IN 227 Control Systems Design

Lectures 7 and 8

Instructor: G R Jayanth
Department of Instrumentation and Applied Physics
Ph: 22933197
E-mail: jayanth@isu.iisc.ernet.in

Bode Plots

20Log(Mag.)
20Log(Mag.)

20log |K1|

Effect of adding a pole

Log()
Slope=-20dB/decade

Phase (deg.)

Effect of scaling

~0.1
0
-45
-90

~10

Effect of adding a zero

Slope=+20dB/decade

20Log(Mag.)

If C(j)P(j) is the open-loop transfer function, Bode plots jointly refer to two graphs, one
depicting 20log|C(j)P(j)|versus log() (the magnitude plot) and the other depicting
the phase of C(j)P(j) versus log() (the phase plot).
Since Bode Plots would be used to visualize the effect of the design, we briefly recount
the nature of the changes introduced by specific kinds of transfer functions on the Bode
plot of a system.
Summation Property: The open-loop transfer function of a plant commanded by a
controller is C(j)P(j) . Thus, the Bode plot of the open-loop transfer function is the sum
of the individual Bode plots of C(j) and P(j). This is one key reason that has made
Bode Plots such popular tools. It indicates that the bode-plots of the controller can be
separately designed with the full awareness of its consequence on the overall Bode plot.
Effect of a Scaling Operation: Multiplying P(j) by a factor K1 (>0) amounts to vertically
displacing the entire Magnitude plot by an amount 20log |K1| while leaving the phase
plot untouched.
Effect of Adding a Pole: To add a pole at a frequency is to multiply the existing
transfer function P(j) by C1(j)=1/(j/+1). The effect of doing this is to attenuate the
magnitude plot at a rate 20dB/10 units of frequency=20dB/decade starting from =.
Likewise, in the phase plot, the function C1(j) adds -90 over a wide frequency range. For
<< the phase is close to 0. For =, the phase is -45 and for >> , the phase is -90.
It is evident from the summation property that adding multiple poles has the same effect
as the sum the effect of each pole.
Effect of Adding a Zero: To add a zero at a frequency is to multiply the existing
transfer function P(j) by C1(j)=(j/+1). The effect of doing this is to amplify the
magnitude plot at a rate 20dB/10 units of frequency=+20dB/decade starting from =.
Likewise, in the phase plot, the function C1(j) adds +90 over a wide frequency range.
for << the phase is close to 0. For =, the phase is +45 and for >> , the phase is
+90. It is evident from the summation property that adding multiple zeros has the same
effect as the sum of the effect of each zero.

Phase (deg.)

Log()
~10

+90
+45
0

~0.1

Design of One Degree of Freedom Control Systems

We have so far familiarized ourselves with the general constraints associated with the
design of linear control systems due to the possibility of instability. The most
important tool to study stability in the frequency domain is the Nyquist plot. We used
it to learn that stability constraints imposed limits on the bandwidth of the control
systems, demanded the use of dynamic controllers and necessitated performance
specifications for the accuracy with which the control objective x(t)=r(t) is achieved
(constraints that we, control engineers, would like to wish away if we could!). The
most important tool to design a controller that achieves the desired performance
based on frequency domain specifications is the Bode Plot.
General Design Philosophy: The goal of this lecture is outline the process of control
design, and to provide an understanding of what each of the control techniques do,
i.e., explain the usefulness of specific tools such as Proportional/ Integral/ Derivative/
Lead/Lag control techniques. How these tools are to be used in practice, and in what
combination, is a judgment that has to be made by the design engineer once he/she
understands what each of these tools can accomplish. Formulae/recipe oriented
approach to design inhibits understanding and thus, is generally avoided in the
lecture. In building a controller, we seek simplicity and understanding, and refrain
from introducing any more complexity to the controller than the situation demands.
So, in introducing each specific tool in this lecture, we proceed from the simplest to
the more complicated, and at each stage justify why we are forced to choose a more r
complicated design over a simpler one.
One Degree of Freedom Control System is one that possesses just a single controller C(j) to design. Once we have designed C(j) to address a few constraints, we have
no more freedom left to independently address any remaining constraints, especially
those that could not be addressed by C(j).

