INDEX

S. No Topic Page No.

Week 1
1 Introduction to Control Systems - Part 1 1
2 Introduction to Control Systems - Part 2 11
3 Overview of Feedback Control Systems - Part 1 20
4 Overview of Feedback Control Systems- Part 2 27
5 Mathematical Preliminaries - Part 1 36
6 Mathematical Preliminaries- Part 2 45
Week 2
7 Transfer Function - Part 1 57
8 Transfer Function- Part 2 62
9 System Response - Part 1 69
10 System Response- Part 2 74
11 BIBO Stability - Part 1 84
12 BIBO Stability- Part 2 90
Week 3
13 Effect of Zeros - Part 1 101
14 Effect of Zeros- Part 2 106
15 Closed Loop System - Part 1 114
16 Closed Loop System- Part 2 118
17 First Order Systems - Part 1 126
18 First Order Systems- Part 2 131
Week 4
19 Second Order Systems - Part 1 136
20 Second Order Systems- Part 2 143
21 Controllers - Part 1 151
22 Controllers- Part 2 156
23 Closed Loop Control - Part 1 166
24 Closed Loop Control- Part 2 173
Week 5
25 Routh's Stabilty Criterion - Part 1 181
26 Routh's Stabilty Criterion- Part 2 188
27 Special Cases of Routh's Stability Criterion - Part 1 196
28 Special Cases of Routh's Stability Criterion- Part 2 200
29 Performance Specifications - Part 1 211
30 Performance Specifications- Part 2 216
Week 6
31 Steady State Error Analysis - Part 1 223
32 Steady State Error Analysis- Part 2 233
33 Root Locus 1 - Part 1 243
34 Root Locus 1- Part 2 250
35 Root Locus 2 - Part 1 260
36 Root Locus 2- Part 2 267
Week 7
37 Root Locus 3 - Part 1 277
38 Root Locus 3- Part 2 284
39 Root Locus 4 - Part 1 293
40 Root Locus 4- Part 2 301
41 Case Study - Modelling - Part 1 308
42 Case Study - Modelling- Part 2 312
Week 8
43 Case Study - Control Design - Part 1 316
44 Case Study - Control Design- Part 2 322
45 State Space Representation - Part 1 332
46 State Space Representation- Part 2 340
47 Frequency Response - Part 1 344
48 Frequency Response- Part 2 350
Week 9
49 Bode Plot 1 - Part 1 354
50 Bode Plot 1- Part 2 360
51 Bode Plot 2 - Part 1 364
52 Bode Plot 2- Part 2 369
53 Bode Plot 3 - Part 1 374
54 Bode Plot 3- Part 2 380
Week 10
55 Bode Plot 4 - Part 1 388
56 Bode Plot 4- Part 2 393
57 Nyquist Plot 1 - Part 1 400
58 Nyquist Plot 1- Part 2 405
59 Nyquist Plot 2 - Part 1 413
60 Nyquist Plot 2- Part 2 417
Week 11
61 Nyquist Stability Criterion - Part 1 426
62 Nyquist Stability Criterion- Part 2 432
63 Relative Stability 1 - Part 1 442
64 Relative Stability 1- Part 2 447
65 Relative Stability 2 - Part 1 454
66 Relative Stability 2- Part 2 459
Week 12
67 Lead Compensation - Part 1 465
68 Lead Compensation- Part 2 471
69 Lead Compensator Design - Part 1 477
70 Lead Compensator Design- Part 2 483
71 Lag and Lag-Lead Compensation - Part 1 490
72 Lag and Lag-Lead Compensation- Part 2 496
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 01
Introduction to Control Systems
Part - 1

Greetings, welcome to this course on Control Systems; this is a course on control

systems.

(Refer Slide Time: 00:26)

And to be more specific, this is a first level undergraduate course on control systems
where we are going to learn the basics behind what is called as classical control theory.
And we would learn how to design controllers for dynamic systems that we encounter in
practice.

Before we go into the details of what we would learn in this particular course, let us just
look at the title of this course. There are two words in this title. The first word is what we
call as a system. Let us discuss what we would refer to as a system in this particular
course. For us a system is an entity, to which we give an input u(t ) and we get an
output y (t) .

1
This is typically called as a Dynamic System, it essentially deals with variables, inputs
and outputs that are functions of time. So, as far as this dynamic system is concerned,
time is the independent variable; and all other variables that are associated with the
system are functions of time.

We are going to deal with the so called dynamic systems, where variables change with
time. And a system or a process or a plant is an entity that is given an input and produces
an output. That is the visualization we would be having for the system. There are
alternative terms for the system, some would call the system as a plant.

If you go to chemical engineering, for example, they would be dealing with the
processes, so a process is also visualized as an entity to which we give an input and we
get an output. So, there are various terms that are typically used. For us, a system is a
collection of objects, let us group it as one phrase, a process and that is understood. That
is what we would refer to as a system.

Although we are not going to repeatedly use various adjectives to denote the type of
systems that we are going to look at. We will first study what class of systems that we are
going to be concentrating on and then we will go forward.

The class of dynamic systems that we would be looking at in this particular course
belong to what are called as linear time invariant causal dynamic systems, and by and
large we will restrict ourselves to what are called single input single output systems. So,
before we get into that, we would see that mathematically we can visualize a dynamic
system as a mapping S() from u(t) to y (t) . So, what we are going to say is
that, we have y (t )=S ( u ( t ) ) .

So, essentially we would picturize a system as a mapping S() which takes an input
u(t) and delivers an output y (t) . That is a mathematical visualization of a
dynamic system. Given this there are various classes of systems that one could
essentially study. So, let us look at some classes of Dynamic Systems.

(Refer Slide Time: 06:24)

2
As far as classification of dynamic systems is concerned; the first classification that we
would consider is based on the number of inputs and outputs, what we abbreviate as
SISO and MIMO. SISO is nothing but Single Input Single Output, and the abbreviation
MIMO stands for Multiple Input Multiple Output systems.

We can visualize, the input and the output as a mapping from the time domain to the
domain of real numbers. For example, if I deal with a single input single output system, I
can say even the input u(t ) is a mapping from the field of real numbers to the field of
real numbers. Because t belongs to the field of a real numbers. Essentially, we take
time and the input is a function of time and in the case of a single input single output
system the input u(t ) is a scalar valued function of time.

So, u(t) would be a scalar valued function of time. So, that is why we write it in this
form. And similarly the output y (t) is also a scalar valued function of time. So, that is
a single input single output system. On the other hand, let us say we have, M inputs and
P outputs with the M > 1 and also P > 1.

What is going to happen is that; we have input to be a vector valued function which will
take time and then map it to a vector of dimension M. Similarly, we have the output
vector to be a vector valued function of time which will map it to a vector of dimension
P. So, that will be a scenario of multiple input multiple output system.

In this particular course we are going to be interested in SISO systems, SISO dynamic
systems. So, the first classification was based on the number of inputs and outputs. The

3
second classification that we are going to look at is whether a particular a system is
linear or non-linear.

(Refer Slide Time: 09:33)

Let us look at the definition of linearity shortly. Let us say we have a system S ( ) and
let us say we provide an input u1 (t) and we get an output y 1 (t) . And to the same
system and let us say we provide an input u2 (t) and we get an output y 2 (t) .

Condition 1: Now, given these two arbitrary input output combinations, the question that
we ask ourselves is that; if, to the system, we provide an input that is the sum of these
two arbitrary inputs u1 (t ) +u2 (t) and if the output is also the sum of the two individual
outputs y 1 ( t )∧ y 2 (t) corresponding to u1 (t ) and u2 (t) .

Condition 2: And if we give any input which is an arbitrary scalar multiple of any
arbitrary input u1 (t) , and if the output is the same scalar multiple of the
corresponding output y 1 (t) .

Then the system is said to be a linear one, provided these two conditions are satisfied,
(condition 1 and condition 2). The first condition what is typically referred to as,
additivity the second condition is referred to as homogeneity. The same definition is the
definition of a linear system, it can be re written in this way; if we have a system S()
and we provide an input which is any arbitrary linear combination of u1 (t ) and

4
 u2 (t) ; that means providing a input like c 1 u1 (t )+c 2 u2 (t ) where c1 and c2
are two real numbers.

The question is, whether we get the same linear combination of the two corresponding
outputs y 1 (t) and y 2 (t) . If this holds true then the system is linear. These are two
alternative and equivalent ways of defining a linear system, typically people call the
second one as the principle of superposition.

We would see in reality that most processes, most plants, most systems are going to be
non-linear. Linearity, we would realize when we work with practical problems is that, it
is going to be an approximation of reality. This should be kept in mind. In this particular
course we are going to deal with linear systems.

Later, we will go through all the classifications and summarize what we are going to deal
with. The third classification that we are going to look at is that of a time invariant
system versus a time varying system. The question is that, what is this concept of time
invariance. Let us demonstrate by a simple diagram. So, let us say we have a system and
we provide some input u(t) to it at time t=0 , let us say we provide a step input.

And we get an output. Now let us say, we now provide the same input u(t) after a
time interval of T . The question becomes do we get the same output. If the output is
the same function which is delayed by T, then the system is said to be time invariant or in
other words the question that we are asking ourselves is that, do we get the same output;
if we provide the same input no matter at what time we provide the input to the system, if
that is the case then the system is said to be time invariant.

5
(Refer Slide Time: 15:56)

A time invariant system is one that provides the same output for the same input
irrespective of when the input is given. For example, we take the example of a ceiling
fan, that most of us encounter every day.

Suppose, if I come and switch on a ceiling fan, I am providing an electrical signal as an

input to the system. And let us say the output of the system is the rpm of the fan, the
speed of rotation of the fan. So, I switch on the fan today, and I give some electrical input
to the fan. And let us say the speed of the fan is, 100 rpm.

And I come tomorrow and switch on the same fan, provided the electrical input is the
same, I would expect the speed to be close to 100 rpm once again. Now if I come after a
month and switch on the same fan there may be some wear and tear in the fan. Given the
same electrical input signal, I may get the rpm as 99.5.

If I come one year down the line and switch on the same fan, giving the same electrical
input signal the rpm may be 99 rpm. For all systems dynamic characteristics change with
time, due to various factors. In the case of a fan there may be some wear and tear, the
components might have become old, then the systems response changes.

Now, the question that we are asking ourselves is that in the time scale of interest, as far
as systems analysis is concerned, do that response characteristics of the system remain
more or less the same. If that is the case, then we approximate the system as a time

6
invariant system. In the case of a fan, let us say I want to analyze the fan only for a few
minutes.

In those few minutes I can reasonably assume that the response characteristic of the fan
is not going to change. What do I mean by that? In a timescale of a few minutes if I
provide the same electrical signal at any instant of time I am going to get close to 100
rpm. Which is what I started off here. On the other hand, if I use a fan in an industrial
application where it is supposed to run for months without any shutdown then I need to
be a bit more careful.

But then my timescale of interest goes in the order of months. So, then we need to ask
ourselves the question, do I need to consider the variation in the systems response
characteristics and model the system as a time varying system. So, there is a question
which should be considered, asked then. In our course, we would work with what are
called time invariant systems. That is the scope of our particular course.

As far as we are concerned, what is the mathematical definition of time invariant

systems; if y (t) is S (u ( t )) then time invariance implies that
S ( u ( t−T ) ) = y (t−T ) , for all T belonging to the field of real numbers. In essence, if I
shift the input by a time of T, I should get the same output shifted by the same amount of
time. That is what time invariance means.

In this particular course we are going to work with time invariant systems. Once again
time invariance is an approximation. I am sure all of us agree that if you look at
processes and systems that surround us you will see that no system provides the same
output, at any point of time in it is entire lifetime right. So, time variance may become
important.

But the question we need to ask ourselves is that: what is the time scale of interest to be
serviced, the time scale over which the systems characteristics change, when we decide
to approximate something as time invariant or time varying.

Another classification that we are going to do is that of causal systems versus non causal
systems. What is a causal system? A causal system is one where the output at any instant
of time depends only on past and current inputs. A causal system is one in effect, that is

7
non anticipative, It does not try to figure out what inputs may come to the system in the
near future and take action right now.

(Refer Slide Time: 22:37)

A causal system is non anticipative, whereas a non-causal system is anticipative.

In this particular course we are going to consider systems that are causal. Let us recall
where we really started off and what we are doing in this particular course. We started off
with the title of this course. The first word we started off looking at is the word
SYSTEMS.

We are going to look at what are called DYNAMIC SYSTEMS. We are going to look at
CAUSAL SYSTEMS. We are going to look at TIME INVARIANT SYSTEMS. We are
going to look at LINEAR SYSTEMS and we are going to look at SINGLE INPUT
SINGLE OUTPUT SYSTEMS.

This is the class of systems that will be under study in this course. We are going to
consider SISO LINEAR TIME INVARIANT CAUSAL DYNAMIC SYSTEMS. I hope
through this discussion it is clear as to what each one of these words mean.

By and large linear time invariant is abbreviated as LTI. Typically, LTV stands for Linear
Time Varying systems. Of course, when I ask the question is this a reasonable class of
systems to study; the answer is yes. This is a useful class of systems to study because we

8
will see that, when we later look at case studies and so on, this is a reasonable
approximation as far as many practical problems are concerned.

Now, having looked at the word system now we come to the word Control. What is
control? The word control essentially means making a system behave as design. That in a
very generic sense is what we would refer to as control.

Now, the questions are how we regulate the system. That is something which we are
going to look at. Let us look at an example, armature control of a DC motor. I provide a
voltage v (t ) to the DC motor and the shaft speed ω( t) is output of the DC motor.
I visualize the DC motor as a system to which I provide a voltage input v (t ) , and get
the speed output ω( t) . Now the question when we want to design controller is that; if
I want to maintain the speed of the DC motors output shaft at a particular value, what
voltage input do I need to provide. If we wish to achieve ω d , desired angular speed
for the DC motor, what is the input voltage that should be provided. This whole course
about a learning how do we do this process, which is the idea behind this.

In a general sense, we are going to learn how to design controllers, for many practical
systems, that belong to the class of interest.

What is our approach that we are going to follow? In a very generic sense, we are first
going to develop a mathematical representation for S ( ) . That is the first step. S()
is the mapping between the input and output, and we follow this process to get a
mapping for S ( ) .

9
(Refer Slide Time: 30:16)

This typically constitutes what is called as a mathematical modeling for the dynamic
systems. There can be various ways of obtaining these mathematical models. In this
particular course we would employ physics based models; of course, we can also have a
data driven models or empirical models and sometimes we use both.

We use physics to essentially derive part of the model, and a part of the model is
obtained by using experimental data. There can be a semi empirical approach also. First
step is to essentially develop a mathematical representation for the mapping S() and
then second step is to analyze the system response. And then the third step is essentially
to design the controller.

So, in a very broad sense, a big picture view perspective, these are the three steps
involved in our approach, which we are going to learn in this particular course. There are
also various categories of models we will see.

10
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 02
Introduction to Control Systems
Part - 2

(Refer Slide Time: 00:17)

What do we mean by these models and how do we get them and what do we use them
for? So, that is something which we need to understand. So, once again I am revisiting
our idea of what we mean by a system. So, for us, a system is a mapping between u (t )
and y (t ) . Now, suppose if, let me call this as problem 1 ok. The first problem I want to
solve is that, given u (t ) and y (t ) , we are asked to find the mapping S, that relates the
input u (t ) and the output y (t ) . This problem is what is called as the problem of
synthesis.

That is, suppose let us say, let me take an example of a room that is being air-
conditioned. So, let us say if I take the example of a room that is being air-conditioned
then suppose, if I want to regulate the temperature of the room, the temperature of air in
the room is the output of the system and the input to the system is the amount of cold air
that I blow into the room.

11
Let us say we have a room air conditioner and let us say the temperature of air in the
room is my output and let us say the amount of cold air provided right to the room
through the ac vents is the input of the system. So, what would this synthesis problem
entail? The problem of synthesis essentially means that I provide varying quantities of
the input to the system and then measure the output and try to get a mapping between the
two. So, that is what I do so that I get a mathematical model for the system.

So, in this case, if I provide various amounts of cold air to my air conditioner and I
measure the corresponding temperature and I get a relationship between the two, then I
will know how the temperature of air in this particular room would vary. So, once I do
this first problem of synthesis, I can go and do problem number 2, which is going to be
the following: Given the mapping S, once we find the mapping S, and given a particular
input u (t ) , find y (t ) .

Suppose, I have got a mathematical model for the temperature of air in this room
equipped with an air conditioner. So, tomorrow let us say, I want to essentially change
the air conditioner and let us say I have 5 choices. Without even purchasing them I can
use the model which I have developed to figure out how temperature would vary and
settle down to a desired value by doing this analysis right, once I have the mathematical
model for this particular process of air conditioning.

So, the advantage is that I can predict the temperature variation without even buying the
air conditioner, installing it and testing it so that I can make a well-informed choice as far
as which AC to buy. So, this problem is what is called the analysis problem or the
prediction problem.

Another problem, which we can do once we complete the synthesis problem is that,
given the mapping S and a desired output y (t ) , find u (t ) . That is the third problem
that we can solve. So, this problem is what is called the control problem.

So, what do I mean by this? Suppose, let us say I come into a room with an air
conditioner and I essentially want to regulate the temperature of air in this room. So, I set
the temperature of the air at 25 degree Celsius. The question is that, what is the amount
of cold air, which I should blow, which would get my temperature to 25 degree Celsius.
That is the problem of control.

12
So, I know what the mapping of the dynamic system is and I want the desired output,
what is the input which I have to provide? So, in a certain sense, it is the inverse problem
of what we do in problem number 2. So, this mapping helps us to find a y (t ) given a
u (t ) . In the problem of control, what we do is that, given y (t ) I want to use this

mapping to find what u (t ) will get me the output y (t ) . So, that is the problem of
control.

So, what are a few examples of control that we have already look at? One is room
temperature control. So, that is, that is one example that we have discussed. Let us say,
we have also looked at an example, where we want to control the speed of a motor. And
our human body is a marvellous controller.

Say, for example, our human body maintains our body temperature in a very-very narrow
band. So, even if the temperature goes to 100 degrees Fahrenheit right, so we are in
trouble, correct. So, essentially our body maintains our internal body temperature, in a
very narrow band irrespective of what that environment temperature is.

So, it does not matter whether we are in winter or in summer you know the human
internal body temperature is maintained at almost a constant value and that is a great
control system. For example, the blood pressure that is maintained by your body. You
know it should be in a good range. Blood sugar level is another marvellous control
mechanism. And in fact, even our heart which is a pump which essentially pumps blood
throughout the body needs to work repeatedly again and again as far as we live. So, that
is extremely important to us. So, our human body is filled with marvellous controllers,
which regulate various variables around the desired values very accurately and for a long
time and that is something which is marvellous.

So, essentially we are going to look at various case studies, as we go along and then we
will see how to formulate practical problems as a control problem and solve them, ok.
So, before we get into the mathematics I just want to give another physical perspective,
as far as how we look at control and so on.

13
(Refer Slide Time: 08:41)

So, essentially in a very broad sense, people classify control as what is called an open
loop control and closed loop control. What are these terms? Let us say we consider the
example of a ceiling fan. So, I have a ceiling fan, I have a regulator and I have a switch,
right. I can come, switch on the fan and I can vary the speed setting, maybe in discrete
quantities by using the fan speed regulator and that is about it, right.

So, I just get some output, which is a fan RPM. So, that is an open loop controller in the
sense that, it cannot account for any disturbances, which come during the operation of
the fan. For example, if there is a voltage fluctuation, then, the output of the fan or the
speed of rotation of the fan is going to fall down or go up depending on whether the
voltage is falling down or going up. So, then, you know like, the question is that like I
lose certain performance right, but the question we need to ask ourselves is that, is it
important, for a ceiling fan? So, typically ceiling fans are open loop control systems,
where the speed settings are calibrated in the factory and the regulator essentially, adjusts
the fan RPM in a discrete set of values.

Essentially in the presence of disturbances, obviously, the system response gets affected.
So, that is the characteristic of open-loop control. So, what are the various characteristics
of open loop control: There is no feedback (we will shortly define what is called as
feedback). So, essentially, it cannot tolerate disturbances; it is not robust to disturbances
and so on, ok. But on the flip side, the cost and complexity are lower.

14
On the other hand, if we want a fan, whose RPM needs to be maintained at a baseline
value, let us say 100 RPM and I cannot afford a huge variation in the RPM, right, for
some industrial application, what would I do? What I would do is that, I would measure
the actual RPM using a speed sensor and then I would take the actual RPM at each and
every instant of time, compare it with what is the desired value and then take the
difference, quantify the difference between what I desire and what is actually being
obtained, take that error and pass it through what is called as a controller and the
controller will then adjust the electrical input to the fan. Then, when that happens we
have what is called as closed-loop control and closed loop control systems have
feedback. Feedback is the process of measuring variables that need to be regulated or
controlled so that corrective action can be taken ok, that is the process of feedback.

So, consequently closed loop control is more tolerant or robust to uncertainties that come
in disturbances and so on (what we call as un-modelled dynamics). So, but anyway let
me group them under uncertainties, ok. But the flipside of closed-loop control is that the
cost and complexity are higher.

In this particular course, we are going to deal with closed-loop control. So, that is what
we are going to deal with in this particular course. So, let me just construct an example to
just explain what we did. Suppose, let us say, we have, let us say, a DC motor right. So,
the DC motor is our system or plant, ok. And to this DC motor, let us say, I provide a
voltage V (t ) as my input and  (t ) is my output.

Suppose, let us say I want a desired output (desired RPM of rotation from the motor) and
this is what we call as a reference input and what we do then is the following. Suppose if
I measure the actual output, then what I do is that I go and compare, then I take the
difference between what I desire and what I measure, which is what is called as the error
and then, let us say, we pass it through an element called as a controller and the
controller then calculates what should be the input that should be provided to the DC
motor, ok.

When the controller calculates the input that should be provided to the system, the
adjective control is added to the term ‘input’. So, the input to the system becomes what is
called as the control input. So, this is feedback, ok. So, this is a typical layout of a closed
loop control system with what is called as negative feedback, ok.

15
So, this is essentially closed loop control system with negative feedback. Why is it
negative feedback? Because at the summing junction you can see that, we are subtracting
the feedback signal from the desired reference input. So, that is why it is called as
negative feedback.

So, we are taking the difference between the reference input and the actual output and
calculating the error. This is what is called a closed loop control system with negative
feedback. So, we can see that this is what we are going to essentially do in this particular
course. So, some more terms that I want to introduce here is that, like, this process of
essentially taking this output measurement and feeding it back is what is called as
feedback. This is what is called a feedback path.

(Refer Slide Time: 17:47)

Typically, we can have a mapping in the feedback path, say we can have some sensor
dynamics coming into play here. So, let us say I use a sensor for measuring speed, the
sensor may have its own dynamic characteristics. I may need to use a filter to filter out
noise, then a mapping comes in the feedback path ok.

So, if a mapping comes in the feedback path (when the mapping is not one ) we call it as
a non-unity feedback. So, when this mapping in the feedback path is 1, we call it as unity
feedback. We will see how this affects our analysis later on right and many times when
the controller calculates and provides a control signal to the system, it is typically
realized by what is called an actuator.

16
Say, for example, let us say you know I have to essentially move, say, design a motion
control system that essentially displaces a work piece along one axis right. Let us say,
simple translation of a work piece in a machining system. So, let us say, I want to
regulate the position of the work piece. I give a, a voltage signal to, let us say, an electric
motor drive system, which essentially provides translational motion to the work piece
right. Suppose, if the controller calculates that, at this point of time, provide 5 volts of
input right to the electric motor or if it calculates and tells me that look, you know like,
provide, you know like, 10 Newton’s of force to the work piece, in order to essentially
move by some distance x , right. So, the controller may say, provide 10 Newton’s at this
instant of time, but then there may be a small, what to say, response time before the 10
Newton’s is actually realized in practice through the electric motor and drive system,
right.

So, that, if that is important then you know, we need to figure out what is called as
actuator dynamics in the design process, ok and sometimes we can have disturbances
coming into the system. Let us say, I have a sudden load that may come on a DC motor,
right, for example, you know that I mean to model as a disturbance and then like see how
we can overcome such disturbances and so on, right.

So, one can see that, you know, we can add varying levels of complexity to this,
feedback system, ok. We are going to study the basic feedback loop in this particular
course and as we go along maybe when we go to some case studies, we will add some
more blocks and then see how the design varies.

To summarize, what we are going to do in this course is the following. So, as a summary
of all this discussion that we did, the title of our course is control systems, right. So, what
we are going to do, what we are going to learn is the following: Closed Loop Feedback
Control of SISO LTI Causal Dynamic Systems. So, that is what we are going to do in
this particular course, ok.

So, that is the class of systems that we are going to do, SISO LTI causal dynamic system
and we are going to do what is called as closed loop feedback control, right, of this class
of systems and it so turns out that this class of systems, SISO LTI causal dynamic
systems, the mathematical models that are typically used to characterize this class of

17
systems usually take the form of a Linear Ordinary Differential Equations, what are
abbreviated as ODEs, with constant coefficients.

So, typically the mathematical models that are used to characterize this class of systems
that we are going to study in this particular course take the form of linear ordinary
differential equations with constant coefficients, ok. So, that is the class of equations that
we would be focusing on.

And, another important aspect is that the kind of mathematical models we are going to
essentially look at are what are called as spatially homogeneous. That means that we do
not consider the variations of variables with space. So, we only consider temporal
variations of variables. For example, let us say, I want to analyze the variation of the
temperature of air in this room, right. So obviously, the temperature of the air in this
room can be different depending on the point where I am measuring and also the time at
which I measure, right. So, but then what we do is that we lump all the points in this
room i.e. the temperature of all the points in this room into a single entity, which
essentially can be represented as a function of time, ok.

So, in a certain sense, what we are doing is that we are lumping the effects as far as
spatial variation is concerned and we are going to assume that the entire room can be
characterized by a single temperature which is only a function of time, ok. So, spatially
homogeneous means we are essentially following sort of a lumped parameter approach
towards our modelling.

Then we are going to have what are called continuous time, dynamic, deterministic,
mathematical models, ok. That is the class of models we are going to use, right. So,
essentially in a certain sense, continuous time means that we treat time as a continuous
variable. So from a pragmatic perspective, it just means that we are going to get ODEs
right, Ordinary Differential Equations. Dynamic means we are going to have derivatives
in the equations.

So, what is a dynamic model? It is something which explicitly considers future states of
the system, right, so, and it essentially incorporates what is going to happen to the system
in the future by incorporating derivatives of the variables. So, that is a dynamic model.
Deterministic means we essentially neglect any stochastic effects, ok. We do not consider

18
variables as random variables. We consider all variables as deterministic variables and
we are going to have deterministic models, ok.

So, this is the class of models that we are going to use. So, in summary, what we are
going to do in this course is closed loop feedback control of SISO LTI causal dynamic
systems using spatially homogeneous, continuous time, dynamic, deterministic
mathematical models, ok. So, that is what we are going to essentially learn how to do in
this course, ok. So, what we would subsequently do is essentially have a brief recap of
what is the mathematical background that is required to do this analysis, ok.

So, that is something which we are going to recap. It is assumed that some mathematical
courses have already been completed before one comes to this particular course,
particularly courses on complex variables, ordinary differential equations and Laplace
transform. So, we would quickly recap some of these tools and then we will move
forward, ok. So, that is going to be the plan of action. Fine.

Thank you.

19
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 03
Over view of Feedback Control Systems
Part – 1

(Refer Slide Time: 00:20)

In the previous class, we had a broad overview of dynamic systems. We learnt about a
dynamic system and what are the different classes of dynamic systems based on certain
attributes. We learned the class of dynamic systems that we would deal in this course
come under linear time invariant causal single input single output dynamic systems.
Additionally, we learned the definitions of each of these words that we use along with
the word system; that is we learned what is meant by a linear system, a time invariant
system, a causal system, and so on. And then we looked at what is meant by an open loop
system and closed loop system. We also looked at the concept of feedback and we drew
the basic layout of a closed loop control system with feedback and that is where we
stopped.

We are going to continue from that point and discuss a few more concepts that would
provide a big picture view of this particular course, before we go into mathematical
preliminaries in the next lecture. If you recall, for us a system is an entity to which we

20
provide an input u(t) and we get an output y (t) . If we want to regulate the value
of the system output y (t) around a desired point or in a desired region, what we do is
that we give what is called as a reference input which is denoted by r (t) . Please note
that in this particular course, we are dealing with the dynamic systems, time is the
independent variable and all other variables are functions of time. We see that the input,
output and the reference input in general can be functions of time. We measure this
output, and then feed it back and compare the value of the output with that of the
reference input. We get an error e (t) , the error is the difference between the desired
value of the output and the actual output that we are obtaining at each and every instant
of time.

We pass this error through an element called as a controller which then calculates what
should be the input that needs to be provided to the system so that it can follow the
desired output r (t) . So, in other words, the controller is the block that takes in the
error between what is desired and what is actually obtained and then calculates what
should be the corresponding input that needs to be provided to the system.

When the controller calculates the system input and provides it to the same, the function
u(t) is then called as a control input. This is a broad visualization of a closed loop
control system with feedback, in this case it is what is called as negative feedback. The
reason why it is called as negative feedback is because at the summing junction where
we are comparing the reference input to the output, here we are taking a difference
between r (t) and y (t) . So, it is called as a negative feedback.

There can be various other attributes that can be added to this basic closed loop structure,
and that is something which we are going to look at now. The path wherein we go from
the controller to the system and then to the output is called as the forward path and the
part wherein we give the measurement of the output back to the summing junction so
that we can calculate the error, which then goes to the controller, is called as the feedback
path. And in this feedback path, one can have other dynamic components, which may
need to be modeled, depending on the system.

For example, one would typically use a sensor to measure the system output, then one
may need to take into account the dynamics of the sensor that is used. In other words,

21
there is an additional attribute called sensor dynamics that may come into picture in the
feedback path. What do we mean by that?

Suppose we want to control the temperature of air in a room. If I use a temperature

sensor to measure the temperature of air in a room, that sensor may have it is own
response characteristics. In other words, if I use the sensor to measure the temperature at
this instant of time, it may not show the temperature value instantaneously. The correct
value at this instant of time may be showed after a certain response time which needs to
be characterized by what is called a sensor dynamics. The question that we need to ask
ourselves when we design this closed loop control system is that: what is the time scale
of the sensor’s response with regards to the time scale of the systems response.

For example; if the system responds in a matter of seconds. And if I have a sensor which
can provide accurate measurements in a matter of milliseconds, then I may not need to
worry about the sensor dynamics. On the other hand, if the system responds in a matter
of seconds and the sensor also has a corresponding response time in the order of seconds,
then I may need to incorporate the effect of sensor dynamics.

So, sensor dynamics may come into picture when we consider the feedback path. And
correspondingly an additional component called an actuator may come in the forward
path. What is actuator dynamics? The controller calculates the control input and provides
the control input to the system, but we need some element to physically realize this
control input. For example; suppose, we want to drive a car autonomously. So, let us
consider the task of steering a car. The angle which is provided at the steering wheel is
the input to the system and the orientation of the car can be considered as the output from
the system.

Now, if I want a certain orientation for the car, then I design a controller to calculate
what should be the steering wheel input that should be provided at each and every instant
of time. Now how do I realize the calculated control input in practice? I may put a motor
at the steering linkages to rotate them by a desired angle whatever the controller
calculates.

Let us say at this instant of time, I want the steering, linkage or steering column to be
rotated by 40 degrees. Now the thing is whether I achieve that 40 degree steering angle
instantaneously. The answer may be most probably not. If I use a motor to rotate the

22
steering column and I give a command now, it may take some time before which the
steering column may be actually rotated by the control input angle of 40 degrees. That
aspect or that attribute needs to be characterized by as actuator dynamics.

Actuator dynamics essentially incorporates the response characteristics of actuators that

take the controller’s control input and realize it in the physical world. So, one may need
to consider actuator dynamics. Once again, it is all relative, if the actuator responds very
fast when compared to the systems dynamic characteristics. Then one need not consider
actuator dynamics. On the other hand, if the system responds in a matter of seconds and
the actuator also responds in a matter of seconds. Then we need to consider actuator
dynamics while modeling the entire closed loop control system. That is something which
we should realize. So, sensor dynamics and actuator dynamics become important when
we design these closed loop control systems for real world applications.

What are the other attributes that we need to consider in this a block diagram? Most
often we can have disturbances coming into play. Disturbances are essentially unwanted
which affect the system output. For example, I am driving a car and there may be some
crosswind which is blowing on the road which can act as a disturbance and it may try to
change the orientation of the car. So, although in the ideal world, I want the orientation
of the car to change only when I turn the steering wheel, a strong crosswind can create a
disturbance which can potentially lead to a change in the car’s orientation. How does one
consider the effect of disturbances and account for them? That is a question one needs to
ask when designing this closed loop control system. The effect of disturbances may play
an important role, and we have to design a controller that is a robust enough to these
disturbances. These are some elements that we need to consider in the feedback diagram,

If you recall the class of systems that we have been considering fall under single input
single output causal linear time invariant dynamic systems. It so happens that, if we want
to model this class of systems using mathematical models, they take the form of a linear
ordinary differential equations with constant coefficients. The mathematics that we use in
this course would compose of related topics. We will briefly recap ordinary differential
equations complex variables, Laplace transform and so on, that one typically learns in a
course involving ordinary differential equations.

23
Depending on how one goes about analyzing these ODEs to study the systems response,
one could have two representations of the mathematical model that characterizes this
class of systems. If one uses the Laplace transform, then we get what is known as a
transfer function representation. This is the representation that we would predominantly
use in this course.

(Refer Slide Time: 15:57)

On the other hand, I do not apply the Laplace transform to the governing equations, but I
stay in the time domain itself then we rewrite the governing ODE as a set of first order
differential equations. And then we get what is called as a state space representation of
this class of systems. In this course we would learn both representations. The Laplace
transform takes functions that are essentially real valued functions of real variables, and
then converts them to complex valued functions in the complex domain. Predominantly
we are going to look at the transfer function representation. All the analysis and control
design is going to be done using the transfer function representation.

There are a few more concepts that we would be discussing with respect to the general
closed loop control system. Let us write down a set of points. Depending on what is the
mapping in the feedback path, we can have unity and non-unity feedback. What is unity
feedback? If the mapping in the feedback path is unity or 1, then it is called as unity
feedback, if not it is called as non-unity feedback. We would have mappings in the
feedback path which are going to be represented by corresponding transfer functions.

24
And if we have unity feedback that particular transfer function for the feedback path
would be 1.

(Refer Slide Time: 19:10)

The second point is that, as far as this course on control systems is concerned, any design
of a control system has two important aspects. If I am designing a controller for a
dynamic system, I always need to focus on two important attributes. What are they? The
first one is what is called a stability and the second one is what is called as performance.
All of us can get a very intuitive idea by looking at these terms.

For a control system designer stability is paramount; that means, we need to design
systems that is stable in the first place. Then we essentially look at the performance
characteristics and try to design controllers so that their performance could be improved.
Later, we are going to carefully define stability and performance, and also identify
quantities and parameters that would essentially be utilized in quantifying our
performance characteristics.

For example, I am driving a car, I give a steering input. If the conditions happened to be
such that my car may spin out of control. So, one would intuitively call that scenario as
the car being unstable. A stable response means I would want the system’s response to be
as we desire in a very finite range for a finite input which is given.

25
What is performance? Performance can essentially be visualized through, how fast the
system responds. For example; I am designing one steering system and someone else is
designing another steering system for the same car. Let us say both of us give the same
input to the steering wheel. My car is able to take a turn faster and more accurately, then
I would say that my steering system is performing better for that particular car.

We would use a notions like these to come up with parameters that can quantify
performance of dynamic system. These are concepts that give us a big picture view of
what we are going to learn in this particular course.

26
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 04
Overview of Feedback Control Systems
Part - 2

(Refer Slide Time: 00:14)

So, to just convey these ideas and also to co-relate what we have learnt, let me take an
example and then we would discuss a few points, ok. So, let me take an example of a
typical mass spring damper system that all of us are familiar with. Let us say I have a
mass, this is a one degree of freedom mechanical system. So, let us say, I have a mass
m which is connected to a linear spring with spring constant of k and a linear damper
with a damping coefficient of c , ok. And let us say I apply a force f (t ) on this
particular mass and let the displacement of this mass from its equilibrium position be
x (t ) .

So the governing equation of motion for this system, derived based on the Newton’s
second law of motion, is
.. .
m x(t )  c x(t )  kx(t )  f (t ) .

27
Of course, in this particular course if I use a dot over a variable, that means, I am taking
. ..
it's time derivative. So, x(t ) means it is the first time derivative of x(t ) , x(t ) means

it is the second derivative of x(t ) with respect to time and so on, ok.\

So this is the governing equation for this mass spring damper system. As we know,
.. .
m x (t ) characterizes the inertia effects that occur in the system; c x (t ) corresponds to

the viscous dissipation that happens due to the presence of damper; kx(t ) is indicative
of the compliance that is present in the system.

So, these are typically used to visualize this class of mass spring damper systems and
even one would use these analogies to develop models for even other systems from other
domains. So, but let us consider this example and we immediately realize that the
governing equation is a second order linear inhomogeneous ordinary differential
equation with constant coefficients, ok.

(Refer Slide Time: 03:06)

So, we can see that, the order of the system is 2, because we have a second order
derivative, in the governing equation. It is linear as is evident from the governing
equation, because the equation is linear in x(t ) and f (t ) . It is inhomogeneous because
we have a forcing term on the right hand side of this ordinary differential equation. And
the ODE has constant coefficients which is reflective of the system being time invariant.

28
So, this is essentially an LTI causal single input single output system, ok, the class of
systems that we are essentially looking up.

So, if we look at the response of the system, we are going to have two components, one
due to the presence of any potential non-zero initial conditions and other due to input
provided. So, for example, I can displace the mass a little bit and then leave it, right. So,
then I am giving a non-zero displacement as an initial condition. This would be called as
the free response. And then we can also have a system response, due to the input
provided. In this case, the force is the input to system.

So in fact, if we want to visualize this using the typical systems perspective, in this case
of the mass spring damper system, the input is the force f (t ) and the output is the
displacement x(t ) . So, in other words, based on the input (force) that is provided I can
have a displacement, right. So, that is what is called as forced response of the system.
The system response can have both components (free and forced response) and this is an
example of a linear time invariant causal SISO system.

On the other hand, let us say, we consider a scenario where the spring constant changes
with time. So, for example, if I keep on operating the system for a long time, the spring
may weaken, right. So, then the spring constant may change with time, right. Let us
consider the scenario where the spring constant changes with time, then the governing
equation could be rewritten as,

.. .
m x(t )  c x (t )  k (t ) x (t )  f (t )

So, this k (t ) is a representative to of a time varying spring constant. A time varying

parameter comes in the system model. So, as a result this becomes what is called as a
linear time varying system. So, this is just an example of when we could potentially get a
linear time varying system instead of a linear time invariant system. We can immediately
realize that the governing equation in this case is once again a second order linear
inhomogeneous ordinary differential equation but with time varying coefficients. So, we
will have time varying coefficients in the governing equation, ok. So, that is something
which will come into play.

29
(Refer Slide Time: 07:45)

Now, let us consider another example. Let us consider a scenario where the spring is
non-linear. This also could be a potential scenario which can be encountered in practice.
So, what do we mean by this. So essentially, in many problems in practice, if we have a
coil spring in some machine element or let us say even in our automotive suspensions,
right. We take a spring and then, we test it by giving force and then plot the force
displacement characteristics.

So, by and large what may happen is that, we may get a curve which may be non-linear,
if we test it in the entire domain, ok. So, what typically happens is the following. Let’s
say in a two wheeler, you know you would have seen a coil spring which essentially
deflects as the two-wheeler is going.

So, let us say we have the spring in an initial state of compression, right. Let us call this
displacement as xi and by and large we would operate the spring in a region around this
xi . It may so happen that the spring may be operated in a region around this initial state,

you know, most of the time. Then, we can immediately observe that in this region the
spring can be approximated as a linear element.

So, then the question that we can ask ourselves is the following, right. Given this range
of operation of the spring, can I use a linear approximation to model the process? If yes, I

30
could use a linearized model and then we can go to the class of linear time invariant
systems. But on the other hand, let us say, I am in an application where the spring is
stretched and compressed over a wide range of displacements, then I may no longer be
able to use a linear model for the spring. Then what is it that I can do?

I am just taking an example, where let us say you know, the governing equation for the
springs response is a quadratic function of the displacement, right. So, what will happen
is that the governing equation for this mass spring damper system will then become

.. .
m x(t )  c x(t )  kx 2 (t )  f (t )

So, you see that the term kx 2 (t ) makes this system non-linear, ok. So, this becomes a
non-linear system. We could immediately observe that this is still a second order ODE,
but now it becomes a non-linear ODE, right, so, non-linear inhomogeneous ODE with
constant coefficients. So that is what happens to this governing equation, ok.

So, we are just considering different examples to just learn how the governing equations
would change depending on the scenario just to get a broad idea about the class of
systems that we are studying. So, I hope just through the simple example the difference
between linear, non-linear systems and time invariant, time varying systems have
become a little bit clear and we just considered a simple example to show how these
attributes are reflected in the governing equations of such systems, ok. So, to conclude
this class, let me just show you some simple examples just simple simulation results to
convey the concepts what we have learnt.

(Refer Slide Time: 14:19)

31
So, let us consider the mass spring damper system that we have been looking at. Let us
say, I consider these following parameters. I consider the mass of the element to be 1
kilogram, the damping coefficient to be 0.5 Ns/m and the spring constant 0.75 N/m, just
take some values to just solve those equations, right. So, let us say that we provide a
force input of 1 Newton at time t=0.

So, if we solve this mass spring damper system, as we have already discussed, we can
have what is called as the free response of the system, that is we can have non-zero
initial conditions and we have what is called as the free response which is essentially
indicated by this blue curve, right. We can see that the initial displacement is at one and
then it essentially oscillates and then goes to a 0 finally, ok.

On the other hand I can have a zero initial condition and then give an input to the system,
which essentially is what is called as a forced response, right. So, that is the response
term due to the input that we provide and that is denoted by the red curve which
essentially goes and settles down at some non-zero final value, ok. So, if we combine
both that is we give non-zero initial condition and the input we will we get the curve
which is indicated black. So, we can see that the response is the sum of both
contributions, that is the free response plus forced response term gives the third response,
right, where we have a response from that mass spring damper system that consists of
both the non-zero initial condition component and the force input component, right.

So, once again please remember, what our system was. For us the mass spring damper
system was visualized as one to which we give the force as the input and the
displacement is output, ok. That is why we are essentially plotting the displacement and
studying it. So, this is just to show whatever we have learnt through a graphical
visualization.

32
(Refer Slide Time: 16:36)

So, let us now look at time variance. Now, let us say instead of having a constant spring
constant of 0.75 N/m, let us say the spring has a varying spring constant, which is of this
form 0.75e 0.001t N/m. So, the spring constant is slowly decreasing from the initial
value of 0.75, right. So, if we plot it, of course, we can see that I have just zoomed in into
a range of 0 to 100 seconds.

So, may be around the 75th second we can see that the spring constant has decreased to
around 0.693 N/m, ok. So, that is what has happened. So, and as we just discussed a few
minutes back the governing equation now becomes

.. .
m x(t )  c x (t )  k (t ) x (t )  f (t )

So, this makes it as a linear time varying system, ok. So, we can see that the system has
now, become LTV right, due to this time varying spring constant. So, let us see what
effect of this on the system response is.

33
(Refer Slide Time: 17:57)

So, we consider zero initial conditions and so let us not worry about the free response.
We only look at the force response here. So, what we do is that we are giving an unit step
force input of 1 Newton in one scenario at t=0 which is the response is indicated by the
blue curve and then what we do is that, we then give the same unit step force input at a
time, instant of 75 seconds. So, as we realize from the previous slide, at time t equals 75
seconds the spring constant decreased. So, because this is a time varying parameter in
this particular case, so you could immediately see that the response, which is plotted by
the red curve is different from the blue curve right with this essentially, going against the
definition of time invariance.

So, if you recall the definition of time invariance, if we give the same input at let us say
time t=0 and time t=75, as in this example I should get the same output right, but that is
not the case. Why? Because, we have a time varying system, ok. So we can immediately
see that the red curve is different from the blue curve, right. So, you can see the peak of
these oscillations that has increased in the red curve when compared to the blue curve,
right. So, in, in certain sense, the responses are becoming, different ok. So, this gives us
an overview of how these factors affect the system response. Through simple examples,
I just wanted to communicate how linearity time invariance and all come into play, ok.

34
So, we would stop here. So, we have essentially looked at a big picture view of what we
are going to learn in this particular course and defined a few terms in this particular class
and then looked at an example. That’s what we have done here and we also looked at
what is meant by sensor dynamics, actuator dynamics and unity, non unity feedback and
so on. So, in the next class, we would essentially meet and start off with mathematical
preliminaries and we would revise mathematical tools that we would be using in this
particular course, ok.

Thank you for your attention.

35
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 05
Mathematics Preliminaries
Part – 1

(Refer Slide Time: 00:18)

Having obtained a big picture view point of what we are going to study in this course; let us
have a brief recap of what mathematical tools that we would be requiring. As we discussed,
we are going to deal single input single output linear time invariant and causal dynamic
systems that we are going to characterize using spatially homogeneous dynamic continuous
time deterministic models.

Typically, these models take the form of an ordinary differential equation. So, the, first set of
a mathematical tools that we would require for this course is the basic working knowledge of
ODEs. Specifically we are going to consider linear ODEs with constant coefficients. Why
linear ODEs with constant coefficients? Typically linearity comes into play because we are
dealing with linear systems and because of time invariance, we are dealing with constant
coefficients. That is why we are considering linear ODEs with constant coefficients.

Let us consider some specific ordinary differential equations. But I suggest that a recap of
engineering mathematics course on ordinary differential equations would really be helpful. I

36
am just going to recap a few important concepts. If we consider a first order homogeneous
ODE of the form, of the form

dy (t )
+ ay ( t )=0,
dt

where a is a constant parameter (real number). The solution to this ODE is of the form

y (t )=c e−at .

We can write c= y (0) , where y (0) is the initial condition.

This is an initial value problem; given that the independent variable is time. I can write the
same solution as y (t )= y( 0)e−at . Why do we have an exponential solution? How do we
get this? We just substitute y (t ) of the form e mt and then we get m=−a . why should
we have exponential solutions? Why should we substitute an exponential function to begin
with? Because it has been shown that, for linear ODEs with constant coefficient, the solution
exists and the unique solution takes the form of exponential functions. In other words the
exponential function is the only solution for linear ODEs with constant coefficients, That is
why we are substituting a solution form that corresponds to an exponential function to solve
this equation.

If we now consider, a first order inhomogeneous ODE of the form

dy (t)
+ ay ( t )=bu (t) ,
dt

What is the solution to this equation?

t
+∫ e−a (t −τ ) bu(τ)dτ
−at
y ( t )= y ( 0 ) e
0

We know that it is going to comprise of two parts. The first one is due to the initial condition.
This part is typically referred to as free response. The second part is the solution, due to the
input provided. This response is what is called as forced response. Please remember we are
looking at a system to which we provide an input u(t) and an output y (t) . So
equations are taken in terms of the input on the output variables.

37
So, we see that in general, in, you know like, we have two components to the output function,
ok, like, when we model systems using this class of equations, like, the first part is, what is
called as a free response, which comes in, you know like, due to the nonzero initial
conditions the second part is what is called as the forced response, which comes due to the
input that is provided to the system, ok.

Given these responses, the question is how we use this to analyze the class of systems under
study. Once we model the systems, we would get a linear ODEs with constant coefficients.
And then we see that given any input, we can use solutions to the ODE to calculate what
would be the corresponding output from the system.

In this process, we are going to use Laplace transform that helps us to go from the time
domain to the complex domain and convert problems involving ordinary differential
equations into problems involving algebraic equations. Then we take the inverse Laplace
transform to come to the time domain.

(Refer Slide Time: 08:51)

Let us look at complex variables, then we will come back to ODEs and we will see how we
use Laplace transform on ODEs and process them. We consider complex variable of the form
s=σ + jω , where σ is some real number ω is a real number and j 2=−1 .

38
Typically, we will draw what is called as an S plane, which can be used to graphically,
represent a complex number. We will have two axis called as the real axis and the imaginary
axis and we use this to graphically represent any complex number.

And a complex function F( s) is a complex valued function of s . We write

F ( s )=F r ( σ , ω ) + j F i ( σ ,ω )

(Refer Slide Time: 10:40)

For example; we consider F ( s )=s 2 . We can rewrite this as

F ( s )=(σ + jω)2 = σ 2−ω2 + j(2 σω)

Here Fr ( σ ,ω )=σ 2−ω 2 and Fi ( σ , ω ) =(2 σω)

A few definitions:

A complex valued function F ( s) is said to be analytic in a given domain if F ( s ) and all

its derivatives exist in the domain. The points in the s domain where F ( s ) is not analytic

1
are called singular points or poles. Consider, F ( s )= ' we can immediately note that,
s+1
s=−1 is a singular point or pole.

39
There are what are called as Cauchy Riemann conditions, which are used to evaluate whether
a given function F ( s ) is analytic. The first condition is

∂ Fr ∂ Fi
=
∂σ ∂ω

The second condition is

∂ F i −∂ Fr
=
∂σ ∂ω

Another set of equations are called Euler’s relationships. They are

e jωt =cos ωt+ j sin ωt and e− jωt =cos ωt− j sin ωt

We can immediately see that

e jωt +e− jωt jωt − jωt

cos ωt = and sin ωt= e −e .
2 2j

(Refer Slide Time: 15:38)

Now we look at Laplace transform. Let us write down the definition of a Laplace transform
of a real valued function f (t) .

∞
L [ f ( t ) ]= ∫ f (t)e−st dt – Bilateral Laplace Transform
−∞

40
 ∞
L [ f ( t ) ]=∫ f (t )e−st dt – Unilateral Laplace Transform
0

In this course, we would use unilateral Laplace transform, because we are going to start
analysis of a system from time t=0 . So, we make the lower limit as 0 and consequently
we will go from 0 to ∞ .

There is something called as the inverse Laplace transform, which maps any given function
in the complex domain back to its function in the time domain. But we calculate the inverse
Laplace transform using what is called as a partial fraction expansion rather than using the
definition per say.

Before we move on to learn how we are going to apply Laplace transform in our course, let
us point out a few important functions that are going to be useful to us. Let us look at some
standard inputs. If you want to have a proper analysis done, we need to give some standard
inputs, so that we evaluate the system response. If I design, n systems for our application, I
can evaluate them provided, I give the same input to all the designs. For that purpose we use
some standard inputs. Let us look at each one of them.

(Refer Slide Time: 19:16)

First we look at unit impulse input. This is also called as a Dirac delta function denoted by
δ (t) . We consider, u(t ) to be a signal of the form shown in figure, around time

41
 −ϵ ϵ
t =0 for a very-very small time interval to , the magnitude of the signal is
2 2

1
. We see that the area of this rectangular signal is 1. That is how the adjective unit
ϵ
comes in to be.

Then, we shrink, this epsilon and make it tend to 0 ϵ → 0 . Now the input is going to just
jump to a very high magnitude instantaneously and come back to 0 instantaneously. If I
provide the unit impulse input as an input to the system. The corresponding output is what is
called as the unit impulse response.

Another input which is typically used in systems analysis is unit step input. As the name
suggests, what we are going to do is that at time t equals 0, we give a step input is of
magnitude 1, to the system. That is what is called as a unit step input. The output that we get
from the system is what is called as a unit step response. We will see that the impulse
response and the step response are going to be very valuable to us.

The third standard input that we would consider is a unit ramp input. Unit ramp input is
u ( t ) =t . The input the scales like t. The slope is 1 and that is why we call it as a unit ramp
input. When we provide the unit ramp as the input, the output that we get is what is called as
a unit ramp response.

The fourth standard input that we provide are sinusoidal inputs. Sinusoidal inputs as the name
suggests, we provide sinusoidal functions as inputs of varying frequencies cos ωt and
sin ωt . If we have a stable LTI system, and we provide a sinusoidal input, the steady state
output is also going to be sinusoidal of the same frequency. This information is used in
analysis called as frequency response analysis.

These are four common inputs that are typically provided impulse, step, ramp, and sinusoids.
In general, an input can be in any arbitrary combination of these inputs.

There are two reasons why we study the response of systems to standard inputs. We have a
uniform baseline to study systems and also we extract and define parameters for system
performance, which are based upon one type of input. And we can quantify system
performance. We are going to use step response and frequency response to define parameters
that quantify system performance, which is one reason.

42
And second, in general, any arbitrary input, can be usually written as a linear combination of
standard inputs. So, since we are dealing with linear systems once we know the output to
these standard inputs, in theory at least, we can calculate the output to any arbitrary input.

There are two important aspects that are critical in control design.

(Refer Slide Time: 26:18)

The first one is what is called as stability. We want to design stable systems and if a system is
unstable to begin with, we want to stabilize systems using feedback. That is one reason why
we design control systems. Whenever we design dynamic systems we want those systems to
be stable. We will shortly define, what notion of stability we are going to use in this course.

The second thing is, once we design stable systems, we want to evaluate how those systems
perform. We are going to be interested in performance. For us stability is paramount; once we
have stable systems, we worry about performance.

The notion of stability that we are going to use in this course is called as bounded input,
bounded output stability. It is abbreviated as BIBO. What has meant by bounded input,
bounded output stability? It means that if I provide any bounded input to the system, the
corresponding output should be bounded in magnitude for all time.

The system is said to be a BIBO stable, if given any input u(t) such that |u(t )|≤ M <∞
∀ t , output of the system is always bounded in magnitude. That is | y (t)|≤ N <∞ ∀t

43
, where M and N are finite positive real numbers. This is the notion of bounded input
bounded output stability that we will consider in this course.

These ideas are extremely important. When we design controllers, we want to ensure that the
closed loop system that we design is first of all stable and then it performs as desired. We will
learn about them as we go along.

44
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 06
Mathematics Preliminaries
Part - 2

(Refer Slide Time: 00:15)

So, now let us come back to Laplace transform. So, let us return to Laplace transform.
So, we learned what are the lap, what to say, the, what is the definition of the unilateral
Laplace transform. So now, let us look at some properties, like which are which we are
going to use. The first one is transform of derivatives.

45
(Refer Slide Time: 00:51)

So, first let us see how derivatives of functions transform when we apply the Laplace
transform. So, for example, if F ( s ) is equal to the Laplace transform of or let me write
the other way, let us say if I take a function f ( t ) , if the Laplace transform of f ( t ) in
the time domain is indicated by capital F ( s ) in the complex domain, then the Laplace

df
of is going to be equal to s F ( s )−f (0) .
dt

So, here what is f (0) ? f (0) is nothing but f ( t ) evaluated at t=0, that is the
initial condition. Similarly the Laplace transform of the second derivative of f which is

d2 f 2 df
is going to be s F ( s )−sf (0)− (0) . So, that is the Laplace transform of the
dt 2 dt
second derivative, ok

Now, we have 2 initial conditions right f and f́ , ok. So, as far as this particular
course is concerned if we use dot over a variable, that means it is one derivative with
respect to time, ok. So, that is the notation we are going to use. Similarly, if we use, if we
use 2 dots it is a second derivative and so on.

So, the number of dots over a variable indicates the order of the derivative, right, with
respect to time. So, you see that, as we keep on going to derivatives of higher order we
are going to get as many initial condition terms, right, in the time domain. So, that is the

46
first property, we, which we would be using. The second property is what is called as the
initial value theorem.

The initial value theorem essentially says that lim f ( t ) = lim sF( s) . That is what
t→0 s→∞

was called as an initial value theorem. And there's another theorem which is called as the

final value theorem. So, what is this final value theorem? lim f ( t ) =lim sF (s) . So,
t→∞ s →0

that is the final value theorem.

(Refer Slide Time: 04:10)

Then the 4th property that we would be using is what is called as a complex

−dF ( s )
differentiation theorem. It just states that L ( tf ( t ) ) = . that is like we take the
ds
derivative in the complex domain, right. That is why it is called as the complex
differentiation theorem.

So, then there is something called as the real integration theorem. So, what is this real

F (s)
f ( t )=¿
s
integration theorem? t . So in fact, like if you use tools like MATLAB and
∫¿
0

Simulink, this there is a reason why we see the block s for derivative and 1/s for integral,

47
integrals, right. So, we if you use Simulink, this is the graphical user interfaces that we
would have for a differentiator and integral block, right. So, that is because of this
Laplace transform because when you take derivative we introduce s in the complex
domain and we multiply by s and then when we integrate we divide by s, right. So, that is
why we have these blocks.

Another theorem which we are going to use is multiplication by an exponential function.

And this is, I am just pointing out what are the properties which are useful, this just says
that Laplace of e power minus a t f ( t ) it is going to be equal to F ( s+a ) . So, let us
say if I multiply the function f (t ) by −at
e and take the Laplace transform of the
product, the complex domain wherever I have s, I just substitute s+a, ok.

So, typically we have a table of Laplace transform, ok, like I am just going to write a
table of Laplace transform. I strongly suggest that you derive this table, based on the
definition of the Laplace transform that we have written down.

So, if you have the unit impulse, the Laplace of the unit impulse is one, and if you have

1
the unit step that applies transform the unit step is . If you have the unit ramp that
s

1 −at
Laplace transform the unit ramp is . If you have an exponential function e the
s2

1
Laplace transform of the exponential function is . So, if you have cos ⁡( ωt ) ,
s+ a

( ωt )
cos ¿ .
¿
L¿

ω
If you have sin ⁡( ωt) , Laplace transform is going to be , all right. So, that is
s + ω2
2

what we are going to have. So, this is just a preliminary table of Laplace transform. Of
course, we can add more functions as we go along, but I am just listing whatever we are
going to use very frequently.

48
(Refer Slide Time: 08:10)

So, just to demonstrate a few theorems, right; So, even if I want to demonstrate this,
suppose let us say let f ( t )=cos ⁡( ωt) , then what will happen is that

( ωt )
−at
L( e cos ( ωt ) )=F (s +a) , right. That is, cos ¿ . So, what this theorem tells me is
¿
L¿
that wherever you have s, substitute s+a. So, I am going to substitute s+a, wherever we

s+a
have s, so the Laplace transform is going to be 2 2 , right.
(s +a) +ω

2
ω
Similarly, we have L ( e−at sin ( ωt ) ) = 2 2 .
(s+a) +ω

We will keep on adding to these properties, and let us say some other Laplace transform
as we go further in this course, but from a big picture level what we are going to do is
that we are just going to apply this Laplace transform operator on linear odes with
constant coefficients, ok. So and how do we get these linear odes with constant
coefficients? As we discussed, the typical mathematical models using the modeling
approach that we use for the class of systems that we are considering in this particular
course take the form of linear odes with constant coefficients, right.

49
So, that is our rationale. So, we use a Laplace transform, and then convert them into
algebraic equations, which we can then, analyze, ok, but there is a catch here, right what
is the catch? When we are dealing with linear odes with constant coefficients we are in
the time domain, but that goes to the s domain, right. So, when we apply the Laplace
transform we convert the problem into a problem in the complex domain, right.

And so, here we will shortly see that we need to have a basic understanding of
polynomials to essentially work with the algebraic equations that we get in the s domain,
right. So, that is also something which we will look at as we go along. So, a couple of
more concepts in Laplace transform that we wish, we would recap is that how to take the
inverse Laplace transform, alright.

(Refer Slide Time: 11:20)

So, how do we take the inverse Laplace, right. So, suppose let us say I am given a

1
Laplace transform function which is F ( s )= 2
. so, then what happens? What
s +3 s +2
we need to do is that like we need to what is called as partial fraction expansion and then
like take the inverse Laplace transform, right. So, we see whether we can factorize the
denominator polynomial. In this case, I can factorize s 2 +3 s+ 2 as (s +1)(s +2) , and

A B
this can be factorized as + , ok. So, that is what we have.
s+1 s +2

50
So, then A and B are what are called the residues of the 2 factors, ok. So, how do we find
A and B? Of course, we can use LCM and then like we can essentially equate the
coefficients of the numerator polynomial on both sides and then we can solve. There is
also another method to solve for A and B which I will just teach you.

So, if I want to find A, what I do is that I multiply both sides by the factor s+1 which is
in the denominator. So, this implies that the left hand side will become one divided by

s+1
s+2 the right hand side will just become A, and then I will have B ( ) .
s+ 2

So, what will happen to this? Now, this equation is valid for any value of s right. So,
what I do that is, I substitute the root of the factor that I have multiplied with, ok. So,
what is the root of s+1? A root of a polynomial is, indicates those values of s at which the
polynomial is 0, right. So, that is, that is essentially the meaning of the word root, right.
So, the root of s+1 is going to be s=−1 , right.

So, I substitute s=−1 , and the left hand side becomes 1, the right hand side I have A
and B(s+1) that term vanishes, right? Because s+1 at s=1 is 0. So, that B term vanishes
so, we see that we very easily get the value of A. Similarly, I hope now you get the point,

1
right so, if I want to find B I multiply both sides by s+2, we are left with on the
s+1
left hand side.

s+2
Then I will have A +B and of course, now I need to substitute the root of s+2
s+1
that is going to be a s = -2. So, what I am going to get is minus 1 on the left hand side.
A(s+2) vanishes at s=-2. So, I am just going to be left at B, ok. So, as a result what is

1 1
going to happen is that this partial fraction expansion is going to be − .
s+1 s+2

Now, once we have broken down into blocks that we can easily process, we look at the
table of Laplace transform, right. Immediately we can see that if I look at the table of

1 −t 1
Laplace transforms, the Laplace inverse of I can write as e and I
s+1 s+ 2
can re-write as e−2t , ok. So, that is the way we will do the inverse Laplace transform,
ok.

51
So, this is what is called as partial fraction expansion. So, these quantities A and B are
what are called the residues of the partial fraction expansion process. And one important
application of Laplace transform for us is to solve initial value problems, ok. So, that is
essentially one important application from our course perspective, because we are going
to deal with models that take the form of linear ordinary differential equations with
constant coefficients.

(Refer Slide Time: 15:48)

So, let me consider an example. Let us say I have a system whose governing equation is
governed by ý (t ) +5 ý (t ) +6 y ( t ) =u(t) . Let us say ( 0 )=0 , ý ( 0 )=0 , right. So,
let us consider a system whose governing equation is, ok, is the following. So, let us say,
that is, find y(t) for u(t) =1, right. So, that is what we want. So, the solution what we do
is that like we take Laplace transform on both sides of this particular equation.

So, what happens? So, for ý (t ) you will get s 2 Y ( s )−sy ( 0 ) − ý (0) . Then we will
get 5 ý ( t ) we will get 5( sY ( s )− y ( 0 ) ) plus for 6 y (t ) we will get 6 Y ( s ) and

1
that is going to be equal to U(s), all right. So, if u(t) is 1 this implies that U ( s )= ,
s

1
right. So, we know that the Laplace transform is of one is , right. Given this, let us
s
process this. So, what we are going to get is the following.

52
(Refer Slide Time: 18:15)

So, if I group terms involving Y (s) , let us say, I will have

( s2 +5 s +6 ) Y ( s )−( s+5 ) y ( 0 )− ý ( 0 ) = 1 . So, this will imply that Y (s) =

s

s+5 y ( 0 ) + ý ( 0) U (s )
2
+ 2 . That is what we are going to have. So, immediately we
s +5 s+ 6 s +5 s+6
see that the first term is the response due to initial conditions, right, I am repeating it
again. So, this is what is called as a free response right.

The second term is the output due to the input that we provide, right, due to input
provided. So, that is called as the forced response.

So now we substitute ( 0 )=0 , ý ( 0 )=0 , the free response term vanishes so, the
force response term is one what we are left with.

53
 s
s (¿¿ 2+5 s +6)
So, we will just get . If I factorize for s 2 +5 s+ 6 can rewrite this as
1
¿

s
1
s (¿¿ 2+5 s +6)= .
(s+2)(s+3). So, s(s+2)( s+3) So, this can be rewritten as
1
¿

A B C
+ + .
s s +2 s+3

So, if I want to find A, I multiply by the factor involved in the denominator of A, which

1 Bs Cs
is s. So, =A + + . And here I substitute the root of s which is s =
(s +2)(s +3) s+ 2 s +3
0 right. So, consequently I will get 1/6 on the left hand side and just A, right. Similarly, if
I want to find B I multiply it by the factor the denominator of B which is s + 2. So, I will

1 A( s+ 2) C (s+2)
get = +B+ .
(s )(s +3) s s+3

−1
And here I substitute the root of s + 2 which is s = -2. So, I am going to get =B
2

So, similarly if I want to find C what I will do is that I will essentially multiply both

1 A(s+3) B( s+ 3)
sides by s + 3, the left hand side becomes = + +C . Now, we
(s )(s +2) s s+2

1
will substitute the root of s + 3 and that is - 3. So, we are going to get C= So, that
3
is what we will have.. So, if I take the inverse Laplace transform just by looking at the

1 1 −1
table of Laplace transform, will give me , right and will give me
6s 6 2( s+ 2)

−1 −2 t 1 1 −3t
e . And will give me e . So, that is what I will have, ok.
2 3 (s +3) 3

So, we see that the process of calculating the solution to the ordinary differential
equation becomes pretty simple because we are converting the ode into an algebraic

54
equation and then we process that. So, that is how we are going to use the Laplace
transform for our advantage, right. So, to once again give a big picture perspective we
are dealing with, SISO LTI causal systems which are going to be modeled using spatially
homogeneous dynamic continuous time deterministic models that will typically result in
linear odes with constant coefficients.

And once we get linear odes with constant coefficients we are going to use Laplace
transform and convert them into algebraic equations. And then we calculate the output or
response of the system to various standard inputs and we analyze them, ok. So, that is the
big picture viewpoint. That is why we have studied all the concepts still now.

So, following this we will go further, we will see how we represent the system, and then
we see whatever I call as transfer functions and then how do we use transfer functions in
our analysis, ok. So, we are going to look at the big picture viewpoint following this. But
I am just going to leave you with a few problems that you can do as exercises, ok.

(Refer Slide Time: 25:52)

So, that like we become familiar with the inverse Laplace transform. So, once again I am
sure all of us would have learnt it in engineering mathematics course, this is just to help
you with the recapping all those concepts, right. So, please go back and let us say find
the inverse Laplace transform of the following functions:

55
 s
1 1 1 (¿¿ 2+1)2 1
a¿ 2 b¿ 2 c¿ 2 e¿ 2
s +2 s+2 s +2 s+ 1 s +s s s +2 s−2
d¿ ¿

Please go back and work out the inverse Laplace transform of these functions, ok. And
then like we would continue from here.

Thank you.

