BEE4413

DIGITAL SIGNAL PROCESSING

Topic 7

Discrete Fourier
Transform (DFT)
Subtopics

7.0 Discrete Fourier Transform (DFT) (6 Hours)

7.1 Introduction to DFT
7.2 Properties and Relationship to z-Transform
7.3 Inverse Discrete Fourier Transform (IDFT)
7.4 Fast Fourier Transform (FFT)
7.5 Inverse Fast Fourier Transform (IFFT)
7.6 Convolution using DFT technique

(BT Level 3: Applying)

Rearrange Subtopis

7.0 Discrete Fourier Transform (DFT) (6 Hours)

7.1 Introduction to DFT
7.2 Properties and Relationship to z-Transform
7.3 Inverse Discrete Fourier Transform (IDFT)
7.4 Fast Fourier Transform (FFT) and Inverse
Fast Fourier Transform (IFFT)
7.5 Convolution using DFT technique
7.1

INTRODUCTION TO DFT
Frequency Domain Sampling

 Frequency analysis has been studied in

Topic 4 where the frequency response of a
discrete LTI system were examine
j
 Recall DTFT: x(n)  X ( )  X (e )
 Exponential is a continuous function
 As for that, the exponential function needs to
be sampled to become discrete
Frequency Domain Sampling

 Since X(ω) is periodic with period 2π, only

samples in fundamental frequency range are
necessary
 As for that, the frequency ω from 0 to 2π can
be sampled by N samples
2k

N
where k is the variable for each sample
Frequency Domain Sampling

 If x(n) is a finite sequence that also has been

sample by N, the DTFT equation now has
becomes DFT:
 N -1 2kn
-j
- jn
X ( )   x ( n )e
n  -
 X ( k )   x ( n )e
n 0
N

 And the inverse of DFT is called IDFT:

N -1 2kn
1 j
x ( n) 
N
 X ( k )e
k 0
N
Example 1

 Let x(n) = {1,2,1,0}. Obtain X(k) by

transforming x(n) using DFT
Solution to Example 1

 The DFT equation is:

N -1 2kn
-j
X ( k )   x ( n )e N

n 0
 N = 4, as the sequence have 4 samples
 X(k) becomes:
4 -1 2kn 3 - jkn
-j
X ( k )   x ( n )e 4
  x ( n )e 2

n 0 n 0
Solution to Example 1

 The results:
3 - j ( 0 ) n 3
X (0)   x(n)e 2
  x ( n)  1  2  1  0  4
n 0 n0
3 - j (1) n 3 - jn - j
X (1)   x(n)e 2
  x ( n)e 2
 1  2e 2
 e - j  0
n 0 n 0

  
 1  2 cos - j sin   cos  - j sin    1  2- j  - 1  - j 2
 2 2
3 - j ( 2 ) n 3
X (2)   x(n)e 2
  x(n)e - jn  1  2e - j  e - j 2  0
n 0 n 0

 1  2cos  - j sin    cos 2 - j sin 2   1  2- 1  1  0

3 - j ( 3) n 3 - j 3 n - j 3
X (3)   x(n)e 2
  x ( n)e 2
 1  2e 2
 e - j 3  0
n 0 n 0

 3 3 
 1  2 cos - j sin   cos 3 - j sin 3   1  2 j   - 1  j 2
 2 2 
Solution to Example 1

 X (k )  {4,- j 2,0, j 2}
Example 2

 Find the 10-point DFT of each of the

following sequence:
a) x(n)   (n)   (n - 5)
b) x(n)  u (n) - u (n - 6)
Solution to Example 2(a)

 The sequence can also be written as:

x(n)  1,0,0,0,0,1
 The DFT required a 10-point sample, the
sequence should be written as:
x(n)  1,0,0,0,0,1,0,0,0,0
 As for that, the DFT for the sequence can be
written as (in variable of k):
N -1 2kn 9 kn
-j -j
X ( k )   x ( n )e N
  x ( n)e 5
 1  e - jk
n 0 n 0
Solution to Example 2(b)

