Frequency-Domain Analysis

Fourier Series
Consider a continuous complex signal

x(t) ∈ [−T /2, T /2].

Represent x(t) using an arbitrary orthonormal basis ϕn(t):

∞
X
x(t) = αnϕn(t)
n=−∞

Orthonormality condition:
Z T /2
1
ϕn(t)ϕ∗k (t)dt = δ(n − k).
T −T /2

Multiplying the above expansion with ϕ∗k (t) and integrating

over the interval, we obtain
∞
1 T /2
Z Z T /2 X
1
x(t)ϕ∗k (t)dt = αnϕn(t)ϕ∗k (t)dt
T −T /2 T −T /2 n=−∞
∞
!
Z T /2
X 1
= αn ϕn(t)ϕ∗k (t)dt
n=−∞
T −T /2
∞
X
= αnδ(n − k) = αk .
n=−∞

EE 524, Fall 2004, # 3 1

Thus, the coefficients of expansion are given by
Z T /2
1
αk = x(t)ϕ∗k (t)dt.
T −T /2

Proposition. The functions ϕn(t) = exp(j2πnt/T ) are

orthonormal at the interval [−T /2, T /2].

Proof.
Z T /2 Z T /2
1 1 2π(n−k)
ϕn(t)ϕ∗k (t)dt = j
e T t
dt
T −T /2 T −T /2

sin[π(n − k)]
= = δ(n − k).
π(n − k)

Thus, we can take exponential functions ϕn(t) =

exp(j2πnt/T ) as orthonormal basis =⇒ we obtain Fourier
series.
Fourier series for a periodic signal x(t) = x(t + T ):
∞
j 2πn
X
x(t) = Xn e T t

n=−∞
Z T /2
1 −j 2πn
Xn = x(t)e T t dt.
T −T /2

EE 524, Fall 2004, # 3 2

Fourier coefficients can be viewed as a signal spectrum:

2πn
Xn ∼ X(Ωn), where Ωn = ⇒
T

Fourier series can be applied to analyze signal spectrum! Also,

this interpretation implies that periodic signals have discrete
spectrum.

Example: Periodic sequence of rectangles:

Z T /2 Z τ /2
1 −j 2πn 1 −j 2πn
Xn = x(t)e T t dt = Ae T t dt
T −T /2 T −τ /2

Aτ sin(πn Tτ )
= · real coefficients.
T πn Tτ

EE 524, Fall 2004, # 3 3

Remarks:

• In general, Fourier coefficients are complex-valued,

• For real signals, X−n = Xn∗ .

• Alternative expressions exist for trigonometric Fourier series,

exploiting summation of sine and cosine functions.

EE 524, Fall 2004, # 3 4

 Convergence of Fourier Series
Dirichlet conditions:

Dirichlet conditions are met for most of the signals encountered

in the real world.
Still, convergence has some interesting characteristics:
N
j 2πn
X
xN (t) = Xn e T t .
n=−N

As N → ∞, xN (t) exhibits Gibbs’ phenomenon at points of

discontinuity. Under the Dirichlet conditions:

EE 524, Fall 2004, # 3 5

• The Fourier series = x(t) at points where x(t) is continuous,

• The Fourier series = “midpoint” at points of discontinuity.

Demo: Fourier series for continuous-time square wave (Gibbs’

phenomenon).

http://www.jhu.edu/~signals/fourier2/index.html.

EE 524, Fall 2004, # 3 6

 Review of Continuous-time Fourier Transform

What about Fourier representations of nonperiodic continuous-

time signals?

Assuming a finite-energy signal and T → ∞ in the Fouries

series, we get limT →∞ Xn = 0.

Trick: To preserve the Fourier coefficients from disappearing

as T → ∞, introduce

Z T /2
−j 2πn
T t
X
e n = T Xn = x(t)e dt.
−T /2

EE 524, Fall 2004, # 3 7

Transition to Fourier transform:

X(Ω ) = lim X
en
T →∞
Z T /2
−j 2πn
T t
= lim x(t)e dt
T →∞ −T /2
Z ∞
= x(t)e−j Ω tdt,
−∞

where the “discrete” frequency 2πn/T becomes the continuous

frequency Ω .

