Prepared by Professor Eliathamby Ambikairajah, Head of

School
Electrical Engineering and Telecommunications
University of New South Wales (UNSW), Australia

Modified and Delivered by Dr. T. Thiruvaran

Discrete Time Fourier Transform (DTFT) ..........................................

Properties of the Fourier Transform of a Discrete Signal (FTD
or DTFT) ..........................................................................................
The Discrete Fourier Transform (DFT) ............................................
Conjugate Symmetry .........................................................
Parseval’s Theorem ............................................................
Padding with Zeros and Frequency Resolution .......................
Inverse Discrete Fourier Transform (IDFT).............................
Computations for Evaluating the DFT .....................................
Summary of DFT/IDFT Calculations ......................................
Comparison of FFT/DFT Complex Multiplications ................
Summary ..................................................................................
Discrete Time Fourier Transform
(DTFT)

Fourier Transform of an arbitrary discrete signal (FTD) is also

called Discrete Time Fourier Transform (DTFT).

For discrete-time signals, the Fourier transform pair is defined by


X ( )   xn e  jn

n  

xn 
1

jn
X ( ) e d
2 
FTD

xn  X ( )
Sequence continuous
variable

Where  is the digital frequency or relative frequency (-   

) and  = T  X() is periodic with period 2. (ie. X(+2) = X
())

Note: The Fourier transform of an analogue signal is not periodic.

Example:
δ[n]
X()
x[n] = [n] n continuous
-2 -1 0 1 2
 fs
X ( )    ne
n  
 jn
1 2

-  
Example:

0 n0
x[n]   n
a n  0, | a | 1
 
X ( )   a e n  jn
  (ae j ) n
n 0 n 0

1

1  ae j
1
X ( ) 
(1  a cos )  ja (sin  )
1
| X ( ) |
1  2a cos  a 2

 a sin 
X ( )   ( )   tan1  
 1  a cos 

|X()|

continuous
variable
x(n)=anu(n) FTD

n - 
-2 -1 0 1 2 3 4 

Note: X() can be obtained using z-transform as well.

xn  a nun  X ( z ) 
1
1  az 1
1
X ( )  X ( z ) z e j 
1  ae j
(Note: X()  X(ej))
Exercise:

x(n)
1

1 nN
xn  
0 n N

0
-5 -4 -3 -2 -1 0 1 2 3 4 5
-N N

 (2 N  1) 
sin 
Determine X(), ans : X ( )   2 
 
sin 
 2
6.1 Properties of the Fourier Transform of a Discrete
Signal (FTD or DTFT)

xn  X ( )
FTD

x1 n  X 1 ( )
FTD

x2 n  X 2 ( )
FTD

Linearity ax1n  bx2 n  aX1 ( )  bX 2 ( )

FTD

xn  k   e jk X ( )
FTD
Time shift
 
 
Z
Proof:  x n  k  z k X (z )
 j
z e
e-jkX(), {we define X(ej) X()}

Frequency shift

xne X (   )  X (e 
FTD
jn 0 j (  0 )
 X (  0 ) 0 )

Example:
The frequency response of an ideal low pass filter, in the
fundamental interval -     is given by

1
 0 |  |  C
H ( )   2
 0  C |  | 

(a) Find the impulse response of the ideal low pass filter
 
(b) Sketch the impulse response for  C 
4
(a) 1
|H()|
2

- -c c  

 C
1 e jC n  e  jC n
hn 
1 1 1 jn
 H ( )e 
jn
d  e d 
2 
2  C
2 4 jn
1 sin  c n

2 n

(b)
n n
sin sin
 1 4 1 4
c  h( n) 
4 2 n 8 n
4
h[n]
1/8

0123 n

Exercise: The frequency response of a band stop filter is given

by
 
 1 |  |
6
  
H d ( )  0 |  |
 6 3
1 
|  | 
 3
Show that the impulse response hd[n] of the band stop filter is
given by
    
sin n  sin n 
hd n   n  
3 
 
6 
n n
Hd()


3  /6 

1  jn 
hd n 
1
2  2    
jn jn jn
H d ( ) e d    e d   e d  e d 
 
 6 3 
 jn 3 jn 6

jn  
1  e   e   e  
    
2  jn    jn    jn   
 6 3

  j 3 n  jn

j n

j n
jn

j n 
1 e e e e
6 6
e e 3 
    
2  jn jn nj 
 
 
sin n sin n
sin n 6  3
 
n n n
sin n
where 1 n  0 and zero elsewhere
n
 
sin n sin n
hd n   n  3  6
n n
6.2 The Discrete Fourier Transform (DFT)

xn  X ( )
DTFT
DTFT
xn X k 
DFT
DFT {DFT mapping x[n] to
another sequence}

The DFT denoted by X[k], allows to evaluate the Fourier

Transform X(). This complex valued sequence X[k] is obtained
by sampling the Fourier Transform X() at a finite number of
frequency points.

This sampling is conventionally performed at equally spaced

points over the period extending over -    .