 The sequence can also be written as:

x(n)  {1,1,1,1,1,1,0,0,0,0}
 The 10-point DFT is defined as:
9 kn
-j
X ( k )   x ( n )e 5

n 0

 By examining the sequence, there will only

be result for n = 0 until n= 5, as n ≥ 6 will
result in 0
Solution to Example 2(b)

 Hence, the DFT can be simplified as:

k 6
 -j 5 
kn k n 1 -  e  -j
 1- e 5
6k
5 -j 6 -1  -j 5 
X (k )   e 5
   e 
   k   k
n 0 n 0    -j 5  - j
1 -  e  1- e 5

 
Matrix Formulation of DFT
- j 2 N
 W
By substituting N  e , the definition of
DFT and IDFT can be written as:
N -1 2kn N -1
-j
DFT : X (k )   x(n)e N
  x(n)WNnk
n 0 n 0

N -1 2kn N -1
1 j 1 - nk
IDFT : x(n) 
N
 X ( k )e
k 0
N

N
 X (k )W
k 0
N
Matrix Formulation of DFT

 In matrix form: DFT : X N  WN x N

0 0
 X (0)  WN WN WN0 ... WN0   x ( 0) 
 X (1)  W 0 W 1 W 2
... WNN -1

x (1) 
   N N N  
2  N -1 
 X (2)  WN0 WN2 WN4 ... WN  x(2) 
 .  . . . . . . . 
    
 .   . . . . .  . 
 .   . . . . .

. 
   
 N -1 N -1  x ( N - 1) 
 X ( N - 1) 0
 WN WN
N -1
WN2  N -1 ... WN   
1 -1
IDFT : x N  W X N  WN X N
N
N
Example 1

 Find the DFT for the sequence x(n) =

{1,2,1,0} by using matrix formulation tehnique
Solution to Example 1

 As N = 4, the substitute WN will be:

kn
-j
nk 2
W N e

-j
2
 
WN  e  cos - j sin -j
2 2
Solution to Example 1

 The matrix formulation would be:

 X (0)  WN0 WN0 WN0 WN0   x(0)  - j 0 - j 0 - j0 - j 0  1 
 X (1)   0 1 2 3    0 1 3  
 W
 N W N W N W N
x (1)  - j - j - j2 - j  2
. .
 X (2) WN WN WN WN   x(2) - j 0 - j 2
0 2 4 6
- j4 6
- j 1 
   0 3 6 9    0 3 9  
 X (3)  WN WN WN WN   x(3)  - j - j - j6 - j  0
1 1 1 1  1   1  2  1   4 
1 - j - 1 j  2 1 - j 2 - 1 - j 2
 .     
1 - 1 1 - 1  1  1 - 2  1   0 
      
1 j - 1 - j 0
   1  j 2 - 1  j 2 
Solution to Example 1

 The result is:

 X (k )  {4,- j 2,0, j 2}
7.2

PROPERTIES AND
RELATIONSHIP TO Z-
TRANSFORM
DFT Properties
DFT Properties

 One of the most useful DFT properties is the

symmetry property
 The DFT of a real sequence possesses
conjugate symmetry about the origin:
X (-k )  X  (k )
 The conjugate symmetry is at k = 0.5N
Example 3

 Find the 8-point DFT of a sequence:

x(n)  1,1,0,0,0,0,0,0
Solution to Example 3

 The DFT would be:

7 kn k
-j -j
X ( k )   x ( n )e 4
 1 e 4

n 0
 Calculate only from X(0) until X(4), as after
that the conjugate symmetry property will
handle it:
X (2)  1 - j
X (0)  2
X (3)  0.2929 - j 0.7071
X (1)  1.7071 - j 0.7071
X (4)  0
Solution to Example 3

