You are on page 1of 149

Dr. Mahmoud M.

Al-Husari

Control Systems Theory


Lecture Notes

This set of lecture notes are never to be considered as a substitute


to the textbook recommended by the lecturer.

ii

Contents

1 Introduction
1.1 Historical review . . . . . . . . . . . . . . . . . . . . .
1.2 Control system fundamentals . . . . . . . . . . . . . .
1.2.1 Concept of a system . . . . . . . . . . . . . . .
1.2.2 Open-loop systems . . . . . . . . . . . . . . . .
1.2.3 Closed-loop systems . . . . . . . . . . . . . . .
1.2.4 The control problem . . . . . . . . . . . . . . .
1.2.5 Examples of control systems . . . . . . . . . . .
1.3 A first analysis of feedback . . . . . . . . . . . . . . .
1.3.1 Effect of feedback on tracking and disturbance
1.3.2 Effect of feedback on sensitivity . . . . . . . . .
1.3.3 Effect of feedback on stability . . . . . . . . . .
1.3.4 The cost of feedback . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

1
1
2
2
3
4
4
5
8
9
9
10
10

2 System Modelling
2.1 Introduction . . . . . . . . . . . . . . . . . . . . .
2.2 The Laplace transform . . . . . . . . . . . . . . .
2.2.1 Definition . . . . . . . . . . . . . . . . . .
2.2.2 Properties . . . . . . . . . . . . . . . . . .
2.2.3 Theorems . . . . . . . . . . . . . . . . . .
2.2.4 More examples . . . . . . . . . . . . . . .
2.2.5 Inverse Laplace transform . . . . . . . . .
2.2.6 Partial fraction expansion . . . . . . . . .
2.3 Transfer Functions . . . . . . . . . . . . . . . . .
2.4 Models of Electric Circuits . . . . . . . . . . . . .
2.5 Models of mechanical systems . . . . . . . . . . .
2.5.1 Translational motion . . . . . . . . . . . .
2.5.2 Rotational motion . . . . . . . . . . . . .
2.6 DC motors in control systems . . . . . . . . . . .
2.7 System modelling diagrams . . . . . . . . . . . .
2.7.1 The block diagram . . . . . . . . . . . . .
2.7.2 Signal-flow graph . . . . . . . . . . . . . .
2.7.3 Conversion from block diagram to SFG .
2.7.4 Construction of block diagrams Examples

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

11
11
11
11
12
13
13
14
14
17
18
21
22
23
25
27
27
34
43
45

iii

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

iv

CONTENTS

3 Time-Domain Analysis of Control Systems


3.1 Introduction . . . . . . . . . . . . . . . . . .
3.2 System Time Responses . . . . . . . . . . .
3.3 Common test Signals . . . . . . . . . . . . .
3.3.1 The unit impulse function . . . . . .
3.3.2 The unit step function . . . . . . . .
3.3.3 The unit ramp function . . . . . . .
3.3.4 The sinusoidal functions . . . . . . .
3.4 Response of first order systems . . . . . . .
3.4.1 Standard form . . . . . . . . . . . .
3.4.2 Step response . . . . . . . . . . . . .
3.4.3 Steady-state response . . . . . . . .
3.5 Response of second order systems . . . . . .
3.5.1 Standard form . . . . . . . . . . . .
3.5.2 Pole locations . . . . . . . . . . . . .
3.5.3 Step response . . . . . . . . . . . . .
3.6 Specifications of a second order system . . .
3.7 Specifications vs. pole locations . . . . . . .
3.7.1 Step response vs. pole location . . .
3.8 Steady-state accuracy . . . . . . . . . . . .
3.8.1 Error signal . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

49
49
49
50
50
50
50
51
51
51
51
52
53
53
54
56
59
62
66
67
68

4 Stability Analysis
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . .
4.2 Bounded-input bounded-output stability . . . . . . .
4.2.1 Relationship between characteristic equation
stability . . . . . . . . . . . . . . . . . . . . .
4.3 Routh-Hurwitz Stability Criterion . . . . . . . . . .
4.3.1 General properties of polynomials . . . . . .
4.3.2 Routh-Hurwitz stability criterion . . . . . . .
4.4 Applications in feedback design . . . . . . . . . . . .

. . . . . . .
. . . . . . .
roots and
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .

5 Root-Locus Analysis and Design


5.1 Root-locus principles . . . . . . . . . . .
5.1.1 Root-locus criterion . . . . . . .
5.2 Rules & steps for plotting the root-locus
5.3 Examples . . . . . . . . . . . . . . . . .
5.4 The complementary root locus . . . . .
5.5 Root locus design . . . . . . . . . . . . .
5.6 Dynamic compensation . . . . . . . . .
5.6.1 Lead compensation . . . . . . . .
5.6.2 Lag compensation . . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

85
85
87
90
102
108
110
111
112
116

6 Frequency Response Analysis


6.1 Frequency response . . . . . . . . . .
6.2 Bode diagrams . . . . . . . . . . . .
6.2.1 Constant gain . . . . . . . . .
6.2.2 Poles and zeros at the origin
6.2.3 Nonzero real poles and zeros
6.2.4 Phase diagrams . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

119
119
122
125
126
127
129

.
.
.
.
.
.

.
.
.
.
.
.

73
73
74
74
77
77
78
82

CONTENTS

6.3

6.2.5 Complex poles and zeros . . . . . . . . . . . . . . . . . . . 132


Nyquist plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.3.1 Nyquist criterion . . . . . . . . . . . . . . . . . . . . . . . 137

vi

CONTENTS

Chapter

Introduction
1.1

Historical review

Control theory has developed rapidly over the last 80 years. A significant
and rapid development due mainly to digital computers. Indeed, recent developments in digital computers especially their increasingly low cost facilitate their
use in controlling complex systems and processes.
The period 1930-1940 was important in the history of control, since remarkable theoretical and practical results, such as those of Nyquist were reported.
During the following years and until about 1960, further significant research and
development was reported, due mainly to Ziegler and Nichols, Bode, Wiener and
Evans. All the results of the last century, and up to about 1960, constitute what
has been termed classical control. Progress from 1960 to date has been especially
impressive, from both the theoretical and the practical point of view. This last
period has been characterized as that of modern control, the most significant
results of which have been due to Astrom, Doyle, Francis, Kailath, Kalman,
Zames, and many others.
The main differences between the classical and the modern control approaches are the following: classical control refers mainly to single input-single
output systems. The design methods are usually graphical (e.g., root locus,
Bode and Nyquist diagrams, etc.) and hence they do not require advanced
mathematics. Modern control refers to complex multi-input multi-output systems. The design methods are usually analytical and require advanced mathematics. In todays technological control applications, both classical and modern
design methods are used. Since classical control is relatively easier to apply than
modern control, a control engineer may adopt the following general approach:
simple cases, where the design specifications are not very demanding, he uses
classical control techniques, while in cases where the design specifications are
very demanding, he uses modern control techniques.
It should be noted that classical control techniques which have existed since
1940s predominate in the overall practice of control engineering today. Despite
the impressive progress since the 1940s, practical applications of modern control
techniques are limited. This is indeed a serious gap between theory and practice.
To reduce this gap, techniques of modern control engineering should be designed
1

CHAPTER 1. INTRODUCTION

with an eye toward applicability, so as to facilitate their use in practice. This


course is concerned with introducing classical control theories to undergraduate
students taking an introductory course in control systems.

1.2
1.2.1

Control system fundamentals


Concept of a system

Before discussing the structure of a control system it is necessary to define what


is meant by a system. A system is a combination of components (appropriately
connected to each other) that act together in order to perform a certain task.
For a system to perform a certain task, it must be excited by a proper input
signal. When one or more excitation signals are applied at one or more system
inputs, the system produces one or more response signals at its outputs. Figure
1.1 shows a block diagram of a single-input, single-output system. Systems with
more than one input and more than one output are called MIMO (Multi-Input
Multi-Output).

Figure 1.1: Block diagram of a simple system.

In control engineering, the way in which the system outputs respond in


changes to the system inputs (i.e. the system response) is very important.
The control system designer will attempt to evaluate the system response by
determining a mathematical model for the system. Knowledge of the system
inputs, together, with the mathematical model, will allow the system outputs
to be calculated.
The system being controlled is usually referred to as the plant. Some inputs,
the engineer will have direct control over, and can be used to control the plant
outputs. These are known as control inputs. There are other inputs over which
the engineer has no control, and these will tend to deflect the plant outputs
from their desired values. These are called disturbance inputs.
Example 1.1

In the case of the ship shown in Figure 1.2, the rudder and engines are the
control inputs, whose values can be adjusted to control certain outputs, for
example heading and forward velocity. The wind, waves and current are disturbance inputs and will induce errors in the outputs (usually called controlled
variables) of position, heading and forward velocity. In addition, the disturbances will introduce increased ship motion (roll, pitch, etc.) which again is not
desirable. 
Generally, the relationship between control input, disturbance input, plant
and controlled variable is shown in Figure 1.3.

1.2. CONTROL SYSTEM FUNDAMENTALS

Figure 1.2: A ship as a dynamic system.

Figure 1.3: Plant inputs and outputs.

Control systems can be divided into two categories: the open-loop and the
closed loop systems.

1.2.2

Open-loop systems

Figure 1.3 represent an open-loop control system. Note that the system input
does not depend on the output, i.e., the input is not a function of the output.
A very simple example of an open-loop system is that of the clothes washing
machine. Here, the control signal is the input to the washing machine and
forces the washing machine to execute the desired preassigned operations, i.e.,
water heating, water changing, etc. The output of the system is the quality
of washing, i.e., how well the clothes have been washed. It is well known that
during the operation of the washing machine, the output (i.e., whether the
clothes are well washed or not) it not taken into consideration. The washing
machine performs only a series of operations contained in the control input
without being influenced at all by the output.
The main problem with open-loop control is that the controlled variable
is sensitive to changes in disturbance inputs. So, for example if a heater is
switched on in a room, and the temperature climbs to 20 C, it will remain at
that value unless there is a disturbance. This could be caused by leaving a door
to the room open, for example. The internal room temperature will change. For
the room temperature to remain constant, a mechanism is required to vary the
energy output from the heater.

CHAPTER 1. INTRODUCTION

1.2.3

Closed-loop systems

A closed-loop system is a system whose input depends on the output, i.e., the
input is a function of the output. For a room temperature control system, the
first requirement is to detect or sense changes in room temperature. The second
requirement is to control or vary the energy output from the heater, if the sensed
room temperature is different from the desired room temperature. In general,
a system that is designed to control the output of a plant must contain at least
one sensor and controller as shown in Figure 1.4.

Figure 1.4: Closed-loop control system.

The controller and plant lie along the forward path, and the sensor in the
feedback path. The measured value of the plant output is compared at the
summing point with desired value. The difference, or error is fed to the controller
which generates a control signal to drive the plant until its output equals the
desired value.

1.2.4

The control problem

We may state the control problem as follows: given a physical system or process that is to be accurately controlled and the desired system response, find a
controller whose output is such that, when applied to the system, the output of
the system is the desired response. Generally speaking to be able to accurately
control a system, closed-loop or feedback operation is required. Typical aims of
feedback are:
disturbance rejection
transient response shaping
sensitivity reduction
closed-loop stability
tracking improvement
Solving a control problem generally involves:
choosing sensors to measure the plant output

1.2. CONTROL SYSTEM FUNDAMENTALS

choosing actuators to drive the system


model building
controller design
simulation testing
hardware testing

1.2.5

Examples of control systems

Automobile steering control system


The driver has the task of keeping the car on track on a desired direction of
travel. The driver uses the difference between the actual and desired direction
of travel to generate a controlled adjustment of the steering wheel as shown in
Figure 1.5.

Figure 1.5: Human control of an automobile.

A simple block diagram of an auomobile steering control system is shown in


Figure 1.6. The desired course is compared with a measurement of the actual
course in order to generate a measure of the error, as shown in Figure 1.5. The
measurement is obtained by visual feedback. In the feedback system of Figure
1.6, the driver is the controller, the actuator is the steering mechanism, the
plant is the car and the sensor is the visual.

Figure 1.6: Closed-loop control system.

CHAPTER 1. INTRODUCTION

Figure 1.7: Block diagram of human speech.

Human speech [Figure 1.7]


As we all know, we use our ears not only to hear others but also to hear ourselves.
Indeed, when we speak, we hear what we are saying and, if we realize that we
didnt say something the way we had in mind to say it, we immediately correct it.
Thus, human speech operates as a closed-loop system, where the reference input
is what we have in mind to say and want to put into words, the system is the
vocal cords, and its output is our voice. The output is continuously monitored
by our ears, which feed back our voice to our brain, where comparison is made
between ourintended (desired) speech and the actual speech that our own ears
hear (measure). If the desired speech and the measured speech are the same, no
correction is necessary, and we keep on talking. If, however, an error is realized,
e.g., in a word or in a number, then we immediately make the correction by
saying the correct word or number.
Teaching [Figure 1.8]
The proper procedure for teaching has the structure of a closed-loop system.
Let the students be the system, the teaching material presented by the teacher
the input, and the degree of understanding of this material by the students
the systems output. Then, teaching can be described with the block diagram
of Figure 1.8. This figure shows that the systems output, i.e., the degree of
understanding by students of the material taught, is fed back to the input, i.e.,
to the teacher. Indeed, an experienced teacher should be able to sense (measure)
if the students understood the material taught. Subsequently, the teacher will

Figure 1.8: Block diagram of teaching.

1.2. CONTROL SYSTEM FUNDAMENTALS

either go on teaching new material, if the students understood the material


taught, or repeat the same material, if they did not. Therefore, proper teaching
has indeed the structure of a closed-loop system.
Taking a shower
Imagine taking a shower from a two-knob faucet as shown in Figure 1.9. You
want to set the rate of water flow and its temperature so that the shower is
effective and comfortable. You can control the hot water flow and the cold
water flow separately.

Figure 1.9: Taking a shower.

So how do you regulate your shower?


1. You test the flow and temperature
2. Decide if it is OK
3. If not OK, decide what adjustment to make
4. Adjust the knobs
5. Repeat from the first step
The system can be presnted by a block diagram as shown in Figure 1.10.

Figure 1.10: A block diagram representing taking a shower.

1.3

CHAPTER 1. INTRODUCTION

A first analysis of feedback

As mentioned earlier in section 1.2.4 closed-loop or feedback control systems


have many advantages. The introduction of feedback enables the engineer to
control a desired output and improve accuracy. Hence, more accurate control
of plant under disturbances, i.e., effective disturbance rejection. A faster response to an input signal is achieved. Systems that are inherently unstable in
the open-loop form can be stabilized using feedback.
In this section the value of feedback can be demonstrated by quantitative
analysis of a simple static system. Consider the open-loop static system shown
in Figure 1.11. The static system G has an input u and and output y. The

Figure 1.11: A block diagram of an open-loop system..

output is possibly disturbed by an additive disturbance d. By simple algebraic


manipulations, it is simple to show that
y = Gu + d

(1.1)

Introducing a controller C with feedback to obtain a closed-loop system shown


in Figure 1.12. The controller C has an input e and an output u. Input e is

Figure 1.12: A block diagram of a closed-loop system..

created by comparing output y of the system G with the reference signal r:


e = r y. Note here that the reference signal r represents the desired response
of the system, i.e., in other words one would like the output y to track the
reference signal r. The controller output u, is given by
u = Ce
Hence, the output y is
y = GCe + d

(1.2)

The tracking error signal e, is given by


e = r (GCe + d)
1
= e =
(r d)
1 + GC

(1.3)

1.3. A FIRST ANALYSIS OF FEEDBACK


Next, substitute (1.3) in (1.2) to obtain


1
y = GC
(r d) + d
1 + GC
GC
1
= y =
r+
d
1 + GC
1 + GC

1.3.1

(1.4)

Effect of feedback on tracking and disturbance

The output of an open-loop system with no control was seen in (1.1) to be


y = Gu + d
On the other hand, with feedback control the output was given in (1.4) as
y=

1
GC
r+
d
1 + GC
1 + GC

Without control the map from the disturbance d to the ouput y is simply 1.
Any disturbance d will be seen directly (without reduction) on the output y.
With control, the map from disturbance d to output y is
1
1 + GC
therefore
any disturbance
d will be seen with a factor on the output y. Hopefully,


1
factor can reduce the disturbance. It can be seen that if C  1
the
1 + GC
(large or high loop gain), then
1
0
1 + GC
which implies that the effect of the disturbance d is eliminated. Similarly, the
map from reference r to output y is
GC
1 + GC
and if C  1 it can be seen that
GC
1
1 + GC
which implies that y = r, i.e., the output is tracking the reference closely. In
conclusion: Good tracking and disturbance rejection both require high loop gain.

1.3.2

Effect of feedback on sensitivity

Generally, any control system will contain parameters that change with temperature, pressure, humidity, time, and so on. However, we prefer that the control
system characteristics not vary as these parameters vary. Of course, the system characteristics are a function of the system parameters, but in some cases
the sensitivity of system characteristics to parameter variations can be reduced.

10

CHAPTER 1. INTRODUCTION

Consider the closed-loop system in Figure 1.12, with d = 0 (i.e., disturbances


are neglected), from (1.4) we obtain
y = Hr
where

GC
(1.5)
1 + GC
Clearly H is the map from reference r to output y, (it is usually referred to
as the overall system gain). The sensitivity of the system gain H to changes
in C for example is defined as a measure of the percentage change in H to a
percentage change in C. One such definition is
H=

S=

H C
H/H
=
C/C
C H

where H is the variation in H caused by C, the variation in C. We will now


find the sensitivity of H with respect to C
H C
C H
G(1 + GC) GCG C
(1 + GC)
=
(1 + GC)2
GC
1
=
1 + GC

S=

(1.6)

For this sensitivity to be small, we require that C  1. In conclusion: Sensitivity


reduction requires high loop gain.

1.3.3

Effect of feedback on stability

In a nonrigorous manner, a system is said to be unstable if its output is out of


control. To investigate the effect of feedback on stability, we can refer to the
expression in (1.5). If GC = 1, the output of the system is infinite for any
finite input, and the system is said to be unstable. therefore, we may state that
feedback can cause a system that is originally stable to become unstable.

1.3.4

The cost of feedback

Almost all the benefits of feedback can be achieved, provided that the loop gain
is sufficiently high. Unfortunately, for most plants, high loop gain tends to drive
the system into instability. Dont forget u = Ce so if C  1, the control signal
u will be large and this gets amplified in the feedback loop over and over causing
instability. Other disadvantages of closed-loop systems but not limited to:
require the use of sensors which increase the system costs
more complex design, harder to build
power costs (due to high gains) are high
It is essential in the design of controllers to trade-off between high gain, disturbance rejection, tracking and stability. For good design trade-off knowledge on
dynamic behavior of the system is essential. In conclusion: In most cases the
advantages of feedback far outweigh the cost and feedback is utilized.

Chapter

System Modelling
2.1

Introduction

If the dynamic behavior of a physical system can be represented by an equation, or a set of equations, this is referred to as the mathematical model of
the system. Such models can be constructed from knowledge of the physical
characteristics of the system, i.e. mass for a mechanical system or resistance for
an electrical system. Because the systems under consideration are dynamic in
nature, the descriptive equations are usually differential equations. Differential
equations are often the initial description of a system. The variables are just
the inputs and outputs. If the differential equations can be linearized, then the
Laplace transform can be utilized to simplify the method of solution. a singleinput single-output process is described by its transfer function: the ratio of the
Laplace transform of output and input.

2.2

The Laplace transform

The Laplace transform is a very powerful analysis tool for a certain class
of systems, namely, linear time-invariant systems. It transforms the problem
from the time (or t) domain to the Laplace (or s) domain. The advantage in
doing this is that complex time-domain differential equations become relatively
simple s-domain algebraic equations. It is then possible to manipulate the
algebraic equation by simple algebraic rules to obtain the solution in the sdomain. When a suitable solution is arrived at, it is inverse transformed back
to the time-domain.

2.2.1

Definition

The Laplace transform of a function of time f (t), 0 t with f (t) = 0 for


t 0 is defined as
Z
F (s) = L [f (t)] =
f (t)est dt
(2.1)
0

where s is a complex variable s = + j and is called the laplace operator.

11

12
Example 2.1

CHAPTER 2. SYSTEM MODELLING

Let f (t) be a unit step function defined as f (t) = 1 for t 0.


 Solution The Laplace transform of f (t) is obtained as

Z

1
1
est dt = est =
F (s) =

s
s
0
0

Example 2.2

Consider the exponential function f (t) = eat for t 0.


