Digital Signal Processing

Discrete-Time Fourier Transform

(Adapted from Oppenheim, Alan V., lecture notes)
x( ] y (t)

x [n] y [n] TRANSPARENCY

7.1
The principle of
superposition for
linear systems.

If: x(t) = a, 0, (t) + a2 0 2 (t) +

ek(t) k(t)

and system is linear

Then: y(t) = a, 0 1 (t) + a2 0 2 (t) +

Identical for discrete-time

If: x= a, 1 +a 2 02 +.
TRANSPARENCY
Then: y = a, 01 + a2+'... 7.2
Criteria for choosing
set of basic signals in
terms of which to
decompose the input
Choose $k (t) or $k [n] so that: to a linear system.

- a broad class of signals can be

constructed as a linear combination

Of $k's

- response to #k's easy to compute

 4 k(t) = h(t - kA)
LTI System
=> Convolution Integral

eD-T: #k [n] = 6 [n - k]

k[n] =h[n-k]
=> Convolution Sum
 Fourier Analysis:
Complex
*C-T:Exponentials
sk ~ jwk
as Base
k (t) = Signals
eikt
$k (t) = eskt sk complex
*D-T: IzkI = 1 = eskt = e 9kn
kksk[n]complex
$k (t)
RENCY k [n] = Zk n zk complex
s
k [n] = Zk=>
skcomplex zk complex
n Laplace transforms
xponentials
basic
zk complex => z-transforms
Fourier Analysis:
Fourier Analysis:
Fourier
*C-T: Analysis
*C-T:sk sk
~ jwk kk(t)(t) = eikt
= eikt
~ jwk
*D-T:
*D-T:IzkI = 1 kkkk[n] = e
e 9kn
9kn
IzkI = 1 [n] =

skcomplex =>
skcomplex Laplacetransforms
=>Laplace transforms
Discrete-Time Freq. Response
Discrete-Time
Fourier Series
 Ex. 1
• Given x[n] = sinwon, calculate ak (3.99)

-time counterpart of the signal x(t) = sin w 0 t of Example 3.3. x[n]

T/w 0 is an integer or a ratio of integers. For the case when 27T/w 0 is
when

27T
wo = N'

fundamental period N, and we obtain a result that is exactly analo-

 an integer N, that is, when 27T
wo = N'
27T
wo = N'
is periodic with fundamental period N, and we obtain a result that is exactly analo-
is periodic with fundamental
x[n]continuous-time period N, andthe
we signal
obtain a as
result that isof
exactly analo-
to the • Given x[n] =case. Expanding
sinwon, calculate a a sum two
gous to the continuous-time case. Expanding the signal ask a sum of two complex expo-
complex expo-
(3.99)
als, nentials,
we get we get
-time counterpart of the signal x(t) = sin w 0 t of Example 3.3. x[n]
_!_ej(21T!N)n
= = _!_
a x[n]
T/w 0 is an integer orx[n] ej(21T!N)n _ __!__!_
.case when 27T/w 0 is (3.100)
e- j(21T!N)n
ratio of
2j integers. 2jFor e-the
j(21T!N)n (3.100)
2j 2j .
when
Comparing eq. (3.100) with eq. (3.94), we see by inspection that
paring eq. (3.100) with eq. (3.94), we see by inspection that
27T 1
wo = N'
at = 2 j'
1
(3.101)

at = 2 j' (3.101)
fundamental period N, and we obtain a result that is exactly analo-
 Example 3. 11
Ex. 2 Consider the signal

Find the Fourier Series coeffs of: x[n] = 1 + sm. (27T)

N n + 3 cos (27T)
N n + cos (47T
N n + 27T).

This signal is periodic with period N, and, as in Example 3.10, we can ex

directly in terms of complex exponentials to obtain

x[n] = 1 + 1j [ej(27TIN)n _ e- j(27TIN)n] + [ej(27TIN)n + e- j(27TIN)n]

2

+ [ej(47Tn/N+7TI2) + e- j(4

Collecting terms, we find that

x[n) I + + 21j )ej(2w/N)• + - )e- j(2w/N)•

 x[n] = 1 + sm N n + 3 cos N n + cos N n +2 .
directly intoterms
complex exponentials obtain of complex exponentials to obtain

Solution
=+1 +[ej(27TIN)n
This signal
TIN)n _ e- j(27TIN)n]
x[n] is 1j
directly in terms
+ e-with
periodic _ period
j(27TIN)n]
[ej(27TIN)n N, and,
e- j(27TIN)n]
2 of complex exponentials to obtain
+ as[ej(27TIN)n
in Example+ e-3.10, we can expand x[n]
j(27TIN)n]

