary pole
placement. In the placement, if a complex number is assigned as a pole, its complex
conjugate must also be assigned . From (10 .4), we see that the pole-placement prob-
lem is equivalent to solving
A(s)D(s) + B(s)N(s) = D (s) (10.5)
This polynomial equation is called a Diophantine equation . In the equation, D(s)
and N(s) are given, the roots of D(s) are the poles of the overall system to be
assigned, and A(s) and B(s) are unknown polynomials to be solved. Note that B(s)
also appears in the numerator N(s) of G(s). Because B(s) is obtained from solving
(10 .5), we have no direct control over it and, consequently, no control over the zeros
of G(s). Note that before solving (10 .5), we don't know what G (s) will be ; therefore
C(s) = B(s)/A(s) cannot be computed from (10 .2).
10.3.1 Diophantine Equations

The crux of pole placement is solving the Diophantine equation in (10 .5) or
A(s)D(s) + B(s)N(s) = D (s)
In this equation, D(s) and N(s) are given, D(s) is to be chosen by the designer. The
questions are: Under what conditions will solutions A(s) and B(s) exist? and will the
compensator C(s) = B(s)/A(s) be proper? First we show that if D(s) and N(s) have
common factors, then D(s) cannot be arbitrarily chosen or, equivalently, arbitrary
pole placement is not possible. For example, if D(s) and N(s) both contain the factor
(s - 2) or D(s) = (s - 2)D(s) and N(s) = (s - 2)N(s), then (10 .5) becomes
A(s)D(s) + B(s)N(s) = (s - 2)[A(s)D(s) + B(s)N(s)] = D(s)
This implies that D(s) must contain the same common factor (s - 2). Thus, if N(s)
and D(s) have common factors, then not every root of D (s) can be arbitrarily as-

10.3 UNITY-FEEDBACK CONFIGURATION-POLE PLACEMENT BY MATCHING COEFFICIENTS 391
signed . Therefore we assume from now on that D(s) and N(s) have no common
factors .
Because G(s) and C(s) are proper, we have deg N(s) :!~; deg D(s) = n and
deg B(s) < deg A(s) = m, where deg stands for "the degree of ." Thus, D0 (s) in
(10 .5) has degree n + m or, equivalently, the unity-feedback system in Figure 10 .1
has (n + m) number of poles . We develop in the following the conditions under
which all (n + m) number of poles can be arbitrarily assigned .
If deg C(s) = 0 (that is, C(s) = k, where k is a real number), then from the
root-locus method, we see immediately that it is not possible to achieve arbitrary
pole placement . We can assign poles only along the root loci . If the degree of C(s)
is 1, that is
B(s) = B 0 + B1s
C(s)
= A(s) A0 + A1 s
then we have four adjustable parameters for pole placement . Thus, the larger the
degree of the compensator, the more parameters we have for pole placement . There-
fore, if the degree of the compensator is sufficiently large, it is possible to achieve
arbitrary pole placement .
Conventionally, the Diophantine equation is solved directly by using polyno-
mials and the solution is expressed as a general solution . The general solution, how-
ever, is not convenient for our application . See Problem 10 .19 and Reference [41] .
In our application, we require deg B(s) < deg A(s) to insure properness of compen-
sators . We also require the degree of compensators to be as small as possible . Instead
of solving (10 .5) directly, we shall transform it into a set of linear algebraic equations .
We write
D(s) : = Do + D1 s + D2s 2 + + Dnsn Dn 0 0 (10 .6a)
N(s) : = No + N1 s + N2S 2 + + Ns' (10 .6b)
A(s) := A 0 + A l s + A2 S 2 + + A ms' (10 .7a)
and
B(s) := B 0 + B 1 s + B 2 S 2 + . . . + Bs' (10 .7b)
where D,, Ni, A;, B; are all real numbers, not necessarily nonzero . Because deg D0(s)
= n + m, we can express D0(s) as
D0(s) = F0 + Fs + F2S2 + . . . + Fn+msn+m (10 .8)
The substitution of these into (10 .5) yields
(A0 + As + + A msm)(D 0 + D l s + + Ds')

+ (B o + B 1 s + + Bmsm)(N0 + N1s + + Nnsn)
= F0 + F1 s + F2S 2 + + Fn+mS n+m

which becomes, after grouping the coefficients associated with the same powers
of s,
(AoD0 + B oNo) + (A oD I + BON, + A ID0 + B1 N o )s +
+ (A mD, + BmN,i)Sn+m = FO + F1s + F2s2 + . . . + Fn+mS n+m
Matching the coefficients of like powers of s yields
A0D0 + B o No = FO
AoDI + B ON, + A 1Do + B 1 No = F l
AmDn + BmNn = Fn+m

There are a total of (n + m + 1) equations. These equations can be arranged in
i
matrix form as
Do No 0 0 ' 0 0 AO _ _
D1 N1 Do No ' Bo FO
0 0 A1 F1
Smcm := Dn Nn Dn_1 Nn-1 Do No B1 F2 (10 .9)
0 0 Dn Nn D1 NI
Am Fn+m
0 0 . 0 0 Dn Nn Bm
This is a set of (n + m + 1) linear algebraic equations . The matrix S m has

(n + m + 1) rows and 2(m + 1) columns and is formed from the coefficients
of D(s) and N(s) . The first two columns are simply the coefficients of D(s) and N(s)
arranged in ascending order. The next two columns are the first two columns shifted
down by one position . We repeat the process until we have (m + 1) sets of coeffi-
cients . We see that solving the Diophantine equation in (10 .5) has now been trans-
formed into solving the linear algebraic equation in (10.9).
As discussed in Appendix B, the equation in (10 .9) has a solution for any F, or,
equivalently, for any D0(s) if and only if the matrix S m has a full row rank. A
necessary condition for Sm to have a full row rank is that S m has more columns than
rows or an equal number of columns and rows:
n + m + 1<2(m+ 1) or n- 1 :m (10 .10)
Thus in order to achieve arbitrary pole placement, the degree of compensators in the
unity-feedback configuration must be n - 1 or higher. If the degree is less than
n - 1, it may be possible to assign some set of poles but not every set of poles.
Therefore, we assume from now on that m ? n - 1 .
With m ? n - 1, it is shown in Reference [15] that the matrix S m has a full
row rank if and only if D(s) and N(s) are coprime or have no common factors . We

mention that if m = n - 1, the matrix Sm is a square matrix of order 2n. In this

case, for every D0(s), the solution of (10 .9) is unique . Ifm ? n, then (10 .9) has more
unknowns than equations and solutions of (10 .9) are not unique .
Next we discuss the condition for the compensator to be proper or deg B(s)
deg A(s) . We consider first G(s) strictly proper and then G(s) biproper . If G(s) is
strictly proper, then Nn = 0 and the last equation of (10 .9) becomes
AmDn + B,,,Nn = A,nDn = Fn+m
which implies
Am = Fn+m (10.11)
Dn
Thus if Fn+m 0 0, then A. = 0 and the compensator C(s) = B(s)/A(s) is proper .
Note that if m = n - 1, the solution of (10 .9) is unique and for any desired poles,
there is a unique proper compensator to achieve the design . If m ? n, then the
solution of (10 .9) is not unique, and some parameters of the compensator can be
used, in addition to arbitrary pole placement, to achieve other design objective, as
will be discussed later .
If G(s) is biproper and if m = n - 1, then Sm in (10 .9) is a square matrix and
the solution of (10 .9) is unique . In this case, there is no guarantee that An _ 1 54 0
and the compensator may become improper. See Reference [15, p. 463 .] . If m ? n,
then solutions of (10.9) are not unique and we can always find a strictly proper
compensator to achieve pole placement . The preceding discussion is summarized as
theorems.
THEOREM 10 .1
Consider the unity-feedback system shown in Figure 10 .1 with a strictly proper
plant transfer function G(s) = N(s)/D(s) with deg N(s) < deg D(s) = n . It is
assumed that N(s) and D(s) are coprime . If m ? n - 1, then for any polynomial
D0 (s) of degree (n + m), a proper compensator C(s) = B(s)/A(s) of degree m
exists to achieve the design . If m = n - 1, the compensator is unique. If m
n, the compensators are not unique and some of the coefficients of the
compensators can be used to achieve other design objectives . Furthermore, the
compensator can be determined from the linear algebraic equation in (10 .9).
THEOREM 10 .2
Consider the unity-feedback system shown in Figure 10 .1 with a biproper plant
transfer function G(s) = N(s)/D(s) with deg N(s) = deg D(s) = n. It is
assumed that N(s) and D(s) are coprime . If m ? n, then for any polynomial D0 (s)
of degree (n + m), a proper compensator C(s) = B(s)/A(s) of degree m exists
to achieve the design . If m = n, and if the compensator is chosen to be strictly
proper, then the compensator is unique . If m ? n + 1, compensators are not
unique and some of the coefficients of the compensators can be used to achieve
other design objectives . Furthermore, the compensator can be determined from
the linear algebraic equation in (10 .9).

Example 10.3.2
N(s) - s - 2 -2 + s + 0 s 2
G(s) = D(s) (10.12)
(s + 1)(s - 1) - 1 + 0 s + S2
Clearly D(s) and N(s) have no common factor and n = 2. Let
B0 + B,s
C(s) =
A0 + As
It is a compensator of degree m = n - 1 = 1 . Arbitrarily, we choose the three
poles of the overall system as - 3, - 2 + j 1 and - 2 - j 1 . Then we have
D(s) :=(s+3)(s+2-jl)(s+2+jl)= 15 + 17s+7s 2 + S 3
We form the linear algebraic equation in (10 .9) as
-1 -2 ~ 0 0 A0 15
0 1 -1 -2 B0 17
(10 .13)
1 0 0 1 A, 7
0 0 1 0 B, 1-
Its solution can easily be obtained as
62 58
A0 = 79 A = 1 B = -
0 3 0 3 3
62 58
-- - -s
3 3 = -(58s + 62) = -(19 .3s + 20.7) .
C(s) - (10 .14)
79 3s+79 s+26 .3
-+s
3
and the resulting overall system is, from (10.4),
62 58
s (s - 2)
B(s)N(s) (_ 3 3 - (58s + 62)(s- 2)
(s)
G ,(s)
D(s) - s3 + 7s2 + 17s + 15 3(s3 + 7s2 + 17s + 15)
Note that the zero 58s + 62 is solved from the Diophantine equation and we have
no control over it . Because G(0) = 124/45, if we apply a unit-step reference input,
the output will approach 124/45 = 2 .76 . See (4.25). Thus the output of this overall
system will not track asymptotically step-reference inputs .

10.3 UNITY-FEEDBACK CONFIGURATION-POLE PLACEMENT BY MATCHING COEFFICIENTS 395
r U Y
k C(s) -1 G (s)
Figure 10.3 Unity-feedback system with a precompensator .
The design in Example 10.3.2 achieves pole placement but not tracking of step-
reference inputs . This problem, however, can be easily corrected by introducing a
constant gain k as shown in Figure 10 .3 . If we choose k so that kGo(0) =
k X 124/45 = 1 or k = 45/124, then the plant output in Figure 10 .3 will track
any step-reference input . We call the constant gain in Figure 10 .3 a precompensator.
In practice, the precompensator may be incorporated into the reference input r by
calibration or by resetting . For example, in temperature control, if ro, which corre-
sponds to 70, yields a steady-state temperature of 67 and rl yields a steady-state
temperature of 70 . We can simply change the scale so that ro corresponds to 67
and rl corresponds to 70 . By so doing, no steady-state error will be introduced in
tracking step-reference inputs .
Example 10 .3.3
Consider a plant with transfer function 1 /s 2 . Find a compensator in Figure 10 .1 so
that the resulting system has all poles at s = - 2. This plant transfer function has
degree 2 . If we choose a compensator of degree m = n - 1 = 2 - 1 = 1, then
we can achieve arbitrary pole placement . Clearly we have
D0(s)=(s+2) 3 =S 3 +6S 2 +12s+8
From the coefficients of
1 1 +0 s +0 s 2
S2 0+0 s + 1 s 2
and D0 (s), we form
0 1 0 0 Ao 8
0 0 0 1 Bo 12
1 0 0 0 At 6
0 0 1 1 0 Bt 1-
Its solution is
A0=6 At =1 I3 =8 Bi = 12
12s+8
C(s) =
s + 6

and the resulting overall system is

B(s)N(s) 12s + 8
G(s) = (10 .15)
D(s) s3 + 6s2 + 12s + 8
Note that the zero 12s + 8 is solved from the Diophantine equation and we have
no control over it . Because the constant and s terms of the numerator and denomi-
nator of G(s) are the same, the overall system has zero position error and zero
velocity error. See Section 6.3.1 . Thus, the system will track asymptotically any
ramp-reference input. For this problem, there is no need to introduce a precompen-
sator as in Figure 10.3. The reason is that the plant transfer function is of type 2 or
has double poles at s = 0 . In this case, the unity-feedback system, if it is stable,
will automatically have zero velocity error .
From the preceding two examples, we see that arbitrary pole placement in the
I
unity-feedback configuration can be used to achieve asymptotic tracking . If a plant
transfer function is of type 0, we need to introduce a precompensator to achieve
tracking of step-reference inputs . In practice, we can simply reset the reference input
rather than introduce the precompensator . If G(s) is of type 1, after pole placement,
the unity-feedback system will automatically track step-reference inputs . If G(s) is
of type 2, the unity-feedback system will track ramp-reference inputs . If G(s) is of
type 3, the unity-feedback system will track acceleration-reference inputs . In pole
placement, some zeros will be introduced. These zeros will affect the transient re-
sponse of the system . Therefore, it is important to check the response of the resulting
system before the system is actually built in practice .
To conclude this section, we mention that the algebraic equation in (10 .9) can
be solved by using MATLAB . For example, to solve (10 .13), we type
a=[-1 -2 0 0 ;0 1 -1 -2 ;1 0 0 1 ;0 0 1 0] ;b=[15 ;17 ;7 ;1] ;
alb
Then MATLAB will yield
26.3333
-20 .6667
1 .0000
-19 .3333
This yields the compensator in (10 .14).
Exercise 10 .3 .1
Redesign the problem in Example 10.3.1 by solving (10 .9).

Exercise 10.3 .2

1
G(s)
s2 - 1
Design a proper compensator C(s) and a gain k such that the overall system in Figure
10.3 has all poles located at - 2 and will track asymptotically step-reference inputs .
[Answers : C(s) = (13s + 14)/(s + 6), k = 4/7 .]
10.3 .2 Pole Placement with Robust Tracking

Consider the design problem in Example 10 .3.2, that is, given G(s) = (s - 2)/
(s + 1)(s - 1) in (10 .12), if we use the compensator in (10 .14) and a precompensator
k = 45/124, then the resulting overall system in Figure 10 .3 will track any step-
reference input. As discussed in Section 6 .4, the transfer function of a plant may
change due to changes of load, aging, wearing or external perturbation such as wind
gust on an antenna . Now suppose, after implementation, the plant transfer function
changes to G(s) = (s - 2.1)/(s + 1)(s - 0.9), the question is: Will the system in
Figure 10 .3 with the perturbed transfer function still track any step-reference input
without an error?
In order to answer this question, we compute the overall transfer function with
G(s) replaced by G(s):
s - 2.1 -(58s + 62)

45 (s + 1)(s - 0.9) 3s + 79
G(s) _ 124
1 + s - 2.1 -(58s + 62)
(s + 1)(s - 0.9) 3s + 79 (10 .16)
_ 45 58s 2 + 59.8s + 130.2
124 3s3 + 21.3s2 + 65s + 59 .1
Because G(0) = (45 X 130 .2)/(124 X 59 .1) = 0.799 0 1, the system no longer
tracks asymptotically step-reference inputs . If the reference input is 1, the output
approaches 0 .799 and the tracking error is about 20% . This type of design is called
nonrobust tracking because plant parameter perturbations destroy the tracking
property .
Now we shall redesign the system so that the tracking property will not be
destroyed by plant parameter perturbations. This is achieved by increasing the degree
of the compensator by one and then using the extra parameters to design a type 1
compensator . The original compensator in (10.14) has degree 1 . We increase its

degree and consider

B0 + B 1s + B2s 2
C(s) = (10.17)
A0 + A 1s + A2s2
Both the plant and compensator have degree 2, therefore the unity-feedback system
in Figure 10 .1 has four poles . We assign the four poles arbitrarily as - 3, - 3,
- 2 j, then we have
D0(s)=(s+3)2(s+2-jl)(s+2+jl)
(10.18)
=s4 + 10s3 +38s 2 +66s+45
The compensator that achieves this pole placement can be solved from the following
linear algebraic equation
-
-1 -2 0 0 0 0 A0 45
0 1 1 -2 0 0 B0 66
A,
1 0 0 1 -1 -2 = 38 (10.19)
B
0 0 1 0 0 1 1 10
0 0 0 0 1 0 A2 1
_B2_
This equation has six unknowns and five equations . After deleting the first column,
the remaining square matrix of order 5 in (10 .19) still has a full row rank . Therefore
A 0 can be arbitrarily assigned . (See Appendix B) . If we assign it as zero, then the
compensator in (10 .17) has a pole at s = 0 or becomes type 1 . With A 0 = 0, the
solution of (10 .19) can be computed as B0 = - 22.5, A1 = 68.83, B, = -78 .67,
A2 = 1, and B2 = - 58 .83 . Therefore, the compensator in (10 .17) becomes
- - (58 .83s 2 + 78.67s + 22.5)
C(s) (10.20)
s(s + 68 .83)
and the overall transfer function is
-(58 .83S 2 + 78 .67s + 22.5) s - 2
s(s + 68 .83) (s + 1)(s - 1)
G0(s) =
(58 .83s 2 + 78.67s + 22.5) s - 2
1 + -
s(s + 68 .83) (s + 1)(s - 1) (10.21)
-58 .83S 3 + 38 .99s2 + 134 .84s + 45
s4 + lOs 3 + 37 .99s2 + 66 .01s + 45
We see that, other than truncation errors, the denominator of G0(s) practically equals
D0(s) in (10 .18) . Thus the compensator in (10 .20) achieves the pole placement.
Because G0 (0) = 45/45 = 1, the unity-feedback system achieves asymptotic track-

ing of step-reference inputs . Note that there is no need to introduce a precompen-

sator as in Figure 10 .3 in this design, because the compensator is designed to be of
type 1 . _
Now we show that even if the plant transfer function changes to G(s) _
(s - 2.1)/(s + 1)(s - 0 .9), the overall system will still achieve tracking . With the
perturbed G(s), the overall transfer function becomes
-(58.83S 2 + 78 .67s + 22.5) s - 2.1
s(s + 68.83) (s + 1)(s - 0.9)
G,(s) -
-(58 .83S 2 + 78.67s + 22.5) s - 2.1
1 +
s(s + 68 .83) (s + 1)(s - 0.9) (10 .22)
_ - 58.83s3 + 44.873s 2 + 142 .707s + 47.25
s4 + 10.1s3 + 50.856s 2 + 80.76s + 47 .25
We first check the stability of G0(s). If G0(s) is not stable , it cannot track any signal .
The application of the Routh test to the denominator of G0(s) yields
s4 1 50 .856 47 .25
1/10 .1 s3 10.1 80.76 [0 42.86 47 .251
10.1/42 .86 s2 42.86 47.25 [0 69.631
5 69.63
1 47.25
All entries in the Routh table are positive, therefore G 0(s) is stable . Because
G0(0) = 47 .25/47 .25 = 1, the overall system with the perturbed plant transfer
function still tracks asymptotically any step-reference input . In fact, because the
compensator is of type 1, no matter how large the changes in the coefficients of the
plant transfer function, so long as the unity-feedback system remains stable, the
system will track asymptotically any step-reference input . Therefore the tracking
property of this design is robust .
Exercise 10 .3.3
Given a plant with transfer function 1/(s - 1), design a compensator of degree 0
so that the pole of the unity-feedback system in Figure 10 .3 is - 2 . Find also a
precompensator so that the overall system will track asymptotically any step-
reference input . If the plant transfer function changes to 1/(s - 1 .1), is the unity-
feedback system still stable? Will it still track any step-reference input without an
error?
[Answers : 3, 2/3, yes, no .]

Exercise 10 .3 .4
Given a plant with transfer function 1/(s - 1), design a compensator of degree I
so that the poles of the unity-feedback system in Figure 10 .3 are - 2 and - 2. In
this design, do you have freedom in choosing some of the coefficients of the com-
pensator? Can you choose compensator coefficients so that the system in Figure 10 .3
will track asymptotically step-reference inputs with k = 1? In this design, will the
system remain stable and track any step-reference input after the plant transfer func-
tion changes to 1 /(s - 1 .1)?
[Answers: [(5 - a)s + (4 + a)]/(s + a) ; yes ; yes by choosing a = 0; yes .]
To conclude this subsection, we remark on the choice of poles in pole placement .

Clearly, the poles chosen must be stable . Furthermore, they should not be very close
to the imaginary axis. As a guide, we may place them evenly inside the region C
shown in Figure 6.13 . Pole placement, however, will introduce some zeros into the
resulting system . These zeros will also affect the response of the system . (See Figure
2.16 .) There is no way to predict where these zeros will be located ; therefore, it
would be difficult to predict from the poles chosen what the final response of the
resulting system will be . It is therefore advisable to simulate the resulting system
before it is actually built .
10.3.3 Pole Placement and Model Matching

In this subsection, we give some remarks regarding the design based on pole place-
ment and the design based on model matching. We use an example to illustrate the
issues . Consider a plant with transfer function G (s) _ (s + 3)/s(s - 1) . Its quadratic
optimal system was computed in (9 .46) as G 0(s) = 10(s + 3)/(s 2 + 12.7s + 30)
under the constraint that the actuating signal due to a unit-step reference input meets
lu(l)l < 10 for all t ? 0. Now we shall redesign the problem by using the method
of pole placement . In other words, given G(s) = (s + 3)/s(s - 1), we shall find a
compensator of degree 1 in the unity-feedback configuration in Figure 10 .1 so that
the resulting system has a set of three desired poles . Because the poles of the quad-
ratic optimal system or, equivalently, the roots of (s 2 + 12.7s + 30) are - 9 .56 and
- 3.14, we choose the three poles as - 9.56, - 3 .14 and - 10. Thus we have
D0(s) = (s 2 + 12.7s + 30)(s + 10) = s 3 + 22.7s 2 + 157s + 300
and the compensator C(s) = (B o + B 1 s)/(A 0 + A l s) to achieve this set of pole
placement can be computed from
0 3 0 0 A0 300
-1 1 0 3 B0 _ 157
1 0 -1 1 A, 22.7
0 0 1 0 B, 1

as
20.175s + 100
C(s) =
s + 3 .525
Thus, from (10.4), the overall transfer function from r to y is
(20 .175s + 100)(s + 3)
G0(s)
- s3 + 22.7s 2 + 157s + 300
and the transfer function from r to u is
U(s) - G0(s) -(20.175s + 100)s(s - 1)
T(s) :
= R(s) G(s) s3 + 22.7s2 + 157s + 300
The unit-step responses of G0(s) and T(s) are plotted in Figure 10 .4 with solid lines .
For comparison, the corresponding responses of the quadratic optimal system are
also plotted with the dashed lines . We see that for the set of poles chosen, the
resulting system has a larger overshoot than that of the quadratic optimal system .
Furthermore, the actuating signal does not meet the constraint Ju(t)l < 10.
It is conceivable that a set of poles can be chosen in pole placement so that the
resulting system has a comparable response as one obtained by model matching .
The problem is that, in pole placement, there is no way to predict the response of
the resulting system from the set of poles chosen, because the response also depends
on the zeros which are yet to be solved from the Diophantine equation . Therefore,
pole placement design should consist of the following steps : (1) choose a set of
poles, (2) compute the required compensator, (3) compute the resulting overall trans-
fer function, and (4) check the response of the resulting system and check whether
the actuating signal meets the constraint . If the design is not satisfactory, we go back
to the first step and repeat the design . In model matching, we can choose an overall
system to meet design specifications . Only after a satisfactory overall system is
chosen do we compute the required compensator . Therefore, it is easier and simpler
to obtain a good control system by model matching than by pole placement .
Y (t) U (t)
A
12 25
20
08 15
"- Model matching
06 t 10
t
0. Pole placement 5
02 0
t -s t
0 .2 0.4 0 .6 0 .8 1 1 .2 1 .4 0 0 .2 0 .4 0 .6 0.8 1 1 .2 1 .4
(a) (b)
Figure 10.4 (a) Unit-step response. (b) Actuating signal .

10 .4 TWO-PARAMETER COMPENSATORS
Although the unity-feedback configuration can be used to achieve arbitrary pole

placement, it generally cannot be used to achieve model matching . In this section,
we introduce a configuration, called the two-parameter configuration, that can be
used to implement any implementable overall transfer function .
The actuating signal in Figure 10 .1 is of the form, in the Laplace transform
domain,
U(s) = C(s)(R(s) - Y(s)) = C(s)R(s) - C(s)Y(s) (10 .23)
That is, the same compensator is applied to the reference input and plant output to
generate the actuating signal . Now we shall generalize it to
U(s) = C 1 (s)R(s) - C2(S)Y(S) (10 .24)
as shown in Figure 10.5(a). This is the most general form of compensators . We call
C 1(s) feedforward compensator and C2(S)feedback compensator. Let C l (s) and C2 (s)
be
L(s) M(s)
C1(s) = A,(s) C2(s)
= A 2(s)
where L(s), M(s), A 1(s), and A2 (s) are polynomials . In general, A 1 (s) and A2(s) need
not be the same . It turns out that even if they are chosen to be the same, the two
compensators can be used to achieve any model matching . Furthermore, simple and
straightforward design procedures can be developed . Therefore we assume A 1 (s) _
A2 (s) = A(s) and the compensators become
C 1(s) = A (sj C2(s) = A ~ j (10 .25)
and the configuration in Figure 10 .5(a) becomes the one in Figure 10 .5(b). If A(s),
which is yet to be designed, contains unstable roots, the signal at the output of
L(s)/A(s) will grow to infinity and the system cannot be totally stable . Therefore the
configuration in Figure 10 .5(b) cannot be used in actual implementation . If we move
L(s)/A(s) into the feedback loop, then the configuration becomes the one shown in
Figure 10.5(c) . This configuration is also not satisfactory for two reasons : First, if
L(s) contains unstable roots, the design will involve unstable pole-zero cancellations
and the system cannot be totally stable . Second, because the two compensators
L(s)/A(s) and M(s)/L(s) have different denominators, if they are implemented using
operational amplifier circuits, they will use twice as many integrators as the one to
be discussed immediately . See Section 5 .6.1 . Therefore the configuration in Figure
10.5(c) should not be used . If we move M(s)/L(s) outside the loop, then the resulting
system is as shown in Figure 10 .5(d). This configuration should not be used for the
same reasons as for the configuration in Figure 10 .5(c). Therefore, the three config-
urations in Figures 10 .5(b), (c), and (d) will not be used in actual implementation .

10 .4 TWO-PARAMETER COMPENSATORS 403
r Y
C 2 (s)
(a)
r L (s) Y
A (s)
M(s)
A (s)
(b)
Y
G (s)
M(s)
L(s)
(c)
r L (s)
M(s) 0- M (S)
A(s)
U
91 G (s)
Y
(d)
Figure 10 .5 Various two-parameter configurations .
We substitute (10 .25) into (10 .24) and rewrite it into matrix form as
R(s) - M(s) [L(s) -M(s) R(s)
U(s) = L(s) Y(s) = (10 .26)
A(s) A(s) A(s) ] Y(s)
Thus the compensator
CL (s) - M(s)]
C(s) := [Cl(s) - Cz(s)l = = A-1(s)[L(s) -M(s)] (10 .27)
A(s) A(s)

is a 1 X 2 rational matrix ; it has two inputs r and y and one output u, and can be
plotted as shown in Figure 10 .6. The minus sign in (10.27) is introduced to take care
of the negative feedback in Figure 10 .6. Mathematically, this configuration is no
different from the ones in Figure 10 .5. However, if we implement C(s) in (10 .27)
as a unit, then the problem of possible unstable pole-zero cancellation will not arise .
Furthermore, its implementation will use the minimum number of integrators . There-
fore, the configuration in Figure 10 .6 will be used exclusively in implementation .
We call the compensator in Figure 10 .6 a two-parameter compensator [63] . The
configurations in Figure 10 .5 are called two-degree-of-freedom structures in [36] .
We can use the procedure in Section 5 .6 .1 to implement a two-parameter com-
pensator as a unit . For example, consider
10(s + 30)2 88.9s2 + 140s + 9000
[C1(s) - C2(s)] =
s(s - 15 .2) s(s - 15.2)
We first expand it as
752s + 9000 -1491.28s - 9000
I C I(S ) _ C2(S)1 = [10 -88 .91 +
C S 2 - 15.2s + 0 s 2 - 15.2s + 0
From its coefficients and (5 .44), we can obtain the following two-dimensional state-
variable equation realization
1 (t)1 15 752 -1491 .28 1 1(t)1
C 2(t) =C 0.2 10 ] [x1(t)1
X2(t) J + 9000 -9000 y(t)
u(t) = [1 0] [10 -88 .9]

Cx2(t)J + Cy(t)J
From this equation, we can easily plot a basic block diagram as shown in Figure
10.7. It consists of two integrators . The block diagram can be built using operational
amplifier circuits . We note that the adder in Figure 10 .6 does not correspond to any
adder in Figure 10.7.
y
L(s) A- ' (s) G (s)
I M(s)
Figure 10 .6 Two-parameter feedback configuration .

L_ _ J
Figure 10 .7 Basic block diagram .
Exercise 10 .4 .1
Find a minimal realization of

10(s + 30) 38.7s + 300
[C1(s) -C2(s)] =
L s + 5.025 s + 5 .025
and then draw a basic block diagram for it.
[Answer : i(t) = - 5.025x(1) + [249 .75 - 105 .53251
r(t)
u(t) = x(t) + [10 -38 .7] .]
y(t)
10 .4 .1 Two-Parameter Configuration-Model Matching

In this subsection, we present a procedure to implement any implementable G 0(s) in
the two-parameter configuration . From Mason's formula, the transfer function from
r to y in Figure 10 .6 is
Y(s) L(s)A-1(s)G(s)
R(s) 1 + A-1(s)M(s)G(s)
which becomes, after substituting G(s) = N(s)/D(s) and multiplying by
A(s)D(s)/A(s)D(s)
Y(s) L(s)N(s)
(10 .28)
R(s) A(s)D(s) + M(s)N(s)

Now we show that this can be used to achieve any model matching . For convenience,
we discuss only the case where G(s) is strictly proper .
Problem Given G(s) = N(s)/D(s), where N(s) and D(s) are coprime, deg N(s) <
deg D(s) = n, and given an implementable G(s) = N(s)/D(s), find proper com-
pensators L(s)/A(s) and M(s)/A(s) such that
N(s) L(s)N(s)
G( s) = D (10 .29)
(s) A(s)D(s) + M(s)N(s)
Procedure :
Step 1 : Compute
G (s) _ N(s) NP(s)
(10.30)
N(s) D(s)N(s) D,(s)
where NP(s) and DP(s) are coprime . Since N(s) and D(s) are coprime by
assumption, common factors may exist only between N (s) and N(s) . Can-
cel all common factors between them and denote the rest NP(s) and DP(s).
Note that if N(s) = N(s), then Dr (s) = D (s) and NP (s) = 1 . Using (10.30),
we rewrite (10.29) as
_ L(s)N(s)
G (s)
,(s) = (10 .31)
DP(s) A(s)D(s) + M(s)N(s)
From this equation, one might be tempted to set L(s) = Na(s) and to solve
for A(s) and M(s) from Dp(s) = A(s)D(s) + M(s)N(s). Unfortunately, the
resulting compensators are generally not proper . Therefore, some more
manipulation is needed .
Step 2 : Introduce an arbitrary Hurwitz polynomial Dp (s) so that the degree of
DP(s)DP (s) is at least 2n - 1. In other words, if deg Dp(s) : = p, then the
degree of DP (s) must be at least 2n - 1 - p. Because the polynomial
Dr(s) will be canceled in the design, its roots should be chosen to lie inside
an acceptable pole-zero cancellation region .
Step 3: Rewrite (10 .31) as

N(s)Np(s) N(s)[Np(s)DD(s)j _ N(s)L(s)
(10 .32)
G'(s) DP(s) Dp(s)Dp(s) A(s)D(s) + M(s)N(s)
Now we set
L(s) = Np(s)Dp(s) (10 .33)
and solve A(s) and M(s) from
A(s)D(s) + M(s)N(s) = Dp(s)Dp(s) = : F(s) (10 .34)

If we write
A(s) = A 0 + A ls + + As (10.35a)
M(s) = M0 + M1s + + Mms m (10 .35b)
and
2 (10.36)
F(s) := DP(s)15,(s) = F0 + F 1s + F2S + + Fn, mSn+m
with m ? n - 1, then A(s) and M(s) in (10 .34) can be solved from the
following linear algebraic equation :
Do No 0 0 0 0 A0
D 1 N1 Do No M0 F0 -
0 0 A1 F1
Dn Nn Dn-1 Nn-1 Do No M1 F2 (10 .37)
0 0 Dn Nn D1 N1
Am Fn+m
0 0 , 0 0 Dn Nn Mm
The solution and (10 .33) yield the compensators L(s)/A(s) and M(s)/A(s) .
Now we show that the compensators are proper . Equation (10.37) becomes
(10 .9) if M ; is replaced by B i . Thus, Theorem 10 .1 is directly applicable to (10 .37).
Because deg N(s) < deg D(s) and deg F(s) ? 2n - 1, Theorem 10 .1 implies the
existence of M(s) and A(s) in (10 .37) or (10 .34) with deg M(s) < deg A(s) . Thus,
the compensator M(s)/A(s) is proper. Furthermore, (10 .34) implies
deg A(s) = deg [DP(s)DP(s)] - deg D(s) = deg F(s) - n
Now we show deg L(s) < deg A(s) . Applying the pole-zero excess inequality of
G0(s) to (10 .32) and using (10 .33), we have
deg [DP(s)DP(s)] - (deg N(s) + deg L(s)) ? deg D(s) - deg N(s)
which implies
deg L(s) <_ deg [DP(s)Dp(s)] - deg D(s) = deg A(s)
Thus the compensator L(s)/A(s) is also proper . _
- The design always involves the pole-zero canc ellation of DP(s). The polynomial
DP(s), however, is chosen by the designer . Thus if DP (s) is chosen to be Hurwitz or
to have its roots lying inside the region C shown in Figure 6 .13, then the two-
parameter system in Figure 10 .6 is totally stable . The condition for the two-parameter
configuration to be well posed is 1 + G(-)C2(cc) 0 0 where C 2(s) = M(s)/A(s) .
This condition is always met if G(s) is strictly proper and C 2 (s) is proper. Thus the
system is well posed . The configuration clearly has no plant leakage . Thus this design

meets all the constraints discussed in Chapter 9 . In conclusion, the two-parameter

configuration in Figure 10 .6 can be used to implement any implementable overall
transfer function.
Example 10 .4.1
Consider the plant with transfer function
= N(s) = s + 3
G(s)
D(s) s(s - 1)
studied in (9 .44). Its ITAE optimal system was found in (9 .59) as
600 .25
G(s) = s (10 .38)
2 + 34.3s + 600 .25
Note that the zero (s + 3) of G(s) does not appear in G0 (s), thus the design will
involve the pole-zero cancellation of s + 3 . Now we implement G0 (s) by using the
two-parameter configuration in Figure 10 .6. We compute
G0 (s) 600.25 Ne (s)
N(s) (s 2 + 34.3s + 600 .25)(s + 3) Dp(s)
Next we choose 5p(s) so that the degree of Dp(s)_DD(s) is at least 2n - I =
3 . Becau se the degree of Dp(s) is 3, the degree of Dp(s) can be chosen as 0 . We
choose Dr(s) = 1 . Thus we have
L(s) = Np(s)Dp(s) = 600.25 X 1 = 600 .25
The polynomials A(s) = A 0 +_ A ls and M(s) = M o + M,s can be solved from
A(s)D(s) + M(s)N(s) = Dp(s)Dp( s) with
F(s) : = Dp(s)Dp(s) = (s 2 + 34.3s + 600 .25)(s + 3)
= s3 + 37 .3s2 + 703.15s + 1800 .75
or from the following linear algebraic equation :
Do No 0 0 A0 0 31 0 0 A0 1800.75
D, N, Do No Mo -1 1 0 3 Mo _ 703.15
D2 N2 D, N, A, 1 0 -1 1 A, 37.3
0 0 D 2 N2 M, 0 0 1 0 M, 1
The solution is A(s) = A 0 + A,s = 3 + s and M(s) = Mo + M,s = 600.25 +
35.3s. Thus the compensator is
600.25 35.3s + 600.25
ICI(s) _C2(S)1 = (10 .39)
s + 3 s + 3

Example 10 .4.2
Consider the same plant transfer function in the preceding example . Now we shall
implement its quadratic optimal transfer function G 0(s) = 10(s + 3)/(s 2 + 12.7s
+ 30) developed in (9 .46). First we compute
G0(s) _ 10(s + 3) _ 10 N,,(s)
N(s) (s 2 + 12 .7s + 30)(s + 3) s2 + 12.7s + 30 Dr(s)
Because the degree of Dp(s) is 2, we must introduce Dr (s) of degree at least 1
so that the degree of DD (s)DD(s) is at least 2n - 1 = 3 . Arbitrarily, we choose
Dr (s) = s + 3 (10 .40)
(This issue will be discussed further in the next section.) Thus we have
L(s) = Np(s)DD(s) = 10(s + 3)
and
F(s) = Dp(s)Dp (s) _ (s 2 + 12.7s + 30)(s + 3)
= s 3 + 15.7s2 + 68.1s + 90
The polynomials A(s) and M(s) can be solved from
0 3 0 0 A0 90
-1 1 0 3 Mo _ 68.1
(10 .41)
1 0 - 1 1 A, 15.7
0 0 1 0 M, 1
as A(s) = A1s + A 0 = s + 3 and M(s) = M l s + Mo = 13.7s + 30. Thus, the
compensator is
10(s + 3) 13.7s + 301
LC 1(s) - C2(S)l Cs + 3
13.7s + 301
s + 33 (10 .42)
_ 10 _
s + 33
This completes the design. Note that C,(s) reduces to 10 because DP (s) was chosen
as s + 3 . For different Dp (s), C 1 (s) is not a constant, as is seen in the next section .
Example 10 .4.3
Consider a plant with transfer function G(s) = 1/s(s + 2) . Implement its ITAE
optimal system G0(s) = 3/(s 2 + 2.4s + 3) . This G0(s) was implemented by using
the unity-feedback system in Example 10 .2.1. The design had the pole-zero cancel-
lation of s + 2, which was dictated by the given plant transfer function . Now we

implement G0(s) in the two-parameter configuration and show that the designer has
the freedom of choosing canceled poles . First we compute
G0(s) 3 NP(s)
N(s) (s2 + 2.4s + 3) . 1 Dr(s)
In this example, we have NP(s) = N0(s) = 3 and Dp(s) = D0(s) = s2 + 2.4s + 3.

We choose a polynomial DP(s) of degree 1 so that the degree of Dp(s)DP(s) is
2n - 1 = 3 . Arbitrarily, we choose Dr(s) = s + 10 . Then we have L(s) _
Np(s)DP(s) = 3(s + 10) and
F(s) = DP(s)DP(s) = (s 2 + 2.4s + 3)(s + 10) = s 3 + 12 .4s2 + 27s + 30

0 1 10 0 A0 30
2 0 0 1 Mo 27
1 0 2 0 A, 12.4
0 0 1 0 M, 1
as A(s) = A,s + A 0 = s + 10 .4 and M(s) = M1 s + Mo = 6.2s + 30 . This

completes the design .
Although this problem can be implemented in the unity-feedback and two-
parameter configurations, the former involves the pole-zero cancellation of
(s + 2), which is dictated by the plant; the latter involves the cancellation of (s +
10), which is chosen by the designer . Therefore, the two-parameter configuration is
more flexible and may be preferable to the unity-feedback configuration in achieving
model matching .
Exercise 10.4 .2
Given G(s) = 1/s(s - 1), implement

a. G0(s) = 4/(s 2 + 2.8s + 4)
b. G0(s) = (13s + 8)/(s 3 + 3.5s2 + 13s + 8)
in the two-parameter configuration. All canceled poles are to be chosen at s = - 4.
[Answers : (a) L(s)/A(s) = 4(s + 4)/(s + 7 .8), M(s)/A(s) = (23s + 16)/(s +
7.8). (b) L(s)/A(s) = (13s + 8)/(s + 4.5), M(s)/A(s) _
(17 .5s + 8)/(s + 4.5).]

10 .5 EFFECT OF D0 (s) ON DISTURBANCE REJECTION AND ROBUSTNESS 411
10 .5 EFFECT OF DF,(s)ON DISTURBANCE REJECTION AND ROBUSTNESS 2
In the two-parameter configuration, we must introduce a Hurwitz polynomial Dr(s)

in (10 .32) or
_ N(s)L(s)
(s)
G ,(s) D
p(s)Dp(s) A(s)D(s) + M(s)N(s)
to insure that the resulting compensators are proper . Because Dr(s) is completely
canceled in G0(s), the tracking of r(t) by the plant output y(t) is not affected by the
choice of Dr(s). Neither is the actuating signal affected by Dp(s), because the transfer
function from r to u is
o(s) N(s)NN(s)DD(s)D(s) NN(s)D(s)

T(s)
(s) : (10.43)
= R(s) = G
G(s) N(s)DD(s)DD(s) Dr(s)
where Dr(s) does not appear directly or indirectly . Therefore the choice of Dr(s)
does not affect the tracking property of the overall system and the magnitude of the
actuating signal .
Although Dr (s) does not appear in G0(s), it will appear in the closed-loop transfer
functions of some input/output pairs . We compute the transfer function from the
disturbance input p to the plant output y in Figure 10 .6:
= Y(s) = G(s)
H(s) :
P(s) 1 + G(s)M(s)A - t (s) (10.44)
N(s)A(s) N(s)A(s)
A(s)D(s) + M(s)N(s) Dp(s)Dp(s)
We see that Dr(s) appears directly in H(s); it also affects A(s) through the Diophantine
equation in (10 .34). Therefore the choice of Dr(s) will affect the disturbance rejection
property of the system. This problem will be studied in this section by using
examples. _
In this section, we also study the effect of Dr(s) on the stability range of the
overall system . As discussed in Section 6 .4, the plant transfer function G(s) may
change due to changes of the load, power supplies, or other reasons . Therefore it is
of practical interest to see how much the coefficients of G(s) may change before the
overall system becomes unstable . The larger the region in which the coefficients of
G(s) are permitted to change, the more robust the overall system is . In the following
examples, we also study this problem .
'May be skipped without loss of continuity .

Example 10 .5 .1
Consider a plant with transfer function G(s) = (s + 3)/s(s - 1). Implement
its quadratic optimal system G0(s) = 10(s + 3)/(s 2 + 12.7s + 30) in the two-
parameter configuration . As shown in Example 10 .4.2, we must choose DP (s) of
degree 1 to achieve the design . If PP(s) is chosen as (s + 3), then the compensator
was computed in (10.42) . For this Dp(s) and compensator, the transfer function from
p to y can be computed as
N(s)A(s) (s + 3)(s + 3) - s + 3
H(s) = (10.45)
Dp(s)DD(s) (s 2 + 12 .7s + 30)(s + 3) s 2 + 12.7s + 30
If Dp(s) is chosen as (s + 30), using the same procedure as in Example 10.4 .2, we
can compute the compensator as
10(s
+ _ 38.675s + 3001
[CI(s) -C2(s)] (10 .46)
E s + 5 .025 s + 5 .025 J
For this Dp(s) and compensator, we have
H(s) =
H( (s + 3)(s + 5.025)
(10 .47)
(s2 + 12 .7s + 30)(s + 30)
Next we choose Dp(s) = s + 300, and compute the compensator as
10(s + 300) _ 288.425s + 3000

[C1 (s) -C2(s)] (10.48)
s + 25 .275 s + 25 .275
For this Dp(s) and compensator, we have
(s + 3)(s + 25.275)
H(s) = (10.49)
(s 2 + 12.7s + 30)(s + 300)
Now we assume the disturbance p to be a unit-step function and compute the plant
outputs for the three cases in (10 .45), ( 10.47), and (10.49). The results are plotted
in Figure 10 .8 with the solid line for Dr(s) = s + 3, the dashed line for Dp(s) _
5
s + 30, and the dotted line for p(s) = s + 300 . We see that the system with
DP(s) = s + 300 attenuates the disturbance most . We plot in Figure 10.9 the am-
plitude characteristics of the three H(s) . The one corresponding to Dp(s) = s + 300
again has the best attenuation property for all w . Therefore, we conclude that, for
this example, the faster the root of Dr (s), the better the disturbance rejection property .
_ Now we study the robustness property of the system . First we consider the case
Dp (s) = s + 300 with the compensator in (10 .48). Suppose that after the imple-

10 .5 EFFECT OF Do(s) ON DISTURBANCE REJECTION AND ROBUSTNESS 413
Y (t)
A
0 .12
0 .1
Dp(s)=s+3
0 .08
0 .06
0 .04
0 .02 s+30
---------------------
s+300
---------------------------------------------------------------------- ---------- ----------------------------
>t
0 0 .1 0.2 0 .3 0 .4 0.5 0 .6 0 .7 0 .8 0 .9 1
Figure 10 .8 Effect of canceled poles on disturbance rejection (time domain) .
mentation, the plant transfer function G(s) changes to
N(s) =s + 3 + eZ
G(s) (10 .50)
= D(s) s(s - 1 + E l )
With this perturbed transfer function, the transfer function from r to y becomes
L(s)N(s)
Go (s) =
A(s)D(s) + M(s)N(s)
I H(jw) I
dB
A
-10-
D (s)=s+3
-20
-30
-40
-50
-60
-70
-80 , . 1,,,, , , . . 1- .1 , . , , I ., ., , , . . . . . >co

10 -1 10 0 10 1 10 , 10 3 10 4
Figure 10 .9 Effect of canceled poles on disturbance rejection (frequency domain) .