 Solution The Laplace transform of f (t) is
Z

F (s) =

eat est dt =

2.2.2


1
e(s+a)t
=
s + a 0
s+a

Properties

The application of the Laplace transform in many instances is simplified by


utilization of the properties of the transform. These properties are presented
here, for which no proofs are given.
Linearity
L[k1 f1 (t) k2 f2 (t)] = k1 F1 (s) k2 F2 (s)
Differentiation

L


df (t)
= sF (s) f (0)
dt

where f (0) is the limit of f (t) as t approaches 0. In general, for higher-order


derivatives of f (t),

 n
d f (t)
= sn F (s) sn1 f (0) sn2 f (1) (0) f (n1) (0)
L
dtn
where f (i) (0) denotes the ith-order derivative of f (t) with respect to t, evaluated
at t = 0.
Integration
Z
L


f ( )d =

F (s)
s

Shift in time
L[f (t T )] = eT s F (s)

for T 0

Shift in frequency
L[et f (t)] = F (s )

2.2. THE LAPLACE TRANSFORM

13

Covolution
L[f1 (t) f2 (t)] = F1 (s)F2 (s)

2.2.3

Theorems

Initial value theorem


A useful property of the Laplace transform known as the initial value theorem
which states that it is always possible to determine the initial value of the time
function f (t) from its Lapalce transform. We may state the theorem in this
way:
lim f (t) = lim sF (s)
s

t0

Final value theorem [More on this property later]


A second valuable Laplace transform theorem is the final value theorem, it
allows us to compute the constant steady-state value of a time function given
its Laplace transform.
lim f (t) = lim sF (s)
t

2.2.4

s0

More examples

In this section more examples are given to demonstrate the utilization of the
Laplace transform properties. For the subjects treated in this course, the direct
evaluation of the Lapalce transform integral is almost never used.
Find the Laplace transform of f (t) = cos t. [Linearity property]

Example 2.3

 Solution The Laplace transform is


Z
F (s) =
(cos t)est dt
0

We substitute the relation


cos t =

1 jt
(e + ejt )
2

into the integral, we find that





1
1
1
+
s j
2 s + j
s
= 2

s + 2

F (s) =

1
2

Find the Laplace transform of f (t) =

d2 f
. [Differentiation property]
dt2

 Solution The Laplace transform is


F (s) = s2 F (s) sf (0)


df
dt t=0

Example 2.4

14
Example 2.5

CHAPTER 2. SYSTEM MODELLING

Find the Laplace transform of f (t) = et cos t. [Shifting in frequency]


 Solution From Example 2.3 we know that
s
L[cos t] = 2
s + 2
s+
= L[et cos t] =
(s + )2 + 2

Table 2.1: Common Laplace transform pairs


Time function f (t)

2.2.5

Laplace transform L[f (t)] = F (s)

1.

unit impulse (t)

2.

unit step u(t)

3.

unit ramp t

4.

tn

n!
sn+1

5.

eat

1
s+a

6.

tn eat

7.

sin t

8.

cos t

9.

eat sin t

10.

eat cos t

1
s
1
s2

n!
(s + a)n+1

s2 + 2
s
s2 + 2

(s + a)2 + 2
s+a
(s + a)2 + 2

Inverse Laplace transform

The inverse transform of a function of s is given by the integral


Z +j
1
1
f (t) = L [F (s)] =
F (s)est ds
2j j
In general, this expression is difficult to evaluate. In practice, inverse transformation is most easily achieved by using partial fractions to break down solutions
into standard components, and then use tables of Laplace transform pairs, as
given in Table 2.1.

2.2.6

Partial fraction expansion

A rational function F (s) can be written as


F (s) =

Q(s)
P (s)

2.2. THE LAPLACE TRANSFORM

15

where P (s) and Q(s) are polynomials of s. Rational functions are defined as the
ratio of two polynomials. It is assumed that the order of P (s) in s is greater
than that of Q(s), F (s) is said to be strictly proper. The polynomial P (s) may
be written
P (s) = sn + an1 sn1 + + a1 s + a0
where a0 , a1 , , an1 are real coefficients. The roots of the polynomial P (s)
are referred to as poles of the function F (s).
Case 1: F (s) has distinct real poles
If all the poles of F (s) are real but distinct, F (s) can be written as
F (s) =

Q(s)
Q(s)
=
P (s)
(s + s1 )(s + s2 ) (s + sn )

which can be rewritten as a partial-fraction expansion


F (s) =

K1
K2
Kn
+
+ +
s + s1
s + s2
s + sn

To be able to determine the coefficients Ki (i = 1, 2, , n) we use the so-called


cover-up method


Ki = (s + si )F (s)
s=si

Find the inverse Laplace transform of


F (s) =

Example 2.6

s+2
s3 + 4s2 + 3s

 Solution We may write F (s) as


F (s) =

s+2
s(s + 1)(s + 3)

and in terms of its partial-fraction expansion:


F (s) =

K1
K2
K3
+
+
s
s+1 s+3

Using the cover-up method, we get






s+2
2

=
=
K1 = sF (s)

(s
+
1)(s
+
3)
3
s=0
s=0
In a similar fashion


K2 = (s + 1)F (s)
and



K3 = (s + 3)F (s)

s=1


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

s=3


s + 2
1
=
=

s(s + 1) s=3
6

With the partial fraction the solution can be looked up in the tables at once to
be
2 1
1

f (t) = et e3t
3 2
6

16

CHAPTER 2. SYSTEM MODELLING

Case 2: F (s) has distinct complex poles


We have to take special care of the quadratic factors in the denominator. The
numerator of the quadratic factor is chosen to be first-order as shown in the
following example.
Example 2.7

Find the inverse Laplace transform of


F (s) =

1
s(s2 + s + 1)

 Solution We rewrite F (s) as


F (s) =

K1
K2 s + K3
+ 2
s
s +s+1

Using the cover-up method, we find K1 to be




=1
K1 = sF (s)
s=0

We equate the numerators,


(s2 + s + 1) + (K2 s + K3 )s = 1
After equating like powers of s on the two sides of this equation, we find that
K2 = 1 and K3 = 1. To make it more suitable for using the tables we use
the method of completing the squares to rewrite the partial fraction as
F (s) =

s + 12 + 21
1

s (s + 12 )2 + 34

From the tables we have,


f (t) = 1 et/2 cos

3
4t

1
et/2 sin
3

3
4t

Case 3: F (s) has multiple-order poles


If r of the n poles of F (s) are identical, or we say that the pole at s = si is of
muliplicity r, F (s) is written
F (s) =

Q(s)
Q(s)
=
P (s)
(s + s1 )(s + s2 ) (s + snr )(s + si )r

(i 6= 1, 2, , n r). Then F (s) can be expanded as


F (s) =

A1
K1
K2
Knr
A2
Ar
+
+
+ +
+
+ +
s + s1
s + s2
s + snr s + si
(s + si )2
(s + si )r
|
{z
} |
{z
}
n r terms of distinct poles

r terms of repeated poles

The (n r) coefficients, K1 , K2 , , Knr , which corresponds to the distinct


poles, may be evaluated by the cover-up method. In, general, we may compute
Ak for a factor with multiplicity r as



1 dk
r

Ark =
(s
+
s
)
F
(s)
k = 0, 1, , r 1
i

k
k! ds
s=si

2.3. TRANSFER FUNCTIONS

17

Find the inverse Laplace transform of


F (s) =

Example 2.8

1
s(s + 1)3 (s + 2)

 Solution We rewrite the partial fraction as


F (s) =

K1
A3
K2
A1
A2
+
+
+
+
s
s + 2 s + 1 (s + 1)2
(s + 1)3

The coefficients corresponding to the distinct poles are




1
=
K1 = sF (s)
2
s=0


1
=
K2 = (s + 2)F (s)
2
s=2
and those of the third-order pole are


A3 = (s + 1) F (s)
= 1
s=1



d
A2 =
(s + 1)3 F (s)
=0
ds
s=1



1 d2
3

= 1
A1 =
(s
+
1)
F
(s)

2
2! ds
3

s=1

The completed partial-fraction expansion is


F (s) =

1
1
1
1
+

2s 2(s + 2) s + 1 (s + 1)3

The function f (t) is


f (t) =

2.3

1
1 1 2t
+ e
et t2 et
2 2
2

Transfer Functions

The classical way of modelling linear time-invariant systems is to use transfer functions to represent input-output relations between variables. A transfer
function is nothing more than the s plane representation of a physical system
that can be described by an ordinary differential equation with constant coefficients.
The transfer function of a linear time-invariant system is the ratio of the Laplace
transform of the output to the Laplace transform of the input, with all initial
conditions assumed to be zero
H(s) =

Y (s)
X(s)

The transfer function of a system represents the relationship describing the dynamics of the system under consideration. Another way of defining the transfer

18

CHAPTER 2. SYSTEM MODELLING

Figure 2.1: The transfer function.

function is to use the impulse response. The transfer function of a linear timeinvariant systems is defined as the Laplace transform of the impulse response,
with all initial conditions set to zero
H(s) = L[h(t)] =
Example 2.9

Y (s)
X(s)

Find the transfer function of the system described by the following differential
equation:
dy
d2 y
+3
+ 2y = 5
dt2
dt
with initial conditions
y(0) = 4 and y(0)

=3
 Solution Take Laplace transform and set all initial conditions to zero
s2 Y (s) + 3sY (s) + 2Y (s) =
Y (s) =

5
s
5
s(s2 + 3s + 2)

which can be rearranged as a ratio of the output to the input


H(s) =

2.4

1
Y (s)
= 2
(5/s)
s + 3s + 2

Models of Electric Circuits

In this section we develop models for simple electrical circuits. We know


that for a resistor, the voltage-current relationship in the time domain is
v(t) = Ri(t)

2.4. MODELS OF ELECTRIC CIRCUITS

19

Figure 2.2: Time-domain and s-domain representation of passive elements under


zero initial conditions.

Taking Laplace transform, we get


V (s) = RI(s)
For an inductor,
di(t)
dt
Taking Laplace and assuming zero initial conditions
v(t) = L

V (s) = sLI(s)
For a capacitor,
dv(t)
dt
which transforms into the s-domain (assuming zero initial conditions) as
i(t) = C

V (s) =

1
I(s)
sC

The s-domain equivalents are shown in Figure 2.2.


Example 2.10

Determine the transfer function H(s) = Vo (s)/Io (s) for the circuit shown in

20

CHAPTER 2. SYSTEM MODELLING

Figure 2.3: For Example 2.10.

Figure 2.3.
 Solution By current division,
I2 =

(s + 4)Io
s+4+2+

But
Vo = 2I2 =

1
2s

2(s + 4)Io
1
s + 6 + 2s

Hence,
H(s) =

Example 2.11

Vo (s)
4s(s + 4)
= 2
Io (s)
2s + 12s + 1

For the circuit shown in Figure 2.4, find the transfer function I2 (s)/V( s).

Figure 2.4: Two loop network for Example 2.11.

 Solution The first step in the solution is to convert the network into Laplace
transform for impedance and circuit variables, assuming zero initial conditions,
as shown in Figure 2.5. The circuit requires two simultaneous equations to
solve for the transfer function. These equations can be found by summing voltages around each mesh through which the assumed currents I1 (s) and I2 (s) flow.
Around Mesh 1,
R1 I1 (s) + LsI1 (s) LsI2 (s) = V (s)

2.5. MODELS OF MECHANICAL SYSTEMS

21

Figure 2.5: s-domain representation of the circuit in Figure 2.4.

Around Mesh 2,
LsI2 (s) + R2 I2 (s) +

1
I2 (s) LsI1 (s) = 0
Cs

By solving the two equations, we get


H(s) =

LCs2
I2 (s)
=
V (s)
(R1 + R2 )LCs2 + (R1 R2 C + L)s + R1

as shown in Figure 2.6.

Figure 2.6: The transfer function.

2.5

Models of mechanical systems

We have shown that electrical networks can be modeled by transfer function


that algebraically relates the Laplace transform of the output to the Laplace
transform of the input. Now, we will do the same for mechanical systems.
The motion of mechanical elements can be described in various dimensions as
translational, rotational, or combined. The equations governing the motion of
mechanical systems are often formulated directly or indirectly from Newtons
laws of motion.
Mechanical systems require one or more differential equations, called the equations of motion, to describe it. We will begin by assuming a positive direction of
motion, for example, to the right. This assumed positive direction of motion is
similar to assuming a current direction in an electrical loop. Using our assumed
direction of positive motion, we first draw a free-body diagram, placing on the
body all forces that act on the body either in the direction of motion or opposite
to it. Next, we use Newton?s law to form a differential equation of motion by
summing the forces and setting the sum equal to zero. Finally, assuming zero
initial conditions, we take the Laplace transform of the differential equation,
separate the variables and arrive at the transfer function.

22

2.5.1

CHAPTER 2. SYSTEM MODELLING

Translational motion

The motion of translation is defined as a motion that takes place along straight
lines. The variables that are used to describe translational motion are acceleration, velocity, and displacement. Newtons law states that the algebraic sum
of forces acting on a rigid body in a given direction is equal to the product
of the mass of the body and its acceleration in the same direction. Table 2.2
shows force-displacement translational relationship for spring, viscous damper
and mass. the constants K, B, and M are called spring constant, coefficient of
viscous friction, and mass, respectively.
Table 2.2: Force displacement translational relationship for spring, viscous damper
and mass.

Component

Force-displacement

f (t) = Kx(t)

f (t) = B

dx(t)
dt

f (t) = M

Example 2.12

d2 x(t)
dt2

Find the transfer function X(s)/F( s) for the system shown in Figure 2.7(a).

1
M s2 + Bs + K

Figure 2.7: (a) Mass, spring and damper system; (b) block diagram.

2.5. MODELS OF MECHANICAL SYSTEMS

23

 Solution Begin the solution by assuming a positive direction of motion.


Then, draw the free-body diagram as shown in Figure 2.8(a). Place on the
mass all forces felt by the mass. Only the applied force points to the right;
all other forces impede the motion and act to oppose it. Hence, the spring,
viscous damper and the mass due to acceleration point to the left. We write

Figure 2.8: (a) Free-body diagram of mass, spring and damper system; (b) transformed free-body diagram (in s domain).

the differential equation of motion using Newtons law to sum to zero all of the
forces shown on the mass in Figure 2.8(a):
M

d2 x(t)
dx(t)
+B
+ Kx(t) = f (t)
2
dt
dt

Taking the Laplace transform, assuming zero initial conditions,


M s2 X(s) + BsX(s) + KX(s) = F (s)
which is represented in Figure 2.8(b). Solving for the transfer function yields
H(s) =

1
X(s)
=
F (s)
M s2 + Bs + K

which is represented in Figure 2.7(b). 

2.5.2

Rotational motion

Rotational mechanical systems are handled the same way as translational mechanical systems, except that torque replaces force and angular displacement
replaces translational displacement. The mechanical components for rotational
systems are the same as those for translational systems, except that the components undergo rotation instead of translation. The rotational motion of a body
can be defined as motion about a fixed axis. The extension of Newtons law
of motion for rotational motion states that the algebraic sum of moments or
torque about a fixed axis is equal to the applied angular force or the product
of the inertia and the angular acceleration about the axis. Table 2.3 shows the
components along with the relationships between torque and angular velocity,
as well as angular displacement.
The constants K, D (or sometimes denoted as B) and J are called spring constant, coefficient of viscous friction, and moment of inertia, respectively.
The rotational system shown in Figure 2.9 consists of a disk mounted Example
on a
2.13

24

CHAPTER 2. SYSTEM MODELLING

Table 2.3: Torqur-angular displacement relationship for spring, viscous damper and
inertia.

Component

Torque-angular displacement

shaft that is fixed at one end. The moment of inertia of the disk about the axis
of rotation is J. The edge of the disk is riding on the surface, and the viscous
friction coefficient between the two surfaces is B. The inertia of the shaft is
negligible, but the torsional spring constant is K. Assume the torque is applied
to the disk, as shown; then the torque or moment of equation about the axis of
the shaft is written from the free-body diagram of Figure 2.9(b) as

T (t) = J

d2 (t)
d(t)
+B
+ K(t)
2
dt
dt

Figure 2.9: Rotational system for Example 2.13.

(2.2)

2.6. DC MOTORS IN CONTROL SYSTEMS

2.6

25

DC motors in control systems

A common actuator in control systems is a DC motor. A motor is an electromechanical component that yields a displacement output for a voltage input.
We will derive the transfer function for one particular kind of electromechanical
system, the armature-controlled dc servomotor. The motors schematic is shown
in Figure 2.10

Figure 2.10: A DC motor schematic.

The armature-controlled DC motor uses the armature current ia as the control


variable. In our analysis we need to include the back emf for the electrical circuit. For the mechanical part of the system we need to include the motor torque
in analyzing the rotor. Since the armature is rotating in a magnetic field, a back
electromotive force1 (back emf), e, is generated
e = Ke m

(2.3)

where Ke is a constant of proprtionality called the back emf constant and m =


m is the angular velocity of the motor. Taking the Laplace transform we have
E(s) = Ke s(s)

(2.4)

The relationship among the armature current ia , the applied armature voltage
va , and the back emf e is found by writing a loop equation around the Laplace
transformed armature circuit:
Ra Ia (s) + La sIa (s) + E(s) = Va (s)

(2.5)

The torque developed by the motor is proprtional to the armature current, thus
Tm (s) = Km Ia (s)

(2.6)

where Tm is the torque developed by the motor and Km is a constant of proportionality, called the motor-torque constant. Rearranging Equation (2.14)
yields
Tm (s)
Ia (s) =
(2.7)
Km
1 Because the generated electromotive force (emf) works against the applied armature voltage, we call it the back emf.

26

CHAPTER 2. SYSTEM MODELLING

To find the transfer function of the motor, we first substitute Equation (2.12)
and Equation (2.7) into Equation (2.13), yielding
(Ra + La s)Tm (s)
+ Ke s(s) = Va (s)
Km

(2.8)

Now we must find Tm (s) in terms of m (s) if we are to seperate the input
(s)
. The free-body
and output variables, and to obtain the transfer function Vma (s)
diagram for the rotor, shown in Figure 2.10, defines the positive direction and
shows the two applied torques, T and bm . Therefore,
(Jm s2 + bs)m (s) = Tm (s)

(2.9)

Substituting Equation (2.9) into Equation (2.8) yiields


(Ra + La s)(Jm s2 + bs)m (s)
+ Ke s(s) = Va (s)
Km

(2.10)

If we assume that the armature inductance La is small compared to the armature


resistance Ra , which is usually the case for a dc motor, Equation (2.10) becomes


Ra (Jm s + b)
+ Ke sm (s) = Va (s)
Km
After simplification, the desired transfer function

m (s)
Va (s)

is found to be

Km
m (s)
=
Va (s)
s [Ra (Jm s + b) + Km Ke ]
Km
=
s [Jm Ra s + bRa + Km Ke ]
The relations of the armature-controlled DC motor are shown schematically in
Figure 2.11

Figure 2.11: Armature-controlled DC motor.

2.7
2.7.1

System modelling diagrams


The block diagram

Block diagrams may be considered as a form of system description that provides a simplified overview schematic diagram of a system. It describes the

2.7. SYSTEM MODELLING DIAGRAMS

27

composition and interconnection of a system, or it can be used together with


the transfer functions to describe the cause-and-effect relationships throughout
the system. The transfer function of each component is placed in a box, and
the input-output relationships between components are indicated by lines and
arrows. We can then solve the equations by graphical simplification, which is
often easier and more informative than algebraic manipulation.
Many practical control systems consist of complicated interconnection of smaller
subsystems. Before tackling a control system design for such systems, it is usually helpful to simplify the complex interconnection of subsystems. Essentially,
we seek a systematic way to eliminate variables (signals) we do not want to
control or measure.
Block diagram algebra
Block diagrams usually consist of (see Figure 2.12):
1. Blocks: these give a description of subsystem dynamics.
2. Summers: add or subtract two or more signals.
3. Arrows: these give the direction of signal propagation.
4. Take off points.
Summing point

Take off point

Figure 2.12: Block diagram of a closed-loop system.

Block diagram transformations and reduction techniques are derived by considering the algebra of the diagram variables. For example, consider the block
diagram shown in Figure 2.12. This negative feedback control system is described by the equation for the error signal, which is
E(s) = R(s) B(s) = R(s) H(s)Y (s)
Because the output is related to the error signal by G(s), we have
Y (s) = G(s)E(s)
thus,
Y (s) = G(s)[R(s) H(s)Y (s)]
Solving for Y (s), we obtain
Y (s)[1 + G(s)H(s)] = G(s)R(s)

28

CHAPTER 2. SYSTEM MODELLING

Therefore, the transfer function relating the output Y (s) to the input R(s) is
Y (s)
G(s)
=
R(s)
1 + G(s)H(s)
The gain of a single-loop negative feedback system is given by the forward gain
divided by the sum of one plus the loop gain. When the feedback is added instead of subtracted, we call it positive feedback. In this case the gain is given by
the forward gain divided by the sum of 1 minus the loop gain.
A control system may have several feedback control loops as the one shown
in Figure 2.13. In principle, the block diagram of a closed-loop system, no matter how complicated it is, it can be reduced to the standard single loop form
shown in Figure 2.12. Reduction of complex block diagrams is facilitated by a
series of easily derivable transformations which are summarized in Table 2.4.

