1

Lecture 12/4-6/19 0.3

Impulse Response of G p(s)

In this lecture we investigated the situation where the plane was 0.25

statically unstable, but had a stable Dutch roll mode. The transfer
0.2
function for this situation was chosen to be:

Amplitude
0.15

0.1
. (1)
0.05

The impulse response is shown at right. Clearly, the system is

0
unstable. 0 2 4
Time (seconds)
6 8 10

Figure 1 Plot of .

Next, we considered the use of unity feedback proportional control to stabilize

k G p (s)
(1). The block diagram is shown at right. In class it was shown that the closed  d (t )  (t )
loop transfer function associated with Figure 2 is: 

. (2) Figure 2 Closed loop block diagram.

The system (2) will be stable only if all three roots of are in the proper Left Half Plane
(LHP). To find the range of stabilizing K-values, one could use the Routh array:

, where .

The first column of the array will have no sign changes for the K-values . Hence, this is the range of
stabilizing K-values. Impulse Response of W(s)for K=15
2

However, the fact that (2) is stable for does not mean that 1.5

it is desirable. Consider the impulse response shown at right for .

1
The response is stable, but totally unacceptable.
Amplitude

0.5

The use of the Routh array, as valuable as it is, is an exercise in algebra.

0
To gain a more visual appreciation, we will now address the use of the
root locus; which is a geometric approach. -0.5

-1
0 5 10 15 20 25
Time (seconds)
Figure 3 Closed loop impulse response for .

We are interested in the CL system roots that are the solution to:

. (3)
 2

For any chosen value of K, (3) will have three roots. Hence, as we vary K, there will be three trajectories of roots in the
complex plane. These trajectories are also called loci; hence, the term root locus.

Now, to arrive at the root locus associated with (3), we could simply use the ‘roots’ command, as you have done in past
homeworks. However, this is what I would call a ‘turn the crank’ method. It lends little geometric insight as to what’s
really going on. We will now approach the problem of constructing the root locus for (1) in a less rote manner. More
importantly, this approach will lead one to naturally posit control structures that can improve matters.

To this end, note that (3) is equivalent to:

. (4)

The only caveat in re-casting (3) as (4) is that (4) is only well-defined for s-values that are not the roots of
. For any one of those three roots, (3) will be zero if and only if .

Now, recall that is the denominator polynomial of the plant transfer function (1). However,
in view of Figure 2, (1) is also the open loop transfer function; i.e. that transfer function one arrives at by opening the
feedback loop at the summing junction, and then gathering up all the transfer functions in that loop. Hence, the subscript
OL in . In gathering up the transfer functions, we arrive at the open loop transfer function:

. (5)

From (5), we see that in this example (4) becomes:

. (6)
Root Locus 30
This leads us to the Matlab command ‘rlocus’. While it is natural to assume 30
0.66 0.52 0.4 0.26 0.12 25

that the required argument for this command would be the polynomial (1), 20
0.8 20
15
whose root locus is desired. However, generally, such a polynomial could
Imaginary Axis (seconds-1 )

0.9
10
10

problematic to express as a function of K. On the other hand, (4), which is 0.97 5

0
equivalent to (3), is not at all problematic. We simply give as the
0.97 5
argument of ‘rlocus’. The commands below resulted in the root locus plot in -10
0.9
10

Figure 4: -20
15

20
0.8

0.66 0.52 0.4 0.26 0.12 25

-30
G=tf(25,[10 19 248 -25]); rlocus(G) grid -30 -25 -20 -15 -10 -5
Real Axis (seconds -1)
300 5 10 15

Figure 4 Root locus associated with (3).

The x’s in Figure 4 are the roots of (3) when . These roots are exactly the poles of the open loop transfer function.
As K is increased, the roots of (3) depart from the OL poles in the manner shown in Figure 4. Specifically, the real root
locus begins at and moves toward as K is increased. The other two roots that are a complex conjugate
pair for retain their complex conjugate structure as . Moreover, they enter the RHP.
 3

Root Locus
We knew that this would be the case from analysis of the first 6
0.23 0.16 0.1 0.05
6

column of the Routh array, since, for sufficiently large K there are 5
5

System: G
two sign changes in the first column. Furthermore, the zoomed plot

Imaginary Axis (seconds -1)

0.32 Gain: 19.9 4
4
Pole: -0.000192 + 4.98i
of Figure 4 given in Figure 5 verifies the Routh array result that all 3
Damping: 3.86e-05
Overshoot (%): 100
3
0.44
the roots of (3) will be in the LHP for . Frequency (rad/s): 4.98
System: G
2
2 0.6 Gain: 1
Pole: -0.000281
Damping: 1 1
1
Specifically, we see that the real root enters the LHP for , and 0.84 Overshoot (%): 0
Frequency (rad/s): 0.000281

the conjugate pair of roots enter the RHP for . 0

-1 0.84
1
-1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4
-1
Real Axis (seconds )

Figure 5 Zoomed plot of Figure 4.

Now, let’s go a little deeper into the rabbit hole . Write (6) as:

. (7)

Furthermore, write the OL transfer function (5) as:

. (8)

Im s
To connect (8) to (7), consider the generic term . This is the difference between
s p
two vectors (i.e. numbers in the complex plane), as is shown at right. We can write
this difference vector in polar form as:
p
. (9) Re
Figure 6 Plot of .
From (9), we can write (8) as:

. (10)

Finally, equating (7) and (10) gives:

. (11)

From (11) we see that two conditions are required for s to be on the root locus:

(C1) The magnitude condition: and (C2) The angle condition: . (12)
 4

The condition (C1) is trivial, in the sense that for any s, regardless of whether or not it is on the root locus, we can satisfy

this condition by setting . It is (C2) that determines the geometric structure of the root locus.

In-Class 1: Use (C2) to prove that is not on the root locus of Figure 4.

In class we computed . In words, the ‘defect angle’ associated with is -340o.

In-Class 2: What would need to be added to so that is on the root locus?

We could add zeros to such that the angles between them and add up to 340o.