56
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 07
Transfer Function
Part – 1

We are dealing with single input single output systems abbreviated as SISO and linear
time invariant causal dynamic systems that are characterized by linear ordinary
differential equations with the constant coefficients.

(Refer Slide Time: 00:28)

We have looked at how one could use the Laplace transform to solve such equations.
What we have done till now is to have a broad overview of what we are going to study in
this course and some mathematical tools that we are going to use.

57
(Refer Slide Time: 00:49)

We will go into the analysis using the classical control approach. One of the main topics
that we are going to learn is that of transfer functions. Let us see what a transfer function
is. Let us consider the mass spring damper system that is governed by

𝑀𝑥(𝑡̈ ) + 𝐶𝑥(𝑡̇ ) + 𝐾𝑥(𝑡) = 𝑓 (𝑡)

This is a linear time invariant system. Let us take the Laplace transform on both sides,
we are going to get

𝑀[𝑠 2 𝑋 (𝑠) − 𝑠𝑥(0) − 𝑥̇ (0)] + 𝐶 [𝑠𝑋(𝑠) − 𝑥(0)] + 𝐾𝑋(𝑠) = 𝐹 (𝑠)

If we just rearrange the terms we are going to get

[𝑀𝑠 2 + Cs + K]𝑋 (𝑠) = [𝑀𝑠 + 𝐶 ]𝑥(0) + 𝑀𝑥̇ (0) + 𝐹 (𝑠)

If we now divide both sides by 𝑀𝑠 2 + Cs + K we are going to get the following.

[𝑀𝑠 + 𝐶 ]𝑥(0) + 𝑀𝑥̇ (0) 𝐹(𝑠)

𝑋 (𝑠 ) = 2
+ 2
[𝑀𝑠 + Cs + K] [𝑀𝑠 + Cs + K]

The first term is the component of the output is due to the non-zero initial condition. One
could immediately observe that it involves 𝑥(0) and 𝑥̇ (0). That is the displacement
which is caused by any non-zero initial condition. If we have the mass initially at rest,
we displace it by a small amount and leave it, it may vibrate depending on the values of

58
K and C and finally, it may settle down back to it is originally equilibrium state. The first
part is the output due to non zero initial conditions. We call it as a free response.

The second part involves the input term 𝐹(𝑠), which is the force. Suppose we apply a
time varying force on the mass, the second term provides us the displacement of the
mass. That is the contribution of the input force to the output. We call it as the forced
response. We are going to have these two components to the output. One due to non-zero
initial conditions and other due to the input that is provided to the system.

(Refer Slide Time: 07:42)

Now, let us consider all initial conditions to be 0. In this case 𝑥 (0) = 0 and 𝑥̇ (0) = 0.
Then

𝐹(𝑠)
𝑋 (𝑠 ) =
[𝑀𝑠 2 + Cs + K]

𝑋 (𝑠 ) 1
= 2
𝐹 (𝑠) [𝑀𝑠 + Cs + K]

𝑋(𝑠)
So, this quantity (𝐹(𝑠) ) which is the ratio of the Laplace transform of the output to the

Laplace transform of the input under 0 initial conditions is called as the transfer function
of the system or plant. We will use 𝑃(𝑠) to denote the transfer function. To summarize,
we wrote down the governing linear ordinary differential equation, then we took Laplace
transform on both sides, we saw that the output term has two components one due to the

59
non-zero initial conditions and one due to the input. We took all initial conditions to be
0. Then, we took the ratio of the Laplace of the output to the Laplace of the input, which
is called as the transfer function of the system or the plant and we are going to denote by
(𝑠). In a certain sense when we use the transfer function approach we are interested in
the forced response of the system because we have taken all initial conditions to be 0.
Now the question becomes will initial conditions be zero all the time in practice? If not
what would happen if we follow this approach? That is a question we would answer
later.

Please remember that, we can take ratios of Laplace of output and input only if we have
a SISO system. If we have a MIMO system, we would have a transfer function matrix.
We can no longer divide a vector output by a vector input. We can still write it in another
product form, where we can say the output vector is going to be equal to a transfer
function matrix multiplying an input vector in the complex domain.

But we are going to deal with single input single output systems in this course. So, the
notations followed holds through for a single input single output system. Consequently,
𝑌 (𝑠 ) = 𝑃 ( 𝑠 )𝑈 (𝑠 ).

(Refer Slide Time: 14:26)

If we take the inverse Laplace transform, we have

𝑡
𝑦(𝑡) = ∫0 𝑝(𝑡 − 𝜏)𝑢(𝜏)𝑑𝜏,

60
𝑦(𝑡) is the output in the time domain if we take the inverse Laplace transform. We can
note certain points. The first point is, if 𝑃(𝑠) is known, we can find the system output for
any input. Another thing we can do is, if we want some output we can find out what
input to give. We can do both control and analysis, what we studied as purpose of
models. Prediction or analysis is where we have an input and we evaluate the output.
Control is where we have a desired output and we find the input to be given. The second
point is, if 𝑢(𝑡) is the unit impulse input, then 𝑈 (𝑠) = 1. Therefore 𝑌(𝑠) = 𝑃(𝑠), this
implies that 𝑦(𝑡) = 𝑝(𝑡)

If you give any input and you calculate the corresponding output, that output is called as
a response and the adjective can be added which corresponds to that specific input. In
this particular case we are providing a unit impulse input to the system. So, consequently
the output that we get is called as the impulse response and 𝑝(𝑡) is the impulse response
function. And the Laplace of the impulse response function 𝑝(𝑡) is the plant transfer
function 𝑃(𝑠).

This gives us a very useful idea, if we want to experimentally determine the plant
transfer function, we just need to give a unit impulse input and then measure what is the
output and that will be the impulse response function. From the impulse response
function, we can calculate the plant transfer function. The advantage is we need to do
only one experiment, but the catch here is, in practice it is very difficult to give an ideal
impulse input. Then the question becomes, what approximations can we get? Is there any
other method to figure out the transfer functions? We will learn as we progress. There is
another method by which we can experimentally determine transfer functions by using
frequency response, ok. So, we would learn that technique also. At the end of the day, if
we want to figure out a plant transfer function it is going to be a mix of both approaches.
For example, it is going to be a mix of deriving equations based on physics and also
performing experiments and using experimental data to fit the transfer function, we need
both approaches.

As far as the definition of the transfer function is concerned it is extremely important that
we keep all initial conditions to be 0, then only we can talk about transfer functions and
use them.

61
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 08
Transfer Function
Part - 2

(Refer Slide Time: 00:16)

We give an input, let us say, you have a system, right, like a black box; let us say you
have a signal generator; you give an ideal impulse input, let us say you have a data
logger at the system output terminal, right. You log the output, you get the output signal,
you have the measurements. So, that is the impulse response, provided you can give an
ideal impulse input, right.

Then you use the data to figure out the transfer function. What we will do is that, like,
when you come towards the end of the semester, I will give you a project where we will
give you real experimental data and you need to figure out the plant transfer function, ok.
We will do a small project which will essentially help you in learning how this can be
done, all these aspects can be done ok.

So, now I am going to tell you something in general. In general, the governing equation
of what we call as a nth order, see the mass spring damper system was a second order
system right, because the highest derivative of the output term was order 2, right.

62
So, in general the governing equation of an nth order SISO LTI causal system can be
written as the following.

𝑑 𝑛 𝑦(𝑡) 𝑑 𝑛−1 𝑦(𝑡) 𝑑𝑦(𝑡)

𝑎0 + 𝑎1 + ⋯ + 𝑎 𝑛−1 + 𝑎𝑛 𝑦(𝑡)
𝑑𝑡 𝑛 𝑑𝑡 𝑛−1 𝑑𝑡
𝑑 𝑚 𝑢(𝑡) 𝑑 𝑚−1 𝑢(𝑡) 𝑑𝑢(𝑡)
= 𝑏0 + 𝑏1 + ⋯ + 𝑏𝑚−1 + 𝑏𝑚 𝑢(𝑡)
𝑑𝑡 𝑚 𝑑𝑡 𝑚−1 𝑑𝑡

So, this, this is how in general the governing equation for the class of systems that we are
going to study would come up, ok. So, please note that this class of systems is modeled
by using a linear ode with a constant coefficient. So, immediately you can see that it is
linear in y and u, ok. It has constant coefficients that is reflective of time invariance. So,
let me write down all the implications, because just by looking at the equation you
should be able to tell things about the system, right.

So, time invariance is reflected by the fact that we have constant coefficients in the
above equation. The fact that it is linear is because the above equation is linear in u(t)
and y(t) and the derivatives of these variables. Causality is reflected by the fact that n is
greater than or equal to m, ok. When n is greater than m some people will call it a strictly
causal. So, I am just introducing you to some terms which you may encounter, right. So,
essentially causality is reflected by the fact that n is greater than or equal to m, ok.

So, that is how causality is reflected in this particular equation, ok. So, this is an nth order
system, where n is the order of the highest derivative of the output term, ok. So, as an
example, for the mass spring damper system, what were the values of n and m?

If you look at the mass spring damper system, the input was the force f(t) right, the
output was the displacement x(t) correct. So, what, what was the value of n.

Student: 2.

2, because we had 𝑥̈ . What about m?

Student: 0.

0, because there was no derivative of the out, force term at all right. On the right hand
side we had just f(t) right. So, immediately you see that n is greater than m, so that was a
causal system, right. So, that is a representation of a causal system; that means, that

63
before you give a force, the mass is not going to get displaced, right. So, only the current
displacement is going to depend on past and current force that you are giving to the
system right. So, that is how the different aspects of LTI causality are reflected in the
governing equation.

Now, this is something I am going to leave it to you as a homework problem, because we

did it in detail for the mass spring damper system, this is just a general case, ok. So, take
the Laplace transform on both sides and apply zero initial conditions to obtain

𝑌(𝑠)
𝑃(𝑠) = 𝑈(𝑠), the plant transfer function, which will turn out as

𝑏0 𝑠 𝑚 + 𝑏1 𝑠 𝑚−1 + ⋯ + 𝑏𝑚−1 𝑠 + 𝑏𝑚
𝑃 (𝑠 ) =
𝑎0 𝑠 𝑛 + 𝑎1 𝑠 𝑛−1 + ⋯ + 𝑎𝑛−1 𝑠 + 𝑎𝑛

So, this I leave it to you as homework, ok.

So, once again pretty straightforward, you just take the Laplace transform on both sides.
If you apply all initial conditions to be zero, when you take the Laplace transform of the
nth derivative the only term which will remain is 𝑠 𝑛 𝑌(𝑠). Similarly when you take the
Laplace transform the n-1 derivative of y, if you apply all the initial conditions to be zero
the term which you will be left with is 𝑠 𝑛−1 𝑌(𝑠).

So, that is what you will be left with. Same story in the right hand side right, if you take
the Laplace transform of the m th derivative of the input, you will have 𝑠 𝑚 𝑈(𝑠). plus a
bunch of terms having initial conditions of the input.

Even that you take as zero, so you will be just left with 𝑠 𝑚 𝑈(𝑠), so you cross multiply.
This is what you would get ok. So, that is the plant transfer function in this particular
case. So, that is what we have as the plant transfer function. So, immediately we will see,
what really is the consequence, right. So, let us look at it further. So, immediately one
can observe that, the plant transfer function is a ratio of two polynomials in s, right.

64
(Refer Slide Time: 09:30)

So, one could say that P of s is a ratio of two polynomials n(s) by d(s), right. So, n(s) is
the numerator polynomial of order m.

So, similarly d(s) is a denominator polynomial and that is of order n. So, we see that the
plant transfer function in general is going to be a ratio of two polynomials in s; a
numerator polynomial and a denominator polynomial. So, for when n is greater than or
equal to m for a causal system, we are going to have what is called as a proper transfer
function, ok.

Transfer function is said to be proper when the order of the denominator polynomial is
greater than or equal to the order of the numerator polynomial, ok. And in this case
anyway we are dealing with causal systems, you know that is going to be true, ok.

So, for the class of systems that we are going to deal with, we are going to get proper
transfer functions. So, when n is strictly greater than m, we add the term strictly ok. We
are going to get what are called, what is called as a strictly proper transfer function ok,
that is what is going to result when n is strictly greater than m, ok.

So, that is essentially the definition of what is called as a proper transfer function and a
𝑛(𝑠)
strictly proper transfer function. So, now we see that 𝑃(𝑠) = 𝑑(𝑠), right. So, we see

immediately that, if I calculate the roots of n(s), that is solve 𝑏0 𝑠 𝑚 + 𝑏1 𝑠 𝑚−1 + ⋯ +

𝑏𝑚−1 𝑠 + 𝑏𝑚 , the roots are called as, what would you call the roots of n of s as? So if

65
you solve this equation the numerator polynomial equals zero and you get particular
values of s, at those values of s what will happen to the transfer function, it will become
zero right, because the numerator becomes zero. So, the roots of n(s) or what are called
as the zeros of the transfer function. I hope it is clear why they are called zeros right, that
transfer function vanishes right.

So, similarly the roots of the polynomial P(s) that is we solved 𝑎0 𝑠 𝑛 + 𝑎1 𝑠 𝑛−1 + ⋯ +
𝑎𝑛−1 𝑠 + 𝑎𝑛 = 0. So, wherever I have the denominator polynomial to be zero, you can
immediately see that the transfer function becomes singular, right. So, the transfer
function is a complex valued function in the variable s.

Now the point is that if I have a complex function f(s) and in the s domain at certain
points the function does not exist and its derivatives do not exist, those values of s are
called as singular points or poles right. We discussed it sometime before when we briefly
discuss complex variables. So, consequently the roots of d(s) are called as the poles of
the transfer function, ok.

Student: So, can we (Refer Time:15:04) y n equal to n greater than (Refer Time;15:07)
causality.

Causality ok so, let me give you a very simple example right. So, just to convey that
point ok, let us let us do it aside, right. So, the question was how one could say that n
greater than or equal to m implies causality. Let me go the other way ok, I will show you
how m greater than n will mean non causality ok, let us let us do that way ok.

66
(Refer Slide Time: 15:37)

𝑑𝑢(𝑡)
So, let us take a very simple case where 𝑦(𝑡) = right. So, what are the values of n
𝑑𝑡

and m here? n is going to be.

Student: 0.

0, m is going to be 1 all right. Do you agree? Ok. So, here n is less than m. So, we have
gone to the other side right. So, now, how is it anticipating, a non causal system is
anticipatory. Why? Because what does the derivative do.

See, suppose just to convey it graphically so that way we have a better idea right. So, let
us say I plot u(t) versus t right, I give some arbitrary input, but let us say I am at this
instant of time, this is my current instant of time t. So, what does my derivative give me?
It is a rate of change, graphically it will give me the slope right and the value of the
slope.

So, in a certain sense I will know what is going to come up in the future. So I am basing
my output as the derivative of the input signal which will tell me what is going to come
in the future. So, my system output is, at this instant of time, is influenced by what input
will come in the next instant. So, it tries to anticipate right and take action. So, you can
see that how its non causal, right.

67
 𝑑𝑢
If let us say we are at this point and if is positive the output will increase, right. If I am
𝑑𝑡

at this point where u(t) has reached what to say maxima, local maxima then the
derivative will be zero, so output will be zero right. So, in a certain sense it tries to
anticipate what is happening that is why it becomes non causal. So, now, I hope you have
an intuitive understanding of why this implies anticipation, right.

So, what I will do is that I am going to leave you with a few problems to do, right. And
then we will come back in the next class and essentially discuss further, right. So, let us
look at a few exercise problems, right. So, what I am going to do is that I am going to
essentially write down governing ordinary differential equations for a few simple LTI
systems right, and then you need to calculate the transfer function and get the unit step
response, ok.

So, that is the exercise I am going to give you ok, or let us make life simpler, let us say
we calculate the unit impulse response. So, you just need to take the inverse Laplace
transform of the transfer function that is all I want, right.

So, in each case determine the plant transfer function, its poles and zeroes and calculate
its unit impulse response. So, let me leave you with a few problems. Of course, the
moment I say transfer function, obviously we should assume all initial conditions to be
zero all right, ok.

1) 𝑦̈ (𝑡) + 5𝑦̇ (𝑡) + 6𝑦(𝑡) = 𝑢(𝑡)

2) 𝑦̈ (𝑡) + 𝑦(𝑡) = 𝑢(𝑡)
3) 𝑦̈ (𝑡) + 𝑦̇ (𝑡) − 2𝑦(𝑡) = 𝑢(𝑡)

So, derive the transfer function, find the poles, zeroes and calculate the unit impulse
response ok, we would start from here in the next class.

68
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 09
System Response
Part – 1

In the last lecture, we were looking at the concept of transfer function. If we have a linear
ODE with constant coefficient that corresponds to a linear time invariant causal SISO
system. We could apply the Laplace transform, take all initial conditions as 0 and then
get the ratio of the Laplace of the output to the Laplace of the input. Once we do that the
quantity that we get is called as a transfer function of the system. And we defined what
are called as poles and zeros of the transfer function. We are going to discuss the
solution to few exercise problems and make a few observations.

(Refer Slide Time: 01:12)

Let us take the first problem. This is the system, whose governing equation is given by

𝑦̈ (𝑡) + 5𝑦̇ (𝑡) + 6𝑦(𝑡) = 𝑢(𝑡)

This is a second order system. If we take the Laplace transform on both sides, we get

𝑠 2 𝑌(𝑠) − 𝑠𝑦(0) − 𝑦̇ (0) + 5[𝑠𝑌(𝑠) − 𝑦(0)] + 6𝑌(𝑠) = 𝑈(𝑠)

69
We collect terms and we will get

[𝑠 2 + 5𝑠 + 6]𝑌(𝑠) = [𝑠𝑦(0) + 5𝑦(0) + 𝑦̇ (0)] + 𝑈 (𝑠)

𝑠𝑦(0) + 5𝑦(0) + 𝑦̇ (0) 𝑈 (𝑠 )

𝑌 (𝑠 ) = +
𝑠 2 + 5𝑠 + 6 𝑠 2 + 5𝑠 + 6

The output term has two components, the first component is due to the initial conditions,
that is what we call as the free response. And the second part is due to the input that is
what is called as forced response of the system. So, now when we want to get the transfer
function, we take all the initial conditions as 0.

(Refer Slide Time: 05:13)

That is, in this particular problem we take 𝑦(0) = 0 and 𝑦̇ (0) = 0. Once we have that,
the first term is going to vanish.

𝑈 (𝑠 )
𝑌 (𝑠 ) =
𝑠 2 + 5𝑠 + 6

𝑌 (𝑠 ) 1
= 2 = 𝑃(𝑠)
𝑈(𝑠) 𝑠 + 5𝑠 + 6

This is going to be the plant transfer function or the system transfer function 𝑃(𝑠). Here
we see that the order of the denominator polynomial of the transfer function 𝑛 = 2 and
the order of the numerator polynomial of the transfer function 𝑚 = 0. 𝑛 > 𝑚 so we have

70
a strictly proper transfer function. And we solve the denominator polynomial equal 0 to
get the poles

𝑠 2 + 5𝑠 + 6 = 0

The poles are -2 and -3 and there are no zeroes for this problem. And typically what we
do is that we plot what is called as an s plane (the complex plane) with the real and the
imaginary axis. The imaginary axis divides the complex plane into two halves called as
the left half plane abbreviated as LHP and right half plane abbreviated as RHP. The
poles are denoted by a cross sign (x) in the complex plane. We locate -2 and -3 on the
negative real axis. The zeros are graphically denoted by a small circle (o) in the complex
plane. There are no zeros in the above problem.

We know that the order of the denominator polynomial is going to be the same as the
order of the system. If we look at the transfer function of the system. The order of the
transfer function is 2, because the denominator polynomial is order 2.

(Refer Slide Time: 10:49)

Now let us calculate the unit step response of the system.

𝑌(𝑠) = 𝑃(𝑠)𝑈(𝑠)

1
𝑌 (𝑠 ) = 𝑈 (𝑠 )
𝑠2 + 5𝑠 + 6

71
 1
For a unit step response 𝑢(𝑡) = 1 and 𝑈(𝑠) = 𝑠 .

We can write

1 1 1 𝐴 𝐵 𝐶
𝑌 (𝑠 ) = = = + +
𝑠2 + 5𝑠 + 6 𝑠 𝑠(𝑠 + 2)(𝑠 + 3) 𝑠 (𝑠 + 2) (𝑠 + 3)

1 1
A, B, C are called the residues. Evaluating the partial fractions, we get 𝐴 = , 𝐵 = − ,
6 2
1
𝐶 = 3.

1 1 1
𝑌 (𝑠 ) = − + .
6𝑠 2(𝑠 + 2) 3(𝑠 + 3)

If we take the inverse Laplace transform

1 1 −2𝑡 1 −3𝑡
𝑦 (𝑡 ) = − 𝑒 + 𝑒
6 2 3

This is the unit step response of the system. Suppose instead of giving 𝑢(𝑡) = 1, we give
𝑢(𝑡) = 5, how will this output change? We can use the property of homogeneity and say
the output (unit step response) gets multiplied by 5.

Let us write down what we observe from this unit step response.

1
1) As,𝑡 → ∞, 𝑦(𝑡) → 6, this is called as the steady state value.

2) Initial value of 𝑦(𝑡), if we substitute to 𝑡 = 0, we get 𝑦(𝑡) = 0, because anyway

we assume all the initial conditions to be 0.
3) What can we say about the exponents of the exponential solution? The exponents
here are -2 and -3 which happened to be the poles of the system.

In this problem the poles happen to be real, but in general poles and zeros can be
complex conjugate pairs. We can have complex numbers as poles and zeros but we need
to have, the conjugate also as a corresponding pole or zero, because we are dealing with
dynamic systems where the governing ordinary differential equation has real numbers as
coefficients. Consequently, when we take that Laplace transform, we get a polynomial in
whose coefficients are real. So, once we have a polynomial with real coefficients, we
could have a complex root, but that conjugate must also be a root

72
The exponents of the exponential terms in general would be the real part of the poles. In
this particular example -2 and -3 are real numbers, the complex part was 0. That is why
we have the terms as 𝑒 −2𝑡 and 𝑒 −3𝑡 .

(Refer Slide Time: 24:26)

4) What about BIBO stability here? Step is a bounded input, and the corresponding
𝑦(𝑡) that we have calculated is bounded. The observation we can make here is
that the system is bounded input bounded output stable.

But then it should be bounded for all possible bounded inputs. But that is a
generalization which we are making here, but we will prove it more carefully later on
when we look at stability.

73
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 10
System Response
Part - 2

So, let me do the second exercise, ok. So, let me go back and figure out what was the
second exercise. So, I gave 𝑦̈ (𝑡) + 𝑦(𝑡) = 𝑢(𝑡) right.

(Refer Slide Time: 00:23)

(Refer Slide Time: 00:29)

74
So, let us do this problem 𝑦̈ (𝑡) + 𝑦(𝑡) = 𝑢(𝑡). So, that is the second problem. So, once
again you can see that it is a second order system and if you take the Laplace transform
and apply zero initial conditions what are we going to get?

So, let us follow the same steps. So, take the Laplace transform on both sides. Of course,
I am going to skip a few steps, because by now you know what to do right. Once again
please write down the complete transform of the derivatives; do not substitute the initial
conditions to be 0 in the first step itself because you never know, ok. In some problem
initial conditions may not be zero. So, we need to be a bit careful.

So, always write down the entire expansion. Now in order to find the transfer function
take all initial conditions to be 0 right, because that is a definition of the term transfer
𝑌(𝑠) 1
function, right. So, this implies = 𝑠 2+1. That is what we will get, if I take all initial
𝑈(𝑠)

conditions to be 0 and I collect terms involving Y(s) and cross multiply, this would be
the plant transfer function P(s).

So, now what can I comment, what are the values of n and m? n is 2, second order
system, m is 0. What about poles?

Student: (Refer Time: 02:28).

You solve for 𝑠 2 + 1 = 0. So, this implies that s is going to be equal to of course, we are
going to use j, you could use i also, but as I told you we are going to use j in this
particular course, ok. So, we will essentially say it is ±j, so zeros, there are none once
again. So, if you plot the s plane and mark these two poles, where are they going to be?
They are going to be at +j and –j. So, those are the two poles for this particular system,
yes.

Student: (Refer Time: 03:18).

Which one?

Student: S plus j a real part may be (Refer Time: 03:23).

Real part is 0. See 𝑠 2 + 1 means s = ±j. So, your real component of that solution is going
to be 0 right, isn’t it? See if you substitute any real component if you say C ± j, where C
is any real number can you satisfy this equation? Please try ok, you cannot right.

75
So, that is why the solution is, we take it to be ±j right ok. So, we can we can try it out
right, let us see. Since this question has been raised, let me also work it out ok. So, let us
say we take it as C ± j, is that your point?

So, C being any real number, let us try it out. So, if I have C ± j, then

(𝑐 + 𝑗)2 + 1 = 0

𝑐 2 + 2𝑐𝑗 + 𝑗 2 + 1 = 0

𝑐 2 + 2𝑐𝑗 = 0

Implies, 𝑐 = 0.

So, this will be satisfied only when C is identically equal to 0 correct. So, that is why we
are taking plus or minus right. So, the real part is anyway 0 ok. Of course, let me put it in
a box. So, that like, whatever I am drawing like this you know like our asides right. So,
that is ±j right, those are the two roots for this one right. So, what we are having, is that
like we are having what are called as a complex conjugate poles right. We are essentially
having a purely imaginary poles here, correct. So, what do we do for the unit step
response?

So, we need to get the unit step response. How do we get this? Unit step response means
1
𝑈(𝑠) = 𝑠 . So, then 𝑌(𝑠) = 𝑃(𝑠)𝑈(𝑠). See this is the advantage; once we find that

transfer function we can find the system output for any given input right. So, 𝑌(𝑠) =
1
𝑠(𝑠 2+1)
. How can I simplify this?.

Student: (Refer Time: 06:12) p by s; P s by (Refer Time: 06:15).

1 𝐴 𝐵𝑠 + 𝐶
= + 2
𝑠(𝑠 2 + 1) 𝑠 𝑠 + 1

So, immediately we see A is going to be 1 right, and if you solve for B and C what are
you going to get?

Student: (Refer Time: 06:37) minus 1.

76
 1 𝑠
Minus 1 and 0; so we will have essentially − 𝑠 2+1. I am sure by now if we have done a
𝑠

few examples, by now I am sure all of you will be familiar with how to calculate these
residues right. I think last week we did a few cases right, where I showed you how to do
even for second order terms and so on right.

So, once we get this, this implies that y(t)=1-cos(t). Do you agree? Ok. So, that is my
step response. So, but then like if we want to make the same observations as before right.
So, what happens here, let us say I want to make the same observations as before right.
So, what happens as t→∞?

Student: (Refer Time: 07:59).

Of course, essentially it oscillates, but is it bounded?

Student: (Refer Time: 08:07).

So the magnitude of y(t) remains bounded, right. So, that is something which we can see.
So, using this; See once again step is a bounded input right is it not? So we are getting a
bounded output; so from this would you conclude that the system is BIBO stable.

Student: No.

No, why not.

Student: (Refer Time: 08:36).

We do not know about all inputs right, so that is something which we need to check out.
And by the way, where are the exponential terms here?

Student: (Refer Time: 08:48).

You could use Euler’s relation. So, please note that the general viewpoint even still
works out right. So, see I wrote that real part of the poles would appear ok.

77
(Refer Slide Time: 09:04)

Some people will write the poles would appear. So, that is why I just wanted to make a
distinction ok, because see if I write poles would appear, yes that is also correct, but then
you see that in this particular problem the real part is 0, so you do not need to necessarily
have exponential solutions appearing all the time, its implicit, its contained within the
cos and the sine. So, we could say that exponential terms will have the real part as the
exponent. If you have complex conjugate you will get cos and sine. You could also write
𝑒 𝑗𝑡 +𝑒 −𝑗𝑡
this as 𝑦(𝑡) = 1 − ( )
2

So, here omega is 1. So, you see that now the poles are appearing as, once again we are
going to have two exponential terms corresponding to two poles right, the exponents are
the poles themselves, right. So both interpretations I am just writing this way right, both
interpretations could be derived out of this exercise, ok. So, you see that the poles are
there right as the exponential terms.

So, but if you have complex exponentials you know you can write as cos and sin, right
using the Euler’s relation. So, that is why I told in general we could also look at for our
purpose, we could look at the real parts being the exponents, right. In this particular term
1 - cos t, you do not have an explicit exponential term, because the real parts of the poles
are 0 ok, so that is why you have this. Is it clear?

78
Now what I want you to do, remaining on this problem, we will go to the next problem
shortly, I am just renumbering, so please do this. So, I just changed only one thing right.
So, instead of 𝑦̈ (𝑡) + 𝑦(𝑡) = 𝑢(𝑡), I just changed the transfer function to be 𝑦̈ (𝑡) +
𝑦̇ (𝑡) = 𝑢(𝑡); that is the difference between problems 2 and 3, alright.

So, now, can you repeat the same process? So, what would be the plant transfer
function? Please take the Laplace transform on both sides, take initial conditions to be 0
1
get the ratio of Y(s) and U(s), what would you get? You would get . So, this I can
𝑠 2+𝑠
1
rewrite as 𝑠(𝑠+1).

So, similarly I can say n = 2, m = 0. What are the poles? The poles are at 0 and - 1
correct and there are no zeros, ok. So, consequently what will happen, I will have this ok.
So, where are my poles here in this example? In the s plane I have a pole of the origin, I
have a pole at minus 1 ok.

So, these are my two poles. So, please find the unit step response and let me know what
do you get?

Student: Sir.

Yes.

Student: (Refer Time: 13:02).

Is it?

Student: It should be because it, (Refer Time: 13:14).

So, to answer that I am going to generalize by doing this problem ok, let us please hold
on, ok. The question was that like I neither said that its BIBO stable nor said its unstable
right, I only said we need to do further analysis.

So, but to give you the answer it is a tricky question; it depends on whom you talk to that
is why I am doing things carefully step by step ok. So, unit step response please, so, unit
1
step response means 𝑈(𝑠) = 𝑠 .

79
(Refer Slide Time: 13:55)

So,

1 𝐴 𝐵 𝐶
𝑌 (𝑠 ) = 𝑃 (𝑠 )𝑈 (𝑠 ) = = + 2+
𝑠 2 (𝑠+ 1) 𝑠 𝑠 𝑠+1

So, what are A, B and C?

Student: (Refer Time: 14:28).

C, I can get as 1 right, B I can get once again as 1, am I correct? So, A you can substitute
any other value you want and if you simplify what will you get? Minus 1, is that correct.
So 𝑦(𝑡) = −1 + 𝑡 + 𝑒 −𝑡 .

So, this is the unit step response ok. Previously also there is a unit step response right.
So, now, what can we observe?. So, what do you think happens to the magnitude of y(t)
as t tends to infinity, what happens to its magnitude?

Student: (Refer Time: 15:52).

It tends to infinity why right, because I have a t term right. So, once again you see that,
the solution has exponential terms the poles are the exponents right. So, the poles are at 0
and minus 1. So, what is 𝑒 0 ?

Student: 1.

80
It is just like what to say 1 right, so that is why you have a term like that and then you
also have 𝑒 −𝑡 − 1 also comes in right, but in addition you have a t term ok, which
essentially tells you that the magnitude of the output is going to go to infinity right, as
the time tends to infinity. So, would you call the system as BIBO stable? Yes or no.

Student: (Refer Time: 16:38).

See what is BIBO stable right?

Student: (Refer Time: 16:40).

So, let me recap the definition very carefully ok, each and every word is very important.
A system is said to be BIBO stable, if the output remains bounded in magnitude for all
time given any bounded input, ok, the key word is any or all bounded inputs right. So,
here if you want to show something is not BIBO stable, you just need to find one
bounded input for which the output is unbounded, is it not. See in fact, in life showing
that something does not work or is not true is easier right, because you just need to have
only one case, where it does not hold true right and for this particular system have we not
found one bounded input for which the output is unbounded, we have right.

So, for this particular system we have found a step input for which the output is
unbounded ok. Now the question is would you call this as unstable or something else?
We will see what that ‘something else’ is in the afternoon ok, so after doing a few more
examples ok. So, just hold on ok. So, we would see that you know like, there are some
more examples that we need to do before we can generalize ok.

So, I hope the flow is clear right. So, as the fourth example let us do this problem ok. So,
𝑦̈ (𝑡) + 𝑦̇ (𝑡) − 2𝑦(𝑡) = 𝑢(𝑡), alright.

So, what is the plant transfer function P(s) here?

Student: (Refer Time: 18:29).

𝑌(𝑠) 1 1
𝑃 (𝑠 ) = = 2 =
𝑈(𝑠) 𝑠 + 𝑠 − 2 (𝑠 + 2)(𝑠 − 1)

I am assuming that all of you can follow with me. Please stop me if anything is unclear
right. So, this I can rewrite it as 1 divided by s plus 2 times s minus 1. Am I correct? Is

81
that correct? So, once again we can see that n = 2, m = 0, the poles are at -2 and +1.
Zeros, there are none.

So, if I draw the s plane for this particular example the poles are going to be at minus 2
and plus 1 ok. So, this is the left of complex plane and this is the right of plane ok. So,
what is the step response, unit step response?

1 𝐴 𝐵 𝐶
𝑌 (𝑠 ) = 𝑃 (𝑠 )𝑈 (𝑠 ) = = + +
𝑠(𝑠 + 2)(𝑠 − 1) 𝑠 𝑠 + 2 𝑠 − 1

−1 1 1
𝐴= ,𝐵 = ,𝐶 =
2 6 3

−1 1 1
So, those are the 3 partial fractions. So, this implies that 𝑦(𝑡) = + 6 𝑒 −2𝑡 + 3 𝑒 𝑡
2

(Refer Slide Time: 21:24)

So, now we immediately note that as t tends to infinity what happens to the output y(t).

Student: (Refer Time: 21:35).

We can immediately see that once again there are two exponential terms: the poles
appear as exponent or the real parts appear as an exponent, you know because anyway
the poles are real here.

82
So, you could see that -2 and + 1 are the two poles, you could see them appearing as
exponents, but then you see that as t→∞, the magnitude tends to infinity why, because
we have et all right, the magnitude of et tends to infinity ok, so that is why this blows off
right to infinity. Now the question is given this unit step response, you know like with
this system be BIBO stable or unstable right.

So, once again you know like, we would figure out the answer right. So, what we are
essentially getting at is the following. So, before you come to afternoon’s class, you
know like anyway you have time right. So, I want you to do this example ok. Take
problem number 2, 𝑦̈ (𝑡) + 𝑦(𝑡) = 𝑢(𝑡), please take that problem ok.

So, as problem 5 consider 𝑦̈ (𝑡) + 𝑦(𝑡) = 𝑢(𝑡) ok; now find y(t) when u(t)=cos(t).

Similarly the 6th problem I am going to give you this, take the third problem 𝑦̈ (𝑡) +
𝑦̇ (𝑡) = 𝑢(𝑡), find y(t) once again when u(t) = cos(t), ok.

So, please do these problems, and you come and tell me what you observed and then we
would continue our discussions from here in the next class right.

So, I hope it is clear what I want you to do right. Take problems 2 and 3. For both
problems we figured out the unit step response right. Now I want you to figure out what
is the response of the system or the output of the system when we give a cosine as the
input to the particular system. Do the same exercise, take inverse Laplace transform, get
y(t) and then we will discuss, what are the consequences thereafter.

83
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 11
BIBO Stability
Part – 1

Let us solve some problems and make some observations and do a general derivation as
far as BIBO stability is concerned. Let us start with problem number 5.

(Refer Slide Time: 01:27)

From the previous discussion, we know that the plant transfer function is

𝑌(𝑠) 1
𝑃 (𝑠 ) = = 2 .
𝑈(𝑠) 𝑠 + 1

System poles at ±𝑗, they are purely imaginary poles. And we saw that when it was
subjected to a unit step input, the output was bounded in magnitude.

Now, let us subject it to a cosine input of a very specific angular frequency 1 rad/s.

In general cos 𝜔𝑡 means 𝜔 is the angular frequency in radians per second. cos 𝑡 means
𝑠
𝜔 = 1 rad/s. If 𝑢(𝑡) = cos 𝑡, then 𝑈 (𝑠) = 𝑠 2 +1.

84
 1 𝑠 𝑠
𝑌 (𝑠 ) = 𝑃 (𝑠 ) 𝑈 (𝑠 ) = = 2
𝑠2 + 1 𝑠 + 1 (𝑠 + 1)2
2

Now, we use a property of Laplace transform, the complex differentiation theorem.

𝑑𝐹(𝑠)
𝐿[𝑡𝑓(𝑡)] = −
𝑑𝑠

Let us take 𝑓 (𝑡) = sin 𝑡,

1
𝐹 (𝑠 ) =
𝑠2 + 1

𝑑𝐹(𝑠) 2𝑠
− = 2 ,
𝑑𝑠 ( 𝑠 + 1)2

1
𝑦(𝑡) = 𝑡 sin 𝑡
2

We can observe that, as 𝑡 → ∞, 𝑦(𝑡) → ∞.

(Refer Slide Time: 05:22)

Now, let us solve problem number 6. We already know that the plant transfer function is

𝑌(𝑠) 1 1
𝑃(𝑠) = 𝑈(𝑠) = 𝑠 2+𝑠 = 𝑠(𝑠+1).

𝑠
𝑢(𝑡) = cos 𝑡, then 𝑈 (𝑠) = 𝑠 2+1.

85
 𝑠 1 𝐴 𝐵𝑠 + 𝐶
𝑌 (𝑠 ) = 𝑃 (𝑠 ) 𝑈 (𝑠 ) = 2
= 2
= + 2
𝑠(𝑠 + 1)(𝑠 + 1) (𝑠 + 1)(𝑠 + 1) 𝑠 + 1 𝑠 + 1

Here the cancellation of 𝑠 in the numerator and denominator was a mathematical

operation which came in due to the specific nature of the input that we have, but in
general we have a more subtle idea which is called as pole zero cancellation. We will
discuss that when we go deeper into control system design.

1 1 1
Solving the partial fractions, we get 𝐴 = 2, 𝐵 = − 2 and 𝐶 = 2.

1 1 1 𝑠 1 1
𝑌 (𝑠 ) = − 2 + 2 .
2𝑠 + 1 2𝑠 + 1 2𝑠 + 1

Taking inverse Laplace, we have

1 1 1
𝑦(𝑡) = 𝑒 −𝑡 − cos 𝑡 + sin 𝑡.
2 2 2

We can observe that, as 𝑡 → ∞, 𝑦(𝑡) is bounded.

Earlier we saw that, for the same system when we give the unit step response, 𝑦(𝑡) → ∞.
Now we gave another bounded input cos 𝑡, we are getting a bounded output. Let us
construct a simple table.

System Poles Input Output

S1 ±𝑗 1 Bounded as 𝑡 → ∞

𝑦̈ (𝑡) + 𝑦(𝑡) = 𝑢(𝑡)

cos 𝑡 Unbounded as 𝑡 → ∞

S2 0,-1 1 Unbounded as 𝑡 → ∞

𝑦̈ (𝑡) + 𝑦̇ (𝑡) = 𝑢(𝑡)

cos 𝑡 Bounded as 𝑡 → ∞

86
It so turns out that if we use the definition of BIBO stability which states that a system is
BIBO stable if its output is bounded for all possible bounded inputs, we should brand
these two systems as unstable.

(Refer Slide Time: 14:14)

But, for the first system (S1) the output is unbounded, only if we give a sin 𝑡 or cos 𝑡 as
the input. For unit step input or any other bounded input like a sin or a cosine of any
angular frequency other than 1 rad/s, we can show that the output is going to be bounded.
It so turns out that only when the input becomes a sinusoid whose angular frequency is
equal to the magnitude of the poles on the imaginary axis, the output becomes
unbounded. For S1 the magnitude of poles on imaginary axis is 1. Some people will call
this class of systems as marginally stable or critically stable; that is the output is bounded
for all possible bounded inputs except one set. In strict parlance the system is unstable if
we use the definition as it is. We do not want to design such systems in practice. If we
encounter such systems we want to stabilize them.

If we have a mass spring damper system where there is no damper, we just have a mass
and the spring, the equation of motion will become 𝑀𝑥̈ + 𝑘𝑥 = 𝑓, the moment we give a
step input, the output will become unbounded.

Now, on the other hand if you look at system S2, it has a pole at the origin. If we have a
non-repeating pole at the origin you give a step input, the output is unbounded. But for
all other sinusoidal inputs, the output is bounded. That is the property of S2. We can see

87
that the output is unbounded only for one set of inputs. This result is true only if we have
for non repeating poles on the imaginary axis.

If you have a repeating pole on the imaginary axis. For example.

1
𝑃 (𝑠 ) =
𝑠 2 (𝑠 + 1)

(Refer Slide Time: 21:53)

The poles are at 0,0,-1, with the repeating poles this system is unstable.

Similarly, if you have another system where the transfer function is

1
𝑃 (𝑠 ) =
(𝑠 2 + 1)2

The poles are +𝑗, +𝑗, −𝑗, −𝑗, this is unstable. The above concept of marginal stability or
critical stability applies, only if we have non repeating poles on the imaginary axis.

1. For BIBO stability, all poles of the plant transfer function must lie in the LHP or
in other words have negative real parts.
2. If there exists even one pole of the plant transfer function in the right half plane,
that is have has a positive real part, then the plant or the system is not BIBO
stable.

88
 3. If there are repeating poles of the plant transfer function on the imaginary axis (j
omega axis) with all remaining poles in the LHP, then system is not BIBO stable.
4. If there are non repeating poles of the plant transfer function on the imaginary
axis with all remaining poles in the LHP then the system is unstable or critically
stable.

Most textbook we will say it is critically or marginally stable, but if we strictly apply the
definition of BIBO stability it is unstable. If we are strict to that extent we should call
this as unstable, but if we can say the output is bounded only for one small group of
bounded inputs so we call it as critically stable or marginally stable.

That is basically some perspectives of how to interpret system stability when you have
poles, non-repeating poles on the imaginary axis, including the origin.

In condition number 1, if we have a system, we find the transfer function and figure out
that all poles of the system transfer function lie in the left half complex plane, is the
system BIBO stable? And answer is yes and we shall prove it later.

89
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 12
BIBO Stability
Part – 2

(Refer Slide Time: 00:15)

So, let us look at BIBO stability in some detail. Any questions on the examples and the
general conclusions which I drew from them? So, I am going to prove one of them ok.
So, that that is my intention now; yes any questions? Anyone? Ok, so ok. So, let us let us
look at BIBO stability right. So, we know that 𝑌(𝑠) = 𝑃(𝑠)𝑈(𝑠) right. So, you give me
any plant transfer function, the output to an input U is going to be P(s) times U(s).

𝑡
This implies that 𝑦(𝑡) = ∫0 𝑝(𝑡 − 𝜏)𝑢(𝜏)𝑑𝜏. So, I am using the convolution integral ok.
So, let |𝑢(𝑡)| ≤ 𝑀 < ∞ ∀ 𝑡 ≥ 0 .Then what will happen?

𝑡 𝑡

|𝑦(𝑡)| = |∫ 𝑝(𝑡 − 𝜏)𝑢(𝜏)𝑑𝜏| ≤ ∫|𝑝(𝑡 − 𝜏)𝑢(𝜏)|𝑑𝜏

0 0

How did I get this? It is a property of the integral right. So just to give an analogy if you
have, let us say two scalars A and B and you add A + B right.

90
Let us say A and B are real numbers right. So, |𝐴 + 𝐵| ≤ |𝐴| + |𝐵|, what we call as
triangle inequality right, so a similar thing right. So, because you have an integral and
you have the integrand, first on the left hand side I am taking the absolute value of the
entire integral right and that is going to be less than or equal to the integral of the
absolute value of the integrand ok. So, that is what we are doing right ok.

Now, this is going to be

𝑡 𝑡 𝑡

|𝑦(𝑡)| = |∫ 𝑝(𝑡 − 𝜏)𝑢(𝜏)𝑑𝜏| ≤ ∫|𝑝(𝑡 − 𝜏)𝑢(𝜏)|𝑑𝜏 ≤ 𝑀 ∫|𝑝(𝑡 − 𝜏)|𝑑𝜏

0 0 0

Student: (Refer Time: 03:07).

Yes.

Student: M is greater than 0.

Yeah M is greater than 0 right.

Student: (Refer Time: 03:13).

Sorry.

Student: M is greater than 0, is not (Refer Time: 03:17).

Yeah this is a magnitude right. Here M is a finite positive real number. Anyway I am
taking the absolute value right; so, absolute value of u(t) right. So, it is going to be a
positive number right. So I hope it is clear.

So, now, please remember what was p(t)? It was what is called as the impulse response
function right, if you recall the physical meaning of p(t) right. So, question becomes, the
question that we would ask ourselves is that, when would the magnitude of y(t) be
bounded, given this expression?

So, given this expression when do you think the magnitude of y(t) will be bounded? See
𝑡
the magnitude of y(t) is bounded by this integral 𝑀 ∫0 |𝑝(𝑡 − 𝜏)|𝑑𝜏. Now suppose let us
say this integral is bounded right, certainly the magnitude of y(t) is bounded. Don’t you
agree?

91
 𝑡
See because |𝑦(𝑡)| ≤ 𝑀 ∫0 |𝑝(𝑡 − 𝜏)|𝑑𝜏, the follow up question that we would ask is that
𝑡
when is ∫0 |𝑝(𝑡 − 𝜏)|𝑑𝜏 bounded? Correct?

So, when do you think, the integral of a non negative integrand would be bounded? Why
am I saying nonnegative integrand? Because I am taking the absolute value of 𝑝(𝑡 −
𝜏) right; so, 𝑝(𝑡 − 𝜏) anyway can be a real valued function right. It can take positive or
negative values, but then once I take the absolute value it is going to be nonnegative.

Why am I saying non negative? It can be greater than 0 or equal to 0 also right. So,
𝑝(𝑡 − 𝜏) the absolute value of it is going to be a nonnegative function. So, I am taking
the integral of a nonnegative integrand. When do you think this integral will be bounded?

Student: (Refer Time: 06:12).

Only when lim𝑡→∞ |𝑝(𝑡)| = 0. See otherwise if this limit does not go to 0; See
integration is like sum right if p(t) does not go to 0 right what do you have? What
happens if you keep on integrating the number? Is only going to increase right.

So, for example, let me just give you a simple illustration right. So, let us say the
magnitude of p(t) goes something like this and then like it settles it down to a
nonnegative value what is going to happen? Integral is just the area, right.

So, what will happen is, although the magnitude of p(t) is bounded, the integral is going
to become unbounded as t tends to infinity, what is the only scenario when the integral
will be bounded, when the area will be bounded? Only when the function itself goes to 0
as t tends to.

Student: Infinity.

Infinity right. So, this is the condition. So, we can immediately see that this implies a
magnitude of y(t) is bounded for all time, if lim𝑡→∞ |𝑝(𝑡)| = 0 , ok. So, this is the
sufficient condition for BIBO stability right. It says that, look you give me the impulse
response and if the impulse response decays to 0 as time t tends to infinity then my
system is BIBO stable ok, but then this is a good concept, but still we are not there right.

So, because if I want to test it in test a system using impulse response, we know the
difficulty that we would face right, because it is very difficult to generate an ideal

92
impulse right. So, then how do I get the impulse response and how do I test the system
for its stability ok.

So, that is going to be a question right. So, can there be an equivalent criteria for this?
Yes there is and it is based on poles ok. So, that is what we are going to look at. So, let
us see how we get an equivalent criteria ok. We are only done with half the derivation
right, I hope it is clear till this point right.

We want to figure out when the system will be BIBO stable for a bounded input right.
For any bounded input when will the output be bounded? We figure out that the output is
bounded if the, what to say, the limit of the magnitude of the impulse response goes to 0
as t tends to infinity ok. Now, let us go further alright ok.

(Refer Slide Time: 09:09)

𝑛(𝑠)
We already know that 𝑃(𝑠) = 𝑑(𝑠) right, this is the plant transfer function alright and the

Laplace of p(t) is going to be equal to P(s), this is also something which we know right.
So, let there be k distinct poles of P(s).

Let the multiplicity of the pole 𝑠𝑖 , 𝑖 = 1, … , 𝑘 𝑏𝑒 𝜇𝑖 ok. So, the system order is n ok. So,
the order of the system is anyway n right; that is a general notation, anyway we are
following right the order of system is n. So, this implies that there are going to be n poles
right of the transfer function; that is something which we already know right, we are
dealing with a generic case, nth order system ok.

93
So, let us say out of these n poles, let k be distinct ok. I am doing a generic case. Of
course, all n can be distinct that is also right. So, in a few examples we would have seen
that the poles were distinct, there are no repeating poles, but there can be repeating poles
also right; we are just doing the generic case.

I am saying like let that be k distinct poles k ≤ n right ok, and let the multiplicity of the
pole be 𝜇𝑖 ok. What is meant by multiplicity? The number of times it repeats; if it is non
repeating multiplicity is 1. So, this implies ∑𝑘𝑖=1 𝜇𝑖 = 𝑛, alright.

Let us say if I have a 10th order system and let us say I have 9 non repeating poles with
the ninth pole repeating twice alright, summation of the multiplicity should be 10 right,
so that I cannot violate right. So, essentially the summation of the multiplicity should be
equal to the order of the system, which is the total number of poles correct. Now
immediately we see that P(s) can be written in this particular form ok.

𝑛(𝑠)
𝑃(𝑠) =
(𝑠 + 𝑠1 )𝜇1 (𝑠 + 𝑠2 )𝜇2 … (𝑠 + 𝑠𝑘 )𝜇𝑘

So, that is how I can split the denominator polynomial right, I can factorize the
denominator polynomial. Now I need to take the partial fraction expansion right. How
would the partial fraction expansion look in general? We did a few examples last time
right, so if you had a repeating pole what did we do? Let us say if we had a factor like 1
𝐴 𝐵
by (𝑠 + 1)2 what did we do? We did like 𝑠+1 + (𝑠+1)2 right. So, you will have two terms,

isn’t it?. So, what is going to happen is that, if when we do the partial fraction expansion
we are going to get something like this.

𝑘 𝜇𝑙 −1
𝑛(𝑠) 𝑐𝑙𝑚
𝜇 𝜇 𝜇
= ∑ ∑
(𝑠 + 𝑠1 ) 1 (𝑠 + 𝑠2 ) 2 … (𝑠 + 𝑠𝑘 ) 𝑘 (𝑠 + 𝑠𝑖 )𝑚+1
𝑙=1 𝑚=0

So, for example, I hope I did it correctly, let us double check ok. See for example, let us
1
say I have a system whose transfer function is this . What is the order of
(𝑠+1)(𝑠+2)(𝑠+3)2

the system?

Student: (Refer Time: 14:47).

What do you think is the order of the system?

94
Student: (Refer Time: 14:53)

See the order of the system is, the order of the denominator polynomial of the transfer
function right. What is the order of the denominator polynomial?.

Student: 4.

So, n is going to be equal to 4 right, and what are the poles? So, its going to be -1, -2, -3
and -3 right, there are four poles, but minus 3 repeats twice. So, what is the number of
unique non repeating poles? k= 3. So, then what, what are the values of µ1 µ2 µ3 which
are the multiplicities of the 3 non repeating poles?

µ1=1, µ2=1, µ3=2

If you add µ1, µ2, µ3 what do you get? You get 4 which is n right.

(Refer Slide Time: 15:57)

Now, if you want to do the partial fraction expansion what we will do?

𝐴 𝐵 𝐶 𝐷
𝑃 (𝑠 ) = + + +
𝑠 + 1 𝑠 + 2 𝑠 + 3 (𝑠 + 3)2

That is what we would do right, that is the same thing I have done in compact form right
using the summation.

95
 1
𝐴 𝐵 𝐶3𝑚
𝑃 (𝑠 ) = + +∑
𝑠+1 𝑠+2 (𝑠 + 3)𝑚+1
𝑚=0

So, the Clm are the residues ok, I am just putting two indices, because I have two
summations right that is it, what you called as A B C D I am just calling it c subscript l m
ok, that is just the change in the notation, because I have two indices l and m. I hope it is
clear how we got it right, yeah.

Now we take the Laplace inverse. So, now, in this particular problem, if I take Laplace
inverse what will I get? Let us continue with this problem right. If I take p(t) I am going
to get 𝐴𝑒 −𝑡 + 𝐵𝑒 −2𝑡 + 𝐶𝑒 −3𝑡 + 𝐷𝑡𝑒 −3𝑡 .

So, now, I am going to generalize from here right. So, this implies that

𝑘 𝜇𝑙−1

𝑝(𝑡) = ∑ ∑ 𝑐𝑙𝑚 𝑡 𝑚 𝑒 𝑠𝑙𝑡

𝑙=1 𝑚=0

I have just generalized from there ok, because you see that when there is no multiplicity
there is no t term ok, because anyway you will not have the summation with respect to
m. You will have a summation with respect to m, only if you have multiplicity greater
than 1. So, then you will have t m. So, for m = 0 which is the case when you do not have
any multiplicity. See for example, for the pole at minus 1, what is µ1? It is 1.

So, you just use m = 0 that is it right. So, you won’t get a t power term at, all, do you
agree?

Student: Yeah.

So, but then like for any case where the multiplicity is 2 and above you would get a
corresponding t term right, so that is what you, we are going to have. Do you agree? Is
everyone in agreement with what I have written? I am just parallely doing as example
and then doing the general case so that we can map them yeah.

Now once I get this, so this means that

𝑘 𝜇𝑙−1 𝑘 𝜇𝑙 −1

|𝑝(𝑡)| = |∑ ∑ 𝑐𝑙𝑚 𝑡 𝑚 𝑒 𝑠𝑙𝑡 | ≤ ∑ ∑ |𝑐𝑙𝑚 ||𝑡 𝑚 ||𝑒 𝑠𝑙𝑡 |

𝑙=1 𝑚=0 𝑙=1 𝑚=0

96
Do you agree? That is what we will have right ok. Now let us let us look at this further

𝑠𝑙 = 𝜎𝑙 + 𝑗𝜔𝑙

𝑒 𝑠𝑙 𝑡 = 𝑒 (𝜎𝑙 +𝑗𝜔𝑙 )𝑡

|𝑒 𝑠𝑙𝑡 | = |𝑒 𝜎𝑙 𝑡 | = 𝑒 𝜎𝑙 𝑡

So, we are almost there ok; of course, this was a long derivation right. So, see by and
large, I think this is the only proof I am going to do ok. So, in future we will make use of
results right, that is going to be the focus for this particular course, but at least I wanted
to show you ok, so that you understand how the inner details of whatever you are
learning works right.

So, essentially this shows that

𝑘 𝜇𝑙−1

|𝑝(𝑡)| ≤ ∑ ∑ |𝑐𝑙𝑚 |𝑡 𝑚 𝑒 𝜎𝑙 𝑡
𝑙=1 𝑚=0

Now, if 𝜎𝑙 < 0 ∀ 𝑙 = 1,2, … , 𝑘 , what can you say about the magnitude of p(t) as t tends
to infinity.

Student: 0.

It will go to 0 right, because please see that, please note that the left hand side is less than
or equal to the right hand side. In the limit t tending to infinity if the right hand side goes
to 0; obviously, the left hand side has to go to 0, because the left hand side cannot take
negative values right magnitude of p of t is nonnegative ok. So, the only choice for that
is, is that like magnitude of p of t should go to 0.

Student: 0.

So this implies that lim𝑡→∞ |𝑝(𝑡)| = 0. Now immediately one may ask the question; hey
what about the tm term? Alright. It so happens that the exponential term with the negative
exponent will dominate the t m term ok. So, the product will go to zero anyway right. So,

97
it so turns out that the sufficient condition for the impulse response to asymptotically go
to zero is that I need to have all poles to have negative real parts ok.

Let us say k there are k distinct poles, k can be less than or equal to n right. So, whatever
poles are there, all poles must have negative real parts or in other words we should have
all poles lying in the left of complex plane yeah, yes please.

Student: How can we take that magnitude of e slp is nothing, but..

Ah ok. So I just skipped one step there. So, if I take

|𝑒 𝑠𝑙𝑡 | = |𝑒 𝜎𝑙 𝑡 𝑒 𝑗𝜔𝑙 𝑡 | = |𝑒 𝜎𝑙 𝑡 ||𝑒 𝑗𝜔𝑙 𝑡 |

𝑒 𝑗𝜔𝑙 𝑡 = cos(𝜔𝑙 𝑡) + 𝑗𝑠𝑖𝑛(𝜔𝑙 𝑡)

So, to summarize we see that this implies that if all poles of the plant transfer function lie
in the LHP; that is have negative real parts then the system is BIBO stable ok. Some
people will say it is asymptotically stable ok, we are going to shortly see what that is.

So, that is that is what it is. So the important condition here is that I find the system
transfer function if all poles are in the left of complex plane, left of the imaginary axis
right left of the imaginary axis in the s plane then we have asymptotic stability or what
we call as BIBO stability. So, is this also necessary?

So, what do I mean by necessary? We have only shown what is called a sufficiency
condition right. So, what we have shown ok, if all poles lie in the left of plane we have
shown that system is BIBO stable. Now the opposite question arises, if a system has to
be BIBO stable, do all poles need to lie in the left of complex plane?

See I hope all of us understand necessary conditions sufficient condition and necessary
and sufficient condition. See if A is necessary for B to happen right; that means, that A
needs to be true for B to happen that need not imply that only A is required for B to
happen right.

So, if we say A is sufficient for B to happen, if A happens; that means, that even B will
happen right, but then B can happen due to other reasons also right. So, we do not know
right. So, essentially what we have shown is there, we have only shown that the
sufficient condition for BIBO stability is that all poles should lie in the left of complex

98
plane. Now what is the other way round ok; that is if a system, what to say, is BIBO
stable then all the poles lie in the left of complex plane right, is that true?

So, is the, what to say the is the condition of all poles lying the left of complex plane a
necessary condition for BIBO stability. Why is necessary condition important? Only then
we can talk about a corollary right, because if I say for BIBO stability all poles being in
the left of complex plane is necessary, then I can say that even a one pole lies in the right
of plane it is unstable. With this statement I cannot show right, because let us let us look
at this derivation right.

Now, if you tell me I have a system where one pole is in the right of complex plane what
will happen to the, what to say this derivation, what conclusion can I draw from this
derivation? Nothing, why? Let us say one pole is in the right of complex plane right, 1
sigma l is greater than 0. So, what will that imply?

Student: (Refer Time: 31:15) infinity

Ah I will the, I will only be able to conclude in the magnitude of p of t is less than or
equal to infinity that does not help me, that does not tell me it is unstable right is it. See 0
is also less than infinity you know like any other positive number is also less than or
equal to infinity right, less than infinity you know. So, it really does not help me alright.
So, what we have shown is only sufficiency ok, what about necessary condition ok.

I leave it to as homework you, it is not, and once again it is not necessary as far as this
course is concerned ok, but for those who are mathematically inclined please go on look
at the proof ok. So, it is available in many standard textbooks on controls. So, I hope my
point is clear right, what we have shown is only sufficiency it. So, turns out that it is also
a necessary condition.

So, let me what to say pose that question, is this condition necessary. What do I mean by
is this condition? All poles in LHP necessary for BIBO stability, the answer is yes, and
that is why if we do not, if we even have one pole lying the RHP or on the imaginary
axis the system is not BIBO stable that we have shown through examples right.

See to disprove BIBO stability I need to figure out only one bounded input for which the
output is unbounded we have already shown it ok, whether we call it critically stable

99
marginally stable or whatever it is, but then we found that it violates BIBO stability
definition. And if you have a pole on the pole in the right half plane, you give me any
bounded input the output will be unbounded that is that is sure.

If you give repeating poles on the imaginary axis, any bounded input will lead to
unbounded output, if you have what are called like non repeating poles on the imaginary
axis, you can always find one bounded input for which the output will be unbounded
right. So, that is why, this is also the, how do we conclude, because this is also necessary.
See I went from examples to a generic conclusion, the best way to do it is also always to
prove the generic one and then like do the examples ok, but since this is the first course
on controls you know like I wanted to do examples to motivate our understanding and
then like we did the general derivation right.

So, this I leave it to you as homework, the answer is yes, you need, it is also necessary;
that is why we are able to conclude that, even if one pole lies in the right of plane, system
is not BIBO stable ok. And if you have non repeating poles on the imaginary axis once
again, you know like to me it is BIBO unstable right, because like I can always find one
bounded input for which the system or to it is unbounded right.

So, although many textbooks will call it as critically stable or marginally stable; that is
because of another notion of stability ok. So, what I will do is that maybe this is a good
place to stop ok, so that you can digest all this information right

100
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 13
Effect of Zeros
Part – 1

In the previous lecture we looked at the effect of system poles on the response. We identified
the fact that the location of the poles influences the stability of the system. And for BIBO
stability all poles of the system transfer function should lie in left half complex plane. In
other words, they should have negative real parts.

In this lecture we are going to do a few examples to convey the impact of zeros on the system
response. To recall, the transfer function the ratio of the Laplace of the output to the Laplace
of the input with all initial conditions being 0. And the transfer function is going to be a ratio

N (s )
of two polynomials ( ) . The roots of
D(s )
N ( s) are called zeros of the transfer

function. The roots of D( s) are called poles of the transfer function. We have seen that are
poles are connected to system stability. Now let us look at zeros.

(Refer Slide Time: 01:39)

Let us consider two systems: the first plant transfer function is

101
 1
P ( s )=
( s+1)(s +10)

The second system is

S+ 2
P ( s )=
(s+1)(s +10)

The order of these systems is 2 because the order of the denominator polynomial is always
going to be the order of the system for the class of systems that we study.

For the first system, n=2 and m=0 .

We want to figure out the unit step response.

1
U ( s )=
s

1 1 A B C
Y ( s )=P ( s ) U ( s )= = + + .
(s +1)(s+10) s s s+1 s+10

1 −1 1
Evaluating the partial fractions, we get A= , B= , C= .
10 9 90

If we take the inverse Laplace transform, we get

1 1 −t 1 −10 t
y ( t )= − e + e
10 9 90

We can observe that the two poles are in the left of complex plane. So the system is BIBO

1
stable. As t → ∞ , y (t) → . The output remains bounded for all time. And we can
10
observe that the output function has exponential terms which correspond to the real part of
the poles as we have discussed.

For the second system, n=2 and m=1 .

The poles are at -1 and -10 there is a zero at -2. The difference between system 1 and system
2 is, we have introduced a zero at -2. Let us now calculate the unit step response.

102
 1
U ( s )=
s

S+2 1 A B C
Y ( s )=P ( s ) U ( s )= = + + .
(s +1)(s+10) s s s+1 s+10

1 −1 −8
Evaluating the partial fractions, we get A= , B= , C= .
5 9 90

If we take the inverse Laplace transform, we get

1 1 −t 8 −10 t
y ( t )= − e − e .
5 9 90

1
Even in this solution we can see that as t → ∞ , y (t ) → . So, it is bounded, the system is
5
BIBO stable because the poles are in the left half plane.

When compared to the first system, coefficients of exponent terms have changed due to the
zero. Let us see what the relative impact due to this is.

(Refer Slide Time: 11:40)

Let us draw the S plane, the poles are at -1 and -10 for both systems. In this case of system 2,
we have a zero at -2. If we look at the forced response in problem number 1, e−t will
dominate the output term for a longer time because as the exponent becomes more negative
the exponential function is going to decay to 0 much faster. If we consider the magnitude of

103
 1 −t 1 −10t 1 −t
the first term, e versus e , e is going to dominate for a longer time.
9 90 9

1
Another point is that the residues multiplying the two exponential terms
( )
9
∧1
90
. So, not

1 −t
only does e
−t
dominate e
−10t
but also the residues are such that e will certainly
9

1 −10t
dominate e . A pictorial representation is given in the slide. For this reason, the
90
poles that are closest to the imaginary axis are called dominant poles since they dominate the
output for longer time. Here -1 is the dominant pole.

Let us now consider system 2; here we introduced a zero which is closer to the pole at -1. As
we can see, it affects the residues. We can see from the pictorial representation that the
dominance of the -1 pole has reduced in comparison with system 1.

(Refer Slide Time: 17:43)

the consequence of the introduction of the zero at -2 which is closer to -1 is that the
dominance of the pole at -1 is reduced. As we keep on moving the zero closer and closer to
-1, the dominance of the pole at -1 will keep on reducing further and further.

104
We can see that the zero is going to affect the dynamic response of the system but it is not
going to have a direct impact on the stability. Because stability is dependent on the location
of the poles. Zeros will affect the coefficients of the response function.

(Refer Slide Time: 21:06)

If we shift the zero to -1, we are going to have,

S+ 1 1
P ( s )= =
( s+1)(s +10) (s +10)

The zero is going to cancel the effect of the pole completely. What was a second order system
to begin with would start acting like a first order system. This effect is what is called as pole-
zero cancellation. Typically do not do this, we will see why.

We will see that, doing control designs, we can introduce system zeros. For example, if we do
a proportional derivative control, we can introduce a system zero. During the design process
then we need to be in a position to how to handle those zeroes. We shall learn this later.

105
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 14
Effect of Zeros
Part – 2

So, I am going to do one more example, I want you to do this example and tell me what the
unit step response of this system is?

(Refer Slide Time: 0:0:18)

So, let us say, I take this system for example

s−2
P ( s )=
( s+ 1 )( s +10)

Here n = 2, m = 1
Poles: -1 and -10.
Zero: 2.

106
(Refer Slide Time: zero1:16)

Previously, the zero was at -2, what we have done is that we have taken at to 2 ok, that is
what we have done. Please calculate the unit step response and let me know what happens.

1
U ( s )=
s

s−2 A B C
Y ( s )=P ( s ) U ( s )= = + +
s ( s +1 ) ( s+10 ) s s +1 s +10

−2 3 12
¿ + −
10 s 9(s +1) 90 (s +10)

Taking inverse Laplace transform,

−2 3 −t 12 −10 t
y (t )= + e − e
10 9 90

Once again, you could observe that the system is stable; poles are in the left of complex
plane. Poles are the same, right see I have kept the poles to be the same in all these problems;
why? Because, we want to investigate the effect of zeros; so, I am not even touching the
poles, right. So we are working with the same poles, right.

So, we have shifted the zero to + 2 to the right of plane. So, the system is still BIBO stable;

−2
As t → ∞ , y (t) → .
10

107
So, the value can be negative also absolutely no problem, it needs to be a finite real number.
And you can once again observe that the exponential terms have the poles as the exponents
and the residues are also affected by the location of the zero. Now, we see that compared to
−t
example 2 the relative importance of e has now increased because we have shifted the
zero away from -1.

