Modern Control Systems

Stability of a Feedback System

Xu Chen

University of Washington

ver. April 30, 2020

Outline

1. Stability of Transfer Functions

2. Arithmetic of Feedback Loops

3. Internal Stability

UW Controls FB stability – 1
Stability of Transfer Functions
I Properness of transfer functions
I proper: the degree of the numerator does not exceed the degree
of the denominator.
I strictly proper: the degree of the numerator is less than that of
the denominator.
I proper transfer function ⇒ causal system
I A transfer function is stable if its zero-input response converges
to zero in the steady state regardless of initial conditions.
I a continuous-time transfer function is stable if its poles are all in
the open left-half plane
I a discrete-time transfer function is stable if its poles are all in
the unit circle
I If all the poles and zeros of a proper continuous/discrete-time
transfer function are on the left-half plane/inside the unit circle,
the corresponding system is called a minimum-phase system.
UW Controls FB stability – 2
Motivating Example

d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n

Stable Gr →y =⇒ closed-loop stability?

Let P(s) = 1/(s 2 − 1), C (s) = (s − 1)/(s + 1), and F (s) = 1. Verify
that the transfer function from r to y is stable while the transfer
function from d to y is not.

I closed-loop stability cannot be judged by a single input-output

transfer function alone
UW Controls FB stability – 3
Arithmetic of Feedback Loops

d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n

P: plant dynamics, C : controller, F : sensor dynamics.

I Signal Relationships:

e = r − yf generate the regulation error

u = Ce control strategy
y = P(u + d) process behavior
yf = Fym = F (y + n) sensor behavior

UW Controls FB stability – 4
Arithmetic of Feedback Loops

d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n
P: plant dynamics, C : controller, F : sensor dynamics.
I Signal Relationships:
    
r F 0 1 ym
 d = 0 1 −C   v 
n 1 −P 0 e
    
ym PC P 1 r
1
inverting ⇒  v  =  C 1 −CF   d 
1 + PCF
e 1 −FP −F n
UW Controls FB stability – 5
The Nine Governing Transfer Functions

d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n

I To analyze all signal relationships, it sufficies to look at the nine

transfer functions from r , d, n to e, ym , v :
    
ym PC P 1 r
1
 v =  C 1 −CF   d  (1)
1 + PCF
e 1 −FP −F n
All transfer functions in the closed-loop system can be obtained
from the above nine transfer functions.
UW Controls FB stability – 6
Terminology: Well-posted Closed-loop Systems

d
+/
◦O − /C /+ v /y
r e u +◦ P

        C 1  
r P 1 v v 1+PC 1+PC r
= ⇒ = 1 −P
d 1 −C e e 1+PC 1+PC
d

I The closed-loop system is well-posed

I if for any signals r and d, there exist unique signals v and e
solving the above loop equation.
I If not well-posed, then for given signals r and y , there are either
an infinite number of solutions or no solutions.
I Example: P(s) = 1, C (s) = −1 ⇒ not well-posed
UW Controls FB stability – 7
Terminology: Well-posted Closed-loop Systems
d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n
    
ym PC P 1 r
1
 v =  C 1 −CF   d  (2)
1 + PCF
e 1 −FP −F n

I The closed-loop system is well-posed if, for P, C , and F

proper, all of the nine closed-loop transfer functions are rational
and proper.
I Example:
I P(s) = 1, C (s) = 1, and F (s) = −1 ⇒ not well-posed
I P(s) = (s + 2)/(s + 1), C (s) = −1, F (s) = 1 ⇒ not well-posed
UW Controls FB stability – 8
Terminology: Well-posted Closed-loop Systems
d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n
    
ym PC P 1 r
1
 v =  C 1 −CF   d  (3)
1 + PCF
e 1 −FP −F n

I A necessary and sufficient condition: 1 + PCF not be strictly

proper, i.e., 1 + P(∞)C (∞)F (∞) 6= 0
I guaranteed if at least one of P, C , and F are strictly proper.
I if 1 + P(∞)C (∞)F (∞) = 0, the relative degree of 1 + PCF is
at least one. 1/(1 + PCF ) then is not proper in (3) —a
contradiction.
UW Controls FB stability – 9
Motivating Example: Revisit

d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n

P(s) = 1/(s 2 − 1), C (s) = (s − 1)/(s + 1), and F (s) = 1.

I not enough to look at a single input-output transfer function

UW Controls FB stability – 10
Internal Stability
d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n
    
ym PC P 1 r
1
 v =  C 1 −CF   d 
1 + PCF
e 1 −FP −F n
I If all the nine transfer functions above are stable, then the
feedback system is said to be internally stable.
I ⇒ if r , d, n are bounded, so are ym , v , e and hence u, y , yf .
I i.e., internal stability guarantees bounded internal signals for all
bounded external input signals.
UW Controls FB stability – 11
Internal Stability Criteria
I Fortunate result: no need to check individual stability of all nine
transfer functions; need only consider the roots of a
characteristic polynomial.
I Write P, C and F as ratios of coprime (no common roots)
polynomials:
NP NC NF
P= , C= , F = (4)
DP DC DF
I the characteristic polynomial of the closed loop is defined as

DP DC DF + NP NC NF

I the roots of the characteristic polynomial are the closed-loop

poles.

UW Controls FB stability – 12
Note on Coprime Polynomials

Example:
NP (s) NC (s) NF (s)
z }| { z}|{ z}|{
s −1 1 1
P(s) = , C (s) = 2 , F (s) = 1 =
+ 1}
s| {z − 1}
s| {z 1
|{z}
DP (s) DC (s) DF (s)

But not
NF (s)
z }| {
s +1
F (s) = 1 =
+ 1}
s| {z
DF (s)

UW Controls FB stability – 13
Internal Stability Criteria
Theorem
The feedback system is internally stable if and only if all the
closed-loop poles are stable.

Proof.
For brevity and without loss of generality, we assume that F = 1.
The essential nine transfer functions in the loop:
    
ym PC P 1 r
1
 v =  C 1 −C   d 
1 + PC
e 1 −P −1 n

UW Controls FB stability – 14
 Cont’d.
Factorizing into coprime numerator-denominator forms
NP NC NF
(P = D P
, C=D C
, F =DF
) gives
    
ym NP NC NP DC DP DC r
 v = 1  NC DP DP DC −NC DP   d 
NP NC + DP DC
e DP DC −NP DC −DP DC n

I ⇐: If all closed-loop poles are stable, then all transfer functions

are stable, hence internal stability holds.
I ⇒: If the closed loop is internally stable, then all nine transfer
functions are stable. All roots of NP NC + DP DC must be stable;
otherwise, there must be cancellation of at least one unstable
zero between NP NC + DP DC and all the polynomials in the 3 × 3
matrix. This is not possible due to the coprime assumption.

UW Controls FB stability – 15
Example
d
+/
◦O − /C /+ v /
r e u +◦ P y

Fo ◦ +
o
+
yf ym n
NP (s) NC (s) NF (s)
z }| { z}|{ z}|{
s −1 1 1
P(s) = , C (s) = 2 , F (s) = 1 =
+ 1}
s| {z − 1}
s| {z 1
|{z}
DP (s) DC (s) DF (s)

Closed-loop characteristic polymonial:

(s − 1) + (s + 1) s 2 − 1 = (s − 1) s 2 + 2s + 2

One unstable closed-loop pole at 1. Exercise: verify stabilty of the

nine transfer functions.
UW Controls FB stability – 16