Stability of

linear time-invariant systems

- causal systems
Negative feedback

A system that maintains the angular

position of a telescope using feedback
input voltage
applied to the
motor that rotates
the platform

The motor is used to

move the platform
Its angle is (t)
Scheme of the system
open loop / open system
Angle of the
platform

Fine tune difficult to achieve

Perturbation (moving the platform) no reaction

Closed system/closed loop

u(t) input
Desired
angle

voltage

error

voltage
Angle of the
platform

input voltage u(t) ~

D - (t)
voltage

Perturbation error correction

Know only D but not the structure of closed system

Closed loop systems advantages

Low sensitivity to perturbations,
detailed knowledge of the system not required
Extended use:
biological, chemical, thermodynamic, aerospace
processes

Time domain analysis

Stable systems: bounded input bounded output.
Wide sense stable systems:
finite duration bounded input bounded output.

Wide sense stable systems examples

1)

u (t ) =

1
i ( )d
C 0

i (t ) = (t ) u (t ) =

t
1
( ) d =
C 0
C

t
C , 0 t T
i (t ) = (t ) (t T ) u (t ) =
T , otherwise
C
2)

Accumulator: h [ n ] = [ n ]
x [ n ] = [ n ] y [ n ] = n + 1.
x [ n] = [ n] [ n N ]

0,

y [ n ] = n + 1,
N + 1,

n<0
0n< N
nN

Analog systems
Strictly stable system
Transfer function : degree of numerator smaller than the
degree of denominator
Poles of the transfer function are in the left half plane
Wide sense stable system
Simple poles on the imaginary axis j or Re{s}=0

Digital systems
Strictly stable system
Transfer function : degree of numerator smaller or equal
than the degree of denominator
Poles of the transfer function are inside the unit circle
Wide sense stable system
Simple poles on the unit circle

Other criteria for BIBO stability

Denominator Q(s) of rational transfer
function H(s)=P(s)/Q(s) is a Hurwitz
polynomial
The polynomial Q ( s ) with real coefficients that has all the
roots in the left half pane is strictly Hurwitz.
Wide sense Hurwitz: simple roots on the imaginary axis
All coefficients of a strictly Hurwitz polynomial are strictly
positive. All coefficients of a wide sense Hurwitz polynomial are
positive.
These conditions are not sufficient.

Hurwitz Stability Criterion

Strictly Hurwitz polynomial (strictly stable system):
Q(s ) = a0 s n + a1s n 1 + ... + an 1s + an = 0 a0 > 0
if and only if all minors from the principal diagonal of the
determinant n are strictly positive.
a1

a3

a5

.. 0

a0
n = 0

a2
a1

a4
a3

0
0
#

..

.. an

Wide sense stable system If one of the minors = 0.

Unstable system If one of the minors <0.
9

Example
d 5 y (t ) d 4 y (t )
d 3 y (t )
d 2 y (t )
dy ( t )
+
+
7
+
4
+ 10
+ 3 y (t ) = 2x (t )
5
4
3
2
dt
dt
dt
dt
dt
with zero initial conditions. The transfer function of the system is:
2
.
s + s + 7 s + 4 s 2 + 10s + 3
The coefficients of the polynomial Q ( s ) are:
H (s) =

a0 = 1, a1 = 1, a2 = 7, a3 = 4, a4 = 10 and a5 = 3, strict positive.

So, the system could be stable.

10

1 4

1 7 10

0 Minors on the principal diagonal:

5 = 0 1

0 1

0 0

1 = 1 > 0, 2 =

4 =

10 0
4

1 4
1 7

1 4 3
1 7 10

0
0

0 1

0 1

10

1 4

a1

a3

a5

= 3 > 0, 3 = a0

a2

a4 = 1 7 10 = ... = 5 > 0,

a1

a3

1 4

0 1
0

1 7 10 0
= ... = 8 > 0. 5 = 0 1 4 3
0 1

0 0

0
0 = 3 4 = 24 > 0.

