Lecture 6 — Describing function analysis Course Outline

Lecture 1-3 Modelling and basic phenomena

Today’s Goal: To be able to (linearization, phase plane, limit cycles)
◮ Derive describing functions for static nonlinearities Lecture 2-6 Analysis methods
◮ Predict stability and existence of periodic solutions through (Lyapunov, circle criterion, describing functions)
describing function analysis
Lecture 7-8 Common nonlinearities
(Saturation, friction, backlash, quantization)

Material: Lecture 9-13 Design methods

(Lyapunov methods, Sliding mode & optimal control)
◮ Chapter 14 in Glad & Ljung
◮ Chapter 7.2 (pp.280–290) in Khalil Lecture 14 Summary
◮ Lecture notes

Example: saturated sinusoidals Motivating Example

4
2

2
y
3

y u
1.5

r e u
G(s)
2 1
1

−
1 0.5
input signal

0
output signal

0 0

−1 −0.5 −1
−2 −1

−2
−3 −1.5

−4
0 1 2 3 4 5 6 7 8 9 10
−2
0 1 2 3 4 5 6 7 8 9 10 0 5 10 15 20
time [s] time [s]

4
G(s) = and u = sat (e) gives stable oscillation for r = 0.
s(s + 1)2

Sine Wave Saturation Scope ◮ How can the oscillation be predicted?

sat(A sin ωt)

The “effective gain” ( the ratio ) varies with the input
A sin ωt
signal amplitude A. Q: What is the amplitude/topvalue of u and y ? What is the frequency?

Recall the Nyquist Theorem Motivating Example (cont’d)

Heuristic reasoning:

−1/N (A)
0
Assume G(s) stable. For what frequency ω and gain k is
0 e u y −2

k · G(iω) = −1? f (·) G(s)

− −4
G(iω)
−6

−8
G(iω)
0 e u
G(s)
y −10
k −1/k
−8 −6 −4 −2 0

−
Let N (A) be a complex number such that
f (A sin ωt) ≈ |N (A)|A sin[ωt + arg N (A)].
Heuristic reasoning: For what frequency and what amplitude is ”the
loop gain” N (A) · G(iw) = −1?
The intersection of the −1/N (A) and the Nyquist curve G(iω)
predicts amplitude and frequency.

Fourier Series

Every periodic function u(t) = u(t + T ) has a Fourier series

expansion
∞
◮ How do we derive the describing function N (A)? a0 X
u(t) = + (an cos nωt + bn sin nωt)
2 n=1
◮ Does the intersection predict a stable oscillation? ∞ q
a0 X
= + a2n + b2n sin[nωt + arctan(an /bn )]
◮ Are the estimated amplitude and frequency accurate? 2 n=1

where ω = 2π/T and

Z Z
2 T
2 T
an = u(t) cos nωt dt bn = u(t) sin nωt dt
T 0 T 0

Note: Sometimes we make the change of variable t → φ/ω

1
 The Fourier Coefficients are Optimal The Key Idea

0 e u y
N.L. G(s)
The finite expansion −
k
a0 X
bk (t) =
u + (an cos nωt + bn sin nωt)
2 n=1
Assume e(t) = A sin ωt and u(t) periodic. Then
solves
∞ q
Z a0 X
2 T  2 2π u(t) = + a2n + b2n sin[nωt + arctan(an /bn )]
min bk (t) dt
u(t) − u T = 2
{ân ,b̂n }1≤n≤k T 0 ω n=1

If |G(inω)| ≪ |G(iω)| for n = 2, 3, . . . and a0 = 0, then

if {an , bn } are the Fourier coefficients.
q
y(t) ≈ |G(iω)| a21 + b21 sin[ωt + arctan(a1 /b1 ) + arg G(iω)]

Find periodic solution by matching coefficients in y = −e.

Definition of Describing Function

e(t) u(t) e(t) = A sin ωt = Im (Aeiωt )

The describing function of N.L. is

e(t) u(t) ∞
N.L. a0 X
b1 (ω) + ia1 (ω) u(t) = + (an cos nωt + bn sin nωt)
N (A, ω) := 2 n=1
A
Z 2π/ω Z 2π/ω
ω ω
a1 (ω) := u(t) cos(ωt)dt b1 (ω) := u(t) sin(ωt)dt e(t) u1 (t) u1 (t) = a1 cos(ωt) + b1 sin(ωt)
π 0 π 0 N (A, ω)
= Im (N (A, ω)Aeiωt )
where u(t) is the output corresponding to e(t) := A sin(ωt)
If G is low pass and a0 = 0, then where the describing function is defined as

b1 (t) = |N (A, ω)|A sin[ωt + arg N (A, ω)]

u b1 (ω) + ia1 (ω)
N (A, ω) = =⇒ U (iω) ≈ N (A, ω)E(iω)
A
can be used instead of u(t) to analyze the system.

