Anda di halaman 1dari 35

# UNIT-II

## 1. DISCRETE FOURIER TRANSFORM

Computation of DFT
Computation of IDFT
Relation b/n DTFT & DFT
Properties of Twiddle Factor

2. PROPERTIES OF DFT
Linear Property
Periodic Property
Time Shifting Property
Frequency Shifting Property
Time Reversal property
Conjugate Property
Parsevalls Theorem
Time Convolution Theorem
Frequency Convolution Theorem

3. CONVOLUTION
Linear Convolution
Circular Convolution
Linear Convolution through Circular Convolution
Response of discrete LTI system through circular convolution
Circular Convolution through DFT & IDFT
Linear Convolution through DFT & IDFT

## 4. FAST FOURIER TRANSFORM

Introduction
Comparison b/n DFT & FFT
Comparison b/n DIT and DIF
Inverse FFT

5. DESCRIPTIVE QUESTIONS

6. OBJECTIVE QUESTIONS

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -1-
DISCRETE FOURIER TRANSFORM:

## Fourier series representation of finite duration sequence is known as Discrete Fourier

Transform (DFT). It is used to obtain discrete frequency domain representation of a given
discrete time domain representation.

COMPUTATION OF DFT:
Discrete frequency domain representation of a given finite duration sequence x(n) of length N
over the range 0 ≤ n ≤ N–1 can be defined as
N−1 nk
DFT[ x(n) ] = X(k) = ∑ x(n) W
n =0 N
Where,
X(k) is Discrete frequency domain of x(n) and range of k is 0 ≤ k ≤ N–1.
2 π nk
nk -j
W =e N
is Phase Factor or Twidle Factor or Complex Quantity.
N
N is duration of x(n) or X(k)

COMPUTATION OF IDFT:
Inverse Discrete Fourier Transform of X(k) of length N over the range 0 ≤ k ≤ N–1 can be
defined as
1 N−1 - nk
IDFT[ X( k) ] = x(n) = ∑ X(k) W
N k =0 N
RELATION BETWEEN DTFT & DFT:
5From basic definition of DFT is
N−1 nk
DFT[ x(n) ] = ∑ x(n) W
n =0 N
N−1 2 π nk
-j
= ∑ x(n) e N

n =0

N−1  2π k 
- j n
= ∑ x(n) e  N 

n =0

= DTFT [ x(n) ]
DFT and DTFT are equal, if 2π k / N = ω

## PROPERTIES OF PHASE FACTOR:

0
Property 1 : W N =1
N
Property 2 : W N =1
nk ±N nk
Property 3 :W =W
N N
n1k n k (n + n 2 )k
Property 4 :W W 2 =W 1
N N N
nk
Property 5 : W =1
N

PROPERTIES OF DFT :

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -2-
LINEAR PROPERTY :
Let x1(n), x2(n) are two finite duration sequences, with a equal duration of N samples and
DFT[ x1(n) ] = X1(k), DFT[ x2(n) ] = X2(k), then according to linear property of DFT,
DFT[ a x1(n) + b x2(n) ] = a X1(k) + b X2(k). Where a & b are arbitrary constants
PROOF:
From basic definition of DFT
N−1 nk
DFT[ x(n) ] = ∑ x(n) W
n =0 N
Replace x(n) by a x1(n) + b x2(n)

N−1 nk
DFT[ a x1 (n) + b x 2 (n) ] = ∑ [ a x 1(n) + b x 2 (n) ] W
n= 0 N
N− 1
 nk nk
= ∑ a x 1(n) W + b x 2 (n) W 
n= 0  N N
N− 1
 nk  N−1  nk 
= ∑ a x 1(n) W  + ∑  x 2 (n) W 
n= 0  N  n= 0  N
N−1 nk N−1 nk
= a ∑ x1(n) W + b ∑ x 2 (n) W
n= 0 N n= 0 N
= a DFT[ x1 (n) ] + b DFT[ x 2 (n) ]
= a X1(k) + b X 2 (k)

PERIODIC PROPERTY :
Let x(n) be a finite duration sequence, with a duration of N samples and DFT[ x(n) ] = X(k), then
according to periodic property of DFT,
(a) X(N + k) = X(k)
(b) x(N + n) = x(n)
PROOF:
From basic definition of DFT
N−1 nk
DFT[ x(n) ] = X(k) = ∑ x(n) W
n =0 N
Replace k by N + k
N−1 n(N + k)
X(N + k) = ∑ x(n) W
n =0 N
N−1 (nN + nk)
= ∑ x(n) W
n =0 N
N−1 nN nk
= ∑ x(n) W W
n =0 N N
N−1 nk
= ∑ x(n) (1) W
n =0 N
N−1 nk
= ∑ x(n) W
n =0 N
= X(k)

## TIME SHIFTING PROPERTY:

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -3-
Let x(n) be a finite duration sequence, with a duration of N samples and DFT[ x(n) ] = X(k), then
n 0k
according to time shifting property of DFT, DFT[ x(n – n0) ] = W X(k) . It is also known as
N
circular time shift property.

