International Islamic University Chittagong

Dept. of Electrical and Electronic Engineering

Segment-6
Discrete Fourier Transform (DFT) &
Fast Fourier Transform (FFT)

EEE-3603 Digital Signal Processing

Prepared By
Mohammed Abdul kader
Assistant Professor, Dept. of EEE, IIUC
 Contents
 Fourier Analysis
 Discrete Fourier Transform (DFT)
 Difference between DTFT and DFT.
 The DFT as linear Transformation.
 Problems related to DFT and IDFT
 Circular convolution
 Fast Fourier Transform
 Decimation in Time FFT Algorithm
Reference Book:
Digital Signal Processing (4th Edition), John G. Proakis, Dimitris K Manolakis
Chapter 7 (Discrete Fourier Transform) and
Chapter 8(Efficient Computation of the DFT)
Digital Signal Processing, Barrie Jervis
Chapter-2 (Discrete Transforms)
2
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Fourier Analysis

Fourier analysis convert a time domain signal into frequency domain signal.
Can be divided into 4 types:
a) Aperiodic continuous.
b) Periodic continuous (Fourier Series).
c) Aperiodic Discrete (DTFT).
d) Periodic Discrete (DFT)

3 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 DFT
Discrete Fourier Transform
The discrete Fourier transform of a discrete-time signal x(n) is defined as

𝑁−1
2𝜋𝑛𝑘
−𝑗 𝑁 K=0, 1, ….., N-1
𝑋 𝑘 = 𝑥(𝑛)𝑒
𝑛=0

The Inverse Discrete Fourier Transform (IDFT) is defined as

𝑁−1
1 2𝜋𝑛𝑘
𝑗 𝑁
𝑥 𝑛 = 𝑋(𝑘)𝑒 , 𝑛 = 0,1, … … … 𝑁 − 1
𝑁
𝑛=0

DFT
𝑥(𝑛) 𝑋(𝑘)
Time Domain IDFT Frequency Domain

4
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Difference between DTFT and DFT
+∞
𝑁−1
𝑋 𝜔 = 𝑥(𝑛)𝑒 −𝑗𝜔𝑛 −𝑗
2𝜋𝑛𝑘
𝑋 𝑘 = 𝑥(𝑛)𝑒 𝑁
𝑛=−∞
𝑛=0

In DTFT frequency domain (𝜔) is 2𝜋𝑘

In DFT frequency domain (𝜔 = 𝑁 ) is
continuous.
𝜔 changes from 0 to 2𝜋, but it is continuous. discrete.
𝜔 changes from 0 to 2𝜋 taking 0 to (N-1)
number of samples.

5
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The DFT as a Linear Transformation
𝑁−1
2𝜋𝑛𝑘
−𝑗
The N-Point DFT is defined as 𝑋 𝑘 = 𝑥(𝑛)𝑒 𝑁
𝑛=0
𝑁−1
The above expression also written as 𝑋 𝑘 = 𝑥(𝑛)𝑊𝑁𝑛𝑘
𝑛=0
2𝜋𝑛𝑘
−𝑗 𝑁
Where, 𝑊𝑁𝑛𝑘 = 𝑒 Is called the twiddle factor

If N=4 we say it is 4-Point DFT. For N=4, the above expression can be written as
3

𝑋 𝑘 = 𝑥(𝑛)𝑊4𝑛𝑘
𝑛=0

𝑋 0 = 𝑥 0 𝑊40 + 𝑥(1)𝑊40 + 𝑥(2)𝑊40 + 𝑥(3)𝑊40

𝑋 1 = 𝑥 0 𝑊40 + 𝑥(1)𝑊41 + 𝑥(2)𝑊42 + 𝑥(3)𝑊43

𝑋 2 = 𝑥 0 𝑊40 + 𝑥(1)𝑊42 + 𝑥(2)𝑊44 + 𝑥(3)𝑊46

𝑋 3 = 𝑥 0 𝑊40 + 𝑥(1)𝑊43 + 𝑥(2)𝑊46 + 𝑥(3)𝑊49

6
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The DFT as a Linear Transformation (cont.)
𝑊43

𝑋(0) 𝑊40 𝑊40 𝑊40 𝑊40 𝑥(0)

