Control Theory

Dhruva Hegde

1 Basics of control systems

What is a control system?
A control system is a meaningful interconnection of devices that produces
desirable output with respect to a given input, which has the ability to mod-
ify its behaviour with respect to requirements.

Measures of a control system

1. Speed : Transient analysis

2. Accuracy : Steady state analysis

3. Stability : Various time domain and frequency domain techniques

Open Loop Control System
Control action does not depend on actual output. It is basically just a direct
function mapping value of input to output.

While designing an open loop system is easier and cheaper, it is not always
reliable and is more prone to disturbances.

Closed Loop Control System

Control action depends on the actual output. This is achieved using the
concept of feedback.

A closed loop system is more difficult to design and can be expensive, but it
is more robust.

1
Feedback is the process of measuring the actual output of the system and
feeding it back to the system for better performance.

For Control Systems, Negative Feedback is used. In negative feedback,

the actual output signal is subtracted from the input signal to obtain the dif-
ference (or error). This will give information about how far the actual output
is when compared to the desired output, using which the control system can
adjust its behaviour.

It is worth noting that Positive Feedback is where the output is added to

the input to produce a sum term, which is not very useful in control theory
as it causes instability. More will be discussed about this later.

1.1 Control Systems terminology

Plant: The system that is to be controlled.
Actuator: The device that drives the plant to provide required output.
Controller: The device that uses input and feedback signal to control the
actuator in such a way that desired output is produced.
Sensor: The device that measures the actual output and feeds it back.

X E U P Y
− Controller Actuator P lant

Yf
Sensor

The input signal, X is just the desired output value.

The difference between input signal and actual output measured by the sen-
sor (also called feedback signal) is called error signal, E.
The controller provides the control signal, U in order to control the actuator.
The actuator provides the actuating/power signal, P to the plant to give
actual output Y .

For simplicity, all the 3 blocks in the forward path are together represented
as G and the feedback block is represented as H (unless specified otherwise).

2
1.2 Impulse Response and Transfer Function
Impulse Response of a system is defined as the output of the system when
unit impulse function is provided as input.

The impulse response model of a control system is defined under relaxed

conditions i.e under the assumption that the initial states of the system are
0 (meaning the system has no initial condition).
This model will be able to provide the output of the system for any given in-
put. Hence, the impulse response model is basically an input-output relation.

If r(t) is the given input such that r(t) = 0 ∀ t < 0 and the impulse response
of the system is g(t), then the output y(t) is given by,
Z t
y(t) = r(τ )h(t − τ )dτ = r(t) ∗ g(t)
τ =0

The Laplace transform of the impulse response of a system is called its Trans-
fer Function.
By taking the Laplace transform on the above equation,

Y (s) = R(s)G(s)
Y (s)
=⇒ G(s) =
R(s)

Transfer Function of a system is defined as the ratio of the Laplace Trans-

form of the output to the Laplace Transform of the input, computed under
zero initial conditions.

From the definition, it can be noted that for transfer function to exist, the
system must be a Linear and Time-Invariant (LTI) System and the
input signal must be a Causal Signal.

The differential equation model is obtained by using basic laws such as New-
ton’s Laws and Kirchhoff’s Laws to obtain equations that describe how the
components interact with each other when connected in specific ways. This
model can be used to find the transfer function as well.

3
If a system input-output relation is defined by the differential equation,

d2 y(t) dy(t)
2
+ a1 + a2 y(t) = bo r(t)
dt dt
then by taken Laplace transform on both sides and manipulating,

Y (s) bo
= 2
R(s) s + a1 s + a2
which is its Transfer function.

Note that the impulse response or the differential equation represent the
behaviour of the system in time domain whereas the transfer function repre-
sents the behaviour of the system in frequency domain.
n(s)
The transfer function is expressed as ratio of 2 polynomials. G(s) = d(s)

Poles of Transfer Function:

The poles are the values of s at which the transfer function tends to infinity.
This corresponds to system being able to give an output when input is zero.
=⇒ poles are solutions to the equation : d(s) = 0

Zeros of Transfer Function:

The zeros are the values of s at which the transfer function becomes zero.
This corresponds to system not giving any output despite non-zero input.
=⇒ zeros are solutions to the equation : n(s) = 0

Open Loop Transfer Function:

OLTF is the transfer function of the system prior to providing feedback.

Closed Loop Transfer Function:

CLTF is the transfer function obtained after simplifying the system with
feedback.

Characteristic Equation:
The denominator polynomial of the CLTF equated to 0 gives the character-
istic equation of the system.
The denominator polynomial that determines the behaviour of the system,

4
while the numerator polynomial only determines amplitudes.

Type of a system:
The number of poles at the origin in the OLTF of a system is called its type.

Order of a system:
The highest degree of ’s’ in the characteristic equation (or denominator poly-
nomial of CLTF) is called its order.

Examples:
s+1
G(s) =
s(s + 5)
=⇒ Type 1, Order 2

2
G(s) =
(s3 − s + 6)
=⇒ Type 0, Order 3

A transfer function can be decomposed such that it can be expressed in

Pole-Zero form as follows.
(s + zo )(s + z1 )....(s + zn )
G(s) =
(s + po )(s + p1 )....(s + pm )
If the highest power of the denominator polynomial is greater than or equal
to the highest power of the numerator polynomial, then the transfer function
is ”Proper”, otherwise it is ”Improper”.

Improper transfer functions are not physically realizable because a pure dif-
ferentiator can’t be physically realized.

If the highest power of the denominator polynomial is greater than the high-
est power of the numerator polynomial, then the transfer function is ”Strictly
Proper”.

The value of the transfer function at s = 0 is called the DC gain of the system.

5
Loading Effect on Transfer Function:

Note that while deriving transfer function, it is assumed that there is no

loading effect i.e no power is drawn at the output of the system.
This assumption must be satisfied even while deriving transfer functions for
each component in a control system. If one component is acting as a load on
another component, then:

• Transfer function of each component cannot be determined separately.

• Transfer function of both components combined should be determined.

• Both components are put in the same block in the representation.

This problem is commonly encountered in cascaded electrical circuits where

impedance matching has not been achieved. The overall transfer function
of the cascaded circuit will not be equal to the product of the individual
transfer functions. This is why impedance matching is extremely important
in circuit design.

6
2 Principles of Feedback
As mentioned earlier, feedback is the process of measuring the actual output
of the plant using a sensor and comparing it with the input (desired output)
and controlling the plant accordingly.

A standard negative feedback system with open loop transfer function G(s)
and feedback element H(s) with input R(s) and output Y (s) is represented
as,
(each of the blocks and signals are expressed in terms of Laplace variable)

R(s) E(s) Y (s)

D(s) G(s)
−

Yf (s)

H(s)

It can be algebraically derived that the overall transfer function of the system
will be
Y (s) D(s)G(s)
T (s) = =
R(s) 1 + D(s)G(s)H(s)

This will be the closed loop transfer function of the system.

Usually, H(s) is taken to be unity and D(s)G(s) together is considered to be

G(s)
G(s). Then the transfer function will be reduced to, T (s) = 1+G(s) .
n(s)
In this case, if the OLTF is expressed as G(s) = d(s)
, then the CLTF will be
n(s)
G(s) = d(s)+n(s)
.

2.1 Block Diagram algebra

The most straightforward way to represent a control system is by using func-
tional block diagrams.
Each block represents the transfer function of the particular subsystem.
The overall transfer function of the system can be found by using block
diagram reduction techniques.

7
• Blocks in cascade

• Blocks in parallel

• Moving take-off point present before a block

• Moving take-off point present after block

• Moving summing point present after block

8
 • Moving summing point present before block

• Feedback loop

Also note that neighbouring take-off points can be interchanged and neigh-
bouring summing points can be interchanged.

The block diagram technique is intuitive and can be used to solve simple sys-
tems. However, if the system becomes increasingly complex (i.e more blocks
and connections), then the process of simplification using block diagram re-
duction becomes cumbersome.

2.2 Signal Flow Graph

The Signal Flow Graph is a graphical representation of a system, consisting
of nodes and branches.
It provides a simpler way to obtain the overall transfer function of a system
when the system consists of more elements.

Nodes represent system variables.

Branches are interconnections between nodes.

Forward Path is any path (of branches) that connects from the input node
to the output note (there can be one or more forward paths in a system).

Note that any node can be considered an output node. It depends on the
requirements of the analysis.

9
However, input nodes are only those nodes which have no incoming branches.
Meaning, the nodes that have outgoing branches only can be input nodes.

Loop is a path which originates and terminates at the same node without
repetition of any node in the middle.
Non-touching loops are a group of loops that have no nodes in common.

