8.

I NVERSE Z-TRANSFORM
113
8. INVERSE Z-TRANSFORM

The process by which a Z-transform of a time series | | { } k x , namely X(z), is returned to
the time domain is called the inverse Z-transform. The inverse Z-transform is defined by:

| | ( ) | | z X Z k x
1
=
Computer study

M-file iztrans.m is used to find inverse Z-transform.
Example 8.1
6 z 5 z
z
) z ( X
2
+ +
=

| | ( ) | | ( ) | | k U 3 k U 2 k x
s
k
s
k
+ =




Four practican techniques can be used to implement an inverse transform. They are:

1. Long divisions
2. Partial fractions
3. Residue theorem
4. Difference equations

8.1 Inverse Z-transforms via long division

For causal sequences, the z-transform X(z) can be expended into a power serises in
1
z

.
For a rational X(z), a convenient way to determine the power serise is an expansion by
long division.
....... z c z c c
a z a ....... z a z
b z b ....... z b z b
) z ( X
2
2
1
1 0
0 1
1 n
1 n
n
0 1
1 n
1 n
n
n
+ + + =
+ + + +
+ + + +
=

where ..... c , c , c
2 1 0
are power serise coefficients.
Example 8.2
1 z 414 . 1 z
z
) z ( X
1 2
+
=





z
z
2
-1.414z+1
z-1.414+z
-1
z
-1
+1.414z
-2
+z
-3
-z
-5
. . .
1.414-z
-1

1.414-2z
-1
+1.414z
-2

z
-1
-1.414z
-2

z
-1
-1.414z
-2
+z
-3

-z
-3

-z
-3
+1.414z
-4
-z
-5
X[k]=o[k-1]+1.414o[k-2]+ o [k-3]+0-o[k-5]. . .
syms z k
x=z/(z^2+5*z+6);
iztrans(x)

ans =

(-2)^k-(-3)^k
8. I NVERSE Z-TRANSFORM
114


Computer study

The inverse of a rational z-transform can also be readly calculated using MATLAB. The
function impz can be utilized for this purpose. Three versions of this function are as
follows:
- [h,t]=impz(num,den)
- [h,t]=impz(num,den, L)
- [h,t]=impz(num,den, L, FT)

Where the input data consists of the vector num and den containing the coefficients of the
numerator and the denominator polynomials of the z-transform given in the descending
powers of z, the output impulse response vector h, and the time index vector t. The first
form, the length L of h is determined automatically by the computer with t=0:L-1,
whereas in the remaining two forms it is supplied by the user through the input data L. In
the last form, the sampling interval is
FT
1
. The default value of FT is 1. The following two
examples show application | | | | den num impz t h , , = file to and plot power.

Example 8.3
1 414 . 1
) (
2
+
=
z z
z
z X



























num=[1 0];
den=[1 -1.414 +1];
L=8;
[x,k]=impz(num,den,L)
x =
1.0000
1.4140
0.9994
-0.0009
-1.0006
-1.4140
-0.9988
0.0017
k =
0
1
2
3
4
5
6
7
stem(k,x,fill,k)


0 1 2 3 4 5 6 7
-1.5
-1
-0.5
0
0.5
1
1.5

k
Power series coefficiens for
1 414 . 1
) (
2
+
=
z z
z
z X
8. I NVERSE Z-TRANSFORM
115


1 3 2
1
) (
2
2
+ +
+
=
z z
z z
z X

Example 8.4
Power series coefficients for
1 3 2
1
) (
2
2
+ +
+
=
z z
z z
z X




















8.2 The Inverse Z-Transform Using Partial Fractions

We now derive the expression for the inverse z-transform and outline the two
methods for its computation.
Recall that, for
e
=
j
re z , the z-transform G(z) given by the equation is merely the Fourier
transform of the modified sequence | |
k
r k g

. Accordingly, by the inverse Fourier
transform, we have:
| |
}
t
t
e e
e
t
= d e ) re ( G
2
1
r k g
k j j k
(8.2)

By making the change of variable
e
=
j
re z , the above equation can be converted into a
contour integral given by :
| | | |

= C inside poles the at z ) z ( G of residues k g
1 k
(8.3)
Note that theequation mentioned above needs to be equated at all values of k which can
be quite complicated in most cases.
A rational G(z) can be expressed as:
) z ( D
) z ( P
) z ( G =

num=[1 1 -1];
den= [2 3 1];
L=10;
[x,t]=impz(num,den,L)
x =
0.5000
-0.2500
-0.3750
0.6875
-0.8438
0.9219
-0.9609
0.9805
-0.9902
0.9951
stem(k,x,fill,k)