The DFT allows us to determine the frequency content of a signal,

that is, to perform spectral analysis.

The DFT plays a central role in the implementation of a variety

of digital signal processing algorithms, as a result of the existence
of the efficient algorithm for the Fast Fourier Transform (FFT).

(1) Fourier transform of a discrete signal (DTFT or FTD) is


X ( )   xne
n  
 jn
,    

(2) X ( )  X ( z ) | z e j
{Evaluate z-transform on the unit circle}

(3) Discrete Fourier Transform (DFT) or N-point DFT

N 1  2k 
 jn  
X k    xne  N 

n 0
i.e. Sampling X() at N equally spaced interval, k = 0,1,2,… N-1.
Note: One of the most important properties of the DFT is implied
periodicity.
Number of time samples = N
Number of frequency samples (k) = N
 The DFT corresponds to sampling the z-transform of X(z) at N-
points equally spaced in angle around the unit circle.
2
N 2
2 N
N
 = -
 = - =0
 = - =0

N points equally spaced

on the unit circle
Note:
x[n] x[n] x[n]

Discrete

DTFT
or
FTD
|X()|
N time samples
DFT
|X()| (or
using
FFT)
Continuous
|X(k)|

- 

|X[k]|
Discrete
N frequency
Samples

-  
  When the DFT of a block of N samples is calculated, the
assumption is that the original signal actually repeats itself
periodically, with period N.
 Clearly, for all real signals this will not be true.
 Even for artificial signals such as a pure sinusoid this will only
be true if N is a multiple of the period of the sinusoid.
 This “windowing” process introduces a slight distortion into
the frequency representation of the signal being analysed.

 For example when a 256 point DFT is taken using a

rectangular window (grouping N samples), one can see that
the waveform does not repeat and as a result slight
frequency distortion will occur when calculating X[k].

6000
window
4000 x[n]
2000

-2000

-4000

-6000

-8000
256 samples
-10000

-12000
50 100 150 200 250 300 350 400 450 500

Example: Let fs = 8000 Hz, Number of sample(N) = 1000

f s 8000
Frequency resolution =   8Hz
N 1000

f0 = 0; f1 = 8Hz, f2 = 16 Hz, …. f999 = 8000Hz

 2k 
999  jn  
X k    xne  1000 

0
k = 0,1,2,3, …
 Example: A speech signal is sampled at a rate of 20000
samples/sec. A sequence of length (N) 1024 samples is selected
and the 1024-point DFT is computed.

(1) What is the time duration of segment of speech?

Duration = no of samples  sampling period.
= 1024 (1/20000) = 51.2 ms

(2) What is the frequency resolution (spacing in Hz) between

the DFT values.
f s 20000
Re solution    19.531Hz
N 1024

DFT calculation
2k 2
N 1  
X k    xn e
 jn
N
k  0,1, 2, 3,.... N N
n 0 f sampling
f 
N 1
2 N 1
2 N
X k    xn cos( nk )  j  xn sin( nk )
n 0 N n 0 N This is a
lengthly
X k   Ak   Bk 
procedure, if
X k   Ak   Bk  k  0,1, 2, 3,.... N for example, N
2 2

= 1024
samples
fk=k(fs/N) = k f
fs=sampling
N=256 frequency
|X[k]|

0 1 2 k
255
f0 f1 f2  f255

The frequency resolution (f) can be made as small as desired

by increasing the value of N (window size being analysed)
 6.1.1 Conjugate Symmetry

The DFT of a real sequence possesses conjugate symmetry about

the origin with

X [k ]  X * [k ]
Since the DFT is periodic,

X [ k ]  X [ N  k ]
N
This implies conjugate symmetry about the index k 
2

𝑋[−𝑘] = 𝑋 ∗ [𝑘] = 𝑋[𝑁 − 𝑘]

N is odd (N=7)

Conjugate
symmetry

k
0 1 2 3 4 5 N-1
N/2=3.5
N is even (N=8)

Conjugate
symmetry

k
0 1 2 3 4 5 6 N-1
N/2=4
Example: The DFT of a real signal is

{1, X1, -1, X2, 0, -j2, X3, -1+j}

Find X1, X2 and X3.

1 X1 -1 X2 0 -j2 X3 -1+j
k
0 1 2 3 4 5 6 7
8/2=4
8 Frequency band = 8 time samples

X 1  x * [1]  1  j
X 2  x * [5]  j 2
X 3  x * [2]  1
Example: Consider the finite length sequence.

x[n] = [n] + 0.2[n-2]

Find the N-point DFT of x[n] for N = 50

Solution: X(z) = 1 + 0.2 z-2  X() = 1 + 0.2 e-j2

The N-point DFT is obtained by evaluating X() at N points

equally spaced around the unit circle.

2k

N
Therefore the DFT is given by:

 j  2Nk 2
X k   1  0.2e
 j  225
k 
 1  0.2e
k = 0, 1, 2, …. 49

Exercise: Compute the N-point DFT H[k] of the sequence h[n].