2.2.1 Mason’s Gain formula

For a signal flow graph with N forward paths, the overall transfer function
is given by,
PN
k=1 Pk ∆k
T (s) =
∆

Pk is the k th forward path gain

∆k is the value of ∆ obtained by removing all loops that are touching the
k th forward path
∆ = 1 - [sum of all individual loop gains] + [sum of gain products of 2
non-touching loops] - [sum of gain products of 3 non-touching loops] + ...

• ∆ is constant for a given SFG. This means, ∆ is independent of the

selection of the output node.

• Self loops immediately next to input node are useless and can be re-
moved.

• If there are multiple inputs and the transfer function with respect to
any one has to be found, then perform the analysis by making all the
other inputs zero.

• If the SFG does not consist of any input node and A/B has to be found,
then create a dummy input node I, find A/I and B/I and divide them.

The typical feedback structure, when converted to SFG is shown.

r(t) 1 G y(t)
−H

10
2.3 Sensitivity analysis
Sensitivity is defined as the change observed in a certain quantity due to
variation (or change) that occurs in another quantity.
In control systems analysis, sensitivity is measured as change in the overall
transfer function of the system with changes in some part of the system (like
the parameter variations in the plant or the sensor).

∆α/α
Sβα =
∆β/β

where α is the quantity under observation and β is the parameter varying.

For an Open Loop System, T (s) = G(s). Hence, SGT = 1.

This means that an OL system is 100% sensitive to changes in system pa-
rameters.
D(s)G(s)
For a Closed Loop System, T (s) = 1+D(s)G(s)H(s)
.
Hence,
1
SGT =
1 + D(s)G(s)H(s)

(considering H(s) as constant) The above relation indicates that by choosing

appropriate value of controller D(s), the sensitivity to changes in the plant
parameters can be made negligible (at the cost of decreasing the overall gain).

It can also be deduced that,

T −D(s)G(s)H(s)
SH =
1 + D(s)G(s)H(s)

(considering G(s) as constant) This relation however tells that the system
will be highly sensitive to the changes in the sensor parameters, especially if
the gain of the controller is high.

Therefore, it must be ensured that the sensor is extremely well designed and
should not be susceptible to changes.
(it is easier to pose restrictions on the design on the sensor than the design
of the plant)

11
2.4 Disturbances
Disturbances are the unwanted signals that might get added to some pro-
cessed signals in the control system due to practical factors.

There are two types of Disturbances, namely Erratic High Frequency distur-
bances and Slow-varying disturbances.

Erratic disturbances (also called Noise) have unpredictable wave-forms and

can’t be analyzed using any standard modelling. Stochastic modelling has
to be used (which is not covered here).

Slow-varying disturbances are easier to observe and understand, and though

not entirely predictable, they can be modelled or approximated using stan-
dard signals and analysed.

2.4.1 Standard Disturbance Models

Impulse Signal
Used to model a sudden (almost instantaneous) jerk.

δ(t) ←→ 1

Step Signal
Used to model a sudden (almost instantaneous) change.
1
u(t) ←→
s

Ramp Signal
Used to model linear change.
1
tu(t) ←→
s2

Parabolic Signal
Used to model quadratic change.
t2 1
u(t) ←→ 3
2 s
12
Pulse Signal
Used to model a sudden change which lasts only for a certain known period.

1 − e−sT
u(t) − u(t − T ) ←→
s

If a certain system gives desired response to these inputs, then it usually

means the system can work accordingly for an arbitrary input. Hence, the
above signals are also called as Standard Test Signals.

2.4.2 Disturbance rejection

External disturbance signal can be modelled into the system by adding a
new input which adds to the output.

Here, the transfer function considering the disturbance signal as input is

given by,

N (s) W (s)
=
Y (s) 1 + D(s)G(s)H(s)

Hence the effect of disturbance can be effectively nullified by choosing ap-

propriate value for the controller D(s).

Such disturbances will directly add to the output and nothing can be done
to compensate for it in the case of open loop control systems.

13
2.4.3 Noise filtering
The major source of noise in a system is the sensor itself. Hence, effect of
noise is something that is added due to feedback which was absent in the
open loop system.
This can be modelled into the block diagram as follows.

Here, the transfer function considering the noise signal as input is given by,

V (s) −D(s)G(s)H(s)
=
Y (s) 1 + D(s)G(s)H(s)

Hence the effect of noise becomes more significant as the controller gain is
increased. This reinforces the intuition that adding a sensor and closing the
loop will lead to noise effects.
To take care of this, a low pass filter must be used in order to avoid high
frequency noise from entering the loop.

14
3 Analysis of Standard Systems
Most practical systems can be modelled using First and Second order sys-
tems. Hence, analysis and characterization of these two will be extremely
useful.

3.1 First order systems

k
Consider an Open Loop transfer function, G(s) = sτ
with unity negative
feedback.

R(s) E(s) 1
Y (s)
− k sτ

Closed Loop transfer function is given by,

Y (s) k
T (s) = =
R(s) 1 + sτ
This is the standard 1st order system transfer function.

k is referred to as System gain and τ is referred to as Time constant

Impulse response is given by,

k −t
h(t) = e τ u(t)
τ
Impulse Response

2.5

1.5
Amplitude

0.5

0
0 2 4 6 8 10 12
Time (sec)

15
Unit step response is given by,
t
s(t) = k[1 − e− τ ]u(t)
Step Response

4.5

3.5

3
Amplitude

2.5

1.5

0.5

0
0 2 4 6 8 10 12
Time (sec)

t
Here, k is the steady state response of the system and ke− τ is the transient
response of the system.
This means, the unit step input is getting multiplied by a factor of k in steady
state, hence it is called the system gain.

The time constant τ defines the speed of the system, since the transient re-
sponse tells how fast the system will reach it’s steady state.

Initial slope of step response is k/τ . This means, if the slope was maintained,
then the system transient would die out at t = τ .
However, since the slope is changing, the following observations can be made.
• at t = τ , the output would have reached 63.2% it’s steady state value.
• at t = 3τ , the output would have reached 95.0% it’s steady state value.
• at t = 4τ , the output would have reached 98.2% it’s steady state value.
Hence, 3τ and 4τ are called Settling Time with 5% and 2% tolerance band
respectively.

By the time t = 5τ , the output would have reached more than 99% of it’s
steady state value and for all practical purposes, it can be said that steady
state has been reached.

Hence, the system transients die faster if the time constant is lesser, which
is why it defines the speed of the system.

16
(note that the time constant is defined only for stable systems, meaning such
systems where the output actually settles down to some value and does not
tend to infinity [more on stability later])

3.2 Second order systems

2
kωn
Consider an Open loop transfer function, G(s) = s(s+2ζωn )
with unity nega-
tive feedback.

R(s) E(s) 2
ωn
Y (s)
− k s2 +2ζωn s

Closed loop transfer function is given by,

Y (s) kωn2
T (s) = = 2
R(s) s + 2ζωn s + ωn2

This is the standard 2nd order system transfer function.

k is the system gain as usual.

ωn is called Natural Frequency
ζ is called Damping Ratio

s2 + 2ζωn s + ωn2 = 0 is the Characteristic equation.

Roots of the Characteristic
p equation are given by,
s =p−ζωn ± jωn 1 − ζ 2
ωn 1 − ζ 2 = ωd , which is the Damped Frequency.

General expression for unit step response will be,

e−ζωn t
s(t) = 1 − p sin(ωd t + θ) where θ = cos−1 ζ
1 − ζ2

17
cosθ = ζp
sinθ = √1 − ζ 2
1−ζ 2
tanθ = ζ

1
Time constant of a second order system is given by, τ = ζωn
.

The response to a second order system can be classified to 4 different cases

depending on the value of the damping ratio. Location of poles in the s-plane
for each of the 4 cases:

Location of poles

Undamped
5 Underdamped
Critically damped
4
Overdamped
3

2
Imaginary axis

−1

−2

−3

−4

−5

−6 −4 −2 0 2 4
Real axis

Undamped Case: ζ = 0
There is no damping factor. Roots of CE are purely imaginary. Hence the
step response will be oscillatory, which is not usually favoured.
Step Response

2.5

2
Amplitude

1.5

0.5

0
0 2 4 6 8 10 12 14 16 18
Time (sec)

Under-damped Case: 0 < ζ < 1

Damping factor causes oscillations to die out and settle to steady state value

18
eventually.
This is the most commonly used configuration in second order systems.

