1

Lecture 5 Inputoutput stability

or
How to make a circle out of the point 1 0i, and different
ways to stay away from it ...

r y
G(s)
f ()
y
k
1
y
k
2
y f (y)

1
k
1

1
k
2
G(i)
Todays Goal
To understand
signal norms
system gain
bounded input bounded output (BIBO) stability
To be able to analyze stability using
the Small Gain Theorem,
the Circle Criterion,
Passivity
Material
[Glad & Ljung]: Ch 1.5-1.6, 12.3
[Khalil]: Ch 57.1; [Slotine & Li]: Ch.4.74.8
lecture slides
History

r y
G(s)
f ()
y
f (y)
For what G(s) and f () is the closed-loop system stable?
Lure and Postnikovs problem (1944)
Aizermans conjecture (1949) (False!)
Kalmans conjecture (1957) (False!)
Solution by Popov (1960) (Led to the Circle Criterion)
Gain
Idea: Generalize static gain to nonlinear dynamical systems
u y
S
The gain of S should tell what is the largest amplication from
u to y
Here S can be a constant, a matrix, a linear time-invariant
system, etc
Question: How should we measure the size of u and y?
Norms
A norm [ [ measures size.
A norm is a function from a space to R

, such that for all

x, y
[x[ 0 and [x[ = 0 = x = 0
[x y[ [x[ [y[
[ x[ = [[ [x[, for all R
Examples
Euclidean norm: [x[ =
_
x
2
1
x
2
n
Max norm: [x[ = max|[x
1
[, . . . , [x
n
[
Signal Norms
A signal x(t) is a function from R

to R.
A signal norm is a way to measure the size of x(t).
Examples
2-norm (energy norm): [x[
2
=
_
_

0
[x(t)[
2
dt
sup-norm: [x[

= sup
tR
[x(t)[
The space of signals with [x[
2
< is denoted L
2
.
Parsevals Theorem
Theorem If x, y L
2
have the Fourier transforms
X(i) =
_

0
e
it
x(t)dt, Y(i) =
_

0
e
it
y(t)dt,
then
_

0
y
T
(t)x(t)dt =
1
2
_

(i)X(i)d.
In particular,
[x[
2
2
=
_

0
[x(t)[
2
dt =
1
2
_

[ X(i)[
2
d.
[x[
2
< corresponds to bounded energy.
System Gain
A system S is a map between two signal spaces: y = S(u).
u y
S
The gain of S is dened as (S) = sup
uL
2
[y[
2
[u[
2
= sup
uL
2
[S(u)[
2
[u[
2
Example The gain of a static relation y(t) =u(t) is
() = sup
uL
2
[u[
2
[u[
2
= sup
uL
2
[[[u[
2
[u[
2
= [[
2
2 minute exercise: Show that (S
1
S
2
) (S
1
) (S
2
).
u y
S
2
S
1
ExampleGain of a Static Nonlinearity
[ f (x)[ K[x[, f (x

) = Kx

u(t) y(t)
f () x
x

Kx
f (x)
[y[
2
2
=
_

0
f
2
_
u(t)
_
dt
_

0
K
2
u
2
(t)dt = K
2
[u[
2
2
u(t) = x

, t (0, ) gives equality =

( f ) = sup
uL
2
[y[
2
[u[
2
= K.
ExampleGain of a Stable Linear System

_
G
_
= sup
uL
2
[Gu[
2
[u[
2
= sup
(0,)
[G(i)[
10
1
10
0
10
1 10
2
10
1
10
0
10
1
(G)
[G(i)[
Proof: Assume [G(i)[ K for (0, ) and [G(i

)[ = K
for some

. Parsevals theorem gives

[y[
2
2
=
1
2
_

[Y(i)[
2
d
=
1
2
_

[G(i)[
2
[U(i)[
2
d K
2
[u[
2
2
Equality by choosing u(t) = sin

t.
BIBO Stability
u y
S
(S) = sup
uL
2
[y[
2
[u[
2
Denition
S is bounded-input bounded-output (BIBO) stable if (S) < .
Example: If x = Ax is asymptotically stable then
G(s) = C(sI A)
1
B D is BIBO stable.
The Small Gain Theorem
r
1
r
2
e
1
e
2
S
1
S
2
Theorem
Assume S
1
and S
2
are BIBO stable. If
(S
1
) (S
2
) < 1
then the closed-loop map from (r
1
, r
2
) to (e
1
, e
2
) is BIBO stable.
Proof of the Small Gain Theorem
Existence of solution (e
1
, e
2
) for every (r
1
, r
2
) has to be veried
separately. Then
[e
1
[
2
[r
1
[
2
(S
2
)|[r
2
[
2
(S
1
)[e
1
[
2
|
gives
[e
1
[
2