+ [ej(47Tn/N+7TI2) + e- j(47Tn/N+7T!2)].
x[n] = 1 + 1j [ej(27TIN)n _ e- j(27TIN)n] + [ej(27TIN)n + e-+j(27TIN)n]
[ej(47Tn/N+7TI2) + e- j(47Tn/N+7T!2)].
2
e find that
Collecting terms, we find that + [ej(47Tn/N+7TI2) + e- j(47Tn/N+7T!2)].
1j )ej(2w/N)• + - )e-
j(2w/N)•

x[n) I+
Collecting terms,+we21jfind
+ (iejwa that
)ej(2w/N)•
)ej2(2w/N)• +
+ )e-
(ie-jwa - )e-
j2(2w/Nl•,
j(2w/N)•

x[n) I+ + 21j )ej(2w/N)•

ries coefficients for this example are
+ - )e-+ (iejwa
j(2w/N)•
)ej2(2w/N)• + (ie-jwa )e- j2(2w/Nl•,

a0 =1, + )ej2(2w/N)• (iejwa + (ie-jwa )e- j2(2w/Nl•,

Thus the
3 Fourier
1 3 series
1 . coefficients for this example are
a = 2 + 2j = 2- 21'
1

Thus the
3 Fourier
1 3series
1 .coefficients
a for
= this
1, example are
0
a-l = 2 - 2j = 2 + 21'
a0 = 1, 3 1 3 1.
a2 =
1 .
21· a1 =
3
2 +1 2j 3= 2-
1.
21'
a 1 = 2 +3 2j =1 2- 321' 1 .
1.
a-2 = -2], a-l =3 2 -1 2j 3= 21+. 21'
a-l = 2 - 2j = 2 + 21'
Discrete-Time Fourier Transform
DISCRETE-TIME FOURIER TRANSFORM

TRANSPARENCY
11.1
x [n] - 1fX(2) ejn dU synthesis The discrete-time
27r 2 Fourier transform.
[As corrected here,
x[n], not x(t), has
+00 Fourier transform
X(fl).J
X(92) x[n] e- jn
analysis
n=-w0

x[n] <TO X(i)

= Re X(n) t + jImiX(S)t

= IX(2)|ei X(2)
To illustrate the discrete-time Fourier transform, let us consider several ex

Example
Example 5.1
Consider the signal
Representation of Aperiodic Signals: The Discrete-Time Fourier Transform
x[n] = a"u[n], lal < 1.
this case,
+co
X(ejw) = L anu[n]e- jwn
n=-oo
GO

= ""'(ae-jwt
L
= ----:--
1- ae-jw'
n=O
he magnitude and phase of X(ejw) are shown in Figure 5.4(a) for a> 0 and in
 n=O
The magnitude and phase of X(ejw) are shown in Figure 5.4(a) for a> 0 and in Fig-
ure 5.4(b) for a< 0. Note that all of these functions are periodic in w with period 27T.
If a > 0

(a)
If a < 0

<l:X(Eh
tan- 1 (lal//1 - a2 )-.........

-tan- 1 (lal/!1 - a2 )
 In this case, X(eiw) is real and is illustrated in Figure 5.5(b), again for 0 <
In this case, X(eiw) is real and is illustrated in Figure 5.5(b), again for 0 <a < 1.

Ex. 2
Example
Example 5.3 5.3
Consider the rectangular pulse
Consider the rectangular pulse
x[n] = { (5.10)

x[n] = {
which is illustrated in Figure 5.6(a) for N 1 2. In this case,
The= Discrete-Time Fourier Transform Chap.5
which is illustratedX(eiw)
in Figure
2_5.6(a)
Nl
e- Jwn.
for N 1 = 2. In this case,
= (5.11)
Using calculations similar to those employed in obtaining eq. (3.104) in Example 3.12,
n=-Nl
we can write Nl
X(eiw) = 2_ e- Jwn.
. sinw
x[n] (N 1 + n=-Nl
(5.12)
X(eJw) = sin(w/2) ·

..................II]II..................
This Fourier transform is sketched in Figure 5.6(b) for N 1 = 2. The function in eq. (5.12)
x[n] in the Fourier trans-
is the discrete-time counterpart of the sine function, which appears
 Nl
X(eiw) = 2_ e- Jwn. (5.11)
n=-Nl

Ex. 2 x[n]

..................II]II..................
-N 1 0 N1 n
(a)

(b)

Figure 5.6 (a) Rectangular pulse signal of Example 5.3 for N1 = 2 and
(b) its Fourier transform.
any behavior like the Gibbs phenomenon in evaluating the discrete-time F
synthesis equation. This is illustrated in the following example.
Ex. 3
Example 5.4
Let x[ n] be the unit impulse; that is,

x[n] = 8[n].