When designing a second order system, the designer has to ensure the system
meets the transient and steady state specifications required. The transient
specifications are generally analysed from step response of the system in time
domain. Frequency domain specifications can also be useful in design.
Step Response

1.8

1.6

1.4

1.2
Amplitude

0.8

0.6

0.4

0.2

0
0 5 10 15
Time (sec)

3.2.1 Time Domain Specifications

(of under-damped response)

19
 • Delay time (td ) is the time required for the response to reach 50% of
it’s steady state value.

• Rise time (tr ) is the time required for the response to reach the steady
state value for the first time.

• Peak time (tp ) is the time required for the response to reach it’s maxi-
mum value.

• Maximum peak overshoot (mp ) is the maximum value the response

reaches, it is also the maximum deviation from the steady state value.
(it is measured in terms of percentage)

• Settling time (ts ) is the time required for the response to reach 95%
of 98% of it’s final value (depending on which is appropriate for the
specific application)

These terms can be derived from the typical step response formulae using
the definitions.
π−θ
tr =
ωd
π
tp =
ωd
√−ζπ
mp = e 1−ζ 2

3
ts (f or 95%) =
ζωn
4
ts (f or 98%) =
ζωn
Number of oscillations completed before settling (N) is given by,
N = ts ω2πd

Peak overshoots occur at π/ωd , 3π/ωd , 5π/ωd , ...

Peak undershoots occur at 2π/ωd , 4π/ωd , 6π/ωd , ...
Similarly, the peak values at the corresponding overshoots and undershoots
−nζπ
√
are governed by, mp = e 1−ζ 2 where n is the integer before π/ωd term.

20
If the input is scaled by a factor, the output will also be scaled by that factor
and the peak overshoot will be same.

It can be noted that mp is a function of ζ and others are all functions of ζ and
ωn . Peak overshoot and rise time are usually chosen as design specifications
for satisfying the transient response required.

Some patterns/conclusion that can be useful:

• mp ∝ 1/ζ

• mp ∝ 1/tp

• ζ ∝ 1/θ

• 5% < mp < 25% (preferred range in general)

• 0.4 < ζ < 0.7 (preferred range in general)

3.2.2 Frequency Domain Specifications

(of under-damped response)

• Resonant peak (Mr ) is the peak magnitude of the frequency response

of an under-damped second order system.

• Resonant frequency (ωr ) is the frequency at which the resonant peak

occurs.

• Cut-off frequency (ωc ) is the frequency at which the magnitude of the

frequency response is √12 (or 0.707) times it’s resonant peak.

21
 k
Mr = p
2ζ 1 − ζ 2
p
ωr = ωn 1 − 2ζ 2
From theprelation, it is obvious that Mr exists only when ωr is real and pos-
itive i.e 1 − 2ζ 2 > 0.
ζ < 0.707 =⇒ Mr > 1
ζ = 0.707 =⇒ Mr = 1
ζ > 0.707 =⇒ Mr does not exist

Mr is a function of ζ and ωc is a function of ζ and ωn . Hence, they directly

related to the time domain specifications and can be used as design specifi-
cations if frequency domain analysis is being carried out.

Critically Damped Case: ζ = 1

Limiting case of damping, meaning the oscillations are effectively killed and
the response is similar to that of first order system.

Over-damped Case: ζ > 1

Damping factor greater than 1 causes the system response to be slow, hence
this configuration is not used.
Step Response Step Response

1.4 1.4

1.2 1.2

1 1
Amplitude

Amplitude

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
Time (sec) Time (sec)

For a second order control system to be stable, the damping ratio has to be
positive. If it is negative, the system is certainly unstable.

3.3 Dominant Poles

First and second order systems are well studied and majority of the practical
systems can be modelled using them. However, it is not out of the question

22
that a designer might have to deal with higher order systems. In analysis of
systems of order 2 or more than 2, the concept of dominant poles becomes
very useful.

Consider a system has 2 poles at s = −s1 and s = −s2 . The location of these
poles determine the response of the system.

k
G(s) = (s+s1 )(s+s 2)

=⇒ G(s) = (1+s/s1k)(1+s/s
o
2)
in time constant form.
A B
=⇒ G(s) = s+s1 + s+s2 using partial fractions.
∴ g(t) = Ae−t/s1 + Be−t/s2 by Inverse Laplace transform.

The time constants of the terms are 1/s1 and 1/s2 . If the values of s1 and s2
are such that s2 ≥ 5s1 , it means that the term with time constant 1/s1 will
decay very slowly when compared to the term with time constant 1/s2 and
hence, the fast decaying pole is considered to be Insignificant pole whereas
the slow decaying pole is considered to be Dominant pole.

In case of higher order systems where the dominant poles are complex con-
jugates, the effective time constant is given by,
−1
τ=
Real [Dominant P ole]
If the dominant pole criterion is satisfied, it effectively reduces higher order
systems to lower order systems because insignificant poles can be neglected
(given the transfer function is expressed in pole-zero form).

3.4 Steady State Error

The accuracy of a control system is given by it’s steady state response as
mentioned earlier. The closer the steady state response is to the desired out-
put, the more accurate the system is. Which is why the accuracy of a system
can be defined in terms of steady state error.

For a general control system, the steady state error is defined as the difference
between the steady state value of the actual output and the desired output.
=⇒ ess = yss − rss

23
Steady State Error for unity feedback systems
If the system is a unity feedback system, it can be observed that the steady
state error is same as the final value of the error signal obtained (which is
fed to the controller).

R(s)
E(s) =
1 + D(s)G(s)
Using Final Value Theorem,
sR(s)
ess = lim
s→0 1 + D(s)G(s)
The following table gives steady state error for different types of inputs
against different types of systems.
Step input Ramp input Parabolic input
1
Type 0 system 1+Kp
∞ ∞
1
Type 1 system 0 Kv
∞
1
Type 2 system 0 0 Ka

Positional Error Constant

Kp = lim G(s)
s→0

Velocity Error Constant

Kv = lim sG(s)
s→0

Acceleration Error Constant

Ka = lim s2 G(s)
s→0

24
Note that all these constants are DC gains of the particular systems. Hence,
by designing the appropriate controller which can manipulate the overall DC
gain, it is possible to reduce the steady state error using feedback.

Steady State Error for non-unity feedback systems

For non-unity feedback systems (H(s) 6= 1), the error signal obtained (as
input to controller) is not the steady state error. It is more suitably called
actuating error.

To find the steady state error, there are 3 methods.

• Find final value of the output (using final value theorem) and find the
difference of it with desired output.

• Convert the non-unity feedback system to equivalent unity feedback

system by manipulating the block diagram.
D(s)G(s)
This will give G0 (s) = 1+D(s)G(s)H(s)−D(s)G(s)
 
1 + D(s)G(s)H(s) − D(s)G(s)
=⇒ ess = lim sR(s)
s→0 1 + D(s)G(s)H(s)

• Use the formulae:

ess = lim sR(s)[1 − T (s)]

s→0

where T (s) is the closed loop transfer function

Note that the steady state error is defined only for stable systems.

25
4 Stability Analysis
Any control system being designed has to be designed under the condition
that it must satisfy stability. Meaning, an unstable system which has all
other desired qualities is still unacceptable.

Stability is generally defined in the following two ways.

Bounded Input Bounded Input (BIBO) Stability:

A system is said to be BIBO stable if for a bounded input, the output always
turns out to be bounded.
This means for any input (command or disturbance), if it is absolutely
summable, the output must also be absolutely summable.
For LTI systems, it can also be stated that BIBO stability means the impulse
response must be absolutely summable.

Zero Input Stability:

A system is said to be Zero-input stable if all the states of the system are
bounded (or finite) when the input to the system is zero.

For most systems, both the above definitions will coincide, meaning for this
class of systems, analysis can be directly done for stability.

4.1 Stability analysis in s-plane

The closed loop stability of any system is analysed using its transfer function,
mainly the characteristic equation. The location of poles of the transfer
function i.e the solutions to the characteristic equation in the s-plane will
determine the stability of the system.
Consider a system with closed loop transfer function as T (s) = (s+z o )(s+z1 )....(s+zm )
(s+po )(s+p1 )....(s+pn )
where n > m.
The partial fraction expansion of the system can be expressed as,
Ao A1 An
T (s) = + + .... +
s + po s + p1 s + pn
By taking inverse Laplace transform to obtain the impulse response,

h(t) = Ao e−po t + A1 e−p1 t + .... + An e−pn t

26
It can be noted that the impulse response will be absolutely summable only
if po , p1 , ....pn are positive.
If even one of them is negative, the impulse response will eventually tend to
infinity and it will be an unstable system.
The unique case in which the impulse response is neither dying nor growing,
but stays as a non-zero constant or oscillates in a fixed range (in steady state)
implies the system is ”Marginally stable”.