𝑋(1) 𝑊40 𝑊41 𝑊42 𝑊43 𝑥(1)
=
𝑋 2 𝑊40 𝑊42 𝑊44 𝑊46 𝑥(2)
𝑋(3) 𝑥(3) 𝑊42 𝑊40
𝑊40 𝑊43 𝑊46 𝑊49

𝑋(0) 1 1 1 1 𝑥(0) 𝑊41

𝑋(1) 1 −𝑗 −1 𝑗 𝑥(1)
= Periodicity property of W
𝑋 2 1 −1 1 −1 𝑥(2)
𝑁
𝑋(3) 1 𝑗 −1 −𝑗 𝑥(3) 𝑘+ 2
𝑊𝑁 = −𝑊𝑁𝑘 ≫ 𝑊4𝑘+2 = −𝑊4𝑘
Note that for 4 point DFT, the computation of each point of DFT can be accomplished by 4 complex
multiplication and (4-1) complex additions.
Total computation for 4-point DFT, Addition= (4-1)X3 and Multiplication= 4X4
Hence the N-point DFT values can be computed in a total of 𝑁 2 multiplications and N(N-1) complex
additions.
7 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The DFT as a Linear Transformation (cont.)

𝑋0 1 1 1 . . 1 𝑥(0)
𝑋1 1 𝑊𝑁1 𝑊𝑁2 . . 𝑊𝑁𝑁−1 𝑥(1)
𝑋2 2(𝑁−1)
1 𝑊𝑁2 𝑊𝑁4 . . 𝑊𝑁 𝑥(2)
𝑋3 =
. . . . . . .
.
. . . . . . .
. 2(𝑁−1) (𝑁−1)(𝑁−1)
1 𝑊𝑁𝑁−1 𝑊𝑁 . . 𝑊𝑁 𝑥(𝑁 − 1)
𝑋(𝑁 − 1)
The N-point DFT may be expressed in matrix form as:
Where 𝑊𝑁 is the matrix of the linear transformation. If we assume that the inverse of 𝑊𝑁 exists,
we obtain

The expression of inverse DTFT can be expressed as

8
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on DFT
Example 7.1.3: Compute the DFT of the 4-point sequence
x(n)={0,1,2,3}
Solution:
The matrix of the linear transformation for 4-point DFT can be expressed as

𝑊40 𝑊40 𝑊40 𝑊40 1 1 1 1

𝑊40 𝑊41 𝑊42 𝑊43
1 −𝑗 −1 𝑗
𝑊4 = =
𝑊40 𝑊42 𝑊44 𝑊46 1 −1 1 −1
𝑊40 𝑊43 𝑊46 𝑊49 1 𝑗 −1 −𝑗

Now, Using the matrix expression 𝑋𝑁 = 𝑊𝑁 𝑥𝑁 , we can calculate DFT of the above sequence as

𝑋(0) 1 1 1 1 0 0+1+2+3 6
𝑋(1) 1 −𝑗 −1 𝑗 1 0 − 𝑗 − 2 + 3𝑗 −2 + 2𝑗
= = =
𝑋 2 1 −1 1 −1 2 0−1+2−3 −2
𝑋(3) 1 𝑗 −1 −𝑗 3 0 + 𝑗 − 2 − 3𝑗 −2 − 2𝑗

9 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on IDFT

Problem: The 4-point DFT of a discrete time sequence x(n) is given below. Determine
x(n).

Solution: X(k)={6,-2+2j,-2,-2-2j}
The matrix of the linear transformation for 4-point DFT can be expressed as

𝑊40 𝑊40 𝑊40 𝑊40 1 1 1 1 1 1 1 1

𝑊40 𝑊41 𝑊42 𝑊43 1 −𝑗 −1 𝑗 1 𝑗 −1 −𝑗
𝑊4 = = So, 𝑊4∗ =
𝑊40 𝑊42 𝑊44 𝑊46 1 −1 1 −1 1 −1 1 −1
𝑊40 𝑊43 𝑊46 𝑊49 1 𝑗 −1 −𝑗 1 −𝑗 −1 𝑗
1
Now, Using the matrix expression 𝑥𝑁 = 𝑁 𝑊𝑁∗ 𝑋𝑁 , we can calculate IDFT of the above sequence as
𝑥(0) 1 1 1 1 6 0 0
𝑥(1) 1 1 𝑗 −1 −𝑗 −2 + 2𝑗 1 4 1
= = =
𝑥 2 4 1 −1 1 −1 −2 4 8 2
𝑥(3) 1 −𝑗 −1 𝑗 −2 − 2𝑗 12 3