The perturbed overall system is stable if

A(s)D(s) + M(s)N(s) = (s + 25 .275) - s(s - 1 + 1 )
+ (288.425s + 3000)(s + 3 + 2)
= S 3 + (312 .7 + 1 )S 2
+ (3840 + 25 .275 1 + 288 .452 )s + 3000(3 + 2 )
(10 .51)
is a Hurwitz polynomial . The application of the Routh test to (10 .51) yields the
following stability conditions:
312 .7 + 1 > 0 3 + 2 > 0 (10 .52a)
and
3000(3 + 21 >
3840 + 25 .275 1 + 288 .4252 - 0 (10 .52b)
312 .7 + ,
(See Exercise 4.6.3.) These conditions can be simplified to
2 > - 3 if 1 > - 117 .7
and
1191768 + 11743 .4925E, + 25.275E 2
2 > - if - 302 .29 < 1 < - 117 .7
87190 .4975 + 288 .425 1
It is plotted in Figure 10 .10 with the dotted line .
15
10 s+3
-5 El
-400 -300 -200 -100 0 100 200 300
Figure 10 .10 Effect of canceled poles on stability range.

10.5 EFFECT OF 5,(s) ON DISTURBANCE REJECTION AND ROBUSTNESS 415
_ In order to make comparisons, we repeat the computation for the cases

Dr (s) = s + 30 and DP (s) = s + 3 . Their stability regions are also plotted in Figure
10.10 respectively with the dashed line and solid line . We see that the region cor-
responding to DP(s) = s + 300 is the largest and the one corresponding to DP (s) =
s + 3 is the smallest . Thus we conclude that for this problem, the faster the root of
Dp(s), the more robust the resulting system is .
Example 10 .5.2
Consider a plant with transfer function G(s) = N(s)/D(s) = (s - 1)/s(s - 2) .
Implement its quadratic optimal system G0(s) = - 10(s - 1)/(s 2 + 11 .14s + 10)
in the two-parameter configuration . First, we compute
G0(s) -10(s - 1) - - 10 NP(s)
(10.53)
N(s) - (s 2 + 11 .14s + 10)(s - 1) s2 + 11 .14s + 10 . DP(s)
Because the degree of DP(s) is 2, which is smaller than 2n - 1 = 3, we must
choose a Hurwitz polynomial DP (s) of degree at least 1 . We choose
DP(s)=s+0
Then we have
Dp(s)Dp(s) = (s 2 + 11.14s + 10)(s + )3)
=s3 +(11 .14+ B)s 2 +(10+ 11 .14/3)s+ 10(3
and
L(s) = Np(s)DP(s) = - 10(s + 6)
0 -1 0 0 A0 10/3
-2 1 0 -1 Mo 10 + 11 .14/3
1 0 ; -2 1 A1 11 .14 + /3
0 0 1 0 M1 1
This can be solved directly . It can also be solved by computing
-1
0 -1 0 0 -1 -1 -1 -2
-2 1 0 -1 -1 0 0 0
1 0 -2 1 0 0 0 1
0 0 1 0 1 1 2 4

Thus we have
A0 -1 -1 -1 -2 10/3 - 23 .14 - 22.14/3
M0 -1 0 0 0 10 + 11 .14/3 _ -10/3
A, = 0 0 0 1 11 .14 + 0 1
M1 1 1 2 4 1 36.28 + 23 .14/3
and the compensator is
F
L(s) M(s)
ICI(S) - C'2(s)l =
A(s) A(s)
-10(s + /3)(36.28 + 23.14/3)s - 10/31
s-23.14-22.14/3 s-23.14-22.14(3
(10.54)
This completes the implementation of the quadratic optimal system . Note that no
matter what value /3 assumes, as long as it is positive, Dr(s) = s + /3 will not affect
the tracking property of the overall system . Neither will it affect the magnitude of
the actuating signal . _
Now we study the effect of Dr (s) on disturbance rejection . The transfer function
from the disturbance p to the plant output y is
Y(s) N(s)A(s) = (s - 1)(s - 23.14 - 22.14/3) '
H(s) : (10.55)
= P(s) = Dp (s)Dp (s) (s 2 + 11.14s + 10)(s + /3)
Let the disturbance be a unit-step function . We compute the unit-step resp onses of
(10.55) on a personal computer and plot the results in Figure 10.11 for Dr(s) =
Y(0
5
-2 t
0 1 2 3 4 5 6
Figure 10.11 Effect of canceled poles on disturbance rejection (time domain) .

10 .5 EFFECT OF Dr(s) ON DISTURBANCE REJECTION AND ROBUSTNESS 417
s + 1 (solid line), Dp(s) = s + 10 (dashed line), and Dp (s) = s + 100 (dotted

line). We see that the choice of Dr (s) does affect the disturbance rejection property
of the system . Although the one corresponding to Dr (s) = s + 100 has the smallest
steady-state value, its undershoot is the largest. We plot in Figure 10 .12 the amplitude
characteristics of H(s) for /3 = 1, 10, and 100 respectively with the solid line, dashed
line, and dotted line . The one corresponding to /3 = 100 has the largest attenuation
for small w, but it has less attenuation for w ? 2. Therefore, for this example, the
choice of Dp (s) is not as clear-cut as in the preceding example . To have a small
steady-state effect, we should choose a large /3 . If the frequency spectrum of dis-
turbance lies mainly between 2 and 1000 radians per second, then we should choose
a small /3 . _
Now we study the effect of Dr (s) on the robustness of the overall system . Sup-
pose after the implementation of G0(s), the plant transfer function G(s) changes to
G(s) (10.56)
D(s) s(s - 2 + el)
With this plant transfer function and the compensators in (10 .54), the overall transfer
function becomes
L(s)N(s)
G0 (s) =
A(s)D(s) + M(s)N(s)
with
A(s)D(s) + M(s)N(s)
_ (s - 23 .14 - 22.14/3) . s(s - 2 + El ) ( 10.57)
+ [(36.28 + 23 .14f3)s - 10/31 . (s - 1 + E2 )
I H(iw) I
dB
Figure 10 .12 Effect of canceled poles on disturbance rejection (frequency domain) .

We compute its stability ranges for the three cases with Dp(s) = s + 1,
Dp(s) = s + 10, and Dp(s) = s + 100. If Dp(s) = s + 1, (10 .57) becomes
A(s)D(s) + M(s)N(s) = (s - 45 .28) s(s - 2 + el )

+ (59 .42s - 10)(s - 1 + E2)
=s3 +(12 .14+1)s 2
+ (21 .14 - 45 .28, + 59.42E2)s + 10(1 - E2)
It is Hurwitz under the following three conditions
12.14 + E, >0 1 - E2 >0
and
10(1 -
(21 .14 - 45.28, + 59.422) - EZ) > 0
(12 .14 + ,)
(See Exercise 4.6.3 .) These conditions can be simplified as
-246.6396 + 528.5592, + 45.28;

< EZ < 1 for - 12.14 < E, < 1 .78
731 .3588 + 59.42,
From these inequalities, the stability range of E, and 2 can be plotted as shown in
Figure 10 .13 with the solid line . We repeat the computation for DP (s) = s + 10
and Dp(s) = s + 100 and plot the results in Figure 10 .13, respectively, with the
E 2
Figure 10 .13 Effect of canceled poles on stability range .

10 .5 EFFECT OF DP(s) ON DISTURBANCE REJECTION AND ROBUSTNESS 419
dashed line and dotted line . We see that, roughly speaking, the larger /3, the larger
the stability range . (Note that in the neighborhood of ei = 0 and E2 = 0, the stability
region correspon ding to DP (s) = s + 100 does not include completely the regions
corresponding to DP(s) = s + 10 and DP(s) = s + 1 .) Therefore we conclude that
the faster the root of Dp(s), the more robust the overall system .
From the preceding two examples, we conclude that the choice of DP(s) in the
two-parameter configuration does affect the disturbance rejection and robustness
properties of the resulting system . For the system
_ in Example 10 .5.1, which has no
non-minimum-phase zeros, the faster the root of DP (s), the better the step disturbance
rejection and the more robust the resulting system . For the system in Example 10.5.2,
which has a non-minimum-phase zero, the choice of DP (s) is no longer clear-cut . In
conclusion, in using the two-parameter configuration to achieve model matching,
although the choice of DP (s) does not affect the tracking property of the system and
the magnitude of the actuating signal, it does affect the disturbance rejection and
robustness properties of the system . Therefore, we should utilize this freedom in the
design . No general rule of choosing DP (s) seems available at present . However, we
may always choose it by trial and error .
10.5.1 Model Matching and Disturbance Rejection

A system is said to achieve step disturbance rejection if the plant output excited by
any step disturbance eventually vanishes, that is,
lim yP(t) = 0 (10 .58)
where yy(t) is the plant output excited by the disturbance p(t) shown in Figure 10 .6.
In the examples in the preceding section, we showed that by choosing the root of
DP (s) appropriately, the effect of step disturbances can be reduced . However, no
matter where the root is chosen, the steady-state effect of step disturbances can never
be completely eliminated . Now we shall show that by increasing the degree of Dr(s),
step disturbances can be completely eliminated as t-->-.
The transfer function from p to y in Figure 10 .6 is
N(s)A(s)
H(s) = (10 .59)
Dp(s)Dp(s)
If p is a step function with magnitude a, then
N(s)A(s) a
YP(s) = H(s)P(s) = (10.60)
DP(s)Dp (s) . s
The application of the final-value theorem to (10 .60) yields
N(s)A(s) a aN(0)A(0)
lim yp( t) = lim s (10.61)
s-o Dp(s)Dp(s) s Do(0)DP(0)

This becomes zero for any a if and only if_N(0) = 0 or A(0) = 0 . Note that
D,(0) :A 0 and Dp(0) =A 0 because Dp(s) and Dp(s) are Hurwitz . The constant N(0)
is given and is often nonzero . Therefore, the only way to achieve disturbance rejec-
tion is to design A(s) with A(0) = 0 . Recall that A(s) is to be solved from the
Diophantine equation in (10 .34) or the linear algebraic equation in (10 .37). If the
degree of Dp(s) is chosen so that the degree of Dp (s)Dp (s) is 2n - 1, where n =
deg D(s), then the solution A(s) is unique and we have no control over A(0) . How-
ever, if we increase the degree of Dp (s), then solutions A(s) are no longer unique
and we may have the freedom of choosing A(0). This will be illustrated by an
example .
Example 10.5.3
Consider a plant with transfer function G(s) = (s + 3)/s(s - 1) . Implement its
quadratic optimal system G0(s) = 10(s + 3)/(s 2 + 12.7s + 30) . This was imple-
mented in Example 10.5.1 by choosing the degree of Dp (s) as 1 . Now we shall
increase the degree of Dr(s) to 2 and repeat the design . First we compute
G0 (s) 10 NP(s)
N(s) s 2 + 12.7s + 30 . Dp(s)
Arbitrarily, we choose
DP(s) = (s + 30)2 (10.62)
Then we have
Dp (s)Dp(s) = (s 2 + 12.7s + 30)(s + 30) 2
= s 4 + 72.7s3 + 1692s 2 + 13,230s + 27,000
and
L(s) = Np (s)Dp(s) = 10(s + 30) 2
The polynomials A(s) = A 0 + A,s + A 2 s 2 and M(s) = M0 + M,s + M2S2 can
be solved from
0 31 0 0~ 0 0 27,000
MO
-1 1i 0 3 0 0 13,230
A,
1 0 -1 1 0 3 1692 (10.63)
M,
0 0 1 0 -1 1 72.7
A2
0 01 0 0 1 0 1
M2 _
This has 5 equations and 6 unknowns . Because the first column of the 5 X 6 matrix
is linearly dependent on the remaining columns, A 0 can be arbitrarily assigned, in
particular, assigned as 0 . With A 0 = 0, the solution of (10 .63) can be computed as

10 .5 EFFECT OF 4,(s) ON DISTURBANCE REJECTION AND ROBUSTNESS 421
Y (t)
0.01
0 .005
-0 .005
-0.01
-0 .015 1
0 0 .1 0 .2 0.3 0 .4 0.5 0 .6 0 .7 0 .8 0 .9 1
Figure 10 .14 Disturbance rejection .
A, = - 15.2, A2 = 1, Mo = 9000, M, = 1410, and M 2 = 88.9. Thus the com-

pensator is
10(s + 30)2 - 88.9s2 + 1410s + 9000
[C,(s) -C2(s)] (10.64)
s(s - 15.2) s(s - 15.2)
This completes the design . With this compensator, the transfer function from r to y
is still G 0 (s) = 10(s + 3)/(s 2 + 12.7s + 30) . Therefore, the tracking property of
the system remains unchanged . Now we compute the transfer function from the
disturbance p to the plant output y :
Y(s) = N(s)A(s) - (s + 3)s(s - 15.2)
H(s) : - (10.65)
P(s) DP(s)Dp(s) (s 2 + 12.7s + 30)(s + 30)2
Because H(0) = 0, if the disturbance is a step function, the excited plant output will
approach zero as t - -, as shown in Figure 10 .14 with the solid line . As a com-
parison, we also show in Figure 10 .14 with the dashed line the plant output due to
a step disturbance for the design using DP (s) = s + 300 . We see that by increasing
the degree of DP(s), it is possible to achieve step disturbance rejection . In actual
design, we may try several Dr(s) of degree 2 and then choose_one which suppresses
most disturbances . In conclusion, by increasing the degree of DP (s), we may achieve
disturbance rejection .
Exercise 10 .5 .1
Repeat the preceding example by choosing DP (s) _ (s + 300)2.

[Answer : H(s) = s(s + 3)(s - 2200)/(s 2 + 12.7s + 30)(s + 300)2.]

10 .6 PLANT INPUT/OUTPUT FEEDBACK CONFIGURATION
Consider the configuration shown in Figure 10 .15(a) in which G(s) is the plant
transfer function and C 1(s), C2(s), and C0(s) are proper compensators . This config-
uration introduces feedback from the plant input and output ; therefore, it is called
the plant input/output feedback configuration or plant I/O feedback configuration
for short . This configuration can be used to implement any implementable GO (s) .
Instead of discussing the general case, we discuss only the case where
deg D0(s) - deg N0 (s) = deg D(s) - deg N(s) (10 .66)
In other words, the pole-zero excess of G0(s) equals that of G(s) . In this case, we
can always assume C0(s) = 1 and
L(s) M(s)
C1 C2 (10 .67)
(s) = A(s) (s) = A(s)
and the plant I/O feedback configuration can be simplified as shown in Figure
10.15(b) . Note that A(s), L(s), and M(s) in Figure 10 .15(b) are different from those
in the two-parameter configuration in Figure 10 .6 . The two compensators enclosed
by the dashed line can be considered as a two-input, one-output compensator and
must be implemented as a unit as discussed in Section 10 .4. The configuration has
two loops, one with loop gain -L(s)/A(s), the other -G(s)M(s)/A(s) . Thus its
r u y
0 -
G (s)
+
C 1 (s)
0 - C2
(s) F
(a)
r U
G (s)
I
L (s) -~O c M(s)
I
I
w
I A -1 (s) E--
I
(b)
Figure 10 .15 Plant I/O feedback system .

10 .6 PLANT INPUT/OUTPUT FEEDBACK CONFIGURATION 423
characteristic function equals

L(s) M(s) L(s) N(s)M(s)
0(s) = 1 - ( - A(s) - G(s) (
A s)
= 1 + A(s) +
D(s)A(s) (10.68)
and the transfer function from r to y equals, using Mason's formula,

G(s) = N(s)A(s)
G0(s) = (10.69)
A(s) A(s)D(s) + L(s)D(s) + M(s)N(s)
Now we develop a procedure to implement G0(s).
Problem Given G(s) = N(s)/D(s), where N(s) and D(s) are coprime and deg N(s)
!!~; deg D(s). Implement an implementable G0(s) = N0(s)/D 0(s) with deg D0(s) -
deg N0(s) = deg D(s) - deg N(s) .
Procedure :
Step 1 : Compute
Go(s) - NN(s) _ Np(s)
(10 .70)
N(s) D0(s)N(s) Dr(s)
where NP(s) and Dp(s) have no common factors .
Step 2 : If deg Na(s) = : m < n - 1, introduce an arbitrary Hurwitz polynomial
A(s) of degree n - 1 - m. If m ? n - 1, set A(s) = 1
Step 3 : Rewrite (10 .70) and equate it with (10 .69) :
N(s)Np(s) N(s)NN(s)A(s)
G0(s) = (10 .71)
Dr(s) Dp(s)A(s)
N(s)A(s)
A(s)D(s) + L(s)D(s) + M(s)N(s)
From this equation, we have
A(s) = Np(s)A(s)
_ (10 .72)
A(s)D(s) + L(s)D(s) + M(s)N(s) = D D(s)A(s)
which becomes, after substituting (10 .72)
L(s)D(s) + M(s)N(s) = DD(s)A(s) - A(s)D(s)
= DD(s)A(s) - NN(s)A(s)D(s) (10 .73)
= A(s)(Dp(s) - Np(s)D(s)) = : F(s)

This Diophantine equation can be used to solve L(s) and M(s). Because of
the introduction of A(s), A(s) in (10 .72) has a degree at least n - 1 and
the degrees of L(s) and M(s) are insured to be at most equal to that of A(s) .
Therefore, the resulting compensators L(s)/A(s) and M(s)/A(s) are proper .
Equation (10 .73) is similar to (10 .34), therefore it can also be solved using
a linear algebraic equation . For example, if deg A(s) = n - 1, then deg

F(s) = 2n - 1 . Let
L(s) = Lo + L 1 s + + Ln_1sn-1 (10 .74a)
M(S) = M0 + M1 S + . . . + Mn-1sn-I (10 .74b)
and
F(s) = F0 + F ls + F2S 2 + . . . + F2n-IS 2n-1 (10 .74c)
Then L(s) and M(s) can be solved from the following linear algebraic
equation :
Do No 0 0 0 0 Lo
D1 N1 Do NO MO F0
L1 F1
Dn Nn Dn-1 Nn-1 M1 F2 (10 .75)
0 0 Dn Nn
Ln-1 F2n -1
0 0 . 0 0 Mn-1
This is illustrated by an example .
Example 10 .6.1
Consider G(s) = (s + 3)/s(s - 1). Implement its quadratic optimal system G0(s)
= 10(s + 3)/(s 2 + 12.7s + 30) . This problem was implemented in the two-
parameter configuration in Example 10.4.2. Now we shall implement it in the plant
I/O feedback configuration shown in Figure 10 .15(b) . First we compute
G0(s) _ 10(s + 3) _ 10 No(s)
N(s) (s2 + 12 .7s + 30)(s + 3) s2 + 12.7s + 30 Dr(s)
Because the degree of NP(s) is 0, we must introduce a Hurwitz polynomial A(s) of
degree at least n - 1 - deg NP(s) = 1 . Arbitrarily, we choose
A(s) = s + 3 (10 .76)
Then we have
A(s) = Np(s)A(s) = 10(s + 3)
and, from (10 .73),
F(s) = A(s)(Dp(s) - Np(s)D(s)) = (s + 3) [s 2 + 12.7s + 30 - 10s(s - 1)]
_ -95 3 -4 .3s2 +98 .1s+90

The polynomials L(s) and M(s) can be solved from

0 31 0 0 Lo 90
-1 1 0 3 Mo 98.1
(10.77)
1 0 -1 1 L, -4.3
0 0 1 0 M1 -9
as L(s) = L,s + L o = -9s - 27 and M(s) = M,s + Mo = 13.7s + 30. Thus
the compensator is
L(s) - -9s - 27 = - 9
C , (s) =
A(s) 10(s + 3) 10 (1078x)
= 13.7s + 30
CZ(s)
2(S) = (10.78b)
A(s) 10(s + 3)
This completes the design. _Note that C I(s) reduces to a constant because A(s) was
chosen as s + 3 . Different A(s) will yield nonconstant C,(s) .
We see that the design using the plant I/O feedback configuration is quite similar
to that of the two-parameter configuration . Because A(s) is completely canceled in
G0 (s), the choice of A(s) will not affect the tracking property and actuating signal of
the system . However, its choice may affect disturbance rejection and stability ro-
bustness of the resulting system . The idea is similar to the two-parameter case and
will not be repeated.
Exercise 10.6 .1
Consider a plant with transfer function G(s) = 1/s(s - 1). Find compensators in
the plant I/O feedback configuration to yield (a) G0(s) = 4/(s 2 + 2.8s + 4) and
(b) G0(s) = (13s + 8)/(s 3 + 3.5s2 + 13s + 8) . All canceled poles are to be chosen
at s = - 4.
[Answers : (a) L(s)/A(s) = (- 3s - 8 .2)/4(s + 4), M(s)/A(s) =
(23s + 16)/4(s + 4) . (b) L(s)/A(s) = (- 12s - 3 .5)/(13s + 8),
M(s)/A(s) = (17 .5s + 8)/(13s + 8).]
10 .7 SUMMARY AND CONCLUDING REMARKS
This chapter discusses the problem of implementing overall transfer functions . We

discuss first the unity-feedback configuration . Generally, the unity-feedback config-
uration cannot be used to achieve model matching . However, it can always be used
to achieve pole placement . If the degree of G(s) is n, the minimum degree of com-
pensators to achieve arbitrary pole placement is n - 1 if G(s) is strictly proper, or
n if G(s) is biproper. If we increase the degree of compensators, then the unity-
feedback configuration can be used to achieve pole placement and robust tracking .
The two-parameter configuration can be used to achieve any model matching .
In this configuration, generally we have freedom in choosing canceled poles . The
choice of these poles will not affect the tracking property of the system and the
magnitude of the actuating signal . Therefore, these canceled poles can be chosen to
suppress the effect of disturbance and to increase the stability robustness of the
system. If we increase the degree of compensators, then it is possible to achieve
model matching and disturbance rejection .
Finally we introduced the plant input/output feedback configuration . This con-
figuration is developed from state estimator (or observer) and state feedback (or
controller) in state-variable equations . See Chapter 11 . The configuration can also
be used to achieve any model matching . For a comparison of the two-parameter
configuration and the plant I/O feedback configuration, see References [16, 44] .
In this chapter all compensators for pole placement and model matching are
obtained by solving sets of linear algebraic equations . Thus the method is referred
to as the linear algebraic method.
We now compare the inward approach and the outward approach . We intro-
duced the root-locus method and the frequency-domain method in the outward ap-
proach . In the root-locus method, we try to shift the poles of overall systems to the
desired pole region . The region is developed from a quadratic transfer function with
a constant numerator. Therefore, if an overall transfer function is not of the form,
even if the poles are shifted into the region, there is no guarantee that the resulting
system has the desired performance . In the frequency-domain method, because the
relationship among the phase margin, gain margin, and time response is not exact,
even if the design meets the requirement on the phase and gain margins, there is no
guarantee that the time response of the resulting system will be satisfactory . Fur-
thermore, if a plant has open right-half-plane poles, the frequency-domain method
is rarely used . The constraint on actuating signals is not considered in the root-locus
method, nor in the frequency-domain method .
In the inward approach, we first choose an overall transfer function ; it can be
chosen to minimize the quadratic or ITAE performance index or simply by engi-
neering judgment. The constraint on actuating signals can be included in the choice .
Once an overall transfer function is chosen, we may implement it in the unity-
feedback configuration . If it cannot be so implemented, we can definitely implement
it in the two-parameter or plant input/output feedback configuration . In the imple-
mentation, we may also choose canceled poles to improve disturbance rejection
property and to increase stability robustness property of the resulting system . Thus,
the inward approach appears to be more general and more versatile than the outward
approach . Therefore, the inward approach should be a viable alternative in the design
of control systems .
We give a brief history about various design methods to conclude this chapter .
The earliest systematic method to design feedback systems was developed by Bode
in 1945 [7] . It is carried out by using frequency plots, which can be obtained by
measurement. Thus, the method is very useful to systems whose mathematical equa-
tions are difficult to develop . The method, however, is difficult to employ if a system,
such as an aircraft, has unstable poles . In order to overcome the unstable poles of
aircrafts, Evans proposed the root-locus method in 1950 [27] . The method has since
been widely used in practice. The inward approach was first discussed by Truxal in
1955 [57] . He called the method synthesis through pole-zero configuration . The
conditions in Corollary 9 .1 were developed for the unity-feedback configuration . In
spite of its importance, the method was mentioned only in a small number of control
texts. ITAE optimal systems were developed by Graham and Lathrop [33] in 1953 .
Newton and colleagues [48] and Chang [10] were among the earliest to develop
quadratic optimal systems by using transfer functions .
The development of implementable transfer functions for any control configu-
ration was attempted in [12]. The conditions of proper compensators and no plant
leakage, which was coined by Horowitz [36], were employed. Total stability was
not considered. Although the condition of well-posedness was implicitly used, the
concept was not fully understood and the proof was incomplete . It was found in [ 14]
that, without imposing well-posedness, the plant input/output feedback configura-
tion can be used to implement G0(s) = 1 using exclusively proper compensators . It
was a clear violation of physical constraints . Thus the well-posedness condition was
explicitly used in [15, 16] to design control systems . By requiring proper compen-
sators, total stability, well-posedness and no plant leakage, the necessity of the im-
plementability conditions follows immediately for any control configuration . Al-
though these constraints were intuitively apparent, it took many years to be fully
understood and be stated without any ambiguity . A similar problem was studied by
Youla, Bongiorno, and Lu [68], where the conditions for G0(s) to be implementable
in the single-loop configuration in Figure P10 .3 by using stable proper compensators
are established . The conditions involve the interlacing of real poles of G0(s) and
G(s), and the problem is referred to as strong stabilization .
The computation of compensators by solving linear algebraic equations was first
suggested by Shipley [55] in 1963 . Conditions for the existence of compensators
were not discussed . The coprimeness condition was used in 1969 [11] to establish
the existence of compensators in the plant input/output feedback system and to
establish its relationship with state feedback [66] and state estimators [45] . The two-
parameter configuration seems to be first introduced by Astrom [2] in 1980 and was
employed in [3] . The design procedure discussed in this text follows [21].
Most results in Chapter 10 are available for multivariable systems, systems with
more than one input or more than one output. The results were developed from the
polynomial fractional approach, which was developed mainly by Rosenbrock [53],
Wolovich [65], Kucera [41], and Callier and Desoer [9] . The polynomial fractional
method was translated into linear algebraic equations in [14, 15] . Using elementary
results in linear algebra, it was possible to develop most results in the polynomial
fractional approach in [15] . More importantly, minimum-degree compensators can
now be easily obtained by solving linear algebraic equations . This method of com-
puting compensators is therefore called the linear algebraic method .
PROBLEMS
10.1 . Consider a plant with transfer function 1/s(s + 3) . Implement the overall
transfer function G0(s) = 4/(s 2 + 4s + 4) in the unity-feedback configu-
ration in Figure 10.1 . Does the implementation involve pole-zero cancella-
tions? Do you have the freedom of choosing canceled poles?
10.2. Consider a plant with transfer function 2/s(s - 3). Can you implement the
overall transfer function G0(s) = 4/(s 2 + 4s + 4) in the unity-feedback
configuration in Figure 10.1? Will the resulting system be well posed? totally
stable?
10.3. Given G(s) = 1/(s - 1). Can G0(s) = 1/(s + 1) be implemented in the
unity-feedback configuration in Figure 10.1? Can G0(s) be implemented in
the single-loop feedback system shown in Figure P10 .3 with C,(s) = 1?
u Y
G (s)
C 2 (s)
Figure P10 .3
10.4. Consider a plant with transfer function 1 /s(s + 2) . Find a proper compensator
of degree 1 in the unity-feedback configuration such that the overall system
has poles at -1 + j, -1 - j, and - 3 . What is the compensator? Use the
root-locus method to give an explanation of the result.
10.5. Consider a plant with transfer function 1/s(s - 2) . Find a proper compensator
of degree 1 in the unity-feedback configuration such that the overall system
has poles at - 1 + j, -1 - j, and - 3. Will the resulting system track
asymptotically every step-reference input? Is this tracking robust?
10.6. a. Consider a plant with transfer function 1/s(s + 2) . Design a compensator
in the unity-feedback configuration such that the overall system has the
dominant poles - 1 .4 j1 .43 and a pole a with a = -4. What is u(0 + )?
b. Repeat (a) for a = -5 . Can you conclude that, for the same dominant
poles, the farther away the pole a, the larger the actuating signal?
10.7. a. Consider a plant with transfer function G(s) = 1/s(s + 2) and its quadratic
optimal system G0(s) = 3/(s 2 + 3 .2s + 3). Can G0(s) be implemented
in Figure P10 .7 by adjusting k1 and k2?
b. Consider a plant with transfer function G(s) _ (s - 1)/(s 2 - 4) and its
quadratic optimal system G0(s) = - 1.8(s - 1)/(s 2 + 5.2s + 5). Can
G0(s) be implemented in Figure P10 .7 by adjusting k1 and k2? Will this
optimal system track asymptotically step-reference inputs? Note that if

PROBLEMS 429
V
G (s)
HH- Figure P10 .7
G(s) is of type 0 as in this example, then the quadratic performance index

equals infinity.
10.8. Consider the single-loop configuration with C, (s) = 1 shown in Figure P10 .3 .
Is it possible to achieve arbitrary pole placement by choosing a suitable proper
compensator C(s)? If yes, develop a design procedure . Do you have control
over the zeros of the resulting overall system?
10.9. a. Consider a plant with transfer function G(s) = (s - 1)/(s 2 - 4) . Find a
compensator C(s) and a precompensator k so that the unity-feedback sys-
tem in Figure 10 .3 has all poles at - 3 and tracks asymptotically step-
reference inputs.
b . If the plant transfer function in (a) becomes G(s) = (s - 0 .9)/(s 2 - 4)
after implementation, is the system still stable? Will the system still track
asymptotically every step-reference input? Is this design robust in track-
ing?
10.10. In Problem 10 .9, is it possible to design a system which is robust in tracking
by increasing the degree of the compensator?
10.11 . Implement G0(s) in Problem 10 .7(a) in the unity-feedback configuration . Also
implement it in the two-parameter configuration . Do you have control over
the canceled pole in both cases?
10.12. Can you implement G0(s) in Problem 10 .7(b) in the unity-feedback configu-
ration? Can you implement it in the two-parameter configuration?
10.13. Consider G(s) = 2/s(s 2 + 0.25s + 6.25). Implement
20
G,(s)
s3 + 3.957s 2 + 14 .048s + 20
in the unity-feedback and two-parameter configurations . Which implemen-
tation is more desirable? Give your reasons .
10.14. a . Consider G(s) = (s + 3)/s(s - 1) and its overall system G0(s) _
784/(s 2 + 50 .4s + 784) . Implement the system in the two-parameter
configuration by choosing Dr(s) = 1 . Will the plant output of the system
reject completely, as t any step disturbance function entering at the
plant input terminal?
b. Repeat the design in (a) by choosing Dp(s) = s + /3 with /3 = 3 . Imple-

ment G0 (s) and at the same time achieve disturbance rejection .
c. Repeat (b) for a = 30 and /3 = 300 . If computer software is available,
simulate and compare the results in (a), (b), and (c).
10.15 . Consider the system in Problem 10.14 . Suppose, after implementation, the
plant transfer function changes to G(s) = (s + 3 + e2 )/s(s - 1 + E l ). Find
the ranges of El and E2 in which the system remains stable for the three cases .
10.16. Implement the systems in Problems 10 .7 and 10.13 in the plant input/output
feedback configuration .
10.17. Implement the system in Problem 10.14(a) in the plant input/output feedback
configuration . Are the compensators proper? Can we apply the procedure in
Section 10 .6 to design this problem?
10.18. a. Considej the unity-feedback system shown in Figure 10 .1 . Show that if
the plant transfer function is of type 1 and if the system is stable, then the
plant output will track asymptotically any step-reference input. The track-
ing is robust with plant and compensator variations as long as the system
remains stable.
b . Consider the two-parameter system shown in Figure 10 .6. Show that if the
plant transfer function is of type 1 and if the overall system is stable, then
the plant output will track asymptotically any step-reference input if and
only if M(0) = L(0) . The tracking is robust with plant and compensator
variations as long as the system remains stable and the dc gains of M(s)
and L(s) remain equal .
c. Consider the plant input/output feedback system shown in Figure
10.15(b) . Show that if the plant transfer function is of type 1 and if the
overall system is stable, then the plant output will track asymptotically any
step-reference input if and only if A(0) = M(0). The tracking is robust
with plant and compensator variations as long as the system remains stable
and the dc gains of A(s) and M(s) remain equal.
10.19 . a. It is known that two polynomials D(s)_and N(s) are coprime if and only if
there exist two polynomials A(s) and B(s) such that
A(s)D(s) + B(s)N(s) = 1
Show that if D(s) = s2 - 1 and N(s) = s - 2, then A(s) = 1/3 and

B(s) = - (s + 2)/3 meet the condition .
b. For any two polynomials D(s) and N(s), there exist two polynomials A(s)
and h(s) such that
A(s)D(s) + B(s)N(s) = 0
Find two such polynomials for D(s) and N(s) in (a).

PROBLEMS 431
c. Consider the Diophantine equation

A(s)D(s) + B(s)N(s) = D 0(s)
where D(s) and N(s) are coprime . Show that, for any polynomial Q(s),
A(s) = A(s)D0 (s) + Q(s)A(s) B(s) = B(s)D0 (s) + Q(s)A(s)
is a general solution of the Diophantine equation . For D(s) and N(s) in (a)
and D0 (s) = s3 + 7s 2 + 17s + 15, show that the following set of two
polynomials
A(s) = (s3 + 7s 2 + 17s + 15) + Q(s)(-s + 2)

3
B(s) _ - (s + 2)(s3 + 7s2 + 17s + 15) + Q(s)(s 2 - 1)
3
is a general solution .
d. Show that if Q(s) = (s 2 + 9s + 32)/3, then the degrees of A(s) and B(s)
are smallest possible . Compare the result with the one in Example 10 .3.2.
Which procedure is simpler? (It is true that many engineering problems
can be solved by applying existing mathematical results . However, em-
phasis in mathematics is often different from that in engineering . Mathe-
maticians are interested in existence conditions of solutions and general
forms of solutions . Engineers are interested in particular solutions and in
efficient methods of solving them . This problem illustrates well their dif-
ferences in emphases .)
State Space Design
11 .1 INTRODUCTION
In this chapter we discuss the design of control systems using state-variable equa-
tions. We use simple networks to introduce the concepts of controllability and ob-
servability, and we develop their conditions intuitively . We then discuss equivalence
transformations . Using an equivalence transformation, we show how to design pole
placement using state feedback under the assumption of controllability. Because the
concept and condition of controllability are dual to those of observability, design of
full-dimensional state estimators can then be established . We also discuss the design
of reduced-dimensional state estimators by solving Lyapunov equations . The con-
nection of state feedback to the output of state estimators is justified by establishing
the separation property . This design is also compared with the transfer function
approach discussed in Section 10 .6. Finally, we introduce the Lyapunov stability
theorem and then apply it to establish the Routh stability test .
The state in state-variable equations forms a linear space, called state space ;
therefore, the design using state-variable equations is also called state space design .
11 .2 CONTROLLABILITY AND OBSERVABILITY

x = Ax + bu (11 .1a)
y=ex+du (11 .1b)
432

11 .2 CONTROLLABILITY AND OBSERVABILITY 433
where A, b, and c are respectively n X n, n X 1, and 1 X n matrices . The constant

d is scalar . Equation (11 .1) is said to be controllable if we can transfer any state to
any other state in a finite time by applying an input . The equation is said to be
observable if we can determine the initial state from the knowledge of the input and
output over a finite time interval . We first use examples to illustrate these concepts .
Example 11 .2.1
Consider the network shown in Figure 11 .1(a). The input is a current source, and
the output is the voltage across the 2-fl resistor shown. The voltage x across the
capacitor is the only state variable of the network . If x(O) = 0, no matter what input
is applied, because of the symmetry of the four resistors, the voltage across the
capacitor is always zero . Therefore, it is not possible to transfer x(0) = 0 to any
nonzero x . Thus, the state-variable equation describing the network is not control-
lable . If x(O) is different from zero, then it will generate a response across the output
y . Thus, it is possible that the equation is observable, as will be established later .
(a) (b)
Figure 11 .1 (a) An uncontrollable network . (b) An unobservable network .
Example 11 .2.2
Consider the network shown in Figure 11 .1(b). The input is a current source and the
output y l is the voltage across the resistor . The network has two state variables : the
voltage x1 across the capacitor, and the current x2 through the inductor . Nonzero
x 1 (0) and/or x 2 (0) will excite a response inside the LC loop. However, the current
i(t) always equals u(t) and the output always equals 2u(t) no matter what x1 (0) and
x2(0) are . Therefore, there is no way to determine the initial state from u(t) and y1 (t).
Thus, the state-variable equation describing the network will not be observable .
Because the LC loop is connected directly to the input, it is possible that the
equation is controllable, as will be shown later . We mention that controllability and
observability depend on what are considered as input and output . If Y2 in Figure
11 .1(b) is considered as the output, then the equation will be observable .

434 CHAPTER 11 STATE SPACE DESIGN
We develop in the following a condition for (11 .1) to be controllable . The

purpose is to show how the condition arises ; therefore, technical details are skipped .
The solution of (11 .1 a) due to any x(O) and u(t) is, as derived in (2.72),
x(t) = e A`x(0) + fo eA (,- T) bu(7)d7
which can be written as, using (2.67),

t _ 2
x(t) - e At x(0) = fo [I + A(t - -r) + A2(t 2 T) + . . .] bu(7)dT
b fo u(T)dT + Ab fo (t - T)u(T)d7
+ A2 b fo 0 .5(t - T) 2 u(T)dT +
u(7)d7
f0t
fo (t - r)u(,r)dT
= [b Ab A2 b . . .]
0 .5(t - r) 2u(T)dT
f0t
Using Theorem B .1, we conclude that for any x(O) and x(t), a solution u(t) exists if
and only if the matrix
[b Ab A2 b . . . An - 'b . . .] (11 .2)
has rank n. The matrix has n rows but infinitely many columns. Fortunately, using
the Cayley-Hamilton theorem (see Problem 2 .37), we can show that (11 .2) has rank
n if and only if the n X n matrix
U = [b Ab A2 b . . . An - 'b] (11 .3)
has rank n . Thus we conclude that (11 .1) is controllable if and only if the matrix in
(11 .3) has rank n or, equivalently, its determinant is different from zero . The matrix
in (11 .3) is called the controllability matrix . The first column is b, the second column
is Ab, and the last column is An - 'b . Similarly, we define the n X n matrix
c
cA
(11 .4)
An-
Its first row is c, second row is cA, and the last row is cAn - ' . Then the n-dimensional
equation in (11 .1) is observable if and only if the matrix V has rank n or, equivalently,

its determinant is nonzero . We call V the observability matrix. Controllability is a

property of (A, b) ; observability is a property of (A, c) . They are independent of the
direct transmission part d . We use examples to illustrate how to use these conditions .
Example 11 .2.3
Consider
N(s) 2s - 1 2s - 1
G(s) = (11 .5)
D(s) s2 - 1 .5s - 1 (s - 2)(s + 0.5)
Its controllable-form realization, as developed in (5.17), is

[
x = 0]x + O ]u (11 .6a)
[15
y = [2 - 1]x (11 .6b)
Because the dimension of (11 .6) equals the number of poles of G(s) in (11 .5), (11 .6)
is a minimal state-variable equation (see Section 2.8) . Every minimal equation is
controllable and observable . We demonstrate this fact for the equation in (11 .6).
To check controllability, we first compute
5 ][ 1]
Ab = = (11 .7)
C1 0 0 [ 1 .5 ]
Thus the controllability matrix of (11 .6) is
1 1.5
U = [b Ab] = (11 .8)
0 1
Its determinant is 1 (see (B .2)). Thus (11.6) is controllable .

To check observability, we first compute
cA = [2 -1] C 1 .5 1 ] = [2 2] (11 .9)
Thus the observability matrix of (11 .6) is

= 2 -1
(11 .10)
V = [cA] [2 2]
Its determinant is 4 + 2 = 6 . Thus the equation is observable .

Example 11 .2.4
2s + 1 2s + 1
G(s) =
s2 - 1 .5s - 1 s - 2)(s + 0 .5)
Its controllable-form realization is
.5 [11 u
x= 1]x+ (11 .12a)
I1
y = [2 1]x (11 .12b)
This equation differs from (11 .6) only in the output equation ; therefore, its control-
lability matrix is the same as (11 .8), and (11 .12) is controllable. To check observa-
bility, we compute
cA = [2 1111 .5 = [4 2]
10]
Thus the observability matrix of (11 .12) is

c 2 1
(11 .13)
V = cA 4 2
Its determinant is 4 - 4 = 0 . Thus, (11 .12) is not observable . Note that the number
of poles of G(s) in (11 .11) is 1 (why?), whereas the dimension of (11 .12) is 2 . Thus
(11 .12) is not a minimal equation . The equation is controllable but not observable.
The observable-form realization of G(s) in (11 .11) is
X 1 .5 [2] u
1]x+ (11 .14a)
y = [1 0]x (11 .14b)-
Its controllability and observability matrices are
U=C1 2] V= C1 .5 1
Because det U = 0 and det V = 1, where det stands for the determinant, the equation
is observable but not controllable .
The preceding discussion can be extended to the general case . Consider

N(s) _ b1 s 3 + b2S 2 + ba s + b4
G( (11 .15)
s) = D(s) s 4 + a1 s3 + a2S2 + a3 s + a4
Note that the assignment of coefficients in (11 .15) is different from the one in Section

11 .2 CONTROLLABILITY AND OBSERVABILI1Y 437
5.5.1 . This is done for convenience in developing the design procedure in Section
11.4. The controllable-form realization of (11 .15) is, as shown in (5.17),
-a1 -a2 -a3 -a4 1
1 0 0 0 0
x + U (11 .16a)
0 1 0 0 0
0 0 1 0 0
y = [b 1 b2 b3 b4 ]x (11 .16b)
Its controllability matrix can be computed as

1 a1 e2 e3
0 1 -a, e2
U =
0 0 1 -a1
0 0 0 1
with e2 = -a2 + ai and e3 = -a3 + 2a1a2 - ai . It is a triangular matrix ; its

determinant is always 1 no matter what the a i are (see (B .2)). Therefore, it is always
controllable, and this is the reason for calling it controllable form . To check whether
(11 .16) is observable, we must first compute the observability matrix V in (11 .4)
and then check its rank. An alternative method is to check whether N(s) and D(s)
have common factors or not . If they have common factors, then the equation cannot
be observable . This is the case in Example 11 .2.4 . If N(s) and D(s) have no common
factors, then (11 .16) will be observable as well . Similarly, the observable-form
realization of (11 .15) is, as discussed in (5 .18),
-a1 1 0 0 bl
-a2 0 1 0 b2
X = x+ u (11 .17a)
-a33 0 0 1 b3
- a4 0 0 0 b4
y = [1 0 0 0]x (11 .17b)
This equation is always observable . It is controllable if N(s) and D(s) have no com-
mon factors ; otherwise, it is not controllable .
Exercise 11 .2.1
Show that the network in Figure 11 .1(a) can be described by the following state-
variable equation
z= -0.75x+0 u y=0 .Sx+u
Is the equation controllable? observable?
[Answers : No, yes.]
Exercise 11 .2.2
Show that the network in Figure 11 .1(b) with y1 as the output can be described by
01
x
=C -0] x + [0] u
y1 = [0 0]x + 2u
[Answers : Yes, no.]
Exercise 11 .2.3
Show that the network in Figure 11 .1(b) with y2 as the output can be described by
0 -
x = ~ x + u
OJ [0]
Y2 = [1 0]x
[Answers : Yes, yes .]
In addition to (11 .3) and (11 .4), there are many other controllability and ob-
servability conditions . Although (11 .3) and (11 .4) are most often cited conditions in
control texts, they are very sensitive to parameter variations, and consequently not
suitable for computer computation . For a discussion of this problem, see Reference
[15, p . 217] . If state-variable equations are in diagonal or, more generally, Jordan
form, then controllability can be determined from b alone and observability, from c
alone . See Problems 11 .2, 11.3, and Reference [15, p . 209] .
11 .2.1 Pole-Zero Cancellations

Controllability and observability are closely related to pole-zero cancellations in
block diagrams . This is illustrated by an example .
Example 11 .2.5
Consider the block diagram in Figure 11 .2(a) with input u and output y . It consists
of two blocks with transfer functions 1/(s - 1) and (s - 1)/(s + 1) . If we assign
the output of the first block as x 1, then we have

u I s-1 y
s-1 s+1
(a)
u s-1 w 1 x 1 y
s+l s-1
(b)
Figure 11 .2 Pole-zero cancellations .
See Section 5 .5.2. The second block has input x, and output y, thus we have
Y(s) = s-1 -2
.=1+
XI (s) s + 1 s + 1
Therefore, it can be realized as
z2 - -x2 -2x, y=x2 +x 1
Thus, the tandem connection of the two blocks can be described by
U
[x2 ] -
[-2 -1][x2] + [0' ]
Y2 = [1 1]x
Because its controllability matrix
U = [b Ab] _
is nonsingular, the equation is controllable . The observability matrix is

- 1]
V =
[cA] - 1 1 1-1
which is singular. Thus, the equation is not observable . The equation has two state
variables, but its overall transfer function
1 s - 1 1
G(s) _ s- 1 s+ 1 s+ 1
has only one pole . Therefore, the equation is not minimal and cannot be both con-
trollable and observable . Similarly, we can show that the equation describing Figure
11 .2(b) is observable but not controllable . In general, if there are pole-zero cancel-
lations in tandem connection, the state-variable equation describing the connection
cannot be both controllable and observable. If poles of the input block are canceled
by zeros of the output block, the equation cannot be observable ; if poles of the output
block are canceled by zeros of the input block, the equation cannot be controllable .