In this case the analysis equation (5.9) is easily evaluated, yielding

What is the Fourier Transform output of x[n]?

In other words, just as in continuous time, the unit impulse has a Fourie
sentation consisting of equal contributions at all frequencies. If we the
to this example, we obtain
Properties of Fourier Transform
PROPERTIES OF THE FOURIER TRANSFORM

x [n] X(g)
TRANSPARENCY
11.2
Periodic: Periodicity and
symmetry properties
of the discrete-time
X(9i) = X( +27r m) Fourier transform.

Symmetry:

x[n] real => x(-i2) X*(92)

RejX(w)t
even
IX(SiZ)I

Imlx()
odd
4 X(W)
 Time shifting:

x[n-n]I X(E)
TRANSPARENCY
11.4 Frequency shifting:
Additional properties
of the discrete-time
Fourier transform.
ejgon x[n] X(92 - 20 )

Linearity:

ax1 [n] +bx2 [n] aX, (92) + bX2(92)

Parseval's relation:

z
+ 00

Ix[n] 12 = 1 fIX(2)2 dE
27r
n=--00
 That is,5.3.4 Conjugation and Conjugate hip[n].
L =-( -lY
jnx[n]e- jwn.
n= - x Symmetry
If
Time Reversal:
The right-hand side of this equation is the Fourier transform of- jnx[n]
5.3.4 both
tiplying Conjugation andsee
sides by j, we Conjugate
that Symmetry
If
5. 3. 7 Time Expansion
then g: dX(ejw)
Differentiation
Because ofin theFrequency:
discrete nature of the
nx[n] time indexj dw discrete-time
for . sign
tweenthentime and frequency scaling in discrete time takes on a somewhat
its continuous-time counterpart. Specifically, x * [n] in Section
X * (e-4.3.5
1.w). we deriv
Theproperty
time usefulness of this property will be illustrated in Example 5.13 in Se
x * [n] X * (e- 1.w).
Conjugation Also,
and if x[n] is real valued, its transform
5.3.9 Parseval's Relation x(at)
g: (jw)
1X(ejw) is conjugate sym
rarx ---;;- .
Conjugate symmetry:
Also, if x[n] is real valued, its transform X(ejw) is conjugate symmetric. T
If x[n] and X(ejw) are a Fourier transform pair, then
However, if we try to define the signal x[an], we run into difficulties i
Therefore, we cannot slow down the signal by choosing a < 1. On th
 IX(eiw)l
Ex. x[n] periodic, real,
even, and/or finite energy? w
_1I. 1I.
2 (a) 2
• Periodicity in time domain corresponds to discrete Fourier output.
• NOT PERIODIC

• Even magnitude, odd phase, i.e. conjugate symmetric

• REAL

• If x[n] is real and even, its Fourier transform is also real and even
• NOT EVEN

• From Parseval’s relation

• FINITE ENERGY
Convolution Property
CONVOLUTION PROPERTY

h[n] * x[n] H(92) X()

TRANSPARENCY
11.5
The convolution
property.
x[n] h [n] h[n] * x[n]

X(92) IH(2) H(M) X(M)

eifon H(UZ1)
 Frequency Response of Ideal Lowpass & Highpass Filter

H1 (n)
ANSPARENCY
quency response of

K7
deal lowpass filter mm, I J
an ideal highpass I
I
r. I
-2ir -R -c 27r

H2 (2)

-27r -9 i 27r
 ave depicted the frequency response H1p(eiw) of a lowpass filter
Discre
We. while in Figure 5.12(b) we have displayed H 1p(ei<w-7T))-that
onse H 1p(eiw) shifted by one-half period, i.e., by 7T. Since high
-Time Fourier
time are Transform
concentrated near 7T (and other odd multiples oflow
7T), the 375
pass
is an ideal highpass filter with cutoff frequency 7T - w c. That is,
x [n] - h [n]
.32) and the frequency-shifting property imply that (5.32)
srom
filters in
eq. (3.122), and as we will discuss again in Section 5.4, the
an LTI system is the Fourier transform of the impulse response
hhp[n]
f h1p[n] and = respectively
hhp[n] high pass (5.33)
eJ7Tn hip[n] denote the impulse responses of
= ( -lY hip[n]. x [n] M (- 1)" h [n] (5.34)