See -1 was the dominant pole when we had the zero at -2 the dominance was reduced. Now
we have shifted the, what to say, pole to away from the dominant pole at -1, you can see that
the relative magnitude has once again increased; the dominance has once again increased;

3 −t
that means that the magnitude of the e
−t
term is e . So, you could immediately
9

3 −t
observe that the e term will dominate ok, fine. So, that is the first observation ok. So,
9
can you observe something else?

So, what is y(0)? Of course, I mean I can always double check see whenever we are using
this analysis, we are assuming initial condition to be zero. So, once you complete the problem
using this transfer function based approach, it is always a good idea to substitute t = 0 and
ensure that y(0) is zero right because that is an implicit assumption that we have made.

So, you see that I start from y(0)=0, right and let us say, this is my final value ok. So, this I
leave it to you, you can go to MATLAB and plot this function, let us say this is minus 2 by
10. So, what is going to happen is that the output function will go something like this and
settle down to minus 2 ok, that is going to be the impact of the zero at 2.

So, you will see that in all those cases the initial value zero and you will see that the output
would monotonically increase to the change to the final value; what do I mean by that? It is
not going to change signs ok. So, for example, the final value is 1/10 and 1/5; the output
function will always be non negative ok, it will go from zero to the final value smoothly
without changing its sign. But in this particular problem, if you plot the output function
which is enclosed in this red box, you will see that it will be at zero, it will start at zero at
time t equals zero. Final value is going to be negative in this particular example it will so
happen that initially, it will go, the values will be positive, then it will change sign and then it
will come to the final value.

108
If I had written 2 - s it will become negative and then it will go to positive, then the effect is
going to be reversed. Either way, you know like the conclusion that we can draw is that if you
have a zero in the right half plane, your step response is going to have such a characteristic;
where the response will start off in one direction, then reverse its direction and go and settle
down in that way. So, let me give an example: let us say, we take a ceiling fan right; suppose
if the ceiling fan had a zero in the right of plane, what will happen? Let us say I switch on
that fan, what will happen? It may start rotating in, let us say clockwise direction after
sometime it may stop and then start rotating in the counterclockwise direction. Picturize that;
that is this case.

Suppose, if I had a door which has a zero in the right of plane, what will happen is that let us
say if I push the door I expect it to go out. Imagine a door if it comes in and then goes out,
right; Doors are not like that, but then like you can imagine right. So, that is going to be effect
of this zero in the right of plane. And zeroes in the right of plane are called as non minimum
phase zeros ok, we will figure out why it is called non minimum phase when we do
frequency response ok, we will see what is the meaning of the adjective non minimum phase
when we come to frequency response ok.

A minimum phase system is one which has all poles and zeros in the left of complex plane
ok. So, that is the definition of minimum phase and non minimum phase is it clear ok?

Student: Also the final value of y of t seems to be the constant in the numerator divided by
the.

Good observation ok. What his observation was that not only this problem; you take problem
number 1, you substitute s equals zero in the transfer function what do you get?

Student: 1 by 10.

1 by 10. Second problem substitute s equals zero, 2 by 10 and what are 1 by 10, 2 by 10
minus 2 by 10 in these problems? It is this steady state output of a stable system when
subjected to unit step input ok, but then we will generalize it using the final value theorem.
Do not always substitute s = 0 to find the steady state value for any input, it does not work
that way ok, it only works for this unit step input and that too for stable systems ok. So,
please remember that right ok.

109
So, anyway we will come to that that is a good observation, but we will come to that ok. Now
the question is when do we get a non minimum phase system in practice? So, I have to give
you an example, right. I can get non minimum phase zeroes right for example, when we have
a system with time delay, ok.

So, let us state let me take a simple example let us say I take

ý (t ) + y ( t ) =u ( t−T d ) T d → time delay

So, that what it means is that I give an input now that is going to affect the output only after
an interval time interval of T d . So, can we encounter time delay in practice, yes, we do
right. Suppose let us say you know I want to push this desk, the force which I am applying is
my input; displacement of the desk is the output. Will the desk start moving instantaneously?

Student: No.

No, right why not?

Student: Inaudible.

Exactly, you have to overcome static friction at the point the region of contact between the
desk legs and the surface right. So, even if you start counting time from the moment I start
applying a force on the desk right; the displacement happens only some time afterwards
alright. So, that is a delay; isn’t it?

So, friction for example, right incorporates a delay in a system ok. So, we will see that many
cases you know like you may have delays; see for example let us say I may give a voltage
input to a motor. The motor shaft may not start spinning immediately, right, it may take a
small interval of time before it starts spinning, right. So, question is that like is that delay time
significantly high when you compare it to the time scale of interest that is a question we need
to ask.

But assuming that is the case then what do we do right. So, I have to factorize timely. Let us
say; now I want to find the transfer function right you take Laplace transform on both sides.

110
(Refer Slide Time: 15:38)

Taking laplace transform,

s
sY ( s ) − y ( 0 ) +Y ( s )=U (s) e−T d

Taking initial conditions to be zero,

Y (s ) e−T s d

P ( s )= =
U (s ) s +1

Now, what are the values of n and m; what is the value of n?

Student: 1.

1. m=?

−T d s −T d s
I have e , right, how can I, can I expand e as a polynomial expansion? Yes, what
will be the order of that polynomial?

Student: Inaudible.

Infinite, right; so, I can just go to any order I want, right. So, I have a problem now because
the system has become infinite dimensional in a sense right. So, because we can apply the
tools that we are going to learn in this course only for proper transfer functions; that means,
that we need to have the order of the numerator polynomial to be less than or equal to the

111
order of the denominator polynomial ok. That is that is when we can apply the tools that we
are going to learn in this particular course ok.

(Refer Slide Time: 17:57)

So, given this limitation we can do approximations of time delay, provided T d is small.

Approximations of Time Delay:

−T s 1
1) e =
d

T d s+1

s
−T d
−T d s e 2 2−T d s
2) e = = → Pad e ' s 1 st order approximation
Td
s
2
2+T d s
e

Now we have introduced a non minimum phase zero through the second approximation; you
look at the numerator right T d is anyway positive what is the zero of the transfer function?
Approximate transfer function 2−T d s right. So, that is it going to be in the right of plane.

If you plot the step response what is going to happen is that, the output will go the other way
and then it will go and settle down assuming that we are stabilizing the system carefully right.
So, this is indicator of the time delay. In reality what is going to happen is that the real system
output may be something like this ok. So, let me show it in blue ok. So, the real system let us
say there may be a time delay and then like it may go and go like this ok. So, in this interval

112
there is no output at all, but then the approximation essentially tells us that the output may go
in the opposite direction and go in the correct way ok.

So, the pink curves only an approximation of reality and why are we doing this approximate,
then only we can design right ah. See I cannot design with this guy the tools that we are going
to use in this learn in this course cannot be used for this transfer function, but I can use it for
this transfer function ok. So, that is why we approximate, but the impact is that we introduce
a non minimum phase zero which is going to behave like this; the dip in this curve is a
fictitious dip, it does not happen in practice right because during that period the actual output
is going to remain zero due to time delay.

So, what it says is that ok, look if I if a model time delay for this desk the actual system
responsively that if I give a step force it is going to remain zero for some time and then like it
is going to move alright. So, that is what is going to happen, but what I am predicting through
this approximation is that if I apply a step force the desk is going to come slightly towards me
and then go backwards; in reality, it is not going to happen, but that is the approximation we
are going to have ok.

And why are we doing it? Because we the tools that we learn will be can be used only when
the transfer function is proper. So, we are using this approximate transfer function ok. So, I
hope it is clear when we can encounter non minimum phase zeros right when we have to
approximate incorporate time delays and do the design process; we have to do this right, we
introduce a non minimum phase zero ok.

So, what we have done in today’s class is essentially look at the influence of zeros, right, that
is what we have done today. So, we have looked at impact of poles yesterday. So, starting
from tomorrow’s class, we are going to look at as I told you stability and performance, right,
we have looked at stability and tomorrow’s class onwards, we look at performance
specifications that will complete the system block.

113
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 15
Closed Loop System
Part – 1

Let us briefly recap where we are, what we are doing and where we are headed.

(Refer Slide Time: 00:39)

We consider a dynamic system or a plant to which we provide an input u(t) and we obtain
an output y (t) . We want to regulate the output of the system to a desired value. So, we
give a reference input, ok. The reference input tells us what is the desired value of y (t) .
Then we sense the output and give it as a feedback and compare it with what we want and
then we get an error. We denote it by e (t) and this error is passed through a controller
which then calculates the control input. This is the overall objective of this course; that is we
want to learn how to design this closed loop feedback system control system. This is also
called as an output regulation problem.

Referring to the block diagram, right now we are in the system block. The class of systems
that we are considering are SISO linear time invariant causal dynamic systems. We used
deterministic continuous time mathematical models to study these systems. We represented
these systems by linear ordinary differential equations with constant coefficients. We then

114
applied Laplace transform to obtain the transfer function representation. We have looked at
the poles and zeros of the transfer function and their implications on BIBO stability. In the
coming lectures, we will look at the controller block and the potential choices for this
controller block.

In the block diagram, we understood the presence of sensor dynamics, actuator dynamics and
disturbances. This is called as a closed loop control system with negative feedback. We can
draw a similar block diagram where we can analyze the blocks with the corresponding
transfer functions. We use the following representation. Plant transfer function ( P( s) ),
controller transfer function ( C( s) ), control input ( U (s) ), output ( Y ( s) ), Laplace
transform of the reference input ( R( s) ), feedback ( H (s) ), error ( E( s) ) and filter in
the feedback path ( W ( s) ).

(Refer Slide Time: 13:56)

The advantage with going to the complex domain and drawing this block diagram is that the
output signal at the plant block ( Y ( s) ) right is going to be the product of the block transfer
function and the input to the block.

Y ( s )=P ( s ) U ( s)

U ( s )=C ( s ) E(s)

Y ( s )=C ( s ) P ( s ) [ R ( s )−W ( s ) ]=C ( s ) P ( s ) [ R ( s )−H ( s ) Y ( s) ]

115
Usually G ( s )=C ( s ) P ( s ) is the product of all the transfer functions is that come in the
forward path.

Y ( s )=G ( s ) R ( s )−G( s) H ( s ) Y (s)

[ 1+G ( s ) H ( s ) ] Y ( s )=G ( s ) R ( s )

Y (s) G(s)
=
R (s) [ 1+G ( s ) H ( s ) ]

This is what is called as the closed loop transfer function because it represents all the
components of the closed loop in a single block with R( s) being the input and Y ( s)
being the output of the block.

(Refer Slide Time: 20:19)

If we had a MIMO system, Y ( s) will be a vector and U (s) will be a vector, then what
we have is a transfer function matrix. In that case we need to be careful about the order of
multiplication of two transfer functions. For example; multiplication of C ( s ) P ( s ) , we
cannot write it as P ( s ) C ( s) because they are matrices.

We are going to abbreviate the closed loop transfer function as CLTF. The poles of CLTF are
called closed loop poles and the zeroes of CLTF are called closed loop zeroes.

116
How do we calculate closed loop poles? Similar to the earlier discussions, poles are the roots
of the denominator polynomial. Here denominator polynomial is 1+G ( s ) H ( s ) . The
equation 1+G ( s ) H ( s )=0 is called as the closed loop characteristic equation. We solve the
closed loop characteristic equation to get the closed loop poles. We are interested in closed
loop poles because when we design a closed loop system, first the closed loop system has to
be stable, for which all the closed loop poles must lie in the left of complex plane.

117
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 16
Closed Loop System
Part – 2

(Refer Slide Time: 00:14)

Now, it can be observed that the closed loop transfer function of a positive feedback system
is,

Y (s) G(s )
=
R (s) 1−G ( s ) H (s)

118
(Refer Slide Time: 06:16)

Now, some more terminology ok; so, suppose if I break open the loop at this point (near
W(s)), ok. Suppose, let us say I have a closed loop and I break open the loop and this point
whatever transfer function I get, which relates the signal at this point which is W(s) to what
goes as the input to the system which is E(s) is what is called as the open loop transfer
function, ok.

W (s ) H ( s ) P ( s ) C( s) E(s)
= =G ( s ) H ( s)
E (s ) E (s )

So, this is what is called as the open loop transfer function, ok. Why is it called open loop
transfer function? As I told look at the ratio of the signals that I am taking right, W and E,
right. So, as I told you imagine that you cut open the loop here, right, you open the loop the
signal that you are getting at this junction is going to be W and the input to the open loop is
going to be E, right. So, that is why we are essentially taking W(s) divided by E(s) as the
open loop transfer function that is a definition, and that is going to be G(s) times H(s).

The poles of the open loop transfer function are called as open loop poles right and the zeros
of the open loop transfer function are called as open loop zeros, ok. So, that is once again
some terminology we should remember, ok. So, if you encounter the adjective open loop we
are talking about G(s)H(s), ok. So, I hope it is clear why it is open loop right. So, I am
opening the loop, alright.

119
So, you may ask me the question, hey is it going to be something which is physically
meaningful? Not necessarily. Ok, but you will see that this is going to be used, right. So, it is
an indicator of what will happen without feedback that is what we have. Suppose, like let us
say I put all these elements in the loop, but I do not close the loop, right which I will never do
in practice by the way, right. So, see then what is the point of designing a closed loop
feedback system, right? Suppose, if I do not close the loop what is the transfer function I am
going to get, ok. So, that is what is called as open loop transfer function. Why is it important?
What is the open loop transfer function? G(s)H(s), right. G(s)=C(s)P(s)H(s).

So, you see that the open loop transfer function is nothing, but the product of all the transfer
functions that you have, right. So, you have three blocks, right. C(s), P(s) and H(s) you see
that immediately the open loop transfer function is formed by the product of these three
transfer functions, right, that is why you will see that it is useful, as we go along, ok.

When we go to advanced tools used in control design you will see the impact of this open
loop transfer function. Immediately you see that the closed loop characteristic equation is one
plus open loop transfer function, one can immediately observe that, right.

Closed loop characteristic equation = 1+ OLTF

In many references we will see that the same feedback diagram is modified in this way. The
controller and plant are combined and written as G(s).

So, now as I told you we are still looking at P(s), right. So, we are still in the plant transfer
function. So, we are once again analyzing the plant or the system, but as we move along we
will extend the concepts to the entire system, ok. So, now, we are looking at whether a plant
is stable or not by looking at the poles of the plant. So, when we design a closed loop
feedback system we want the closed loop to be stable. So, we look at the closed loop poles
we will extend it to the closed loop transfer function, ok. So, that is something which we are
going to do.

120
(Refer Slide Time: 10:56)

And, one important thing is that in any control design ok, we, as control designer should look
at two important aspects, in this order. The first one is that of stability ok, the second aspect is
that of performance, ok. Both are important, but, stability is more important than
performance, right. See, if I do not have a stable system to begin with I cannot even talk
about it is performance right first I should ensure that the system I am designing is stable,
then we talk about performance, ok.

So, what I am going to do is that like I am going to draw a dashed arrow to indicate that when
you design both are important for a control designer, but you first you ensure stability then
we talk about performance, ok. So, now, what we are going to start discussing from now is
that like how can I quantify performance, right. So, that is something which we are going to
learn, right.

So, we are going to learn about performance specification, right. So, what we have done is
that like we have looked at stability you know the notion of stability that we are going to
follow in this particular course is that of bounded input bounded output stability. And we
have seen that you know the criteria for ensuring bibo stability is that all transfer system
transfer function poles should be in the left of complex plane ok. So, that is the criteria right
for bibo stability, right.

let us say we want to develop a controller for regulating the temperature of air in this room,
right. Now, let us say all 60 of us design stable controllers, ok. Now, I have 60 controllers in

121
my hand how do I evaluate which is the best one, right? So, from a practical perspective I
need to evaluate, right. So, then what can be some parameters which I can use for evaluation.
Now, what do you think I can use as criteria for performance evaluation?

So, let us say I come into this room in summer, let us say that room temperature is let us say
38 degree Celsius, right. I wanted to go to 25, right. I look at each controller, right and then
like I see I set, I set my reference as 25 degree Celsius I may evaluate how fast it will go to
25; that is one.

Then, how I would say accurately it can maintain it around 25 how fast it can settle around 25
and how close will it remain to 25, right. See, if I have an air conditioner let us say which
goes from 38 to 25 in 10 minutes, but then it remains what to say between 23 and 27 you
know like, I may not be very happy right. So, let us say it cuts off at 23 degrees Celsius and
cuts in at 27 degree Celsius, right. So, I may not feel very happy right, because I am going to
have a four degree Celsius variation right, but on the other hand let us say an air conditioner
maintains the temperature between 24.5 and 25.5, I may be very happy, right. So, that is one
performance specification, right.

So, then how fast it will settle, right. How fast it may overshoot the value that is what we
have also looked at, right. So, there is I am given a reference value how fast or how far away
will it go from the desired value, right that is another performance specifications right what
are the steady state errors all right. So, the speed with which it will go to the final desired
value, the band to which it will regulate around the final value right, how much it will
overshoot the final value or the desired value all those are performance specifications that we
would be interested in, right and that is what we are going to learn and quantify, right.

So, typically the prerequisite is that the system is stable, that is why I discussed stability first,
right. So, because without stability there is no point in discussing performance, right. So, as a
prerequisite you know like I want my system I can only do performance analysis of those
systems that are stable to begin with, and typically parameters that are used to specify
performance or quantify performance I should say are extracted from the step response of
first order and second order systems, ok. So, we would see why, right.

So, typically the parameters that quantify what to say performance requirements are
motivated and calculated, the formulae are extracted from the step response or the unit step
response of what are called first order and second order systems, we would see why, as we go

122
along ok. The motivation will become clear when we go to advanced control design you will
see that we learnt about dominant poles right yesterday. So, by and large even if you have a
tenth order system, ok, by and large you are going to have a dominant dynamics that
corresponds to a first order system or a second order system by and large ok, there can always
be exceptions, ok.

Typically you are going to have the dominant dynamics which can be captured by a first
order or a second order system, ok. That is why we are motivated in using these systems for
extracting performance parameters ok, we will revisit this point later right ok. So,
consequently let us look at let us start today with first order systems.

(Refer Slide Time: 18:07)

First Order Systems:

So, I am going to start with a typical first order system model and I am going to leave you
with a set of questions, ok. So, what is a first order system the governing equation of a typical
first order system? See, in fact, like it really depends on whom you talk to suppose let us say
you are discussing this what to say this with a chemical engineer for example, right. They
would say a model for a first order process, ok. They will always be looking at processes,
right.

123
So, the plant for them is a process, ok. So, if you if you look at references that are written by
chemical engineer they will say that oh it is a first order process. So, please do not get
confused; it is the same thing, ok. So, we for us we are looking at systems, right.

The governing equation of a typical first order system is,

dy (t)
T + y ( t )=u ( t ) ,T ∈ R
dt

Take Laplace transform on both sides,

T ( sY ( s )− y ( 0 ) ) +Y ( s )=U ( s )

Assuming zero initial conditions,

Y (s) 1
= =P( s)
U (s) Ts+1

(Refer Slide Time: 21:09)

Here n=1; m=0.

So, we have a pole at -1/T, right. So, you need to solve TS + 1 = 0, As far as zeros are
concerned there are none. So, this implies that for stability what should happen?

Student: T must be positive.

124
T must be positive, ok. So, you will see that stability requires that it should be positive, ok.
So, we are going to have a pole at minus 1 over capital T. So, now, what I am going to leave
you with are these questions, ok. So, you calculate the unit step response ok, unit ramp
response and the unit impulse response of the system ok, that is a homework which I am
going to leave you with. We would restart from the same point on in the next class, right. So,
calculate the units unit step response means what do you use for U of s?

Student: (Refer Time: 23:20).

1
U ( s )= . Unit ramp means?
s

Student: 1 by S square.

1
U ( s )=
s2

For unit impulse U(s)=1.

So, please use this U(s) pretty straightforward, ok. It should not take you more than 5 minutes
to calculate all the three responses. Do please get expressions, make observations ok, what do
you observe from them and then like you come back to the next class we will restart our
discussions from here fine, ok.

Thank you.

125
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 17
First Order Systems
Part – 1

(Refer Slide Time: 00:28)

In this lecture we will discuss about performance of a first order system and obtain some
quantifiable parameters that can be used to evaluate a systems response. The general
governing equation of a first order system is given by

dy
T + y ( t )=u ( t ) ,
dt

−1
where T is a real number in general. This system has a pole at . We can immediately
T
see that system is stable if T > 0 .

Let us look at the unit impulse response. That means

u ( t ) =δ ( t ) , U ( s )=1.

126
 1 1
Y ( s )=P ( s ) U ( s )= = .
Ts+1 1
( )
T s+
T

Taking inverse Laplace transform,

−t
1 T
y ( t )= e .
T

That is the unit impulse response. If we want to plot this we look at the following,

1
t → ∞ , y ( t ) → 0 and t=0, y ( t )= .
T

Based on these, the plot of y (t ) can be drawn as shown.

Here at t=0, y ( t ) ≠ 0 , but according to the definition of transfer function the initial
condition should be 0. The reason for this is that we are giving an impulse input at t=0 .
Ideal impulse instantaneously goes to a very high magnitude at t=0 and comes back

1
again. During this the output increases from 0 to . That is how it should be interpreted.
T

1
So, the unit impulse response of a first order system increases from 0 to almost
T

−t
instantaneously and then decays exponentially with the exponent being , and then goes
T
to 0 as t → ∞ .

Let us calculate the unit step response.

1
u ( t ) =1,U ( s )= .
s

1 1 A B
Y ( s )=P ( s ) U ( s )= = +
s Ts+1 s Ts+ 1

Solving the partial fractions, we get A=1 and B=−T .

127
(Refer Slide Time: 06:22)

Taking inverse Laplace, we get

−t
T
y (t )=1−e .

As t → ∞ , y ( t ) →1 and t=0, y ( t )=0 .

y (t ) is going to increase from 0 and go to 1 as t →∞ , based on this we can plot the

response as shown.

We can observe that at t=T , y (t )=0.63 2.

0.632 is 63.2% of 1, which is the steady state output. So, t=T happens when the output
reaches 63.2% of the steady state value. T is called as time constant and it is defined for a
stable first order linear time invariant system as the time at which the output magnitude is
63.2 percent of its final steady state value. The time constant is a single most important
parameter characterizing the first order system dynamics. The time constant quantifies how
fast the system responds.

Now, if we substitute t=4 T , the value of the output is going to be 0.982. This is close to
98% of the final value. We draw a band around the final value which is usually taken as
±2 or ±5 . We consider ±2 , that means the output is 98% of the final value. That
happens to be four times the time constant and this parameter is called as the settling time.

128
Settling time answers the question; what is the time taken for the output magnitude to settle
within 2% of it is final value? This is important to evaluate how fast a system is.

These parameters are defined based on the unit step response. If we want to evaluate time
constant and settling time, we you have to use step as the input. We need not use unit step all
the time. We can always scale the input and the output accordingly.

(Refer Slide Time: 17:16)

Let us look at the unit ramp response.

1
u ( t ) =t , U ( s )= .
s2

1 1 A B C
Y ( s )=P ( s ) U ( s )= = + 2+
s Ts+1 s s Ts+1
2

Solving the partial fractions, we get A=−1 , B=1 and C=T 2 .

Taking inverse Laplace, we get

−t −t
y (t )=−T + t+T e T
=t−T (1−e ) .
T

As t → ∞ , y ( t ) → t−T and t=0, y ( t )=0 . The plot of y (t ) is shown in the diagram.

The gap between the function t which is your input and the output is going to be T

129
graphically. Please note that unit ramp is an unbounded signal. So BIBO stability is not
applicable in this case.

(Refer Slide Time: 25:39)

If we look at all the three responses, we can see that the step response is the time derivative of
the ramp response. And the step input is the time derivative of the ramp input. This is a very
nice feature of stable LTI systems. If we give u (t) and we get y (t ) , this implies that to

du(t ) dy (t)
the same system, if we give we get .
dt dt

130
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 18
First Order Systems
Part – 2

(Refer Slide Time: 00:18)

Other variants of 1st Order system:

dy (t)
1) T + y ( t )=bu ( t ) , b∈ R , b ≠1
dt

Y ( s) b
→ P ( s )= =
U (s) Ts+1

−t
1 b
Unit step response: U ( s )= , →Y ( s )=P ( s ) U ( s )= → y ( t )=b(1−e T )
s s (Ts+ 1)

131
(Refer Slide Time: 00:30)

b
y (t )=lim sY ( s )=¿ lim s =b
The steady state value of unit step response = s→ 0 s →0 s (Ts+1)
lim ¿
t→∞

The dynamic characteristics remains the same; the settling time is still 4T, but now what will
happen you will see that if you substitute t = T, y(t) will be b times 0.632, that is why the
definition was 63.2 percent of the final value not just 0.632, if you look at the definition.
Similarly, if you calculate what is y at t = 4 t you will get 0.982 b. So, the output at our reach
98.2 percent of the final value which is b. So, that is what is called by steady state gain.

So, immediately you will see that this is a 2 parameter model; 2 parameter model or 2
parameter process, ok. So, this people call first order model with non unity steady state gain
right. So, that is the variant and a third variant which people can have which essentially
occurs in many practical processes you will see you will look at this variant even when you
come to the lab in many practical systems you may have a system like this.

dy (t)
2) T + y ( t )=bu ( t−T d ) ,b ∈ R , b≠ 1 , T d > 0,T d ∈ R
dt

So, Td is the time delay all right. So, we have encountered what was time delay before,
right. So, that is you give an input; output starts from 0 only after the interval of T d , right.
So, that is time delay.

132
 Y (s ) b e−T s
d
b(2−T d s)
Plant Transfer Function: P ( s )= = = (using Pade’s
U (s ) Ts+1 (Ts+1)(2+T d s)
approximation)

Now, how many parameters do we have? We have 3 parameters. So, this becomes a 3
parameter model ok. So, that is that is another variant, ok. So, this people will call as a first
order system with the non unity gain and time delay, ok. Some in chemical engineering
references and all you will see first order process plus time delay, ok. So, that is a term which
they will use ok. So, this is the model which is used and this is the transfer function which
comes over as a result ok. So, you see that you have 3 parameters.

So, once again the dynamic characteristics remain, the same the time constant definition is
the same. The steady state gain is once again b, only difference is and now we have added a
time delay factor, ok. So, that is that is the change that happens ok, is it clear?

(Refer Slide Time: 09:51)

So, now, we go to second order systems. So, once again to recap the big picture as to why we
are studying first order and second order systems the reason is that by and large when you
want to talk about performance right we talk about the fact that we have the dominant
dynamics or dominant poles to be either one or two other is like what is close to the
imaginary axis, right. That is why when you want to specify performance we look at either
first order or second order systems you know like that is the motivation behind looking at

133
first order and second order systems for deriving performance parameters, ok. So, that is the
big picture once again.

Second Order Systems:

So, a typically second order systems are motivated from vibration problems. Due to that
reason their governing equation is written in a certain structure ok, we will see how it is
written, ok. So, the governing equation of second order systems is typically taken as the
following,

d 2 y (t) dy (t) 2
2
+2 ζ ωn + ωn y ( t )=ω2n u ( t )
dt dt

So, immediately you can see that n = 2, right. So, m for this particular case happens to be 0,
you can also have = 1, you can also have a derivative of u on the right hand side. can happen
as I told you we we would take it of to be of this particular structure and we will see what
happens when we have m to be 1, that is we have a term on the right hand side what really
happens to the systems response, that is something which we look at as we go along, ok.

So, we can see that immediately of course, before we calculate the poles you know like the
value ζ is what is called as the damping ratio ok, ωn is what is called as the natural
frequency of the system, ok. We would revisit these parameters and then I would ask you to
define these parameters like we are going to look at the physics behind these choices, right.
So, behind this choice of model, right, we will we will look at them. But, then like for
tomorrows or let us say for the next class you know I should say you know like please figure
out what is the definition of natural frequency and come, ok. So, a clear crisp definition of
what you mean by natural frequency, we discuss this as we go along.

And it so happens that we will we will figure out later that for stability of the system please
remember we can only talk about performance only if you have a stable system to begin with
right. It so happens that for stability of this class of systems the sufficient condition is ζ >
0, ω n> 0 , ok. So, that is the sufficient condition of course, in general we will see that you

know like the system will be stable if ζ and ω n are real numbers of the same sign, ok.

Now, taking Laplace transform on both sides,

2 2 2
s Y ( s )−sy ( 0 ) − ý ( 0 )+ 2ζ ω n ( sY ( s )− y ( 0 ) )+ ωn Y ( s )=ω n U ( s)

134
Apply zero initial conditions,

s
[¿ ¿2+2 ζ ω n s+ ω2n ]Y ( s )=ω 2n U (s)
¿

2
Y ( s) ωn
→ P ( s )= = 2
U (s) s + 2ζ ω n s+ ω2n

So, what are the poles of this transfer function? How do you get the poles? We solve the
denominator polynomial equals 0

Poles: s=−ζ ω n ± ω n √ ζ 2−1 .

Now, of course, it is a second order system value of n is 2; obviously, you get two poles,
right. Now, what can you say about these nature of these poles?

Immediately we can observe that the nature of the poles is depend on the value of zeta, right.
If zeta is between 0 and 1, what can I say about the poles?

If 0<ζ < 1 the poles are complex conjugates

If ζ =1 Repeated poles at −ωn

If ζ >1 Distinct real poles.

So, those are three cases that we would look at depending on the value of zeta the location of
the poles changes and consequently the system behavior changes a lot, ok.

So, that is something which we are going to do in the next class. So, we look at what are
called under damped critically damped and over damped systems, depending on whether the
value of ζ is less than 1, exactly equal to 1 or greater than 1, respectively, ok. So, we look
at what is the physical what to say implication of those three cases and then how do we do the
analysis, ok. So, that is something which we will do in the next class, but before that you
know like please figure out the definitions of a natural frequency and damping ratio, and
come.

135
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 19
Second Order systems
Part – 1

In the previous lecture was stopped with the derivation of the transfer function of second
order systems.

(Refer Slide Time: 00:21)

We are considering the general governing equation of a second order system to be given by

2
d y (t) dy (t)
2
+2 ζ ωn + ωn2 y ( t )=ω n2 u ( t )
dt dt

And applying the Laplace transform and using 0 initial condition we obtain the transfer
function to be

2
Y (s ) ωn
P ( s )= = 2
U (s ) s +2 ζ ωn s +ω n2

We observe that the poles are going to be at −ζ ωn ± ωn √ζ 2−1 .

136
(Refer Slide Time: 01:01)

Case 1: 0<ζ < 1 , Under damped second order system.

Since ζ <1 , √ ζ 2−1 will be negative, now we can rewrite the poles as complex
conjugates.

−ζ ωn ± jωn √ 1−ζ 2

And ω n √ 1−ζ 2=ω d , called damped natural frequency. So the poles become

−ζ ωn ± jωd

If we plot it on the complex plane, the real part is going to be −ζ ωn and the imaginary
part is going to be ± jωd .

Case 2: ζ =1 , Critically damped second order system.

The poles are going to be at −ωn and −ωn . We have a pair of repeating poles at
−ωn . This is because when ζ =1 ,

ωn2
P ( s )= 2
( s +ωn )

The denominator becomes a perfect square. So we have a pair of repeating poles.

137
Case 3: ζ >1. Over damped second order system.

Now, the poles are going to be at

−ζ ωn ± ωn √ζ 2−1 .

We have two a real poles

If we qualitatively plot these two poles, one pole (−ζ ωn +ω n √ ζ 2−1) is going to be closer
to the imaginary axis compared to the other. We know this pole is still in the left half plane
because −ζ ωn would be dominating the sum. Since this pole is closer to the origin, it is
the dominant pole.

(Refer Slide Time: 07:26)

Let us see the physical significance associated with each of these cases. Let us consider the
mass spring damper system. Its transfer function is

1
X (s) 1 m
= 2 =
F(s) m s + cs+k c k
s 2+ s+
m m

Comparing this with the second order transfer function, we have

2
ωn =
k
m
, ω n=
k
m √
138
 c c c
2 ζ ωn= ,ζ= =
m 2m ωn 2 √mk

If we have a mass spring damper system, we can correlate the damping ratio and the natural
frequency to the parameters m , c and k . The damping ratio is indicate of the
amount of damping that we put in. Natural frequency is the frequency with which the systems
response would oscillate if there were an initial perturbation with no damping and no external
excitation.

Let us consider the case of the under damped system.

2
Y (s ) ωn
P ( s )= = 2 , Poles are at −ζ ωn ± jωd .
U (s ) s +2 ζ ωn s +ω n2

If we plot the two poles in the s plane, the real part is −ζ ωn and the imaginary parts are
jωd . The length of this line segment which is drawn from the origin to any one of these
two poles will be ω n . This can be obtained using the Pythagoras theorem. Now we define
an angle β as shown in the figure. We can see that cos β=ζ .

Let us calculate the unit step response of this under damped system.

1
U ( s )=
s

2
ωn 1 A Bs+ C
Y ( s )=P ( s ) U ( s )= 2 2
= + 2
s +2 ζ ωn s+ω n s s s +2 ζ ω n s+ω n2

Solving partial fractions,

1 s+2 ζ ω n 1 s +ζ ω n ζ ωn ωd
Y ( s )= − 2 = − −
s s +2 ζ ωn s+ω n2 s ( s +ζ ω n )2+ ωd 2 ( s+ ζ ωn )2 +ω d2 ωd

Taking inverse Laplace transform,

−ζ ω n t ζ −ζ ω t
y (t )=1−e cos ωd t− e n
sin ω d t .
√ ζ −1
2

139
(Refer Slide Time: 24:27)

The expression which we derived for unit step response has two exponential terms and the
exponential terms have the real part as exponent. The system is stable because the real part is
negative. As t → ∞ , y (t )→ 1 . The response is going to be oscillatory with frequency
ωd because of the presence of cos and sin. That is why it is called damped natural
frequency.

The performance specifications of second order systems are motivated/derived from the
above expression. Let us see what these parameters are and how they are calculated. If we
plot unit step response as shown in the figure, it will first overshoot the steady state value of
one is going to oscillate around it and then finally, settle down. Since the cos and the sin
functions are multiplied by a decaying exponential the oscillations is going to decay. Now we
look at performance specifications. Please note that the premise behind these parameters is
that we are looking at the unit step response of a stable under damped second order system.

1) Rise time ( t r ¿ - it is time taken for the system to reach 100 % of it is final value for
the first time. But if we have a critically damped or over damped system, the output
reaches the final value or 100 % of it is final value only as t → ∞ . It will never
overshoot. Then, the rise time definition will change. Some say it is the time taken for
the output to become 90 % of the final value, some say 75 %, some say 10 to 90 % or
10 to 75 %. There are various conventions. To get an expression for rise time we
equate the response to one.

140
 y ( t=t r )=1=1−e
−ζ ω n tr
[ cos ωd t r −
ζ
√ζ 2−1 ]
sin ω d t r ,

e
−ζ ω n t r
[ cos ωd t r −
ζ
√ ζ 2 −1
sin ω d t r =0 ,
]
The exponential term cannot be 0 for a finite time.

ζ
cos ω d t r− sin ωd t r=0 ,
√ ζ 2−1

tan ω d t r= √
ζ 2−1
,
ζ

tan−1 √
− ζ 2−1 π −β
t r=
1
ωd ζ
= (
ωd
. )
(Refer Slide Time: 33:39)

We know that, cos β=ζ

ωd √ ζ 2 −1 √ ζ 2−1
tan β=
ζ ωn
=
ζ
; tan −1 ( ζ )
=β

Therefore,

141
 −√ ζ 2−1
tan −1 ( ζ )=π −β

We can graphically plot the poles of the under damped second order system and compute
β . β should be in radians, ω n and ω d in radians per second.

From the expression of t r ; we can observe that as ζ increases, rise time increases and
β decreases. This means as we make the system more damped, the rise time increases or
in other words the system becomes slower to respond. Conversely, as we make this system
damping lower, the system becomes faster. That is the rise time, starts to decrease.

142
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 20
Second Order Systems Part – 2

(Refer Slide Time: 00:14)

What is the second parameter that we can obtain? The second parameter which you are going
to get it is what is called as a peak time t p ok. So, what is the peak time? The peak time is that
instant of time at which the output reaches the first peak of its overshoot ok. So, the peak time
of the under damped 2nd order system is the instant of time at which the output or response
reaches the first peak as far as overshoot is concerned ok.

So, obviously, you see that the output value is the maximum for all time right at peak time,
because you see that the output at the second peak is lower than the first peak, because you
have a decaying exponential multiplying the cos and sin terms right. So, the output to the first
peak is going to be the maximum value. Then how can I get peak time?

Student: Differentiate expression.

Differentiate that expression equate it to 0, then solve, then

Student: Inaudible

143
What you do? You get the second derivate, and see if it is negative right for maximum right.
We need to do those two steps. See here, why are those two steps important, here you have a
graphical interpretation, so what he says make sense right. See here, I have drawn the graph.
Suppose, if I did not draw the graph, and ask to find the maxima right.

What will you do? You will differentiate it once equate the first derivative to 0 solve, then
you take the second derivative, and substitute the roots for which the first derivative was 0,
and find out whether the second derivative was negative or positive right. If it is negative, it
was a maxima right, positive means it is a minima right. So, please do that ok.

(Refer Slide Time: 04:08)

Am I correct? Right? The cos ω d t terms cancel of each other. So, I am only left with the
sin ω d t term. And the term within this square bracket is also non-zero, because it is
positive right.

So, what can be zero, sin ωd t is 0, so that means, that at the solution for this thing is

nπ
right. Obviously, n = 0 does not make sense right. So, we said n = 1, 2, 3 and so on ok,
ωd

nπ
so that is what we have right.
ωd

144
 d2 y
Now, of course, you need to calculate the second derivative. So, is going to be let us
dt2
put this constant term outside, then what we are going to have is that we will first
differentiate the exponential term, then we differentiate the sin term. So, what can you say
about the second derivative at t=nπ ? At t=nπ the first term is gone right, second term
will be some exponential, which is going to be positive right times cos ⁡( nπ ) ok. For what
value of n will be cos ⁡( nπ ) be negative. The first value is n = 1 right; cos ⁡π is - 1
right. I hope it is clear, what I am doing right.

Now, we see that for n = 1, the second derivative becomes negative. See for n = 0 or 2 or
something the second derivative is positive right, so that that take that indicates a local
minimum right. So, for n = 1, you know like the second derivative is negative, so that is a
maxima. And from our argument, we see that that is the global maxima for this particular

π
function right. So, as a result, the peak time is going to be equal to ok,
ωd

Now, the third parameter is what is called as maximum peak overshoot ok, what is this
maximum peak overshoot ok. So, immediately you see that this value is y at t p right, this one,
I can write as y at infinity right, because as t tends to infinity, what is the value of y right at
steady state value. Now, what is the maximum peak overshoot? What you mean by overshoot,
overshoot is the value beyond which the output goes over the steady state value right.

Now, the question that we are asking ourselves is by how much I am overshooting the final
value to the maximum possible extent, so that means that I should first look at this number
right, what is y(tp) - y(∞) right that is the maximum overshoot right that I am having. So, this
maximum peak overshoot typically is normalized by the final value and indicated as a
percentage measure. That is how it is given right.

y ( t p ) − y (∞ )
M p= ×100
y (∞)

−ζ ωn π −ζ π

=1+ e √1−ζ
ωd 2
y ( t p ) =1+e

y ( ∞ )=1

145
 −ζ π
Therefore, M p=e √ 1−ζ 2
×100

(Refer Slide Time: 11:31)

Immediately, we can see that as ζ increases, Mp decreases right. See we already saw that as ζ
increases, the rise time increases right; the system becomes slower, but the advantage is that
the peak overshoot reduces. Why is that important? Because peak overshoot is indicative of
by what magnitude my does my output exceed the final value.

See, just imagine this, let us say you open a door, which is automatically closing, and then
you pull it enter and then you leave it, you may see that is oscillating. Ideally, do you want
the oscillations to be very high or as small as possible? As small as possible right, because
like without want to hit the next person right. Then what will I do, I will essentially ensure
that my damping is large. But if you make the damping large, you will observe that it takes
longer for the door to close that is indicative of rise time.

So, you would have observed that in many doors if you want fast closure, the oscillation will
be larger right, because you will make the damping to be smaller. But then the price that we
pay is it the oscillations are larger in magnitude. But if I do not want this oscillations, or I
want the oscillations to be smaller, the price I need to pay is that I need to compromise on the
rise time ok.

146
So rise time and peak overshoot are two conflicting requirements. And considering these
requirements we design second order systems with damping ratios between 0.4 and 0.8 ok,
this is know, this is just a rule of thumb ok, which is common control design practice.

By and large we want to design under damped second order systems that have damping ratios
between 0.4 and 0.8. We do not want damping ratios lower than 0.4, because the overshoot
will be very high ok, although the rise time will be very small. We do not want damping ratio
is beyond 0.8, because the system will be sluggish, the overshoot will be low, but then rise
time will be high then the systems response takes a longer time ok, so that is the tradeoff
between rise time and peak overshoot, is it clear.

So, the last parameter, which I am going to touch upon is settling time. This I am going to
leave it to you as homework. I am is going to give you the expressions, you figure out how
we got them ok. The settling time, once again it is the time taken for the output to settle

4
within a band right. So if you take a ±2 band, the expression for settling time is .
ζ ωn

3
If you take it 5 percent band, settling time is plus ok.
ζ ωn

By default in this course, if I say settling time, we are dealing with the 2 percent band ok.
Settling time for a 1st order system will be 4 times time constant that is the settling time we
are going to use in this particular course, Is it clear? So, how did we get this expression? I
leave it to you as homework ok. We derived for three expressions that is what where peak
time, rise time, and peak overshoot; settling time you please do it.

So, once again, how do we visualize settling time, let me go back to this graph. Let us say I
take the 2 percent band, I draw the upper band upper limit and the lower limit using this pink
dotted line, and then you figure out what is the time after which the output settles to within
this band right that is my settling time, is it clear. I hope it is graphically also clear as to what
we define by settling time right.

So, theses four parameters are based upon the transient response of under damped, stable,
second order, linear time invariant, causal systems ok. So, these are typical parameters that
will be specified ok, like when you want to specify the responses, is it clear, yeah.

Student: [Inaudible].

147
No not first peak per se what you can say is that it is the first instance after which the output
settles within the band. See for example, let say the output settles here,

Student: [Inaudible]

Yeah. So, see that thing is what is the first instant of time, you know like after which no part
of the output touches the band that is how you to did yeah.

Student: [Inaudible]

We are going there ok. Let me preemptively give you the reason, the reason why either time
constant or these parameters are used is because once again the dominant dynamics that you
typically find in higher order systems are either correspond to a 1st order system or an under
damped second order system. What do I mean by that is that dominant poles will be either a
pair of complex conjugate poles, or a pole on the real axis, which corresponds to either a
under damped second order system, or a 1st order system respectively that is why we use this.
Of course, it is one approximation, but it is an useful approximation nonetheless ok.

Student: [Inaudible]

Yeah I just indicated qualitatively ok. The idea is that like after the settling time point, you
know the output should remain within the band, that’s all ok, fine.

So, now that we have look at under damped second order systems, let us qualitatively plot the
step response of the three class of systems, and then I am going to leave you the homework
problem ok. So, let us do this. So, if I plot y(t) versus t, and we are looking at the unit step
response of the three class of systems right, we are looking at the unit step response of the
three categories of systems. So, if you have under damped second order system, what is going
to happen is that the output is going to be oscillatory and then it will settle down right. If you
had a critically damped, so this is an under damped system ok. If you had a critically damped
system, what is going happen is that the output is going to start like this, and it is going to
essentially never going to oscillate ok, never going to cross one ok, but it will monotonically
tend to the final value. So, a critically damped system has no overshoot, but it is the fastest
among 2nd order system without any overshoot ok.

148
On the other hand, if you have an over damped system, the over damped system will be
extremely sluggish ok. So, not a great diagram ok, let me let me draw a nice smooth curve ok,
it will it is going to be extremely sluggish ok. So, we hardly design over damped systems ok

To give an example, if you want to prevent collision between vehicles right, if you have two
vehicles following each other you do not want the vehicle spacing to oscillate like an under
damped system. See you do not want to be in a vehicle, which is travelling behind another
vehicle, and it coming and going forward and backward suddenly all right, when is stopping
you want a smooth nice motion right.

So, then, but then like you also want the system to act as fast as possible. So, you have to
conflicting requirements, then what you do that you would design as if it is a critically
damped 2nd order system. But, of course, as I told you critically and 2nd order system means
zeta is exactly 1, and in real life, it is very difficult to make it exactly equal to 1 ok, we will
see when you design right.

But, on other hand, if you make zeta to be greater than 1, practically it is going to be very
sluggish ok. I am sure, you would have seen many of these automatic closing doors right. You
open it, it will close slowly right, that means that is over damped ok, it is very sluggish ok.
So, you see that if the damping ratio is very high, things will be very sluggish right, so that is
an over damped system ok.

So, as I told, by and large, we are going to design as control designers we would be under
damped systems with damping rations between 0.4 and 0.8; that is just a rule of thumb ok,
there is no hard and fast rule.

So, what I am going to leave you with are two problems. First, find the unit step response of
critically damped and over damped systems. Second, find the unit impulse response of under
damped, critically damped, and over damped second order systems. Of course, it goes
without saying that we are dealing with 2nd order systems ok, so that is what we I want you
to do ok.

So, please do these two exercises ok. And so, this essentially gives a broad overview of what
we wanted to look at in the systems block right as I told you right. So, starting tomorrow
what I will do that I will provide you with an overview of what options are available in a

149
controller block, before we go for further analysis ok, so that is what we are going to do
tomorrow right.

150
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 21
Controllers
Part – 1

Let us recall what we have been looking at. We started off with the block diagram of the
negative feedback closed loop control system.

(Refer Slide Time: 00:34)

The first block that we looked at was the system or the plant block. We identified the class of
systems that we are going to discuss as far as this particular course is concerned and the
mathematical equations we would use to model those systems which happened to be linear
ODE’s with constant coefficients. We used Laplace transform to get the transfer function
representation. We evaluated the system response and observed that it is comprised of free
response and forced response. We focus on forced response because we are more interested in
the effect of input on the output.

151
(Refer Slide Time: 03:53)

Also, in the above example from previous lectures, if we look at the response due to the
initial condition, which is the first part, and take the partial fraction expansion and inverse
Laplace. We are going to get terms which have e−2t and e−3t , which means that the free
response decays to 0 with time. So we give more importance to forced response.

Also the notion of stability that we looked was bounded input bounded output stability. We
want to all the system poles to be in the left of complex plane for BIBO stability. We also
discussed the effect of location of poles on system response and stability.

(Refer Slide Time: 07:14)

152
In the second order systems, when we look at the unit step response and unit impulse
response, we see that for an under damped system we are going to have an oscillatory
response because we had complex conjugate poles, but for over damped and critically
damped systems there are no oscillations because we had real poles on the negative real axis
right. So, in a certain sense poles affect the nature of response. And zeros affect the dynamic
response by influencing the coefficients of the exponential. We can say that poles affect the
nature of response and zeros affect the shape of the response. We also learned that the poles
located closer to the jω axis dominate the system response and if we have zeros close to
these dominant poles their dominance is going to be reduced.

(Refer Slide Time: 08:59)

Now, we look at the controller block in our diagram. We will start with the discussion of
typical choices for a controller.

(Refer Slide Time: 12:00)

153
If we draw a similar diagram for the controller as we did for the system, the input is e (t)
and output from the controller is a control input u(t) . If we go to the complex domain, we
call the controller transfer function as C( s) . And then the Laplace of error e (t) and
the control input (t) , we call them as E( s) and U (s) respectively. We can see that
U ( s )=E ( s ) C(s)

We use P (s) for plant transfer function and C( s) for controller transfer function. In
some cases people also use G p(s ) and Gc(s) respectively. But according to our notation
G ( s )=C ( s ) P ( s ) .

Let us see the different choices of controller;

1. On-off controller (Bang-Bang controller): For example, consider aan air conditioning
system. Let us say we keep the temperature at 25 o C. When the temperature increases
to 26o C the compressor cuts in and blows cold air. And when the temperature goes to
24o C it cuts off. That is on-off control. There are only two actions either on or off that
is about it. The control input structure looks like,

u (t)=
{
u 1if e (t )<e
u 2Otherwise
1

}
We can have multiple variants and finer regions to this. This is not very effective,
although it is very intuitive, it may not be the best for the system because it does not
regulate continuously. We are not going to consider such controllers in this course.

154
 2. Proportional controller (P controller): As the name indicates the control input u(t)
is directly proportional to the error which is given. The structure of the controller is

u ( t ) =K p e ( t )

Kp is the proportional controller parameter very commonly called as the

proportional gain. It can take any value in (−∞ , ∞ ) at least from a mathematical
perspective.

If we take the Laplace transform,

U ( s )=K p E( s)

K p is the controller transfer.

The advantages of the proportional controller is that it is very simple in structure and
implementation is going to be pretty straightforward. The error is just multiplied by a
constant. By and large this is the first controller we try out when we want to design systems.
But on the flip side, it so turns out that for certain class of systems a proportional controller
alone would not be able to remove steady state errors. We will see what those scenarios are as
we go along.

155
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 22
Controllers
Part – 2

So, what is the next type of control that we can look at to essentially alleviate this problem?

Student: [Inaudible]

I am sorry what is that?

Student: [Inaudible]

Sorry suppose if I if I want to essentially set my design temperature as 25 degree Celsius

right, let us say the temperature is coming down from 30 degree Celsius right.

(Refer Slide Time: 00:17)

If the room’s structure is such that it fits in a certain class of systems, which will learn about
later short maybe in the next maybe after 4 or 5 lectures, what will happen is that the
proportional controller may bring the temperature to 25.5 degree Celsius down from 30, but it
will never go to 25 exactly. So, that is a steady state error right. So, for example, let me draw
a diagram let us say I plot output versus input right ok.

156
So, let us say I want a temperature of 25 degree Celsius right, what the proportional controller
may do is that like starting at 30 degree Celsius it may essentially bring it down, but then it
will settle down like this. So, as time increases you are going to have a non-zero steady state
error ok. So, that is what I called as a tracking error, what is tracking? I want to follow a
certain set point right or even a time varying reference input right so, that is tracking right.
So, essentially we may not be able to achieve zero steady state error that is a limitation.

Student: [Inaudible]

You can have, as I told you can have different variants, absolutely, you can have absolute
value of E or one may say that look even if it is negative I will do this, if it is positive do this
one depending on the value of e1, e1 can be negative. So, that is why I said there can be many
variants you can give range.

But then the thing is you know you take these discrete actions, it is not like continuous action
based on the error magnitude ok, you have like a discrete binary operation right that is it, but
you look at a proportional control it is a continuous action right at each instant of time
depending on the error, you take action right. And the biggest advantage of proportional
control is its simplicity, because the control input at any instant depends only on the error at
that instant that is it does not worry about the past, neither do you worry about what may
come in the future ok. It only lives in the present that is it ok, that is a big advantage of
proportional control right ok.

Then let us go to integral control ok, integral control is abbreviated by I ok. So, that is the
abbreviation used for integral control ok. So, what is this integral control? As the name
suggests so, you just take the control input u(t) to be integral of the error right over a time

interval. So, you just take u ( t ) =K i∫ e ( τ ) dτ

0

Ki is once again a real number ok, Ki is what is called as an integral gain ok.

So, if we take the Laplace transform what happens to this?

Ki
U ( s )= E (s )
s

157
 Ki
And, C ( s )=
s

So, this is the structure in the time domain for this particular controller, this is a transfer
function of this particular controller, the integral controller right ok.

(Refer Slide Time: 05:08)

Now, what are the pros and cons? It can lead to zero steady state errors ok, that is big
advantage why we use integral control ok. So, we will see how it does it. So, that is big
advantage of integral control, but then like it should be used judiciously why? Because if you
look at the integral controller transfer function itself, you will see that it has a pole at zero
see, if you look at C(s) right. Ultimately what is going to happen we discuss open loop pole
and closed loop poles right, open loop poles are poles of the G(s)H(s), C(s) is going to go in

Ki
G(s) isn’t it right. So, the moment I put a transfer function of the form , I am going to
s
introduce an open loop pole at s = 0 right.

So, then we will see that the control design for the closed loop system becomes a bit
challenging ok, we need to be a bit more careful because see the control we are only going to
look at the open loop transfer function, my plant may not have even have a pole at the origin
right to begin with the plant may be stable. But the controller structure is i am choosing can
make it unstable right, that is the product of the plant and the controller transfer function will
now have a pole at the origin; that means, I will have an open loop pole at the origin right.

158
So, then I need to be careful in stabilizing the closed loop control system right. So, this
introduces an open loop pole at origin ok, which essentially needs to be accounted for and
more interestingly integral control can lead to what to say actuator saturation ok. So, that
that is something which also we need to worry about ok, like when we do integral control. So,
what do I mean by this, let me demonstrate it by means of a graph. Let us say suppose I
consider a scenario right.

Let us say once again of this temperature control ok, let us say my initial temperature is 30
degree Celsius I come in I set my temperature at 25 degree Celsius: Design 1. So, let us say
initially the error is 5 degree Celsius right, what happens from 5 degree Celsius the error is
going to keep on decreasing ok. Let us assume that the error remains non negative all the time
just for the sake of argument ok.

Now, what would integral control do, let us say now this current time t let us say t 1 let me
take an instant t1 right, what will be the control input calculated by the integral controller at

t1

time t equals t1, it will be ∫ e ( τ ) dτ and graphically what will be that?

0

Student: Inaudible

It would be the area right. So, now, let us say I go to an instant t 2 what will that be, it will be
this area.

So, now what can you observe assuming that the error does not vary between positive and
negative let us say it remains the same sign right what can happen is that this area can keep
on increasing if the area keeps on increasing what do you think will happen to u? It keeps on
increasing right of course, you multiply by K i, but then as time progresses ok, you are
decreasing the errors magnitude. But the integral action takes into account the past error also
all right, because integration in a certain sense takes into account the history of errors right
that is what it is doing right.

So, essentially we are take into account past errors and, we are carrying them forward with
us, then what happens is that we may face a situation, where the area become so high that the
control input may become very high that it can exceed the capacity of the actuator see for
example, in this air conditioner what is the actuator I may use I will be using a compressor
based cooling system right

159
Let us say I have a certain capacity for the compressor and the fan assembly right and, I keep
on calculating this control input even as the temperature error reach goes close and close to 0,
there may be a point where I may have a saturation in the control input ok, I may ask for let
us say x liters per minute ok, but the capacity of the system or actuator may be less than x
liters per minute right that may be possible ok.

So, that was also another limitation with integral control. So, we need to be a bit more careful
ok. So, this is what is called as actuator saturation when you design integral control. So, you
need to plot the control input and check whether you are what to say you are able to maintain
your control inputs within the allowable bounds of operation. Anyway when we do control
design we will do all these things ok. So, when we do control design not only we are
interested in making sure the error goes to 0, but also in ensuring that the control input that
we are giving to the system is within practical bounds right otherwise, there is no point right.

See for example, if I want to have a linear motion system right, in some machining assembly
right and I have a I purchase a motor system based actuator electromechanical actuator,
which can provide a maximum of 100 Newton’s. Now, I program a controller which at an
instant of time asked me for 1000 Newton’s, I cannot realize it in practice right, then what
will happen I reach actuator saturation.

Then with such a controller although in theory I make be satisfying some performance
requirements, in practice I would not be able to reproduce those performance requirements
right, because I am saying settling time would be so much with 1000 Newtons; peak
overshoot would be lower than let us say 5 percent if you give 1000 Newton’s, but my
capacity of the actuator is only 100.

So, in theory while designing I might have I can put in whatever value I want, but we need to
be realistic right as far as what to say maintaining the control input within the actuator range
is concerned ok. So, that is something which we will also look at when we do integral control
fine. So, that is the second block.

Now, what is the third building block which we can use which is what is called as the
derivative controller ok, derivative controller abbreviated as D ok. So, what does what do you
think derivative control does? It differentiates right, as the name indicates, it differentiates the
error at each and every instant of time.

160
And then multiplies it by a constant Kd which we call has a derivative gain and of course, K d
is once again a real number right and, then calculates the control input ok. So, that is what it
does. So, if you take the Laplace transform of course, we neglect the initial conditions, then
what will we get is this ok. So, then the controller transfer function becomes K dS that is what
happens ok. So, the controller transfer function is KdS.

So, if you look at the derivative controller as a block itself transfer function, you see that this
is not a proper transfer function see, if you incorporate the derivative controller as a block
C(s) in your block diagram, if you look at the controller transfer function as a standalone
transfer function in its own right you see that K dS is not a proper transfer function right. what
is a proper transfer function?

Student: Inaudible.

The order of the denominator must be greater than or equal to the order of the numerator so,
here the order of numerator is 1 order denominator is 0. So, this is not a proper transfer
function, but then anyway when you put it together, we want the open loop transfer function
to be a proper transfer function that typically would work out ok. So, that should be kept in
mind so, for example, if your plant transfer function had n equals m and, you do not have
H(s) you have to be 1 you cannot; obviously, use the derivative controller why.

Student: Inaudible.

Because the open loop transfer function will not be proper right. So, we need to keep that in
mind so, but that is something which I want you want to alert you all (Refer Time: 14:47).

So, let me repeat that suppose if you had your plant transfer function P(s) to be a proper
transfer function with n equals m, there can be scenarios like that right, the order and the
order of the numerator and denominator polynomials of the plant transfer function P(s) let us
say they are equal right. And H(s) happens to be 1, then if you put C(s) to be K dS, then you
calculate the open loop transfer function G(s) and H(s) you would have the order of the
numerator to be one order more than the denominator right due to the KdS, then we need to be
a bit careful right. So, we cannot really use that right yeah.

161
So, but typically the open loop transfer function would become proper and, there is what the
derivative action would lead to right. So, essentially it would lead to this transfer function
right, but physically what does the derivative controller do?

Student: [Inaudible]

Ok very good. So, if you look at the derivative action itself right. So, let us say you take any
time t, what is going to happen is that the derivative is going to calculate the slope of the error
right. So, in a certain sense it is peeking into the future right, the derivative the sign of the
derivative and also the magnitude of the derivative would tell me, what is going to happen to
the error in the next instant all right.

So, in a certain sense the derivative controller leads to an anticipatory response and like so, it
tends to peek into the future and take corrective action as he said you know, let us say if the
error is going to increase a lot then maybe it may take action now itself.

So, it will not allow the error to increase beyond a certain level right, if the error if we know
that the error is on a downward trend, then I will progressively reduce the control input now
why should I have see control input has to do with cost ultimately in real life right say. Let us
say you are running a compressor you know we are spending for electricity right is it not
right, if I know that my temperature error is decreasing I would essentially run my
compressor at a lower setting so, that I can save cost right.

So obviously, the derivative controller can do these what you say you can satisfy these
functionalities, right but what are potential limitations with derivative control

Student: It can it can need to non-zero tracking error.

Non-zero tracking error?

Student: Non-zero taken steady state error.

Non-zero, steady state error.

Student: Inaudible

So, in other words you will see the derivative what you say controller will not take any
action, suppose if the error reaches a constant nonzero value, see let us say let us say the

162
temperature error due to some reason falls down to two degree Celsius and remains there,
whatever reason right in practice you know we cannot control what may happen right.
Imagine a scenario, where it becomes constant then what is the derivative of a constant
becomes 0 right. So, that is one perspective what he told right.

So, but more importantly any other in your concern that we may have from derivative
control, we look at response characteristics perspective later, but we are trying to get an
intuitive understanding right.

Student: It will make it unstable [Inaudible].

It will make it unstable?

Student: [Inaudible]

Oh that once again maybe you have sharp changes in the error, you once again you may
exceed the actuators limits. Agreed

See more importantly if you have constant error the derivative controller will not take any
action or will take negligible action right. Let us say if it what to say essentially hangs around
a constant value derivative controller may not be able to take sufficient action and as he told,
if you have a discontinuity in the error not even a discontinuous sudden change in the error to
do whatever reason ok, the derivative controller may essentially give very high inputs ok,
which may be beyond the actuators limits.

See for example, let us say I have an air conditioned room suddenly I, it is in the peak of
summer , suddenly I keep the door open for 2 minutes due to what are reason what is going
to happen, the thermal load is going to increase a lot right the error is going to shoot up right.
Then the derivative error will the derivative control will try to use the maximum capacity of
the compressor right.

So, question is whether is desirable or not that is a different question to ask ok, but that is the
characteristic of the derivative controller, but I am going to essentially leave you with one
more limitation of derivative controller with this example ok, once again it is just a numerical
example to just convey the point ok.

163
Let us say your error was something like let us say sin t + 0.01 sin(100 t). So, what this
means is that I have an error signal ok, once again this is just a notional example to convey
the issues with derivative control right.

So, let us say my error essentially oscillates as sin t ok, but what is sin 100 t it is high
frequency noise right, in practice you always have noise coming in through measurements
right in real systems, see even if you take a oscilloscope you measure a constant voltage
signal do you get a nice smooth line? No you will get a band right. So, let us say you have a
band which is your noise right. So, I am just to just demonstrate the concept I am just I have
to say doing this.

So, now if you let us say you take the derivative of this, what will happen is that you will get
cos t plus, what will happen to the other term, you will get cos 100 t ok. So, what should I
have been my derivative signal in the ideal world.

Student: Inaudible

It should have been only cos t which will be like this that should I mean ideal derivative
signal, which should have been the control input, but what does cos 100 t, extremely fast
cosine signal right and cosine 100 t is going to mask this cos t. Do you agree? So, cos 100 t is
going to be something like let us say it is going to be like a band ok, we will not even be
able to I am just drawing relative to this ok.

So, but you get the point right so, it is going to just mask this signal. So, as a result you are
your controller is going to fluctuate very fast which is not which is not the intended result
right, but that is due to the fact that the derivative controller may amplify high frequency low
magnitude noise ok. So, this is we may consider it to be low magnitude, but the issue for us
that it is a low magnitude high frequency noise that is the problem for us ok. So, on the other
hand the control action, just mix amplifies it right and we can lead into that can lead to
problems ok.

So, I will I will stop here please think about all these issues, we will come back and then
look at all these points. Once again tomorrow and then like will I will start doing some
examples, but also like just go to MATLAB and simulink and figure out how this derivative
controller is realized in practice do, they use it as K dS or something else please look at it and
come to tomorrows class ok. So, what is one way by which you can remove this effect.

164
Student: Inaudible

Integration, see I want to remove the effect of this low magnitude high frequency noise right
what can I do?

Student: Inaudible

You can use a low pass filter ok. So, just check out how it is how this derivative control is
realized in MATLAB and come and tell me tomorrow ok. So, then you will understand what
this one is.

165
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 23
Closed Loop Control
Part – 1

In the previous lectures, we discussed about the controller block and three types fundamental
controllers which are proportional control, integral control and derivative control.

(Refer Slide Time: 00:45)

166
(Refer Slide Time: 00:59)

If we look at practical implementation of derivative control, in many cases, people would use

Kd 1
. Here would act like a low pass filter to filter of the noise signals because
1+Ts 1+Ts
one limitation of derivative control is that it can accidentally amplify high frequency noise
that comes up during practical implementation.

Proportional Integral (P I) controller:

One starts with proportional control and if things do not work out, then we try out
proportional and an integral control or P I controller. We can try the integral block alone but
that introduces an open loop pole at the origin. There are going to be some difficulties in
designing the closed loop system and making it stable. Consequently one alternative is to use
the P I controller. It has two components. The control input is essentially the sum of the
proportional part and the integral part. The transfer function is going to be

K i K p s +Ki
K p+ = . We see that we introduce an open loop pole at zero, but at the same
s s

−K i
time we introduce an open loop zero at .
Kp

167
(Refer Slide Time: 04:50)

There is something called as a controller parameter space or control gain region. It is used to
indicate the feasible region of the controller gains to achieve closed loop stability and
performance. It is just a graphical representation. In the case of P or I or D, there is only one
controller parameter that is the real number. This controller parameter space will be one
dimensional - a line.

(Refer Slide Time: 06:08)

168
But then we comes to P I the controller parameter space becomes 2 dimensional. We need to
plot a two dimensional plot of Kp and Ki where we graphically represent all the
feasible values.

Proportional Derivative (P D) controller:

P D controller, as the name suggests is proportional plus derivative. Consequently the

controller transfer function is just the sum of the two individual transfer functions of
proportional and derivative control which is K p + sK d . We can see that this introduces and

−K p
open loop zero at . We call it open loop zero because open loop transfer function is
Kd
G ( s ) H ( s )=C ( s ) P ( s ) H (s) , so when we change C ( s ) , open loop transfer function is
affected. In this case the controller parameter space will be a two dimensional plot of Kp

and Kd .

(Refer Slide Time: 11:14)

Proportional Integral Derivative (P I D) controller:

169
Then in general we can combine all the three blocks; proportional, integral and derivative,

de(t )
e ( t ) +¿ K d
dt
which is called as PID. As the name indicates t .
u ( t ) =K p e ( t )+ K i∫ ¿
o

(Refer Slide Time: 12:48)

Ki K p s + K i + K d s2
The controller transfer function will be C ( s )=K p+ + K d s= . We can see
s s
that this introduces two open loop zeros, which depend on the values of Kp , Ki and
Kd .

Now there are three parameters to design, Kp , K i and K d . The complexity keeps on
increasing. Note that each of these parameters is a real number that can vary between −∞
to ∞ . We would be focusing on this controller for most of our discussions, for the simple
reason that an overwhelming majority of controllers that are used in practical applications,
belong to this class of PID controllers.

Let us see some examples, we have a plant transfer function

1
P ( s )=
s−1

170
The order of the system is 1 and it is unstable since it has a pole at 1 which is in the right of
complex plane. One purpose of feedback is also to ensure that we stabilize an unstable
system. Let us try to design a closed loop unity negative feedback system. As a common
condition for all these examples; we consider unity negative feedback closed loop control
system thought not always necessary. This means H ( s )=1 .

Let us try a proportional controller. So we have C ( s )=K p .

Kp
G ( s )=C ( s ) P ( s )=
s−1

(Refer Slide Time: 19:49)

The closed loop transfer function is going to be

Kp
Y (s) G(s) s−1 Kp
= = =
R (s) 1+G ( s ) H ( s) K s + K p−1
1+ p
s−1

Solving the closed loop characteristic polynomial which is the denominator polynomial of the
closed loop transfer function, we find the closed loop poles.

s + K p−1=0

Therefore the closed loop pole is going to be at 1−K p .

171
For closed loop stability we want all closed loop poles to be strictly in the left of complex
plane. That means we want 1−K p to be negative. That is possible when K p >1 . The

feasible region for K p can be drawn as shown in the figure.

By this analysis, we have converted a parameter which can take any value between −∞ to
∞ into a parameter whose range is between 1 and ∞ , we have narrowed down the
range. We will later see that if we want to incorporate closed loop performance this range will
reduce further.

