4-23-2010

Frequency Response Example:

& 12 #
For the Open-Loop Transfer Function KGH = K $ !
% (s + 1)(s + 2 )(s + 3)"
At zero frequency, this system has a DC (steady-state) gain of
& 12 #
GH( 0 ) = $ !=2
% 1* 2* 3 "
In a decibel (dB) scale, this gain is expressed.
GH (0) = 20 log10 (2 ) = 20( 0.3010 ) = 6.02 dB
The two are equivalent but one is on a linear scale and the other on a logarithmic scale

Note: For the record, “deci-Bel” dB is the 10*log( power ratio)

& P # & 2 # & 2 # ,& #
2
)
dB = 10 log10 $ ! = 10 log10 $ ky ! = 10 log10 $ y ! = 10 log10 *$ y ! '
$P ! $ ky 2 ! $ y2 ! *$% y ref ! '
% ref " % ref " % ref " + " (
& y #
= 20 log10 $ ! = 20 log10 (y ) for y ref = 1 because log10 (1) = 0
$y !
% ref "

The Root Locus showing the closed-loop pole locations is

2.5

1.5

0.5
Imag
Axis 0

-0.5

-1

-1.5

-2

-2.5
-4 -3 -2 -1 0 1 2
Real Axis

The Open-Loop Frequency Response (Bode Diagram) is plotted with the command

EDU» num=[12]; den=conv([1 1],conv([1 2], [1 3])); G=tf(num,den)

EDU» bode(G)

1 of 11
 4-23-2010

And the Gain and Phase margins are computed with the command
EDU» margin(G)

Yielding the plot...

Bode Diagram
Gm=13.979 dB (at 3.3166 rad/sec), Pm=75.636 deg. (at 1.2232 rad/sec)
20

-20
Magnitude (dB)
-40

-60

-80

-100

-50

-100
Phase (deg)
-150

-200

-250

-300
10-1 100 101 102

Frequency (rad/sec)

This plot show a Gain Margin (GM) = 13.979 dB and a Phase Margin of 75.636 degrees.

The Gain Margin indicates that the control gain can be increased by 13.979 dB
(10(14/20) = 5) from K = 1 = 0dB before the system will go unstable.

2 of 11
 4-23-2010

The Phase Margin indicates that at the current gain (K = 1), the system can absorb an
addition phase lag of 75.636 before it will go unstable.

For the system

& 12 #
KGH = K $ !
% (s + 1)(s + 2 )(s + 3)"
the GM=14. dB and Phase Margin = 76 degrees with K=1indicate a very stable system.

The step response for the Closed-Loop system

12
T= 3
s + 6 s + 11s + 18
2

confirms this prediction.

Step Response
0.9

0.8

0.7

0.6

0.5
Amplitude
0.4

0.3

0.2

0.1

0
0 1.6 3.2 4.8 6.4 8
Time (sec.)

Although the step response is stable, the steady state error is 33%. Could we reduce it by
increasing the gain K from K = 1 (0 dB) to K = 2 (6.02 dB) ? Yes, just add 6.02 dB to the
frequency response shift it up 6.02 dB at all frequencies. Let’s plot it with the Matlab
commands

EDU» num=[24]; den=[1 6 11 6]; G=tf(num,den);margin(G)

The new open-loop transfer function

24
G2 ( s) =
s + 6 s + 11s + 6
3 2

has a DC gain of G2 ( s ) = 4 (= 12 dB) reducing the error to e(!) = 1 (1 + 4) = 0.2 = 20%

but sacrifices stability with the new GM=8 dB and PM = 35 degrees.

The system’s closed-loop transfer function

24
T2 =
s + 6 s + 11s + 30
3 2

3 of 11
 4-23-2010

has the predicted, less stable response with better steady-state error.

Bode Diagram
Gm=7.9588 dB (at 3.3166 rad/sec), Pm=35.425 deg. (at 2.0639 rad/sec)

50

Magnitude
(dB)

-50

-100

-50

-100
Phase
(Deg)
-150

-200

-250

-300 -1 0 1 2
10 10 10 10
Frequency
(rad/sec)

StepResponse
1.4

1.2

Amplitude 0.8

0.6

0.4

0.2

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

4 of 11
 4-23-2010

More Advanced Control Design:

Our system still has inadequate steady-state accuracy – even at the high gain (K=2) where
the limiting gain and phase margins are achieved. To increase steady-state accuracy,
apply an integral control.

Integral Controller:
The transfer function of an integral controller is:
Gc ( s ) = K s
The frequency response of this controller is plotted using the Matlab command

and is shown below for K=1.

Adding the magnitude and phase of the Integral controller to the magnitude and phase of
the original controller can be done either graphically or analytically. The result is the
Bode diagram of the open-loop transfer function
& K #& 12 # & 12 #
KGH = $ ! $ ! = K$ !
% s " % (s + 1)(s + 2 )(s + 3)" % s (s + 1)(s + 2 )(s + 3)"
of the system shown in the figure below.