Transition to inverse Fourier transform:

∞ ∞
X
j 2πn
X X
en 2πn
x(t) = lim X e n T t = lim ej T t
T →∞
n=−∞
T →∞
n=−∞
T
Z ∞
1 2π 2πn
= X(Ω )ej Ω tdΩ ⇐ dΩ = , Ω= .
2π −∞ T T

EE 524, Fall 2004, # 3 8

Continuous-time Fourier transform (CTFT):

Z ∞
X(Ω ) = x(t)e−j Ω tdt,
−∞
Z ∞
1
x(t) = X(Ω )ej Ω tdΩ .
2π −∞

Example: Finite-energy rectangular signal:

Z ∞
X(Ω ) = x(t)e−j Ω tdt
−∞
Z τ /2
= Ae−j Ω tdt
−τ /2

sin(Ω τ /2)
= Aτ real spectrum.
Ω τ /2

EE 524, Fall 2004, # 3 9

Remarks:

• In general, Fourier spectrum is complex-valued,

• For real signals, X(−Ω ) = X ∗(Ω ).

EE 524, Fall 2004, # 3 10

 Dirac Delta Function
Definition:
 Z ∞
∞, t = 0,
δ(t) = , δ(t)dt = 1.
6 0
0, t = −∞

Do not confuse continuous-time δ(t) with discrete-time δ(n)!

Sifting property:
Z ∞
f (t)δ(t − τ )dt = f (τ ).
−∞

The spectrum of δ(t − t0):

Z ∞
X(Ω ) = x(t)e−j Ω tdt
−∞
Z ∞
= δ(t − t0)e−j Ω tdt
−∞

= e−j Ω t0 .

Delta function in time domain:

Z ∞
1
δ(t) = ej Ω tdΩ .
2π −∞

EE 524, Fall 2004, # 3 11

Delta function in frequency domain:
Z ∞ 
1 ∞, Ω = 0,
δ(Ω ) = e−j Ω tdt = .
2π −∞ 6 0
0, Ω =

For signal
x(t) = Aej Ω0t
we have
Z ∞
X(Ω ) = x(t)e−j Ω tdt
−∞
Z ∞
= A e−j(Ω −Ω0)tdt
−∞
= A 2π δ(Ω − Ω0).

EE 524, Fall 2004, # 3 12

 Harmonic Fourier Pairs

Delta function in frequency domain:

ej Ω0t ↔ 2π δ(Ω − Ω0),

cos(Ω0t) ↔ π [δ(Ω − Ω0) + δ(Ω + Ω0)],
π
sin(Ω0t) ↔ [δ(Ω − Ω0) − δ(Ω + Ω0)].
j

EE 524, Fall 2004, # 3 13

 Parseval’s Theorem for CTFT

Z ∞ Z ∞
2 1
|x(t)| dt = |X(Ω )|2dΩ
−∞ 2π −∞

Z ∞
1
|X(Ω )|2dΩ
2π −∞
Z ∞ Z ∞ Z ∞ 
1
= x(t)x∗(τ )e−j Ω (t−τ )dtdτ dΩ
2π −∞ −∞ −∞
Z ∞Z ∞  Z ∞ 
1
= x(t)x∗(τ ) ej Ω (τ −t)dΩ dtdτ
−∞ −∞ 2π −∞
| {z }
δ(τ −t)
Z ∞
= |x(t)|2dt.
−∞

EE 524, Fall 2004, # 3 14

 Discrete-time Fourier Transform (DTFT)
Represent continuous signal x(t) via discrete sequence x(n):
∞
X
x(t) = x(n)δ(t − nT ).
n=−∞

Substituting this equation into the CTFT formula, we obtain:

Z ∞ ∞
X
X(Ω ) = x(n)δ(t − nT )e−j Ω tdt
−∞ n=−∞
∞
X Z ∞
= x(n) δ(t − nT )e−j Ω tdt
n=−∞ −∞
∞
X
= x(n)e−j Ω nT .
n=−∞

Switch to the discrete-time frequency, i.e. use ω = Ω T :