172
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 24
Closed Loop Control Part – 2

(Refer Slide Time: 00:17)

So, I hope this first example is clear, pretty simple, right. So, now, let us go and do another
example, ok. So, since I could stabilize with a proportional controller itself we stop here, that
is it, ok. So, I am not even trying other controllers, right. So, of course, one could try them,
ok, but we are just stopping here, ok.

So, as a second example let us say we want to choose the plant transfer function as

1
P ( s )= 2
s +1

So, is this plant stable? So, what is order of this plant?

Student: Inaudible

Second order, right. So, is it bounded input bounded output stable?

Student: No.

173
No, all right. So, as I told you we will introduce another notion of stability called stability in
the sense of Lyapunov. It will so turn out that as per that notion of stability this stable, as long
as I do not I have simple roots on the imaginary axis. But as we have already seen that we
here the poles are at ± j, if I have non repeating or what we call a simple poles, right on the
imaginary axis, right what will happen is that if I provide a sinusoidal input whose angular
frequency is equal to the magnitude of the pole the corresponding output essentially goes to
infinity in magnitude as time progresses.

So, we see that if through the system if we provide cos t or sin t the output essentially blows
off, right to infinity, ok. So, this is BIBO unstable ok, and of course, people will some
people will call it as marginal marginally or critically stable ok. So, it depends on which
school of thought you belong to right. So, that is what we have here, right.

So, now once again we want to stabilize. You know next see, yeah one may call it marginally
critically stable, but I do not think anyone of us want to have this as a transfer function of a
practical system, for multiple reasons, ok.

First reason is the following, right. Let us say you know just to give an example, let us say we
have an automotive suspension whose transfer function is like this. Would that car even sell?
Can you sell that car? Imagine that you have an automotive suspension you know whose
transfer function is this, right. And suppose I drive that car on a road whose frequency is
which at a speed such that the road input which comes into the car from the tires has an
angular frequency of 1 radians per second. Let us say I excite that I can always do that, right.
Suppose if I excite it what is going to happen? The car is going to essentially vibrate by a
huge amount, right. So, I will not even buy that car, right. So, because that is going to be very
uncomfortable, ok.

Second thing is please note that all this analysis only approximate. So, if I say that the poles
are ± j ok, that is only an approximation ok. In reality the poles can be in a region around + j
and - j. So, it can just because the system parameters may vary whatever I have done is an
approximation, right. So, essentially I might have not taken into account some dynamics. So,
there may be unmodeled dynamics there may be uncertain parametric uncertainties and so on,
right. So, the poles may be around a region, of ± j.

So, depending on my luck sometimes it can be in the right half plane marginally, but still it is
in the right half plane. The moment that happens, obviously, the system becomes unstable,

174
right. So, we in practice due to these reasons we do not want to place any poles on the
imaginary axis by design, ok. If it comes to us naturally, if a plant naturally has poles on the
imaginary axis we stabilize by using closed loop feedback, which is what we are going to do,
ok. And even we will talk about what is called relative stability later on, right.

Even we do not want the poles to be for this reason we do not want the poles to be very close
to the imaginary axis. Why? Because once again let us say I place poles which are very very
close to the imaginary axis, right, once again this pole may be in a domain or region around
the whatever I have specified as one particular location. So, we do not want it to be too close,
within quotes to the imaginary axis, right.

So, what we call as relative stability would be used to quantify those effects, we will come
there, so obviously, we do not want poles on the imaginary axis from a practical perspective.
So, what we do is the following, we once again stabilize right. So, let us say you know I am
drawing this block diagrams now, in future you should be able to visualize, right. So, for a
few examples I am drawing this block diagram ok. So, we are considering unity negative
feedback. So, this is the diagram. So, we have E(s) and we have the controller transfer
function C(s), right.

So, once again we start with proportional control ok, Kp, right. So, this implies that

Kp
G ( s )=C ( s ) P ( s )=
s 2+1

The closed loop transfer function becomes,

G(s) K
CLTF= = 2 p
1+G ( s ) H (s ) s + K p +1

Closed loop characteristic equation, s 2 + K p +1=0

→ s=± √−( K p+ 1)

175
(Refer Slide Time: 07:40)

If K p >−1 , Closed loop poles are on the imaginary axis

If K p=−1 , Closed loop poles at 0,0

If K p <−1, One closed loop pole each on the –ve real axis and +ve real axis

So, either, see irrespective of what range of K p we choose between minus infinity plus
infinity, is there any value of Kp that will satisfy that will essentially achieve closed loop
stability for this system?

Student: No.

No. So, this implies a proportional control would not work, would not achieve closed loop
stability for this plant ok. And the way we have done closed loop for this plant under, with
unity negative feedback, that is important, right. So, because that is what we have done right,
ok. So, I am writing it once, in future I will just say it does not work that is it, right. But then I
want to write very carefully, right, so for this particular plant and when we do unity negative
feedback proportional controller would not stabilize the closed loop system, ok. That is a
conclusion we need to draw, right from this example. Then what can we do? Now I go and try
maybe.

Student: [Inaudible]

176
What can I try?

Student: [Inaudible]

I can try PI for example, right. So, let us say we try PI. So that means, that we are going to
have the controller transfer function to be

K i K p s+ K i
C ( s )=K p+ =
s s

K p s +Ki
→G ( s )=C ( s ) P ( s )= 3
s +s

(Refer Slide Time: 13:06)

G(s) K p s+ K i
Closed loop Transfer function: = 3
1+G ( s ) H (s) s + ( 1+ K p ) + K i

3
Closed loop characteristic equation: s + ( 1+ K p ) + K i =0

Now, what can you say about the closed loop poles of course, now you see that the closed
loop system is of order 3, right. See the order of the closed loop system is the order of the
denominator polynomial of the closed loop transfer function, right. So, that can keep on
changing; the plant may be the same the plant may be a second order plant, but once we
introduce a controller we are introducing its own dynamics, all right. So, consequently the
order of the closed loop system can change with the controller. So, here we have obtained a

177
third order closed loop system. When we tried with a proportional controller the closed loop
system was still order 2, now it has become order 3. So obviously, there are going to have 3
roots, right 3 closed loop poles. Where are they?

So, you can see that the problem becomes as you keep on going ahead things are becoming
more challenging, right isnt it, ok. So, we in fact, we started off the premise that oh we do not
want to work with odes, let us apply Laplace transform and work with algebraic equations,
even algebraic equations have their own challenges, ok. So, it is not that just because we have
gone to polynomials things become very straightforward yes, yes it is easier to do by hand,
ok, but still you know we need to learn some theory, ok. So, I am going to digress here and
teach you some general theory and then we will come back to this, ok.

So, now you will understand that how mathematics can be used for practical design
applications, ok. So, let us go learn some general polynomial theory which leads to what is
called as the Rouths criteria and then we will see how it is used for control design, ok. So,
that is something which we are going to look at. So, I am I am pausing the example here we
will come back to this, right after we learn the Routh’s criteria, right.

(Refer Slide Time: 16:28)

So, let us consider a polynomial d ( s )=a 0 sn +a 1 sn−1 +…+ an−1 s+ an

So, a polynomial obviously, this is an nth order polynomial, right. So, it will have n roots,
right, if I solve it, right. So, this is a definition; a polynomial is said to be Hurwitz if all its

178
roots have negative real parts that is when a polynomial is said to be a Hurwitz polynomial,
ok.

So, in other words if I have a nth order polynomial all n roots must be in the left of complex
plane, all right then that polynomial is said to be Hurwitz, ok. Of course, we are considering
what to say, of course, in this particular field what to say for our particular application we
consider only a real polynomial; that means, that all coefficients are a real numbers, a0

through an are real numbers you know for us, ok. That means, that if I have a complex root
the conjugate also must be the root must be a root, ok. So, please remember that, ok.

So, how is this related to our control design? See we want all closed loop poles to be in a left
half plane for closed loop stability, right, we have we have figured it out from what to say
basically our derivations, right previously, now we correlated the physics behind BIBO
stability to the mathematics. So, now how do I get the closed loop poles? By solving for the
roots of the closed loop characteristic polynomial which was 1 + G(s)H(s), right. The
denominator polynomial or the closed loop transfer function is the closed loop characteristic
polynomial, ok. So, if I ensure that my closed loop characteristic polynomial is Hurwitz I am
done, right I ensure that the closed loop system is stable do you agree? Ok.

Now, we ask the question when or for what values of these controller parameters if they exist,
ok, what would the closed loop characteristic polynomial be Hurwitz, ok. So, that is the
question we are going to answer in general in the next class, ok.

So, we are going to stop with this question, ok, how do we evaluate whether a polynomial
d(s) is Hurwitz?

Student: Sir, is that all coefficients could be positive?

Ok. I think I will leave you with this question once again. I partially have given you the
answer. So, a very good point, right. So, I will just take one minute of your time, sorry about
that, right.

179
(Refer Slide Time: 21:25)

So, let us say in this example hey why I should even do closed loop what to say design, right.

1
Let us say I have ok, what you do is that you just do an open loop control design and
s−1

s−1
then multiply by which is a proper transfer function and that is about it, right. So,
s +1

Y (s) 1
you have = . Very simple you know, why should I even worry about feedback
R (s) s+1
and so on, but why should we not do it, right please think about it, ok. We will also discuss
this on in the next class, right.

So, partially the answer has to do with the answer which I have already given you today in
today’s class, ok. And there are more reasons to that, ok. We will discuss them.

180
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 25
Routh’s Stability Criterion
Part – 1

(Refer Slide Time: 00:20)

In the last class we looked at evaluation of stability, we learned that if we are given a transfer
function and if we want to comment on the stability of the system, we need to look at the
denominator polynomial of the transfer function and all the roots of the denominator
polynomial must be in the left of complex plane, which is the criteria for bibo stability.

We wanted to answer the question: if we have any system any generic system, how do we
essentially figure out when that system would be bibo stable.

181
(Refer Slide Time: 01:30)

We solved a problem where the plant transfer function was

1
P ( s )= 2
s +1

we saw that it has poles on the imaginary axis at ± j . So we know, if we give an input like
cos t or sin t , the output is going to become unbounded.

(Refer Slide Time: 01:52)

182
(Refer Slide Time: 02:00)

We wanted to design a unity negative feedback system. And, we tried with the proportional
controller and found out that the proportional controller could not help us there. When we
tried the PI controller, we ended up with a third order polynomial. Now, we have two
parameters that we can change and a third order polynomial. With this can we find out the
ranges of values of these parameters for which the close poles would lie in the left half
complex plane.

In general when would a polynomial be Hurwitz? If we consider a polynomial

D ( s )=a0 sn +a 1 sn−1 +…+ an ,

where all the coefficients a0 , a1 … an are all real numbers. This polynomial is said to be
Hurwitz if all it is roots have negative real parts, In our discussion, we want the denominator
polynomial of the closed loop transfer function to be Hurwitz. Then we are assured that the
closed loop system is going to be stable. How do we evaluate whether a given polynomial
D ( s ) is going to be Hurwitz? This is the question we are going to answer.

Let us first consider a first order polynomial (n=1) .

D ( s )=a0 s+a 1

183
This would be Hurwitz when a0 ≠ 0, a1 ≠ 0, and both a0 and a1 have the same sign.

−a 1
They can both be positive or both be negative. In this case the root is going to be . If
a0
we want this real root to be negative then both the real numbers must be of the same sign.

Now, let us first consider a first order polynomial ( n=2 )

D ( s )=a0 s2 +a1 s +a2 .

It so turns out that this polynomial would be Hurwitz when a0 , a1 and a2 are all
nonzero and of the same sign. The roots are going to be in the left of complex plane. We can
tell this for sure even without calculating the roots.

Now let us go to n=3 . The polynomial becomes

D ( s )=a0 s3 +a 1 s 2 +a2 s +a3 .

th
There are four coefficients. For an n order polynomial we are going to have n+1
coefficients. In any general, in any nth order polynomial if a0 =1 , that is called as a
monic polynomial. That means that the coefficient of the highest order term is 1.

Now, for the roots to lie in the left of complex plane we still need a0, a 1 , a2 and a3 to be
non-zero and of the same sign. But for polynomials of order 3 and more that is only a
necessary condition. For polynomials of order 1 and 2, the necessary condition is also
sufficient but not far polynomials of order 3 and more. Let us demonstrate it through two
examples ok.

184
(Refer Slide Time: 09:49)

Let us consider the first polynomial to be

D 1=s3 +2 s 2+3 s+1.

And the second polynomial the polynomial to be

D 2=s3 +2 s 2 +s +3 .

If we determine the roots of the above equations, we can see that, all the roots of D1 have

negative real parts and lie in the left of complex plane. And D 2 has one root on the left of
complex plane and two roots in the right of complex plane.

The only change between D 1 and D 2 is the coefficient of s term and constant term
are interchanged. In these two polynomials all coefficients are nonzero and of the same sign.
But we see a drastic change in the way the roots of the two polynomials are distributed. We
can see that what was necessary and sufficient for polynomial till order 2 is not sufficient
here.

For sufficiency we construct the Routh’s array. The Routh’s array is constructed as shown
below.

185
(Refer Slide Time: 17:31)

sn a0 a2 a4 a6 …

s
n−1
a1 a3 a5 a7 …

s n−2 b1 b2 b3 …

s n−3 c1 c2 c3 …

⋮ ⋮ ⋮ ⋮

s1 ⋮ ⋮

s0 ⋮

We write the column s n till s 0 and then we start filling in the first two rows of this array
with the coefficients of the polynomial which we are given in the order of the decreasing
powers as seen in the array. We fill the first two rows till the coefficients get exhausted. Then
we start calculating bi and c i terms.

Let us come to b1 . So, if we want to calculate matter any coefficient in any row we
consider the coefficients which are available in the two rows that are immediately above that
row. In this case b1 appears in the s n−2 row. So, we consider the two rows immediately
above that which are s n−1 and s n . Then b1 is calculated as

a 1 a2 −a0 a3
b1=
a1

a 1 a 4−a0 a5
Similarly, b2=
a1

186
 b1 a3−a1 b 2
c 1=
b1

b1 a5−a1 b 3
c 2=
b1

We continue this process until we finish constructing the Routh’s array. Once we finish this
process, we look at the first column of the Routh’s array. The sufficiency condition is that
there should not be any sign changes in the first column of the Routh’s array for the
polynomial to be Hurwitz. That is that the number of sign changes in the first column of the
Routh’s array should be 0.

So for any general polynomial of order n with real coefficients, the necessary condition is that
all the coefficient must be non-zero and of the same sign. The sufficient condition is that the
first column of the Routh’s array should not have any sign changes

(Refer Slide Time: 26:02)

Suppose, if there is a sign change, it is interpreted as follows.

The number of sign changes in the first column of the Routh’s array is equal to the number of
roots of the polynomial in the right of complex plane.

If we have a 0 in the first column of the Routh’s array then we need to follow a different
process and that may imply roots on the imaginary axis. We are going to test those scenarios.

187
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 26
Routh’s Stability Criterion
Part – 2

(Refer Slide Time: 00:17)

Now, let us go back and look at the two polynomials that we looked at and let us apply these
conditions to the two polynomials right.

So, d 1 ( s )=s 3 +2 s 2+3 s+1

So, now how do I construct the Routh’s array? I write s 3 , s 2 , s 1 , s 0 right. So, that
is what I do. Then I take the first two rows right, then I start filling the coefficients right.
What are the coefficients? 1, 2, 3, 1 alright that is it, I have exhausted my coefficients, ok.

1
Now, if I want to go to the s row what do I do? I multiply 2 with 3; I get 6, I multiply 1
and 1 I get 1. So, 6 minus 1 is going to be 5 and I divide 5 by 2 so, I get 5 by 2 ok. So, would
1
I have any other entry in the s row? No, because there is nothing to the right of the
0
column 3 and 1; that is it I stop here ok. Now, I go to the s row so, now then I consider
the s 2 and s 1 row right, for the s 0 row.

188
So, what do I do? I do the same process; 5 by 2 times 1 minus 2 times, you know when you
do not have an entry you take it as 0 ok; that is once not the first entry, but then the other
entries right. When you have reached s 1 row you do not have any entry here right so, you
take it as a null entry. So, 5 by 2 times 1 minus 2 times 0 is 5 by 2 and then 5 by 2 you divide
by 5 by 2 itself, what will you get? You will get 1.

Rouths array:

s3 1 3

2
s 2 1

s1 5/2 -

0
s 1 -

So now, let us look at the first column of the Routh’s array right. What can you observe?

Student: No sign changes.

No sign changes right because you one can observe that all entries are positive. So, this
implies that all 3 roots of d1( s ) have negative real parts. So, I hope it is clear how we
applied it here right. So, please note you know I am not calculating the exact roots per se
right. Did I ask you to solve the polynomial for the roots? No, right that is what I am
reiterating here right, at this stage we are only interested in figuring out the location of the
roots ok. So, we did not solve for the roots. So, I am we are only concluding that all 3 roots
are in the left half complex plane, where we need we need to solve the equation right to get
the exact values of the roots ok.

So, that is that is the first polynomial. So, now, let us go to the second polynomial. So, what
happened in d 2 ( s ) ?

d 2 ( s )=s 3 +2 s 2+ s+ 3

So, let us do the Routh’s array construction once again s 3 , s2 , s 1 , s0 right. So, we have once
again we fill in the first two rows 1 2 1 and 3 right.

189
Now, we want to calculate for the s 1 row. What do we do? We multiply 2 with 1 subtract 1
and 3. What are we going to get?

Student: [Inaudible]

-1 right, -1 I divide by 2 I will get -1 by 2 and that is it right I I have exhausted my

0
coefficients. So, we come to the s row, for the s power 0 row we consider these two rows
right. So, minus 1 by 2 times 3 is going to be minus 3 by 2, minus 3 by 2 minus of 2 times
null entry is going to be just minus 3 by 2 and minus 2 by 2 divided by minus 1 by 2 is going
to be.

Student: 3.

3. So, we immediately look at the first column of the Routh’s array. What can we see?

Rouths array:

s3 1 1

2
s 2 3

s1 -1/2 -

0
s 3 -

Student: 2 sign changes.

There are 2 sign changes. How come there are 2 sign changes? You see that from 2 to -1 by 2
is 1 sign change, from -1 by 2 to 3 here is another sign change ok. So, that is how we get 2
sign changes here ok. What does it imply?

Student: [Inaudible]

So, this implies that there are 2 roots in RHP and obviously, where should the third root be?
In LHP right so, plus 1 root in LHP ok. Yes please.

190
Student: [Inaudible]

No, why not? That is a good point ok. How come in this example, we are sure that it is not in
the on the jω axis ok. So, let us argue from two perspectives ok. So, that is a good
question, let me repeat this question. So, if you if you logically apply whatever we have
learnt till now, you know like the only conclusion we can draw based on the facts that we
have written under the Routh’s criteria is that there are 2 sign changes. So, 2 roots should be
in the right of plane. So, this is a third order polynomial right so, totally there are going to be
3 roots. So, 2 roots are in the right of complex plane so, we are left with only 1 root ok.

Where does the 1 remaining root lie? Obviously, there are two ways to eliminate the jω
axis; first is by looking in the Routh’s array you see that there are no 0 entries ok. So, as I told
you tomorrow we look at special cases you know that rules out the jω axis. Second thing
you can look at the polynomial itself, see this is a third order polynomial right you have 1
remaining root. So, if I want the 1 remaining root to be on the jω axis right, I can have
two possibilities. Either it should be the origin or it should be a complex conjugate pair on the
jω axis right. Complex conjugate pair is ruled out, why?

Student: Because only 1.

Because I have only 1 remaining root, complex conjugate pair means I need 2 more
remaining roots right that is gone and so, the only possibility which is left is at the origin. If
you have a root of a polynomial at the origin the constant term must be 0. Is the constant term
0 here? No alright. So, that is the second line of argument so, we are ruling out j omega axis.

Student: Because we got 2 roots, we got 1 root [Inaudible]

You may have, but once again that is a good point. Let us say we have an example, there is
only 1 sign change right, let us say a third order polynomial we have only 1 sign change.
Question is I am left with 2 remaining roots, I can still have let us say a complex conjugate
pair on the jω axis; possible. But still in that example if you in your first column of the
Routh’s array, if you do not have a 0 that you are you rule out that possibility ok. As I told
you we are going to do that tomorrow right. So, yeah I think that is a good question ok, I
hope you do not follow that result right yeah ok.

191
(Refer Slide Time: 08:26)

So, let me write down the general Routh’s stability criteria then we will go back to our
example that we are doing it last class and complete it ok. So, let me write down the Routh’s
stability criteria ok. So, you see that whatever we have been discussing till now applies to any
polynomial right; you can encode polynomials in any application. Now, we are going to apply
it to control theory ok.

So, I hope you can see how mathematics is applied to our subject of discussion right, which is
control design right. So, we will now see how we are going to apply it to our thing.

Routh’s Stability Criteria:

n(s)
Consider a system whose transfer function is . This system would be BIBO stable
d (s)
(asymptotically stable) iff:

a) All coefficients of d(s) are non zero and of the same sign.

b) The first column of the Rouths array constructed using the coefficients of d(s) has all
entries being non-zero and of the same sign.

So, once again please note that BIBO stability means you know like you cannot have poles
on the imaginary axis also right, that is ruled out. So, I cannot say I will have a 0 on the first
column; no I should have non-zero entries in the first column and they should all be of the

192
same sign ok. So, those are the two conditions as far as the Routh’s stability criteria is
concerned ok. This is what is called as Routh’s stability criteria. So, we will go back and now
apply this Routh’s stability criteria to the problem that we were doing ok.

We are doing this problem right, where we were looking at the PI controller and we figured
3
out that the closed loop characteristic equation is s + ( K p +1 ) s + K i=0 right.

(Refer Slide Time: 13:00)

So, now, the roots of this polynomial will be the closed loop poles. So, do you think the PI
controller will stabilize this closed loop system? No why not?

I do not even need to go any further right, why not the necessary condition is not satisfied
right. the coefficient of s2 is zero right. So, this implies that there are no values of K p and Ki
can stabilize the closed loop system that is the implication of this result correct. So, the PI
controller will not work. So, now, you see why we digressed and learn the Routh’s stability
criteria and came here right.

So, you can see that our analysis becomes more simple right. So, now, as the next step let us
try the PD controller because, for this system let us go ahead and try the PD controller ok. So,
PD controller means its transfer function is Kp + Kd s. So, we follow the same framework ok.

K d s+K p
G ( s )=C ( s ) P ( s )=
s 2+ 1

193
(Refer Slide Time: 15:03)

K d s+ K p
G(s) s2 +1 Kd s + K p
C.L.T.F = = = 2
1+G ( s ) H (s) K d s + K p s + K d s+(K p +1)
1+
s 2 +1

So, immediately we see that this is a polynomial of order 2 right; the closed loop system with
a derivative PD controller is still the second order system ok. With this you know, like
immediately we see that as I discussed for polynomial still order 2 the necessary and
sufficient condition is all coefficients must be non-zero and of the same sign ok. So, this
implies that the closed loop system would be BIBO stable if and only if the all the
2
coefficients must be non-zero and the same sign right; s term the coefficient is plus 1. So,
already 1 coefficient is positive. So, we must have K d > 0, Kp + 1 > 0. So, this implies that K d
> 0 and Kp > -1 is the feasible region for stability.

So, if I want to plot this in the controller parameter space, we have K p on the horizontal axis,
let us say Kd on the in the vertical axis right. So, how will I plot K d > 0? Kd greater than 0 is
this half plane right, isn’t it right. How do I plot K p > - 1? First I draw K p equals minus 1 and
any region to the right of it is going to be K p greater than minus 1 right. So, consequently the
feasible region is going to be the intersection of the two regions right correct; this is the
feasible region. Alright this is what I called as a controller parameter space right.

194
If I say calculate the feasible region in the controller parameter space, this is what you need to
do for the controller design, fine. So, as a very simple exercise for you to do, consider the
general second order polynomial a0 s 2 +a1 s +a2 . Apply the Routh’s array to that and
convince yourself that the necessary condition is also sufficient ok. You will see that the first
column will have a0 a1 and a2 ok. So, what was necessary will also become sufficient for
polynomial still order 2 ok. So, that is the, what to say, criteria for the for orders polynomial
still order 2 ok. So, is it clear? Yeah.

195
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 27
Special Cases of Routh’s Stability Criterion
Part – 1

Let us continue with our discussion on the Routh’s criteria. In the last class, we started
off with the question as to how one could obtain the distribution of roots of a polynomial.
We figured out that we can construct the Routh’s array and check the number of sign
changes the first column of the Routh’s array. This can be used for control system design
to check for what values of controller parameters the closed loop system would become
stable. We form the Routh’s array for the closed loop characteristic polynomial and then
figure out the feasible regions for the controller gains. Let us see a few examples to point
out some special cases that may come up with the construction of the Routh’s array and
then how to handle them.

(Refer Slide Time: 01:20)

The first case is when the first element in any row of the Routh’s array is zero with the
other remaining elements or entries being non-zero. Let us consider the polynomial,

𝐷(𝑠) = 𝑠 4 + 3𝑠 3 + 3𝑠 2 + 3𝑠 + 2

196
We can observe that all coefficients are non-zero and of the same sign. So the necessary
condition is satisfied in this particular polynomial. Let us construct the Routh’s array.

If we want only to figure out the distribution of roots we can continue even when that
necessary condition is not satisfied. That is the necessary condition for stability.

𝑠4 1 3 2

𝑠3 3 3

𝑠2 2 2

𝑠1 0 (𝜀)

𝑠0 2

We can see that the first entry in 𝑠1 row is zero. Now we follow the procedure below.
Replace the zero entry with a “small” positive real number 𝜀 and continue the process.
And from here, once we continue this process, if there is a sign change from the entry
above epsilon in the 1st column of the Routh’s array to the entry below epsilon then
there are roots in the RHP. For example, if the entry above epsilon is positive and the
entry below epsilon is negative, then there are roots in the right half plane. If there is no
sign change from the entry above epsilon to the entry below epsilon then there is a pair
of roots on the 𝑗𝜔 axis or the imaginary axis

Let us see what happens in the current example. We are going to replace the 0 with a
small positive number positive real number 𝜀 and continue the calculation. We get 2 in
the 𝑠 0 row. If we look at the entry above epsilon and the entry below epsilon, we can
observe that there is no sign change from the entry above epsilon to the one below
epsilon in the first column of the Routh’s array. This implies that there is a pair of roots
on the imaginary axis. If we calculate the roots of the polynomial, we get −1, −2, ±𝑗.
We can see that out of the 4 roots there are 2 roots on the 𝑗𝜔 axis the remaining 2 roots

197
are in the left half plane. Without solving for the roots, we can get the distribution of
roots.

Let us do one more example;

𝐷(𝑠) = 𝑠 3 − 3𝑠 + 2

We see that not all the coefficients are 0 and of the same sign, but find out the
distribution of roots. Let us carry out the construction of the Routh’s array.

𝑠3 1 -3

𝑠2 0 (𝜀) 2

𝑠1 −3𝜀 − 2
2

𝑠0 2

Now, if we look at the elements below and above epsilon, the element above epsilon has
a positive sign and the element below epsilon has a negative sign. There is a sign change
below and above epsilon. We need to count that as one sign change. Then from 𝑠1 row to
𝑠 0 row in the first column, there is another sign change. So there are 2 sign changes. This
implies that there are 2 roots in the right half plane with the remaining root in the left
half plane. That is the conclusion we can draw here. The roots of the polynomial are
−2,1,1.

198
(Refer Slide Time: 14:13)

199
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 28
Special Cases of Routh’s Stability Criterion Part – 2

Now, I am going to do a second special case where we ask ourselves is a question.

(Refer Slide Time: 00:19)

What happens when an entire row of the Routh’s array is 0 ok; so, we asked ourselves the
question what happens when the first entry in any row is 0, but let us say the entire row is 0
ok, then what do we interpret from that ok?

So, let me do an example to convey that point ok. So, let us consider a polynomial; so, that is
why in order to essentially address his question ok, I am going to give you a polynomial,
where the necessary condition is satisfied see I just deviated from it to just show you an
example ok, but that is why you know I wanted to show you that just by observation
sometimes you can’t generalize you know for polynomials of order 3 and more ok.

So, let us consider the polynomial d ( s )=s7 +6 s6 +11 s5 +6 s 4 +4 s3 +24 s 2+ 44 s+24 . So, of
course, I could have taken a simpler polynomial and it would have appeared a simpler right.
So, with lot of coefficients being 0, but I purposefully constructed an example where, the
necessary condition is satisfied right. So, do you agree that all coefficients are non zero and

200
of the same sign, yes. So, let us see suppose if you have a 7th order system ok, whose transfer
function denominator polynomial is this ok, then how do you find out its stable or not right.

So, now we see the usefulness of the Routh’s criteria ok, for control system design right
ultimately we wanted see the, this course is not about what to say general math right, so how
do we apply those mathematics to control design right. So, that is what we are interested in
right. So, let us let us construct the Routh’s array and then like see what happens right.

So, let us write down the powers of S and you help me in constructing the Routh’s array ok.
So, what do we do, we fill in the entries in the first two rows right. So, that is going to be 1 6
11 6 4 24 44 and 24 ok, so that is what is going to happen. Now can you help me with the
numbers starting from the s5 row what am I going to get.

Student: [Inaudible]

If I want to get this you know like I multiply 6 and 11 subtract 6 and 1 divided by 6 what will
that give me?

Student: 10.

Ten. Then next entry I multiply 6 and 4 subtract the product of 1 and 24 divided by 6 what
will that get me, zero. Then what will what will I get here I have multiply 6 with 44 subtract a
product of 1 and 24 divided by 6 what does that give me.

Student: 40.

5
40 right. So, that is about it right, so as far as a s row is concerned. Now let us go to the
4
s row. So, what would I get in the s 4 row.

Student: 6.

Ah, I would get 6 as a first entry, then I would get 0 then 24 right. Now we go to the s3
row what will I get.

Student: 0.

0 and.

Student: 0.

201
0. So, now, we have an issue right. So, that we have come to a stage where we have the entire
row of the Routh’s array to be 0, what do we do ok. So, before we figure out what to do ok,
let us learn a term which I am going to use here ok, what are called as radially opposite roots
ok. So, what can one mean by a radially opposite roots.

See for example, let us say I take the S plane. So, if I consider the S plane and let us say I
consider -3 and +3 that is a pair of radially opposite roots right. So, essentially, you look at
the origin you draw a circle right. So, any pair of roots which lies on the circle, relevant pair
is a radially opposite pair right, -3 and +3 form a radially opposite pair right.

Similarly I can have let us say 2j and -2j form a radially opposite pair. Do you agree? So, that
is a radially opposite pair right. So, on the imaginary axis, same condition holds true. Then I
can also have minus 3 + 2j and 3 - 2j form a radially opposite pair right correct, but the
moment we since we are dealing with polynomials which have real coefficients, we can have
only complex roots that are whose conjugates are also roots.

So, if I have such a scenario, I need to have -3 - 2j and 3 + 2j to also be roots of that
particular polynomial ok. So, these are three ways in which we can have radially opposite
roots ok. So, radially opposite roots can either be multiples of in general multiples of 2 they
can occur as 2 roots or 4 roots ok. When can they occur as 2 roots, of course, non repeating
pairs right.

So for example, if they lie on the real axis which are radially opposite or on the imaginary
axis and if they, if you have radially opposite roots essentially lying anywhere else what will
happen is that like you are going to have like this right - 3±2 j and 3 ±2 j that is also a radially
opposite set of 4 roots ok, but essentially they are radially opposite to each other right ok. So,
that is the definition of radially opposite roots.

Now when an entire row of the Routh’s array happens to be 0 that implies that the polynomial
is going to have radially opposite roots, which of these patterns we are going to evaluate ok,
using the Routh’s array, is it clear? let me restate it once again, if the entire row happens to be
0 this implies that we are going to have radially opposite roots ok, so that is what is the
implication ok. So, then how do we process this right. So, that is what we are going to do.

So, let us let us learn what we do. So; obviously, I cannot proceed any further right. So, I i am
stuck here in s3 row. So, what I do is that, I construct what is called as an auxiliary

202
polynomial. Auxiliary polynomial Pa(s) is equal to. For looking at the auxiliary polynomial I
look at the row immediately above, the row in which I have all those entries to be 0 ok. So,
let me write down their auxiliary polynomial then we will understand how I wrote it right.

4
So, here the coefficient 6 corresponds to s ok, and as I go one coefficient to the right the
power of S decreases by 2 ok, that is how we wrote down the coefficients of the Routh’s
7
array, because see if you look at the very first row, the coefficient one corresponds to s ,
11 corresponds to s 5 all right. So, and then like 4 corresponds to s 3 , 44 corresponds to S
right, so as you keep on shifting one element to the right in any row, you know like in essence
we are decreasing the power of the corresponding S term by 2.

So, in the auxiliary polynomial I start with 6 which corresponds to s 4 and 0s2. So, I just do not
write anything, and then from s2 I just go to a constant term, s 0 is constant right. So, the
auxiliary polynomial becomes,

Pa ( s )=6 s4 + 24

And once again I am just only stating the process and the results. I am not proving anything
here right, so these have been well established ok.

So, and it so turns out that the roots of the auxiliary polynomial give us the radially opposite
roots ok. So, if you solve for the roots of the auxiliary polynomial we get the radially
opposite roots. So, in this case what do you think the radially opposite roots, if you calculate
6 s 4 +24=0 . So, that gives me s 4 +4=0 right. What is the solution of s 4 + 4=0 ?
Anyway we will come there ok.

So, as I told you we are one first interest in location then we will double check right. So, that
that is the general result ok. So, if you have the entire row of the Routh’s array to be 0 what
you do is that, you look at the previous row; extract the coefficients from that and then form
the auxiliary polynomial. Then what you do is that, you take the derivative of the auxiliary
polynomial with respect to s.

d Pa ( s)
So, we do =24 s3 . So, that is going to give me 3
24 s +0. that is it all right, when
ds
I take the derivative with respect to s. Do you agree ok? So, what I need to do is that, I will
replace in the s3 row with 24. Of course, this zero I will keep it as it is, because the

203
 3
constant term is 0 right. So, what I have done using this process is that in the s row the
first element has been replaced with the 24, but the second element still remains 0 ok. So, I
have to basically continue with that right. So, let us continue the process, let us go to the
s 2 row what will I get? If I go to the s 2 row what will I get? 24 times 0 minus 6 times 0
by 24 what is it, 0 right let us continue and what will come here 24 times 24 minus 6 times
null entry by 24 what will that be, 24 right. So, now, what should I do?

Student: [Inaudible]

Epsilon ( ε ) right, what we just learned. So, now, you see that we have come to a case
where the only the first element is zero, there are nonzero elements in the row all right. So, so
I replace the first element zero by an ε and then continue right. So, what will I get? ε
being a small positive real number right.

So, I will get ε times 0 minus 24 times 24 that is going to be 576 divided by ε . So, I
am going to get -576/ ε all right, and then that is it right in s 1 row if I come to the s0 row
what am I going to get?

Student: 24

I will just get 24 right. I have completed the construction of the Routh’s array, now we look at
the first column. Now if you look at the element above and below ε , what can you tell
about their signs. The element above ε is 24, the element below ε is going to be
negative in sign right -576/ ε is going to be negative right. So, do you think there is a sign
change above and below ε ? So, that implies that there are roots in the right of plane.

Student: [Inaudible]

So, this implies that there are two sign changes, which implies that 2 roots in the right of
plane.

Student: [Inaudible]

At this 24, yeah you multiplied 24 with 24 minus 6 with null entry divided by 24 right. So, is
it clear yeah. So, I i hope my calculations are correct right. So, right I hope this array is
correct right. So, yeah, so that is what we get. So, we get 2 roots in the R H P right, and it so
turns out that if you essentially solve s 4 + 4=0 you will get -1±j and +1±j as the 4 roots ok.

204
 s
s
So, how can I factorize s 4 + 4 , can I do this, can I write it as (¿¿ 2−2 s+2) So, is this
(¿¿ 2+2 s+ 2)¿
¿
correct; so, what are the roots of this polynomial: -1±j and +1±j.

So, you see that we are going to have a case where we have 4 roots which are going to form a
radially opposite set. So, that is why we have this sort of array. So, consequently two roots
are in the R H P and of course, the remaining 5 roots are going to be in the left of complex
plane, because we have ruled out roots on the imaginary axis and there is no root at the origin
also. How can you conclude that there is no root at the origin? We discussed this point
yesterday right.

Student: [Inaudible]

Constant term is nonzero. See if you have a root of the origin constant term is zero right so,
but then the constant term is nonzero here right. So, there is, no root at the origin right. So,
that is how we figure out things. Is it clear?

Student: [Inaudible]

No ε ok. So, -576/ ε times 24 right, times −ε with null entry. So, you will just get
-576/ ε divided by 24, you divide by -576/ ε cancels off right, you are just left at 24
right yeah ok. So, these are two special cases which I wanted to look ok; any questions on
this?

So, please try out like, kindly try out more examples and the more you try and more you
come and discuss with me you know like, I am sure both of us will learn together. See I am
also trying out cases, you also try out cases right you construct examples where you have
radially opposite pairs roots on the imaginary axis and then like a mix of both right roots on
the imaginary axis and roots on the right of plane right and then like try things out yeah fine.

So, this is the Routh’s stability criterion; now we are going to come back and we are going to
see how to apply it to our case ok. So, I am essentially going to talk about application of the
Routh’s stability criteria to our case.

205
(Refer Slide Time: 19:21)

So, we are going to talk about the application of on the Routh’s stability criteria ok. So, we
already did one example yesterday right, wherein we essentially looked at how do we get the
feasible region of the controller parameters right, to give us close loop stability right, so that
is something which we did yesterday right. So, if I go down this is the example which we did
right, so we essentially used the Routh’s criteria to figure this one out right.

(Refer Slide Time: 19:51)

So, let, today let us do one more example and then we will see how to go forward right. So,
let us say I want you to design a negative feedback control system for a plant which has a

206
 1
transfer function like this; 2 ok, let us say we consider the unity negative feedback ok.
s −1
Yes please.

Student: [Inaudible]

7 roots yes exactly.

Student: [Inaudible]

That is only for the auxiliary polynomial

Student: [Inaudible]

I did not get you.

Student: [Inaudible]

Two rows for sign changes yeah. So, 2 out of the 4 roots are going to be the right of plane
right that is why you have two sign changes, they are accommodated right. So, that is why I
made a statement that there are 5 roots are in the left half complex plane, for the original
polynomial. See the auxiliary polynomial if you solve it that will give you the radially
opposite roots right. So, essentially you have four radially opposite roots; two in the left half
plane two in the right of plane ok.

Now there are three remaining roots, for the original polynomial they are going to be in the
left half plane, because you are already accounted for two right half plane roots because there
are two sign changes all right. So, that is why I made a statement that this, from this I can
observe that there are two roots in the right half plane, 5 roots in the left half plane ok.

Student: [Inaudible]

Let me repeat it ok. So, I hope the construction of the Routh’s array is clear right and all of us
can agree that there are two sign changes right. Two sign changes means there are two roots
in the right half plane right. So, yes please pay attention.

So, you can see that the auxiliary polynomial is of fourth order. So, you are going to have 4
radially opposite roots. Here we can see that you know the 4 radially opposite routes are
going to be -1±j and +1±j right.

207
(Refer Slide Time: 22:33)

So, there are two roots in the right of plane, two roots in the left half plane for the auxiliary
polynomial right. So, that which constitutes a radially opposite set. There are 3 remaining
roots which are once again going to be the left half plane why, because if you look at the
epsilon entry and you look at the entries below and above epsilon since there is a sign
change , we conclude that there are roots in the right half plane ok.

So, if the first entry is a 0 and you replace it with epsilon, we discussed that there are two
cases right. If there is no sign change that indicates a pair of roots on the imaginary axis, there
is a sign change then roots in the right half plane; that is the criteria used

Student: [Inaudible]

The roots of the auxiliary polynomial forms a subset of the roots of the original polynomial.

Student: [Inaudible]

You can factorize it. So, yeah I am not proving it ok, I am going to using the result ok, good
question ah. How do we show that the roots of the auxiliary polynomial are the roots of this
one; you can prove it in general ok.

So, I am not doing it doing it here, but if you want to convince yourself you can substitute the
roots in the polynomial, you will see that it will go to 0 right, you can do that check right
yeah that is a best way to check right whether some value is a root or not right yeah

208
Student: [Inaudible]

Yeah

Student: [Inaudible]

7 roots right; totally there are 7 roots, 2 roots in the R H P right

Student: [Inaudible]

Which are these remaining 5 roots in the LHP, out of which 2 are -1±j, I do not know about
other three yeah, but of course, I can write here ok. Of course, since I know the answer right.
So, now, I am going to write the roots -1, -2, -3, -1±j, 1±j.

Of course, that is how I construct the polynomial, its just not that you know I just wrote the
polynomial arbitrarily, I use the poly command in MATLAB right to generate this polynomial
by trial and error ok, yeah fine yeah. So, please check all the questions that you are getting
are very common which I also have and I have essentially what to say figured them out right.

So, I want you to do more examples to essentially figure these things that is how we learn
fine ok. So, I think what I am, what I am going to leave you with homework is that like look

1
at this plant ok. The plant transfer function is 2 ; obviously, its an unstable plant or
s −1
system right, because you have roots at +1 and -1, trying to stabilize it using P PI PD PID
whichever controller works, figure things out and then like use the Routh’s stability criteria to
essentially figure out which, what ranges of controller gains or parameters would give you
closed loop stability ok. So, we would come back in the next class and then like look at
closed loop performance, now that we have figured out how to evaluate closed loop stability
we would figure out how to evaluate closed loop performance right. And see, in many
references you will find the term absolute stability ok, the condition that we need to have all
the roots of the closed loop poles of the closed loop transfer function in the left of complex
plane is what is called as absolute stability ok.

Later on we will look at what is called as a relative stability that is we want roots in the left of
complex plane, but even a little bit further to the left from the jω axis. As I discussed
already poles are once again an approximation right. So, you can have unmodeled dynamics,

209
parametric uncertainties and all so the poles can shift a little bit right. So, I do not want a pair
of poles at –0.01±j right; that is going to be very very close to the imaginary axis right.

So, relative stability talks about that those questions ok. So, can we push the poles a little bit
farther to the left, but there is a price to pay ok, we will see what is the price and all when we
look at relative stability ok. So, we will stop here.

210
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 29
Performance Specifications Part-1

Let us discuss on consideration of performance specifications during control design. Till now,
we have looked at the plant or system block and the controller block. Then we have seen how
to characterize the plant or system using mathematical models and obtain the transfer
function of the system. And then how to calculate the response or the output of the system for
various standard inputs. And based on those inputs how to quantify performance parameters
for various types of systems. And then we looked at stability, we figure out that for BIBO
stability, we require all the poles to be in the left of complex plane. And the Routh’s criteria to
get ranges of controller parameters that would provide closed loop stability. And we looked at
various choices for controllers that we have and we did a few examples on how to control the
closed loop system.

(Refer Slide Time: 01:39)

The question that we are trying to answer today is that how can we convert performance
specifications into desired regions of closed loop poles. We figured out that if we want closed
loop stability, we need all the closed loop poles in the left half complex plane. Now, in
addition to closed loop stability if we have certain closed loop performance specifications to

211
be met, how do we select a region in the left of complex plane where we should have my
closed loop poles?

We are going to look at it from two perspectives. We will take an example and then
demonstrate.

Let us say that it is desired that the “dominant dynamics” of a closed loop control system
behave like that of an underdamped 2nd order system with a settling time of at most 4 s, and
a maximum peak overshoot of at most 10%. By dominant dynamics here means that we want
two poles which are closest to the imaginary axis to behave like those of an underdamped 2nd
order system with a settling time of at most 4 s, and maximum overshoot of at most 10%.

(Refer Slide Time: 05:18)

Recall that for an underdamped second order system, the settling time corresponding to the 2
percent settling criteria was given by

4
t s=
ζ ωn

So, according the condition,

4
≤ 4, ζ ω n ≥ 1
ζ ωn

−ζ ωn ≤−1

212
So, recall that the real part of the poles of an underdamped 2nd order system is −ζ ωn .
This means that if we have dominant poles with real part, which is ≤−1 , we can meet the
settling time criteria. This is drawn as a vertical line, which is parallel to the imaginary axis at
-1 in the s-plane as shown in the figure. The conclusion means that if we have dominant poles
to the left of the line passing through -1, then settling time is going to be less than 4 s. Now,
we have constructed a sub region of the left half complex plane, in which we want closed
loop poles to lie.

Now, let us use the second criteria of maximum percentage overshoot. The maximum peak
overshoot is given by

−ζπ
√ 1−ζ 2
M p=e ∗10 0 .

This should be less than or equal to 10%.

−ζπ

e √1−ζ ≤0. 1 .
2

Simplifying the above relation,

ζ ≥ 0.5911 .

(Refer Slide Time: 17:40)

213
Recall that ζ is related to an angle called β . β was the angle made by the two poles
with the negative real axis. The relationship is ζ =cos β .

cos β ≥0.5911, β ≤53.76 °

The graphically visualization of beta is shown in the figure above. It is drawn as an open
cone which makes this angle of 53.76 degrees on both sides of course. This means is that as
long as the poles lie in this cone, the constraint on maximum peak over shoot will be
satisfied. We graphically visualize the damping ratio in the complex plane by using this angle
β .

Now we want to meet both conditions. We have to take the intersection of the two regions.
This is the shaded region in the above figure that is going to be like an open trapezoid. As
long as we place the closed loop poles in this region, we will be guaranteed both conditions
on settling time and maximum peak overshoot. That is the manner in which we have
converted conditions on performance to a desired region of closed loop poles. Note that we
can place the closed loop poles anywhere in the region for our conditions on performance
requirements to be met. And note that this feasible region is a subset of the left half plane. We
always talk about stability first and then performance. So, if we place the closed loop poles in
this desired region, we guarantee stability anyway because it is a subset left half complex
plane.

Now we have a challenge to make sure that all our closed loop poles lie in this open
trapezoidal region. Till now we just did examples where we just wanted closed loop stability,
we use the Routh’s criteria and figure out what conditions on the controller parameters would
give all roots of the characteristic polynomial to be in the left half complex plane. Now we
want to have conditions on the controller parameters which would place the closed loop poles
in this desired region. We are going to make use of Root locus methods for this purpose.

This can be also used to evaluate what is called as a relative stability. There are broadly two
notions. What we have been looking at is called as absolute stability. Relative stability means
we do have stability but then we do not want to be too close to the imaginary axis because
there can be un-modeled dynamics that can be parametric variations and so on. So the
locations of the poles may change a little bit. So, then what happens is that like we introduce
what is called as a relative stability, where we want poles to be in the left half complex plane,
but I do not want them too close to the left or complex plane.

214
For example, that place all closed loop poles to the left of -0.5. Then we are drawing a
vertical line at -0.5 and placing the closed loop poles on the left of that. The desired region is
offset. If we want relative stability, we want all poles to be to the left of -0.5. We can use Root
locus or Routh’s criteria for this. We will revisit these later in the course.

(Refer Slide Time: 27:17)

Let us see a simple example exercise to do. Let us say the closed loop characteristic
polynomial is

s 2 + ( K p−3 ) s+ 2 .

For absolute stability, we need K p >3.

If we want all poles to be to the left of minus 0.5. We substitute s= s^ + 0.5 and evaluate for
s^ . From this we will get the range of K p which will ensure this relative stability.

215
 Control Systems
Prof. C. S. Shankar Ram
Indian Institute of Technology, Madras
Department of Engineering Design

Lecture – 30
Performance Specifications
Part 2

So, that is one concept which I wanted to introduce, yes.

Student: [Inaudible]

(Refer Slide Time: 00:34)

All right, so, see what I mean is that see let us say you have a tenth order system for example,
right. So, you need to design a closed loop control system which is order of 10, but still I can
always say look I will specify performance in this manner; see the reason is a following,
right. So, we have explicit expressions for specific performance specification for systems of
order 1 and 2 and by and large the dominant dynamics is going to be of that order, all right.

So, consequently, I can essentially create regions like this and if I am able to place all my
poles in this region, I am done ok, we will double check ok. Obviously, I would place the
dominant pair as close to the imaginary axis as possible right is it not, then only they will
remain dominant see think about it. We will do all this design process when we learn about
root locus and do the design examples; see when I am using maximum peak overshoot and

216
settling time for an under damped second order system, I am implicitly assuming that my
dominant poles are going to be a pair of a complex conjugate poles, right, which should be
the closest to the imaginary axis right. So, when I design I should ensure that the dominant
poles basically are complex conjugate and they are the closest to the imaginary axis that is
something which I should ensure, right. So, by and large it works that way. You can say look I
do not want dominant complex conjugate poles, I will have a dominant poles on the negative
real axis fine you are still satisfying the performance requirements.

But what will happen is that the system response will become sluggish that is what will
happen because like your rise time value will go up as we discussed earlier right, yeah. So,
this is just what to say a I would say an engineering solution right to visualize a domain in the
s plane where you can place the closed loop poles; that is the application right. Of course, if
someone says look design a closed loop control system whose time constant is let us say less
than 4 seconds what will we do? The moment I say time constant I am telling you that the
dominant dynamics will that will be that of a first order system and if time constant is 4
seconds what should happen what was the pole of a first order system: -1/T. right for a stable

1 −1
first order system. So, if I want time constant of at most four seconds, = . So, I want
T 4
all my poles to be to the left of -0.25 line ok. So, that is how we should visualize, is it clear.
So, that is an alternative example, right.

So, if I want my performance specification I am saying that look I want a time constant of let
us say less than 4 seconds; then automatically we look at the step response of a stable first
order system whose pole is at -1/T. So, time constant of at most four means my poles should
be to the left of minus 0.25 line; that is it, right. So, that will ensure that I satisfy that criteria
ok. So, that is the way we construct and then like we look at perform specification ok.

217
(Refer Slide Time: 04:13)

So, one more thing which we are going to do today at least start today is analysis of steady
state errors ok. So, before we go we put everything together and then like start doing
controller design. So, this is something which we are going to discuss today ok. So, how do
you analyze steady state errors because by and large you know like we design controllers to
do some let us say regulation or tracking right; follow a desired behavior. See for example, I
would design the air conditioner in this room to meet a set temperature right. So, I will come
and set the temperature as 25 degree Celsius.

I want my controller to maintain the temperature of air near to 25 degree Celsius right;
obviously, I would be happy if as time progresses I stay near 25 degree Celsius, yes, right is it
not, that is when I will tell that my system or controller is performing well, right. So, steady
state errors become very important, right; so, and they become very critical, right.

So, say for example, I am developing an autonomous vehicle, right, I wanted to essentially
follow a particular trajectory I wanted to be an autonomous car which has to travel within a
lane, alright. So, I do not want the car to oscillate in the lane and also like deviate from the
path, right; that is going to have serious implications right. So, as time progresses what is
going to happen to that vehicle right. So, that is something which we should figure out. So,
let us let us look at steady state errors. So, how do we analyze steady state errors? Let us
construct the standard closed loop negative feedback system that we have been looking at.

218
So, this is the feedback path transfer function and this is the feedback diagram right. So,
please note what was G(s) called that is what is called as a forward path transfer function
right H(s) is the feedback path transfer function, right. So, recall that the closed loop transfer

Y (s) G(s)
function is essentially equal to =
R (s) 1+G ( s ) H (s)

So, that was the expression for the closed loop transfer function ok. So, we are going to
essentially make an important assumption ok, let us assume that the closed loop system is
stable ok. So, of course, then only we can talk about steady state errors right think about it
right, let us say if I do not stabilize the closed loop system there is no point in talking about
steady state performance right.

So, let us assume that the closed loop system is stable ok. So, that first requirement of
stability the critical requirement of stability has been met right that is an assumption we will

E (s ) 1
make before we proceed right. So, immediately we note that =
R (s ) 1+G ( s ) H (s)

So, what is the name that we associate to the function G ( s ) H ( s) ; open loop transfer
function, right. So, G(s) is what is called as a forward path transfer function right H(s) is what
is called as a feedback path transfer function right. So, G ( s ) H ( s ) is what is called as the
open loop transfer function.

Please remember these terms. see of course, in your quizzes and exams I will not give an nice
block diagram and then like give you expressions right I would say design a negative
feedback system and if I say unity; that means, H(s) = 1 right, I may say the plant transfer
function is something controller transfer function is something. So, then you should be able to
figure out G(s) and so on right correct.

So, please remember these terms, right. So, this implies that the steady state error ok, let me

sR (s)
call it as e ss =lim e ( t )=lim sE ( s ) =lim
t→∞ s→0 s →0 1+ G ( s ) H ( s)

So, that is what we have for the steady state error .

219
(Refer Slide Time: 10:29)

b0 s m +b 1 sm−1 +…+ bm−1 s+ bm

Now let G ( s ) H ( s )=
a 0 sn +a 1 sn−1 +…+ an−1 s+ an

So, let us say G( s ) H (s ) comes out to be in this generic form. Please note that I have
already have closed loop stability ok, I am reiterating that point again and again yeah. So,
now, let us say we can rewrite this as

a
¿
(¿ 0 ¿ s p + a^1 s p −1 +…+ a^ p−1 s+ 1)
s n ¿^ So, I what I
b0 s m +b 1 sm−1 +…+ bm−1 s+ bm K ( b^0 sm + b^1 sm −1+ …+ b^ m −1 s +1)
G ( s ) H ( s )= =
n n−1
a 0 s +a 1 s +…+ an−1 s+ an ¿

am doing is that I am doing two things, first thing is and I am seeing whether there are some
open loop poles at the origin.

See for example, what are open loop poles? There the poles are the open loop transfer
function right. So, if I look at the denominator, if I have poles at the origin then a n is 0;
obviously, right. So, I can pull some s term out, common. So, I can have 1 pole at the origin 2
poles at the origin 3 poles of the origin or none ok.

220
So, the value of capital N can be from 0, 1, 2, 3 and so on right. So, I am sure all of us agree
correct, once I pull it out what I do is that remaining factors also what I am doing is that I am
making the constant term as 1, I do some algebraic manipulation and then introduce a factor
K such that the constant term becomes 1 we will see why right shortly.

So, this is the general structure of the open loop transfer function.

When, N=0 → Type 0 System

N=1 → Type 1 System

N=2 → Type 1 System

So, when a type 0 system does not have any open loop poles at the origin ok, yes.

Student: [Inaudible]

Yes, they are different coefficients right that is why I am putting a hat over it, right.

So, I am just doing some algebraic operations I am just writing in a general form; so, that you
will see why I am doing it because I want the constants to be 0.

But because in the two polynomials when I substitute s = 0 because that is where I want how
I would use my final value theorem right I want to end up with just 1 and 1. So, consequently
I am putting whatever that ratio which comes out as K right I would say ok. So, depending on
the value of capital N, we have type 0, type 1, type 2 type 3 and so on and see for example, if
you use an integral controller we automatically get a type 1 system.

I am sure all of us agree right see what is the transfer function of an integral controller?

Ki Ki
If you use in our open loop transfer function, I automatically have a s in the
s s
open loop transfer function denominator right. So, I automatically get a type 1 system, I am
just giving an example ok. So, please remember this ok. So, that is the terminology which we
use.

Student: (Refer Time: 17:46).

221
see denominator of course, I am assuming that there are no pole 0 cancellation to the origin,
right. So, I am assuming that there is no 0 at the origin ok.

Student: (Refer Time: 17:58).

Yeah yes.

Student: (Refer Time: 18:03).

So, for type 1 and higher my original a n should have been 0, as you can see from here ok, I
am only writing a pth order polynomial here, right, the only thing is an I am making the
constant term of the pth order polynomial as 1.

Student: (Refer Time: 18:33).

We will see.

Student: (Refer Time: 18:43).

Yeah. So, I pre-emptively told you why am I taking out let me once again do as do another
example then we will I think, we will stop here and then we will continue right let me do an
example right. So, let us say, I have

s+4
G ( s ) H ( s )=
s +3 s 2+5 s
3

So, what I do is that I take 4 out in the numerator and then I take 5s common in the
denominator. Therefore,

s+ 4 4( 0.25 s+ 1) 0.8(0.25 s+1)

G ( s ) H ( s )= = =
s +3 s +5 s 5 s (0.2 s +0.6 s +1) s(0.2 s2 +0.6 s+1)
3 2 2

So, here N is 1, p is 2 , m is 1 and K happens to be 0.8; that is what is called as the open loop
gain like we will come back shortly next class.

222
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 31
Steady State Error Analysis
Part – 1

(Refer Slide Time: 00:20)

In the last lecture, we were looking at study state error analysis. We considered the standard
negative feedback system whose block diagram is shown in the figure. Here G(s) was
essentially C (s)timesP(s). G(s) is the forward path transfer function and H (s) is the
feedback path transfer function and G ( s ) H ( s )is the open loop transfer function and

G(s) is the closed loop transfer function. Those are the two critical transfer
1+G(s) H (s)
functions that we are going to be interested in.

Using this we got

R(s)
E ( s )=
1+G(s) H (s)

Using the final value theorem, the steady state error can be obtained as,

223
 s R( s)
e ss=lim e ( t ) =¿ lim sE ( s ) =lim ¿
t→ ∞ s→0 s→ 0 1+G(s) H (s)

We discussed in the previous lecture that we can write G ( s ) H ( s )as

m m−1
K ( b^ 0 s + b^ 1 s …+ b^ m−1 s+1)
G ( s ) H ( s )=
s N ( a^ 0 s p + a^ 1 s p−1 …+ a^ p−1 s+1)

(Refer Slide Time: 02:20)

Depending on the values of N, we call the corresponding system as a type N system in

general. If N=0 it is called ‘Type 0’, ifN=1, it is called ‘Type 1’ and so on. That is where
we stopped in the last lecture.

224
(Refer Slide Time: 03:26)

Let us continue from where we left off. We are looking at the steady state error. L et us first
do the steady state error analysis to a unit step reference input, ok. So, that is what we are
going to do to begin with. So, let us take r of t to be 1 and let us observe what happens, right.
So, that is the first analysis we are going to do as far as steady state error analysis concern.
So, if r of t is 1, what is R of s?

Student: (Refer Time: 03:27).

It is going to be 1 divided by s, right. So, this immediately implies that the steady state error
which is limit s tending to 0 s times capital R of s divided by 1 plus G of s H of s that is going
to be equal to limit s tending to 0, 1 divided by 1 plus G of s H of s, right this can be re
written as 1 divided by 1 plus G 0 H 0, ok. So, essentially where G 0 H 0 is nothing, but the
value of the open loop transfer function evaluated at s equal 0, right. So, that is what it is,
right.

So, using this we define what is called as is static position error constant which is denoted by
K pe that is defined as limit s tending to 0 G of s H of s, ok. So, that is the definition, ok, that
is a parameter. As we progress we will see why it is called as a static position error constant,
ok. So, right now we will just define it. So, the static position error constant is nothing but the
limit s tending to 0 G of s H of s.

225
(Refer Slide Time: 05:06)

So, this would immediately imply that the steady state error is going to be nothing, but 1
divided by 1 plus K pe, right. So, that is what you will get, ok, for the steady state error to a
unit step reference input.

So, now let us look at what happens when N equal 0, when N equals 1, when N equals 2 and
so on, right. So, what do you think happens to K pe when N is equal to 0? When capital N is
equal to 0, so that means, that in the open loop transfer function I am not going to have any
just factor of s or any higher orders of just s in the denominator, right. So, if I take the limit as
s tending to 0 of G of s, H of s what will happen? I will get only K, right. So, because s power
N is not there, right because capital N is 0, am I correct? So, when capital N is 0, so, I will
just get K times b naught hat s power m all the way till 1 divided by a naught hat s power p
all the way till 1, ok. So, of course, p becomes equal to small n there, right.

So, then you immediately see that if I take the limit s tending to 0, what happens? I get K, ok,
which is what is called as open loop gain, ok; we will see real encounter is later, ok. So, that
is what we are going to get and as a result what is going to happen to the steady state error we
have going to get it as 1 divided by 1 plus K, is it clear? How we got it as 1 divided by 1 plus
K because K pe becomes equal to capital K.

Now, if N is equal to 1 what do you think happens to the value of K pe? If N is equal to 1
what do you think happens to the value of K pe as s tends to 0? So, immediately you see that
the limit of s tending to 0 G of s H of s will become infinity, right. So, what will happen to

226
the steady state error, it will asymptotically tend to 0, right and that is going to be the same
case if N is equal to 2 and higher, right. So, that is what we will get for different values of N,
when we want to track a follow a step input, yes.

Student: Is that condition was m by n (Refer Time: 08:04) numerator and denominator?

Anyway, we are considering causal systems, right. So, m is less than or equal to small n ok,
capital N plus p is equal to small n, ok. So, that is those are the constraints that we are. So,
please note, you know like, see of course, from here onwards I am not going to tell all the
adjectives before the word system because. So, that we looked in the first few classes, right,
throughout this course we are going to be interested in single input, single output, linear time
invariant causal dynamic systems, that is frozen, right, unless otherwise I say so, right. So, so
obviously, like we are a going to work with proper transfer functions. So, that means, small m
is less than or equal to small n, ok, yeah.

So, yeah so, this is the steady state error for a unit step input. So, what can we observe from
this? What can we conclude? So, we can immediately see that of course, when we write this
please note that capital K is what is called as the open loop gain we will see why it is called
that way, you know like, ok. So, it is called as open loop gain, ok. So, then so, what can we
observe, right so, from this analysis? So, what we have done what we have given a, of course,
please note that we are assuming that the closed loop system is stable, right. So, then only we
can do all this analysis, right. So, that is the presumption, right.

So, when we provide a unit step reference input, right, so, when do you think we would be
able to track the or follow the unit step reference input perfectly as time increases we need at
least a type 1 system, right, because if it there a type 0 system I can follow the unit step
reference input, but with a steady state error of one divided by 1 plus capital K of course, one
can say that look let me increase capital K, you know, by some means, but you will see that
you cannot increase capital K, because capital K will be tied with the control effort that you
need to put in. So, we cannot just increase the open loop gain to whatever value we want, ok.
When we in fact, like when you come to automotive controls and other courses would be able
to physically realize what capital K is, right. So, you will see that that will have something to
do with the actuator specifications.

For example, if you want to steer a vehicle, what is the capacity of the steering system, right,
for example, right. If you want to brake a vehicle, you know: what is the capacity of the brake

227
system, for example, right you have a limit on that, right. So, I cannot keep on increasing
capital K mathematically I can do it, but physically I am not be able to do that, right. So, one
can see that if I give a unit step reference input if I have a type 0 system, ok. So, let us say,
you know like, this is my reference input r of t. So, let us say, I give this is my r of t. right,
what will happen is that my y of t if I have a type 1 system may essentially settle down like
this with the study state error, let us say, sorry if it is a type 0 system, ok. So, you see that the
steady state error is going to be 1 divided by 1 plus capital K, ok.

So, there is going to be some finite error, right, as time tends to infinity. So, but on the other
hand if I have a type 1 system or higher, I can at least theoretically guarantee that the steady
state error is going to go 0 as time increases, ok. So, then that means, that, you know like, this
is indicative of, in this case, you know, what we call as tracking error; that means, that, you
know like, I give a step input to track that is that is what I want to follow.

So, if the word tracking means following some desired signal or desired profile, right. So, I
want to follow as unit step input unit step reference input I give it see for example, see let us
say, you know like, if you want to essentially steer a vehicle, right. So, for example, let us
say, you want to give a unit step change in the orientation of the vehicle for some reason,
right. So, then the question is that, can your vehicle essentially achieve that as t tends to
infinity? Ok. So, that is what we are looking at, right.

So, so if my plant does not have a pole at the origin; that means, it does not have a 1 by s
term how can I introduce a 1 by s term in the open loop transfer function? See typical P of s
is given to us it is what the system is, let us say, the P of s plant transfer function does not
have a. So, as an example, let us say, P of s, let us say, something like 1 by s squared plus 1
another is no pole as the origin, then let us say, H of s is 1, then how do I introduce, what to
say, let us let me if put it in this way s squared plus s plus 1 ok, for example, right. So, then
how do I introduce, a pole at the origin or a 1 by s term to make a type 1? I need to use an
integrator, right, in the controller block, ok. So, that is what we discussed, right. So, I told
you that integral control can guarantee you 0 steady state errors if you want to essentially
follow step input, right. So, now, you can observe why, right.

Suppose so, that is why we need to take it to the take it with a pinch of salt, we need to look
at this is very carefully if my plant transfer function already had, if P of s where like 1 by s
plus s plus 1, I do not need an integral controller to make the steady state error go to 0 I can

228
use, let us say, proportional control itself, ok. So, the main key thing is that if I want to track
unit step input with zero steady state error, the only thing I need to check is it like that is the
open loop transfer function have 1 by s term that can come from the plant it can come from
the controller, ok. So, if the plant already has 1 by s term I do not need to use integral control,
but if it does not I need to use integral control, ok. So, now, you know, the you understand the
reason as to why people say that integral control can drive the steady state error to 0.

So, if your plant does not have an 1 by s term if you just use proportional control you are
going to get a steady state error like a type 0 system, right, that is what we discuss when we
discuss various controllers, right. So, I am. So, slowly all those points are getting clarified,
right, as we go further and farther in the course, ok. So, I hope this point is now clear, right,
ok. So, that is that is the, what to say, so, if P of s is 1 by s times s plus 1 ok. So, we do not
require integral term I in C of s for driving the steady state error to 0, ok. So, that that is what
the completes the sentence, is it clear? Ok.

So, but then have you shortly see, you know like, when you when we go to the root locus and
other analysis that see purposefully introducing a pole is the origin, you know like, has a
counteracting effect, nothing comes free, ok. So, if we introduce 1 by s term through our
integral term for this purpose what I am going have is I am going to have an open loop at the
origin, then stabilizing the closed loop system and getting closed loop performance are going
to be relatively more challenging. So, there is a price to pay, right. So, the control design
needs to be done more carefully fine, ok. So, that is to unit step reference input, ok. So, the
second input which I am going to consider is steady state error to a unit ramp reference input.

Student: (Refer Time: 16:54) we need different controller for different (Refer Time: 16:56)
input?

I am sorry, what is that?

Student: We need different controller for different input?

And, not in that sense, you know like, but the thing is it, like, the question boils down to, you
know like, what are the typical inputs you need to follow, right, you need to track that is a
question we need to ask ourselves, right; based on that sometimes we can intuitively adjust
the open loop transfer function and design, ok. So, the design goes both ways, you know like,
see for example, one of the recent work we did, you know, we looked at the data and then

229
hypothesize the transfer function, you know like, because based on this analysis we know
what structural satisfy fit the data, right. So, it is a two way process, right. So, just not that,
you know, we always go one way in, you know like, and do a universal control design, it is
really depend on what you want to try, ok.

