105

Sec. 1.6. Z-Transform

 1.6 Z-Transform
The z-transform is an important tool for filter design and for analyzing the stability of
systems.
It is defined as

X
X(z) =
x(n)z n ,
(1.44)
n=

where z is a complex variable, i.e. z = |z|ej .

|z| = 1

j
s |z|e

l
l
1
l
l '$

-1

&%

<

-1

Figure 1.70. Z-transform.

DTFT can be interpreted as the z-transform evaluated on the unit circle,

X(ej ) = X(z)|z=ej

(1.45)

 1.6.1 Rational Z-Transform

We will mostly be interested in z-transforms that are rational functions of z, i.e. ratios
of two polynomials,
P (z)
,
(1.46)
H(z) =
Q(z)
where P () and Q() are polynomials. Rational z-transforms are transfer functions of
LTI systems described by linear, constant-coefficient difference equations such as:
y(n) =

N
1
X

bi x(n i)

i=0

M
X

ak y(n k).

(1.47)

k=1

We now introduce some terminology.

If there is no second term in the right-hand side of Eq. (1.47), then the system is
called non-recursive, or finite-duration impulse response (FIR) system.

106

CHAPTER 1. ANALYSIS OF DISCRETE-TIME LINEAR TIME-INVARIANT SYSTEMS

Otherwise, it is recursive (because the current output sample y(n) is expressed in

terms of the past output samples). If it cannot be written as non-recursive, then
it is an infinite-duration impulse response system.
ZT {x(n i)} = z i X(z)
Y (z) =
Y (z) 1 +

M
X

bi z X(z)

i=0

!
ak z k

N
1
X

(1.48)
M
X

ak z k Y (z)

k=1

= X(z)

N
1
X

bi z i

i=0

k=1

Therefore, the transfer function can be written as

H(z)

Y (z)
X(z)
N
1
X

bi z i

i=0
M
X

1+

ak z k

k=1

z N +1
=
z M

N
1
X

bi z n1i

i=0
M
X

zM +

!
ak z M k

k=1
N
1
Y
assume

b0 6=0

(z zi )

b0 z M N +1 i=1
M
Y

(1.49)

(z pk )

k=1

where the last equality comes from the fact that polynomials of degrees N 1 and M
have N 1 and M roots, respectively. The values of z for which H(z) = 0 (i.e., the roots
of the numerator z1 , z2 , , zN 1 ) are called the zeros of H(z); whereas the values for
which H(z) is infinite (i.e., zeros of the denominator p1 , p2 , , pM ) are called the poles
if M > (N 1), then there are M (N 1) zeros at z = 0;
of H(z). In addition,
if M < (N 1), then there are (N 1) M poles at z = 0.
Poles and zeros may also occur at z = :
zero at means lim H(z) = 0,
|z|

poles at means lim H(z) = .

|z|

107

Sec. 1.6. Z-Transform

Example 1.30. Let us consider the z-transform of an u(n):

ZT {an u(n)} =
=

X
n=0

an z n
az 1

n

n=0

=
=



if az 1  < 1,

1
1 az 1
z
za

The region of convergence (ROC) is the set of all values of z for which the z-transform
converges. In this example, it is |z| > |a|.
Zero
Pole
ROC

Im(z)

Re(z)
a

Figure 1.71. The region of convergence of the series in Example 1.30.

Let us now consider the filter whose transfer function is the z-transform of Example 1.30.
Example 1.31. Let
H(z) =

z
.
za

(a) Difference Equation To see how the z-transform is related to the time-domain
representation, we expand it :
Y (z)
1
=
X(z)
1 az 1
Y (z) az 1 Y (z) = X(z)
Y (z) = az 1 Y (z) + X(z)
Inverting the z-transform, we have:
y(n) = ay(n 1) + x(n).

108

CHAPTER 1. ANALYSIS OF DISCRETE-TIME LINEAR TIME-INVARIANT SYSTEMS

(b) System Diagram The system diagram is depicted in Fig. 1.72.

x(n)



y(n)



z 1

Figure 1.72. The system diagram of Example 1.31. z 1 delays a signal by one time unit.

(c) Frequency Response The frequency response of the system can be obtained by
replacing z with ej in the transfer function:
H ej




H ej  =
=

|H(ej)|

1
1 aej
1
p
(1 aej )(1 aej )
1

1 2a cos + a2
1
p
(1 a)2 + 2a(1 cos )

10
a=0.6
9
a=0.8
8
a=0.9
7
6
5
4
3
2
1
0
4 3 2 1

Figure 1.73. Magnitude response of the filter in Example 1.31 for several values of the parameter a.

