ECE

EE 8443
3512 – PatternContinuous
– Signals: Recognition
and Discrete

LECTURE 16: DISCRETE-TIME FOURIER TRANSFORM

• Objectives:
Derivation of the DTFT
Transforms of Common Signals
Properties of the DTFT
Lowpass and Highpass Filters

• Resources:
Wiki: Discete-Time Fourier Transform
JOS: DTFT Derivation
MIT 6.003: Lecture 9
CNX: DTFT Properties
SKM: DTFT Properties

URL: Audio:
Discrete-Time Fourier Series
• Assume x[n] is a discrete-time periodic signal. We want to represent it as a
weighted-sum of complex exponentials:
xn   c k e jk ( 2 / T ) n   c k e jk0 n ,  0 2 / T
k N  k N 
Note that the notation <N> refers to performing the summation over an N
samples which constitute exactly one period.
• We can derive an expression for the coefficients by using a property of
orthogonal functions (which applies to the complex exponential above):
 N , k 0, N , 2 N , ...
 e jk ( 2 / N ) n
 
n N   0, otherwise
• Multiplying both sides by e  jr ( 2 / N ) n and summing over N terms:
 xne  jr ( 2 / N ) n    ck e jk ( 2 / N ) n e  jr ( 2 / N ) n    ck e j ( k  r )( 2 / N ) n
n N  n N  k N  n N  k N 

• Interchanging the order of summation on the right side:

 xne  jr ( 2 / N ) n   ck  e j ( k  r )( 2 / N ) n
n N  k N  n N 

• We can show that the second sum on the right equals N if k = r and 0 if k  r:
1
ck   xn e  jk ( 2 / N ) n

N n N 

EE 3512: Lecture 16, Slide 2

Discrete-Time Fourier Series (Cont.)
• This provides a closed-form expression for the Discrete-Time Fourier Series:

xn  c e k
jk ( 2 / T ) n
 c e k
jk0 n
,  0 2 / T (synthesis)
k N  k N 

1 1
ck 
N
 xne  jk ( 2 / N ) n 
n N  N
 xne
n N 
 jk0 n
(analysis)

• Note that the DT Fourier Series coefficients can be interpreted as in the CT

case – they can be plotted as a function of frequency (k0) and demonstrate
the a periodic signal has a line spectrum.
• As expected, this DT Fourier Series (DTFS) obeys the same properties we
have seen for the CTFS and CTFT.
• Next, we will apply the DTFS to nonperiodic signals to produce the Discrete-
Time Fourier Transform (DTFT).

EE 3512: Lecture 16, Slide 3

Periodic Extension
• Assume x[n] is an aperiodic,
finite duration signal.
• Define a periodic extension of x[n]:
~ N N
x n  x[n], n 
2 2
~ ~
x n  N   x [n]
• Note that: ~ x n  x[n] as N  
• We can apply the complex Fourier series:
N /2
~
x n  c e
k  N / 2
k
jk0 n
,  0 2 / T
N /2 N2 
1 ~ 1 ~ 1
cn 
N
 x [n]e  jk0 n 
n  N / 2 N
 x [n]e  jk0 n 
n  N1 N
 x[n]e
n 
 jk0 n


1
• Define: X e   j

N
 x[n]e  j n
. Note this is periodic in  with period 2.
n 

• This implies: c n 
N
1
X e jk0
 
. Note that these are evenly spaced samples of
our definition Xfore
j
.  
EE 3512: Lecture 16, Slide 4
The Discrete-Time Fourier Transform
• Derive an inverse transform:
N 
1 1 N 
2 1 N 
~ 
x n   ( ) X e jk0 e jk0 n 
2
  X e ejk0 jk0 n
( )
2
 X e  e
jk 0 jk0 n
0
k N  N k N  N k N 

 2 
• As N  , ~ x [n]  x[n],  0    0,   0  d
 N 
• This results in our Discrete-Time Fourier Transform:
1
 X e e d

j  j n
x[n]  (synthesis equation)
2 2

X e    x[n]e
j  j n
(analysis equation)
n 

Notes:
• The DTFT and inverse DTFT are not symmetric. One is integration over a
finite interval (2π), and the other is summation over infinite terms.
• The signal, x[n] is aperiodic, and hence, the transform is a continuous
function of frequency. (Recall, periodic signals have a line spectrum.)
• The DTFT is periodic with period 2. Later we will exploit this property to
develop a faster way to compute this transform.
EE 3512: Lecture 16, Slide 5
Example: Unit Pulse and Unit Step
• Unit Pulse:
x[n]  [n]
 
X e j
   x[n]e  j n
   [n]e  jn 1
n  n 

The spectrum is a constant (and periodic over the range [-,].