So, suppose if you want to, please note that unit ramp is not a bounded input, right, that is
point number one, kindly remember see if ever, you know like, we figure out the value of a
function of y of t for a unit ramp and we observe that output is going to infinity naturally it
would, right, because it is not a?

Student: Bounded.

Bounded input, ok so, please remember that, right.

(Refer Slide Time: 18:14)

So, so r of t is t, right; so, that means, that R of s is going to be 1 divided by s square. So, this
immediately implies that I am just repeating the same calculations limit s tending 0 s times r
of s divided by one plus G of s H of s, sorry what is going to happen. So, what will happen
here? I will get limit s tending to 0 s times 1 by s squared, right. So, one s will get cancelled I
will be left one more s term. So, this will give me limit s tending to 0, 1 divided by s times G
of s H of s, ok. So, then the terms static velocity error constant K ve is defined as limit s
tending to 0, s times G of s H of s, ok.

230
So, that is the static velocity error constant that is once again a definition. So, immediately we
see that the steady state error for a unit ramp input is going to be just one divided by K ve that
is what we have, right, ok. So, then we do a similar analysis now for N equal 0, N equals 1, N
equals 2 and higher, right. So, what do you think about happened now when N equal 0, 1 and
2 what happens to K ve? So, K ve is going to be s times G of s H of s, right when for a type 0
system. Please note that there is no s term in the denominator, right, what will happen to this
limit will be 0, right; Please remember what is the structure of G of s H of s, this. So, for type
0 system s power n capital N term does not exist, right. You multiply by s take the limit s
tending to be 0, you are going to get a limit s tending to 0 s times K, that is going to be 0,
right.

So, that is why we got K ve as 0 for a type 0 system. So, the steady state error is going to tend
to infinity, right. So, it is going to become very very large as we go on. Yes?

Student: Sir why are sending steady state error for an unbounded input?

Unbounded input, ok. So, sometimes you may want to track it, ok. So, sometimes you may
want to track ramp, ok, the question is that can be what structure I need, ok. So, that is the
object of this analysis, right. So, we will see that, you know like, it is not a very simple
straight forward problem write it really depends on what you want to do in practice in some
applications, let us say you want to track signal, you know like, which goes like a ramp and
maybe then studies with a steady state what structure would you want to essentially track it
very closely, right.

So, you can see that whether type 0 system we are not going to essentially we are going to
have a significant study state error, right. So, and as t if you maintain the ramp forever your
steady state error is going to go to infinity that is what we can observe is it clear? Now, with
type 1 system what will happened to K ve?

Student: (Refer Time: 22:01).

Type 1 system means there is a s term in the denominator. So, you multiply by s, the s and s
will get cancelled, right. So, as you take the limit s tending to 0 what will happened to this K
ve, it will become capital K, the open loop gain, right. So, what will happen to the steady
state error it will become 1 by K, right, correct. So, type 2 K ve will be infinity, type 2 and
higher and consequently the steady state error will be 0, ok. So, what this means is it like if I

231
want to track a signal which goes like a ramp if I use a type 0 system then I am going to get
very high errors, right, tracking errors, ok.

Of course, if the ramp is only for a finite duration then we may still live with that error, right,
but if I have a type 1 system at least I can get reasonable tracking, with a steady state error as
1 by K capital K, ok. Of course, higher the value open loop gain lower will be the steady state
error, but once again the same constraint comes in, right, if you increase capital K the there is
a cost as per as the control effort is concerned and if you have type 2 and more theoretically
or guarantee if the steady state error will tend to 0, ok.

So, that is that is the, right. So, But obviously, if you want to track a ramp with 0 steady state
error you need a 1 by s square term in the open loop transfer function, right. So, that is like a
double integrator, right and you introduce two open loop poles in the.

Student: Origin.

Origin, right so, then we will see that, you know, it becomes progressively more and more
difficult to, in fact stabilize the closed loop system, ok. So, we will we will look at those
constraints, right so, but this what you have. But, another reason why I am doing this analysis
is also to introduce you to this terms static velocity error constant, static position error
constant and what is called as static acceleration error constant, which we are going to come
to now, right.

232
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 32
Steady State Error Analysis
Part – 2

(Refer Slide Time: 00:15)

So, now, similarly the final error analysis that we are going to do is for a unit parabolic
reference input, ok. So, what is a unit parabolic reference input? This means that r(t)=t 2/2 ok,
this So, R(s)=1/s3, ok. So, that is what will happen, ok. So, this implies that

lim 1 lim 1
sR (s) s→0
e ss=lim = 2 = 2 s→ 0
s→0 1+G ( s ) H (s) s (1+G ( s ) H (s)) s G ( s ) H (s)

233
(Refer Slide Time: 01:55)

2
So, similarly the static acceleration error constant, which is defined as K ae =lim s G ( s ) H ( s).
s→0

1
So, this immediately implies that e ss= , ok. So, that is what we have, ok.
K ae

Type Kae ess

0 0 ∞

1 0 ∞

2 K 1/K

3 ∞ 0

Student: [Inaudible]

So, essentially it is we divided by 2. So, that like it comes like 1 by s cube.

234
Student: [Inaudible] what is unit parabolic?

So, essentially what we are doing is that we are instead of just calling at t squared we are
doing at t squared by 2, so that like it comes like factor like 1 by s cube, that is.

Student: But (Refer Time: 04:55) it is not a unit parabolic.

That is a definition, that is how it is defined.

Student: It is defined as (Refer Time: 05:01).

Yeah.

Student: (Refer Time: 05:03) unit.

It is called that way yeah, yeah, yeah. If it is if you have a unit parabolic input the r(t)
function time domain function is t2/2, yeah. So, of course, since we are dealing with linear
systems you can still consider a t2 right. So, you will just have a factor 2 that is it ok. So, does
not matter. So, that is the steady state error to a unit parabolic input.

But, see my aim here was twofold the first thing for us because by and large we will be
interested in tracking step inputs all, right. So, I wanted to show that we need at least one
integral term in the open loop transfer function that was point one; 1/s can come either from
the plant or the controller all right, please remember that. So, you the statement that integral
controller will remove steady state error is correct, but then you do not need an integral
controller always to remove steady state error to a unit step input that is you need an integral
controller to remove nonzero steady state error I am sorry you know to only if we do not have
a 1/s term in the plant transfer function to begin with, right.

So, at the end of the day we need a type 1 system, all right. So, if you want to have zero
steady state error while tracking a step input, ok. So, that is the that is one thing and I also
wanted to introduce these terms right, definitions of these terms static position, error constant
velocity, error constant and acceleration error constant, ok. So, that was the aim and the
structure of the open loop transfer function which is also like quite useful for us how we
wrote it down right. So, K times, in this way, ok. So, this is a structure which we will utilize
later also, right. So, you give me any open loop transfer function I can rewrite it in this
manner, right. So, that is something which we already looked at,

235
(Refer Slide Time: 07:18)

So, the next topic which we are going to start today is that of root locus.

Student: Sir.

Yes, please.

Student: Why cannot we just use type 2 system for the [Inaudible] (Refer Time: 07:22)

Type 2 systems, ok. So, that we are going to answer when we will get a clearer picture when
we do root locus, ok. So, his question was you know why can we not use a type 2 system to
begin with? The problem is that the moment you have a type 2 system you have a 1 by s
square term, right. You have two open loop poles at the origin, ok. Even you, we will shortly
see that even stabilizing with one open loop pole of the origin is going to be challenging to
open loop poles it is going to be even more difficult.

So, the moment you keep on increasing the number of open loop poles at the origin stability,
closed loop stability becomes a challenge, alright. So, in this analysis we are assuming that
closed loop system is stable, but practically stabilizing it is going to be difficult, ok.

Let us so, let us do root locus, ok. So, what this what is what do we mean by root locus?
Right. So, as the term indicates it is a locus of some root, right. See, all of us know what
locus means, right. So, it is just a trace of something right as a parameter is being varied,
right. So, that is what we call as a locus, right. So, what is root locus? This is the locus of the

236
closed loop pole closed loop poles as a system parameter is varied. Ok, see for example, if I
vary the open loop gain capital K which came in the previous slide right how does the, what
to say, the closed loop poles vary, ok. So, once again we are going to consider the same block
diagram which we looked at; see this is the reason I spent quite a bit of time in constructing
these block diagrams, right and explaining what each block represented, ok.

So, so, we take this block diagram, ok. So, immediately the closed loop transfer function
which is nothing but

Y (s) G(s)
cltf = =
R(s) 1+G ( s ) H ( s)

closed loop characteristic equation : 1+G ( s ) H ( s ) =0

So, please note that the roots of this equation are the close loop poles, ok. So, that is
something which we already know, alright? So, now, let us look at this closed loop
characteristic equation in more detail, right.

(Refer Slide Time: 11:06)

1+G ( s ) H ( s )=0→ G ( s ) H ( s ) =−1

G ( s ) H ( s ) is what is called as the open loop transfer function.

So, what can you infer from this, that any value of s that makes the value of the open loop
transfer function to be minus 1 is a closed loop pole, do you agree? Correct. So, that is

237
another way of interpreting when a value of s would qualify as a closed loop pole and we are
going to use this repeatedly to our advantage as we will see shortly please note that s is a
complex variable right and G ( s ) H ( s )is a complex valued function. In general G ( s ) H ( s )can be
a complex valued what to say can take a complex value right because that is a complex
valued function of s, right. So, the moment we have this equation I can rewrite in two terms,
because any complex variable will have a magnitude and a phase.

So, if I have G ( s ) H ( s )=−1; that means, that at that value of s,|G ( s ) H ( s )|=1 (Magnitude

condition) and angle ( G ( s ) H ( s ) ) =±180 °(2k +1) (Angle condition). So, of course, some
people will call it as a phase condition, right. So you see that from the closed loop
characteristic equation we have got two conditions, right and any value of s which satisfies
these two conditions is a closed loop pole, I am sure we are still an agreement, right through
this process, ok.

Now, the question is how are we going to use it to our advantage? That is what we are going
to learn, right when we do root locus. So, let us consider that the open loop transfer function

K ( s+ z 1 ) … (s+ zm)
takes the form G ( s ) H ( s )=
( s+ p1 ) …( s+ pn)

So, by and large you can rewrite in this form. please remember that in the previous discussion
on steady state error we used a K such that the constant terms in the numerator and the
denominator polynomial were one; here it would not be please note that; that K was known as
the open loop gain; this K is something else it is a parameter ok, but unfortunately the same
symbol is used. So, please let us we should not get confused between the same symbol I
could have called it something else, but in most textbooks you will see that this will be the
structure will be followed ok. So, I do not want us to essentially follow a different notation ok
that is my only motivation ok.

So, of course, you see that I have factorized the denominator and the numerator polynomials.
So, you can see that the numerator polynomial is of order m, denominator polynomial is of
order n, right. So, that structure is something which we are still keeping, right. So,
immediately we can see that there are m open loop zeroes, right. What are the m open loop
zeroes? Minus z1, minus z2, all the way till minus zm, right. So, there are m open loop zeroes
do you agree, ok.

238
(Refer Slide Time: 16:58)

And, there are n open loop poles. So, I am sure all of us remember what are meant by open
loop zeroes and open loop poles, right. Open loop zeros are the zero are the open loop
transfer function, open loop poles are the poles are the open loop transfer function, ok.

So, what are the n open loop poles they are essentially -P1, -P2 all the way till -Pn, ok. So,
those are the n open loop poles, ok. Now, what is the question we are going to ask ourselves,
right. So, the question we are going to ask ourselves is the following, ok. So, before that, this
essentially implies that you can immediately see that there would be n closed loop poles.
Why? Because the moment you calculate 1+G ( s ) H ( s )=0and take LCM you will get an n-th
order polynomial, do you agree?

See, because if the order of the open loop transfer function is n the order of the closed loop
transfer function will also be n if you evaluate 1+G ( s ) H ( s )you would see that that will be an
n-th order polynomial, right. So, you would be solving for n roots. So, you will have n closed
loop poles I hope that is that is clear to everyone? Right.

So, the question that we are going to ask ourselves is that how would the n closed loop poles
vary with respect to the open loop poles and open loop zeros as the parameter K is varied, of
course, K we consider to be a real parameter, right. So, as the value of the parameter K is
varied, ok so.

Student: Sir.

239
So, yes.

Student: [Inaudible]

Why would the poles be different, right? So, because to answer that question let us go here
right. So, what will be 1+G ( s ) H ( s )here? It will be 1 plus this, right. So, I will have s + P1.

Student: The denominator remains the same, right.

A denominator remains the same, ok, but we are evaluating for the numerator, right. So, so,
the closed loop characteristic equation will be this of course, you will have divided by (s +
P1) (s + Pn) that you anyway like take it other side will become zero. So, essentially if you
solve this you are solving this so, obviously, you see that the values of s are different, right.
They depend on the value of K ok, that is a good point. So, I am going to leave you are
preemptively given part of the answer to questions I am going to leave you in, ok.

So, essentially yeah this is the question we are going to ask ourselves and the answer is the
root locus provides a graphical representation, ok. We will see that the root locus provides us
with the corresponding graphical visualization graphical visualization, ok.

So, what we are going to consider is the following, ok. So, we would first consider the case
when the parameter K is non negative, then we will see what happens for the other case also,
ok. We will we will go on case at time, right. So, that means, that K is greater than or equal to
0, of course, strictly greater than 0 because equal to 0 would be it does not make sense, ok.
So, essentially K should be a positive parameter right so, but as we will take the limit also all
right.

So, essentially what we need to say is that what happens when K goes from 0 to infinity, that
is what we are going to look at first, ok. So, I am going to leave you with a couple of
questions, alright.

240
(Refer Slide Time: 22:00)

So, what happens to the closed loop poles as K tends to 0 and K tends to infinity, ok. The
answer is very evident from the thing that we wrote down to your question, right. So, what
we wrote down here immediately you can figure out what happens as K tends to 0 and K
tends to infinity. What do you think will happen as K tends to 0?

Student: Open loop pole (Refer Time: 22:34).

The closed loop poles will tend to the open loop poles and what happens as K tends to
infinity?

Student: Open loop pole (Refer Slide Time: 22:00).

m closed loop poles will tend to m open loop zeroes. What will happen to the other n minus
m closed loop poles? We will see what happens, ok. So, that is something which we are going
to you think through and then like we will answer it in the next class, ok. The second question
which I am going to leave you with is anyway we will revisit the answers, ok. So, please
think about it we are going to revisit these questions and then like discuss all the answers.

And, how do we check. So, I think let me put it this way how do we check or when a
polynomial has multiple roots, right, essentially repeating roots what characteristics would
the polynomial have, right. So, let me put it this way how do we check if a root of a
polynomial has a multiplicity greater than 1. So, that means, that we are we are having
repeating roots.

241
So, how do I check if I give a polynomial I tell you that something is a root all right how do
you check whether what characteristic will you get, right? If you have a repeating root
obviously, if you give any polynomial you give me any root at that root the value of the
polynomial should be 0, right that is the definition of a root of a polynomial, but that is true
for any root, right. So, but then if I know that some root is repeating you know like how do I
check that, right. So, that is something which we will which you think about and come
because these answers are going to become pretty important when we what to say learn the
root locus fine.

So, I will stop here and then we will meet on Monday, right so, yeah.

242
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 33
Root Locus 1
Part – 1

In the last class, we stopped stop with the brief introduction to root locus. We considered
the standard feedback block diagram.

(Refer Slide Time: 00:29)

We know that the closed loop transfer function is

Y (s) G( s)
= ,
R(s) 1+G ( s ) H (s)

G ( s )=C ( s ) P ( s ) .

1+G ( s ) H ( s )=0 is the closed loop characteristic equation.

if we rearrange the terms in this equation, we get

G ( s ) H ( s )=−1.

243
Since sis a complex variable and G ( s ) H ( s ) is going to be a complex valued function of s,
this gives me two conditions.

1. Magnitude condition: The magnitude of G ( s ) H ( s ) should be equal to 1,

|G ( s ) H ( s )|=1.
2. Angle condition: The phase of G ( s ) H ( s ) should be an odd multiple of 180

degrees,∠ G ( s ) H ( s ) =±180 ( 2 k +1 ) , k=0 , 1, 2 … .

Any value of sthat satisfies these two conditions is a closed loop pole.

(Refer Slide Time: 03:11)

We also discussed that the open loop transfer function can be expressed as

K ( s+Z 1 )( s+ Z2 ) …(s+ Zm )
G ( s ) H ( s )= ,
( s+ P1 )( s+ P2 ) … (s+ Pn )

K being a constant real number.

Now, the question is, how would the closed loop poles change as K is varied?

Let us first consider the scenario where K is a non-negative scalar, K ≥0.

244
Of course K=0 does not make sense because in practice it is going to be greater than 0.
But we are going to look at the limiting value of K → 0 andK → ∞. The closed loop
characteristic equation is going to be in the form,

K ( s+Z 1 )( s+ Z2 ) …( s+Z m )
1+ =0
( s+ P1 )( s+ P 2 ) …(s+ Pn )

The root locus plot gives us the locus of the closed loop poles as K is varied, right. We
are going to learn the construction of the root locus in this exercise. The handout given
contains the details the list of steps involved in the construction of the root locus plot.
Now we are going to list those steps and get an intuitive understanding of those steps.

(Refer Slide Time: 07:48)

First thing to note is that the root locus will have nbranches because there are going to be
n closed loop poles. Before we go into the process, let us take a simple example.
Consider a plant transfer function

1
G ( s )= ,
( s+ 4)(s−1)

Let us consider unity feedback and a proportional controller. Then the open loop transfer
function to becomes

K
G ( s ) H ( s )= ,
(s+ 4)(s−1)

245
Where K the proportional gain and it is a non-negative real number. From the open loop
transfer function we see that n=2and m=0. The open loop poles are at - 4 and 1 and
there are no open loop zeros. Now we construct the s plane with the real axis, the
imaginary axis and the origin. We mark the two open loop poles there at - 4 and + 1 as in
the figure above.

Let us take a test point st . We want to figure out whether the point lies on the root locus
or not. If the test point lies on the root locus, it should satisfy two conditions. If any value
of s has to be a closed loop pole, it has to satisfy the magnitude condition and the angle
condition, right. Please note that, though we are interested in the closed loop poles we
are evaluating the open loop transfer function here. And the angle condition is the one
which is going to be helpful to us.

So let us figure out the phase of G ( s ) H ( s ) at the test point.

K
∠ G ( st ) H ( st ) =∠
( )
(st +4)( st −1)
=∠ K −∠ ( st + 4 ) −∠(st −1)

The parameter K is taken to be positive. Therefore∠ K=0°. And to get the phase of a
complex number we draw a vector from the origin to the corresponding point. Then
measure the angle that it makes with the positive real axis in the counterclockwise
direction. The counterclockwise direction is take as positive by convention.

To figure out the phase of st + 4, we can write st + 4as st −¿). The vector st is the vector
from origin to st and vector −4is going to be vector from origin to−4. Therefore vector
st + 4 is going to be a vector which starts from −4 and goes towardsst . It is drawn as
shown in figure. Let us call the angle made by st −¿) with the direction of the positive
real axis as θ 1. Now we can get the phase of st −1by the same logic. The vector st −1 will
be the vector from +1 to st . Let us call the angle made by st −¿) with the direction of the
positive real axis as θ 2. Therefore

∠ G ( st ) H ( st ) =−θ 1−θ 2

We are interested in knowing if the following condition is satisfied. If it is satisfies the

point lies on root locus.

246
∠ G ( st ) H ( st ) =−θ 1−θ 2 =±180 ° ( 2 k +1) , k=0 , 1, 2…

By and large the angle condition is used to plot their root locus and the magnitude
condition is used to get the value of gainK . We shall solve more examples related to this.

Let us go to the steps involved in the construction of the root locus (You can use the
command ‘rlocus’ in Matlab to plot the root locus of a given open loop transfer
function.). Please recall that we are considering a closed loop negative feedback system
whose characteristic equation is

K ( s+Z 1 )( s+ Z2 ) …( s+Z m)
1+ =0.
( s+ P1 )( s+ P2 ) …(s+ Pn )

(Refer Slide Time: 20:00)

We have considered the case m<n and K being a positive parameter. Z1to Zmare the open
loop zeros and P 1to P nare the open loop poles. The root locus traces the evolution of the
roots of the characteristic equation with variation inK and the root locus will haven
branches or curves.

Following are the steps to construct root locus for a given open loop transfer function.

The first step is to locate the open loop zeros and poles in the complex plane. Each of the
n curves in the root locus will start from open loop pole for K=0. Out of these m curves

247
will terminate at an open loop zero as K → ∞and the remaining n−m curves will to ∞ in
the complex plane along asymptotes, as K → ∞.

(Refer Slide Time: 21:36)

To better understand these steps, we will start constructing the root locus for the system
under consideration. Step one is to locate the open loop poles and open loop zeros. Let us
mark them in the s plane. One open loop pole is at −4 and another one is at −1 and
there are no open loop zeros.

There will be 2 branches in the root locus because n=2. The closed loop characteristic
equation is going to be

K
1+ =0
( s+4)(s−1)

As K starts increasing from 0 each branch will start from an open loop pole. This
happens because when we substitute K=0in ( s+4 ) ( s−1 ) + K=0, the roots are −4 and 1.
So, we are going to have 2 branches starting from−4 and 1 when K=0. The branch
which starts at −4 would go towards the right on the real axis and the branch starting at
1 would go to the left on the real axis.

Please note that −4 is on the real axis. A branch starting from −4 cannot go anywhere
else other than on the real axis. This is because we are dealing polynomials with real
coefficients which means if we have a complex root the conjugate also must be a root. If

248
the branch was to go to the complex region there should also be another branch
symmetric to it. Since there is only one pole at −4, two branches cannot start from a
single pole. This rules out the evolution in the complex plane. There are only 2 options,
either it can go on the real axis to the left or to the right. It should go to the right because
of the angle condition. The angle condition is satisfied only between −4 and 1.

Let us take a test point st in between −4 and 1. Now θ 1=0° and θ 2=18 0° . The sum is
180 °which is an odd multiple of 180 °. So the angle condition is satisfied. That is how
the two branches evolve.

249
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 34
Root Locus 1
Part – 2

(Refer Slide Time: 00:15)

Now, comes the question as what happens as K tends to infinity right.

(Refer Slide Time: 00:22)

250
So, if we go back to this, I hope it is clear why each one of the n curves will start from open
loop pole at K equals 0 right. Now, out of these m curves will terminate at an open loop 0 as
K tends to infinity, and the remaining n minus m curves will go to infinity in the complex
plane along what are call asymptotes right.

(Refer Slide Time: 00:44)

So, now what is going to happen right? So, the same equation I can rewrite this as

K
1+ =0→ ( s+ 4 ) ( s−1 ) + K=0
( s+4 ) ( s−1 )

1
( s+ 4 ) ( s−1 ) +1=0
K

Now, if I put K to be infinity, I get 1 = 0 right that is not exactly what we want to convey
right. On the other hand, that is why I am saying I am not doing a very tight proof you know I
am doing a very logical argument here right.

Suppose, on the other hand, you write this one as (0s+1)(0s+1), can I do that right? So, in a
certain sense when K tends to infinity, I am going to have the closed loop pole S tending to
infinity in the complex plane; it can be in any direction all right, so that is why the second
statement comes into play right. So, if we had m to be nonzero in this case m was 0 right. In
our example m was 0, n was 2 right. Since, m was 0 you know like I do not have any branch
of the root locus terminating at an open loop zero. Suppose, m were 1 suppose let me write it

251
as an example suppose if we had this equation to be like this rather than this way minus 1
would have been an open loop zero.

So, immediately you can see that if you rearrange this equation, we would have got this right.
So, as K tends to infinity, S will tend to -1 right that would be one of the closed loop poles, so
that is why you know like there was a statement out of the n branches, m branches will tend
to open loop zeros if there were any right.

Other n minus m branches will tend to infinity along what are called asymptotes ok; we are
going to going to come there right. So, I hope the logic is clear right. So, as to why we have
that what to say behavior. In this particular example, we do not have any what to say open
loop zero. So, we are going to have two branches, both the branches going to infinity in the
complex plane as K tends to infinity ok, so that is what is going to happen right ok. Is it clear,
the first step?

(Refer Slide Time: 03:13)

Yeah.

Student: [Inaudible]

I only took an arbitrary test point to get the angle condition right, is it not? Now, we are
starting the construction on the root locus. Let us say we focus at minus 4 right. So, we agree
that one branch with start from minus 4 right from this equation from the equation in the red
box substitute K equals 0, what will you get one of the closed loop poles as K tends to 0. I

252
would not say substitute K equals 0 because it does not have any meaning, as such as K tends
to 0 you will agree that the closed loop pole tends to minus 4 right one of the closed loop
poles.

So, let us say I look at minus 4. I can have a test point going in any direction right, the branch
starting off at any direction. We eliminated the, what to say entire space other than the real
axis by arguing that there is only one branch right. If there were a repeating set of open loop
poles at minus 4, we cannot dismiss that possibility right, so but since there was a non-
repeating open loop pole at minus 4, we have only two options; one to the left, one to the
right and then I applied this angle condition that is it. So, now we have a set of test points to
be limited right in this step. So, to show you how it started I use this. Is it clear? Yeah.

Student: [Inaudible].

We are going to come there right. So, what it means is that see as K tends to infinity, the first
term goes away right. So, as I told you this is not a very tight way of showing things, but then
like imagine that is why I told you this example right. So, in this case, there was one open
loop 0. So, what happened is that. As K tends to infinity S + 1 = 0, s = -1. So, you have one
branch going to an open loop zero.

(Refer Slide Time: 05:41)

But then there is one more branch right what would happen to this. So, the way people will
demonstrate it you know like without going through the proof and so on is that like let us say

253
we multiply this with 0s + 1 the product remains the same it is as if like you make the number
of terms to be equal. And you see that s tending to infinity is another route that is what
happens here right. So, because what I did was that I replace this one with 0s + 1 and 0s + 1
ok. If you do not prefer 0, you put epsilon, epsilon tending to 0 because we are looking at the
limit analysis right.

So, you write (εs + 1)( εs + 1), ε being very small right, the product will tend to 1 right. So,
essentially you will see that the value of s tends to infinity right. How it goes to infinity we
are going to find out right. So, as I told you, I am not doing the derivation I am only giving a
logical intuitive process, I am not writing things very carefully right please remember that ok.
So, we are only good doing a limit analysis here right, so that is something which we should
remember. Is it clear? Ok.

Now, let us go to the second step. So, root locus is all about following these steps like
logically that is all there is to it right. The second step is to locate the root loci that lie on the
real axis ok. This is determined by the open loop poles and open loop zeros that lie on the real
axis alone ok, we will see why. Choose a test point on the real axis, if the total number of real
open loop poles and real open loop zero is that lie to the right of the test point is odd then that
point lies on the root locus that is the condition ok.

Now, let us look at the first statement here right. This is determined by the open loop poles
and open loop zeros that lie on the real axis right. Why is that true? So, I can have complex
conjugate open loop poles and complex conjugate open loop zeros I can have right? But then
it so turns out that if I have complex conjugate open loop poles and complex conjugate open
loop zeros their angle contribution to any test point on the real axis would be a would you can
call it either 0 or an even multiple of pi, it will cancel off right, think about it.

Let us say as an aside I do this. Let us say I had two open loop poles like this right. And let us
say I take a test point like this. So, the first test point is this way, the second test point is this
way right. What is the angle contribution from this test point is this all right, what is the angle
contribution from this test point; it is this. So, what do you think will be the sum of the angle
contribution of these two vectors?

Student: (Refer Time: 08:35).

254
Depending on how you look at it either they will cancel off or it will become they either 0
degrees or ±360 degrees right. So, depending on how you count right, but certainly you see
that it is going to become an it is not going to matter as far as of locating points on the real
axis is concerned that is why we are only going to consider real open loop poles and real open
loop zeros first step 2 ok. So, step 2 is to locate the parts of the real axis that lie on the root
locus ok. So, for this what we do is that in this example there are no complex conjugate open
loop poles and open of zeros, but as we looked at it, what we need to do is that we need to
consider only the real axis ok.

(Refer Slide Time: 09:54)

So, I am going to draw the real axis and then like consider just the real axis of our problem
right. So, we consider an open loop pole at 1 and -4. So, what happens is that we are going to
consider three test points let us say st1, st2 and st3 right. So, you see that the open loop poles
and open loop zeros divide the real axis into 3 regions right minus infinity to minus 4, minus
4 to 1, and 1 to infinity. Do you agree? The real axis is divided into three chunks right in this
example by the real open loop poles and real open loop zeros right, so minus infinity to minus
4, minus 4 to 1 and 1 to infinity right.

Now, the questions is that which of these points lie on the real axis right. Once again you
need to apply the angle condition. So, if I take test point st1 which is in the region minus
infinity to minus 4, do you think the angle condition is satisfied? No, right theta 1 is 180;
theta 2 is 180 degrees, so it becomes an even multiple of pi. So, this region does not lie on the

255
real axis. What about minus 4 to plus 1, see what I am using in my arguments, this diagram
which we drew right, this condition. We are we are evaluating this question right ok.

So, for a test point which is between minus 4 and 1 is angle condition satisfied?

Student: Yes.

If you take st2, you see that theta 1 will be 0; theta 2 will be 180, so this is tick right. Now,
what about 1 to infinity?

Student: (Refer Time: 11:51).

No, because the angle will be 0 ok, which is not an odd multiple of 180 degrees. So, so this
consequently tells me that the only portion which lies on the root locus is the region between
minus 4 to plus 1 on the real axis ok. Of course, we are going to come we are going to do
more steps, but as far the real axis is concerned for this example minus 4 to one is the only
portion that lies on the real axis.

(Refer Slide Time: 12:38)

How do we generalize it? You know from this example that is what this document is about
right. What was the condition choose a test point on the real axis if the total number of real
open loop poles and real open loop zeros that lie to the right of the test point is odd, then that
point lies on the root locus. If it were 0 or an even number then that point is removed. So, let

256
us go to our example once again let us say you take st1 look to the right of st1 on the real
axis. How many real open loop poles and real open loop zeros can we count?

Student: (Refer Time: 13:11).

Two right, minus 4 and plus 1 to the right of st1. Please remember the condition right. You
take a test point on the real axis look at the look towards the right of the test point on the real
axis consider only the real axis ok, not the complex plane entirely right.

Now, you count the number of real open loop poles and real open loop zeros to the right of
the test point right. How many do we get? 2. Is that odd? No, that is why we can eliminate
right ok. Of course, this calculation I did by using this theta 1 and theta 2, but imagine I can
have a general system right, how do we simplify, how do we generalize right that is what we
are doing.

So, but then if you take a test point st2, you look towards the right of the test point how many
real open loop poles and real open loop zeros do we have.

Student: 1.

Only one right. There is one open loop pole at plus 1. So, is that an odd number?

Student: (Refer Time: 14:10).

It is right so that is why we have ticked the region from minus 4 to 1, is it clear ok. Now, 1 to
infinity, you take st3, you look towards the right of st3, there are no open loop poles and no
open loop zeros to the right of st3 right. So, we can immediately discount that region 1 to
infinity ok, so that is how we get this condition right yeah. So, basically that is the
generalization of what we discussed right as far as locating the root loci on the real axis is
concerned. Is it clear? What we did that is step 2. Yes.

Student: Inaudible

Does not matter it should be some odd number ok. So, we are not worried about n and m here
because I am only looking at open loop poles and open of zeros on the real axis. In this
particular case n happens to be two and both open loop poles lie on the real axis, but that need
not be the case always right this condition you just need to look at the fact that when you take
a test point in the real axis look towards the right and the count should be odd because I can

257
have an arbitrary distribution of real and complex open loop poles and open loop zeros right
ok. So, let us not relate them to n minus m or n minus m minus 1 here right so that that will
not hold true in general ok. So, we just want any anyway odd and even.

(Refer Slide Time: 15:47)

And I will tell you the logic like if for people were interested in generalizing it. The logic has
to do with the angle condition right. So, because in this particular example we do not have a
zero, but think about it if we had a zero that would contribute a positive angle right is it not
right and this will be a negative angle, so that is why we are only looking at the whether it is
an we want the difference to be odd right.

So that is why because any test point the real axis will contribute either plus 180 or minus
180 with the open loop poles and open loop zeros that are on the real axis right that is why we
want an odd number because we want an odd multiple of 180 degrees here ok but please note
at this juncture please note that the steps will be different if K were negative why because
here we took K the phase contribution of K to be 0 right.

If K were negative, what will be the phase contribution?

Student: (Refer Time: 16:41).

It will be?

Student: (Refer Time: 16:47).

258
180 degrees right, so things will change a little bit anyway will come there you know like
later on ok. Yeah.

Student: Sir.

Yes please.

Student: If there was 0 in between or somewhere.

Same condition, no change that is why these are general steps ok. Suppose, if there were zero
right, let us say we had this example right. So, let us say to answer that question. Let us say
we consider the real and the imaginary axis this is plus 1, minus 4 and minus 1 was the 0
right so same steps, but now the only thing is that you will have four regions minus minus
infinity to minus 4 right, minus 4 to minus 1, minus 1 to plus 1, and 1 to infinity

So, if you use this you will see that minus infinity to minus 4 will now on the will now be on
the real axis, because any test point here will have three open loop poles and open loop zeros
to the right. This will no longer be there; minus 1 to plus 1 will be giving you an odd number
this will not so that is that is the difference which you get right. So, once you have the zero so
that step is generic enough right ok.

So, we will do steps three and further tomorrow in the next class because I think explaining
these steps also what to say and discussing them are pretty important ok. So, but then what I
request you to do is it like kindly look at this document from the course website. Please read
through the steps and come prepared for tomorrow’s class ok. So, at least you have read them
read them once and think about the questions that I asked of you right ok. We will continue
tomorrow. Yeah.

259
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 35
Root Locus 2
Part- 1

We were looking at Root Locus and we have completed two steps. The first step is to locate
the open loop poles and open loop zeros in the complex plane and the second step is to locate
the parts of the real axis that lie on the Root Locus.

(Refer Slide Time: 00:40)

Let us start from step number 3, this deals with determining the asymptotes of the root locus.
This step applies only when m<n.

260
(Refer Slide Time: 00:55)

As we discussed the root locus will have n branches. Each branch will start from an open
loop pole and m branches will terminate at the m open loop zeroes as k → ∞and the remaining
n−m branches will go to infinity along asymptotes. Therefore asymptotes are going to exist
only when n>m.

Let us take the same example as before where n=2 and m=0. We are going to have n−m=2
asymptotes.

By definition asymptotes is the trace of the root locus or the closed loop poles as k → ∞. If
we push the test point st away towards infinity, farther and farther away from the origin, θ 1
and θ 2 will become equal. It does not matter which direction we go. So, you see that the angle
contribution from each open loop pole and each open loop zero would become equal. That is
how we get the equation for angle of asymptotes. If the test point is chosen very far away
from the origin, then the angle contribution of each open loop pole cancels with that of an
open loop zero. Because the phase contribution cancels off, between the open loop pole and
the open loop zero because they are equal in magnitude.

Hence the asymptotes will be straight lines which make an angle with the positive real axis of

±180 °(2 k+1)

n−m

261
(Refer Slide Time: 04:41)

The asymptotes which exist only when m<n will be straight lines which will make an angle

±180 °(2 k+1)

with the positive real axis of .
n−m

The point of intersection of the asymptotes on the real axis is given by

( p1 + p2 …+ pn ) −( z 1 + z2 …+ z m )
.
( n−m )

This is same as calculating sum of the open loop poles minus the sum of the open loop of
zeros is divided by ( n−m ).

262
(Refer Slide Time: 06:47)

Let us look at the example, in this problem, n−m=2−0=2. This implies that there are two
asymptotes and angle made by the two asymptotes is going to be equal to ± 90 °.

Now, the point of intersection of the two asymptotes on the real axis is going to be

(−4+1) −(0)
=−1.5
2

At −1.5, the two asymptotes meet each other. One asymptote is going to go or like +90
another asymptote is going to go like−90.

(Refer Slide Time: 1 1:54)

263
(Refer Slide Time: 12:37)

Let us go to step 4, It says determine the break-away points and break-in points, A break-
away point for example, will exist on the root locus between two successive open loop poles
on the real axis with no real open loop zeros in between them. Similarly a break-in point will
exist on the root locus between two consecutive open loop zeros on the real axis with no real
open loop poles in between them.

(Refer Slide Time: 13:32)

When will we have a break-away point? In the example, on the real axis we have −4 and1.
These are two open loop poles on the real axis without any open loop zero in between. In this

264
example, we have one branch starting from −4 and coming to the right; one branch starting
from 1 and coming to the left. The root locus will start at the open loop poles as k → 0 and as
k keeps on increasing, one branch is going to come towards the, other branch is going to
come towards the left and at some point, they are going to meet each other. At this point they
are going to break-away from the real axis because they cannot turn back or cross each other.
The two branches hit each other and just break-away from the real axis. This is called as a
break-away point. Please note that break-away point need not always happen, but in this
example it happens.

Similarly if there are two open loop zeros on the real axis with no open loop poles in between
them, the root loci may travel each other and some point hit each other and break into the real
axis. This is called as a break-in point.

One unique feature about a break-away or a break-in point is that the two branches intersect
for the same value ofkwhich means we are going to have a repeating pole.

We know the fact that: If you have a root to have a multiplicity of μ till the μ−1 derivative,
the derivative would be made 0 when you substitute that repeating root. Now since at the
break-away or a break-in point, we have a repeating pole, we can say that the first derivative
is zero.

Let G ( s ) H (s) be written as

A (s)
G ( s ) H ( s )=K
B(s)

A ( s ) =( s+ z1 )( s+ z 2 ) …(s+ zm )

B ( s )=( s+ p1 )( s+ p1 ) … ( s+ pn )

The closed loop characteristic equation is going to be

A(s)
1+ K =0
B( s)

The closed loop characteristic polynomial has a repeating root, so the derivative is going to
be 0.

265
(Refer Slide Time: 22:57)

A(s)
P cl ( s )=1+ K
B( s)

' d P cl ( s )
P cl ( s )= =0
ds

Solving this, we get

' '
A ( s ) B ( s ) −B ( s ) A ( s ) =0

The roots of this equation are potential break-away, break-in points. We are saying potential
because we still need to do one more check. We need to figure out the value of K
corresponding to these points and check whether they are positive. Only then the points will
lie on the root locus.

There are two steps:

1. First we solve this equation.

2. Then we find the corresponding values of K from the closed loop characteristic

equation.

266
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 36
Root Locus 2
Part – 2

(Refer Slide Time: 00:15)

So, the question was at how does this step what is a how does this calculation of break-away
break-in point, you know like translate to the final construction of the root locus in the
sequence of steps that we are doing, ok. We are going to look at it, ok. So, let us complete all
the steps then we will put everything together, ok, yeah.

Student: [Inaudible]

Why we are making the derivative zero? Because of this, right.

267
(Refer Slide Time: 00:41)

So, when you when at a break-away or break-in point right what is going to happen two
branches of the root locus are going to intersect, right. So, that means, that the corresponding
closed loop characteristic equation would have a repeating root for that particular value of K,
because when two branches intersect each branch represents one closed loop pole, right. So,
you are going to have a common repeating root right at that point when you have a common
repeating root for a polynomial a repeating root with a multiplicity greater than 1, you are
assured that the first derivative is going to be 0, right, that is why I did this aside right. So,
that is why we are differentiating the closed loop characteristic equation or closed loop
characteristic polynomial to be more accurate ok, yeah.

Student: [Inaudible]

That is the definition of a break-away point, right and a break-in point, right. So, I just
illustrated it to you graphically. So, there may be break-away points then may not be ok, but
this is the process to check if there were break-away or break-in points they would imply that
you have a repeating closed loop pole that is why you are checking. See, we do not know a
priori whether a given open loop transfer function would result in a break-away break-in
point always, right it may not, ok. That is why we are following this general process where

A(s)
we differentiate the closed loop characteristic polynomial which is 1+ K =0and then we
B( s)
process it right. So, we get this equation and we check the values of K, right, ok.

268
Student: [Inaudible]

No, it will be ok, we are assured of that ok. So, maybe for those who are interested I will give
you a reference where you can go and look at the general proof and so on, right yeah. So, we
are not proving the results we are only taking the results directly I am only giving intuitive
explanation right to understand each of these steps, yeah.

Student: [Inaudible]

You may have, you may not have.

Student: Ok.

Need, not be there.

Student: [Inaudible]

Please remember, your closed loop poles are going to vary depending on the value of K right
you are essentially figuring out is there a value of K between 0 and infinity where I have a
repeating closed loop pole, you can interpret this step in that manner, ok. So, that is an
exercise which we are doing right from another perspective that is correct. So, the question
that we are answering through the step is that can I find a value of K, if it exists where I can
have repeating closed loop poles, right for the given system right that is what we are doing
ok, yeah.

So, yeah for the given open loop transfer function if I scroll so, please remember that

K , right.
G ( s ) H ( s )=
(s+ 4)(s−1)

269
(Refer Slide Time: 03:53)

That is that is our open loop transfer function, right. So, what are A(s) and B(s), here right?

Student: (Refer Time: 04:05).

A(s)=1, right. B(s) = (s+4)(s-1).

(Refer Slide Time: 04:28)

So, this implies that A’(s)=0 and B’(s)=2s+3, right, that is what we are going to have. So,
immediately if you evaluate these equation; A’(s)B(s) – B’(s)A(s) = 0 this would be
essentially –(2s+3)=0, right that is what we would end up there.

270
So, consequently a potential break-away point is -1.5, ok. So, once again in this particular
example the potential break-away point turns out to be the same as the point of intersection of
the asymptotes; need not be in general, ok. Please, what to say do not think that oh all break-
away points or break-in point should be the same as the point of intersection of the
asymptotes, this is a specific example where it just happens to be the same as that of the point
of intersection of the asymptotes. But, we are not yet done, right. So, we need to check the
value of K at s = sb, that is going to be

B( s) ( sb+4 ) ( sb−1 )
K s=sb−
A ( s ) s=sb [
=−
1 ]=−( 2.5 ) (−2.5 ) =6.25>0

So, consequently this lies on the root locus. So, this is a break-away point for the system that
we have studied because the corresponding value of K is positive; right? That is what we
could figure out. Yes please.

Student: [Inaudible]

Why did I say this is a break-away point because you see that in this particular example there
are two branches are coming towards each other because I know that -4 and one are two open
loop poles, right. So, I know that each branch of the root locus is going to start from an open
loop pole, they are going to come towards each other and then they are going to break away,
right. Would they not?

Student: [Inaudible]

No break-away point exactly, so.

Student: (Refer Time: 07:37).

Yes. So, if the value of K happened to be negative in this particular case right. So, you could
have concluded that -1.5 is not a break-away or break-in point ok. So, that is what it is, fine.
So, that is that is the fourth step, ok, fine.

271
(Refer Slide Time: 08:02)

So, let us go to the fifth step, right. So, we have done what to say the construction of
asymptotes and calculation of break-away break-in point’s right those are the two new steps
that we have done today right. Let us go to the fifth step, right. So, what is the fifth step? We
have to locate what is called as the angle of departure or angle of arrival. So, let me read it
once again then we will figure out what it means. So, locate the angle of departure from an
open loop complex pole, ok, that is how we should read it or locate the angle of arrival at an
open loop complex zero, ok.

So, what does this mean, right. So, what does step-6 mean really? So, let us look at step-6.
So, suppose if I had a scenario like this in this example which we are considering we do not
have the scenario, but let us say we had a scenario where I had to open loop poles like this,
ok. So, I know that from each open loop pole I am going to have one branch of the root locus
starting as K increases from 0. So, I will have one branch starting from this open loop pole
and another branch starting from this open loop pole. Now, the question is that along which
direction will this open loop pole start? Ok, that is a question .

See, when we had a non-repeating pole on the real axis we argued that it has to be only on
the real axis, why? Because I can’t go to the complex plane there is outside the real axis
because I do not have a conjugate you remember, but when my open loop pole itself is a
complex conjugate pole to begin with I can have the root locus branch going in any direction.
How do I determine that? Right, that is the question we are asking ourselves, right and angle

272
with which it starts initially as it leaves the open loop pole is what is called as an angle of
departure, right. So, I hope it is clear why it is called as the angle of departure, right.

So, why do we call this angle as angle of departure from an complex open loop pole? Is it
clear why it is called an angle of departure? Because I have a branch departing from an open
loop pole right; the question is at what angle does it make? Similarly, if I have an, what to say
an open loop zero which is complex conjugate, same logic right I have the real and the
imaginary axis in the s plane. So, let us say I had two open loop zeros which are complex
conjugates here, right I know that we are going to have two branches coming in to this open
loop 0 as K tends to infinity, right correct.

Now, the question is that like at what angle does it arrive at this open loop zero, ok. That is
why it is called as an angle of arrival at a complex open loop zero, that is the reason it is
called angle of arrival, ok. Now, the question becomes how do we calculate this angle of
departure and angle of arrival.

(Refer Slide Time: 11:47)

But, before this step-6 applies only when we have complex open loop poles or open loop
zeros, right ok. So, otherwise step-6 does not even apply because the example that we are
doing, please remember what are the open loop poles and open loop zeroes, right. See for us
if you if you scroll up, the open loop poles are -4 and 1 open loop zeroes where none, right.
So, we only had real numbers as open loop poles, right. So, step 6 does not even apply for
this particular example, but suppose if you had a complex conjugate open loop pole or open

273
loop zero pair right what do we do right; how do we calculate the angle of departure and
angle of arrival, ok. So, that is what we are going to learn, is it clear what is meant by angle
of departure and angle of arrival, ok?

So, with this understanding we will go to this step and then read through, it, ok. The angle of
departure from an open loop complex pole is going to be 180 degrees minus sum of the
angles made by the vectors from the other open loop poles to this pole, plus sum of the angles
made by the vectors from open loop zeros to this pole, ok; that is the equation for calculating
the angle of departure, right. So, how do we get this formula, right? So, let me let me take an
example I am going to lead you to this answer, ok. I will leave you with a small exercise to
do before the next class.

(Refer Slide Time: 13:42)

Let us say I have a case where I have three open loop poles and one open loop zero, right. So,
let us say that one open loop pole -P1 is here and let us say -P2 and -P3 are complex
conjugate open loop poles and let us say -Z1 which is real zero, right. Now, I need to figure
out the angle of departure from this let us say open loop pole -P2, right. So, we want to figure
out the angle of departure from -P2, ok. How, do we do this for this let us take a test point
which is very very close to st.

Now, what is the angle condition? Angle(G(st) H(st)) = ±180°(2k+1); k = 0, 1, 2…, right. So,
that means, that angle(K) + angle(st + Z1) – {angle(st + P1) + angle(st + P2) + angle(st +
P3) }= ±180°(2k+1), right. Do you agree? This something which we have already done, ok, I

274
am not doing anything new here. So, this guy is already zero, alright because we are
considering K to be positive alright the phase of K is already 0.

Now, what is angle of st + Z1. It is going to be the angle made by this vector right from the
open loop zero to this to the test point which is very very close to the pole minus P2. What is
angle of st + P1? It is the angle made by the vector drawn from minus P1 to s t. What is an
angle of st + P2? It is this vector. What is angle of st + P3? It is angle made by this vector.
So, immediately we can see that this angle of st + P2 is the departure angle right because st +
P2 that is the direction in which the root locus will start, ok. So, that is the departure angle
and please note that as st tends to -P2 because st is you know like very close to minus P2,
yYou can see that all the other angles become like this, right st plus P 1 can be approximated
as minus P2 plus P1. What is minus P2 plus P1? It is the angle made by the vector drawn
from minus P1 to minus P2, do you agree?

θ dep (−P 2) =180+ angle (−P 2+ Z 1 )−¿

So, with this understanding let us go and read this step and we will stop there. The angle of
departure from open loop complex pole is 180 degrees minus sum of the angles made by the
vectors from other open loop poles to this pole, that is this was; plus sum of the angles made
by the vectors from open loop zeroes to this pole. In the example that we consider there was
only one open loop pole zero. So, that is why we had only one term ok. So, that is what
happened here.

Now, what do you think will be the angle of departure from -P3? It will just be the negative
of this angle or you subtract it from 360 degrees. Why? Because the two thing two should add
to 360 right because these are two complex conjugate poles and any root locus should be a
mirror image about the real axis, right. So, that is the, that is another important point.

So, this statement calculation you do it as homework. The angle of arrival at an open loop
complex zero is 180 degrees minus sum of the angles made by the vectors from other open
loop zeros to the 0 plus sum of the angles made by the vectors from all over all open loop
poles to this 0, it is very very complimentary, right. Now it is pretty obvious how we get this,
once again from the angle condition right, ok. So, this step once again applies only if we have
complex open loop poles and complex open loop zeros, otherwise it does not apply, ok.

275
So, I will stop here we would we would start from this step complete step-6 tomorrow and
then we will construct the entire root locus maybe do another example, which will make
things clear to us, fine, ok.

Thank you.

276
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 37
Root Locus 3
Part-1

(Refer Slide Time: 00:22)

(Refer Slide Time: 00:30)

277
Root Locus (Continued):

Following is the recap for root locus construction steps covered previously.

Step 1: Given an open loop transfer function, the first step was to locate the open loop poles
and zeros in the s-plane. The root locus will have n branches in general, each of those n
branches will start from an open loop pole, and they will end at an open loop zero, if it exists.
And the remaining n−m branches will go to infinity along asymptotes.

Step 2: To get the parts of the root locus that lies on the real axis, the real axis is divided into
sub regions based on the location of open loop poles and zeros. In each sub region a test point
is taken, and if the number of real open loop poles and real open loop zeros is odd on the
right of the test point, then that region lies on the root locus.

Step 3: Asymptotes indicate the locus of the closed loop poles as K → ∞. The angle made by
the asymptotes with the real axis is given by,

±180 °(2 k+1)

, k=0 , 1, 2 … .
(n−m)

The point of intersection of the asymptotes on the real axis is given by,

∑ of openloop poles−∑ of openloop zeros .

(n−m)

(Refer Slide Time: 02:19)

278
Step 4: At a break-away or break-in point, there is going to be a repeated pole. Equating the
first derivative with 0 is the concept in determining a potential break-away and break-in
points.

(Refer Slide Time: 02:50)

Step 5: This step applies only when there is a complex conjugate open loop pole pair, or
complex conjugate open loop zero pair.

The remaining steps for the root locus construction are discussed next.

Step 6: This step is to determine the crossover points. The imaginary axis divides the S-plane
into RHP and LHP. A cross over point, as the name indicates, is a point at which the root
locus transitions from one half plane to other, i.e. from right half plane to left half plane, or
left half plane to right half plane (this transition is important as far as stability is concerned).
In other words, crossover points are points where the root locus cuts the imaginary axis. To
determine the crossover points, substitute s= jω in the closed loop characteristic equation and
solve for the corresponding value of K and ω.

Let us consider the previous example.

We have,

K K
G ( s ) H ( s )= = .
(s+ 4)(s−1) s2 +3 s−4

279
The closed loop characteristic equation is given by,

s +3 s−4=0.
2

(Refer Slide Time: 06:14)

Substituting s= jω we get,

−ω + K−4 + j ( 3 ω )=0 , ⇒ ω=0∧K=4 .

2

Hence, s=0 is a potential solution of the closed loop characteristic equation for a crossover
point. Since K is assumed to be positive, this is a valid solution for this problem. Thus, the
origin is a valid crossover point.

280
(Refer Slide Time: 09:46)

Step 7: The final step is to sketch the root locus using steps 1 to 6.

(Note: ‘rlocus’ command can be used in MATLAB to sketch the root locus)

Consider the construction of the root locus based on above steps for the example. We know
that one branch of the root locus is going to start from −4. The root locus starts at an open
loop pole as K → 0. One branch of the root locus starts from −4, and another from 1 with
crossover at 0. At the crossover point, K=4.

These two branches intersect at −1.5 which is the break-away point.

281
(Refer Slide Time: 12:30)

Previously it was found that the value of K was 6.25. Also, the asymptotes are going to
intersect at −1.5 for this problem (the point of intersection of the asymptotes on the real axis
and the break-away point happened to be the same in this example but is not always the case).

As angle of the asymptotes is ± 90 °, one branch will take off vertically upwards and another
branch will go vertically downwards.

(Refer Slide Time: 14:06)

We can observe from the root locus that the closed loop system is stable when K >4, because
the branch which starts from 1 migrates into the left half plane only after K >4.

282
Next, consider the problem of finding the range of K satisfying the following performance
requirements of settling time, t s <4 s.

Settling time expression is given by (using the expression for the step response of an under
damped 2nd order system),

4 .
ts=
ξ ωn

Which implies,

−ξ ω n <−1, to satisfy the given requirement.

As previously discussed, if the dominant dynamics is going to be that of an under damped

2nd order system and all the closed loop poles lie to the left of the vertical line at −1, will
assure that the settling time t s <4 s. Substituting s=−1 in the closed loop characteristic
equation, we get K=6.

Hence, for any value of K >6, the settling time t s <4 s is achieved.

When K ∈ (6 ,6.25), the system is over damped as the two branches of the root locus are on
the real axis and both closed loop poles are going to be real. At K=6.25, system becomes
critically damped and for K >6.25 it becomes under damped. As response like an under
damped system is desirable, K >6.25 would be a suitable range.

Thus, root locus gives a graphical representation from which these interpretations can be
obtained.

Note: The root locus plot does not scale linearly with K .

Questions:

1. What are the changes in root locus construction when K is negative?

2. Why and how would the root locus change in case of positive feedback?

283
 Control Systems
Prof. C.S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 38
Root Locus 3
Part – 2

(Refer Slide Time: 00:17)

So, now I am going to do an example where we just, to begin with, we consider a mass spring
system ok. Let us say I apply a force f(t) and let us say we get displacement x(t) the spring
being some k ok. Let us say we have a frictionless surface right. We already know that the
governing equation is going to be

m ẍ ( t ) +kx ( t )=f ( t )

X ( s) 1
Plant Transfer function: P ( s )= = 2
F (s) ms +k

1
Let m = 1kg and k=1 N/m, then P ( s )= 2
s +1

The poles are at ±j

284
So, is this system BIBO stable? So, if you have non-repeating poles on the imaginary axis for
one class of inputs the system output is going to blow off to infinity right. So, in this example
for what input your system output will go to infinity? cos t and sin t right.

So, for imaginary roots if you have a sinusoidal input whose angular frequency is going to be
the magnitude of the pole, our output magnitude goes to infinity as t tends to infinity. So, we
want to stabilize the system right. So, let us design a PD controller. So that means,

Kp K∧K p
(
C ( s ) =K d s+ K p =K d s+
Kd )
=K ( s+10 ) ; Let K d =
Kd
=10

K ( s+10)
This implies, G(s) H ( s )= 2
s +1

Now, plot the locus of closed loop poles for K > 0. So, this will be pretty useful for us right
because, I want to see how the root locus varies right. The closed loop poles vary as a change
a controller parameter. See previously using Routh’s criteria what we did, we only got a
ranges of the controller parameters that stabilized the closed loop system. But now, I want to
look at more things right. I want to see how the branches traverse in the complex plane as I
vary a parameter so, that I can design my controller in a better manner right. So, that is the
advantage of a root locus ok that is how we are going to use it for control design ok.

So, I am taking very simple examples so, that we can work them out by hand right. So, see if
I take a higher order system you know hand calculations are going to become more
challenging ok; that is why I am taking second order systems to illustrate the point. So, let us
run through the steps ok, like you tell me what would be step 1.

285
(Refer Slide Time: 06:27)

What is step 1? So, step 1 is to locate the open loop poles and open loop zeros. So, in this
example the value of n is 2. What is the value of m? 1. And what are the open loop poles? ±j
and we have an open loop zero at -10.

So, first let us let us mark those ok. So, let us say I have a zero at -10 and then I have an open
loop pole at + j and - j ok. Those are my open loop poles and open loop zeros.

Now, what is step 2? Step 2 is locate the part of the real axis that lie on the root locus. So, if
you look at the real axis are there any open loop poles or open loop zeros on it, there is one
right. There are no open loop poles on the real axis, but there is an open loop zero on the real
axis. And that cuts the real axis into two sub-regions right. What are they?
(−∞ ,−10 )∧(−10 , ∞). Now, the question is that like which part lies on the root locus? To
figure it out what will I do, I choose a test point st1 here and I look to the right.

How many open loop poles and open loop zeros do I have? 1, which is an odd number. So,
(−∞ ,−10 )lies on the root locus. What about (−10 , ∞ )? I look to the right, let us say I take a
test point st2 in the region from (−10 , ∞ )I look to the right and there are 0 open loop poles and
open loop zeros right to the right of it. So, this region does not lie on the root locus.

So, the only region that lies on the root locus is the region between (−∞ ,−10 ), right so, that
is the region. What is step 3? Asymptotes right so, there will be n-m asymptotes in this case

286
n-m=1. So, there is going to be 1 asymptote ok. And so, what is the angle of the asymptote?

±180 °(2 k+1)

That is going to be equal to =180°.
n−m

You can call it +180 or -180, anyway the angle will be the same. Now, the question is that
like where is the asymptote. See minus 180 degree means, I can have lines parallel to the
negative real axis. But where in the s plane, that will be answered by the point of intersection
although, it is trivial let us do this right.

(Refer Slide Time: 10:17)

So, point of intersection of the asymptote with the negative real axis is going to be the sum of
open loop poles minus sum of open loop zeros divided by n minus m right.

( + j− j ) −(−10)
Point of intersection of the asymptote = =10
1

So, what does this really mean? It really means that the negative real axis is the asymptotes
right? See the asymptote goes along the negative real axis instead of any line parallel to the
negative real axis, the negative real axis is an asymptote right. So, the asymptote is along the
negative real axis. So, that is step 3.

Now, what is step 4? What is step 4? Break-away or break-in. So, how do we calculate break-
away, break-in points? We look at the open loop transfer function, right, the open loop

287
 K A (s)
transfer function is of the form . So A ( s ) =s+10 , B ( s )=s2 +1. See as I told you just
B(s)
need to follow the steps ok; if you follow the steps methodically we will be able to get the
root locus right. So, that is it ok.

Student: [Inaudible]

Negative real axis because the point of intersection of the asymptotes with the real axis is
plus 10. So, what is the only line which will intersect the real axis at plus 10 and still have an
angle of 180 degrees with the positive real axis? So, the asymptote has to necessarily go
along the negative real axis right.

Student: [Inaudible]

See, what is an asymptote? The asymptote is indicates the line along which the root locus will
tend to as it goes to infinity right, that is the meaning of asymptotes. Now, when we say the
asymptote makes an angle of 180 degrees with the positive real axis and intersects the real
axis at +10; the only choice is the real axis itself right. So, that is why the asymptote goes
along the negative real axis right.

Student: Plus minus tend to be angle.

Yeah, plus minus both are the same right, I just took one that is it right. You can take ±180
degrees, ±540 degrees does not matter ok, you will get the same direction right. So
consequently,

' '
A ( s )=1 , B ( s ) =2 s

' ' 2 2
A ( s ) B ( s ) − A ( s ) B ( s ) =0→ s +1−( s+10 ) 2 s=0 →−s −20 s+1=0

The root are -20.05 and 0.05

So, sb =−20.05would imply K=40.1and sb =0.05 implies K=−0.1, right. How do we get

−B(s)
this? K ( s )= .If you solve that you will get this. So, sb =−20.05is a breakaway/break-in
A(s)
point. In this case it is a break-in point, I will explain why.

288
Now, the question becomes why is this a break-in point? Because, you can see that see I am
going have two branches starting from +j and -j ok. They are going to be complex conjugates
right and ultimately one branch is going to come and terminate at -10 because that is one
open loop zero. Another branch is going to go to infinity along the negative real axis. So
obviously, both points should intersect somewhere on the real axis right and that is going to
be at -20.05. So, two branches will come from the complex conjugate plane, hit at -20.05 and
then go to one to the right one, one to the left ok. So, that is what is going to happen.

So, in this case what is going to happen is the following. So, you have an open loop zero at -
10 ok; I am not drawing to scale, but you get the idea right. And what is going to happen is
that you are going to have two branches which are going to start from the two open loop
poles and come and hit the negative real axis at -20.05. And then one branch will go like this
and one branch will go like this ok, that is why it is called as a break-in point right because it
breaks into the real axis ok. So, that is why it is a break-in point ok.

Now, what is the next step? 5th step. What is step 5? It is to calculate the angle of departure
right. So, in this case we need to calculate the angle of departure because we have a bunch
pair of complex open loop poles right.

(Refer Slide Time: 19:01)

So, let us calculate the angle of departure from +j. What is the formula for angle of departure
from plus from an open loop pole? It is 180 degrees minus the sum of the angles made by the
vectors from other open loop poles to this open loop pole right. So, if you look at what is the

289
only other open loop pole, see we are considering +j right. What is the only other open loop
pole?

Student: Minus.

-j so, if I draw a vector from -j to +j; what is the angle made by that vector? 90 degrees. So, I
just write 90 there are no other open loop poles plus sum of the angles made by the vectors
drawn from open loop zeros to this open loop pole. What is the only open loop zero here? -
10, you draw a vector from -10 to +j. And what is the angle made by this vector?

Student: Tan inverse of 1 by 10.

Angle of departure from +j = 180° - 90° +tan-1(0.1) = 95.71°

So, this is the angle of departure from +j. So, what it really means is that, if I consider the s
plane once again and if I consider the pole at +j; the root locus is going to depart at an angle
which is 95.71, right. So, then what can you say about -j, it is going to be symmetric right, it
is a mirror image. Why? Because you can only have complex conjugate poles right. So, you
can immediately figure it out that the angle of departure from -j is going to be either 360
minus this or -95.71. But anyway I will calculate so, that like we will convince ourselves
right. So, you have a zero at -10 now, we are looking at -j right.

So, what is the angle made by the vector drawn from the other open loop poles to this pole? It
is going to be -90 or 270 you can write it either way, let us write it as 270, right. Then what is
the angle made by the vector drawn from the open loop zero to this pole? It is going to be this
angle right, which I can write as 360- tan-1(0.1), right. Because, I want to look at this angle
right I take angles to be positive counter clockwise from the positive real axis. You can write
it as either positive thing or -tan-1(0.1).

So, if you calculate this you will get the answer as 264.29° which is the same as calling it as -
95.71°. So, I am just doing this to convince you that when you have a complex conjugate
open loop pair the angles of the departure are just going to be negatives of each other or they
will sum to and even multiple of pi alright. So, that is important ok. So, they are just mirror
images about the real axis ok. So, that is something which we need to remember ok.

290
So, what I want you to do is do the final step and then construct the root locus. Step 6 is the
cross-over point right. So, how do we get the cross-over point? Consider the characteristic
equation right, close loop characteristic equation is going to be

K (s+10) 2
1+ 2 =0→ s + Ks+10 K +1=0
s +1

So, if I substitute s= jωin this equation. What am I going to get?

We get, (−ω2 +10 K +1 ) + j ( Kω )=0

So now, if you equate both real parts and imaginary parts to 0, what do you get?

K =0→ ω=±1
Kω=0 →
ω=0→ K=−0.1,

(Refer Slide Time: 24:20)

Now, K=−0.1is not under consideration because we are considering K to be positive. So,
anyway these are the open loop poles right ±j. So, in essence there are no further cross-over
points rather than the open loop poles. So, the root locus starts from the open loop poles and
it remains in the left half plane always; for all K greater than 0 ok. So, that is what we can
figure out ok. So, I will quickly draw the root locus; you can go back and think as to how it
came in this shape. And we will start our discussion from this point in the next class ok.

291
So, let me quickly finish there and then we will stop ok. So, the open loop poles are at +j and
-j and we have an open loop zero at -10. So, what is going to happen to the root locus is that
we are going to have two branches ok; which are going to come like this of course, they are
mirror images of each other ok. They intersect at -20.05 then I have one branch going like
this, another branch going like this ok. So, here K tends to 0 at the two open loop poles. The
value of K at the break-in point was 40.01. So, the direction is like this here and it comes to
an open loop zero at K as K tends to infinity about this cases ok. So, this is the root locus.

Why should it take this shape? As I told you, with experience you will get it. You know I can
also, please note that the closed loop characteristic equation is a second order polynomial
right. So, it will take this particular shape ok. So, see we can’t reproduce the exact shape by
hand calculation, but you draw an approximate shape ok, marking the critical points right ok.
So, I will stop here. But please think about the other two questions and come to the class for
Monday ok.

Thank you.

292
 Control Systems
Prof. C.S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 39
Root Locus 4
Part-1

We are looking at Root Locus. To recall, consider the basic block diagram, which has
been our focus.

(Refer Slide Time: 00:38)

G(s)G ( s ) is the forward path transfer function, H (s)G ( s ) is the feedback path transfer
function, and we have a negative feedback.

So, the closed loop transfer function was,

G(s) ,
1+G(s) H (s)

and the closed loop characteristic equation was,

1+G ( s ) H ( s )=0.

We looked at the roots of this equation.

293
Suppose, we have positive feedback with everything else remaining the same, that is,

K ( s+ z1 ) …(s+ z m )
G ( s )= .
( s+ p1 ) …(s+ pn )

In the block diagram, everything else remains the same, other than the fact that at the
summing junction, the two signals are added rather than subtracting them.

Following is an example to examine the steps involved in the construction of the root
locus. In this case, the closed loop transfer function is,

G(s) .
1−G(s) H (s)

The closed loop characteristic equation is,

1−G ( s ) H ( s )=0.

Any value of sthat satisfies the above equation is a closed loop pole.

(Refer Slide Time: 04:35)

This implies that any value of s which makes the open loop transfer function to be equal
to 1 will be a closed loop pole.

This will give us 2 conditions. The magnitude condition being,

294
|G(s) H (s)|=1, and

the phase condition,

phase ( G ( s ) H ( s ) ) =±360 ° k , k=0 , 1, 2 …,

The same example is repeated to understand the root locus.

(Refer Slide Time: 07:06)

Step1:

The first step to locate the open loop zeros and open loop poles in the complex plane
remains the same.

As K → 0 ,the root locus starts from an open loop pole, and go to the mopen loop zeros as
K → ∞ , and along asymptotes ifn>m.

Step 2:

The step of locating the root locus portion of the root loci that lie on the real axis
changes, because it depends on the angle condition.

295
(Refer Slide Time: 09:25)

Considering a test point on the real axis, the number of real open loop poles and real
open loop zeros to the right of it should be even for the test point to lie on the root locus.

Step 3:

The asymptotes angle depends on the angle condition. Hence, the angles in this case are
given by,

± k 360 °
,k=0 ,1 ,2….
(n−m)

There are n−masymptotes, where the point of intersection of the asymptotes on the real
axis is calculated by the same formula given by,

∑ of openloop poles−∑ of open loop zeros .

