184 5.

Z-TRANSFORMS

Solution: Using the definition (5.1.2), we find:

5 H(z)= h0 + h1 z−1 + h2 z−2 + h3 z−3 = 2 + 3z−1 + 5z−2 + 2z−3

z-Transforms for case (a), and

H(z)= h0 + h1 z−1 + h2 z−2 + h3 z−3 + h4 z−4 = 1 − z−4

for case (b). ⊔

⊓

The three most important properties of z-transforms that facilitate the analysis and
synthesis of linear systems are:

• linearity property
• delay property
5.1 Basic Properties • convolution property

Here, we review briefly z-transforms and their properties. We assume that the reader The linearity property simply states that the z-transform of a linear combination of
already has a basic familiarity with the subject. Our usage of z-transforms in this book signals is equal to the linear combination of z-transforms, that is, if X1 (z) and X2 (z)
is basically as a tool for the analysis, design, and implementation of digital filters. are the z transforms of the signals x1 (n) and x2 (n), then the z-transform of the linear
Given a discrete-time signal x(n), its z-transform is defined as the following series: combination a1 x1 (n)+a2 x2 (n) is

∞
Z
X(z)= x(n)z−n (z-transform) (5.1.1) a1 x1 (n)+a2 x2 (n)−→ a1 X1 (z)+a2 X2 (z) (linearity) (5.1.3)
n=−∞
The delay property states that the effect of delaying a signal by D sampling units is
or, writing explicitly a few of the terms: equivalent to multiplying its z-transform by a factor z−D , namely,

X(z)= · · · + x(−2)z2 + x(−1)z + x(0)+x(1)z−1 + x(2)z−2 + · · · Z Z

x(n)−→ X(z) ⇒ x(n − D)−→ z−D X(z) (delay) (5.1.4)
−n
There are as many terms as nonzero signal values x(n). The terms z can be Note that D can also be negative, representing a time advance. Finally, the convolu-
thought of as place holders for the values x(n). If the signal x(n) is causal, only negative tion property states that convolution in the time domain becomes multiplication in the
powers z−n , n ≥ 0 appear in the expansion. If x(n) is strictly anticausal, being nonzero z-domain:
for n ≤ −1, only positive powers will appear in the expansion, that is, z−n = z|n| , for
n ≤ −1. And if x(n) is mixed with both causal and anticausal parts, then both negative y(n)= h(n)∗x(n) ⇒ Y(z)= H(z)X(z) (convolution) (5.1.5)
and positive powers of z will appear.
that is, the z-transform of the convolution of two sequences is equal to the product of
The definition (5.1.1) can also be applied to the impulse response sequence h(n) of
the z-transforms of the sequences.
a digital filter. The z-transform of h(n) is called the transfer function of the filter and
is defined by:
Example 5.1.2: The two filters of the above example and of Example 3.4.3 can also be written
in the following “closed” forms, valid for all n:
∞
H(z)= h(n)z−n (transfer function) (5.1.2)
n=−∞ (a) h(n)= 2δ(n)+3δ(n − 1)+5δ(n − 2)+2δ(n − 3), (b) h(n)= δ(n)−δ(n − 4)

Example 5.1.1: Determine the transfer function H(z) of the two causal filters of Example 3.4.3, Their transfer functions can be obtained using the linearity and delay properties as follows.
namely, First, note that the z-transform of δ(n) is unity:

(a) h = [h0 , h1 , h2 , h3 ]= [2, 3, 5, 2] Z

∞
δ(n)−→ δ(n)z−n = δ(0)z−0 = 1
(b) h = [h0 , h1 , h2 , h3 , h4 ]= [1, 0, 0, 0, −1] n=−∞

183
5.1. BASIC PROPERTIES 185 186 5. Z-TRANSFORMS

Then, from the delay property, we have Solution: The two sequences

Z Z Z
δ(n − 1)−→ z−1 · 1 = z−1 , δ(n − 2)−→ z−2 , δ(n − 3)−→ z−3 , etc. h = [1, 2, −1, 1], x = [1, 1, 2, 1, 2, 2, 1, 1]

Using linearity, we obtain have z-transforms:

Z H(z) = 1 + 2z−1 − z−2 + z−3

2δ(n)+3δ(n − 1)+5δ(n − 2)+2δ(n − 3)−→ 2 + 3z−1 + 5z−2 + 2z−3
X(z) = 1 + z−1 + 2z−2 + z−3 + 2z−4 + 2z−5 + z−6 + z−7
for case (a), and
Multiplying these polynomials, we find for the product Y(z)= H(z)X(z):
Z −4
h(n)= δ(n)−δ(n − 4)−→ H(z)= 1 − z
Y(z)= 1 + 3z−1 + 3z−2 + 5z−3 + 3z−4 + 7z−5 + 4z−6 + 3z−7 + 3z−8 + z−10
for case (b). ⊔
⊓
The coefficients of the powers of z are the convolution output samples:
Example 5.1.3: Using the unit-step identity u(n)−u(n − 1)= δ(n), valid for all n, and the
z-transform properties, determine the z-transforms of the two signals: y = h ∗ x = [1, 3, 3, 5, 3, 7, 4, 3, 3, 0, 1]

(a) x(n)= u(n), (b) x(n)= −u(−n − 1) Note that the term z−9 is absent, which means that its coefficient is zero. ⊔
⊓

Solution: For case (a), we have the difference equation:

5.2 Region of Convergence
x(n)−x(n − 1)= u(n)−u(n − 1)= δ(n)
If x(n) has infinite duration, Eq. (5.1.1) becomes an infinite series, and it is possible that
certain values of the complex variable z might render it divergent.
Taking z-transforms of both sides and using the linearity and delay properties, we obtain
The region of convergence (ROC) of the z-transform X(z) is defined to be that subset
Z 1 of the complex z-plane C for which the series (5.1.1) converges, that is,
x(n)−x(n − 1)= δ(n)−→ X(z)−z−1 X(z)= 1 ⇒ X(z)=
1 − z−1 ∞
)  *
Region of Convergence = z ∈ C  X(z)= x(n)z−n ≠ ∞ (5.2.1)
Similarly, for case (b) we have the difference equation: n=−∞

  The ROC is an important concept in many respects: It allows the unique inversion of
x(n)−x(n − 1)= −u(−n − 1)+u −(n − 1)−1 = u(−n)−u(−n − 1)= δ(−n)
the z-transform and provides convenient characterizations of the causality and stability
properties of a signal or system.
where in the last equation we used the given identity with n replaced by −n. Noting that
The ROC depends on the signal x(n) being transformed. As an example, consider
δ(−n)= δ(n), and taking z-transforms of both sides, we find
the following causal signal:
Z 1
x(n)−x(n − 1)= δ(−n)−→ X(z)−z−1 X(z)= 1 ⇒ X(z)= x(n)= (0.5)n u(n)= {1, 0.5, 0.52 , 0.53 , . . . }
1 − z−1

Thus, even though the two signals u(n) and −u(−n − 1) are completely different in Its z-transform will be:
the time domain (one is causal, the other anticausal), their z-transforms are the same. ∞ ∞ ∞
 n
We will see in the next section that they can be distinguished in terms of their region of X(z)= (0.5)n u(n)z−n = (0.5)n z−n = 0.5z−1
convergence. ⊔
⊓ n=−∞ n=0 n=0

Example 5.1.4: Compute the output of Example 4.1.1 by carrying out the convolution operation where the summation was restricted over n ≥ 0 because of the causality of x(n). This
as multiplication in the z-domain. infinite sum can be done with the help of the infinite geometric series formula:

∞
1
1 + x + x2 + x3 + · · · = xn = (5.2.2)
n=0
1−x
5.2. REGION OF CONVERGENCE 187 188 5. Z-TRANSFORMS

which is valid only for |x| < 1 and diverges otherwise. Setting x = 0.5z−1 we find the which is valid only for |x| < 1 and diverges otherwise. Setting x = (0.5)−1 z, we have
sum: ∞ ∞
 m x 0.5−1 z
∞ ∞ X(z)= − (0.5)−1 z =− xm = − =− , or,
 n 1 1−x 1 − 0.5−1 z
X(z)= 0.5z−1 = xn = , or, m=1 m=1

n=0 n=0
1−x
z 1
X(z)= =
z − 0.5 1 − 0.5z−1
1 z
X(z)= = which is the same as the causal example above. However, the ROC in this case is differ-
1 − 0.5z−1 z − 0.5
ent. It is determined by the geometric series convergence condition
where the convergence of the geometric series requires
|x| = |0.5−1 z| < 1 ⇒ |z| < 0.5
−1
|x| = |0.5z |<1 ⇒ |z| > 0.5 which is the set of z’s that lie strictly inside the circle of radius 0.5, as shown below:

Thus, the ROC is the set of z’s in the z-plane that lie strictly outside the circle of z-plane
radius 0.5, as shown below:
 0.5 z
ROC = {z ∈ C  |z| < 0.5}
0.5
ROC z-plane
ROC
z pole
 0.5
|z|
ROC = {z ∈ C  |z| > 0.5}
0.5

pole To summarize, we have determined the z-transforms:

Z 1
(0.5)n u(n) −→ , with |z| > 0.5
1 − 0.5z−1
Note, that the z-transform has a pole at z = 0.5. In summary, we have Z 1
−(0.5)n u(−n − 1) −→ , with |z| < 0.5
1 1 − 0.5z−1
n Z
(0.5) u(n)−→ , with |z| > 0.5 The two signals have the same z-transform but completely disjoint ROCs. More
1 − 0.5z−1
generally, we have the result:
A z-transform and its ROC are uniquely determined by the time signal x(n). How-
ever, it is possible for two different time signals x(n) to have the same z-transform, as Z 1
was the case in Example 5.1.3. Such signals can only be distinguished in the z-domain an u(n) −→ , with |z| > |a|
1 − az−1
by their region of convergence. Consider, for example, the anticausal signal (5.2.3)
Z 1
−an u(−n − 1) −→ , with |z| < |a|
1 − az−1
x(n)= −(0.5)n u(−n − 1)
where a is any complex number. Their ROCs are shown below.
The presence of the anti-unit step u(−n − 1) restricts n to be −n − 1 ≥ 0 or, n ≤ −1.
Its z-transform will be: ROC z-plane z-plane

−1 −1 ∞
z pole pole
n −n
 −n  m |z| z
X(z)= − (0.5) z =− (0.5)−1 z =− (0.5)−1 z |a | a |a| a
n=−∞ n=−∞ m=1
ROC
where we changed summation variables from n to m = −n. To sum it, we use the
following variant of the infinite geometric series:
causal case anticausal case
∞
x
x + x2 + x3 + · · · = xm = The z-transforms (5.2.3), together with the linearity and delay properties, can be
m=1
1−x
used to construct more complicated transforms.
5.2. REGION OF CONVERGENCE 189 190 5. Z-TRANSFORMS

Example 5.2.1: Setting a = ±1 in Eq. (5.2.3), we obtain the z-transforms of the causal and
anticausal unit-steps and alternating unit-steps:
1 z−10 1 − (−0.8)10 z−10
X(z)= − (−0.8)10 =
Z 1 1 + 0.8z−1 1 + 0.8z−1 1 + 0.8z−1
u(n) −→ , with |z| > 1
1 − z−1
Here, the ROC is not |z| > 0.8 as might appear at first glance. Rather it is the set of all
Z 1 nonzero z’s, z = 0. This follows by recognizing x(n) to be a length-10 finite sequence.
−u(−n − 1) −→ −1
, with |z| < 1
1−z Indeed, setting a = −0.8, we have

Z 1 x(n)= an [u(n)−u(n − 10)]= {1, a, a2 , a3 , a4 , a5 , a6 , a7 , a8 , a9 , 0, 0, 0, . . . }

(−1)n u(n) −→ , with |z| > 1
1 + z−1
Z 1 and therefore, its z-transform can be computed by the finite sum
−(−1)n u(−n − 1) −→ , with |z| < 1
1 + z−1
X(z)= 1 + az−1 + a2 z−2 + · · · + a9 z−9
which agree with Example 5.1.3. ⊔
⊓

Example 5.2.2: Determine the z-transform and corresponding region of convergence of the which exists for any z = 0. Using the finite geometric series
following signals:
1 − xN
1. x(n)= u(n − 10) 1 + x + x2 + · · · + xN−1 =
1−x
2. x(n)= (−0.8)n u(n)
we may sum the above series to
3. x(n)= (−0.8)n [u(n)−u(n − 10)]
1 1 − a10 z−10 1 − (−0.8)10 z−10
4. x(n)= [u(n)+(−1)n u(n)]= {1, 0, 1, 0, 1, 0, 1, 0, . . . } X(z)= 1 + az−1 + a2 z−2 + · · · + a9 z−9 = =
2 1 − az−1 1 + 0.8z−1
1
5. x(n)= [(0.8)n u(n)+(−0.8)n u(n)]
2 For case (4), we have, using linearity and Eq. (5.2.3) with a = 1 and a = −1:
πn
6. x(n)= cos( )u(n)= {1, 0, −1, 0, 1, 0, −1, 0, 1, 0, −1, 0, . . . }
2 1 1 1 1
πn X(z)= + =
7. x(n)= (0.8)n cos( )u(n) 2 1 − z−1 1 + z−1 1 − z−2
2
1 with ROC |z| > 1. The same result can be obtained using the definition (5.1.1) and summing
8. x(n)= [(0.8j)n u(n)+(−0.8j)n u(n)]
2 the series:
9. x(n)= cos(ω0 n)u(n) and x(n)= sin(ω0 n)u(n)
10. x(n)= {1, 2, 3, 1, 2, 3, 1, 2, 3, . . . }, periodically repeating {1, 2, 3} X(z)= 1 + 0z−1 + z−2 + 0z−3 + z−4 + · · · = 1 + z−2 + z−4 + z−6 + · · ·

Solution: Using the delay property, we have in case (1): which is an infinite geometric series of the type of Eq. (5.2.2) with x = z−2 . Therefore,

z−10 
X(z)= z−10 U(z)= 1  1
X(z)=  =
1 − z−1 
1−x x=z−2 1 − z−2
with ROC |z| > 1. In case (2), we apply Eq. (5.2.3) with a = −0.8 to get
The convergence of the series requires |x| = |z−2 | < 1 or equivalently, |z| > 1. In case (5),
1 we find again using linearity and Eq. (5.2.3):
X(z)= , with ROC: |z| > | − 0.8| = 0.8
1 + 0.8z−1
1 1 1 1
X(z)= + =
For case (3), we write 2 1 − 0.8z−1 1 + 0.8z−1 1 − 0.64z−2

x(n)= (−0.8)n u(n)−(−0.8)10 (−0.8)n−10 u(n − 10) with ROC |z| > 0.8. Case (6) can be handled directly by the definition (5.1.1):

where in the second term we multiplied and divided by the factor (−0.8)10 in order to X(z)= 1 − z−2 + z−4 − z−6 + z−8 + · · · = 1 + x + x2 + x3 + x4 + · · ·
reproduce the delayed (by 10 units) version of the first term. Thus, using the linearity and
delay properties and the results of case (2), we get where x = −z−2 . The series will converge to
5.2. REGION OF CONVERGENCE 191 192 5. Z-TRANSFORMS