PROOF:
From basic definition of DFT
N−1 nk
DFT[ x(n) ] = ∑ x(n) W
n =0 N
Replace k by x(n) by x(n – n0)
N−1 nk
DFT[ x(n - n 0 ) ] = ∑ x(n - n 0 ) W , Let n - n 0 = m
n =0 N
N−1-n0
(n 0 + m)k
= ∑
m = − n0
x(m) W
N
N−1 n 0 k + mk
= ∑ x(m) W
m =0 N
N−1 n0k mk
= ∑ x(m) W W
m =0 N N
n 0 k N−1 mk
=W ∑
N m =0
x(m) W
N
n k
= W 0 X(k)
N
FREQUENCY SHIFTING PROPERTY:
Let x(n) be a finite duration sequence, with a duration of N samples and DFT[ x(n) ] = X(k), then
- nk 0
according to time shifting property of DFT, DFT[ W x(n) ] = X(k – k0). It is also known as
N
modulation theorem.
PROOF:
From basic definition of DFT
N−1 nk
DFT[ x(n) ] = ∑ x(n) W
n =0 N
- nk 0
Replace x(n) by W x(n)
N
- nk 0 N−1 - nk 0 nk
DFT[ W x(n) ] = ∑ W x(n) W
N n =0 N N
N−1 nk - nk 0
= ∑ x(n) W W
n=0 N N
N−1 n(k - k 0 )
= ∑ x(n) W
n=0 N
= DFT[ x(n) ] at k = k − k 0 .
= X(k − k 0 )

## TIME REVERSAL PROPERTY:

Let x(n) be a finite duration sequence, with a duration of N samples and DFT[ x(n) ] = X(k), then
according to time reversal property of DFT, DFT[ x(N – n) ] = X(N – k).
PROOF:

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -4-
From basic definition of DFT
N−1 nk
DFT[ x(n) ] = ∑ x(n) W
n =0 N
Replace x(n) by x(N – n)
N−1 nk
DFT[ x(N - n) ] = ∑ x(N - n) W , Let N − n = m
n =0 N
1 (N - m)k
= ∑ x(m) W
m =N N
1 Nk - mk
= ∑ x(m) W W
m =N N N
1 - mk
= ∑ x(m) (1) W
m =N N
1 - mk
= ∑ x(m) W (1)
m =N N
N−1 - mk mN
= ∑ x(m) W W
m =0 N N
N−1 m(N - k)
= ∑ x(m) W
m =0 N
= DFT[ x(n) ] at k = N - k.
= X(N - k)

CONJUGATE PROPERTY:
Let x(n) be a finite duration sequence, with a duration of N samples and DFT[ x(n) ] = X(k), then
according to conjugate property of DFT,
(a) DFT[ x*(n) ] = X*(N – k).
(b) DFT[ x*(N-n) ] = X*(N + k).
PROOF:
(a) From basic definition of DFT
N−1 nk
DFT[ x(n) ] = ∑ x(n) W
n =0 N
Replace x(n) by x*(n)
N −1 nk
DFT[ x * (n) ] = ∑ x * (n) W
n =0 N
*
N −1
 - nk 
=∑ x * (n) W
n =0  N 
*
N −1
 - nk 
=∑ x(n) W N 
n =0  
*
N−1 - nk 
= ∑ x(n) W
n = 0 N 

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -5-
*
 N −1 - nk 
DFT[ x * (n) ] =  ∑ x(n) (1) W
n = 0 N 
*
 N −1 nN - nk 
=  ∑ x(n) (W ) W
n = 0 N N 
*
 N −1 n(N - k)
=  ∑ x(n) W
n = 0 N 
= [ DFT of x(n) at k = N − k ] *
= [ X(N - k) ] *
= X * (N - k)