Fig. 1.73 shows the magnitude response of the filter for different values of a. In

109

Sec. 1.6. Z-Transform

particular, note that the height of the peak is determined only by a, since the term
with the cosine is removed when = 0. The closer a is to the unit circle, the
sharper the peak, and the thinner the passband.
As we have seen from Example 1.31, in general, a pole near the unit circle will cause
the frequency response to increase in the neighborhood of that pole; a zero will cause
the frequency response to decrease in the neighborhood of that zero (Fig. 1.74).
Im(z)
ej
'$
e
Re(z)

|H(ej )|

1
e
&%
ej

(a) Zeros on the unit circle.

(b) A bandstop filter corresponding to (a).

Im(z)
'$
rej
Re(z)

|H(ej )|

1
&%
rej

(c) Poles near the unit circle.

(d) A bandpass filter corresponding to (c).

Figure 1.74. The effects of zeros and poles near the unit circle.

 1.6.2 Region of Convergence (ROC) of the Z-Transform

Example 1.32. Let x(n) = 2n u(n). Then,
X(z) =

X
n=0

n n

2 z

 n
X
2
n=0

When can we sum this series? In other words, when does this series converge? Consider
these two cases:
 
1. If |z| 2, then  z2  1. This means that every term in the series has an absolute
value greater than or equal to 1. If it is greater than 1, then every successive terms
grows larger. If it is equal to 1, then we are just adding 1 an infinite number of
times. In both cases, the series diverges.

110

CHAPTER 1. ANALYSIS OF DISCRETE-TIME LINEAR TIME-INVARIANT SYSTEMS

 
2. On the other hand, if |z| > 2, then the geometric series converges because  z2  < 1,
and we have
1
X(z) =
.
(1.50)
1 z2
Thus, the z-transform of x(n) = 2n u(n) is

|z| > 2

1 z2
X(z) =

undefined |z| 2
Im(z)

ROC

2
Re(z)
2

2
2

Figure 1.75. The ROC of Example 1.32.

Usually, |z| > 2 is called the region of convergence (ROC) of the z-transform,
because when z lies in this region, the series actually converges to the function (1.50).
A slightly more accurate term would be the region of definition, since the z-transform
is undefined outside of this region.
Example 1.33. Let y(n) = 2n u(n 1). Then
Y (z) =

2n z n

n=1

2m1 z m+1

m=0

z
2

m=0

1. If |z| 2, the series diverges.

 z m
2

where m = n 1

111

Sec. 1.6. Z-Transform

2. if |z| < 2, the series converges, and

z 1
2 1 z2
z
=
2z
1
=
.
1 z2

Y (z) =

Putting it all together,

Y (z) =

undefined |z| 2

1
1 z2

|z| < 2

We get the same expression for the z-transform as in Example 1.32 but the ROC is
different and in fact does not intersect the ROC from Example 1.32.
Example 1.34. Let w(n) = 2n for < n < . Then
2n u(n) + 2n u(n 1) = x(n) y(n),
where x(n) and y(n) are from Examples 1.32 and 1.33, respectively. Hence,
W (z) = X(z) Y (z)
1
1
=
2
1 z
1 z2
= 0???
What is wrong with this derivation? As we saw in the two previous examples, X(z) and
Y (z) have no common ROC:
X(z) is undefined for |z| 2.
Y (z) is undefined for |z| 2.
This means that W (z) is not defined for any z.
Definition 1.12. The region of convergence of

x(n)z n

n=

is the set of all z for which this series is absolutely convergent, i.e.,

X


x(n)z n  < .
n=

112

CHAPTER 1. ANALYSIS OF DISCRETE-TIME LINEAR TIME-INVARIANT SYSTEMS

We use absolute convergence to avoid certain pathological series which converge,

but do not converge absolutely. An example of this is

1
(1)n u(n) z n
n
n=

at z = 1.

 1.6.3 Properties of ROC, Poles and Zeros

The following is a list of several important properties of the z-transform.
1. The ROC is a ring or a disc centered at the origin: r1 < z < r2 . Note that r1 or
r2 could be 0 or .
ROC

Im(z)

Im(z)

Im(z)

Re(z)

Re(z)

(a) x(n) left-sided.

(b) Generic x(n).

Re(z)

(c) x(n) right-sided.

Figure 1.76. The geometries of the ROC.

2. The ROC cannot contain any poles.

3. If x(n) is a finite-duration sequence (i.e., x(n) = 0 except for N1 n N2 ), then
the ROC is the whole z-plane except possibly z = 0.
4. If x(n) is a right-sided sequence (x(n) = 0 for n < N1 ), then the ROC is
|z| > |pmax |,
where pmax is the outermost finite pole of X(z).
5. If x(n) is a left-sided sequence (x(n) = 0 for n N2 ), then the ROC is
|z| < |pmin |,
where pmin is the innermost non-zero pole of X(z).
6. Generalizing Example 1.32 and Example 1.33, we have:
an u(n)

1
, where the ROC is |z| > |a|,
1 az 1

an u(n 1)

1
, where the ROC is |z| < |a|.
1 az 1

(1.51)
(1.52)

113

Sec. 1.6. Z-Transform

7. An LTI system is BIBO stable if and only if the ROC of its transfer function H(z)
includes z = 1, (i.e., includes the unit circle):
"
#
X

h(n)z n 
<
n

z=1

m
|h(n)| <

BIBO stability.