Finally, we consider case (10). Using the definition (5.1.1) and grouping the terms in groups
of 3, we obtain:
1 1
X(z)= =
1−x 1 + z−2
X(z) = (1 + 2z−1 + 3z−2 )+(1 + 2z−1 + 3z−2 )z−3 +
provided |x| = |−z−2 | < 1, or equivalently, |z| > 1. The same result can be obtained using + (1 + 2z−1 + 3z−2 )z−6 + (1 + 2z−1 + 3z−2 )z−9 + · · ·
Euler’s formula to split the cosine into exponential signals of the type (5.2.3):
= (1 + 2z−1 + 3z−2 )(1 + z−3 + z−6 + z−9 + · · · )
πn 1  jπn/2  1 n 
x(n)= cos( )u(n)= e u(n)+e−jπn/2 u(n) = a u(n)+a∗n u(n) 1 + 2z−1 + 3z−2
2 2 2 =
1 − z−3
where a = ejπ/2 = j and a∗ = e−jπ/2 = −j. Thus, we find
The infinite geometric series converges for |z−3 | < 1 or |z| > 1. An alternative method is
! " to delay x(n) by one period, that is, 3 time units
1 1 1 1
X(z)= + =
2 1 − jz−1 1 + jz−1 1 + z−2
x(n − 3)= {0, 0, 0, 1, 2, 3, 1, 2, 3, . . . }
In case (7), using Euler’s formula as above, we find
and subtract it from x(n) to get the difference equation
πn 1 
x(n)= (0.8)n cos( )u(n)= (0.8)n ejπn/2 u(n)+(0.8)n e−jπn/2 u(n)
2 2 x(n)−x(n − 3)= {1, 2, 3, 0, 0, 0, 0, 0, 0, . . . } = δ(n)+2δ(n − 1)+3δ(n − 2)

which can be written as the signal of case (8):

Then, taking z-transforms of both sides, we get

1 
x(n)= (0.8j)n u(n)+(−0.8j)n u(n) 1 + 2z−1 + 3z−2
2 X(z)−z−3 X(z)= 1 + 2z−1 + 3z−2 ⇒ X(z)=
1 − z−3
Thus, cases (7) and (8) are the same. Their z-transform is obtained using a = ±0.8j in
This technique can be generalized to any periodic sequence. It will be used later to imple-
Eq. (5.2.3):
ment digital periodic waveform generators. ⊔
⊓
! "
1 1 1 1 Example 5.2.3: Determine the z-transform and corresponding region of convergence of the
X(z)= + =
2 1 − 0.8jz−1 1 + 0.8jz−1 1 + 0.64z−2 following signals:

with ROC |z| > |0.8j| = 0.8. The cosinewave in case (9) can be handled in a similar fashion. 1. x(n)= (0.8)n u(n)+(1.25)n u(n)
We write 2. x(n)= (0.8)n u(n)−(1.25)n u(−n − 1)

1  jω0 n  Z 1 1 1 3. x(n)= −(0.8)n u(−n − 1)−(1.25)n u(−n − 1)

cos(ω0 n)u(n)= e + e−jω0 n u(n)−→ +
2 2 1− ejω0 z−1 1− e−jω0 z−1 4. x(n)= −(0.8)n u(−n − 1)+(1.25)n u(n)

which combines to give:

Solution: Using Eq. (5.2.3) with a = 0.8 and a = 1.25, we note that the first three cases have
exactly the same z-transform, namely,
1 − cos(ω0 )z−1
X(z)=
1 − 2 cos(ω0 )z−1 + z−2
1 1 2 − 2.05z−1
X(z)= −
+ −
=
1 − 0.8z 1 1 − 1.25z 1 1 − 2.05z−1 + z−2
Setting ω0 = π/2, we recover case (6). Similarly, for a sinewave we have
The three cases differ only in their ROCs. In case (1), both terms are causal, and therefore,
1  jω0 n  Z 1 1 1
sin(ω0 n)u(n)= e − e−jω0 n u(n)−→ − we must have simultaneously |z| > 0.8 and |z| > 1.25. Thus, the ROC is |z| > 1.25.
2j 2j 1− ejω0 z−1 1− e−jω0 z−1
In case (2), the second term is anticausal and therefore we must have the simultaneous
inequalities: |z| > 0.8 and |z| < 1.25. Thus, the ROC is 0.8 < |z| < 1.25. In case (3), both
which combines to give:
terms are anticausal requiring |z| < 0.8 and |z| < 1.25. Therefore, the ROC is |z| < 0.8.
The three ROCs are shown below:
sin(ω0 )z−1
X(z)=
1 − 2 cos(ω0 )z−1 + z−2
5.3. CAUSALITY AND STABILITY 193 194 5. Z-TRANSFORMS

z-plane z-plane z-plane

I I I |z| < min |pi | (5.3.5)
II II II i
III III III that is, the inside of the circle defined by the pole of minimum magnitude. The ROCs of
1
0.8 1.25 0.8 1.25 0.8 1.25 these two cases are shown in Fig. 5.3.1.

z-plane z-plane
unit circle
p4 p4
p1 p1
The two poles of X(z) at z = 0.8 and z = 1.25 divide the z-plane into three non-
p2 p2
overlapping regions which are the three possible ROCs.
Note that case (1) is causal but unstable because the term (1.25)n diverges for large positive p3 p3
n. Case (2) is stable, because the term (0.8)n converges to zero exponentially for large
positive n, and the term (1.25)n converges to zero exponentially for large negative n. And, anticausal ROC
case (3) is anticausal and unstable because the term (0.8)n diverges for large negative n.
causal ROC
The unit circle is contained entirely within the ROC of case (2), in accordance with the
general criterion of stability of Section 5.3.
The fourth case, which is unstable both for n → ∞ and n → −∞, does not have a z- Fig. 5.3.1 Causal and anticausal ROCs.
transform because convergence requires |z| < 0.8 for the anticausal term and |z| > 1.25
for the causal term. Thus, there is no z for which X(z) is converges. The ROC is the empty In summary, causal signals are characterized by ROCs that are outside the maximum
set. ⊔
⊓ pole circle. Anticausal signals have ROCs that are inside the minimum pole circle. Mixed
signals have ROCs that are the annular region between two circles—with the poles that
lie inside the inner circle contributing causally and the poles that lie outside the outer
5.3 Causality and Stability circle contributing anticausally.
Stability can also be characterized in the z-domain in terms of the choice of the ROC.
The z-domain characterizations of causality and stability can be obtained with the help
It can be shown that a necessary and sufficient condition for the stability of a signal x(n)
of the basic result (5.2.3). A causal signal of the form
is that the ROC of the corresponding z-transform contain the unit circle. For a system
h(n), it can be shown that this condition is equivalent to the condition (3.5.4) discussed
x(n)= A1 pn1 u(n)+A2 pn2 u(n)+ · · · (5.3.1)
in Chapter 3.
will have z-transform Stability is not necessarily compatible with causality. For a signal or system to be
simultaneously stable and causal, it is necessary that all its poles lie strictly inside the
A1 A2 unit circle in the z-plane. This follows from Eq. (5.3.3) which is required for a causal
X(z)= + + ··· (5.3.2)
1 − p1 z−1 1 − p2 z−1 ROC. If this ROC is to also correspond to a stable signal, then it must contain the unit
with the restrictions |z| > |p1 |, |z| > |p2 |, and so forth. Therefore, the common ROC circle. In other words, we may set |z| = 1 in Eq. (5.3.3):
of all the terms will be
1 > max |pi |
i
|z| > max |pi | (5.3.3)
i
which implies that all poles must have magnitude less than one. A signal or system
that is, the outside of the circle defined by the pole of maximum magnitude. Similarly, can also be simultaneously stable and anticausal, but in this case all its poles must lie
if the signal is completely anticausal strictly outside the unit circle. Indeed, the anticausality condition Eq. (5.3.5), together
with the stability condition that the ROC contain the points |z| = 1, imply
x(n)= −A1 pn1 u(−n − 1)−A2 pn2 u(−n − 1)− · · · (5.3.4)
1 < min |pi |
its z-transform will be the same as Eq. (5.3.2), but the ROC restrictions on z will be i
|z| < |p1 |, |z| < |p2 |, and so forth. Thus, the ROC is in this case:
which means that all poles must have magnitude greater than one. If some of the poles
have magnitude less than one and some greater than one, then it is possible to have a
5.3. CAUSALITY AND STABILITY 195 196 5. Z-TRANSFORMS