∞
X
X(ω) = x(n)e−jωn.
n=−∞

X(ω) is periodic with period 2π:

∞
X
X(ω) = x(n)e−jωn e|−j2πn
{z }
n=−∞ 1

EE 524, Fall 2004, # 3 15

 ∞
X
= x(n)e−j(ω+2π)n = X(ω + 2π).
n=−∞

Trick: in computing inverse DTFT, use only one period of

X(ω):

∞
X
X(ω) = x(n)e−jωn DTFT,
n=−∞
Z π
1
x(n) = X(ω)ejωndω Inverse DTFT.
2π −π

Inverse DTFT:
Proof.
Z π
1
x(n) = X(ω)ejωndω
2π −π
Z π ∞
1 X
= x(m)ejω(n−m)dω
2π −π m=−∞
∞ Z π
X 1
= x(m) ejω(n−m)dω = x(n).
m=−∞
2π −π {z
| }
δ(n−m)

EE 524, Fall 2004, # 3 16

 Fourier Series vs. DTFT

∞ Z T /2
X
j 2πn 1 2πn t
x(t) = X e
n T t , Xn = x(t)e−j T dt FS
n=−∞
T −T /2
∞ Z π
X
−jωn 1
X(ω) = x(n)e , x(n) = X(ω)ejωndω DTFT
n=−∞
2π −π

Observation: Replacing, in Fourier Series

x(t) → X(ω),
Xn → x(n),
t → −ω,
T → 2π,

we obtain DTFT!

An important conclusion: DTFT is equivalent to Fourier

series but applied to the “opposite” domain. In Fourier
series, a periodic continuous signal is represented as a sum of
exponentials weighted by discrete Fourier (spectral) coefficients.
In DTFT, a periodic continuous spectrum is represented as a
sum of exponentials, weighted by discrete signal values.

EE 524, Fall 2004, # 3 17

Remarks:

• DTFT can be derived directly from the Fourier series,

• All Fourier series results can be applied to DTFT

• Duality between time and frequency domains.

EE 524, Fall 2004, # 3 18

 Parseval’s Theorem for DTFT

∞ Z π
X
2 1
|x(n)| = |X(ω)|2dω
n=−∞
2π −π

Z π
1
|X(ω)|2dω
2π −π
Z π ∞ ∞
1 X X
= x(n)x∗(m)e−jω(n−m)dω
2π −π n=−∞ m=−∞
∞ ∞ Z π
X X
∗ 1
x(n)x (m) e−jω(n−m)dω
n=−∞ m=−∞
2π −π {z
| }
δ(n−m)
∞
X
= |x(n)|2.
n=−∞

EE 524, Fall 2004, # 3 19

 When Does DTFT Exist (i.e. |X(ω)| < ∞)?

Sufficient condition:
∞
X
|x(n)| < ∞.
n=−∞

∞
X
|X(ω)| = x(n)e−jωn
n=−∞
∞
X
≤ |x(n)| |e−jωn|
| {z }
n=−∞ 1
∞
X
= |x(n)| < ∞.
n=−∞

Example: Finite-energy rectangular signal:

N/2 N/2
X X
−jωn
|X(ω)| = Ae =A e−jωn
n=−N/2 n=−N/2

sin( N2+1 ω)
= A(N + 1)
(N + 1) sin( ω2 )

EE 524, Fall 2004, # 3 20

 sin( N2+1 ω)
≈ A(N + 1) for ω  π
(N + 1) ω2
| {z }
well-known function

Both functions look very similar in their “mainlobe” domain.

EE 524, Fall 2004, # 3 21

 DTFT — Convolution Theorem
If X(ω) = F{x(n)} and H(ω) = F{h(n)} and
∞
X
y(n) = x(k)h(n − k) = {x(n)} ? {h(n)}
k=−∞

then Y (ω) = F{y(n)} = X(ω)H(ω).

 
X∞  X ∞ 
−jωn
Y (ω) = F{y(n)} = x(k)h(n − k )
| {z }  e
m
 n=−∞ k=−∞
 
∞
X ∞
 X 
= x(k)h(m)e−jω(m+k)
m=−∞
 
k=−∞
 (
∞ ∞
)
 X  X
= x(k)e−jωk h(m)e−jωm
  m=−∞
k=−∞

= X(ω)H(ω).