5 of 11
 4-23-2010

R (s ) E (s ) K A(s ) 12 C (s )
s (s + 1)(s + 2)(s + 3)

Closed Loop system with Integral Controller

The Bode diagram is …

, K=1

The resulting system has a Gain Margin, GM = -1.6 dB and the Phase Margin,
PM = -7 degrees so it is unstable. To stabilize the system at an acceptable GM = 8 dB
and PM = 45degrees requires a reduction in integral control gain by a factor of at least
8+1.6 = 9.6 dB to meet the gain margin requirement. The new control gain

K = !9.6 dB = 10 ( !9.6 / 20) = 0.33

The resulting bode diagram has a GM=8 dB and a PM=36 degrees and is shown below.
This reduced gain now meets the gain margin (GM) requirement but does not achieve the
required phase margin, PM = 45 degrees.

6 of 11
 4-23-2010

To achieve the require phase margin, decrease the gain K further. How much? Look at
the Bode Diagram for K = 0.33and make an estimate. The 0 dB crossing point needs to
move from about ! = 0.55 to about ! = 0.42. This requires a further magnitude
adjustment of about 3 dB = 0.7. Try K=0.33*0.7 = 0.23. Matlab Computes this system’s
GM = 11.2 dB and PM = 48. degrees. The 0 dB crossing shows a bandwidth of about 0.4
rad/sec yielding a time constant of approximately ! = 2.5 seconds.

7 of 11
 4-23-2010

The system integral controller design is Gc = 0.23 s and has the closed-loop step
response given by the closed-loop transfer function

T (s) =
Gc G p
=
(0.23 *12)
1 + Gc G p s 4 + 6 s 3 + 11 s 2 + 6s + (0.23 * 12 )
This system has a steady-state gain T (0) = 1 and no steady-state error at any gain because
the open-loop system is type 1. The closed-loop system step response is given below.

The system’s overall time constant is slower than predicted, Ts approximately 20 seconds
yielding an equivalent time constant ! = about 5 seconds.

Proportional-Derivative Control:
A Proportional-Derivative (PD) control generates a “lead” action. This control will both
increase the speed of response and further stabilize the system.
& K #
K p + K d s = K d $$ s + p !
Kd !
% "
Propose a controller with a “zero” at K p K d = ! = 0.4 rad/sec. This additional control
makes the controller transfer function

Gc = K
(s + 0.4)
s

8 of 11
 4-23-2010

yielding the Open-Loop system transfer function for the P-D plus Integral controller
& (s + 0.4 )# & 12 # & 12( s + 0.4) #
KGH = K $ $ ! = K$ !
% s " % (s + 1)(s + 2 )(s + 3)" % s (s + 1)(s + 2 )(s + 3)"
!
The Bode diagram (open-loop frequency response) of this system with the modified
controller for K=1 shows a decrease in phase lag about the zero “corner” frequency.

This new “P-D+I” controlled system has a gain margin GM = 11.9 dB and a phase
margin PM = 55 degrees while retaining zero steady-state error. Furthermore, the
bandwidth of the system ! = about 1.3 rad/sec (look at the 0 dB crossing) yielding an
expected time constant ! = about 0.76 seconds and a settling time Ts = about 3 seconds.
It should be substantially faster than with only “I” action. The step response for the
closed-loop system

T (s) =
Gc G p
=
(1.0 *12)(s + 0.4)
1 + Gc G p 4
s +6s 3
+ 11 s 2 + 6s + (1.0 * 12 )(s + 0.4 )
12s + 0.48
=
s 4 + 6 s 3 + 11 s 2 + 18s + 4.8

verifies this expectation.

9 of 11
 4-23-2010

The result is a closed-loop system with settling time Ts = 10 seconds, corresponding to a

time constant ! = 2.5 sec. The system has about 10% overshoot because we designed it
with plenty of gain and phase margin. This compares very favorably to the “Integral
only” control that had a settling time of about 20 seconds. The “PD” action has
decreased the time of response of our system substantially. Looks like a good controller.

You should see that although we have used the analytical model to explain what is
happening and allow easy Matlab plotting of results, WE DID NOT NEED THE PLANT
MODEL to do the design. In fact, the use of a Bode diagram allows the simple addition
of the controller frequency response to a measured plant frequency response.
Understanding with a transfer function model is important BUT you can simply add the
controller transfer function to a measured plant transfer function if you do not have a
transfer function model available.

A Final Note:

The addition of the “Proportional-Derivative” term to the Integral controller meant that
we did not need to add a “pole” to the PD controller to implement it. If we wanted to add
a “PD” term to the original system, we would have had to use something like

10 of 11
 4-23-2010

' K $
%% s + p "
Kd "
Gc = K d & #
(s + ! LP )
where ! LP is some frequency above the frequency of the “zero” K p K d = ! = 0.4
rad/sec. I might try something like ! LP = 4 rad/sec.

Gc = K
(s + 0.4)
(s + 4.0)
which has the Bode diagram,

This diagram shows explicitly the phase lead generated by the pole-zero pair on the real
axis with the pole to the left (high frequency ! = 4.0 rad/sec) of the zero (low
frequency ! = 0.4 rad/sec). This figure also shows the 20 dB = 10 (linear) magnitude
reduction that results at zero frequency normally causing reduced system steady-state
accuracy through the resulting larger steady-state error.

11 of 11