11 .3 EQUIVALENT STATE-VARIABLE EQUATIONS
We first use an example to illustrate the concept of equivalent state-variable equa-

tions. Consider the network shown in Figure 11 .3. First we choose the inductor
current x 1 and the capacitor voltage x 2 as state variables, as shown in Figure 11 .3(a).
Then the voltage across the inductor is 2z1 and the current through the capacitor is
.x2 . Because the voltage across the resistor is x2, the current through it is x2/2 =
0.5x2. From the outer loop, we have
2z1 + x2 - u = 0 or X l = -0.5x2 +,
From the node denoted by A, we have
= x 1 - 0.5x2
x2
Thus, the network can be described by the following state-variable equation
-0
I I_ I .51 x+ [0 (11 .18a)
r2 1 -0.5 0 .5] u
y = [0 l]x (11 .18b)
For the same network, we now choose the two loop currents x 1 and x2 shown
in Figure 11 .3(b) as state variables . Then the voltages across the inductor, resistor,
and capacitor are respectively 2x 1 , 2(x 1 - x2) and fo x2 (T)dT. From the left-hand
side loop, we have
u = 2x 1 + 2(x1 - x2)
which implies
x,= -xl +x2 +0.5u
From the right-hand side loop, we have
fo x2(T)dT = 2(x1 x2) = y
which implies
X2 = 2(x 1 - x2)
~--ZX~--~
(a) (b)
Figure 11 .3 Network with different choices of state variables .

11 .3 EQUIVALENT STATE-VARIABLE EQUATIONS 441
This becomes, after the substitution for Xl,

x2 = - 0.5x2 +(-x 1 +xx2 +0.5u)= -x1 +0.5x2 +0.5u
Thus, the network in Figure 11 .3 can also be described by the following state-
variable equation
X, 1 1 [ 0.5
x + u (11 .19a)
x[ -1
[2] _1 0.5] 0.5]
y = [2 -2]x (11 .19b)
Both (11 .18) and (11 .19) describe the same network ; therefore, they must be related
in some way . Indeed, from the currents through the inductor and resistor in Figure
11 .3(a) and (b), we have x1 = xl and x2/2 = x l - x2. Thus we have
x := x (11 .20)
X21 1
1X = C2 - 0 ]Cx2] C2 -2]
The square matrix in (11 .20) is nonsingular because its determinant is -2 . Two
state-variable equations are said to be equivalent if their states can be related by a
nonsingular matrix such as in (11 .20) . Thus (11 .18) and (11 .19) are equivalent.
i = Ax + bu (11 .21 a)
y = cx + du (11 .21b)
Let P be an arbitrary nonsingular matrix . Define x = Px. The substitution of x =
P - ' x and i = P - 'x into (11 .21) yields
P - ' x = AP - 'x + bu
y=cP - 'x+du
which become
x = Ax + bu (11 .22a)
y = cx + du (11 .22b)
with
A : = PAP - ' b : = Pb c : = cP -1 d := d (11 .22c)
Equations (11 .21) and (11 .22) are said to be equivalent, and P is called an equiva-
lence transformation . For (11 .18) and (11 .19), we have, from (11 .20),
P-1
= and
12 -02] P = [2 -20] 1 - [1 -0.5]

and
- 0] - -1 1
PAP -1 = ]=A
[1 -0.5][0 -0 .5][2 2 - 1 0 .5
Pb = = b
[1 -0.5] [0 5] - [0.5]
cP - ' = [0 1][ 1 - 0] = [2 -21 = c

Thus (11 .18) and (11 .19) are indeed related by (11 .22c) .
The transformation in A = PAP -1 is called a similarity transformation. Any
similarity transformation will not change the eigenvalues of a matrix . Indeed, using
(B.3), we have
det P det P -1 = det (PP- ') = det I = 1
and
det (sI - A) = det (sPP - ' - PAP- ) = det [P(sI - A)P- ']
(11 .23)
= det P det (sI - A) det P -1 = det (sI - A)
Thus A and A have the same characteristic polynomial and, consequently, the same
eigenvalues . Equivalent state-variable equations also have the same transfer function .
Indeed, the transfer function of (11 .22) is
G(s) = c(sI - A) -'b = cP- '[P(sI - A)P - '] - 'Pb
= cP- 'P(sI - A) - 'P - 'Pb = c(sI - A) - 'b = G(s)
Thus any equivalence transformation will not change the transfer function . The prop-
erties of controllability and observability are also preserved . Indeed, using
A2 = AA = PAP- 'PAP -1 = PA2P - '
and, in general, An = PA'P -1 , we have
U :_ [b Ab A2 b A" - 'b]
_ [Pb PAP - 'Pb PA2P- 'Pb . . . PA"-'P-'Pb] (11 .24)
= P[A Ab A2b . . . A" - 'b] = PU
Because P is nonsingular, the rank of U equals the rank of U . Thus, (11 .22) is
controllable if and only if (11 .21) is controllable . This shows that the property of
controllability is invariant under any equivalence transformation . The observability
part can be similarly established .
11 .4 POLE PLACEMENT
In this section, we discuss the design of pole placement by using state feedback . We
first use an example to illustrate the basic idea .

11 .4 POLE PLACEMENT 443
Example 11 .4.1
10
G(s) _
s(s + 1)
The plant could be an armature-controlled dc motor driving an antenna and can be
represented by the block diagram enclosed by the dotted line in Figure 11 .4(a). If
we assign the output of each block as a state variable as shown, then we can readily
obtain
and
z2 = -x2 + 10u
(see Section 5.5.2). Thus, the plant can be described by the following state-variable
equation
x+ (11 .25a)
[0 -1] [10] u
y = [1 0]x (11 .25b)
Now we introduce feedback from x 1 and x2 as shown in Figure 11 .4(a) with real
constant gains k1 and k2. This is called state feedback. With the feedback, the transfer
function from r to y becomes, using Mason's formula,
10
s(s + 1) 10
G(s) _ 1 +
10k2 + IOk1 s(s + 1) + lOk2s + 10k 1
s + 1 s(s + 1) (11 .26)
_ 10
S 2 + ( 1 + 10k2)s + 10k 1
We see that by choosing k1 and k2, the poles of G0(s) can be arbitrarily assigned
1
(a) (b)
Figure 11 .4 Block diagrams .

provided complex-conjugate poles are assigned in pairs . For example, if we assign

the poles of G0(s) as -2 j2, then k, and k2 can be determined from
(s+2+j2)(s+2-j2)=s 2 +4s+8
(11 .27)
= s2 + ( 1 + 10k2)s + 10k,
or
1 + IOk2 = 4 10k, = 8
ask, = 0.8 and k2 = 0 .3. This shows that by state feedback, the poles of the resulting
system can be arbitrarily assigned .
Because the feedback paths consist of only gains (transfer functions with degree
0), the number of the poles of the resulting system remains the same as the original
plant . Thus, the ,introduction of constant-gain state feedback does not increase the
number of poles of the system, it merely shifts the poles of the plant to new positions .
Note that the numerator of G0(s) equals that of G(s) . Thus, the state feedback does
not affect the numerator or the zeros of G(s). This is a general property and will be
established later.
Exercise 11 .4.1
Consider a plant with transfer function G(s) = 10/(s + 1)(s - 2) as shown in

Figure 11 .4(b). Find two feedback gains k, and k 2 such that the poles of the resulting
system are located at -2 j2 .
[Answer : k, = 2, k2 = 0.5.]
Although Example 11 .4.1 illustrates the basic idea of pole placement by state
feedback, the procedure cannot easily be extended to general G(s) = N(s)/D(s) . In
the following, we use state-variable equations to discuss the design . Consider the
n-dimensional state-variable equation
x = Ax + bu (11 .28a)
y = cx (11 .28b)
G(s) = c(sI - A) - 'b (11 .28c)
To simplify discussion, we have assumed the direct transmission part to be zero .
The equation is plotted in Figure 11 .5, where single lines denote single signals and
double lines denote two or more signals . Let
u(t) = r(t) - kx(t) (11 .29)

11 .4 POLE PLACEMENT 445
Figure 11.5 State feedback.
where k = [k, k2 . . . k] is a 1 X n real constant vector and r(t) is the scalar

reference signal . Equation (11 .29) is called constant-gain state feedback, or simply
state feedback, and is plotted in Figure 11 .5. The substitution of (11 .29) into (11 .28)
yields
x = Ax - bkx + br = (A - bk)x + br (11 .30a)
y = cx (11 .30b)

G0(s) = c(sI - A + bk) - 'b (11 .30c)
We show that if (A, b) is controllable, then the eigenvalues of (A - bk) can be

arbitrarily assigned by choosing a real constant feedback gain k . Because A and b
are implicitly assumed to be real matrices and k is required to be real, if a complex
number is assigned as an eigenvalue of (A - bk), its complex conjugate must also
be assigned . To simplify discussion, we assume that (11 .28) is of dimension 4 and
its characteristic polynomial and transfer function from u to y are
0(S) = s 4 + a,s 3 + a2S2 + a3s + a4 (11 .31)
and
_ N(s) b,s 3 + b2s2 + bas + b4
G (s) (11 .32)
D(s) s4 + a,s 3 + a2 S2 + a3s + a 4
We show in the following that if (11 .28) is controllable, then it can be transformed
by an equivalence transformation into
-a, -a 2 -a3 -a4 1

1 0 0 0 0
x=Ax+bu= x+ 0 u (11 .33a)
0 1 0 0
0 0 1 0 -0-
y = cx = [b, b2 b3 b4 ]x (11 .33b)

The controllability matrix of (11 .33) can be computed as

1 -a, e2 e3
0 1 -a, e 2
U = (11 .34a)
0 0 1 -a,
0 0 0 1
with e2 = -a2 + a, and e 3 -a3 + 2a1a2 a,. It is triangular and its inverse
is also triangular and equals
U -1 = (11 .34b)
The easiest way to verify this is to show U - 1 U = I. Now, as developed in (11 .24),
the controllability matrices of (11 .28) and (11 .33) are related by
U = PU (11 .35)
which implies, for n = 4,
S := P-1 = UU -1 = [b Ab A2b A3 b] (11 .36)
Now if (11 .28) is controllable, then U is nonsingular and, consequently, S = P-1

is nonsingular . Thus, by the equivalence transformation x = Px or x = P-1 x =
Si, (11 .28) can be transformed into (11 .33).
Consider the controllable-form equation in (11 .33). If we introduce the state
feedback
u=r-kx=r-[k1 k2 k3 k4]x (11 .37)
then the equation becomes

z=(A-bk)+br
-a, - k1 - a2 - k2 - a3 - k3 - a4 - k4- 1
1 0 0 0 0
x + r (11 .38a)
0 1 0 0 0
0 0 1 0 0
Y = [b 1 b2 b3 b 4 ]x (11 .38b)
We see that (11 .38) is still of the controllable form, so its transfer function from r
to y is

11 . 4 POLE PLACEMENT 447
b 1s3 + b2S2 + bas + b4

GG(s) = (11 .39)
S 4 + (a, + k,)s 3 + (a2 + k2)s2 + (a3 + k3)s + (a4 + k4)
Now by choosing k the denominator of G0(s), and consequently the poles of G 0(s),
can be arbitrarily assigned . We also see that the numerator of G0(s) equals that of
G(s) . Thus the_state feedback does not affect the zeros of G(s).
To relate k in (11 .37) with k in (11 .29), we substitute X = Px into (11 .37) to
yield
u=r-kx=r-kPx
Thus we have k = kP. The preceding procedure is summarized in the following .
Procedure for Designing Pole Placement

Given a 4-dimensional controllable (A, b) and a set of eigenvalues A1, i = 1, 2, 3,
4. Find a real 1 X 4 vector k so that the matrix (A - bk) has the set as its
eigenvalues .
1. Compute the characteristic polynomial of A : 0(s) = det (sI - A) = s4 + a,s 3
+ a2s2 + a3s + a4-
2. Compute the desired characteristic polynomial
L(s) = (s - A,)(s - A2)(s - A3)(s - A4)
= s4 + als 3 + a2sZ + a3s + a4
3 . Compute the feedback gain for the equivalent controllable-form equation
k=[a,-a, a2 -a 2 a3 -a3 a4 -a4]
4. Compute the equivalence transformation
1 a, a2 3
0 1 a, a2
S : = P - ' = [b Ab A2b A3b] (11 .40)
0 0 1 a,
0 0 0 1
5. Compute the feedback gain k = kP = KS -1 .
It is clear that the preceding procedure can be easily extended to the general
case. There are other design procedures . For example, the following formula
k = [0 0 0 1][b Ab A2 b A3 b] -'0(A)
= [0 0 0 1][b Ab A2b A3b] - '[A4 + a1 A3 + a2A2 + a3A + a4I]
called the Ackermann formula, is widely quoted in control texts . Note that 0(s)
is not the characteristic polynomial of A (it is the characteristic polynomial of
(A - bk)), thus 0(A) = 0 . The derivation of the Ackermann formula is more
complex ; its computation is no simpler. This is why we introduced the preceding
procedure . The procedure is probably the easiest to introduce but is not necessarily
the best for computer computation . See Reference [15] .

Example 11 .4 .2
Consider the equation in (11 .25) or

1]
X + u (11 .41 a)
x = [x2] = [00 1
y = [1 0]x (11 .41 b)

10
G(s) (11 .41c)
_ s+s
2
Find the feedback gain k in u = r - kx such that the resulting equation has
-2 j2 as its eigenvalues .
We compute the characteristic polynomial of A
det(sI-A)=detEs -= s(s+ 1)=s 2 +s+0

0 s + i1 ]
and the desired characteristic polynomial
(s+2+j2)(s+2-j2)=s 2 +4s+8
Thus, the feedback gain for the equivalent controllable-form equation is
k= [a 1 -a1 a2 -a2]= [4- 1 8-0]=[3 8]
Next we compute the equivalence transformation
S = P - ' = [b Ab]
10 11 1 [10 -10][0 1] = [10 1 01
Thus we have
k = kP = [3 81 - [0 .8 0.3]
1 10 10] 1 = [3 8][00 .1 0-1] 0
As we expected, this result is the same as the one in Example 11 .4.1. To verify the
result, we compute
s 0 0 -1] 0] s -1
sI-A+bk= 0 s .8 0
0 -1] [ 10 [0 .3]= 8 s+4
Thus, the transfer function of the resulting system is
_ -1
GG(s) _ [1
0] [g s + 14 10
1 s+4 1][0] _ 10
_ [1 0]
5 2 +4s+8 -8 s 10 s2 +4s+8

11 .5 QUADRATIC OPTIMAL REGULATOR 449
as expected. We see that the state feedback shifts the poles of G(s) from 0 and -1
to - 2 j2 . However, it has no effect on the numerator of G (s) .
If (A, b) is not controllable, then the matrix in (11 .36) is not nonsingular and
(11 .28) cannot be transformed into the controllable-form equation in (11 .33). In this
case, it is not possible to assign all eigenvalues of (A - bk); however, it is possible
to assign some of them. See Problem 11 .8 and Reference [15] .
To conclude this section, we mention that state feedback gain can be obtained
by using MATLAB . For the problem in Example 11 .4.2, we type
a=[0 1 ;0 -1] ;b=[0 ;10] ;
i=sgrt(-1) ;
p = [ - 2 + 2*i ; - 2 - 2*i] ;
k=place(a,b,p)
Then k = [0 .8 0 .3] will appear on the screen . The command acker(a,b,p), which
uses the Ackermann formula, will also yield the same result . But the MATLAB
manual states that acker(a,b,p) is not numerically reliable and starts to break down
rapidly for equations with dimension larger than 10 .
11 .5 QUADRATIC OPTIMAL REGULATOR
Consider
x = Ax + bu (11 .42a)
y = cx (11 .42b)
In the preceding section, the input u was expressed as r - kx, where r is the
reference input and k is the feedback gain, also called the control law . Now we
assume that the reference input r is zero and that the response of the system is excited
by nonzero initial state x(0), which in turn may be excited by external disturbances .
The problem is then to find a feedback gain to force the response to zero as quickly
as possible . This is called the regulator problem. If r = 0, then (11 .29) reduces to
u = -kx, and (11 .42a) becomes
x = (A - bk)x (11 .43)
Clearly, the response of (11 .43) depends on the eigenvalues of (A - bk). If all
eigenvalues have negative real parts or lie inside the open left half s-plane, then for
any nonzero x(0), the response of (11 .43) approaches zero as t -* o. The larger the
negative real parts, the faster the response approaches zero . Now if (A, b) is con-
trollable, we can find a k to place all eigenvalues of (11 .43) in any desired positions .
This is one way of designing the regulator problem .
How to choose the eigenvalues of (11 .43) is not simple, however . One possi-
bility is to use the pole patterns in Figure 9 .15 . The radius of the semicircles in
Figure 9 .15 is to be determined from the constraint on the magnitude of u(t) . Gen-

erally, the larger the radius, the larger the feedback gain k and the larger the ampli-
tude of u(t) . Therefore, the constraint on u(t) will limit the rate for the response to
approach zero . The most systematic and popular method is to find k to minimize
the quadratic performance index
J = [x'(t)Qx(t) + Ru2(t)]dt (11 .44)

Jo
where the prime denotes the transpose, Q is a symmetric positive semidefinite matrix,
and R is a positive constant . Before proceeding, we digress to discuss positive definite
and semidefinite matrices .
A matrix Q is symmetric if its transpose equals itself, that is, Q' = Q . All
eigenvalues of symmetric matrices are real, that is, no symmetric matrices can have
complex eigenvalues . See Reference [15, p . 566 .] . A symmetric matrix is positive
definite if x'Qx > 0 for all nonzero x ; it is positive semidefinite if x'Qx ? 0 for all
x and the equality.holds for some nonzero x . Then we have the following theorem .
See Reference [15, p. 413 .].
THEOREM 11 .1
A symmetric matrix Q of order n is positive definite (positive semidefinite) if
and only if any one of the following conditions holds :
1. All n eigenvalues of Q are positive (zero or positive) .
2. It is possible to decompose Q as Q = N'N, where N is a nonsingular square
matrix (where N is an m X n matrix with 0 < m < n) .
3. All the leading principal minors of Q are positive (all the principal minors
of Q are zero or positive) .
If Q is symmetric and of order 3, or

q11 q21 q31
Q - q21 q22 q32
q31 q32 q33
then the leading principal minors are
q11 det q11 q21 det Q

q21 q22
that is, the determinants of the submatrices by deleting the last k rows and the last
k columns for k = 2, 1, 0. The principal minors of Q are
q11 q21
q11 q22 q33 det
q21 q22
q11 q31 q22 q32

det det det Q
q31 q33 q32 q33

11 .5 QUADRATIC OPTIMAL REGULATOR 451
that is, the determinants of all submatrices whose diagonal elements are also diagonal
elements of Q . Principal minors include all leading principal minors but not con-
versely . To check positive definiteness, we check only the leading principal minors .
To check positive semidefiniteness, however, it is not enough to check only the
leading principal minors . We must check all principal minors . For example, the
leading principal minors of
1 0 1
1
0 0 0
1 0 0
are 1, 0, and 0, which are zero or positive, but the matrix is not positive semidefinite
because one principal minor is - 1 (which one?) .
If Q is positive semidefinite and R is positive, then the two integrands in (11 .44)
will not cancel each other and J is a good performance criterion . The reasons for
choosing the quadratic index in (11 .44) are similar to those in (9.13). It yields a
simple analytical solution, and if Q and R are chosen properly, the solution is ac-
ceptable in practice .
If Q is chosen as c'c, then (11 .44) becomes
J = Jo [x'(t)c'cx(t) + Ru 2(t)]dt = Jo [y 2(t) + Ru2(t)]dt (11 .45)
This performance index is the same as (9 .13) with r(t) = 0 and R = 1 /q . Now, if
the state-variable equation in (11 .42) is controllable and observable, then the feed-
back gain that minimizes (4 .45) is given by
k = R - 'b'K (11 .46)
where K is the symmetric and positive definite matrix meeting
-KA-A'K+KbR - 'b'K-c'c=0 (11 .47)
This is called the algebraic Riccati equation . This equation may have one or more
solutions, but only one solution is symmetric and positive definite. The derivation
of (4.46) and (4 .47) is beyond the scope of this text and can be found in References
[1, 5] . We show in the following its application .
Example 11 .5.1
Consider the plant with transfer function 1/s(s + 2) studied in Example 9 .4.1 . Its
controllable form realization is
X = 1 + u (11 .48a)
0] x [0]
y = [0 l]x (11 .48b)
Find the feedback gain to minimize the performance index

= fo [Y2(t) + 9 U2(t) dt
j (11 .49)
1
The comparison of (11 .45) and (11 .49) yields Q = c'c and R = 1/9 . Let
kz1
k21 k22
It is a symmetric matrix . For this problem, (11 .47) becomes
[k u k2i ] [-2 0] [-2 1 ] 1k11l k21]
k2 l k22 1 0 0 0 k2
k21][1][1 0][k1,
+ 9 k11
k21 kzz 0 C21 -
k21 kzz] 1 0 1 [0
1] =
Equating the corresponding entries yields

4k11 - 2 k21 + 9k11 = 0 (11 .50a)
2k21 - k22 + 9k 1 1 k 21 = 0 (11 .50b)
and
9k21 - 1 = 0 (11 .50c)
From (11.50c), we have k21 = 1/3 . If k21 = -1/3, then the resulting K will not
be positive definite . Thus we choose k 21 = 1/3 . The substitution of k21 = 1/3 into
(11 .50a) yields
9k11 + 4k 11 - = 0
3
whose solutions are 0 .129 and -0.68. If k1 l = - 0.68, then the resulting K will
not be positive definite . Thus we choose k11 = 0.129 . From (11 .50b), we can solve
k22 as 1 .05 . Therefore, we have
K _ 0.129 0 .333
0 .333 1 .05
which can be easily verified as positive definite . Thus the feedback gain is given by
.129 0.333] = [1
k = R -1 b'K = 9[1 0] 0 .2 31 (11 .51)
0.333 1.05
and (11 .43) becomes
z = (A - bk)x - 1 ][1 .2 31~ x

_ ([ 1 0] [ (11 .52)
= X
1 0

11 .6 STATE ESTIMATORS 453
The characteristic polynomial of the matrix in (11 .52) is

s+3 .2 3I
det =s 2 +3.2s+3
C I ss
which equals the denominator of the optimal transfer function in (9.24). This is not
surprising, because the performance index in (11 .49) is essentially the same as (9 .21)
with zero reference input . Therefore the quadratic optimal regulator problem using
state-variable equations is closely related to the quadratic optimal transfer function
in Chapter 9 . In fact, it can be shown that D0(s) obtained by spectral factorization
in Chapter 9 equals the characteristic polynomial of (A - bk). See Reference [1] .
We also mention that the conditions of controllability and observability are essential
here . These conditions are equivalent to the requirement in Chapter 9 that N(s) and
D(s) in G(s) = N(s)/D(s) have no common factors .
The optimal gain in quadratic regulators, also called linear quadratic regulator
or lqr, can be obtained by using MATLAB . For the example, we type
a=[-2 0 ;1 0] ;b=[1 ;0] ;
q=[0 0 ;0 1] ;r=1/9 ;
k= Igr(a,b,q,r)
then k=[1 .1623 3.000] will appear on the screen . Thus the use of MATLAB is
very simple .
11 .6 STATE ESTIMATORS
The state feedback in the preceding sections is introduced under the assumption that
all state variables are available for connection to a gain . This assumption may or
may not hold in practice . For example, for the dc motor discussed in Figure 11 .4,
the two state variables can be generated by using a potentiometer and a tachometer .
However, if no tachometer is available or if it is available but is very expensive and
we have decided to use only a potentiometer in the design, then the state feedback
cannot be directly applied . In this case, we must design a state estimator . This and
the following sections will discuss this problem .
Consider
x = Ax + bu (11 .53a)
y = cx (11 .53b)
with known A, b, and c . The problem is to use the available input u and output y to
drive a system, called a state estimator, whose output x approaches the actual state
x. The easiest way of building such an estimator is to simulate the system, as shown
in Figure 11 .6. Note that the original system could be an electromechanical one, and
the estimator in Figure 11 .6 may be built using operational amplifier circuits . Be-

Figure 11 .6 Open-loop state estimator .
cause the original system and the estimator are driven by the same input, their states
x(t) and x(t) should be equal for all t if their initial states are the same . Now if
(11 .53) is observable, its initial state can be computed and then applied to the esti-
mator . Therefore, in theory, the estimator in Figure 11 .6 can be used, especially if
both systems start with x(O) = x(0) = 0 . We call the estimator in Figure 11 .6 the
open-loop state estimator.
Let the output of the estimator in Figure 11 .6 be denoted by x. Then it is
described by
X=Ax+bu (11 .54)
Subtracting this equation from (11 .53a) yields
x - X = A(x - R)
Define e(t) : = x(t) - i(t) . It is the error between the actual state and the estimated
state at time t. Then it is governed by
e = Ae (11 .55)
and its solution is
e(t) = eAè(0) = eA`(x(0) - x(0))
Although it is possible to estimate x(0) and then set x(0) = x(0), in practice e(0) is
often nonzero due to estimation error or disturbance . Now if A has eigenvalues in
the open right half plane, then the error e(t) will grow with time . Even if all eigen-
values of A have negative real parts, we have no control over the rate at which e(t)
approaches zero . Thus, the open-loop state estimator in Figure 11 .6 is not desirable
in practice.
Although the output y is available, it is not utilized in the open-loop estimator
in Figure 11 .6. Now we shall compare it with cx and use the difference to drive an
estimator through a constant vector I as shown in Figure 11 .7(a). Then the output x
of the estimator is governed by
x = Ax + bu + 1(y - cx)

or
x=(A-lc)x+bu+ly (11 .56)
and is replotted in Figure 11 .7(b) . We see that the estimator is now driven by u as
well as y . We show in the following that if (A, c) is observable, then (11 .56) can be
designed so that the estimated state x will approach the actual state x as quickly as
desired.
The subtraction of (11 .56) from (11 .53a) yields, using y = cx,
x-x=Ax+bu-(A-lc)X-bu-lcx
= (A - Ic)x - (A - Ic)z = (A - lc)(x - X)
which becomes, after the substitution of e = x - x,
e = (A - Ic)e (11 .57)
(sI - A) -1
L A
(a)
u Y
I I
I I
I I
I I
L . J
(b)
Figure 11 .7 State estimator .

456 CHAPTER 1 1 STATE SPACE DESIGN
Now we show that if (A, c) is observable, then the eigenvalues of (A - lc) can
be arbitrarily assigned (provided complex-conjugate eigenvalues are assigned in
pairs) by choosing a suitable vector 1 . If (A, c) is observable, its observability matrix
in (11 .4) has rank n . The transpose of (11 .4) is
V' = [c' A'c' (A')'c' (A' )n-' c ']
and is also of rank n . By comparing this with (11 .3), we conclude that (A', c') is
controllable. Consequently, the eigenvalues of (A' - c'1') or its transpose (A - lc)
can be arbitrarily assigned by choosing a suitable 1 . This completes the argument .
We list in the following a procedure of designing 1 . It is a simple modification of
the procedure for pole placement .
Procedure for Designing Estimators

Given a 4-dimensional observable (A, c) and a set of eigenvalues 1t,, i = 1, 2, 3, 4,
find a real 4 X I vector I so that the matrix (A - lc) has the set as its eigenvalues.
1. Compute the characteristic polynomial of A : det (sI - A) = s4 + a1s3 + a2s2
+a3s+a4.
2. Compute the desired characteristic polynomial of the estimator in (11 .56):
(S - A 1 )(S - A2)(S - A 3)(s - A4) = S 4 + a1s 3 + a2s2 + a3 S + a4
3. Compute 1' = [ a1 - a 1 a 2 - a2 a3 - a3 a4 - a4] .
4. Compute the equivalence transformation
1 0 0 0 c
P-1 a1 1 0 0 cA
S := = (11 .58)
a2 a1 1 0 cA2
a3 a2 a1 1 cA3
5. Compute 1 = Pi = S - '1.
Now if the eigenvalues of (11 .57) are designed to have large negative real parts
by choosing a suitable 1, then no matter what e(0) is, e(t) will approach zero rapidly .
Therefore, in using the estimator in Figure 11 .7, there is no need to estimate x(0) .
The state estimator in (11 .56) has the same dimension as (11 .53) and is called a full-
dimensional estimator.
11 .6.1 Reduced-Dimensional Estimators

Consider the n-dimensional equation
x = Ax + bu (11 .59a)
y = cx (11 .59b)
The estimator in (11 .56) has dimension n and is thus a full-dimensional estimator .
In this section, we discuss the design of (n - 1)-dimensional estimators . Such an

estimator can be designed using similarity transformations as in the preceding sec-

tion. See Reference [15]. In this section, we discuss a different approach that does
not require any transformation .
Consider the (n - 1)-dimensional state-variable equation
z=Fz+gy+hu (11 .60)
where F, g, and h are respectively (n - 1) X (n - 1), (n - 1) X 1 and (n - 1)
X 1 constant matrices. The equation is driven by y and u. The state vector z(t) is
called an estimate of Tx(t), where T is an (n - 1) X n constant matrix, if
lim Iz(t) - Tx(t)l = 0
t-.
for any x(0), z(0), and u(t). The conditions for (11 .60) to be an estimate of Tx are
1. TA - FT = gc
2. h = Tb
3. All eigenvalues of F have negative real parts .
Indeed, if we define e : = z - Tx, then
e=z-Tx=Fz+gy+hu-T(Ax+bu)
which becomes, after the substitution of z = e + Tx and y = cx,
e=Fe+(FT-TA+gc)x+(h-Tb)u
This equation reduces to, after using the conditions in 1 and 2,
e - Fe
If all eigenvalues of F have negative parts, then e(t) = e Fte(0) approaches zero for
any e (0) . This shows that under the three conditions, (11 .60) is an estimate of Tx .
The matrix equation TA - FT = gc is called a Lyapunov equation . Now we list
the design procedure in the following .
Procedure for Designing Reduced-Dimensional Estimators

1. Choose an (n - 1) X (n - 1) matrix F so that all its eigenvalues have negative
real parts and are different from those of A .
2. Choose an (n - 1) X 1 vector g so that (F, g) is controllable .
3. Solve the (n - 1) X n matrix T from the Lyapunov equation TA - FT =
gc.
4. If the square matrix of order n
is singular, go back to step 1 and/or step 2 and repeat the process . If P is

nonsingular, compute h = Tb . Then
z=Fz+gy+hu (11 .61 a)

x = (11 .61b)
[T] 1[Z]
is an estimator of x in (11 .59).
We give some remarks about the procedure . The conditions that the eigenvalues
of F differ from those of A and that (F, g) be controllable are introduced to insure
that a solution T in TA - FT = gc exists and has full rank. The procedure can
also be used to design a full-dimensional estimator if F is chosen to be of order n.
For a more detailed discussion, see Reference [15]. In this design, we have freedom
in choosing the form of F . It can be chosen as one of the companion forms shown
in (2.77); it can also be chosen as a diagonal matrix . In this case, all eigenvalues
must be distinct, otherwise no g exists so that (F, g) is controllable . See Problem
11 .2. If F is chosen as a Jordan-form matrix, then its eigenvalues can be repeated .
If F is of Jordan form, all solutions of TA - FT = gc can be parameterized . See
Reference [59] . We mention that Lyapunov equations can be solved in MATLAB
by calling the command Iyap .
Example 11 .6.1
Consider the equation in (11 .41) or
0 0
% _ x + U (11 .62a)
[ack,e] = [0 - 1 ] 110]
y = [1 0]x (11 .62b)
Design a reduced-dimensional state estimator with eigenvalue - 4. Equation (11 .62)

has dimension n = 2 . Thus, its reduced-dimensional estimator has dimension 1 and
F reduces to a scalar. We set F = -4 and choose g = 1 . Clearly (F, g) is con-
trollable . The matrix T is a 1 X 2 matrix . Let T = [t1 t2]. Then it can be solved
from
[t1 t2][0 -1] -(-4)[t1 t2] = 1 X [1 0]

or
[0 t1 - t2] + [4t1 4t2] = [1 0]
Thus, we have 4t1 = 1 and t 1 - t2 + 4t2 = 0, which imply k1 = 0 .25 and t2 =

-t1 /3 = -1/12 = -0 .083. The matrix
P
[TC
] -1[0.25 -0.083]
0

11 .7 CONNECTION OF STATE FEEDBACK AND STATE ESTIMATORS 459
is clearly nonsingular and its inverse is
[0.25 -0.083 1 - [3 - 012]

We compute
h = Tb = [0.25 -0.08311 = -0.83
Thus the one-dimensional state estimator is

i= -4z+y-0 .83u (11 .63a)
y (11 .636)
x = [3 -1 02] [z] = [3y 12z]
Exercise 11 .6 .1
In Example 11 .6.1, if F is chosen as -1 and g = 1, can you find a T to meet the

Lyapunov equation?
[Answer : No. In this case, the eigenvalue of F coincides with one of the
eigenvalues of A, and the first condition of the procedure is not met .]
11 .7 CONNECTION OF STATE FEEDBACK AND STATE ESTIMATORS

i = Ax + bu (11 .64a)
y = cx (11 .64b)
If (11 .64) is controllable, by introducing the state feedback

u=r kx
the eigenvalues of (A - bk) can be arbitrarily assigned . If the state is not available,
we must design a state estimator such as
z=(A-Ic)i+bu+ly (11 .65)
to generate an estimate i of x . We then apply the state feedback to i as shown in

Figure 11 .8, that is,
u=r-ki (11 .66)

Figure 11 .8 State feedback and state estimator.
The feedback gain is designed for the original state x . Now it is connected to the
estimated state z . Will the resulting system still have the desired eigenvalues? This
will be answered in the following .
The substitution of (11 .66) and y = cx into (11 .64a) and (11 .65) yields
X=Ax+b(r-ki)
and
x=(A-Ic)i+lcx+b(r-ki)
They can be combined as
r (11 .67a)
[z] - [lc A - I ck bk] [i] + [b]
y = cx = [c (11 .67b)
x
0] [x]
This equation describes the overall system in Figure 11 .8. Consider
-0]
P = II P -1 (11 .68)
I =
The inverse of P happens to equal itself. By applying the equivalence transformation
[i] [I -I] [i]

to (11 .67), we will finally obtain, using (11 .22c),
z A - bk bk x b]
+ r (11 .69a)
[X] [ 0 A - IC][~]
X [0
Y = [c 0][X] (11 .69b)

i
Because any equivalence transformation will not change the characteristic poly-
nomial and transfer function, the characteristic polynomial of (11 .67) equals that

of (11 .69) and, using (B.3), is

sI-A+bk -bk
det
LC 0 sI - A + Ic (11 .70)
= det (sI - A + bk) det (sl - A + Ic)
Similarly, the transfer function of (11 .67) equals that of (11 .69) and, using (B .6), is
A - A + bk -bk 1 b
[c 0]C
0 sI - A + Ic] [0]
(sI - A + bk) - ' a b]
_ [c U]L (11 .711
0 (sI - A + lc)-1]C0
[b]
_ [c(sI - A + bk) -1 ca] = c(sI - A + bk) -1b
where a = (sI - A + bk) -1bk(sI - A + lc) -1 . From (11 .70), we see that the
eigenvalues of the overall system in Figure 11 .8 consist of the eigenvalues of the
state feedback and the eigenvalues of the state estimator . Thus the connection of the
feedback gain to the output of the estimator does not change the original designs .
Thus the state feedback and the state estimator can be designed separately . This is
often referred to as the separation property .
The transfer function of the overall system in Figure 11 .8 is computed in (11 .71).
It equals (11 .30c) and has only n poles . The overall system, however, has 2n eigen-
values . Therefore, the state-variable equations in (11 .67) and (11 .69) are not minimal
equations. In fact, they can be shown to be uncontrollable and unobservable . The
transfer function of the state feedback system with a state estimator in Figure 11 .8
equals the transfer function of the state feedback system without a state estimator in
Figure 11 .5 ; thus, the state estimator is hidden from the input r and output y and its
transfer function is canceled in the design . This can be explained physically as
follows. In computing transfer functions, all initial states are assumed to be zero,
therefore, we have x(0) = i(0) and, consequently, x(t) = i(t) for all t and for any
u(t). Thus the estimator does not appear in (11 .71). This situation is similar to the
pole-zero cancellation design in Chapter 10 .
We have shown the separation property by using the full-dimensional estimator.
The property still holds if we use reduced-dimensional estimators . The proof, how-
ever, is slightly more complicated . See Reference [15] .
11 .7 .1 Comparison with Linear Algebraic Method

Consider a minimal state-variable equation with transfer function
G(s) = N(s)
D(s)
After introducing state feedback and the state estimator, the transfer function of the
resulting overall system will be of the form

N(s)
G ,(s)
(s) D0 (s)
where D0(s) has the same degree and the same leading coefficient as D(s). Note that
the numerator of G0 (s) is the same as that of G(s) . Clearly G0 (s) is implementable
for the given G(s) (see Section 9.2) and the linear algebraic methods discussed in
Chapter 10 can be used to implement such G0(s). In this subsection, we use an
example to compare the design using the state-variable method with that using the
linear algebraic method .
Consider the minimal equation in Example 11 .4.2 or
j =
[x2] - [0 -1] x + [10] u

N(s) = 10
G(s) = (11 .72)
D(s) s2 + s
It is computed in Example 11 .4.2 that the feedback gain k = [0 .8 0.3] in u
r - kx will shift the poles of G(s) to -2 j2 and that the resulting transfer
function from r to y is
N0(s) _ 10
G0(s) = (11 .73)
D0 (s) s 2 + 4s + 8
Now if the state is not available for feedback, we must design a state estimator .
A reduced-dimensional state estimator with eigenvalue -4 is designed in Example
11.6.1 as
z= -4z + y - 0.83u
x y
[3y - 12z
Its basic block diagram is plotted in Figure 11 .9(a). Now we apply the feedback gain
k to x as shown in Figure 11 .9(b). This completes the state-variable design .
In order to compare with the linear algebraic method, we compute the transfer
functions from u to w and y to w of the block bounded by the dashed line in Figure
11 .9(b). There is no loop inside the block ; therefore, using Mason's formula, we
have
C1 (s)
= U(s)
= (-0.83) .
s + 4
. (- 12) (0.3)
= s + 4
and
_ (-12) (0.3) 1 .7s + 3.2
C 2(s)
2(S) = Y(s) + 0.3 3 + 0.8 =
s+4 s+4

u Y
I
I I
I
I I
I
I I
I
I I
I
I I
I
I I
i
I I
I I I
L 1 J
State feedback State estimator
(b)
Figure 11 .9 (a) State estimator. (b) State feedback and state estimator .
These two transfer functions are plotted in Figure 11 .10(a) and then rearranged
in Figure 11 .10(b) . It is the plant input/output feedback configuration studied in
Figure 10 .15.
Now we shall redesign the problem using the method in Section 10.6, namely,
given G(s) in (11 .72) and implementable G 0(s) in (11 .73), find two compensators of
the form
C1(s)A(s) CZ(s) A( s)
so that the resulting system in Figure 10 .15(b) or Figure 11 .10 has G0(s) as its transfer

10 Y
s(s + 1)
3
s+4
1 .7s + 3 .2
s + 4
(a)
r M 10
s(s + 1)
w
r
I
I
1 .7s + 3 .2 F-
I
I
I
I (s + 4) -t
L ) I
Figure 11 .10 (a) Compensators . (b) Plant I/O feedback configuration .
function. Using the procedure in Section 10 .6, we first compute

G0(s) _ 10 _ 1 Np(s)
N(s) (s2 + 4s + 8) 10 s2 + 4s + 8 Dr (s)
with NP(s) = l and Dp(s) = s 2 + 4s + 8 . Because deg Na(s) = 0, we must introduce
a polynomial A(s) of degree 2 - 1 = 1 . We choose it as A(s) = s + 4 for the
eigenvalue of the estimator is chosen as -4 in the state-variable approach . Thus we
have
A(s) = Np(s)A(s) = 1 (s + 4) = s + 4
The polynomials L(s) = Lls + L o and M(s) = M,s + Mo can be determined, using
(10.73), from
L(s)D(s) + M(s)N(s) = A(s)(Dp(s) - NN(s)D(s))
=(s+4)(s 2 +4s+8-s 2 -s)=3s 2 +20s+32
or from the following linear algebraic equation, using (10.75),
0 10 1 0 0 Lo 32
1 0 ;0 10 Mo 20
(11 .74)
1 0 ;1 0 L, 3
0 011 0 M1 0

11 .8 LYAPUNOV STABILITY THEOREM 465
The first equation of (11 .74) yields 10M0 = 32 or Mo = 3 .2. The fourth equation
of (11 .74) yields L 1 = 0 . The second and third equations of (11 .74) are Lo +
10M1 = 20 and Lo + L 1 = 3, which yield Lo = 3 and M 1 = 1.7. Thus the
compensators are
= L(s) .3 M(s) 1 .7s + 3 .2
C 1 (s)
A(s) s + 4 C2(s) A(s) s + 4
They are the same as those computed using state-variable equations .
Now we compare the state-variable approach and the transfer function approach
in designing this problem . The state-variable approach requires the concepts of con-
trollability and observability . The design requires computing similarity transforma-
tions and solving Lyapunov matrix equations . In the transfer function approach, we
require the concept of coprimeness (that is, two polynomials have no common fac-
tors) . The design is completed by solving a set of linear algebraic equations . There-
fore, the transfer function approach is simpler in concept and computation than the
state-variable approach . The transfer function approach can be used to design any
implementable transfer function . The design of any implementable transfer function
by using state-variable equations would be more complicated and has not yet ap-
peared in any control text .
11 .8 LYAPUNOV STABILITY THEOREM
A square matrix A is said to be stable if all its eigenvalues have negative real parts .
One way to check this is to compute its characteristic polynomial
0(s) = det (sI - A)
We then apply the Routh test to check whether or not 0(s) is a Hurwitz polynomial.
If it is, A is stable ; otherwise, it is not stable .
In addition to the preceding method, there is another popular method of checking
the stability of A . It is stated as a theorem .
THEOREM 11 .2 (Lyapunov Theorem)
All eigenvalues of A have negative real parts if for any symmetric positive
definite matrix N, the Lyapunov equation
A'M + MA = -N
has a symmetric positive definite solution M .
COROLLARY 11 .2
All eigenvalues of A have negative real parts if, for any symmetric positive
semidefinite matrix N = n'n with the property that (A, n) is observable, the
Lyapunov equation
A'M + MA = -N (11 .75)
has a symmetric positive definite solution M .