Windowing theorem: If X(ω) = F{x(n)}, W (ω) =

F{w(n)}, and y(n) = x(n)w(n), then
Z π
1
Y (ω) = F{y(n)} = X1(λ)X2(ω − λ)dλ.
2π −π

EE 524, Fall 2004, # 3 22

 Frequency-Domain Characteristics
of LTI Systems

Recall impulse response h(n) of an LTI system:

∞
X
y(n) = h(k)x(n − k).
k=−∞

Consider input sequence x(n) = ejωn, −∞ < n < ∞.

∞
X ∞
X
y(n) = h(k)ejω(n−k) = ejωn h(k)e−jωk = ejωnH(ω).
k=−∞
|k=−∞ {z }
H(ω)

The complex function

∞
X
h(k)e−jωk
k=−∞

is called the frequency response or the transfer function of the

system.

• Impulse response and transfer function represent a DTFT

pair =⇒ H(ω) is a periodic function.

EE 524, Fall 2004, # 3 23

• Transfer function shows how different input frequency
components are changed (e.g. attenuated) at system output.

• Y (ω) = X(ω)H(ω) implies that an LTI system cannot

generate any new frequencies, i.e. it can only amplify
or reduce/remove frequency components of the input.
Conversely, if a system generates new frequencies, then
it is not LTI!

• Systems that are not LTI do not have a meaningful frequency

response.

EE 524, Fall 2004, # 3 24

 Elements of Sampling Theory
Preliminaries
How are CTFT and Fourier series related for periodic signals?
Consider a continuous-time signal xc(t) with CTFT

X(Ω ) = 2πδ(Ω − Ω0).

Then
Z ∞
1
x(t) = 2πδ(Ω − Ω0)ej Ω tdt = ej Ω0t.
2π −∞

We know: periodic signal has line equispaced spectrum. Let

X(Ω ) be a linear combination of impulses equally spaced in
frequency:
∞
X
X(Ω ) = 2πXnδ(Ω − nΩ0). (1)
n=−∞

Using inverse CTFT, i.e. applying it to each term in the sum,

we obtain:
∞ Z ∞
X 1
x(t) = 2πXnδ(Ω − nΩ0)ej Ω tdΩ
n=−∞
2π −∞
∞
X
= XnejnΩ0t Fourier series!
n=−∞

EE 524, Fall 2004, # 3 25

CTFT of a periodic signal with Fourier-series coefficients {Xn}
can be interpreted as a train of impulses occurring at the
harmonically-related frequencies with the weights {2πXn}.

How about the following signal (periodic impulse train), defined

as ∞
X
s(t) = δ(t − nT )?
n=−∞

Note: The above periodic impulse train does not satisfy

the Dirichlet conditions. Hence, its CTFT is introduced and
understood in a limiting sense.

Here are couple of useful expressions for the periodic impulse

train:
∞ ∞
X 1 X jn(2π/T )t
δ(t − kT ) = e ,
T n=−∞
k=−∞
∞ ∞
2π X  2πn  X
δ Ω− = e−j Ω kT .
T n=−∞ T
k=−∞

Also, the Fourier transform of a periodic impulse train is a

periodic impulse train:

∞ ∞
X 2π X  2πn 
δ(t − kT ) ↔ δ Ω− .
T n=−∞ T
k=−∞

EE 524, Fall 2004, # 3 26

Proof.
P∞
The impulse train s(t) = k=−∞ δ(t − kT ) is a periodic signal
with period T =⇒ we can apply Fourier series and find the
Fourier coefficients:

Z T /2
1 −j 2πn 1
Sn = δ(t)e T t dt = .
T −T /2 T

Hence
∞
1 X jn(2π/T )t
s(t) = e .
T n=−∞

Also
Z ∞ ∞
X
S(Ω ) = s(t)e−j Ω tdt = e−j Ω kT .
−∞ k=−∞

EE 524, Fall 2004, # 3 27

 From (1), we obtain

∞ ∞
X 2π X  2πn 
S(Ω ) = Sn δ(ω − n |{z}
2π |{z} Ω0 ) = δ Ω− .
n=−∞ 2π
T n=−∞ T
1/T
T