Note that this stability criterion is applicable to LTI systems only.

Example 1.35. Find all sequences whose z-transform is
X(z) =

1 4z 1
.
1 3z 1 + 2z 2

Solution. We decompose X(z) into partial fractions,

X(z) =

A1
A2
1 4z 1
=
+
.
1
1
1
(1 z )(1 2z )
1z
1 2z 1

(1.53)

Then we solve for A1 and A2 .

Method 1. Rearranging the equation, we have
X(z) =

A1 (1 2z 1 ) + A2 (1 z 1 )
A1 + A2 (2A1 + A2 )z 1
=
.
(1 z 1 )(1 2z 1 )
(1 z 1 )(1 2z 1 )

Comparing terms, we solve for A1 and A2 :



A1 + A2 = 1
A1 = 3

2A1 + A2 = 4
A2 = 2
Method 2. Using (1.53),
[X(z)(1 z

[X(z)(1 2z

)]z=1

1 4z 1
=
1 2z 1

A2 (1 z 1 )
= A1 +
1 2z 1
z=1
3 = A1 .


z=1




1 4z 1
1 4z 1 )
=
= A1 +
1 1z 1 z=2
1 z 1 z=2
2 = A2 .


)]z=2

114

CHAPTER 1. ANALYSIS OF DISCRETE-TIME LINEAR TIME-INVARIANT SYSTEMS

Thus, we have
2
3

.
1 z 1 1 2z 1
We now consider the three possible ROCs that this z-transform can have.
X(z) =

Case 1. ROC |z| > 2. Using (1.51),

x(n) = 3 1n u(n) 2 2n u(n)
= (3 2n+1 )u(n).
Case 2. ROC 1 < |z| < 2. Using (1.51) and (1.52),
x(n) = 3 1n u(n) + 2 2n u(n 1).
Case 3. ROC |z| < 1. Using (1.52),
x(n) = (3 + 2n+1 )u(n 1).
Im(z)
Case 1

Re(z)
Case 2
Case 3

10

30
50

x(n)

10

x(n)

x(n)

Figure 1.77. ROC for the cases in Example 1.35.

1
3

70
5 3 1

1
n

(a) |z| > 2

1
3

5
5 3 1

1
n

(b) 1 < |z| < 2

5
5 3 1

1
n

(c) |z| < 1

Figure 1.78. The inverse z-transforms for the 3 possible ROCs of Example 1.35.

115

Sec. 1.6. Z-Transform

 1.6.4 Discrete-Time Exponentials z0n are Eigenfunctions of Discrete-Time

LTI systems
Suppose that a discrete-time complex exponential x(n) = z0n is the input signal to an
LTI system with impulse response h(n). Then the output is:
y(n) =
=

X
k=

h(k)x(n k)
h(k)z0nk

k=

X
n
z0
h(k)z0k
k=

If z0 is in the ROC of H(z), we can write

y(n) = z0n ZT {h(k)} |z=z0 = z0n H(z0 ),
as shown in Fig. 1.79.

y(n) = H(z0 )z0n ,

x(n) = z0n

LTI system
with impulse
response h(n)

if z0 ROC of H(z).

Figure 1.79. An LTI system with z0n as the input.

The transfer function H(z) is the z-transform of the impulse response h(n). It is
also the scaling factor of z n when z n goes through the system.
Recall that we have already considered the case of z0 = ej0 . The frequency response
H(ej0 ) is:
the DTFT of the impulse response h(n);
the scaling factor when ej0 n is the input, as shown in Fig. 1.80.

x(n) = ej0 n

LTI system
with impulse
response h(n)

y(n) = H(ej0 )ej0 n

Figure 1.80. An LTI system with ej0 n as the input.

116

CHAPTER 1. ANALYSIS OF DISCRETE-TIME LINEAR TIME-INVARIANT SYSTEMS

X(ej )
@
@
@
@
@
@

fs

@
@
@
@
@
@

@
@
@

(a) Input signal spectrum before interpolation.

Xu (ej )

B
 B
 B

B

B

B
2

fs

B
B
B
 B
 B
 B
 B
 B
 B

B 
B
B 

B 
B
B 

B
B
B

L
2
L

2
L

B
 B
 B

B

B

B
2

(b) The spectrum of the upsampled signal.

HLP F (ej )
L

(c) Frequency response of the low-pass filter.

Xint (ej )

fs L
B
 B
 B

B

B

B
2

B
 B
 B

B

B

B

B
 B
 B

B

B

B
2

(d) The spectrum of the interpolated signal.

Figure 1.65. The effect of interpolation on the spectrum of the input signal x.