 As N = 8 is an even number, the values will

be conjugate symmetry at 0.5N = 4:

 The values for X(5), X(6) and X(7):

X (5)  X  (3)  0.2929  j 0.7071
X (6)  X  (2)  1  j
X (7)  X  (1)  1.7071  j 0.7071
Solution to Example 3

 Hence, the overall X(k) are:

 X (k )  {2,1.7071 - j 0.7071,1 - j ,0.2929 - j 0.7071,
0,0.2929  j 0.7071,1  j ,1.7071  j 0.7071}
DFT and z-Transform Relationship

 Recall z-transform definition:


-n
X ( z)   x ( n) z
n  -
 Recall frequency response using DTFT:
j
ze

- j n
 X ( )   x ( n )e
n  -
DFT and z-Transform Relationship

 Recall frequency sampling in introduction of

2k
DFT: 2k  - j n
 ,  X ( k )   x ( n )e N
N n  -
 As for that, the direct transformation from z-
transform to DFT is:
j 2k N
ze
 
-n - j 2kn N
 X ( z)   x ( n) z
n  -
 X (k )   x ( n)e
n  -
7.3

INVERSE DISCRETE FOURIER

TRANSFORM (IDFT)
IDFT

 IDFT is defined by:

N -1 2kn
1 j
x ( n) 
N
 X ( k )e
k 0
N

 The sequence x(n) will results in the same

number (point) as in X(k)
Example 4

 Let X(k) = {4,-j2,0,j2}. Find its IDFT.

Solution to Example 4

 Using the IDFT equation:

N -1 2kn kn
1 j 1 3 j
x ( n) 
N
 X ( k )e
k 0
N
  X ( k )e 2
4 k 0
1 3 1
x(0)   X (k )  4 - j 2  j 2   1
4 k 0 4
k  3
1 3 j 1  j j 
x(1)   X (k )e 2   4 - j 2e 2  j 2e 2 2

4 k 0 4 
1 3 1

x(2)   X (k )e jk  4 - j 2e j  j 2e j 3
4 k 0 4
1
3k 3 9
1 3 j 1  j j 
x(3)   X (k )e 2   4 - j 2e 2  j 2e 2 0

4 k 0 4 
Solution to Example 4

 Hence, the overall x(n) are:

 x(n)  {1,2,1,0}
IDFT by Using Matrix Formulation

 Recall the IDFT equation by using matrix

formulation:
-1 1 
x N  W X N  WN X N
N
N
 WN* means that the matrix WN needs to be
transpose and complex conjugate (conjugate
transpose)
 This is a very good advantage as for IDFT
can be obtained from its DFT
Example 4

 Let X(k) = {4,-j2,0,j2}. Find its IDFT by using

matrix formulation.
Solution to Example 4

 By using the same WN as in Example 1, the

DFT can be obtained, where WN needs to be
conjugate transpose
WN  WN   WN
T

complex
transpose conjugate

1 1 1 1  1 1 1 1  1 1 1 1
1 - j - 1 j  1 - j - 1 j  1 j - 1 - j 
   
1 - 1 1 - 1  1 - 1 1 - 1  1 - 1 1 - 1 
     
1 j - 1 - j  1 j - 1 - j  1 - j - 1 j 
Solution to Example 4

 Evaluate IDFT using the matrix formulation

equation: 1 
xN  WN X N
N
 x(0)  1 1 1 1   4  1 
 x(1)  1 j - 1 - j  - j 2 2
 1 .  
 x(2) 4 1 - 1 1 - 1   0  1 
      
 x(3)  1 - j - 1 j   j 2  0
 x(n)  {1,2,1,0}
7.4

FAST FOURIER TRANSFORM

(FFT) AND INVERSE FAST
FOURIER TRANSFORM (IFFT)
Introduction to FFT

 Even though DFT is a very useful and

implement-friendly transform of a discrete
sequence, it has one major disadvantage:
– Heavy computation when N is increase
 FFT is developed to overcome the DFT
problem, it is not another kind of transforms
 To simplified, FFT is a fast (computationally
efficient) way to calculate DFT
FFT Functionality

