2 The z and Fourier transforms for discrete time signals

Continuous time: Laplace transform, and (C.T.)Fourier transform as special case

Discrete time: z transform, and (D.T.) Fourier transform as special case

Laplace transf.: 2D function of a 2D variable,sometimes less practical,

but converges for more signals than Fourier (e.g. unstable systems/signals).
2.1 Introduction Fourier transf.: 2D function of a 1D variable, and |X(w)| is a 1D function
of a 1D variable, easier to use (or visualize) for some applications

In Chapter 1, we studied linear time-invariant systems, using both impulse responses

and difference equations to characterize them. In this chapter, we study another very
useful way to characterize discrete-time systems. It is linked with the fact that, when
"eigenfunction"
an exponential function is input to a linear time-invariant system, its output is an
exponential function of the same type, but with a different amplitude. This can be
deduced by considering that, from equation (1.38), a linear time-invariant discrete-time
system with impulse response h(n), when excited by an exponential x(n) = z n ,
produces at its output a signal y(n) such that
∞
 ∞
 ∞

y(n) = x(n − k)h(k) = z n−k h(k) = z n h(k)z −k (2.1)
k=−∞ k=−∞ k=−∞

that is, the signal at the output is also an exponential z n , but with an amplitude
multiplied by the complex function
∞

H (z) = h(k)z −k (2.2)
k=−∞
simple scalar, In this chapter, we characterize linear time-invariant systems using the quantity
depends on z H (z) in the above equation, commonly known as the z transform of the discrete-time
value sequence h(n). As we will see later in this chapter, with the help of the z transform,
linear convolutions can be transformed into simple algebraic equations. The impor-
tance of this for discrete-time systems parallels that of the Laplace transform for
continuous-time systems.
The case when z n is a complex sinusoid with frequency ω, that is, z = e jω , is of
particular importance. In this case, equation (2.2) becomes
∞ special case where z magnitude is unity

jω − jωk
H (e ) = h(k)e (2.3)
k=−∞

which can be represented in polar form as H (e jω ) = |H (e jω )|e j(ω) , yielding, from

equation (2.1), an output signal y(n) such that

y(n) = H (e jω )e jωn = |H (e jω )|e j(ω) e jωn = |H (e jω )|e jωn+ j(ω) (2.4)

change of magnitude and phase,

36
same for cos(wn) and sin(wn) functions, since they
can be expressed in terms of exp(jwn).
37 2.2 Definition of the z transform

This relationship implies that the effect of a linear system characterized by H (e jω ) on a

complex sinusoid is to multiply its amplitude by |H (e jω )| and to add (ω) to its phase.
For this reason, the descriptions of |H (e jω )| and (ω) as functions of ω are widely
used to characterize linear time-invariant systems, and are known as their magnitude
and phase responses, respectively. The complex function H (e jω ) in equation (2.4) is
also known as the Fourier transform of the discrete-time sequence h(n). The Fourier
transform is as important for discrete-time systems as it is for continuous-time systems.
In this chapter, we will study the z and Fourier transforms for discrete-time signals.
We begin by defining the z transform, discussing issues related to its convergence and
its relation to the stability of discrete-time systems. Then we present the inverse z
transform, as well as several z-transform properties. Next, we show how to transform
discrete-time convolutions into algebraic equations, and introduce the concept of
a transfer function. We then present an algorithm to determine, given the transfer
function of a discrete-time system, whether the system is stable or not and go on to
discuss how the frequency response of a system is related to its transfer function.
At this point, we give a formal definition of the Fourier transform of discrete-time
signals, highlighting its relations to the Fourier transform of continuous-time signals.
An expression for the inverse Fourier transform is also presented. Its main properties
are then shown as particular cases of those of the z transform. We close the chapter by
presenting some MATLAB functions which are related to z and Fourier transforms,
and which aid in the analysis of transfer functions of discrete-time systems.