We give an intuitive argument to establish these results . Consider

x(t) = Ax(t) (11 .76)
Its solution is x(t) = eA`x(0). If the eigenvalues of A are c1 , c2, and c 3, then every
component of x(t) is a linear combination of e`1`, e`21, and e`3`. These time functions
will approach zero as t--*- if and only if every c ; has a negative real part . Thus we
conclude that the response of (11 .76) due to any nonzero initial state will approach
zero if and only if A is stable .
We define
V(x) : = x'Mx (11 .77)
If M is symmetric positive definite, V(x) is positive for any nonzero x and is zero
only at x = 0 . Thus the plot of V(x) will be bowl-shaped, as is shown in Figure
11.11 . Such a V(x) is called a Lyapunov function .
Now we consider the time history of V(x(t)) along the trajectory of (11 .76).
Using x' = x'A', we compute,
dt
V(x(t)) = dt (x'Mx) = x'Mx + x'Mx
= x'A'Mx + x'MAx = x'(A'M + MA)x
which becomes, after the substitution of (11 .75),
V(x(t)) _ -x'Nx (11 .78)

dt
If N is positive definite, then dV(x)/dt is strictly negative for all nonzero x . Thus,
for any initial state x(0), the Lyapunov function V(x(t)) decreases monotonically
until it reaches the origin as shown in Figure 11 .11(a) . Thus x(t) approaches zero as
t-*oc and A is stable . This establishes the Lyapunov theorem .
If N is symmetric positive semidefinite, then dV(x)/dt < 0, and V(x(t)) may not
decrease monotonically to zero . It may stay constant along some part of a trajectory
such as AB shown in Figure 11 .11(b) . The condition that (A, n) is observable,
however, will prevent dV(x(t))/dt = 0 for all t. Therefore, even if dV(x(t))/dt = 0
V( x) V(x)
(a) (b)
Figure 11 .11 Lyapunov functions .

for some t, V(x(t)) will eventually continue to decrease until it reaches zero as t---> - .
This establishes the corollary . For a more detailed discussion of these results, see
Reference [15].
11 .8.1 Application-A Proof of the Routh Test

Consider the polynomial of degree 3
D(s) = s 3 + a,s 2 + a2s + a 3 (11 .79)
We form the Routh table :
S 3 I a2
a
k3 =1
a,
s2 0 a2 - 3
al
=: C
a
k2 = i s
C
C
k, = - 1
a3
Then the Routh test states that the polynomial is Hurwitz if and only if
a3
a, > 0 a3 > 0 c : = a2 - >0 (11 .80)
a,
Now we use Corollary 11 .2 to establish the conditions in (11 .80). Consider the
block diagram in Figure 11 .12 with ki defined in the Routh table . The block diagram
has three loops with loop gains -1/k 3 s, - 1/k 1 k2s2, and -I /k 2k 3 s2, where we
have assumed implicitly that all k; are different from 0 . The loop with loop gain
- l/k 3s and the one with - 1/k l k 2s2 do not touch each other . Therefore, the char-
acteristic function in Mason's formula is
1 1
=1-(- 1 1- 1)+
k 3s k,k 2s2 k 2 k3 s 2 k 3s (k 1T2 s 2)
Let us consider x 3 as the output, that is, y = x3 . Then the forward path gain from u
toy is
1
P1 = k 3s
u + 1 x 3 =y + 1 x 2 klzl
1
k3s k2s k1 s
Figure 11 .12 Block diagram.

and, because the path does not touch the loop with loop gain -1/k,k 2 s 2 , the cor-
responding 0, is,
1
0,=I- -
kl 2)
Thus the transfer function from u to y is
k3 s (1 + k1kz s2 )
G(s) =
1 1 1 1
1 + k
3s + k,kzs2 + +
k2k3sz k,k2k3s 3
1 1
- s2 +
G(s) = k3 k1k2k3
3 1 k, + k3 1
s + k s2 s
3 + k1k2k3 + kl kzk3
From k, in the Routh table, we can readily show 1/k 3 = a,, 1/k 1k2k3 = a 3 , and
(k 1 + k3)/k 1k2k3 = a 2 . Thus, the transfer function from u to y of the block diagram
is
O a,s z + a3 = N(s) ( )
G s = 1181
s 3 +a 1 s 2 +a 2s+a 3 D(s)
Clearly, N(s) and D(s) have no common factor . Thus G(s) has three poles . Now we
develop a state-variable equation to describe the system in Figure 11 .12 . We assign
the state variables as shown . Then we have
k1-i1 = x 2 k2x2 = x 1 + x3
k3X3 = _X2 - X3 + u y = x3
These can be expressed in matrix form as
0
0
-1
X = x + 0 u (11 .82a)
k2 1
0 k3
k3 k3
y = [0 0 l]x (11 .82b)
Both (11 .81) and (11 .82) describe the same block diagram, thus (11 .81) is the transfer
function of the state-variable equation in (11 .82). Because the dimension of (11 .82)
equals the number of poles of G(s) in (11 .81), (11 .82) is a minimal realization

of G(s) and the characteristic polynomial of A in (11 .82) equals D(s) = s 3 + a1s2
+ a2s + a 3.
Now we use Corollary 11 .2 to develop a condition for A to be stable or, equiv-
alently, for D(s) to be Hurwitz . Define
0 0 0 0
N= n'n = 0 [0 0 V-21 = 0 0 0
0 0 2
It is symmetric and positive semidefinite . Because the matrix
k3 k3
N/2 V-2 d
_k2k3 k2k3
with d = - \(k 3 - k2)/k 2k3, is nonsingular, the pair (A, n) is observable . There-
fore Corollary 11 .2 can be used to establish a stability condition for A . It is straight-
forward to verify the following
-1
0 0
k2 k1 0 0
0 k2 0
0 k3
(11 .83)
1
0
k1 0 0 k1 0 0 0
-1 1
+ 0 k2 0 0 0 0 0
k2 k2
0 0 k3 -1 -1 0 0 2
--
k3 k3
Therefore the symmetric matrix
k1 0 0
M = 0 k2 0
0 0 k3
is a solution of the Lyapunov equation in (11 .83). Consequently, the condition for
A in (11 .82) to be stable is M positive definite or, equivalently,

k, > 0 k, k, > 0 k,k2 k3 > 0

which implies
k,= c >0 k 2 = a'>0 k3 = 1 >0

a3 c a,
This set of conditions implies
a,>0 c=a2 -a3 >0 a3 >0

a,
which is the same as (11 .80). This is one way to establish the Routh stability test .
For a more general discussion, see Reference [15] .
11 .9 SUMMARY AND CONCLUDING REMARKS
This chapter introduced state space designs . We first introduced the concepts of
controllability and observability . We showed by examples that a state-variable equa-
tion is minimal if and only if it is controllable and observable . We then introduced
equivalent state-variable equations ; they are obtained by using a nonsingular matrix
as an equivalence transformation . Any equivalent transformation will not change the
eigenvalues of the original equation or its transfer function . Neither are the properties
of controllability and observability affected by any equivalence transformation .
If a state-variable equation is controllable, then it can be transformed, using an
equivalence transformation, into the controllable-form equation . Using this form, we
developed a procedure to achieve arbitrary pole placement by using constant-gain
state feedback. Although state feedback will shift the eigenvalues of the original
system, it does not affect the numerator of the system's transfer function .
If (A, c) is observable, then (A', c'), where the prime denotes the transpose, is
controllable . Using this property, o - + that if (A, c) is observable, a state
estimator with any eigenvalue can be designed . We also discussed a method of
designing estimators by solv g Lyapunov equations . The connection of state feed-
back gains to estimated st tes, rather than to the original state, was justified by
establishing the separating property . We then compared the state space design with
the linear algebraic metho developed in Chapter 10 . It was shown that the transfer
function approach is simpler, in both concept and computation, than the state-
variable approach. Finally, we introduced the Lyapunov stability theorem.
To conclude this chapter, we discuss constant gain output feedback. In constant-
gain state feedback, we can assign all n poles arbitrarily . In constant-gain output
feedback of single-variable systems, we can arbitrarily assign only one pole ; the
remaining poles cannot be assigned . For example, consider the constant-gain output
feedback system shown in Figure 11 .13(a) . We can assign one pole in any place .
For example, if we assign it at - 3, then from the root loci shown in Figure 11.13(b),
we can see that the other two poles will move into the unstable region . Therefore,
the design is useless . For constant-gain output feedback, it is better to use the root-

PROBLEMS 471
IM S
i
I( I
I I > - Res
-3 -2 -1 v 0 1 2 3
(a) (b)
Figure 11 .13 (a) Constant gain output feedback. (b) Root loci .
locus method in Chapter 7 to carry out the design . The design using a compensator
of degree 1 or higher in the feedback path is called dynamic output feedback . The
design of dynamic output feedback is essentially the same as the design of state
estimators and the design in Chapter 10 . Therefore, it will not be discussed .
PROBLEMS
11 .1 . Check the controllability and observability of the following state-variable

equations:
a. z= -x+u y = 2x
b.x = 1 x+
CO J C 0J u
y = [2 -2]x
c. x = [-2 + [ 4] u
0] x
y = [1 O]x
-3 1 0 2
d. x= -2 0 0 x+ 4 u
1 0 0 1
y = [1 0 0]x
Which are minimal equations?
11 .2. Show that the equation
% = C A,0 A20 1
y = [2 0] x

is controllable if and only if A, =?~= A 2. Show that the equation is always not
observable .
11 .3. Show that the equation
X= A1Jx+
[0 [b2]
y = [2 0] x
1.
is controllable if and only if b2 0 0. It is independent of b Show that the
equation is always observable .
11.4. Let X = Px with
P - [-1 3]
Find equivalent state-variable equations for the equations in Problem 11 .1(b)
and (c).
11 .5. Check the controllability and observability of the equations in Problem 11 .4.
Also compute their transfer functions . Does the equivalence transformation
change these properties and transfer functions?
11 .6. Given a plant with transfer function
10
G(s) =
s(s - 1)(s + 2)
use the procedure in Example 11 .4.1 to find the feedback gain such that the
resulting system has poles at - 2, - 3, and - 4.
11 .7. Redesign Problem 11 .6 using the state-variable method. Is the feedback gain
the same?
11 .8. Consider the state-variable equation
x= I All0 A12x+
A 221 0 1 1
u
Show that the equation is not controllable . Show also that the eigenvalues of
A22 will not be affected by any state feedback . If all eigenvalues of A22 have
11 , 1
negative real parts and if (A b ) is controllable, then the equation is said
to be stabilizable .
11 .9. Consider
x= [1 1Jx+
L O]
y = [2 -1]x

PROBLEMS 473
Find the feedback gain k in u = r - kx such that the resulting system has
eigenvalues at - 2 and - 3 .
11 .10. Consider
x+
[-2 0] [-2] u
y = [1 0]x
Find the feedback gain k in u = r - kx such that the resulting system has
eigenvalues at - 2 2j.
11 .11 . Design a full-dimensional state estimator with eigenvalues - 3 and - 4 for
the state-variable equation in Problem 11 .9.
11 .12 . Design a full-dimensional state estimator with eigenvalues - 3 and - 4 for
the state-variable equation in Problem 11 .10.
11 .13 . Design a reduced-dimensional state estimator with eigenvalue - 3 for the
state-variable equation in Problem 11 .9.
11 .14. Design a reduced-dimensional state estimator with eigenvalue - 3 for the
state-variable equation in Problem 11 .10 .
11 .15. Consider a co ntrollable n-dimensional (A, b) . Let F be an arbitrary n X n
matrix and let k_be an arbitrary n X 1 vector. Show that if the solution T of
AT - TF = bk is nonsingular, then (A - bkT -1) has the same eigenvalues
as F .
11 .16. Connect the state feedback in Problem 11 .9 to the estimator designed in Prob-
lem 11.11 . Compute the compensators from u to w and from y to w in Figure
11 .8. Also compute the overall transfer function from r to y . Does the overall
transfer function completely characterize the overall system? What are the
missing poles?
11 .17. Repeat Problem 11 .16 by using the estimator in Problem 11 .13. Does the
overall transfer function equal the one in Problem 11 .16?
11 .18. Connect the state feedback in Problem 11 .10 to the estimator designed in
Problem 11.12 . Compute the compensators from u to w and from y to w in
Figure 11 .8. Also compute the overall transfer function from r to y . Does the
overall transfer function completely characterize the overall system? What are
the missing poles?
11 .19. Repeat Problem 11 .18 by using the estimator in Problem 11 .14 . Does the
overall transfer function equal the one in Problem 11 .18?
11 .20 . Redesign Problem 11 .17 using the linear algebraic method in Section 10 .6.
Which method is simpler?
11 .21 . Redesign Problem 11 .19 using the linear algebraic method in Section 10 .6.
Which method is simpler?

11 .22. Check whether the following matrices are positive definite or semidefinite .
0 0 -2 5 0 -2
11
Jo0 3 0 Jo0 3 0
C21 11 -2 0 1 -2 0 1
11 .23. Compute the eigenvalues of the matrices in Problem 11 .22. Are they all real?
From the eigenvalues, check whether the matrices are positive definite or
semidefinite.
11 .24. Consider the system in Problem 11 .9 . Find the state feedback gain to minimize
the quadratic performance index in (11 .45) with R = 1 .
11 .25. Consider
0 1 0
A2 = 0 0 1
-2 -3 -1
a. Use the Routh test to check their stability .
b. Use the Lyapunov theorem to check their stability .
Discrete-Time
Systcm Analysis
12.1 INTRODUCTION
Signals can be classified as continuous-time and discrete-time . A continuous-time

signal is defined for all time, whereas a discrete-time signal is defined only at discrete
instants of time . Similarly, systems are classified as analog and digital . Analog sys-
tems are excited by continuous-time signals and generate continuous-time signals as
outputs . The input and output of digital systems are discrete-time signals, or
sequences of numbers . A system with an analog input and a digital output or vice
versa can be modeled as either an analog or a digital system, depending on conven-
ience of analysis and design .
A control system generally consists of a plant, a transducer, and a compensator
or controller, as shown in Figure 12 .1(a) . The plant may consist of an object, such
as an antenna, to be controlled, a motor, and possibly a power amplifier . Most plants
in control systems are analog systems . The transducers and the compensators dis-
cussed in the preceding chapters are analog devices . However, because of their
reliability, flexibility, and accuracy, digital transducers and compensators are in-
creasingly being used to control analog plants, as is shown in Figure 12 .1(b). In this
and the following chapters, we discuss how to design digital compensators to control
analog plants.
The organization of this chapter is as follows . In Section 12 .2 we discuss the
reasons for using digital compensators . Section 12 .3 discusses A/D and D/A con-
verters; they are needed to connect digital compensators with analog plants and vice
versa. We then introduce the z-transform ; its role is similar to the Laplace transform
in the continuous-time case . The z-transform is then used to solve difference equa-
475

476 CHAPTER 12 DISCRETE-TIME SYSTEM ANALYSIS
r +e u y
C (s) G (s)
(a)
Clock
r +e Digital y
A/D System D/A ---I- G (s)
(b)
Figure 12 .1 (a) Analog control system . (b) Digital control system.
tions and to develop transfer functions . Discrete-time state-variable equations to-

gether with the properties of controllability and observability are then discussed . We
also discuss basic block diagrams and the realization problem . Finally we discuss
stability, the Jury test, the Lyapunov Theorem, and frequency responses . Most con-
cepts and results in the discrete-time case are quite similar to those in the continuous-
time case; therefore, some results are stated without much discussion .
12 .2 WHY DIGITAL COMPENSATORS?
A signal is called a continuous-time or analog signal if it is defined at every instant

of time, as shown in Figure 12 .2(a). Note that a continuous-time signal is not nec-
essarily a continuous function of time ; as is shown, it may be discontinuous .
The temperature in a room is a continuous-time signal . If it is recorded or
sampled only at 2 P .M. and 2 A .M . of each day, then the data will appear as shown
y (t) At)
11111 1 1
\~Sampling period 1 1
(a) (b)
Figure 12 .2 Continuous-time and discrete-time signals .

12 .2 WHY DIGITAL COMPENSATORS? 477
y(kT) y(kT)
k
1 2 3 4
(a) (b)
Figure 12 .3 Digital signal and its binary representation .
in Figure 12.2(b). It is called a discrete-time signal, for it is defined only at discrete

instants of time . The instants at which the data appear are called the sampling in-
stants; the time intervals between two subsequent sampling instants are called the
sampling intervals. In this text, we study only discrete-time signals with equal sam-
pling intervals, in which case the sampling interval is called the sampling period.
The amplitude of a discrete-time signal can assume any value in a continuous
range . If its amplitude can assume values only from a finite set, then it is called a
digital signal, as shown in Figure 12 .3(a). The gap between assumable values is
called the quantization step . For example, if the temperature is recorded with a digital
thermometer with integers read out, then the amplitude can assume only integers,
and the quantization step is 1 . Thus, a digital signal is discretized in time and quan-
tized in amplitude, whereas a discrete-time signal is discretized in time but not quan-
tized in amplitude . Signals processed on digital computers are all digital signals .
Clearly, an error arises when a discrete-time signal is quantized . The error depends
on the number of bits used to represent the digital signal . The error may be appre-
ciable for a 4-bit representation . However, if 16 or more bits are used to represent
digital signals, then the errors between discrete-time and digital signals are often
negligible .
Consider digital signals that are limited up to the first decimal point-such as
10.1, 0.7, and 2 .9 . The product of two such nonzero numbers may become zero . For
example, 0 .2 X 0.1 = 0.0, which is indistinguishable from 0 .2 X 0 = 0 . For this
and other reasons, it is difficult to analyze digital signals . Therefore, digital signals
are often considered to be discrete-time signals in analysis and design . The discrep-
ancies between them are then studied separately as an error problem . Representation
of discrete-time signals, such as 1/3, may require an infinite number of bits, which
is not possible in practice; therefore, in implementation, discrete-time signals must
be quantized to become digital signals . In conclusion, in analysis, we consider only
"discrete-time signals ; in implementation, we consider only digital signals . In this
text, we make no distinction between these two types of signals, and discrete-time
and digital are often used interchangeably .
In processing and transmission, digital signals are expressed in binary form, a
string of zeros and ones or bits, as shown in Figure 12 .3(b). Because strings of zeros
and ones do not resemble the waveforms of original signals, we may call a digital

signal a nonanalog signal . A continuous-time signal usually has the same waveform
as the physical variable, thus it is also called an analog signal .
Systems that receive and generate analog signals are called analog systems .
Systems that receive and generate digital signals are called digital systems. However,
an analog system can be modeled as a digital system for convenience of analysis
and design . For example, the system described by (2 .90) is an analog system . How-
ever, if the input is stepwise, as shown in Figure 2 .23 (which is still an analog signal),
and if we consider the output only at sampling instants, then the system can be
modeled as a digital system and described by the discrete-time equation in (2.92).
This type of modeling is used widely in digital control systems . A system that has
an analog input and generates a digital output, such as the transducer in Problem
3 .11, can be modeled as either an analog system or a digital system .
We compare analog and digital techniques in the following :
1. Digital signals are coded in sequences of 0 and 1, which in terms are represented
by ranges of voltages (for example, 0 from 0 to 1 volt and 1 from 2 to 4 volts) .
This representation is less susceptible to noise and drift of power supply .
2. The accuracy of analog systems is often limited. For example, if an analog
system is to be built using a resistor with resistance 980 .5 ohms and a capacitor
with capacitance 81 .33 microfarads, it would be difficult and expensive to obtain
components with exactly these values . The accuracy of analog transducers is
also limited . It is difficult to read an exact value if it is less than 0 .1 % of the
full scale . In digital systems, there is no such problem, however. The accuracy
of a digital device can be increased simply by increasing the number of bits .
Thus, digital systems are generally more accurate and more reliable than analog
systems.
3. Digital systems are more flexible than analog systems . Once an analog system
is built, there is no way to alter it without replacing some components or the
entire system. Except for special digital hardware, digital systems can often be
changed by programming . If a digital computer is used, it can be used not only
as a compensator but also to collect data, to carry out complex computation,
and to monitor the status of the control system . Thus, a digital system is much
more flexible and versatile than an analog system .
4. Because of the advance of very large scale integration (VLSI) technology, the
price of digital systems has been constantly decreasing during the last decade .
Now the use of a digital computer or microprocessor is cost effective even for
small control systems.
For these reasons, it is often desirable to design digital compensators in control
systems.
12.3 A/D AND D/A CONVERSIONS
Although compensators are becoming digital, most plants are still analog . In order
to connect digital compensators and analog plants, analog signals must be converted
into digital signals and vice versa . These conversions can be achieved by using

12 .3 A/D AND D/A CONVERSIONS 479
xp x2 x3
eout
- E - 2R 4R 8R
T T
1 0
(a) (b)
Figure 12.4 D/A converter .
analog-to-digital (A/D) and digital-to-analog (D/A) converters . We now discuss

how these conversions are achieved .
Consider the operational amplifier circuit shown in Figure 12 .4. It is essentially
the operational amplifier circuit shown in Figure 3 .15 . As in (3.39), the output can
be shown to be
R R R R
U _ - (x' 2R + x2 4R + x3 8R + x4 16R E
= -(x,2 - ' + x2 2 -2 + x 3 2 -3 + x42 -4 )E
where E is the supplied voltage, and xt is either 1 or 0, closed or open . The bit x0 is
called the sign bit. If x0 = 0; then E > 0; if x0 = 1, then E < 0. If xo x 1x2x3 x4 =
11011, and if E = 10 volts, then
2-3
U = -(1 . 2 -1 + 1 + 1 . 2 -4) . (- 10) = 0 .6875
The circuit will hold this value until the next set of x; is received . Thus the circuit
changes a five-bit digital signal xox1 x2x3x4 into an analog signal of magnitude
0.6875, as shown in Figure 12 .4(b). Thus the circuit can convert a digital signal into
an analog signal, and is called a D/A converter . The D/A converter in Figure 12 .4
is used only to illustrate the basic idea of conversion; practical D/A converters
usually use different circuit arrangements so that resistors have resistances closer to
each other and are easier to fabricate . The output of a D/A converter is discontinuous,
as is shown in Figure 12 .4. It can be smoothed by passing through a low-pass filter.
This may not be necessary if the converter is connected to a plant, because most
plants are low-pass in nature and can act as low-pass filters .
The analog-to-digital conversion can be achieved by using the circuit shown in
Figure 12.5(a). The circuit consists of a D/A converter, a counter, a comparator, and
control logic . In the conversion, the counter starts to drive the D/A converter . The
output of the converter is compared with the analog signal to be converted . The
counter is stopped when the output of the D/A converter exceeds the value of the
analog signal, as shown in Figure 12 .5(b). The value of the counter is then transferred
to the output register and is the digital representation of the analog signal .

Comparator
Analog signal u
D/A Converter
Counter and Control y- t

e ' I
output registers logic
Conversion time
(a) (b)
Figure 12 .5 A/D converter and conversion time.
We see from Figure 12 .5(b) that the A/D conversion cannot be achieved in-
stantaneously ; it takes a small amount of time to complete the conversion (for ex-
ample, 2 microseconds for a 12-bit A/D converter) . Because of this conversion time,
if an analog signal changes rapidly, then the value converted may be different from
the value intended for conversion . This problem can be resolved by connecting a
sample-and-hold circuit in front of an A/D converter . Such a circuit is shown in
Figure 12.6. The field-effect transistor (FET) is used as a switch ; its on and off states
are controlled by a control logic . The voltage followers [see Figure 3 .14(a)] in front
and in back of the switch are used to eliminate the loading problem or to shield the
capacitor from other parts of circuits . When the switch is closed, the input voltage
will rapidly charge the capacitor to the input voltage . When the switch is off, the
capacitor voltage remains almost constant. Hence the output of the circuit is stepwise
as shown . Using this device, the problem due to the conversion time can be elimi-
nated . Therefore, a sample-and-hold circuit is often used together with an A/D
converter.
With A/D and D/A converters, the analog control system in Figure 12 .1 (a) can
be implemented as shown in Figure 12 .1(b) . We call the system in Figure 12 .1(b) a
digital control system . In the remainder of this chapter, we discuss digital system
analysis ; design will be discussed in the next chapter .
X (t) FET
x()
u (t) u (t)
Voltage
follower
Control
logic 1 ~ t
Figure 12 .6 Sample-and-hold circuit .

12.4 THE z-TRANSFORM 481
u (k) y (k)
Discrete
system
Figure 12.7 Discrete-time system .
12 .4 THE z-TRANSFORM
Consider the discrete-time system shown in Figure 12.7. If we apply an input se-
quence u(k) : = u(kT ), k = 0, 1, 2, . . . , then the system will generate an output
sequence y(k) : = y(kT) . This text studies only the class of discrete-time systems
whose inputs and outputs can be described by linear difference equations with con-
stant real coefficients such as
3y(k + 2) + 2y(k + 1) - y(k) = 2u(k + 1) - 3u(k) (12.1)
or
3y(k) + 2y(k - 1) - y(k - 2) = 2u(k - 1) - 3u(k - 2) (12.2)
In order to be describable by such an equation, the system must be linear, time-
invariant, and lumped (LTIL) . Difference equations are considerably simpler than
the differential equations discussed in Chapter 2. For example, if we write (12 .2) as
1
y(k) = 3 [-2y(k - 1) + y(k - 2) + 2u(k - 1) - 3u(k - 2)]
then its response due to the initial conditions y( - 2) = 1, y( - 1) = - 2 and the

unit-step input sequence u(k) = 1, for k = 0, 1, 2, . . . , and u(k) = 0 for k < 0,
can be computed recursively as
1
y(O) = [-2y(- l) + y(-2) + 2u(-1) - 3u(-2)]
3
1 5
3 [-2x(-2)+ 1+2x0-3x0]=3
1
y(1) [-2y(0) + y(-1) + 2u(0) - 3u(-1)]
3
1 5 -10
-2 X--2+2 =
3 3 9
1
y(2) [-2y(1) + y(O) + 2u(1) - 3u(0)]
3
1 -10 5 26
- -2X + - + 2 - 3 =-
3 9 3 27
and so forth . Thus, the solution of difference equations can be obtained by direct
substitution . The solution obtained by this process is generally not in closed form,

and it is difficult to abstract from the solution general properties of the equation . For
this and other reasons, we introduce the z-transform.
Consider a sequence f (k) . The z-transform of f (k) is defined as
F(z) := Z[f(k)] :_ I f(k)z -k (12 .3)

0
where z is a complex variable . The z-transform is defined for f (k) with k ? 0 ; f (k)
with k < 0 does not appear in F(z) . If we write F(z) explicitly as
F(z) = f(O) + f(l)z -1 + f(2)z -2 + f(3)z -3 + . . .
then z - `can be used to indicate the ith sampling instant . In other words, z indicates
-;
the initial time k = 0 ; z- 1 indicates the time instant at k = 1 and, in general, z
z-1
indicates the ith sampling instant . For this reason, is called the unit-delay
element.
The infinite power series in (12 .3) is not easy to manipulate . Fortunately, the
z-transforms of sequences encountered in this text can be expressed in closed form .
This will be obtained by using the following formula
1
1 + r + r2 + r 3 + =I
r k = ( 12 .4)
0 1 - r
where r is a real or complex constant with amplitude less than 1, or Irl < 1 .
Example 12 .4 .1
Consider f(k) = e-akT, k = 0, 1, 2 Its z-transform is
e-akTZ -k = (e-aTZ-1)k = 1 1 z
F(z) _ e-aTZ-1 _ (12 .5)
0 0 - Z - e-aT
This holds only if le - aTZ -1 1 < 1 or Ie-aTl < ~ zJ . This condition, called the region
of convergence, will be disregarded, however, and (12 .5) is considered to be defined
for all z except at z = e - aT. See Reference [18] for a justification .
If a = 0, a - akT equals 1 for all positive k. This is called the unit-step sequence,
as is shown, in Figure 12 .8(a), and will be denoted by q(k) . Thus we have
q (k) S(k - 3)
1 1
'-k >-k
-2 -1 0 2 3 4 5 -1 1 2 3 4 5
(a) (b)
Figure 12.8 (a) Step sequence . (b) Impulse sequence at k = 3.

12 .4 THE z-TRANSFORM 483
1 z
2[q(k)] = 1 -z -1 z- 1
If we define b = e - T, then (12 .5) becomes
1 z
Z[bk] = _
1 -bz - ' z-b
The z-transform has the following linearity property

Z[a lfl( k) + a2f2(k)] = a1Z[f 1(k)] + a2Z[f2(k)]
Using this property, we have
e ' ' - e-'~kT 1 z z
2 [sin wkT] = Z = - -
2j 2j z - e'' z - e - 'wT
z(z - e-jmr - z + e'wT)
2j(z - e j,T)(z - e - JwT )
z(e'wT - e -'wT) zsin wT
2j(z2 - (ei-T + e -j-T )z + 1) z2 - 2(cos wT)z + 1
An impulse sequence or a Kronecker sequence is defined as
1 if k=0
8(k) = (12 .6a)
t 0 if k#0
or, more generally,
8(k - n) _ {0 (12 .6b)

if k ~ n
as shown in Figure 12 .8(b). The impulse sequence 8(k - n) equals 1 at k = n and
zero elsewhere. The z-transforms of (12 .6) are, from definition,
Z[6(k)] = 1 Z[6(k - n)] = z -"
All sequences, except the impulse sequence, studied in this text will be obtained
from sampling of analog signals. For example, the sequence f(kT) = e -,IT in
Example 12.4.1 is the sampled sequence of f(t) = e -at with sampling period T. Let
F(s) be the Laplace transform of f (t) and F(z) be the z-transform of f(kT). Note that
we must use different notations to denote the Laplace transform and z-transform, or
confusion will arise . If f(kT) is the sample of f (t), then we have
F(z) = Z[f(kT)] = Z[f(t)I(=kT] = Z[[Y-'F(s)]Ir=kT] (1 2 .7a)
This is often written simply as

F(z) = Z[F(s)] (12 .7b)

Example 12 .4 .2
e-at.
Consider f(t) = Then we have
1
F(s) = -T[f(t)] = -T[e -at ] _
s + a
and
z
F(z) = Z[f(kT)] = Z[e-akT] =
z - e -aT
Thus we have
Z
z - e-aT - [s + a]
Exercise 12 .4 .1
Let f (t) = sin wt. Verify

z sin wT - w
Z
(z - e iwT)(z - e -'wT) -
(s j(o)(s + jw)
From the preceding example and exercise, we see that a pole a in F(s) is mapped
into the pole e~T in F(z) by the analog-to-digital transformation in (12.7). This prop-
erty will be further established in the next_subsection . We list in Table 12 .1 some
pairs of F(s) and F(z) . In the table, we use 6(t) to denote the impulse defined for the
continuous-time case in Appendix A and 6(k) to denote the impulse sequence defined
for the discrete-time case in (12 .6). Because 6(t) is not defined at t = 0, 6(k) is not
the sample of 6(t) . The sixth and eighth pairs of the table are obtained by using
Z [kf(k)] _ - z dF(z)
dz
For example, because Z[bk] = z/(z - b), we have
z -z . (z - b) - z = bz
Z[kb k ] _ -z d (Z b)2
dz z -b (z-b)2
12 .4 .1 The Laplace Transform and the z-Transform
The Laplace transform is defined for continuous-time signals and the z-transform is
defined for discrete-time signals . If we apply the Laplace transform directly to a
discrete-time signal f(kT ), then the result will be zero . Now we shall modify f (kT )

Table 12 .1 z-Transform Pairs .
F(s) f(t) f(kT) F(z)
1 6(t)
e- T, S(t - T )
S(kT) I
Z-
S((k - n)T)
Z
1 1
Z - 1
TZ
kT
(z - 1) 2
e -akT z
s + a z - e -aT
1 Tze - aT
to -a ` kTe -kT
(s + a)2 (z - e -aT)2
w z sin wT
sin wt sin wkT
s 2 + w2 z2 - 2z(cos (oT) + 1
S z(z - cos wT)
cos wt cos wkT
s2+ w2 z2 - 2z(cos wT) + 1
w e-at ze -aT sin wT
sin wt e -akT sin wkT
(s + a) 2 + w2 Z2 - 2ze - cT(cos wT) + e -2aT
S + a Z2 - ze - aT(cos wT)
ea t cos wt e - a kT cos wkT
(s + a) 2 + w 2 Z2 - 2ze - aT(cos wT) + e -2aT
so that the Laplace transform can be applied . Consider f (kT ), for integer k ? 0 and
positive sampling period T > 0 . We define
f*(t) : =
k=0
I f(kT)S(t - kT) (12.8)
where 8(t - kT) is the impulse defined in Appendix A . It is zero everywhere except
at t = kT. Therefore, f *(t) is zero everywhere except at sampling instants kT, where
it is an impulse with weight f(kT) . Thus, we may consider f *(t) to be a continuous-
time representation of the discrete-time sequence f (kT ) . The Laplace transform of
f *(t) is, using (A.21),
F*(s) _ `~[f*(t)] _ f(kT)E[S(t - kT)] _ f(kT)e -kTs (12.9)

k=0 k=0

If we define z = e Ts, then (12 .9) becomes
F*(s)1 Z _ eTS =
E
f(kT)z - k
k=0
Its right-hand side is the z-transform of f (kT). Thus the z-transform of f (kT) is the
Laplace transform of f*(t) with the substitution of z = e Ts or
Z[f(kT)I = .[f* (t)]I Z=eTs (12.10)
This is an important relationship between the Laplace transform and z-transform .
We now discuss the implication of
= Ts 1
z
e
or s = In z (12.11)
-
where In stands for the natural logarithm . If s = 0, then z = e0 = 1 ; that is, the
origin of the s-plane is mapped into z = 1 in the z-plane as shown in Figure 12.9.
In fact, because
Jm2rr
lT
e T l = e jm2~ = 1
for all positive and negative integer m, the points s = 0, j21T/T, -j277/T, j4i-/T,
. . . are all mapped into z = 1 . Thus, the mapping is not one-to-one . If s = jm, then
Izl = I e j&TI = 1
for all cw . This implies that the imaginary axis of the s-plane is mapped into the unit
circle on the z-plane . If s = a + jw, then
IzI = l e(a+1 TI = IeaTI I e j-T I =
e
aT
Im z
Im z
Ad Re z Re z
<
Am
Figure 12.9 Mapping between s-plane and z-plane .

Thus, a vertical line in the s-plane is mapped into the circle in the z-plane with radius
eaT. If a < 0, the vertical line is in the left half s-plane and the radius of the circle
is smaller than 1 ; if a > 0, the radius is larger than 1 . Thus the entire open left half
s-plane is mapped into the interior of the unit circle on the z-plane . To be more
specific, the strip between - ar/T and it/T shown in Figure 12.9 is mapped into the
unit circle . The upper and lower strips shown will also be mapped into the unit circle .
We call the strip between - lr/T and 1r/T the primary strip .
12 .4 .2 Inverse z-Transform
Consider a z-transform F(z) = N(z)/D(z), where N(z) and D(z) are polynomials of
z. One way to obtain the inverse z-transform of G(z) is to express G(z) as a power
series as in (12.3). This can be done by direct division . For example, consider
2z - 3
F(z) _ 3z + 2z (12 .12)
2 - 1
If we compute
13 -2 + 2327Z _3 + . . .
gz -1 - 7z
3z2 +2z- 112z-3
2z+g-gz -1
33
+ 9z -1
33 - 96 Z -1 + 13 Z -2
9z - 2
32 -1 -
z
then we have
2z - 3 2 1 13 2 + 32 z 3
F(z) = =0+ z - +
3z2 +2z- 1 3 9z 27
Thus, the inverse z-transform of F(z) is
2 13 32
f(0) = 0, f(l) = 3, f(2) _ - 9 , f(3) = 27, . . . (12 .13)
Therefore, the inverse z-transform of F(z) can be easily obtained by direct division .
The inverse z-transform can also be obtained by partial fraction expansion and
table look-up. We use the z-transform pairs in Table 12 .1 . Although the procedure
is similar to the Laplace transform case, we must make one modification . Instead of
expanding F(z), we expand F(z)/z . For example, for F(z) in (12 .12), we expand
F(z) 2z - 3 _ 2z - 3
z z(3z2 + 2z - 1) z(3z - 1)(z + 1) (12 .14)
= k1 + k2 + k3
z 3z- 1 z+ 1

with
2z - 3 -3
_-= 3
- 1)(z + 1) Z=0
-1
2
- - 3
2z - 3 3 -21
kz
z(z + 1) 1 4 4
3 3
and
2z - 3 -5 _ -5
k 3 = z(3z - 1)
Z--1 (-1)(-4) 4
We then multiply (12.14) by z to yield
z 21 z 5 z _-3 7 z 5 z
F(z)=3
Z - 4 3z- 1 4z+ 1 4 1 4z+1
Z - -
3
Therefore, the inverse z-transform of F(z) is, using Table 12 .1,
f(k) = 35(k) - k - 4 (-I)k

4 C3
fork = 0, 1, 2, 3, . . . . For example, we have
7 - 5
f(0) = 3 - = 0
4 4
7 1 5 -7 5 -7 + 15 2
f(1)=0-4 .3-- .(-1)=
12 + 4 12 3
f(2)=0-7- (1) -5=93

9 4
and so forth . The result is the same as (12 .13).
From the preceding example, we see that the procedure here is quite similar to
that in the inverse Laplace transform . The only difference is that we expand F(z)/z
in partial fraction expansion and then multiply the expansion by z . The reason
for doing this is that the z-transform pairs in Table 12.1 are mostly of the form
z/(z - b).
12.4.3 Time Delay and Time Advance
Consider the time sequence f (k) shown in Figure 12 .10(a). It is not necessarily zero
for k :5 0. Its z-transform is
F(z) _ Z[f(k)] _ I f(k)z -k (12.15)
0

f(k) f(k - 1)
111 I I -k
-2-1 01 2 3 4 5
f(k + 1)
01 2 3 4
(c)
Figure 12 .10 (a) Sequence. (b) Time delay . (c) Time advance .
It is defined only for f (k) with k ? 0, and f ( - 1), f( - 2) . . . . do not appear in F(z) .
Consider f(k - 1). It is f(k) shifted to the right or delayed by one sampling period
as shown in Figure 12 .10(b) . Its z-transform is
Z[f(k - 1)] = I0 f(k - 1)z -k = z- ' E f(k - l)z -(k-1)

k=0
which becomes, after the substitution of k = k - 1,
f(k)z-1
Z[f(k - 1)] = z - ' _~ f(k)z-k = z - ' f(-1)z +
k= -1 k=o (12.16a)
= z - '[f(-I)z + F(z)]
This has a simple physical interpretation . F(z) consists of f (k) with k ? 0 . If f (k) is
delayed by one sampling period, f(- 1) will move into k = 0 and must be included
in the z-transform of f(k - 1) . Thus we add f(- 1)z to F(z) and then delay it
(multiplying by z - ') to yield Z[f(k - 1)] . Using the same argument, we have
Z[f(k - 2)] = z -2 [f(-2)z 2 + f(-l)z + F(z)] (12.16b)
Z[f(k - 3)] = z - 3[f(-3)z 3 + f(-2)z 2 + f( - 1)z + F(z)] (12 .16c)
and so forth.
Now we consider f (k + 1) . It is the shifting of f (k) to the left (or advancing
by one sampling period) as shown in Figure 12 .10(c) . Because f (O) is moved to
k = - 1, it will not be included in the z-transform of f(k + 1), so it must be excluded
from F(z) . Thus, we subtract f (O) from F(z) and then advance it (multiplying by z),

to yield Z[f(k + 1)] or

Z[f(k + 1)] = z[F(z) - f(0)] (12 .17a)
Similarly, we have
Z[f(k + 2)] 2
= z [F(z) - f(0) - f(1)z -1] ( 12 .17b)
Z[f(k + 3)] = z3 [F(z) - f(0) - f(l)Z-t - f(2)z -2] ( 12 .17c)
and so forth .
12 .5 SOLVING LTIL DIFFERENCE EQUATIONS
Consider the LTIL difference equation in (12 .2) or

3y(k) +. 2y(k - 1) - y(k - 2) = 2u(k - 1) - 3u(k - 2) (12 .18)
As was discussed earlier, its solution can be obtained by direct substitution . The
solution, however, will not be in closed form, and it will be difficult to develop from
the solution general properties of the equation . Now we apply the z-transform to
study the equation . The equation is of second order ; therefore, the response y(k)
depends on the input u(k) and two initial conditions . To simplify discussion, we
assume that u(k) = 0 for k < 0 and that the two initial conditions are y(-1) and
y( - 2). The application of the z-transform to (12 .18) yields, using (12.16),
3Y(z) + 2z- '[Y(z) + y(- 1)z] - z -2[Y(z) + y(-1)z + y(-2)z2]
(12 .19)
= 2z - 'U(z) - 3z-2U(z)
which can be grouped as
(3 + 2z - ' - z -2)Y(z) _ [-2y(-1) + y(-1)z -1 + y( - 2)]
+ (2z -1 - 3z-2)U(z)
Thus we have
Y(z) = [-2y(-1) + y(-1)z -1 + +
(2z - ' - 3z-2)
y(-2)]
3 + 2z - ' - z -2 U(z) (12 .20a)

3 + 2z' - z -2
- [-2y(-1)z2 + y(-1)z + y(-2)z2] + (2z-3)
U(z) (12 .20b)
3z2 +2z- 1 3z2 +2z- 1
Zero-Input Response Zero-State Response
Now if y( - 2) = 1, y( - 1) _ - 2, and if u(k) is a unit-step sequence, then U(z) _
z/(z - 1) and
_ 5 2z - ' (2z - ' - 3z 2) z
Y(z)
3+2z -1 -z-2+3+2z-'-z-2 z- 1
5z2 - z + z(2z - 3) - z(5z2 - 4z - 2)
3z2 + 2z - 1 (3z - 1)(z + 1)(z - 1) (3z - 1)(z + 1)(z - 1)

12 .5 SOLVING LTIL DIFFERENCE EQUATIONS 491
To find its time response, we expand Y(z)/z as

Y(z) 5z2 - 4z - 2 19 1 9 1 1 1
z (3z - 1)(z + 1)(z - 1) 8 3z - 1 8 z + 1 4 z - 1
Thus we have
19 z 9 z 1 z
Y(z) _
8 3z - 1 + 8 z+ 1 4 z- 1
and its inverse z-transform is, using Table 12 .1,
19 1
y(k) k+ 8(-1)k-4(1)k (12 .21)
24 (3 )
for k = 0, 1, 2 We see that using the z-transform, we can obtain closed-form
solutions of LTIL difference equations .
12.5.1 Characteristic Polynomials and Transfer Functions

The response of any LTIL difference equations can be decomposed into the zero-
state response and zero-input response as shown in (12 .20). Consider the nth order
LTIL difference equation
ay(k + n) + a_ 1y(k + n - 1) +
+ a 1 y(k + 1) + a 0y(k)=b
(12 .22)
mu(k + m) + bm _ 1u(k + m - 1) +
. + b lu(k + 1) + bou(k)
We define
D(p) := apn + an l pr -1 + + a lp + a o (12 .23a)
and
N(p) := bm pm + bm l p'n - ' + . . + blp + bo (12 .23b)
where the variable p is the unit-time advance operator, defined as

py(k) : = y(k + 1) p2y(k) : = y(k + 2) p3y(k) : = y(k + 3) (12 .24)
and so forth . Using this notation, (12 .22) can be written as

D(p)y(k) = N(p)u(k) (12 .25)
In the study of the zero-input response, we assume u(k) _- 0 . Then (12.25) becomes
D(p)y(k) = 0 (12 .26)
This is the homogeneous equation . Its solution is excited exclusively by initial con-
ditions . The application of the z-transform to (12 .26) yields, as in (12 .20),
Y(z) = I(z)
D(z)

where D(z) is defined in (12 .23a) with p replaced by z and I(z) is a polynomial of z
depending on initial conditions y( - k), k = 1, 2, . . . , n. As in the continuous-time
case, we call D(z) the characteristic polynomial of (12 .25) because it governs the
free, unforced, or natural response of (12 .25). The roots of the polynomial D(s) are
called the modes. For example, if
D(z)=(z-2)(z+1)2(z+2-j3)(z+2+ j3)
then the modes are 2, -1, - 1, -2 + j3 and -2 -j3. Root 2 and complex roots
-2 j3 are simple modes and root - 1 is a repeated mode with multiplicity 2.
Thus, for any initial conditions, the zero-input response due to any initial conditions
will be of the form
y(k) = k, (2)k + k2 (-2 + j3)k + k3(-2 - j3) k + c1(-1) k + c 2k(-1)k
for k = 0, 1, 2, . . . . Thus the zero-input response is essentially determined by the
modes of the system .
Consider the difference equation in (12 .18). If all initial conditions are zero,
then the response is excited exclusively by the input and is called the zero-state
response. In the z-transform domain, the zero-state response of (12 .18) is governed
by, as computed in (12 .20),
2z - ' 3z 2 2z 3
Y(z) = -1 U(z) = : G(z)U(z) (12 .27)
+ 2z - z - ~ U(z) 3z2 + 2z - 1
where the rational function G(z) is called the discrete, digital, pulse, or sampled
transfer function or simply the transfer function . It is the ratio of the z-transforms
of the output and input when all initial conditions are zero or
Y(z) Z[Output]
G(z) = _ (12 .28)
U(z) Initial conditions-0 Z[Input] Initial conditions =0
The transfer function describes only the zero-state responses of LTIL systems .
The transfer function can easily be obtained from difference equations . For
example, if a system is described by the difference equation
D(p)y(k) = N(p)u(k)
where D(p) and N(p) are defined as in (12 .23), then its transfer function is
G(z) = N(z) (12 .29)

D(z)
Poles and zeros of G(z) are defined exactly as in the continuous-time case . For
example, given
G(z) _ N(z) - 2(z + 3)(z - 1)(z + 1)3

1) 2(z + 3)
( 12 .30)
D(z) (z - 1)(z + 2)(z + (z + 2)(z + 1) z
Its poles are - 2, - 1 and - 1 ; its zero is - 3 . Thus, if N(z) and D(z) in G(z) =
N(z)/D(z) have no common factors, then the roots of N(z) are the zeros and the roots
of D(z) are the poles of G(z) .

12 .5 SOLVING LTIL DIFFERENCE EQUATIONS 493
Consider a discrete-time system described by the difference equation

D(p)y(k) = N(p)u(k) (12.31)
with D(p) and N(p) defined in (12.23). The zero-input response of the system is
governed by the modes, the roots of the characteristic polynomial D(z). If N(p) and
D(p) have no common factors, then the set of the poles of the transfer function in
(12 .29) equals the set of the modes . In this case, the system is said to be completely
characterized by its transfer function and there is no loss of essential information in
using the transfer function to study the system . On the other hand, if D(z) and N(z)
have common factors-say, R(s)-then
N(z) N(z)R(z) N(z)
G(z)
D(z) D(z)R(z) D(z)
In this case, the poles of G(z) consist of only the roots of D(z) . The roots of R(z) are
not poles of G(z), even though they are modes of the system . Therefore, if D(z) and
N(z) have common factors, not every mode will be a pole of G(z) (G(z) is said to
have missing poles), and the system is not completely characterized by the transfer
function . In this case, we cannot disregard the zero-input response, and care must
be exercised in using the transfer function .
To conclude this section, we plot in Figure 12 .11 the time responses of some
poles . If a simple or repeated pole lies inside the unit circle of the z-plane, its time
response will approach zero as k - o. If a simple or repeated pole lies outside the
unit circle, its time response will approach infinity . The time response of a simple
pole at z = 1 is a constant ; the time response of a simple pole on the unit circle
other than z = 1 is a sustained oscillation . The time response of a repeated pole on
the unit circle will approach infinity as k --> -. In conclusion, the time response of
a simple or repeated pole approaches zero if and only if the pole lies inside the unit
circle. The time response of a pole approaches a nonzero constant if and only if the
pole is simple and is located at z = 1 .
AN,
(a) (b)
Figure 12.11 Time responses of poles.

12.5.2 Causality and Time Delay

Consider the digital transfer function
G (z) = N(z)
D(z)
with deg N(z) = m and deg D(z) = n. The transfer function is improper if m > n
and proper if n ? m. A system with an improper transfer function is called a non-
causal or an anticipatory system because the output of the system may appear before
the application of an input . For example, if G(z) = z2 /(z - 0 .5), then its unit-step
response is
z2 z
Y(z) = G(z)U(z) 1 = z + 1 .5 + 1 .75z -1 + 1.875z -2 +
= z - 0.5 z -
and is plotted in Figure 12 .12(a) . We see that the output appears at k = - 1, before
the application of the input at k = 0. Thus the system can predict what will be
applied in the future . No physical system has such capability . Therefore no physical
discrete-time system can have an improper digital transfer function .
The output of a noncausal system depends on future input . For example, if
G(z) = (z3 + 1)/(z - 0.1), then y(k) depends on past input u(m) with m < k and
future input u(k + 1) and u(k + 2). Therefore, a noncausal system cannot operate
on real time . If we store the input on a tape and start to compute y(k) after receiving
u(k + 2), then the transfer function can be used . However, in this case, we are not
using G(z), but rather G(z)/z 2 = (z 3 + 1)/z 2 (z - 0.1), which is no longer improper .
If we introduce enough delay to make an improper transfer function proper, then it
can be used to process signals . Therefore, strictly speaking, transfer functions used
in practice are all proper transfer functions.
If a system has a proper transfer function, then no output can appear before the
application of an input and the output y(k) depends on the input u(mT), with
u (k) u (k)
yin . k . .
01 2 3 4
y (k) a (k)
. .
>-k . . . . 1 >- k
-1 01 2 3 4 - 1 01 2 3 4 5 6
(a) (b)
Figure 12.12 (a) Response of noncausal system . (b) Response of causal system with r = 5 .