 FFT works by making use of periodicities and

symmetries in the exponential function e-j2πnk/N
 FFT will decomposes the set of data to be
transformed into a series of smaller data sets to
be transformed
 Then, it decomposes those smaller sets into
even smaller sets
 At each stage of processing, the results of the
previous stage are combined in special way
 Finally, it calculates the DFT of each small data
set
FFT Functionality

 For example, an FFT of size 16 is broken into 2 FFTs of

size 8, which are broken into 4 FFTs of size 4, which are
broken into 8
FFTs of size
2, which are
broken into
16 FFTs of
size 1
FFT Functionality

 It is possible to take the DFT of the first N/2

points and combine them in a special way with
the DFT of the second N/2 points to produce a
single N-point DFT (symmetry property)
 Each of these N/2-point DFTs can be calculated
using smaller DFTs in the same way
 This is called radix-2 FFT where r = 2 for rm = N
FFT Efficiency

 Normal DFT calculation needs N2 operations

for N points
 FFT total computation (for radix-2) takes only
Nlog2N
 Therefore, the ratio between a DFT
computation and an FFT computation for the
same N is proportional to N/log2N
 FFT becomes very efficient compared to
normal DFT calculation when N is increased
The Radix of FFT

 Radix is the size of FFT decomposition

 The most common and familiar FFT
implementation is radix-2
 There are some other radix that have been
used such as radix-4, where it has an
interestingly "twiddle factors“, which are all 1,
-1, j, or -j, that can be applied without any
multiplications at all (observe Example 1
using matrix formulation)
IFFT

 The inverse Fast Fourier Transform (IFFT)

will have the same implementation and
computational method as in FFT, except for
the multiplication with 1/N and the transpose
conjugate of the previous FFT
7.5

CONVOLUTION USING DFT

TECHNIQUE
Convolution in DFT

 Usually, when convolution involve in time-domain, it

will be converted into as multiplication in that
transform domain (s-domain, ω-domain, z-domain,
etc). Same for DFT
 Recall the convolution steps in transform-domain (in
k-domain)
1. x1(n)  X1(k)
2. x2(n)  X2(k)
3. X3(k) = X1(k) X2(k)
4. X3(k)  x3(n)
Convolution in DFT

 Interestingly, in DFT there will always be two

kind of convolution occurred:
1. Circular convolution
2. Linear convolution (the regular convolution)
 This is due to restricting the convolution to N-
point convolution
 Changing the value of N-point will resulted in
a very different convolution results
Circular Convolution in DFT

 Usually, in a regular (linear) convolution, the length

of the resulted sequence, N will be:
N  L  M -1
where L is the length for first sequence and M is the
length of the second sequence
 Circular convolution in DFT occurred when two
discrete sequence with the same length L = M are
being convolved in the same length of the resulted
sequence:
N LM
Example 4

 Determine the 4-point circular convolution of

the two sequence x1(n) and x2(n) given by:
x1 (n)  {1,2,0,1} & x2 (n)  {2,2,1,1}
Solution to Example 4

 The convolution needs to be a 4-point in results, as

for that circular convolution will be provoked
 Step 1: Transform from x1(n) to X1(k) using DFT
3 - jkn - jk - j 3k
X 1 (k )   x1 (n)e 2
 1  2e 2
e 2

n 0

X 1 ( 0)  1  2  1  4
- j - j 3
2 2
X 1 (1)  1  2e e  1- j
X 1 (2)  1  2e - j  e - j 3k  -2
X 1 (3)  1  j using the conjugate
 X 1 (k )  {4,1 - j ,-2,1  j} symmetry property
Solution to Example 4