## (b) From basic definition of DFT

N−1 nk
DFT[ x(n) ] = ∑ x(n) W
n =0 N
Replace x(n) by x*(N-n)
N−1 nk
DFT[ x * (N - n) ] = ∑ x * (N - n) W , Let N − n = m
n =0 N
1 (N - m)k
= ∑ x * (m) W
m =N N
N Nk - mk
= ∑ x * (m) W W
m =1 N N
N−1 - mk
= ∑ x * (m) (1) W
m =0 N
*
1
 mk 
= ∑ x * (m) W
N 
m =N 
*
N−1
 mk 
=∑ x(m) W N 
m =0  
*
N−1 nk 
= ∑ x(n) W 
n=0 N
*
N−1 nk 
= ∑ x(n) (1) W 
n=0 N
*
N−1 nN nk 
= ∑ x(n) (W )W 
n=0 N N
*
N−1 n(N + k) 
= ∑ x(n) W
n=0 N 
= [ DFT of x(n) at k = N + k ] *
= [ X(N + k) ] *
= X * (N + k)

PARSEVALLS THEOREM:
Let x(n) be a finite duration sequence, with a duration of N samples and N-Point
DFT[ x(n) ] = X(k), then Parsevalls theorem provides the relation between x(n) and its frequency
domain X(k) as
B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -6-
N−1 N−1
1
∑ ∑
2 2
x(n) = X(k)
n =0 N k =0
PROOF:
N −1

2
LHS = x(n)
n= 0
N− 1
= ∑ x(n)[ x(n)]
*

n= 0
N− 1
= ∑ x(n)[IDFT[ X(k) ] ]
*

n= 0
*
N− 1
 1 N− 1 - nk
= ∑ x(n)  ∑ X(k) W
n= 0  N k =0 N 
N− 1
1 N −1 nk
= ∑ x(n) ∑ X * (k) W
n= 0 N k =0 N
Change the order of two sums
1 N− 1 N −1 nk
= ∑ X * (k) ∑ x(n)W
N k =0 n= 0 N
1 N− 1
= ∑ X * (k) DFT[ x(n)]
N k =0
1 N− 1
= ∑ X * (k) X(k)
N k =0
1 N− 1
= ∑
N k =0
X(k) 2

= RHS

## TIME CONVOLUTION THEOREM :

Let x1(n), x2(n) are two finite duration sequences, with a equal duration of N samples and
DFT[ x1(n) ] = X1(k), DFT[ x2(n) ] = X2(k), then according to time convolution theorem,
DFT[ x1(n) ⊗ x2(n) ] = X1(k) X2(k). i.e "Convolution in time domain leads to multiplication in
frequency domain"
PROOF:
From basic definition of DFT
N−1 nk
DFT[ x(n) ] = ∑ x(n) W
n =0 N
Replace x(n) by x1(n) ⊗ x2(n)
N−1 nk
DFT[ x 1 (n) ⊗ x 2 (n)] = ∑ [ x 1 (n) ⊗ x 2 (n)] W
n=0 N
N−1 N−1 nk
= ∑ ∑ [ x 1 (m) x 2 (n - m)] W
n=0 m=0 N
Change the order of summation

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -7-
N−1  nk 
 N−1
DFT[ x 1 (n) ⊗ x 2 (n)] = ∑ x 1 (m)  ∑ x 2 (n - m)] W 
m =0   n=0 N 
N−1
= ∑ [ x 1 (m) ( DFT(x 2 (n - m) ) ]
m =0
N−1 
 mk 
= ∑ x 1 (m)  W X 2 (K) 
m =0   N 
N−1
 mk 
= X 2 (K) ∑ x 1 (m) W
m =0  N 
= X 2 (K) X1 (K)
= X1 (K)X 2 (K)

