EE3202B Control Systems

Lecture 5: Stability of LTI Systems

Sei Zhen Khong

National Sun Yat-sen University

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 1 / 28
Part I
Disturbance Rejection and Internal Stability

Part II
Root Locus

This lecture covers material in Chapters 3.6, 4.1, 5.1, 5.2, 5.3 of the textbook.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 2 / 28
Feedback control

Preliminary examples have demonstrated that feedback can:

reduce the effect of disturbances
decrease sensitivity to model errors
stabilize an unstable system.
However, if applied incorrectly feedback can:
make a previously stable system unstable
add oscillatory behavior to a previously smooth response
result in high sensitivity to measurement noise.
Analysis of the control loop can be done to obtain the desired behavior while
avoiding the unsavory aspects of feedback. The next few lectures will introduce
some useful analysis techniques.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 3 / 28
A typical feedback loop
di x0 do
+ +
P u P P
r C (s) G (s) y
−

+
dm
x0 : initial condition
di : input disturbance
do : output disturbance
dm : measurement disturbance

C (s) C (s)G (s) C (s)

U(s) = R(s) − Di (s) − Do (s)
1 + C (s)G (s) 1 + C (s)G (s) 1 + C (s)G (s)
C (s) C (s)
− Dm (s) − F (s; x0 )
1 + C (s)G (s) 1 + C (s)G (s)

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 4 / 28
Degree of freedom

C (s)G (s) G (s) 1

Y (s) = R(s) + Di (s) + Do (s)
1 + C (s)G (s) 1 + C (s)G (s) 1 + C (s)G (s)
C (s)G (s) 1
− Dm (s) + F (s; x0 )
1 + C (s)G (s) 1 + C (s)G (s)

The system response to the reference input r (t) is shaped by the choice of
C (s).
However, once C (s) is fixed, The system response to the disturbances is
determined. Thus there is only one degree-of-freedom here, i.e., the choice of
C (s).
One can have an additional freedom by using a reference filter. This improves
our ability to shape the system response to the reference input r (t).

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 5 / 28
Two degree-of-freedom systems
di x0 do
+P +
r˜ P u P
r H(s) C (s) G (s) y
−

+
dm
H(s)C (s)G (s) G (s) 1
Y (s) = R(s) + Di (s) + Do (s)
1 + C (s)G (s) 1 + C (s)G (s) 1 + C (s)G (s)
C (s)G (s) 1
− Dm (s) + F (s; x0 )
1 + C (s)G (s) 1 + C (s)G (s)
H(s)C (s) C (s)G (s) C (s)
U(s) = R(s) − Di (s) − Do (s)
1 + C (s)G (s) 1 + C (s)G (s) 1 + C (s)G (s)
C (s) C (s)
− Dm (s) − F (s; x0 )
1 + C (s)G (s) 1 + C (s)G (s)

Design C (s) so that the disturbances are sufficiently suppressed; then design
H(s) to make the response to r (t) satisfactory.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 6 / 28
Terminology

C (s)G (s)
• T (s) = complementary sensitivity function
1 + C (s)G (s)

1
• S(s) = sensitivity function
1 + C (s)G (s)

G (s)
• Si (s) = input-disturbance sensitivity function
1 + C (s)G (s)

C (s)
• Su (s) = control sensitivity function
1 + C (s)G (s)

Y (s) = T (s) (R(s) − Dm (s)) + Si (s)Di (s) + S(s) (Do (s) + F (s; x0 ))

U(s) = Su (s) (R(s) − Dm (s) − Do (s) − F (s; x0 )) − T (s)Di (s)

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 7 / 28
Algebraic relationships

The sensitivity functions are related as follows:

T (s) + S(s) = 1
Si (s) = S(s)G (s) = T (s)/C (s)
Su (s) = S(s)G (s) = T (s)/G (s)

These relationships can be used to shape the response so that disturbances are
attenuated while the response to the reference is accentuated.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 8 / 28
Internal stability

Definition: The feedback system shown on previous frames is said to be stable if

all the sensitivity functions are stable

B(s) P(s)
Let G (s) = , C (s) = . Suppose A(s) and B(s) do not have
A(s) L(s)
common roots. Suppose P(s) and L(s) do not have common roots. Then the
poles of any of the sensitivity functions are the roots, or a part of the roots,
of the characteristic equation

L(s)A(s) + P(s)B(s) = 0

For all sensitivity functions to be stable, the roots of

L(s)A(s) + P(s)B(s) = 0 must reside in the open left-half complex plane.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 9 / 28
Example: unstable pole-zero cancelation

3 −s + 2
Consider a feedback system with G (s) = , C (s) = . Note
(s + 4)(−s + 2) s
that when forming C (s)G (s), an unstable pole of G (s) is canceled by the unstable
zero of C (s).