2.2 Definition of the z transform

The z transform of a sequence x(n) is defined as

∞

X (z) = Z {x(n)} = x(n)z −n (2.5)
n=−∞

where z is a complex variable. Note that X (z) is only defined for the regions of the
complex plane in which the summation on the right converges. ROC, "region of convergence"
Very often, the signals we work with start only at n = 0, that is, they are nonzero
only for n ≥ 0. Because of that, some textbooks define the z transform as
mostly useful for solving difference equations with non-zero
∞
initial conditions, like the unilateral

−n
X U (z) = x(n)z Laplace transform
(2.6)
n=0
unilateral
which is commonly known as the one-sided z transform, while equation (2.5) is
referred to as the two-sided z transform. Clearly, if the signal x(n) is nonzero for
n < 0, then the one-sided and two-sided z transforms are different. In this text, we
work only with the two-sided z transform, which is referred to, without any risk of
ambiguity, just as the z transform.
bilateral
38 The z and Fourier transforms

As mentioned above, the z transform of a sequence exists only for those regions of
the complex plane in which the summation in equation (2.5) converges. The example
below clarifies this point.

E XAM PL E 2.1
Compute the z transform of the sequence x(n) = K u(n).

SOL UT I O N
By definition, the z transform of K u(n) is
∞
 ∞ 
 n
−n
X (z) = K z =K z −1 (2.7)
n=0 n=0

Thus, X (z) is the sum of a power series which converges only if |z −1 | < 1. In such a
case, X (z) can be expressed as
K Kz
X (z) = −1
= , |z| > 1 (2.8)
1−z z−1

Note that for |z| < 1, the nth term of the summation, z −n , tends to infinity as
n → ∞, and therefore X (z) is not defined. For z = 1, the summation is also infinite.
For z = −1, the summation oscillates between 1 and 0. In none of these cases does
the z transform converge.
△

It is important to note that the z transform of a sequence is a Laurent series in the

complex variable z (Churchill, 1975). Therefore, the properties of Laurent series apply
directly to the z transform. As a general rule, we can apply a result from series theory
stating that, given a series of the complex variable z
∞

S(z) = f i (z) (2.9)
i=0

such that | f i (z)| < ∞, i = 0, 1, . . ., and given the quantity

 
 f n+1 (z) 
α(z) = lim    (2.10)
n→∞ f n (z) 

then the series converges absolutely if α(z) < 1, and diverges if α(z) > 1 (Kreyszig,
1979). Note that, for α(z) = 1, the above procedure tells us nothing about the
convergence of the series, which must be investigated by other means. One can justify
this by noting that, if α(z) < 1, the terms of the series are under an exponential a n for
some a < 1, and therefore their sum converges as n → ∞. One should clearly note
that, if | f i (z)| = ∞, for some i, then the series is not convergent.
39 2.2 Definition of the z transform

The above result can be extended for the case of two-sided series as in the equation
below
∞

S(z) = f i (z) (2.11)
i=−∞

if we express S(z) above as the sum of two series S1 (z) and S2 (z) such that
∞
 −1

S1 (z) = f i (z) and S2 (z) = f i (z) (2.12)
i=0 i=−∞

then S(z) converges if the two series S1 (z) and S2 (z) converge. Therefore, in this case,
we have to compute the two quantities
   
 f n+1 (z)   f n+1 (z) 
α1 (z) = lim   and α2 (z) = lim   (2.13)
n→∞ f n (z)  n→−∞  f n (z) 

Naturally, S(z) converges absolutely if α1 (z) < 1 and α2 (z) > 1. The condition
α1 (z) < 1 is equivalent to saying that, for n → ∞, the terms of the series are
under a n for some a < 1. The condition α2 (z) > 1 is equivalent to saying that,
for n → −∞, the terms of the series are under bn for some b > 1. One should note
that, for convergence, we must also have | f i (z)| < ∞, ∀i.
Applying these convergence results to the z-transform definition given in equa-
tion (2.5), we conclude that the z transform converges if
 
 x(n + 1)z −n−1     
 x(n + 1) 
  −1 
α1 = lim  = z lim  <1 (2.14)

n→∞  x(n)z −n  n→∞  x(n) 
 
 x(n + 1)z −n−1     
 x(n + 1) 
  −1 
α2 = lim  = z lim  >1 (2.15)

n→−∞  x(n)z −n  n→−∞  x(n) 

Defining
 
 x(n + 1) 
r1 = lim   (2.16)
n→∞ x(n) 
 
 x(n + 1) 
r2 = lim   (2.17)
n→−∞ x(n) 
then equations (2.14) and (2.15) are equivalent to
r1 < |z| < r2 (2.18)
That is, the z transform of a sequence exists in an annular region of the complex plane
defined by equation (2.18) and illustrated in Figure 2.1. It is important to note that, for
some sequences, r1 = 0 or r2 → ∞. In these cases, the region of convergence may or
may not include z = 0 or |z| = ∞, respectively.
We now take a closer look at the convergence of z transforms for four important
classes of sequences.
40 The z and Fourier transforms

Im{z}

r2 r1
Re{z}

Figure 2.1 General region of convergence of the z transform.