H,p(<,iw)
onjugate Symmetry
 .Example
11 5. 11
onsider an LTI system with impulse response
Consider
n LTI systeman LTI system
with with impulse
impulse response
response
Ex. 4H(eiw) = +"'
L h[n] = 8[n- no].
8[n- no]e-
h[n] = 8[n- no]. jwn = e- jwno.
he frequency response is h[n] = 8[n- no].
n=-x
The frequency response is

ncy response
y input x[n] iswith FourierL
+"'+"'
transform X(eiw), the Fourier transfo
H(eiw) = L 8[n- no]e- jwn = e- jwno.
H(eiw) = 8[n- no]e- jwn = e- jwno.
n=-x
n=-x

hus,Thus, for input

for any
utputoutput
is is
any input
H(eiw)
x[n]x[n] L
+"'
=withwith Fourier
Fourier
8[n- transform
transform jwn =the
no]e-X(eiw),
X(eiw), e-Fourier
the jwno. transform
Fourier of of
transform thethe
n=-x
(5.49)
(5.49)
,nyforinput
this x[n]
example, y[n] = transform
with Fourier ] and eq.
x[n - n 0X(eiw), the(5.49)
Fourier is transform
consisten
We note that, for this example, y[n] = x[n - n 0 ] and eq. (5.49) is consistent with the
Weproperty.
note Note
that, for
time-shifting alsoNote
this example,
property. that the
y[n]
also frequency
that= the
x[n - n 0 ] and
frequency response
H(eiw)H(eiw)
eq. (5.49)
response is =consistent e-the
with
e- Jwno of=a pure Jwn
me-shifting property. Note also that the frequency response H(eiw) = e--wn Jwno of a pure
smeunity
time magnitude
shift has unity at all
magnitude frequencies
at all frequencies and
and a a phase
phase characteristic
characteristic
shift has unity magnitude at all frequencies and a phase characteristic -wn that is
0 that is -w
linear with frequency. 0
h l/31 < 1. Evaluating the Fourier transforms of h[n] an
Consider an LTI system with
ystem with impulse response
at theEx. 5 to this system is
input with Ia I < 1, and suppose tha
is=I < 1, and1suppose
with Ia
suppose that the input to this H(ejw)
system . th
x[n] = f3nu[n], 1-ae-Jw
x[n] to= this
suppose that the input with l/is
f3nu[n],
system 31 < 1. Evaluating the F
Fourier transforms of h[n] and x[n], we< have
with l/31 1. Evaluating the
luating the Fourierx[n]
transforms of h[n] and x[n], we hav
= f3nu[n],
H(ejw) = 1 . (5
luating the Fourier transforms
1-ae-Jw
H(ejw) = of
1 .h[n]
and and x[n], we ha
1-ae-Jw
and
hat H(ejw) = 1 .
1-ae-Jw
 on (together, perhaps, with the use A of the frequency B differentiation property
r 5.3.8).
terms ofThethis Y(ejw)
type
general so = we procedure
that
algebraic can find the+ inverse
for .. ofiseach
rationaltransform
transforms term b
described
ion (together, perhaps, with 1-use
the ae-jw 1- f3e-1w
endix. For this example, if a ¥- f3, the partial fractiondifferentiation
of the frequency expansion of property
Y ( ejw) is o
Ex. 5
m 5.3.8). The general algebraic procedure for rational transforms is described
ting the right-hand sides of eqs (5.53) and (5.54), we find that
endix. For this example, if a ¥- f3, the partial fraction expansion of Y ( ejw) is o
m Y(ejw) =
A + B .. (5.5
A= _a_ 1- ae-jwB = 1- _f3e-1w
_/3_.
As with ExampleY(ejw) 4.19, determining
a-
=
A
f3' the
+ inverse B
a-{3 transform
.. of Y(ejw) is mo
quating the right-hand sides ofmethod
eqs (5.53) and (5.54
by expanding Y(ejw) by the1- ae-jw 1- (5.54),
of partialf3e-1w we findSpecifically,
fractions. that Y(e
from Examplein5.1powers
f polynomials and the of linearity
e- jw, andproperty,
we wouldwe likecan
to obtain
expressthe thisinve
as a
Equating the right-hand
er(5.54)
termsby A=
of inspection:
sides
_a_
this type so that
of eqs (5.53)
B = and_ _/3_.
(5.54), we find that
a- we f3' can find a-{3
the inverse transform of each
tion (together, perhaps, A= with_a_ the use
B =of_ the_ frequency differentiation pro
/3_.
nre,5.3.8).
from Example
The y[n] 5.1 and
general the linearity
algebraic
a- property,
f3' procedure
= _a_anu[n]- _f3_f3 u[n] for
a-{3 we can11 obtain
rational the inverse
transforms is tran
desc
eq. (5.54)
pendix. Forbythis
inspection:
example,a-{3 if a ¥- f3, the partial
a-{3fraction expansion of Y ( e
ore, from Example 5.1 and the linearity property, we can obtain the inverse tran
m
f eq. (5.54) by inspection: 1
or a = f3, the partial-fraction expansion in eq. (5.54) is not valid. However, in this
which can be expressed as
whichFor
ase, = f3, as
can bea expressed the partial-fraction expansion in eq.Chap.5
(5.54
The Discrete-Time Fourier Transform
case, Y(eiw) = j jw d (
Y ( ejw) = ( 1 1 - . )2 ' 1 ) (5.56
Y(eiw) = j jw- aedw
d (JW1 - ae-
1 Jw ) · (5.5
which can be expressed as dw 1 - ae- Jw ·
As in Example 4.19, we can use the frequency differentiation property, eq. (5.46
As with
together in Example 4.19,transform
the Fourier we can= use
Y(eiw) j the
pair Y
jw frequency
d (
dw 1 - ae- Jw ·
) = (
1 (differentiation
ejw) 1 1
property,
- ae JW
-
eq. .(5.46
(5.56) )2
together with the Fourier transform pair
As in Example 4.19, we cananu[n] ---,----
use the frequency differentiation property, eq. (5.46),
1-ae-Jw'
anu[n]
together with the Fourier transform pair ---,----
1-ae-Jw'
to conclude that
to conclude that anu[n] ---,----
.d( 1)
1-ae-Jw'