stable signal but it will be of the mixed kind. Those poles that lie inside the unit circle As we emphasized in Chapter 3, stability is more important in DSP than causality in
will contribute causally and those that lie outside will contribute anticausally. order to avoid numerically divergent computations. Causality can be reconciled exactly
Figure 5.3.2 illustrates three such possible stable cases. In all cases, the z-transform if all the poles are inside the unit circle, but only approximately if some of the poles are
has the same form, namely, outside. We will discuss this issue later.
An important class of signals are the so-called marginally stable signals, which nei-
A1 A2 A3 A4
X(z)= + + + ther diverge nor converge to zero for large n. Rather, they remain bounded. The unit-
1 − p1 z−1 1 − p2 z−1 1 − p3 z−1 1 − p4 z−1 step, alternating unit-step, and more general sinusoidal signals fall in this class. Such
signals have poles that lie on the unit circle.
stable/causal ROC stable/anticausal ROC stable/mixed ROC Some examples were cases (1,4,6,9,10) of Example 5.2.2. A simpler example is the
case of a complex sinusoid of frequency ω0

p4 p4 p4 (causal) x(n)= ejω0 n u(n)

p1 p1
p2 1 p2 p1 1
(anticausal) x(n)= −ejω0 n u(−n − 1)
unit p3 p3 p2
circle unit p3 which is a special case of Eq. (5.2.3) with a = ejω0 . Note that the plain unit-step u(n)
circle and alternating step (−1)n u(n) are special cases of this with ω0 = 0 and ω0 = π. The
unit
circle corresponding z-transform follows from Eq. (5.2.3):

z-plane e jω0 pole

Fig. 5.3.2 Stable ROCs.
1 ω0 1
X(z)=
In the stable and causal case, all poles must have magnitude less than one, that is, 1 − ejω0 z−1 0
unit
|pi | < 1, i = 1, 2, 3, 4 and the signal x(n) will be circle
 
x(n)= A1 pn1 + A2 pn2 + A3 pn3 + A4 pn4 u(n) with ROC being either |z| > 1 for the causal case, or |z| < 1 for the anticausal one.

with all terms converging to zero exponentially for large positive n. In the stable/anticausal
case, all poles have magnitude greater than one, |pi | > 1, i = 1, 2, 3, 4, and x(n) will be: 5.4 Frequency Spectrum
 
x(n)= − A1 pn1 + A2 pn2 + A3 pn3 + A4 pn4 u(−n − 1) The frequency spectrum, frequency content, or discrete-time Fourier transform (DTFT)
of a signal x(n) is defined by
where because n is negative, each term will tend to zero exponentially for large negative
n. This can be seen more clearly by writing a typical term as ∞
X(ω)= x(n)e−jωn (DTFT) (5.4.1)
 |n|
n=−∞
−|n| 1
−A1 pn1 u(−n − 1)= −A1 p1 u(−n − 1)= −A1 u(−n − 1)
p1 It is recognized as the evaluation of the z-transform on the unit circle, that is, at the
z points:
where we set n = −|n| for negative n. Because |p1 | > 1 it follows that |1/p1 | < 1
and its successive powers will tend to zero exponentially. In the mixed case, we have jω
z=e (5.4.2)
|p1 | < |p2 | < 1 and |p4 | > |p3 | > 1. Therefore, the stable signal will be
    Indeed, we have:†
x(n)= A1 pn1 + A2 pn2 u(n)− A3 pn3 + A4 pn4 u(−n − 1)
∞  ∞
 
with p1 , p2 contributing causally, and p3 , p4 anticausally. An example of such a stable X(z)z=ejω = x(n)z−n 
 = x(n)e−jωn = X(ω)
n=−∞ z=ejω n=−∞
but mixed signal was given in the second case of Example 5.2.3, namely,
† Here, we abuse the notation and write X(ω) instead of X(ejω ).
x(n)= (0.8)n u(n)−(1.25)n u(−n − 1)
5.4. FREQUENCY SPECTRUM 197 198 5. Z-TRANSFORMS

The frequency response H(ω) of a linear system h(n) with transfer function H(z)
is defined in the same way, namely,
x(n)= ejω0 n , −∞ < n < ∞
∞
H(ω)= h(n)e−jωn (frequency response) (5.4.3) Then, its DTFT will be given by
n=−∞

and it is also the evaluation of H(z) on the unit circle: X(ω)= 2πδ(ω − ω0 )+(Nyquist replicas)

 where the term “Nyquist replicas” refers to the periodic replication of the first term at
H(ω)= H(z)z=ejω
intervals of 2π. This is needed in order to make X(ω) periodic with period 2π. More
As discussed in Chapter 1, the digital frequency ω is in units of [radians/sample] precisely, the full expression will be
and is related to the physical frequency f in Hz by ∞
X(ω)= 2π δ(ω − ω0 − 2πm)
2πf m=−∞
ω= (digital frequency) (5.4.4)
fs
To verify it, we insert it into the inverse DTFT equation (5.4.6) and recover the given
The Nyquist interval [−fs /2, fs /2] is the following interval in units of ω: sinusoid. It was also discussed in Chapter 1, Example 1.5.1. Assuming that ω0 lies in
the Nyquist interval [−π, π], then the restriction of X(ω) within it will be given only
by the m = 0 term, that is:
−π ≤ ω ≤ π (Nyquist interval) (5.4.5)

In Chapter 1, the quantity X(ω) was denoted by X(ω)= 2πδ(ω − ω0 ), −π ≤ ω ≤ π

∞ Therefore, Eq. (5.4.6) gives

 )=
X(f x(nT)e−2πjf n/fs
n=−∞
π π
1 1
x(n)= X(ω)ejωn dω = 2πδ(ω − ω0 )ejωn dω = ejω0 n
It was the Fourier spectrum of the sampled signal x(nT) and was given by the peri- 2π −π 2π −π

odic replication of the original analog spectrum at multiples of fs . Similarly, for a linear combination of two sinusoids we have:
In units of ω, periodicity in f with period fs becomes periodicity in ω with period
2π. Therefore, X(ω) may be considered only over one period, such as the Nyquist x(n)= A1 ejω1 n + A2 ejω2 n −→ X(ω)= 2πA1 δ(ω − ω1 )+2πA2 δ(ω − ω2 )
interval (5.4.5).
The inverse DTFT recovers the time sequence x(n) from its spectrum X(ω) over This can be verified in the same way, if we assume that both ω1 and ω2 lie in the
the Nyquist interval: Nyquist interval. In particular, for real-valued cosine and sine signals, we have:

π
1 cos(ω0 n) −→ πδ(ω − ω0 )+πδ(ω + ω0 )
x(n)= X(ω)ejωn dω (inverse DTFT) (5.4.6)
2π −π
sin(ω0 n) −→ −jπδ(ω − ω0 )+jπδ(ω + ω0 )
It expresses x(n) as a linear combination of discrete-time sinusoids ejωn of different
frequencies. The relative amplitudes and phases of these sinusoidal components are Another useful relationship is Parseval’s equation, which relates the total energy of
given by the DTFT X(ω). One quick way to prove Eq. (5.4.6) is to think of Eq. (5.4.1) a sequence to its spectrum:
as the Fourier series expansion of the periodic function X(ω). Then, Eq. (5.4.6) gives
∞ π
simply the Fourier series expansion coefficients. In terms of the physical frequency f in 1
Hertz, the inverse DTFT reads as |x(n)|2 = |X(ω)|2 dω (Parseval) (5.4.7)
n=−∞ 2π −π

 fs /2
1 The DTFT can be given a geometric interpretation by recognizing that the points
x(n)= X(f )e2πjf n/fs df
fs −fs /2 z = ejω lie on the unit circle on the z-plane. As ω varies over the Nyquist interval
[−π, π], the complex point z = ejω moves around the unit circle, as shown in Fig. 5.4.1.
As an example, consider a (double-sided) complex sinusoid of frequency ω0 : The phase angle of z is ω.
5.4. FREQUENCY SPECTRUM 199 200 5. Z-TRANSFORMS

The corresponding spectrum and its magnitude are obtained by replacing z by ejω :
e jω
ejω − z1 |ejω − z1 |
X(ω)= ⇒ |X(ω)| =
ejω − p1 |ejω − p1 |
ω= π ω ω=0
Figure 5.4.2 shows the relative locations of the fixed points z1 , p1 and the moving
0 1
point z = ejω . A rough plot of |X(ω)| based on this pole/zero pattern is also shown.
The magnitude spectrum |X(ω)| is the ratio of the distance of the point ejω to the zero
z1 , namely, |ejω − z1 | divided by the distance of ejω to the pole p1 , namely, |ejω − p1 |.
unit
circle |z-p1|
pole
peak
Fig. 5.4.1 Evaluation of z-transform on the unit circle. p1 e jω |z-z1| |X(ω)|

In order for the spectrum X(ω) to exist,† the ROC of the z-transform X(z) must ω
z1 zero
contain the unit circle; otherwise the z-transform will diverge at the unit circle points ω1 dip
ϕ1
z = ejω . But if the ROC contains the unit circle, the signal x(n) must be stable. Thus, 0 1
the Fourier transform X(ω) exists only for stable signals.
Marginally stable signals, such as sinusoids, strictly speaking do not have a spectrum ω
because their poles lie on the unit circle and therefore the evaluation of X(z) on the 0 ϕ1 ω1 π
unit
unit circle will cause X(z) to diverge at certain z’s. However, it is intuitively useful to circle
consider their spectra. For example, for the causal complex sinusoid of the previous
section we have:
Fig. 5.4.2 Geometric interpretation of frequency spectrum.
Z 1
x(n)= ejω0 n u(n)−→ X(z)=
1 − ejω0 z−1 As ejω moves around the unit circle, these distances will vary. As ejω passes near
jω the pole, the denominator distance will become small causing the value of |X(ω)| to
and therefore the formal replacement of z by e will yield
increase. If ω1 is the phase angle of the pole p1 , then the point of closest approach
|X(ω)| ∞ to p1 will occur at ω = ω1 causing a peak in |X(ω)| there. The closer the pole is to
the unit circle, the smaller the denominator distance will become at ω = ω1 , and the
1 1 sharper the peak of |X(ω)|.
X(ω)= =
1 − ejω0 e−jω 1 − ej(ω0 −ω) Similarly, as ejω passes near the zero z1 , the numerator distance will become small,
causing |X(ω)| to decrease. At the zero’s phase angle, say ω = φ1 , this distance will
0 ω be smallest, causing a dip in |X(ω)| there. The closer the zero to the unit circle, the
ω0 π
sharper the dip. The zero z1 can also lie on the unit circle, in which case |X(ω)| will
which diverges at ω = ω0 . However, this is to be expected because if the signal were vanish exactly at ω = φ1 .
a pure sinusoid x(n)= ejω0 n , its spectrum would be a single spectral line concentrated In summary, we can draw a rough sketch of the spectrum |X(ω)| by letting ejω
at ω = ω0 , that is, X(ω)= 2πδ(ω − ω0 ) (plus its Nyquist replicas). Here, the signal trace the unit circle and draw peaks as ejω passes near poles, and dips as it passes near
is not a pure sinusoid; it is a causal, truncated version of a pure sinusoid and therefore zeros. By proper location of the zeros and poles of X(z) or H(z), one can design any
additional frequencies are present. However, the dominant frequency is still ω0 . desired shape for X(ω) or H(ω).
The shape of the spectrum X(ω) or H(ω) is affected by the pole/zero pattern of It is convenient to divide the unit circle into low-, medium-, and high-frequency
the z-transform X(z) or H(z), that is, by the relative geometric locations of the poles wedge regions, as shown in Fig. 5.4.3. This subdivision is somewhat arbitrary because
and zeros on the z-plane. To see this, consider a simple z-transform having a single what is “low” or “high” frequency depends on the application. It aids, however, in the
pole at z = p1 and a single zero at z = z1 . placement of poles and zeros. For example, to make a lowpass filter that emphasizes low
frequencies and attenuates high ones, one would place poles inside the circle somewhere
1 − z1 z−1 z − z1
X(z)= = within the low-frequency wedge and/or zeros within the high-frequency wedge.
1 − p1 z−1 z − p1
Such filter design methods are somewhat crude and are used in practice only for
† That is, to be finite X(ω)≠ ∞ for all ω. the design of simple and/or specialized filters, such as resonator or notch filters or
5.4. FREQUENCY SPECTRUM 201 202 5. Z-TRANSFORMS

medium 5.5 Inverse z-Transforms

j The problem of inverting a given z-transform X(z) is to find the time signal x(n) whose
ω= π/2 z-transform is X(z). As we saw already, the answer for x(n) is not necessarily unique.
But it can be made unique by specifying the corresponding ROC.
In inverting a z-transform, it is convenient to break it into its partial fraction (PF)
expansion form, that is, into a sum of individual pole terms of the type (5.3.2).
high -1 ω= π 0 ω=0 1 low
Once X(z) is written in the form (5.3.2), one still needs to know how to invert each
term, that is, causally or anticausally. This depends on the choice of the ROC.
In general, the circles through the poles at z = p1 , z = p2 , and so on, divide the
ω=-π/2 z-plane into non-overlapping regions, which are all possible candidates for ROCs. Any
-j one of these ROC regions will result into a different x(n). Among all possible x(n),
there will be a unique one that is stable, because the unit circle lies in exactly one of the
medium possible ROCs.

Fig. 5.4.3 Low-, medium-, and high-frequency parts of the unit circle. Example 5.5.1: In Example (5.2.3), the first three signals had a common z-transform:

1 1
X(z)= +
biquadratic filter sections for digital audio graphic and parametric equalizers. Such 1 − 0.8z−1 1 − 1.25z−1
design examples will be considered later on.
The two circles through the poles at z = 0.8 and z = 1.25 divide the z-plane into the
The DTFT X(ω) of a signal x(n) is a complex-valued quantity and therefore, it can
three regions I, II, III, shown in Example 5.2.3. There are therefore three possible inverse
be characterized also by its real and imaginary parts Re X(ω), Im X(ω) or, in its polar
z-transforms, that is, three different signals x(n) corresponding to the three ROC choices.
form, by its magnitude and phase responses |X(ω)|, arg X(ω). Thus, But, only II is stable. ⊔
⊓