10 0
4

____

k > 0, k = 1,5 strictly stable system

11

Linear negative feedback systems

Forward-path transfer function: H(s) or H(z)

Negative feedback (feedback path): G(s) or G(z)
Open loop transfer function: H(s)G(s) or H(z)G(z)
Closed loop transfer function:

Q (s) =

Y (s)
H (s)
=
X (s) 1+ H (s)G (s)

Q( z) =

Y ( z)

X ( z)

H ( z)

1+ H ( z)G ( z)

Closed loop system is strictly stable :

- its poles are in the closed left half plane (analog system)
- its poles are inside the unit disc (digital system)

12

 Open loop:

Open loop transfer function

L(s) = H (s)G (s)

L( z) = H ( z)G ( z)

Applications and consequences of

negative feedback
The Inverse System
Known: direct system P(s); Required: inverse system 1/ P(s).
-linear feedback system with: H(s)=K (frequency independent
gain) and G(s)=P(s).
Closed loop transfer function:

Q (s) =

K
1 + KP ( s )

KP ( s ) >>1

1
P (s)

The closed loop system is the inverse system for a value

of the gain K large enough

14

Example
Direct system: differentiation, implemented using a capacitor (the
current trough the capacitor is the derivative of the voltage across
the device)
P(s)=sC
Inverse system: integrator
Q (s) =

Y (s)
1
=
X (s)
sRC

15

Compensation for nonideal elements

Amplifier H(j) with non-constant frequency response in a frequency
band Constant gain amplifier Q(j) in that bandwidth
H(s) response of the amplifier and G(s)=K
H ( s ) s = j
H ( j )
G (s) = K Q (s) =
Q ( j) =
1 + KH ( s )
1 + KH ( j)
Open loop gain: KH ( )
If KH ( )  1, closed loop gain: Q ( ) =

1
ct
K

Usually K<1 (a constant gain is obtained for attenuators) Q()>1.

This implies that
H ( ) 

1
Q ( ) so the open loop gain  closed loop gain
K

16

Stabilization of unstable
systems. Examples
unstable systems included in a closed loop to make
them stable (for example, aircraft flight)
Example #1 Proportional feedback system
b
, a > 0 and G ( s ) = K
sa
H (s)
b
Q (s) =
=
1 + KH ( s ) s ( a Kb )

H (s) =

Pole: s p = a Kb LHP if

Kb > a

stable system. The feedback variable is proportional

with the output variable (G(s)=K)
17

Example #2 Oscillator, simple poles on the imaginary axis j.

H (s) =

b
b
and G ( s ) = K Q ( s ) = 2
s2 + a
s + ( a + Kb )

Second order systems are stable if 0>0 and >0

Q (s) =

02
s 2 + 20 s + 02

only 02 can be influenced by proportional feedback because =0

can not be stabilized by proportional feedback, unless a derivative
component is included into the feedback loop
b
G ( s ) = K1 + K 2 s Q ( s ) = 2
s + bK 2 s + ( a + K1b )
Stable if
a + K1b > 0 and bK 2 > 0

18

Example #3
H ( z) =

H ( z)
1
1
=
; G ( z ) = 2z 1 Q ( z ) =
1
1 2z
1 + H ( z ) G ( z ) 1 2 (1 ) z 1

z p = 2 (1 ) ,0 < < 1
Stable if z p < 1 1/ 2 < < 1

+
x[n]

1
1-2z-1

+
-

e[n]

y[n]

2z-1
19

Tracking systems
Automatic pilot: input is the desired route.
The output is the actual route of the plane.

20

For: H ( z ) = H c ( z ) H p ( z )
Y ( z) =

No disturbance.

H ( z)
X ( z ) ; with E ( z ) H ( z ) = Y ( z )
1+ H ( z)

E ( z) =

X ( z)
1 + H ( z)

On the unit circle: E ( e

X ( e j )

) = 1+ H

(e )
j

The error needs to be negligible:

e [ n ] 0, E ( e j ) 0 large H ( e j )
A good tracking performance is obtained for a large global gain.
21

Errors are modeled by d[n] (disturbance)

Y ( z) =

H ( z)
H ( z)
X ( z)
D ( z)
1+ H ( z)
1+ H ( z)

Small errors means small gain: H ( e j )

The gain should be large (at low
frequencies) and small (high frequencies)

22

Instabilities caused by feedback

Audio signal (voice) perturbed by an undesired signal from the
speaker at the input of the microphone.
A feedback loop is produced.
Signals in phase amplification of the signal from the speaker.
K1 - gain,
K2 - attenuation due to sound propagation,
T - duration of the propagation between speaker and microphone
K2e-sT
Amplifier
Speaker

Microphone
K1

Q (s) =

23

K1
1 K1 K 2 e sT = 0 e sT = K1 K 2
sT
1 K1 K 2 e

( s > 0 K1 K 2 > 1 instability condition )