12.6 DISCRETE-TIME STATE EQUATIONS 495
m < k . Such systems are called causal systems . We study in the remainder of this
text only causal systems with proper digital transfer functions . Recall that in the
continuous-time case, we also study only systems with proper transfer functions .
However, the reasons are different . First, an improper analog transfer function cannot
be easily built in practice . Second, it will amplify high-frequency noise, which often
exists in analog systems . In the discrete-time case, we study proper digital transfer
functions because of causality .
Consider a proper (biproper or strictly proper) transfer function G(z) _
N(z)/D(z) . Let r = deg D(z) - deg N(z). It is the difference between the degrees
of the denominator and numerator . We call r the pole-zero excess of G(z) because
it equals the difference between the number of poles and the number of zeros . Let
y(k) be the step response of G(z). If r = 0 or G(z) is biproper, then y(O) 0 0 . If
r = 1, then y(O) = 0 and y(l) = 0 . In general, the step response of a digital transfer
function with pole-zero excess r has the property
y(O) = 0 y(l) = 0 . . . y(r - 1) = 0 y(r) 0 0
as shown in Figure 12 .12(b) . In other words, a pole-zero excess of r will introduce
a delay of r sampling instants . This phenomenon does not arise in the continuous-
time case.
12 .6 DISCRETE-TIME STATE EQUATIONS
Consider the discrete-time state-variable equation

x(k + 1) = Ax(k) + bu(k) (12.32a)
y(k) = cx(k) + du(k) (12.32b)
This is a set of algebraic equations . Therefore the solution of the equation due to
x(0) and u(k), k ? 0, can be obtained by direct substitution as
x(1) = Ax(O) + bu(O)
x(2) = Ax(1) + bu(1) = A[Ax(O) + bu(O)] + bu(1)
= A2x(O) + [Abu(0) + bu(1)]
x(3) = Ax(2) + bu(2)
= A3x(0) + [A2bx(0) + Abu(l) + bu(2)]
and, in general,
k-t
x(k) = Akx(0) + Ak-'-m bu(m) (12.33)
m=0
Response Response

The application of the z-transform to (12 .32a) yields

z[X(z) - x(0)] = AX(z) + bU(z)
which implies
(zI - A)X(z) = zx(O) + bU(z)
Thus we have
X(z) = (zI - A) -lzx(0) + (zI - A)- 'bU(z) (12 .34a)
The substitution of (12 .34a) into the z-transform of (12 .32b) yields
Y(z) = c(zI - A) -lzx(0) + [c(zl - A) -1b + d]U(z) (12 .34b)
If x(O) = 0, then (12.34b) reduces to
Y(z) = [c(zI - A) -' b + d]U(z) (12 .35)
Thus the transfer function of (12 .32) is
G(z) = Y(z) = c(zI - A) - 'b + d (12 .36)

U(z)
This is identical to the continuous-time case in (2 .75) if z is replaced by s . The
characteristic polynomial of A is defined, as in (2.76), as
0(z) = det (zI - A)
Its roots are called the eigenvalues of A .
12.6.1 Controllability and Observability

x(k + 1) = Ax(k) + bu(k) (12 .37a)
y(k) = cx(k) + du(k) (1 2 .37b)
The equation is controllable if we can transfer any state to any other state in a finite
number of sampling instants by applying an input . The equation is observable if we
can determine the initial state from the knowledge of the input and output over a
finite number of sampling instants. The discrete-time equation is controllable if and
only if the controllability matrix
U = [b Ab A2 b . . . A" - 'b] (12 .38)
has rank n . The equation is observable if and only if the observability matrix
V = (12 .39)

12 .7 BASIC BLOCK DIAGRAMS AND REALIZATIONS 497
has rank n. These conditions are identical to the continuous-time case . We prove the
controllability part in the following. We write (12 .33) explicitly fork = n as
n-1
x(n) - AN(0) = An-l- mbu(m)
E
M=0
= bu(n - 1) + Abu(n - 2) + A2 bu(n - 3) (12.40)

+ + An - 'bu(0)
= [b Ab A%b . . . An - 'b]
u(0) -
For any x(0) and x(n), a solution u(k), k = 0, 1, . . . , n - 1, exists in (12 .40) if and
only if the matrix U has rank n (Theorem B.1). This completes the proof . If an
equation is controllable, then the transfer of a state to any other state can be achieved
in n sampling periods and the input sequence can be computed from (12 .40). Thus,
the discrete-time case is considerably simpler than the continuous-time case . The
observability part can be similarly established . See Problem 12 .13 .
If a state-variable equation is controllable and observable, then the equation is
said to be a minimal equation . In this case, if we write
N(z)
c(zI - A) - 'b + d = :
D(z)
with
D(z) = 0(z) = det (zI - A)
then there are no common factors between N(z) and D(z), and the set of the eigen-
values of A [the roots of the characteristic polynomial A (z)] equals the set of the
poles of the transfer function . This situation is identical to the continuous-time case .
12 .7 BASIC BLOCK DIAGRAMS AND REALIZATIONS
Every discrete-time state-variable equation can be easily built using the three ele-
ments shown in Figure 12.13 . They are called multipliers, summers or adders, and
unit-delay elements . The gain a of a multiplier can be positive or negative, larger or
x
x(k) ax(k) +x 2 +x 3 x(k+ 1) I x (k)
> x2 -~ z
x3
Figure 12 .13 Three discrete basic elements .

z (k)
0
Figure 12 .14 Basic block diagram of (12 .41) .
smaller than 1 . An adder has two or more inputs and one and only one output . The
output is simply the sum of all inputs . If the output of the unit-delay element is x(k),
then the input is x(k + 1) . A unit-delay element will be denoted by z -1. These
elements are quite similar to those in Figure 5 .3. A block diagram which consists of
only these three types of elements is called a basic block diagram .
We use an example to illustrate how to draw a basic block diagram for a discrete-
time state-variable equation . Consider
x k+ 1 _ 2 -0.3 x [- 2]
u(k) (12 .41 a)
2 (k + 1)] [0 -8 ] [x2(k)]
kx + 0
y(k) _ [-2 31 [ xl(k) I

+ 5u(k) (12 .41b)
x2(k)
It has dimension 2 and therefore needs two unit-delay elements . The outputs of the
two elements are assigned as x1 (k) and x2(k), as shown in Figure 12.14. Their inputs
are x1 (k + 1) and x2 (k + 1). A basic block diagram of (12 .41) can be obtained as
shown in Figure 12.14 . The procedure of developing the diagram is the same as the
one in developing Figure 5 .5 . These two diagrams are in fact identical if integrators
are replaced by delay elements or every 1/s is replaced by 1/z .
Consider a transfer function G(z) . If we can find a state-variable equation
x(k + 1) = Ax(k) + bu(k) (12 .42a)
y(k) = cx(k) + du(k) (12 .42b)
such that its transfer function from u to y equals G(z), or

G(z) = c(zI - A) -1 b + d (12 .43)
then G(z) is said to be realizable and (12 .42) is a realization of G(z). As in the
continuous-time case, G(z) is realizable if and only if G(z) is a proper rational
function . If G(z) is improper, then any realization of G(z) must be of the form

12 .7 BASIC BLOCK DIAGRAMS AND REALIZATIONS 499
x(k + 1) = Ax(k) + bu(k) (12 .44a)
y(k) = c x(k) + du(k) + d,u(k + 1) + d 2u(k + 2) + (12 .44b)
This means that y(k) will depend on the future input u(k + 1) with 1 ? 1 and the
system is not causal . Thus, we study only proper rational G(z).
12.7.1 Realizations of N(z)/D(z)

Instead of discussing the general case, we use a transfer function of degree 4 to
illustrate the realization procedure . The realization procedure is identical to the
continuous-time case ; therefore, we state only the result . Consider
Y(z) _ b4z4 + b3z 3 + b2z2 + b,z + bo - . N(z)
G z = (12 .45)
U(z) a4z4 + a3z3 + a2z2 + a,z + ao D(z)
where a; and b; are real constants and a4 =A 0 . The transfer function is biproper if
b4 0 0, and strictly proper if b4 = 0. First, we express (12 .45) as
Y(z) b3 z 3 + b2z2 + b1 z_+TO N(z)
G(z) _ = G(oo) + d + (12 .46)
z4 + a 3z3 + a2Z
2 + a z + ao
U(z) 1 D(z)
Then the following state-variable equation, similar to (5.17),
-a 3 l -ao
-a
1 0 0 0
x(k) + u(k) (12 .47a)
0 1 0 0
.0 0 1 0
y(k) = [b 3 b2 b, bo ]x(k) + du(k) (12 .47b)
with d = G(oo), is a realization of (5 .12). The value of G(oo) yields the direct
transmission part . If G(z) is strictly proper, then d = 0. Equation (12.47) is always
controllable and is therefore called the controllable form realization . If N(z) and
D(z) in (12 .45) have no common factors, then (12 .46) is observable as well and the
equation is called a minimal equation . Otherwise, the equation is not observable.
The following equation, which is similar to (5.18),
- a3 1 0 0 b3
-a2 0 1 0 x(k) + b2 u(k)
x(k + 1) = _ (12 .48a)
-a, 0 0 i
- ao 0 0 0 bo
y(k) = [1 0 0 O]x(k) + du(k) (12 .48b)
is a different realization of (12 .45). The equation is observable whether or not N(z)
and D(z) have common factors . IfN(z) and D(z) have no common factors, the equa-
tion is controllable as well and is a minimal equation . Equation (12 .48) is called the
observable form realization .

Exercise 12 .7 .1
Find controllable- and observable-form realizations of

2z + 10
a.
z + 2
3z2 -z+2
b. z
3 + 2z2 + 1
4z3 + 2z + 1
C.
2z3 + 3z2 + 2
5
d
. 2z2 + 4z + 3
The tandem and parallel realizations discussed in Section 5 .5.2 can again be
applied directly to discrete transfer functions, and the discussion will not be repeated.
To conclude this section, we mention that the same command tf2ss in MATLAB
can be used to realize analog transfer functions and digital transfer functions . For
example, if
3z2 -z+2 0+3z-1 -z -2 +2z -3
G(z) = (12.49)
Z3 + 2Z2 + 1 1+ 2z-1 + 0 . Z-2 + z -3
then the following
num=[3 -1 2] ;den=[1 2 0 1] ;
[a,b,c,d] =tf2ss(num,den)
will generate its controllable-form realization . The command to compute the re-
sponse of G(s) due to a unit-step function is "step" ; and G(s) is expressed in de-
scending powers of s. The command to compute the response of G(z) due to a unit-
step sequence is "dstep" . Furthermore, G(z) must be expressed in ascending powers
of z -1. Therefore,
num=[0 3 -1 2] ;den=[1 2 0 1] ;
y = dstep(num,den,20) ;
plot(y)
will generate 20 points of the unit-step response of (12 .49).
12 .8 STABILITY
A discrete-time system is said to be bounded-input, bounded-output stable, or simply

stable, if every bounded input sequence excites a bounded output sequence . The
condition for a system with digital transfer function G(z) to be stable is that every

12 .8 STABILITY 50 1
Table 12 .2 The Jury Test
ao a, a 2 a3 a4
a4 a 3 a2 a, ao ki a4
a
o
b, b2 b 3 0 (1st a row) - ki ( 2nd a row)

b3
b 3 b 2 b, bo k
z = b0
C2 0 (1st brow) - k2(2nd brow)
- C2
C2 C1 CO k3 -
Co
d, 0 (1st crow) - k3 (2nd crow)

d,
d, do k4 = d0
(1st d row) - k4 (2nd d row)
pole of G(z) must lie inside the unit circle of the z-plane or have a magnitude less
than 1 . This condition can be deduced from the continuous-time case, where stability
requires every pole to lie inside the open left half s-plane . Because z = es T maps
the open left half s-plane into the interior of the unit circle in the z-plane, discrete
stability requires every pole to lie inside the unit circle of the z-plane .
In the continuous-time case, we can use the Routh test to check whether all
roots of a polynomial lie inside the open left half s-plane . In the discrete-time case,
we have a similar test, called the Jury test . Consider the polynomial
D(z) = a o z4 + a,z3 + a2 Z2 + a 3z + a4 ao > 0 (12.50)
It is a polynomial of degree 4 with a positive leading coefficient . We form the table
in Table 12.2. The first row is simply the coefficients of D(z) arranged in the de-
scending power of z . The second row is the reversal of the order of the coefficients
in the first row . We then take the ratio k, of the last elements of the first two rows
as shown in the table . The subtraction of the product of k, and the second row from
the first row yields the first b row . The last element of the b row will be zero and
will be disregarded in the subsequent development . We reverse the order of the
coefficients of the b row and repeat the process as shown in Table 12 .2. If D(z) is
of degree n, then the table consists of 2n + 1 rows .
The Jury Test

All roots of the polynomial of degree 4 and with a positive leading coefficient
in (12 .50) lie inside the unit circle if and only if the four leading coefficients
(bo, co , do, eo) in Table 12 .2 are all positive .

Although this test is stated for a polynomial of degree 4, it can be easily extended
to the general case . We use an example to illustrate its application .
Example 12 .8 .1
Consider a system with transfer function
(z - 2)(z + 10)
G (z) _ (12.51)
z3 - O.IZ2 - 0.122 - 0.4
To check its stability, we use the denominator to form the table
1 -0 .1 -0.12 -0.4
-0.4 -0.12 -0.1 1 k, = - 0.4/1 = - 0.4
0.84} -0.148 -0.16 0 (1st row) + 0 .4(2nd row)
-0.16 -0.148 0.84 k2 = -0.16/0 .84 = -0 .19
,0.8096 ; -0.176 0 (3rd row) + 0.19(4th row)
-0.176 0.8096 k3 = - 0.176/0 .8096 = -0 .217
0.771 ; (5th row) + 0 .217(6th row)
The three leading coefficients 0 .84, 0.8096, and 0.771 are all positive ; thus, all roots
of the denominator of G(z) lie inside the unit circle . Thus the system is stable .
12.8 .1 The Final-Value and Initial-Value Theorems

Let F(z) be the z-transform of f (k) and be a proper rational function. If f (k) ap-
proaches a constant, zero or nonzero, then the constant can be obtained as
lim f (k) = lim (z - 1)F(z) (12.52)
k-
.ao z-1
This is called the final-value theorem . The theorem holds only if f (k) approaches a
constant . For example, if f(k) = 2k, then F(z) = z/(z - 2) . For this z-transform
pair, we have ft-) = -, but
z
lim (z - 1) . = 0 . (-1) = 0
z-1 z-2
Thus (12.52) does not hold. The condition for f(k) to approach a constant is that
(z - 1)F(z) is stable or, equivalently, all poles of (z - 1)F(z) lie inside the unit

12 .9 STEADY-STATE RESPONSES OF STABLE SYSTEMS 503
circle. This implies that all poles of F(z), except for a possible simple pole at z =
1, must lie inside the unit circle . As discussed in Figure 12.11, if all poles of F(z)
lie inside the unit circle, then the time response will approach zero . In this case, F(z)
has no pole (z - 1) to cancel the factor (z - 1), and the right-hand side of (12 .52)
is zero . If F(z) has one pole at z = 1 and remaining poles inside the unit circle such
as
N(z)
F(z) =
(z - 1)(z - a)(z - b)
then it can be expanded, using partial fraction expansion, as
z z z
+k2 z-a +k3 z-b
F(z) =klz- 1
Z - 1
with k1 = lim F(z) = lim (z - 1) F(z) . The inverse z-transform of F(z) is
Z-1 Z Z-1
f(k) = k 1 (1)k + k2ak + k3bk

which approaches k1 as k-c because jal < 1 and JbI < 1 . This establishes intuitively
(12 .52). For a different proof, see Reference [18] .
Let F(z) be the z-transform of f (k) and be a proper rational function . Then we
have
F(z) = f(O) + f(1 )z -1 + f(2)z -2 + f(3)z -3 + . . .
which implies
f (0) = lim F(z) (12.53)
z-
This is called the initial-value theorem .
12 .9 STEADY STATE RESPONSES OF STABLE SYSTEMS
Consider a discrete-time system with transfer function G(z). The response of G(z)
as k - - is called the steady-state response of the system. If the system is stable,
then the steady-state response of a step sequence will be a step sequence, not nec-
essarily of the same magnitude . The steady-state response of a ramp sequence will
be a ramp sequence ; the steady-state response of a sinusoidal sequence will be a
sinusoidal sequence with the same frequency . We establish these in the following .
Consider a system with discrete transfer function G(z) . Let the input be a step
sequence with magnitude a-that is, u(k) = a, for k = 0, 1, 2, . . . . Then U(z) _
az/(z - 1) and the output y(k) is given by
az
Y(z) = G (z) U(z) = G (z)
.z-1

To find the time response of Y(z), we expand, using partial fraction expansion,
Yzz) = aG(z) = aG(l)
+ (Terms due to poles of G(z))
which implies
z
Y(z) = aG(l) + (Terms due to poles of G(z))
z - 1
If G(z) is stable, then every pole of G(z) lies inside the unit circle of the z-plane and
its time response approaches zero as k -> c . Thus we have
ys (k) : = lim y(k) = aG(1)(1) k = aG(1) (12.54)
k--
Thus, the steady-state response of a stable system with transfer function G(z) due to
a unit-step sequence equals G(1) . This is similar to (4 .25) in the continuous-time
case . Equation (12'.54) can also be obtained by applying the final-value theorem . In
order for the final-value theorem to be applicable, the poles of
aG(z)z
(z - 1) Y z = (z - 1) z - 1 = azG(z)
must all lie inside the unit circle. This is the case because G(z) is stable by assump-
tion. Thus we have
y,, (k) : = lim y(k) = lim (z - 1) Y(z) = lim azG (z) = aG (1)
k-- z-->1 z-1
This once again establishes (12 .54).
Example 12 .9.1
G(z) - (z - 2)(z + 10)
z3 - O.1z2 - 0.12z - 0.4
It is stable as shown in Example 12 .8.1. If u(k) = 1, then the steady-state output is

(1- 2)(1 + 10) - -11
YS(k) = G(1) = = -28 .95
1 - 0.1 - 0.12 - 0.4 0.38
If u(k) = akT, for k = 0, 1, 2, . . . , a ramp sequence, then

aTz
U(z) =
(z - 1)2

12 .9 STEADY-STATE RESPONSES OF STABLE SYSTEMS 505
and
aTz
Y(z) = G(z)U(z) = G(z) (z
- 1)2
We expand, using (A.8) in Appendix A,
Y(z) _ aTG(z) _ aTG(1) + aTG'(1)
+ Terms due to poles of G z
z (z - 1)2 (z - 1)2 z - 1 ( ( ))
Tz
Y(z) = aG(1) 2 + aTG'(1) z
(z - 1) z - 1 (12 .55)
+ (Terms due to poles of G(z))
where
dG(z)
dz z=1
Thus we conclude from (12.55) that if G(z) is stable, then

ys (k) = aG(1)kT + aTG'(1)
This equation is similar to (4 .26a) in the continuous-time case . In the continuous-

time case, the equation can also be expressed succinctly in terms of the coefficients
of G(s) as in (4.26b) . This is, however, not possible in the discrete-time case .
If G(z) is stable and if u(k) = a sin kw0 T, then we have
ys (k) = aA(wo ) sin (kw0 T + 0((w 0 )) (12 .56)
with
A(coo) = IG(eJ'o T)I and 0(coo) = IG(eJ',,T) ( 12 .57)
In other words, if G (z) is stable, its steady-state response due to a sinusoidal sequence
approaches a sinusoidal sequence with the same frequency ; its amplitude is modified
by A((oo ) and its phase by 0(w0 ). The derivation of (12 .56) is similar to (4 .32) and
will not be repeated .
The plot of G(e',T ) with respect to w is called the frequency response of the
discrete-time system . The plot of its amplitude A(w) is called the amplitude char-
acteristic and the plot of 0(w), the phase characteristic. Because
z-7l
+ T r
ej = e 1-Tej2- = e 1wT
e',T is periodic with period 27T/T . Consequently, so are G(e',T), A(w), and 0(w).
Therefore, we plot A(w) and 0((o) only for w from - it/T to 7r/T . If all coefficients
of G(z) are real, as is always the case in practice, then A(w) is symmetric and 0((0)
is antisymmetric with respect to w as shown in Figure 12 .15. Therefore, we usually
plot A(w) and 0(w) only for w from 0 to rr/T or, equivalently, we plot G(z) only
along the upper circumference of the unit circle on the z-plane .

A(co) 6((0)
U) - CU
n 0 it Jr n
T T T
(a) (b)
Figure 12 .15 (a) Symmetric of A(w). (b) Antisymmetric of 6(w) .
12 .9 .1 Frequency Responses of Analog and Digital Systems

Consider a stable analog system with transfer function G(s). Let g(t) be the inverse
Laplace transform of G(s). The time function g(t) is called the impulse response of
the system . Let G(z) denote the z-transform of g(kT), the sample of g(t) with sam-
pling period T-that is,
G(z) = Z[g(kT )] = Z[[-T - ' G(s)] t =kT]
(12 .58a)
For convenience, we write this simply as

G(z) = Z[G(s)] (12 .58b)
This is the transfer function of the discrete system whose impulse response equals
the sample of the impulse response of the analog system . Now we discuss the re-
lationship between the frequency response G(j(o) of the analog system and the fre-
quency response G(e'l T) of the corresponding discrete system . It turns out that they
are related by
G(e'"T) _ 1 7r
G (~ (w - k 2 ~~ 1 G(1(w - kwS)) (12 .59)
T k=-~ T T k=-~
where co, = 21r/T is called the sampling frequency . See Reference [18, p . 371 ; 13,
p. 71 .]. G(j(w - w,)) is the shifting of G(j(o) to w, and G(j(w + w,)) is the shifting
of G(jw) to - w, . Thus, except for the factor 1 /T, G(ej" T) is the sum of repetitions
of G(jw) at kw, for all integers k. For example, if G(jw) is as shown in Figure
12.16(a), then the sum will be as shown in Figure 12 .16(b) . Note that the factor
11T is not included_in the sum, thus the vertical coordinate of Figure 12 .16(b) is
TG(e'WT). The plot G(j(o) in Figure 12 .16(a) is zero for 1w1 ? 1r/T, and its repetitions
do not overlap with each other . In this case, sampling does not introduce aliasing
and we have
G(j(o) = TG(ei -T ) for Iwl ar/T (12 .60)
The plot G(jw) in Figure 12 .16(c) is not zero for 1wl ? IT/T, and its repetitions do
overlap with each other as shown in Figure 12 .16(d) . In this case, the sampling is
said to cause aliasing and (12.60) does not hold . However, if the sampling period T
is chosen to be sufficiently small, we have
G(jw) -~- TG(ej(T ) for wl : TT/T (12 .61)

G(j(0) TG(ejwT)
7r
11 It
T
(a)
G(jw) TG(ej wT)
(0
Ir 0 n Jr 0 jr 2ir
T T T T T
(c) (d)
Figure 12 .16 Frequency responses of analog and digital systems .
In conclusion, it is possible to obtain a discrete-time system with frequency response

as close as desired to the frequency response of an analog system by choosing a
sufficiently small sampling period or a sufficiently large sampling frequency .
12.10 LYAPUNOV STABILITY THEOREM
In this section, we study the stability of

x(k + 1) = Ax(k) (12 .62)
If all eigenvalues of A lie inside the unit circle, then A is said to be stable. If we
compute its characteristic polynomial
0(z) = det (zI - A) (12 .63)
then the stability of A can be determined by applying the Jury test . Another way of
checking the stability of A is applying the following theorem .
THEOREM 12 .1 (Lyapunov Theorem)

All eigenvalues of A have magnitude less than 1 if and only if for any given
positive definite symmetric matrix N or any given positive semidefinite symmetric
matrix N with the property that (A, N) is observable, the matrix equation
A'MA - M = -N (12 .64)
has a unique symmetric solution M and M is positive definite .
To prove this theorem, we define the Lyapunov function

V(x(k)) = x'(k)Mx(k) (12 .65)
We then use (12 .62) to compute

VV(x(k)) := V(x(k + 1)) - V(x(k)) = x'(k + 1)Mx(k + 1) - x(k)Mx(k)
= x'(k)A'MAx(k) - x'(k)Mx(k) = x'(k)[A'MA - M]x(k)
_ -x'(k)Nx(k)
where we have substituted (12 .64). The rest of the proof is similar to the continuous-
time case in Section 11 .8 and will not be repeated . We call (12 .64) the discrete
Lyapunov equation . The theorem can be used to establish the Jury test just as the
continuous-time Lyapunov theorem can be used to establish the Routh test . The
proof, however, is less transparent than in the continuous-time case . See Reference
[15, p . 421] .
PROBLEMS
12 .1 . Find the z-transforms of the following sequences, for k = 0, 1, 2, . . . ,

a. 38(k - 3) + (-2)k
b. sin 2k + e -o.2k
c. k(0.2)k + (0.2)k
12.2 . Find the z-transforms of the sequences obtained from sampling the following
continuous-time signals with sampling period T = 0 .1 :
a. e -02t sin 3t + cos 3t
b. teo.1 t
12 .3. Use the direct division method and partial fraction method to find the inverse
z-transforms of
z - 10
0.
(z + 1)(z - 0.1)
z
b. (z + 0
.2)(z - 0.3)
1
C.
z3 (Z - 0 .5)
12 .4 . Find the solution of the difference equation

y(k) + y(k - 1) - 2y(k - 2) = u(k - 1) + 3u(k - 2)
due to the initial conditions y( - 1) = 2, y(-2) = 1, and the unit-step input
sequence.
12.5 . Repeat Problem 12.4 for the difference equation
y(k + 2) + y(k + 1) - 2y(k) = u(k + 1) + 3u(k)
Is the result the same as the one in Problem 12.4?

PROBLEMS 509
12.6. Find the solution of the difference equation

y(k) + y(k - 1) - 2y(k - 2) = u(k - 1) + 3u(k - 2)
due to zero initial conditions (that is, y( - 1) = 0, y( - 2) = 0) and the unit-
step input sequence. This is called the unit-step response .
12.7 . Repeat Problem 12 .6 for the difference equation
y(k + 2) + y(k + 1) - 2y(k) = u(k + 1) + 3u(k)
Is the result the same as the one in Problem 12 .6?
12.8 . Find the unit-step response of a system with transfer function
z2 -z+ 1
G (z) _
z + 0 .9
Will the response appear before the application of the input? A system with
improper transfer function is a noncausal system . The output y(k) of such a
system depends on u(l) with l ? k-that is, present output depends on future
input.
12 .9 . Consider
[x2 (k + 1)] -
[1 1] x(k) + ~ 0i u(k)
y(k) _ [2 1]x(k)
Compute its transfer function .
12.10 . Consider
x l (k + 1) 0 1 0 0
x2(k + 1) = 0 0 1 x(k) + 0 u(k)
x 3(k + 1) 1 1 0 1 11
1
y(k) = [2 1 1]x(k)
Compute its transfer function .
12 .11 . Is the equation in Problem 12 .9 controllable? observable?
12 .12 . Is the equation in Problem 12 .10 controllable? observable?
12 .13 . Consider (12 .32) with d = 0 . Show
y(0) c
y(l) - cbu(0) cA
x(0)
y(n - 1) - cAi -2 bu(0) - - cbu(n - 2) cA" - '

Use this equation to establish the observability condition of (12.32) .

12 .14 . Draw basic block diagrams for the equations in Problems 12 .9 and 12.10.
12 .15. Find realizations for the following transfer functions
z 2 +2
a. 4z3
3z4 + 1
b. 2z4 +3z3 +4z 2 +z+ 1
(z + 3)2
C. (z + 1)2 (z + 2)
12 .16. Are the transfer functions in Problems 12 .15 stable?
12 .17 . Plot the frequency responses of
1 z
G(s) = s + 1 G(z) =
for T = 1 and T = 0.1.

12.18 . Check the stability of
.2 0.5
A = [0
1 0
using the Jury test and Lyapunov Theorem .
Discrete- Tivne
System Design
13.1 INTRODUCTION
Plants of control systems are mostly analog systems . However, because digital com-
pensators have many advantages over analog ones, we may be asked to design digital
compensators to control analog plants . In this chapter, we study the design of such
compensators . There are two approaches to carrying out the design . The first ap-
proach uses the design methods discussed in the preceding chapters to design an
analog compensator and then transform it into a digital one . The second approach
first transforms analog plants into digital plants and then carries out design using
digital techniques . The first approach performs discretization after design ; the second
approach performs discretization before design . We discuss the two approaches in
order.
In this chapter, we encounter both analog and digital systems . To differentiate
them, we use variables with an overbar to denote analog systems or signals and
variables without an overbar to denote digital systems or signals . For example,
G(s) is an analog transfer function and G(z) is a digital transfer function; y(t) is an
analog output and y(kT) is a digital output . However, if y(kT) is a sample of y(t),
then y(kT) = y(kT) and the overbar will be dropped . If the same input is applied to
an analog and a digital system, then we use u(t) and u(kT) to denote the inputs ; no
overbar will be used .
511

512 CHAPTER 13 DISCRETE-TIME SYSTEM DESIGN
13 .2 DIGITAL IMPLEMENTATIONS OF ANALOG

COMPENSATORS-TIME-DOMAIN INVARIANCE
Consider the analog compensator with proper transfer function C(s) shown in Figure
13.1(a). The arrangement in Figure 13 .1(b) implements the analog compensator digi-
tally. It consists of three parts : an A/D converter, a digital system or an algorithm,
and a D/A converter . The problem is to find a digital system such that for any input
e(t), the output u(t) of the analog compensator and the output u(t) of the digital
compensator are roughly equal . From Figure 13 .1(b), we see that the output of the
A/D converter equals e(kT ), the sample of e(t) with sampling period T. We then
search for a digital system which operates on e(kT) to yield a sequence u(kT). The
D/A converter then holds the value of u constant until the arrival of next data . Thus
the output u(t) of the digital compensator is stepwise as shown . The output of the
analog compensator is generally not stepwise ; therefore, the best we can achieve is
that u(t) approximately equals u(t).
In designing a digital system, ideally, for any input e(t), u(kT) in Figure 13 .1(b)
should equal the sample of u(t) . It is difficult, if not impossible, to design such a
digital compensator that holds for all e(t). It is, however, quite simple to design such
a digital compensator for specific e(t) . In this section, we design such compensators
for e(t) to be an impulse and a step function .
Impulse-Invariance Method
Consider an _analog compensator with a strictly proper transfer function Qs). If
the input of CS (s) is an impulse (its Laplace transform is 1), then the output is
U(s) = CS (S) 1 = CS (S)
Its inverse Laplace transform is actually the impulse response of the analog com-
pensator . The z-transform of the sample of the impulse response yields a digital
compensator with discrete transfer function
C(z) _ [~- t[cS(s)]It=k1]
or, using the notation in (12 .7),
C(z) = 2[CS (s)] (13.1)
u (t) u (t)
loc
-t 3-t
v A I
e (t) it (t) e(t) e (kT) Digital u (kT) u (t)
C(s) > -) A/D D/A
system
(a) (b)
Figure 13.1 (a) Analog compensator . (b) Digital compensator .

13 .2 DIGITAL IMPLEMENTATIONS OF ANALOG COMPENSATORS 513
As discussed in Section 12 .9.1, if the sampling period is very small and the aliasing
is negligible, then the frequency responses of C s(s) and C(z) will be of the same
form but differ by the factor 1/T in the frequency range Im1 < it/T . To take care of
this factor, we introduce T into (13 .1) to yield
C(z) = TZ[CS(s)]
This yields an impulse-invariant digital compensator for a strictly proper C s(s). If
C(s) is biproper, then C(s) can be decomposed as
C(s) = k + C s(s)
The inverse Laplace transform of k is H(t) . The value of 6(t) is not defined at
t = 0 ; therefore, its sample is meaningless . If we require the frequency response of
C(z) to equal the frequency response of k, then_ C(z) is simply k. Thus the impulse-
invariant digital compensator of C(s) = k + Cs(s) is
C(z) = k + TZ[CS(s)] (13 .2)
Note that the poles of C(z) are obtained from the poles of C(s) or C s(s) by the
transformation z = es T which maps the open left half s-plane into the interior of the
unit circle of the z-plane ; therefore, if all poles of C(s) lie inside the open left half
s-plane, then all poles of C(z) will lie inside the unit circle on the z-plane . Thus, if
C(s) is stable, so is C(z) .
Example 13.2.1
Consider an analog compensator with transfer function
= 2(s - 2)
C(s) (13 .3)
(s + 1)(s + 3)
To find its impulse response, we expand it, using partial fraction expansion, as
5 3
C(s)=
s+3 S + 1
Thus, using (13 .2) and Table 12 .1, the impulse-invariant digital compensator is
C(z)=T 5 z _ T1 (13 .4)

C z - z3T-3
e- z - e
The compensator depends on the sampling period . Different sampling periods yield
different digital compensators . For example, if T = 0.5, then (13 .4) becomes

C(z) = 0 .5
5z - 3z 1 0 .5z(2z - 2 .366)
(13 .5)
Cz - 0.223 z - 0.607 J (z - 0.223)(z - 0.607)
It is a biproper digital transfer function .

Step-Invariance Method
Consider an analog compensator with transfer function C(s) . We now develop a
digital compensator C(z) whose step response equals the samples of the step response
of C(s) . The Laplace transform of a unit-step function is 1/s ; the z-transform of a
unit-step sequence is z/(z - 1) . Thus, the step responses of both systems in the
transform domains are
C(s)
S
1
and C(z)
z -
z
1
If they are equal at sampling instants, then
Z-' C(z) . z = è-' C(s) ~
C z 1J C s t=kT
which, using the notation of (12 .7), implies

z - 1 C(s)
C(z) =
z
[C(s)]
s = (1 - z-1)Z
Cs J
(13 .6)
This is called the step-invariant digital compensator of C(s) .
Example 13 .2 .2
Find the step-invariant digital compensator of the analog compensator in (13 .3). We
first compute
C(s) - 2(s - 2) - 4 + 3 5
s s(s + 1)(s + 3) 3s s + I 3(s + 3)
Using Table 12 .1, we have
C(s) 4z 3z 5z
Z s 3(z - 1) + z - e-T 3(z - e -3T)
which becomes, after lengthy manipulation,
= z[(9e -T - 5e- 3T - 4)z - (4e -4T - 9e -3T + 5e -T )l

Z C(s)
s 3(z - 1)(z - e-T )(z - e- 3T)
Thus the step-invariant digital compensator of (13 .1) is
(9e -T - 5e-3T - 4)z - (4e-4T - 9e-3T + 5e-T)
C(z) = ( 13 .7 )
3(z - e -T )(z - e -3T)
If the sampling period is 0 .5, then the compensator is
. 0.607 - 5 . 0.223 - 4)z - (4 . 0.135 - 9 . 0.223 + 5 . 0.607)
C(z) = (9
3(z - 0.607)(z - 0.223)
0.116z - 0.523 (13 .8)
z2 - 0.830z + 0 .135

This is a strictly proper transfer function . Although (13.5) and (13 .8) have the same
set of poles, their numerators are quite different . Thus the impulse-invariance and
step-invariance methods implement a same analog compensator differently.
As can be seen from this example that the z-transform of C(s)/s will introduce
an unstable pole at 1, which, however, will be cancelled by (1 - z - '). Thus the
poles of C(z) in (13 .6) consist of only the transformations of the poles of C(s) by
z = esT. Thus if C(s) is stable, so is the step-invariant digital compensator.
The step-invariant digital compensator of C(s) can also be obtained using state-
variable equations . Let
X(t) = Ax(t) + be(t) (13 .9a)
u(t) = cx(t) + de(t) (13 .9b)
be a realization of C(s) . Note that the input of the analog compensator is e(t) and
the output is u(t) . If the input is stepwise as shown in Figure 2 .23(a), then the
continuous-time state-variable equation in (13 .9) can be described by, as derived in
(2 .89),
x(k + 1) = Ax(k) + be(k) (13 .1Oa)
u(k) = cx(k) + de(k) (13 .1Ob)
with
T
A=eAT b= f0 eATdr b c=c d = d (13 .1Oc)
The output u(k) of (13 .10) equals the sample of (13 .9) if the input e(t) is stepwise .
Because a unit-step function is stepwise, the discrete-time state-variable equation in
(13 .10) describes the step-invariant digital compensator . The discrete transfer func-
tion of the compensator is
G(z)=c(zI-A)-'b+d (13 .11)
This is an alternative way of computing step-invariant digital compensators . This

compensator can easily be obtained using MATLAB, as is illustrated by the follow-
ing example .
Example 13 .2.3
Find the step-invariant digital comp ensator for the analog compensator in (13 .3).
The controllable-form realization of C(s) = (2s - 4)/(s 2 + 4s + 3) is
-4
x(t) = x(t) + e(t) (13 .12o)
0] L ]
u(t) = [2 - 41 x(t) (13 .12b)

This can also be obtained on MATLAB by typing

nu=[2 -4] ;de=[1 4 3] ;
[a,b,c,d] = tf2ss(nu,de)
Next we discretize (13 .12a) with sampling period 0.5. The command
[da,db] = c2d(a,b,0 .5)
will yield
0.0314 -0.5751 0.1917
da = db =
0.1917 0.7982 0 .0673
where da and db denote discrete a and b. See Section 5 .3. Thus, the step-invariant
digital compensator is given by
1 0.0314 -0.5751 [0.1917 ,I
x(k +, 1) = x(k) + e(k) ( 13 .13a)
0.1917 0.7982 ] 0.0673 J
u(k) = [2 -4]x(k) (13 .13b)
To compute its transfer function, we type

[num,den] = ss2tf (da,db,c,d, 1)
Then MATLAB will yield
0.1144z - 0.5219
C(z) (13 .14)
z2 - 0.8297z + 0 .1353
This transfer function is the same as (13 .8), other than the discrepancy due to trun-
cation errors . Therefore, step-invariant digital compensators can be obtained using
either transfer functions or state-variable equations . In actual implementation, digital
transfer functions must be realized as state-variable equations . Therefore, in using
state-variable equations, we may stop after obtaining (13 .13). There is no need to
compute its transfer function .
13.2.1 Frequency-Domain Transformations

In addition to the time-domain invariance methods discussed in the preceding sec-
tion, there are many other methods of implementing analog systems digitally . These
methods will be obtained by transformations between s and z ; therefore, they are
grouped under the heading of frequency-domain transforma tions .
Consider an analog compensator with transfer function C(s) . Let
x(t) = Ax(t) + be(t) (13 .15a)
u(t) = cx(t) + de(t) (13 .15b)
be its realization . In discretization of C(s) or (13 .15), we are approximating an in-

tegration by a summation. Different approximations yield different discretizations

and, consequently, different digital compensators . For convenience of discussion,

we assume e(t) = 0 and A scalar .
Forward Approximation
The integration of (13 .15a) from to to to + T yields, with e(t) = 0,
rto +T dx(t) td +T
dt = x(to + T) - x(to) = ~ Ax(t)dt (13 .16)

Jto to
Let Ax(t) be as shown in Figure 13 .2. If the integration is approximated by the
shaded area shown in Figure 13 .2(a), then (13.16) becomes
x(to + T) - x(to) = Ax(to )T
or
x(tp + T) - x(to)
= Ax(t o ) ( 13 .17)
T
This is the same as approximating the differentiation in (13 .15a) by
x(t + T) - x(t)
X(t) (13 .18)
Because -T[i] = sX(s) and

x(t + T) - x(t) - zX(z) - X(z) = z - 1
X(z)
T T T
in the transform domains, Equation (13.18) is equivalent to
s = z T I (Forward difference) (13 .19)
Using this transformation, an analog compensator can easily be changed into a digital
compensator. This is called the forward-difference or Euler's method. This trans-
formation may not preserve the stability of C(s) . For example, if C(s) = 1/(s + 2),
then
1 T
C(z) =
z- 1 z- 1 +2T
+ 2
T
Ax(t) Ax(t) Ax(t)
I I
I- t I t I- t
to to + T
(a) (b) (c)
Figure 13 .2 Various approximations of integration .

which is unstable for T > 1 . Therefore, forward difference may not preserve the
stability of C(s) . In general, if the sampling period is sufficiently large, C(z) may
become unstable even if C(s) is stable .
The forward-difference transformation can easily be achieved using state-
variable equations as in Section 5 .2. Let
X(t) = Ax(t) + be(t) (13 .20a)
u(t) = cx(t) + de(t) (13 .20b)
be a realization of C(s), then
x(k + 1) = (I + TA)x(k) + Tbe(k) (13 .21 a)
u(k) = cx(k) + de(k) (13 .21b)
is the digital compensator obtained by using the forward-difference method .
Example 13 .2.4
Consider C(s) = 2(s - 2)/(s 2 + 4s + 3) . Then
z T
2 2I
C /
C(z) = C(s)Is-(Z-u/T = z 2
- 1 z- 1
+ 4 T + 3
T / (13 .22)
2T(z - 1 - 2T)
z 2 +(4T-2)z+(3T2 -4T+ 1)
is a digital compensator obtained using the forward-difference method . If T = 0 .5,
then (13 .22) becomes
2 . 0.5(z - 1 - 1)
C(z) = a
Z + (4 . 0.5 - 2)z + (3 . 0.25 - 2 + 1) (13 .23)
z - 2 z - 2
Z2 - 0.25 (z + 0.5)(z - 0.5)
This is the digital compensator . If we realize C(s) as
-4 3 1
x(t) = 1 0 x(t) + 0 e(t)
u(t) = [2 - 4] x(t)
Then
T 4T -13T] x(k) +
e(k) (13 .24a)
I [o ]
u(k) = [2 - 4] x(k) (13 .24b)

is the digital compensator . It can be shown that the transfer function of (13 .24) equals
(13 .22) . See Problem 13 .5.
Backward Approximation
In the backward approximation, the integration in (13 .16) is approximated by the
shaded area shown in Figure 13 .2(b). In this approximation, (13 .16) becomes
x(to + T) - x(to) = Ax(to + T )T
x(to) - x(to -T)
T = Ax(t0)
Thus the differentiation in (13 .15a) is approximated by

x(to) - x(to - T)
*(t)
T
This is equivalent to, in the transform domains,
1-z -1 z-1
(Backward difference) (13.25)
s = T = Tz
This is called the backward-difference method. A pole (s + a + j,13) in C(s) is
transformed by (13 .25) into
1 _ 1
Tz +a+j/3=T [(1 +aT)+j/3T-z-1]
(13 .26)
(1 +aT)+jJ3T~ 1
z
zT (1 + aT) + j/3T
If the pole is stable (that is, a > 0), then the magnitude of (1 + aT) + j/3T is
always larger than 1 ; thus, the pole in (13 .26) always lies inside the unit circle . Thus
the transformatio n in (13 .25) will transform a stable pole in C(s) into a stable pole
in C(z) . Thus, if C(s) is stable, so is C(z) .
Trapezoid Approximation
In this method, the integration in (13 .16) is approximated by the trapezoid shown in
Figure 13 .2(c) and (13 .16) becomes
x(to + T) + x(to)
x(to + T) - x(to) -= A T
Its z-transform domain is
zX(z) - X(z) =
T
A[zX(z) + X(z)] =
T(z 21) AX(z)

which implies
z
+ 1 X(z) = AX(z)
T
The Laplace transform of (13 .15a) with e(t) = 0 is A(s) = AX(s) . Thus, the
approximation can be achieved by setting
2z- 1
s = (Trapezoidal approximation) (13 .27)
Tz + 1
This is called the trapezoidal approximation method . Equation (13.27) implies
(z+1)s2=(z-1) or 2-)z=C1+ 2)
C 1-
Thus we have
Ts
1 +-
2
Z = (13 .28)
Ts
1
2
For every z, we can compute a unique s from (13 .27); for every s, we can compute
a unique z from (13 .28). Thus the mapping in (13 .27) is a one-to-one mapping,
called a bilinear transformation . Let s = o- + jco. Then (13 .28) becomes
To- TW
1 + + j
Z = 2 2
To- T (O
1 - 2 I 2
and
Tor
1 +-
~( 2 )2 ( T2w ) 2
Izl = ( 13 .29)
To-)2 (Tco)2
+
1 - 2
J( 2
This equation implies jzi = 1 if o- = 0, and jzj < 1 if o- < 0. Thus, the j&) -axis on
the s-plane is mapped onto the unit circle on the z-plane and the open left half
s-plane is mapped onto the interior of the unit circle on the z-plane . To develop a
precise relationship between the frequency in analog systems and the frequency in
digital systems, we define s = jce and z = ejw T. Then (13 .27) becomes
2e jwT - 1 2 e jo.SwT( ejo .SwT _ e -jo .5wT)
jw T ejwT + 1 T ej0.5wT(e jO .5wT + e -j0 .5wT)
2 2 j sin 0.5wT _ 2j wT
tan
T 2 cos 0 .5cwT T 2

which implies
2 coT
to = tan 2 (13 .30)
-
This is plotted in Figure 13 .3 . We see that the analog frequency from w = 0 to
W = oc is compressed into the digital frequency from to = 0 to co = it/T . This is
called frequency warping . Because of this warping and the nonlinear relationship
between w and ce, some simple manipulation, called prewarping, is needed in using
bilinear transformations . This is a standard technique in digital filterdesign. See, for
example, Reference [13].
Pole-Zero Mapping
Consider an analog compensator with pole p, and zero q,. In the pole-zero mapping,
pole pi is mapped into epiT and zero q, is mapped into egiT. For example, the
compensator
- 2 2(s - 2) - 2(s - 2) - 2(s - 2)
C(s) 1) (s-(-3))(s-(-1))
S + 4s + 3 (s + 3)(s +
is mapped into
2(z - e2T )
C(z) = (z
- e -3T )(z - e- T)
Thus the transformation is very simple .
There is, however, one problem with this transformation . Consider
b b
C(s) _ = (13 .31)
D(s) (s - a i)(s - a2)(s - a3)
Then its corresponding digital compensator is
b
C(z) = (z (13 .32)
- ea, T)(z - eazT)(z - ea3T)
Let y(t) be the step response of (13 .31). Because C(s) has three more poles than
tv
CO
Figure 13 .3 Analog and digital frequencies in (13 .30) .