C ( j)
e

P( j)

The Plant

20log|P(j)|

A plant, in general, refers to the system that we want to control. A well designed
plant would have been constructed so as to minimize the amount of feedback
control necessary. Here we assume that the plant cannot be altered any further to
improve the feedback control performance and the controller has to make do with
what has been given.
Modeling a plant: Mathematically, a linear plant is represented by the transfer
function P(j). P(j) can be obtained in two ways (a) apply a known input u(t), such
as a sinusoid, or a step, obtain the response x(t) and evaluate X(j)/U(j). (b)
Understand the physics of each of the subsystems that constitute the plant and use
engineering intuition to pick those subsystems that are relevant to the overall
dynamics (slowest dynamics are generally the most important because they limit
the overall speed of the system) and mathematically model their resultant
response. In order to gain an understanding of the overall system, it is
recommended to adopt (b) to the extent possible before turning to a more accurate 0dB
model using (a).
General plant: While illustrating the design procedures, we assume that a general
plant can be modeled as P( j) K 0 ( j / z 1 1)( j / z 2 1)..( j / z m 1) where, n>m, and
( j / p 1 1)( j / p 1 1)..( j / p n 1)
z1,.. zm, p1,.. pm,K0>0.
-180
The Bode plot of a general plant looks as shown on the right.

It is desired that the output of the plant track a specified set of reference inputs or
be insensitive to certain disturbances by the specified amount. Our job as control
engineers is to design a feedback network around the plant that ensures that these
specifications are met. The important steps in the design process are outlined first.

Log()

One Degree of Freedom Control Design: Determination of closed-loop

specifications

D
|S|

|T|

R(j)

Control design begins with gathering together, all the requirements of the closedloop system: the desired maximum tracking errors for different inputs and the
desired maximum suppression of typical disturbances that affect the system.
These are typically time domain specifications. However, since all the useful tools
of analysis are in the frequency domain, it is necessary to translate the timedomain specs into frequency domain specs. For example, if the frequency
spectrum of the reference inputs or the output disturbances extend up to
T1
frequencies R and D respectively, it is necessary to achieve T=CP/(1+CP) to be
close to 1 and S=1/(1+CP) to be close to zero for frequencies up to R and D
respectively. Their exact values depend on the permissible tracking error or
desired rejection of disturbance.

However, since the exact spectrum of the inputs or disturbance are specific to
particular applications, it is desirable to choose a standard input whose response
can be employed to quantify the performance of a general control system.

A little thought reveals that a step input is an ideal candidate for such a standard
input: since its magnitude does not change beyond the instant of application, it
clearly showcases the transient response of the plant itself. Likewise, since it
1
possesses a non-zero steady-state value, it showcases steady-state errors in the 0.9 k
response, if any.

While a lot of features can be defined on the step response of the closed-loop
tr
system, and guidelines can be derived to relate them to closed-loop frequency 0.1 k
domain requirements on the transmission function T=CP/(1+CP), we shall choose
to restrict ourselves to defining three features. They are:
Rise time (tr): Time taken for the response to rise from 10% to 90% of its steady-state
value
Peak overshoot (Mp): largest overshoot of the response beyond its steady-state value.
Steady state error (ess): The difference between the reference(1) and output (k) after
the transients have disappeared, i.e., ess=1-k.