If the distance speaker-microphone increases attenuation

increases K2 decreases
the system can become stable.
24

Root-locus method
Poles of the closed loop transfer function are
plotted as a function of the gain K.
Simple case: poles are known
Example #1, digital system
1
z
2
=
; G ( z ) = 2z 1 =
1
1 2z
z2
z
1
z
Q( z) =
=
1 2 (1 ) z 1 z 2 (1 )
H (z) =

z p = 2 (1 )
Stable system if z p < 1

1
3
<<
2
2

25

Example #2, analog system

s
2
; G (s) =
s2
s
s
Q (s) =
; s p = 2 (1 )
s 2 (1 )
H (s) =

Stable system if Re {s p } < 0 > 1

26

More complicated: poles are not

known
feedback systems with variable gain

Y ( s)

X (s)

Y (s)

X (s)

KH ( s )

1 + KH ( s ) G ( s )

H ( s)

1 + KH ( s ) G ( s )
27

End points of root-locus: closed

loop poles for K=0 and |K|=

The poles are given by 1 + KH ( s ) G ( s ) = 0

For the closed loop system, the poles depend on K:

H (s)G (s) =

1
K

For K0: H(s)G(s), solution are the poles of H(s)G(s)

for: H ( s ) =

s
2
2
; G (s) =
H ( s) G ( s) =
s2
s
s2

2
1
=
s2

=0s =2

For K, 1/K0, solution are the zeros of H(s)G(s)

28

Argument (angle) criterion

K-real, s0 - closed loop pole then:
jArg {G ( s ) H ( s )}
G ( s0 ) H ( s0 ) = G ( s0 ) H ( s0 ) e
\
0

jArg {G ( s0 ) H ( s0 )}

= 1

Arg {G ( s0 ) H ( s0 )} multiple of

odd multiple of K =

1
>0
G ( s0 ) H ( s0 )

even multiple of K =

for: H ( s ) G ( s ) =

1
<0
G ( s0 ) H ( s0 )

29

2
2
1

s2
s2

2
s0 < 2 Arg
=-
s0 2
2 s0
1
gain =
=
2
2
s0 2
s0 = 2 (1 )

Root locus: s0 for which the argument of

the open loop transfer function is
-odd multiple of for positive gain K>0
-even multiple of for negative gain K<0

30

H (s) =

1
1
;G ( s ) =
s +1
s+3

Real poles:
s0 \ ,s0 > 1 : Arg {G ( s0 ) H ( s0 )} = 0 K < 0
s0 \ , 3 < s0 < 1 : Arg {G ( s0 ) H ( s0 )} = - K > 0
s0 \ ,s0 < 3 : Arg {G ( s0 ) H ( s0 )} = - 2 K < 0

Complex poles. For positive : K>0, for

negative : the same
> 0 : 2 < Arg {G ( s0 ) H ( s0 )} < 0 Arg {G ( s0 ) H ( s0 )} = K > 0
<0:K >0

31

Root locus for

K>0
(stable
system)

K<0 (may be
unstable)
32

G (s) H (s) =

s 1
( s + 1)( s + 3)

33

G (z) H (z) =

z 1
z
=
1
1
1 1 1 1
1 z 1 z z z
2
4
2
4

34

G (s) H (s) =

1
s ( s + 2 )( s + 4 )

Stable system for

K(0,48)

K>0

Nyquist Stability Criterion

The closed-loop system is strictly stable if and only if the
number of complete rotations around the point (- 1/K,0) made
by Nyquist hodograph of the open loop system, H(j)G(j), in
trigonometric sense for from - to , is the number of
poles from the right half plane

36

Observations
1. If the open loop system is stable then H(s)G(s) doesnt have
poles in the right half plane or on the imaginary axis.
So, the open loop Nyquists hodograph doesnt make complete
rotations around the point (-1/K,0)
2. Since h(t) and g(t) are real functions, Nyquists hodograph for
(-,0) is obtained by symmetry with respect to the real axis
of the complex plane H(s)G(s) from the Nyquists hodograph for
(0,)

37

Example #1
G (s) =

1
1
1
; H (s) =
H (s)G ( s) =
2
s +1
0.5s + 1
0.5s + 1.5s + 1

38

The open loop system is stable the number of

complete rotations of Nyquist hodograph around
(-1/K,0) requested by Nyquist stability criterion is zero.
This critical point must be outside the open loop
Nyquist hodograph:

1
<0
K

or

1
>1 K > 0
K

39

Example #2
- open loop system is unstable, a pole
in the right half plane:

G (s) H (s) =

2 ( s + 1)

( s 1)( s + 2 )

- closed loop system stable: open

loop Nyquists hodograph should
make a complete rotation around the
critical point = -1/K in
trigonometric sense.
1 <

1
< 0 K >1
K

40

Example #3 Acoustical system

sT + j )
K = K1K 2 , G ( s ) H ( s ) = e sT = e (

stable open loop system

No complete rotations around the critical point for Nyquist hodograph

K <1
K1, K2 are acoustical attenuations
=> positive.
stable closed loop system if
K1K2<1

41

Discrete-time systems
Closed loop system is stable if all the zeros of :
R(z)=1/K + G(z)H(z)=0,
are inside the unit circle.

1
R ( z ) = R
z
z0 zero(pole) of R ( z ) 1 / z0 zero(pole) of R ( z )
z0 > 1

1
<1
z0

Any zero (pole) of R(z) located outside the unit circle is a

( z ) located inside the unit circle.
zero (pole) of R
42

Nyquist Stability Criterion for

Discrete-time Systems
The closed loop system is stable if and only if is the
number of complete rotations in the trigonometric
sense made by the Nyquist hodograph of the open loop
system around the point (-1/K,0) for [0, 2) is the
number of poles of H(z)G(z) located outside the unit
circle.

43

Example #1
z 2
1
=
1 1
1

1+ z
z z +
2
2

1
1 j ( 1 +2 ) 1 j ( 1 +2 )
H e j =
e
=
= e
1 v v
v2

e j e j + 1 2
2

G( z) H ( z) =

( ) ( )

G e j

maximum and minimum values of 2

( ) ( )
G ( e j ) H ( e j ) = 2

= 0,2: G e j H e j = 2 / 3
= :

44

{ ( ) ( )} = ( + )
Arg {G ( e ) H ( e )} = 4
Arg G e j H e j
j

the sense on the hodograph is clock wise

2 complete rotations around (0,0).
stable open loop system complete rotation of the critical point
is not required.
closed system is stable for:

1 / K < 1 or 1 / K > 2 0 < K < 1 or 1 / 2 < K < 0

K ( 1 / 2 ,0 ) ( 0 ,1)

45

Example #2
G( z) H ( z) =

1
: pole on the unit circle
z ( z 1)

( ) ( )

G e j H e j =

The contour of the unit circle C is

modified by adding a semi-circle of
radius 0, that keep the poles inside
the new contour.

1 j ( +) 1 j ( +)
e
= e
v1v2
v2

The part of the new contour

corresponding to the unit circle 46


/3

2/3

Magnitude

Argument

1
1/2

-
-2

Observations
x=-3/2 vertical
asymptote

{ ( ) ( )} = ( / 2 + / 2) =

Arg G e j H e j

closed loop system stable if:

-1/K<1 or 0<K<1.

47

Gain and phase margins

For a stable system it is interesting to know the

measure in which the gain or phase modifications
affect its stability.

48

Stable system

The gain bound of the first system is the minimum gain K for which
the second system, for =0, becomes unstable.
The phase bound of the first system is the minimum phase , for
which the second system, for K=1, becomes unstable .

49

Instability condition of the second system is:

( ) ( )

1 + H e j G e j K e j = 0
Modify K or one of the poles of the closed loop second
system goes on the imaginary axis, in the position j0.

50

 1
4 1 + s
2
Example. G ( s ) H ( s ) =
.
2
s (1 + 2 s ) 1 + 0, 05s + ( 0,125s )

-stable open loop system.

-the critical point must be outside the Nyquist's hodograph.
-gain margin = distance, on the real axis, between critical point
and the point located at the intersection between Nyquist's hodograph
and the negative real axis.
= 0, the instability condition: K G ( j0 ) H ( j0 ) e
e

jArg {G ( j1 ) H ( j 1 )}

jArg {G ( j 0 ) H ( j0 )}

= 1.

= 1 Arg {G ( j1 ) H ( j1 )} = .

At 1 : intersection of Nyquist's hodograph with the negative

real axis.
51

Gain margin:
1
K=
G ( j1 ) H ( j1 )
2 : G ( j2 ) H ( j2 ) = 1.
Phase margin:
= Arg {G ( j2 ) H ( j2 )} .

52

Nyquists hodograph of the open loop system

53