X(ω)= Re X(ω)+j Im X(ω)= |X(ω)| ej arg X(ω) The partial fraction expansion method can be applied to z-transforms that are ratios
of two polynomials in z−1 of the form:
For real-valued signals x(n), the quantity X(ω) satisfies the following so-called
hermitian property: N(z)
X(z)=
D(z)
X(ω)∗ = X(−ω) (5.4.8)
The zeros of the denominator polynomial D(z) are the poles of X(z). Assuming
which translates to the following relationships for the magnitude and phase responses: D(z) has degree M, there will be M denominator zeros, say at p1 , p2 , . . . , pM , and D(z)
may be assumed to be in the factored form
|X(ω)| = |X(−ω)|
(5.4.9) D(z)= (1 − p1 z−1 )(1 − p2 z−1 )· · · (1 − pM z−1 )
arg X(ω) = − arg X(−ω)
The partial fraction expansion of X(z) is given by†
that is, the magnitude response is even in ω and the phase response odd. Similar defini-
tions and results apply to the frequency response H(ω) of a real-valued system h(n). N(z) N(z)
We note finally that the multiplicative filtering property Y(z)= H(z)X(z) evaluated X(z) = =
D(z) (1 − p1 z−1 )(1 − p2 z−1 )· · · (1 − pM z−1 )
on the unit circle takes the following frequency-domain form: (5.5.1)
A1 A2 AM
= + + ··· +
Y(ω)= H(ω)X(ω) (filtering in frequency domain) (5.4.10) 1 − p1 z−1 1 − p2 z−1 1 − pM z−1
For this expansion to be possible as an identity in z−1 , the degree of the numerator
Its consequences will be explored later on.
polynomial N(z) must be strictly less than the degree M of the denominator polynomial.
The PF expansion coefficients Ai can be computed by the formulas:
† We have assumed that all the poles are single poles.
5.5. INVERSE Z-TRANSFORMS 203 204 5. Z-TRANSFORMS

⎡ ⎤
  ⎢ N(z) ⎥ N(z) Q(z)D(z)+R(z) R(z)
⎢ ⎥ X(z)= = = Q(z)+
Ai = (1 − pi z−1 )X(z) z=pi = ⎢+ ⎥ (5.5.2) D(z) D(z) D(z)
⎣ (1 − p z ) ⎦
−1
j
j=i z=pi where now the second term will admit an ordinary PF expansion of the form (5.5.1) be-
−1 cause the degree of the remainder polynomial R(z) is strictly less than M. Alternatively,
for i = 1, 2, . . . , M. In words, the factor (1 − pi z ) is deleted from the denominator
one may simply remove the numerator polynomial N(z) altogether, then carry out an
and the remaining expression is evaluated at the pole z = pi .
ordinary PF expansion of the quantity
Example 5.5.2: In Example 5.2.3 the z-transform was written in the form 1
W(z)=
D(z)
2 − 2.05z−1 2 − 2.05z−1
X(z)= =
1 − 2.05z−1 + z−2 (1 − 0.8z−1 )(1 − 1.25z−1 ) and finally restore the numerator by writing

Because the numerator polynomial has degree one in the variable z−1 , there is a PF expan- X(z)= N(z)W(z)
sion of the form:
We may refer to this method as the “remove/restore” method. Some examples will
2 − 2.05z−1 A1 A2 illustrate these techniques.
X(z)= = +
(1 − 0.8z−1 )(1 − 1.25z−1 ) 1 − 0.8z−1 1 − 1.25z−1
Example 5.5.3: We emphasize that a PF expansion may exist in one independent variable, say
The two coefficients are obtained by Eq. (5.5.2) as follows: z−1 , but not in another, say z. For example, the z-transform
! "
  2 − 2.05z−1 2 − 2.05/0.8 2 − 2.05z−1 z(2z − 2.05)
A1 = (1 − 0.8z−1 )X(z) z=0.8 = = =1 X(z)= =
1 − 1.25z−1 1 − 1.25/0.8 (1 − 0.8z−1 )(1 − 1.25z−1 ) (z − 0.8)(z − 1.25)
z=0.8
! "
  2 − 2.05z−1 has numerator of degree one with respect to the variable z−1 , but degree two with respect
A2 = (1 − 1.25z−1 )X(z) z=1.25 = =1
1 − 0.8z−1 to z. Thus, it admits an expansion of the form (5.5.1) with respect to z−1 , but not with
z=1.25
respect to z.
which are as expected. ⊔
⊓ Many texts prefer to work with z and therefore to make the PF expansion possible, a factor
z is divided out to lower the degree of the numerator and then restored at the end, that is,
If the degree of the numerator polynomial N(z) is exactly equal to the degree M
of the denominator D(z), then the PF expansion (5.5.1) must be modified by adding an X(z) (2z − 2.05) A1 A2
= = +
extra term of the form: z (z − 0.8)(z − 1.25) z − 0.8 z − 1.25

N(z) N(z) When z is restored, one gets

X(z) = =
D(z) (1 − p1 z−1 )(1 − p2 z−1 )· · · (1 − pM z−1 )
(5.5.3) zA1 zA2 A1 A2
X(z)= + = +
A1 A2 AM z − 0.8 z − 1.25 1 − 0.8z−1 1 − 1.25z−1
= A0 + + + ··· +
1 − p1 z−1 1 − p2 z−1 1 − pM z−1
It is easily verified that the PF expansion coefficients will be the same in the two approaches.
The coefficients Ai , i = 1, 2, . . . , M are computed in exactly the same way by Eq. (5.5.2). In this book, we prefer to work directly with z−1 and avoid the extra algebraic steps required
The extra coefficient A0 is computed by evaluating the z-transform at z = 0, that is, to write everything in terms of z, divide by z, restore z, and rewrite the final answer in
terms of z−1 . ⊔
⊓

A0 = X(z)z=0 (5.5.4)
Example 5.5.4: Compute all possible inverse z-transforms of
If the degree of N(z) is strictly greater than M, one may divide the polynomial D(z)
into N(z), finding the quotient and remainder polynomials, so that z-plane I
6+z −1 II
N(z)= Q(z)D(z)+R(z) X(z)= -0.5 0.5
1 − 0.25z−2 0
and then writing
5.5. INVERSE Z-TRANSFORMS 205 206 5. Z-TRANSFORMS

Solution: Because the numerator has degree one in z−1 , we have the PF expansion: Again, there are only two ROCs I and II: |z| > 0.5 and |z| < 0.5. For the first ROC, the A1
and A2 terms are inverted causally, and the A0 term inverts into a simple δ(n):
6 + z−1 6 + z−1 A1 A2
X(z)= = = +
1 − 0.25z−2 (1 − 0.5z−1 )(1 + 0.5z−1 ) 1 − 0.5z−1 1 + 0.5z−1 x(n)= A0 δ(n)+A1 (0.5)n u(n)+A2 (−0.5)n u(n)

where
For the second ROC, we have:
! " ! "
−1 −1
6+z 6+z
A1 = = 4, A2 = =2 x(n)= A0 δ(n)−A1 (0.5)n u(−n − 1)−A2 (−0.5)n u(−n − 1)
1 + 0.5z−1 z=0.5
1 − 0.5z−1 z=−0.5

The two poles at ±0.5 have the same magnitude and therefore divide the z-plane into two Only the first inverse is stable because its ROC contains the unit circle. ⊔
⊓
ROC regions I and II: |z| > 0.5 and |z| < 0.5. For the first ROC, both terms in the PF
expansion are inverted causally giving: Example 5.5.6: Determine the causal inverse z-transform of

x(n)= A1 (0.5)n u(n)+A2 (−0.5)n u(n)

z-plane I
6+z −5 II
X(z)= -0.5 0.5
Because this ROC also contains the unit circle the signal x(n) will be stable. For the second 1 − 0.25z−2 0
ROC, both PF expansion terms are inverted anticausally giving:

x(n)= −A1 (0.5)n u(−n − 1)−A2 (−0.5)n u(−n − 1)

Solution: Here, the degree of the numerator is strictly greater than that of the denominator.
This answer is unstable, because the ROC does not contain the unit circle. ⊔
⊓ The first technique is to divide the denominator into the numerator, giving