 Step 2: Transform from x2(n) to X2(k) using DFT

3 - jkn - jk - j 3k
- jk
X 2 ( k )   x2 ( n ) e 2
 2  2e 2
e e 2

n0

X 2 (0)  2  2  1  1  6
- j - j 3
2 - j 2
X 2 (1)  2  2e e e  1- j
X 2 (2)  2  2e - j  e - j 2  e - j 3  0
X 2 (3)  1  j
 X 2 (k )  {6,1 - j ,0,1  j}
 Step 3: Multiply X1(k) with X2(k)
X 3 (k )  X 1 (k ) X 2 (k )  {4,1 - j ,-2,1  j}  {6,1 - j ,0,1  j}  {24,- j 2,0, j 2}
Solution to Example 4

 Step 3: Inverse X3(k)

kn n 3n
1 3 j 1 j j 
x ( n )   X ( k )e 2  2
  24 - j 2e  j 2e 2 

4 k 0 4 
1
x(0)  24 - j 2  j 2  6
4
 3
1 j j 
x(1)   24 - j 2e 2  j 2e 2   7
4 
1
 
x(2)  24 - j 2e j  j 2e j 3  6
4
3 9
1 j j 
x(3)   24 - j 2e  j 2e 2   5
2
4 
 x(n)  {6,7,6,5}
Problem 5.14

 Consider the sequences

x1 (n)  {0,1,2,3,4} x2 (n)  {0,1} s (n)  {1}
 For 5-point DFT,
a) Determine a sequence y(n) so that Y(k)=X1(k) X2(k)
b) Obtain the sequence x3(n) such that S(k)=X1(k) X3(k)
Solution to Problem 5.14

 As x2(n) and s(n) is not a 5-point sequence,

zero should be padded to fulfill the emptiness
of both sequences
 After zero-padding both of the sequences:
x1 (n)  {0,1,2,3,4}
x2 (n)  {0,1,0,0,0}
s (n)  {1,0,0,0,0}
Solution to Problem 5.14(a)

 To find y(n), x1(n) needs to be circular convolved

with x2(n):
4 - j 2nk - j 2k - j 4k - j 6k - j 8k
X 1 ( k )   x1 (n)e 5
e 5
 2e 5
 3e 5
 4e 5

n 0

X 1 (0)  1  2  3  4  10
- j 2 - j 4 - j 6 - j 8
5 5 5 5
X 1 (1)  e  2e  3e  4e  -2.5  j 3.441
- j 4 - j 8 - j12 - j16
5 5 5 5
X 1 ( 2)  e  2e  3e  4e  -2.5  j 0.8123
X 1 (3)  -2.5 - j 0.8123
X 1 ( 4)  -2.5 - j 3.441
 X 1 ( k )  {10,-2.5  j 3.441,-2.5  j 0.8123,-2.5 - j 0.8123,-2.5 - j 3.441}
Solution to Problem 5.14(a)
4 - j 2nk - j 2k
X 2 ( k )   x2 ( n )e 5
e 5

n 0

X 2 ( 0)  1
- j 2
5
X 2 (1)  e  0.309 - j 0.9511
- j 4
5
X 2 ( 2)  e  -0.809 - j 0.5878
X 2 (3)  -0.809  j 0.5878
X 2 ( 4)  0.309 - j 0.9511

 X 2 ( k )  {1,0.309 - j 0.9511,-0.809 - j 0.5878,

- 0.809  j 0.5878,0.309  j 0.9511}
Solution to Problem 5.14(a)

Y (k )  X 1 (k ) X 2 (k )
 {10,2.5  j3.441,2.5  j 0.8123,2.5 - j 0.8123,2.5 - j 3.441}

2kn
1 5 j
y ( n )   Y ( k )e 5
4 k 0
2k 4k
 j j 
1 10  2.5  j 3.441e
 5
 2.5  j 0.8123e 5 
  6k 8k