• Right-handed, one-sided sequences: These are sequences such that x(n) = 0, for
n < n 0 , that is
includes causal signals (n0>=0)
∞

X (z) = x(n)z −n (2.19)
n=n 0

In this case, the z transform converges for |z| > r1 , where r1 is given by
equation (2.16). Since |x(n)z −n | must be finite, then, if n 0 < 0, the convergence
region excludes |z| = ∞.
• Left-handed, one-sided sequences: These are sequences such that x(n) = 0, for
n > n 0 , that is
includes anti-causal signals (no<=0)
n0

X (z) = x(n)z −n (2.20)
n=−∞

In this case, the z transform converges for |z| < r2 , where r2 is given by
equation (2.17). Since |x(n)z −n | must be finite, then, if n 0 > 0, the convergence
region excludes |z| = 0.
• Two-sided sequences: In this case,
∞

X (z) = x(n)z −n (2.21)
n=−∞
41 2.2 Definition of the z transform

and the z transform converges for r1 < |z| < r2 , where r1 and r2 are given by
equations (2.16) and (2.17). Clearly, if r1 > r2 , then the z transform does not exist.
• Finite-length sequences: These are sequences such that x(n) = 0, for n < n 0 and
n > n 1 , that is
n1

X (z) = x(n)z −n (2.22)
n=n 0

In such cases, the z transform converges everywhere except at the points such that
|x(n)z −n | = ∞. This implies that the convergence region excludes the point z = 0
if n 1 > 0 and |z| = ∞ if n 0 < 0.

E XAM PL E 2.2
Compute the z transforms of the following sequences, specifying their region of
convergence:

(a) x(n) = k2n u(n)

(b) x(n) = u(−n + 1)
(c) x(n) = −k2n u(−n − 1)
(d) x(n) = 0.5n u(n) + 3n u(−n)
(e) x(n) = 4−n u(n) + 5−n u(n + 1)

SOL UT I O N
∞

(a) X (z) = k2n z −n
n=0
This series converges if |2z −1 | < 1, that is, for |z| > 2. In this case, X (z) is the
sum of a geometric series, and therefore
k kz
X (z) = −1
= , for 2 < |z| ≤ ∞ (2.23)
1 − 2z z−2
1

(b) X (z) = z −n
n=−∞
This series converges if |z −1 | > 1, that is, for |z| < 1. Also, in order for the term
z −1 to be finite, |z|
= 0. In this case, X (z) is the sum of a geometric series, such
that
z −1 1
X (z) = = , for 0 < |z| < 1 (2.24)
1−z z − z2
−1

(c) X (z) = −k2n z −n
n=−∞
This series converges if | 2z | < 1, that is, for |z| < 2. In this case, X (z) is the sum
42 The z and Fourier transforms

of a geometric series, such that

−k 2z kz
X (z) = z = , for 0 ≤ |z| < 2 (2.25)
1− 2 z−2

∞
 0

(d) X (z) = 0.5n z −n + 3n z −n
n=0 n=−∞
This series converges if |0.5z −1 |
< 1 and |3z −1 | > 1, that is, for 0.5 < |z| < 3. In
this case, X (z) is the sum of two geometric series, and therefore
1 1 z 3
X (z) = + = + , for 0.5 < |z| < 3 (2.26)
1 − 0.5z −1 1 − 31 z z − 0.5 3 − z
∞
 ∞

−n −n
(e) X (z) = 4 z + 5−n z −n
n=0 n=−1
This series converges if | 14 z −1 | < 1 and | 15 z −1 | < 1, that is, for |z| > 41 . Also, the
−1
term for n = −1, 15 z −1

= 5z, is finite only for |z| < ∞. In this case, X (z) is
the sum of two geometric series, resulting in