EE 524, Fall 2004, # 3 28

 The Sampling Theorem

Introduce the “modulated” signal

∞
X
xs(t) = xc(t)s(t) = xc(t) δ(t − nT ).
n=−∞

Since xc(t) δ(t − t0) = xc(t0) δ(t − t0), we obtain

∞
X ∞
X
xs(t) = x(nT ) δ(t − nT ) = x(n) δ(t − nT ).
n=−∞ n=−∞

Using this “modulated” signal, we describe the sampling

operation.

EE 524, Fall 2004, # 3 29

 ∞
1 X jn(2π/T )t
xs(t) = xc(t) s(t) = xc(t) · e
T n=−∞
∞
1 X
= xc(t) ejn(2π/T )t.
T n=−∞

It turns out that the problem is much easier to understand

in the frequency domain. Hence, we compute the Fourier
transform of xs(t). Looking at each term of the summation,

EE 524, Fall 2004, # 3 30

we have from the frequency-shift theorem:
 2π 
xc(t) ejn(2π/T )t ↔ Xc Ω − n .
T

Hence, the Fourier transform of the sum is

P∞  
1
Xs(Ω ) = T n=−∞ Xc Ω − 2πn
T .

Recall that:
∞
X ∞
X
xs(t) = x(nT ) δ(t − nT ) = x(n) δ(t − nT ).
n=−∞ n=−∞

Taking the FT of the above expression, we obtain another

expression for Xs(Ω ):

∞
X Z ∞
Xs(Ω ) = x(n) δ(t − nT )e−j Ω tdt
n=−∞ −∞
∞
X
= x(n) e−j Ω N T = X(ω) .
| {z } ω=Ω T
n=−∞ DTFT{x(n)}

By sampling, we throw out a lot of information: all values of

x(t) between the sampling points are lost.

EE 524, Fall 2004, # 3 31

Question: Under which conditions can we reconstruct the
original continuous-time signal x(t) from the sampled signal
xs(t)?

Theorem. Suppose x(t) is bandlimited, so that X(Ω ) = 0

for |Ω | > ΩM. Then x(t) is uniquely determined by its samples
{x(nT )} = {x(n)} if

2π
Ωs = T > 2ΩM ≡ the Nyquist rate.

EE 524, Fall 2004, # 3 32

Frequency-Domain Effect for Nyquist Sampling
(2π/T = 2ΩM)

EE 524, Fall 2004, # 3 33

 Frequency-domain Effect for Sampling Faster
than Nyquist (2π/T > 2ΩM)

EE 524, Fall 2004, # 3 34

 Frequency-domain Effect for Sampling Slower
than Nyquist (2π/T < 2ΩM)

EE 524, Fall 2004, # 3 35

 Elements of Sampling Theory (cont.)

Introduce a lowpass filtering operation. The spectrum of the

filtered signal:
Xf (Ω ) = HLP(Ω )Xs(Ω )
where HLP(Ω ) ≡ ideal lowpass filter:

T, −Ωc ≤ Ω ≤ Ωc,
HLP(Ω ) = ,
0, otherwise

with the cut-off frequency Ωc. How to reconstruct a

bandlimited signal from its samples in the time domain?

Having a signal sampled at a rate higher than the Nyquist rate

and infinite number of its discrete values, the signal can be
exactly recovered as

∞
X sin[π(t − nT )/T ]
xf (t) = x(n) ideal interpolation formula.
n=−∞
π(t − nT )/T

Proof. Start from: Xf (Ω ) = Xs(Ω )HLP(Ω ). In time domain

xf (t) = {xs(t)} ? {hLP(t)}

( ∞ )
X
= xc(nT )δ(t − nT ) ? {hLP(t)}
n=−∞

EE 524, Fall 2004, # 3 36

 Z ∞ ∞
X
= xc(nT )δ(τ − nT )hLP(t − τ )dτ
−∞ n=−∞
∞
X
= x(n)hLP(t − nT ).
n=−∞

Ideal transfer function:

T, − Tπ ≤ Ω ≤ Tπ ,

HLP(Ω ) = ,
0, otherwise

Ideal impulse response:

sin(πt/T )
hLP(t) = .
πt/T

Now, insert hLP(t) into the equation for xf (t). 2

EE 524, Fall 2004, # 3 37

 Representations of Narrowband Signals

Narrowband signals have small bandwidth compared to the

band center (carrier) frequency.

