Lecture 6: Frequency response experiment

Z-transform, poles and zeros

1 1

0.5 0.5
• Z-transform
0 0
• Pulse transfer function −0.5
u y −0.5

• Poles and zeros −1 −1

0 10 20 30 40 0 10 20 30 40

• Transformations between system representations

 
u( kh) = sin(ω kh) [ y( kh) = A(ω ) sin ω kh + φ (ω )

Read section 2.7 to 2.9 except for Table 2.1.

Z-transform Example — Step

Let y( kh) = 1 for k ≥ 0. Then
Given f (0), f (h), f (2h), . . ., define the Z-transform
z
∞ Y ( z) = 1 + z−1 + z−2 + ⋅ ⋅ ⋅ =
X z−1
Z f ( z) = F ( z) = f ( kh) z−k
k=0
Example — Ramp
Compare to the Laplace transform
Z ∞
Let y( kh) = kh for k ≥ 0. Then
−ts
L f ( s) = f (t) e dt hz
0 Y ( z) = 0 + hz−1 + 2hz−2 + 3hz−3 ⋅ ⋅ ⋅ =
( z − 1)2

F
c Dept. of Automatic Control, Lund 1
 Mini-problem Frequency response revisited
1 1

For the step and the ramp examples, for what values of z is the 0.5 0.5

series in the Z-transform definition convergent? 0 0

−0.5
u y −0.5

−1 −1
0 10 20 30 40 0 10 20 30 40

With u( k) = sin(ω k) we get

k
X k
X
y( k) = (h ∗ u)( k) = h( j )u( k − j ) = h( j ) sin ω ( k − j )
j =0 j =0
P   P 
k
= Im j =0 h( j ) eiω (k− j ) = Im eiω k kj=0 h( j ) e−iω j

= [ k → ∞] = Im [ eiω k H ( eiω )]
 
= p H ( eiω )p sin ω k + arg H ( eiω )

Forward shift Properties of the Z-transform

2 2
f qf
1 1 Z (q−n f ) = z−n F ( z)
0 0

−1 −1
0 2 4 0 2 4
Z (q f ) = z( F ( z) − f (0))
q f ( kh) = f ( kh + h) k = 0, 1, 2, . . .
 
Xk 
Z ( f ∗ ˆ) = Z f ( j )ˆ( k − j ) = (Z f ) ⋅ (Z ˆ)
Backward shift 
j =0


q−1 f
2 2
f
1 1

0 0

−1 −1
0 2 4 0 2 4

(
−1 0 k=0
q f ( kh) =
f ( kh − h) k = 1, 2, . . .

F
c Dept. of Automatic Control, Lund 2
 Bode diagram for continuous transfer function 1/(s2 + 1.4s + 1)
From state space to pulse transfer function (dashed) and for the sampled counterpart with Z-O-hold (solid)

log p H ( eiω )p
x( k + 1) = Φ x( k) + Γ u( k)
y( k) = Cx( k) + Du( k) −10

−20

( −30
z( X ( z) − x(0)) = Φ X ( z) + Γ U ( z)
−40
Y ( z) = C X ( z) + DU ( z)

arg H ( eiω )
−45

−1 −1
Y ( z) = C ( zI − Φ) zx(0) + [ C ( zI − Φ) Γ + D ] U ( z) −90

−135
The rational function H ( z) = C( zI − Φ)−1 Γ + D is called the
−180
pulse transfer function from u to y. It is the Z-transform of the
−225
pulse response. Compare to the time domain formula log ω
k
y( k) = C Φ x(0) + (h ∗ u)( k) The magnitude and phase of H ( eiω h ) are plotted for ω h ∈
[0, π ]

Nyquist diagram Poles and zeros

0.8

0.6 Y ( z) = [ C ( zI − Φ)−1 Γ + D ] U ( z)
| {z }
Imaginary axis

0.4 H ( z)
0.2

0
H ( eiω ) The points p ∈ C where H ( z) = ∞ are called poles of H .
−0.2
They are eigenvalues of Φ and determine stability.
−0.4 The poles of H ( z)−1 are called zeros of H .
−0.6

−0.8

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

The complex number H ( eiω h ) is plotted for ω h ∈ [−π , π ].

The original continuous time Nyquist plot is dashed.