C (s)G (s) 3
T (s) = = 2 stable
1 + C (s)G (s) s + 4s + 3
1 s 2 + 4s
S(s) = = 2 stable
1 + C (s)G (s) s + 4s + 3
G (s) 3s
Si (s) = = unstable
1 + C (s)G (s) (−s + 2)(s 2 + 4s + 3)
C (s) (s + 4)(−s + 2)
Su (s) = = stable
1 + C (s)G (s) s 2 + 4s + 3

System is NOT internally stable.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 10 / 28
Assessing stability

The internal stability of a system can be determined from the roots of the
characteristic polynomial.
Computer programs can easily find the roots of polynomials of arbitrary
order. However, it is desirable to be assess system stability without having to
find the roots.
Several techniques have been developed for obtaining this information from
the characteristic polynomial; e.g., Routh array, root locus, Nyquist plot. On
the following frames, we look at one of these methods.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 11 / 28
Algebraic analysis

Consider a polynomial of the form: p(s) := s n + an−1 s n−1 + · · · + a1 s + a0 .

We would like to determine whether all roots of p(s) have negative real parts
(such polynomials are called strictly Hurwitz).
A necessary condition: if all roots of p(s) have negative real part, then all
coefficients of p(s) must be strictly larger than zero; i.e., ai > 0, for
i = 0, 1, · · · , n − 1.
The above is not sufficient in general. However, it is also sufficient for
quadratic polynomials p(s) = s 2 + a1 s + a0 .
For polynomial of degree higher than 2, it is not possible to determine
whether all roots have negative real parts by inspecting the sign of the
coefficients. The Routh algorithm addresses this problem.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 12 / 28
Routh algorithm

Build the Routh array

The array: consider p(s) = s n + an−1 s n−1 + · · · + a1 s + a0
sn 1 an−2 an−4 ··· ← coefficients of p(s)
n−1
s an−1 an−3 an−5 ··· ← coefficients of p(s)
 
1 an−2k
s n−2 γ11 γ12 γ13 ··· ← γ1k = −a1n−1 det
 an−1 an−2k−1
a an−2k−1
s n−3 γ21 γ22 γ23 ··· ← γ2k = −γ11 det n−1
1
γ11 γ1k+1
.. .. .. ..
. . . .
s1 * * *
s0 * * *
The number of roots of the polynomial p(s) with real parts greater than zero
is equal to the number of sign changes in the first column of the Routh array.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 13 / 28
Example of the Routh algorithm

Determine if p(s) = s 4 + 2s 3 + 9s 2 + 4s + 5 has any roots with positive real

parts.
The Routh array
s4 1 9 5 ← coefficients a4 , a2 , a0
s3 2 4 0 ← coefficients a3 , a1 , and a 0
1 1 9 1 1 5
s2 7 5 ← 7 = −2 ; 5 = −2
2 4 2 0
2 4
s1 18/7 0 ← 18 1
7 = −7 7 5

1 7 5
s0 5 ← 5 = −18/7
18/7 0
All elements of the first column are positive therefore all roots of p(s) have
negative real parts.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 14 / 28
Summary of Part I

Internal stability of a feedback loop requires that the system response to all
bounded disturbances and bounded reference input is bounded.
This condition will be satisfied if all roots of the characteristic polynomial of
the system have negative real parts.
Routh algorithm is a useful technique for checking for roots with positive real
parts without having to calculate the roots. This makes it useful for assessing
system stability.
In the next couple of lectures, we will see more techniques for assessing
stability and performance of feedback systems.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 15 / 28
Part I
Disturbance Rejection and Internal Stability

Part II
Root Locus

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 16 / 28
Introduction

The location of the closed-loop poles determines system stability and has a
significant effect on the dynamic response.
We have seen that the Routh array can be used to determine whether there
are unstable poles.
More specific information can be obtained from a plot of the closed-loop pole
locations as a chosen parameter varies. This plot is called the root locus.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 17 / 28
A second-order example

Consider the second-order system shown below for K > 0:

r e u 1 y
K (s+2)(s+4)
−

We wish to determine the locations of the closed-loop poles as a function of K .

The characteristic polynomial has the form: s 2 + 6s + K + 8 = 0 The roots are
√
p = −3 ± 1 − K

For 0 ≤ K ≤ 1 the closed-loop poles are real-valued

For K > 1 the closed-loop poles are complex-valued

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 18 / 28
Root locus plot and the step response

Root Locus

0.8 1

0.6

0.4
0.8

0.2
Imag Axis

0.6
0
Green: Open loop
Blue: K=5
−0.2 Black: K=20
0.4
−0.4

−0.6
0.2
−0.8

−1
0
−4 −3 −2 −1 0 0 1 2 3 4 5
Real Axis

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 19 / 28
Some comments

The second-order system just considered permits a closed-form expression for

the system poles.
This is not possible for higher-order systems.
The root locus can be produced using computer programs (see rlocus in
MATLAB).
Alternatively, a set of guidelines can be used for manual construction of the
root locus (this was the only way before computers became widely available).
A better understanding of the computer generated results can be obtained if
you are familiar with the guidelines used to draw the root locus.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 20 / 28
Rules governing root loci (0)

b(s) (s − z1 )(s − z2 ) · · · (s − zm )
Let G (s) := = , where n ≥ m.
a(s) (s − p1 )(s − p2 ) · · · (s − pn )

We are interested in plotting the root locus of the characteristic polynomial, i.e.
the roots of a(s) + Kb(s) = 0, as K varies from 0 to infinity.

b(s)
a(s) + Kb(s) = 0 ⇐⇒ 1 + K =0
a(s)

One may view G (s) as the open-loop plant and the constant K as the controller
(proportional feedback controller).