Problem: Compute the convolution of the sequences 𝒙𝟏 𝒏 = {𝟏, 𝟐, 𝟑, 𝟏} and

𝒙𝟐 𝒏 = {𝟒, 𝟑, 𝟐, 𝟐} using DFT and IDFT
10 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Multiplication of Two DFTs and Circular Convolution
Suppose that we have two finite-duration sequences of length N, 𝑥1 (𝑛) and 𝑥2 (𝑛). Their respective
N-point DFTs are

If we multiply the two DFTs together, the result is a DFT say 𝑋3 𝑘 , of a sequence 𝑥3 (𝑛) of length N.
Let us determine the relationship between 𝑥3 (𝑛) and the sequences 𝑥1 (𝑛) and 𝑥2 (𝑛).

The IDFT of 𝑋3 𝑘 is:

11
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Multiplication of Two DFTs and Circular Convolution (Cont.)
Putting the value 𝑋1 (𝑘) and 𝑋2 (𝑘)

We can write
Where,

if (m-n-l) is a multiple of N, i.e.

𝑚 − 𝑛 − 𝑙 = 𝑝𝑁, 𝑝 𝑖𝑠 𝑎𝑛 𝑖𝑛𝑡𝑒𝑔𝑒𝑟
a becomes 𝑎 = 𝑒 𝑗2𝜋𝑝 = 1

12 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Multiplication of Two DFTs and Circular Convolution (Cont.)

𝑎𝑁 = 1, for any value 𝑎 ≠ 0

𝑎𝑁 = 𝑒 𝑗2𝜋(𝑚−𝑛−𝑙) = cos 2𝜋 𝑚 − 𝑛 − 𝑙 + 𝑗𝑠𝑖𝑛 2𝜋 𝑚 − 𝑛 − 𝑙 =1

Finally we can write,

The above expression has the form of a convolution sum. However it is not the ordinary linear
convolution which relates the output sequence y(n) of a linear system to the input sequence x(n) and
impulse response h(n). The above expression involves the index (𝒎 − 𝒏)𝑵 and is called
circular convolution. Multiplication of DFTs of two sequences is equivalent to the
circular convolution of the two sequences in the time domain.
13
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on Circular Convolution
Problem: Compute the circular convolution of following two sequences

Solution:

We can represent the sequences in graph where the samples are placed in counterclockwise direction
in a circle

14 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on Circular Convolution (Cont.)

𝐱 𝟑 𝟎 = 𝟐 + 𝟒 + 𝟔 + 𝟐 = 𝟏𝟒
15 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on Circular Convolution (Cont.)

𝐱 𝟑 𝟏 = 𝟒 + 𝟏 + 𝟖 + 𝟑 = 𝟏𝟔

𝐱 𝟑 𝟐 = 𝟔 + 𝟐 + 𝟐 + 𝟒 = 𝟏𝟒
16
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on Circular Convolution (Cont.)

𝐱 𝟑 𝟑 = 𝟖 + 𝟑 + 𝟒 + 𝟏 = 𝟏𝟔

𝑥3 (𝑛) = {14,16, 14,16}

↑

17 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Alternative method of computing Circular Convolution

Problem: Compute the circular convolution of following two sequences

Solution:
𝑥3 (0) 2 1 2 1 1 2+2+6+4 14
𝑥3 (1) 1 2 1 2 2 1+4+3+8 16
= = =
𝑥3 (2) 2 1 2 1 3 2+2+6+4 14
𝑥3 (3) 1 2 1 2 4 1+4+3+8 16

𝑥3 (𝑛) = {14,16, 14,16}

↑

18 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Zero padding
Zero padding is a simple concept, it simply refers to adding zeros to end of a time domain signals to
increase its length.