## FREQUENCY CONVOLUTION THEOREM :

Let x1(n), x2(n) are two finite duration sequences, with a equal duration of N samples and
DFT[ x1(n) ] = X1(k), DFT[ x2(n) ] = X2(k), then according to frequency convolution theorem, DFT[
x1(n) x2(n) ] = [ X1(k) ⊗ X2(k) ] / N. i.e "Convolution in frequency domain leads to multiplication in
time domain"
PROOF:
From basic definition of IDFT
1 N−1 - nk
IDFT[ X( k) ] = ∑ X(k) W
N k =0 N
Replace X(k) by X1(k) ⊗ X2(k)
1 N−1 - nk
IDFT[ X1( k) ⊗ X 2 ( k) ] = ∑ [ X1 ( k) ⊗ X 2 ( k) ] W
N k =0 N
1 N−1  N−1  - nk
= ∑  ∑ [ X1 ( m)X 2 ( k - m) ]  W
N k =0  m=0  N
Change the order of two sums
1 N−1  N−1 - nk 
IDFT[ X1 ( k) ⊗ X 2 ( k) ] = ∑  X1 ( m) ∑ X 2 ( k - m) W 
N m =0  k =0 N 
1 N−1   1 N−1 - nk  
= ∑ 
N m=0 
X 1 ( m) N 
 ∑ X 2 ( k - m) W 
N  
 N k =0
1 N−1
= ∑ ( X1( m) N (IDFT(X 2 (k − m)) )
N m =0
1 N−1   - mn 
= ∑  X1 ( m) N  W
N m=0  N
x 2 (n)  
 
1 N−1  - mn 
= N x 2 (n) ∑  X1 ( m) W
N m=0  N


= N x 1 (n) x 2 (n)

X1 (k) ⊗ X 2 (k)
DFT [ x 1 (n) x 2 ( n) ] =
N

CONVOLUTION :

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -8-
Convolution is a special operation like addition (+), Subtraction (-), multiplication (x) and it may
be operated between two or more signals or sequences. Convolution includes Folding, Shifting,
Multiplication and Summation and is represented by * or ⊗. It is used to obtain the response of
discrete LTI system having input x(n) and impulse response h(n). Convoluted sequence or
response [ y(n) ] of the system can be defined as

y(n) = ∑x(m)
m =−

h(n − m)

## x(n) Impulse Response h(n) y(n)

Where
x(n) : Input Sequence
h(n) : Impulse Response
y(n) : Convoluted sequence or Response of the system

Convolution is also used to obtain the response of cascaded system having individual system
impulse responses h1(n), h2(n), h3(n)………,
y(n) = x(n) ⊗ h1(n) ⊗ h2(n) ⊗ h3(n)………

## There are two types of convolution

Linear Convolution
Circular Convolution

LINEAR CONVOLUTION:
Linear convolution of a first sequence x1(n) having N1 samples [0 ≤ n ≤ N1 – 1] and a second
sequence x2(n) having N2 samples [0 ≤ n ≤ N2 – 1] can be defined as
N −1
x(n) = ∑x
m =0
1 (m) x 2 (n − m)

Where
x(n) : Linear convoluted sequence, with a duration of N = N1 + N2 – 1
N1 : Duration of x1(n), 0 ≤ n ≤ N1 – 1
N2 : Duration of x2(n), 0 ≤ n ≤ N2 – 1
N : Duration of linear convoluted sequence x(n),
Duration of linear convoluted sequence x(n) and x1(n) or x2(n) are different, therefore linear
convolution is also known as aperiodic convolution. DFT does not support linear convolution,
due to unequal durations.

## Procedure For Evaluating Linear Convolution:

The following four steps were required to compute linear convolution
1. Folding : Fold x2(m) about k=0, to obtain x2(-m)
2. Shifting : Shift the folded sequence x2(-m) by n units left and/or right,
to obtain x2(n-m) .
3. Multiplication : Multiply x1(m) and x2(n-m),
to obtain the product sequence x1(m) . x2(n-m),
4. Summation : Sum all the values of product sequence at every instant,
N −1
to obtain x(n) = ∑x
m =0
1 (m) x 2 (n − m)

CIRCULAR CONVOLUTION:

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No -9-
Circular convolution of a first sequence x1(n) having N samples [0 ≤ n ≤ N – 1] and a second
sequence x2(n) having N samples [0 ≤ n ≤ N – 1] can be defined as
N −1
x(n) = ∑x
m =0
1 (m) x 2 (n − m)

Where
x(n) : Circular convoluted sequence, with a duration of N = N – 1
N : Duration of x1(n) or x2(n) or x(n), 0 ≤ n ≤ N – 1
Durations of circular convoluted sequence x(n), first sequence x 1(n) and second sequence x2(n)
are equal, therefore circular convolution is also known as periodic convolution. DFT supports
circular convolution, due to equal durations.