Example 5.5.5: Determine all inverse z-transforms of (6 + z−5 )= (1 − 0.25z−2 )(−16z−1 − 4z−3 )+(6 + 16z−1 )

z-plane I where (6 + 16z−1 ) is the remainder polynomial and (−16z−1 − 4z−3 ) the quotient. Then,

10 + z−1 − z−2 II
X(z)= -0.5 0.5 6 + z−5 6 + 16z−1
1 − 0.25z−2 0 X(z)= = −16z−1 − 4z−3 +
1 − 0.25z−2 1 − 0.25z−2

and expanding the last term in PF expansion:

Solution: Ordinary partial fraction expansion is not valid in this case because the degree of the 19 13
numerator is the same as the degree of the denominator. However, we may still have an X(z)= −16z−1 − 4z−3 + −
1 − 0.5z−1 1 + 0.5z−1
expansion of the form (5.5.3)
The causal inverse, having ROC |z| > 0.5, will be:
10 + z−1 − z−2 10 + z−1 − z−2
X(z) = =
1 − 0.25z−2 (1 − 0.5z−1 )(1 + 0.5z−1 )
x(n)= −16δ(n − 1)−4δ(n − 3)+19(0.5)n u(n)−13(−0.5)n u(n)
A1 A2
= A0 + +
1 − 0.5z−1 1 + 0.5z−1 The second technique is the “remove/restore” method. Ignoring the numerator we have

where A1 and A2 are determined in the usual manner and A0 is determined by evaluating 1 0. 5 0.5
X(z) at z = 0: W(z)= = +
1 − 0.25z−2 1 − 0.5z−1 1 + 0.5z−1
! " ! "
10 + z−1 − z−2 10z2 + z − 1 −1 which has the causal inverse
A0 = = = =4
1 − 0.25z−2 z=0
z2 − 0.25 z=0
−0.25
w(n)= 0.5(0.5)n u(n)+0.5(−0.5)n u(n)
! " ! "
10 + z−1 − z−2 10 + z−1 − z−2 Once w(n) is known, one can obtain x(n) by restoring the numerator:
A1 = = 4, A2 = =2
1 + 0.5z−1 z=0.5
1 − 0.5z−1 z=−0.5
5.5. INVERSE Z-TRANSFORMS 207 208 5. Z-TRANSFORMS

The assumption that the numerator and denominator polynomials N(z) and D(z)
have real-valued coefficients implies that the complex-valued poles of X(z) come in
X(z)= (6 + z−5 )W(z)= 6W(z)+z−5 W(z)
complex-conjugate pairs. In that case, the PF expansion takes the form
Taking inverse z-transforms of both sides and using the delay property, we find z-plane p1
p2
n
x(n) = 6w(n)+w(n − 5)= 3(0.5) u(n)+3(−0.5) u(n) n ∗
A1 A1 A2 0 ω1
X(z)= + + + ··· p3
+ 0.5(0.5)n−5 u(n − 5)+0.5(−0.5)n−5 u(n − 5) 1 − p1 z−1 1 − p∗
1z
−1 1 − p2 z−1 R1
p*
1
The two expressions for x(n) from the two techniques are equivalent. ⊔
⊓
where the PF expansion coefficients also come in conjugate pairs. Thus, it is necessary
Example 5.5.7: Determine all possible inverse z-transforms of
to determine only one of them, not both. The corresponding inverse z-transform will
be real-valued; indeed, considering the causal case we have
z-plane
IV
III x(n)= A1 pn1 u(n)+A∗ ∗n n
1 p1 u(n)+A2 p2 u(n)+ · · ·
II
7 − 9.5z−1 − 3.5z−2 + 5.5z−3 -1 I 1 1.5
X(z)= Because the first two terms are complex conjugates of each other, we may use the
(1 − z−2 )(1 − 0.5z−1 )(1 − 1.5z−1 ) 0.5 result that C + C∗ = 2Re(C), for any complex number C, to write the first term as
 
A1 pn1 + A∗ ∗n n
1 p1 = 2Re A1 p1

Writing A1 and p1 in their polar form, say, A1 = B1 ejα1 and p1 = R1 ejω1 , with
Solution: X(z) admits the PF expansion:
B1 > 0 and R1 > 0, we have
1 1 3 2      
X(z)= + + +
1 − z−1 1 + z−1 1 − 0.5z−1 1 − 1.5z−1 Re A1 pn
1 = Re B1 e
jα1 n jω1 n
R1 e = B1 Rn1 Re ejω1 n+jα1

where the PF expansion coefficients are easily found. The four poles at z = 0.5, 1, −1, and taking the real part of the exponential, we find
1.5 divide the z-plane into the four ROC regions I, II, III, IV. Region I corresponds to the
completely anticausal inverse and region IV to the completely causal one. For region II, the  
A1 pn1 + A∗ ∗n n n
1 p1 = 2Re A1 p1 = 2B1 R1 cos(ω1 n + α1 )
pole at z = 0.5 will be inverted causally and the rest anticausally. For region III, z = 0.5
and z = ±1 will be inverted causally and z = 1.5 anticausally. Thus, the four possible and for x(n)
inverse z-transforms are:
  x(n)= 2B1 Rn1 cos(ω1 n + α1 )u(n)+A2 pn2 u(n)+ · · ·
x1 (n) = − 1 + (−1)n +3(0.5)n +2(1.5)n u(−n − 1)
 
x2 (n) = 3(0.5)n u(n)− 1 + (−1)n +2(1.5)n u(−n − 1) Thus, complex-valued poles correspond to exponentially decaying sinusoids (if R1 <
  1). The decay envelope Rn
x3 (n) = 1 + (−1)n +3(0.5)n u(n)−2(1.5)n u(−n − 1) 1 and the frequency ω1 depend on the complex pole by p1 =

  R1 ejω1 .
x4 (n) = 1 + (−1)n +3(0.5)n +2(1.5)n u(n) The first-order terms in the partial fraction expansion corresponding to complex con-
jugate poles can be reassembled into second-order terms with real-valued coefficients,
Strictly speaking there is no stable answer because two of the poles, z = ±1, lie on the
as follows:
unit circle. However, x2 (n) and x3 (n) are marginally stable, that is, neither diverging nor
converging to zero for large n. In both cases, the anticausal term (1.5)n tends to zero for A1 A∗1 (A1 + A∗ ∗ ∗
1 )−(A1 p1 + A1 p1 )z
−1
large negative n. Indeed, because n is negative, we write n = −|n| and −1
+ ∗ −1 = −1 ∗ −1
1 − p1 z 1 − p1 z (1 − p1 z )(1 − p1 z )

(1.5)n = (1.5)−|n| → 0 as n → −∞ Using the identities

The terms due to the poles z = ±1 are causal or anticausal in cases III and II, but they (1 − p1 z−1 )(1 − p∗
1z
−1
)= 1 − 2Re(p1 )z−1 + |p1 |2 z−2
remain bounded. The other two signals x1 (n) and x4 (n) are unstable because the unit
circle does not lie in their ROCs. ⊔
⊓ or,
5.5. INVERSE Z-TRANSFORMS 209 210 5. Z-TRANSFORMS