Amplitude dependent gain and phase shift!

Existence of Limit Cycles Describing Function for a Relay

u 1

u
H 0.5
e
G(iω)
0 e u y e
f (·)
0
G(s)
− −0.5
−1/N (A) −H
−1
A 0 1 2 3
φ
4 5 6

Z
1 1 2π
y = G(iω)u ≈ −G(iω)N (A)y ⇒ G(iω) = − a1 = u(φ) cos φ dφ = 0
N (A) π 0
Z Z
1 2π 2 π 4H
The intersections of G(iω) and −1/N (A) give ω and A for possible b1 = u(φ) sin φ dφ = H sin φ dφ =
π 0 π 0 π
limit cycles.
4H
The describing function for a relay is thus N (A) = .
πA

Describing Function for Odd Static Nonlinearities Limit Cycle in Relay Feedback System

0.1 −1/N (A)

r e u y 0

G(s) −0.1

− −0.2

−0.3 G(iω)
−0.4

−0.5

−1 −0.8 −0.6 −0.4 −0.2 0

Assume f (·) and g(·) are odd static nonlinearities (i.e.,

f (−e) = −f (e)) with describing functions Nf and Ng . Then, 3
G(s) = with feedback u = −sgn y
(s + 1)3
• Im Nf (A, ω) = 0, coeff. (a1 ≡ 0) −3/8 = −1/N (A) = −πA/4 ⇒ A = 12/8π ≈ 0.48
• Nf (A, ω) = Nf (A) √
• Nαf (A) = αNf (A) G(iω) = −3/8 ⇒ ω = 3 ≈ 1.7, T = 2π/ω ≈ 3.6
• Nf +g (A) = Nf (A) + Ng (A)

2
 Limit Cycle in Relay Feedback System (cont’d) Describing Function for a Saturation

u
The prediction via the describing function agrees very well with the 1

H
0.8 e

true oscillations: 0.6

0.4
u

−D D e 0.2

1
u −0.2

y −0.4

−0.6

0.5 −H −0.8

φ
−1
0 1 2 3 4 5 6 7

Let e(t) = A sin ωt = A sin φ. First set H = D . If A ≤ D then

−0.5 N (A) = 1, if A > D then for φ ∈ (0, π)
(
−1 A sin φ, φ ∈ (0, φ0 ) ∪ (π − φ0 , π)
u(φ) =
φ ∈ (φ0 , π − φ0 )
0 2 4 6 8 10
D,

G filters out almost all higher-order harmonics. where φ0 = arcsin D/A. Moreover, u(φ) = −u(2π − φ).

Describing Function for a Saturation (cont’d) Describing Function for a Saturation (cont’d)

If H = D then
 
1
1 2π
Z N (A) = 2φ0 + 2 sin φ0 cos φ0 , A≥D
a1 = u(φ) cos φ dφ π
 s 
π 0
1
Z π Z
1 π 2 D D D2 
= u(φ) cos φ dφ + u(2π − φ) cos φ dφ = 0 = arcsin + 1−
π 0 π 0 π A A A2

Z Z
1 2π 4 π/2 1.1

b1 = u(φ) sin φ dφ = u(φ) sin φ dφ 1

π 0 π 0
Z φ0 Z π/2
0.9
N (A) for H = D = 1
4A 4D
sin2 φ dφ +
0.8

= sin φ dφ 0.7
π 0 π φ0 NOTE: dependance of A shows up
0.6
  in φ0 = arcsin D/A
A
= 2φ0 + 2 sin φ0 cos φ0
0.5

0.4
π 0.3

0.2

0.1
0 2 4 6 8 10

3 minute exercise: The Nyquist Theorem

Assume G(s) stable, and k is positive gain.

◮ The closed-loop system is unstable if the point −1/k is encircled

by G(iω)
What oscillation amplitude and frequency do the describing function ◮ The closed-loop system is stable if the point −1/k is not
analysis predict for the “Motivating Example”? encircled by G(iω)

G(iω)
0 e u
G(s)
y
k −1/k
−

How to Predict Stability of Limit Cycles How to Predict Stability of Limit Cycles

Assume G(s) stable. For a given A = A0 :

G(Ω)
◮ A increases if the point −1/Nf (A0 ) is encircled by G(iω)
◮ A decreases otherwise

G(iω) −1/N (A)

0 e u y
f G(s)
− −1/N (A)
An unstable limit cycle is predicted

A stable limit cycle is predicted An intersection with amplitude A0 is unstable if A < A0 gives
decreasing amplitude and A > A0 gives increasing.

