CHAPTER IV - Nonlinear systems

FREQUENCY DOMAIN ANALYSIS OF

DYNAMIC SYSTEMS

JOSÉ C. GEROMEL

DSCE / School of Electrical and Computer Engineering

UNICAMP, CP 6101, 13083 - 970, Campinas, SP, Brazil,
[email protected]

Campinas, Brazil, August 2006

1 / 24
CHAPTER IV - Nonlinear systems

Contents

1 CHAPTER IV - Nonlinear systems

Lur’e systems
Passivity
Popov criterion
Frequency domain interpretation
Example
Sector optimization using LMIs
Control design
Problems

2 / 24
CHAPTER IV - Nonlinear systems

Lur’e systems

Lur’e systems
Lur’e systems are those presenting the following state space
model
ẋ(t) = Ax(t) + Bw (t)
z(t) = Cx(t) + Dw (t)
w (t) = −φ(z(t))
where x(t) ∈ Rn , w (t) ∈ R m , z(t) ∈ Rr and
φ(·) : Rr → Rm
is a vector-valued nonlinear function belonging to some set.
The above model is a linear system with impulse response
L(h(t)) = C (sI − A)−1 B + D with a nonlinear feedback
defined by w (t) = −φ(z(t)). The linear part is supposed to
be asymptotically stable.

3 / 24
CHAPTER IV - Nonlinear systems

Passivity

Passivity

Consider the linear part of a previously defined Lur’e system

with r = m. Passivity and Positive Realness are equivalent
concepts.
Definition (Time domain characterization)
It is said to be Passive if for x(0) = 0 the following inequality holds
Z T
z(t)′ w (t)dt ≥ 0 , ∀ T ≥ 0
0

Definition (Frequency domain characterization)

Its transfer function is Positive Real. That is, it satisfies

H(jω)∼ + H(jω) ≥ 0 , ∀ ω ∈ R

4 / 24
CHAPTER IV - Nonlinear systems

Passivity

Passivity

Passivity can be verified with LMIs. Using the Lyapunov

function v (x) = x ′ Px with P = P ′ > 0, from the time domain
characterization we have that
Z T
(w (t)′ z(t) + z(t)′ w (t))dt > v (x(T )) > 0
0

implies (strictly) passivity. Hence, with x(0) = 0 we have to

impose for all (x, w ) 6= (0, 0) that

v̇ (x) < w ′ z + z ′ w

⇑
(Ax + Bw )′ Px+x ′ P (Ax + Bw ) < w ′ (Cx + Dw ) + (Cx + Dw )′ w
| {z } | {z } | {z } | {z }
ẋ ′ ẋ z z′
5 / 24
CHAPTER IV - Nonlinear systems

Passivity

Passivity

Main result on Passivity and Positive Real transfer functions :

Theorem (Passivity and Positive Real)
The transfer function H(s) = C (sI − A)−1 B + D is (strictly)
Positive Real if and only if there exists P = P ′ > 0 such that
 ′ 
A P + PA PB − C ′
<0
B ′ P − C −D − D ′

Sufficiency follows from the previous quadratic Lyapunov

function.
Necessity follows from the KYP Lemma.

6 / 24
CHAPTER IV - Nonlinear systems

Passivity

Passivity

Indeed, the transfer function H(s) is positive real whenever

 ∼  
I I
Π <0, ∀ω∈R
H(jω) H(jω)

where the multiplier Π is given by

 
0 −I
Π=
−I 0

Hence, the KYP Lemma provides the necessity part of the

previous theorem.

Passivity is more restrictive than simple asymptotical stability.

However, stability is preserved for all φ(·) such that φ(0) = 0
and w ′ z = −φ(z)′ z ≤ 0.
7 / 24
CHAPTER IV - Nonlinear systems

Passivity

Passivity

Important remarks :
The previous theorem implies D + D ′ > 0 that is, H(s) must
be proper. This is required for extended passivity. Passivity
(not extended) is imposed by making D → 0 and

A′ P + PA < 0 , PB = C ′ , P > 0

in this case, the LMI given before is not strict.

For SISO systems, Passivity is equivalent to

Re(H(jω)) ≥ 0 , ∀ω ∈ R

For SISO systems, the graphic of the nonlinear function

φ(z) : R → R with respect to z ∈ R must belong to the first
and third quadrants.