[r
1
[
2
(S
2
)[r
2
[
2
1 (S
2
) (S
1
)
(S
2
) (S
1
) < 1, [r
1
[
2
< , [r
2
[
2
< give [e
1
[
2
< .
Similarly we get
[e
2
[
2

[r
2
[
2
(S
1
)[r
1
[
2
1 (S
1
) (S
2
)
so also e
2
is bounded.
Linear System with Static Nonlinear Feedback (1)

r y
G(s)
f ()
y
Ky
f (y)
G(s) =
2
(s 1)
2
and 0
f (y)
y
K
(G) = 2 and ( f ) K.
The small gain theorem gives that K |0, 1]2) implies BIBO
stability.
The Nyquist Theorem

G(s)

2 1.5 1 0.5 0 0.5 1

2
1.5
1
0.5
0
0.5
1
1.5
2

G()
Theorem
The closed loop system is stable iff the number of
counter-clockwise encirclements of 1 by G() (note:
increasing) equals the number of open loop unstable poles.
3
The Small Gain Theorem can be Conservative
Let f (y) = Ky for the previous system.
1 0.5 0 0.5 1 1.5 2
1
0.5
0
0.5
1

G(i)
The Nyquist Theorem proves stability when K |0, ).
The Small Gain Theorem proves stability when K |0, 1]2).
The Circle Criterion
Case 1: 0 < k
1
k
2
<

r y
G(s)
f ()
lac
y
k
1
y
k
2
y f (y)

1
k
1

1
k
2
G(i)
Theorem Consider a feedback loop with y = Gu and
u = f (y) r. Assume G(s) is stable and that
0 < k
1

f (y)
y
k
2
.
If the Nyquist curve of G(s) does not intersect or encircle the
circle dened by the points 1]k
1
and 1]k
2
, then the
closed-loop system is BIBO stable from r to y.
Other cases
G: stable system
0 < k
1
< k
2
: Stay outside circle
0 = k
1
< k
2
: Stay to the right of the line Re s = 1]k
2
k
1
< 0 < k
2
: Stay inside the circle
Other cases: Multiply f and G with 1.
G: Unstable system
To be able to guarantee stability, k
1
and k
2
must have same
sign (otherwise unstable for k = 0)
0 < k
1
< k
2
: Encircle the circle p times counter-clockwise
(if increasing)
k
1
< k
2
< 0: Encircle the circle p times counter-clockwise
(if increasing)
Linear System with Static Nonlinear Feedback (2)
y
Ky
f (y)
1 0.5 0 0.5 1 1.5 2
1
0.5
0
0.5
1

1
K
G(i)
The circle is dened by 1]k
1
= and 1]k
2
= 1]K.
min Re G(i) = 1]4
so the Circle Criterion gives that if K |0, 4) the system is
BIBO stable.
Proof of the Circle Criterion
Let k = (k
1
k
2
)]2 and

f (y) = f (y) ky. Then

f (y)
y

k
2
k
1
2
=: R
y
1
y
2
r
1
r
2
e
1
e
2
G(s)
f ()
r
1

G
G
k
y
1
r
2

f ()
r
1
= r
1
kr
2
Proof of the Circle Criterion (contd)
r
1
r
2

G(s)

f ()
k
R
1
G(i)
SGT gives stability for [

G(i)[R < 1 with

G =
G
1 kG
.
R <
1
[

G(i)[
=

1
G(i)
k

Transform this expression through z 1]z.

Lyapunov revisited
Original idea: Energy is decreasing
x = f (x), x(0) = x
0
V(x(T)) V(x(0)) 0
(some other conditions on V)
New idea: Increase in stored energy added energy
x = f (x, u), x(0) = x
0
y = h(x)
V(x(T)) V(x(0))
_
T
0
(y, u)
. .
external power
dt (1)
Motivation
Will assume the external power has the form (y, u) = y
T
u.
Only interested in BIBO behavior. Note that
V 0 with V(x(0)) = 0 and (??)
==
_
T
0
y
T
u dt 0
Motivated by this we make the following denition
4
Passive System
u y
S
Denition The system S is passive from u to y if
_
T
0
y
T
u dt 0, for all u and all T 0
and strictly passive from u to y if there 0 such that
_
T
0
y
T
u dt ([y[
2
T
[u[
2
T
), for all u and all T 0
A Useful Notation
Dene the scalar product
y, u
T
=
_
T
0
y
T
(t)u(t) dt
u y
S
Cauchy-Schwarz inequality:
y, u
T
[y[
T
[u[
T
where [y[
T
=
_
y, y
T
. Note that [y[

= [y[
2
.
2 minute exercise:
u y
S
1
S
2
P
u y
S
1
S
2
S
1
Feedback of Passive Systems is Passive

r
1
r
2
y
1
y
2
e
1
e
2
S
1
S
2
If S
1
and S
2
are passive, then the closed-loop system from
(r
1
, r
2
) to (y
1
, y
2
) is also passive.
Proof: y, r
T
= y
1
, r
1