(n−m)

Step 4:

Step 4 remains the same.

296
(Refer Slide Time: 10:55)

Note that the characteristic equation is,

A(s) .
1−K =0
B ( s)

(Refer Slide Time: 12:03)

The values of K at the potential breakaway break-in points would now be determined by

B(s)
K |s=s =
b |
A (s) s=s
.
b

297
Step 5:

As the angle of departure and angle of arrival uses the angle condition, the formulas for
this case becomes:

Angle of Departure=0°−( ∑ of the angles made by the vectors¿other open loop poles¿this pole )+(∑ of t
.

Angle of Arrival=0 °−(∑ of the angles made by the vectors ¿other open loop zeros¿this zero ) +(∑ of the a
.

Step 6:

Step 6 remains the same, that is, substitute s= jω∈¿the closed loop characteristic
equation.

This concludes the changes in steps for constructing the root locus in the case of a
positive feedback.

Example:

K
G ( s ) H ( s )= , K >0.
(s+ 4)(s−1)

Step 1:

n=2 , m=0 ,

Consequently, 2 branches will start from -4 and 1 as k → 0, and the 2 branches will go to
infinity along asymptotes as k → ∞.

Step 2:

The real axis is divided into 3 regions:

298
(−∞ ,−4 ) , (−4 , 1 ) ,(1, ∞).

Taking test points in these three regions we can conclude the following –

(−∞ ,−4 ) : Lies on the root locus.

(−4 ,1 ) : Does not lie on the root locus.

(1, ∞ ) : Lies on the root locus.

(Refer Slide Time: 18:52)

Step 3:

Number of asymptotes: n−m=2

± k 360 °
Angle made by the asymptotes: , k=0 ,1 , 2…, = 0 °, 180 °.
2

Point of intersection on real axis: −1.5

Hence, 2 asymptotes will intersect at−1.5, and make an angle of 0 °and 180 ° with the
positive real axis.

Step 4:

A ( s ) =1, B ( s )=( s+4 ) ( s−1 ) , A ( s )=0 , B ( s )=2 s+3,

' '

299
A ( s ) B ( s ) −B ( s ) A ( s ) =0 ⇒ sb =−1.5.
' '

B(s)
K |s=−1.5 =
A (s) |
s=−1.5
=−6.25 ,

(Refer Slide Time: 23:09)

As K >0 is assumed, there is no break-away or break-in point.

Step 5:

Since we have only real open loop poles in this example, step 5 is not applicable.

Step 6:

1−G ( s ) H ( s )=0 ⇒ s +3 s−K −4=0,

2

Substituting s= jω we get,

[−ω2−K −4 ] + j [3 ω]=0 ⇒ K=−4,

Hence, there are no crossover points.

300
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 40
Root Locus 4
Part-2

(Refer Slide Time: 00:22)

So, if we sketch the root locus now what do we see? So, let us say this is the real axis,
imaginary axis, ok. The open loop poles were at -4 and +1, right. Let us say I use a red color
to construct the root locus with negative feedback. If you recall what happened with negative
feedback, the root locus was something like this. So, I think the point of intersection was at -
1.5 and then, it broke out and it went in two branches, right. This was K tending to 0, right
and this happened as K tends to infinity. This was with negative feedback, right. So, the red
line indicates with negative feedback, right.

Now, let us look at what happens with positive feedback. Based on the root locus
construction that we did, we can immediately see that with positive feedback, I have one
branch which just goes along which starts from plus 1 and then, in goes along the positive
real axis as K tends to infinity. That is it, right. Another branch will just start at -4 and goes
along the negative real axis as K tends to infinity. That is about it, right. So, the green line
indicates the root locus with positive feedback, ok. So, of course I think here the value of K

301
was 6.25, right. So, please check this. I think from the previous example, you have to indicate
the corresponding points and what was the value of K at the crossover point? K was 4.

So, you can immediately observe that once you have positive feedback, what can you say
about the stability of the closed loop system? For all K > 0, it is unstable, right. We can
immediately observe this, because with positive feedback, you see that one branch of the root
locus that starts from +1 is always in the right half plane no matter what we do with the value
of K, right.

So, one branch is on the left half plane, but then we need both to be in the left half plane, for
closed loop stability, correct. So, consequently we have a problem, right. So, with positive
feedback we are not even able to stabilize the closed loop system with a positive value of K,
all right. So, you can immediately see that how the root locus has changed drastically, right.
Once we change the nature of the feedback that we introduce ok, but I hope the steps are
clear and how we constructed the root locus for this particular case is clear, right ok.

So, now what I want you to do is that I want you to go and repeat the same exercise for the
second problem that we did, right. Do you remember what was the open loop transfer
function?

K (s+10)
G ( s ) H ( s )= 2 , K >0
s +1

So, plot the root locus in the case of positive feedback, ok. So, that is an exercise. I leave it to
you as homework, ok. Once again notice the differences between what happens when we
have a negative feedback and positive feedback in this example also, right. That is something
which you have to do as an exercise. Is it clear? Any questions on this?

We are coming. Good point. So, I am going to ask you that question next, but any questions
on these steps? I am sure everything is pretty clear, right.

Now, we will come to our friends point, right. What happens when K is negative. So, see till
now I have been sticking to a case, where K is positive, but it need not be right because say
let us say K is a controller gain. See like in the example that we did last class, right with the
vibrating system, let us say K be pulled out of the proportional derivative controller and
basically it was a controller gain, right. So, now the question is it is a real number, right. So,
it can have both positive and negative values, but till now we have only plotted the root locus

302
only for positive values of the parameter K. You know what happens when you have negative
values, ok; so that something which you are going to ask ourselves now, right.

(Refer Slide Time: 06:54)

So, the question that we need to ask is what happens when K is negative in the case of
negative feedback?

K ( s+ z1 ) … ( s+ z m )
G ( s ) H ( s )= , K <0
( s+ p1 ) … ( s+ pn )

So, then what happens to the closed loop characteristic equation? With negative feedback
loop closed loop characteristic equation is 1+G ( s ) H ( s )=0.

K ( s+ z1 ) … ( s+ zm )
1+ =0
( s+ p1 ) … ( s+ pn )

Now, what do I do? What I can do is the following, all right. I can say let a parameter
K ≔−K . I can do that, all right. So, I can essentially define a variable K hat which is the
^
negative of K. So, what do you think happens to the range of ^
K ? So, the same closed loop
characteristic equation now becomes

^
K ( s+ z 1 ) … ( s+ z m )
1− =0 , ^
K >0
( s+ p1 ) … ( s+ pn )

303
Now, the answer is obvious, right. So, what do we need to do? We need to follow the steps
that we discussed today. See this looks like a closed loop characteristic equation of a pseudo
positive feedback system with a positive parameter ^
K , ok. Physically the original system is
negative feedback, but this is basically a pseudo positive feedback system with a positive
gain ^
K . You follow the same steps that we learnt today, ok.

In the case of positive feedback and positive gain, you repeat the same stuff, determine the
answers of ^
K , then your root locus will be the same ok, but then what you need to do is, your
answer whatever the answers you get for stability and performance, change the sign. So,
when you quote as in terms of values of K, it should be minus of ^
K . That is all you need to
do, when you have, when you want to go the other way, ok. So, is it clear?

So, now using this I want you to do the following as homework, right. Repeat problem 1 and
2, all right.

K
1¿ G ( s ) H ( s ) = , K <0
( s+ 4 ) ( s−1 )

K (s+10)
2¿ G ( s ) H ( s )= 2 , K <0
s +1

Plot the root locus in this case, ok. Is it clear? Now, let me ask you the follow up question,
ok. Well I think the answer would become obvious to you by now, right.

(Refer Slide Time: 12:15)

304
So, the follow up question is that I am going to ask is that what happens when K < 0 in the
case of positive feedback? We can, answer is pretty obvious, right. If you have positive
feedback, what do you think will happen to the closed loop characteristic equation?

K ( s+ z 1 ) … ( s+ z m ) ^ ( s+ z1 ) … ( s+ zm )
K
1− =0 →1+ =0 , ^
K >0
( s+ p1 ) … ( s+ pn ) ( s+ p1 ) … ( s+ pn )

So, one can immediately observe that this is like a pseudo negative feedback system with a
positive gain ^
K.

So, you repeat the steps with the, what to say odd multiple of π, right. What we learnt
originally construct the root locus, find the values of ^
K , then in order to get the value of K
multiply by -1, ok. Is it clear? So, your homework is once again repeat in this case, ok. So,
that like you practice all the four cases, right that is what are the four cases?

Positive feedback, negative feedback was one attribute, positive K negative K was one
attribute, right. So, then you essentially need to use a correct set of steps, fine. So, if I call the
set of steps that we discussed in the previous discussion as the first set and what we discussed
today as a second set, there is let us say you know we call set 1,

(Refer Slide Time: 15:07)

305
as the set of steps with the angle condition involving odd multiples of πor multiples of 180
degrees and set 2 is the set of steps, the angle condition involving even multiples of 180
degrees, right we can immediately construct a grid like this, right.

Negative Feedback Positive Feedback

K>0 SET 1 SET 2

K<0 SET 2 ( ^
K ≔−K ) SET 1 ( ^
K ≔−K )

K ( s+ z1 ) … ( s+ z m )
G ( s ) H ( s )=
( s+ p1 ) … ( s+ pn )

So, depending on your design choice, you know like as far as see sometimes, many times
what will happen is that we will do a case study tomorrow, like that we will essentially run
through the entire process, ok.

Now, that we have done with root locus design, I will take a physical problem. We will start
from the beginning you know. We will derive a mathematical model, get that plant transfer
function, then we will try to design controllers, then we will try to find out what ranges of
controller gains will stabilize, the closed loop system, then how to construct the root locus to
satisfy certain performance requirements, right. As we would have already realized the root
locus provides us with a graphical representation, right.

So, we can immediately see how easily we can get the ranges of controller gains that would
satisfy performance requirements if you recall performance requirements were expressed or
visualized as an open trapezoid, right. For example, in the S plane left half plane, right. So,
with the root locus, we can immediately get a what the value of K. If at all they exist would
place my closed loop poles in the desired region, right so things become much easier to
visualize and swap, right.

So, that is something which is the entire process. I will run through once tomorrow and
maybe in the next class, right so that you know like we are almost halfway through the
course. You know at this point, it is better to pause and take a stock of what we have done
and how it can be used for in practical design, right. That is my motivation for doing a case

306
study. Tomorrow we will do a case study, then we will go and will have a brief introduction
to state space representation and then, we will go to frequency response which should keep us
engaged to the rest of the semester.

So, what we have been doing is broadly what is called as Time Domain Analysis. Although
like we are analyzing a frequency domain, but by and large you know like the performance
specifications that we are putting forth is in the time domain, right like settling time, rise time
you know and peak overshoot. That is based on transient response, right unit step response in
the time domain, ok. So, some people will call it as you know like control design using time
domain based specifications, all right. We look at equivalent frequency domain based
specifications when you do frequency response, ok. I will stop here and then, like we will
continue tomorrow.

307
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 41
Case Study-Modelling
Part-1

Case Study

Task: Derive the equation of motion for the system to obtain the transfer function. Design a
suitable controller to satisfy performance and stability. Constructing the root locus and using
the root locus for designing the controller.

(Refer Slide Time: 00:52)

A gear transmission is considered for the study.

Let input gear has N 1 teeth and an output gear have N 2 teeth. An input torque of T m is given
to the input shaft (driving torque from a motor) and T l is the load torque.

Let,

J 1 = The moment of inertia of the entire input shaft and input gear assembly.

J 2 = The moment of inertia of the output shaft assembly along with the output gear.

308
θ 1 = Angular displacement of the input shaft.

θ 2 = Angular displacement of the output shaft.

f 1=¿ Viscous friction coefficient for bearing supporting the input shaft.

f 1=¿ Viscous friction coefficient for bearing supporting the output shaft.

The first step is to derive the governing equation of motion. Consider the entire entity as a
system to which the torque T m (t) is the input and the angular displacement of the output
shaft, θ 2 (t ) is the output.

The objective is to control the angular displacement of the output shaft (position control).

Let T 1 (t ) be the reaction torque from the output shaft to the input shaft due to the fact that the
input shaft is driving the output shaft.

Considering the free body diagram of the input shaft we get,

J 1 θ̈ 1 +f 1 θ̇ 1=T m ( t ) −T 1 (t ).

Let T 2 (t ) be the driving torque to the output shaft coming from the input shaft. Considering
the free body diagram for the output shaft we get,

J 2 θ̈ 2 +f 2 θ̇ 2=T 2 ( t ) −T l (t ).

As we require a consolidated governing linear ode that relates the input to the output, we
need to eliminate T 1 (t ), T 2 (t ), θ 1 (t ), and T l (t ) can be modelled as a disturbance.

Assuming that there is no slip at the gear interface we have,

309
(Refer Slide Time: 10:41)

θ 2 (t) N 1
= .
θ 1 (t) N 2

This implies,

θ̇ 2 (t) N 1
= , and
θ̇ 1 (t) N 2

θ̈ 2 (t) N 1
= .
θ̈ 1 (t) N 2

Neglecting the energy loss, we get,

N2
T 2 (t )=T 1 (t ) .
N1

Rearranging terms to eliminate T 1 (t ), T 2 (t ), and θ1 (t ) we get,

N2 2 N2 2 N2
[( ) ] [( ) ] ( )
J1
N1
+ J 2 θ̈ 2 t + f 1
( )
N1
+ f 2 θ̇ 2 ( t )= T ( t )−T l (t).
N1 m

310
 (Refer Slide Time: 17:18)

This concludes the derivation of the governing equation for the gear transmission system.

311
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 42
Case Study-Modelling
Part-2

(Refer Slide Time: 00:21)

So, let us get the system transfer function. So, to get the system transfer function, obviously,
what should we do? Take Laplace transform on both sides and apply zero initial conditions
right. So, this is a very important step and since we have done this step many times, I am just,
going to skip and go directly to the final answer,

2
L [ θ̈ 2 ( t ) ]=s θ 2 ( s )−sθ 2 ( 0 )−θ̇ 2 (0)

L [ θ̇ 2 ( t ) ]=sθ 2 ( s ) −θ2 ( 0 )

So, once we apply the Laplace transform and substitute initial conditions to be 0 and collect
the terms and so on. What are we going to get?

θ 2 ( s )=¿

So, this is what we will get right. Is this correct? Did I miss with something? I think it is fine
right ok. So, you see that the finally, the output of the system θ 2 has two components right.

312
(Refer Slide Time: 04:16)

So, the first term is the one due to the input. The second term is due to the load torque right.
So, please note that if I want to move the output shaft to any position you can immediately
see that the load torque is going to play a role all right. So, we need to keep that in mind. So,
just for further analysis and design, momentarily, we will neglect the load torque ok. So, that
we can carry on with the analysis, then we will revisit towards the end and see how our
design is affected by the load torque ok. That is something we will come back and look at it
ok. So, for the time being, let us neglect the load torque TL ok.

θ 2 (s) N2/ N1
P ( s )= =
T m (s) [J ¿¿ 1 ( N 2 / N 1 ) 2 +J 2 ] s2 +[ f 1 ( N 2 / N 1 )2 + f 2 ]s ¿

N 2/ N 1
a a= 2
This, this I can rewrite it as 2 where 2 [f 1 ( N 2 / N 1 ) +f 2 ]
s + bs [J ¿¿ 1 ( N 2 / N 1 ) +J 2 ], b= 2 ¿
[ J ¿ ¿1 ( N 2 / N 1 ) + J 2 ]¿

So, immediately we can see that this is a second order system. One can immediately note that
the parameters a and b are positive and what are the system poles? We have a pole at the
origin, we have pole at -b right. So, is this plant going to be stable? No right, because of
course, it depends on how we look at it as we discussed some people will call as critically or
marginally stable, but we know that when we have a non repeating pole at the origin and you
give a step input, the output will become unbounded right.

313
So, you can immediately picturize it, let us say you have a motor shaft and a gear
transmission. You give a step voltage, you will see the output shift shaft will keep on
rotating; You might have observed it even in your experiment on DC motor. So, essentially
the output shaft might have just kept on a rotating right. So, if you do not apply any control to
the system right that is what is going to happen. So, now, the question is, what do we do?

So, for the sake of simplicity, a=2 and b=10 ok. So, this implies that the plant transfer

2
function is P ( s )= . So, now the question is for us to design a closed loop feedback
s(s+10)
system and then ensure that the closed loop system is stable and then ensure that it satisfies
some performance requirements ok. So, that is the second task which I am going to pose ok.

(Refer Slide Time: 10:03)

So, task 2 is to design a stable unity, negative feedback control system that satisfies the
following ok. Settling time less than 2 seconds; peak overshoot Mp less than 10 percent ok.
So, the task for us is to design not only a stable closed loop feedback system, but also one
that will satisfy this performance requirements right of settling time less than 2 seconds and
peak overshoot less than 10 percent ok.

So, what I am going to do is that I am going to stop here ok. I will let you think about it and
then we will complete this task tomorrow ok. So, that we go through, the entire control
design process once right. As I told you we are at the halfway mark before going into, the
next chunk of topics.

314
Now, I thought let us work out a problem from start to finish right, based on whatever we
have learnt. So, that we understand. So, just try out of proportional control, you know like,
you will see that in this case even a proportional controller would stabilize the closed loop
system and then think about how you would, you can use root locus to ensure that you can
satisfy these performance requirements ok. Please, think about that and then like we would
continue in the next class.

315
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 43
Case Study-Control Design
Part I

Case Study of a Gear Transmission (Continued):

(Refer Slide Time: 00:29)

An input shaft is provided a torque to drive an output shaft. The transfer function is of order 2
and have a pole at the origin.

316
(Refer Slide Time: 00:55

(Refer Slide Time: 01:02)

Aim is to design a closed loop feedback control system that would stabilize the closed loop
system and satisfy the performance requirements of the settling time t s < 2 s and peak
overshoot M p< 10 %.

317
(Refer Slide Time: 01:43)

In this section, a controller is designed to satisfy performance requirements.

Plant transfer function:

2 .
P ( s )=
s(s+10)

Let us choose the controller transfer function as a proportional controller. The controller
transfer function is,

C ( s ) =K p.

The closed loop transfer function is,

Y (s) 2 Kp
= 2 .
R(s) s +10 s+2 K p

Closed loop characteristic equation is,

s +10 s+2 K p.
2

For the 2 roots to lie in the left half plane for closed stability, all coefficients need to be non-
zero and of the same sign. This implies that,

K p > 0.

318
For performance requirements consider the open loop transfer function,

2 Kp
G ( s ) H ( s )= , K >0 .
s (s+10) p

To design for performance, root locus techniques can be utilized.

(Refer Slide Time: 06:47)

To satisfy the requirement of the settling time t s < 4 s, and comparing with that of an under
damped second order system; implies,

4
<2 ⇒−ξω<−2,
ξω

For satisfying the requirement of maximum peak overshoot M p < 10 %,

ξ> 0.591, ⟹ β<53.76 °.

The net feasible region that will satisfy both performance conditions will be open trapezoidal
region.

Root Locus Construction Steps ( K p > 0):

Step 1: Locate the open loop poles and open loop zeros in the S-plane.

Open loop poles: 0 ,−10.

319
Open loop zeros: Nil.

n=2 , m=0.

(Refer Slide Time: 12:40)

Step 2:

Regions of the real axis that lie on the root locus:

(−10 , 0).

Step 3:

No. of asymptotes: n−m=2

±180 °(2 K +1)

Angle of the asymptotes: =±90 ° (¿90 ° ,270 °).
n−m

Point of intersection of the asymptotes: −5.

Step 4: Consider the open loop transfer function.

320
(Refer Slide Time: 18:04)

In this case, A ( s ) −2, B ( s )=s ( s+10 ) , A ( s ) =0 , B ' ( s )=2s+10 .

Hence,

' '
A ( sb ) B ( sb )−B ( sb ) A ( sb ) =0 ⇒ sb=−5 .

Thus,

B ( s)
K p|s=s =−
b |
A ( s ) s=s
=12.5 >0⇒ sbis break away point.
b

321
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 44
Case Study-Control Design
Part - 2

(Refer Slide Time: 00:16)

So, s b equals minus 5 is a breakaway point right. So, that is what we have done.

322
(Refer Slide Time: 00:20)

So, that is why it breaks away and anyway it goes along asymptotes right, but anyway we
will come and plot it as we go along right.

(Refer Slide Time: 00:32)

So, what is Step 5? So, we will go step by step right. So, step 5 was to.

Student: [Inaudible]

323
Get the angle of departure or angle of arrival right, but it does not apply here right because
why? there are no complex open loop poles or complex open loop zeros here right. So, step 5
does not apply right for this problem right.

So, let us look at Step 6. So, step 6 is about getting the crossover points. So, for the cross,
determination of the crossover points please note that you need to have a look at the closed
loop characteristic equation right. So, what is a closed loop characteristic equation? It is
going to be 1 plus G of S, F of S equals 0, right. So, that is what we need to look at. So, that
is going to be in this particular case 1 plus 2 K P divided by S times S plus 10. So, that is so,
we are going to get S square plus 10 S plus 2 K P is equal to 0.

Here what we do? We are looking for solutions of the form S equal j omega right, because
that’s a crossover point right; crossover point at some point on the imaginary axis where a
root locus branch may cross from the left of plane to the right of plane or vice versa. So, that
is a crossover point right. So, we substitute S equal j omega. So, what do we get? We get
minus omega square plus j times 10 omega plus 2 K P is equal to 0. So, this will give us 2 K
P minus omega square that is the real part plus j times 10 omega equals 0 right.

So, this immediately implies that from the imaginary part omega is equal to 0 is the only
possibility; that means, the origin right, and but already we know that the open loop pole is at
the origin alright. So, you from the, if omega is 0 that is what we get from the imaginary part
immediately see that the real part will give us K P is equal to 0 correct, see 2 K P minus
omega square is equal to 0, you substitute omega is equal to 0 what will you get; 2 K P is
equal to 0 or in other words K P should be 0.

So, omega equals 0 K P equals 0 is the only crossover point; that means, that the root locus
does not cross the imaginary axis except when it starts at the open loop pole that is it. So, that
is what we can interpret from this calculation. So, there are no further crossover points.

So, it starts at the origin when K is 0, when K tends to 0 because that is an open loop pole
right. So, for any K P greater than 0, so, the conclusion we can draw is that for all K P greater
than 0 the root locus does not cross the j omega axis. So, that is the interpretation we can
have, yes.

Student: [Inaudible].

324
Zero does not lie on the root locus.

Student: [Inaudible].

Yes absolutely, but we need to be careful right. So, I got your question.

So, I think we might have discussed this point because you are right if zero does not lie on the
root locus from step number 2, we can straight away say that the origin is not a crossover
point, but I can have a complex conjugate crossover point once again; see imagine this you
know like we can have two branches which are cutting the imaginary axis.

In fact, which happened if you recall last week right we considered another example where
we had a open loop poles at plus or minus j it was not a crossover point, but the 2 branches
started on the imaginary axis you are correct, but my what I am adding to a statement is that,
that does not that only says that zero is not a crossover point it does not exclude the fact that
we can have crossover points on the imaginary axis right.

So, let us go back and complete the root locus. So, what is going to happen is that like we are
going to have 2 branches starting from a one starting from minus 10, one starting from 0, they
break away at minus 5 and then they just go to infinity along the asymptotes. So, that is what
is going to happen.

So, now, the question is for what values of K P this is the root locus right we have plotted the
root locus. So, now, you can see the advantage of using this method right. So, we have a
visual characterization right or visualization of the two branches of the closed loop poles
right. So, 2 branches of the root locus each one corresponding to one closed loop pole right.

So, you can immediately see that for all K P greater than 0 both branches lie in the left half
plane. So, as we already know K P greater than 0 implies closed loop stability, now we want
closed loop performance right. So, the question is for what range of K P would both branches
lie in that open trapezoidal region right, that is the question we need to ask ourselves. So, how
can I determine that? see where is the open trapezoidal region let me redraw it highlight it in
green once again. So, it is this right. So, for what values of K P would the 2 branches lie
within this region, how do I find it?

First I need to find the value of K P at minus 2 right, because K one branch of the root locus
starts from 0 keeps on increasing sorry keeps on coming to the left as K P keeps on increasing

325
and it comes with in this performance region only when it crosses minus 2 right S equals
minus 2 right. So, how do I find the value of K P at s equals minus 2; obviously, I substitute S
equals minus 2 in the close loop characteristic equation, what was the closed loop
characteristic equation, it is this.

So, if you substitute S equals minus 2 here what do you get for K P, K P should be.

Student: 8.

8 right, you substitute S equals minus 2 right. So, what we are going to do is, that so, consider
the closed loop characteristic equation right S square plus 10 S plus 2 K P is equal to 0. So,
here if you substitute S equals minus 2 we will get K P to be 8 alright. So, that is pretty
straightforward algebra correct.

So, K P is going to be 8 here. So, now, you can see what we have done right, for stability we
wanted K P to be greater than 0 right, for performance K P should be greater than 8 right only
then both branches will enter into this trapezoid, but not only that we have to find out what is
the value of K P when the root locus hits this point and this point right because beyond this
the two branches of the root locus go outside the desired region right. So, how can I find the
value of K P at this point?

First I need to figure out you know like what is that point all right, how do I find out the
coordinates of that point, I hope it is clear what I am doing right I am figuring out for what
range of K P would the two branches or root locus lie in this green what to say the region
open trapezoidal region which is bounded by this green colored lines right. So, how do I
calculate this point of point I can use angle right. So, can you tell me what is the coordinate
of course, it is going to be my real part is minus 5 what is going to be the imaginary part
yeah, what is going what is that going to be 5 tan 53.76 right.

So, what do you think you will get. So, what is this value and similarly what is this one, what
is 5 times tan of 53.76, please take tan 53.76 multiplied by 5 what do you get?

Student: 6.82.

6.82, so, this is going to be j times 6.82 and this is going to be minus j times 6.82, now how
do you find the value of k at that point or K P at that point? Once again we substitute the
corresponding value of S in the closed loop characteristic equation right alright. See what is

326
the value of S here? Minus 5 plus 6.82 j please plug it into closed loop characteristic
equation, if you do that you know like please check it you know like I got the answer as
35.76.

So, that is what I got the answers like, so, if you substitute it in that closed loop characteristic
equation 35.76. So, what does this imply? This implies that for performance I need to
essentially have my proportional gain to be between 8 and 35.76 that is it right. So, what I
can conclude is that, for satisfying closed loop stability and performance K P should be in the
region 8 and 35.76.

So, please note that see initially to begin with K P was a real number right between minus
infinity to plus infinity stability essentially reduced that region into half only from 0 to
infinity, but that is that is that is once again a big region right. So, through this analysis what
we have done, we have figured out that look you know like 0 to infinity will give me closed
loop stability, but then 8 to 35.76 will give me performance also the way I desire it right. So,
do not you think this is a very useful tool to know, you know like we. So, please note that
what we have done we have progressively reduced a huge region of proportional gain minus
infinity plus infinity first we reduce to 0 to infinity, then 0 to infinity now has been reduced to
8 to 35.76 right.

And even if I go back to the root locus and look at this in fact, I would design be for some K
P between 12.5 and 35.76 right, because by and large we would want our closed loop system
or control systems to be slightly under damped right. So, if I want to satisfy the performance
specifications close and closed loop stability and still the design and under damped second
order system right I would choose K P between 12.5 and 35.76.

I can also choose K P between 8 and 12.5 that will also satisfy performed specification, but
the system will be over damped, see if you have an over-damped system where is maximum
peak overshoot. So, maximum peak overshoot anyway is not there right and yet I can satisfy
the condition on setting time right so, that is essentially one; this one.

Of course, it is going to be slightly sluggish right, because the settling time expression
assumes that you know like you are going to have an underdamped second order system. So,
in a certain sense you know like what I would do is that like I would pick K P between 12.5
and 35.76. So, that it is consistent with whatever analysis we have done, because please

327
remember how did we get the desired region of closed loop poles to begin with by using the
step response of an under damped second order system right.

For an under damped second order systems the poles should be complex conjugates right,
when are the poles complex conjugates as for the root locus here when K P is greater than
12.5 right. So, I would choose K P between 12.5 and 35.76 and tune the final system. So, this
way the analysis theoretical analysis can help us figure out a range of controller gains, then
we need to go to the actual experimental system and fine tune. So, that is what ultimately we
would end up doing, is it clear?

Student: (Refer Time: 16:59).

Because why is 12.5 the transition point right, because please note that that 2 branches of
root locus are starting from minus 10 and 0 all right and when K P increases from 0 you can
immediately see what is happening right. So, one branch is going to come on this real axis
right in this manner, another branch is going to start from 0 and travel to the left.

So, what do these branches indicate they are the closed loop poles? So, till K P equals 12.5
you can immediately see that both branches are on the real axis. So, then what can you say
about the closed loop poles they are real. If you have two real distinct poles for a second
order system what is it called, over damped right. So, at 12.5 they intersect they become
critically damped and then beyond 12.5 you see that the two branches go into the move away
from the real axis. So, you have complex conjugate closed loop poles right. So, the system
becomes under damped yeah.

Student: [Inaudible]

Because we are dealing with a symmetric root locus right. So, because if you are have going
to have a what to say two closed loop poles you know like at the same value of K P think
about it this way right K P is a parameter right. So, let us say you substitute K P equals 35.76
you are going to have one closed loop characteristic equation right you are going to have
some constants as parameters.

So, when you calculate the roots you know since we are dealing with real polynomial that is
polynomials with real coefficients, if you have a complex root the conjugate also must be
root. So, conversely if you have if you parameterize the polynomial using one parameter you

328
choose any two complex conjugate roots which are solutions to the polynomial the value of
the parameter should be the same right otherwise there will be a dicot in right is it not.

That is why K P is the same right. So, it does not matter whether it is minus 5 plus 6.8 to j or
minus 5 minus 6.82 j fine so, I hope it is clear right. So, what we have done is that like we
have just ran through one study where we can generate models and do this process right so, of
control design right. So, that is just to give a flavor of what we could do with whatever we
have learnt.

And you can immediately see that root locus is a very handy tool to know right because like
you can have a graphical visualization. See if I had asked you know like without a root locus
how do you satisfy these conditions performance conditions would have been very difficult
right it would have been more challenging to get these bounds right on K P; with root locus it
was quite straight forward and also very intuitive right because we can see the graphical
visualization right.

(Refer Slide Time: 20:30)

So, what I am going to leave you is a homework problem where this is just for your own
learning process right. So, plot the root locus for the same system when K P is negative. So,
this is just an exercise for you. So, that like you get familiar with the other ways of plotting
the root locus also. So, you immediately; obviously, you will see that when K P is negative,
already we know that you know like closed loop system is not going to be stable, you will see

329
that one branch will always go in the right half plane. So, that is what will happen when K P
is negative.

But the reason I am asking you to do it is because see many times people would want the root
locus to be plotted for a parameter that is varied from minus infinity to plus infinity, then you
cannot essentially plot at one go what you need to do is that if you give me a transfer function
like this and you ask me to plot the root locus for K P going from minus infinity to plus
infinity right, because why what was the open loop transfer function, it was 2 K P divided by
S times S plus 10.

Now, you asked me plot the root locus you know for K P for all real values of K P right. So, I
need to go from minus infinity plus infinity right. So, what I do is, that I divide that region
into two sub regions minus infinity to 0, 0 to infinity then I plot the root locus for one region
let us say minus infinity to zero; that means, K P is negative and then 0 to infinity; that
means, K P is positive, then I get the complete root locus. So, that is what I want you to do
right, we have plotted the root locus for a K P greater than 0 as homework, what I want you
to do is, I want you to plot the root locus or for K P less than 0. So, there you get the
complete picture; obviously, you will realize that the closed loop system is not going to be
stable for K P less than 0, right.

So, we will just a confirm that or support that result you know like using the root locus fine
so, please do this. So, this completes our discussion on root locus. So, what we are going to
do is it like from the next class you know like I am just going to go to frequency response,
but before we go to frequency response in the next class I am going to give you a brief
introduction to what is called as the state space representation right, which forms the basis for
what is called as modern control theory you know which you will learn maybe as the next
course and controls if you chose to go into control systems.

And in fact, we will use those tools also in or advanced control courses like whichever
courses you take right on advanced controls. So, I will just introduce that to you and show
you the equivalence between the two approaches, but this course is going to be about transfer
function right. But since we are at the halfway point I thought that is a good time to introduce
state space to you right then we will go to what is called as frequency response analysis, we
will do frequency response that is a response of this class of systems to sinusoidal inputs we
will study those characteristics and then we learn how to analyze systems for the frequency

330
response and then like design, controllers using frequency response. So, that is going to be
the second half of our course fine.

So, I will stop here and then like we will meet on Monday.

Thank you.

331
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology Madras

Lecture – 45
State Space Representation
Part – I

We are going to learn about the State Space Representation. This is an alternative
representation of the class of dynamic systems that we are studying. This course is more
towards analyzing systems, where we represent them using what are called as transfer
functions.

The state space representation is another commonly used method to essentially do the same
thing as what we are doing with the transfer function. The main difference is that we stay in
the time domain rather than going to the complex domain when we talk about the transfer
function of the system.

(Refer Slide Time: 01:18)

Let us start with an example. Let’s say that I have a mass-spring damper system. The
governing equation of the mass-spring damper system is

m ẍ ( t ) +c ẋ ( t ) +kx ( t )=f (t )

332
For this system or plant, the input is a force f (t ) given to the mass m. Let us consider that the
output of the system is the displacement x (t).

X ( s)
We have already derived its transfer function. The plant transfer function was P ( s )= and
F (s)
all the initial conditions were 0 and for this particular mass-spring damper system, the

1
transfer function is 2 .
m s +cs +k

Now what is this state space representation? The basic idea behind the state space
representation is to rewrite an nth order ODE (Ordinary Differential Equation) as a set of n 1st
order ODEs.

To do that we define what are called state variables. So, what are state variables? For
example, let us say in this mass-spring damper system, I can consider the displacement of the
mass m and the speed of the mass ẋ ( t ) as state variables.

(Refer Slide Time: 04:58)

Suppose someone gives me information about the displacement of the mass and the speed of
the mass at each and every instant of time, I would be able to completely characterize, what is
happening to that mass. So, state variables are those variables whose knowledge at each and
every instant of time is sufficient to completely characterize the system. This is the definition

333
of state variables. In this particular mass-spring damper system, I could take the displacement
of the mass and the speed of the mass as the 2 state variables to characterize the system.

In advanced courses and controls, we would realize that the state space representation for the
same system can be non-unique and that is used to our advantage when we do analysis.

Q: How do I get the state space representation of the system? (or) If I want to write the
governing equation of any system in the state space representation, what do I do? (or) How
does one represent an nth order system in the state space form?

A: The first step is to select n state variables because, in general, we are dealing with an n th
order system. The second step is to write n 1 st order ODEs, each one characterizing the time
evolution of the n state variables.

(Refer Slide Time: 08:42)

We have n state variables. You essentially write down ‘n’ 1 st order ODEs each one
characterizing the time evolution of each state variable. The main mathematical tool which
will be useful in the state space representation or analysis using state space representation is
linear algebra. A good working knowledge of linear algebra is important.

Let us go back to the example. Why am I selecting 2 state variables? Because, the value of n
is 2 here right, so for our mass-spring damper system the value of n is 2.

334
State variables are denoted by the character x. Let the first state variable be x 1. Do not
confuse this x with the displacement x. It is a convention to write the state variable using the
character x. The first state variable x 1, let us take it as the displacement x ( t ) and let the second
state variable x2 be the speed ẋ ( t ). Then, we can immediately observe that ẋ1 ( t ), which is the
first derivative of the first state variable x 1, is going to be equal to ẋ ( t ). If I differentiate the
first state variable, I get ẋ ( t ). But what is ẋ ( t )? ẋ ( t ) dot is going to be equal to x2 ( t ). The idea
is to write these time evolution equations in terms of the state variables themselves and the
inputs.

ẋ1 ( t )= ẋ ( t )=x 2 ( t )

I defined 2 state variables as the displacement x and the speed ẋ ( t ) and I am taking the first
derivative of each state variable. So first state via first derivative of x1 (t ) is ẋ1 (t ), that is ẋ ( t )
by our definition and ẋ ( t ) is nothing but x2 (t ) once again by the choice of our state variables.

Then we have ẋ2 (t ). What is ẋ2 (t ) going to become? ẍ (t ).

Now I need to rewrite ẍ, which is the acceleration, in terms of the state variables themselves.
How can I do that? The idea is to write equations which characterize a time evolution of the
state variables in terms of first order ODEs. So, how can I write ẍ (t )? I can use the governing
equation.

How did we get the transfer function? The transfer function came from the governing
equation of motion; we use the governing equation of motion for the system to derive the
transfer function. Similarly, we have to use it for the state space representation also. From the

−k c 1
governing equation of motion I can write ẍ as x ( t )− ẋ ( t ) + f (t) by rearranging.
m m m

If I take c ẋ andkx to the other side, you get a negative sign and then you divide throughout

−k c 1
by m. So, if I do that I get ẍ as x ( t )− ẋ ( t ) + f (t).
m m m

Now what is x ( t )? It is the first state variable. I can replace x ( t ) as x1 ( t ) and ẋ ( t ) is going to be
x2 ( t ). f (t ) is the input and I retain it as it is. So, these two equations, equation 1 and equation
2 are the 2 state equations.

335
What are the state equations? State equations are those first order ODEs that talk about or
characterize the time evolution of each state variable. So those are what are called as state
equations. We need to write the ‘n’ 1 st order ODEs, each one characterizing the time
evolution of the ‘n’ state variables.

Now, we define let the state vector. It is called as a state variables vector or the state vector or
the vector of state variables. Here, we have a state vector x ( t ). In my notation, vectors are
denoted by lowercase characters in bold font.

Let us choose the state vector as a column vector where I arrange the 2 state variables x 1 and
x2 .

(Refer Slide Time: 16:46)

ẋ1 ( t )
Now ẋ, which is a first derivative of the state vector, is
[ ]
ẋ 2 ( t )
. That is nothing but something

x1 ( t )
multiplying
[ ]
x2 (t )
plus something multiplying the input f (t ).

I have collected the n state variables as an nth order state vector. In general, the dimension of
the state vector x (t ) is ‘n’.

336
ẋ1 =x2. The first row is [ 0 1 ] and in the term multiplying the input f (t ), I write 0 as a first

x1
entry. Now, ẋ1 =x2. So, the row [ 0 1 ] multiplies the column
[]
x2
. You get x2 . 0 multiplying

f ( t ) will not contribute anything additional. So we are getting the first state equation.

−k −c 1
How will I get ẋ2 ? I can rewrite this as , , and . If I do this can I get the second
m m m

1
equation? I do; If you multiply the second row with the column state vector and then
m
multiplies f (t ), you get the second equation. So, this is what is called as a state equation.

0 1 0
ẋ ( t )=
[ ] [(
x˙1 ( t )
x˙2 ( t )
= −k
m ) ( )[
−c
m ] ] [ ( )]
x1 ( t )
x2 ( t )
+ 1 f (t )
m

This state equation is a vector state equation because x is a vector. This is a vector ODE.

In general, what happens is that this column vector (state vector x (t )) is multiplied by a
matrix A which is called as the STATE MATRIX. In general, the dimension of the state
matrix will be n ×n. For SISO systems, the vector multiplying f (t ) is denoted by lowercase b
which is called as the INPUT VECTOR.

The general state equation for a SISO LTI causal dynamic system is

ẋ ( t )= Ax ( t ) +b u(t )

State equation is a vector ODE. But a first order vector ODE. It tells me how the state vector
would evolve with respect to time in relation to the input that we are providing to the system.
ẋ (t ), which is the time derivative of x (t ), is related to the state vector itself and the input
provided to the system.

If you have some initial condition for the state and you give an input to the system, then what
happens to the state variables with time? This is what the state equation will tell. The state
equation is only the first equation in the state space representation. The second equation is the
output equation.

(Refer Slide Time: 23:14)

337
Let us say we call the output as a value y (t). For our example, the output is the displacement
x . So, what is x in terms of the state variable? It is x 1. So how can I get x 1 from the state
vector? In this example, my output is displacement x, which is the first state variable. This

can be obtained as a dot product of a vector and the state vector. I choose the first vector [10 ]
appropriately depending on what my output is. Since we are dealing with a linear system, the
output is going to be a linear combination of the state variables. So, the vector with which I

take the dot product will be appropriately varying. In this example it is [10 ]. This is called as
the output equation. This vector is denoted by a vector c which is called as the output vector.

The general output equation for a SISO LTI causal dynamic system is going to be

y ( t ) =c ∙ x ( t ) +d u(t)

What is this d u ( t ) ? d u ( t ) is what is called as a direct transmission term which comes in the
output equation when m=n.

What are m and n? If you recall the general governing equation of the class of system under
study, the ODE, ‘n’ is the highest derivative of the output term on the left hand side and ‘m’
is the highest derivative of the input term. In this example, m=0 and n¿ 2. So, we did not get
a direct transmission term in the output equation.

But in general the output equation is y ( t ) =c ∙ x ( t ) +d u(t).

338
Direct transmission term is called so because, when you have m=n, the state vector will be
affected by the input that you are getting. But the output term b, in addition to being
influenced by the state vector, will have one component which will be directly influenced by
the input.

If m=n for any transfer function, when you take the partial fraction expansion and inverse
Laplace transform, you will have a constant which will come in. The numerator and
denominator will have second order polynomials. The quotient will be a constant and you
will have a remainder. The quotient will automatically multiply the input. You have one term
which will directly come from the input and the other term will be from the dynamics of the
system.

The direct transmission term will essentially imply the term that is directly affecting the
output through the input. Almost all textbooks references in controls would write this output
equation as y ( t ) =cT x ( t ) + d u(t ). They will take the output vector to be a row vector. It is a
matrix operation. We have c T =[ 1 0 ] .

(Refer Slide Time: 29:54)

339
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 46
State Space Representation
Part – 2

Summary of State Space Representation of a SISO LTI causal dynamic:

(Refer Slide Time: 00:22)

State equation: ẋ ( t )= A x ( t ) +b u ( t ) ,

Output equation: y ( t ) =c . x ( t ) +d u ( t ) .

Unlike Laplace (or transfer function) representation where we can directly relate input to
output, for state space representation, we first solve for the state variables and then get the
output from the state variables. For this reason, transfer function representation is called as an
external representation of the system because it directly relates the input and the output. The
state space representation is called as an internal representation of the system because we go
through the state vector.

The set {A, b, c, d} is called as a realization of a system.

The state space representation of a MIMO LTI causal dynamic system is as follows.

340
State equation: ẋ ( t )= A x ( t ) +B u ( t ) ,

Output equation: y ( t ) =C x ( t ) + D u ( t ) .

Here, u and y are vectors.

If, for example, we have ‘p’ inputs and ‘q’ outputs, p and q > 1, the size for A will be nxn, B
(input matrix) will be nxp, C (output matrix) will be qxn, and D (direct transmission matrix)
will be qxp.

In the MIMO domain, the matrices {A, B, C, D} form a realization.

(Refer Slide Time: 09:37)

In summary, for LTI causal dynamic systems, once we get the model in the form of linear
ODE’s with constant coefficients, application of Laplace transform with zero initial
conditions gives the transfer function representation. If the model is re-written as n first order
ODEs, we get the state space representation. For transfer function representation, analysis is
done in the complex domain whereas for state space representation, analysis is done in the
time domain. Transfer function representation, once the model is fixed, is unique, whereas
state space representation is non-unique.

The equivalence between transfer function and state space representation is as follows.

(Refer Slide Time: 15:43)

341
Consider a SISO LTI causal and dynamic systems. The state space representation is

ẋ ( t )= A x ( t ) +b u ( t ) ,

y ( t ) =c . x ( t ) +d u ( t ) .

Taking the Laplace transform, we get

s X ( s )− x ( 0 )= A X ( s ) +b U ( s ) .

Let the initial conditions be zero, i.e., x (0) = 0. The above equation simplifies to

s X ( s )= A X ( s ) +b U ( s ) .

Taking X (s) common on the left-hand side,

[ s I − A ] X ( s )=b U ( s ) .

Assuming the inverse exists,

X ( s )=( s I − A)−1 b U ( s ) .

Similarly, taking the Laplace transform of the output equation,

−1
Y ( s ) =c . X ( s ) +d U ( s )=c . ( s I − A ) b U ( s ) +d U ( s )

¿>Y ( s )= [ c . ( s I − A )−1 b+d ] U ( s)

342
(Refer Slide Time: 19:46)

In the above equation, the term in square brackets is the plant transfer function (P (s)). For
the general class of SISO LTI causal system, the plant transfer function is related to the state
space representation through

P ( s ) =[ c . ( s I −A )−1 b+d ]

HW: Check this relationship for the mass spring damper system.

HW: Calculate the poles of the transfer function and the eigenvalues of the state matrix and
comment.

343
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 47
Frequency Response
Part-1

We were looking at the state space representation. We rewrote the mass spring damper
system governing equation in the state space form.

(Refer Slide Time: 01:18)

Then, we got an expression between the transfer function and the state space representation.
−1
P ( s )=c ⋅ ( s I − A ) b+d

I asked you to check the relationship for the mass spring damper system. If you just work it
out, you will see that you get the transfer function itself. I asked you to calculate the poles of
the transfer function and the eigen-values of the state matrix. The poles of the transfer

−c ± √ c 2−4 mk
function are . It is observed that the eigen-values of the state matrix also turn
2m
out to be the same. When the poles of the transfer function and the eigenvalues of the state
matrix are the same, it is what is called as minimal realization. A minimal realization is one

344
which is both completely controllable and completely observable This is dealt in detail in
advanced courses on control theory.

345
(Refer Slide Time: 01:23)

(Refer Slide Time: 03:15)

For BIBO stability, we wanted all the poles to be the left of the complex plane when we use
the transfer function approach. Similarly, if I use the state space approach, for asymptotic
stability, I should have all the eigen-values of the state matrix to be in the left half complex
plane. So, the task of ensuring asymptotic stability boils down to the location of the poles of
the transfer function or the eigen-values of the state matrix. That is the equivalence which I
wanted to show you here. Our course is mainly focused on transfer-function based analysis,
but I just wanted to give you an introduction to state space and we would learn more about
this in advanced courses on control.

346
(Refer Slide Time: 08:08)

We are going to discuss what is called as frequency response. Till now, predominantly we
have looked at the step response. We have looked at the step response of dynamic systems to
essentially come up with performance parameters like time constant, rise time, settling time,
peak overshoot and so on. Now, we are going to analyze what happens when we provide a
sinusoidal input to the system. Frequency response deals with the response of the system
when a sinusoidal input is provided to it. By and large, we are going to look at the steady
state response.

Let us consider a stable LTI causal SISO dynamic system whose plant transfer function is P
(s). We provide an input u(t), whose Laplace transform is U ( s) and we get an output y(t),
whose Laplace transform is Y (s). We know that Y ( s )=P ( s ) U (s).

Let us consider u ( t ) =U 0 sin ωt. The input has a frequency ω and a magnitude of U 0 . The

U oω n( s)
laplace transform U ( s)= 2 2 . Let P ( s )= . This can be rewritten as
s +ω d (s)

n(s)
. We are assuming a stable system to begin with. So, all poles of P(s)
( s+ p1 )( s+ p2 ) … (s+ pn )
lie in the LHP (Left Half Plane).

Let us process this equation. Let us write s2 +ω2 as (s+ jω)(s− jω) and do the partial fraction
expansion.

347
 n(s) U o ω a a2 n (s)
Y ( s )=P ( s ) U ( s )= = 1 + + 1 .
d (s) (s +ω ) s+ jω s− jω d (s)
2 2

(Refer Slide Time: 13:06)

To find a1, we multiply both sides by (s+ jω). I get the following:

P(s)U o ω a (s+ jω) n1 (s)(s+ jω)

=a1 + 2 + .
s− jω (s− jω) d(s)

Now, we substitute the root of that factor, which is s=− jω . Then, we get

a1 =− ( U2 j ) P (− jω) .
o

P (− jω ) is the plant transfer function evaluated at s=− jω . That is going to be a complex

quantity. Any complex variable has a magnitude and a phase. There are various
representations of a complex variable. I can have a representation in terms of real and
complex part, I can have a magnitude and a phase representation for the same complex
variable.

So, I can rewrite the complex variable P( jω) as |P ( jω)|e jϕ, where ϕ is the phase of P( jω).
This is one representation of a complex variable. If I have P(− jω), I can write it as

|P ( jω)|e− jϕ.

−U o
This implies that a1 = |P( jω)|e− jϕ. Please remember that ϕ is the phase of P ( jω ) .
2j
Obviously, ϕ depends on ω. When you vary the frequency, obviously the phase also varies.

348
(Refer Slide Time: 19:02)

Similarly, let us figure out a2. If I want the expression for a 2, I multiply both sides by (s− jω)
. I get the following:

P(s)U o ω a1 (s− jω) n (s)(s− jω)

= +a2 + 1 .
s+ jω (s+ jω) d (s)

I substitute s= jω, which is the root of this particular factor. Then, we get

a2 = ( U2 j ) P ( jω)= U2 j |P( jω)|e .

o o jϕ

a1 a2 n (s)
Now, let us go back here.Y ( s ) is going to be + + 1 . If I take the inverse
s+ jω s− jω d (s)

Laplace transform, I will get a1 e− jωt for the first term and a2 e jωt for the second term. What is
going to happen to the next term? Let there be k distinct poles with μi being the multiplicity
of each pole. This is something which we have already done for an nth order system. We get

k μi−1
− jωt
y ( t ) =a1 e +a2 e + ∑ ∑ c ℑ t m eP t .
jωt i

i=1 m=0

The important thing is we have the poles as the exponents. This is the result we used to figure
out that if my poles lie in the left half complex plane, the system is going to be BIBO stable.

349
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 48
Frequency Response
Part – 2

(Refer Slide Time: 00:15)

As t tends to infinity, the exponential terms (third term in the equation shown in the figure
above) tend to zero since the system is stable, i.e., all poles of the plant transfer function are
on the left half complex plane.

Therefore, the steady state output is

−U 0 U
lim y ( t ) =a1 e− jωt +a2 e jωt = |P ( j ω )| e− jϕ e− jωt + 0 |P ( jω )|e jϕ e jωt
t →∞ 2j 2j

The above expression simplifies to

e j(ωt +ϕ)−e− j(ωt +ϕ )

lim y ( t ) =U 0|P ( jω )| =U 0|P ( jω )|sin ⁡(ωt +ϕ)
t →∞ 2j

350
An important feature is that the steady state output is a sinusoidal signal with the same
frequency (ω) as that of the input, but scaled in magnitude by |P(jω)| and shifted in phase by
phase of P(jω). This is a property of stable linear time invariant systems. Here, P(jω) is the
sinusoidal transfer function.

Frequency response is the response of the system to a sinusoidal input.

The sinusoidal transfer function P(jω) can be obtained from the transfer function P(s) by
substituting s = jω.

1
Example: P ( s ) =
s+1

1 1− jω 1 ω
P ( jω )= = 2
= 2
−j
j ω+1 1+ω 1+ω 1+ω2

1
|P ( jω )|= 2
,⦟ P ( jω ) =−tan−1 (ω)
√ 1+ω

351
(Refer Slide Time: 12:52)

For a general case, with n1 (jω) till nm (jω) in the numerator, and d1 (jω) till dn (jω) in the
denominator, the magnitude of P (jω) is the same expression for P (jω) but with magnitude of
each term taken, and the phase of P (jω) is the sum of phase of terms in the numerator minus
the sum of phase of terms in the denominator.

How can one visualize P (jω) as ω is varied?

There are broadly three types of visualization that are used for this purpose, which are as
follows:

1. Bode plot – it comprises of a magnitude plot, where |P (jω)| is plotted against ω, and a
phase plot, where phase of P (jω) is plotted against ω.
2. Nyquist plot – it is a plot of real part of P (jω) versus the imaginary part of P (jω) in
the complex plane (P (s) plane).
3. Nichols plot – HW.

(Refer Slide Time: 20:19)

352
353
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 49
Bode Plot 1
Part 1

Frequency response is used to denote the response of systems to sinusoidal inputs. An

important property of linear time invariant systems is that, if we give a sinusoidal input of
frequency ω, the steady state output is also going to be of the same frequency ω, but the
magnitude of the input is going to be scaled by ¿ P ( jω )∨¿, P ( s )being the plant transfer
function. The steady state output is going to be a sinusoid, but it is going to be shifted in
phase by ϕ, which is also the phase of P ( jω).

(Refer Slide Time: 00:44)

Depending on ω, the magnitude and the phase shift would be different. Thus, consequently,
the frequency response is characterized by a P ( jω )which is called as a sinusoidal transfer
function. The questions that arise are: What is a complex valued function of ω? How does
one visualize it?

There are multiple visualizations that are possible; the most commonly used one is what is
called as a Bode Plot and there is something called as a Nyquist plot and also a Nichols plot.

354
How to plot the Bode plot? Once given a Bode plot, how do we interpret system
characteristics from that plot? These are studied now.

(Refer Slide Time: 02:15)

We know that P ( jω )has two components: one is the magnitude of P ( jω )i.e ¿ P ( jω )∨¿, and
another is the phase of P ( jω ). Both the magnitude and phase are functions of ω. In the Bode
plot, there are two plots: the first one is what is called as a Magnitude plot. In the magnitude
plot, we plot ωon the abscissa and ¿ P ( jω )∨¿ on the ordinate. We use a logarithmic scale for
ωon the abscissa ok, and |P ( jω )|is plotted in decibel measure (dB).

How do we get the decibel measure? Given P ( jω ), we take|P ( jω )|and then we take the
logarithm to the base 10 and then multiply it by 20 to get the equivalent magnitude in dB.

Why do we do this? What do you think is the difference between a logarithmic scale and a
linear scale?

Let us say I have three points ω1 ,ω2 ,and ω3. The distance between ω1 and ω2 is the same as
the distance between ω2 and ω3. If I use a linear scale, you know like what can I say about the
distance between these two sets of points? They are equidistant. But what does it
characterize?

355
It means is that ω2 −ω1is going to be equal to ω3 −ω2. On the other hand, if I use a
logarithmic scale, what can I interpret?

ω2
If I use a logarithmic scale, the ratios are going to be the same i.e is going to be equal to
ω1

ω3
.
ω2

In the Bode plot, in the abscissa, what happens is it if I have 3 points which are equidistant
from one other? One point is 10 times the previous one. That is how we define a decade.

Why do we use a logarithmic scale? We can plot a wide range of ω. In a finite scale, I can
cover a vast range of ω. If I use a linear scale, it is going to be really difficult to plot such a
vast scale.

What is the second advantage of using logarithmic scale? We can also expand the low
frequency range. In many practical applications, you will see that the low frequency range is
also of equal importance as the high frequency.

(Refer Slide Time: 09:34)

What is a limitation of the logarithmic scale? We cannot plot 0. On a linear scale, I can mark
a 0. However, from a practical perspective, what does it mean by giving a 0 frequency signal?
sin ωt or cos ωt , when ω = 0, become either a 0 or a 1. That does not make sense because I

356
want to see sinusoidal inputs. Thus, it is not a very serious limitation as far as a logarithmic
scale is concerned. So, that is why, we use a logarithmic scale on the abscissa for the
frequency.

Why do we use a decibel (dB) scale for the magnitude? We take the logarithm to the base 10
and multiply it by 20 and convert it to dB. Why should I use a logarithmic scale for the
magnitude? Similar reasons would apply. I can cover a larger range of magnitudes in a finite
scale. I can focus on lower magnitude values also as that region is also going to be expanded.

P1 ( jω) P 2 ( jω)
Let P ( jω )= . Let P1 and P2 be some factors. If you take the logarithm, we get
P3 ( jω)

log10|P ( jω )|=log 10|P 1 ( jω )|+log 10|P2 ( jω )|−log 10|P 3 ( jω )| .

We observe that if you have a product/ratio of factors, they become the algebraic sum when
logarithm is taken. This is a pretty good advantage which we would exploit. Given a transfer
function, I break it into smaller terms. I plot the magnitude plot of the individual terms. Then
I just add or subtract like as the case may be. So, that way the Bode plots can be plotted
easily.

Similarly, you can see that the phase of P ( jω ) is also going to follow a similar trend.

If you can break down a given transfer function into simpler building blocks, we just plot the
bode diagrams of the individual blocks and then take the algebraic That is a very good
advantage of using a logarithmic scale.

The second plot is what is called as a Phase Plot. Once again ω is plotted on the abscissa
using a logarithmic scale and the phase of P ( jω ) is plotted in degrees using a linear scale. We
just use a linear scale as there is no need for any logarithmic scale. This is because phase of
the complex valued function already satisfies this property.

Thus, the Bode Plot has two plots: the Magnitude Plot and the Phase Plot. The Magnitude
Plot essentially, plots the magnitude of the sinusoidal transfer function in dB vs. ω on the x-
axis using a logarithmic scale. The Phase Plot plots the phase of that sinusoidal transfer
function in degrees vs. ω, once again on a logarithmic scale on the abscissa. One advantage
of doing this way is that you can really use certain building blocks. If we learn how to plot

357
the bode diagram for the building blocks, it is easy to plot it for any particular transfer
function.

(Refer Slide Time: 15:57)

What do you think are the potential building blocks? These are the basic factors which would
which can make up a transfer function.

1
A transfer function can have some constant k. It can have an integral term . It can have a
s

1
derivative term s. It can have a first order term of the form . It can also have Ts+1. We
Ts+1

ωn 2
can also have a complex conjugate factor of the form . Further, we can have
s 2 +2 ξ ωn s +ωn 2

s 2 +2 ξ ωn s +ωn 2
a second order factor in the numerator, as in 2 . We can immediately observe
ωn
that, I can essentially rewrite any transfer function in this particular form. So, let us
understand how to plot the bode diagram for these blocks.

Let us consider the symbol G (s) to say that it is any generic complex valued transfer function
. The first factor is G ( s ) = K, where K is a real number. K >0 so that it is possible to take
logarithm.

358
What happens if I want to plot the bode diagram if G ( s ) = K? G ( jω ) is going to be just K.

|G ( jω )|is going to be 20 log10 K , which is just K in dB.

What can you say about the phase of G ( jω )? It is going to be 0 o if K >0 and 180o if K <0. So,
it really depends on the sign of K ok.

So, what is going to happen to the magnitude plot? The magnitude plot will just have a
straight line with respect to ω, where this value is going to be 20 log10 K .

So, in the phase plot, phase is going to be 0 o if K >0 and 180o if K <0. However, typically we
take a negative value i.e., we take it as −180o .I will explain why we take the negative value
here later on. So, that is going to be the Bode Plot for the constant term.

(Refer Slide Time: 24:03)

Typically, on the abscissa we plot/mark frequencies which are a decade apart. What is a
decade of frequency? Let us define what a decade is at a given frequency ω. The range ω to
10 ω is one decade above ω. Similarly, the range 0.1 ω to ω is called as one decade below ω.
If you consider 1 as the base frequency, the frequency 10 is one decade above 1. The region
from 0.1 to 1 is one decade below 1.

359
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 50
Bode Plot 1
Part - 2

(Refer Slide Time: 00:15)

Q. What is the advantage of using the logarithmic scale?

Ans. We’ll figure out when we plot the Bode diagram.

Octave is another band which is used. Let us say a frequency omega is given. The region
omega to 2 omega is what is called as one octave above omega. Similarly, from 0.5 omega to
omega is one octave below omega.

Slopes of the curves are typically expressed in terms of decades or octaves.

(Refer Slide Time: 02:43)

360
 1
Example: Plot the Bode diagram for G (s) = .
s

1 1
G (s)= ⇒G ( jω)=
s jω

1
⇒∨G ( jω)∨¿ ,20 log10∨G ( jω)∨dB=−20 log10 ω d B
ω

∠ G( jω)=0−90=−90 °

Hence, the phase is constant and independent of ω, whereas the magnitude is a function of
omega.

To plot the magnitude and phase plot, first setup the ωaxis in the logarithmic scale (..., 0.1, 1,
10, …).

The phase diagram is going to be a horizontal line with a value of −90 °.

For the magnitude diagram, substitute ω= 0.1, 1, 10 rad/s to get the value of the magnitude of
G ( jω). Then plot. At ω= 0.1, the value is 20 dB; at ω= 1, the value is 0 dB; and at ω= 10,
the value is -20 dB. The resultant plot will be line with a negative slope.

To calculate the slope, one can compute it in terms of decades or in terms of octaves. If it is
in terms of decade, the slope is -20 dB/decade. This is because as we increase ω by one
decade, for example, from ωto 10 ω, the value decreases by -20 dB. If it is in terms of octave,

361
the slope is approximately -6 dB/octave. This is because as we increase ω by one octave, for
example, from ωto 2 ω, the value decreases by -20 log10 2 dB, which is approximately -6 dB.

(Refer Slide Time: 10:04)

Example: Plot the Bode diagram for G (s) = s.

G (s)= s ⇒ G ( jω)= j ω

⇒∨G ( jω)∨¿ ω ,20 log 10∨G( jω)∨dB=20 log 10 ω d B

∠ G( jω)=90 °

The phase diagram is going to be a horizontal line with a value of +90 °.

For the magnitude diagram, substitute ω= 0.1, 1, 10 rad/s to get the value of the magnitude of
G ( jω). Then plot. At ω= 0.1, the value is -20 dB; at ω= 1, the value is 0 dB; and at ω= 10,
the value is 20 dB. The resultant plot will be line with a positive slope with a value of +20
dB/decade or 6 dB/octave.

Note that 1/s and s are reciprocals of each other. The magnitude plot and phase plot for these
two are mirror images. But about what axis is it mirrored?

Note that if you have a factor and its reciprocal in a transfer function, their effects will cancel
one another. Similarly, the sum (since in logarithmic scale) of the magnitude and phase plot

362
of the factors will be zero at any ω. Therefore, the two plots will be a mirror image about a
horizontal line passing through zero.

(Refer Slide Time: 18:11)

To recap,

Q: What do you observe about factors that are reciprocals of each other?

s+2
Example: G (s)= . How to plot the Bode plot of this transfer function?
s ( s¿¿ 2+4 s+3)¿

The transfer function can be rewritten as

s+2
G (s)=
s
s ( s¿¿ 2+4 s+3)=
s+2
=
( ) 2 +1
2 2 s 1
= ( +1)( )(
1
)(
1
)¿
.
s ( s +1 ) ( s +3 ) s 3 2 s s+1 s
( )
3 s ( s+1 ) +1
3 3
+1

We’ll see how to put together the Bode plots of the individual factors to plot the Bode plot of
the transfer function.

363
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 51
Bode Plot 2
Part - 1

(Refer Slide Time: 00:17)

We are looking at the Bode Plot or the Bode diagram. We saw that the Bode Plot has two
plots; a magnitude plot and a phase plot. One of the big advantages of plotting the magnitude
in decibels is that if we have the transfer function as a product or a ratios of individual blocks,
we could plot the magnitude plot of the individual blocks and then just add them. And same
story with the phase plot. And the reason we why we want to choose the frequency in a
logarithmic scale is to essentially cover a wider range.

The basic factors that we looked at were a constant K , the derivative term s and the integral

1
term . Let us proceed to the next factor.
s

364
(Refer Slide Time: 01:43)

1
G ( s )=
Ts+1

We will consider that T is are a positive real number.

1 1−Tjω
G ( jω )= =
1+Tjω 1+T 2 ω2

1
|G ( jω )|= 2 2
,∠ G ( jω )=−tan−1 (Tω¿) ¿
√1+T ω
1
At low frequencies, ω ≪ , |G ( jω )|→ 1(0 dB)
T

The magnitude is going to tend to a line which corresponds to the 0 decibel line in the
magnitude plot as frequencies become smaller. For this reason, this particular value or locus
is going to be denoted as the low frequency asymptote.

1 1
At high frequencies, ω ≫ , |G ( jω )|→ (−20 log 10 (Tω) dB)
T Tω

This is the value to which the magnitude will tend to at high frequencies, That is why it is
called as a high frequency asymptote.

365
(Refer Slide Time: 08:29)

The high frequency asymptote is going to be a straight line in the Bode Plot. Because T is a
scalar and we are going to plot ω in a logarithmic scale, so obviously going to be a straight
line. The low frequency asymptote, the 0 decibel line is a horizontal straight line. Both of
them are going to intersect wherever the high frequency asymptote value will become 0

1
decibel that is when ω= . This is what is called as a Corner Frequency or Break Frequency.
T

Let is find the slope of the high frequency asymptote. Slope is going to be expressed in
Decibels per Decade or Decibels per Octave. Typically, we express in decibels per decade.
We choose a particular ω and go to 10 ω omega and then find out what is the difference in
magnitude. The difference is the value of the slope in Decibels per Decade.

Let us say, we go from T ω to T (10 ω). The slope of the high frequency asymptote is going to
be −20 decibels per decade. So, consequently the slope of the high frequency asymptote is
minus 20 decibels per decade. The slope of the low frequency asymptote is 0 because it is a
flat line.

(Refer Slide Time: 14:08)

366
The magnitude plots are as shown in the figure. Low frequency asymptote is the pink dotted

1
line and high frequency asymptote is the orange dotted line. Both of them intersected atω=
T
. The purple solid line gives the approximate bode plot. Please note that the high frequency
asymptote makes sense only when ω → ∞.

Let us create a table

ω |G ( jω )| (dB) Asymptotic value (dB)

1 -3.01 0
T

2 -6.99 -6.02
T

0.5 -0.97 0
T

10 -20.04 -20
T

0.1 -0.04 0
T

We can see the level to which the asymptotes and the actual curve will differ from each other.
We get the maximum error at the corner frequency and then the errors are going to just
decrease.

367
(Refer Slide Time: 26:15)

1
If we plot the error versus frequency. We are going to have the maximum error at ω= and
T
beyond that, on either side a decreasing error profile, which will asymptotically tend to 0.

Drawing the asymptotes is only an approximation, but a reasonable approximation. We are

going to get a maximum error of 3 decibels.

368
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 52
Bode Plot 2
Part – 2

(Refer Slide Time: 00:20)

Phase of the factor is ⦟ G ( j ω ) =−tan−1 (T ω).

1
As ω tends to zero, ⦟ G ( jω ) tends to zero. As ω tends to , ⦟ G ( jω ) tends to - 45˚. As ω
T
tends to infinity, ⦟ G ( jω ) tends to - 90˚. These values can be used to plot the phase plot.

(Refer Slide Time: 03:53)

369
In the magnitude plot, if the actual plot is plotted, there’ll be an error is -3 dB at the corner
frequency. The magnitude plot has the characteristics of a low pass filter. It can be observed
by passing sinusoids of different frequencies. If the frequency is less than the corner
frequency, the magnitude of the input signal is not attenuated. Whereas, if the frequency is
greater than the corner frequency, the magnitude of the signal is attenuated; higher the
frequency, higher is the attenuation. High frequency roll-off is a term used to indicate at what
rate the magnitude curve is dropping at high frequency.