8 / 24
CHAPTER IV - Nonlinear systems

Passivity

Passivity

For SISO systems stability is preserved whenever the locus of

the linear part H(s) and the graphic of the nonlinear part are
as indicated :

Im(H(jω)) × Re(H(jω)) φ(z) × z

9 / 24
CHAPTER IV - Nonlinear systems

Popov criterion

Popov criterion

For simplicity we only consider SISO systems. The basic

stability criterion applies to Lur’e systems with r = m = 1.
The number of states n is arbitrary. The nonlinear function:
satisfies the condition φ(0) = 0.
belongs to the sector (0, κ > 0) that is (φ(z) − κz)φ(z) ≤ 0.

κz

φ(z) × z

10 / 24
CHAPTER IV - Nonlinear systems

Popov criterion

Popov criterion

Quadratic Lyapunov function : Consider v (x) = x ′ Px with

P = P ′ > 0 and impose

v̇(x) < (φ(z) − κz)′ φ(z) + φ(z)′ (φ(z) − κz)

Taking into account that w = −φ(z), this inequality is

enforced by the existence of P > 0 such that
 ′ 
A P + PA PB − κC ′
<0
B ′ P − κC −(I + κD)′ − (I + κD)

From the Passivity Theorem this is possible if and only if the

transfer function G (s) := κH(s) + I is strictly positive real.

11 / 24
CHAPTER IV - Nonlinear systems

Popov criterion

Popov criterion
Hence global asymptotic stability is preserved whenever
1
Re(H(jω)) > − , ∀ ω ∈ R
κ

−1/κ

Im(H(jω)) × Re(H(jω))

This condition is more stringent than the Nyquist criterion

applied to the linear system defined by φ(z) = βz with
0 ≤ β < κ.
12 / 24
CHAPTER IV - Nonlinear systems

Popov criterion

Popov criterion

We now assume that the transfer function of the Lur’e system

is strictly proper, that is D = 0. We introduce the
Popov-Lyapunov function :
Z z
′
V (x) = x Px + 2θ φ(ξ)dξ
0

where P = P ′ > 0 and θ ≥ 0. It is important to keep in mind

that, under our assumptions, this function is positive definite
and radially unbounded. Hence is time derivative is given by

V̇ (x) = ẋ ′ Px + x ′ P ẋ + θφ(z)′ ż + θ ż ′ φ(z)

= (Ax + Bw )′ (Px − θC ′ w ) +
+(x ′ P − θw ′ C )(Ax + Bw )

13 / 24
CHAPTER IV - Nonlinear systems

Popov criterion

Popov criterion

Finally, the constraint V̇ (x) < (w + κz)′ w + w ′ (w + κz) is

assured from the existence of P > 0 and θ ≥ 0 satisfying the
LMI
 
A′ P + PA PB − θA′ C ′ − κC ′
<0
B ′ P − θCA − κC −(I + θCB) − (I + θCB)′

Once again, applying the Passivity Theorem this is possible if

and only if the transfer function

G (s) = (κC + θCA)(sI − A)−1 B + (I + θCB)

is strictly positive real. The Popov criterion follows from the

equality A(sI − A)−1 = −I + s(sI − A)−1 which yields

G (s) = (κ + sθ)H(s) + I
14 / 24
CHAPTER IV - Nonlinear systems

Popov criterion

Popov criterion

The previous calculations lead to the celebrated Popov

criterion :
Theorem (Popov criterion)
Stability is preserved whenever there exits a scalar θ ≥ 0 such that
  
θ 1
Re 1 + j ω H(jω) > −
κ κ

holds for all ω ∈ R.

Notice that in the affirmative case the parameter θ ≥ 0 is

readily determined and allows the construction of the
Popov-Lyapunov function V (x).
The previous and more conservative stability condition is
obtained by imposing θ = 0 implying that V (x) = v (x).
15 / 24
CHAPTER IV - Nonlinear systems

Frequency domain interpretation

Frequency domain interpretation

Simple algebraic manipulations show that the Popov stability
condition can be rewritten as
θ 1
Re(H(jω)) − ωIm(H(jω)) > − ∀ω ∈ R
κ κ

1/θ
−1/κ

ωIm(H(jω)) × Re(H(jω))

Notice that the complex plane used to apply the Popov

criterion in not the same adopted by the Nyquist criterion.
16 / 24
CHAPTER IV - Nonlinear systems

Example

Example

Consider the asymptotically stable transfer function

s +2
H(s) =
s 4 + 6s 3 + 13s 2 + 14s + 6
We have obtained the following values :
Nyquist criterion applied for φ(z) linear provides κ ≈ 17.36.
Popov criterion with quadratic Lyapunov function applied for
φ(z) nonlinear provides κ ≈ 9.38.
Propov criterion applied for φ(z) nonlinear provides κ ≈ 17.33.
This example puts in evidence the quality of the result
obtained from the Popov-Lyapunov function yielding the
celebrated Popov criterion.