1 2 3 4 5 6 7 8 9
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
8. I NVERSE Z-TRANSFORM
116
where P(z) and D(z) are the polynomials in
1
z

. If the degree M of the numerator

where the degree of the polynomial ) z ( P
1
is less that that of D(z). The rational function
) z ( D
) z ( 1 P
is called a proper fraction.
The expression of Eq (8.2) can be computed in a number of ways. Consider the following
cases:
Case 1: G(z) is a proper fraction with simple poles. Let the poles of G(z) be at
k
p z = ,
k=1,2,3,,N, where p
k
are distinct. A partial-fraction expansion of G(z) then is of the
form :

=

=
N
1 k i
i
p z
z a
) z ( G (8.4)
where the constants
l
in the above expression, called the residues, are given by:
i
p z
i i
z
) z ( G
) p z ( a
=
= (8.5)
Each term of the sum on the right-hand side of Eq.(8.4) has an ROC given by z > ,p
k
,,
therefore, the inverse transform g[k] of G(z) is given by
| | | | k U p a k g
s
k
i
N
1 k
i
=
=
Note that the above approach with slight modifications can also be used to determine the
inverse z-transform of a noncausal sequence with a rational z-transform.
Example 8.5
Let the z-transform of a causal sequence g[k] be given by :

) 6 . 0 z )( 2 . 0 z (
) 0 . 2 z ( z
) z ( G
+
+
=
6 . 0 z
a
2 . 0 z
z a
) z ( G
z 2 1
+
+

=
2 . 0 z
1
) 6 . 0 z )( 2 . 0 z (
0 . 2 z
) 2 . 0 z ( a
=
+
+
= 75 . 2
8 . 0
2 . 2
= =
75 . 1
8 . 0
4 . 1
) 6 . 0 z )( 2 . 0 z (
) 0 . 2 z (
) 6 . 0 z ( a
6 . 0 z 2
=

=
+
+
+ =
=

| | | | | | k U ) 6 . 0 ( 75 . 1 ) 2 . 0 ( 75 . 2 k g
s
k k
=
Example 8.6
Using MATLAB determine the partial fraction expansion of X(z):

5 . 1 z 5 . 3 z 2
12 z 3
) z ( X
2 3
3

+
=



8. I NVERSE Z-TRANSFORM
117










5 . 0 z
625 . 2
1 z
6 . 3
5 . 1 z
47 . 2
) z ( X
+
+
+

=

75 . 0 z 75 . 1 z
z 6 z 5 . 1
) z ( X
2 3
2 3

+
=







Multiplying the numerator and the denominator by 2.

5 . 1 z 5 . 3 z 2
12 z 3
) z ( G
2 3
3

+
=

Case 2. G(z) has multiple poles, for example, if the pole at v = z is of multiplicity r and
the remaining N-r poles are simple and at , r N ,..... 3 , 2 , 1 k , p z
k
= = then the general
partial-fraction expansion of G(z) takes the form


=

v
+

+ =
r
1 i
i 1
ri
r N
i k k
k
N M
0 k
k
k
) z 1 (
z
a
p z
z a
z a ) z ( G

where the constant a
ri
(no longer called the residues for i=1) are computed using the
formula:
v =

=
z
r
i r
i r
ri
z
) z ( G
) z (
) z ( d
d
)! i r (
1
a , i=1,2,3,..,r
Example 8.7
2
2
) 1 z (
z
) z ( X

= ;
2
22 21
) 1 z (
z a
) 1 z (
z a
) z ( X

=
2
22 21
) 1 z (
a
) 1 z (
a
z
) z ( X

= ; 0
z
) z ( zX
a
0
= = ; 1
z
) z ( X ) 1 z (
a
1 z
2
21
=

=
=

( )
1 z
dz
d
z
) z ( X 1 z
dz
d
a
1 z
1 z
2
22
= =

=
=
=

num=[3 0 0 12];den=[2 -3.5 0 -1.5];
[r,p,k]=residuez(num,den)
r =
1.9219
3.7891 - 0.3013i
3.7891 + 0.3013i
p =
1.9477
-0.0989 - 0.6126i
-0.0989 + 0.6126i
k =
-8
r=[ 1.9219 3.7891 - 0.3013i 3.7891 + 0.3013i];
p=[ 1.9477 -0.0989 - 0.6126i -0.0989 +
0.6126i];
[num,den]=residuez(r,p,k)
num =
1.5001 -0.0000 0.0008 5.9999
den =
1.0000 -1.7499 -0.0000 -0.7500
8. I NVERSE Z-TRANSFORM
118