Therefore, it can be concluded that the system will be stable if and only if
all of its poles lie on the left half of s-plane.

4.1.1 Routh-Hurwitz Criteria

Routh-Hurwitz is a method that can provide both absolute as well as relative
stability of a system by analysing it’s characteristic equation.
This method also points out the general location of the poles but not the
exact points.
(note that the same technique can also be used to gather information about
the zeros, though it is of no use for stability analysis)

Routh-Hurwitz method needs the construction of the Routh table, which is

done using the characteristic equation.

Let the equation under consideration be,

P (s) = a0 sn + a1 sn−1 + a2 sn−2 + ....an−1 s + an
For convenience, assume that an 6= 0 and a0 > 0 (so as to maintain unifor-
mity in sign and to make sure s = 0 factor is not present).

The characteristic equation directly gives the coefficients of first two rows of
the Routh table. The rest of the table is filled according to the pattern shown.

Routh Table

27
 sn a0 a2 a4 ...
sn−1 a1 a3 a5 ...
sn−2 b0 b1 b2 ...
sn−3 c0 c1 c2 ...
.. .. .. .. ..
. . . . .
s0 h0 0 0 0
where,
a1 a2 − a3 a0 a1 a4 − a5 a0
b0 = ; b1 =
a1 a1
b 0 a3 − a1 b 1 b 0 a5 − a1 b 2
c0 = ; c1 =
b0 b0
and so on...

Once the Routh Table is constructed, the sign changes in the first column of
the table are to be noted.

For the system to be stable, all coefficients in the first column of the Routh
table must be strictly positive.

If there are one or more sign changes, then the system is unstable.
The number of sign changes in the first row of the Routh table is equal to
the number of roots (of the equation) that lie in the right half of s-plane.

Cases with coefficients turning out to be 0-

1. If one coefficient in the first column is zero, with its corresponding row
coefficients being non-zero, it means there is at least one root on the right
hand side of s-plane.
Replace zero with some small positive number  and perform normal analysis.

2. If row/s of zeros occurs, it means symmetric roots about the origin are
present.
Obtain Auxiliary Equation which is constructed by using the coefficients
of the row above the row of zeros.
Then differentiate this equation (w.r.t s) to obtain new coefficients for the
row of zeros and perform the usual analysis. Note that power of Auxiliary

28
equation must be even since it will give number of roots that are symmetric
about the origin.
• Number of sign changes above AE = number of roots on RHP

• Number of sign changes below AE = number of symmetric roots on

RHP (each such root on RHP will have a corresponding symmetric
root in LHP)
If there is only a single row of zeros, with no sign change, it means the system
has simple poles on the jω axis and is marginally stable.
If there are multiple rows of zeros, with no sign change, it means the system
has repeated poles on the jω axis and is unstable.

The Routh-Hurwitz criteria can be reduced to simple short-cuts for second

and third order characteristic equations.

Conditions for stability of second order system:

as2 + bs + c = 0

a > 0, b > 0, c > 0

Conditions for stability of third order system:

as3 + bs2 + cs + d = 0

a > 0, b > 0, c > 0, d > 0, bc > ad

(note that if bc = ad, it means the system is marginally stable)

Conditional Stability

The most important utility of the Routh-Hurwitz criteria is to determine the

range of values of the system gain K for which the closed loop system will
be stable.

In general, if a system with open loop transfer function G(s) is closed using
unity feedback and controller gain K. The characteristic equation of the
closed loop system will be 1 + KG(s) = 0.
Using Routh-Hurwitz criteria, the constraint equations on K can be found.

29
The value of K where the system is marginally stable is Km .
The conclusion for range of values of K can be-

• Stable for 0 < K < Km ; Unstable for K > Km

• Unstable for 0 < K < Km ; Stable for K > Km

The value of frequency ω at K = Km is the frequency of oscillation ωn .

Relative Stability

So far, Routh-Hurwitz criteria has been used to evaluate absolute stability

by analysing position of poles with respect to the jω axis i.e s = 0 line. The
same criteria can also be used to evaluate relative stability of a system with
respect to any vertical line in the s-plane.

To find stability of a system with respect to the line s = −a, let s + a = so .

Hence, s = so − a. Substitute so − a for s in the characteristic equation of the
system and apply Routh-Hurwitz criteria for the new equation in so -plane.

This technique can be used to find the number of roots (poles or zeroes from
the equation) that lie on the s = −a line, between s = 0 and s = −a lines,
RHS of s = −a line, LHS of s = −a line.

30
4.1.2 Root Locus
The Root Locus technique provides a graphical method of plotting the locus
of the roots (poles and zeros) of a closed loop transfer function in the s-plane
as a given system parameter (typically gain E) is varied over the complete
range of values (0 to ∞).

The Root Locus plot is extensively used in control systems design because
once the sketch is obtained, the roots corresponding to a particular value of
the system parameter can be obtained or the value of a desired root location
can be determined.

Consider a system with open loop transfer function G(s), is closed using
feedback element H(s) with controller gain K. Then the closed loop transfer
KG(s)
function will be T (s) = 1+KG(s)H(s) . Hence the characteristic equation of the
closed loop system will be 1 + KG(s)H(s) = 0. =⇒ KG(s)H(s) = −1

From the above result, two conditions can be obtained for any point s to lie
on the root locus.

Magnitude Criterion:

|KG(s)H(s)| = 1

Angle Criterion:

∠G(s)H(s) = ±(2q + 1)180o

It can be noted that since K can take any value, every point in the s-plane
inherently satisfies the magnitude criterion. Hence, the magnitude criterion
is used to find the value of K given a particular value of s.

Since K does not contribute any angle, angle criterion is used to check if any
given point s = s1 lies on the root locus plot of the given closed loop system.

The construction of the root locus plot for any closed loop system for stabil-
ity and design purpose is done using the following steps/rules.

31
Root Locus Construction Rules
1. The root locus is symmetrical about the σ axis (jω = 0).
2. As K increases from 0 to ∞, each branch of the root locus originates
from an open loop pole with K = 0 and terminates at either an open
loop zero or infinity with K = ∞. The number of branches terminating
at ∞ is equal to the difference between open loop poles and zeros
(n − m).
3. A point on the real axis lies on the root locus if the number of open
loop poles and zeros on the real axis to the right of this point is odd.
4. The (n − m) branches that terminate at ∞ do so along straight line
o
asymptotes whose angles are given by, φA = (2q+1)180
n−m
; q = 0, 1, 2, ...n−
m − 1.
5. The asymptotes cross the real axis at a point known as centroid, deter-
mined by the relationship, C = (ΣReal[poles] − ΣReal[zeros])/(n − m).
6. Break away points are those points at which the root locus of two
different branches meet and break away towards ∞, hence they are the
points where multiple roots occur. The break away points of the root
locus are the solutions to dK/ds = 0.
Break away points are points on the real axis only. Meaning, if the
solutions to dK/ds = 0 are complex, then there are no points at which
the closed loop poles meet for breaking away.
7. In case of existence of complex break poles or zeros, the break away
directions are given by,
Angle of departure from an open loop pole side: φd = ±180o + φ;
Angle of arrival from an open loop zero side: φa = ±180o − φ;
where φ = Σθz − Σθp (i.e net angle contributed at the open loop root
by all other open loop roots)
8. The intersection of root locus branches with the jω axis can be deter-
mined by using Routh-Hurwitz criteria. At K = Km where the root
locus branches intersect the jω axis, a row of 0s will occur in the Routh
table. The row above this row gives auxiliary equation, whose roots
will give the points of intersection.

32
 k
Consider a system with open loop transfer function G(s) = s(s+4) .
The root locus plot will be as indicated in the following figure.
Root Locus

2.5

1.5

1
Imaginary Axis

0.5

−0.5

−1

−1.5

−2

−2.5
−4.5 −4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5
Real Axis

The addition of a pole to an open loop system will drive the system towards
instability if the loop is closed. This is illustrated in the root locus plot of
k
G(s) = s(s+2)(s+4) .
Root Locus

10

4
Imaginary Axis

−2

−4

−6

−8

−10
−14 −12 −10 −8 −6 −4 −2 0 2 4
Real Axis

33
The addition of a zero to an open loop system will increase the relative sta-
bility of the closed loop system. This is illustrated in the root locus plot of
k(s+1)
G(s) = s(s+2)(s+4) .
Root Locus

2
Imaginary Axis

−2

−4

−6

−8
−4.5 −4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5
Real Axis

The root locus plot of a system with complex poles is shown.