Bandwidth for this transfer function is the frequency range between zero and the cutoff
frequency.

HW. Plot the phase plot using a linear scale and logarithmic scale for ω.

(Refer Slide Time: 09:50)

370
Next, we’ll plot the magnitude and phase plot for the factor G ( s ) =Ts+1.

G ( jω )=T jω+1

|G ( jω )|= √1+T 2 ω2=20 log 10 √1+T 2 ω2 dB

⦟ G ( jω ) =tan−1 (Tω)

For the magnitude plot, the low frequency asymptote is 0 dB and the high frequency

1
asymptote is 20 log10 ( T ω ) dB. They intersect at ω= (corner frequency). The slope of the
T
high frequency asymptote is 20 dB/decade.

(Refer Slide Time: 12:18)

371
HW. Calculate the error in magnitude (as before).

At the corner frequency, there’ll be an error of 3 dB, and one octave above and below there’ll
be an error of 0.97 dB, and one decade above and below, there’ll be an error of 0.04 dB.

It can be observed that the magnitude plot of Ts+1 is just a reflection about the horizontal line
passing through the 0 dB point. It is just a mirror image right.

(Refer Slide Time: 18:23)

372
The phase plot is also a reflection about the 0˚ line. As ω tends to zero, ⦟ G ( jω ) tends to zero.

1
As ω tends to , ⦟ G ( jω ) tends to 45˚. As ω tends to infinity, ⦟ G ( jω ) tends to 90˚. These
T
values can be used to plot the phase plot.

The objective is to write the transfer function as a product of individual building blocks (such
as those covered in the lectures) and use the Bode plots of the individual factors to obtain the
Bode plots of the transfer function.

373
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 53
Bode Plot 3
Part - 1

We are looking at bode diagrams and we looked at the bode plots of a few individual factors.
The last set of factors that we are going to look at are second order factors. We write the
second order factor as:

ωn 2 1
G ( s )= 2 2 = 2
s +2 ζ ωn s +ωn
( ωs )n
+
2ζ
ωn
s+1

1
G ( jω )=
ω 2 2 ζω
1−
( )
ωn
+j
ωn
1
¿ G ( jω )∨¿
2 2 2

√( 1−
ω
( )) ( )
ωn
2 ζω
+
2 ζω
ωn

∠ G ( jω ) =−tan−1
( ( ))
1−
ω
ωn
ωn
2

374
(Refer Slide Time: 00:48)

Let us look at the magnitude of this factor and then try to figure out what are the asymptotes,
The magnitude of this particular factor in decibels is going to be:

2 2 2
|G ( jω )|=−20 log10
( √( 1−
( )) ( ))
ω
ωn
+
2 ζω
ωn

When ω ≪ ωn, |G ( jω )|→ 0 dB (low frequency asymptote).

ω
When ω ≫ ωn, |G ( jω )|→− 40 log 10 dB (high frequency asymptote).
ωn

375
(Refer Slide Time: 08:26)

Where do the two asymptotes intersect? We can immediately figure out the answer as ω=ωn .
It is because when we substitute ω=ωn in the high frequency asymptote, we get 0 dB. So,
consequently ω=ωn is the corner frequency. The natural frequency is the corner frequency.

What is the slope of the high frequency asymptote? The slope of the high frequency
asymptote is going to be -40 dB/decade.

At what frequency is the magnitude of |G ( jω )| maximum?

We differentiate and equate to zero and then take the second derivative, the second
derivative is negative for a maximum. But here we can use a simpler method because this in
the numerator is 1. We will have a maximum where the denominator is going to be a
minimum.

376
(Refer Slide Time: 11:57)

So, at whatever frequency the term within the square root in the denominator is a minimum,
that is the frequency at which the|G ( jω )| is going to be maximum. Let

2 2 2
ω 2 ζω
( ( )) ( )
f ( ω ) = 1−
ωn
+
ωn

2
d f ( ω) ω
dω ( ( ( )) )
= − 1−
ωn
+2 ζ 2 ω=0

This has two solutions.

ω 2
2
ω=0 and 2 ζ =1− ( ) ωn
,

ω=ωr =ωn √ 1−2 ζ 2

1
This exists only when ζ < .
√2

377
(Refer Slide Time: 17:13)

This frequency ωr is called as the resonant frequency. We have the resonant frequency giving
the maximum value of the amplitude or magnitude of the sinusoidal transfer function for the
second order factor.

What is natural frequency?

If we take a second order system and remove the damping, then give a perturbation, the
frequency at which the undamped system oscillates is what is called as the natural frequency.
Undamped second order system, the plant transfer function is going to be,

ωn 2
P ( s )=
s2 +ωn 2

ω n 2 ωn
Let u ( t ) =sin ( ωn t ) , y ( s )=¿ 2 ¿
( s 2+ωn2 )
¿ y ( t ) ∨→ ∞ as t → ∞.

Damped natural frequency was the frequency at which the damped second order system
oscillatedwhen the damping ratio was between 0 and 1 in response to an step input.

ωd =ωn √ 1−ζ 2

378
Even if you give a sinusoidal input equal to the damped natural frequency, the system output
would be bounded because you have an under damped stable second order system.

The third frequency is the resonant frequency ωr .

ωr =ωn √ 1−2 ζ 2

This means that I have a stable under damped second order system with damping ratio

1
between 0 and . If we have such a system, ωr the frequency at which the amplitude of the
√2
corresponding sinusoidal transfer function will be maximum. In other words, if you give a
sinusoidal input equal to whose frequency is ωr , you would get the maximum amplification
of the input. The output amplitude which is a sine wave that will also be a maximum there; it
is still bounded mathematically, but physically it may create issues. The system is still stable
no doubt about it, but then resonance can lead to oscillations of reasonably high magnitude
which can cause structural damage. One famous example which is typically given to illustrate
this is Tacoma Narrows Bridge Collapse.

As homework, take the second derivative and ensure that f ' ' ( ω) is going to be positive at
this frequency.

(Refer Slide Time: 27:08)

379
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 54
Bode Plot 3
Part - 2

(Refer Slide Time: 00:15)

So, now let me go and draw the bode plot, with all this background. Let us go and now
construct bode diagram for this particular second order transfer function, ok. Sohow does it
look? So, let us draw the bode plot.

(Refer Slide Time: 00:30)

380
Let me first draw the magnitude plot. So, of course we are going to consider omega on a
logarithmic scale and on the y axis, we have the magnitude of G(jω) in decibels, ok.

So, let us say this is the 0 decibel line. Let us say this is the 0 decibel line, let say this is
minus 20, let say this is minus 40, ok. So, that is what we have and let us say this is ω n, let us
say this is 10ωn, let us say this is 0.1ω n because I am considering the natural frequency or the
corner frequency, ok.

So, let me use the same color code as yesterday for the asymptotes. Let me just scroll up and
see what colors are used. So, for the low frequency asymptote, I use pink and yellow for the
high frequency asymptote, great. So, the low frequency asymptote here is going to be the 0
decibel line. So, this is the, pink line is the low frequency asymptote. So, the high frequency
asymptote is going to have a slope of minus 40 decibels per decade, right.

So, at ω=ωn, the high frequency asymptote will be 0; at ω=10ω n what should be its value? -
40, right because in one decade it decreases by -40, right. So, as a result the high frequency
asymptote is going to be something like this, ok.

So, I am just extending it to show it graphically ok, but this is the high frequency asymptote.
The slope is -40 decibels per decade, ok. Now, the point is the actual curve depends on the

value of zeta, ok. So, please note that the maximum amplitude occurs at ω=ωn √ 1−2 ζ 2, ok.
That is the resonant frequency. So, as zeta increases from 0, what do you think will happen to
this resonant frequency? Will it go towards the left of ωn. Yes because as zeta increases, you
see that √ 1−2 ζ 2 decreases.

So, as a result as zeta increases from 0, the further the increase in zeta, the further is the shift
of the resonant frequency to the left from the natural frequency. So, I am just going to draw a
family of curves which will essentially tell us how the natural frequency curve looks like, ok.
So, this is the actual curve. So, let us say at low damping ratios I have this; ultimately it will
go like this. As the damping ratio increases, the peak essentially shifts further and further to
the left, but anyway after some time everything goes to the high frequency asymptote, right.

So, I am just drawing a freehand sketch. So, please pardon the approximate diagram, ok. So,
what happens is that this is how the peak changes, right as zeta increases. Of course, after

381
zeta greater than 1/ √ 2, you see that the maximum value never crosses zero decibels, ok. That
means, that there is no peak as such for zeta greater than 1/ √ 2.

So, that is why the maximum value is 0 decibels, right. So, that is it. The low frequency
asymptote is the highest value, ok. So, that is what happens. Of course, what happens to this
magnitude plot for an undamped system, what do you think will happen? You will see that
once again low frequency asymptote will be 0 decibels, high frequency asymptote will have a
slope of minus 40 decibels per decade, but what will happen at ω=ω n? You will see that the
amplitude will blow up to infinity, right.

So, essentially that is what we have, right. So, essentially it is going to be a very high
amplitude. So, what do I mean by that. Let us say we consider the undamped system

ω2n
G ( s )=
s2 +ω2n

1
G ( jω )=
Substituting s=jω, ω2
1+ 2
ωn

1
|G ( jω )|=
2 2

√( ω
1− 2
ωn )
So you can immediately see that as ω tends to ω n, the amplitude just blows off to infinity,
right. So, that is what you have when you have what to say for an undamped system, ok. So,
this is just an aside, right ok. Now, this is the magnitude plot, ok. So, this is the magnitude
plot of the second order term that we have been considering.

Now, what about the phase?

ω
2ζ
Phase of G ( jω ) =−tan−1
( )
ωn
ω2
1− 2
ωn

382
So, one can immediately figure out that this phase goes to 0 degrees as omega tends to 0. This
goes to minus 90 degrees as ω=ωn, ok. This essentially goes to minus 180 degrees as omega
tends to infinity, ok.

So, this is what happens to this to the phase plot of this particular factor. So, if you want to
plot the phase plot; what will happen is the following?

(Refer Slide Time: 08:25)

So, let me plot the phase plot of this particular transfer function. So, this we are going to plot
in degrees. So, let us say this is 0, minus 90, then minus 180, then we have ω n, 10ωn and
0.1ωn, ok.

So, what is going to happen is the following. So, 0 to minus 180 are the two bounds and at
ω=ωn, you are going to have minus 90, right. So, what will happen to the phase plot is
following. So, as zeta starts to increase from 0, we are going to have the phase plot go
something like this and as zeta keeps on increasing, it becomes like this, ok. We are going to
get a family of curves, right, depending on the value of zeta. So this is for a lower zeta, ok.
So, this is what happens as zeta increases, ok. So, that is just a family of curves, you know
just to show you how the phase changes for this particular factor, as zeta increases from 0, ok.
So, that is the phase of this particular second order transfer function and if you consider the
reciprocal of that, what will happen? You will just get the mirror image, right about the 0
decibel line, but I leave that to as homework, ok. So, the seventh factor is essentially this

383
which is essentially the reciprocal of the factor that we just considered. So, what I want you
to do is that do it as homework, ok.

So, it just will be a mirror image of this one, right. Any questions here; fine. So, I am just
going to leave you with one more question to think about right and then, like we will continue
and discuss it in the next class, right and then, we will do an example, ok. The next class we
are going to do an example of the construction of the bode diagram, but I am just going to
leave you with a question to essentially think about. So, we are done with this, ok.

(Refer Slide Time: 11:41)

Next discussion is on minimum phase and non-minimum phase systems, ok. So, I told you
that when we come to frequency response, I will revisit minimum phase and non-minimum
phase systems, right. So, do you remember what a non-minimum phase system is? Of course,
we are assuming stable systems. What was when did you use the adjective non minimum
phase, do you remember?

Student: [Inaudible].

When there was zero on the right half plane, right that was a non-minimum phase, right.
When all the zeros are in the left half plane, it is a minimum phase, right. So, I told you that I
will revisit and then, like tell you the reason behind this terminology from the perspective of
frequency response, right. I am just going to leave you with a question from which you can
figure out the answer and then, we will anyway discuss answer tomorrow, right.

384
So, let us say consider two systems, of course two stable systems right to begin with, all right.
So, two stable systems whose transfer functions are ok.

(Refer Slide Time: 13:02)

1+Ts
G 1 ( s )= ,T >0 , T 1 >0∧¿
1+T 1 s

1−Ts
G 2 ( s )= ,T >0 , T 1 >0
1+T 1 s

So, immediately you see that both transfer functions have a pole -1/T 1 which is in the left of
plane, right. No issues regarding the stability of either systems, right. So, what is the
magnitude of the sinusoidal transfer function, but before we calculate the magnitude let us
write,

1+ jTω 1− jTω
G 1 ( jω ) = ∧G 2 ( jω )=
1+ jT 1 ω 1+ jT 1 ω

So, what are the magnitudes? It is going to be

1+T 2 ω2 √ 1+T 2 ω2
|G1 ( jω)|= √ ∧|G2 ( jω)|=
√ 1+T 21 ω2 √ 1+T 21 ω2
So, what can you say about the two magnitudes? Same, right. They are equal in magnitude.

385
So, you give me any omega, you can see that the magnitude of G 1 ( jω ) and magnitude of
G 2 ( jω ) are the same. Now, let us look at phase of G 1 ( jω ), ok. What is the phase of G 1 ( jω )? It

is going to be tan−1 ( Tω ) −tan−1 (T 1 ω), right. So, then what is the phase of G 2 ( jω )? It is going

to be −tan−1 ( Tω ) −tan−1 (T 1 ω), right.

So, what do you think happens to the phase of G 1 ( jω )as omega tends to infinity? Tends to 0,
right. What can you say about the phase of G 2 ( jω ) as omega tends to infinity, it tends to -180
which has the minimum phase G1, right and see G 2 ( jω )phase tends to minus 180, G1 tends to
0, right. So, I hope it is clear.

Now, why the system with the transfer function G 1(s) is called minimum phase and why the
system is the transfer function G2(s) is called non-minimum phase? So, in essence what we
are saying is that if you give me two systems whose transfer functions have the same
magnitude characteristics, the minimum phase system will have the lower phase. Of course,
the thing is that like we are only looking at the magnitude per se, right with this, with the
sign, right.

So, essentially it will have the minimal phase of the two, ok. That is why it is called as a
minimum phase system. The question I am going to leave you with is that how can we
generalize it for any nth order system? Let us say what was our general transfer function, you
remember G(s) was n(s)/d(s), right. N(s) was a polynomial of order m, d(s) was a polynomial
of order n, right.

How can you generalize it for any nth order system? So, that is a question I leave you with or
if I give you bode plot of a transfer function by looking at it, can you get the relative degree
of the system. What I mean by the relative degree is n - m. n - m is the relative degree of the
transfer function and also, figure out whether its minimum phase or non-minimum phase, ok.
Can we do that?

So, those are questions I am going to leave you with, I am going to come and discuss the
answers tomorrow, but please think about them, ok. We will discuss those answers tomorrow
and then, we will also do an example problem, right and tomorrow's class for the bode
diagram which would complete our discussion on bode plots fine, ok. So, I will stop here. I
will see you in the next class.

386
Thank you.

387
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 55
Bode Plot 4
Part - 1

In Bode diagrams, we saw the difference between a minimum phase and a non-minimum
phase system, as far as asymptotic value of the magnitude plot and the phase plot were
concerned. In general, a transfer function G (s), which is a ratio of N (s ) and D (s ) with the
numerator polynomial of order m and the denominator polynomial of order n.

If we have a minimum phase system, which means all the zeros are in the left half of complex
plane. We can write down some general statements.

(Refer Slide Time: 01:25)

If we look at the transfer function, we will see that the slope of the magnitude plot of G ( jω)
would tend to −20(n−m) decibels per decade as ω → ∞. This is going to be true for both
minimum phase and non-minimum phase systems.

388
(Refer Slide Time: 02:31)

We can check from the example in previous class example. We saw that the magnitudes were
essentially the same. If you look at the transfer function,

1+Ts 1
G 1 ( s )= =(1+Ts)
1+T 1 s 1+T 1 s

We already know from the building blocks that asω → ∞, the slope of (1+Ts) is going to be
20 decibels per decade and the slope of high frequency asymptotes is going to be -20 decibels
per decade for (1+T 1 s). As ω → ∞ omega tends to infinity the net slope this is the sum,
which is 0. In this case, n=1 ,m=1andn−m=0. So, we would have a slope of 0 decibels per
decade at high frequency. The magnitude curve will flatten out. This is true for both
minimum phase and non minimum phase systems. But ∠ G( jω) → 90 ° (n−m) only for
minimum phase systems. In the previous example, n−m=0. So, the phase went to 0 ° as
ω → ∞.

If you conduct experiments and plot the magnitude and the phase plots, you can use this
information to figure out the relative degree of the transfer function. From the slope of the
high frequency curve you can get n−m, which gives the relative degree of the transfer
function. From the high frequency value for the phase, you will be able to determine whether
it is a minimum phase system or a non-minimum phase system.

389
(Refer Slide Time: 06:43)

Let us do one example today where we construct the bode plot of a transfer function from
start. This will illustrate the entire process as to how to construct the magnitude plot and the
phase plot. Let us consider a transfer function which is of the form

s
G ( s )=
(s+1)( s+10)

We want to plot the bode diagram for this particular transfer function. We need to rewrite this
transfer function in terms of the factors.

s 0.1 s
G ( s )= =
( s +1 ) (0.1 s+1) ( )( 0.11s+1 )
=(0.1)(s )
1
s+1
( 10s +1)
10 (s+1)

We have to plot the individual bode plots of the four factors and then get the net bode
diagram.

Consider 0.1. The magnitude is given by 20 log10 0.1=−20 decibels. The phase is 0 °, it is on
the positive real axis.

The second factor is s, its magnitude in decibels is 20 log10 ω. The phase of the s term is 90 °.

390
 1 1 1
The third term is
s +1
. This is of term of the form
Ts+1
. HereT =1. So corner frequency( )
T
is 1 rad/s.

1
The fourth term is . HereT =0. 1. The corner frequency is 10 rad/s.
0.1 s+1

(Refer Slide Time: 11:21)

Let us start from the magnitude plot. 1 and 10 are the corner frequencies. Typically, we
consider one decade above and one decade below the corner frequencies. We can go for a
larger range also. On the ordinate we are going to plot the magnitude in decibels.

Let us plot each term one by one. Let us first take the constant term, 0.1. The magnitude of
0.1 is -20 decibels. That will remain the case irrespective of the frequency. The magnitude
plot of 0.1 is going to be a just a horizontal line at -20.

The next factor is s, that has magnitude plot of a straight line with a slope of 20 decibels per
decade. At ω=1, the magnitude is 0. At ω=1 0, the magnitude is 20. At ω=0. 1, the
magnitude is -20. The magnitude plot is a straight line a slope of 20 decibels per decade
passing through these points.

391
 1
The third factor is . The low frequency asymptote in the magnitude plot is 0 decibels. We
s +1
are going to have this till the corner frequency of 1. The high frequency asymptote is going to
be a straight line with a slope of -20 decibels per decade.

1
The fourth factor is . The corner frequency is going to be 10. The low frequency
0.1 s+1
asymptote is going to be the 0 decibel line till the corner frequency of 10. The high frequency
asymptote is a straight line with a slope of - 20 decibels per decade.

Once we have plotted all the four graphs, we need to add them. The s term is going to affect
throughout all frequencies. The minus 20 decibels per decade term also will affect throughout

1 1
all frequencies. And the term and term the effect will come from the
s +1 0.1 s+1
corresponding corner frequencies.

The final bode plot is shown in sold red colour. The plot for this s term but that is going to be
shifted down by - 20 decibels. Because of the effect of 0.1 term till the corner frequency of 1.

1
After the corner frequency of 1, the effect of s term and term cancel each other and the
s +1
plot remains horizontal till corner frequency of 10. After the corner frequency of 10, due to

1
the effect of term, the bode plot has a slope of -20 decibels per decade.
0.1 s+1

You can use the command bode in MATLAB to get the Bode plot.

392
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 56
Bode Plot 4
Part - 2

So now, having plotted the magnitude plot; once again please note that I am only plotting the
asymptotes right. We are not plotting the actual curve as such right. So, now let us go ahead
and plot the phase plot.

(Refer Slide Time: 00:38)

So, let us do the phase plot. So, the phase of G(jω) I am plotting in degrees. So, let us say I
start with 0, and then let us say this is minus 90, let us say this is plus 90, fine.

So, once again we have 0.1, 1, 10, 100 as the four regions. Now, let us just plot the phase
plot. So, what is going to happen let us do it for each one factor at a time right that is what we
are going to do. So, let me use the same colour code.

0.1 s
So, what happens to the, see ultimately what are we plotting for? G (s)= , so
( s+1 ) (0.1 s +1)
that is what we are doing ok. So, now what is going to be the phase contribution of 0.1, the
phase contribution of 0.1 is going to be 0 degrees right. So, I draw a horizontal line at 0
degrees. So, this is going to be the contribution from 0.1.

393
What is going to be the contribution from s? plus 90 right. So that is going to be the same
throughout the frequency range. So, this is the phase contribution of s. What about the phase
contribution of 1/(s+1)? So, if you recall what happens with 1/(s+1) is that the low frequency
value is 0, it has a phase of minus 45 degrees at the corner frequency which is 1 and then it
tends to 90 degree minus 90 degrees as omega tends to infinity right that is the phase plot of
1/(s+1) right. So, that is what I am going to plot. So, at the corner frequency of 1, we are
going to have a phase of minus 45.

So, what is going to happen is the following. So, 1/(s+1) will have a phase plot like this ok.
This is going to be the phase plot of 1/(s+1) in green. Now, what is going to happen with
1/(0.1s+1)? At the corner frequency of 10, we are going to have a phase of minus 45. So,
what is going to happen is the following. So, we are going to have a phase diagram of
1/(0.1s+1) coming like this ok. So, this is the phase diagram of 1/(0.1s+1).

So, now what we need to do is that if I want a phase diagram of the complete transfer
function, I just need to add the four curves right. But this is a little bit complex because these
are not straight lines right. So, it is ok, but the thing is that like at low frequencies you will
start from plus 90 degrees that is obvious right because at very low frequencies the only
phase contribution will come from s, because 0.1 anyway will give you phase contribution of
0 degrees throughout, 1/(s+1) and 1/(0.1s+1) will give you a phase contribution of close to 0
degrees at low frequencies.

So, at low frequencies, anyway I will start from the value of 90 degrees. So, if you in a sense
look at the phase of G(jω), the essentially the formula for this is going to be
90 °−tan−1 (ω)−tan−1 (0.1ω).

So, I think this is something which we have discussed before right you have a transfer
function as essentially products of the you just add the individual take the algebraic sum of
the individual phase contributions right so that is what we have.

So, what we do is that in order to plot the net phase plot, we just calculate the phase at the
corner frequency, so that we can mark them and draw a smooth curve through those
frequencies right. So, the two corner frequencies are at ω=1 and ω=10. So, at ω=1, what is
going to happen to this value, we are going to have 90 °−tan−1 (1)−tan−1 (0.1) all right so
that is what we are going to have.

394
At ω=1,

Phase G ( jω )=90 ° −tan−1 (1)−tan−1 (0.1)=39.29 °

At ω=10,

Phase G ( jω )=90 ° −tan−1 (10)−tan−1 (1)=−39.29°

So, I am going to get essentially the negative of the same number, I am going to get minus
39.29 degrees that is what I am going to get alright. So, at 10, I am going to have something
like this. So, this is going to be minus 39.29.

So, what I need to draw is that I need to draw a phase curve which essentially starts from 90,
and then like passes through these two points. And then ultimately goes to minus 90 as
omega tends to infinity ok, so that is what will happen.

As ω → ∞, Phase G ( jω ) →−90 °

As ω → 0, Phase G ( jω ) → 9 0 °

So, you see that the phase curve goes from + 90 to -90 in this manner ok. So, this is the phase
curve of this net transfer function.

And you immediately see that because this has a zero at 0, it does not have a zero on the right
half plane, it essentially follows the points that we discussed at the beginning of this class
right. What did we discuss, you give me a transfer function right, and I plot the bode diagram,
you immediately see that the magnitude plot is going to tend to a slope of -20(n-m) decibels
per decade. What was the n - m in this problem? What is n? 2. m? 1 in this particular problem
right, because this is a transfer function we started off with is not it ok. So, you see that n - m
is 1 in this case.

395
(Refer Slide Time: 11:21)

So, what did the magnitude plot slope tend to at high frequencies -20 decibels per decade one
could see that right, so that was the general concept that we learned. And since this is a
minimum phase system, what did the phase tend to minus 90(n – m). You see that it tends to -
90 degrees.

So, this is the procedure for calculating the bode plot right. So, given the particular transfer
function, we should be also familiar with the inverse process. Let us say I give you this bode
plot. Can you figure out the transfer function that plays a very important role like as far as
experimental determination of transfer function is concerned. And towards the end of the
course, we will do a case study in that regard ok. So, I will give you, the reason I am letting
you do all the experiments I want to see you know like how much you gain out of those
experiments.

So, I want you to independently think and then like what I will do is that I will give you
actual data from one of those experiments and then ask you to get the transfer function ok.
Some techniques you may be already reading what to say learning in some of the experiments
so, but then like let us see towards the end like. So, how the theory and practice come
together. See right now we are looking at many mathematical what to say formulations and
ideas and so on right.

So, using frequency response, towards the end you will see how everything comes together
for practical applications so that is something which we look at in that maybe the last two

396
weeks of the course right. So, we will do a few case studies. So, that is that is something
which we will keep for the end ok.

So, this is the bode plot and it becomes very important because what happens is that like
many times if you want to experimentally determine the transfer function of the system right,
let us say what we have been doing in this course, we have been using physics to derive,
mathematical equations in the form of linear ODEs, then take Laplace transform then get the
transfer function right. Suppose, you know like we want to do the same process empirically
that is I do not want to derive equations of motion or I am not in a position to derive
equations of motion because my system which is given to me is like a black box, I do not
know what components are there. So, I cannot I am in a position not to apply physics to
model the system

Then what I do is a I can use frequency response to my advantage right isn’t it I can use
experience to my advantage right. I can give sinusoids of varying frequency, I look at the
steady state output. Of I also get sinusoids of the same frequency, then I know that in the
frequency range of interest, I can model the system as an LTI system right that is point
number one.

And then what you do is that then you record the input sinusoid, output sinusoid you know
what is a scaling ok. You know the input amplitude output amplitude you will get the
magnitude at particular frequencies you can also measure the phase shift all right then you get
the phase plot then you plot the magnitude plot, plot the phase plot you get the bode plot
right. And you do the inverse problem that is given the bode plot, how do you get the transfer
function that is something which you can figure out right, so that is that is an important
practical application of what we are learning as far as the bode diagram is concerned right.

So, anyway I will give you a few more bode diagrams to plot as homework ok. So but now,
you know like what I am going to leave you with is a bunch of question is a set of questions
right.

397
(Refer Slide Time: 15:15)

So, I am going to ask you a question ok, please think about it and try to figure out the answer.
Using the bode plot bode plot, how can one calculate the values of Kpe, Kve, and Kae ok. I
think I am using; I hope I am using the same symbols right.

So, Kpe was a static position error constant; Kve was a static velocity error constant; Kae was a
static acceleration error constant. You remember when we did steady state error analysis.
Please refer to those notes right. I am hoping that using the same symbols ok.

So, the question is are like I am going to leave you with is it like how do we calculate these
constants right when we have the bode plot. Let’s say I give you the bode plot, can you use a
bode plot to graphically figure out these constants. And what is the advantage of this, because
using the bode plot you can tell me what is going to be the steady state error right when you
have certain types of inputs right, so that is something which I want you to figure out ok.

So, I am just going to stop here. So, with this we complete our discussion of bode plots. So,
in the big picture perspective what we are doing, we are doing frequency response right
essentially what happens when we give sinusoidal inputs to stable LTA systems we saw that
the steady state output is also going to be a sinusoid of the same frequency, but scaled in
magnitude and shifted in phase right. And the term which is important is what is called a
sinusoidal transfer function right so which we get by substituting s equals j omega in the
system transfer function.

398
Now, the question is how do we visualize the sinusoidal transfer function, so that we can do
frequency response analysis. One of the main tools which you will use in multiple domains
including automotive and biomedical which is which is the which are the groups of interest
for our department, we one of the main tools we use is what is called the bode plot that is
why I go ahead spend a quite a bit of time in explaining the bode plots.

So, next week what I am going to do is that like I am going to briefly discuss what is called as
a Nyquist plot and the Nyquist stability criteria. I am not going to spend so much time on that
ok, but I want you to know what they are right. And then like we will go to what to say
essentially controller design using frequency response methods ok. We did control design
using time response methods. Now we are going to use these frequency response methods to
design controllers. And then we look at both ok, so that is what our action plan is fine.

399
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 57
Nyquist Plot 1
Part – 1

Let us look at Nyquist plot, right. We have been looking at frequency response, which is the
response of the class of systems under study to sinusoidal inputs, right. If we have stable LTI
systems and if we give a sinusoidal input of frequency omega, the steady state output will
also be a sinusoid of the same frequency, but scaled in amplitude and shifted in phase.

The entity that becomes extremely important in that case is a sinusoidal transfer function
G ( jω). We learned about the bode plot, where we figured out how to graphically illustrate
the sinusoidal transfer function in terms of the magnitude and phase as ω varies from very
low frequency to very high frequencies. The Nyquist plot is alao a visualization of the
sinusoidal transfer function.

(Refer Slide Time: 01:35)

Nyquist plot is the plot of a G ( jω) which is a sinusoidal transfer function in the complex
plane. We essentially plot the real part of the sinusoidal transfer function versus the
imaginary part of that sinusoidal transfer function in the complex plane.

400
Sometimes it is called G (s) plane to indicate that it is no longer the s-plane. We are not
tracing the variable s, but rather we are tracing the transfer function G (s) that is a complex
valued function. So it is called as the G (s) plane or G ( jω) omega plane. It is essentially the
plot of the real part of G ( jω) versus the imaginary part of G ( jω), as omega is varied from 0
to ∞.

We are going to look at the application of this Nyquist plot in stability analysis and control
design. We will construct the Nyquist plot of individual blocks that we have been learning
and identify some difference between the Nyquist plot and Bode plot.

So, let us start with first building block as constant K. K is going to be either plotted on the
positive real axis or the negative real axis depending on its sign.

1 1 −j −1 1
Let us consider , G ( jω )= = =0+ j = ∠ 90 ° .
s jω ω ω ω

The function is always going to be on the negative imaginary axis. As omega starts from 0,
the plot is going to start from − j ∞. And as ω → ∞, it ends up at the origin.

Let us plot for G ( s ) =s, so then G ( jω )= jω=0+ jω=ω ∠ 90 °.

This function is going to be on the positive imaginary axis. As omega starts from 0, the plot is
going to start from the origin. And as ω → ∞, it goes to infinity along the imaginary axis,

1
Let us consider G ( s ) =
Ts+1

1 1
G ( jω )= = ∠−tan−1 ωT
1+ jTω √1+T 2 ω2

Let us plot Nyquist plot for this transfer function. Since plotting Nyquist plot by hand is
complex, we are going to look at simple examples when we discuss things analytically.

401
(Refer Slide Time: 11:10)

One process is to figure out what happens at critical frequencies and the limiting frequencies
and even some frequencies in between. Let us do that. We can construct a table to tabulate
what happens to the real part of G ( jω ), the imaginary part of G ( jω ), magnitude and phase as
we vary it at various critical ω. We construct the following table:

ω Re[G ( jω ) ] Im [G ( jω ) ¿ |G ( jω )∨¿ ∠ G ( jω )

→0 1 0 1 0°

1 1 −1 1 −45 °
T 2 2 √2
(corner frequency)

→∞ 0 0 0 −90 °

Please note that, the imaginary part is always going to be negative and the real part is going
to be positive. So, in the complex plane this Nyquist plot is going to lie in the fourth
quadrant.

402
Let us construct the Nyquist plot. We mark the coordinates of the obtained points

1 −1
(1 ,0)( , )(0 ,0) on the real and imaginary axes. We need to draw a curve which passes
2 2
through these three points. But we need to determine the structure. We evaluate the
following:

2 2 2
1 1 −Tω 1
[ 1+T ω 2
2 2
− −
] [
2 2
1+T ω
=
2 ]()
This equation will trace out a circle, right, which is centred at ( 12 ,0) and of radius 12 . That is
what is the Nyquist plot here. It is a semicircle that starts at ( 1 ,0 ) and ends at (0 , 0), with its

centre at ( 12 ,0). The arrow indicates the direction of increasing ω.

(Refer Slide Time: 21:49)

Let us consider the reciprocal of the previous example. G ( s ) =Ts+1. When we plotted Bode
diagrams, we observed, if we have two factors, which are the reciprocals of one another, the
Bode diagrams happen to be just mirror images about the 0 decibel line.

G ( s ) =Ts+1

G ( jω )=1+ jTω= √ 1+T 2 ω2 ∠ tan−1 Tω

403
And let us construct a similar table,

ω Re[G ( jω ) ] Im [G ( jω ) ¿ |G ( jω )∨¿ ∠ G ( jω )

→0 1 0 1 0°

1 1 1 √2 45 °
T

(corner frequency)

→∞ 1 ∞ ∞ 90 °

The real part is always 1 for this particular factor. If we plot the real part verses the
imaginary part in the G (s) plane, we are going to get a vertical line starting at (1,0) and then
just going up parallel to the imaginary axis. This is Nyquist plot of the factor (Ts+1). Look at

1
the Nyquist plot of the factor . One is a semi-circle, another is a straight line. We
Ts+1
understand that drawing a Nyquist plot is a different ballgame and a bit of a challenge.

404
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 58
Nyquist Plot 1
Part 2

(Refer Slide Time: 00:23)

Before we go to second order factors, I am just going to do another important factor which
becomes a very useful for engineering applications right. Because if you recall when we did
transient response analysis, I talk to you about time delay right. In many physical systems,
you are always going to have a time delay. So, what is meant by a time delay, I give an input
now, the output starts increasing from 0 after a finite time interval which I cannot neglect. So,
then there is time delay.

So, for example, I can have friction resulting in a time delay right. So, let us say I take this
desk, I give a force, the desk does not start to move immediately right. So, I may need to
overcome the static friction before the desk starts to move right that is the time delay. In fluid
systems; sometimes I may give an input I you will have some transmission delay before the
fluid pressure starts to build up right. Then what happens is that like we have a finite time
delay and we need to incorporate that, correct.

405
So, if you recall, what was the transfer function of the time delay? Some people call it as a
transport lag. So, because if you look at process control and so on, they will call it as
transport lag, because in process control they are interested in some fluid systems, fertilizer
plants, chemical plans and so on right, they call it as a transport lag right. So, essentially it is
a time delay right.

So, what is the transfer function of a time delay, e−T s . So, this is the transfer function of the
d

time delay. How did we get it? Because if you recall we have if you had a time delay my
input excitation becomes something like this right u(t-Td). Then if you take the Laplace of
this, what is going to happen you are going to get U ( s)e−T s that is why the transfer function
d

of the time delay is taken as e−T s right.

d

Let us construct the Nyquist plot for this right. So, what will happen to G(jω) then this will
become e−T ωright. And what is this, how can I rewrite this; this is cos ( T d ω ) − jsin (T d ω)
d

right that is the real part and the imaginary part correct. I am just using the Euler’s relation
right. And then like what can you say about its magnitude?

Student: 1.

One. Phase?

Student: −T d ω

−T d ω right, correct. Do you agree right? So, anyway I need not have even written the
imaginary and real part, I can just observe the exponential function this is just a complex
exponential. And the exponent does not have any real part right pure imaginary exponent
right. So, you can immediately get this right.

So, what do you think is going to be the shape of this particular sinusoidal transfer function.
If I want to plot the real part and imaginary part right, what do you think is going to be the
shape as omega goes from 0 to infinity. I think this thing should give you the answer, the
magnitude is always one. So, what can you say?

Student: circle.

Circle right, it is a unit circle right, isn’t it. The magnitude is always 1 right and the phase
goes like −T d ω right. So, if you plot it, so I am just going to try my best to draw a circle

406
freehand. This is omega tending to 0 as omega tends to infinity, once again I am going to
come back to the same point right, so that is that is what is going to happen to my circle all
right. This is the Nyquist plot of this transport lag.

Now, let us on top of this, let me superimpose the Nyquist plot of this first order factor. So,

1
let us consider the first order factor right which we have already drawn. What was?
1+ j (Tω)
That that was a semi circle of radius 1/2 right which was centred at oh sorry right. So, this is
1/2 right, and let us say this is sorry that is j/2 it is a minus j/2 this is - j this is let maybe a
careful let me use the same colour, so that this is 1, right this is - j, this is - 1, and this is + j
right.

1
So, so essentially we had a semicircle as the plot of right that is what we have right.
1+ j (Tω)
Why am I making this comparison? You can see that immediately at low frequencies, you see
that the two curves approximately are the same. Do you agree?

1
So, you see that at low frequencies, the Nyquist plot of e−T ωand d
or almost the
1+ j (Tω)
same. So, at low frequencies I am going to have what to say this thing to be true. So, you can
immediately see that for T d ω very much less than 1, you can immediately see that

1
the function can be approximated it as 1+ j (T d ω) right, if T d ω is very small.
1+ j (T d ω)

Similarly, what can I say about e−T ω? e−T ωcan also be approximated as 1− j (T d ω) for low
d d

frequencies right, so that is what we have correct. So, you can immediately see that at low
frequencies the functions become the same. And why did I what to say point this what to say
essentially highlight this point here, I told you that the function e−T s in fact, an infinite order
d

factor right. If you recall our discussion then right so then we have to do some finite order
approximations right; so, that we can get proper transfer functions if you recall when we want
to analyze systems with time delay.

What were the couple of approximations that we discussed, one was this, another was the
pade approximation right. So, I hope now it is clear how why this first order approximation
make sense right. And it is also very clear when you can apply them. See suppose if I make

407
this first order approximation, I approximate e−T s =1−T d s right. I can use it as long as I am
d

subjecting the system to free inputs whose frequencies are much lower than 1/Td ok, so that is
something which we need to analyze.

See, for example, let us say Td = 0.01, then I need to ensure that I am essentially subjecting
the system to frequencies which are lower than 1/0.01 alright, so that something which I need
to be careful about; So, then that approximations very reasonable ok.

So, why am I pointing this because we will see that in many engineering systems, we may
need to approximate time delay and that is where this discussion helps us right. It helps us in
taking a call as to physically when you can approximate use these approximations, is it clear.
So, I just use Nyquist plot to show you that when this approximation is reasonable. So, of
course, the at high frequencies two curves deviate from one another as you can clearly
observe fine. Any questions here, yeah.

Student: Sir, taking to omega tending to 0.

Omega tending to 0 where?

Student: (Refer Time: 10:59).

On the top ok, no omega tending to 0 it starts from see at omega equals 0 it starts from 1 0, it
comes into the fourth quadrant; as omega tends to infinity once again it goes and finishes at 1
comma 0. So, for this transport lag, the starting point and the ending point are the same.

Student: (Refer Time: 11:23).

Student: You get this phase angle plus minus 9.

Which one?

Student: Sir, (Refer Time: 11:29).

No, see that is where see you need to essentially use this visualization along with this
mathematical block ok, because this tan inverse function is also periodic right. We need to
look at it from that perspective. If you plot the what to say if you keep on plotting the values
of the real part and the imaginary part, this is what you will get ok, so that is where we end up
with ok.

408
So, as far as the Nyquist plot of the transfer function which corresponds to the transport lag or
time delay is concerned, it will be a unit circle so that is what will happen, fine yeah. So, now
I am going to just briefly discuss the second order factor, but I am going to leave it to you as
homework ok.

(Refer Slide Time: 12:27)

ωn 2
So, the detailed plot so let us go and plot for G ( s ) = 2 2 . So,
s +2 ζ ωn s +ωn

ω2 2 ζω
1−
1 ωn 2 ωn
G ( jω )= 2 = 2
−j 2
ω 2 ζω ω2 2 ζω 2
ω2 2 ζω 2
1− 2 + j(
ωn ωn
)
( ωn )( ) (
1− 2 +
ωn
1− 2 +
ωn )( )
ωn

So, this part I am going to leave it to you as homework as to how we got this Nyquist plot for
the second order factor. So, I am just going to plot the Nyquist plot you just need to complete
it based on what we have discussed now ok. I just adopt a similar approach like just plot the
sorry this is not that omega, this is real part ok, this is the imaginary part. And this is the G(s
plane, this is the origin.

So, you will observe that all curves are going to start at 1, when omega tends to 0. And what
happens is that the Nyquist plot takes a takes this shape and then like comes like this and ends
like this so that is what happens for a second order system. And of course, we are going to

409
have a family of Nyquist plots for different values of zeta ok, zeta keeps on increasing it
comes closer and closer ok.

So, this is the trend as zeta increases. And the, it cuts the imaginary axis when ω=ωn because
that you can immediately see that that the, what to say the real part is 0 right. If you look at
the real part for any value of zeta at ω=ω n, the real part is 0. So, it cuts all curves cut the
imaginary axis at ω=ωn ok. So, this is what happens at omega tends to infinity, but these are
the points where ω=ωn is the corner frequency if you remember in the bode plot right.

So, what I suggest you to do is to essentially construct a similar table as omega tends to 0,
omega tends to omega n, omega tends to infinity, and also ω=ωr. See ωr is the frequency at
which the magnitude is the maximum. Magnitude is a maximum means the point should be
the farthest from the origin.

So, for example, here this will be ω r right, so that is where it is farthest from the origin right,
you get the maximum magnitude. So, you can also identify resonant frequency from here
right. So, you please look at the resonance frequency also account for it ok. Of course, please
remember the resonant frequency is what to say reasonable one is holds only when there is

1
damping ratio is between 0 and . So, this generating this plot I leave it as a homework ok.
√2
So, I have told you the answer I have also told you the process right. I’ve given you the real
part imaginary part phase magnitude you know you just need to construct the table and draw
this right. It just takes the shape.

410
(Refer Slide Time: 18:00)

So, then you also need to do the reciprocal. What is a reciprocal? It is going to be

s2 +2 ζ ωn s +ωn 2 ω2 2 ζω
G ( s )= . So, this implies that G ( jω )=1− + j( ).
ωn
2
ωn 2
ωn

So, so once again if you plot this you can immediately observe that if you look at the real part
and the imaginary part what can you observe as far as this particular factor is concerned.
What can you say about the imaginary part for any zeta?

Student: (Refer Time: 19:26).

It is always non-negative. So, it is always going to be non negative. What can you say about
the real part?

Student: (Refer Time: 19:42).

It is first going to be positive, and then it becomes 0 at ω n then becomes negative. But you
can see that the real part monotonically becomes more and more negative after that ok; and
the imaginary part becomes more and more positive after that. So, you will see that the
Nyquist plot of this will look something like this.

So, this will go like this. So, this is what will happen as omega tends to 0; this is what
happens as omega tends to infinity at this point ω=ωm. So, I am just giving you the structure

411
of the Nyquist plot. So, please complete once again for both these factors draw the table and
then observe so that is what I want you to do. Is it clear?

So, you will see that once again the Nyquist plot of the factor and its reciprocal right does
not have a correlation as was the case in the bode plot ok. So, please remember that that is an
important observation here. So, this is the Nyquist plot. So, I just a given you an exposure to
introduction to Nyquist plot. So, a Nyquist plot is also a visualization of the sinusoidal
transfer function. So, where we plot the real part of G(jω) versus the imaginary part of G(jω)
in the complex plane right.

So, and important thing to notice is that like unlike the bode plot, I cannot split the transfer
function into individual blocks and then draw the individual Nyquist plots and just add them
in a very straightforward manner you know, we cannot do that ok. So, essentially I just
wanted to expose you to this. What we are going to do tomorrow is that we are going to see
how one could use this Nyquist plot to our advantage in stability analysis which is relevant to
this particular course ok.

So, I am going to discuss what is called as a Nyquist stability criteria; And we will see how
that is going to be used in stability analysis of closed loop systems right, so that is another
technique ok. So, I am just going to discuss that technique with you tomorrow before we go
to control design based on frequency response methods right, so that is that is going to be the
action plant ok. So, we will meet in the next class and continue this discussion.

412
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 59
Nyquist Plot 2
Part -1

We were looking at the Nyquist plot of some basic factors. Let us do one example and start
discussing about Nyquist stability criteria.

(Refer Slide Time: 01:17)

s+1
Consider a transfer function, G ( s ) =
s+10

1+ jω (1+ jω)(10− jω) 10+ω2 9ω

G ( jω )= = = +j
10+ jω 2
ω +100 100+ω2
100+ω2

10+ω2 2 9ω 2
|G ( jω)|=
√ 100+ω
2
+
100+ω
2

9ω
∠ G ( jω ) =tan−1
10+ω2

Let us create a table,

413
 ω Re (G ( jω ) ¿ Im (G ( jω ) ¿

→0 0.1 0

→∞ 1 0

We locate the points (0.1,0) and (1,0) on the G ( s ) plane.

(Refer Slide Time: 06:45)

What happens to the locus of this Nyquist plot as ω goes from 0 to ∞. We can see that the
real part is always going to be positive, for all omega between 0 to ∞. And the imaginary part
is also going to be greater than 0. The Nyquist plot is going to lie in the first quadrant.

To get the locus, sometimes, when you have this first order factors, you could try out an
equation of a circle. In this case, as shown the Nyquist plot is a semicircle with centre at
(0.55,0) and radius of 0.45. This is left as an exercise.

We can use the Nyquist plot to find out at what omega, the phase angle of this particular
sinusoidal transfer function is maximum. Phase angle at a point is measured by drawing a
vector from the origin to that point and calculating the angle made by that vector from the

414
positive real axis in the counter clockwise direction. If we want to figure out the maximum
angle, we need to draw the tangent.

(Refer Slide Time: 10:46)

Exercise: Using geometry, we can obtain the maximum phase ∅m and using the equation for
the phase angle we can get the corresponding frequency ωm .

In general, if the maximum phase angle is positive, that means there is a phase lead. And if
the phase angle is negative, that means there is a phase lag.

(Refer Slide Time: 16:06)

415
Exercise: plot the Bode diagram of the above transfer function and compare.

You see that the frequency at which you get the maximum phase contribution is going to be
at the geometric mean of the two corner frequencies (may not hold always).

The Bode diagram and the Nyquist plot are two methods to visualize the frequency response
of a transfer functions. There is a third plot which is called as Nikolas plot. We are not going
to use that in this course.

(Refer Slide Time: 23:54)

Here n =1 and m =1. This is a minimum phase transfer function.

There are two ways in which people look at a minimum phase transfer function or minimum
phase system. Some people associated only with the zeros because they implicitly they
assume a stable system to begin with. Another school of thought essentially tells that
minimum transfer function is one where all its zeros and poles are in the left of complex
plane.

416
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 60
Nyquist Plot 2
Part -2

(Refer Slide Time: 00:14)

Ok. So, if not the next concept that we are going to look at is what is called as the Nyquist
stability criteria.

Student: [Inaudible]

So, do you agree that this transfer function is a minimum phase transfer function? So, as we
know as omega tends to infinity, the phase goes to -90 (n – m). So, you substitute n and m
what do you get from this problem? 0.

417
(Refer Slide Time: 00:57)

And then you go to the Nyquist plot, where is the transfer function as omega tends to
infinity?

Student: 1

Plus 1. You draw a vector from 0 to 1, what is the angle made with a vector, 0 degrees all
right.

(Refer Slide Time: 01:10)

418
So, I am just using the Nyquist plot to just if you draw the bode plot, you will once again
observe the same thing. I am just interpreting it from the Nyquist plot that is all I am doing
right yeah ok. So, one important application of the Nyquist plot is to figure out the stability of
the closed loop system right.

So, so if you recall our block diagram, you know like of the closed loop system that we have
been looking at is that like the forward path transfer function was a G(s) if you recall ok. So,
this is E(s) right, and this is Y(s). And the feedback path transfer function is going to be H(s).
And this is going to close the loop let us say we have negative feedback let me call the signal
as W(s) right. So, this is the block diagram we have.

Y (s) G ( s)
What is a closed loop transfer function? Recall that = . So, this is the
R (s) 1+G ( s ) H (s )
closed loop transfer function ok. So, now how do determine whether the closed loop system
is stable say I am only asking the same questions again right, we are revisiting the same
points right. How do we check out the stability of the closed loop system? We look at the
closed loop characteristic equation. The roots of this equation are the closed loop poles right.
So, this is the closed loop characteristic equation ok.

And you immediately see that the closed loop characteristic polynomial is going to be

n o (s)
1+G ( s ) H ( s). That is the closed loop characteristic polynomial right. Let G ( s ) H ( s )= .
d o ( s)

See, in general rewrite any transfer function as a ratio of two polynomials in s numerator
polynomial n(s) and denominator polynomial d(s). So, I am just saying it as no(s)/do(s), so

no ( s) d o ( s ) +no ( s)
that is what we have ok. So, this implies that 1 1+G ( s ) H ( s ) =1+ = . So, we
d o (s) d o (s )
will see why I am doing all these things, so because I want to set some terminology in order
before we go in go further right.

419
(Refer Slide Time: 04:49)

So, this immediately tells me that you can immediately see that the poles of 1+G(s)H(s) are;
what are they? So, if you consider the closed loop characteristic polynomial 1+G(s)H(s) right,
that is going to be a ratio of two polynomials in s, is it not. What is the denominator
polynomial?

Student: (Refer Time: 05:28).

do(s). What are the roots of do(s) called as?

Student: (Refer Time: 05:33).

Poles of what?

Student: (Refer Time: 05:35).

Open loop poles. Please remember that because do(s) is the denominator polynomial of the
open loop transfer function. So, immediately you see that the poles of 1+G(s)H(s) are the
open loop poles. Are these known, in general? By and large they are under our control right
because what do the open loop poles constitute? They constitute the plant poles controller
transfer function pole and the feedback path transfer function pole.

So, in general at least we will be able to tune them even if the controller contribute some
open loop poles, we will be able to tune them. So, we know what these poles are right yeah.
So, what can you say about the zeros of 1+G(s)H(s).

420
Student: (Refer Time: 07:00).

Closed loop poles that is it right so yeah. So, I am recapping this terminology once again
because we are going to mix poles and zeros right, but I want us to be very clear. I am
considering the polynomial 1+G(s)H(s) which is a ratio of two polynomials. It so happens
that the denominator polynomial of 1+G(s)H(s) is the same as the denominator polynomial of
the open loop transfer function.

So, the poles of 1+G(s)H(s) are nothing but the open loop poles. The zeros of 1+G(s)H(s) are
the closed loop poles. This is the one which is of importance for us right, because we are
when we design the closed loop system, we are worried about the closed loop poles, open
poles are already given to us right. We are worried about what are the closed loop poles. In
other words what are the zeros of 1+G(s)H(s), is it clear?

Now, please remember this, what to say, terminology, we are going to pose a question. Yes,
please.

Student: (Refer Time: 08:04)

Why should they be closed loop zeros? No, but 1+G(s)H(s) is the denominator of the closed
loop transfer function right. Isnt it? I am extracting that is why I am doing it carefully step by
step. So, please go back and of course, very natural question because I am extracting the
denominator then I am writing the denominator of the closed loop transfer function as a ratio
of one another numerator denominator then I am processing it right, so that is why I am doing
it step by step. But from here onwards, I will just use this terminology that is why we came to
this all right.

Now, the question that we are asking ourselves, we are going to ask ourselves; and the
answer to which would be provided by the Nyquist stability criteria is the following.

421
(Refer Slide Time: 10:09)

Given the open loop transfer function G(s)H(s), that means we know the open loop poles can
we comment about the stability of the closed loop system using the Nyquist plot of the open
loop transfer function G(s)H(s). This is a question we are going to ask ourselves ok.

So, let me repeat the question. So, let us read the question, that is you are given the open loop
transfer function G(s)H(s) right by and large it is known right. So, we know the open loop
poles right. So, open loop poles are nothing but the poles of 1+G(s)H(s). Please remember
that right, so as we see in the green box right. So, open loop poles are nothing but poles of
1+G(s)H(s) we know them right.

Given the open loop transfer function can we then use the Nyquist plot of the open loop
transfer function G(s)H(s) to comment about this close loop stability that is what we are
asking ourselves. And of course, the answer is yes that is why we have the Nyquist stability
criteria. So, but the general result on which this stability criteria rests on is what is called as
the mapping theorem in complex algebra there is something called as a mapping theorem.

Once again I am not going to prove any of this, we are just going to use them. The mapping
theorem reads to of course, this is the answer of course, the answer is yes. How? We will visit
what is called as a mapping theorem. And the mapping theorem leads to what is called as a
Nyquist stability criterion. So, this is what I have typed in that handout ok. So, what I want
you to do is that now I want you to read through the handout, if you are not already done.

422
So, if you go to the course website below root locus construction, I have uploaded another
file called Nyquist stability criteria. I hope you are able to access that file right. So, in that I
have clearly typed what is the mapping theorem and how it is applied to the Nyquist stability
criteria. Kindly go through the document and come to the next class we would discuss the
same all right in the next class. But I am going to leave you with another question; please
think through these questions.

Let us say let the plot of G(jω)H(jω) that is the open loop sinusoidal transfer function. So,
G(s)H(s) you substitute s=jω you get the open loop sinusoidal transfer function. And let this
be given for omega belonging to 0 to infinity ok, so that is what we have plotted as the
Nyquist plot right. You give me any transfer function that is a Nyquist plot.

(Refer Slide Time: 13:46)

s+1
See for example let us say just for sake of example. Let us say the transfer function was
s +10
the open loop transfer function, let us say right. We have plotted the Nyquist plot of that as
omega varies from 0 to infinity.

So, till now we have been looking at omega as a physical parameter right, where it is
associated with frequency and it can take only positive values. But when we go to this
mapping theorem and Nyquist stability criteria, we are going to look at omega as a real
parameter and it can take values from minus infinity to plus infinity ok. We will learn that in

423
the next class. So, if that is the case, how can one obtain the plot of G(jω)H(jω) for omega
between minus infinity to 0.

So, the Nyquist plot you now like by definition it is a plot of any transfer function by
substituting s equals j omega and you plot from omega going from 0 to infinity that is the
definition. But let us say instead let us now look at omega as a parameter or rather than
associating the meaning of frequency.

So, let us say I substitute s=jω, omega is now a parameter and that parameter can vary from
let us say minus infinity to plus infinity. So, I am given already the Nyquist plot of the
transfer function when omega varies from 0 to infinity. How can I get the Nyquist plot of the
same transfer function, when it when I go from minus infinity to 0 ok. Please try this one out
for all the transfer functions that we have plotted till now, and then like figure things out.

So, let me do for one example and then of course, I am going to let you extrapolate ok. So, I

1
am just going to do one example and then we will see. See, for example, we did G ( s ) = . So,
s

1 −j
this means G ( jω )= = .
jω ω

So, now if I have omega to be non positive that is it comes from minus infinity to 0, what do
you think will happen? Let us plot. So, what was the Nyquist plot when it came from 0 to
infinity, as it reached infinity where it would go?

Student: (Refer Time: 16:49).

We did it yesterday right. The Nyquist plot was the negative real imaginary axis this was
omega tending to 0; this is omega tending to infinity. Let us say it is the other way around,
what will happen? When it goes from minus infinity to 0 what do you think will happen, what
is the value as it tends to minus infinity anyway it will be 0. And for as it goes to 0, it will
blow off to plus infinity because it is coming from the negative side.

So, what will happen is then it will also start from this as it goes from minus ready and then it
will go to plus this is omega tending to 0 minus. So, I am just going to put a 0 plus and 0
minus just to indicate that from which side am I coming to 0 all right. 0 plus means I am

424
coming going to 0 from the right ok; 0 minus means I am coming to 0 from the left from the
negative real axis right negative side.

But of course, this is a very special example, so you will see that you will have a what to say
radius of a sorry semicircle of infinite radius as your Nyquist plot, but this is a very special
example. I am just giving you one example for to show you how to do that right.

So, kindly plot the first order factor that we looked at and then tell me what you will get when
you have omega as a parameter which varies from minus infinity to plus infinity so that is
something which I am going to leave it to you as homework fine. So, when we meet in the
next class you know like we will look at the mapping theorem and then like we will also
discuss the Nyquist stability criterion.

425
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 61
Nyquist Stability Criterion
Part – 1

We started off with the Nyquist Stability Criteria. We are trying to figure out information
about the stability of the closed loop system by looking at the frequency response of the open
loop system.

(Refer Slide Time: 00:34)

Let us consider the standard negative feedback system. G (s) being the forward path transfer
function, H ( s) being the feedback path transfer function. G (s) H (s) being the open loop
transfer function. And the closed loop transfer function is going to be

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

Consequently the closed loop characteristic polynomial is going to be 1+G ( s) H ( s). The
roots of this closed loop characteristic polynomial are the closed loop poles. Let us consider
the open loop transfer function to be represented as:

426
 n 0 ( s)
G ( s ) H ( s )=
d 0 (s )

n0 ( s) d 0 ( s ) +n0 (s)
1+G ( s ) H ( s ) =1+ =
d 0 ( s) d 0 (s)

We are particularly interested in the numerator because the roots of the numerator polynomial
are the closed loop poles, right.

(Refer Slide Time: 02:10)

The zeros of 1+G ( s ) H ( s ) are the closed loop poles. The poles of 1+G ( s ) H ( s ) are the open
loop poles.

The question that we are asking ourselves is, can we comment on the location of the closed
loop poles and in turn can we comment on the stability of the closed loop system by plotting
the frequency response of the open loop system.

Can we plot G ( jω ) G ( jω ) as a Nyquist plot and then from that comment on the stability of the
corresponding closed loop system. The first concept we are going to look at is called as a
mapping theorem

427
(Refer Slide Time: 03:59)

Let us go to the handout and look at the mapping theorem.

Let F ( s ) be a ratio of two polynomials in s. Let a closed contour in the s-plane contain Z
zeros, and P poles of F ( s )without passing over any of the poles and zeros of F ( s ). Then this
closed contour in the s-plane is mapped to a corresponding closed contour in the F ( s ) plane.
If you have a path in the s-plane obviously we are going to have at each and every point in
the contour a particular value of s, that we can substitute in F ( s ) and get a corresponding
value in the F ( s ) plane. We count the number of clockwise encirclements of the origin by the
contour in the F ( s ) plane. The number of clockwise encirclements of the origin by the
contour in the F ( s ) plane that is going to be equal to Z minus P.

428
(Refer Slide Time: 06:49)

First we need to mark the number of poles and zeros of F ( s) on the s-plane and take any
contour. This is going to be mapped into a corresponding contour in the F ( s) plane. Let us
say F ( s) has two poles and a zero. Let us take some arbitrary contour (green) for the sake of
argument. The value of Z-P in this case is 0. So the number of encirclements of the origin it is
going to be 0. Let us consider another contour (black) which encircles both zeros and pole.
The value of Z-P is -1, which means that there is a clockwise encirclement of the origin in
F ( s) plane.

(Refer Slide Time: 12:30)

429
(Refer Slide Time: 14:51)

Let us look at some examples from presented in the handout. We take F ( s) to be this
function, which has two zeros at -1.5±2.4j and two poles at -1 and -2.

(Refer Slide Time: 15:33)

If we look at the first figure, we can see the two poles at -1 and -2. And the two zeros are at
are complex conjugate at -1.5±2.4j. If we take a contour ABCD in the s-plane, which is not a
closed contour. We can see that they get mapped to contours A’B’C’D’. We get a different
shape altogether, that depends on the function, right. Similarly if you have the contour DEFA
that gets mapped to a contour D’E’F’A’.

430
(Refer Slide Time: 16:52)

We take a closed contour ABCDEFA that gets mapped to the contour A’B’C’D’E’F’A’. The
value of Z-P=-2. Therefore we can see that there are two clockwise encirclements around the
origin in the F ( s) plane, A’B’C’D’ and D’E’F’A’

(Refer Slide Time: 21:46)

There are a few more examples in this case that we can look at and verify the Mapping
theorem. We can also use MATLAB to get an understanding of mapping of various contours.

431
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 62
Nyquist Stability Criterion
Part – 2

(Refer Slide Time: 00:15)

Now, the important question is that how is it useful to us right. So, in what we are going to
use it.

Student: [Inaudible]

We are coming to that ok. So, we will discuss all those things right. So, in a certain sense,
yes, of course, one can also argue the other way that is why I am coming to this one right, so
that is where I am going to say, application of mapping theorem to stability analysis that is
what we are going to do.

432
(Refer Slide Time: 00:43)

So, application of mapping theorem to stability analysis, so that is what we are going to do.
So, now, how is it essentially applied let us look at that question.

(Refer Slide Time: 01:38)

So, the way it is applied is the following ok. Now, we will go back to our closed loop transfer
function right and the closed loop characteristic polynomial which is 1+G(s)H(s) correct. So,

d 0 ( s ) +n 0 ( s)
consider F ( s )=1+G ( s ) H ( s ) = . Now, the question is that like how do I apply the
d 0 ( s)
mapping theorem to this polynomial which is a ratio of two different polynomials right. We

433
are interested in the open loop sorry closed loop poles right, their location ok. Now, for
closed loop stability what is it that we want?

Student: [Inaudible]

Which poles?

Student: [Inaudible]

Closed loop poles right if I want my closed loop system to be stable I want all the closed loop
poles to be in the left of complex that means, that I do not want any closed loop pole to be
either on the imaginary axis or the right half plane ok. So, consequently we will see whether
we can use them to our advantage ok.

So, let us take the s-plane. Suppose let us say I take a contour where I start from the origin I
go along the imaginary axis to infinity along the imaginary axis that sweep the entire right of
complex plane like an arc in the form of a semicircle of infinite radius and then come back to
the come back to the origin ok. Suppose, let us say I consider such a contour in the s-plane.
So, my contour the closed contour in the s-plane is a semicircle of infinite radius that
essentially covers the entire right half complex plane.

So, first let me convey the concept then we will discuss any questions right. So, let us say this
is the contour we are considering in the s-plane ok. So, let the closed contour of course, one
can ask look is this contour really closed? So, would you say that look you know like when
you take a arc of infinite radius right would you consider it to be really closed or open. So,
we will not get into that level of debate here right.

So, we will say that look let the contour in the s-plane be a semicircle of infinite radius that
sweeps the entire right half plane ok, so that is the closed contour in the s-plane ok. Now, we
want essentially no closed loop poles within this contour do you agree? See for closed loop
stability, we do not want any closed loop poles will to lie within this contour right that is
what we want right. We do not want any closed loop poles on the imaginary axis neither do
we want any closed loop pole in the right of complex plane right.

So, and if we assume that to as a first step we assume that none of the open loop poles and
open loop zeros lie on the jω axis to begin with ok. Then I will ask I will leave with a
question what happens if it, it does not happen right. So, see I can have open loop poles and

434
open loop zeros in the right half complex plane that is allowed right, but I assume that look
for the open loop poles an open loop zeroes right none of them lie on the Imaginary axis ok,
so that is an assumption that we are essentially looking at. Now, further we assume that

lim G ( s ) H (s) is either zero or a non-zero constant ok.

s →∞

(Refer Slide Time: 07:24)

So, that is what we are looking at which is essentially saying that the system is causal ok. We
have a proper transfer function to begin with you know that is a class of system anyway we
are dealing with.

Why am I looking at this because then this will imply that when I am going on a circle of
infinite radius s is tending to infinity right is it not, then that means, that at that value of s the
all the points will be mapped to one point in the G(s)H(s) plane ok, so that is the implication
of the system being causal right ok. So, that these are some assumptions that we took. Now,
this mapping is now this contour is now mapped into a corresponding contour in the
1+G(s)H(s) plane ok.

So, let us say I map this contour in the F(s) plane, but what is this F(s) plane for stability
analysis? F(s) was 1+G(s)H(s) right ok. So, let us say that gets mapped into some arbitrary
contour in the F(s) plane. Now, recall that Nc = Z – P.

435
(Refer Slide Time: 08:57)

So, once again what is Z? Z is the number of zeros of F(s) within the contour in the s-plane
right. Now, for closed loop stability we want Z to be.

Student: [Inaudible]

0.

Z is to be 0. Please remember that F(s) was 1+G(s)H(s). The zeros of this F(s) are the closed
loop poles. We do not want any closed loop pole in this contour all right so that that is why
we want Z to be 0. So, what should be then Nc? N c should be -P. What is P? P is the
number of open loop poles right which lies within this contour in the s-plane right.

So, you know the location of open loop poles anyway, we are given a open loop transfer
function, we know where are the open loop poles within this contour and I know the value of
P right, but of course, there is an important assumption right, so that is where this assumption
comes right we assume that none of the open loop poles and zeros lie on the jω axis because
the original mapping theorem states that you consider a contour where none the contour does
not go along go on pass through any of the poles and zeros of the function F(s). That is an
assumption that we are doing.

We will then see what happens when we; see sometimes we may given we may be given
open loop transfer function where the open loop poles are on the imaginary axis. Let us say
you take a PI controller you will have a pole at the origin right we have already seen it then

436
can I choose this control we will ask ourselves that question later. For the time being we will
first deal with the initial simpler case where there are no open loop poles on this j omega axis
all right, so that is what we are.

So, now what does it say. So, this essentially tells that for closed loop stability the contour in
the 1+G(s)H(s) plane should encircle the origin P times in the counterclockwise direction. Do
you agree? That is what is required for stability right. We are not done yet we are only
halfway through right. Do you agree? Ok, so that is what this equation tells me right.

Now, now, I am going to essentially go from this 1+G(s)H(s) to the G(s)H(s)plane. So, please
note you know like we are talking about closed loop stability. So, what it means is that for
closed loop stability the contour in the 1+G(s)H(s) plane should encircle the origin P times in
the counterclockwise direction that is the condition. If it does not the closed of a system is
unstable that is what we can conclude right yeah.

So, now from the 1+G(s)H(s) plane, we go to the G(s)H(s) plane. Now, how do I draw a
contour from the 1+G(s)H(s) plane to the G(s)H(s) plane? I subtract one; see how do I get
G(s)H(s) from 1+G(s)H(s)? minus 1 right. So, say of course, this very simple why am I
writing it essentially what it means is that the contour in the 1+G(s)H(s) plane is shifted to the
left by one unit along as far as the real part is concerned. So, you take all the points in the
1+G(s)H(s) contour you just shift it shift the real part of all the points by 1 to the left; that
means, you are subtracting 1 then you will get the corresponding contour in the 1+G(s)H(s)
plane let us say that contour is something this similar shape ok, but that is something like this
ok.