3
 Stable Periodic Solution in Relay System Periodic Solutions in Relay System
The relay gain N (A) is higher for small A:

0.2

0.15 G(iω)
r e u y 0.1

G(s) 0.05
− 0 Big amplitudes Small amplitudes
−0.05 −1/N (A) Small relay gain High relay gain
−0.1
No encirclement One encirclement No encirclement
−0.15
Shrinking amplitudes Growing amplitudes Shrinking amplitudes
−0.2
−5 −4 −3 −2 −1 0
Growing relay gain Shrinking relay gain Growing relay gain
(s + 10)2
G(s) = with feedback u = −sgn y
(s + 1)3
gives one stable and one unstable limit cycle. The left most Stable Unstable
intersection corresponds to the stable one. periodic periodic
orbit orbit

Automatic Tuning of PID Controller Describing Function for a dead-zone relay

Period and amplitude of relay feedback limit cycle can be used for u 1

autotuning. D 0.8 e
0.6

0.4
1
y u e 0.2 u
0

−0.2
0
D −0.4

−0.6

−1 −0.8

−1
0 5 10 0 1 2 3 4 5 6
Time

Let e(t) = A sin ωt = A sin φ. Then for φ ∈ (0, π)

PID (
0, φ ∈ (0, φ0 )
u(φ) =
A y
Σ u
Process

φ ∈ (φ0 , π − φ0 )
T
Relay D,

−1
where φ0 = arcsin D/A (if A ≥ D )

Describing Function for a dead-zone relay–cont’d. Plot of Describing Function for dead-zone relay

0.7

0.6
Z
1 2π
a1 = u(φ) cos φ dφ = 0 0.5 N (A) for D = 1
π 0
Z 2π Z
1 4 π/2
b1 = u(φ) sin φdφ = D sin φdφ 0.4
π 0 π φ0
4D 4D q 0.3
= cos φ0 = 1 − D2 /A2
π π 0.2

 0.1
 0, A<D
N (A) = 4 p
 1 − D2 /A2 , A ≥ D 0
0 2 4 6 8 10
πA

Notice that N (A) ≈ 1.3/A for large amplitudes

Pitfalls Example

The control of output power x(t) from a mobile telephone is critical for good
performance. One does not want to use too large power since other channels
are affected and the battery length is decreased. Information about received
power is sent back to the transmitter and is used for power control. A very
simple scheme is given by
Describing function analysis can give erroneous results.
ẋ(t) = αu(t)
◮ DF analysis may predict a limit cycle, even if it does not exist.
u(t) = −sign y(t − L), α, β > 0
◮ A limit cycle may exist, even if DF analysis does not predict it.
y(t) = βx(t).
◮ The predicted amplitude and frequency are only approximations
and can be far from the true values. Use describing function analysis to predict possible limit cycles.

y(t) = βx(t)
α
x(t) y(t) y(t − L)
x(t) s β e−sL

u(t)
y(t − L)

4
 Accuracy of Describing Function Analysis Accuracy of Describing Function Analysis

Control loop with friction F = sgn y :

1.2
b = 1/3
1

F 0.8
b = 4/3 DF predicts period times and
Friction
ampl. (T, A)b=4/3 = (11.4, 1.00)
0.6

0.4

and (T, A)b=1/3 =(17.3,0.23)

0.2
_
yref u y 0

C G −0.2

−0.4
_ −1 −0.5 0 0.5

1
y b = 4/3 Simulation:
0

Corresponds to
−1
0 5 10 15 20 25 30
(T, A)b=4/3 = (12, 1.1)
b = 1/3
0.4
y
G s(s − b) 0.2
(T, A)b=1/3 = (22, 0.28)
= 3 with feedback u = −sgn y 0

1 + GC s + 2s2 + 2s + 1 −0.2
−0.4
0 5 10 15 20 25 30

The oscillation depends on the zero at s = b.

Accurate results only if y is sinusoidal!

Analysis of Oscillations—A summary Today’s Goal

There exist both time-domain and frequency-domain methods to

analyze oscillations.
Time-domain:
To be able to
◮ Poincaré maps and Lyapunov functions
◮ Derive describing functions for static nonlinearities
◮ Rigorous results but hard to use for large problems
◮ Predict stability and existence of periodic solutions through
Frequency-domain: describing function analysis

◮ Describing function analysis

◮ Approximate results
◮ Powerful graphical methods

Next Lecture

◮ Saturation and antiwindup compensation

◮ Lyapunov analysis of phase locked loops

◮ Friction compensation