G(s) = s(s+6)(sk2 +4s+13
Root Locus

10

4
Imaginary Axis

−2

−4

−6

−8

−10
−12 −10 −8 −6 −4 −2 0 2 4 6 8
Real Axis

34
If the root locus of a system traces
√ the path of a circle, then the break away
√
points will be of the form a + b where a is the center of the circle and b
is it’s radius. G(s) = k(s+2)(s+4)
s(s+1)

Root Locus

1.5

0.5
Imaginary Axis

−0.5

−1

−1.5
−4.5 −4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5
Real Axis

The root locus plot of a system with symmetric poles will have break away
points at the centroid. G(s) = s(s+2)(sk2 +2s+2)
Root Locus

2.5

1.5

1
Imaginary Axis

0.5

−0.5

−1

−1.5

−2

−2.5
−3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5
Real Axis

35
Transportation Lag
The transportation lag is the delay between the time an input signal is applied
to a system and the time the system reacts to that input signal. Meaning if
input is r(t), the output can be represented as y(t-T).
In the Laplace domain, the transfer function representing the transportation
lag will be: YR(s)
(s)
= e−sT .
To incorporate the exponential term in the root locus, it can be approximated
as e−sT = 1 − sT using first order approximation of Taylor’s series expansion.

If transportation lag occurs in a system, then by using the above approxi-

mation, it can be inferred that the effective characteristic equation will be 1
- G(s)H(s) = 0 i.e it has to be modelled as a positive feedback system. To
analyse such systems in the root locus, the magnitude criterion will remain
the same but angle criterion will change to
∠G(s)H(s) = ±2q(180o ) q = 0, 1, 2, ...
. The root locus plot derived from this angle criterion is called Comple-
mentary Root Locus and is used for positive feedback systems.

The rules that change in the complementary root locus from the original are-

• A point on the real axis lies on the complementary root locus if the
number of open loop poles and zeros on the real axis to the right of
this point is even.

• The (n − m) branches that terminate at ∞ do so along straight line

o
asymptotes whose angles are given by, φA = (2q)180
n−m
; q = 0, 1, 2, ...n −
m − 1.

• In case of existence of complex break poles or zeros, the break away

directions are given by,
Angle of departure from an open loop pole side: φp = φ;
Angle of arrival from an open loop zero side: φz = −φ;
where φ = Σθz − Σθp (i.e net angle contributed at the open loop root
by all other open loop roots)

The rest of the rules are same as usual.

36
4.2 Stability analysis using Frequency response
If a sinusoidal signal is applied as an input to a Linear Time-Invariant (LTI)
system, then it produces the steady state output, which is also a sinusoidal
signal. The input and output sinusoidal signals have the same frequency, but
differ in amplitudes and phase angles.

Consider a system with transfer function T(s) with input r(t) = Asin(ωo t+θ)
and output y(t) = Bsin(ωo t+φ) where B = A|G(jωo )| and φ = θ +∠G(jωo ).
Here, |G(jωo )| is the magnitude of the frequency response and ∠G(jωo ) is
the angle of the frequency response.
=⇒ G(ω) = |G(jω)|∠G(jω)

Stability of a system can be analysed using its frequency response. One of

the major advantages of this type of analysis in design is, there is no need to
come up with an actual model for the system. The frequency response can
be obtained by giving the system sample inputs and analysing the outputs.

Gain Cross-over Frequency is the frequency at which the magnitude of

the transfer function is one. It is denoted by ωgc .

Phase Cross-over Frequency is the frequency at which the phase of the

transfer function is 180o . It is denoted by ωpc .

Gain Margin is the magnitude factor by which a stable system can be in-
creased before making it marginally stable (or driving it towards instability).
Mathematically, the gain margin (GM) is equal to the reciprocal of the mag-
nitude of the Nyquist plot at the phase cross over frequency.
=⇒ GM = M1pc
Gain is usually expressed in decibels for convenience, hence GM = 20 log10 M1pc
Where, Mpc is the magnitude in normal scale at ω = ωpc .

Phase Margin is the additional phase which can be added to a stable sys-
tem before making it marginally stable (or driving it towards instability).
Mathematically, the phase margin (PM) is equal to the sum of 180o and the
phase angle at the gain cross over frequency.
=⇒ P M = 180o + Φgc
Where, Φgc is the phase angle at ω = ωgc .

37
Frequency domain stability criteria:

GM & PM ωgc & ωpc

o
Stable GM > 0dB & P M > 0 ωgc < ωpc
Marginally Stable GM = 0dB & P M = 0o ωgc = ωpc
Unstable GM < 0dB & P M < 0o ωgc > ωpc
(this criteria is for closed loop system, assuming open loop system is stable)

4.2.1 Polar Plot

The Polar plot is a plot, which can be drawn between the magnitude and the
phase angle of G(jω)H(jω) by varying ω from 0 to ∞.
Hence, it provides a mapping from the s-plane to G(s)H(s)-plane.

Steps to obtain polar plot:

1. Substitute, s = jω in the open loop transfer function and write the

expressions for magnitude and the phase of G(jω)H(jω).

2. Find the starting magnitude and the phase of G(jω)H(jω) by substi-

tuting ω = 0. So, the polar plot starts with this magnitude and the
phase angle.

3. Find the ending magnitude and the phase of G(jω)H(jω) by substi-

tuting ω = ∞. So, the polar plot ends with this magnitude and the
phase angle.

4. Check whether the polar plot intersects the real axis, by making the
imaginary term of G(jω)H(jω) equal to zero and find the value of ω.

5. Check whether the polar plot intersects the imaginary axis, by making
real term of G(jω)H(jω) equal to zero and find the value of ω.

6. For drawing polar plot more clearly, find the magnitude and phase of
G(jω)H(jω) by considering the other value of ω.

Standard polar plot structure for different types and orders of systems is
shown.

38
The blue words (type of system) indicate the starting direction and the green
words (order of system) indicate the ending direction of the polar plot.

Polar plot of an open loop transfer function can be used for stability analysis
of its corresponding closed loop system.

• If the plot encloses the critical point (−1 + 0j), the system is unstable
since ωgc > ωpc i.e GM and PM are negative.

• If the plot intersects the critical point (−1+0j), the system is marginally
stable since ωgc = ωpc i.e GM and PM are zero.

• If the plot does not enclose the critical point (−1 + 0j), the system is
stable since ωgc < ωpc i.e GM and PM are positive.

39
The figure shows the polar plot of a stable system and illustrates when it can
reach instability using gain and phase margins.

The polar plot is used to analyse stability of closed loop system in more detail
the Nyquist stability criteria.

4.2.2 Nyquist Criterion

Nyquist plot is the continuation of polar plot for finding the stability of the
closed loop control systems by varying ω from -∞ to ∞.

The Nyquist stability criterion works on the Principle of Argument.

Any closed contour defined in the s-plane will map to another closed contour
in the 1 + G(s)-plane, which is called as a plot.
• If the contour in the s-plane encircles a pole in the clockwise direction,
the plot encircles the origin in anti-clockwise direction
• If the contour in the s-plane encircles a zero in the clockwise direction,
the plot encircles the origin in clockwise direction.
Therefore, the number of anti-clockwise encirclements of the origin in 1 +
G(s)-plane is equal to the difference between number of poles and number of
zeros encircled by the contour in s-plane.

40
This principle can be applied to analyse stability.
Consider a system with open loop transfer function G(s), which is closed
using unity negative feedback. The closed loop transfer function will be
G(s)
T (s) = 1+G(s) .
The stability of open loop system is known, because it is given by the poles of
G(s) that are on right half of s-plane, denoted by P . To analyse the stability
of the closed loop system, it is necessary to get the number of poles of T (s)
present on the right half of s-plane which is same as the number of zeros of
the characteristic equation 1 + G(s), denoted by Z.

Hence, a contour on the s-plane is defined such that it covers the entire right
half of s-plane (starting from origin to +j∞ (imaginary axis), extends to ∞
on positive real axis to −j∞ (imaginary axis) and ends back at origin. This
is called ”Nyquist Contour”.
Instead of taking 1+G(s)-plane and considering anti-clockwise encirclements
of 0 + 0j (origin) for mapping, it is more convenient to take G(s)-plane and
consider anti-clockwise encirclements of −1 + 0j instead (which is called
”Critical Point”). When the Nyquist Contour is mapped to the G(s)-plane,
the obtained graph is called ”Nyquist Plot”.
N =P −Z
where P is known and N is found using the Nyquist plot.
Stability is determined using the value of Z.
• Z > 0 =⇒ closed loop system is unstable (or marginally stable)
• Z = 0 is zero =⇒ closed loop system is stable
A few examples implying the Nyquist stability criterion are given.
5(s+1)
(i) Consider an open loop transfer function G(s) = (s−2)(s+3) in unity nega-
tive feedback configuration.
There is 1 open loop pole on RHP (s = 2), hence P = 1. The characteristic
equation will be s2 + 6s − 1 = 0, which is unstable with 1 root on RHP, hence
Z = 1. N = Z − P = 0, meaning the Nyquist plot must encircle the critical
point 0 times.
4(s+2)
(ii) Next consider a different open loop transfer function G(s) = (s−1)(s−2)
in
unity negative feedback configuration.