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 21 / 28
Rules governing root loci (1)

R1 When K = 0, the roots are located at the roots of a(s) = 0; i.e., the poles of
G (s)
R2 The number of roots of a(s) + Kb(s) = 0 are equal to n; thus the root locus
has n branches
R3 Suppose s0 belongs to the root locus. Then there must exist a K0 ≥ 0 such
1
that G (s0 ) = − , i.e. ∠G (s0 ) = (2k + 1)π for some integer k.
K0
R4 By Rule 3, a point s0 on the real axis is part of the root locus if and only if it
is located to the left of an odd number of poles and zeros.
R5 When K is close to infinity, m of the roots are located at the roots of
b(s) = 0; i.e. the zeros of G (s), and, if n > m, the other n − m roots are
located at infinity.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 22 / 28
Rules governing root loci (2)

R6 The n − m roots that go to infinity as K tends to infinity, follow n − m

asymptotes. These asymptotes intersect at

sum of poles of G (s) - sum of zeros of G (s)

σ :=
n−m
The angles of these asymptotes are η1 , · · · , ηn−m , where

(2k − 1)π
ηk = , k = 1, 2, · · · , n − m.
n−m

dG −1 (s)
R7 Breakaway and breakin points (if any) on the real axis satisfy = 0.
ds
R8 If the root locus crosses the imaginary axis, say at s = jwc , then wc can be
found using Routh algorithm, or using the fact that s 2 + wc2 is a factor of
a(s) + Kc b(s) for some positive Kc .

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 23 / 28
A third-order example (1)

Plot the root locus for the system with characteristic equation
s(s 2 + 6s + 13) + K = 0 ⇐⇒ a(s) = s(s 2 + 6s + 13), b(s) = 1,
1
G (s) = .
s(s 2 + 6s + 13)
The open-loop poles are p1 = 0 and p2,3 = −3 ± 2j.
The root locus will occupy the part of the real axis to the left of the pole at
s = 0; i.e., the region (−∞, 0).
The intersection point for the asymptotes is located on the real axis at
0 − 3 + 2j − 3 − 2j
σ= = −2
3
The angles of the asymptotes are

(2k − 1)π π 5π
ηk = = , π,
3 3 3

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 24 / 28
A third-order example (2)

The angles of departure (AOD): use rule R3

For p1 = 0:
−(AOD of p1 + ∠(p1 − p2 ) + ∠(p1 − p3 )) = (2k + 1)π, for some integer k.
AOD of p1 = π rad
For p2 = −3 + 2j: −(AOD of p2 + ∠(p2 − p1 ) + ∠(p2 − p3 )) = (2k + 1)π, for
some integer k. AOD of p2 = −0.98 rad
For p3 = −3 − 2j: −(AOD of p3 + ∠(p3 − p1 ) + ∠(p3 − p2 )) = (2k + 1)π, for
some integer k. AOD of p2 = 0.98 rad
cross-over points on the imaginary axis: suppose the root locus crosses the
imaginary axis at ±ωc j. Then

(s 2 + ωc2 )(s + a) = s(s 2 + 6s + 13) + K

Matching the coefficients, we obtain

√
a = 6, ωc2 = 13, aω 2 = K ⇔ ωc = 13, K = 78

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 25 / 28
A third-order example (3)

Locate points at which the root locus leaves (breakaway points) or meets
(breakin points) the real axis (use R7): find out the real roots of

d(s(s 2 + 6s + 13))
= 3s 2 + 12s + 13 = 0
ds
if any. In this case, there is no such root; thus there is no breakaway/breakin
points on the real axis.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 26 / 28
A third-order example (4)

Root Locus

4
3.61j
3

2
−0.98 rad
1
Imag Axis

−1

0.98 rad
−2

−3
−3.61j
−4

−5

−8 −7 −6 −5 −4 −3 −2 −1 0 1
Real Axis

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 27 / 28
Summary of Part II

The root locus is a plot of the closed-loop pole locations as one system
parameter varies.
The parameter under investigation is often, although need not be, a
controller gain.
Although the root locus can be obtained via computer programs, it is useful
to be familiar with the rules used to construct a root locus.
The root locus can be used in control system design, where one varies a
system parameter to obtain a certain closed-loop pole configuration.

Sei Zhen Khong EE3202B Control Systems Lecture 5: Stability of LTI Systems 28 / 28