19 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Computation of linear convolution by circular convolution
Compute the linear convolution of the following two sequences by circular
convolution

Solution:
Length of convolution sequence = 4+41=7
Start of sequence= min (min(x(n)) min(h(n))) =-1
After zero padding: 𝑥 (𝑛) = {1,2, 3,1,0,0,0} ℎ (𝑛) = {1,2, 1, −1,0,0,0}
↑ ↑

𝑦 −1 1 0 0 0 1 3 2 1 1
𝑦 0 2 1 0 0 0 1 3 2 4
𝑦 1 3 2 1 0 0 0 1 1 8
𝑦 2 = 1 3 2 1 0 0 0 −1 = 8
𝑦(3) 0 1 3 2 1 0 0 0 3
𝑦(4) 0 0 1 3 2 1 0 0 −2
𝑦(5) 0 0 0 1 3 2 1 0 −1

20 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Fast Fourier Transform (FFT)
A Fast Fourier Transform (FFT) is an efficient algorithm to compute the Discrete Fourier Transform
(DFT) and inverse of DFT. FFT requires a smaller number of arithmetic operations such as
multiplications and addition than DFT (i.e. FFT requires lesser computation time than DFT).

No of computations in direct DFT Computations in FFT

Multiplications: 𝑁 2 𝑁
Multiplications: 2 𝑙𝑜𝑔2 (𝑁)
Additions: N(N-1)
Additions: 𝑁 𝑙𝑜𝑔2 (𝑁)
For, 𝑁 = 106
Total mathematical operations required to find DFT
Direct DFT: 1012 + 106 106 − 1 ≈ 2 × 1012
FFT: (5 × 105 × 𝑙𝑜𝑔2 (106 ) + 106 × 𝑙𝑜𝑔2 (106 ) ≅ 24 × 106
If each mathematical operation needs 1 ns to compute by digital computer,
Direct DFT needs 2 × 1012 𝑛𝑠 = 2 × 103 𝑠 = 2000𝑠
Whether FFT algorithm needs 24 × 106 ns = 24 × 10−3 𝑠 = 0.024 𝑠
21
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Fast Fourier Transform (FFT) Algorithm
 Direct computation of the DFT is less efficient because it does not exploit the properties of
symmetry and periodicity of the phase factor 𝑊𝑁𝑛𝑘

 FFT algorithms exploit the above properties of phase factor to reduce the number of mathematical
calculations to compute DFT. There are many FFT algorithm which involves a wide range of
mathematics.
 On the basis of decimation (decimation means decomposition into decimal parts) process FFT
algorithms are two types.
 Decimation-in-Time FFT algorithm: The sequence x(n) will be broken up into odd numbered
and even numbered subsequences. This algorithm was first proposed by Cooley and Tukey in 1965.
 Decimation-in-Frequency algorithm: The sequence x(n) will be broken up into two equal
halves. This algorithm was first proposed by Gentlemen and Sande in 1966.

22
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The decimation in time FFT Algorithm
Cooley-Tukey Algorithm
X(k)
𝑁−1
2𝜋𝑛𝑘
−𝑗 𝑁 ,
𝑋 𝑘 = 𝑥𝑛 𝑒 𝑘 = 0,1, … , 𝑁 − 1 Xe(k) Xo(k)
𝑛=0

𝑁−1 Xee(k) Xoe(k)

2𝜋
= 𝑥𝑛 𝑊𝑁𝑛𝑘 Where, 𝑊 = 𝑒 −𝑗 𝑁

𝑛=0 Xeo(k) Xoo(k)

𝑵 𝑵 Some relationship involving 𝑾𝑵
−𝟏 −𝟏
𝟐 𝟐
(𝟐𝒏+𝟏)𝒌
= 𝒙𝟐𝒏 𝑾𝟐𝒏𝒌
𝑵 + 𝒙𝟐𝒏+𝟏 𝑾𝑵 𝟐𝝅 𝟐𝝅
−𝒋 −𝒋
𝒏=𝟎 𝒏=𝟎 𝑾𝟐𝑵 = (𝒆 𝑵 )𝟐 = 𝒆 𝑵/𝟐 = 𝑾𝑵/𝟐
𝑵
(𝒌+ ) 𝑵/𝟐
Symmetry: 𝑾𝑵 𝟐 = 𝑾𝒌𝑵 𝑾𝑵
Even sequence Odd sequence
= 𝑾𝒌𝑵 𝒆−𝒋𝝅 = −𝑾𝒌𝑵
𝑥0 , 𝑥2 , 𝑥4 , 𝑥6 𝑥1 , 𝑥3 , 𝑥5 , 𝑥7 (𝒌+𝑵)
Periodicity: 𝑾𝑵 = −𝑾𝒌𝑵
𝑁 𝑁
2 −1 2 −1
2𝑛𝑘 𝑛𝑘
= 𝑥2𝑛 𝑊𝑁/2 +𝑊𝑁𝑘 𝑥2𝑛+1 𝑊𝑁/2 = 𝑋𝑒 𝑘 + 𝑊𝑁𝑘 𝑋𝑜 𝑘 , 𝑘 = 0,1, … , 𝑁 − 1
𝑛=0 𝑛=0