T
y
2
, r
2

T
= y
1
, r
1
y
2

T
y
2
, r
2
y
1

T
= y
1
, e
1

T
y
2
, e
2

T
0
Hence, y, r
T
0 if y
1
, e
1

T
0 and y
2
, e
2

T
0
Passivity of Linear Systems
Theorem An asymptotically stable linear system G(s) is
passive if and only if
Re G(i) 0, 0
It is strictly passive if and only if there exists 0 such that
Re G(i ) 0, 0
Proof: See Slotine and Li p. 139 for the rst part.
Example
G(s) =
1
s 1
is passive and
strictly passive,
G(s) =
1
s
is passive but not
strictly passive.
0 0.2 0.4 0.6 0.8 1
0.4
0.2
0
0.2
0.4
0.6


G(i)
A Strictly Passive System Has Finite Gain
u y
S
If S is strictly passive, then (S) < .
Proof: Note that [y[
2
= lim
T
[y[
T
.
([y[
2
T
[u[
2
T
) y, u
T
[y[
T
[u[
T
[y[
2
[u[
2
Hence, [y[
2
T
[y[
2
[u[
2
, so letting T gives
[y[
2

1

[u[
2
The Passivity Theorem

r
1
r
2
y
1
y
2
e
1
e
2
S
1
S
2
Theorem If S
1
is strictly passive and S
2
is passive, then the
closed-loop system is BIBO stable from r to y.
Proof of the Passivity Theorem
S
1
strictly passive and S
2
passive give

_
[y
1
[
2
T
[e
1
[
2
T
_
y
1
, e
1

T
y
2
, e
2

T
= y, r
T
Therefore
[y
1
[
2
T
r
1
y
2
, r
1
y
2

T

1

y, r
T
or
[y
1
[
2
T
[y
2
[
2
T
2y
2
, r
2

T
[r
1
[
2
T

1

y, r
T
Finally
[y[
2
T
2y
2
, r
2

T

1

y, r
T

_
2
1

_
[y[
T
[r[
T
Letting T gives [y[
2
C[r[
2
and the result follows
5
Passivity Theorem is a Small Phase Theorem

r
1
r
2
y
1
y
2
e
1
e
2
S
1
S
2

1
ExampleGain Adaptation
Applications in channel estimation in telecommunication, noise
cancelling etc.
lac
Model
Process
u

(t)
G(s)
G(s)
y
y
m
Adaptation law:
d
dt
= u(t)|y
m
(t) y(t)|, 0.
Gain AdaptationClosed-Loop System
u

(t)
G(s)
G(s)
y
y
m

Gain Adaptation is BIBO Stable

u S

)u y
m
y

s
G(s)
S is passive (Exercise 4.12), so the closed-loop system is BIBO
stable if G(s) is strictly passive.
Simulation of Gain Adaptation
Let G(s) =
1
s 1
, = 1, u = sin t, (0) = 0 and

= 1
0 5 10 15 20
2
0
2
0 5 10 15 20
0
0.5
1
1.5
y, y
m

Storage Function
Consider the nonlinear control system
x = f (x, u), y = h(x)
A storage function is a C
1
function V : R
n
R such that
V(0) = 0 and V(x) 0, x ,= 0

V(x) u
T
y, x, u
Remark:
V(T) represents the stored energy in the system
V(x(T))
. .
stored energy at t = T

_
T
0
y(t)u(t)dt
. .
absorbed energy
V(x(0))
. .
stored energy at t = 0
,
T 0
Storage Function and Passivity
Lemma: If there exists a storage function V for a system
x = f (x, u), y = h(x)
with x(0) = 0, then the system is passive.
Proof: For all T 0,
y, u
T
=
_
T
0
y(t)u(t)dt V(x(T)) V(x(0)) = V(x(T)) 0
Lyapunov vs. Passivity
Storage function is a generalization of Lyapunov function
Lyapunov idea: Energy is decreasing

V 0
Passivity idea: Increase in stored energy Added energy

V u
T
y
6
Example KYP Lemma
Consider an asymptotically stable linear system
x = Ax Bu, y = Cx
Assume there exists positive denite symmetric matrices P, Q
such that
A
T
P PA = Q, and B
T
P = C
Consider V = 0.5x
T
Px. Then

V = 0.5( x
T
Px x
T
P x) = 0.5x
T
(A
T
P PA)x u
T
B
T
Px
= 0.5x
T
Qx u
T
y < u
T
y, x ,= 0
(2)
and hence the system is strictly passive. This fact is part of the
Kalman-Yakubovich-Popov lemma.
Next Lecture
Describing functions (analysis of oscillations)