## Procedure For Evaluating Circular Convolution:

The following four steps were required to compute circular convolution
1. Folding : Fold x2(m) about k=0 and take periodic extension,
to obtain x2(-m)
2. Shifting : Shift the folded sequence x2(-m) by n units left and/or right,
to obtain x2(n-m) .
3. Multiplication : Multiply x1(m) and x2(n-m),
to obtain the product sequence x1(m) . x2(n-m),
4. Summation : Sum all the values of product sequence at every instant,
N −1
to obtain x(n) = ∑x
m =0
1 (m) x 2 (n − m)

## Derivation for Circular Convolution:

Let x1(n) and x2(n) are two finite duration sequences with a equal duration of N samples,
assume x(n) be the circular convoluted sequence with a duration of N samples
x(n) = x1(n) ⊗ x2(n), convolution in time domain leads to multiplication in frequency domain. i.e
X(k) = X1(k) X2(k).
IDFT of X(k) can be defined as
1 N−1 - nk
IDFT[ X( k) ] = ∑ X(k) W
N k =0 N
Replace X(k) = X1(k) X2(k)
1 N− 1 -nk
ID F T[X ( k) ] = ∑ X1(k) X 2 (k)W
N k=0 N
1 N− 1 - nk
x(n)= ∑
N k=0
X1(k)] D FT[x2 (n)] W
N
1 N− 1   N− 1 m k - n k
= ∑
N k=0 
 1  ∑ x 2 (m ) W  W 
X (k)
N N
 m= 0
Change the order of two sums
1 N−1   mk - nk  
N−1
x(n) = ∑  x 2 (m) 
N m=0 
∑ X (k)W
N
1W 
N  
k =0

N−1  - (n - m)k  
 1 N−1
= ∑  x 2 (m)  ∑ X1 (k)W  
m=0   N k =0 N 
N−1 N−1
x(n) = ∑ [ x 2 (m) x 1 (n - m) ], or ∑ [ x 1 (m) x 2 (n - m) ]
m =0 m =0

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 10 -

Let x1(n) and x2(n) are two finite duration sequences, with a duration of N1and N2 samples, then
the duration of linear convoluted sequence N1 + N2 – 1. Following steps were required to
compute the linear convolution through circular convolution
1. Pad the first sequences x1(n) by "N2 – 1" number of zeroes, to obtain N1 + N2 – 1 length.
2. Pad the second sequences x2(n) by "N1 – 1" number of zeroes, to obtain N1+N2–1 length.
3. Now determine the circular convolution of x1(n) and x2(n) called linear convolution.

## RESPONSE OF DISCRETE LTI SYSTEM THROUGH CIRCULAR CONVOLUTION:

Response of the system can be defined as the Linear convolution of input x(n) and impulse
response h(n). Let N1 be the duration of input sequence x(n) and N2 be the duration of impulse
response h(n), then the duration of response of the system is N 1+N2–1. Following steps were
required to compute the response of the discrete LTI system through circular convolution
1. Pad the input sequences x(n) by "N2 – 1" number of zeroes, to obtain N1 + N2 – 1 length.
2. Pad the impulse response sequences h(n) by "N1 – 1" number of zeroes, to obtain N 1+N2–1
length.
3. Now determine the circular convolution of x(n) and h(n) called Response of the given
discrete LTI system.

## CIRCULAR CONVOLUTION THROUGH DFT & IDFT:

Let x1(n) and x2(n) are two finite duration sequences, with a equal duration of N samples, then
the duration of circular convoluted sequence is also N samples. Following steps were required
to compute the circular convolution through DFT approach.
1. Compute N –Point DFT of x1(n), i.e X1(k).
2. Compute N –Point DFT of x2(n), i.e is X2(k).
3. Now determine N samples of X(k) from X(k) = X1(k). X2(k).
4. Compute N –Point IDFT of X(k), i.e x(n) called circular convoluted sequence

## RESPONSE OF THE DISCRETE LTI SYSTEM THROUGH DFT & IDFT:

Response of the system can be defined as the Linear convolution of input x(n) and impulse
response h(n). Let N1 be the duration of input sequence x(n) and N2 be the duration of impulse
response h(n), then the duration of response of the system is N 1+N2–1. Following steps were
required to compute the response of the discrete LTI system through DFT approach
1. Pad the input sequences x(n) by "N2 – 1" number of zeroes, to obtain N1 + N2 – 1 length.
2. Pad the impulse response sequences h(n) by "N1 – 1" number of zeroes, to obtain N 1+N2–1
length.
3. Compute N –Point DFT of x(n), i.e X(k).
4. Compute N –Point DFT of h(n), i.e is H(k).
5. Now determine N samples of Y(k) from Y(k) = X(k). H(k).
6. Compute N –Point IDFT of Y(k), i.e y(n) called Response of the given discrete LTI system.