• Shifted Unit Pulse:
x[n]  [n  n0 ]
 
X e j
   x[n]e  j n
   [n  n0 ]e  jn e  jn0
n  n 

X e j   e  jn0 1 and X e j  e  jn0  n0

Time delay produces a phase shift linearly proportional to . Note that these
functions are also periodic over the range [-,].
• Unit Step: x[n] u[n]
Since this is not a time-limited function, it has
1
no DTFT in the ordinary sense. However, it can  
X e j   j
  ( )
be shown that the inverse of this function is 1 e
a unit step:
EE 3512: Lecture 16, Slide 6
Example: Exponential Decay
• Consider an exponentially decaying signal:
x[n] a n u[n] a  1

  
X e    x[n]e
j  j n
 a e n  j n
 (ae  j ) n
n  n 0 n 0

1

1  ae  j
1 1
X e j   
1  ae  j (1  a cos( ))  ja sin( )
1

(1  a cos( )) 2  (a sin( )) 2

Note :
(1  a cos( )) 2  (a sin( )) 2 1  2a cos( )  a 2 cos 2 ( )  a 2 sin 2 ( ) 1  2a cos( )  a 2
Hence,
1
X e j  
1  2a cos( )  a 2
EE 3512: Lecture 16, Slide 7
The Spectrum of an Exponentially Decaying Signal
1
X e j   Lowpass Filter:
 0
1  2a cos(0)  a 2
1

(1  2a  a 2 )
1

(1  a ) 2
1

1 a
1
X e j  
 
1  2a cos( )  a 2 Highpass Filter:
1

(1  2a  a 2 )
1

(1  a ) 2
1

1 a

EE 3512: Lecture 16, Slide 8

Finite Impulse Response Lowpass Filter

0, n   N1

h[n] 1,  N 1 n  N 1
0, n  N1


N1 N1
sin( ( N 1  1 / 2))
X e j
  e  j n
  ( e  j ) n 
n  N1 n  N1 sin( / 2)

• The frequency response of

a time-limited pulse is a
lowpass filter.
• We refer to this type of
filter as a finite impulse
response (FIR) filter.
• In the CT case, we obtained
a sinc function (sin(x)/x) for the frequency response. This is close to a sinc
function, and is periodic with period 2.
EE 3512: Lecture 16, Slide 9
Example: Ideal Lowpass Filter (Inverse)

sin  c n 
c
1

j n
h[n]  (1) e d 
2  c
n

The Ideal Lowpass

Filter Is Noncausal!

EE 3512: Lecture 16, Slide 10

Properties of the DTFT
• Periodicity:  
  (different from the CTFT)
X e j (   2  )  X e j
• Linearity: ax[n]  by[n]  aX e  bY e  j j

• Time Shifting: xn  n   e

0 X e 
 j  n0 j

• Frequency Shifting: e xn  X e

 j 0 n
 j (  0 )

Example: y[n] e jn xn ( 1) n x[n]

 Note the role

periodicity
plays in the
result.

EE 3512: Lecture 16, Slide 11

Properties of the DTFT (Cont.)
• Time Reversal: x[ n]  X e  j 
• Conjugate Symmetry: x[n] real  X e  j   X * e j 
Also: X e j  and ReX e j  are even functions
X e j  and ImX e j  are odd functions

 
x[n] is real and even, X e j is real and even
x[n] is real and odd, X e  is purely imaginary and odd
j

• Differentiation in d
nx[n]  j [ X e j ]
Frequency: d

1
• Parseval’s Relation:  x[ n ]
2

2 2 X e j
 2
d
n 

• Convolution: y[n]  x[n] * h[n]  Y (e j )  H (e j ) X (e j )

EE 3512: Lecture 16, Slide 12

Properties of the DTFT (Cont.)
• Time-Scaling:  x[n / k ] if n is a multiple of k
x k [ n]   X (e jk )
 0 otherwise

EE 3512: Lecture 16, Slide 13

Example: Convolution For A Sinewave

EE 3512: Lecture 16, Slide 14

Summary
• Introduced the Discrete-Time Fourier Transform (DTFT) that is the analog of
the Continuous-Time Fourier Transform.
• Worked several examples.
• Discussed properties:

EE 3512: Lecture 16, Slide 15