B  F0 .

xc(t) = a(t) cos[2πF

| {z }0 t + |{z}
θ(t) ].
|{z}
Ω0
amplitude phase
modulation modulation

EE 524, Fall 2004, # 3 38

The above representation can be used to describe any signal,
but it makes sense only if a(t) and θ(t) vary slowly compared
with cos(2πF0t), or, equivalently, B  F0.

• Complex-envelope and

• Quadrature-component

representations of narrowband signals.

Complex-envelope representation:

xc(t) = Re{a(t) exp(j[Ω0t + θ(t)])}

= Re{a(t) exp[jθ(t)] exp(jΩ0t)}.
| {z }
x
ec (t)

The complex-valued signal xec(t) contains both the amplitude

and phase variations of xc(t), and is hence referred to as the
complex envelope of the narrowband signal xc(t).

EE 524, Fall 2004, # 3 39

 Quadrature-component representation:

xc(t) = a(t) cos θ(t) cos(Ω0t) − a(t) sin θ(t) sin(Ω0t).

| {z } | {z }
xcI (t) xcQ (t)

xcI(t) and xcQ(t) are termed the in-phase and quadrature

components of narrowband signal xc(t), respectively.

Note that
x
ec(t) = xcI(t) + jxcQ(t).
If we “blindly” apply the Nyquist theorem, we would choose

B
FN = 2(F0 + ) ≈ 2F0 for B  F0.
2

However, since the effective bandwidth of xc(t) [and x

ec(t)] is
B/2, the optimal rate should be B!

Recall

xc(t) = a(t) cos[2πF0t + θ(t)]

exp{j[Ω0t + θ(t)]} + exp{−j[Ω0t + θ(t)]}
= a(t) ·
2
a(t) exp[jθ(t)] a(t) exp[−jθ(t)]
= exp(jΩ0t) + exp(−jΩ0t)
| 2
{z } | 2
{z }
1 1 ∗
2xec (t) 2x
ec (t)

EE 524, Fall 2004, # 3 40

and hence

Xc(Ω ) = 21 [X ec∗(−Ω − Ω0)],

ec(Ω − Ω0) + X

implying that x ec is a baseband complex-valued signal

(occupying the band [−B/2, B/2]):

∞
X  n  sin[πB(t − n/B)]
x
ec(t) = x
ec
n=−∞
B πB(t − n/B)

EE 524, Fall 2004, # 3 41

Now,

xc(t) = Re{e
xc(t) exp(j2πF0t)}
n X∞  n  sin[πB(t − n/B)] o
= Re x
ec exp(j2πF0t)
n=−∞
B πB(t − n/B)
∞
n X n sin[πB(t − n/B)] o
= Re a exp{j[θ(n/B) + 2πF0t]}
n=−∞
B πB(t − n/B)
∞
X n sin[πB(t − n/B)]
= a cos[2πF0t + θ(n/B)]
n=−∞
B πB(t − n/B)
∞ h
X n n i
= xcI cos(2πF0t) − xcQ sin(2πF0t)
n=−∞
B B
sin[πB(t − n/B)]
· .
πB(t − n/B)

EE 524, Fall 2004, # 3 42

 Z Transform

∞
X
X(z) = Z{x(n)} = x(n)z −n.
n=−∞

Relationship between the z transform and DTFT: substitute

z = rejω ,

∞
X
X(z)|z=rejω = x(n)(rejω )−n
n=−∞
∞
X
= {x(n)r−n}e−jωn
n=−∞

= F{x(n)r−n} =⇒

The z transform of an arbitrary sequence x(n) is equivalent to

DTFT of the exponentially weighted sequence x(n)r−n.