Figure 2.13: Block diagram of a closed-loop system.

The following steps may be used to simplify complicated block diagrams:


1. Combine all cascade blocks.
2. Combine all parallel blocks.
3. Eliminate all minor (interior) feedback loops.
4. Shift summing points to left.
5. Shift take off points to the right.
6. Repeat steps 1 to 5 if necessary.
Block diagram transformations will be illustrated by examples using block diagram reduction.

2.7. SYSTEM MODELLING DIAGRAMS

29

A block diagram of a multiple-loop feedback control system is shown in Figure


2.13. It is interesting to note that the feedback signal H1 (s)Y (s) is a positive
feedback signal, and the loop G3 (s)G4 (s)H1 (s) is a positive feedback loop. First,
to eliminate the minor loop G3 G4 H4 , we move H2 behind block G4 by using
rule 10 (see Table 2.4), and therefore obtain Figure 2.14.

Figure 2.14: Block diagram reduction of the system of Figure 2.13.

Eliminating the loop G3 G4 H1 by using rule 4, we obtain Fogure 2.15.

Figure 2.15: Block diagram simplification.

Then eliminating the inner loop containing H2 /G4 , we obtain Figure 2.16(a).
Finally, by reducing the loop containing H3 , we obtain the closed-loop system
transfer function as shown in Figure 2.16(b).

Example 2.14

30

CHAPTER 2. SYSTEM MODELLING

Figure 2.16: Reduced block diagram.

Example 2.15

Find the transfer function of the system shown in Figure 2.17.

Figure 2.17: Block diagram of Example 2.15.

 Solution Moving the first summing point ahead of G1 , and the final take
off point beyond G4 gives a modified block diagram shown in Figure 2.18(a).
The block diagram in Figure 2.18(a) is then reduced to the form given in Figure
2.18(b).

Figure 2.18: Stages of block diagram reduction.

2.7. SYSTEM MODELLING DIAGRAMS

31

The overall closed-loop transfer function is then


Y (s)
=
R(s)
1+
=

G1 G2 G3 G4
(1+G1 G2 H1 )(1+G3 G4 H2 )
G 1 G 2 G 3 G 4 H3
(G1 G4 )(1+G1 G2 H1 )(1+G3 G4 H2 )

G1 G2 G3 G4
(1 + G1 G2 H1 )(1 + G3 G4 H2 ) + G2 G3 H3

Table 2.4: Block diagram transformations.

Transformation

Equation

1.

Cascaded blocks

Y = (P1 P2 )X

2.

Combining
blocks in parallel

Y = P1 X P2 X

3.

Removing a block
from a forward
loop

Y = P1 X P2 X

4.

Eliminating feedback loop

Y = P1 (X P2 Y )

5.

Removing a block
from a feedback
loop

Y = P1 (X P2 Y )

6.

Rearranging
summing
junctions

Z =W X Y

7.

Moving a summing junction in


front of a block

Z = PX Y

Block diagram

Equivalent block diagram

32

CHAPTER 2. SYSTEM MODELLING


Table 2.4: .... continued.

8.

Moving a summing junction beyond a block

Z = P (X Y )

9.

Moving a takeoff
point in front of a
block

Y = PX

10.

Moving a takeoff
point beyond a
block

Y = PX

11.

Moving a takeoff
point in front of
a summing junction

Z =X Y

12.

Moving a takeoff point beyond


a summing junction

Z =X Y

Systems with multiple inputs


In feedback control systems, we often encounter multiple inputs. For a linear
system, we can apply the principle of superposition to solve this type of problems, i.e. to treat each input one at a time while setting all other inputs to
zeros, and then algebraically add all the outputs as follows:
1. Set all inputs except one equal to zero.
2. Transform the block diagram to solvable form.
3. Find the output response due to the chosen input action alone.
4. Repeat steps 1 to 3 for each of the remaining inputs.
5. Algebraically summ all the output responses found in steps 1 to 5.

2.7. SYSTEM MODELLING DIAGRAMS

33

Find the complete output for the system shown in Figure 2.19 when both inputs
act simultaneously.
Y (s)

Figure 2.19: System with multiple inputs.

 Solution The block diagram shown in Figure 2.19 can be reduced and simplified to the form given in Figure 2.20.
Y (s)

Figure 2.20: Reduced and simplified block diagram.

Putting R2 (s) = 0 and replacing the summing point by +1 gives the block diagram shown in Figure 2.21. In Figure 2.21 nothe that Y1 (s) is response to R1 (s)
acting alone.
Y1 (s)

Figure 2.21: Block diagram for R1 (s) acting alone.

The closed-loop transfer function is therefore


G G

1 2
Y1 (s)
2 H2
= 1+G
1 G 2 H1
R1 (s)
1+ G
1+G2 H2

Example 2.16

34

CHAPTER 2. SYSTEM MODELLING

or
Y1 (s) =

G1 G2 R1
1 + G2 H2 + G1 G2 H1

Now if R1 (s) = 0 and the summing point is replaced by 1, then the response
Y2 (s) to input R2 (s) acting alone is given by Figure 2.22. The choice as to
whether the summing point is replaced by +1 or 1 depends upon the sign at
the summing point.

Y2 (s)

Figure 2.22: Block diagram for R2 (s) acting alone.

Note that in Figure 2.22 there is a positive feedback loop. Hence the closed-loop
transfer functionn relating R2 (s) and Y2 (s) is
G G H

1 2 1
Y2 (s)
1+G2 H2 
=
R2 (s)
1 G2 H1
1 G
1+G2 H2

or
Y2 (s) =

G1 G2 H1 R2
1 + G2 H2 + G1 G2 H1

Using the principle of superposition, the complete response is given by


Y (s) = Y1 (s) + Y2 (s)
or
Y (s) =

2.7.2

G1 G2 R1 G1 G2 H1 R2
1 + G2 H2 + G1 G2 H1

Signal-flow graph

Block diagrams are adequate for the representation of the interrelationships of


controlled and input variables. However, for a system with reasonably complex interrelationships, the block diagram reduction technique is cumbersome
and often quite difficult to complete. An alternative method for determining
the relationship between system variables has been developed by Mason and is
based on a representation of the linear system by line segments called SignalFlow Graph (SFG). The advantage of the SFG method is the availability of a
flow graph gain formula, which provides the relation between system variables
without requiring any reduction procedure or manipulation of the flow graph.
Basic elements of an SFG
A signal flow graph is a pictorial representation a set of simultaneous linear
algebraic equations describing a system. When constructing a SFG, junction

2.7. SYSTEM MODELLING DIAGRAMS

35

points or nodes are used to represent variables. The nodes are connected by
line segments, called branches. A signal can transmit through a branch only in
the direction of the arrow. As an example consider a linear system represented
by a simple algebraic equation
y2 = a12 y1
where y1 is the input, y2 is the output, and a12 is the gain between the two
variables. The SFG representation is shown in Figure 2.23

Figure 2.23: Signal-flow graph of y2 = a12 y1 .

Construct a SFG to the following set of algebraic equations:


y2 = a12 y1 + a32 y3
y3 = a23 y2 + a43 y4
y4 = a24 y2 + a34 y3 + a44 y4
y5 = a25 y2 + a45 y4
 Solution Figure 2.24 shows a step-by-step construction of the signal-flow
graph. The complete SFG is shown in Figure 2.25.


Figure 2.24: Step-by-step construction of the signal-flow graph.

Basic properties of SFG


The important properties of the SFG are summarised as follows:

Example 2.17

36

CHAPTER 2. SYSTEM MODELLING

Figure 2.25: Complete signal-flow graph.

1. SFG applies only to linear systems.


2. The equations of which SFG is drawn must be algebraic equations in the
form of cause and effect.
3. Nodes are used to represent variables. Normally, the nodes are arranged
from left to right, from the input to the output, following succession of
cause-and-effect relations through the system.
4. Signals travel along branches only in the direction described by the arrows
of the branches.
5. The branch directing from yk to yj represents the dependence of yj upon
yk but not the reverse.
6. A signal yk traveling along a branch between yk and yj is multiplied by
the gain of the branch akj , so that a signal akj yk is delivered at yj .
Definitions of the SFG terms
Before proceeding further, we define some terms which we will need later:
A source is a node which has outgoing branches only (Example: y1 in
Figure 2.25).
A sink is a node which has only incoming branches (Example: y5 in Figure
2.25).
A path is a set of branches having the same sense of direction.
A forward path originates from a source and terminates in a sink. No
node may be encountered more than once. In the SFG of Figure 2.25, there
are three forward paths between y1 and y5 . One such path for example is
from y1 to y2 to y5 (through the branch with gain a25 ).
The path gain is the product of the coefficients associated with the
branches of the path. For example the path gain for the path y1 y2 y5
in Figure 2.25 is a12 a25 .
A feedback loop is a path that begins and ends at the same node; additionally no node may be encountered more than once. For example there
are four loops in the SFG of Figure 2.25. These are shown in Figure 2.26.
The loop gain is the path gain of a feedback loop. For example, the loop
gain of the loop y2 y4 y3 y2 in Figure 2.26 is a24 a43 a32 .

2.7. SYSTEM MODELLING DIAGRAMS

37

Figure 2.26: Four loops in the signal-flow graph of Figure 2.25.

Masons Rule
Given an SFG or block diagram, the task of solving for the input-output relations by algebraic manipulation could be quite tedious. Fortunately, there is a
general gain formula available that allows the determination of the input-output
relations of an SFG by inspection.
Masons states that the input-output transfer function associated with a signalflow graph is given by
P
Pk k
G= k
(2.11)

where
P
P
P
P
= 1 L1 + L2 L3 + + (1)m Lm
and
Pk = gain of the k th forward path
L1 = gain of each closed loop in the graph
L2 = product of loop gains of any two nontouching loops (loops are called nontouching if they have no node in common)
..
.
Lm = product of loop gains of any m nontouching loops
k = the value of remaining with the loops touching the path Pk are removed
Procedures to solve SFG by using Masons rule:
1. Identify the no. of forward paths and determine the forward-path gains.
2. Identify the no. of loops and determine the loop gains.
3. Identify the non-touching loops taken two at a time, three at a time and
so on. Determine the product of the non-touching loop gains.

38

CHAPTER 2. SYSTEM MODELLING


4. Determine and k .
5. Substitute all of the above information into the Masons gain formula.

Example 2.18

Determine the closed-loop transfer function Y (s)/R(s) of the SFG shown in


Figure 2.27.

Figure 2.27: Signal-flow graph for example 2.18.

 Solution
1. Forward path: There is only one forward path between R(s) and Y (s),
and the forward-path gain is
P1 = G(s)
2. Closed loops: There is only one loop; the loop gain
L1 :

G(s)H(s)

3. Non-touch loops: There are no non-touching loops since there is only one
loop.
4. and 1 : The forward path is in touch with the only loop. Thus, 1 = 1,
and
= 1 L1 = 1 + G(s)H(s)
5. Using (2.11), the closed-loop transfer function is written as
P1 1
G(s)
=

1 + G(s)H(s)

2.7. SYSTEM MODELLING DIAGRAMS

39

Determine the gain between y1 and y5 using the gain formula for the SFG shown
in Figure 2.25.
 Solution
1. Forward path: There are three forward paths between y1 and y5 and
forward-path gains are
P1 = a12 a23 a34 a45

Forward path:

y1 y2 y3 y4 y5

P2 = a12 a25

Forward path:

y1 y2 y5

P3 = a12 a24 a45

Forward path:

y1 y2 y4 y5

2. Closed loops: The four loops of the SFG are shown in Figure 2.26. The
loop gains are
L1 :

a23 a32

a34 a43

a24 a43 a32

a44

hence
X

L1 = a23 a32 + a34 a43 + a24 a43 a32 + a44

3. Non-touch loops: There is only one pair of non-touching loops; that is, the
two loops
y2 y3 y2
and
y4 y4
Thus the product of the gains of the two non-touching loops is
L2 :
4. and k : = 1

L1 +

a23 a32 a44


L2 hence

= 1 (a23 a32 + a34 a43 + a24 a43 a32 + a44 ) + a23 a32 a44
All the loops are in touch with forward paths P1 and P3 . Thus, 1 =
3 = 1. Two of the loops are not in touch with forward path P2 . These
loops are: y3 y4 y3 and y4 y4 . Thus,
2 = 1 a34 a43 a44
5. Using (2.11), the closed-loop transfer function is written as
P1 1 + P2 2 + P3 3

a12 a23 a34 a45 + (a12 a25 )(1 a34 a43 a44 ) + a12 a24 a45
=
1 (a23 a32 + a34 a43 + a24 a43 a32 + a44 ) + a23 a32 a44

G=

Application of Masons rule between a source and a non-sink node


In the previous example, Example 2.19, we basically determined the transfer
function between y1 (source node) and y5 (sink node). Often, it is of interest to
find the relation between a source and a non-sink node. For example, in the SFG
of Figure 2.25, it may be of interest to find the relation y2 /y1 , which represents
dependence of y2 upon y1 ; noting that y2 is not a sink node. To make y2 a sink

Example 2.19

40

CHAPTER 2. SYSTEM MODELLING

Figure 2.28: Modification of signal-flow graph so that y2 satisfy the condition as


sink node.

node, we simply connect a branch with unity gain from the existing node y2 to
a new node also designated as y2 , as shown in Figure 2.28.
Example 2.20

Determine the gain between y1 and y2 using the gain formula for the SFG
shown in Figure 2.28.
 Solution
1. Forward path: There is only one forward path between y1 and y2 and
forward-path gain is
P1 = a12

Forward path:

y1 y2

2. Closed loops: The four loops of the SFG are shown in Figure 2.26. The
loop gains are
L1 :

a23 a32

a34 a43

a24 a43 a32

a44

hence
X

L1 = a23 a32 + a34 a43 + a24 a43 a32 + a44

3. Non-touch loops: There is only one pair of nontouching loops; that is, the
two loops
y2 y3 y2
and
y4 y4
Thus the product of the gains of the two nontouching loops is
L2 :

a23 a32 a44

4. and k : = 1 (a23 a32 + a34 a43 + a24 a43 a32 + a44 ) + a23 a32 a44
Two of the loops are not in touch with forward path P1 . These loops
are: y3 y4 y3 and y4 y4 . Thus,
1 = 1 a34 a43 a44
5. Using (2.11), the transfer function between y1 and y2 is written as
G=
=

P1 1

a12 (1 a34 a43 a44 )


1 (a23 a32 + a34 a43 + a24 a43 a32 + a44 ) + a23 a32 a44

Note that is the same as in Example 2.19 regardless of which sink node
is chosen.


2.7. SYSTEM MODELLING DIAGRAMS

41

Application of Masons rule between a non-source and sink node


We have seen earlier how to determine the gain between a source and a non-sink
node. Another situation of interest is find the relation between a non-source and
a sink node. For example, in the SFG of Figure 2.29, it may be of interest to find
the relation y7 /y2 , which represents the dependence of y7 upon y2 , the latter is
not a source node.

Figure 2.29: Signal-flow graph for Example 2.21.

We can show that by including a source node (y1 in this case), we may write
y7 /y2 as

P

k Pk k


y7 /y1
y7
from y1 to y7
=
= P

y2
y2 /y1
k Pk k


from y1 to y2

Since is independent of the sources and sinks, the last equation is written


P

P

k k k
y7
y7 /y1
from y1 to y7
=
=

P
y2
y2 /y1

k Pk k
from y1 to y2

Note that does not appear in the last equation. However, you must evaluate
it to be able to find k .
Determine the gain between y7 and y2 for the SFG shown in Figure 2.29.
 Solution We start by determining y2 /y1 :
1. Forward path: There is only one forward path between y1 and y2 and
forward-path gain is
P1 = 1

Forward path:

y1 y2

2. Closed loops: There are four loops


y2 y3 y2

y4 y5 y4

y2 y3 y4 y5 y2

y4

with gains
L1 :

G1 H1

G3 H2

G1 G2 G3 H3

H4

Example 2.21

42

CHAPTER 2. SYSTEM MODELLING


Hence,
X

L1 = (G1 H1 + G3 H2 + G1 G2 G3 H3 + H4 )

3. Non-touch loops:
Product of loop gains of any two nontouching loops (there are four
possible combinations), thus:
L2 :
and
X

G1 G3 H1 H2

G1 H1 H4

G3 H2 H4

G1 G2 G3 H3 H4

L2 = G1 G3 H1 H2 + G1 H1 H4 + G3 H2 H4 + G1 G2 G3 H3 H4

Product of loop gains of any three nontouching loops:

4. and k : = 1

L3 :
G1 G3 H1 H2 H4
P
P
L1 + L2 L3 , therefore,

= 1 + G1 H1 + G3 H2 + G1 G2 G3 H3 + H4 + G1 G3 H1 H2
+ G1 H1 H4 + G3 H2 H4 + G1 G2 G3 H3 H4 + G1 G3 H1 H2 H4
Two of the loops are not in touch with forward path P1 . These loops are:
y4 y5 y4 and y4 y4 . Thus,
1 = 1 + G3 H2 + H4 + G3 H2 H4
5. Using (2.11), the gain between y1 and y2 is written as
1 + G3 H2 + H4 + G3 H2 H4
P1 1
y2
=
=
y1

In a similar fashion we now determine the gain between y1 and y7 :


1. Forward path: There are two forward path between y1 and y7 , the forwardpaths gains are
P1 = G1 G2 G3 G4

Forward path:

y1 y7

P2 = G1 G5

Forward path:

y1 y2 y3 y6 y7

2. Closed loops: As before.


3. Non-touch loops: As before
4. and k : The forward path P1 is in touch with all loops. Thus, 1 = 1.
One loop is not in touch with forward path P2 , y4 y5 y4 , thus
2 = 1 + G3 H2
5. Using (2.11), the gain between y1 and y7 is written as
y7
P1 1 + P2 2
G1 G2 G3 G4 + G1 G5 (1 + G3 H2 )
=
=
y1

Finally, the gain between y2 and y7 is


y7
y7 /y1
G1 G2 G3 G4 + G1 G5 (1 + G3 H2 )
=
=
y2
y2 /y1
1 + G3 H2 + H4 + G3 H2 H4

2.7. SYSTEM MODELLING DIAGRAMS

2.7.3

43

Conversion from block diagram to SFG

An equivalent SFG for a block diagram can be drawn by performing the following
steps:
1. Identify the input/output signals, summing junctions & pickoff pointsthey
are replaced with nodes.
2. Interconnect the nodes & indicate the directions of signal flow by using
arrows.
3. Identify the blocks they are replaced with branches. For each negative
sum, a negative sign is included with the branch.
4. Label the input/output signals and the branches accordingly.
5. Add unity branches as needed for clarity or to make connections.
6. Simplify the SFG eliminate redundant nodes/branches (only if the node
is connected to branches of a single flow in & a single flow out with unity
gain).

Figure 2.30: Block diagrams and corresponding signal-flow graphs.

44
Example 2.22

CHAPTER 2. SYSTEM MODELLING

Convert the block diagram in Figure 2.31 to a signal flow graph and determine
the transfer function using Masons gain formula.

Figure 2.31: (a) Block diagram a control system. (b) Equivalent signal-flow graph.

 Solution
1. Forward path:
P1 = G1 G2 G3
P2 = G1 G4
2. Closed loops:
L1 :

G1 G2 H1

G2 G3 H2

G1 G2 G3

G1 G4

G4 H2

3. Non-touch loops: There are no nontouching loops.


P
P
P
4. and k : = 1 L1 + L2 L3 , therefore,
= 1 + G1 G2 H1 + G2 G3 H2 + G1 G2 G3 + G1 G4 + G4 H2
All the loops are in touch with P1 and P2 , thus 1 = 2 = 1.
5. Using (2.11), the transfer function between Y (s) and R(s) is written as
P1 1 + P2 2
G1 G2 G3 + G1 G4
Y (s)
=
=
R(s)

2.7. SYSTEM MODELLING DIAGRAMS

2.7.4

45

Construction of block diagrams Examples

Construct a block diagram for the mechanical system described by the following
set of equations
B2 s + K
F (s)
X3 (s) +
m1 s2 + (B1 + B2 )s + k
m1 s2 + (B1 + B2 )s + k
B3
X2 (s) =
X3 (s)
m2 s + B3 + B4
B2 s + K
B3 s
X3 (s) =
X1 (s) +
X2 (s)
(B2 + B3 )s + k
(B2 + B3 )s + k
X1 (s) =

(2.12)
(2.13)
(2.14)

 Solution Each dynamic equation represents a subsystem. Its block diagram


is constructed by a simple principle: treat the right hand side signals as the
input and the left hand side as the output. Equation (2.12) can be represented
as

Figure 2.32: Block diagram of Equation (2.12).

