The z-Transform

ECON 397
Macroeconometrics
Cunningham

1
 z-Transform
 The z-transform is the most general concept for the
transformation of discrete-time series.
 The Laplace transform is the more general concept for
the transformation of continuous time processes.
 For example, the Laplace transform allows you to
transform a differential equation, and its corresponding
initial and boundary value problems, into a space in
which the equation can be solved by ordinary algebra.
 The switching of spaces to transform calculus problems
into algebraic operations on transforms is called
operational calculus. The Laplace and z transforms are
the most important methods for this purpose.

2
 The Transforms
The Laplace transform of a function f(t):

F (s ) 
0
f (t )e  st dt

The one-sided z-transform of a function x(n):


X (z)  
n 0
x (n )z  n

The two-sided z-transform of a function x(n):


X (z)  
n  
x (n )z  n

3
 Relationship to Fourier Transform
Note that expressing the complex variable z in
polar form reveals the relationship to the
Fourier transform:

X (re i )  
n  
x (n )(re i ) n , or


X (re i )  
n  
x (n )r n e in , and if r  1,


X (e i )  X ( )  
n  
x(n )e in

which is the Fourier transform of x(n).

4
 Region of Convergence
The z-transform of x(n) can be viewed as the Fourier
transform of x(n) multiplied by an exponential
sequence r-n, and the z-transform may converge
even when the Fourier transform does not.
By redefining convergence, it is possible that the
Fourier transform may converge when the z-
transform does not.
For the Fourier transform to converge, the
sequence must have finite energy, or:



n  
x (n )r  n  

5
 Convergence, continued
The power series for the z-transform is called a
Laurent series:

X (z)  
n  
x (n )z  n

The Laurent series, and therefore the z-transform,

represents an analytic function at every point inside
the region of convergence, and therefore the z-
transform and all its derivatives must be continuous
functions of z inside the region of convergence.
In general, the Laurent series will converge in an
annular region of the z-plane.
6
 Some Special Functions
First we introduce the Dirac delta function (or unit
sample function):
0, n  0 or 0, t  0
 (n )    (t )  
1, n  0 1, t  0
This allows an arbitrary sequence x(n) or
continuous-time function f(t) to be expressed as:

x (n )   x(k ) (n  k )
k  



f (t )  f ( x ) ( x  t )dt


7
 Convolution, Unit Step
These are referred to as discrete-time or
continuous-time convolution, and are denoted by:
x(n )  x(n ) *  (n )
f (t )  f (t ) *  (t )
We also introduce the unit step function:
1, n  0 1, t  0
u( n )   or u(t )  
0, n  0 0, t  0
Note also: 
u(n )   (k )
k  
8
 Poles and Zeros
When X(z) is a rational function, i.e., a ration of
polynomials in z, then:
1. The roots of the numerator polynomial are
referred to as the zeros of X(z), and
2. The roots of the denominator polynomial are
referred to as the poles of X(z).
Note that no poles of X(z) can occur within the region
of convergence since the z-transform does not
converge at a pole.
Furthermore, the region of convergence is bounded
by poles.
9
 Example
Region of convergence
n
x (n )  a u(n )

The z-transform is given by: a
 
X (z)  
n  
a nu(n )z  n  
n 0
(az 1)n

Which converges to:

1 z
X (z)  1
 for z  a
1  az z a
Clearly, X(z) has a zero at z = 0 and a pole at z = a.

10
Convergence of Finite Sequences
Suppose that only a finite number of sequence values are
nonzero, so that: n 2
X (z)  
n  n1
x (n )z n

Where n1 and n2 are finite integers. Convergence requires

x (n )   for n1  n  n2 .

So that finite-length sequences have a region of convergence

that is at least 0 < |z| < , and may include either z = 0 or z =
.