If r = 1 then

X(z)|z=ejω = X(ω) = F{x(n)} =⇒

DTFT corresponds to z transform with |z| = 1.

Notation: Observe that X(ejω ) ≡ X(ω).

EE 524, Fall 2004, # 3 43

The z transform reduces to the DTFT for values of z on the
unit circle:

Question: When does the z transform converge?

Region of convergence (ROC) ≡ range of values of z for which

|X(z)| < ∞.

Example: The z transform of the signal x(n) = anu(n) is

∞
X ∞
X
X(z) = anu(n)z −n = (az −1)n.
n=−∞ n=0

EE 524, Fall 2004, # 3 44

For convergence, we require that
∞
X
|az −1|n < ∞,
n=0

which holds if |az −1| < 1 or, equivalently, |z| > |a|. Note:

1
X(z) = .
1 − az −1

Example: The z transform of the signal


0, n ≥ 0,
x(n) = −anu(−n − 1) =
−an, n ≤ −1

is
−1
X ∞
X ∞
X
X(z) = − anz −n = − a−nz n = − (a−1z)n
n=−∞ n=1 n=1

a−1z 1
= − −1
= −1
≡ same as previous ex.
1 − a z 1 − az

But, ROC is now |z| < |a|.

Remark: a discrete-time signal x(n) is uniquely determined by

its z transform X(z) and its ROC.

EE 524, Fall 2004, # 3 45

 ROC Properties

• The ROC of X(z) consists of a ring in the z plane centered

about the origin,

• The ROC does not contain any poles,

• If x(n) is of finite duration, then the ROC is the entire z

plane, except possibly z = 0 and/or z = ∞,

• If x(n) is a right-sided sequence, and if the circle |z| = r0

is in the ROC, then all finite values of z for which |z| > r0
will also be in the ROC (need to check z = ∞),

• If x(n) is a left-sided sequence, and if the circle |z| = r0 is

in the ROC, then all finite values of z for which 0 < |z| < r0
will also be in the ROC (need to check z = 0),

• If x(n) is a two-sided sequence, and if the circle |z| = r0 is

in the ROC, then the ROC will be a ring in the z plane that
includes the circle |z| = r0 (we can represent this sequence
as right-sided sequence + left-sided sequence).

EE 524, Fall 2004, # 3 46

 Z Transform (cont.)
Inverse Z transform:
Recall that
X(z) = F{x[n]r−n}.
z=rejω
Applying the inverse DTFT, we get

x[n] = rnF −1{X(rejω )}

Z π
1
= rn · X(rejω )ejωn dω
2π −π
Z π
1 jω jω n
= X(re )(re ) dω
2π −π |{z}
z
|{z}
z
I
1
= X(z)z n−1 dz ←− dz = jrejω dω.
2πj

Comments:
H
• · · · dz denotes integration around a closed circular contour
centered at the origin and having radius r,

• r must be chosen so that the contour of integration |z| = r

belongs to the ROC,

• contour integration in complex plane may be a complicated

task; simpler alternative procedures exist for obtaining a
sequence from a Z transform.

EE 524, Fall 2004, # 3 47

LTI system analysis:

y(n) = {h(n)} ? {x(n)} ↔ Y (z) = H(z)X(z)

Results:

• A discrete-time LTI system is causal if and only if the ROC

of its transfer function is the exterior of a circle including
infinity.

• A discrete-time LTI system is stable if and only if the ROC

of its transfer function includes the unit circle |z| = 1.

EE 524, Fall 2004, # 3 48

 Rational Z transforms

Recall LCCD equations of ARMA processes

N
X M
X
ak y(n − k) = bk x(n − k).
k=0 k=0

Taking z-transforms of both sides, we get

N
X M
X
Z{ak y(n − k)} = bk Z{x(n − k)},
k=0 k=0

yielding
N
X M
X
Y (z) ak z −k = X(z) bk z −k .
k=0 k=0
Hence, the transfer function of an ARMA process is
PM −k
Y (z) k=0 b k z
H(z) = = PN .
X(z) k=0 ak z
−k

EE 524, Fall 2004, # 3 49