41
There are 2 open loop poles on RHP (s = 1, 2), hence P = 2. The charac-
teristic equation will be s2 + s + 10 = 0, which is stable i.e no roots on RHP,
hence Z = 0. N = Z − P = −2, meaning the Nyquist plot must encircle the
critical point in clockwise direction 2 times.

10
(iii) Finally, consider another open loop transfer function G(s) = (s−1)(s+2)(s+3)
in unity negative feedback configuration.
There is 1 open loop pole on RHP (s = 1), hence P = 1. The characteristic
equation will be s3 + 4s2 + s + 4 = 0, which is marginally stable with a pair of
complex conjugate roots on jω axis, hence the Nyquist plot should intersect
the critical point twice.

The following steps are to be followed to obtain Nyquist plot from the Nyquist
contour i.e from s-plane to G(s)-plane.

The Nyquist contour is divided to 4 parts.

1. First part is mapped by obtaining the polar plot.

2. For the second part, substitute s = lims→∞ Rejθ ; θ goes from 90o to
−90o . Usually this part maps to a single point in the G(s)-plane.

42
 3. Third part is mapped by obtaining the inverse polar plot i.e mirror of
polar plot with respect to real axis.

4. For the fourth part, substitute s = lims→0 Rejθ ; θ goes from −90o to
90o . However, this is necessary only if there are open loop poles on the
jω axis.

Once the complete Nyquist plot is obtained, the number of anti-clockwise

encirclements around the origin can be found and closed loop stability is de-
termined using Z = N − P from the criterion explained earlier.

The Nyquist plots corresponding to the transfer functions given in examples

(i), (ii) and (iii) respectively are illustrated.

It can be verified that the number of anti-clockwise encirclements counted

from each of the plots match with the value of N obtained through analysis.

For Nyquist plots that go to ∞ for some values of ω, in order to find the line
asymptotic to the plot, split the transfer function to real and imaginary parts
and then substitute of ω. This is because using the magnitude directly will
not give information regarding which part is causing the transfer function to
go to ∞.

1
For example, consider G(s) = s(s+1)(s+2)
. |G(jω)| → ∞ at ω = 0.

1 −j(−jω + 1)(−jω + 2)
G(ω) = =
jω(jω + 1)(jω + 2) ω(jω + 1)(jω + 2)(−jω + 1)(−jω + 2)

43
 −3 −2 + ω 2
G(jω) = + j
(ω 2 + 1)(ω 2 + 4) (ω 2 + 1)(ω 2 + 4)
=⇒ Re[G(jω)] → −3/4 when ω → 0 and hence the plot is asymptotic to
Re[G(jω)] = −3/4 at ω = 0.

Observations regarding Nyquist plots of some standard systems:

Systems having neither poles nor zeros in the right side of s-plane are called
”Minimum phase systems”. Minimum phase systems will have Nyquist plot
only on right side of GH-plane.

Systems with poles in left side of s-plane and zeros in right side of s-plane that
are symmetric about jω axis are called ”All pass systems”. All pass systems
will have symmetric Nyquist plot with respect to both real and imaginary
axes of GH-plane.

A strictly proper transfer function implies the number of poles is greater than
the number of zeros. Hence, the Nyquist plot of a system having a strictly
proper transfer function will go to 0 as ω → ∞.

44
4.2.3 Bode Plot
Bode plot is another tool used to analyse stability of a closed loop system in
the frequency domain. It actually consists of two plots, namely the Magni-
tude Bode plot and the Phase Bode plot.
Magnitude Bode plot is drawn between 20 log10 |G(jω)| (i.e magnitude in
dB) against ω in log scale. Phase Bode plot is drawn between ∠|G(jω)| (i.e
phase in degrees) against ω in log scale.

The gain and phase margins of the closed loop system can be obtained by
the two graphs and hence stability can be analysed.

Construction of Bode plot:

The magnitude and phase plot construction is made simple by taking different
individual building blocks that appear in a system and analysing their effects.
1. Constant gain: G(s) = K
|G(jω)| = 20 log10 K = C dB → Constant line in magnitude plot
∠G(jω) = tan−1 (0/K) = 0o → No effect in phase plot
2. Poles and Zeros at origin: G(s) = s±n
|G(jω)| = 20 log10 (jω)±n = ±n 20 dB
∠G(jω) = ±n tan−1 ω/0 = ±n 90o

First order pole −20 dB/dec −90o

Second order pole −40 dB/dec −180o
First order zero 20 dB/dec 90o
Second order zero 40 dB/dec 180o
k
For a type ’n’ system G(s) = sn
, Bode magnitude plot cuts the ω-axis
at ω = k 1/n
3. First order factors : G(s) = (1 + sτ )±n
(
±n 20 dB/dec : ωτ > 1
|G(jω)| = ±n 20 log10 (1 + jωτ )±n =
0 : ωτ < 1

o
±0
 : ωτ → 0
−1 o
∠G(jω) = ±n tan ωτ = ±n 45 : ωτ = 1

±n 90o : ωτ → ∞

where ω = 1/τ is Corner Frequency.

45
 4. Second order factors : G(s) = (s2 + 2ζωn s + ωn2 )±n (
q ±n 40 dB/dec : ω > ωn
|G(jω)| = ±20 n log10 ( (1 − ( ωωn )2 )2 + ( 2ζω
ωn
)2 =
0 : ω < ωn

o
±0
 :ω→0
−1 ω o
∠G(jω) ≈ ±n tan ωn = ±n 90 : ω = ωn
 o
±n 180 : ω → ∞

where ω = ωn is Corner Frequency.
Note that the above methods are to obtain approximate Bode plot. The
approximations will not hold at the corner frequencies and around it. If cal-
culated, at the corner frequency, the magnitude will be 3n dB more or less
than the approximated magnitude.
And also, Bode plot analysis is generally applicable only to Minimum phase
systems and where the open loop system is known to be stable.

Stability analysis using Bode Plot:

The graphs below indicate the bode magnitude and phase plots of stable and
unstable systems respectively.
• Stable systems will have positive values of GM & P M as ωgc < ωpc .
• Unstable systems will have negative values of GM & P M as ωgc > ωpc .
Bode Diagram Bode Diagram
Gm = 7.96 dB (at 1.73 rad/sec) , Pm = 76.3 deg (at 0.859 rad/sec) Gm = −12 dB (at 1.73 rad/sec) , Pm = −37.8 deg (at 2.27 rad/sec)
50
20

0
0
Magnitude (dB)

−20
Magnitude (dB)

−50 −40

−60
−100
−80

−150 −100
0 0
Phase (deg)

−90 −90
Phase (deg)

−180 −180

−270 −270
−2 −1 0 1 2 −2 −1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Frequency (rad/sec)

(for marginally stable systems, the crossover frequency will be same, hence
there will be no margins i.e GM = 0 dB and P M = 0o )

46
Obtaining transfer function from given Bode Plot:
For a minimum phase system, the transfer function can be uniquely deter-
mined from the magnitude curve alone.

The following steps are to be followed to estimate the minimum phase transfer
function from a given magnitude Bode plot.

• Observe starting slope: Will give information regarding poles or zeros

at origin.

• Observe change in slope at each corner frequency: Will give information

about 1st and 2nd order factors. Express in time constant form.

• Find DC gain i.e the constant k using Y = mX + C where Y is the

magnitude (in dB) at the frequency X, m is the slope of the curve
at the frequency X and C is the intercept on the magnitude axis.
=⇒ C = 20 log10 (k).

Bode plots and Nyquist plots are extremely useful in designing of controllers
and compensators using frequency domain representation.

5 Industrial Systems
As studied in the introduction, the plant by itself will not give desired re-
sponse. It is to be controlled by another system which gives signal to the
plant based on command input and actual output (i.e error signal). These
systems that modify the response of the plant to desired response based on
the error signal are either called controllers or compensators.

5.1 Controllers
Controllers are systems that are cascaded before the plant in order to use the
error signal and modify the action of the plant to satisfy design requirements
such as steady state error, rise time, settling time, maximum peak overshoot,
etc.
There are really 4 types of controllers that are used in practical systems, but
theoretically there can be 6 types which are all discussed in detail.