F
c Dept. of Automatic Control, Lund 3
 Poles determine stability Interpretation of poles and zeros
Poles:
All poles of H ( z) = C( zI − Φ)−1 Γ + D are eigenvalues of Φ.
• A pole z = a is associated with the time function z( k) = a k
The matrix Φ can always be written on the form
  • A pole z = a is an eigenvalue of Φ
λ1k
 
λ1 ∗ ∗

Φ=U ..  −1
Hence Φ k = U 
 ..  −1 Zeros:
. U .  . U .

0 λn 0 λ nk • A zero z = a implies that the transmission of the input
u( k) = a k is blocked by the system
The diagonal elements are the eigenvalues of Φ. • A zero is related to how inputs and outputs are coupled to
Φ k decays exponentially if and only if pλ kp < 1 for all k. the states
1 1

0.8
0.5
0.6
0
0.4
−0.5
u y 0.2

0
−1
0 10 20 30 40 0 5 10

Transformation of poles λ i(Φ) = eλ i ( A)h New evidence of alias problem

s z
ωN z = esh
Several points in the s-plane is mapped into the same point in
the z-plane. The map is not bijective
−ω N

3π / h
ωN

p2 x
π/h
p1 x px
−ω N

S0 p1 x x
ωN −π /h
p2 x

−ω N −3 π / h

F
c Dept. of Automatic Control, Lund 4
 Sampling of a second order system Transformation of zeros
ω 02 • More difficult than poles
s2 + 2ζ ω 0 s + ω 02
• In general more sampled zeros than continuous
Poles of the discrete-time system are given by z2 + a1 z + a2 = 0
p  • For short sampling periods
where a1 = −2e−ζ ω 0 h cos 1 − ζ 2 ω 0 h , a2 = e−2ζ ω 0 h
zi ( esi h
1
ζ= 0

ζ = 0.2
Imaginary axis

ζ = 0.4
0.5
ζ = 0.6

ζ = 0.8

0 ζ = 1.0

−1 −0.5 0 0.5 1
Real axis

From pulse transfer function to state space Calculation of H ( z) given G (s)

Taking the Z-transform of both sides in the difference equation Make state realization of G (s). Sample to get Φ and Γ.
Then H ( z) = C( zI − Φ)−1 Γ + D .
y( k) + a1 y( k − 1) + ⋅ ⋅ ⋅ + an y( k − n) = b1 u( k − 1) + ⋅ ⋅ ⋅ + bn u( k − n)

n n−1 n−1
Example 1. For G (s) = 1/s2 , the previous lecture gave
gives [ z + a1 z
| {z
+ ⋅ ⋅ ⋅ + an ] Y ( z) = [b1 z
} |
+ ⋅ ⋅ ⋅ + bn−1 z + bn ] U ( z)
{z }  2
  h   
A( z) B ( z) 1 h 
 
Φ=
 
 Γ=



 2 
 C = 1 0
0 1 
Hence B ( z)/ A( z) is the pulse transfer function from u to y. h

A corresponding state space representation was derived in the With h = 1, this gives
previous lecture  z− 1 −1  h2 

−1

 −h  
 
    H ( z) = C ( zI − Φ) Γ = 1 0  
   
  
2 
−a1 −a2 . . . − an 1 0 z−1

 

 1  0 h
 0 ... 0      h2 
x( k + 1) =  ..  x( k) +  .  u( k)    ( z − 1)−1 h( z − 1)−2   2
..  = h ( z + 1)
  .   

 . .  . = 1 0
  
 

 2 

0 ( z − 1)−1 
h
 2( z − 1)2
1 0 0
y( k) = [ b1 b2 ... bn ] x( k)
F
c Dept. of Automatic Control, Lund 5
 Calculation of H ( z) given G (s) Summary
Example 2. For G (s) = e−τ s/s2 , the previous lecture gave • Z-transform
x( kh + h) = Φ x( kh) + Γ 1 u( kh − h) + Γ 0 u( kh) • Pulse transfer function
 
    τ  (h − τ )2
 1 h  τ h −  
 
 • Poles and zeros
Φ=  
 Γ1 = 
 2 

 Γ0 = 

 2



0 1 τ  
h −τ • Transformations between system representations
With h = 1 and τ = 0.5, this gives
0.125( z2 + 6z + 1)
H ( z) = C ( zI − Φ)−1 (Γ 0 + Γ 1 z−1 ) =
z( z2 − 2z + 1)

Order: 3
Poles: 0, 1, and
√ 1
Zeros: −3 ± 8

F
c Dept. of Automatic Control, Lund 6