23
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The decimation in time FFT Algorithm (Cont.)
𝑁
𝑋𝑒 𝑘 = 𝑋𝑒𝑒 𝑘 + 𝑊𝑁𝑘 𝑋𝑒𝑜 𝑘 , 𝑘 = 0 𝑡𝑜 −1
2 2

Even sequence Odd sequence

𝑥0 , 𝑥4 𝑥2 , 𝑥6
𝑋𝑒𝑒 𝑘 = 𝑥𝑜 + 𝑊𝑁𝑘 𝑥4 𝑋𝑒𝑜 𝑘 = 𝑥2 + 𝑊𝑁𝑘 𝑥6
4 4

𝑋𝑒𝑒 0 = 𝑥𝑜 + 𝑥4 𝑋𝑒𝑜 0 = 𝑥2 + 𝑥6 2𝜋
Stage-1

−𝑗 2 .1
𝑊𝑁1 = 𝑊21 = 𝑒
𝑋𝑒𝑒 1 = 𝑥𝑜 − 𝑥4 𝑋𝑒𝑜 1 = 𝑥2 − 𝑥6 4
= −1
𝑁
𝑋𝑜 𝑘 = 𝑋𝑜𝑒 𝑘 + 𝑊𝑁𝑘 𝑋𝑜𝑜 𝑘 , 𝑘 = 0 𝑡𝑜 −1
2
2

Even sequence Odd sequence

𝑥1 , 𝑥5 𝑥3 , 𝑥7
𝑋𝑜𝑒 0 = 𝑥1 + 𝑥5 𝑋𝑜𝑜 0 = 𝑥3 + 𝑥7

𝑋𝑜𝑒 1 = 𝑥1 − 𝑥5 𝑋𝑜𝑜 1 = 𝑥3 − 𝑥7

24 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The decimation in time FFT Algorithm (Cont.)
𝑁
𝑋𝑒 𝑘 = 𝑋𝑒𝑒 𝑘 + 𝑊𝑁𝑘 𝑋𝑒𝑜 𝑘 , 𝑘 = 0 𝑡𝑜 −1
2 2
0 𝑋𝑒𝑒 2 = 𝑥𝑜 + 𝑊𝑁2 𝑥4
𝑋𝑒 0 = 𝑋𝑒𝑒 0 + 𝑊8/2 𝑋𝑒𝑜 0 = 𝑋𝑒𝑒 0 + 𝑊80 𝑋𝑒𝑜 0 4
= 𝑥𝑜 + 𝑊22
𝑥4 = 𝑥𝑜 + 𝑥4
𝑋𝑒 1 = 𝑋𝑒𝑒 1 + 𝑊81 𝑋𝑒𝑜 1 = 𝑋𝑒𝑒 1 + 𝑊82 𝑋𝑒𝑜 1 = 𝑋𝑒𝑒 0
2
𝑋𝑒𝑜 2 = 𝑋𝑒𝑜 0
𝑋𝑒 2 = 𝑋𝑒𝑒 2 + 𝑊82 𝑋𝑒𝑜 2 = 𝑋𝑒𝑒 0 − 𝑊80 𝑋𝑒𝑜 0 2
2𝜋4
−𝑗 8
2 𝑊8 = 𝑒 = −1
2
Stage-2

𝑋𝑒 3 = 𝑋𝑒𝑒 3 + 𝑊83 𝑋𝑒𝑜 3 = 𝑋𝑒𝑒 1 − 𝑊82 𝑋𝑒𝑜 1 N

(k+ 2 )
2 WN = −𝑾𝒌𝑵
Similarly,
0
𝑋𝑜 0 = 𝑋𝑜𝑒 0 + 𝑊8/2 𝑋𝑜𝑜 0 = 𝑋𝑜𝑒 0 + 𝑊80 𝑋𝑜𝑜 0

𝑋𝑜 1 = 𝑋𝑜𝑒 1 + 𝑊81 𝑋𝑜𝑜 1 = 𝑋𝑜𝑒 1 + 𝑊82 𝑋𝑜𝑜 1

2

𝑋𝑜 2 = 𝑋𝑜𝑒 2 + 𝑊82 𝑋𝑜𝑜 2 = 𝑋𝑜𝑒 0 − 𝑊80 𝑋𝑜𝑜 0

2

𝑋𝑜 3 = 𝑋𝑜𝑒 3 + 𝑊83 𝑋𝑜𝑜 3 = 𝑋𝑜𝑒 1 − 𝑊82 𝑋𝑜𝑜 1

2
25 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The decimation in time FFT Algorithm (Cont.)