1 5z 4z 25z 2 1
X (z) = + = + , for < |z| < ∞ (2.27)
1 − 41 z −1 1 − 51 z −1 4z − 1 5z − 1 4

In this example, although the sequences in items (a) and (c) are distinct, the
expressions for their z transforms are the same, the difference being only in their
regions of convergence. This highlights the important fact that, in order to completely
specify a z transform, its region of convergence must be supplied. In Section 2.3, when
we study the inverse z transform, this issue is dealt with in more detail.
△

In several cases we deal with causal and stable systems. Since for a causal system
its impulse response h(n) is zero for n < n 0 , then, from equation (1.48), we have that
a causal system is also BIBO stable if
∞

|h(n)| < ∞ (2.28)
n=n 0

Applying the series convergence criterion seen above, we have that the system is stable
only if
 
 h(n + 1) 
lim  =r <1 (2.29)
n→∞  h(n)  for causality,
ROC is outside
This is equivalent to saying that H (z), the z transform of h(n), converges for |z| > r .
Since, for stability, r < 1, then we conclude that the convergence region of the z
for stability, ROC includes
the unit circle
 for causality and stability:
43 2.3 Inverse z transform

transform of the impulse response of a stable causal system includes the region outside
the unit circle and the unit circle itself (in fact, if n 0 < 0, then this region excludes
|z| = ∞).
A very important case is when X (z) can be expressed as a ratio of polynomials, in
the form rational form

N (z)
X (z) = (2.30)
D(z)
We refer to the roots of N (z) as the zeros of X (z) and to the roots of D(z) as the poles
of X (z). More specifically, in this case X (z) can be expressed as
N (z)
X (z) = (2.31)
K

(z − pk )m k
k=1

where pk is a pole of multiplicity m k , and K is the total number of distinct poles.

Since X (z) is not defined at its poles, its convergence region must not include them.
Therefore, given X (z) as in equation (2.31), there is an easy way of determining its
convergence region, depending on the type of sequence x(n):

• Right-handed, one-sided sequences: The convergence region of X (z) is |z| > r1 .

Since X (z) is not convergent at its poles, then its poles must be inside the circle
|z| = r1 (except for poles at |z| = ∞), and r1 = max {| pk |}. This is illustrated in
1≤k≤K
Figure 2.2a.
• Left-handed, one-sided sequences: The convergence region of X (z) is |z| < r2 .
Therefore, its poles must be outside the circle |z| = r2 (except for poles at |z| = 0),
and r2 = min {| pk |}. This is illustrated in Figure 2.2b.
1≤k≤K
• Two-sided sequences: The convergence region of X (z) is r1 < |z| < r2 , and
therefore some of its poles are inside the circle |z| = r1 and some outside the
circle |z| = r2 . In this case, the convergence region needs to be further specified.
This is illustrated in Figure 2.2c.
For X(z) with N poles of different radius, there are N+1 possible ROC in general,
with N+1 corresponding signals x(n). But only one x(n) signal will be right-sided:
the one with the ROC outside all poles. Also, only one x(n) signal can be stable
(or "absolutely summable"): the one with ROC including the unit circle.
2.3 Inverse z transform

Very often one needs to determine which sequence corresponds to a given z transform.
A formula for the inverse z transform can be obtained from the residue theorem, which
we state next.

T H E ORE M 2.1 (RE SI DUE T H E O R E M)

Let X (z) be a complex function that is analytic inside a closed contour C, including
44 The z and Fourier transforms

Im{z} Im{z}

Re{z} Re{z}

(a) (b)

Im{z}

Re{z}

(c)

Figure 2.2 Regions of convergence of a z transform in relation to its poles: (a) right-handed,
one-sided sequences; (b) left-handed, one-sided sequences; (c) two-sided sequences.

the contour itself, except in a finite number of singular points pn inside C. In this case,
the following equality holds:
 K

X (z)dz = 2π j res {X (z)} (2.32)
C z= pk
k=1

with the integral evaluated counterclockwise around C.

If pk is a pole of multiplicity m k of X (z), that is, if X (z) can be written as
Pk (z)
X (z) = (2.33)
(z − pk )m k
where Pk (z) is analytic at z = pk , then the residue of X (z) with respect to pk is
given by