y (t) y (kt)
>k
0 0 1 2 3 4 5 6 7
(a) (b)
Figure 13 .4 (a) Step response of (13 .31). (b) Step response of (13 .32) .
zeros, it can be shown that y(0) = 0, y(0) = 0, and y(0) = 0 (Problem 13 .7) and
the unit-step response will be as shown in Figure 13 .4(a). Let y(kT) be the response
of (13 .32) due to a unit-step sequence . Then, because C(z) has three more poles than
zeros, the response will start from k = 3, as shown in Figure 13 .4(b). In other words,
there is a delay of 3 sampling instants . In general, if there is a pole-zero excess
of r in C(z), then there is a delay of r sampling instants and the response will start
from k = r . In order to eliminate this delay, a polynomial of degree r - 1 is
introduced into the numerator of C(z) so that the response of C(z) will start at
k = 1 . For example, we may modify (13 .32) as
bN(z)
C(z) = (13 .33)
(z - eaIT )(z - e a2T )(z - ea3T)
with deg N(z) = 2. If the zeros at s = - in C(s) are considered to be mapped into
z = 0, then we may choose N(z) = z2. It is suggested to choose N(z) = (z + 1) 2
in Reference [52] and N(z) = z 2 + 4z + 1 in Reference [3]. Note that the steady-
state response of C(s) in (13 .31) due to a unit-step input is in general different from
the steady-state response of C(z) in (13 .32) or (13 .33). See Problem 13 .8. If they are
required to be equal, we may modify b in (13 .33) to achieve this .
13 .3 AN EXAMPLE
Consider the control system shown in Figure 13 .5(a) . The plant transfer function is
1
G(s) = (13 .34)
s(s + 2)
The overall transfer function is required to minimize a quadratic performance index
and is computed in (9 .38) as
3 = 3
G0(s) = s (13 .35)
2 + 3.2s + 3 (s + 1 .6 + jO.65)(s + 1 .6 - jO.65)
The compensator can be computed as
Go(s) 3(s + 2) = 3 - 3 .6 = :
C(s) = - k + CS (s) (13 .36)
G(s)(l - G0(s)) s + 3 .2 s + 3 .2

13.3 AN EXAMPLE 523
(a)
Clock
u Y
AID C(z) D/A G (s)
(b)
Figure 13 .5 (a) Analog control system . (b) Digital control system .
Now we shall implement the analog compensator as a digital compensator as

shown in Figure 13 .5(b). The impulse-invariant digital compensator is, using (13 .2),
3 .6Tz
CQ (z) = k + TZ[C,(s)] = 3 - (13.37)
z - e- 3 .2T
The step-invariant digital compensator is, using (13 .6),
Ch(z) _ (( 1- z-')Z CC(s) I

S
1 .875 1 .125
s + s+3.2J (13.38)
(1 - Z)
_ 1.875z
+
.125z
1 1
z - 1 z - e -3 2T]
3z - 1 .125 - 1.875e -3.2T
z - e- 3 .2T
The forward-difference digital compensator is, by substituting s = (z - 1)/T,
3(z - 1 + 2T)
cc(z)
`(z) = (13 .39)
z - 1 + 3.2T
The backward-difference digital compensator is, by substituting s = (z - 1)/Tz,
3(z - 1 + 2Tz)
Cd(z) = (13.40)
z - 1 + 3.2Tz

The digital compensator obtained by the bilinear transformation s = 2(z - 1)/

T(z + 1) is
CQ(z) = 6((1 + T)z - 1 + T)
(13 .41)
(2 + 3.2T)z - 2 + 3.2T
The digital compensator obtained by pole-zero mapping is
Cf(z) = 3(z - e -2T)
3 .2T (13 .42)
z - e
Figure 13 .6(a) through (f) shows the unit-step responses of the overall system in
Figure 13 .5(b) with C;(z) in (13 .37) through (13 .42) for T = 0.1, 0.2, 0.4, 0.6, and
0.8. For comparison, the response of the analog system in Figure 13 .5(a) is also
plotted and is denoted by T = 0 . We compare the compensators in the following
T = 0.1 0.2 0.4 0.6 0.8

Impulse-invariant A A C C F
Step-invariant A A B B B
Forward difference B B B F F
Backward difference C C F F F
Bilinear A- A- B B B+
Pole-zero B+ B C C C
where A denotes the best or closest to the analog system and F the worst or not
acceptable . For T = 0 .1 and 0 .2, the responses in Figure 13 .6(a) and (b) are very
close to the analog response; therefore, they are given a grade of A . Although the
responses in Figure 13 .6(e) are quite good, they show some overshoot; therefore
they are given a grade of A - . If T is large, the responses in Figure 13 .6(a), (c), and
(d) are unacceptable . Overall, the compensators obtained by using the step-invariant
and bilinear methods yield the best results . The compensator obtained by pole-zero
transformation is acceptable but not as good as the previous two . In conclusion,
if the sampling period is sufficiently small, then any method can be used to digitize
analog compensators . However, for a larger sampling period, it is better to use
the step-invariant and bilinear transformation methods to digitize an analog
compensator.
13.3.1 Selection of Sampling Periods

Although it is always possible to design a digital compensator to approximate an
analog compensator by choosing a sufficiently small sampling period, it is not de-
sirable, in practice, to choose an unnecessarily small one . The smaller the sampling
period, the more computation it requires . Using a small sampling period may also

13 .3 AN EXAMPLE 525
2.0 2 .0
1 .5 1 .5
0 .6
0 .1 T=0 0.8 . 0 4 --------------
1 .0 02 1 .0
---- 0.2
0.6 0 .1
.4 T=0
0 .5 0.5
0 :- ---- Q=8
0 .0 2.
0" 4.0 6 .0 8 .0 10.0 0 .0 20
. 4.0 6 .0 8 .0 10.0
(a) (b)
2 .0 A
2 .0
Figure 13 .6 Unit-step responses of analog system with various digital compensators .
introduce computational problems, as is discussed in the next section . See also Ref-
erence [46] . How to choose an adequate sampling period has been widely discussed
in the literature . References [46, 52] suggest that the sampling frequency or l/T be
chosen about ten times the bandwidth of the closed-loop transfer function . Reference
[3] suggests the following rules : If the pole of an overall system is real-say,

(s + a) with a > 0-then the sampling period may be chosen as

1
(13.43)
T = (2 - 4) X a
Because 1/a is the time constant, (13 .43) implies that two to four sampling points
be chosen in one time constant . If the poles of an overall system are complex and
have a damping ratio in the neighborhood of 0 .7, then the sampling period can be
chosen as
0.5 - 1
T = (13.44)
(on
where wn is the natural frequency in radians per second . If an overall transfer function
has more than one real pole and/or more than one pair of complex-conjugate poles,
then we may use the real or complex-conjugate poles that are closest to the imaginary
axis as a guide in choosing T.
The bandwidth of G0(s) in (13 .35) is found, using MATLAB, as B = 1 .2 ra-
dians . If the sampling frequency 1/T is chosen as ten times of 1 .2/27x, then T =
27r/12 = 0 .52 . The poles of the closed-loop system in (13 .35) are -1 .6 j0.65.
Its natural frequency is to,, = \ = 1 .73, and its damping ratio is = 3.2/2cwn =
0.46 . The damping ratio is not in the neighborhood of 0 .7; therefore, strictly speak-
ing, (13 .44) cannot be used . However, as a comparison, we use it to compute the
sampling period which ranges from 0.29 to 0 .58. It is comparable to T = 0 .52. If
we use T = 0 .5 for the system in the preceding section, then, as shown in Figure
3 .6, the system with the digital compensator in (13 .38) or (13 .41) has a step response
close to that of the original analog system but with a larger overshoot . If overshoot
is not desirable, then the sampling period for the system should be chosen as 0 .2,
about 20 times the closed-loop bandwidth . If T = 0 .2, the compensators in (13 .37),
(13 .39), and (13 .42) can also be used . Thus the selection of a sampling period de-
pends on which digital compensator is used . In any case, the safest way of choosing
a sampling period is to use computer simulations .
13 .4 EQUIVALENT DIGITAL PLANTS
Digital compensators in the preceding sections are obtained by discretization of

analog compensators . In the remainder of this chapter we discuss how to design
digital compensators directly without first designing analog ones . In order to do this,
we must first find an equivalent digital plant for a given analog plant. In digital
control, a digital compensator generates a sequence of numbers, as shown in Figure
13.7(a). This digital signal is then transformed into an analog signal to drive the
analog plant. There are a number of ways to change the digital signal into an analog
one . If a number is held constant until the arrival of the next number, as shown in
Figure 13 .7(b), the conversion is called the zero-order hold . If a number is extrap-
olated by using the slope from the number and its previous value, as shown in Figure

13 .4 EQUIVALENT DIGITAL PLANTS 527
A
(k) u (t) u (t)
>- k t t
(a) (b) (e)
Figure 13 .7 D/A conversions .
13 .7(c), the conversion is called the first-order hold. Clearly, higher-order holds are
also possible . However, the zero-order is the hold most widely used . The D/A
converter discussed in Section 12 .3 implements zero-order hold.
The Laplace transform of the pulse p(t) with height 1 and width T shown in
Figure A.3 is computed in (A .23) as
1 (1 - e - ST )
s
If the input of a zero-order hold is 1, then the output is p(t) . Therefore, the zero-
order hold can be considered to have transfer function
1 - e -Ts
Gh(s) _ (13 .45)
S
In digital control, a plant is always connected to a zero-order hold (or a D/A con-
verter) and the transfer function of the plant and hold becomes
e -Ts) _
(I - G(s) (13 .46)
S
as shown in Figure 13 .8(a). Note that the input u(kT) of Figure 13 .8(a) must be
modified as
u*(t) _ u(kT)6(t - kT )
k=0
as in (12 .8) in order for the representation to be mathematically correct . The output
y(t) of the plant and hold is an analog signal . If we add an A/D converter after the
plant as shown in Figure 13 .8(a), then the output is y(kT) . If we consider y(kT) as
the output and u(kT) as the input, then the analog plant and hold in Figure 13 .8(a)
can be modeled as a digital system with discrete transfer function
G(s)] = [G(s)] G(s)1
G(z) = Z C(1 - e Ts) J Z - Z e Ts J (13 .47) s
S S
Because e-Ts introduces only a time delay, we may move it outside the z-transform
as
Z, [e-Ts G(s) = e-TSB L
S s J
G(s)

1 _ e -T` y (t) y(kT) u(kT) z-1 Gs)

.31 G (s) A/D
S
z
(a) (b)
Figure 13.8 (a) Equivalent analog plant . (b) Equivalent digital plant .
Using z = e Ts, (13 .47) becomes
G(z) z-')Z G(s)1 = z - 1 Z [ G(s) ] (13 .48)

C sJ z s
This is a discrete transfer function . Its input is u(kT) and its output is y(kT ), the
sample of the analog plant output in Figure 13 .8(a) . The discrete system shown in
Figure 13 .8(b) is called the equivalent discrete or digital plant, and its discrete
transfer function is given by (13 .48).
By comparing (13 .48) with (13 .6), we see that the equivalent discrete plant
transfer function G(z) and the original analog plant transfer function G(s) are step
invariant . As discussed in Secti on 13 .2, the step-invariant digital plant G(z) can also
be obtained from analog plant G(s) by using state-variable equations . Let
i(t) = Ax(t) + bu(t) (13.49a)
y(t) = cx(t) + du(t) (13.49b)
be a realization of G(s) . If the input is stepwise as in digital control, then the input
y(t) and output u(t) at t = kT can be described, as derived in (2.89), by
x(k + 1) = Ax(k) + 6u(k) (13 .50a)
y(k) = cx(k) + du(k) (1 3 .50b)
with
T
A = e AT b = Jo eA dT Ib T c = c d = d (13 .50c)
The transfer function of (13 .50) equals (13 .48). As shown in Example 13 .2.3, this
computation can easily be carried out by using the commands tf2ss, c2d, and ss2tf
in MATLAB .
13 .4 .1 Hidden Dynamics and Non-Minimum-Phase Zeros

Once an equivalent digital plant is obtained, we can design a digital compensator to
control the plant as shown in Figure 13 .9(a) or, more generally, design an algorithm
and use a digital computer to control the plant as shown in Figure 13 .9(b). In the
remainder of this chapter, we shall discuss how to carry out the design . Before
proceeding, we use examples to illustrate the problems that may arise in using equiv-
alent digital plants .

r(kT) + u(kT) y(kT) r(kT) u(kT) y(kT)

C(z) G (z) Digital G (z)
" Y" (a)

computer
c
(b)
Figure 13.9 Direct design of digital control system .
Example 13 .4.1
Consider an analog plant with transfer function
101 _ 101
G(s) _ (s + 1) 2 + 100 (13.51)
s2 + 2s + 101
with its poles plotted in Figure 13.10. Its step-invariant digital transfer function is
given by
101
G(z) = (1 - z-) IS((s + 1)2 + 100)
z- 1 z 1 s+ 1 1
z s (s + 1)2 + 100 (s + 1)2 + 100
z - 1 z z
2 - ze-T cos lOT
z C z - 1 z2 - 2ze -T cos lOT + e- 2T
(13.52)
ze -T sin lOT
10(z 2 - 2ze -T cos 10T + e -2T ) ]
where we have used the z-transform pairs in Table 12.1 . Now if the sampling period
T is chosen as lOT = 2ar, then cos 10T = 1, sin 10T = 0 and e -T = e- 0.21r -
IM s
X 10
7r/T
Y~V_
2 2
0
z_/_ -7c/T
I - Res
X -10
Figure 13.10 Poles of (13 .51).

0.53 . Thus (13 .52) can be reduced as
z- 1 .z z(z-e T)
G(z) =
z [z - 1 z2 - 2ze -T + e -2T
z - 1 z z(z-e - T )
(13 .53)
z z - 1 (z
z- 1 z z _ 1- e-T 0.47
z [z- 1 z-e -T z-e-T z-0.53
It is a transfer function with only one real pole, whereas the original analog plant
transfer function in (13 .51) has a pair of complex-conjugate poles . Figure 13 .11
shows the unit-step responses of (13 .51) and (13 .53) . We see that the oscillation in
the analog plant does not appear in its step-invariant digital plant . Thus, some dy-
namics of an analog plant may disappear from or become hidden in its equivalent
digital plant.
1 .8
1 .6
1 .4
1 .2
1
0 .8
0.6
0 .4
0 .2
0
6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
(a) (b)
Figure 13 .11 Unit-step responses of G(s) and G(z) .
The reason for the disappearance of the dynamics can easily be explained from
the plot in Figure 13 .10 . Recall from Figure 12 .9 that the mapping z = e T is not a
one-to-one mapping . If the sampling period T is chosen so that rr/T equals half of
the imaginary part of the complex poles as shown in Figure 13 .10, then the complex
poles will be mapped into real poles . Furthermore, the two poles are mapped into
the same location. This is the reason for the disappearance of the dynamics . Knowing
the reason, it becomes simple to avoid the problem . If the sampling period is chosen
to be small enough that the primary strip (the region bounded between - ar/T and
Ir/T as shown in Figure 12 .9) covers all poles of G(s), then no dynamic will be lost
in the sampling and its equivalent digital plant can be used in design .

Example 13 .4.2
Loss of dynamics can also be explained using state-variable equations . Consider the
transfer function in Example 13.4.1 or
101 101
G(s) = 1)2 _ I
(s + + 100 (s + + j lO)(s + 1 - j lO) (13 .54)
5.05j 5.05j
s+1+10j s+1-10j
It is plotted in Figure 13 .12 . If we assign state variables as shown, then we can

readily obtain the following state-variable equation
-1 10j
x(t) = 1 u(t) (13 .55a)
0 - 1 0+ 10']
J x(t) + C ]
y(t) _ [5 .05j -5 .05j]x(t) (13 .55b)
This is a minimal realization of (13 .54) and is controllable and observable .' Because
matrix A is diagonal, the discretized equation of (13 .55) with sampling period T can
be easily obtained as
I (-t-lOi)T )
(1 - e
Ce(-t-10j)T 0 lOj
x(k) + 1 + u(k)
0 e(- t + tOj)T e(-t+lo;)T)
1 (I -
1 - lOj
y(k) = [5 .05j -5 .05j]x(k)
Figure 13 .12 Block diagram of (13 .54) .
'Although we consider state-variable equations with only real coefficients in the preceding chapters, all
concepts and results are equally applicable to equations with complex coefficients without any modifi-
cation . See Reference [15] .

If T is chosen as lOT = 27r or T = 0.2ar, then e - 1OfT = 1, and the equation reduces
to
1
- e -T)
e T 0 1 1 + 10j (1
x(k + 1) = x(k) + u(k) (13 .560)
0 e -T I 1
- e - T)
1 - lOj (1
y(k) = [5 .05j -5.05j]x(k) (13.56b)
Because A is diagonal and has the same eigenvalues, the equation is neither con-
trollable nor observable . See Problem 11 .2. This can also be verified by checking
the ranks of the controllability and observability matrices . The transfer function of
(13 .56) can be computed as (1 - e-T )/(z - e-T ), which is the same as (13.53) .
Thus controllability and observability of a state-variable equation may be destroyed
after sampling. For a more general discussion, see Reference [15, p. 559] .
_ We now discuss a different problem due to discretization . Consider G(s) =

N(s)/D(s) . Let G(z) = N(z)/D(z) be its equivalent digital plant transfer function .
Pole-zero excess is defined as the difference between the n umber of poles and the
number of zeros . It turns out that if the pole-zero excess of G(s) is zero, so is G(z).
That is, if G(s) is biproper, so is G(z). However, if G(s) is strictly proper, then the
pole-zero excess of G(z) is always 1 no matter what the pole-zero excess of G(s) is.
Thus sampling will introduce zeros . The number of additional zeros introduced in
G(z) equals r - 1 where r is the pole-zero excess of G(s) . We first give an example
and then establish the assertion .
Example 13 .4 .3
Consider
1
G(s) (13 .57)
= s(s z + 2s + 2)
Its pole-zero excess is 3; therefore, the step-invarient discretization of G(s) will
introduce two zeros into G(z) . We use MATLAB and state-variable equations to
carry out discretization . The commands
nu =1 ;de = [1 2 2 0] ; (Express G(s) in numerator and denominator.)
[a,b,c,d] = tf2ss(nu,de) ; (Yield controllable-form realization of G(s).)
[da,db]=c2d(a,b,0 .5) ; (Discretize a and b with sampling period 0.5.)
[dn,dd]=ss2tf(da,db,c,d,1) ; (Compute the discrete transfer function G(z).)
[z,p,k]=tf2zp(dn,dd) (Express G(z) in zero and pole form .)

yield2
0.0161(z + 2.8829)(z + 0.2099)
G(z) (13 .58)
(z - 1 .0000)(z - 0.5323 + 0 .2908j)(z - 0.5323 - 0 .2908j)
The discretization does introduce two zeros . If the sampling period is chosen as
T = 0.1, then
1 .585 10 -4 (z + 3 .549)(z + 0.255)
G(z) = (13 .59)
(z - 1)(z - 0.9003 + 0 .0903j)(z - 0.9003 - 0.0903j)
It also introduces two zeros .
Now we discuss why sampling will introduce additional zeros . The Laplace
transform of the unit-step response y(t) of G(s) is G(s)/s . Using the initial value
theorem in Appendix A, its value at t = 0 can be computed as
G(s)
y(0) = lim s = G(oc)
t->= S
Let y(kT) be the unit-step response of G(z), that is,

z
Y(z) = G(z)U(z) = G(z)
Z - 1
Using the initial value theorem in (12 .53), the value of y(kT) at k = 0 can be obtained
from Y(z) as
z
y(O) = lim Y(z) = lim G(z) z 1 = G(oc)
Z_. Z-m
Because y(kT) is the sample of y(t), we have y(O) = y(O) and, consequently, G(oo)
= G() . Now if G(s)is biproper, that is, G(oc) 0 0, then G(-) ~ 0 and G(z) is
biproper. Similarly, if G(s) is strictly proper, so is G(z) . Let G(z) be a strictly proper
transfer function of degree 4 :
b,z 3 + b2z2 + biz + b4
G(z) - (13 .60)
z4 + a,z3 + a2 z 2 + a3z + a4
Then its unit-step response is

b1 z 3 + b2z2 + biz + b4 z
Y(z) = Z3 Z2
z4 + a, + a2 + a3z + a4 z - 1 (13 .61)
= 0 + b,z -1 + [b2 - b, (a, - 1)]z -2 +
2The two steps ss2tf and tf2zp can be combined as ss2zp . However, I was not able to obtain correct
answers using ss2zp for this problem, even though ss2zp is successfully used in Example 13 .5 .1 .
Therefore, the difficulty may be caused by numerical problems .

Thus we have y(O) = 0 and y(T) = b, . The unit-step response y(t) of the analog
system G(s) is generally of the form shown in Figure 13 .4(a) . No matter what its
pole-zero excess is, generally y(T) is different from zero . If y(kT) is a sample of
y(t), then y(T) = b, = y(T) 0 . Therefore, the pole-zero excess of G(z) in (13 .60)
=,Z=
is 1 . This establishes the assertion that if the pole-zero excess of G(s) is r ? 1, the
pole-zero excess of G(z) is always 1 . Thus, sampling will generate r - 1 additional
zeros into G(z) .
In the analog case, a zero is called a minimum-phase zero if it lies inside the
open left half s-plane, a non-minimum-phase zero if it lies inside the closed right
half s-plane . Following this terminology, we call a zero inside the interior of the unit
circle on the z-plane a minimum-phase zero, a zero on or outside the unit circle a
non-minimum-phase zero . For the analog plant transfer function in (13 .57), sampling
introduces a minimum- and a non-minimum-phase zero as shown in (13 .58) and
(13 .59) . Non-minimum-phase zeros will introduce constraints in design, as will be
discussed in a later section .
To conclude'this section, we mention that the poles of G(z) are transformed
from the poles of G(s) by z = eS T. Once the poles of G(z) and, consequently, the
coefficients a; in (13 .60) are computed, then the coefficients b; in (13 .60) can be
computed from al and the samples of the unit-step response y(t) as
b, 1 0 0 0 y(T)
b2 - a, - 1 1 0 0 y(2T)
(13.62)
b3 a 2 - a, a, - 1 1 0 y(3T)
b4 a3 - a 2 a2 - a, a, 1 1 y(4T)
See Problem 13 .11 . Thus, the numerator of G (z) in (13 .60) is determined by the first
four samples of y(t). If T is small, then these four samples are hardly distinguishable,
as can be seen from Figure 13 .4(a). Therefore, the possibility of introducing errors
in b; is large . Thus, using an unnecessarily small sampling period will not only
increase the amount of computer computation, it will also increase computational
error. Therefore, selection of a sampling period is not simple .
13.5 ROOT-LOCUS METHOD
In this section we discuss how to use the root-locus method to design a digital
compensator directly from a given equivalent digital plant . The root-locus design
method actually consists of two parts : searching for a desired pole region, and plot-
ting the roots of p(s) + kq(s) as a function of real k. The plot of root loci discussed
in Section 7 .4 for the continuous-time case is directly applicable to the discrete-time
case ; therefore, we discuss only the desired pole region in the digital case .
The desired pole region in Figure 7 .4 for analog systems is developed from the
specifications on the settling time, overshoot, and rise time . Settling time requires
closed-loop poles to lie on the left-hand side of the vertical line passing through
-a = -4.5/t s, where is denotes the settling time . The vertical line is transformed
by z = esT into a circle with radius e _ T as shown in Figure 13 .13(a). Note that the

Im s lm z
z/T
4
e
I 0
,
-a2-a Res Rez
-zc/T
(a)
Im s Im z
OnNwomm
., Re
(b)
INPF
Im s Im z
Res ,fAh Re z
(c)
i
Im axis
90 72
1
108
Sam-7r 21r 3n
2 lOT 54-
126'p7np _ ~
~` 5T
144 Fr 5T
l
IOT,~0~~~_0 .1 ~-0
U~ '
,
I
__
A~ 1~
36
~If-/ 4, ÌV
16 I OT
A-
M~~~~~~
~~' 'I
NEWS ~~
'1
,5 p~ \
~~`,` ~'"``
0.9
!, 1 . V1 ~~ ~\ ar
2OT
CUn
I ,`
T i~~
-1 .0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1 .0
(d)
Liwira 14 1 4 TlpcirPil nnla raninn

mapping z = es T is not one-to-one, therefore we map only the primary strip (or
- ar/T :5 (o it/T) in the s-plane into the interior of the unit circle on the z-plane .
The poles denoted by X on the s-plane are mapped into the positive real axis of the
z-plane inside the unit circle . The poles with imaginary part ar/T shown with small
squares are mapped into the negative real axis inside the unit circle .
The overshoot is governed by the damping ratio ~ or the angle 0 in the analog
case . If we substitute s = rej e into z = esT and plot z as a function of r (for a fixed
0) and as a function of 0 (for a fixed r), then we will obtained the solid lines and
dotted lines in Figure 13 .13(b) . Because the overshoot is governed by 0, the solid
line in Figure 13 .13(b) determines the overshoot . The distance from the origin or r
is inversely proportional to the rise time ; therefore, the dotted line in Figure 13 .13(b)
determines the rise time . Consequently, the desired pole region in the analog case
can be mapped into the one shown in Figure 13 .13(c) for the digital case . For con-
venience of design, the detailed relationship in Figure 13 .13(b) is plotted in Figure
13.13(d) . With the preceding discussion, we are ready to discuss design of digital
compensators using the root-locus method . We use an example to illustrate the
design .
Example 13 .5.1
Consider the problem in Section 7.2 .2-that is, given a plant with transfer function
1
(13 .63)
G(s) = s(s + 2)
use the root-locus method to design an overall system to meet
1. Position error = 0 .
2. Overshoot < 5% .
3. Settling time < 9 seconds .
First we compute the equivalent digital plant transfer function of (13 .63):
G(z) z-i)2 = (1 - z-1)2 [s

CGss)] 2(s + 2)]
2 - 0.25 s0+52
z-')2 O (13.64)
z - 1 0.5Tz 0 .25z 0.25z

z (z-1) 2 z-1 z-e -2T
If the sampling period is chosen as T = 1, then e- 2T = 0.1353 and (13 .64) can be
simplified as
0.2838(z + 0.5232)
G(z) = (13 .65)
(z - 1)(z - 0.1353)

13 .5 ROOT-LOCUS METHOD 537
u (k) y (k)
h G (z)
This can also be obtained using MATLAB by typing nu = [1 ]; de= [1 2 0] ;

[a,b,c,d]=tf2ss(nu,de) ; [da,db]=c2d(a,b,1) ; [z,p,k]=ss2zp(da,db,c,d). The
result is z = -0 .5232, p = 1, 0 . 1353, and k = 0.2838, and is the same as (13 .65).
Next we choose the unity-feedback configuration in Figure 13 .14 and find, if pos-
sible, a gain h to meet the design specifications . First we compute the overall transfer
function :
hG(z) - 0.2838h(z + 0.5232)
G,(z) - (13 .66)
1 + hG(z) (z - 1)(z - 0.1353) + 0.2838h(z + 0 .5232)
Because of the presence of factor (z - 1) in the denominator, we have G0(1) = 1
for all h. Thus if G0 (z) is stable, and if r (k) = a, then
ys(k) := lim y(k) = Go(1)a = a
k--
Thus the position error, as defined in (6.3), is

a - a
ep = lim r(k) - ys(k) = 0
k-poe a a
and the overall system will automatically meet the specification on the position error
so long as the system is stable . This situation is similar to the continuous-time case
because the analog plant transfer function is of type 1 . Thus if a digital plant transfer
function has one pole at z = 1, and if the unity-feedback system in Figure 13 .14 is
stable, then the plant output will track asymptotically any step-reference input .
As discussed in Section 7 .2.1, in order to have the overshoot less than 5%, the
damping ratio ~ must be larger than 0 .7. This can be translated into the curve denoted
by 0 .7 in Figure 13 .15 . In order to have the settling time less than 9 seconds, we
require a > 4 .5/9 = 0 .5 . This can be translated into the circle with radius e_OST
= 0.606 as shown in Figure 13 .15. We plot in the figure also the root loci of
-1 _ 0.2838(z + 0.5232)
(13 .67)
h (z - 1)(z - 0.1353)
The one in Figure 13 .15(a) is the complete root loci ; the one in Figure 13 .15(b)
shows only the critical part . The root loci have two breakaway points at 0 .48 and
- 1 .52 and consist of a circle centered at - 0.5232 and with radius 1 . From the root
loci, we see that if
h l <h !5 h2

CHAPTER 13 DISCRETE-TIME SYSTEM DESIGN
1M Z
Re z
-1 0
-0 .523
(a)
Im z
= 0 .7
i
1
1h
i
0 I I
)( V I )( - Re z
-0 .523 0 0 .2 0 .4 1 0.6 1
0 .48
(b)

where h 1 and h2 are indicated on the plot, the system will meet the specifications on
overshoot and settling time. Now the system is required to have a rise time as small
as possible, therefore the closest pole should be as far away as possible from the
origin of the s-plane or on the dotted line in Figure 13 .13(d) with cw n as large as
possible . Thus we choose h = h2 . By drawing vectors from the two poles and one
zero to h2 as shown in Figure 13 .15(b) and measuring their magnitudes, we obtain,
from (13 .67),
_ 0.67 X 0.42
h2 .03
0.2838 X 0 .96 = 1
Thus by choosing h = 1 .03, the overall system will meet all design specifications.
This example shows that the root-locus method discussed in Chapter 7 can be
directly applied to design digital control systems . Note that the result h = 1 .03 in
digital design is quite different from the result h = 2 obtained in analog design in
Section 7 .2.2 . This discrepancy may be caused by the selection of the sampling

13 .5 ROOT-LOCUS METHOD 539
period T = 1 . To see the effect of the sampling period, we repeat the design by
choosing T = 0 .2. Then the equivalent digital transfer function is
0.0176(z + 0.8753) - 0.0176z + 0.0154

G(z) - (13.68)
(z - 1)(z - 0.6703) z2 - 1.6703z + 0.6703
From its root loci and using the same argument, the gain h that meets all the spec-
ifications can be found as h = 1 .71 . This is closer to the result of the analog design .
To compare the analog and digital designs, we plot in Figure 13 .16 the plant outputs
and actuating signals of Figure 13 .14 due to a unit-step reference input for T = I
with h = 1 .03 and T = 0.2 with h = 1 .71 . We also plot in Figure 13 .16 the unit-
step response and actuating signal for the analog design, denoted by T = 0 and
h = 2. We see that the digital design with T = 0.2 and h = 1 .71 is almost identical
to the analog design with T = 0 and h = 2. The maximum value of the actuating
signal in the digital design, however, is smaller .
In conclusion, the root-locus method can be applied to design digital control
systems. The result, however, depends on the sampling period . If the sampling period
is sufficiently small, then the result will be close to the one obtained by analog design .
There is one problem in digital design, however . If the sampling period is small,
then the possibility of introducing numerical error will be larger . For example, for
the problem in the preceding example, as T decreases, the design will be carried out
in a region closer to z = 1, where the solid lines in Figure 13 .13(d) are more
clustered . Therefore, the design will be more sensitive to numerical errors .
2.0
u(t) (T=0,h=2)
u(k) (T = 0 .2, h = 1 .71)
1 .0
y(t) (T=1 h=1)

/ u (k)
0 .0
y(t) (T=0,h=2)
y(t) (T = 0 .2, h = 1 .71)
1 .0 0
2 .0 4 .0 6.0 8.0 10.0
Figure 13 .16 Comparison of analog and digital designs .

13.6 FREQUENCY DOMAIN DESIGN
Frequency-domain design methods, specially, the Bode plot method, are useful in
designing analog control systems . Because the frequency response G(jw) of analog
transfer functions is a rational function of jw and because the Bode plot of linear
factors can be approximated by two asymptotes-which are obtained by considering
two extreme cases : w -* 0 and w - --the plot of Bode plots is relatively simple .
In the digital case, the frequency response G(e'"T) is an irrational function of w.
Furthermore, the frequency of interest is limited to the range from 0 to 7r/T . There-
fore, the procedure of plotting Bode plots in Chapter 8 cannot be directly applied to
the discrete-time case .
One way to overcome this difficulty is to carry out a transformation . Consider
2z - 1
w = (13 .69a)
T z + 1
or
Z = (13 .69b)
This is the bilinear transformation studied in (13 .27). Let z = e 3`"T and w = jv .
Then, we have, similar to (13 .30),
2 wT
v = -tan (13 .70)
T
Thus the transformation transforms w from 0 to Ir/T to v from 0 to - as shown in
Figure 13 .3 . Define
G(w) = G(z)Iz=(1+wT12)1(1-wT/2) (13 .71)
Then G(jv) will be a rational function of v and v ranges from 0 to -. Thus the Bode
design method can be applied . In conclusion, design of digital compensators using
the Bode plot method consists of the following steps : (1) Sample the analog plant
to obtain an equivalent digital plant transfer function G(z). (2) Transform G(z) to
G(w) using (13 .69). (3) Use the Bode method on G(jv) to find a compensator
C(w). (4) Transform C(w) to C(z) using (13 .69). This completes the design .
The key criteria in the Bode design method are the phase margin, gain margin,
and gain-crossover frequency . Because the transformations from G(s) to G(z) by
z = esT and from G(z) to G(w) by (13 .69) are not linear, considerable distortions
occur. Furthermore, the sampling of G(s) may introduce non-minimum-phase zeros
into G(z); this may also cause difficulty in using the Bode method . Thus, even though
the Bode method can be used to design digital compensators, care must be exer-
cised and the result must be simulated before actual implementation of digital
compensators .

13 .7 STATE FEEDBACK, STATE ESTIMATOR, AND DEAD-BEAT DESIGN 541
13 .7 STATE FEEDBACK, STATE ESTIMATOR, AND DEAD-BEAT DESIGN
The design methods in Sections 11 .4 and 11 .6 for analog systems can be applied to
digital systems without any modification . Consider the discrete-time state-variable
equation
x(k + 1) = Ax(k) + bu(k) (13 .72a)
y(k) = cx(k)' (13 .72b)
If we introduce the state feedback
u(k) = r(k) - kx(k)
x(k + 1) = (A - bk)x(k) + br(k) (13 .73a)
y(k) = cx(k) (13 .73b)
G0(z) = c(zI - A + bk) -lb (13 .74)
If (A, b) is controllable, by choosing a real feedback gain k, all eigenvalues of
(A - bk) or, equivalently, all poles of G0(z) can be arbitrarily assigned, provided
complex-conjugate poles are assigned in pairs . The design procedure in Section 11 .4
can be used to find k without any modification. If (A, c) is observable, a full-
dimensional or a reduced-dimensional state estimator can be designed to generate
an estimated state ; the rate for the estimated state to approach the original state can
be arbitrarily chosen by selecting the eigenvalues of the estimator . The procedure
for designing a full-dimensional estimator in Section 11 .6 and the procedure for
designing a reduced-dimensional estimator in Section 11 .6.1 can be used without
any modification . The only difference is that poles are chosen to lie inside the open
left half plane in the analog case and to lie inside the unit circle in the digital case.
In the digital case, if all poles of G 0(z) are chosen to be located at the origin, it
is called the dead-beat design . In this case, the overall transfer function is of the
form
G0(z) = N(z) (13 .75)

Z
with deg N0 (z) < n. The unit-step response of (13 .75) is

(z)
Y(z) = N
Zn Z - I (13 .76)
_2 -n+1 No (1)z
CD + C1Z -1 + CZZ + . . . Cn-1Z + CnZ-n +
z - 1
The transient response consists of a finite number of terms ; it becomes identically
zero after the nth sampling instant . Thus, if the input is a step sequence, the output

of (13 .75) will become a step sequence after the nth sampling instant . This cannot
happen in analog systems ; the step response of any stable analog system will become
a step function only at t - oo. Note that a pole at z = 0 in digital systems corresponds
to a pole at s = -- in analog systems . It is not possible to design an analog system
with a proper transfer function that has all poles at s = - oc . Thus dead-beat design
is not possible in analog systems .
Example 13 .7 .1
Consider the discrete-time state-variable equation
x(k + 1)
=C O
-1
J
x(k) + [ lOJ u(k) (13 .77a)
y(k) = [1 0] x (13 .77b)
with transfer function G(z) = 10/z(z + 1) . This equation is identical to (11 .41)
except that it is in the discrete-time domain . Find a feedback gain k in u = r - kx
so that the resulting system has all eigenvalues at z = 0 . We use the procedure in
Section 11 .4 to carry out the design . We compute
4(z) = det (zI - A) = det z -1 = z(z + 1) = z 2 + z + 0

C0 z + 1 J
and
4(z)=z 2 +0 z +0
Thus we have
k = [0 -1 0 - 01 = [-1 01
The similarity transformation is identical to the one in Example 11 .4.2 and equals
0 10
P_1 =
10 0
Thus the feedback gain is
k = kP = [-1 -0.1] (13 .78)

01 10 .1 01] = [0
This completes the design of state feedback. Next we use the procedure in Section
11 .6.1 to design a reduced-dimensional state estimator . The procedure is identical
to the one in Example 11 .6.1 . Because (13 .77) has dimension n = 2, its reduced-
dimensional estimator has dimension 1 . Because the eigenvalue of F is required to
be different from those of A, we cannot choose F = 0 . If we choose F = 0, then
TA - FT = gc has no solution . See Problem 13 .14 . We choose, rather arbitrarily,

13 .7 STATE FEEDBACK, STATE ESTIMATOR, AND DEAD-BEAT DESIGN 543
F = 0.1 and choose g = 1 . Clearly (F, g) is controllable . The matrix T is 1 X 2 .

Let T = [t 1 tz ] . Then it can be solved from
0 1
[t 1 t2] -1 -0 .1 X [t 1 t2] = 1 X [1 0]
[0
or
[0 0 .ltz] = [1 0]
t1 - t2 ] - [0.lt 1
Thus we have -0 .1t1 = 1 and t 1 - 1 .1t2 = 0 which imply t 1 = - 10 and t2 =
t 1/ 1 .1 = - 9.09 . The matrix
P
[T] - [-10 -9 .09]
is clearly nonsingular and its inverse is
t
-9.09] [-1.1 -0.11
We compute
h = Tb = [-10 -9.09] = - 90 .9
[10]
Thus the 1-dimensional state estimator is
z(k + 1) = 0 .lz(k) + y(k) - 90 .9u(k) (13 .79a)
1 0 y(k) y(k)
x(k) = (13.79b)
[ - 1 .1 -0.11] [z(k)] = [ -1 .ly(k) - 0 .l lz(k)
This completes the design . We see that the design procedure is identical to the analog
case . We plot in Figure 13 .17 the state estimator in (13 .79) and the state feedback
r(k) +O u(k) 10 y (k)

z(z+1)
9 .09
State feedback State estimator

Figure 13.17 State feedback and state estimator .

in (13 .78) applying at the output of the estimator . To verify the result, we use Ma-
son's formula to compute the transfer function from r to y. It is computed as
10
G0 (z)
zZ
Thus the result is correct .
To conclude this section, we mention that in dead-beat design, the actuating

signal usually will be large and the plant may saturate . This is especially true if the
sampling period is chosen to be small . If the sampling period is large, the possibility
of saturation will be smaller, but the design may be poorer . Therefore, computer
simulation of dead-beat design is recommended before it is implemented in practice .
13.8 MODEL MATCHING '
Pole placement and model matching, discussed in Chapter 10 for analog systems,
can be applied directly to digital systems . To discuss model matching, we must
discuss physical constraints in implementation . As in the analog case, we require
1. All compensators used have proper rational transfer functions .
2. The resulting system is well posed .
3. The resulting system is totally stable .
4. There is no plant leakage in the sense that all forward paths from r to y pass
through the plant .
As was discussed in Section 12 .5.2, a compensator with an improper transfer
function is not a causal system and cannot be built to operate on real time . If a
resulting system is not well posed, then the system has at least one improper closed-
loop transfer function and the system can predict what input will be applied in the
future . This is not possible in the real world . If a resulting system is not totally stable,
then the response of the system may grow without bound if noise or disturbance
enters the system . The fourth constraint implies that all power must pass through
the plant and that no compensator may be introduced in parallel with the plant . For
a more detailed discussion of these constraints, see Chapter 6 .
Given a digital plant transfer function
N(z)
G(z) = (13.80)
D(z)
an overall transfer function
G0(z) = N(z) (13.81)

D0(z)

13 .8 MODEL MATCHING 545
is said to be implementable if there exists a control configuration such that G 0(z) can
be implemented without violating any of the preceding four constraints . This defi-
nition is similar to the analog case . The necessary and sufficient conditions for G(z)
= N(z)/D(z) to be implementable are
1. deg D (z) - deg N (z) ? deg D(z) - deg N(z) (pole-zero excess inequality) .
2. All zeros of N(z) on and outside the unit circle must be retained in N (z) (re-
tainment of non-minimum-phase zeros) .
3. All roots of D (z) must lie inside the unit circle .
Condition (1) implies that if the plant has a delay of r : = [deg D(z) - deg N(z)]
sampling instants, then G (z) must have a delay of at least r sampling instants . If a
control configuration has no plant leakage, then the roots of N(z) will not be affected
by feedback . Therefore, the only way to eliminate zeros of N(z) is by direct pole-
zero cancellations . Consequently, zeros on or outside the unit circle of the z-plane
should be retained in N(z) . In fact, zeros that do not lie inside the desired pole region
discussed in Figure 13 .13(c) should be retained even if they lie inside the unit circle .
Thus the definition and conditions of implementable transfer functions in the digital
case are identical to the analog case discussed in Section 9 .2.
As in the analog case, the unity-feedback configuration in Figure 10.1 can be
used to implement every pole placement but not every model matching . The two-
parameter configuration in Figure 10 .6 and the plant input/output feedback config-
uration in Figure 10 .15 can be used to implement any model matching . The design
procedures in Chapter 10 for the analog case are directly applicable to the digital
case . We use an example to illustrate the procedures .
Example 13 .8.1
Consider
z + 0.6 = N(z)
G(z) = (13 .82)
(z - 1)(z + 0.5) D(z)
Implement the overall transfer function
_ 0.25(z + 0.6) N(z)
Gz) (13 .83)
z2- 0.8z + 0 .2 D (z)
Because G(1) = 0 .25(1 + 0 .6)/(1 - 0.8 + 0.2) = 1, the output of G (z)

will track any step-reference input without an error . Note that even though zero
(z + 0 .6) lies inside the unit circle, it lies outside the desired pole region shown in
Figure 13 .13(c); therefore, it is retained in G(z). We first use the unity-feedback
configuration shown in Figure 13.18(a) to implement (13 .83). Using (10 .2) with s

CHAPTER 13 DISCRETE-TIME SYSTEM DESIGN
y (k)
C (Z) G (z)
(a)
L (z) A - '( )
(b)
Figure 13 .18 (a) Unity-feedback system . (b) Two-parameter feedback system.
replaced by z, we have
_ G0(z)
C(z)
G(z)(1 - G0 (z))
0.25(z + 0.6)
zz -0.8z+0 .2
(z + 0.6) 0.25(z + 0.6)
(z - 1)(z + 0.5) 1 z~ - 0.8z + 0.2 (13 .84)
0.25(z - 1)(z + 0.5)
zZ - 0.8z + 0.2 - 0.25(z + 0.6)
0.25(z - 1)(z + 0.5) 0.25(z + 0.5)

(z - 1)(z - 0.05) z - 0.05
This is a proper compensator . We mention that the design has a pole-zero cancel-
lation of (z + 0.5). The pole is dictated by the plant transfer function . Because the
pole is stable, the system is totally stable.
Next we implement G 0(z) in the two-parameter configuration shown in Figure
13 .18(b) . We use the procedure in Section 10 .4.1 with s replaced by z . We compute
G0(z) 0.25(z + 0 .6) _ 0.25 Ni,(z)
N(z) (z 2 - 0.8z + 0 .2)(z + 0 .6) 0.8z + 0
(z2 - .2) - DP (z)
Next we find a DP (z) such that the degree of .Pp(z)DP(z)
DP (z)DP (z) is at least 2n - 1 = 3 .
Because the degree of Dp ( z) is 2, the degree of can be chosen as 1 . We choose
DP(s) = z. Then we have
L(z) = NP (z)DP(z) = 0.25 X z = 0.25z

13 .9 CONCLUDING REMARKS 547
The polynomials A(z) = A 0 + A,z and M(z) = Mo + M1z can be solved from
A(z)D(z) + M(z)N(z) = DD(z)5,,(z)
with
F(z) : = DP(z)D,(z) = (z 2 - 0.8z + 0 .2)z = z3 - 0 .8z2 + 0 .2z
or from the following linear algebraic equation
Do No 0 0 A0 -0.5 0.6 : 0 0 A0 0
D1 N1 Do No Mo -0.5 1 -0.5 0.6 Mo _ 0.2
D2 N2 D, N, A, 1 0 -0.5 1 A, -0.8
0 0 1D Z N Ml 0 0 1 0 Ml I
The solution is A(z) = A 0 + A 1 z = -3.3 + z and M(z) = Mo + M,z =
-2.75 + 3z . Thus the compensator is
0.25z 3z - 2.75
[C](z) -C2(z)1 - (13 .85)
[z - 3.3 z - 3.3
This completes the design . As for the unity-feedback configuration, this design also
involves one pole-zero cancellation . However, this pole is chosen by the designer .
We have chosen it as z = 0 for this problem .
From this example, we see that the design procedures in Chapter 10 for analog
systems can be directly applied to design digital systems .
This chapter introduced two approaches to design digital compensators . The first
approach is to design an analog compensator and then transform it into a digital one .
We discussed six methods in Section 13.2 to carry out the transformation . Among
these, the step-invariant and bilinear transformation methods appear to yield the best
results . The second approach is to transform an analog plant into an equivalent step-
invariant digital plant . We then design digital compensators directly by using the
root-locus method, state-space method, or linear algebraic method . The Bode design
method cannot be directly applied because G(e' ") is an irrational function of w
and because w is limited to (- IT/T, ar/T ). However, it can be so used after a bilinear
transformation.
A question may be raised at this point: Which approach is simpler and yields a
better result? In the second approach, we first discretize the plant and then carry out
design . If the result is found to be unsatisfactory, we must select a different sampling
period, again discretize the analog plant, and then repeat the design . In the first
approach, we carry out analog design and then discretize the analog compensator .
If the discretized compensator does not yield a satisfactory result, we select a dif-
ferent sampling period and again discretize the analog compensator . There is no
need to repeat the design ; therefore, the first approach appears to be simpler . There
is another problem with the second approach . In discretization of an analog plant, if
the sampling period is small, the resulting digital plant will be more prone to nu-
merical error. In analog design, the desired pole region consists of a good portion
of the left half s-plane . In digital design, the desired pole region consists of only a
portion of the unit circle of the z-plane . Therefore, digital design is clustered in a
much smaller region and, consequently, the possibility of introducing numerical error
is larger . In conclusion, the first approach (discretization after design) may be simpler
than the second approach (discretization before design) .
PROBLEMS
13 .1 . Find the impulse-invariant and step-invariant digital compensators of the fol-

lowing two analog compensators with sampling period T = 1 and T = 0.5 :
2
a
. (s + 2)(s + 4)
2s - 3
b.
s+4
13 .2 . Use state-variable equations to find step-invariant digital compensators for
Problem 13 .1.
13 .3 . Use forward-difference, backward-difference, and bilinear transformation
methods to find digital compensators for Problem 13 .1.
13 .4 . Use state-variable equations to find forward-difference digital compensators
for Problem 13.1 .
13 .5 . a. Find the controllable-form realization for the transfer function in (13 .22).
b . Show that the realization and the state-variable equation in (13 .24) are
equivalent . Can you conclude that the transfer function of (13 .24) equals
(13 .22)? [Hint : The similarity transformation is
T T
0 T2
13.6 . Use pole-zero mapping to find digital compensators for Problem 13 .1 .
13.7 . Use the analog initial value theorem to show that the step response of (13 .31)
has the property y(0) = 0, y(0) = 0, and y(O) = 0 . Thus the value of y(T)
is small if T is small.
13.8 . a. Compute the steady-state response of the analog compensator in (13 .31)
due to a unit-step input .