Equation (2.13) can be represented as

Figure 2.33: Block diagram of Equation (2.13).

and Equation (2.14)

Figure 2.34: Block diagram of Equation (2.14).

After construction of block diagrams for individual equations, we connect these

Example 2.23

46

CHAPTER 2. SYSTEM MODELLING

Figure 2.35: Block diagram of the system in Example 2.23.

block diagrams to form a block diagram for the entire system:


Example 2.24

Construct a block diagram for a system described by the following set of equations
1
Pc (s)
CRd s + 1
1
Pi (s) =
Pc (s)
CRi s + 1
Pc (s) = KX(s)
a
b
E(s) +
Y (s)
X(s) =
a+b
a+b
A
Y (s) =
[Pi (s) Pd (s)]
Ks

Pd (s) =

(2.15)
(2.16)
(2.17)
(2.18)
(2.19)

 Solution Each dynamic equation represents a subsystem. Its block diagram


is constructed by a simple principle: treat the right hand side signals as the
input and the left hand side as the output.

Figure 2.36: Block diagram of Equation (2.15).

2.7. SYSTEM MODELLING DIAGRAMS

Figure 2.37: Block diagram of Equation (2.16).

Figure 2.38: Block diagram of Equation (2.17).

Figure 2.39: Block diagram of Equation (2.18).

Figure 2.40: Block diagram of Equation (2.19).

Figure 2.41: Block diagram of the system in Example 2.24.

47

48

CHAPTER 2. SYSTEM MODELLING

Chapter

Time-Domain Analysis of Control


Systems
3.1

Introduction

This chapter refers to the time-domain analysis of linear time-invariant control systems. The problem of time-domain analysis may be briefly stated as
follows: given the system (i.e., given a specific description of the system) and
its input, determine the time-domain behavior of the output of the system.
In the analysis problem, we will use selected input signals to test the response
of control systems. This response will be characterized by a selected set of
response measures. The basic motivation for system analysis is that one can
predict (theoretically) the systems behavior.

3.2

System Time Responses

The manner in which a dynamic system responds to an input, expressed as


a function of time, is called the time response. It is possible to compute the
time response of a system if the following is known:
the nature of the input, expressed as a function of time
the mathematical model of the system.
The time response of any control system has two components:
(a) Transient response: Is that particular part of the response of the system
which tends to zero as time increases. It is a function only of the system
dynamics, and is independent of the input signal.
(b) Steady-state response: Is that particular part of the response of the system that remains after the transient component has reached zero. It is a
function of both the system dynamics and the input signal.
49

50

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS

The total response of the system is always the sum of the transient and steadystate components. In the design problem, specifications are usually given in
terms of the transient and steady-state performance, and controllers are designed so that the specifications are all met by the design system.
For control system design, we need a basis of comparison of performance of
different designs. One way of setting up this basis is to specify particular test
signals and compare the response of different designs to these signals.

3.3

Common test Signals

All the following signals are defined for t 0.

3.3.1

The unit impulse function

The unit impulse function is based on a rectangular function  (t) such that
 (t) =

1
,


0 t ,

>0

As  0,  (t) approaches the unit impulse (t). The major properties of this
function are
Z
1.
(t)dt = 1
0

g(t)(t a)dt = g(a)

2.
0

3. L[(t)] = 1
It is very useful for modeling shock inputs.

3.3.2

The unit step function

The unit step function is defined as:


r(t) = 1,

t0

= L[r(t)] =

1
s

and is useful for modeling sudden disturbances.

3.3.3

The unit ramp function

The unit ramp function is defined as:


r(t) = t,

t0

= L[r(t)] =

It is useful for modeling gradually changing inputs.

1
s2

3.4. RESPONSE OF FIRST ORDER SYSTEMS

3.3.4

51

The sinusoidal functions

The sinusoidal functions


s
s2 + 2

= L[sin t] = 2
s + 2

cos t = Re ejt = L[cos t] =


sin t = Im ejt

are very important in frequancy response techniques.

3.4
3.4.1

Response of first order systems


Standard form

Consider a first-order differential equation


a

dy(t)
+ by(t) = cr(t)
dt

Take Laplace transoform with zero initial conditions


asY (s) + bY (s) = c R(s)
(as + b)Y (s) = c R(s)
The first order transfer function is
G(s) =

Y (s)
c
=
R(s)
as + b

To obtain the standard form, divide by b


G(s) =

c/b
1 + (a/b)s

which is written as
G(s) =

3.4.2

K
1 + s

Step response

The response of the control system to the unit-step input is called the unit-step
response. When r(t) is a unit step


K
1
1
Y (s) =
=K

s( s + 1)
s s + 1/
= y(t) = K(1 et/ ) = K Ket/

(3.1)

The first term is called the steady-state response and K is called the steady-state
value. The second term is called the transient response. If > 0 the transient
response tends to 0 as t . The step response of a first-order system as given
in (3.1) is shown in Figure 3.1. Note that the exponentially decaying term has
an initial slope of K/ ; that is,



d 
K
K
K Ket/
= et/
=
dt

t=0
t=0

52

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS


y(t)
K
K(1 et/ )
0.632K

Figure 3.1: Step response of first -order systems.

Mathematically, the exponential term does not decay to zero in a finite length
of time. However, if the term continued to decay at its initial rate, it would
reach a value of zero in seconds. The parameter is called the time constant
and has the units of seconds. The exponential function decays to about 2% of
its initial value within 4 time constants. The output y(t) reaches about 63% of
its final vaue when t = .
Example 3.1

An example of a first order system is provided by the following RC circuit.

Figure 3.2: A simple resistor-capacitor circuit.

Vout (s)
1
=
Vin (s)
1 + sRC

3.4.3

(K = 1, = RC) 

Steady-state response

The concept of finding the steady-state response to a unit step for a system of
any order is now developed. Suppose that
Y (s) = G(s)R(s)
where G(s) is a given transfer function. From Section 2.2.3, the final-value
theorem of the Laplace transform is
lim y(t) = lim sY (s) = lim sG(s)R(s)

s0

s0

provided that the limit


lim y(t)

3.5. RESPONSE OF SECOND ORDER SYSTEMS

53

exists, i.e., y(t) has a final value. For the case that the input is a unit step, R(s)
is equal to 1/s and
lim y(t) = lim sG(s)

s0

1
= lim G(s)
s s0

(3.2)

= G(0)
G(0) is often called the dc gain of the system and is defined as the ratio of the
output of a system to a constant input after all transients has decayed. Care
must be taken to apply the Final value Theorem only to stable systems (i.e.,
y(t) is bounded) and with at most a single pole at s = 0.
Find the dc gain of the system whose transfer function is
G(s) =

3(s + 2)
(s2 + 2s + 10)

 Solution Applying (3.2), we get




dc gain = G(s)

s=0

3.5
3.5.1

(3)(2)
= 0.6
(10)

Response of second order systems


Standard form

Consider a second-order differential equation


a

dy(t)
d2 y(t)
+b
+ cy(t) = er(t)
2
dt
dt

Take Laplace transforms, with zero initial conditions


as2 Y (s) + bsY (s) + CY (s) = eR(s)
(as2 + bs + c)Y (s) = eR(s)
The transfer function is
G(s) =

Y (s)
e
= 2
R(s)
as + bs + c

To obtain the standard form, divide by c


G(s) =

(a/c)s2

e/c
+ (b/c)s + 1

which is written as
G(s) =

K
(1/n2 )s2 + (2/n )s + 1

and with the s2 coefficient normalized to unity


G(s) =
where

Example 3.2

s2

Kn2
+ 2n s + n2

54

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS


: damping ratio

n : undamped natural frequency


K : dc gain
Since K only determines the dc magnitude of the response, we will set K = 1
and so
G(s) =

n2
s2 + 2n s + n2

(3.3)

where
s1 , s2 = n n

2 1

are the roots of the characteristic equation and are the poles of G(s).

3.5.2

Pole locations

All system characteristics of the standard second-order system are functions


of only and n , since and n are the only parameters that appear in the
transfer function (3.3). There are four cases of practical interest:
Case 1: > 1 (Overdamped system)
p
When > 1 the roots s1 , s2 = n n 2 1 are real, negative and unequal.

>1

Figure 3.3: Pole locations in the s-plane when > 1.

Case 2: = 1 (Critically damped system)


When = 1 the roots s1 , s2 = n are real, negative and equal.

=1

Figure 3.4: Pole locations in the s-plane when = 1.

3.5. RESPONSE OF SECOND ORDER SYSTEMS

55

Case 3: 0 < 1 (Underdamped system)


When 0 < 1 the roots
s1 , s2 = n jn

1 2

are complex conjugate and have negative real parts. The real parts are zero if
= 0 and the system is called undamped.

0<1

Figure 3.5: Pole locations in the s-plane when 0 < 1.

Case 4: < 0 (Unstable system)


When < 0 the roots s1 , s2 have positive real parts and will be covered later.

Figure 3.6: Pole locations in the s-plane when < 0.

Figure 3.7 summarizes all four cases.

Figure 3.7: Pole locations in the s-plane and the corresponding transient response
type.

56

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS

3.5.3

Step response

Consider the second-order system described by the transfer function given in


(3.3), the step response is
Y (s) = G(s)R(s) =

n2
s(s2 + 2n s + n2 )

Expanding in partial fraction


Y (s) =

K1
K2 s + K3
+ 2
s
s + 2n s + n2

Using the cover-up method, we find




K1 = sY (s)

=1
s=0

Equating the numerators,


(s2 + 2n s + n2 ) + K2 s2 + K3 s = n2
After equating the powers of s on the two sides of the above equation, we find
that K2 = 1 and K3 = 2n , hence
Y (s) =

1
s + 2n

s s2 + 2n s + n2

Completing the square


s + 2n
1

s (s + n )2 + n2 2 n2
1
s + 2n
=
 p
2
s
(s + n )2 + n 1 2

Y (s) =

and writing it in the standard forms in Table 2.1 we have




s + n
n
1
d

Y (s) =
s (s + n )2 + d2
d (s + n )2 + d2

(3.4)

where
d = n

p
1 2

is called the damped natural frequency. Taking the inverse Laplace transform
 n t


y(t) = 1 en t cos d t p
e
sin d t
2
1
!

n t
=1e
cos d t + p
sin d t
1 2

(3.5)

Using the trigonometric identity, sin(a + b) = sin a cos b + cos a sin b, (3.5) can
be written as
1
y(t) = 1 p
en t sin(d t + )
(3.6)
1 2

3.5. RESPONSE OF SECOND ORDER SYSTEMS

57

where
1

1 2
(3.7)

p
is as shown (note that cos = and sin = 1 2 ). The step responses for a
second-order system are shown in Figure 3.8 for several values of as a function
of n t.
1

= tan

y(t)

n t
Figure 3.8: Step response for second-order system.

With reference to Figure 3.8:


If = 0, from (3.5), noting that d = n
y(t) = 1 cos n t
the frequency of the sinusoid is n , called the undamped natural frequency.
The step input will cause the system to oscillate continuously at n .
p
If 0 < < 1, d = n 1 2 is the frequency of the sinusoid, hence,
damped natural frequency. The response is a damped sinusoid, and is
called damped (or underdamped) system. The time constant of the exponential envelope is = 1n . As increases from 0 to 1, the response
becomes less oscillatory, and hence the name damping ratio.
If 1, the oscillations have ceased and hence the system is called an
overdamped system when > 1 and critically damped when = 1. For
the case = 1 it is not clear from (3.5) how the oscillations in y(t) have
ceased. However, before taking the inverse Laplace to Y (s) in (3.4) let
= 1 to obtain


1
s + n
n
d
Y (s) =

s (s + n )2 + 0
d (s + n )2 + 0
1
1
n
=

s s + n
(s + n )2

p
1 2

58

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS


Taking inverse Laplace transform
y(t) = 1 en t n ten t
= 1 en t (1 + n t)
which clearly shows that no sinusoidal term exist. For the overdamped
case, we have two real poles at n d . The corresponding response
is easily obtained from
Y (s) =

K2
K3
1
+
+
s s + n + d
s + n d

as
y(t) = 1 + K2 e(n +d )t + K3 e(n d )t
As an exercise find K2 and K3 .
The effect of the characteristic equation roots on the damping of the secondorder system is further illustrated by Figure 3.9

Figure 3.9: Step response comparison for various pole locations in the s-plane.

3.6. SPECIFICATIONS OF A SECOND ORDER SYSTEM

3.6

59

Specifications of a second order system

In designing control systems, specifications must be developed that describes


the characteristics the system should posses. Usually system design specifications or standard performance measures can be described in terms of the step
response of a system as shown in Figure 3.10.

Figure 3.10: Step response of a control system.

For underdamped systems, we define the following specifications:


Tr : rise time (0%-100%)
Tr1 : rise time (10%-90%)
Mp : peak value
Tp : time to first peak
yss : steady state value
Mp yss
100%
% overshoot :
yss
Ts : settling time
The speed of the response is measured by the rise time, Tr , and the peak
time, Tp . For underdamped systems with an overshoot, the 0 100% rise time
is a useful index. If the system is overdamped, then the peak time is not defined,
and the 10 90% rise time, Tr1 , is normally used.
The tracking properties, i.e., the similarity with which the actual response
matches the step input is measured by the percent overshoot and settling
time, Ts . The % overshoot is defined as
% overshoot =

Mp yss
100%
yss

(3.8)

60

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS

Figure 3.11: Step response of a control system.

where Mp is the peak value of the time response, and yss is the steady state
value or the final value of the response.
The settling time, Ts , is defined as the time required for the system to settle within a certain percentage, , of the input amplitude. This band of is
shown in Figure 3.10. In other words it is the time required for the transient
to decay to a small value so that y(t) is almost in the steady state. For second
order systems we usually determine the time, Ts , for which the response remains
within 2% of the final value. From Figure 3.11 we can determine an approximate
to Ts by computing the time when the decaying exponential en t reaches 2%:
en Ts = 0.02
or
n Ts 4
Therefore, we have
4
n
The steady-state error of the system may be measured on the step response of
the system as shown in Figure 3.10. To obtain analytic expressions for Tp we
first differentiate (3.6) to obtain
p
n
en t sin 1 2 n t
y(t)
=p
1 2
Ts 4 =

and equating y(t)


= 0 gives
p
1 2 n t = n,
from which we get
t=

n = 1, 2, 3,

n
p
n 1 2

3.6. SPECIFICATIONS OF A SECOND ORDER SYSTEM

61

Tp occurs when n = 1, therefore


Tp =

p
=
2

d
n 1

(3.9)

To determine Mp we substitute (3.9) back in (3.6)




y(t)

t=Tp




n /n 1 2

p
n 1 2

sin d
+
e
=1 p
1 2
2
1
=1 p
e/ 1 sin( + )
1 2
2
p
1
e/ 1 sin()
=1+ p
(noting that sin = 1 2 )
2
1
2
= 1 + e/ 1 = Mp

Therefore, since yss = 1, the percent overshoot is, from (3.8)


2
% overshoot = 100e/ 1
The percent overshoot is thus a function only of and is plotted versus in
Figure 3.12. We can express (3.9) as
n Tp = p

1 2

Thus the product n Tp is also a function of only , and this is also plotted in
Figure 3.12. Since Tp is an approximate indication of the rise time, Figure 3.12
also roughly indicates rise time. As increases from 0 to 1, Mp , Ts , and the
percent overshoot decreases, while Tp and Tr increase.

Figure 3.12: Percent overshoot and normalized peak time versus damping ratio .

62

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS

Comment
The transient response of the system may be described in terms of two factors:
1. The speed of response, measured by the rise time, Tr , and the peak
time, Tp .
2. The tracking properties, i.e., the closeness of the response to the desired
response, is measured by the % overshoot and the settling time, Ts .
It is important to realize that the two factors are contradictory requirements;
thus, a compromise must be obtained.
Example 3.3

Consider the following transfer function


G(s) =

4
s2 + 2s + 4

Determine, Ts , Tp and the % overshoot.


 Solution We have
n2 = 4 = n = 2 rad/s, 2n = 2 = = 0.5
p

d = n 1 2 = 3 rad/s

= = 1.82s.
d
3
4
The settling time is found to be Ts =
= 4s.
n
2
The maximum % overshoot = 100e/ 1 = 16.3%

The peak time is obtained as Tp =

3.7

Specifications vs. pole locations

Figure 3.13 illustrates the relationships between the location of the characteristic equation roots and , n , and d . For the complex conjugate roots
shown,

Figure 3.13: Relationship between poles of second-order system and , n , and d .

3.7. SPECIFICATIONS VS. POLE LOCATIONS

63

n is the radial distance from the roots to the origin of the s-plane.
n is the real part of the roots.
d is the imaginary part of the roots.
is the cosine of the angle between the radial line to the roots and the
negative axis when the roots are in the left-halp s-plane, or = cos .
The effect of increasing n and on pole locations in the s-plane is shown in
Figure 3.14.

Figure 3.14: Pole locations tin the s-plane as n and increases respectively.

Settling time and pole locations


The settling time Ts is related to the roots in Figure 3.14 by the relation
Ts =

4
4
=
n

The settling time is then inversely related to the real parts of the poles. If in

Faster response

Figure 3.15: Pole locations to achieve desired settling time.

design the settling is specified to be less than or equal to some value Ts,desired ,

64

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS

n 4/Ts,desired , and the pole locations are then restricted to the region of
the s-plane indicated in Figure 3.15. Hence the speed of response is increased
by moving the poles to the left in the s-plane.
% Overshoot and pole locations
The angle in Figure 3.13 satisfies the relationship
p
1 2
1
= tan
= cos1

(3.10)

The percent overshoot is given by


2
% overshoot = e/ 1 100
Hence this equation can be expressed as
% overshoot = e/ tan

(3.11)

Decreasing the angle reduces the percent overshoot. Hence, specifying the
percent overshoot to be less than a particular value restricts the pole locations
to the region of the s-planre, as shown in Figure 3.16

Figure 3.16: Pole locations and their relation to % overshoot.

Peak time and pole locations


From (3.9) the peak time Tp is related to the roots in Figure 3.14 by the relation
Tp =

The peak time is inversely proportional to the imaginary part of the pole.

3.7. SPECIFICATIONS VS. POLE LOCATIONS

65

Lines of constant Tp , % overshoot, and Ts are shown in Figure 3.17. Note that
Ts2 < Ts1 ; Tp2 < Tp1 ; and %OS1 < %OS2 .

Figure 3.17: Lines of constant Tp , Ts , and % overshoot.

Suppose that, in the design of a second-order system, the %overshoot in a step


response is limited to 4.32%. A maximum settling time of 2s is also required.
On an s-plane show the region to which the pole locations are limited to.

Example 3.4

 Solution To find , from (3.11)


4.32 = e/ tan 100
= 0.0432 = e/ tan

= 3.14 =
after taking natural log of both sides
tan
tan = 1
Thus, = 45o , hence = cos 45o = 0.707. A settling time of 2s implies n 2.
Hence the pole locations are limited to the regions of the s-plane shown in Figure
3.18. The pole locations that exactly satisfy the limits of the specifications are
s = 2 j2.


Figure 3.18: Pole plot for Example 3.4.

Consider the pole plot shown in Figure 3.19. Find , n , Tp , and Ts .

Example 3.5

66

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS

Figure 3.19: Pole plot for Example 3.5.

 Solution The damping ratio is given by




= cos = cos tan1 (7/3) = 0.394
The natural frequency n is given by n =

72 + 32 = 7.616 rad/s

= = 0.449s
d
7
4
4
The settling time Ts is Ts =
= = 1.33s
n
3
The peak time Tp is Tp =

3.7.1

Step response vs. pole location

In Figure 3.20 the step responses are shown as the poles are moved in vertical
direction, keeping the real part the same. We see that the frequency changes, but
the envelope remains the same. Since all curves fit under the same exponential
decay curve, the settling time is virtually the same for all waveforms.

Figure 3.20: Step response as poles move with constant real part.

In Figure 3.21 the step responses are shown as the poles are moved in horizontal

3.8. STEADY-STATE ACCURACY

67

direction, keeping the imaginary part the same. As the poles move to the left,
the response damps out rapidly, while the frequency remains the same. Notice
that the peak time is the same for all waveforms.