Writing jn+1 = ejπ(n+1)/2 and taking real parts we find

(1 − R1 ejω1 z−1 )(1 − R1 e−jω1 z−1 )= 1 − 2R1 cos(ω1 )z−1 + R21 z−2    π(n + 1)   πn 
Re jn+1 = cos = − sin
2 2
and writing
and
A1 + A∗
1 = 2Re(A1 )= 2B1 cos(α1 )  πn 
x(n)= 4δ(n)−6(0.5)n sin u(n)
∗ ∗ ∗ 2
A1 p1 + A1 p1 = 2Re(A1 p1 )= 2B1 R1 cos(α1 − ω1 )
Similarly, we find
we find
 πn 
A1 A∗1 2B1 cos(α1 )−2B1 R1 cos(α1 − ω1 )z−1 x(n)= 4δ(n)+6(0.5)n sin u(−n − 1)
+ ∗ −1 = 2
1 − p1 z−1 1 − p1 z 1 − 2R1 cos(ω1 )z−1 + R21 z−2
for the anticausal version with ROC |z| < 0.5. Some additional examples with complex
having real-valued coefficients.
conjugate poles were cases (6-9) of Example 5.2.2. ⊔
⊓
Example 5.5.8: Determine all possible inverse z-transforms of

z-plane
5.6 Problems
0.5j
I
5.1 Prove the linearity, delay, and convolution properties of z-transforms given by Eqs. (5.1.3)–
4 − 3z−1 + z−2 II
X(z)= (5.1.5).
1 + 0.25z−2 0
5.2 Compute the z-transform of the following sequences and determine the corresponding re-
-0.5j gion of convergence:

a. x(n)= δ(n − 5)
Solution: We write
b. x(n)= δ(n + 5)
4 − 3z−1 + z−2 4 − 3z−1 + z−2
X(z) = = c. x(n)= u(n − 5)
1 + 0.25z−2 (1 − 0.5jz−1 )(1 + 0.5jz−1 )
d. x(n)= u(−n + 5)
A1 A∗
1
= A0 + + 5.3 Compute the z-transform of the following sequences and determine the corresponding re-
1 − 0.5jz−1 1 + 0.5jz−1
gion of convergence:
with the numerical values:
a. x(n)= (−0.5)n u(n)
! " ! "  
−1
4 − 3z + z −2 −1
4 − 3z + z −2 b. x(n)= (−0.5)n u(n)−u(n − 10)
A0 = = 4, A1 = = 3j
1 + 0.25z−2 z=0
1 + 0.5jz−1 z=0.5j c. x(n)= (0.5)n u(n)+(−0.5)n u(n)

5.4 Compute the z-transform of the following sequences and determine the corresponding re-
Therefore,
gion of convergence:

3j 3j 3z−1 a. x(n)= 2(0.8)n cos(πn/2)u(n)

X(z)= 4 + −
− −
=4−
1 − 0.5jz 1 1 + 0.5jz 1 1 + 0.25z−2
b. x(n)= (0.8j)n u(n)+(−0.8j)n u(n)

The causal ROC is |z| > |0.5j| = 0.5, resulting in 5.5 Compute the z-transform of the following sequences and determine the corresponding re-
gion of convergence:
x(n)= 4δ(n)+3j(0.5j)n u(n)−3j(−0.5j)n u(n) a. x(n)= (0.25)n u(n)+4n u(n)
b. x(n)= (0.25)n u(n)−4n u(−n − 1)
Because the last two terms are complex conjugates of each other, we may write them as
c. x(n)= −(0.25)n u(−n − 1)−4n u(−n − 1)
   
x(n)= 4δ(n)+2Re 3j(0.5j) u(n) = 4δ(n)+6(0.5) u(n)Re jn+1
n n d. Explain why x(n)= −(0.25)n u(−n − 1)+4n u(n) does not have a z-transform.
5.6. PROBLEMS 211 212 5. Z-TRANSFORMS

5.6 Using the power series definition of z-transforms, derive the z-transform and its ROC of the 5.15 Consider the z-transform for |z| > 1:
signal x(n)= cos(πn/2)u(n).
5.7 Using partial fractions or power series expansions, compute the inverse z-transform of the X(z)= 1 − z−2 + z−4 − z−6 + z−8 − · · ·
following z-transforms and determine whether the answer is causal and/or stable:
Derive a rational expression for X(z) in two ways: (a) by summing the above series, and (b)
a. X(z)= (1 − 4z−2 )(1 + 3z−1 ) by showing that it satisfies the equation X(z)= 1 − z−2 X(z).
b. X(z)= 5 + 3z3 + 2z−2 Derive also the inverse z-transform x(n) for all n.

5.8 Using partial fractions or power series expansions, determine all possible inverse z-transforms 5.16 Without using partial fractions, determine the causal inverse z-transforms of:
of the following z-transforms, sketch their ROCs, and discuss their stability and causality
1
properties: a. X(z)=
1 + z−4
3(1 + 0.3z−1 ) 1
a. X(z)= b. X(z)=
1 − 0.81z−2 1 − z−4
6 − 3z−1 − 2z−2 1
b. X(z)= c. X(z)=
1 − 0.25z−2 1 + z−8
6 + z−5 1
c. X(z)= d. X(z)=
1 − 0.64z−2 1 − z−8

10 + z−2 5.17 Using partial fraction expansions, determine the inverse z-transforms of Problem 5.16. Ver-
d. X(z)=
1 + 0.25z−2 ify that you get the same answers as in that problem.
6 − 2z−1 − z−2 5.18 Consider a transfer function H(z)= N(z)/D(z), where the numerator and denominator
e. X(z)= , ROC |z| > 1
(1 − z−1 )(1 − 0.25z−2 ) polynomials have real-valued coefficients and degrees L and M in z−1 , and assume L > M.
1 Show that H(z) can be written in the form:
f. X(z)= −4 +
1 + 4z−2
K
4 − 0.6z−1 + 0.2z−2 bi0 + z−1 bi1
g. X(z)= H(z)= Q(z)+
(1 − 0.5z−1 )(1 + 0.4z−1 ) i=1
1 + ai1 z−1 + ai2 z−2

5.9 Consider the z-transform pair:

where Q(z) is a polynomial of degree L − M in z−1 and the second-order sections have
1 real coefficients. The number of sections K is related to M by K = M/2 if M is even and
xa (n)= an u(n) ⇔ Xa (z)=
1 − az−1 K = (M − 1)/2 if M is odd. This result forms the basis of the parallel realization form of
H(z).
Applying the derivative operator ∂/∂a to the pair, derive the z-transform of the sequence
x(n)= nan u(n). 5.19 Determine the factorization into first-order zeros of:
∂ D−
5.10 Consider the differential operator D = a . First, show that its k-fold application gives +1
∂a 1 − z−D = (1 − zk z−1 )
Dk an = nk an . Then, use this result to obtain the z-transform of x(n)= nk an u(n). Derive k=0
the explicit transforms for the cases k = 1, 2, 3, 4.
D−
+1
5.11 Show the z-transform of a triangular signal: 1 + z−D = (1 − zk z−1 )
k=0
L
! "2
 |n|  −n 1 1 − z−L
1− z = zL−1 where D is an integer. What are the zeros zk in the two cases? For D = 4 and D = 8, place
n=−L
L L 1 − z−1
these zeros on the z-plane with respect to the unit circle.

5.12 Using Euler’s formula and the z-transform pair of Problem 5.9, derive the z-transforms of 5.20 Given a > 0 and integer D, repeat the previous problem for:
the signals x(n)= Rn cos(ω0 n)u(n) and x(n)= Rn sin(ω0 n)u(n).
D−
+1
5.13 Consider the causal sequence x(n)= {a0 , a1 , a2 , a3 , a0 , a1 , a2 , a3 , · · · }, where the dots indi- 1 − az−D = (1 − zk z−1 )
cate the periodic repetition of the four samples {a0 , a1 , a2 , a3 }. Determine the z-transform k=0
of x(n) and the corresponding ROC. D−
+1
5.14 Using partial fraction expansions, determine the inverse z-transform of the z-transform of 1 + az−D = (1 − zk z−1 )
Problem 5.13. Verify that the sum of the PFE terms generate the periodic sequence x(n) of k=0

that problem.