PROBLEMS 549
b . Compute the steady-state response of the digital compensator in (13 .33)

with N(z) = (z + 1)2 due to a unit-step sequence.
c. Are the results in (a) and (b) the same? If not, modify b in (13 .33) so that
they have the same steady-state responses .
13.9. Find equivalent digital plant transfer functions for the following analog plant
transfer functions :
1
a 2
s
10
b
. s(s + 2)
8
C.
s(s + 4)(s - 2)
What are their pole-zero excesses?
13 .10. Consider an analog plant with transfer function
1
G(s) _
s(s+ 1 +jl0)(s+ 1 -j10)
Find its equivalent digital plant transfer functions for T = 0.27r and T = 0.1 .
Can they both be used in design?
13.11 . Use (13.61) to establish (13 .62).
13.12 . Given a digital plant
Z + 1
G(z) =
(z - 0.3)(z - 1)
Use the root-locus method to design a system to meet (1) position error = 0,
(2) overshoot :10%, (3) settling time <- 5 seconds, and (4) rise time as small
as possible.
13.13 . Design a state feedback and a full-dimensional state estimator for the plant
in Problem 13 .12 such that all poles of the state feedback and state estimator
are at z = 0.
13.14 . In the design of a reduced-dimensional state estimator for the digital plant
in Example 13 .7.1, show that if F = 0 and g = 1, then the equation TA -
FT = gc has no solution T.
13.15. Design a state feedback and a reduced-dimensional state estimator for the
plant in Problem 13 .12 such that all poles of the state feedback are at z = 0 .
Can you choose the eigenvalue of the estimator as z = 0? If not, choose it
as z = 0.1.

13 .16 . Repeat Problem 13 .15 using the linear algebraic method . Are the results the
same? Which method is simpler?
13 .17. Consider the plant in Problem 13 .12 . Use the unity-feedback configuration to
find a compensator such that all poles are located at z = 0 . What zero, if any,
is introduced in G0(z)? Will G0(z) track any step-reference input without an
error?
13.18 . Consider the plant in Problem 13 .12. Implement the following overall transfer
function
0.5(z z+ 1)
G0(z) =
Z
Will this system track any step-reference input?

13 .19. Consider
z+2 h(z+2)
G(z) =
(z - 1)(z - 3) G(z) z(z + 0 .2)
Find h so that G0(1) = 1 . Implement G0(z) in the unity-feedback configuration
and the two-parameter feedback configuration . Are they both acceptable in
practice?
PID Controllers
14 .1 INTRODUCTION
In this text, we have introduced a number of methods to design control systems . If

a plant is single-variable and can be modeled as linear time-invariant and lumped,
then the methods can be used to design a good control system . If a plant cannot be
so modeled (for example, if it is nonlinear), then the design methods cannot be used .
What should we do in this situation? Basically, there are three approaches :
1. Use nonlinear techniques . Because of the complexity of nonlinear problems, no
simple and general design method is available to design all nonlinear control
systems. However, a number of special techniques such as the phase-plane
method, the describing function method, and the Lyapunov stability method are
available for analysis and design . This is outside the scope of this text . The
reader is referred, for example, to References [6, 56] .
2. Find a linearized model, even if it is very crude, and carry out linear design,
then apply the computed compensator to the actual plant . Clearly, there is no
guarantee that the resulting system will be satisfactory . Then adjust the param-
eters of the compensator and hope that a satisfactory system can be obtained . If
the nonlinear plant can be simulated on a computer, carry out the adjustment on
the computer . This method may become unmanageable if the number of param-
eters to be adjusted is large .
3 . Use the proportional-integral-derivative (PID) controller, adjust the parameters,
and hope that a satisfactory overall system can be obtained . This approach is
widely used in industry, and a number of PID controllers are available on the
market .
551

552 CHAPTER 14 PID CONTROLLERS
In this chapter, we discuss first analog PID controllers and the adjustment of
their parameters . We then discuss digital implementations of these controllers . Be-
fore proceeding, we mention that what will be discussed for nonlinear control sys-
tems is mostly extrapolated from linear control systems . Therefore, basic knowledge
of linear control systems is essential in studying nonlinear systems . We also mention
that, unlike linear systems, a result in a nonlinear system in a particular situation is
not necessarily extendible to other situations . For example, the response of the non-
linear system in Figure 6 .14 due to r(t) = 0.3 approaches 0 .3, but the response due
to r(t) = 4 X 0.3 does not approach 4 X 0 .3. Instead, it approaches infinity . Thus
the response of nonlinear systems depends highly on the magnitude of the input . In
this chapter, all initial conditions are assumed to be zero and the reference input is
a unit-step function .
14 .2 PID CONTROLLERS IN INDUSTRIAL PROCESSES
Consider the system shown in Figure 14 .1. It is assumed that the plant cannot be
adequately modeled as a linear time-invariant lumped system . This is often the case
if the plant is a chemical process . Most chemical processes have a long time delay
in responses and measurements ; therefore, they cannot be modeled as lumped sys-
tems . Chemical reactions and liquid flow in pipes may not be describable by linear
equations. Control valves are nonlinear elements ; they saturate when fully opened .
Therefore, many chemical processes cannot be adequately modeled as linear time-
invariant and lumped systems .
Many industrial processes, including chemical processes, can be controlled in
the open-loop configuration shown in Figure 14 .1(a). In this case, the plant and the
controller must be stable . The parameters of the controller can be adjusted manually
or by a computer . In the open-loop system, the parameters generally must be read-
justed whenever there are changes in the set point and load . If we introduce feedback
around the nonlinear plant as shown in Figure 14 .1(b), then the readjustment may
not be needed . Feedback may also improve performance, as in the linear case, and
make the resulting systems less sensitive to plant perturbations and external disturb-
ances . Therefore, it is desirable to introduce feedback for nonlinear plants .
r(t) Nonlinear Y (t)

Controller plant
(a)
Nonlinear Y (t)
Controller plant
(b)
Figure 14 .1 (a) Open-loop nonlinear system . (b) Unity-feedback nonlinear system.
14 .2 PID CONTROLLERS IN INDUSTRIAL PROCESSES 553
The design of nonlinear feedback systems is usually carried out by trial and
error. Whenever we use a trial-and-error method, we begin with simple control
configurations such as the unity-feedback configuration shown in Figure 14 .1(b) and
use simple compensators or controllers . If not successful, we then try more complex
configurations and controllers . We discuss in the following some simple controllers .
Proportional Controller
The transfer function of a proportional controller is simply a gain, say kP. If the input
of the controller is e(t), then the output is u(t) = kpe(t) or, in the Laplace transform
domain, U(s) = kPE(s) . To abuse the terminology, we call a nonlinear system stable'
if its output excited by a unit-step input is bounded and approaches a constant as
t - oo . Now if a plant is stable, as is often the case for chemical processes, the
feedback system in Figure 14.1 will remain stable for a range of kP. As kP increases,
the unit-step response may become faster and eventually the feedback system may
become unstable . This is illustrated by an example .
Example 14 .2 .1
Consider the system shown in Figure 14 .2. The plant consists of a saturation non-
linearity, such as a valve, followed by a linear system with transfer function
-s + 1
G(s) =
(5s + 1)(s + 1)
The controller is a proportional controller with gain kP. We use Protoblock 2 to simu-
late the feedback system . Its unit-step responses with kP = 0 .1, 1, 4 .5, 9, are shown
i G (s)
L A
Plant
Figure 14.2 Unity-feedback system with nonlinear plant .
'The stability of nonlinear systems is much more complex than that of linear time-invariant lumped
systems . The response of a nonlinear system depends on whether or not initial conditions are zero and
on the magnitude of the input . Therefore, the concept of bounded-input, bounded-output stability defined
for linear systems is generally not used for nonlinear systems . Instead, we have asymptotic stability,
Lyapunov stability, and stability of limit cycles . This is outside the scope of this text and will not be
discussed.
2 A trademark of the Grumman Corporation . This author is grateful to Dr . Chien Y . Huang for carrying
out the simulation .

1 .4
1 .2
0.8
0 .6
0 .4
0 .2
-0.2
0 2 4 6 8 10 12 14 ' 16 18 20
Figure 14.3 Unit-step responses .
in Figure 14.3. We see that as kP increases, the response becomes faster but more
oscillatory . If kP is less than 9, the response will eventually approach a constant, and
the system is stable . If kP is larger than 9, the response will approach a sustained
oscillation and the system is not stable . Recall that we have defined a nonlinear
system to be unstable if its unit-step response does not approach a constant . The
smallest kP at which the system becomes unstable is called the ultimate gain.
For some nonlinear plants, it may be possible to obtain good control systems
by employing only proportional controllers . There is, however, one problem in using
such controllers . We see from Figure 14 .3 that, for the same unit-step reference
input, the steady-state plant outputs are different for different kP. Therefore, if the
steady-state plant output is required to be, for example, 1, then for different kP, the
magnitude of the step-reference input must be different . Consequently, we must
adjust the magnitude of the reference input or reset the set point for each kP. There-
fore, manual resetting of the set point is often required in using proportional
controllers .
The preceding discussion is in fact extrapolated from the linear time-invariant
lumped systems discussed in Section 6 .3.2. In the unity-feedback configuration in
Figure 6 .4(a), if the plant is stable and the compensator is a proportional controller,
then the loop transfer function is of type 0 . In this case, the position error is different
from zero and calibration or readjustment of the magnitude of the step-reference
input is needed to have the plant output reach a desired value . If the loop transfer

function is of type 1, then the position error is zero and the readjustment of the
reference input is not necessary . This is the case if the controller consists of an
integrator, as will be discussed in the next paragraph .
Integral Controller
If the input of an integral controller with gain ki is e(t), then the output is
u(t) = ki Jo e(T)dT (14.1)
or U(s) = kiE(s)/s . Thus the transfer function of the integral controller is ki/s . For
linear unity-feedback systems, if the forward path has a pole at s = 0 or of type 1,
and if the feedback system is stable, then the steady-state error due to any step-
reference input is zero for any ki. We may have the same situation for nonlinear
unity-feedback systems . In other words, if we introduce an integral controller in
Figure 14 .1(b) and if the system is stable, then for every ki, the steady-state error
may be zero, and there is no need to reset the set point . For this reason, the integral
control is also called the reset control and ki is called the reset rate .
Although the integral controller will eliminate the need of resetting the set point,
its presence will make the stabilization of feedback systems more difficult . Even if
it is stable, the speed of response may decrease and the response may be more
oscillatory, as is generally the case for linear time-invariant lumped systems . In
addition, it may generate the phenomenon of integral windup or reset windup . Con-
sider the system shown in Figure 14 .4(a) with an integral controller and a valve
saturation nonlinearity with saturation level u,,,. Suppose the error signal e(t) is as
shown in Figure 14.4(a) . Then the corresponding controller output u(t) and valve
output u(t) are as shown . Ideally, if the error signal changes from positive to negative
at to, its effect on u(t) will appear at the same instant to. However, because of the
integration, the value of u(t) at t = to is quite large as shown, and it will take awhile,
say until time t2, for the signal to unwind to um. Therefore, the effect of e(t) on
u(t) will appear only after t2 as shown in Figure 14 .4(a) and there is a delay of
t2 - to. In conclusion, because of the integration, the signal u(t) winds up over the
saturation level and must be unwound before the error signal can affect on u(t). This
is called integral windup or reset windup .
If the error signal e(t) in Figure 14.4 is small and approaches zero rapidly,
integral windup may not occur . If integral windup does occur and makes the feedback
system unsatisfactory, it is possible to eliminate the problem . The basic idea is to
disable the integrator when the signal u(t) reaches the saturation level . For example,
consider the arrangement shown Figure 14.4(b). The input e(t) of the integral con-
troller is e(t) = e(t) X w(t), where w(t) = 1 if Ju(t)l < u, and w(t) = 0 if Ju(t)l
> u,,,. Such w(t) can be generated by a computer program or by the arrangement
shown in Figure 14 .4(b) . The arrangement disables the integrator when its output
reaches the saturation level, thus u(t) will not wind up over the saturation level, and
when e(t) changes sign at to, its effect appears immediately at u(t) . Hence, the per-
formance of the feedback system may be improved . This is called antiwindup or
integral windup prevention.

e (t) U u
>. t I I >- t
0 t t0 t2
Um
U If
S
Plant
m
(a)
e (t) U U
um I I\ um
.- t i , t
0 0 t
1 t0 t0
e Um
u Y
r 0-`- Plant
W O 1
0 b <
t
F>u m
(b)
Figure 14.4 (a) Integral windup. (b) Anti-integral windup .
Derivative Controller
If the input of a derivative controller with derivative constant kd is e(t), then its
output is kd de(t)/dt or, in the Laplace transform domain, kd sE(s) . Therefore, the
transfer function of the derivative controller is kd s . This is an improper transfer
function and is difficult to implement . In practice, it is built as
kd s
(14 .2)
I + kd s
N

where N, ranging from 3 to 10, is determined by the manufacturer and is called the
taming factor. This taming factor makes the controller easier to build . It also limits
high-frequency gain ; therefore, high-frequency noise will not be unduly amplified .
For control signals, which are generally of low frequency, the transfer function in
(14 .3) can be approximated as kd s.
The derivative controller is rarely used by itself in feedback control systems .
Suppose the error signal e(t) is very large and changes slowly or, in the extreme
case, is a constant . In this case, a good controller should generate a large actuating
signal to force the plant output to catch up with the reference signal so that the error
signal will be reduced . However, if we use the derivative controller, the actuating
signal will be zero, and the error signal will remain large . For this reason, the deriv-
ative controller is not used by itself in practice . If we write the derivative of e(t) at
t=to as
de(t) e(to + a) - e(to)
km
dt t-1o a-
.o a
with a > 0, then its value depends on the future e(t). Thus the derivative or rate
controller is also called the anticipatory controller .
The combination of the proportional, integral, and derivative controllers is called
a PID controller . Its transfer function is given by
kds=kp ( 1+Ts +Td sl

+sl +
kr (14.3)
/~
where Tz is called the integral time constant and Td the derivative time constant . The
PID controller can be arranged as shown in Figure 14 .5(a). This arrangement is
discussed in most control texts and is called the "textbook PID controller" in Ref-
erence [3] . The arrangement is not desirable if the reference input r contains dis-
continuities, such as in a step function . In this case, e(t) will be discontinuous and
its differentiation will generate an impulse or a very large actuating signal . An al-
ternative arrangement, called the derivative-of-output controller, is shown in Figure
14.5(b) where only the plant output y(t) is differentiated . In this arrangement, the
discontinuity of r will appear at u through the proportional gain but will not be
r
PID
PD
(a) (b) (c)
Figure 14 .5 (a) Textbook PID controller . (b) Derivative-of-output controller . (c) Set-point-
on-1 controller .

amplified by differentiation . Yet another arrangement is shown in Figure 14 .5(c),

where the error e is integrated . It is called the set-point-on-I-only controller. In this
case, the discontinuity of r will be smoothed by the integration .
There are three parameters to be adjusted or tuned in a PID controller . It may
be tuned by trial and error . We may first set k; = 0 and kd = 0 and vary kP to see
whether a satisfactory feedback system can be obtained . If not, we may then also
vary k7. If we still cannot obtain a satisfactory system, we then vary all three param-
eters. This is a trial-and-error method .
14.2.1 Rules of Ziegler and Nichols

The PID controller has been widely used in industry, especially in chemical proc-
esses . Therefore, various rules for adjusting the three parameters have been devel-
oped . We discuss in this subsection two sets of rules developed by Ziegler and
Nichols [69] . They, found that if the unit-step response of a system is of the form
shown in Figure 14 .6(a)-that is, the second overshoot is roughly 25% of the first
overshoot-then the integral of the absolute error (IAE) or
J le(t)I dt Jr(t) - y(t)I dt (14.4)

= I = f
with r(t) = 1, is minimized . This is called the quarter-decay criterion. Ziegler and
Nichols used this criterion to develop their rules . These rules were developed mainly
from experiment .
Closed-Loop Method Consider the system shown in Figure 14 .1(b). The controller
consists of only a proportional controller with gain kP. It is assumed that the system
is stable for 0 < kP < k and the unit-step response of the system with kP = k is
of the form shown in Figure 14 .6(b). It has a sustained oscillation with period T,,.
We call k the ultimate gain and T the ultimate period. Then the rules of Ziegler
and Nichols for tuning the PID controller are as shown in Table 14 .1 .
e (t) e (t)
t
(a) (b)
Figure 14 .6 (a) Quarter-decay response . (b) Ultimate gain and period .

Table 14 .1 Closed-Loop Method
Controller k, Td
P 0.5Ku
PI 0.4Ku 0 .8Tu
PID 0.6Ku 0 .5Tu 0 .12T
Open-Loop Method In this method, we measure the unit-step response of the

plant without closing the loop . It is assumed that the response is of the form shown
in Figure 14.7 . If the response is not of that form, the method is not applicable . The
response is approximated by straight lines, with K, L, and T indicated as shown.
Then the response can be approximated by the unit-step response of a system with
transfer function
Ke -L s
G (s) = Ts + 1 (14.5)
This transfer function contains an irrational function e -LS, which is due to the time
delay of L seconds. The rest is a linear time-invariant lumped system with gain K
and time constant T. Then the rules of Ziegler and Nichols for tuning the PID con-
troller are as shown in Table 14 .2. These two sets of rules are quite simple and can
be easily applied . Certainly, there is no guarantee that they will yield good control
systems, but they can always be used as an initial set of parameters for subsequent
adjustment or tuning .
14 .2 .2 Rational Approximations of Time Delays

Industrial processes often contain time delays, which are also called dead times or
transport lags . The transfer function of the plants will then contain the factor e -L
as shown in (14 .5). The function ea can be expressed as
a a2 a3 a"
e" = 1 +-+-+-+ +-+
1! 2! 3! n!
0 t
->I L Imo- T ---I
Figure 14.7 Unit-step response .

Table 14.2 Open-Loop Method
Controller k,, T; Td
P TL
PI 0 .9TL 0.3L
PID 1 .2TL 2L 0.5L
where n! = n(n - 1)(n - 2) . . . 2 1 . Taking the first two or three terms, we

obtain the following rational function approximations :
1 N 1
e -Ls = (14 .6a)
e Ls 1 + Ls
Ls
-Ls/2 1 - 2
e 2 - Ls
e -Ls = e Ls/2 (14 .6b)
2 + Ls
1 + Ls
2
and
2
e Ls12z 1 - Ls + (Ls)
e -Ls - _ 2 8 8 - 4Ls + L2s2
e /2 1 + Ls + (Ls) 2 = 8 + 4Ls + L 2 s 2 (14 .6c)
2 8
Note that these approximations are good for s small or s approaching zero . Because
s small governs the time response as t approaches infinity, as can be seen from the
final-value theorem, these equations give good approximations for the steady-state
response but, generally, poor approximations for the transient response . For example,
Figure 14.8 shows the unit-step responses of
e- 3s
G(s) _
2s + 1
and its rational function approximations
1 2 3s
G,(s) = ( 2s + 1)(1 + 3s) G2(s)
(2s + 1)(2 + 3s)
8- 12s+9s 2
G3(s) _
(2s + 1)(8 + 12s + 9S 2)
All of them approach the same steady-state value, but their transient responses are
quite different. The unit-step response of G 3 (s) is closest to the one of G(s).
Once a plant transfer function with time delay is approximated by a rational
transfer function, then all methods introduced in this text can be used to carry out
the design . This is the second approach mentioned in Section 14 .1 .

14.3 PID CONTROLLERS FOR LINEAR TIME-INVARIANT LUMPED SYSTEMS 561
0.8
0.6
0.4
0.2
G 2 (s)
-0.2
0 2 4 6 8 10 12 14 16 18 20
Figure 14 .8 Unit-step response of rational function approximations .
14 .3 PID Controllers for Linear Time-Invariant Lumped Systems
The PID controller certainly can also be used in the design of linear time-invariant
lumped systems . In fact, the proportional controller is always the first controller to
be tried in using the root-locus and Bode plot design methods . If we use the PID
controller for linear time-invariant lumped systems, then the tuning formula in Table
14.1 can also be used. In this case, the ultimate gain K u and ultimate period T can
be obtained by measurement or from the Bode plot or root loci . If the Bode plot of
the plant is as shown in Figure 14 .9(a) with phase-crossover frequency (Op and gain
margin a dB, then the ultimate gain and ultimate period are given by
= 21r
Ku = l0a120 Tu
lop
If the root loci of the plant are as shown in Figure 14 .9(b), then from the intersection
of the root loci with the imaginary axis, we can readily obtain the ultimate gain K
and the ultimate period as T,, = 27r/co, Once Ku and T are obtained, then the tuning
rule in Table 14 .1 can be employed.
Even though PID controllers can be directly applied to linear time-invariant
lumped systems, there seems no reason to restrict compensators to them . The transfer
function of PI controllers is
k. kps + k
kp +==
S S
This is a special case of the phase-lag compensator with transfer function k(s + b)/
(s + a) and b > a ? 0 . Therefore, phase-lag compensators are more general than
PI compensators and it should be possible to design better systems without restricting
a = 0 . This is indeed the case for the system in Example 8 .10.1 . See the responses

dB Res
K
X co u
> IM S
0
CO
I P
(a) (b)
Figure 14.9 (a) Bode plot . (b) Root loci .
in Figure 8 .39 . The phase-lag controller yields a better system than the PI controller
does.
If we use the more realistic derivative controller in (14 .2), then the transfer
function of PD controllers is
Nk,
(N+k,)Is+
kds kd(N + k)
P k(s + ac)
k, + = (14 .7)
1 + kd s
N
S + Nk s + a
d
where k = N + kk, a = N/k d, and c = kp/(N + kp) < 1 . This is a special case
of the phase-lead compensator with transfer function k(s + b)/(s + a) and
0 :5 b < a . Similarly, the transfer function of realistic PID controllers is a special
case of the following compensator of degree 2 :
k(s 2 + b1s + bo)
(14 .8)
S 2 + as + ao
Thus if we use general controllers, then the resulting systems should be at least as
good as those obtained by using PID controllers . Furthermore, systematic design
methods are available to compute general controllers . Therefore, for linear time-
invariant lumped systems, there seems no reason to restrict controllers to PID
controllers .
There is, however, one situation where PID controllers are useful even if a plant
can be adequately modeled as an LITL system. PID controllers can be built using
hydraulic or pneumatic devices ; general proper transfer functions, however, cannot
be so built . If control systems are required to use hydraulic systems, such as in all
existing Boeing commercial aircrafts, then we may have to use PID controllers . A

14 .4 DIGITAL PID CONTROLLERS 563
new model of AirBus uses control by wire (by electrical wire) rather than hydraulic
tube . In such a case there seems no reason to restrict controllers to PID controllers,
because controllers with any proper transfer functions can be easily built using elec-
trical circuits as discussed in Chapter 5 . Using control by wire, the controller can be
more complex and the performance of the system can be improved .
14 .4 DIGITAL PID CONTROLLERS
In this section we discuss digital implementations of PID controllers . As in Chapter

13, we use G(s) to denote analog PID controllers and G(z), digital PID controllers .
Consider the analog PID controller discussed in (14 .3)
G(s) = kp +1 + Tds (14.9)

C1 T;s J
We have discussed a number of discretization methods in Section 13 .2, all of which

can be used to discretize (14 .9) . If we use the forward difference in (13 .19) for both
the integrator and differentiator, then (14 .9) becomes
Td (z 7-1)1
G(z) = kP 1 + T(z T 1) + J (14.10)
1
This is the simplest digital implementation of the PID controller . Another possibility
is to use the trapezoidal approximation in (13 .27) for the integrator and the backward
difference in (13 .25) for the differentiator ; then (14.9) becomes
T z + 1 Td (z - 1)1
+ 2T,z-1 + T Jz
_1)l
+ T 2 (1 - z- 1 ) + Td(1 - z
J (14 .11
2Tt 1 - z- 1 T )
T + T I + Td (l - Z 1 )1
2Tj T, 1 - z -1 T
If we define
T
(Proportional gain) (14 .12a)
2Tt
(Integral gain) (14 .12b)
(Derivative gain) (14 .12c)

T

G(z) kp + 1 - z 1 +
kd (1 - z -1) ( 14 .13)
This is one commonly used digital PID controller . Note that digital kp differs from
analog kP by the amount kk T/2T 1 which is small if the sampling period T is small .
However, if T is small, then the derivative gain kd will be large . This problem will
not arise if the analog differentiator is implemented as in (14 .2). In this case, the
transfer function of analog PID controllers becomes
Td s
G(s) = kP + I + (14 .14a)
T;s Tds
I +
N
+ I + N(s - 0)
= k,, (14 .14b)
Tis N
S- --
Td
If we use the impulse-invariant method for the integrator and the pole-zero mapping
for the differentiator, then we have
T + N(z - 1)
G(z) = kp I + (14 .15a)
Tj(z - 1) z - a
with
NT
T,,
a = e (14 .15b)
If we use the forward difference for the integrator and the backward difference for
the differentiator, then we have
G(z)=kP I +T
T.(z
Td NTz- I
1 ) +1+T
(14 .16a)
with
Td
(14 .16b)
Td + NT
This is a commonly used digital PID controller . We mention that if T is very small,
then (14 .15) and (14.16) yield roughly the same transfer function . In (14 .16a), if we
use forward difference for both the integrator and differentiator, then the resulting
digital differentiator may become unstable. This is the reason we use forward dif-
ference for the integrator and backward difference for the differentiator .

14 .4 DIGITAL PID CONTROLLERS 565
The digital PID controllers in (14 .10), (14 .13), (14.15), and (14 .16) are said to
be in position form . Now we develop a different form, called velocity form . Let the
input and output of the digital PID controller in (14 .13) be e(k) : = e(kT) and
u(k) : = u(kT) . Then we have
U(z) + I - z -, + k,(1 - z-1) E(z) (14 .17)

I kp I
(1 - z - ')U(z) = kp(1 - z - ')E(z) + kE(z) + k d(l - 2z' + Z-2)E(z)
In the time domain this becomes
u(k) - u(k - 1) = kd [e(k) - e(k - 1)] + kie(k)
(14 .18)
+ kd[e(k) - 2e(k - 1) + e(k - 2)]
In the unity-feedback configuration, we have
e(k) = r(k) - y(k) (14 .19)
where r(k) is the reference input sequence and y(k) is the plant output . If the reference
input is a step sequence, then r(k) = r(k - 1) = r(k - 2) and
e(k) - e(k - 1) = r(k) - y(k) - [r(k - 1) - y(k - 1)]
(14.20)
_ - [y(k) - y(k - 1)l
and
e(k) - 2e(k - 1) + e(k - 2) = r(k) - y(k) - 2[r(k - 1) - y(k - 1)]
+ r(k - 2) - y(k - 2) (14 .21)
- [y(k) - 2y(k - 1) + y(k - 2)]
The substitution of (14 .19) through (14 .21) into (14.18) yields
u(k) - u(k - 1) = - kp[y(k) - y(k -I)] + k [r(k)
(14 .22)
- y(k)] - kd[e(k) - 2y(k - 1) + y(k - 2)]
The z-transform of (14 .22) is
(1 - z - ')U(z) = -kp(1 - z - ')Y(z) + J [R(z) - Y(z)] - kd(1 - z -1 )2 Y(z)
which implies
U(z) _ - kkY(z) + I k`z _ i E(z) - kd(1 - z - ')Y(z)
This is plotted in Figure 14.10 . We see that only the integration acts on the error
signal ; the proportional and derivative actions act only on the plant output . This is
called the velocity form PID controller. This is the set-point-on-I-only controller
shown in Figure 14.5(c) .

r(k) e(k) u(k) y(k)

Plant
kd (1 -z i )
Figure 14.10 Velocity-form PID controller .
If the sampling period is small, then the effects of analog and digital PID con-
trollers will be close ; therefore, the tuning methods discussed for the analog case
can be used to tune digital PID controllers . Because the dynamics of industrial
processes are complex and not necessarily linear, no analytical methods are available
to determine parameters of PID controllers ; therefore, their determinations will in-
volve trial and error . At present, active research has been going on to tune these
parameters automatically . See, for example, References [3, 31] .

The Laplace
Transform
A.1 DEFINITION
In this appendix, we give a brief introduction of the Laplace transform and discuss
its application in solving linear time-invariant differential equations . The introduc-
tion is not intended to be complete ; it covers only the material used in this text .
Consider a function f (t) defined for t ? 0. The Laplace transform of f (t),
denoted by F(s), is defined as
F(s) :_ ~[f(t)] := j f(t)e -st dt (A.1)

o
where s is a complex variable and is often referred to as the Laplace-transform
variable . The lower limit 0 - of the integral denotes that the limit approaches zero
from a negative value. There are two reasons for using 0 - rather than 0 as the lower
limit, as will be explained later .
Example A.1 .1
Consider f(t) = e- at, for t > 0 . Its Laplace transform is
F(s) = c2 [e-at ] = J ~ e-ate-st dt = - 1 e -(Q+s)t
o- s + a t=0 -
567

568 APPENDIX A THE LAPLACE TRANSFORM
- 1 [ e -(a+s)t - e -(a+s)tl (A.2)

c= r=0 - ]
s+a
-1 1
s+ a [0 - 1]= s+a
where we have used e-(s + a)t I, = 0. This holds only if Re s > Re (-a), where
Re stands for the real part . This condition, called the region of convergence, is often
disregarded, however . See Reference [18] for a justification . Thus, the Laplace trans-
form of e - a` is I/(s + a). This transform holds whether a is real or complex .
Consider the two functions defined in Figure A .1 (a) and (b). The one in Figure
A.1(a) is a pulse with width e and height 1/E . Thus the pulse has area 1, for all
e > 0. The function in Figure A .1(b) consists of two triangles with total area equal
to 1, for all E > 0., The impulse or delta function is defined as
8(t) = lim SE(t)
E->0
where 8E(t) can be either the function in Figure A .1 (a) or that in (b) . The impulse is
customarily denoted by an arrow, as shown in Figure A.1(c). If the area of the
function in Figure A.1 (a) or in (b) is 1, then the impulse is said to have weight 1 .
Note that 8(t) = 0, for t 0 0. Because 8(0) may assume the value of -, if Figure
A.1 (a) is used, or 0, if Figure A .1(b) is used, 8(t) is not defined at t = 0 .
The impulse has the following properties
E
6(t)dt =f E
8(t)dt = 1 (A.3)
for every e > 0, and
f- f(t)8(t)dt =f f(t)8(t - 0)dt = f(t)It o = f(0) (A.a)
8E(t) 6 (t) S(t)
1
1 1
E
W
, t >- t t
0\ E 0 E 0
(a) 2 (b) (c)
Figure A . 1 Pulses and impulse

A.1 DEFINITION 569
if f (t) is continuous at t = 0. Strictly speaking, the impulse should be defined using

(A.3) and (A.4).
Using (A.4), the -Laplace transform of the impulse is
0(s) S(t)e - 't dt = e-s`lt_o = 1 (A.5)

:= -T[8(t)] = f -
Note that if the lower limit of (A.1) is 0 rather than 0 - , then the Laplace transform
of 8(t) could be 0 .5 or some other value . If we use 0 - , then the impulse will be
included wholly in the Laplace transform and no ambiguity will arise in defining
0(s) . This is one of the reasons for using 0 - as the lower limit of (A.1).
The Laplace transform is defined as an integral . Because the integral is a linear
operator, so is the Laplace transform-that is,
Z[a1f1(t) + a2f2(t)] = a1_T[f1(t)] + a2 E fa(t)]
for any constants a t and a 2 . We list in Table A . 1 some of the often used Laplace-
transform pairs .
Table A. 1 Laplace-Transform Pairs
f(t), t ? 0 F(s)
8(t) (impulse) 1
1
1 (unit-step function) -
S
n!
t" (n = positive integer) s' + i
1
e - " (a = real or complex)
s + a
t"e " n!
(S + a)" + 1
CJ
sin wt
S
S2 + w2
w
(s + a)2 + w2
s + a
e - "' cos wt
(s + a)2 + w2
f(t)e"' F(s - a)

A.2 INVERSE LAPLACE TRANSFORM-PARTIAL FRACTION EXPANSION
The computation of f (t) from its Laplace transform F(s) is called the inverse Laplace
transform. Although f(t) can be computed from
c+jOO
f(t) = 1 F(s)e''t ds
27rj J c-ice
the formula is rarely used in engineering . It is much simpler to find the inverse of
F(s) by looking it up in a table . However, before using a table, we must express F(s)
as a sum of terms available in the table .
Consider the Laplace transform
F(s) a)(s- (A.6)
D(s) (s - b)215(s)
where N(s) and D(s) are two polynomials with deg N(s) < deg D(s), where deg
stands for the degree . We assume that D(s) has a simple root at s = a and a repeated
root with multiplicity 2 at s = b, as shown in (A .6). Then F(s) can be expanded as
F(s) =ko + ka a + s kbi b + kb2 b)2

s (s (A .7)
+ (Terms due to the roots of D(s))
with
ko = F(co) (A.8a)
ka = F(s)(s - a)I S-a (A.8b)
kb2 = F(s)(s - b)21, = b (A .8c)
and
d [F(s)(s - b)2 ]IS=b

kb1 = ds (A.8d)
This procedure is called partial fraction expansion . Using Table A. 1, the inverse
Laplace transform of F(s) is
f(t) = k0 (t) + kaea` + kbi eb` + kb2 te bt + (Terms due to the roots of D(s))
Note that (A .8b) is applicable for any simple root ; (A.8c) and (A .8d) are applicable
for any repeated root with multiplicity 2. Formulas for repeated roots with multi-
plicity 3 or higher and alternative formulas are available in Reference [18].
Example A .2 .1
Find the inverse Laplace transform of
s3 -2s+3
F(s) =
s2(s + 1)(s - 2)

A.3 SOME PROPERTIES OF THE LAPLACE TRANSFORM 57 1
We expand it as
F(s)=ko+s+L 1 + s k2 2 s' + 2s2

+ k
with
ko = F(c) = 0
S3 - 2s + 3 _ - 1 + 2 + 3 _ -4
k, = F(s)(s + 1)I s , -
s2(s - 2) s=-1
1 (-3) 3
s 3 -2s+3 8 - 4 + 3
= 7
k2 = F(s)(s - 2)~ s=2 = _ -
5 2 (5 + 1) 43 12
s=2
s3 -2s+3 3
k32 = F(s)s21,=0 = = - 1.5
(s - 2)(s + 1) S=0 -2
d
k3i = ds [F(s) s2]l s =o
(s + 1)(s - 2)(3s 2 - 2) - (s3 - 2s + 3)(2s - 1) 7

(s + 1) 2 (s - 2)2 S-0 4
Thus the inverse Laplace transform of F(s) is
4 7 3 7
f(t) = et + 12 e tc - 2 t + 4
3
for t ? 0 .
Exercise A .2 .1
Find the inverse Laplace transforms of 2/s(s + 2) and (s - 1)/s(s + 1) .

[Answers : 1 - e-2r, t ? 0; -1 + 2e -t, t ? 0 .1
For a more detailed discussion of partial fraction expansion, see Reference [18] .
A .3 SOME PROPERTIES OF THE LAPLACE TRANSFORM
In this section we discuss some properties of the Laplace transform .
Differentiation in Time
Let F(s) = `S[f(t)] . Then
-T = sF(s) - f(0-) (A.9a)

Idt f (t) J

z
dt2 f(t) = s 2F(s) - sf(0 ) - f"')(0 ) (A.9b)
and, in general
n
= s?F(s) - sn - 'f(0 - )
è
C nn f(t)
f(n-1)(0 - )
(A.9c)
S n-2f(')(0-)
where f(`)(0 - ) denotes the ith derivative of f(t) at t J= 0- . We see that if

f(`)(0- ) = 0 for i = 0, 1, 2, . . . , then differentiation in the time domain is converted
into multiplication by s in the Laplace-transform domain .
Integration in Time
[It]
The Laplace transform of the integral of f(t) is given by
f(t)dt= s F(s)
Hence integration in the time domain is converted into division by s in the Laplace
transform domain .
Final-Value Theorem
Let f (t) be a function defined for t ? 0. If f (t) approaches a constant as t or
if sF(s) has no pole in the closed right half s-plane, then
t-.
lim f (t) = lim sF(s)
s-o
This theorem is applicable only if f(t) approaches a constant as t --> 00 . If f(t)
(A. 10)
becomes infinite or remains oscillatory as t -, then the theorem cannot be used .

For example, the Laplace transform of f(t) = et is F(s) = 1/(s - 1). Clearly we
have
lim sF(s) = 0
S-0
However, the function f (t) approaches infinity as t and the equality in (A . 10)
does not hold .
Initial-Value Theorem
Let f(t) be a function defined for t ? 0, and let F(s) be its Laplace transform . It is
assumed that F(s) is a rational function of s . If F(s) = N(s)/D(s) is strictly proper,
that is, deg D(s) > deg N(s), then
f(0 +) = lim sF(s)

A .4 SOLVING LTIL DIFFERENTIAL EQUATIONS 573
Consider
s + 3
F,(s)
s 3 +2s2 +4s+2
and
2s2 +3s+ 1
F2(S) = s 3 +2s 2 +4s+2
They are both strictly proper. Therefore the initial-value theorem can be applied .
The application of the theorem yields
f ,(O') = lim sF1 (s) = 0
and
2s 3 + 3s2 + s 2
f2 (0 + ) =
I'M sF2(s) = =
S + 2s2 + 4s + 2
The rational function F 3(s) = (2s + 1)/(s + 1) is not strictly proper. The application
of the initial-value theorem yields
s(2s + 1)
f3(0 + ) = lim sF3 (s) = lim = 00 (A.11)
S-= S_. S + 1
The inverse Laplace transform of

2s+1 -1
=2+
F3(s) _ s+ 1 s+ 1
is, using Table A .1,
f3(t) = 25(t) - e -`
Because 6(t) = 0 if t =A 0, we have f3(0+ ) = -1, which is different from (A.11).
Thus, if F(s) is not strictly proper, the initial-value theorem cannot be directly
employed.
A .4 SOLVING LTIL DIFFERENTIAL EQUATIONS
In this section we apply the Laplace transform to solve linear differential equations
with constant coefficients . This is illustrated by examples .
Example A .4 .1
Consider the first-order differential equation
d
dt y(t) + 2y(t) - 3 d u(t) + 2u(t) (A. 12)
d

The problem is to find y(t) due to the initial condition y(0 - ) = 2 and the input
u(t) = 1, fort ? 0. The application of the Laplace transform to (A .12) yields, using
(A.9a),
sY(s) - y(0 - ) + 2Y(s) = 3sU(s) - 3u(0- ) + 2U(s) (A.13)
The Laplace transform of u(t) is 1/s . The substitution of y(0 - ) = 2, u(0 - ) = 0,
and U(s) = 1 /s into (A .13) yields
(s + 2)Y(s) = y(0 - ) + 3sU(s) - 3u(0- ) + 2U(s)
2 2
= 2 + 3 + - = 5 + -
s s
which implies
_ 5 2
Y(s)
s + 2 + s(s + 2)
which can be simplified as, after partial fraction expansion of the second term,
_ 5 1 1 4 1
Y(s)
s + 2 + s s+2 s + 2 + s
Thus we have, using Table A . 1,
y(t) = 4e-21 + 1 (A. 14)
for t ? 0.
This example gives another reason for using 0 - rather than 0 as the lower limit
in defining the Laplace transform . From (A.14), we have y(0) = 5, which is different
from the initial condition y(0 - ) = 2. If we had used y(0) = 2, then confusion would
have arisen . In conclusion, the reason for using 0 - as the lower limit in (A.1) is
twofold: First, to include impulses at t = 0 in the Laplace transform, and second,
to avoid possible confusion in using initial conditions . If f (t) does not contain im-
pulses at t = 0 and is continuous at t = 0, then there is no difference in using either
0 or 0 - in (A .1).
Exercise A.4 .1
Find the solution of (A .12) due to y(0 -) = 2, u(0 -) = 0, and u(t) = 8(t) .
[Answer: y(t) = 38(t) - 2e-2', t ? 0.]
We give one more example to conclude this section .

A.4 SOLVING LTIL DIFFERENTIAL EQUATIONS 575
Example A .4 .2
Consider the second-order differential equation
d2 d _ d
dt2 y(t) + 2 y(t) + 5y(t)
dt dt u(t) (A .15)
It is assumed that all initial conditions are zero . Find the response y(t) due to
u(t) = e - `, t ? 0. The application of the Laplace transform to (A. 15) yields
s2Y(s) - sy(0 - ) - y (')(0- ) + 2(sY(s) - y(0 - )) + 5Y(s) = sU(s) - u(0 - )
which, because all initial conditions are zero, reduces to
(s2 + 2s + 5)Y(s) = sU(s)
or
S
Y(s) = s (A.16)
2 + 2s + 5 U(s)
The substitution of U(s) = è[u(t)] = 1/(s + 1) into (A.15) yields
s = S
Y(s) = (A.17)
(s2 + 2s + 5)(s + 1) (s + 1)(s + 1 - j2)(s + 1 + j2)
Thus, the remaining task is to compute the inverse Laplace transform of Y(s). We
expand it as
O _ k, k2 k3
Ys
S + 1 + s+ 1- j2+ s+ 1+ j2
with
S
ki = Y(s)(s + 1)Js _ _ 1 =
(s + 1 - j2)(s + 1 + j2) s=-i
1 + j2 - 1 - j2
k2 = Y(s)(s + 1 - J2)1 s - 1 +j2 = -
(j2)(j4) 8
and
-1 - j2 1 + j2
k3 = Y(s)(s + 1 + j2)1 s- _ 1 j2 = _
( -j2)( -j4) 8
The computation of k3 is in fact unnecessary once k2 is computed, because k3 equals
the complex conjugate of k2-that is, if k2 = a + jb, then
k3 =k2* := a-jb

In the subsequent development, it is simpler to use polar form .

re'' The polar form
of a + jb is
x = re
where r = \/ a 2 + b 2 and B = tan - ' (b/a) . For example, if x = - 2 + j 1, then
x = V4 + I e 1ta 'ni~~-2)1 = A eita-'(-0.5)
Because tan (-26 .5') = tan 153 .5 = -0.5, one may incorrectly write x as
e-j26 .5'. The correct x, however, should be
x = \ e '153.5'
as can be seen from Figure A.2. In the complex plane, x is a vector with real part
- 2 and imaginary part 1 as shown . Thus its phase is 153 .5 rather than - 26.5. In
computing the polar form of a complex number, it is advisable to draw a rough graph
to insure that we obtain the correct phase .
Now we shall express k2 and k3 in polar form as
8
v
k2 = _ e -635 _ _ e -ii .i
8
(radians)
= k3
Therefore the inverse Laplace transform of Y(s) in (A. 17) is
y(t) _ -0 .25e - t + e -j' . ' e-(I -j2)t + ell .1e-(i+j2)t (A.18)

8 8
If all coefficients of Y(s) are real, then y(t) must also be real . The second and third
terms in (A. 18) are complex-valued . However, their sum must be real . Indeed, we
have
y(t) = -0.25e -t + e -t [ej(2t-1 .1) + e -1(2t-1 .1)1

8
Im
A
Figure A.2 Vector.

A.5 TIME DELAY 577
Using the identity

eja + e-j'
cos a =
2
we obtain
y(t) = -0 .25e - ` + e t cos(2t - 1 .1) (A .19)

4
It is a real-valued function . Because the unit of to in cos(wt + 0) is in radians per
second, the phase or angle 0 should be expressed in radians rather than in degrees,
as in (A . 19) . Otherwise cot and 0 cannot be directly added.
Exercise A .4 .2
Find the solution of

d2y(t) du(t)
+ 2 dy(t) + 5y(t) = - u(t)
dt dt dt
due to u(t) = e -`, t ? 0. It is assumed that all initial conditions are zero .
[Answer : y(t) = 0.5e - ` + 0.5e- ` cos 2t + 0.5e - ` sin 2t.]
A .5 TIME DELAY
Consider a time function f (t) which is zero for t < 0. Then f (t - T) with T ? 0 is
a delay of f (t) by T seconds as shown in Figure A .3. Let F(s) be the Laplace
transform of f (t) . Then we have
2[f(t - T)] = e - TSF(s) (A .20)
f(t) f(t - T)
t t
0 0 T
Figure A .3 Time function and its time delay .