Figure 3.21: Step response as poles move with constant imaginary part.

In Figure 3.22 the poles are moved along a constant radial line. We see that the
% overshoot remains the same. The farther the poles are from origin, the more
rapid the response.

Figure 3.22: Step response as poles move with constant damping ratio.

3.8

Steady-state accuracy

It has been observed, in the previous sections, that much information about
a system can be obtained from the analysis of its response to test inputs. In
control system design, one of the major objectives is to be able to track reference
inputs precisely, and to maintain this precision in the face of disturbances. In
many cases an error between the desired and resulted final value does occur. In
this section, we consider how such precision can be achieved.

3.8.1

Error signal

Consider the feedback system in Figure 3.23. Assume the feedback loop is stable. We have seen in Section 1.3 how a feedback control system can reduce

68

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS

Figure 3.23: Closed loop control system.

sensitivity of the system. Furthermore, the effect of disturbances can also be


reduced significantly. However, as a further requirement, we must examine and
compare the final steady-state error of the closed loop system.
The output Y (s) is required to track the reference input R(s). The input into
plant G(s) is the tracking error
E(s) = R(s) Y (s)
=

R(s)
1 + G(s)

Let e(t) denote its inverse Laplace transform. The limit


ess = lim e(t)
t

is referred to as the steady-state error. The value of ess characterises the


final value of error as a difference between the value of the input r(t) and the
final value of the output y(t). In other words, ess is the value of error after
transients have died out.
To calculate the steady-state error, we utilize the final value theorem, as long
as E(s) does not have any poles in the right half of the s-plane, except maybe,
at s = 0, then
ess = lim e(t) = lim sE(s) = lim
t

s0

s0

sR(s)
1 + G(s)

Step Input The steady-state error for a step input is


ess = lim

s0

sR(s)
1
=
1 + G(s)
1 + G(0)

Note that in this case, the steady-state error is determined by the dc gain of
G(s). The larger is dc gain, the smaller is the steady-state error. Furthermore,
if G(s) has one or more poles at s = 0, then lims0 G(s) = . In this case,
ess = 0. A pole at s = 0 implies G(s) includes an integrator. The number of
poles at the origin of the loop gain (i.e. the number of integrators)
defines the systems type number, N.
Type zero systems (N = 0) do not include an integrator and therefore have
a finite dc gain G(0). The constant G(0) is denoted by Kp , the position error
constant, and is given by
Kp = lim G(s)
s0

3.8. STEADY-STATE ACCURACY

69

The steady-state error for a step input is thus given by


ess =

1
1 + Kp

Note that a large position error constant corresponds to a small steady-state


error. Furthermore, if N 1 the steady-state error ess = 0.

1
1

Figure 3.24: Steady-state error for step input.

Ramp Input Now consider the steady -state error in response to a ramp input
r(t) = t. The Laplace transform of the ramp input is R(s) = 1/s2 and
s(1/s2 )
1
1
= lim
= lim
s0 1 + G(s)
s0 s + sG(s)
s0 sG(s)

ess = lim

Again, the steady-state error depends upon the number of integrators, N . For
a type zero system, N = 0, the steady-state error is infinite. For a type one
system, N = 1, the error is
1
ess =
Kv
where Kv the velocity error constant and is computed as
Kv = lim sG(s)
s0

For N 2, Kv is infinite and the steady-state error for a ramp input is zero.
Parabolic Input for a parabolic input r(t) = t2 /2, we take R(s) = 1/s3 ,
the steady-state error is
s(1/s3 )
1
= lim 2
s0 1 + G(s)
s0 s G(s)

ess = lim

The steady-sate error is infinite for type zero and type one systems. For type
two, N = 2, and we obtain
1
ess =
Ka

70

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS


1

Figure 3.25: Steady-state error for ramp input.

where Ka is called the acceleration error constant. The acceleration error


constant is
Ka = lim s2 G(s)
s0

The error constants and the steady-state error for the three inputs are summarized in Table 3.1.
Table 3.1: Summary of steady-state errors.

Type

0
1
2

Example 3.6

R(s) =

ess =

1
s

R(s) =

1
1 + Kp

ess = 0
ess = 0

1
s2

R(s) =

1
s3

Error constants

Kp = lim G(s)

1
Kv

Kv = lim sG(s)

1
Ka

Ka = lim s2 G(s)

s0

s0

s0

Consider the control system in Figure 3.23 with


G(s) =

200(s + 1)2
(s + 2)(s + 3)(s + 4)

Find the steady-state error when (a) r(t) is a unit step, (b) r(t) is a unit ramp.
 Solution G(s) is a type zero system, therefore for a step input, the position error constant, Kp = G(0) = 200/24 = 8.333. Then
ess =

1
1
=
= 0.1071
1 + Kp
9.333

3.8. STEADY-STATE ACCURACY

71

For a ramp input Kv = lim sG(s) = 0. Then,


s0

ess =

1
=
Kv

72

CHAPTER 3. TIME-DOMAIN ANALYSIS OF CONTROL SYSTEMS

Chapter

Stability Analysis
Stability is the most crucial issue in designing any control system. One of the
most common control problems is the design of a closed loop system such that
its output follows its input as closely as possible. If the system is unstable such
behavior is not guaranteed. Unstable systems exhibit an unbounded output,
i.e., a response blowing up to infinity as time increases. This usually cause the
system to suffer serious damage such as burn out, break down or it may even
explode. Therefore, for such reasons our primary goal is to guarantee stability.
As soon as stability is achieved one seeks to satisfy other design requirements,
such as speed of response, settling time, steady state error, etc.

4.1

Introduction

To help make the later mathematical treatment of stability more intuitive let
us begin with a general discussion of stability concepts and equilibrium points.
Consider the ball which is free to roll on the surface shown in Figure 4.1. The
ball could be made to rest at points A, E, F , and G and anywhere between
points B and D, such as at C. Each of these points is an equilibrium point of
the system.
A small perturbation away from points A or F will cause the ball to diverge
from these points. This behavior justifies labeling points A and F as unstable

Figure 4.1: Equilibrium points

73

74

CHAPTER 4. STABILITY ANALYSIS

equilibrium points. After small perturbations away from E and G, the ball will
eventually return to rest at these points. Thus E and G are labeled as stable
equilibrium points. If the ball is displaced slightly from point C, it will normally
stay at the new position. Points like C are sometimes said to be neutrally stable.
So far we assumed small perturbations, if the ball was displaced sufficiently
far from point G, it would not return to that point. We say the system is stable
locally. Stability therefore depends on the size of the original perturbation and
on the nature of any disturbances.
Stability deals with the following questions. If at time t0 the system is perturbed from its equilibrium point, does the system return to that point, or
remain close to it, or diverge from it?
As we shall see in this chapter, stability of a feedback system is directly related to the location of the roots of the characteristic equation of the system
transfer function.

4.2

Bounded-input bounded-output stability

A continuous-time system is stable if and only if every bounded input produces a bounded output. Consider a bounded input x(t) such that |x(t)| < B
for all t. Suppose that this input is applied to an LTI system with impulse
response h(t). Then
Z




|y(t)| =
h( )x(t )d
Z 0

|h( )| |x(t )| d
0
Z
B
|h( )| d
0

Therefore, because B is finite, y(t) is bounded, hence, the system is stable if


Z
|h( )| d <
(4.1)
0

Example 4.1

For an LTI system with impulse response h(t) = e3t u(t), determine the stability of this causal LTI system.
 Solution Using (4.1), hence

Z
1 3t
1
3t
e dt = e = <
3
3
0
0
and this system is stable.

4.2.1

Relationship between characteristic equation roots


and stability

To show the relation between the roots of the characteristic equation and stability, let G(s) be a transfer function representation of an LTI system. Then,

4.2. BOUNDED-INPUT BOUNDED-OUTPUT STABILITY

75

G(s) can be written as


G(s) =

bm sm + bm1 sm1 + + b1 s + b0
an sn + an1 sn1 + + a1 s + a0

where the denominator polynomial is called the characteristic polynomial of


G(s). The equation
an sn + an1 sn1 + + a1 s + a0 = 0

(4.2)

is called the characteristic equation of G(s). The roots of the characteristic


equation are called the poles of G(s). The order of the system G(s) is defined
to be the degree of the characteristic polynomial.
Assume that the roots pi of the characteristic equation are real or complex,
but are distinct. The solution to the differential equation whose characteristic
equation is given by (4.2) may be written using partial-fraction expansion as
y(t) =

n
X

Ki epi t

(4.3)

i=1

where pi are the roots of (4.2). The system is stable if and only if every term
in (4.3) goes to zero as t :
e pi t 0

for all pi

This will happen if all the poles of the system are strictly in the LHP, where
Re{pi } < 0
If any poles are repeated, the response must be changed from that of (4.3)
by including a polunomial in t in place of Ki , but the conclusion is the same.

Figure 4.2: Time functions associated with pole locations in the s-plane.

76

CHAPTER 4. STABILITY ANALYSIS

Therefore, the stability of a system can be determined by computing the location


of the roots of the characteristic equation and determining whether they are all
in the LHP. If the system has any poles in the RHP, it is unstable. Hence the
j axis is the stability boundary between BIBO stable and unstable responses.
A system is said to be marginally stable if all its poles lie in the LHP, and at
least one pole on the j axis. If the system has repeated poles on the j axis,
then it is unstable. For example, repeated poles at s = 0
 
1
L1 2 = t
s
results in an unbounded response. The following table illustrates the stability
conditions of an LTI system with reference to the locations of the roots of the
characteristic equation. Figure 4.2 summarizes all the results.
Table 4.1: Stability conditions of a LTI system.
Stability conditions

Locations of the roots

Stable

All the roots are in the LHP

Marginally stable

At least one root and no multiple


roots on the j-axis; and no roots
in the RHP.

Unstable

At least one root in the RHP or at


least one multiple roots on the jaxis.

The following examples illustrates the stability conditions of systems with reference to the poles of the transfer function G(s).
Table 4.2: Stability examples.
Transfer function G(s)

Stability condition

G(s) =

20
(s + 1)(s + 2)(s + 3)

Stable

G(s) =

20(s + 1)
(s 1)(s2 + 2s + 2)

Unstable due to the pole at s = 1

G(s) =

20(s 1)
(s + 2)(s2 + 4)

Marginally stable due to s = j2

G(s) =

10
(s2 + 4)2 (s + 10)

Unstable due to the multiple poles


at s = j2

As shown above, one way of determining stability is to calculate the roots of


the characteristic equation. The disadvantage of this is that system parameters must be assigned numerical values, which makes it difficult to find the

4.3. ROUTH-HURWITZ STABILITY CRITERION

77

range of values of a parameter that ensures stability. For example in the springmass-damper system the characteristic equation is M s2 + Bs + K, one can not
determine the ranges of M, B and K to ensure stability.
An alternative to locating the roots of the characteristic equation is given by
Routh-Hurwitz stability criterion, which is presented next.

4.3
4.3.1

Routh-Hurwitz Stability Criterion


General properties of polynomials

Consider the second order polynomial, assuming all coefficients are real
P (s) = s2 + a1 s + a0 = (s p1 )(s p2 )
2

= s (p1 + p2 )s + p1 p2

(4.4)
(4.5)

By comparing (4.4) with (4.5) implies


a1 = (p1 + p2 )

and a0 = p1 p2

If p1 and p2 are stable, we have a1 > 0 and a0 > 0. Consider next the third
order polynomial
P (s) = s3 + a2 s2 + a1 s + a0 = (s p1 )(s p2 )(s p3 )
= s3 (p1 + p2 + p3 )s2 + (p1 p2 + p2 p3 + p1 p3 )s p1 p2 p3
and by comparing the two equations we have
a2 = (p1 + p2 + p3 )
a1 = (p1 p2 + p2 p3 + p1 p3 )
a0 = p1 p2 p3
Again if all roots are stable, all the coefficients will have the same sign. However, this condition is not sufficient, for it is quite possible that an
equation with all its coefficients nonzero and of the same sign still
will not have all the roots in the left half of the s-plane. Consider for
example the polynomial s3 + s2 + 2s + 8, clearly all the coefficients have the
same sign however not all roots are in the LHP.
Consider a general nth order polynomial
P (s) = sn + an1 sn1 + + a1 s + a0
= (s p1 )(s p2 ) (s pn )

78

CHAPTER 4. STABILITY ANALYSIS

By analogy with the previous examples, we have


an1 =

n
X

pi

i=1

an2 =

product of roots taken 2 at a time


X
=
product of roots taken 3 at a time

an3
..
.

a0 = (1)n

n
Y

pi

i=1

We conclude
If all roots are stable, all the polynomial coefficients will be positive.
If any coefficient is negative, at least one root is unstable.
If any coefficient is zero, not all roots are stable.

4.3.2

Routh-Hurwitz stability criterion

The Routh-Hurwitz criterion represents an analytical procedure of determining


if all roots of a polynomial with constant real coefficients lie in the left half of
the s-plane, without actually solving for the roots.
Consider the nth order polynomial
P (s) = an sn + an1 sn1 + + a1 s + a0
in which we can always assume a0 6= 0. If a+0 = 0, we can write:
P (s) = s (an sn1 + an1 sn2 + + a1 )
{z
}
|
P (s)

and work with P (s) instead. Note that in the case a0 = 0, P (s) will have at
least one root at the origin and we conclude that the LTI system is marginally
stable or unstable.
The Routh array
The first step in the Routh-Hurwitz criterion is to arrange the coefficients of the
characteristic polynomial into an array as follows
sn
n1

an

an2

an4

an6

an1

an3

an5

an7

Further rows are then completed as

4.3. ROUTH-HURWITZ STABILITY CRITERION


sn

79

an

an2

an4

an6

n1

an1

an3

an5

an7

n2

b1

b2

b3

b4

n3

..
.

c1
..
.

c2
..
.

c3
..
.

c4
..
.

..
.

s2

k1

k2

l1

m1


an

an1

s
s

s
s
where
b1 =

1
an1


an

an1


an2
an3

b2 =



1 an1 an3
c1 =
b2
b1 b1
..
.


1 k1 k2
m1 =
l1 l1 0

an1


1 an1
c2 =
b1 b1


an4
an5


an5
b3

Once the Rouths array has been completed, we investigate the signs of the
coefficients in the first column of the array. The roots of the equation are all in
the left half of the s-plane if all the elements of the first column of the Rouths
array are of the same sign. The number of unstable roots is equal to the number
of sign changes in the first column of the array.
Consider the third order polynomial:

Example 4.2

P (s) = s3 s2 + s + 6
This equation has one negative coefficient. Thus, we know without applying
Rouths test that not all the roots of the equation are in the LHP.

Example 4.3

Consider the third order polynomial:


P (s) = s3 + s2 + 2s + 8
The Routh array is:
s3

-6

The two sign changes (from +1 to -6 and from -6 to +8) indicates two unstable
roots.


80

CHAPTER 4. STABILITY ANALYSIS

Special cases when the Rouths array terminates prematurely


Not all arrays can be completed as the one shown in Example 4.3. Depending on
the coefficients of the equation, the following difficulties may occur that prevent
the Rouths array from completing properly:
1. The first element of a row is zero, with at least one nonzero element in the
same row, the procedure is modified by replacing that first element with a
small number  such that ||  1 and proceeding as before.
2. Every entry in a row is zero, the last modification will not give useful
information and another modification is needed.
Example 4.4

To demonstrate the first case consider the polynomial:


P (s) = s5 + 2s4 + 2s3 + 4s2 + 11s + 10
The Routh array is then
s5

11

10

s
s

Since the first element of the s3 row is zero, the elements in the s2 row would
all be infinite. To overcome this difficulty, we replace the zero in the s3 row by
a small positive number  and then proceed with the array
s5

11

10

0

s
s

The first element of the s2 row is calculated as follows




1
12
12
1 2 4
= (12 4) = 4
w

  6



with a similar procedure we calculate the remaining rows
s5

11

s4

10

s3

0

s2

12/

10

10

There are two sign changes (irrespective of the sign of ) indicating two unstable
roots.

Special Case: Zero rows. If all the coefficients in a row are zero, a pair
of roots of equal magnitude and opposite sign is indicated. These could be two
real roots with equal magnitudes and opposite signs or two conjugate imaginary

4.3. ROUTH-HURWITZ STABILITY CRITERION

81

roots. The zero row is replaced by taking the coefficients of dPa (s)/ds, where
Pa (s), called the auxiliary polynomial, is obtained from the values in the row
above the zero row. Why? A zero row implies that a polynomial Pa (s) has only
even or odd powers. It turns out in this case, Pa (s) and Pa (s) + dPa (s)/ds have
exactly the same number of RHP poles (proof beyond the scope of theis course).
As the goal is just to find the number of RHP poles, we can use dPa (s)/ds as a
surrogate to continue the procedure. The pair of roots can be found by solving
dPa (s)/ds = 0. The roots of Pa (s) are also the roots of the the ploynomial P (s).
Example 4.5

As an example of case 2 consider the the polynomial:


P (s) = s3 + s2 + 2s + 2
The Routh array is then
s3

auxiliary polynomial Pa (s) = s2 + 2

The auxiliary polynomial Pa (s) indicates that P (s) = 0 must have two pairs of
roots of equal magnitude and opposite sign, which are also roots of the auxiliary
polynomial equation Pa (s) = 0. Taking the derivative of Pa (s) with respect to
s we obtain
dPa (s)
= 2s
ds
so the s row is as shown below and the Routh array is
s3

Coefficients of dPa (s)/ds

The absence of a sign change indicates no unstable roots, so all roots are on the
imaginary axis. We conclude the system is marginally stable.

The following example combines case 1 and case 2 problems: polynomial:
P (s) = s4 + 4
The Routh array is then
s4

4 Pa (s) = s4 + 4

s3

04

0 Coefficients of

s2

0

16/

The two sign changes indicate two unstable roots.

dPa (s)
= 4s3
ds

In Summary, the three cases that occur in the application of the RouthHurwitx criterion are as follows:

Example 4.6

82

CHAPTER 4. STABILITY ANALYSIS


Case 1. No elements in the first column are zero. There are no problems
in completing the array.
Case 2. There is at least one nonzero element in a row, with the first
element equal to zero. This always indicates an unstable system. The
first element (which is zero) is replaced with the value ,   1, and the
calculation of the array continues.
Case 3. All elements in a row are zero. This always indicates a system
that is not stable, but it may be marginally stable. This case can be
analyzed through the use of the auxiliary equation, as described earlier. If
the system is marginally stable the roots on the j axis are also the roots
of the auxiliary equation.

4.4

Applications in feedback design

Consider the following feedback system involving a plant G(s) and a compensator (or controller) K(s)

K(s)

G(s)

Figure 4.3: Feedback control system.

The closed-loop transfer function H(s) is


H(s) =

G(s)K(s)
1 + G(s)K(s)

The closed-loop poles are given by the roots of the charactersistic equation
1 + G(s)K(s) = 0
Suppose we write
G(s) =

Ng (s)
Dg (s)

and

K(s) =

Nk (s)
Dk (s)

where Ng (s), Dg (s), Nk (s), and Dk (s) are all polynomials. Then, the closed
loop transfer function is given
Ng (s)Nk (s)
Dg (s)Dk (s)
H(s) =
Ng (s)Nk (s)
1+
Dg (s)Dk (s)
=

Ng (s)Nk (s)
Ng (s)Nk (s) + Dg (s)Dk (s)

4.4. APPLICATIONS IN FEEDBACK DESIGN

83

The poles of the closed-loop system are also given by the roots of the characteristic polynomial
Ng (s)Nk (s) + Dg (s)Dk (s)
The poles of the open-loop system G(s) are the roots of its characteristic polynomial
Dg (s) = 0
which are generally different from the closed-loop poles.
A fundamental design objective in control is the stabilization of unstable systems. The Routh-Hurwitz stability criterion can be used as an aid in feedback
design.
Example 4.7

Let G(s) the be given by


G(s) =

s3

1
+ 2s 8

5s2

Let K(s) = K be a constant controller. Find the range of values of K for which
the closed-loop is stable.
 Solution Since the coefficients of the characteristic polynomial do not have
the same sign, we conclude G(s) is unstable. The closed-loop poles are the roots
of the characteristic equation
1 + KG(s) = 1 +

s3

K
=0
+ 2s 8

5s2

which implies
s3 + 5s2 + 2s + (K 8) = 0
The Routh array is
s3

K 8

0.2(18 K)

K 8

For stability we need


8 < K < 18

Let G(s) the be given by

Example 4.8
G(s) =

s3

s2

1
10s 8

Find the range of values of K for which the closed-loop is stable.