𝑋 𝑘 = 𝑋𝑒 𝑘 + 𝑊𝑁𝑘 𝑋𝑜 𝑘 ,

𝑋 0 = 𝑋𝑒 0 + 𝑊80 𝑋𝑜 0 𝑋𝑒𝑒 4 = 𝑥𝑜 + 𝑊24 𝑥4

𝑋 1 = 𝑋𝑒 1 + 𝑊81 𝑋𝑜 1 = 𝑥𝑜 + 𝑥4 =𝑋𝑒𝑒 0

𝑋 2 = 𝑋𝑒 2 + 𝑊82 𝑋𝑜 2 Similarly, 𝑋𝑒𝑜 =𝑋𝑒𝑜 0

Stage-3

𝑋 3 = 𝑋𝑒 3 + 𝑊83 𝑋𝑜 3 𝑋𝑒 4 = 𝑋𝑒𝑒 4 + 𝑊84 𝑋𝑒𝑜 4

2
𝑋 4 = 𝑋𝑒 4 + 𝑊84 𝑋𝑜 4 = 𝑋𝑒 0 − 𝑊80 𝑋𝑜 0 = 𝑋𝑒𝑒 0 + 𝑋𝑒𝑜 0
= 𝑋𝑒 0
𝑋 5 = 𝑋𝑒 5 + 𝑊85 𝑋𝑜 5 = 𝑋𝑒 1 − 𝑊81 𝑋𝑜 1 Similarly, 𝑋𝑜 4 = 𝑋0 0
8
(0+2)
𝑋 6 = 𝑋𝑒 6 + 𝑊86 𝑋𝑜 6 = 𝑋𝑒 2 − 𝑊82 𝑋𝑜 2 W8 = −𝑾𝟎𝟖
𝑋 7 = 𝑋𝑒 7 + 𝑊87 𝑋𝑜 7 = 𝑋𝑒 3 − 𝑊83 𝑋𝑜 3

26 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The decimation in time FFT Algorithm (Cont.)
8-point decimation in time FFT Algorithm
𝑿𝒆𝒆 (𝟎) 𝑿𝒆 (𝟎)
𝒙𝒐
𝑿𝒆𝒆 (𝟏) 𝑿𝒆 (𝟏)
𝒙𝟒
𝑿𝒆𝒐 (𝟎) 𝑿𝒆 (𝟐)
𝒙𝟐
𝑊80
𝑿𝒆𝒐 (𝟏) 𝑿𝒆 (𝟑)
𝒙𝟔
𝑊82
𝑿𝒐𝒆 (𝟎) 𝑿𝒐 (𝟎)
𝒙𝟏
𝑊80
𝑿𝒐𝒆 (𝟏) 𝑿𝒐 (𝟏)
𝒙𝟓
𝑊81
𝑿𝒐𝒐 (𝟎) 𝑿𝒐 (𝟐)
𝒙𝟑 𝑊80 𝑊82
𝑿𝒐𝒐 (𝟏) 𝑿𝒐 (𝟑)
𝒙𝟕 𝑊82 𝑊83
27
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The decimation in time FFT Algorithm (Cont.)
Reduction of computational complexity
The basic computation performed at every stage
(previous fig: 8-point DIT FFT algorithm) is to take two
complex numbers, say the pair (a,b), multiply b by 𝑊𝑁𝑟 ,
and then add and subtract the product from a to form
Fig: Basic butterfly computation in
two new complex numbers (A,B). This basic
the decimation-in-time FFT
computation is called a butterfly because the flow algorithm
graph resembles a butterfly.
In general, each butterfly involves one complex multiplication and two complex additions. For N
point FFT, there are N/2 butterflies per stage of the computation process and 𝑙𝑜𝑔2 (𝑁) stages.
Therefore,
𝑁
Total number of complex multiplications is 𝑙𝑜𝑔2 (𝑁)
2