Indeed, the Laplace transform of f(t - T) is, by definition,

o o
e_Ts - T)
~[f(t - T)] fo
f(t - T)e - s` dt = f0 f(t - T)e-
s(t
dt
which becomes, by defining v = t - T and using f (t) = 0 for t < 0,

-T TS
f f(v)e-s dv = e- fo f(v)e-vs dv = e -TSF(s)
Y[f(t - T)] = e-Ts
This shows (A.20). For example, because 0(s) = è[S(t)] = 1, we have
1'[8(t - T)] = e -Ts (A.21)
Consider the pulse p(t) shown in Figure A.4. Let q(t) be a unit-step function,
that is, q(t) = I for t ? 0 and q(t) = 0, for t < 0 . Then p(t) can be expressed as
p(t) = q(t) - q(t - T) (A.22)
Thus we have
1- e-Ts = 1 - e-Ts
P(s) : _ ~[p(t)] = `[q(t)] - _T [q(t - T)] = (A.23)
S S S
This formula will be used in Chapter 12 .
p(t)
0 T
Figure A.4 Pulse .

Linear Algebraic
Equations
In this appendix, we give a brief introduction to matrices and solutions of linear

algebraic equations . It is introduced to the extent sufficient to solve the problems in
this text . We also discuss briefly the problem of ill-conditioning and numerical sta-
bility of algorithms on computer computation . For a more detailed discussion, the
reader is referred to References [15, 18] .
B. 1 MATRICES
A matrix is a rectangular array of elements such as

[a11 a12 alm
azl a22 a2m
A = = [a,~] (B.1)
and ant anm

All alb are real numbers . The matrix has n rows and m columns and is called an
n X m matrix . The element a;i is located at the ith row and jth column and is called
the (i, j)th element or entry. The matrix is called a square matrix of order n if m =
n, a column vector or simply a column if m = 1, a row vector or a row if n = 1 .
Two n X m matrices are equal if and only if all corresponding elements are the
same. The addition and multiplication of matrices is defined as follows :
579

580 APPENDIX B LINEAR ALGEBRAIC EQUATIONS
A + B = [ aij + bij]nxm
nXm nXm
C A = A c = [caij]nxm
1X1 nXm nXm ] X1
m
A B = E, aikbkj
n Xm mxp k=1 nXp
For a square matrix A = [a ij], the entries a ii , i = 1, 2, 3, . . . , on the diagonal
are called the diagonal elements of A . A square matrix is called a lower triangular
matrix if all entries above the diagonal elements are zero ; an upper triangular matrix
if all entries below the diagonal elements are zero . A square matrix is called a
diagonal matrix if all entries, except the diagonal entries, are zero . A diagonal matrix
is called a unit matrix if all diagonal entries equal 1 . The transpose of A, denoted
by A', interchanges the rows and columns . For example, if
all a12 a 11 a 21
2l 31
A = a21 a 22
a31 a32
then A'
C a12 a22 a32
Therefore, if A is n X m, then A' is m X n. To save space, the vector

1
-4
X = 3
is often written as x' = [1 -4 3] . In general, matrices do not commute, that is
AB 0 BA. But we can move a scalar such as cAB = AcB = ABc .
B .2 DETERMINANT AND INVERSE
The determinant of a square matrix of order n is defined as

det A = E (-1)ja 11 a 212 . . . anjn
where jl, j2, . . . , j,, are all possible orderings of the second subscripts 1, 2, . . . , n,
and the integer j is the number of interchanges of two digits required to bring the
ordering ji, j2, . . . , in into the natural ordering 1, 2, . . . , n . For example, we have
det C all a12 j = (- l)a11a22 + ( - 1 )'a12a21 = a11a22 a12a21

a21 a22
and
all a12 a13
det a21 a22 a23 = a11a22a33 + a12a23a31 + a13a21a32 a13a22a31
a31 a32 a33
- a12a21a33 - a11a23a32

B .2 DETERMINANT AND INVERSE 581
The determinant of upper or lower triangular matrices equals the product of

diagonal entries . For example, we have
0 0 all a12 a13
a22 0 = det 0 a22 a23 = a,la22a33
a32 a33 0 0 a33
Let A and B be square matrices . Then we have
det (AB) = det A det B (B.3)
where A and B have the same order, and
det C I= det [ = det A det B (B .4)

C B, A B ]
where A and B need not be of the same order. For example, if A is n X n and B is
m X m, then C is m X n and D is n X m. The composite matrices in (B .4) may be
called block diagonal matrices.
A square matrix is called nonsingular if its determinant is nonzero ; singular if
it is zero. For nonsingular matrices, we may define the inverse . The inverse of A,
denoted by A -1, has the property A - 'A = AA -1 = I . It can be computed by using
the formula
A-'
det A Add A det A [c".-
where
cij _ ( - 1)"i (Determinant of the submatrix of A
by deleting its jth row and ith column .)
The matrix [c,j] is called the adjoint of A. For example, we have
all a,2 1 a22 - a12
C a2, a22 I a,1a22 - a,2a21 I- a21 all
The inverse of 2 X 2 matrices is simple ; we compute the determinant, interchange
the diagonal elements, and change the signs of off-diagonal elements . The compu-
tation of the inverse of matrices of order 3 or higher is generally complicated .
The inverse of a triangular matrix is again a triangular matrix . For example, the
inverse of
0 0
A = a22 0
a31 a32 a33
is of the form
b11 0 0
B := A-1 = [b21 b22 0
b31 b32 b33

By definition, we have
b 0 0 0
all 0 1 0 0
[b 21 b22 0 a21 a22 0 = 0 1 0
b31 b32 b 33 a 313l 32 a33 0 0 1
Equating the element (1, 1) yields b11a11 = 1 . Thus we have b11 = a11'. Equating
elements (2, 2) and (3, 3) yields b22 = a2z' and b33 s' .
= a3 Equating element
(2, 1) yields
b21a11 + b22a21 = 0
which implies
b21 - - a21
- b22 - - a21
all a11a22
Proceeding forward, the inverse of triangular matrices can be easily computed .
The preceding procedure can be used to compute the inverse of block triangular
matrices . For example, we have
A D ' A -' a
(e.b)
[0 B] [ 0 B-'
where Aa + DB - ' = 0. Thus we have
a = -A - 'DB - '
B .3 THE RANK OF MATRICES
Consider the matrix in (B.1). Let air_ denote its ith row, that is,
air = [a,1 a .2 . . . aim]
The set of n row vectors in (B.1) is said to be linearly dependent if there exist n real
numbers a 1 , a 2 , . . . . an, not all zero, such that
a lair + a2a2r + + a nanr = 0 (B.7)
where 0 is a 1 X m vector with 0 as entries. If we cannot find n real numbers a 1 ,
a2 a n , not all zero, to meet (B .7), then the set is said to be linearly independent.
Note that if a; = 0 for all i, then (B .7) always holds . Therefore the crucial point is
whether we can find a ;, not all zero, to meet (B .7). For example, consider
alr 1 2 3 4
a2r = 2 - 1 0 0 (B .8)
a3r 2 4 6 8
We have
1 Xa, r +OXa 2r +(-0 .5)Xa 3r =[0 0 0 01

B.3 THE RANK OF MATRICES 583
Therefore the three row vectors in (B.8) are linearly dependent . Consider
air [1 2 3 4
a2r = 2 -1 0 0 (B.9)
la3r 1 2 0 0
We have
aiair + a2a2r + a3a3r
= [a i + 2a2 + a3 2a 1 - a 2 + 2a3 3a 1 4a1] ( B .10)
= [0 0 0 01
The only a i meeting (B .10) are ai = 0 for i = 1, 2, and 3. Therefore, the three row
vectors in (B .9) are linearly independent.
If a set of vectors is linearly dependent, then at least one of them can be ex-
pressed as a linear combination of the others . For example, the first row of (B .8) can
be expressed as
air =0Xa 2r +0.5 Xa3r
This first row is a dependent row . If we delete all dependent rows in a matrix, the
remainder will be linearly independent . The maximum number of linearly inde-
pendent rows in a matrix is called the rank of the matrix. Thus, the matrix in (B.8)
has rank 2 and the matrix in (B.9) has rank 3 .
The rank of a matrix can also be defined as the maximum number of linearly
independent columns in the matrix . Additionally, it can also be defined from deter-
minants as follows : An n X m matrix has rank r if the matrix has an r X r submatrix
with nonzero determinant and all square submatrices with higher order have deter-
minants zero . Of course, these definitions all lead to the same rank . A consequence
of these definitions is that for an n X m matrix, we have
Rank (A) min (n, m) (B .11)
An n X m matrix is said to have a full row rank if it has rank n or all its rows
are linearly independent . A necessary condition for the matrix to have a full row
rank is m ? n . Thus, if a matrix has fewer rows than columns, then it cannot have
a full row rank . If a square matrix has a full row rank, then it also has a full column
rank and is called nonsingular .
To conclude this section, we discuss the use of MATLAB to compute the rank
of matrices . Matrices are represented in MATLAB row by row, separated by semi-
colon. For example, the matrix in (B.8) is represented as
a=[1 2 3 4 ;2 -1 0 0 ;2 4 6 8] ;
The command
rank(a)
yields 2, the rank of the matrix in (B.8). The command
rank([1 2 3 4 ;2 -1 0 0 ;1 2 0 0])

yields 3, the rank of the matrix in (B .9). Thus the use of the computer software is
very simple.
The number of bits used in digital computers is finite, therefore numerical errors
always occur in computer computation . As a result, two issues are important in
computer computation. The first issue is whether the problem is ill conditioned or
not . For example, we have
0113 0.31333 _ 2
Rank 2 = 1 Rank
61 1 2]
We see that small changes in parameters yield an entirely different result . Such a
problem is said to be ill conditioned. Thus, the computation of the rank is not a
simple problem on a digital computer . The second issue is the computational method .
A method is said to be numerically stable if the method will suppress numerical
errors in the process of computation . It is numerically unstable if it will amplify
numerical errors and yield erroneous results . The most reliable method of computing
the rank is to use the singular value decomposition . See Reference [15] .
B .4 LINEAR ALGEBRAIC EQUATIONS
Consider the set of linear algebraic equations

a,,x1 + a12x2 + . . . + almxm = yl
a21x1 + a22x2 + . . . + a2mxm = Y2
a n1 xl + a n2 x2 + + anmxm = Yn
where a ll and y, are known and x; are unknown . This set of equations can be written
in matrix form as
Ax = y (B.12)
where
alm
a2m
A = [aid] x =
and ant a nm Yn_

The set has n equations and m unknowns . A is an n X m matrix, x is an m X 1
vector, and y is an n X 1 vector.
THEOREM B .1
For every y, a solution x exists in Ax = y if and only if A has a full row
rank.

B.4 LINEAR ALGEBRAIC EQUATIONS 585
For a proof of this theorem, see Reference [15] . We use an example to illustrate
its implication. Consider
x,
1 2 3 4 yl
x2
2 -1 0 0 = y2
x3
3 4 6 8 y3
x4
Although this equation has a solution (x' = [1 1 0 1]) for y' = [7 1 141,
it does not have a solution for y' = [0 0 1]. In other words, the equation has
solutions for some y, but not for every y . This follows from Theorem 13 .1 because
the 3 X 4 matrix does not have a full row rank .
Consider again (B .12) . It is assumed that A has a full row rank . Then for any
y, there exists an x to meet the equation . Now if n = m, the solution is unique . If
n < m or, equivalently, (B .12) has more unknowns than equations, then solutions
are not unique; (m - n) number of the parameters of the solutions can be arbitrarily
assigned . For example, consider
xl
Ax = x2 (B .13)
[-1 0 2] [-1]
x3
The matrix A in (B .13) has a full row rank . It has three unknowns and two equations,
therefore one of x 1 , x2, and x3 can be arbitrarily assigned . It is important to mention
that not every one of x1, x2, or x3 can be assigned . For example, if we assign x 2 =
3, then (B .13) becomes
2x1 + 3 - 4x3 = 3 -x1 + 2x3 = -1
or
2x1 - 4x3 = 0 -x1 + 2x3 = - 1
These equations are inconsistent, therefore we cannot assign x 2 arbitrarily . It turns
out that either x1 or x3 in (B .13) can be arbitrarily assigned . The reason is as follows :
The first column of A in (B .13) is linearly dependent on the remaining two columns .
If we delete the first column, the remaining matrix still has a full row rank . Therefore
the coefficient corresponding to the first column-namely, x 1 -can be arbitrarily
assigned. If we assign it as x 1 = 10, then (B .13) becomes
20+x2 -4x3 =3 -10+2x3 = -1
or
x2 -4x3 = - 17 2x3 =9
which imply x3 = 4.5 and x2 = 1 . Thus x' = [10 1 4 .5] is a solution . If we
choose a different x 1 , we will obtain a different solution . Similarly, we can assign
x3 arbitrarily, but we cannot assign both x 1 and x 3 arbitrarily .

Exercise B.1
In (B.13), if we assign x3 = 10, what is the solution? If we assign x 1 = 1 and

x3 = 10, does (B .13) have a solution?
[Answers : x1 = 21, x2 = 1, x3 = 10; no.]
Consider again (B.12) with A square . If y = 0, (B .12) reduces to

Ax = 0 (B.14)
This is called a homogeneous equation . It is clear that x = 0 is always a solution

of (B .14) whether or not A is nonsingular . This solution is called the trivial solution .
A nonzero x meeting Ax = 0 is called a nontrivial solution.
THEOREM B .2
A nontrivial solution exists in Ax = 0 if and only if A is singular . Or,
equivalently, x = 0 is the only solution of Ax = 0 if and only if A is
nonsingular.
B .5 ELIMINATION AND SUBSTITUTION
There are many ways to compute the solution of Ax = y . We discuss in this section
the method of Gaussian elimination . This method is applicable no matter whether
or not A is nonsingular . It can also be used to compute nontrivial solutions of
Ax = 0. This is illustrated by examples .
Example B .1
Find a solution of
x1 + 2x2 + x 3 = 10 (B .15)
2x1 + 5x2 - 2x3 = 3 (B .16)
x, + 3x 2 = 0 (B .17)
Subtraction of the product of 2 and (B .15) from (B .16), and subtraction of (B. 15)
from (B.17) yield
x, +2x 2 +x3 = 10 (B.15')
X2 - 4x3 = 17 (B.16')
x2 - x 3 = - 10 (B.17')

B .5 ELIMINATION AND SUBSTITUTION 587
Subtraction of (B .16') from (B.17') yields

x, + 2x2 + x3 = 10 (B .15")
X2 - 4x3 = 17 (B.16")
3x3 = 7 (B.17")
This process is called Gaussian elimination. Once this step is completed, the solution
can easily be obtained as follows . From (B. 17"), we have
7
X3 3
Substitution of x3 into (B . 16") yields
28 23
x2 = -17+4x3 = -17+-= --
3 3
Substitution of x3 and x2 into (B.15") yields
x,= 10-2x 2 -x3 = 10+
36 -3=23
This process is called back substitution . Thus, the solution of linear algebraic equa-
tions can be obtained by Gaussian elimination and then back substitution .
Example B .2
Find a nontrivial solution, if it exists, of
x, + 2x 2 + 3x3 = 0 (B .18)
2x, + 5x2 - 2x3 = 0 (B .19)
3x, + 7x2 + x3 = 0 (B .20)
Subtraction of the product of 2 and (B . 18) from (B . 19) and subtraction of the product
of 3 and (B.18) from (B.20) yield
x, + 2x2 + 3x3 = 0 (B.18')
X2 - 8x3 = 0 (B .19')
x2 - 8x3 = 0 (B.20')
We see that (B . 19') and (B.20') are identical . In other words, the two unknowns x2
and x3 are governed by only one equation . Thus either one can be arbitrarily assigned .
Let us choose x3 = 1 . Then x2 = 8. The substitution of x 3 = 1 and x2 = 8 into
(B.18') yields
x,= -2x2 -3x3 = -16-3= -19
Thus x, _ -19, x2 = 8, x3 = 1 is a nontrivial solution .

Gaussian elimination is not a numerically stable method and should not be used
on computer computation . The procedure, however, is useful in hand computation .
In hand calculation, there is no need to eliminate x; in the order of x 1, x2, and x 3.
They should be eliminated in the order which requires less computation . In Example
B .1, for instance, x 3 does not appear in (B.17) . Therefore we should use (B .15) and
(B. 16) to eliminate x3 to yield
4x 1 + 9x2 = 23
We use this equation and (B . 17) to eliminate x 2 to yield x 1 = 23. The substitution
of x, into (B .17) yields x2 = - 23/3 . The substitution of x, and x2 into (B .15) yields
x3 = 7/3 . This modified procedure is simpler than Gaussian elimination and is
suitable for hand calculation . For a more detailed discussion, see Reference [18].
B.6 GAUSSIAN ELIMINATION WITH PARTIAL PIVOTING
In this section, we modify Gaussian elimination to yield a numerically stable method .

Consider
a, , a12 a 13 a 14 x1 Y1
a21 a22 a23 a24 x2 Y2

(B.21)
a31 a32 a33 a34 x3 Y3
a41 a42 a43 a44 x4 Y4
Before carrying out elimination in the first column (corresponding to the elimination
of x1 from the 2nd, 3rd, and 4th equations), we search for the element with the largest
magnitude in the first column, say a31 , and then interchange the first and third equa-
tions. This step is called partial pivoting . The element a 31 , which is now located at
position (1, 1), is called the pivot . We then divide the first equation by a 31 to nor-
malize the pivot to 1 . After partial pivoting and normalization, we carry out elimi-
nation to yield
1 a12(1) (1) (1) -

a13 a14
-----------------
0 a22(I) (1)
a23(1) a 24 x2 (B.22)
(1) (1) (1)
a32 a33 a 34 x3
(1) (1) a(1 )
a42 a43 44 - _X4
In the elimination, the same operations must be applied to y i. Next we repeat the
same procedure to the submatrix bounded by the dashed lines . If the element with
the largest magnitude among a22, a (32l) , and a42 is nonzero, we bring it to position
(2, 2), normalize it to 1, and then carry out elimination to yield

B .6 GAUSSIAN ELIMINATION WITH PARTIAL PIVOTING 589
(1) (L) (1) (~)

1 -----------------
a12 a 13 a14 x1 y,
(2) (
0 23 24 x2 y (2)
2
(2) (B .23a)
0 0 a(2) 33 a (2)
34 X3 Y3
0 ' 0 a43 a (2)
(2)
44 x4 y42 )
If the three elements a22, a3z, and a42 in (B .22) are all zero, (B .22) becomes
12 a(1)
1 a (1) (1)
13 a14 X,
(1)
y1
0 --- (1) (1)
0 a23 a24 x2 y2
(B .23b)
0 0 a(1) 33
(1)
a 34 x3
- (1)
y3
(1) (1) (1)
0 i 0 a43 a44 X4 y4
In this case, no elimination is needed and the pivot is zero . We repeat the same
procedure to the submatrix bounded by the solid lines in (B .23). Finally we can
obtain
1 p p p x1 Y(1 n
0 0 p p x2 Y22)
(3) (B .24)
0 0 1 p X3 Y3
0 0 0 1 x4 y4 )
where p denotes possible nonzero elements . The transformation of (B .21) into the
form in (B .24) is called Gaussian elimination with partial pivoting . It is a numeri-
cally stable method . It can easily be programmed on a digital computer and is widely
used in practice . Once (B .21) is transformed into (B .24), the solution x, can be easily
obtained by back substitution .
Now we discuss the use of MATLAB to solve a set of linear algebraic equations .
We rewrite (B.15), (B .16), and (B .17) in matrix form as
1 2 1 x1 10
2 5 -2 x 2 = 3
1
1 3 0 x3 0
The commands
a=[1 2 1 ;2 5 -2 ;1 3 0] ; b=[10 ;3 ;0] ; alb
yield x, = 23 .000, x2 = - 7.6667, x 3 = 2.3333 . Thus the use of MATLAB is very
simple.
References
[1] Anderson, B . D. O . and J . B . Moore. Linear Optimal Control, Englewood

Cliffs, NJ : Prentice-Hall, 1971 .
[2] Astrom, K . J. "Robustness of a design based on assignment of poles and
zeros," IEEE Trans . Automatic Control, Vol . AC-25, pp. 588-591, 1980 .
[3] Astrom, K . J. and B . Wittenmark. Computer Controlled Systems : Theory and
Design, Englewood Cliffs, NJ : Prentice-Hall, 1984 .
.[4] Adaptive Control, Reading, MA: Addison-Wesley, 1989 .
[5] Athans, M . and P . L. Falb . Optimal Control, New York: McGraw-Hill, 1966.
[6] Atherton, D. P . Nonlinear Control Engineering, London : Van Nostrand
Reinhold, 1982 .
[7] Bode, H . W. Network Analysis and Feedback Amplifier Design, Princeton,
NJ: Van Nostrand, 1945 .
[8] Boyd, S. P. et al . "A new CAD method and associated architectures for linear
controllers," IEEE Trans . Automatic Control, Vol . 33, pp . 268-283, 1988 .
[9] Callier, F. M. and C . A. Desoer . Multivariable Feedback Systems, New York:
Springer-Verlag, 1982 .
[10] Chang, S. S. L . Synthesis of Optimal Control Systems, New York : McGraw-
Hill, 1961 .
[11] Chen, C. T . "Design of feedback control systems," Proc . Natl . Electronic
Conf., Vol. 57, pp . 46-51, 1969 .
[12] Analysis and Synthesis of Linear Control Systems, New York: Holt,
Rinehart and Winston, 1975 .
.
[13] One-Dimensional Digital Signal Processing, New York: Dekker,
1979.
590

REFERENCES 591
[14] . "A contribution to the design of linear time-invariant multivariable

systems," Proc . Am. Automatic Control Conf., June, 1983.
[15] . Linear System Theory and Design, New York : Holt, Rinehart and
Winston, 1984 .
[16] . Control System Design : Conventional, Algebraic and Optimal Meth-
ods, Stony Brook, NY : Pond Woods Press, 1987 .
[17] . "Introduction to the Linear Algebraic Method for Control System
Design," IEEE Control Systems Magazine, Vol . 7, No . 5, pp. 36-42, 1987 .
[18] . System and Signal Analysis, New York : Holt, Rinehart and Winston,
1989 .
[19] Chen, C . T. and B . Seo . "Applications of the Linear Algebraic Method for
Control System Design," IEEE Control Systems Magazine, Vol . 10, No . 1,
pp. 43-47, 1989 .
[20] . "The Inward Approach in the Design of Control Systems," IEEE
Trans . on Education, Vol . 33, pp. 270-278, 1990 .
[21] Chen, C . T . and S . Y . Zhang. "Various implementations of implementable
transfer matrices," IEEE Trans . Automatic Control, Vol . AC-30,
pp. 1115-1118, 1985 .
[22] Coughanowr, D . R . Process Systems Analysis and Control, 2nd ed ., New
York : McGraw-Hill, 1991 .
[23] Craig, J . J. Introduction to Robotics, 2nd ed., Reading MA : Addison-Wesley,
1989 .
[24] Daryanani, G. Principles of Active Network Synthesis and Design, New York:
Wiley, 1976 .
[25] Doebelin, E . O . Control System Principles and Design, New York : John
Wiley, 1985 .
[26] Doyle, J . and G. Stein. "Multivariable feedback design : Concepts for clas-
sical/modern analysis," IEEE Trans . on Automatic Control, Vol . 26, No. 1,
pp. 4-16, 1981 .
[27] Evans, W. R . "Control system synthesis by the root locus method," Trans .
AIEE, Vol . 69, pp . 67-69, 1950 .
[28] Franklin, G . F., J. D . Powell, and M . L. Workman. Digital Control of Dynamic
Systems, 2nd ed., Reading, MA: Addison-Wesley, 1990 .
[29] Franklin, G . F ., J . D . Powell, and A . Emami-Naeini . Feedback Control of
Dynamic Systems, Reading, MA : Addison-Wesley, 1986 .
[30] Frederick, D . K ., C. J. Herget, R . Kool, and C. M . Rimvall. "The extended
list of control software," Electronics, Vol . 61, No . 6, p. 77, 1988 .
[31] Gawthrop, P . J. and P . E . Nomikos . "Automatic tuning of commercial PID
controllers for single-loop and multiloop applications," IEEE Control Systems
Magazine, Vol . 10, No . 1, pp. 34-42, 1990.
[32] Gayakwad, R. and L . Sokoloff. Analog and Digital Control Systems, Engle-
wood Cliffs, NJ : Prentice-Hall, 1988 .
[33] Graham, D. and R . C. Lathrop . "The synthesis of optimum response : Criteria
and standard forms," AIEE, Vol . 72, Pt. II, pp. 273-288, 1953 .

592 REFERENCES
[34] Hang, C . C. "The choice of controller zeros," IEEE Control Systems

Magazine, Vol. 9, No . 1, pp . 72-75, 1989 .
[35] Hang, C . C . and K. K . Sin. "A comparative performance study of PID auto-
tuners," IEEE Control Systems Magazine, Vol . 11, No. 5, pp . 41-47, 1991 .
[36] Horowitz, I . M. Synthesis of Feedback Systems, New York: Academic Press,
1963.
[37] Jacob, J . M. Industrial Control Electronics : Application and Design, Engle-
wood Cliffs, NJ : Prentice-Hall, 1989 .
[38] Jury, E. I. Sampled-Data Control Systems, New York : Wiley, 1958 .
[39] Killian, M . J . "Analysis and Synthesis of Phase-locked Loop Motor Speed
Control Systems," Master Thesis, State University of New York, 1979 .
[40] Ku, Y . H. Analysis and Control of Linear Systems, Scranton, PA: International
Textbook, 1962.
[41] Kucera, V. Discrete Linear Control-The Polynomial Equation Approach,
Chichester, UK : Wiley, 1979 .
[42] Kuo, B . C. Automatic Control Systems, 3rd ed . Englewood Cliffs, NJ :
Prentice-Hall, 1975 .
[43] Kuo, B . C. Digital Control Systems, New York: Holt, Rinehart and Winston,
1980 .
[44] Lin, J . L., J . J. Sheen, and C . T. Chen . "Comparisons of various implemen-
tations of quadratic optimal systems," Technical Report, National Cheng-
Kung University, Taiwan, 1986 .
[45] Luenberger, D . G. "Observing the state of a linear system," IEEE Trans .
Military Electronics, Vol. MIL-8, pp . 74-80, 1964 .
[46] Middleton, R . H . and G. C . Goodwin. Digital Control and Estimation, A
Unified Approach, Englewood Cliffs, NJ : Prentice-Hall, 1990 .
[47] Moler, C . and C. F . Van Loan . "Nineteen dubious ways to compute the
exponential of a matrix," SIAM Review, Vol. 20, pp . 801-836, 1978 .
[48] Newton, G . C. Jr., L. A . Gould, and J . F. Kaiser. Analytical Design of Linear
Feedback Controls, New York: Wiley, 1959 .
[49] Norimatsu, T . and M. Ito . "On the zero non-regular control system," J. Inst.
Elec. Eng . Japan, Vol . 81, pp . 566-575, 1961 . See also B . Porter, "Com-
ments on Òn undershoot and nonminimum phase zeros'," IEEE Trans . on
Automatic Control, Vol . AC-32, pp . 271-272, 1987 .
[50] Northrop, R . B. Analog Electronic Circuits, Analysis and Application, Read-
ing, MA : Addison-Wesley, 1990 .
[51] Nyquist, H . "Regeneration theory," Bell Syst. Tech . J., Vol. 11, pp. 126-147,
1932 .
[52] Ogata, K. Discrete-Time Control Systems, Englewood Cliffs, NJ : Prentice-
Hall, 1987 .
[53] Rosenbrock, H . H. State-Space and Multivariable Theory, New York: Wiley-
Interscience, 1970 .
[54] Seo, B. "Computer-Aided Design of Control Systems" Ph .D. dissertation,
SUNY, Stony Brook, NY, 1989 .

REFERENCES 593
[55]
Shipley, D . P. "A unified approach to synthesis of linear systems," IEEE
Trans. on Automatic Control, Vol . AC-8, pp. 114-125, April, 1963 .
[56] Slotine, J-J . E. and W. Li. Applied Nonlinear Control, Englewood Cliffs, NJ :
[57] Truxal, J . G. Automatic Feedback Control System Synthesis, New York :
McGraw-Hill, 1955 .
[58] . Introductory System Engineering, New York : McGraw-Hill, 1972 .
[59] Tsui, C . C . "An analytical solution to the equation TA-FT = LC and its ap-
plications," IEEE Trans . on Automatic Control, Vol . 32, No. 8, pp . 742-744,
1987 .
[60] . "On preserving the robustness of optimal control system with ob-
servers," IEEE Trans . on Automatic Control, Vol . 32, No . 9, pp. 823-826,
1987 .
[61] Van de Vegte, J . Feedback Control Systems, 2nd ed., Englewood Cliffs, NJ :
[62] Vidyasagar, M. "On the well-posedness of large-scale interconnected sys-
tems," IEEE Trans . on Automatic Control, Vol . AC-25, pp . 413-421, 1980 .
[63] . Control System Synthesis : A Factorization Approach, Cambridge,
MA: MIT Press, 1985 .
[64] Willems, J. C . The Analysis of Feedback Systems, Cambridge, MA: MIT
Press, 1971 .
[65] Wolovich, W . A . Linear Multivariable Systems, New York : Springer-Verlag,
1974 .
[66] Wonham, W. M . "On pole assignment in multi-input controllable linear sy-
tems," IEEE Trans . on Automatic Control, Vol . AC-12, pp . 660-665, 1967 .
[67] . Linear Multivariable Control, New York : Springer-Verlag, 1985 .
[68] Youla, D . C ., J. J. Bongiorno, Jr ., and C . N . Lu . "Single-loop feedback-
stabilization of linear multivariable dynamical plants," Automatica, Vol . 10,
pp. 159-173, 1974 .
[69] Ziegler, J . G . and N. B . Nichols . "Optimum settings for automatic control-
lers," Trans . ASME, Vol . 64, pp . 759-768, 1942 .
[70] . "Process lags in automatic control circuits," Trans . ASME, Vol. 65,
pp. 433-442, 1943 .
Index
A/D converter, 480 Bilinear transformation, 520

Acceleration error, 193 Block diagram, 39, 94
Acceleration error constant, 306 basic, 162, 497
Acceleration function, 139 equivalent, 95
Ackermann formula, 447 loading of, 43, 65
Actual frequency, 117 Bode plot, 272, 275, 310, 540
Actuating signal, 10, 207 Breakaway point, 242
constraint on, 209, 348
Algebraic Riccati equation, 451 Causality, 494
Amplifier, 84 Cayley-Hamilton theorem, 68
differential, 105 Centroid, 239, 241
feedback, 210 Characteristic function, 100
isolating, 44, 85, 480 Characteristic polynomial, 26, 54, 491
open-loop, 210 Closed-loop properness, 200
operational, 84, 335 Closed-loop system, 10, 210
Amplitude characteristic, 143, 505 Common factor, 35
Analytical method, 1 nontrivial, 35
Angle of arrival, 244 trivial, 35
Angle of departure, 244 Companion form, 55
Anticipatory controller, 557 Compensator, 10
Antiwindup, 555 digital, 514
Asymptote, 239, 276 Complete characterization, 40, 42, 56
Asymptotic tracking, 139, 193, 345 Configuration, 209
Automobile suspension system, 15, 19 closed-loop, 210
feedback, 210
BIBO stability, 126 open-loop, 10, 210
Back substitution, 587 plant I/O, 422
Backlash, 78 two-parameter, 402, 405, 546
Backward-difference method, 519 unity-feedback, 305
Bandwidth, 144, 302 Constant-M loci, 308
Basic block diagram, 162, 497 Control signal, 10
595

596 INDEX
Controllability, 433, 496 Feedback, 115

Controllability matrix, 434, 496 Feedback compensator, 402
Controllable-form realization, 168, 187, 437, 499 Feedback system, 10
Controlled variable, 10 Feedforward compensator, 402
Controller, 10 Final-value theorem, 502, 572
Coprimeness, 35 First-order hold, 527
Comer frequency, 276 Forward path, 100
Coulomb friction, 17 Forward-difference method, 517
Cutoff frequency, 302 Free response, 26, 492
Frequency, 116
D/A converter, 479 actual, 117
corner, 276
DC motor, 70
armature-controlled, 72 cutoff, 302
electrical time constant, 72, 74 damped, 117
field-controlled, 70 gain crossover, 298, 299
mechanical time constant, 72 natural, 116
motor gain constant, 74 phase crossover, 298, 299
Damped frequency, 117 Frequency response, 143, 505, 506
Damping factor, 116 Frequency warping, 521
Damping ratio, 116, 120, 225 Full-dimensional estimator, 456
Dead time, 559 Function, 35
Dead-beat design, 541 acceleration, 139
polynomial, 138
Decibel (dB), 271
ramp, 192
Delta function, 568
step, 35, 191
Derivative controller, 556
Derivative time constant, 557 sinusoidal, 141
Desired pole region, 227, 231, 535
Desired signal, 10 Gain crossover frequency, 298, 310
Determinant, 580 Gain margin, 298, 308
Difference equation, 481, 490 Gaussian elimination, 587
Differential amplifier, 105 with partial pivoting, 589
Differential equation, 16, 26, 573 Gears, 78, 104
Differentiator, 32 backlash, 78
Digital compensator, 476 Generator, 102
Diophantine equation, 390
Half-power point, 303
Direct transmission part, 46, 65
Hidden dynamics, 530
Discrete transfer function, 492
Hidden pole, 203
Discrete-time state-variable equation, 161, 495
Homogeneous equation, 53
Discretization, 477
Hurwitz polynomial, 130
Disturbance, 198
Disturbance rejection, 214, 411, 419 Hydraulic tanks, 22, 66
Dominant pole, 376
IAE, 347
ISE, 348
Empirical method, 1 ITAE, 348, 367
Equivalence transformation, 441 Identification, 77, 286
Equivalent digital plant, 528 Impedance, 29
Eqivalent equation, 441 Implementable transfer function, 340, 342, 379, 545
Error constant, 307 Impulse, 568
acceleration, 306 Impulse response, 506
position, 306 Impulse sequence, 483
velocity, 306 Impulse-invariant method, 512
Error detector, 83, 87 Initial-value theorem, 503, 572
Euler forward algorithm, 156 Integral controller, 555

INDEX 597
Integral of absolute error (IAE), 347 Mason's formula, 100

Integral time constant, 557 Matrix, 579
Integral windup, 555 adjoint, 581
Integration step size, 155 block diagonal, 581, 582
Internal model principle, 218 determinant, 580
Inverse Laplace transform, 570 diagonal, 580
Inverse z-transform, 487 inverse, 581
Inverting amplifier, 87 nonsingular, 581
Inward approach, 216 rank, 583
Isolating amplifier, 45 singular, 581
square, 579
Jury test, 501 symmetric, 450
transpose, 580
Kronecker sequence, 483 triangular, 580
unit, 580
Lag-lead network, 327 Minimal equation, 56, 435, 497, 499
Laplace transform, 50, 484, 567 Minimal state-variable equation, 56
inverse, 570 Minimum-phase transfer function, 285
Left half plane, 123 Minimum-phase zero, 285, 534
closed, 123 Missing pole, 203
open, 123, 127 Mode, 26, 40, 492
Linear algebraic equation, 392, 407, 424, 584 Model, 14, 62
homogeneous, 586 Model matching, 344, 385, 400, 406, 419, 423, 544
Linear algebraic method, 384, 426, 461 Motor, 70
Linear factor, 276 armature-controlled, 72
Linear operational range, 18 electrical time constant, 72
Linear time-invariant system, 16, 40 field-controlled, 70
Loading problem, 43, 65, 80 gain constant, 74
Log magnitude-phase plot, 271 mechanical time constant, 72
Loop, 99 time constant, 74
nontouching, 99 transfer function of, 72, 73, 74
Loop gain, 99 Multivariable system, 16
Loop transfer function, 194, 305
Lyapunov theorem, 465, 507 Natural frequency, 116, 118, 225
Lyapunov equation, 457, 465, 508 Nichols chart, 308
Lyapunov function, 466 Noise, 32, 83, 197
Nonminimum-phase transfer function, 285
MATLAB, 160 Nonminimum-phase zero, 285, 342, 534
bode, 274, 275 Nonsingular matrix, 441
c2d, 161, 532 Nontrivial solution, 586
dstep, 500 Numerical control, 107
lqr, 453 Nyquist plot, 290
lyap, 458 Nyquist stability criterion, 294, 296
nyquist, 274, 275
place, 449 Observability, 433, 496
poly, 354, 363 Observability matrix, 434, 496
rank, 583 Observable-form realization, 170, 187, 437, 499
rlocus, 234 Op amp. See operational amplifier.
roots, 165, 354, 362 Open-loop state estimator, 454
ss2tf, 532 Open-loop system, 10, 210
step, 160, 172 Operational amplifier, 84, 162, 335
tf2ss, 171, 532 Optimal system, 346
tf2zp, 236, 532 ITAE, 367
Magnitude condition, 235, 247 quadratic, 350

598 INDEX
Output equation, 46 Proper transfer function, 340

Outward approach, 216 Proportional controller, 553
Overshoot, 119, 196, 225, 304 Proportional-integral (PI) compensator, 327
PD controller, 257 Quadratic factor, 281

PI controller, 327 Quadratic optimal regulator, 450
PID controller, 551, 557, 561 Quadratic optimal system, 352
digital, 563, 565 Quadratic performance index, 350
position form, 565 Quadratic transfer function, 116, 225, 281, 303
velocity form, 565 step response of, 118
Parallel realization, 174, 177 Quantization, 477
Parameter estimation, 77
Partial fraction expansion, 570
Peak resonance, 302, 308 Radar antenna, 4
Performance criterion, 189, 346 Ramp function, 192
steady-state, 191, 301, 305 Rank, 583
transient, 195, 302, 307 Reaction wheel, 92
Perturbation, 198, 213 Realization, 166, 498
Phase characteristic, 143, 505 controllable-form, 168, 187, 437
Phase condition, 236 minimal, 178, 179, 182, 404
Phase crossover frequency, 298 nonminimal, 178
Phase margin, 298, 308 observable-form, 170, 187, 437
Phase-lag network, 22, 260, 317, 327 parallel, 175
Phase-lead network, 260, 321, 327 tandem, 173
Physical system, 15 Reduced-dimensional estimator, 457, 542
Pivot, 588 Reference signal, 10
Plant, 10, 189 Region of convergence, 482, 568
equivalent digital, 528 Regulating problem, 193, 345
Plant input/output feedback configuration, 422 Reset windup, 555
Plant leakage, 341 Right half plane, 123
Plant output, 10 closed, 123
Plant perturbation, 198, 213 open, 123
Polar plot, 271 Rise time, 196, 228
Pole, 34, 38, 40, 124, 492 Robotic control, 94, 108
missing pole, 40, 42, 56 Robust tracking, 397
stable, 128 Robustness, 194, 218, 411, 414, 418
Pole placement, 384, 388, 393, 397, 400, 447 Root loci, 236, 246
Pole-and-zero placement, 384 magnitude condition, 247
Pole-zero cancellation, 204, 206, 345, 438 phase condition, 236
Pole-zero excess inequality, 342 symmetric, 365
Pole-zero mapping, 521 Root-locus method, 236, 534
Polynomial fractional approach, 427 Routh table, 132, 152
Position control system, 2, 88, 116 Routh test, 132
Position error, 191, 301, 306 proof of, 467
Position error constant, 306
Positive definite matrix, 450 Sample-and-hold circuit, 480
Positive semidefinite matrix, 450 Sampling frequency, 506
Potentiometer, 81 Sampling period, 57, 524
Primary strip, 487, 530 selection of, 524
Principal minor, 450 Saturation, 208
leading, 450 Separation property, 461
Principle of argument, 290 Set point, 139
Proper compensator, 199, 340 Settling time, 196, 227
INDEX 599
Signal, 57 Tandem realization, 173

analog, 57, 476 Temperature control system, 6, 90, 106
bounded,126 Time constant, 74, 76, 112, 145
continuous-time, 57, 476 derivative, 557
digital, 477, 478 integral, 557
discrete-time, 57, 477 motor, 74
Similarity transformation, 442, 548 Time delay, 577
Single-variable system, 16 Total stability, 203, 340
Spectral factorization, 351, 354 Tracking problem, 193
Speed control system, 89, 112 Transducer, 81
Speed of response, 195 optical, 106
Spring constant, 17 Transfer function, 28, 54, 101, 492
Stability, 126 biproper, 31
BIBO, 126 closed-loop, 102
Nyquist test, 294, 296 digital, 492
range of, 135 discrete-time, 492
relative, 298 improper, 30
Routh test of, 132 irreducible, 35
total, 203 loop, 99, 194, 305
Stabilization, 472 measurement of, 75
State, 46 minimum-phase, 285
State equation, 46 nonminimum-phase, 285
State estimator, 453, 456, 459, 541 open-loop, 102
full-dimensional, 456 optimal, 352, 367-371
open-loop, 454 proper, 31
reduced-dimensional, 456 strictly proper, 31
State feedback, 445, 459, 541 vector, 179
State variable, 46 Transducer, 81
State-variable equation, 46, 57 Transient performance, 195, 302, 307
computer computation of, 155 Transient response, 145
discrete-time, 58 Transport lag, 559
equivalent, 441 Trivial solution, 586
op-amp circuit realization of, 162 Trapezoidal approximation method, 520
Static friction, 17 Two-parameter configuration, 402, 405, 546
Steady-state performance, 191, 301, 305
Steady-state response, 139, 140, 142, 145, 503 Ultimate gain, 554, 558
Steady-state speed, 76, 112 Ultimate period, 558
Step function, 191 Undershoot, 232
Step-invariant method, 514 Unit-step response, 35, 112
Strong stabilization, 427 Unity-feedback configuration, 305
Symmetric matrix, 450
Symmetric root loci, 365 Velocity control system, 4
System, 15 Velocity error, 192, 301, 306
analog, 57, 475 Velocity error constant, 306
continuous-time, 57, 475 Viscous friction, 17
digital, 475, 481 Voltage follower, 85
discrete-time, 475, 481 Voltage regulator, 103
System matrix, 46 closed-loop, 103, 148
System type, 194, 217, 218 feedback, 103
open-loop, 103, 147
Tachometer, 82, 110
use of, 82 Ward-Leonard system, 103
Taming factor, 557 Weighting factor, 350, 358

600 INDEX
Well-posedness, 200, 340 infinite, 35, 241

minimum-phase, 128
z-transform, 482, 484 nonminimum-phase, 128
inverse, 487 Zero-input response, 24, 25, 51, 53, 490, 495
Zero, 34, 38, 492 Zero-order hold, 526
finite, 35 Zero-state response, 24, 27, 38, 51, 53, 490, 495

Chen - Analog and Digital Control System Design Ocr PDF

Uploaded by

Copyright:

Available Formats

Chen - Analog and Digital Control System Design Ocr PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chen - Analog and Digital Control System Design Ocr PDF

Uploaded by

Copyright:

Available Formats

Analog (rDzqitai

Saunders College Publishing

Chapter 2 Mathematical Preliminary 14

2.4 Zero-State Response-Transfer Function 27

Chapter 3 Development of Block Diagrams for Control Systems 69

Chapter 4 Quantitative and Qualitative Analyses of Control Systems 111

4.6 The Routh Test 129

Chapter 5 Computer Simulation and Realization 154

5.1 Introduction 154

Chapter 6 Design Criteria, Constraints, and Feedback 188

6.1 Introduction 188

Chapter 7 The Root-Locus Method 223

7.1 Introduction 223

Chapter 8 Frequency-Domain Techniques 270

8.1 Introduction 270

8 .9 Phase-Lead Compensation 321

Chapter 9 The Inward Approach-Choice of Overall Transfer Functions 339

9.1 Introduction 339

Chapter 10 Implementation-Linear Algebraic Method 384

10.6 Plant Input/Output Feedback Configuration 422

Chapter 11 State-Space Design 432

Chapter 12 Discrete-Time System Analysis 475

12.8 Stability 500

Chapter 13 Discrete-Time System Design 511

Chapter 14 PID Controllers 551

14.1 Introduction 551

Appendix A The Laplace Transform 567

A.1 Definition 567

A .4 Solving LTIL Differential Equations 573

Appendix B Linear Algebraic Equations 579

B.! Matrices 579

1 .1 EMPIRICAL AND ANALYTICAL METHODS

The ultimate goal of engineering-in particular, that of control engineering-is to

1 .2 .1 Position Control Systems

Motor and load

Figure 1 .2 Radar antenna. (Courtesy of MIT Lincoln Laboratory.)

1 .2.2 Velocity Control Systems

Housing for signal

Figure 1 .4 Nuclear power plant .

Figure 1 .5 Numerical control system .

1 .2.3 Temperature Control Systems

1 .2.4 Trajectory Control and Autopilot

Control device Switch

Thermocouple and Amplifier

Figure 1 .8 Desired landing trajectory of space shuttle .

1 .2.5 Miscellaneous Examples

1 .3 PROBLEM FORMULATION AND BASIC TERMINOLOGY

Figure 1 .11 Control design problem .

Figure 1 .12 Feedback control system .

1 .4 SCOPE OF THE TEXT

1 .4 SCOPE OF THE TEXT 13

1940s . The root-locus method is introduced in Chapter 7 and the frequency-domain

2 .1 PHYSICAL SYSTEMS AND MODELS

2 .1 PHYSICAL SYSTEMS AND MODELS 15

Figure 2 .1 Model of operational amplifier .

Figure 2.2 Model of automobile suspension system .

(a) (b) (c)

Figure 2 .3 Models of arm .

different models at high and low frequencies . A spaceship may be modeled as a