Show that when N = 8 the value of H(2) = -1.

1  3  n  3
hn  
0 otherwise
Example:

Find FTD and DFT of the three samples average.

1
 0n2
hn   3
 0 otherwise
h[n]

1/3

0 1 2 n

N 1 2 2 2
H k    hne
 jn ( 1  jn ( k )
 e N
k)
N

n0 n0 3

2k
2 (1  2 cos( ))
H k   e N
 j( k)
N
3
2k
H k   [1  2 cos(
1
)] where k  0,1,2
3 N
 6.1.2 Parseval’s Theorem

DFT is an energy-conserving transformation that allows to find

the signal energy from the signal or its spectrum. Therefore, the
sum of squares of the signal samples is related to the sum of
squares of the magnitude of the DFT samples.

Example:

The DFT of a real signal is {1, X1, -1, X2, -7, -j2, X3, -1+j}. What
is the signal energy?
k=0 k=7
X1, X2 and X3 can be found first using conjugate symmetry
property

1
Signal Energy  
2
X [k ]
8
 12  ( 2 ) 2  ( 1) 2  2 2  7 2  2 2  12 
1
8
 2  
2

 1  2  1  4  49  4  1  2
1
8
8
6.3 Padding with Zeros and Frequency Resolution
N 1 2k
X k    xne
 jn
DFT : N
k  0,1, 2, 3,.... N
n 0

 To obtain more points in the DFT sequence, we can always

increase the duration of x[n] by adding additional zero-valued
elements. This procedure is called padding with zeros.

 These zero-valued elements contribute nothing to the sum in

the above equation, but act to decrease the frequency spacing
(2/N).

 The zero padding gives us a high-density spectrum and

provided a better displayed version for plotting. But it does not
give us a high resolution spectrum because no new information
is added to the signal. Only additional zeros are added in the
data.

Three Sample Averager

Padding with zeros

6.4 Inverse Discrete Fourier Transform (IDFT)
The inverse DFT equation comes directly from the Inverse
Fourier Transform equation

N 1 2
1
 X ( k )e
jn k
x[ n]  N
0  n  N-1
N k 0

However, the signal which it produces will be a periodic signal

repeating every N samples

Example:

If X[k] = {4, -j2, 0, j2}, find its IDFT

N=4
2kn
1 3
 x[ n]   X [ k ]e
j
4
4 k 0
kn
1 3
  X [ k ]e
j
2
4 k 0
1
n0 x[0]  [ 4  j 2  0  j 2]  1
4
 3
1 j j
n  1 x[1]  [ 4  j 2e  0  j 2e 2 ]  2
2
4
1
n  2 x[ 2]  [ 4  j 2e j  0  j 2e j 3 ]  1
4
3 9
1 j j
n  3 x[3]  [ 4  j 2e  0  j 2e 2 ]  0
2
4

The IDFT is x[n]={1, 2, 1, 0}

n=0
Use of DFT

• Clearly, the DFT is useful in that it allows the spectral

content of a signal to be determined
• Additionally, once in the frequency domain the DFT of a
signal can be processed in order to “filter” or “alter” the
signal is some desired fashion
• The IDFT can then be used to regenerate the processed
signal

Fast Fourier Transform

• The Fast Fourier Transform (FFT) is simply a mathematical

technique to accelerate the calculation of the DFT. It was
developed by Cooley and Tukey (1965) requires N to be a
power of 2.

• Typically, if the DFT is calculated for a block of 2n samples

e.g. 512 or 1024 samples (N) it would make the calculation
of the DFT quite demanding.

• The FFT simply uses repetition and redundancy in the

calculation to speed it up.

• The FFT is simply a TECHNIQUE to calculate the DFT,

NOT a different transform.

6.5 Comparison of FFT/DFT Complex

Multiplications
As already stated, the FFT algorithm is most efficient when the
number of time samples N to be transformed is a power of 2. If N
is not a power of 2 then it may be padded out with zero samples
to the nearest power of two before transformation.
A comparison of the number of complex multiplications required
for direct evaluation of the DFT and the number needed for the
Cooley-Tukey FFT is given below

No. of Complex FFT Times faster

Points Multiplications in algorithm than direct
DFT (N2) (N/2)log2(N) evaluation
16 256 32 8
256 65,536 1024 64
1024 1,048,576 5120 205

6.6 Summary
At the end of this chapter, it is expected that you should know:

 The definition of the discrete time Fourier transform and its

inverse.
 Given a filter magnitude response, find the impulse response.
 The properties of the DTFT and its applications.
 When to use the discrete time Fourier transform (DTFT) and
when to use the discrete Fourier transform (DFT).
 When and how to use windowed discrete Fourier transforms.
 Zero-padding and its relationship to frequency resolution.
 The zero padding gives us a high-density spectrum but does
not give us a high resolution spectrum
 The Fast Fourier Transform (FFT): the relative complexity
between the DFT and FFT in terms of multiplications.