Which results in the following time-series
| | ( ) | | k u k 1 k x
s
+ =
Example 8.8
2
) 1 z )( 5 . 0 z (
z
) z ( G

=
4
z
) z ( G
) 5 . 0 z ( a
5 . 0 z
1
= =
=
; 2
z
) z ( G
) 1 z ( a
1 z
2
22
= =
=
;
4
z
) z ( G
) 1 z (
dz
d
a
1 z
2
21
=
(

=
=

2
) 1 z (
z
2
1 z
z
4
5 . 0 z
z
4 ) z ( G

=
Consider the following three cases:
1) |z|>1
g[k] | | | | | | k kU 2 k U 4 k U ) 5 . 0 ( 4
k
+ =
2) |z|<
2
1

| | | | 1 k U k U
s s

g[k] | | | | | | 1 k kU 2 1 k U 4 1 k U ) 5 . 0 ( 4
s s s
k
+ =
3)
2
1
<z<1
g[k] | | | | | | 1 k kU 2 1 k U 4 k U ) 5 . 0 ( 4
s s s
k
+ =

Example 8.9
) 5 . 0 z ( ) 1 z (
z 3 z 5 z 3
) z ( X
2
2 3

+
=
2
22 21 11
) 1 z (
z a
1 z
z a
5 . 0 z
z a
) z ( X

=

where ) 5 . 0 z /( z a
11
an exponential, ) 1 z /( z a
21
a step function, and
2
22
) 1 z /( z a a ramp
function. What is desired, however, is the partial fraction expansion of X(z)/z, where:

2
22 21 11
) 1 z (
a
1 z
a
5 . 0 z
a
z
) z ( X

=
where
5
) 1 z ( z
z 3 z 5 z 3
z
) z ( X ) 5 . 0 z (
a
5 . 0 z
2
2 3
5 . 0 z
1
=

+
=

=
=
=

2
) 2 / 1 z ( z
z 3 z 5 z 3
z
) z ( X ) 1 z (
a
1 z
2 3
1 z
2
22
=

+
=

=
= =

2
)) 2 / 1 z ( z (
) 4 / 1 z )( 3 z 5 z 3 ( 2
) 2 / 1 z ( z
3 10 z 9
z
) z ( X ) 1 z (
dz
d
a
1 z
2
2 3
1 z
2
21
=
(

+
=

=
= =

which results in

|z|>1
|z|<
2
1

2
1
<z<1
0.5 1.0
0.5 1.0
0.5 1.0
8. I NVERSE Z-TRANSFORM
119
| | | | | | k u k 2 2 ) 5 . 0 ( 5 k x
s
k
+ =
Example 8.10
Solve using Matlab:
1 z 4 z 3 z 18
z 18
) z ( H
2 3
3
+
=
5 . 0 z
36 . 0
) z 33 . 0 1 (
4 . 0
z 33 . 0 1
24 . 0
) z ( H
2 1 1

+
+
+
+
=





















Using the numerator and the denominator coefficients we have:

0556 . 0 z 2222 . 0 z 1667 . 0 z
z
) z ( X
2 3
3
+
=

It can be seen that the coefficients will be same as in the equation of the question if we
multiply each coefficient by 18.

Example 8.11. Find the inverse Z-transform of


2
3
) 5 . 0 z )( 5 . 0 z (
) 1 z (
) z ( X
+
+
=

2
22 21 1 0
) 5 . 0 z (
a
) 5 . 0 z (
a
) 5 . 0 z (
a
z
a
z
) z ( X

+
+
+ =

8
z
) z ( zX
a
0 z 0
= =
=

num=[18]; den=[18 3 -4 -1];
[r,p,k]=residuez(num,den)
r =
0.2400
0.4000
0.3600
p =
-0.3333
-0.3333
0.5000
k =[]
[num,den]=residuez(r,p,k)
num =
1.0000 0.0000 0.0000
den =
1.0000 0.1667 -0.2222 -0.0556
8. I NVERSE Z-TRANSFORM
120

4
27
) 5 . 0 z (
) 1 z (
dz
d
z
) z ( X ) 5 . 0 z (
dz
d
a
4
27
) 5 . 0 z (
) 1 z (
z
) z ( X ) 5 . 0 z (
a
4
1
) 5 . 0 z (
) 1 z )( 5 . 0 z (
z
) z ( X ) 5 . 0 z (
a
5 . 0 z
3
5 . 0 z
2
21
5 . 0 z
3
5 . 0 z
2
22
5 . 0 z
2
3
5 . 0 z 1
=
+
+
=