 Solution The characteristic equation is given by
1 + KG(s) = 1 +

s3

s2

K
=0
10s 8

84

CHAPTER 4. STABILITY ANALYSIS

which implies
s3 s2 10s + (K 8) = 0
It follows that no choice of K can ensure all coefficients have the same sign. We
conclude that G(s) cannot be stabilized by a constant controller and dynamic
compensation is need.


Chapter

Root-Locus Analysis and Design


In this chapter we introduce one of the major analysis and design methods
discussed in this course. The method is the root-locus procedure; it indicates
to us the characteristics of a control systems transient response. We have seen
that the response of an LTI system is largely determined by the location of its
poles.

5.1

Root-locus principles

We introduce the root locus through an example. Consider the feedback


control system shown in Figure 5.1

1
s(s + 2)

Figure 5.1: Feedback control system.

The closed-loop transfer function is given by


K
K
s(s + 2)
H(s) =
= 2
K
s + 2s + K
1+
s(s + 2)
Hence the characteristic equation, which is the denominator of the closed-loop
transfer function set to zero, is
s2 + 2s + k = 0
We see, since the polynomial is second order, that the system is stable for all
positive values of K. It is not evident for this example exactly how the value of
85

86

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN


Table 5.1: Characteristic equation roots for different values of K.
K

Characteristic equation

Roots

s2 + 2s = 0

s = 0, 2

s2 + 2s + 1 = 0

s = 1 j0

s2 + 2s + 2 = 0

s = 1 j

Note these are the


open loop poles

K affects the transient response. Table 5.2 show the roots of the characteristic
equation for different values of K. To investigate some of the effects of choosing
different values of K, we plot the roots of the system characteristic equation in
the s-plane. These roots are plotted in Figure 5.2 for 0 K .

Figure 5.2: Plot of characteristic equation roots.

We can see from the plot that for 0 < K < 1, the roots are real with different
time constants. For K = 1, the roots are real and equal, and the system is
critically damped. For K > 1, the roots are complex with a time constant of
1s, with the value of decreasing as K increases. Hence, as K increases with
the roots being complex, the overshoot in the transient response increases.
The plot in Figure 5.2 is called the root locus of the system in Figure 5.1.
The root-locus of a system is a plot of the roots of the system characteristic equation (closed-loop poles) as K varies from 0 to .
For an nth order system, the root-locus is a family of n curves traced out by the
n closed-loop poles as K is varied from zero to infinity. Plotting the root locus
for negative values of K will be considered later.

5.1. ROOT-LOCUS PRINCIPLES

5.1.1

87

Root-locus criterion

We generally consider the system of Figure 5.3 in discussing the root locus, with
0 K < .

G(s)

Figure 5.3: Feedback control system.

The characteristic equation for this system is given by


1 + KG(s) = 0

(5.1)

A point s1 lies on the root-locus if and only if s1 satisifies (5.1) for a real value
of K, with 0 K < . Equation (5.1) can be written as
1
G(s)
1
G(s) =
K
K=

KG(s) = 1

(5.2)
(5.3)
(5.4)

Equations (5.1) to (5.4) are all equivalent.


The magnitude citerion
Since K is real and positive, (5.2) is equivalent to
K=

1
|G(s)|

(5.5)

We call (5.5) the magnitude criterion of the root locus and can be used to
find K corresponding to a point on the root-locus.
Assume the point s1 = 2 lies on the root-locus find K if
G(s) =

s+4
(s + 1)(s + 3)

 Solution Evaluate |G(s)|s=2 , we have


|G(2)| =

| 2 + 4|
=2
| 2 + 1|| 2 + 3|

Hence,
1
= 0.5
|G(2)|
The value K = 0.5 can be interpreted as the gain needed, in the feedback control
system of Figure 5.3, that places the locus at the point s = 2.

K=

Example 5.1

88

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

The angle criterion


In general G(s) is complex and can be expressed in polar form as magnitude
and phase as |G(s)| G(s). If G(s) is to satisfy (5.4) we note the following
KG(s) = 180
but K is real and positive which means K = 0, hence,
G(s) = 180

(5.6)

and in general
G(s) = r(180 )

r = 1, 3, 5,

Equation (5.6) called the angle criterion may be interpreted as follows: For
a point s1 to be on the root-locus, the sum of all angles for vectors between
open-loop poles and zeros to point s1 must be equal 180 . The angle criterion is
illustrated in Figure 5.4 for the function
G(s) =

s z1
(s p1 )(s p2 )

In Figure 5.4 the poles of G(s) are marked and the zero is marked . Suppose
that the point s1 is to be tested to determine if it is is on the root-locus. For
this point to be on the locus, we must have G(s1 ) = 180 or equivalently
(s1 z1 ) (s1 p1 ) (s1 p2 ) = r(180 )

r = 1, 3, 5,

The angle from the zero term sz1 can be computed by drawing a line from the
location of the zero at z1 to the test point s1 . In this case the line has a phase
angle marked 1 on Figure 5.4. In a similar fashion, the vector from the pole
s = p1 to the test point s1 is shown with angle 2 , and the angle of the vector
from the pole s = p2 to s1 is shown with angle 3 . Thus the angle condition
(5.6) becomes
1 2 3 = r(180 )
for the point s1 to be on the root-locus.

Figure 5.4: Illustration of the angle criterion.

5.1. ROOT-LOCUS PRINCIPLES

89

In general the condition for a point in the s-plane to be on the root-locus is that
X
X
(angles form zi )
(angles from pi ) = r(180 )
r = 1, 3, 5,
i

Check whether the point s0 = 1 + 2j lies on the root-locus for some value of
K if
s+1
G(s) =
s[(s + 2)2 + 4](s + 5)
 Solution For s0 to be on the locus, we must have G(s0 ) = 180 . Therefore,
G(s0 ) = (s0 + 1) s0 [(s0 + 2)2 + 4] (s0 + 5)
= (s0 + 1) s0 (s0 + 2 2j) (s0 + 2 + 2j) (s0 + 5)
= (2j) (1 + 2j) (1) (1 + 4j) (4 + 2j)
= 90 116.6 0 76 26.6
= 129.2
Alternatively, we could have marked the poles and zeros of G(s) on the s-plane
as shown in Figure 5.5. The angles from the poles and zeros could be computed
by drawing a line from each pole and zero to the test point s0 as shown in Figure
5.5. The point s0 is on the root locus if
X
X
(angles form zi )
(angles from pi ) = 180
i

and by inspecting Figure 5.5 yields


G(s0 ) = 1 1 2 3 4
= 90 116.6 0 76 26.6
= 129.2
Since the phase of G(s0 ) is not 180 , we conclude that s0 is not on the root-locus,
so we must select another point and try again.


Figure 5.5: Measuring the phases of Example 5.2.

Example 5.2

90

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

Example 5.2 demonstrates measuring phase is easy, however measuring phase


at every point in the s-plane is not practical. Therefore, we need some general
rules for determining where the root locus is.

5.2

Rules & steps for plotting the root-locus

The following transfer function is used for illustrating the steps for plotting
the root-locus
1
G(s) =
s[(s + 4)2 + 16]

RULE 1. The root-locus is symmetric with respect to the real axis.


This follows from the assumption the G(s) is a ratio of two polynomials with
real coefficients. So, the characteristic polynomial roots are either real or occur
in complex conjugate pair.
Draw the axes of the s-plane to a suitable scale and enter

STEP 1. for each pole of G(s) and a for each zero. See Figure 5.6.

Figure 5.6: Step 1: Mark the poles and zeros.

RULE 2. The root-locus includes all points on the real axis to the
left of an odd number of poles and zeros.
This follows from the angle criterion, we consider first that all poles and zeros
of the open-loop transfer function are on the real axis, and we test points on
the real axis to determine if these points are on the locus.
Consider an open-loop transfer function G(s) of two poles and one zero as
illustrated in Figure 5.7. If we take a test point s on the real axis to the right of
the zero z1 as shown in Figure 5.7(a) we find that G(s) = 0. Hence, the angle
criterion is not satisfied, and we can see that any point to the right of the zero
z1 cannot be on the root locus.
Consider now a point s between the zero z1 and the pole p1 . In this case

5.2. RULES & STEPS FOR PLOTTING THE ROOT-LOCUS

91

z1 s p 1 s p 2
(a) G(s)= |s {z
} | {z } | {z }
0

Angle criterion is not satisfied!

z1 s p1 s p2
(b) G(s)= |s {z
} | {z } | {z }
180

Angle criterion is satisfied!

z1 s p1 s p2
(c) G(s)= |s {z
} | {z } | {z }
180

180

Angle criterion is not satisfied!

z1 s p1 s p2
(d) G(s)= |s {z
} | {z } | {z }
180

180

180

Angle criterion is satisfied!


Figure 5.7: Real axis locus.

G(s) = 180 , as shown in Figure 5.7(b). However the angles from the poles
p1 and p2 are still 0 . Thus the angle requirement is satisfied, and any point
between z1 and p1 is on the locus.
For a point s between p1 and p2 , (see Figure 5.7(c)) the angle from z1 is still
180 , as now is the angle from p1 . The angle from p2 is still 0 ; hence the angle
requirement is not satisfied and no points between p1 and p2 are on the locus.
If the point s is to the left of the pole p2 , the angles from z1 , p1 , and p2 are all
180 , and the angle criterion is satisfied as shown in Figure 5.7(d).
For the case that we have complex poles or zeros, the preceding discussion
still applies. For example, two complex conjugate poles are shown in Figure 5.8.
Since complex poles (and zeros) must occur in conjugate pairs, the sum of the
angles from a pair of poles (or zeros) to a point on the real axis will always be
0 (or 360 ). Hence complex poles and zeros do not affect the part of the root
locus that lies on the real axis.
Summary:
The angle contribution from a pole or zero to the left of s is 0 .
The angle contribution from a pole or zero to the right of s is 180 .
The angle contribution from a pair of complex conjugate poles or zeros
cancels out.

92

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

Figure 5.8: Real axis locus.

STEP 2. Find the real axis portion of the locus.

Figure 5.9: The real axis parts of the locus are to the left of an odd number of poles
and zeros.

RULE 3. The root-locus originate on the open-loop poles for K = 0


and terminate at the open-loop zeros when they exist, otherwise it
terminates at infinity.
Suppose that
G(s) =

n(s)
(s z1 )(s z2 ) (s zm )
=
d(s)
(s p1 )(s p2 ) (s pn )
=

where

n>m

sm + b1 sm1 + + bm
sn + aa sn1 + + an

We assume that n(s) and d(s) are monic polynomials (monic means the coefficient of the highest power of s is 1). The closed-loop characteristic equation

5.2. RULES & STEPS FOR PLOTTING THE ROOT-LOCUS

93

can be written as
1+K

1
n(s)
n(s)
= 0 d(s) + Kn(s) = 0
+
=0
d(s)
K
d(s)

(5.7)

If K = 0, then from (5.7) implies d(s) = 0 (i.e., the poles of G(s)). Therefore,
for K = 0 the roots of the closed-loop characteristic equation 1 + KG(s) = 0 are
the open-loop poles. The points of the root-locus where K = 0 are sometimes
called the starting or departure points of the root-locus.
As k approaches infinity but s remains finite (the case s will be discussed in Rule 4), (5.7) implies n(s)/d(s) = 0 which in turn implies n(s) = 0.
Therefore, as K approaches infinity the roots of 1 + KG(s) are the open-loop
zeros. The points of the root-locus where K = are sometimes called the ending or arrival points of the root-locus.
Note from the above discussion that we have n poles and m zeros. If m of
the n poles will terminate at m zeros, where will the n m poles terminate. As
Rule 3 states they will terminate at infinity, the question remains which infinity,
Rule 4 next clarifies the matter.

RULE 4. If G(s) has zeros at infinity, the root-locus will approach


asymptotes as K . The angles of asymptotes are
(2l + 1)

l = 0, 1, 2,

The asymptotes intersect the real axis at


P
P
poles zeros
A =
number of poles number of zeros
A =

Recall from the discussion of Rule 3 for K , G(s) = 0 if n(s) is zero for
a finite s. The root locus will approach the open-loop zeros. To see a second
manner in which G(s) may go to zero, we express the characteristic equation
1 + KG(s) = 0 as
1+K

sm + b1 sm1 + + bm
=0
sn + aa sn1 + + an

(5.8)

Since n > m, it is clear that G(s) goes to zero as s . In fact, for very large
values of s (5.8) can be approximated by
1+K

1
=0
(s A )nm

(5.9)

To see why (5.9) is a good approximation to (5.8), try to imagine what would
we see if we could observe the locations of poles and zeros from a distance point
near infinity: They would appear to cluster near the s-plane origin as shown in
Figure 5.10(a). Thus m zeros would cancel the effects of m poles, and the other
n m poles would appear to be in the same place, namely at s = A as shown
in Figure 5.10(b). If = n m we may write (5.9) as
1+K

1
=0
(s A )

94

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

n m Poles
(a)

A
(b)

Figure 5.10: Determination of angles of asymptotes.

We say that the locus of (5.8) is asymptotic to the locus of (5.9) for large values
of K and s.
To find the locus, no matter how far away the point s is on the s-plane it
must satisfy the angle criterion. Since all poles appear to be in the same
place the angle condition gives
A = r(180 )
= A = r

r = 1, 3, 5,

(180 )

The angles A are the angles of asymptotes of the root-locus. Table 5.2 gives
these angles for small values of .
Table 5.2: Angles of asymptotes.

Angles

No asymptotes

180

90

60 , 180

45 , 135

36 , 108 , 180

5.2. RULES & STEPS FOR PLOTTING THE ROOT-LOCUS

95

Finding A
To determine A we make use of polynomial properties discussed in Section
4.3.1. Write G(s) as
G(s) =

(s z1 )(s z2 ) (s zm )
(s p1 )(s p2 ) (s pn )
m
X

!
sm1 +

zi

i=1

=
sn

n
X

!
sn1 +

pi

i=1

Dividing both the numerator and the denominator by the numerator gives
G(s) =
snm

n
X

1
!
m
X
pi
zi snm1 +

i=1

(5.10)

i=1

For very large values of s, G(s) was approximated by


1
(s A )nm

(5.11)

The polynomial (s A )nm in (5.11) can be written as


snm + an1 snm1 + an2 snm2 +
where
an1 =

nm
X

pi = (n m)A

i=1

Hence (5.11) may be written as


1
snm (n m)A snm1 +

(5.12)

Comparing (5.12) and (5.10) to order snm1 yields


(n m)A =

n
X

pi

i=1

m
X

zi

i=1

Hence,
X

pi

zi
A =
nm
P
P
Notice that in the sum
pi and
zi the imaginary parts always add to zero
since complex poles and zeros always occur in complex conjugate pairs.
In summary the loci proceed to the zeros at infinity along asymptotes centered
at A and with angles A . when the number of m finite zeros is less than the n
number of poles, then n m sections of loci must end at zeros at infinity. these
sections of loci proceed to the zeros at infinity along asymptotes as k approaches
infinity. These linear asymptotes are centered at the point A on the real axis.

96

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

STEP 3. Draw the asymptotes for large values of K.


For our example,
A =

4 4 + 0
8
= = 2.67
30
3

and = 3 = asymptotes at 60 and an asymptote at 180 . The asymptotes


at 60 are shown dashed in Figure 5.11. Notice that they cross the imaginary
axis at 4.62j. The asymptote at 180 was already found in Step 2.

Figure 5.11: Draw the asymptotes.

RULE 5. The root-locus departs from a complex pole pj at an angle


d given by
d =

zi

pi r(180 )

r = 1, 3, 5,

i6=j

The root-locus arrives at a complex zero zj at an angle a given by


a =

X
i

pi

zi r(180 )

r = 1, 3, 5,

i6=j

The most important use of this rule is to compute the angle of departure from a
complex pole. This angle of departure can sometimes be an aid in determining
the final shape of the root locus. To illustrate this rule consider the poles and
the zero shown in Figure 5.12. The vector angles at one complex pole p1 is also
shown in Figure 5.12. The radius of the circle around the pole p1 is actually very
small in relation to the distance to the zero and the other pole. The angles at a
test point s0 , an infinitesimal distance from p1 , must meet the angle criterion.
Therefore,
1 2 = 180

5.2. RULES & STEPS FOR PLOTTING THE ROOT-LOCUS

97

s0
p1

Figure 5.12: Determining the angle of departure.

and the angle of departure at pole p1 is


1 = 180 2
Note that we select either +180 or 180 above to ensure that 1 is always
selected to be in the range 180 < 1 < +180 .

STEP 4. Compute the departure and arrival angles.


First we take a test point s0 very near pole 2 at 4 + 4j and compute the
angle of G(s0 ). This situation is sketched in Figure 5.13. We select the test

Figure 5.13: Compute the angle of departure.

point close enough to pole 2 that the angles 1 and 3 to the test point can be

98

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

considered the same as those angles to pole 2. Thus 1 = 90 and 3 = 135 ,


and 2 can be calculated from the angle condition
90 2 135 = 180
To ensure 180 < 2 < +180 we have
2 = 90 + 135 180
2 = 45
as shown in Figure 5.14. By the complex conjugate symmetry of the plots, the
angle of departure of the locus near pole 1 at 4 4j will be +45 . Note for
a multiple pole of order q we must count the angle from the pole q times.

Figure 5.14: Actual angle of departure.

STEP 5. Estimate (or compute) the points where the locus crosses
the imaginary axis.

The points where the root-locus intersect the imaginary axis of the s-plane, and
the corresponding values of K, may be determined by means of the RouthHurwitz criterion. For the third-order example we are using, the characteristic
equation is
1+

K
=0
s[(s + 4)2 + 16]

which is equivalent to
s3 + 8s2 + 32s + K = 0
The Routh array for this polynomial is

5.2. RULES & STEPS FOR PLOTTING THE ROOT-LOCUS


s3

32

s1

256 K
8

99

In this case we see that the s1 row coefficients are all zeros when K = 256
indicating a root on the imaginary axis. Thus K = 256 must correspond to
a solution at s = j0 for some 0 . Substituting this data into the auxiliary
equation gives us
802 + 256 = 0

Clearly the solution is 0 = 32 = 5.66, which is plotted in Figure 5.15.

Figure 5.15: Find the imaginary axis crossing points.

RULE 5. The breakaway points b are among the real roots of


dG(s)
=0
ds
or, equivalently,
n(s)

d
d
[d(s)] d(s) [n(s)] = 0
ds
ds

where n(s) and d(s) are the numerator and denominator polynomials, respectively, of G(s). The breakaway points are points at which two (or more) branches
of the root locus leave the real axis. The example in Figure 5.2 provides an illustration of a breakaway point. In this case, there is a breakaway point at s = 1.
Figure 5.16 shows both the root-locus and a plot of K as a function of real
values of s between 0 and 2. The maximum occurs at s = 1 for K = 1. At
the point where K = 1 the characteristic equation has a double root at s = 1.
This is the maximum gain for which the poles are real; higher gains result in

100

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

Breakaway at s = 1

Maximum gain
at s = 1

Figure 5.16: Gain K as a function of s along the real axis.

complex roots. Notice in Figure 5.16 that the gain K, as a function of the real
roots s, must have a local maximum at the breakaway points, so that, with
K=

1
G(s)

and s considered a real variable, we require


dK
=0
ds

(5.13)

If we express G(s) as a ratio of two polynomials n(s) and d(s) the above equation
can be written as




dK
d
1
d
d(s)
=

=0
ds
ds
G(s)
ds
n(s)
The differentiation with respect to s, yields




d
d(s)
1 d
1 d

= d(s)(1) 2
[n(s)] +
[d(s)]
ds
n(s)
n (s) ds
n(s) ds
Equating the right hand side of the equation above to zero implies
n(s)

d
d
[d(s)] d(s) [n(s)] = 0
ds
ds

It is important to point out that the condition for a breakaway point given in
(5.13) is necessary but not sufficient. In other words, all breakaway points on

5.2. RULES & STEPS FOR PLOTTING THE ROOT-LOCUS

101

the root-locus must satisfy (5.13), but not all solutions of (5.13) are breakaway
points.

STEP 6. Determine the breakaway point b .


In our example, G(s) is
G(s) =

1
s[(s + 4)2 + 16]

and we have
n(s) = 1

d n(s)
=0
ds

d(s) = s3 + 8s2 + 32s

d d(s)
= 3s2 + 16s + 32
ds

the points of possible multiple roots or breakaway are given by


3s2 + 16s + 32 = 0
or
s0 = 2.67 1.89j
The breakaway point must be real and lies on the root-locus, hence, for this
example there is no breakaway point.

STEP 7. Complete the sketch.


The complete locus for our third-order example is drawn in Figure 5.17. It
combines all the results found so far, that is, the real axis segment, the number
of asymptotes and their angles, the angles of departure from the poles, and the
imaginary axis crossing points.