Now, where is the origin shifted? see let us say this is my origin which is the critical point
right as far as the mapping theorem is concerned what is this critical point shifted to. So, the
origin of the 1+G(s)H(s) plane is shifted to , you are subtracting 1 right what is 0 minus 1.

Student: Minus 1.

Minus 1 right. So, it is shifted to minus 1 in the G(s)H(s) plane ok. So, we can say that the
number of encirclements of the origin in the 1+G(s)H(s) plane is the same as the number of
encirclements of the origin of the minus 1 point in the G(s)H(s)plane. Do you agree? So, let
me repeat that statement once again. The number of encirclements of the origin in the
1+G(s)H(s) plane is the same as the number of encirclements of the minus 1 point in the

437
G(s)H(s) plane. I do not change the function. I have just shifted the contour by one to the left
all right yeah.

Student: in this case p is a nonzero number?

If P is a nonzero number then that means, the open loop system is unstable. See P indicates
open loop poles right within the contour, yeah you can have an unstable open loop system
question is that by feedback are you stabilizing it right that is what you are asking you can
right ok. So, I hope how the critical point becomes minus 1 in the G(s)H(s).

So, the same condition one can rewrite this as the contour that is for closed loop stability, the
contour in the G(s)H(s)plane should encircle the minus 1 point not origin minus 1 point P
times in the counterclockwise direction ok. This becomes the Nyquist stability criteria now
ok. So, we are not done yet you know like we are just going step by step this is this is a
parallel to next step ok. Let me go. I hope it is clear right.

So, because whatever is the number of encirclements of the contour in the 1+G(s)H(s), when
you map it to G(s)H(s) plane, it will be the same as the number of encirclements of the minus
1 point because you are just subtracting one that is it you are just shifting everything to the
left by one right. So, the critical point which is origin in the 1+G(s)H(s) plane becomes the
minus 1 point in the G(s)H(s) plane right. Is it clear? Now, you are seeing that this contour in
the s-plane is also like very special, it is not an arbitrary contour right like the general
example of the mapping theorem that we looked at.

438
(Refer Slide Time: 18:12)

So, if you look at the s-plane, what is the contour that we are considering it is a semicircle of
infinite radius that goes from -j∞ to +j∞ then sweeps the entire right of plane and then comes
back all right. So, this is like infinite radius right LHP and RHP ok. So, this is an semicircle
of infinite radius that sweeps the entire right of complex plane is it not ok.

Now, the question that arises is that what sort of form that s takes in this contour. So, note
that S is of the form jω all right, omega belonging to minus infinity to plus infinity right in
this contour. Is it not, anyway when s sweeps in the arc of infinite radius G(s)H(s) is a finite
number. It is going to be the same point that is our assumption from that is the consequence
of causality right.

We are assuming that the system is causal, so as a result when I go on the arc of infinite
radius G(s)H(s) will be at a point. That is it right, it is not going to vary. The variation is
going to come only in the part where S goes from -j∞ +j∞. In that region in that contour s is
purely imaginary, is it not. So, this implies that we are going to be interested in the plot of
rather than saying G(s)H(s) because s is of the form jω.

We are going to be interested in the plot of G(jω)H(jω) as ω varies from minus infinity to
plus infinity right. And what is this plot called as? The Nyquist plot of the open loop transfer
function right if you give me any transfer function you substitute s= jω and see how omega
varies from 0 to infinity that is the Nyquist plot of course, here we are treating omega as a
parameter.

439
So, I am making omega to be negative also. And we I asked you a question last class right
what happens when omega becomes negative right we figured out that in a certain sense it
becomes a mirror image right about the origin right. So, essentially that is what is going to
happen here, so that is why we are essentially going to be interested in the Nyquist plot of the
open loop transfer function to figure out the closed loop stability, so that is what leads to the
statement ok.

So, let me go back we will read this statement. And we will come back and what to say
essentially discuss it in the next class and I will allow you to also like think through these
details right. We are discussed lot of points you know like we will recap once again right.

(Refer Slide Time: 21:28)

So, if you look at the Nyquist stability criteria that is the Nyquist stability criteria finally, of
course, this is a criteria when G(s)H(s) does not have any poles and zeros on the imaginary
axis that is the case we started off it. If the open loop transfer function G(s)H(s) as k poles in

the right of complex plane, and if slim G ( s ) H (s) is either zero or a nonzero constant, then for
→∞

stability of the closed loop system the locus of G(jω)H(jω) as omega is varied from minus
infinity to plus infinity must encircle the minus 1 point k times in the clock counter clockwise
direction.

So, essentially what we are saying is that we are saying P = k right. And then we want Z to be
0. So, what should be Nc? It should be -k and that is k counter clockwise encirclements of the

440
origin in the 1+G(s)H(s) plane and that is going to be k and counter clockwise encirclements
of the minus 1 point in the G(s)H(s) plane. And we are writing G(jω)H(jω) because like s =
jω ok, so that is why we are looking at this.

Student:[Inaudible]

Yeah, that case I will I will discuss what to do in the next class right. So, so essentially that is
the first case we are dealing with right. Essentially what happens when no open loop pole or
open of zero lies on the imaginary axis that is the case we are still at. Now, we will then see
what happens if it does what should be changed right, so that is something which we will
discuss.

441
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 63
Relative Stability 1
Part – 1

We were looking at the Nyquist stability criteria and we studied what the mapping theorem.

(Refer Slide Time: 01:42)

We are going to apply it to stability analysis. We consider a contour in the s-plane, which
essentially goes from − j ∞ to j ∞, traversing the entire imaginary axis. This contour is
typically called as the Nyquist contour. For closed loop stability, we want Z to be 0. If there
were P open loop poles and K open loop poles within the Nyquist contour. Then the contour
in the G ( s ) H ( s) plane must encircle the -1 point K times in the counterclockwise direction.
Because we looked at F ( s) to be 1+G ( s ) H ( s), we shifted everything by -1. Then -1 becomes
the critical point. That is a brief recap of Nyquist stability criterion.

(Refer Slide Time: 03:36)

442
(Refer Slide Time: 04:55)

Let us read the criteria once again. When G ( s ) H ( s) does not have any open loop poles or
open loop zeros on the imaginary axis. If the open loop transfer function G ( s ) H ( s) has K

poles in the right of s plane. And if slim G ( s ) H (s)is either 0 or a non-zero constant, then for
→∞

stability of the closed loop system the locus of G ( jω ) H ( jω) as omega is varied from −∞ to
∞, must encircle the -1 point K times in the counterclockwise direction. If it does not, that
means the closed loop system is unstable.

443
 (Refer Slide Time: 06:08)

What happens if the open loop transfer function G ( s ) H ( s) has zeros and poles on the
imaginary axis?

1
Consider the open loop transfer function G ( s ) H ( s )= .
s ( s+1)

1
G ( jω ) H ( jω )=
jω( jω+1)

There is an open loop pole at the origin. We construct what is called as the modified Nyquist
contour. We start from − j ∞ and move closer to the origin. Since we have an open loop pole
at the origin, we essentially a modified Nyquist contour, which takes a semicircular path
around the open loop pole at the origin of infinitesimal radius (ϵ ).

For s=¿, s is of the form ϵ e jθ , θ=[−90 ° , 90 ° ]. We are representing in the polar form. We
have the magnitude to be ϵ and the phase to be θ.

1
G ( ϵ e jθ ) H ( ϵ e jθ ) ≅
ϵ e jθ

Because s+1 ≅ 1. The magnitude of the open loop transfer function very high because ϵ is
small. We can conclude that we have a counterclockwise semicircle of infinitesimal radius in
the s plane.

444
We apply the Nyquist stability criteria. Here P = 0. Also Z – P = 0, so the Nyquist should not
encircle the -1 point in G (s) plane.

(Refer Slide Time: 15:58)

This completes our discussion on Nyquist stability criteria. The next concept related to
frequency response is called as relative stability.

(Refer Slide Time: 19:41)

Let us take two underdamped second order systems, System 1 (red) and System 2 (green).
System 1 has poles which are closer to the imaginary axis compared to the system 2. Both are
stable since all the poles are in the left half complex plane.

445
There are approximations when we model systems. There are going to be some un-molded
dynamics and modelling errors. There are going to be parametric variations. So the locations
of the poles can change a little bit. We may have a band or a zone in which the pole may lie.

Let us construct a region around the nominal value of the pole. The poles of system 1, if
perturbed they go closer to the imaginary axis and maybe at some point hit the imaginary
axis, then we are going to lose closed loop stability. In that sense, relatively system 2 is more
stable. This is what is called as a relative stability. What we learnt before was absolute
stability. Both systems are stable as far as absolute stability is concerned.

But, just because we want to make a system more relatively stable, we cannot keep on
pushing the poles to the left. It will increase the rise time of the system.

446
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 64
Relative Stability 1
Part – 2

(Refer Slide Time: 00:14)

So, now that we understand what is relative stability right, what do we mean by that right. So,
the question that we need to ask ourselves, see this is a very qualitative discussion is it not?

See, you plot the open loop poles that is a poles of the transfer function of the system transfer
function, and say hey look you know this system is relatively less stable than the other one
right, but then we need to have quantitative measures which essentially help us in taking a
decision, right when we do control design right. So, then the questions that we need to ask
ourselves is that are the following.

So, the one question we can ask is the following you know what are measures of the degree
of stability of a closed loop system? That is a question that we need to ask ourselves right. To
answer this let us revisit our standard feedback system that we have been considering where
the forward path transfer function is G(s) ok.

447
(Refer Slide Time: 01:36)

The feedback path transfer function is H(s) ok. And there is an error E(s) right. So, then

Y (s) G ( s)
please note that = so that is what we are going to have right ok. We are
R (s) 1+G ( s ) H (s )
going to look at the closed loop characteristic equation ok, because the roots of this equation
are the closed loop poles.

Suppose, let us say the closed loop system is stable to begin with right then we know that all
closed loop poles are in the left of complex plane ok. Now, due to some parametric
uncertainties the closed loop system what to say is on the verge of instability, but then you
see that if my closed loop poles are in the left half plane to begin with, they can essentially
transition to the right of plane only when they go through the. What is the boundary between
the left half plane on the right of plane?

Student: jω

jω axis right, so that is essentially that is why it is called as a stability boundary right. So,
essentially the jω axis or the imaginary axis is the boundary between the left half plane and
the right half plane. So, if my closed loop system is stable to begin with you know the closed
loop system poles if they are migrating closer and closer to the right of plane, they have to
necessarily pass through the jω axis right which is a stability boundary before they go to the
right half plane. Do you agree?

448
So, we can use that to our advantage ok. So, the point is the system or the closed loop system
would become unstable when its poles cross the jω axis right. So, of course, the presumption
is that to begin with the closed loop system is stable all right that is when all this discussion
makes sense right, so that is on the j omega axis what will happen the closed loop
characteristic equation will become this right, because s is of the form jω. Do you agree
right? So, this is what will happen to the closed loop characteristic equation.

So, so in this scenario then G(jω)H(jω) = -1. So that is when the closed loop poles are in the
left half plane and they are migrating to the right half plane due to whatever reason right
some parametric concern at disturbances and so on. So, then they need to cross the jω axis let
us say you stop when they when they are crossing the j omega axis some closed loop pole
right. So, the closed loop characteristic equation is going to satisfy 1+ G(jω)H(jω) = 0. I can
rewrite it as G(jω)H(jω) = -1. So, what is G(jω)H(jω)?

Student: [Inaudible]

That is what we have been reading under frequency response right, it is the open loop
sinusoidal transfer function right. So, G(jω)H(jω) is the open loop sinusoidal transfer function
so that is what it is ok. Now, what are we interested in then. So, the question is the system
would become unstable when the Nyquist plot of G(s)H(s) right passes through the minus 1
point because if there is some value of S, which makes the value of G(s)H(s) to be minus 1
right, then the system would become unstable right.

(Refer Slide Time: 06:50)

449
So, or in other words, I can reinterpret this as the following. The closeness of the G(jω)H(jω)
to the minus 1 point is indicative of relative stability, so that is what one could conclude right
from this. So, see what does it mean? Let us say I have the G(s)H(s) plane right. So, let us
say this is my minus 1 point right. So, ok.

So, let us say you know like I have let us say some I want to say a Nyquist plot right. So, let
us say what some systems Nyquist plot which essentially let us say starts from the origin and
then goes like this. And or let us let us do it in this way right. So, let us say comes like this
and goes like this ok. So, another system which essentially comes and goes like this ok. So,
you see that the second system is going to be closer to the minus 1 point, you perturb it, you
stretch the Nyquist plot a little bit it can pass through the minus 1 point maybe right.

So, the closer the locus is to the minus 1 point the less relatively stable that system is right, so
that is what we can conclude about conclude from this particular discussion on relative
stability. Now, note that the minus 1 point in the G(s)H(s) plane means the magnitude is 1,
the phase is minus 180. I am writing minus 180, because right you see that typically we deal
with proper transfer functions you will understand it shortly. So, so of course, minus 1 point
in the in G(s)H(s) plane indicates that the magnitude of the open loop transfer function is one
at that point and the phase of the open loop transfer function is minus 180 degree. Do you
agree? So, I am just writing in terms of magnitude and phase. We are going to use this idea to
define a few parameters that can quantify relative stability ok, so that is what we are going to
do now ok. So, are there any questions now? Yes.

Student: [Inaudible]

Sorry which one.

Student: [Inaudible]

So, that follows from this because you see that see what is the scenario we are considering,
let us say we go back to this right. You plot the Nyquist plot of system 1 and system 2, it may
look like what I have plotted right. So, because the closer it is to the minus 1 point you know
like if you perturb it, it can pass through minus 1. What does passing through minus 1 mean
as far as the Nyquist plot of the open loop transfer function is? That means, that s=jω satisfies
the closed loop characteristic equation that means that a closed loop pole is going to be of the
form j omega.

450
In other words you are going to have closed loop poles on the imaginary axis and that is what
we are discussing right, so that is why we are breaking down the problem or not breaking
down or converting the problem of analyzing closed loop stability once again into a problem
of analyzing the open loop transfer function, see that is a big advantage with frequency
response methods.

So, one advantage of frequency response methods if you go back and think through is that
even in Nyquist stability criteria you know like we are looking only at the open loop transfer
function. Even here I am looking only at the open loop transfer function not at the closed loop
transfer function right. So, we dont need to calculate the closed loop transfer function. You
give me the open loop transfer function I can analyze its Nyquist plot, and then tell you some
information of the closed loop system ok, so that is the impact here ok.

So, you see that I can rewrite this closed loop characteristic equation in this form all right. So,
the as G(jω)H(jω) = -1 if there were a closed loop pole on the jω axis that is a scenario we are
considering ok, yeah ok. So, a minus 1 point in the G(s)H(s) plane means at the magnitude of
G(s)H(s) is 1 and the phase of G(s)H(s) is minus 180 degrees ok, is it clear ok, those are two
important ideas ok. So, based on this we are going to essentially have a few definitions ok.

(Refer Slide Time: 12:15)

So, let me write down those definitions ok. The first definition is the following ok. So, we are
going to define what is called as a gain cross over frequency ok. I will define these two

451
frequencies then we will come back in the next class and then define two important
parameters ok.

So, what is this gain crossover frequency? The gain crossover frequency ω git is the frequency
at which the open loop sinusoidal transfer function magnitude becomes equal to 1. So, the
magnitude of the open loop transfer function should become unity or 0 decibels ok. If you
convert to 0 decibel measure obviously you if you take logarithm of one you will get 0 right,
so that is the definition of the gain crossover frequency. It is the frequency at which the
magnitude of the open loop transfer function becomes 1 right.

Then we have another definition all these are definitions please remember that ok. We have
another definition which is called as a phase crossover frequency ω p . What is the definition
of phase crossover frequency? It is the frequency at which the phase of the open loop transfer
function becomes minus 180 degrees ok, so that is the definition.

So, of course, you can immediately see how these definitions are motivated right, because
why the one point has a magnitude of one and a phase of minus 180 degrees right that is
motivating these two definitions ok. So, of course, if at if ω g =ω p for some particular open
loop transfer function, what does it immediately imply that the open loop transfer function is
passing through the minus 1 point at that frequency, is it not? think about it right. If ω g =ω p
for some open loop transfer function, then at that frequency the magnitude of open or transfer
function is 1, phase is minus 180. What point does it correspond to? The minus 1 point in the
G(s)H(s) plane right.

So, so let us quickly do this example at the example that we considered today. So, let us say

1
we consider G ( s ) H ( s )= .
s ( s+1)

1
G ( jω ) H ( jω )=
jω( jω+1)

So, what is the gain crossover frequency, can you calculate it please? So, how do you get the
gain crossover frequency? Gain crossover frequency, you equate the magnitude to 1 right.

1
2
=1→ ω g √ ω g2 +1 → ωg 2 +ωg 4 =1
ωg √ ωg +1

452
→ ωg =0.7862 rad / s

(Refer Slide Time: 16:52)

So that is what you will have, so that is the gain crossover frequency for this particular open
loop transfer function. So, this is the only frequency at which the magnitude becomes one for
this particular open loop transfer function. Now, how do I calculate the phase crossover
frequency? What is the phase of G(jω)H(jω)? It is going to be 0° −[90° +tan−1 ω]. Do
you agree? I think by now I think all of us should be familiar with these concepts right.

So, the question becomes when will this phase become minus 180 degrees that means,

tan ω pshould be 90 degree right. When will that happen? only when omega tends to
−1

infinity. So, my phase crossover frequency is infinity or in other words this essentially means
at you know like the particular open loop transfer function phase becomes minus 180 degrees
only when omega tends asymptotically to infinity which you can easily figure visualize even
from the bode plot. See think about how the phase plot of this particular factor will look like;
the phase will go only to minus 180 only when omega tends to infinity right so that is why
the phase sorry the phase crossover frequency is infinity ok.

So, we will stop here. So, we will build on these two frequencies in the next class; and then
like I will define parameters which can be used to quantify relative stability and we would
use that ok. We will discuss that in the next class fine.

Thank you.

453
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 65
Relative Stability 2
Part – 1

We were discussing relative stability.

(Refer Slide Time: 00:21)

Closed loop transfer function of the class of systems that we are considering is

Y (s) G ( s)
= .
R (s) 1+G ( s ) H (s )

We discussed the notion of relative stability. And we figured out that the closeness of the
G ( jω ) H ( jω) locus to the -1 point in the G ( s ) H ( s) plane is indicate of relative stability.

We define two terms what was called as a gain crossover frequency and the phase crossover
frequency. The gain crossover frequency ω p is the frequency at which the open loop
sinusoidal transfer function magnitude becomes 1 or in other words |G ( jω ) H ( jω )∨¿ 1.

The phase crossover frequency ω p is the value of frequency at which the phase of the open
loop sinusoidal transfer function becomes−180 °.

454
(Refer Slide Time: 03:16)

We have drawn the G ( s ) H ( s) plane and located the origin, -1 point and drawn the locus of
the transfer function. How can we graphically locate ω g and ω p in this particular plot?

We draw a circle of unit radius and then figure out where this circle cuts the Nyquist plot.
That is ω g. The system becomes unstable if we rotate the entire Nyquist plot by the angle γ.
Or scale the plot such that it passes through -1. This leads to two definitions.

The gain margin K g . It is the reciprocal of the magnitude of the open loop sinusoidal transfer
function G ( jω ) H ( jω) at ω=ω p . If we take the reciprocal and scale the entire open loop
transfer function by the reciprocal, the plot will pass through the -1 point. That is the
motivation behind the definition of the gain margin. It indicates how much margin we have
before the system can become unstable.

1
K g= (decibels)
¿ G ( jω ) H ( jω )∨¿=−20 log 10 ¿ G ( jω ) H ( jω )∨¿ ¿ ¿

Student: (Refer Time: 11:32) omega equal to omega p?

455
(Refer Slide Time: 11:56)

The next parameter which we are going to discuss is what is called as phase marginγ. Phase
margin is the additional amount of phase lag that needs to be added to the phase of the system
at ω=ω gto bring the system to the verge of instability. We calculate the phase of the open
loop transfer function at the gain crossover frequency and subtract it from -180.

γ=∠ G ( j ωg ) H ( j ωg ) −(−180 °)

Note: For a given system whose open loop transfer function does not have poles, zeros in
RHP, both the gain margin in decibel units and the phase margin in degrees should be
positive to ensure the stability of the closed loop system.

Another interpretation of the gain margin and the phase margin using the Bode diagram is
this following.

456
(Refer Slide Time: 18:11)

We draw the Bode diagram of the open loop transfer function and draw the 0 decibel line and
-180 line. We draw the magnitude plot and the phase plot in this manner. We mark the gain
crossover frequency as the frequency at which the magnitude of the open loop transfer
function becomes 1 or 0 decibels. And the phase crossover frequency as the frequency where
the phase becomes−180 °. We obtain the gain margin and phase margin by extrapolating the
plots as shown in the slide. Phase margin is obtained as the phase at gain cross over
frequency and subtracting −180 ° from the phase margin obtained from the Bode plot. Gain
margin is obtained by finding the magnitude at phase crossover frequency and taking the
reciprocal of that.

(Refer Slide Time: 23:40)

457
Let us do an example which we saw previously: The sinusoidal open loop transfer function

1
was . Recall that, ω g =0.7862 and ω p =∞ .
jω( jω+1)

(Refer Slide Time: 24:45)

Phase margin:

−1
γ=180+∠G ( j ω g ) H ( j ωg )=180−90− tan 0.7862=51.83 °

Gain margin:

At ω=ω p=∞, magnitude of this sinusoidal transfer function in decibels tends to −∞.
Therefore gain margin is K g =∞ dB.

That is the calculation of the gain margin and the phase margin.

458
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture - 66
Relative Stability 2
Part - 2

So, now we are going to define a few more terms and all these terms put together define
performance specifications in the frequency domain right. So, let us once again consider the

Y (s) G ( s)
closed loop transfer function right. Consider = . Of course, we assume that
R (s) 1+G ( s ) H (s )
the closed loop system is stable right.

(Refer Slide Time: 00:34)

So, let us say I plot the magnitude plot of the closed loop sinusoidal transfer function ok. See
one important point which I want to highlight here is that gain margin and phase margin were
defined for the closed loop system using the open loop systems frequency response that is
very important, right. Kindly recall that we only looked at the transfer function G(jω)H(jω)
which was the open loop systems sinusoidal transfer function right to comment on some
parameters related to the closed loop system that is something which is critical for us to
remember ok.

459
So, we are done with gain margin and phase margin. I am just moving to two more
definitions of parameters that are important. So, let us say I plot the magnitude plot of the
closed loop frequency response transfer function in decibels. So, let us say the low frequency
asymptote is some constant value and then it goes like this some arbitrary curve right.

Now, what is this value? It is this value at the low frequency asymptote. So, this value I can

Y ( j 0)
call it as the magnitude of . Essentially I am saying that it is the asymptotic value as
R ( j 0)
omega tends to 0. Now what we do is that we take a number which is 3 decibels lower than
this number whatever that number is the low frequency gain some people will call it as the
DC gain or the low frequency gain and so on right.

So, essentially you take a number value of the magnitude which is lower than this number by
3 decibels and find out as to at what frequency ok, the magnitude plot the magnitude of this
closed loop sinusoidal transfer function becomes lower than the low frequency value by 3
decibels. This frequency is what is called as a cutoff frequency ok. So, cutoff frequency is
this frequency. So, let me write down the definition for cutoff frequency.

So, the cutoff frequency ωc is the frequency at which the magnitude of the closed loop
frequency response transfer function ok, the magnitude of Y(jω)R(jω) is 3 decibels below its
zero frequency value ok, zero frequency value is the low frequency asymptotic value that is
that is what it is ok. Please note that the low frequency asymptotic value need not be 0
decibels all the time it is just that for the building blocks we considered it was 0 ok, but it
need not be zero. You would have realized it when we did an example in for the bode
diagram right. The low frequency asymptote was non zero decibels; it was some nonzero
number it can happen no issues at all ok.

So, we are only looking at what is the what to say frequency which is at which the magnitude
is 3 decibels lower than the low frequency asymptote value right. So, with this definition the
frequency range 0 till ωc is called the bandwidth of the system. This is an important
definition.

So, because you can notice from the magnitude curve right, the system allows frequency
components till ωc to pass through it right. After ω c it sort of attenuates all the other
frequencies right. So, that is what happens right. So, the frequency range 0 to ω c is called as

460
the bandwidth of the system. Now one may ask the question hey what is so unique about this
number of minus 3 decibels. What do you think is unique about 3 decibels? You please find
out and tell me ok.

So, why should I take consider 3 decibels as my threshold right; say I like 5 decibels why can
I not define ωc as a number, frequency at which the magnitude is 5 decibels lower than the
low frequency value I am not doing it right. So, I am defining it as 3 decibels lower than the
low frequency value. Why 3 decibels?

Student: [Inaudible]

Say yes, definition I agree I am just asking you what is the physics behind it?

Student: [Inaudible]

Corner frequency? Please find out that is the second question for the day right So, please find
out and tell me right why that three why a number of 3 decibels right or minus 3 decibels in
this case right; so please ok. So, another parameter which we want to essentially define is
what is called as a cutoff rate.

(Refer Slide Time: 08:23)

So cutoff rate it is the slope of the log magnitude curve at the cutoff frequency ok. It
essentially tells me how fast the magnitude is decreasing right. So, that is the cutoff rate and
by the way ideally we want as high a bandwidth as possible ok. So, it does not matter whether

461
I am designing a practical system or I am building a sensor with some dynamic
characteristics or an actuator with some dynamic characteristics, higher bandwidth means
you will have a faster transient response right.

So, that is essentially something which is what to say desirable right, so see because suppose
if I have a high bandwidth actuator right. So that means, that I ask you to let us say move an
object all right then it can do that action much faster right so, than a lower bandwidth
actuator, but at the same time leads to increased cost. So, there is always a tradeoff between a
cost and performance higher bandwidth typically means that better faster transient response
better response characteristics, but then the cost is more ok.

So, that is the price we pay see for example, if you go for actuation systems right if you use
an electromechanical actuator you are going to get a very fast response right so, but then the
cost of an electromechanical actuator is going to be high; complexity is going to be high
right. On the other hand if you use a pneumatic actuator the bandwidth of the pneumatic
actuator is going to be relatively lower, but the cost is also lower right. So, there is a tradeoff
between a cost and performance.

Student: [Inaudible]

We will essentially look at it right. So, how is bandwidth related to response? So, we will ok.
So, that is essentially what to say a thing and so, once again if you use frequency response
methods. So, the performance specifications are typically going to be a given in terms of
these parameters ok, that is gain margin phase margin, phase margin and then like bandwidth,
cutoff rate and then like of course, cutoff frequency ok. These are all parameters and then
resonant peak you remember that Mr right for that second order factors that we looked at right
the resonant frequency all those things are these parameters these are parameters which can
be used.

See in using when you what to say did design using transient response right this using design
using frequency response ok, we are going to learn controllers from tomorrow right. So,
when we did design using transient response what are the parameters that were used to
quantify performance? Rise time, settling time, peak overshoot, peak time and so, on right.

Those are all the parameters that we use right. So, when we will use frequency response we
are going to look at what to say these parameters. The question that arises is whether how are

462
the two related right. So, see how is the bandwidth related to other parameters right? How can
I say bandwidth means faster response? See for example, if I show that higher bandwidth
means a lower rise time would you agree that the system responds faster ok.

So, the there are some relationships we would address them right. So, these are performance
specifications in the frequency domain ok. So, I would stop here and we would continue with
a controller design using frequency response tomorrow, but I am going to leave you with a
question right. So, I hope I we have not discussed this question till now. Let us say right I

1
have this as my system ok, let us say this is my system; plant transfer function is . Is it
s−1
stable or unstable?

Student: Unstable.

Unstable right, the pole is at +1 right. So, one aspect of this course is to essentially stabilize
unstable systems by doing feedback right, but rather than taking all this effect effort right can
I not do this, that is it that is say I pre multiply by a block which essentially does this then
what will happen?

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

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

So, have you not stabilized an unstable system? So, what I have done is what is called as a
pole zero cancellation right. We briefly might have discussed this before, but more
importantly I have done pole zero cancellation in the right half plane right because I am
cancelling a pole in the right half plane with a zero in the right half plane right. So, this solves
all our problems right then I do not need to have expensive closed loop feedback systems and
so, on right.

Student: [Inaudible]

So, I hope everyone got what he was trying to say right. In real systems that were always
going to be some parametric uncertainties like these are all like idealized approaches right.

463
So, you never know like, see for example what essentially it means is that is a following
right.

(Refer Slide Time: 16:31)

1
So, this term can be Y(s) can be right and then yes even if you build a perfect
s−1+ϵ
compensating block ok, this is my compensating block this is my plant, my plant may have
some uncertainties. So, my pole may be around 1 plus or minus epsilon can happen then you
will not have a perfect pole zero cancellation that is very difficult to achieve, right.

And even if you have, if you say epsilon is very small if you take the Inverse Laplace
Transform, you will see that you will have a term like e ϵt , epsilon may be very small, but as t
progresses it will blow off to infinity. So, initially the system may appear to be stable, but
then as time progresses it will become unstable; so not a great idea ok. So, one should never
do pole zero cancellation is in the right half plane for this reason ok. So, that is something
which I wanted to convey as an aside right.

So, anyway we will continue with what to say design of controllers tomorrow based on the
frequency responses.

464
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 67
Lead Compensation
Part – 1

We have been looking at frequency response methods. We looked at stability margins,

relative stability, gain margins, phase margins and so on.

(Refer Slide Time: 00:34)

We will essentially learn how to design a controller, which is called as a lead compensator
using frequency response methods

We will considered this standard feedback loop with unity negative feedback that we have
been dealing with in this particular course. Typically what a lead compensator does is that it
adds a sufficient amount of phase lead to reduce the excessive phase lag if any, that is
associated with the uncompensated system.

We can have a plant which does not have a significantly high phase margin for example. The
lead compensator adds phase to the system or the plant and ensures that the phase margins are
essentially driven to a level which is acceptable. And by and large improves the transient
response. It may also amplify high frequency signals in some cases.

465
(Refer Slide Time: 03:12)

The transfer function of a lead compensator takes the form

1
s+
K α (Ts+1) T
C ( s )= c =Kc ,
(αTs+1) 1
s+
αT

where α is a positive real number between 0 and 1 and T >0 , K c >0.

−1
We can observe that the lead compensator introduces an open loop pole at and open loop
αT

−1
zero at . Let us analyze the sinusoidal transfer function associated with this lead
T
compensators.

K c α ( 1+α T 2 ω2 ) K c α Tω ( 1−α )
C ( jω ) = +j
( 1+α 2 T 2 ω2 ) (1+α 2 T 2 ω2 )

¿ K c α∨ √ T 2 ω2 +1
|C ( jω )|= ,∠ C ¿
√ 1+α 2 T 2 ω2

466
We can see that ∠ C ( jω ) ≥ 0.

1 1
Corner frequencies: ,
T αT

C ( j 0 ) =K c α ,C ( j ∞ ) = K c

At ω=0 and ω=∞, the imaginary part completely vanishes.

We can observe that the sinusoidal transfer function C ( jω ) starts on the positive real axis for
ω=0 and ends on the positive real axis as ω → ∞.

By doing the HW exercise, we can tell that the locus of the sinusoidal transfer function is

Kc
going to be a semi-circle with its center on the positive real axis at [ ( 1+α ) , 0] and having a
2

Kc
radius of (1−α ). (Refer Slide Time: 11:20)
2

467
Let us plot the Nyquist plot for this particular transfer function.

(Refer Slide Time: 12:26)

We can readily observe that for all ω ,the real part is always going to be positive and the
imaginary part is also going to be positive. So the Nyquist plot is going to be in the first
quadrant.

In the C ( s) plane, we can see that, the value of the sinusoidal transfer function starts at K c α
and goes to K c as ω → ∞. We know that it is a semicircle and its center and radius. The plot
looks as shown in the figure.

468
The maximum phase angle ∅m can be obtained by drawing a tangent from the origin to the
semi-circle. ∅m is going to be the maximum phase lead provided by the lead compensator at a
frequency ωm.

Kc
(1−α)
2 (1−α)
sin ∅m= =
Kc (1+α)
(1+α )
2

Kc
(1−α )
2 (1−α)
tan ∅m = =
K c √α 2 √α

(Refer Slide Time: 16:50).

From the phase of the sinusoidal transfer function, ∅m can be written as:

−1 −1
∅m =tan T ωm− tan α T ωm

We can write:

T ωm−αT ωm (1−α)
2 2
=tan ∅m=
1+α T ωm 2 √α

We can get a quadratic equation

469
 2 2
ωm ( 1−α ) α T −ωm [ 2 T √ α ( 1−α ) ] +( 1−α ) =0

1
Solving this equation, we get ωm = , which is the geometric mean of two corner
T √α
frequencies. (Refer Slide Time: 20:03)

470
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 68
Lead Compensation
Part – 2

(Refer Slide Time: 00:14)

This is the expression for the frequency at which we get the maximum phase. So, let us do an
example ok let us essentially take some values for the parameters involved in the lead
compensator. So, you see that in the lead compensator, there are three parameters which have
to be figured out, its Kc, α and T right.

So, those are the three parameters involved in the controlled transfer function. So, let us, let
us say we consider Kc to be 1 ok. We consider alpha to be let us say 0.1 and T to be equal to
1.

So, let us plot the Bode diagram of this controller transfer function C(s) ok. So, the exact
Bode diagram I will leave it to you as homework, but the construction process, because by
now we are familiar with how to construct bode diagrams so, but I am just going to give you
the answer and also discuss some interesting observations from that right.

471
 K c ∝(Ts+1) 0.1( s+1) 1
C ( s )= = = ( 0.1 ) ( s+1)
∝ Ts+1 (0.1 s+1) 0.1 s+1

So, those are the three factors that are involved in this particular transfer functions. So, now,
if I want to plot the log magnitude plot right for the bode diagram.

(Refer Slide Time: 02:12)

So, let us do that right. So, let us plot the log magnitude plot. So, we can immediately observe
that the magnitude of C(jω) if I plot it in decibels. So, the two corner frequencies are 1 and 10
alright.

So, that is something which we can immediately observe right those are the two corner
frequencies and we can immediately observe that the Bode diagram is going to look
something like this ok. I am drawing the final bode diagram; of course, only the asymptotes.

So, we are, we are going to have a contribution of -20 due to the factor 0.1, tell the first
corner frequency of 1, then I we are going to have a straight line with the slope of + 20
decibels per decade between the two corner frequencies. So, as a result I will go from -20
decibels to 0 decibels as long, as for as the log magnitude is concerned.

1
Then after that I will settle down at 0 decibels, because the factor will come in to
0.1 s+1
play right. So, that is going to be their log magnitude curve for this particular transfer
function right. So, once again you know like plotting the three individual, what to say log

472
magnitude curves and adding them, is left to as an homework assignment, but it is pretty
straight forward to see that this will be the final asymptotic log magnitude plot, that is what I
am drawing. I am only drawing the asymptotes, not the actual magnitude plot right.

So, one could immediately observe the following from the log magnitude plot. So, what do
we observe? We immediately observe that this particular lead compensator is going to
essentially attenuate or decrease the magnitude of the low frequency components right,
because we can immediately observe that the low frequency components of the signal are
going to be attenuated by the lead compensator right. So, that is something we can
immediately observe right. The low frequencies are attenuated here.

So, consequently the magnitude of the open loop transfer functions, Please note that the open
loop transfer function is C(s)P(s) right, because we have unity feedback. In general the open
loop transfer function is going to be G(s) H(s). Here H(s)=1 and G(s) = C(s)P(s). So, the open
loop transfer function at low frequencies is decreased ok. So, that is, that something which
happens.

So, this would immediately imply that this can potentially lead to increase in steady state
error right, because please remember, you know like if the low frequency values are reduced
the value of Kpe, Kve and all essentially are effected right, so they go down.

So, as a result you know like the steady state errors may increase right, due to what to say this
particular aspect of the lead compensator. Of course, this depends on the type of the system,
whether we have type 0, type 1, type 2 and so on right, which we have already looked at
right, depending on the type of the system. The steady state errors corresponding to certain
inputs may increase with this particular aspect.

Now, if I want to ensure that you know like that is not the case, then what should I do? What
I should is that, I should push up this magnitude plot. Let us say up by 20 decibels right for
example, all right, considering this particular example. Suppose if I shift this entire log
magnitude curve by 20 decibels by multiplying by a factor of 10 as far as the controller is
concerned, what will happen is that, the low frequency components will now have a
magnitude of 0 decibels as far as the controller transfer function is concerned.

So, consequently the values of Kpe Kve Kae are going to be not affected right. So, the steady
state error characteristics will still be retained as far as the original system is concerned, but

473
what will happen, we immediately see that the high frequency components will now be
magnified right. There will be amplified, because I have to shift the entire bode plot up by 20
decibels.

So, consequently the high frequency components, gets shifted out by also 20 decibels right in
this example. So, consequently the magnitude of the high frequency components will increase
right. So, that is something which can be observed from here.

So, we can immediately note that the second observation is that, if we increase Kc right to
address the first point, then note that the high frequency components would be amplified ok,
that is the flip side of shifting this factor right.

So, if you, if we essentially increase a value of Kc to address this problem we see that the
high frequency components are also amplified right. So, that is something which is an
observation.

(Refer Slide Time: 08:59).

So, in this example note that

1
ωn = = √ 10 rad / s
√ ∝T

1−∝ 0.9
sin φm = = → φ m=54.9°
1+∝ 1.1

474
So, that is the maximum phase which is added by this particular lead compensator ok, the
example that we have considered.

So, if I plot phase plot for this particular example right. So, this, this is also something which
I leave it to you as some work exercise so, but I am just going to plot the final plot. So, if you
plot the phase plot of C(jω) in degrees what we are going to get is, the following. So, 1, this is
10 so on.

A logarithmic scale square root of 10 which is a geometry mean is going to be in the, in

between the middle of 1 and 10 right, and let us say we start from 0 degrees and then let us
say you know like we reach a peak. So, if we plot the phase plot, what is going to happen is?
It is going to look something like this, it is going to start from very low values right and then
as frequency increases ok, it will reach a maximum at square root of 10 and then it will start
to decrease and ultimately it will go to 0.

So, at square root of 10, whatever is the maximum value that we have calculated to be 54.9
degrees ok. So, that is a maximum phase for this particular lead compensator.

So, to summarize a lead compensator, is typically used when the given system or plant does
not have a sufficient phase margin to begin with. So, what we intent to do is, essentially to
add a positive phase to the system, to ensure that we increase the stability margins right, but
what was the trade off here. You know like we immediately saw that the magnitude
characteristics is like a high pass filter.

So, essentially what happen happens is that, like it amplifies high frequency noise, if we a, if
we are not very careful right, but there is a tradeoff once again right in order to not amplify
the high frequency noise. We keep the low frequency components at a high lower magnitude
then the study state error characteristics are going to be affected right, the study state errors
may increase.

So, in order to avoid that, if I push the magnitude plot up, the high frequency components are
going to get amplified right. So, there is a tradeoff here, but the, but the good thing is, it will
essentially give us a positive phase lead which will increase the phase margin of the system.

475
So, this is the theory behind lead compensator. In the next class we will essentially do a
problem, where we will start with the given system and we will see how to design a lead
compensator for a given set of performance, specification, requirements right.

We will go through all the steps and then like calculate the values of Kc, α and T to meet the
given performance requirements ok. So, that is that is all with today's class, you know like we
will earn and example of lead compensation in the next class.

Thank you.

476
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 69
Lead Compensator Design
Part – 1

We are going to look at an example of how to design a Lead Compensator.

(Refer Slide Time: 00:26)

4
Consider a system or a plant whose transfer function is P ( s ) = . Design a lead
s( s+2)
compensator such that the static velocity error constant K ve =20 and the phase margin γ is at
least 50 degrees and the gain margin is at least 10 decibels. Consider unity negative feedback.

The plant has poles at 0 and -2. We call it either marginally stable or unstable also. We want
to stabilize the close loop system and at the same time ensure that some performance
specifications are met. The closed loop feedback path is shown in the figure.

477
(Refer Slide Time: 04:22)

4
The plant transfer function: P ( s ) =
s( s+2)

H ( s ) =1

K c α (Ts+1)
Lead compensator transfer function: C ( s )=
(αTs+1)

The task for us is to find the values of K c,α and T that would stabilize the close loop system
while meeting the above performance requirements.

K c α (Ts+1) 4
G ( s ) =C ( s ) P ( s ) =
(αTs+1) s( s+2)

Let K c α= K and P1 ( s ) =KP( s)

( Ts+1 )
G ( s )= P (s)
( αTs+1 ) 1

478
(Refer Slide Time: 08:16)

The step first step is to adjust K such that the desired value of K ve is obtained. K ve was

defined as: lim

s →0
sG ( s ) H (s)

We get lim
s →0
sG ( s ) H (s)=2 K ,

K ve=2 K =2 0, K = K c α=10

40
P1 ( s ) =KP ( s ) =
s( s+2)

( Ts+1 ) 40
G ( s )=
( αTs+1 ) s (s +2)

Now, we are given that the phase margin should be at least 50°. We have to calculate 180
degrees plus the phase of the open loop transfer function and the gain cross over frequency,

γ=180+∠G ( j ωg ) H ( j ω g )

In this case H ( s ) =1. So we considering only G (s). The phase of the open loop transfer

( Ts+1 )
function comes from two components inG (s). The first component (factor 1) is .
( αTs+1 )

479
 40 ( Ts+1 )
And the second component (factor 2) is . The factor would provide a
s ( s+2) ( αTs+1 )
positive phase to the open loop transfer function which will essentially it is phase margin. But
the question is that by how much we should increase the phase margin.

For that I should figure out the phase margin of the uncompensated system. If we calculate

40
what will be the phase margin of , we can know how much additional positive phase
s ( s+2)
we need to provide by using the first factor. The second step is to find α.

(Refer Slide Time: 16:08)

We have to find the gain cross over frequency for factor 2. By definition,

40
| |
j ωg ( j ω g +1)
=1 .

ω g =6.17 rad/s.

The phase of factor 2 is

ωg
∠ P1 ( j ω g ) =−90 °−tan−1 =−162.04 °
2

480
Phase margin = 180 ° +( −162.04 ° ) =17.96 °

The uncompensated system does not meet the phase margin specification.

Does uncompensated system satisfy the gain margin specification? The gain margin be at
least 10 decibels. We can see that the phase crossover frequency for factor 2 is infinity. So,
the gain margin is going to be infinity decibels. It does meet the gain margin specification.

The factor gives a phase margin around 18°. To make the phase margin 50°, factor 1 should
contribute 32°. But there is a catch here. Let us let us go back and look at the Bode diagram
that we constructed earlier.

(Refer Slide Time: 26:33)

( Ts+1 )
If we just consider , the magnitude curve is going to start at 0 decibels, it is going to
( αTs+1 )
increase and then settle down at 20 decibels. We are going to get the maximum phase at the
geometric mean of the two corner frequencies.

The factor 1 would have already added some magnitude in positive decibels. The open loop
transfer function which is now the product of factors 1 and 2 will essentially be influenced by
this magnitude of 1 and consequently gain crossover frequency would increase. For the
uncompensated system, the gain crossover frequency was 6.17 rad/s Now, with
compensation, the magnitude of G ( jω) is going to be the magnitude of factor 1 plus
magnitude of factor 2.

481
Factor 2 has a gain crossover frequency of 6.17 rad/s, but at that frequency the magnitude of
1 is not 0. It is going to be a positive value. Therefore the magnitude of the overall transfer
function is not 0. So 6.17 rad/s cannot be the gain crossover frequency of the overall system.
The value of the gain crossover frequency would increase as compared to factor 1 alone.

(Refer Slide Time: 29:04)

If the value of gain cross over frequency increases, the phase of this factor 2 will become
more negative at the increased gain cross over frequency. Although we may want to provide
a 50 ° phase margin, the effective phase margin would be lower because of the shift in gain
crossover frequency. If we just keep the phase contribution of the lead compensated to be 32 °
, we will not be in a position to meet the phase margin because it will be less than 50 °. But
as a rule of thumb, we say that the uncompensated system does not meet the phase margin
specification. The lead compensator should provide a phase of 50 °−17.96° =32.04 °, that is
what we can observe at first glance.

482
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 70
Lead Compensator Design
Part – 2

(Refer Slide Time: 00:14)

With compensation, the magnitude of G(jω) is going to be magnitude of 1 plus magnitude of

2 in decibels right. So, with compensation consequently the value of the gain cross over
frequency would increase ok. Hence, the phase of 2 would become more negative at the new
gain cross over frequency right.

Thus, what we do is that in order to avoid any conflict, you know like, because the phase
margin may decrease below 50 degrees due to this phenomena. Thus, we would design the
lead compensator to provide an additional phase of 5 degrees to 6 degrees to, account for this
effect ok. So, that is what we are going to do ok, because the phase of the second factor is
going to become more negative.

So, if I still keep the phase contribution of 1 to be 32 degrees what will happen is that the sum
will fall below 50 degrees right; so, that is what is going to happen. So, what we, we do not
know the new gain cross over frequency yet, so, but as a rule of thumb what we do is that we
typically add 5 to 6 degrees to the number, which we have calculated just now.

483
So, instead of 32, we design the, gain sorry, lead compensator to provide a phase of around,
let us say 38 degrees ok; so, that is what we typically do ok. I, I hope it is clear, what
essentially we are doing.

(Refer Slide Time: 02:37)

Hence, let us design the lead compensator, such that it provides a maximum phase of φ m, you
remember what φ m was right, φ m was the maximum phase contribution for the for the lead
compensator factor.

So, of 38 degrees, let us say we add 6 degrees to 32 degrees. Let us say we know, we say that
the lead compensator, let it provide a maximum phase of 38 degrees that is what we are going
to design for. So, this implies that φ m is 38 this implies that what was sin ⁡φm?

1−α
φ m=38 ° → sin φ m= → α=0.238
1+α

So, we essentially calculate what was the existing phase margin without compensation then
we find out what is the additional phase that should be provided to the lead compensator by
looking at the required phase margin and then what we do is that we add a correction term,
because the gain cross over frequency is going to be shifted to the right with compensation
alright.

484
So, that is why we just added a correction term. Is it clear? Now, let us go to step 3, step 3 is
to find, the value of T. So, in a sense we have, we have found alpha, if you have found alpha
have we found Kc.

Student: Yes

Yes. So, if alpha is 0.238 immediately you note that K = K c ∝. What was K?

Student: 10.

10.

Student: 10.

Right. So, this implies that K c is going to be 10 divided by alpha. So, immediately you will
see that, that value comes to around 42.02. So, what were the three parameters, that are
involved in the lead compensator Kc, alpha and T. So, we have found K c and alpha. What is
left behind?

Student: T

T. So, let us find T. That is what we are going to do. So, let us say find T ok.

1
Recall that ωm =
√α T

Now, |1+1+jωαT
jωT
| ω=
1
√αT
=
1
√∝
So, this expression I leave you to you as homework. The magnitude of 1 plus j omega T
divided by 1 plus j alpha omega t at omega equals 1 divided by square root of alpha times T
that is going to equal to 1 by square root of alpha ok.

So, that is something I leave it to you as homework ok, pretty straight forward right.

485
(Refer Slide Time: 06:31)

So, what is this magnitude I am calculating? I am calculating the magnitude of the sinusoidal

1+sT
transfer function corresponding to that is what I am doing right, that is this ok.
1+αsT

So, this I leave to you as homework pretty straight forward you just substitute the, what to
say, the frequency at which you get the maximum phase right and please note that this should
be our new gain cross over frequency, why? Because our, what to say lead compensator is
going to provide the maximum phase at omega equals ωm and that is what I want as the new
gain cross over frequency would I not? So, we want, we want ωm to be the new gain cross
over frequency.

Ts+1
( ∝Ts+1
G ( s )= )( s ( s40+2) )
G ( jω )= ( ∝jωT +1
) 40
Tjω+1 jω ( jω+2 ) )
(
So, we want ωm to be the new gain crossover frequency ok. So, let us say we call, this to be
ω gn. Now, at the new gain cross over frequency ωm the value of 1 over, what is it? This
particular factor is going to be equal to this value is going to be equal to 1 divided by square
root of alpha, which is essentially calculated as 2.05 right.

486
So, thus this implies that at the new gain cross over frequency ω gn,

40 1
| | =
j ωgn ( j ωgn +1) 2.05
→ ωgn =8.9456 rad / s

Please, check it out ok. So, once again this I leave it to as one intermediate homework ok,
pretty straight forward, we already done for two cases. So, please do that. So, that is it yeah.

Student: Sir, that sir at least 50 degrees sir.

Student: And working out we should use the equality or based on.

So, we are just, we are just adding a factor of safety and we are essentially dealing with
equalities yes ok. You are right, it says at least 50 degrees, if you want you can work with
inequalities. So, in order to get specific values, you know like for design purpose. You know
like we just, what to say put some additional buffer and then we are working with equalities
ok. Finally, we will come back and check that I am going to leave it you as homework.

So, we you have to go back and check whether it is satisfying the performance requirements
right, So, we are almost there right.

(Refer Slide Time: 11:33)

1
So, now, note that we want ωm =ω gn. What is ωm? ωm = =8.9456→ T =0.229
√∝ T

So, that is it we are done right. So, this implies that the controller transfer function is

487
 K c ∝(Ts+1) 10 (0.229 s+1)
C ( s )= =
(∝ Ts+1) (0.0545 s +1)

So, this is the controller transfer function. So, then the open loop transfer function of the
compensated closed loop system is,

10(0.229 s +1) 4
G ( s ) H ( s )=C ( s ) P ( s )=
(0.0545 s+1) s (s +2)

(Refer Slide Time: 14:19)

So, as homework check that the above design satisfies the prescribed performance
requirements ok, so that is the check which I want you to do. So, you go back and check
whether Kv= 20 right correct. So, you can immediately see that Kv=20.

So, check whether gain margin is at least 50 degrees and phase margin is at least 10 decibels
please, do that as, homework ok. You will see that, the design will be as specified right. So,
but you can make one more observation right. Please note that the value of ω gn has increased,
which would mean that the band width of the system has increased right. So, but then what
was your observation if the band width increased.

Student: [Inaudible]

So, note that ω gnincreased implies the bandwidth as increased with compensation ok, but
what is the price we need to pay.

488
Student: [Inaudible]

So, the price was that, if you go back and think about it, but we needed to increase the open
loop gain ok, open loop. What do I mean by open loop gain is, if you substitute s equals, what
you say you take the steady state value ok., sorry, you take Kve right.

So, you initially, we had it as two for the plant right. We had to increase it to 20 to
essentially, ensure that the low frequency components were not attenuated. So, that that is
something which, which we discussed yesterday right, because with a, if you recall, if you go
back to yesterdays, class. We saw that one, limitation with the lead compensator was that the
low frequency components where attenuated right.

So, you see that the low frequency components where attenuated to. In fact, what to say,
essentially address this we need to push this curve up. So, that is why we had an extra factor
of 10. You remember, we had a gain K, you know like; so, which was 10 right.

So, we needed to introduce I would say, we needed to yeah introduce a, a high open loop gain
ok, to ensure that low frequency components were not attenuated, but what is the flip side.

So, that is why we calculated K to be 10, if you recall right, the first step right. So, this has
two effects. One is cost increases when you increase the gain of the system. Second thing is
also the high frequency noise gets slightly amplified right. So, that is that is something which
we need to remember right.

So, I just want to reiterate those two points. Once again right, through this example fine. So,
this is an example which essentially shows you the process for designing a, lead compensator.
We will stop here for, this class and when we come back, what we are going to do is that like,
the, final theory class. What we are going to learn, is about, lag compensator and lag lead
compensator ok. I am just going to tell you what is the, purpose and structure of those
compensators and then we will also see you know like, when to use them right, then we will
recap the entire course once in the next class fine.

Thank you.

489
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 71
Lag and Lag-Lead Compensation
Part - 1

We looked at Lead Compensation. In this lecture, we are going to look at lag compensation,
its uses, structure, features, effects. We will also learn lag lead compensator.

(Refer Slide Time: 00:56)

Let us look at lag compensation. Once again we consider unity negative feedback. The same
structure that we adopted when we discussed lead compensation is taken here. The question
becomes when do we use a lag compensator? Lag compensator is typically used to attenuate
or essentially reduce the magnitude of high frequency components of the systems response.

As the name indicates it introduces a phase lag, but by and large when you use a lag
compensator, we are not unduly worried about the phase margins. So, the base system itself
will have enough phase margin that the addition of a phase lag will not make it unstable.

The controller transfer function, or the transfer function corresponding to a lag compensator
is

490
 1
s+
K β (Ts+1) T
C ( s )= c =Kc ,
( βTs+1) 1
s+
βT

β >1 , T >0 , K c >0.

(Refer Slide Time: 04:01)

−1 −1
This introduces an open loop pole at minus and an open loop zero at .
βT T

1 1
Corner frequencies: ,
T βT

So, now let us look at the sinusoidal transfer function for this lag compensators.

K c β ( 1+β T 2 ω2 ) K c βTω ( β−1 )

C ( jω ) = +j
( 1+β 2 T 2 ω2 ) ( 1+ β 2 T 2 ω2 )

C ( j 0 ) =K c β ,C ( j ∞ ) = K c

491
 (Refer Slide Time: 06:48)

Let us plot the Nyquist plot. If we look at the real component and the imaginary component,
for all ω we can see that the real component is going to be positive. And the imaginary
component is going to be non-positive. The Nyquist plot les in the fourth quadrant. The locus

Kc K
is going to be in the form of a semi-circle with center at at [ ( β+1 ) , 0] and radius c ( β−1)
2 2
.

We shall have a few exercise problems pretty similar to what we did in the lead compensator.

1. Take K c =1 , β=10 ,T =1and plot Bode diagram.

10( s+1) 1
C ( s )= =(10)( s+1)
(10 s+1) 10 s+1

(Refer Slide Time: 11:54)

492
We are going to take the answers straight away so that we discuss it. This curve will start at
20 decibels. The corner frequencies are 0.1 and 1. Initially we will have 20 decibels till we
reach 0.1. Then we have a down slope at -20 decibels per decade till we reach the other
corner frequency, after which it is a horizontal line. We can see that the lag compensator by
and large acts like a low pass filter. Plotting the phase diagram is left as an exercise.

The low frequency gains are going to be increased, right. If we multiply with the plan transfer
function, at low frequencies, we are adding 20 decibels, which is like multiplying the gain by
10.

The steady state errors will decrease. Because the low frequency gains will increase, which
will reduce the steady state errors.

Increasing the low frequency gain means, the cost of the system also increases. But then if we
want to attenuate the high frequency components, we should push the curve down by 20
decibels. This would also decrease the gain crossover frequency. The gain cross over
frequency of the combined system will decrease. This will mean reduced bandwidth. A low
pass filter is supposed to attenuate the high frequency components. But the flip side is that the
bandwidth decreases. Similarly lead compensator acts as a high pass filter.

We will choose the corner frequencies in a frequency range of interest that depends on the
plant. The corner frequencies are chosen in a way that it will be around the original gain cross
over frequency of the system. (Refer Slide Time: 20:59)

493
If you want to combine, the advantages of both a lag compensator and a lead compensator,
while addressing their limitation, we use a lag lead compensator.

A lead compensator improves the stability margins but may decrease a steady state accuracy.
On the other hand, a lag compensator attenuates the high frequency components but
decreases the bandwidth.

To the combine the benefits of both these compensators, we use a lag lead compensator. The
Nyquist plot and bode plot of the lag lead compensator are left as home work problem.

The transfer function of lag lead compensator is given by

1 1
s+ s+
T1 T2
C ( s )=K c
γ 1
s+ s+
T1 β T2

γ >1 , β >1 , T 1 >0 ,T 2 >0 , K c >0

(Refer Slide Time: 23:48)

We can immediately see that the first fraction is the lead compensator and second fraction
corresponds to the lag compensator. The complexity of the controller increases because we
are having two open loop zeros, and two open loop poles being introduced by the controller
transfer function.

494
Corner frequencies for Lead compensator:

1 γ
,
T1 T1

Corner frequencies for Lag compensator

1 1
,
β T2 T2

If we chooseγ= β, the range is going to be pretty much just equal in the logarithm scale. As
long as we design T 1 and T 2 carefully

495
 Control Systems
Prof. C. S. Shankar Ram
Department of Engineering Design
Indian Institute of Technology, Madras

Lecture – 72
Lag and Lag - Lead Compensation
Part - 2

(Refer Slide Time: 00:29)

So, frequentlyγ= β. So, what I am going to do is that I am going to give you a homework
problem ok where you essentially draw the Nyquist plot and Bode plot of this sinusoidal
transfer function, and then comment. So, let us say, we choose the following values, we chose
Kc=1, and γ=10, β=10, and let us say T1 =1, and T2 = 10 ok.

So, what we are doing is that like I am I am choosing 1/T1 and 1/T2 to be 1 decade away right,
so that is what I am doing right. And then like the lead compensator will essentially have a
slope as far as the magnitude curve is concerned one decade from 1/T 1. And similarly, the lag
compensator will have a frequency range of interest 1 decade below 1/T 2 ok, so that is why
that is how we just split them as three equal components ok, I will give you the answer you
just solve it right.

So, let us say you essentially choose these values right, so what you need do is that plot the
Nyquist plot, and the Bode plot of the of C ( jω) ok, so that is your homework. So, I am just

496
going to draw the Bode plot, the log magnitude curve to be more exact, so that we will
visualize how this going to look like right.

So, let me draw the magnitude of C ( jω), of course, in decibels ok. So, what is what are the
critical frequencies? We are going to have 0.01 K will have. So, since 1/ꞵT 2 will be 0.01 that
is why I am taking from 0.01 right, 1/T 2 will be 0.1 ok, then I have 1, then I have 10 right. So,
these are the four frequencies right. So, what I am doing is that, I am just calculating the
values of 1/T1, γ/T1, 1/ꞵT2, and 1/T2 right that is what I have marked for this particular
example. So, what is 0.01 for this particular example, how do you get 0.01?

This is nothing but 1/ꞵT2 right, 0.1 is essentially 1/T2, this is 1/T1; this is gamma by T 1. I
hope now it is clear, how I got these four values ok, these are the two four frequencies. And
the way, we are chosen these numbers, we have ensured that that the gaps in the logarithmic
scale are the same that is just done by convention, you know that is it ok.

So, I am just going to give you the answer, and then discuss. So, what will happen in this
particular case is that the asymptotic plot, obviously, it will start from 0 ok, then it will reduce
to -20, then stay at -20, and then it will go back to 0 ok. So, this is what happens with this
with the log magnitude curve of this particular transfer function the lag lead transfer function
right.

(Refer Slide Time: 05:06)

497
So, you can immediately see that the lag part kicks in first right. In the lag lead compensator,
the lag part kicks in first right, because you would you can immediately correlate it to
whatever we learnt right. The lag compensators log magnitude plot is going to have a
structure like this right. So, it is going to stay constant decrease, and then stay constant right,
so that is what you are going to have in the first part, due to the lag part. And then the lead
part kicks in, and then the magnitude value increases, and then becomes constant ok, so that
is the structure for this lag lead compensator right. Yes please.

Student: [Inaudible]

Yeah, it depends on the value chosen exactly right. So, if you look at it, what is the, what
can you say about the nature of this particular compensator what does it do.

Student: [Inaudible]

Yeah, it takes an intermediate frequency range right, and then attenuates that frequency range
alone right, so that is structure, which is taken by this lag lead compensator right. So, you can
immediately see that that is what is going to happen right.

So, please plot the Nyquist plot, and also the phase plot ok, corresponding to this particular
transfer function ok. And just make observations as before right, and that is something, which
I am going to leave it you as an exercise or a home work, is it clear fine ok.

(Refer Slide Time: 07:02)

498
So, this completes our discussion as far as the theory part for this course on control system
is concerned. So, what I am going to do is that like, I am just going to summarize whatever
we learnt in this course, so that we get the big picture view point right.

So, and then you can also help in summarizing what we really learnt in this in these
lectures, and how we will use them right. So, let us let us do that. So, let us summarize all the
topics that we learnt right. So, where did we start if you recall?

Student: Stability.

Sorry.

Student: [Inaudible]

Stability, see think about from the first class right.

Student: [Inaudible]

Ok. So, I am just what you say essentially giving as bullet point’s right. So, we first started
with what is the system drawing, what is the dynamic system correct. So, and then like what
are the classes of dynamic systems based on whether it is linear, or non-linear, what is the
definition of time invariance, causality, and so on right, single input, single output systems.
And all those classifications, were first learnt right, so that is something, which we looked at
ok. Then what did we do?

Student: Laplace transforms.

Laplace transforms ok, before that?

Student: Properties.

Properties of what?

Student: Systems.

Systems, of course, yeah. So, we looked at the classification, we also identified the class of
systems that we are going to study right. So, now everything whatever we learnt, you know
as far as this particular course is concerned, is applied to what class of systems?

499
Student: [Inaudible]

Yeah. So, if you recall, we identify the class of systems to be studied as linear time invariant,
causal, single input, single output.

Student: Dynamic.

Dynamic systems right, so that is the class of systems that we have analyzed throughout this
particular course right.

So, then what did we did, we looked at characterization of dynamic systems using
mathematical models right, so that is what we learnt next right, how one can use essentially
models to characterize.

And then we also learnt, what was that what is that typical class of models that we will use. If
you remember, lump parameter or spatially homogenous, dynamic continuous time right,
deterministic models right that is the class of models, we essentially use. So, ultimately what
we figured out as that like this essentially comes in the form of ordinary differential equations
right.

So, consequently what we did, we essentially revised a few important mathematical tools
particularly related to complex variables, and Laplace transform ok. We learnt how Laplace
transform can be used to solve ODEs right. And since, for this particular class of systems, we
will we will get linear ODEs with constant coefficient, there is only one solution, which is the
exponential function right, so that is why exponential functions are pretty important. Then
what did we learn?

Student: Laplace transform.

Sorry.

Student: Laplace transform.

Yeah of course, by Laplace transform, I refer to all it is properties examples that we did and
so on right. So, then we looked at of course, we also did one more thing right. So, of course,
in this is not in the same order, but then we learnt how to essentially represent the model,
which is given in terms of ODEs in two representations right, the transform function
representation, and the state space representation.

500
Of course, throughout this course, we looked mainly at the transfer function representation.
Although we had a brief discussion on what is the state space representation, how can it be
obtained, and what is the equivalence between two right that is something, which we
discussed.

Ah But we took the transfer function representation, then what did we do, we learnt about
poles, zeros right, and then we also looked use the Laplace transform to look at free response
right, forced response.

And we learnt that the transfer function is typically used to characterize the force response
right, free response is essentially removed, because the initial conditions are taken to be 0.
But, is that viable approach because in reality we may have a non-zero initial condition right.

But, then by now you would have realize that if we design stable systems, the coefficient that
come in the solution, that depend on the initial conditions will be multiplied by exponential
functions with the decaying, or a with a negative exponent.

So, what really happens is that if you consider the free response of a stable system that
anyway decays to 0, and what persists the steady state value, and also the steady state
response and even the transience to a large extent would be influenced by the forced response
by the input that we get right.

The free response would affect the input, if the initial conditions are non-zero, but the
argument is just typically given is that like look any way the contribution of the free response
is any way going to go to 0 right if you design a stable system, that is why we are looking at
the forced response predominantly right. Then what did we learn? We learnt two important
topics right that of.

Student: Stability.

Stability, and performance right.

So, we learnt the definition of BIBO stability, and course by stability , then we added an
adjective there, we learn something called relative stability. So, people call this as absolute
stability right. So, we learnt, what is called as a BIBO stability, what is the notion.

501
And then like we learnt stability criteria, we figured out that for BIBO stability, a necessary
and sufficient condition was for all the poles to lie in the left of complex plane right. And
then like we learnt the Routh’s stability criteria to figure out you know, like how 1 could
evaluate whether a given system is stable or not or conversely, how could one design
controllers to ensure close loop stability right, so that is something we leant correct.

Then we also looked at performance, how did we quantify performance, if you recall, we
looked at 1st order systems right, 2nd order systems. And then like looked at various
parameters like time constant, rise time, peak time, over shoot and so on right, settling time.
So, all those parameters, where used to essentially quantify performance, correct.

So, then we looked at predominantly PID controllers right. So, we learned the structures,
what is the structure of the PID controller, what was the a contribution of each part
proportional, integral, and derivative, what are the pros and cons of each component in the
PID control control structure and essentially we did a few examples right and we learnt,
when we got an intuitive feel of when to use what right, because when we have poles on
poles at the origin or poles on the imaginary axis for the plan, what to do and when we have
poles in the right of plane, which controller would suffice and so on right. So, those things we
figure out by doing a few examples.

(Refer Slide Time: 15:46)

Then if you recall, we studied about the root locus right. So, root locus essentially was used
to plot the locus of the closed loop poles, and then this was used for control design correct.

502
So, if I if I am given the open loop transfer function, if I vary one parameter, which may be a
controller parameter, how do the locus of the closed loop pole change, and that was used to
essentially design the closed loop system for stability and performance right. So, after this we
moved to frequency response.

Student: I think you missed the steady state steady state error.

Yeah fine that is right, so where would I put it, let me put it under performance ok, because
any way thank you. So, I think we yeah I missed out writing steady state error analysis right.
We learnt various definitions of definitions of various parameters Kpe, Kve, Kea and so on
right, and when to use what. And we looked at different types of system, type 0, type 1, type
2 and so on right and what is the implication of those systems.

We looked at frequency response, then we learnt what it was, and then how it can be
visualized using Bode plot, Nyquist plot right. So, and then we learned the Nyquist stability
criteria and so on ok, so that that was another thing.

(Refer Slide Time: 17:40)

And finally, what we did was that like we essentially looked at relative stability right. We
looked at various parameters that can be used for quantifying performance in the frequency
domain like gain margin, phase margin, and so on right. We then, we looked at cut off
frequency, bandwidth and so on right, so that is something, which we did.

503
And finally, of course, we looked at lead, lag, and lag-lead compensation right, so that is
another just choice of controller transfer functions that you have right, depending on the type
of system that you have you choose either. And I hope that this course essentially also serves
as a starting point right. If you want to essentially learn, advance control theory, this
whatever we learnt in this particular course would be the fundamentals based on, which you
would build right, so that is the expectation of this particular course.

So, if you go back, and look at what were the learning outcome outcomes of the course, you
know that we discussed in the very first class, essentially that is been listed under the
summary right. So, we have learned what is called as classical control theory right, for a
particular group of systems fine. So that is about it as far as the theory is concerned. So, I
thank you for your attention.

504
 THIS BOOK
IS NOT FOR
SALE
NOR COMMERCIAL USE

(044) 2257 5905/08

nptel.ac.in
swayam.gov.in