D(j)

S0

Mp
ess

One Degree of Freedom Control Design: Determination of open-loop

specifications

Since we have not yet designed the controller C, it is not possible to exactly relate
T
the time-domain closed-loop specifications on the step response described earlier
to the closed-loop transfer function T(j). However, reasonable approximations can
Mr
be made by employing the notion of dominant poles: in general the closed-loop
k
system is assumed to have at least one pair of complex conjugate poles, and that
0.707 k
this pair is assumed to be closer to the imaginary axis than all the others. Thus, the
closed-loop system is assumed to look very much like a second order system.
The steady-state error is related to the DC value of the transmission function, i.e.,
T(0)=k=1-ess.
For a second order system, the rise time is indicative of the speed of response and is
therefore related also to the closed-loop bandwidth b. It is given by t 0.4 2.5
r

fb

The peak overshoot Mp is related to the amount of damping in the closed-loop

system. In the absence of any damping the overshoot is 100%. For more reasonable
cases, the relationship between Mp and the damping factor is given by M p exp(

1 2

Convert time domain closedloop requirements to closed

loop frequency domain
requirements

In an underdamped system (<1), the response oscillates before it dies down. The
oscillation frequency is close to the resonant frequency r of the system. The

resonant frequency itself is close to the closed-loop bandwidth and is given by r b

Convert closed loop

frequency domain
requirements to open-loop
frequency domain specs

Once we have gathered the closed-loop frequency domain requirements, we derive

the corresponding open-loop specifications. The flow chart for the entire design
process is given on the right.
From the equation T=CP/(1+CP), it is seen that the steady-state error is related to
ess 1
the loop gain L(j0)=CP(j0) as

Perform design using Bode

plots and study the response
to a standard (step) input

1.5

r b

CP( j 0)

ess

Study the response to the

actual inputs/ disturbances
Stop

20log|P(j)|

The Plant and the Controller

Gain cross-over
frequency

From the point of view of stability, there are four variables of interest in the Bode
plot of a plant (or more generally, the open-loop transfer function C(j)P(j)).
The Gain Cross-Over Frequency is the frequency at which the magnitude graph
Gain
Log()
of 20log|C(j)P(j)| crosses the 0dB line. The gain cross-over frequency is
margin
related to the closed-loop resonant frequency as gc r. The Phase Cross-Over
Frequency is the frequency at which the phase of C(j)P(j) crosses -180.
Their significance becomes evident in the Nyquist plot of the open-loop system.
Phase margin
The Nyquist plot of C(j)P(j) is plotted on the right. Since the Nyquist plot does
-180
not encircle -1, we know that the system is stable. The effect of C(j) is to alter
Phase cross-over
the magnitude of P(j) and/or its phase. It is seen that if C(j) scales up the
frequency
magnitude by a value more than 1/|OM| then the system will encircle -1 and
Nyquist plot
thus, become unstable. Likewise, if C(j) adds a phase less than , again the
system becomes unstable. OM is equal to|C(j)P(j)| of the open-loop system
Unit circle
at the phase cross-over frequency. -180+ is the phase of the system at the gain
cross-over frequency. Thus, for the sake of stability, OM<1 and >0. -20log|OM|
-1
M O
is defined as the gain margin and is defined as the phase margin. If the gain
margin and the phase margin are both greater than zero, then the closed loop
system is stable. If one of them is less than 0, the closed loop system is unstable.
C(j)P(j)
Significance of Phase Margin: For closed-loop damping factors <0.6, there exists
a proportional relationship between the open-loop phase margin and the
closed-loop damping ratio and is given by =/100, when is expressed in
Dominant pole
Dominant pole
degrees. Thus, specifying the phase margin in the frequency domain is
locations for >0 locations for =0
equivalent to specifying a particular damping ratio in the time domain. Note
that this proportional relationship holds for second order systems. For higher
order systems, the relationship holds to the extent that a pair of complex
conjugate poles are the dominant closed-loop poles of the system.