to conclude that .d( 1)

w 1 - ae
w 1 - ae
_
_
.
.
.
.
JW

JW
 .d( 1)
_ . .
.d( 1)
.d(
_..
w 1- ae JW
1)
Ex 5. _JW
..
w 1 - ae JW
w 1- ae
o account for the factor eiw, we use the time-shifting property to obtain
ccount for the factor
To account the we
foreiw, useeiw,
factor the time-shifting property to property
we use the time-shifting obtain to obtain

(n +
(n + 1)an+ (n
1 1)an+
u[n +
1 u[n + 1]
1] jelw_ .d( .d(
+ 1)an+ 1u[n + 1] dw 1-
1)
jelw_
.d(
jelw_
.
ae-Jw
,
dw 1- ae-Jw
. ,
. , 1)
1)
dw 1- ae-Jw

nd finally,
finally, and accounting
accounting
finally, for the for the
factor
accounting factor
for1/a,
the in
factor 1/a,in
eq.1/a, in eq.
(5.56), we (5.56),wewe
eq. obtain
(5.56), obtain
obtain
y[n] = (n + l)anu[n
y[n]
y[n] = =(n(n +++ l)anu[n
1].
l)anu[n ++1].1]. (5.57) (5.
worth noting that, noting
It is worth although
that,thealthough
right-hand side is multiplied
the right-hand side is by a step that
multiplied by abegins
step that beg
=is -1,
worththe noting
sequence that,
(n + although
1)anu[n + the
1] isright-hand
still zero side
prior ton is multiplied
0, since
at n = -1, the sequence (n + 1)anu[n + 1] is still zero prior ton = 0, since the fac
= the by a
factor step th
1nis =zero-1,+ n1the
n at sequence
is= zero
-1. Thus,
at (ncan
n = we
-1. +Thus, we can+alternatively
1)anu[n
alternatively 1] is still
express zero
y[n] as prior
express y[n] ton
as = 0, since
+ 1 is zero at n = -1. y[n] Thus, we can alternatively express y[n] as
= (n + 1)anu[n].
y[n] = (n + 1)anu[n]. (5.58) (5.
Modulation Property
MODULATION PROPERTY
TRA
Discrete-time: 11.9
A co
Fou
mod
x, [n] x2 [n] 27r 2ir
X1 () X2 (E2- 0) dO for c
and

periodic convolution

Continuous-time:
+00