47
5.1.1 Proportional Controller
R(s) E(s) P (s) Y (s)
kp G(s)
−

H(s)

The simplest controller, where the plant response is modified by changing

it’s open loop gain.
Hence, the controller output is proportional to the error signal.

P (s) = kp E(s)

p(t) = kp e(t)
In majority of the practical systems, controlling the system gain alone will
not be able to meet all design requirements. There will be a trade-off between
speed and accuracy of the response.
When the error signal eventually reaches 0, then the control signal also be-
comes 0, which means there is nothing to drive the plant to stay in that state
and steady state output of the system will not be same as desired output. In
such a case, the gain can be increased to reduce the steady state error, but
it will never go to 0. (note that for systems that don’t need any power to
stay at the steady state value, proportional control will work).
Also, increasing the gain to solve this problem can lead to instability of the
closed loop system.

5.1.2 Integral Controller

R(s) E(s) P (s) Y (s)
ki /s G(s)
−

H(s)

48
The integral controller is used to increase the type of a system, which will
eliminate steady state error as it adds an additional pole. Hence, the con-
troller output is proportional to the sum of accumulated error signals i.e
integral of the error signal.
ki
P (s) = E(s)
s
Z t
p(t) = ki e(τ )dτ
τ =0
However, adding a pole to the system will adversely affect stability.

5.1.3 Derivative Controller

R(s) E(s) P (s) Y (s)
skd G(s)
−

H(s)

The differential controller is used to make the system transient faster and
better since it adds a zero. Hence, the controller output is proportional to
the derivative of the error signal.
P (s) = skd E(s)
de(t)
p(t) = kd
dt
However, adding a zero to the system will adversely affect steady state ac-
curacy, and also a pure differentiator is not practically realizable.

5.1.4 PI Controller
R(s) E(s) P (s) Y (s)
kp + ki /s G(s)
−

H(s)

49
As discussed, a drawback of P controller is that if the error signal is 0, then
the control signal is 0 and there is no input to the plant to drive the system.
To avoid this, the control signal is made to be dependent on the accumulation
of error signal along with only the error signal, so the plant will never get
zero input. This will cause the plant to be driven exactly to the required
steady state and also ensure it has the energy to maintain it.
Hence, PI controller will effectively improve the steady state response of the
plant and is used when P controller alone would give some steady state error
for any value of the gain.

 
ki
P (s) = kp + E(s)
s
Z t
p(t) = kp e(t) + ki e(τ )dτ
τ =0

5.1.5 PD Controller
R(s) E(s) P (s) Y (s)
kp + skd G(s)
−

H(s)

P controller does not take into account the changes in the error signal which
should actually have an impact on the plant. For example, if the error is
decreasing too fast, then the plant needs to slow down to avoid overshoot
of the steady state value. Hence PD controller is used, to provide better
transient response. Note that PD controller also prevents any potential in-
stability issues due to high gain of P controller.
It should be noted that due to derivative action, high frequency or high am-
plitude signals can be introduced into the system, which have to be filtered
out or limited.

P (s) = kp + skd E(s)

50
 de(t)
p(t) = kp e(t) + kd
dt

5.1.6 PID Controller

R(s) E(s) P (s) Y (s)
kp + ki /s + skd G(s)
−

H(s)

 
ki
P (s) = kp + + skd E(s)
s
Z t
de(t)
p(t) = kp e(t) + ki e(τ )dτ + kd
τ =0 dt
The PID controller is the optimal controller which generates control signal
proportional to the error, its integral and its derivative. This controller can
almost always be used to ensure both steady state and transient response
requirements are met since the derivative part ensures gives higher speed
and the integral part ensures low steady state error, while stability is also
maintained.

Effect of increasing kp , ki and kd in PID controller:

Parameter Peak overshoot Settling time Steady state error
kp Increases Low impact Decreases
ki Increases Increases Eliminated
kd Decreases Decreases No impact

5.2 Compensators
Compensators are more practical systems used to modify the behaviour of
the plant to satisfy the design specifications. Compensators are systems that
add extra poles and zeros in appropriate points in s-plane to get the desired
behaviour.

51
5.2.1 Lead Compensator
The lead compensator is an electrical network which produces an output
having phase lead for the input applied. The lead compensator circuit in the
‘s’ domain is shown.

α[1 + T s] R2
Gc (s) = α= ; T = R1 C
1 + αT s R1 + R2

Pole at s = −1/αT and Zero at s = −1/T Since α < 1, the zero is closer to
the origin than the pole.

The lead compensator provides slope of +20 dB/decade for the magnitude
response and a positive phase for the phase response of the uncompensated
system.

52
Maximum phase lead provided by the lead compensator is given by
 
−1 1 − α
θm = sin
1+α
The frequency at which maximum phase occurs is given by the geometric
mean of the corner frequencies.
1
ωm = √
T α

5.2.2 Lag Compensator

The lag compensator is an electrical network which produces an output hav-
ing the phase lag for the input applied. The lag compensator circuit in the
‘s’ domain is shown.

[1 + T s] R1 + R2
Gc (s) = β= ; T = R2 C
1 + βT s R2

53
Pole at s = −1/βT and Zero at s = −1/T Since β > 1, the pole is closer to
the origin than the zero.

The lag compensator provides slope of -20 dB/decade for the magnitude
response and a negative phase for the phase response of the uncompensated
system.

Maximum phase lag provided by the lag compensator is given by

 
−1 1 − β
θm = sin
1+β
The frequency at which minimum phase occurs is given by the geometric
mean of the corner frequencies.
1
ωm = √
T β

54
5.2.3 Lag-Lead Compensator
Lag-Lead compensator is an electrical network which produces phase lag
at one frequency region and phase lead at other frequency region. It is
a combination of both the lag and the lead compensators. The lag-lead
compensator circuit in the ‘s’ domain is shown.

α[1 + T1 s][1 + T2 s]
Gc (s) =
[1 + αT1 s][1 + βT2 s]

R2 R1 + R2
α= ; T1 = R1 C; β= ; T2 = R2 C
R1 + R2 R2
The lag-lead compensator can work both ways i.e provide phase lag first and
then phase lead or vice versa.

If the lag compensator pair is closer to origin, then the compensator behaves
like a band stop filter as shown in the magnitude Bode plot.

55
 Bode Diagram

24

23

Magnitude (dB)
22

21

20

19
20

10
Phase (deg)

−10

−20
−1 0 1 2
10 10 10 10
Frequency (rad/sec)

If the lead compensator pair is closer to origin, then the compensator behaves
like a band pass filter as shown in the magnitude Bode plot.

Bode Diagram

28

27
Magnitude (dB)

26

25

24

23
20

10
Phase (deg)

−10

−20
−1 0 1 2
10 10 10 10
Frequency (rad/sec)

56
6 State Space Analysis
Disadvantages of classical control theory:

1. Applicable only to linear time invariant systems.

2. Applicable only if initial conditions are zero.

3. Design of systems with multiple inputs and multiple outputs becomes

cumbersome.

4. Does not consider internal states of the system, considers only input
and output.

5. Higher order system analysis becomes cumbersome.

Advantages of modern control theory:

1. Applicable to linear and non-linear systems.

2. Applicable to time invariant and time varying systems.

3. Applicable even if initial conditions are non-zero.

4. Design of systems with multiple inputs and multiple outputs is easier.

5. Any higher order system can be represented using first order equations
only.

6. Considers internal states of the system, along with input and output.

The state of a system corresponds to the minimum set of variables such that
the knowledge of these variables at time t = to and input at time t ≥ to
completely determines the behaviour of the system for any time t ≥ to .
These variables are called ”State Variables”. Since a system will have mul-
tiple state variables (say ’n’), then the vector consisting of all the n state
variables is called the ”State Vector”. The vector space consisting of all pos-
sible states of a system is called it’s ”State Space”.

57
The above system consists of r input variables (u1 , u2 , ....ur ), m output vari-
ables (y1 , y2 , ....ym ) and n state variables (x1 , x2 , ....xn ) all being functions
of time.

The state variable representation of such a system is given by n first order

differential equations.
~x˙ (t) = f (~x(t), ~u(t), t) for time varying systems
~x˙ (t) = f (~x(t), ~u(t)) for time invariant systems
where ~x˙ (t) is the vector of first derivatives of the state variables.

The outputs are also represented as functions of the states and inputs.
~y (t) = f (~x(t), ~u(t), t) for time varying systems
~y (t) = f (~x(t), ~u(t)) for time invariant systems

For linear time invariant systems, due to property of superposition and ho-
mogeneity, it can be found that the state equations and output equations
can be represented as follows.