4 j j 
  2 .5 - j 0 . 8123e 5
 2 . 5 - j 3 .441e 5


 y (n)  {4,0,1,2,3}
Solution to Problem 5.14(b)
4 - j 2nk
 Obtain the DFT of s(n):
S ( k )   s ( n )e 5
1
n 0

S (0)  S (1)  S ( 2)  S (3)  S ( 4)  1

 S ( k )  {1,1,1,1,1}
 Calculate X3(k):
S (k ) {1,1,1,1,1}
X 3 (k )  
X 1 (k ) 10,-2.5  j 3.441,-2.5  j 0.8123,
 
 - 2 . 5 - j 0 . 8123 , - 2 . 5 - j 3 . 441 
0.1,-0.1382 - j 0.1902,-0.3618 - j 0.1176,
 
 - 0 . 3618  j 0 . 1176 , - 0 . 1382 - j 0 . 1902 
Solution to Problem 5.14(b)

 Inversing X3(k) with IDFT to find x3(n):

x3 (n)
2kn
1 5 j
  X 3 ( k )e 5
4 k 0
2k 4k
 j j 
1 0.1  - 0.1382 - j 0.1902e
 5
 - 0.3618 - j 0.1176e 5 
  6k 8k

4 j j 
  - 0. 3618  j 0 . 1176 e 5
 - 0 . 1382 - j 0 . 1902 e 5


 x3 (n)  {- 0.18,0.22,0.02,0.02,0.02}
Linear Convolution in DFT

 To perform the linear convolution, the length

of the resulted sequence N must be:
N  L  M -1
where L is the length for first sequence and
M is the length of the second sequence
 The process is the same as in circular
convolution, where both sequences need to
be padded with zeros for length to become N
Example 5

 Determine the linear convolution of the two

sequence x1(n) and x2(n) given by:
x1 (n)  {1,2,0,1} & x2 (n)  {2,2,1,1}
Solution to Example 5

 Calculate N: N  4  4 - 1  7
 Padding zeros to both sequences:
x1 (n)  {1,2,0,1,0,0,0}
x2 (n)  {2,2,1,1,0,0,0}
 Obtain the DFT for both sequences:
4,1.346 - j1.9975,1.1784 - j1.168,-1.0245 - j1.8427,
X 1 (k )   
- 1.0245  j1.8427,1.1784  j1.168,1.346  j1.9975 
6,2.1235 - j 2.9725,1.2775 - j 0.7341,0.599 - j1.0609,
X 2 (k )   
0.599  j1.0609,1.2775  j 0.7341,2.1235  j 2.9725 
Solution to Example 5

 The result from linear convolution:

X 3 (k )
 X 1 (k ) X 2 (k )
24,-3.0794 - j8.2428,0.6479 - j 2.3573,-2.5685 - j 0.017,
 
 - 2. 5685  j 0. 017, 0. 6479  j 2 . 3573, - 3. 0794  j 8 . 2428 
 Inversing the result to obtain its sequence in
time-domain:
 x3 (n)  {2,6,5,5,4,1,1}
Example 5.3.1

 By means of the DFT and IDFT, determine

the response of the FIR filter with impulse
response
h(n)  {1,2,3}
to the input sequence
x(n)  {1,2,2,1}
Solution to Example 5.3.1

 First, determine the linear convolution length:

N  3  4 -1  6
 Padding h (n) and x(n) with zeros:
h(n)  {1,2,3,0,0,0}
x(n)  {1,2,2,1,0,0}
 Obtaining the DFTs:
6,0.5 - j 4.3301,-1.5  j 0.866,
H (k )   
2,-1.5 - j 0.866,0.5  j 4.3301 
X (k )  6,- j 3.4641,0,0,0,- j 3.4641
Solution to Example 5.3.1

 Calculating Y(k):
Y (k )  H (k ) X (k )  36,-15 - j1.7321,0,0,0,-15  j1.7321
 Obtaining y(n) by using IDFT:
 y (n)  {1,4,9,11,8,3}