17 / 24
CHAPTER IV - Nonlinear systems

Sector optimization using LMIs

Sector optimization

Using the LMI representation of the Popov criterion, there is

no difficulty to solve the convex problem

sup κ
P>0,θ>0,κ>0

subject to
 
A′ P + PA PB − θA′ C ′ − κC ′
′ <0
B P − θCA − κC −(I + θCB) − (I + θCB)′

which provides :
The parameters P > 0 and θ > 0 of the function V (x).
The biggest sector defined by the optimal value of κ > 0 for
which the Popov criterion holds.

18 / 24
CHAPTER IV - Nonlinear systems

Control design

State feedback

Consider the Lur’e system with state feedback :

ẋ(t) = Ax(t) + B1 w (t) + B2 u(t)

z(t) = Cx(t)
u(t) = K x(t)
w (t) = −φ(z(t))

the goal is to determine the state feedback gain such that the
sector defined by κ > 0 is maximized. The Popov criterion is
used to check global asymptotic stability.
The key observation is that the Popov criterion can be
rewritten in terms of W := P −1 > 0.

19 / 24
CHAPTER IV - Nonlinear systems

Control design

State feedback

with products of variables θW and κW , that is

 
AW + WA′ B − θWA′ C ′ − κWC ′
<0
B ′ − θCAW − κCW −(I + θCB) − (I + θCB)′

However, from the definition of the new variable K = LW −1

the closed-loop system is such that

AW → (A + B2 K )W = AW + B2 L
B → B1

The maximization of κ is an LMI with respect to matrices

W > 0 and L for θ > 0 and κ > 0 fixed.
The optimization requires a two-dimensional parameter search.
Setting θ = 0, the nonlinear term κW remains.
20 / 24
CHAPTER IV - Nonlinear systems

Problems

Problems

1. Consider the nonlinear system ẋ(t) = Aφ(x(t)) where

A ∈ Rn×n and φ(·) : Rn → Rn is such that

φ(0) = 0 , φ(x) = [φ1 (x1 ), · · · , φn (xn )]′

Show that the equilibrium point x = 0 is GAS wherever there

exists a diagonal positive matrix P such that A′ P + PA < 0.
To this end, make use of the so called Persidiskii-Lyapunov
function
Xn Z xi
v (x) = Pii φi (ξ)dξ
i =1 0

and verify under which conditions it is positive definite and

radially unbounded.

21 / 24
CHAPTER IV - Nonlinear systems

Problems

Problems

2. Consider a Lur’e system with

s + 10
H(s) =
(s + 1)(s 2 + 2s + 2)

and determine:
The maximum value of κ > 0 such that stability is preserved
with φ(z) linear.
The maximum value of κ > 0 such that stability is preserved
with φ(z) nonlinear, provided by a quadratic Lyapunov
function.
The maximum value of κ > 0 such that stability is preserved
with φ(z) nonlinear, provided by the Popov-Lyapunov function.
The maximum value of κ > 0 using H∞ theory.

22 / 24
CHAPTER IV - Nonlinear systems

Problems

Problems

3. Consider a linear plant with transfer function H(s) plus a

linear feedback w (t) = −δz(t) where δ ∈ R. For H(s) given
by
(s − 1)(s − 2)
H(s) =
(s + 1)2 (s + 4)
determine the upper bound δmax such that asymptotic
stability is preserved for all 0 ≤ δ ≤ δmax using :
The H∞ theory
The Popov criterion
The Routh criterion
Compare the obtained results from the root locus plot (with
respect to δ) of the closed-loop system.

23 / 24
CHAPTER IV - Nonlinear systems

Problems

Problems

4. Consider a MIMO Lur’e system such that

H(s) = C (sI − A)−1 B ∈ Cm×m

Determine :
the stability conditions provided by a quadratic Lyapunov
function.
the stability conditions provided by a Popov-Lyapunov type
function.
5. Consider a SISO Lur’e system. Generalize the Popov criterion
to cope with nonlinear functions belonging to the sector
(−κ, κ).

24 / 24