Figure 5.17: Complete the sketch.

102

5.3

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

Examples

In all the examples to follow, G(s) is the transfer function of a system to be


controlled using constant-gain in the forward path, with K 0.
Example 5.3

Find the root locus for


G(s) =

s+1
s2 (s + 9)

 Solution
STEPS 1 and 2. Mark the poles and zeros on the s-palne and draw the real axis
portion of the locus:

STEP 3. Draw the n m = 3 1 = 2 asymptotes:


P
P
(180 )
p i z1
A = r
A =
nm
nm
=

(180 )
2

= 90

9 0 (1)
31

= 4

STEP 4. We compute the departure angles from the poles. We draw a small
circle around the two poles at s = 0. The angles from the zero at 1 and from
the pole at 10 are both zero, and the angles from the two poles at the origin
are the same. Therefore, the root locus condition is
2d = r180 = 90

5.3. EXAMPLES

103

STEP 5. We compute the points where the locus crosses the imaginary axis:
1+K

s+1
=0
s2 (s + 9)

s3 + 9s2 + Ks + K = 0
The Routh array for this polynomial is
s3

s1

9K K
9

The entries in the first column are all positive if K > 0, so the equation has no
roots in the RHP for positive values of K.
STEP 6. We locate the points of multiple roots, which will include breakaway
and break-in points:
n(s) = s + 1

d n(s)
=1
ds

d(s) = s3 + 9s2

d d(s)
= 3s2 + 18s
ds

The possible multiple roots are at


(s + 1)(3s2 + 18s) (s3 + 9s2 )(1) = 0
2s3 + 12s2 + 18s = 0
s = 0, 3, 3
The points of multiple roots are on the locus, but we have repeated roots in the
derivative, which indicates that we have three roots at the same place. Note we
can apply the rule of departure angles to the triple root at s = 3, we find that
180 3d = r180
d = 0 , 120

104

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

Figure 5.18: Root locus for G(s) =

(s + 1)
.
s2 (s + 9)

STEP 7. The complete sketch is given in Figure 5.18.


Example 5.4

Find the root locus for


G(s) =

1
s(s + 2)[(s + 1)2 + 4]

 Solution
STEPS 1 and 2. Mark the poles and zeros on the s-palne and draw the real axis
portion of the locus:

STEP 3. Draw the asymptotes:


A = r

(180 )
40

= 45 , 135

A =

2 1 1 0 + 0
40

= 1

5.3. EXAMPLES

105

STEP 4. The departure angle at the complex pole at 1 + 2j is


0 116.6 63.4 90 d = 180
d = 90
We can observe at once that, along the line s = 1 + j the angle criterion is
always satisfied. This is a special case since the angles of the real axis poles to
any point on this line will form an isosceles triangle and always add to 180 .
STEP 5. We compute the crossings of the imaginary axis. The characteristic equation is
s4 + 4s3 + 9s2 + 10s + K = 0
The Routh array for this polynomial is
s4

10

s2

6.5
65 4K
6.5

s1
1

In this case we see that the s1 row coefficients are all zeros when K = 16.25
indicating a root on the imaginary axis. Thus K = 16.25 must correspond to
a solution at s = j0 for some 0 . Substituting this data into the auxiliary
equation gives us
6.502 + 16.25 = 0

Clearly the solution is 0 = 2.5 = 1.58.

106

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

STEP 6. We locate possible multiple roots:


n(s) = 1

d n(s)
=0
ds

d(s) = s4 + 4s3 + 9s2 + 10s

d d(s)
= 4s3 + 12s2 + 18s + 10
ds

The possible multiple roots are at


4s3 + 12s2 + 18s + 10 = 0
From step 4 we notice that the line at s = 1 + j is on the locus, so there
must be a breakaway point at s = 1, which can be divided out. That is , we
can easily show that
4s3 + 12s2 + 18s + 10 = (s + 1)(4s2 + 8s + 10) = 0
The quadratic has roots 1 1.22j. Since these points are on the line between
the complex poles, they are points of multiple roots on the locus.
STEP 7. The complete sketch is given in Figure 5.19. Notice that we have
complex multiple roots. Branches of the locus come together at 1 1.22j and
break away at 0 and 180 .

Sketching root loci relies heavily on experience. Figure 5.20 gives several loci
for low-order systems; these should be studied to familiarize yourself with some
of the characteristics of root loci.

Figure 5.19: Root locus for G(s) =

1
.
s(s + 2)[(s + 1)2 + 4]

5.3. EXAMPLES

107

Figure 5.20: Loci for low-order systems.

108

5.4

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

The complementary root locus

We now consider modifying the root-locus procedure to permit analysis of


negative values of K. Following are the rules for plotting a complementary
root-locus:
1. The root-locus is symmetric with respect to the real axis.
2. The root-locus includes all points on the real-axis to the left of an even
number of poles and zeros.
3. The root-locus originates on the poles (for K = 0) and terminate at the
zeros (as K ) of G(s).
4. If G(s) has zeros at infinity, the root-locus will approach asymptotes
as K approach . The angles of asymptotes are
A =

2l

l = 0, 1, 2,

and the asymptotes intersect the real axis at


X
X
pi
zi
A =

5. The angle of departure from a pole pj is given by


X
X
d =
zi
pi + 2l
i

i6=j

and arrives at a zero zj at an angle


X
X
a =
pi
zi + 2l
i

i6=j

6. The breakaway points b are roots of


dG(s)
=0
ds

Example 5.5

Consider
G(s) =

s
(s 0.5 2j)(s 0.5 + 2j)

 Solution The root locus and the complementary root locus are shown in Figure 5.21. In the complementary root-locus, the locus on the real axis occurs to
the left of an even count of poles and zeros. Since zero is considered even, root
locus on the real axis will occur only to the right of the zero at the origin. The
break-in points are s = 2.06. Thus the break-in point for the complementary
root locus is at s = 2.06. After the break-in, one closed-loop pole migrates to
the zero at the origin and the other to the right toward infinity.

5.4. THE COMPLEMENTARY ROOT LOCUS

Figure 5.21: Root locus for G(s) =

Consider
G(s) =

109

s
, < K < .
(s 0.5 2j)(s 0.5 + 2j)

s+1
s(s + 4)(s + 10)

 Solution The complementary root locus on the real axis occurs to the right
of the pole at the origin, between the zero at s = 1 and the pole at s = 4,
and to the left of the pole at s = 10. The number of zeros at infinity, = 2, so
there are two asymptotes at 0 and 180 . The root locus and the complementary
root locus are shown in Figure 5.22.

Figure 5.22: Root locus for G(s) =

s+1
, < K < .
s(s + 4)(s + 10)

Example 5.6

110

5.5

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

Root locus design

The root-locus is a plot of all possible locations for roots to the equation
1 + KG(s) = 0 for some real positive value of K. The purpose of design is to
select a particular value of K that will meet the design specifications. Consider
for example the locus of
G(s) =

1
s[(s + 4)2 + 16]

For this transfer function, the locus was plotted in Figure 5.17 and is repeated
here in Figure 5.23. On Figure 5.23 the lines corresponding to a damping ratio
of = 0.5 are sketched, and the points where the locus crosses these lines are
marked with (). Suppose we wish to set the gain so that the poles are located
at the dots. This corresponds to selecting the gain so that two of the closed-loop
poles have a damping ratio of = 0.5. What is the value of K when a root is

Figure 5.23: Root locus for G(s) =

1
showing calculations of gain K.
s[(s + 4)2 + 16]

5.6. DYNAMIC COMPENSATION

111

at the dot? The value of K is given by


K=

1
|G(s0 )|

where s0 is the coordinate of the dot. On the figure we have plotted three vectors
marked s0 s1 , s0 s1 , and s0 s3 , which are the vectors from the poles of
G(s) to the point s0 . (Since s1 = 0, the first vector equals s0 .) Therefore,
K=

1
= |s0 ||s0 s2 ||s0 s3 |
|G(s0 )|

By measuring the lengths of these vectors and multiplying the lengths together,
provided that the scale of the imaginary and real axes is identical, we can
compute the gain to place the roots at the dot (s = s0 ). Using the scale of the
figure we estimate that
|s0 | 4
|s0 s2 | 2.1
|s0 s3 | 7.7
Thus the gain is estimated to be
K = 4(2.1)(7.7) 65
We conclude that if K is set to the value 65, then a root of 1 + KG(s) will be
at s0 , which has a damping ratio of 0.5. Another root is at the conjugate of s0 .
Where is the third root? The third root lies on the branch of the locus along the
negative real axis. Usually we take a test point, compute a trial gain, and repeat
this process until we found the point where K = 65. However, in this case we
can make use of polynomial properties that the open-loop and closed-loop sum
is the same if m < n 1. If G(s) has at least two more poles than zeros, we
have
X
X
open-loop poles =
closed-loop poles
(5.14)
From Figure 5.23 we estimate that s0 = 2 + 3.5j. Since the starting point
was at s = 4 + 4j, the root has moved approximately two units to the right.
The conjugate has moved an equal distance. The third root must be moved far
enough to the left to keep the sum in (5.14) fixed, so the third root must have
moved 2 + 2 units to the left of where it began at s = 0. We have marked the
new location at 4 with the third dot. Considering this point as a test point
one can check if the gain at this point is K = 65.
If the closed-loop dynamic response as determined by the root locations is satisfactory, then the design can be completed by gain selection alone. However if no
value of K satisfies all the constraints, as is typically the case, then additional
modifications are necessary to meet the system specifications.

5.6

Dynamic compensation

If the plant dynamics are of such a nature that a satisfactory design cannot
be obtained by a gain adjustment alone, then some modification or compensation of the plant dynamics are needed. A variety of compensation techniques are

112

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

available, only two such techniques that have been found simple and effective
will be discussed here. These are lead and lag compensation. Lead compensation acts mainly to speed up a response by lowering rise time and decreasing
the transient overshoot. Lag compensation is usually used to improve steadystate accuracy of the system.
Compensation with a transfer function of the form
D(s) = K

s+z
s+p

is called lead compensation if |z| < |p| and lag compensation if |z| > |p|. Compensation is typically placed in series with the plant in the feedforward path, as
shown in Figure 5.24.

Figure 5.24: Feedback system with compensation.

The characteristic equation of the system in Figure 5.24 is


1 + D(s)G(s) = 0
1 + KL(s) = 0
where K and L(s) are selected to put the equation in root-locus form as before.

5.6.1

Lead compensation

To see the basic effect of lead compensation on a system, we first consider a simplified proportional control for which D(s) = K. If we apply this compensation
to a second order system with transfer function
G(s) =

1
s(s + 1)

The root locus with respect to K is shown as the solid-line portion of the locus
in Figure 5.25. Also shown in Figure 5.25 is the locus produced by proportional
plus derivative control, where D(s) = K(s + 2). The modified locus is the circle
sketched with dashed lines. Notice that the effect of the zero is to move the
locus to the left, toward the more stable part of the s-plane.
The trouble with choosing D(s) based on only a zero is that the physical realization would contain a differentiator that would greatly amplify the high
frequency noise present from the sensor signal. To remedy this we simply add
a high frequency pole, perhaps at s = 20 to give
D(s) = K

s+2
s + 20

5.6. DYNAMIC COMPENSATION

113

1
without compensation (solid line), and
s(s + 1)
with compensation D(s) = s + 2 (dashed lines).

Figure 5.25: Root locus for G(s) =

The resulting transfer function is thus lead compensation. The root locus with
such compensator is shown in Figure 5.26.

Figure 5.26: Root locus for G(s) =

1
s+2
with lead compensation D(s) =
.
s(s + 1)
s + 20

To see the effect of the pole on the compensation consider moving the pole further to the right at s = 10, i.e, nearer to the zero. The root locus is shown
in Figure 5.27. Notice the effect of moving the pole nearer to the zero, we are
reducing the effect of the zero we placed earlier. In fact, we are moving back

114

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

Figure 5.27: Root locus for G(s) =

s+2
1
with lead compensation D(s) =
.
s(s + 1)
s + 10

to the uncompensated shape. If we move the pole too far to the left, the magnification of noise at the output of D(s) is too great, since the differentiator
will dominate the compensator. Therefore, the choice of pole location is a compromise between the conflicting effects of noise suppression and compensation
effectivness.
Example 5.7

Find a compensation for


1
s(s + 1)
that will provide two closed-loop dominant poles having damping ratio = 0.707
and settling time of 2s. Furthermore, determine the value of the gain K to
achieve this.
G(s) =

 Solution The uncompensated root locus is shown in Figure 5.25. We need


Ts = 2, which implies that n = 2. Hence, the first requirement is satisfied
if we force
the root locus to pass through the point 2 + j2 corresponding to
n = 2 2 and = 0.707. Notice that we need to move the root locus to the
left. This is achieved with a lead compensator of the form
D(s) = K

s+z
s+p

Selecting values of z and p is done by trial and error. In general the compensator zero is placed in the neighborhood of the real part of closed-loop pole and
the compensator pole is placed at a distance 3 to 20 times the value of the zero
location.
The compensator pole position can now be determined by the angle criterion as
shown in Figure 5.28
(1 + 2 + ) = 180
90 (135 + 116 ) 180 =

5.6. DYNAMIC COMPENSATION

115

s0 = 2 + j2

Figure 5.28: Angles for lead compensation.

Hence, = 19 , and to determine the pole location p, we have


tan(19 ) =

2
|p 2|

which implies p = 7.8. Finally, the value of the gain K can be determined
using the magnitude criterion
K=

1
|G(s0 )|

|s0 ||s0 + 1||s0 + p|


|s0 + z|

8 5 37.64
=
2
19.4
=

The final design is then


D(s) = 19.4

s+2
s + 7.8

Although the design is complete and two of the closed-loop poles are already
known, namely, the poles at s = 2 j2. However the lead compensator introduces a third closed-loop pole. In this case the easiest way to locate this third
pole is to make use of (5.14) since m < n 1. Thus, the third closed-loop pole
is at s = 4.8.
Design a lead compensator for the system given by the transfer function
G(s) =

1
s(s + 1)

that will provide a closed-loop damping ratio = 0.5 and natural frequency
n > 7 rad/sec.

Example 5.8

116

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

 SolutionpSince = 0.5 and n = 7, the closed loop poles are given by


n jn 1 2 = 3.5j6.062. Let us choose z = 3.5. The compensator
pole position can now be determined by the angle criterion as shown in Figure
5.29
90 112.41 120 p = 180
Hence the angle subtended by the compensator pole with the closed loop pole
is 37.59 .

3.5

Figure 5.29: Angles for lead compensation.

To determine the pole location, we have


tan(37.59 ) =

2
x

which implies p = 11.37. Finally we use the magnitude criterion to calculate


the gain K 75.


5.6.2

Lag compensation

In this section we consider the design of lag compensators. As in the preceding


sections, we assume that the compensator transfer function is first order and is
given by
(s + z)
D(s) = K
|z| > |p|
(s + p)
It was shown that the effect of the addition of lead compensation is to shift
the root locus to the left in the s-plane. However, lag compensators will tend
to shift the root locus to the right in the s-plane, that is, toward the unstable
region. Thus, in general, the shift to the right must be minimal to minimize the
destabilizing effects. This small shift is assured by placing the pole and the zero
of the compensator very close to each other. In fact we choose to place the compensator pole near the origin to approximate a perfect integrator. This should
increase the system type by 1 that might be needed to improve steady-state
error constants. The compensator zero is placed nearby so that the pole-zero
pair does not significantly interfere with the dynamic response of the overall
system. Thus, we want an expression for D(s) that will yield a significant gain

5.6. DYNAMIC COMPENSATION

117

at s = 0 to raise an error constant and that is nearly unity (no effect) at the
higher frequencies.
We now illustrate lag design with an example.
Consider a system whose open loop transfer function is given by
G(s) =

K
s(s + 2)

Design a lag compensator so that the dominant poles of the closed loop system
are located at s = 1 j and the steady state error to a unit ramp input is less
than 0.2.
 Solution For the specification that the steady state error of the system
must not exceed 0.2, we have
Kv = lim sKD(s)G(s)
s0

= lim sK
s0

1
(s + z)
(s + p) s(s + 2)

Kz
2p

2p
We require the steady state error to be less than 0.2, i.e.,
< 0.2. Let us
Kz
choose p = 0.01, therefore we have Kz = 0.1. We know that the closed loop
poles s = 1 j lie on the root locus, hence

s(s + 2)(s + 0.01)
K=

(s + z)
s=1+j
Solving for K and z, we get K = 1.88 and z = 0.0532. Therefore, the lag
compensator is given by
D(s) = 1.88

(s + 0.532)
(s + 0.01)

Example 5.9

118

CHAPTER 5. ROOT-LOCUS ANALYSIS AND DESIGN

Chapter

Frequency Response Analysis


Frequency response methods are among the most useful techniques available for
control system analysis and design. There is no one systematic design procedure for all control problems, rather, the different techniques complement each
other. Root-locus techniques give powerful indicators for closed-loop transient
response. Unfortunately, we need accurate, hence expensive, plant models to
benefit from the root locus techniques. One of the advantages of the frequency
response methods is that the response of the system can be obtained from measurements on the physical system without deriving the system transfer function.
In fact, it is possible to design a control system without the need for a transfer
function model.
In this chapter we consider frequency-response analysis methods, the important tools of Bode and Nyquist plots are presented.

6.1

Frequency response

In Chapter 3, the time responses of first and second order systems were
considered. In this section we give meaning to steady-state response of systems
to sinusoidal inputs, which is called the frequency response. Suppose that the
input to a system with transfer function G(s) is the sinusoid
r(t) = A cos t
Then
R(s) =

As
s2 + 2

and
Y (s) = G(s)R(s) = G(s)

As
(s j)(s + j)

We can expand this expression into partial fractions of the form


Y (s) =

k1
k2
+
+ F (s)
s j s + j
119

(6.1)

120

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

where F (s) is the collection of all terms in the partial fraction that originate
in the denominator of G(s). It is assumed the system poles are real, distinct
and are in the LHP implying that the terms in F (s) will decay to zero with
increasing time. Therefore, only the first two terms in (6.1) contribute to the
steady-state response. Using the cover-up method to find k1 and k2 we have

1
k1 = (s j)Y (s) s=j = AG(j)
2

1
k2 = (s + j)Y (s) s=j = AG(j)
2
and k2 is seen to be the complex conjugate of k1 . For any given value of ,
k1 and k2 are complex numbers and will find it convenient to express them in
polar form as
A
k1 = |G(j)|ej
2
where |G(j)| is the magnitude and = G(j). Then y(t) = L1 {Y (s)} and
its sinusoidal steady-state value (i.e. lim y(t)) is
t

yss (t) = k1 ejt + k2 ejt


A
A
= |G(j)|ej ejt + |G(j)|ej ejt
2
2
ej(t+) + ej(t+)
= A|G(j)| cos(t + )
= A|G(j)|
2
since |G(j)| = |G(j)|.
This analysis can be summarized as follows. If a sinusoid input is applied to a
system all of whose poles have a negative real part, the steady state response
is a scaled, phase-shifted version of the input. The scaling factor called the
steady-state gain is |G(j)| and the phase shift is the phase of G(j).
Example 6.1

Consider the system with the transfer function


G(s) =

5
s+2

and an input 7 cos 3t. Then



G(s) s=j3 =

5
= 1.387 56.3
2 + j3

and the steady-state output is given by


yss (t) = (1.387)(7) cos(3t 56.3 ) = 9.79 cos(3t 56.3 )

We see then that, from the complex function G(j), we can obtain the steadystate response for any sinusoidal input, provided that the system is stable. We
call G(j), 0 , the frequency response function. We usually plot G(j)
versus in some form to characterize the frequency response. We illustrate two
forms by a simple example. Suppose we have a system with transfer function
G(s) =

1
s+1

6.1. FREQUENCY RESPONSE

121

The frequency response function of this system is given


G(j) =

1
1
=
tan1 ()
1 + j
1 + 2

(6.2)

One common method of displaying this frequency response is in the form of a


polar plot. In such a plot, the magnitude and angle of the frequency response
function are plotted in the complex plane as the frequency, , is varied. For
the function of (6.2), to construct a polar plot we first evaluate the function for
values of . As an example, a table of these values is given in Table 6.1. Next
Table 6.1: Frequency response

G(j)

1.000 0

0.5

0.894 26.6

1.0

0.707 45

1.5

0.555 56.3

2.0

0.447 63.4

3.0

0.316 71.6

5.0

0.196 78.7

10

0.100 84.3

0.000 90

these values are plotted in the complex plane, as shown in Figure 6.1.

Figure 6.1: Frequency response.