=
=
+
+
=

=
=

+ +
=
+
=
= =
= =
= =

2
) 5 . 0 z (
z
4
27
) 5 . 0 z (
z
4
27
) 5 . 0 z (
z
4
1
8
z
) z ( X

+
=
| | | | | | k U ) 5 . 0 ( k
4
27
) 5 . 0 (
4
27
) 5 . 0 (
4
1
k 8 k x
s
k k k
(

+ o =


















Case 3. X(z) has a complex pole
Example 8.12. The second-order ) 4 / 1 z )
6
cos( z /( ) z 5 . 1 z 3 ( ) z ( X
2 2
+
t
= has non
repeated complex roots. The partial expansion of X(z) is defined by:

) z (
z
a
) z (
z
a ) z ( X
*
2 1
o
+
o
=
) z (
a
) z (
a
z
) z ( X
*
2 1
o
+
o
=
where o=0.433 j0.25 and
o =
o =
o

=
o
=
z
*
2
z
1
) z ( z
) z 5 . 1 z 3 (
z
) z ( X ) z (
a


num=[1 3 3 1]; den=poly([0 -0.5 0.5 0.5])
den =
1.0000 -0.5000 -0.2500 0.1250 0
[r,p,k]=residue(num,den)
r =
-0.2500
-6.7500
6.7500
8.0000
p =
-0.5000
0.5000
0.5000
0
k = []



8. I NVERSE Z-TRANSFORM
121
*
1
* z * z
*
2
a
) z ( z
) z 5 . 1 z 3 (
z
) z ( X ) z (
a =
o

=
o
=
o = o =

Also note that
| | | | k u
z
1
Z k x
k 1
1
o =
(

o
=


and | | | | k u ) (
z
1
Z k x
k *
*
1
2
o =
(

o
=



Where o=0.433013-j0.25=0.5exp(-jt/6). Therefore,
) 6 / j exp( 5 . 0 z
z
a
) 6 / j exp( 5 . 0 z
z
a ) z ( X
*
1 1
t
+
t
=

which corresponds to a time-series, for k>0

| |
) 12 / 6 / k cos(
2
1
1 . 3
)) 12 / 6 / k j exp( ) 12 / j 6 / k j (exp(
2
1
55 . 1
) 6 / k j exp(
2
1
) 12 / j exp( 55 . 1 ) 6 / k j exp(
2
1
) 12 / j exp( 55 . 1 k x
k
k
k k
t + t
|
.
|

\
|
=
t t + t t
|
.
|

\
|
=
t
|
.
|

\
|
t + t
|
.
|

\
|
t =


which is seen to be a causal phase-shifted cosine wave with an exponentially descending
envelope. Also observe that | | 3 ) 12 / cos( 1 . 3 0 x = t = , which can be verified using the
initial value theorem.


8.3 Difference Equations

Long division can be intensive and tedious computational process. If a computer-based
signal processing is desired, the use of difference equation is generally more efficient.
Assume that the Z-transform of a time series is | | k x is X(z), where

=
=

N
0 i
i
i
M
0 i
i
i
z a
z b
) z ( X


Recall that the | | | | 1 k Z = o and | | | |
n
z n k Z

= o . Therefore, it follows that:

| | | | | | | |
| | | | | | | | N k b ) 1 N ( k b .... 1 k b k b
M k x a ) 1 M ( k x a .... 1 k x a k x a
N 1 N 1 0
M 1 M 1 0
o + o + + o + o
= + + + +

8. I NVERSE Z-TRANSFORM
122
The response | | k x can be simulated by implementing the difference equation.



Example 8.13
Consider causal
) 5 . 0 z )( 1 z (
z 3 z 5 z 3
) z ( X
2 3

+
=

from example 11.

3 2 1
2 1
1 2 1
2 1
2 3
2 3
z 5 . 0 z 2 z 5 . 2 1
z 3 z 5 3
) z 5 . 0 1 ( ) z 1 (
z 3 z 5 3
5 . 0 z 2 z 5 . 2 z
z 3 z 5 z 3
) z ( X




+
+
=

+
=
+
+
=

Which produces a time-serises

| | | | k u ) k 2 2 ) 5 . 0 ( 5 ( k x
s
k
+ =

Then | | k x , for 0 k > , can be simulated using

| | | | | | | | | | | | | | 2 k 3 1 k 5 k 3 3 k x 5 . 0 2 k x 2 1 k x 5 . 2 k x o + o o + + =