~x˙ (t) = A~x(t) + B~u(t)

~y (t) = C~x(t) + D~u(t)

A → System matrix; B → Input matrix;
C → Output matrix; D → Transmission matrix;

Block diagram representation

58
Selection of state variables
• The number of state variables must be minimized.

• The state variables must be linearly independent.

• In a practical system, the number of state variables should be equal to

the number of independent energy storage elements.

• The number of state variables will be equal to the order of the differ-
ential equations and equal to the number of integrators used.

6.1 State Model Formation

State model of a system can be obtained from general models such as dif-
ferential equation, transfer function, state diagram (block diagram or signal
flow graph).

6.1.1 State model from differential equations

Most physical systems can be expressed in terms of differential equations as
mentioned earlier. The state model can be obtained using these equations.
Consider an SISO system defined by a second order differential equation
d2 y(t) dy(t)
2
+a + by(t) = ku(t)
dt dt
where y(t) is the output and u(t) is the input.
Since order of the equation is 2, this system has to be defined using 2 state

59
variables.
Take y(t) = x1 (t), then
dy(t)
dt
= x1˙(t) = x2 (t) and
d2 y(t)
dt2
= x2˙(t) = ku(t) − ax2 (t) − bx1 (t)
where x1 (t) and x2 (t) are the state variables.

From this, the state equations and output equations can be expressed in ma-
trix
" form # to obtain the 4 defining matrices.
x1˙(t)
    
0 1 x1 (t) 0
˙ = + u(t)
x2 (t) −b −a x 2 (t) k
 
  x1 (t)
y(t) = 1 0 + 0u(t)
x2 (t)
Same concept can be applied to MIMO systems of higher order. Any order
differential equation with any number of inputs and outputs can be repre-
sented in the same manner.

6.1.2 State model from transfer function

Transfer function of a system will be of the form G(s) = Y (s)/U (s).
This can be expressed as G(s) = YX(s)U
(s)X(s)
(s)
Y (s)
where X(s) = n(s) i.e numerator polynomial and X(s)
U (s)
1
= d(s) i.e denominator
polynomial.
Y (s) = n(s)X(s) and U (s) = d(s)X(s).
By taking inverse Laplace transform of the above two expressions, differential
equations are obtained, and that can be used to obtain state model from the
method explained earlier.

For example,
Y (s) s+l
G(s) = = 2
U (s) s + cs + d

=⇒ Y (s) = (s + l)X(s); (s2 + cs + d)X(s) = U (s)

dx(t)
y(t) = + px(t)
dt
dx2 (t) dx(t)
2
+c + dx(t) = u(t)
dt dt
60
 dx(t)
The state variables are x1 (t) = x(t) and x2 (t) = dt

These
" #equations expressed in matrix will give the 4 defining matrices.
˙
x1 (t)

0 1

x1 (t)
  
0
˙ = + u(t)
x2 (t) −d −c x2 (t) 1
 
  x1 (t)
y(t) = l 1 + 0u(t)
x2 (t)
For MIMO systems, a unique transfer function will be defined for each com-
bination of input and output. Similar exercise for each of those transfer
functions provides overall state model.

6.1.3 State model from state diagram

Using the signal flow graph or block diagram, the transfer function can be
found, which can be used to obtain the state model as explained.
Another technique would be to assign state variables in the diagram and get
the state equations. Here, the number of integrators will give the number
of state variables. The outputs of integrators are assigned as state variables
and equations are obtained from incoming branches at each node.

−r

1 y1 p ÿ2 s−1 ẏ2 = x2 s−1 y 2 = x1

u(t)
−q

In the above state diagram (SFG) there are 2 integrators, whose outputs are
taken to be the state variables.
State equations:
x1˙(t) = x2 (t)
x2˙(t) = (p − q)x1 (t) − rx2 (t) + pu(t)

Output equation:
y2 (t) = x1 (t)

61
The above
" # equations expressed in matrix will give the 4 defining matrices.
˙
x1 (t) 0 1

x1 (t)
  
0
˙ = + u(t)
x2 (t) (p − q) −r x 2 (t) p
 
  x1 (t)
y2 (t) = 1 0 + 0u(t)
x2 (t)

6.2 Transfer Function from State Model

Consider a system with the state equations ẋ = Ax + Bu and output equa-
tion being y = Cx + Du (where y is a single output; since different outputs
will have different transfer functions).
Taking Laplace Transform, sX(s) = AX(s) + Bu(s) and Y (s) = CX(s) =
Du(s).
=⇒ X(s) = [SI − A]−1 Bu(s)

Y (s)
∴ = C[SI − A]−1 B + D
U (s)
Characteristic equation of the system will be given by |SI − A| = 0
since the determinant of that matrix is to be divided why finding the inverse.
|SI −A| = 0 will give the locations of the poles, which determine the stability
and characteristic behaviour of the system.

6.3 Controllability and Observability

Controllabability of a system:
A system is said to be completely state controllable if it is possible to transfer
the system from any initial state x(to ) to any desired state x(t) in specified
finite time by a control vector.

A general nth order MIMO LTI system with state equations defined by ẋ =
Ax + Bu and output equations defined by y = Cx + Du is controllable if
and only if the rank of the matrix Qc is n i.e determinant of the matrix Qc
is non-zero.

Qc = [B : AB : .... : An−1 B]

62
Observability of a system:
A system is said to be completely state observable if every state x(to ) can be
completely identified by measurements of the outputs y(t) over a finite time
interval.

A general nth order MIMO LTI system with state equations defined by ẋ =
Ax + Bu and output equations defined by y = Cx + Du is observable if and
only if the rank of the matrix Qo is n i.e determinant of the matrix Qo is
non-zero.
Qo = [C T : AT C T : .... : (AT )n−1 C T ]

6.4 Solution to State Equations

˙ = Ax(t) + Bu(t).
The state equations are expressed as x(t)
The overall response of the system is given by the sum of free response and
forced response.
Free response is due to the initial conditions of the system and forced re-
sponse is due to the input provided.

Free Response (or zero input response)

x(t) = L−1 ([SI − A]−1 )x(0)
where x(0) gives the initial conditions
Forced Response (or zero state response)
x(t) = L−1 ([SI − A]−1 )Bu(t)
Overall Response
∴ x(t) = L−1 ([SI − A]−1 )[x(0) + Bu(t)]
L−1 ([SI − A]−1 ) = eAt = φ(t)
φ(t) is called State transition matrix.
• φ(0) = I
• φ0 (0) = AI
• φ(t1 + t2 ) = φ(t1 )φ(t2 )
• φ−1 (t) = φ(−t)

63
6.4.1 State space transformations
A state space transformation can be obtained using a linear transformation
~
which links the old state vector ~x(t) with the new vector X(t).
~
X(t) = T ~x(t).
Applying this to the state equations and output equations, a new equivalent
set of equations are obtained.
~˙
X(t) ~
= A1 X(t) + B1~u(t)
~y (t) = C1~x(t) + D~u(t)
Where A1 = T −1 AT ; B1 = T −1 B; C1 = CT

By properly choosing matrix T , it is possible to obtain mathematical descrip-

tions of the given system characterized by matrices A1 , B1 , C1 and D which
have particularly simple structures.

For each choice of matrix T , a different but equivalent mathematical descrip-

tion of the given system is obtained. All these different mathematical models
maintain the basic physical properties of the given dynamic system: stability,
controllability and observability.

If the matrix T can be selected such that A1 = T −1 AT is a diagonal matrix,

then T is represented as S and is called the diagonalizing matrix and A1 is
now written as Λ.

6.4.2 Eigenvalues and Eigenvectors

For any system matrix A, there exists some vectors x̄ that satisfies Ax̄ = λx̄
where λ is a scaling factor i.e multiplying the vector with the matrix will give
the scaled version of the same vector.
These vectors are called Eigenvectors and their respective scaling factors
are called Eigenvalues.

This gives (A − λI)x̄ = 0̄.

=⇒ |A − λI| = 0
The above equation gives the characteristic equation of the given matrix and
it is used to find the eigenvalues of A.

64
The matrix Λ obtained by diagonalizing the system matrix A is the eigenvalue
matrix which consists of the eigenvalues as its principle diagonal elements.
The matrix that diagonalizes A i.e the matrix S is formed by placing the
eigenvectors of A as columns of the matrix. S = [x¯1 x¯2 ... x¯n ]

The eigenvalues and eigenvectors provide an alternate method to find the

state transition matrix.
 λt 
e 1 0 ... 0
 0 eλ2 t . . . 0 
eAt = S  . ..  S −1
 
.. ..
 .. . . . 
0 0 . . . eλn t

65