122

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

Note that mathematically, the frequency response is a mapping from the s-plane
to the G(j)-plane. The upper half of the j-axis which is a straight line, is
mapped into the complex plane via the mapping G(j). A second form for
displaying the frequency response is to plot the magnitude and phase of G(j)
versus . These plots for the example above are shown in Figure 6.2.

Figure 6.2: Frequency response.

6.2

Bode diagrams

This section presents a method for plotting a frequency response that is


different from the two methods given in the first section of this chapter. This
method results in a plot of magnitude versus frequency and phase versus frequency, but the frequency scale is logarithmic. In addition, the magnitude is
also plotted on a logarithmic scale. The plot that is presented here is called a
Bode plot, or a Bode diagram.
We develop the Bode diagram by using as an example the second-order transfer
function
K(1 + 3 s)
K(1 + s/3 )
G(s) =
=
(6.3)
(1 + 1 s)(1 + 2 s)
(1 + s/1 )(1 + s/2 )
where it is assumed that both poles and the zeros are real. Note that we have
defined a constant i equal to the reciprocal of i . The reason for using the

6.2. BODE DIAGRAMS

123

symbol for frequency will become evident later. Also, we call the value i a
break frequency, for a reason to be explained later.
First we form the magnitude of G(j)
|G(j)| =

|K||1 + j/3 |
|1 + j/1 ||1 + j/2 |

(6.4)

Next we use the property of logarithms given by


 
ab
log
= log ab log cd = log a + log b log c log d
cd
Also we define the unit decibel1 (dB) as dB = 20 log a, where a is a gain. A
number-decibel conversion line is given in Figure 6.3. As a number increases by
a factor of 10, the corresponding decibel value increases by a factor of 20. This

Figure 6.3: Number-decibel conversion.

may be seen from the following


20 log(a 10n ) = 20 log a + 20n
Note that when expressed in decibels, the reciprocal of a number differs from
its value only in sign; that is,
20 log a = 20 log

1
a

1 The unit was first defined as bel, however, this unit proved to be too large, and hence a
decibel (1/10 of a bel) was selected as a more useful unit.

124

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

We plot the magnitude of the frequency response in decibels; that is we plot


20 log |G(j)|. For the transfer function of (6.4),



j
20 log |G(j)| = 20 log |K| + 20 log 1 +
3







j
1 + j
20 log 1 +

20
log
(6.5)


1
2
The effect of plotting in decibels is then to cause the individual factors in the
numerator to add to the total magnitude and the individual factors in the denominator to subtract from the total magnitude.
Consider now the general frequency dependent term in (6.5)
v"
u


 2 #
u


j
= 20 log t 1 +
20 log 1 +
i
i

(6.6)

This term is plotted versus log in Figure 6.4. Note that the value
of the term at
the frequency i (called the break frequency) is equal to 20 log 2 = 3.0103. We

Figure 6.4: First-order term.

usually approximate this value as 3dB and say that, for a general first-oder numerator term, the value of the magnitude is equal to 3dB at the break frequency.
For a first-order denominator term, the value is equal to 3dB
at its break frequency. Note that the first-order term has a value of 20 log 101 = 20.04, or
approximately 20dB, at the frequency 10i .
Accurate Bode diagrams are usually done using digital computers. However,
there are situations in which approximate sketches of a Bode diagram are adequate. We now develop the approximations for the first-order terms. Consider
the first order term of (6.6)
v"
u
 2 #
u

20 log t 1 +
i
For frequencies very small compared to the break frequency i , we have
20 log(1) = 0

 i

6.2. BODE DIAGRAMS

125

Figure 6.5: First-order approximation.

and for frequencies very large compared to i ,


 

= 20 log 20 log i
20 log
i

 i

For low frequencies the term is approximated by a straight line (the -axis). For
high frequencies and if = 10i the difference between the logarithmic gains is
20dB. This represents a line that has a slope of 20dB per decade of frequency.
Equating the above high-frequency and low-frequency expressions shows that
the two straight lines intersect at = i . The two terms are plotted in Figure
6.5(a). Comparing this figure with the exact curve of Figure 6.4, we see that the
exact curve approaches the straight lines asymptotically, as is shown in Figure
6.5(b). As an approximation in sketching, we quite often extend the straight
lines to the intersection at = i and use this straight line approximation instead of the exact curve. The frequency i is called the break frequency because
of the break in the slope at that frequency, as shown in Figure 6.5(b).
In constructing frequency responses, we consider the following types of transfer
function factors:
1. Constant gain
2. Poles and zeros at the origin
3. Real poles and zeros not at the origin
4. Complex poles and zeros
We now consider each of these factors in order. First we develop the magnitude
plots, and then we develop the phase plots.

6.2.1

Constant gain

For the case of a constant gain the magnitude is


20 log |K|
this term does not vary with frequency. The two possible cases are shown in
Figure 6.6. If |K| is greater than unity, the magnitude is positive; if |K| is less
than unity, the magnitude is negative. In either case, the magnitude plot is a
straight line with a slope of zero.

126

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

Figure 6.6: Constant-gain term.

6.2.2

Poles and zeros at the origin

For the case that the transfer function has a zero at the origin, the magnitude
of this term is given by
20 log |j| = 20 log
Hence a plot of this term is a straight line, with a slope of 20dB per decade of
frequency, that intersects the -axis at = 1. The plot of this term is shown
in Figure 6.7(a).

Figure 6.7: Zero and pole at s = 0.

For the case that the transfer function has a pole at the origin, the magnitude
of the term is given by

1
20 log = 20 log
j
and the curve is the negative of that for a zero at the origin. Thus the curve is
a straight line with a slope of 20dB/decade that intersects the log axis at
= 1. This curve is shown in Figure 6.7(b). For the case of N th-order zeros at
the origin, the magnitude is
20 log |(j)N | = 20 log N = 20N log
Thus the curve is still a straight line that intersects the -axis at = 1, but
the slope is now 20N dB per decade. For example if we have two zeros at the
origin the slope is 40dB/decade and -40dB/decade if two poles are at s = 0.

6.2. BODE DIAGRAMS

6.2.3

127

Nonzero real poles and zeros

The case of real poles and zeros was considered previously. For a term of this
type
v"
u


 2 #
u


j
= 20 log t 1 +
20 log 1 +
i
i
(
0
i

20 log 20 log i w > i


This straight line approximation is shown in Figure 6.8(a) for a zero and in
Figure 6.8(b) for a pole. Note that the terms have been normalized to have
a dc gain of unity, or 0 dB. This is convenient otherwise each term will have
a different low-frequency gain, and the Bode diagram is somewhat difficult to
plot.
Suppose that first-order term is repeated, that is, suppose that we have an
N th-order term of the form (1 + s/i )N . The magnitude term is then given by
"
20 log 1 +

2 #N/2

(
0

20N log /i

Figure 6.8: First-order terms.

 i
 i

(6.7)

128

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

The straight line approximation for this term is shown in Figure 6.9 for the case
of a numerator term. It is seen that for > i , the line has a slope of 20N .
For a given denominator term, the slope is 20N .

Figure 6.9: Bode diagram for repeated zeros.

Example 6.2

Plot the Bode diagram for the system with the transfer function
G(s) =

10(s + 1)
(s + 10)

 Solution First we convert the function to the form of (6.3)


G(s) =

(1 + s)
(1 + s/10)

The break frequency of the numerator is = 1, and the freak frequency of the
denominator is = 10. The numerator term, the denominator term, and the
total magnitude (which, from (6.5), is the sum of the two terms) are shown in
Figure 6.10.


Figure 6.10: Bode diagram for Example 6.2.

Example 6.3

As a second example, consider the transfer function


G(s) =

200(s + 1)
(s + 10)2

 Solution We rewrite the transfer function as


G(s) =

2(1 + s)
(1 + s/10)2

6.2. BODE DIAGRAMS

129

The bode diagram has three terms. The first term is the constant gain, which
adds a term of value 20 log 2 = 6dB at all frequencies. The second term is the
zero term with break frequency at = 1, and the third term is the second-order
pole at = 10. The three terms are plotted in Figure 6.11.


Figure 6.11: Bode diagram for Example 6.3.

Example 6.4

A third example, with


1000(s + 3)
G(s) =
s(s + 12)(s + 50)
 Solution The transfer function is rewritten as
5(1 + s/3)
G(s) =
s(1 + s/12)(1 + s/50)
The constant gain term is obtained from
(1000)(3)
=5
(12)(50)
The constant gain term is now 20 log 5 = 14dB, and the five terms of the Bode
diagram are as shown in Figure 6.12.

6.2.4

Phase diagrams

Before we consider the final type of terms that can appear in a Bode diagram,
we construct the phase diagrams for the three types of terms already considered.
First, for the constant gain term, the phase angle is either 0 or 180 . If the
gain term is positive, the phase angle is 0 ; if the gain term is negative, the
phase angle can be plotted as either 180 or 180 . For a zero of the transfer
function at the origin, the phase angle is 90 , since
s|s=j = j = 90
In a like manner, a pole at the origin gives a phase angle of 90 , since

1
1
1
=
=
90

s s=j
j

130

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

Figure 6.12: Bode diagram for Example 6.4.

For a real zero of the transfer function, with the zero not at the origin, the term
is given by
s


 2
s
j

1+
=
1
+
=
1
+
()

i s=j
i
i
where
() = tan

Figure 6.13 shows the phase plotted for various values of the ratio /i . The
exact curve is approximatted with the straight line shown in Figure 6.13. The
straight line approximation for the phase characteristic breaks from 0 at the
frequency 0.1i and breaks back to the constant value of 90 at 10i . Note that
the phase characteristic for a pole is the negative of that for a zero, since, for a
pole,


1
1
1

=
=p
()

1 + s/i s=j
1 + j/i
1 + (/i )2
where
() = tan

Consider again the system of Example 6.1


G(s) =

(1 + s)
(1 + s/10)

Example 6.5

6.2. BODE DIAGRAMS

131

Figure 6.13: Phase characteristics of a real zero.

 Solution For the zero i = 1, hence the straight line approximation to


the phase characteristic breaks at = 0.1(1) = 0.1 and breaks back at =
10(1) = 10. The pole has i = 10 and breaks at = 0.1(10) = 1 and breaks
at = 10(10) = 100. These characteristics, along with the total phase diagram
(which is the sum of the two characteristics), are plotted in Figure 6.14.
As a second example illustrating the phase characteristic of the Bode diagram,
consider the system of Example 6.3, with the transfer function
G(s) =

5(1 + s/3)
s(1 + s/12)(1 + s/50)

The phase characteristics of the various terms, along with the total phase characteristic of the system, are given in Figure 6.15.


Figure 6.14: Example 6.5.

Example 6.6

132

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

Figure 6.15: Bode phase diagram for Example 6.6.

Example 6.7

As a final example, the complete Bode diagram will be constructed for the
transfer function
(1 s)
G(s) =
(1 + s/10)
For the zero at s = 1,
1 j =

p
(1 + 2 ) ()

() = tan1 ()

The magnitude characteristic is as shown in Figure 6.16. The phase of the


zero term varies from 0 to 90 , because of the minus sign on the imaginary
part. The total phase characteristic is then as shown in Figure 6.16. The
characteristics at the extremes in frequency are verified through the calculations
lim G(j) = 1 0

lim G(j) =

6.2.5

j
= 10 = 10 180
j/10

Complex poles and zeros

In this section we consider an additional term that can be encountered in constructing a Bode diagram. We consider poles or zeros of the form
s2 + 2n s + n2

0<1

(6.8)

For convenience in plotting we normalize (6.8), to have a dc gain of unity; this


is accomplished by factoring out n2 . Hence, we consider
 2
s
s
+
(6.9)
1 + 2
n
n
The magnitude and phase of this expression for s = j is an involved function
of the damping ratio , and in general it does not lend itself to approximations

6.2. BODE DIAGRAMS

133

Figure 6.16: Bode diagram of Example 6.7.

by straight lines.
Consider first the case that = 1. For this case, (6.9) has two real equal
zeros

 2
2

s
s
s

= 1+
1 + 2
+
(6.10)

n
n
n
=1
Since the zeros are real, this case is covered by the methods given in the preceding sections. The straight-line approximations for this case are given in Figure
6.17, along with the exact curves. For cases in which 0 < < 1, the asymptotic
approximations to the frequency response curves are not accurate and the errors
can be large for low values of . This is because the magnitude and phase of
(6.9) depend on both the break frequency and the damping ratio . Noting that
the exact magnitude of (6.9) in dB is
s

 2
2 
2


j
j
2

= 20 log
+
2
20 log 1 + 2
+
1


n
n
n2
n

134

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

Figure 6.17: Bode diagrams of a repeated zero.

The approximation to the frequency response may be obtained as follows: for


low frequencies such that  n , the magnitude becomes
20 log 1 = 0 dB
The low-frequency asymptote is thus a horizontal line at 0 dB. For  n , the
magnitude becomes
2

20 log 2 = 40 log
n
n
the equation for the high frequency asymptote is a straight line having the slope
40 dB/decade. The high-frequency asymptote intersects the low-frequency one
at = n . The two asymptotes just derived are independent of the value of .
The phase for second order zeros is

(
2

0
 n
n
1

tan
2

180  n
1 2
n

6.2. BODE DIAGRAMS

135

Figure 6.18 illustrates some exact curves for several values of between zero and
unity for complex zeros. Once again, the curves for complex poles are obtained
by inverting these curves.

Figure 6.18: Bode diagrams of complex zeros.

For the case that < 0.3, the straight line approximations are very inaccurate
and are seldom used. Instead exact curves such as in Figure 6.18 are used. An
example is now given to illustrate complex terms ina Bode diagram.
Consider the transfer function
200(s + 1)
2(s + 1)
G(s) = 2
=
2
s + 4s + 100
(s/10) + 2(0.2)(s/10) + 1

Example 6.8

For the complex poles, = 0.2 and n = 10. We do not expect the straight-line
approximation to be very accurate. Both the straight line approximation and

136

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

the exact Bode diagram are given in Figure 6.19. The maximum error in the
magnitude diagram for the straight line approximation is seen to be approximately 8 dB. Note also the very large errors in the straight line approximation
for the phase.

6.3

Nyquist plots

Frequency response information can be presented in a various forms of which


Bode plot is one. The same information can be presented by the Nyquist plot also
known as the polar plot. The Nyquist plot of the frequency response function
G(j) is a plot of the magnitude of G(j) versus the phase angle of G(j) on
polar coordinates as is varied from zero to infinity. An example of such a plot
is shown in Figure 6.20. The projection of G(j) on the real and imaginary
axes are its real and imaginary components. An advantage in using a Nyquist
plot is that it depicts the frequency response characteristics of a system over

Figure 6.19: Bode diagram for Example 6.8.

6.3. NYQUIST PLOTS

137

Figure 6.20: Nyquist plot.

the entire frequency range in a single plot. Table 6.2 shows examples of Nyquist
plots of simple transfer functions.

6.3.1

Nyquist criterion

In designing a control system, we require that the system is stable. In what


follows we shall show that the Nyquist plot indicates not only whether a system
is stable but also the degree of stability of a stable system. In this section we
consider closed loop systems of the type shown in Figure 6.21. The closed loop

Figure 6.21: Closed loop system.

transfer function is given by


H(s) =

G(s)
1 + G(s)

and the characteristic equation is given by


1 + G(s) = 0

138

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

We can write the closed loop transfer function as


G(s)
n(s)/d(s)
=
1 + G(s)
1 + n(s)/d(s)
n(s)
=
d(s) + n(s)

H(s) =

Table 6.2: Nyquist plots of simple transfer functions.

6.3. NYQUIST PLOTS

139

Furthermore, we can write the characteristic equation as


F (s) = 1 + G(s) = 1 +
= F (s) =

n(s)
=0
d(s)

d(s) + n(s)
=0
d(s)

That is
1 The poles of F (s) are the open loop poles (i.e., poles of G(s)).
2 The zeros of F (s) are the closed loop poles (i.e., poles of H(s)).
Therefore, to determine closed loop stability, we need to know the number of
right-half plane zeros of F (s).
In order to introduce the Nyquist criterion, we consider some mappings from
the complex s-plane to the F (s)-plane. For example consider the function
F (s) =

s 0.5
s(s 1)(s + 4)

Suppose F is evaluated around the simple, circular closed contour of radius


2 in the s-plane as shown in Figure 6.22(a). Evaluating F at each point on
generates the closed loop contour shown in Figure 6.22(b). Table 6.3 provides
the values of F at some key points along the contour . The closed contour
could then be approximated by simply plotting and connecting these points. It
is woth noting thet as shown in Figure 6.22 is not drawn to scale.
Note that the contour in the s-plane, where F was evaluated, was traversed
in the counterclockwise direction, and enclosed the circular region in the splane. Further, the contour generated by the evaluation of F along evolves
in the clockwise direction. Also note the contour encircles the origin of the
F -plane.

(a)

(b)

Figure 6.22: (a) Curve in the s-plane and (b) resulting curve in the F -plane.

140

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS


Table 6.3: Magnitude and phase of F (s) along .

F (2 )

0.125 0

20

0.13 20

40

0.16 48

60

0.17 81

80

0.18 180

100

0.2 138

120

0.19 162

140

0.19 175

160

0.2 179.7

180

0.21 180

Based on the above observations, one can ask is there a relationship between
the number of poles and zeros encircled by in the s-plane and the number
and direction of encirclements of the origin in the F -plane. In the above mapping, the counterclockwise encirclement of two poles and one zero resulted in
one clockwise encirclement of the origin.
The relationship between the contours in the two complex planes is given by
Cauchys theorem (known as Cauchys principle of argument) twhich states
(given here without proof) for a given contour in the s-plane that encircles
P poles and Z zeros of the function F (s) in a clockwise direction, the resulting
contour in the F-plane encircles the origin a total of N times in a clockwise
direction, where N = Z P . This theorem explains the mapping in Figure
6.22, since Z = 1 and P = 2, hence, N = 1. Therefore, the contour encircles
the origin once and the negative sign implies opposite direction to the contour .
We now develop the Nyquist criterion. Suppose that we let the mapping of
F (s) be the characteristic polynomial of the closed-loop system of Figure 6.21;
that is;
F (s) = 1 + G(s)
Furthermore, let the curve be as shown in Figure 6.23(a). This curve, which
is composed of the imaginary axis and an arc of in finite radius, completely
encircles the right half of the s-plane. Then, in Cauchys principle of argument,
Z is the number of zeros of the system characteristic polynomial in the right
half of the s-plane. Also recall that Z is the number of poles of the closed-loop

6.3. NYQUIST PLOTS

141

Figure 6.23: Nyquist diagram.

system in the RHP. Therefore, Z must be zero for the system to be stable. P
is the number of poles of the characteristic polynomial in the right half of the
s-plane and thus is the number of poles of the open loop function G(s) in the
right half plane, since the poles of 1 + G(s) are also those of G(s).
The curve in Figure 6.23(a) is called the Nyquist path, and a typical mapping
is shown in Figure 6.23(b). The mapping encircles the origin two times in the
clockwise direction, and from Cauchys principle
N =2=Z P
or
Z =2+P
Since P is the number of poles of a function inside the Nyquist path, it cannot
be a negative number. This in this example, Z is greater than or equal to 2,
and the closed loop system is unstable.

142

CHAPTER 6. FREQUENCY RESPONSE ANALYSIS

To simplify the application of the Nyquist criterion a modification is usually


made. Instead of plotting 1 + G(s), as in Figure 6.23(b), we plot just G(s).
The resulting plot has the same shape but is shifted one unit to the left, as
shown in Figure 6.23(c). Hence, rather than plotting 1 + G(s) and counting encirclements of the origin, we get the same result by plotting G(s) and counting
encirclements of the point 1 + j0. The resultant plot of the open-loop function
G(s) is called the Nyquist diagram. Note that we are plotting the open-loop
function to determine closed-loop stability.
A simple example is given to illustrate the Nyquist criterion.
Example 6.9

Consider the system with open-loop transfer function


G(s) =
Then
G(j) =

5
(s + 1)3
5
(1 + j)3

An evaluation of this function is given in Table 6.4 for certain values of , and a
plot of theses values is shown in Figure 6.24 The dc gain, G(0), is equal to 5 and
is shown as part I. The solid curve, part II, is obtained directly by plotting the
values of Table 6.4. However, note as is increased the magnitude of each factor
in the denominator has an increasing magnitude. Therefore |G(j)| decreases
from 5 to 0.

Table 6.4: Frequency response

G(j)

5.00 0

0.5

3.58 79.8

1.0

1.77 135

1.5

0.85 169

2.0

0.45 190.3

5.0

0.038 236.1

20

0.0006 261.3

6.3. NYQUIST PLOTS

Figure 6.24: Nyquist diagram for Example 6.9.

143