Goals of Design of the Control Systems

Achieving the control objective: Before we plunge into investigation of the

specific design tools, it is good to remind ourselves what we expect from
the design. The goal of a control engineer is to ensure x=r. The block
diagram and the associated input-output relationship on the right indicates
that this is achieved when |C(j)P(j)|>>1, i.e., when loop gain is high.
Thus, if the plant does not already possess sufficiently high gain, our
controller should provide it. However, we learnt from Nyquist analysis that
we are limited in our ability to supply arbitrarily high gain when the phase
of the plant is close to -180. Thus, our best bet to achieve high loop gain,
given the plant on the right (below), is at low frequencies when the plant
gain itself is maximum and the phase is close to zero. At higher frequencies,
the attenuating plant gain and the specter of instability force us to be more
conservative in increasing loop gain. This makes us less of control engineers
and more of stability engineers at these frequencies. As stability engineers,
we want to make sure that the antics we play out to achieve high gain at
low frequencies do not spill over too much into the high frequency region
and destabilize the overall system.
Phase margin specification: It was evident from the discussion on phasemargins that ensuring sufficient phase-margin for the open-loop system is
equivalent to ensuring sufficient damping for the closed-loop system, which
is clearly a desirable trait. Thus, we decide that our open-loop systems must
have a predetermined satisfactory phase margin at least of 0.
The plant: In order to throw full spotlight on the controllers and not on the
plant, we assume that our plant has no specific characteristic that demands
special attentioni.e., it is very normal. Its parameters are assumed to be
fixed and its phase margin is greater than 0. (Fig. on the right).

C ( j)

G( j)

1
1
1
C ( j)G ( j)

Region of interest for

control purposes

Region of compromise in
control objective for
stabilitys sake

20log|P(j)|

0dB

-180

Log()

20log|P(j)|

Tools for Design of the Control Systems

Tool#1 The Proportional Controller: A proportional controller is a simple gain K1. It
pushes up the magnitude plot but leaves the phase plot untouched.
It is evident from the figure on the right that the best we can do to increase the loop
gain using this strategy is by the amount K10 for which the system phase margin at
0dB
the gain-cross-over frequency is 0. Any larger gain and the system phase margin
would be less than 0, thereby leading to excessive transients.
Thus, the improvement in loop gain is rather modest over all frequencies and thus, at
no frequency can the controller engineer achieve adequately high gain. For this
-180
reason, a plain proportional controller is not an attractive design choice.
There is one case where proportional controllers can be used: if the plant is of first
order, then its phase does not dip below -90 and thus never crosses 0. For this case
alone proportional controllers work very well. Of course, no plant is strictly first order.
However, many systems, particularly thermal systems, possess their higher poles
much farther away than the first pole and therefore resemble first order systems.

20log |K10|

Highest
possible gain
cross-over
frequency

Log()

Tools for Design of the Control Systems

Tool#2 The Integral Controller: A integral controller has a transfer

function K1/s. The figure on the right is a Bode plot of an integral
controller.
Magnitude
0dB
Log()
It is evident from the bode plot that the really useful range for this
type of controller is at low frequencies, when its gain is large. Indeed,
its gain tends to as 0. Thus, we can achieve close-to-perfect
-90 Phase
control around =0.
On the other end, it is not at all useful at high frequencies. As seen
from the figure below, it drastically reduces the gain cross-over
frequency. Furthermore it adds a constant phase of -90 at all
Highest
frequencies thereby reducing the phase cross-over frequency as well.
possible gain
cross-over
If the >0 at the new gain cross-over frequency, then we have
20log |K10|
frequency
excessive phase margin, which may be unnecessary and too
conservative. Thus, the gain of integral control can be scaled up at
every frequency by an amount K10 by using a proportional controller,
New gain cross-over
such that the final phase-margin just satisfies the minimum
Original plant
specifications. If <0 then the proportional controller should scale
0dB
Log()
down the gain until the desired phase-margin is achieved.
An integrators performance can be safely declared to be adequate
only around =0. At other low frequencies, it continues to be valuable
-90
in terms of increasing the gain, but if a specified minimum gain needs

to achieved at a particular frequency that is higher than what the

-180 0
integrator can supply, the demand cannot be met. This is because we
have only one parameter to tune, viz., K1, and this gets fixed by the
need to maintain sufficient phase margin.