11
 Inverse z-Transform
The inverse z-transform can be derived by using Cauchy’s
integral theorem. Start with the z-transform

X (z)  
n  
x( n )z n

Multiply both sides by zk-1 and integrate with a contour integral

for which the contour of integration encloses the origin and lies
entirely within the region of convergence of X(z):



1 1
2i 
X ( z )z k 1dz 
C
2i
x (n )z n  k 1dz
C n  



1

n  
x(n )
2i 
z n  k 1dz
C
1
2i 
C
X ( z )z k 1dz  x (n ) is the inverse z - transform.
12
 Properties
 z-transforms are linear:
Z  ax (n )  by (n )  aX ( z )  bY ( z )
 The transform of a shifted sequence:
Z  x (n  n0 )  z n X ( z )
0

 Multiplication:
 
Z a n x (n )  Z (a 1z )
But multiplication will affect the region of
convergence and all the pole-zero
locations will be scaled by a factor of a.
13
 Convolution of Sequences

w (n )   x ( k )y ( n  k )
k  
Then
   
W (z)   

n   k  
x( k )y (n  k ) z n

 
 
k  
x(k ) 
n  
y (n  k ) z n

let m  n  k
   
W (z)  
k  
x ( k ) 
m  
y (m )z  z k
 m

W ( z )  X ( z )Y ( z ) for values of z inside the regions of convergenc e of both.
14
 More Definitions
Definition. Periodic. A sequence x(n) is periodic with
period  if and only if x(n) = x(n + ) for all n.

Definition. Shift invariant or time-invariant. Consider a

sequence y(n) as the result of a transformation T of x(n).
Another interpretation is that T is a system that responds
to an input or stimulus x(n):
y(n) = T[x(n)].
The transformation T is said to be shift-invariant or time-
invariant if:
y(n) = T [x(n)] implies that y(n - k) = T [x(n – k)]
For all k. “Shift invariant” is the same thing as “time
invariant” when n is time (t).
15
Let hk (n ) be the response of the system to  (n  k ), a " spike" or shock
occurring at n  k. Then :
   
y (n )  T  
k  
x(k ) (n  k )  
 k  
x (k )T  (n  k )


y (n )   x(k )h (n).
k  
k

If we have time invariance of the transform T , then


y (n )   x(k )h(n  k )  x(n) * h(n).
k  

This implies that the system can be completely characterized by

its impulse response h(n). This obviously hinges on the stationarity
of the series.

16
Definition. Stable System. A system is stable if


 h(k )  
k  

Which means that a bounded input will not yield an

unbounded output.

Definition. Causal System. A causal system is one in

which changes in output do not precede changes in input.
In other words,
If x1(n )  x 2 (n ) for n  n0
then T  x1(n )  T  x 2 (n ) for n  n0 .

Linear, shift-invariant systems are causal iff h(n) = 0 for n < 0.

17
 
Given y (n )   x(k )h (n) let x(n) be sinusoidal. That is,
k  
k

in
let x(n )  e for    n  . Then
 
y (n )  
k  
h( k )e i ( n k )  e in 
k  
h(k )e ik


Let H (e i )  
k  
h(k )e ik so that

i in
y (n )  H (e )e .

Here H(ei) is called the frequency response of the system

whose impulse response is h(n). Note that H(ei) is the Fourier
transform of h(n).

18
We can generalize this state that:

X (e i )  
n  
x (n )e in These are the Fourier
transform pair.
1 
x (n ) 
2 

X (e i )e in d

If  x(n)  , then the transform is absolutely convergent and
n  
converges uniformly to a continuous function of .

This implies that the frequency response of a stable system always

converges, and the Fourier transform exists.

19
If x(n) is constructed from some continuous function xC(t) by
sampling at regular periods T (called “the sampling period”),
then x(n) = xC(nT) and 1/T is called the sampling frequency or
sampling rate.
If 0 is the highest radial frequency of sinusoids comprising
x(nT), then 2 1 w0
0  or 
T T 2

Is the sampling rate required to guarantee that xC(nT) can be

used to fully recover xC(t), This sampling rate 0 is called the
Nyquist rate (or frequency). Sampling at less than this rate will
involve losing information from the time series.
Assume that the sampling rate is at least the Nyquist rate.

20
 1  
X (e iT ) X c (i ),  
T T T
From the continuous time Fourier transform :

1
x c (t ) 
2 
T

T
X c (i )e it d.

Combining :

1
x c (t ) 
2  T TX (e iT

T
)e it d.


Since X (e iT
)  x c (kT )e iTk , we have
k  

  
 
T T   i Tk it
x c (t )  
x c ( kT )e  e d
2  
T k  
21
Changing the order of summation and integratio n,
   

T
x c (t ) 
k  
x c (kT )

 T e i ( t  kT )d 
 2 
T


Evaluating the integral :
  
sin   t  kT )
x c (t )   x c (kt )
 
T
  t  kT 
.
k   T

NOTE: This equation allows for recovering the

continuous time series from its samples. This is valid
only for bandlimited functions.

22