And complex addition is 𝑁 𝑙𝑜𝑔2 (𝑁)

28
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 The decimation in time FFT Algorithm (Cont.)
Comparison of Computational Complexity for the Direct Computation of the DFT
Versus the FFT Algorithm

29 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on FFT
Problem: For the discrete time sequence x(n)={1,1,1,0,0,0,0,0}, find the 8-point
DFT using DIT-FFT algorithm.
Solution:
Even 𝑥0 𝑥4 𝑥2 𝑥6
Part 1 0 1 0
Odd 𝑥1 𝑥5 𝑥3 𝑥7
Part 1 0 0 0
Stage-1: 𝑋𝑒𝑒 0 = 𝑥𝑜 + 𝑥4 =1
𝑋𝑒𝑒 1 = 𝑥𝑜 − 𝑥4 = 1

𝑋𝑒𝑜 0 = 𝑥2 + 𝑥6 = 1
𝑋𝑒𝑜 1 = 𝑥2 − 𝑥6 = 1

𝑋𝑜𝑒 0 = 𝑥1 + 𝑥5 = 1
𝑋𝑜𝑒 1 = 𝑥1 − 𝑥5 = 1

𝑋𝑜𝑜 0 = 𝑥3 + 𝑥7 = 0
𝑋𝑜𝑜 1 = 𝑥3 − 𝑥7 = 0
30
Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on FFT (Cont.)
𝜋
−𝑗 2
Stage-2 𝑊82 = 𝑒 = −𝑗

𝑋𝑒 0 = 𝑋𝑒𝑒 0 + 𝑊80 𝑋𝑒𝑜 0 = 2

𝑋𝑒 1 = 𝑋𝑒𝑒 1 + 𝑊82 𝑋𝑒𝑜 1 = 1 − 𝑗

𝑋𝑒 2 = 𝑋𝑒𝑒 0 − 𝑊80 𝑋𝑒𝑜 0 = 0

𝑋𝑒 3 = 𝑋𝑒𝑒 1 − 𝑊82 𝑋𝑒𝑜 1 = 1 + 𝑗

𝑋𝑜 0 = 𝑋𝑜𝑒 0 + 𝑊80 𝑋𝑜𝑜 0 = 1

𝑋𝑜 1 = 𝑋𝑜𝑒 1 + 𝑊82 𝑋𝑜𝑜 1 = 1

𝑋𝑜 2 = 𝑋𝑜𝑒 0 − 𝑊80 𝑋𝑜𝑜 0 = 1

𝑋𝑜 3 = 𝑋𝑜𝑒 1 − 𝑊82 𝑋𝑜𝑜 1 = 1

31 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC
 Problem on FFT (Cont.)
Stage-3
−𝑗
𝜋 1 1 3𝜋 1 1
𝑊81 = 𝑒 4 = −𝑗 𝑊83 = 𝑒 −𝑗 4
=− −𝑗
√2 √2 √2 √2

𝑋 0 = 𝑋𝑒 0 + 𝑊80 𝑋𝑜 0 =3
1 1
𝑋 1 = 𝑋𝑒 1 + 𝑊81 𝑋𝑜 1 =1−j+( −𝑗 )
2 2
𝑋 2 = 𝑋𝑒 2 + 𝑊82 𝑋𝑜 2 = −𝑗
1 1
𝑋 3 = 𝑋𝑒 3 + 𝑊83 𝑋𝑜 3 = 1 + 𝑗 + (− −𝑗 )
2 2
𝑋 4 = 𝑋𝑒 0 − 𝑊80 𝑋𝑜 0 = 2 − 1 = 1
1
1 1
𝑋 5 = 𝑋𝑒 1 − 𝑊8 𝑋𝑜 1 = 1 − 𝑗 − ( − 𝑗 )
2 2
𝑋 6 = 𝑋𝑒 2 − 𝑊82 𝑋𝑜 2 = 𝑗

1 1
𝑋 7 = 𝑋𝑒 3 − 𝑊83 𝑋𝑜 3 = 1 + 𝑗 − (− −𝑗 )
2 2
32 Lecture materials on "Discrete Fourier Transform" By- Mohammed abdul kader, Assistant Professor, EEE, IIUC