Tools for Design of the Control Systems

Tool#3 The Proportional-Integral Controller: It is evident from the previous

Magnitude
slide that the use of an integral controller at high frequencies runs against the
0dB
Log()
credo of control engineers since this results in deliberate attenuation of the
loop gain.
Phase
The frequency at which this tendency begins is when the controller crosses the
-45
0dB line (if K1=1, this happens at =1rad/s). Therefore, it would be wiser to
-90
curb the tendency beyond this frequency by adding a zero at the frequency
where integral control is no longer useful for increasing the loop gain. The
s 1
resulting Bode plot is shown on the right. In general it has a transfer function C (s) K1
s
The Bode plot reveals another unintended attraction of cutting of integratorcharacteristics at high frequencies. The phase of the controller returns to 0.
This tends to restore the phase-cross-over frequency to that of the original
plant. Indeed, since the controller also has unity gain at high frequencies, it is
as though there is no specific form of control for the plant at frequencies >.
Having recovered the original plants dynamics at mid-and high-frequencies,
Highest
possible gain
we can inspect to see if the phase margin >0 . If yes, we can crank up the
cross-over
gain (K1) to a value at which the phase margin just equals 0.
frequency
20log |K10|
However, if <0 , we have two options: the first option is, as before, to crank
down the gain until the phase margin increases to 0. However, there is a
possibility that the poor phase-margin is because the controller phase was not
restored to 0 early enough and the residual negative phase has leaked into
0dB
Log()
the high frequency region and reduced the phase margin of the overall system.
Thus, the second option is to shift the zero to the left so that the negative
phase introduced by the controller returns to zero at a much lesser frequency.
-90
-180

Tools for Design of the Control Systems

Tool#4 The Proportional-Integral-Derivative Controller: The best loop gain that

we can achieve with a PI controller is shown on the right. Suppose we desire an
even higher gain in the mid frequency region than what is supplied by the best
PI-design, our only option is to increase the proportional gain K1 even further
until the desired loop gain is achieved.
However, in increasing the gain, we pay a heavy (and possibly fatal) price in
0dB
terms of the phase margin. We see from the schematic that our phase margin
(the new gain-cross-over frequency is named 0) has not only reduced below 0
but could have possibly become negative. This implies that for this gain, the
closed-loop system is unstable.
-90
-180
In this situation, we desire a controller that adds positive phase so that we can
pull the phase curve back up and achieve a phase margin 0. Furthermore, we
desire that the controller does not add any gain since any additional gain will
push the gain cross-over frequency even more to the right and make the phase
margin even lesser.
We notice that if we add a zero at to the controller situated slightly to the
0dB
right of 0, then we can add a phase of approximately +45 at 0 but no gain
(Fig. on the right). This is exactly what we want!
If we include this new transfer function to the original PI-controller design, the
overall transfer function looks as follows: C (s) K (s 1) ( s 1)
1

Desired mid-frequency gain

Best possible
PI-design

0
Log()

=0

<0 !!

Magnitude

If this is expanded out we see that C (s) 1 s K1 ( ) K1 . Since it can be

s
expressed as a summation of proportional, integral and derivative controllers
this is called Proportional-Integral-Derivative (PID) control.
Of course we cannot add a stand-alone zero at because we would have a non( s 1) ( s 1)
causal transfer function. So, we add a far-off pole at >> 0. Thus, in practice C (s) K1
s ( s 1)

Log()
+90

+45

Some notes on PID Controllers

The overall Bode Plot of the Loop Gain shaped by means of a PID controller
looks as shown on the right.
The role of each control is specific and independent of the others. The
Integral-controller is meant to achieve very high gain at very low frequencies,
thereby ensuring perfect input-tracking and/or disturbance rejection at
these frequencies. Its negative effects on high-frequency range is minimized
by adding a zero around the frequency at which its gain dips below 0dB. At
the mid-frequencies, compromise in the achievable loop gain is inevitable
and the loop gain is in general much less than that at low frequencies.
Nevertheless, we try to maximize the mid-frequency gain to the extent
possible by using Proportional-control and thereby also extend the control
bandwidth. Finally comes the derivative part of the controller, which does
not worry about the gain (its low/mid-frequency gain is 1). Its prime concern
is to ensure stability. Its role is to add phase alone near the cross-over
frequency and thus improve the phase margin.
Conventionally, PID controllers are introduced as though they are designed
using the Zeigler-Nichols tuning rules. However, these rules are meant for a
specific type of plant model P(s)=K0e-sT/(s+a), which happens to be a suitable
approximation for most plants found in industries. For other plants their
performance is sub-optimal.
PID controllers are today the most prevalent type controllers. However, this
is not because of the enduring relevance of Zeigler-Nichols (Z-N) tuning
rules. Rather, it is because PID represents a family of controllers (and not a
specific set of control gains according to Z-N prescription) whose general
characteristics enable achieving high loop gain in every relevant part of a
plants frequency response while simultaneously addressing the stability
constraint.

Desired lowfrequency gain

Desired midfrequency gain

0
0dB

-90
-180

Log()

=0

Desired highfrequency phase

Tools for Design of the Control Systems

Having briefly paused to highlight the significance of PID control, we now cover
two more control schemes that derive inspiration from PID based control.
Tool#5 Lag Compensator: The input to any controller cannot be arbitrarily large.
It eventually has to saturate at a certain value. If this happens for a proportionalintegral controller, we see that the output of the integral controller, due to a
saturated input, keeps on building up until feedback pulls the error back into the
linear region. This phenomenon is called wind-up. While anti-wind up straegies
exist, it can be altogether avoided if the output of the controller, to a constant DC
input, does not build up. Thus, in order to curb the characteristics of 1/s at DC,
we replace it with a very low bandwidth low-pass filter 1/(s+). For >, this
function behaves exactly as an integrator. However, for <<, it outputs a
constant gain, independent of frequency. This is called a lag compensator. Its
transfer function is
s 1

C ( s) K1

The bode plot of a lag compensator is shown on the right. It is intended to mimic
PI-control. Thus, >>. However, having invented it, we see that a lag
compensator can also be used for other reasons than just as replacements for PI
control. In particular, we can use it to add gain below any specified frequency .
The amount of gain we add is determined (and if the freedom exists, by K1).
Thus, as an example of the use of a lag compensator, if we cannot achieve
sufficiently high gain with an integrator at a particular low frequency (0), we
can introduce a lag compensator whose corner sufficiently to the right of so
that the extra gain necessary is added. Likewise, if we do not desire perfect
tracking at DC, we can altogether replace PI control by a lag compensator.
The lag compensator gets its name from the fact that its phase is always
negative.

0dB

Log()
C ( s) K1

Magnitude

s 1
s
Phase

Min. Phase>-90

Tools for Design of the Control Systems

Tool#6 Lead Compensator: In the case of PID control, we added a zero at in

order to add phase. To ensure causal implementation, we added a pole at >>.
Between frequencies and , the controller behaves as a differentiator. Thus, if
is excessively large, our controller would be rendered extremely susceptible to
high frequency noise. Instead, if we include also as a design parameter and
limit its value in order to cutoff the derivative action early enough, we get a lead
compensator. The general transfer function of a lead compensator is C ( s) s 1
s 1

The bode plot of a lead compensator is shown on the right. The purpose of a lead
compensator is exactly the same as that of its equivalent in PID controller: it is to
add phase near the gain cross-over frequency. It is called a lead compensator
because its phase is always positive.
As with PID control, lead lag controllers operate at two different frequency
regimes and achieve two different goals. For the most versatile design (i.e., one
that achieves the maximum gain bandwidth), it is suggested that both lag as well
as lead compensators be used.

0dB

Log()

Magnitude

Max. Phase<+90

Phase

C (s)

s 1
s 1