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 11 -

FAST FOURIER TRANSFORM (FFT):

INTRODUCTION:
Fast Fourier Transform is a method or algorithm is used to compute Discrete Fourier Transform
with reduced number of calculations (Complex additions & multipliers). From basic definition of
DFT of a N – point sequence x(n) is
N−1 nk
X(k) = ∑ x(n) W
n =0 N
0k 1k 2k (N - 1)k
X(k) = x(0) W + x(1) W + x(2) W +.......... .......... ..... + x(N - 1) W
N N N N
0 0 0 0
k = 0 ⇒X(0) = x(0) W + x(1) W + x(2) W +.......... .......... ..... + x(N - 1) W
N N N N
0 1 2 N-1
k =1 ⇒X(1) = x(0) W + x(1) W + x(2) W +.......... .......... ..... + x(N - 1) W
N N N N
0 2 4 (N - 1)2
k = 2 ⇒X(2) = x(0) W + x(1) W + x(2) W +.......... .......... ..... + x(N - 1) W
N N N N
.
.
.
0 N-1 2(N - 1) (N - 1)(N - 1)
k = N −1 ⇒X(N - 1) = x(0) W + x(1) W + x(2) W +.......... .. + x(N - 1) W
N N N N
It is very clear from above N-point DFT computation, for each line requires "N" number of
complex multiplications and "N – 1" number of complex additions. Such lines are N, therefore
for the complete computation of N-point DFT
• Number of Complex additions used = N(N – 1).
• Number of Complex multiplications used = N2.

To reduce above number of calculations, FFT technique is used. In a FFT method, N can be
expressed as N = r m.
Where, r : Radix number (minimum value is 2).
m : Required number of stages to compute N-Point DFT
In a FFT method,
• Number of Complex additions used = N Log2 N.
• Number of Complex multiplications used = (N/2) Log2 N.

N – point DFT – FFT is a indirect method, where the total computation is divided into m number
of stages and all these m stages are cascaded to compute N-point DFT. The total computation
consists of
Two N/2-point DFT's
Four N/4-point DFT's
Eight N/8-point DFT's and so on.

The process of converting N-point DFT into smaller point DFT,s is known as decimation. The
decimation process may starts from time domain or from frequency domain. Based on the
decimation in time or in frequency, FFT algorithms are classified into two types.
1. Decimation in Time(DIT) Radix-2 FFT algorithm.
2. Decimation in Frequency(DIF) Radix-2 FFT algorithm.

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 12 -

Direct Computation of Indirect Computation of N-Point DFT
N-Point DFT (DIT Radix 2 – FFT / DIF Radix 2 – FFT)
No. of
Complex Complex No. of Complex Complex
points
N
Required Required Required Required Required
A=N(N–1) B=N2 m=Log2N a=N.m b=Nm/2
4 12 16 2 8 4
8 56 64 3 24 12
16 240 256 4 64 32
32 992 1,024 5 160 80
64 4,032 4,096 6 384 192
128 16,256 16,384 7 896 448
256 65,280 65,536 8 2,048 1,024
512 2,61,632 2,62,144 9 4,608 2,304
1024 10,47,552 10,48,576 10 10,240 5,120

## % of saving due to Speed of FFT due to

No. of
points Complex Complex Complex Complex
(1 – a / A) x 100 (1 – b / B ) x 100 A/a B/b
4 33.33% 75.00% 1.50 4.00
8 57.14% 81.25% 2.33 5.33
16 73.33% 87.50% 3.75 8.00
32 83.87% 92.19% 6.20 12.80
64 90.48% 95.31% 10.50 21.33
128 94.49% 97.27% 18.14 36.57
256 96.86% 98.44% 31.86 64.00
512 98.24% 99.12% 56.78 113.78
1024 99.02% 99.51% 102.3 204.8

From above two tables it can be observed that for a large value of N (if N increases),
• Number of calculations decreases.
• The percentage saving increases.
• Speed of FFT method increases.

## DIT RADIX – 2 FFT ALGORITHM:

In DIT Radix – 2 FFT, the time domain N-point sequence x(n) is decimated into possible
number of 2-point sequences. Now compute possible number of 4-point DFT's by combining

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 13 -

two 2-point DFT's. Then compute possible number of 8-point DFT's by combining two 4-point
DFT's . This process is continued until we get N-point DFT. Following steps were required to
implement the DIT-FFT algorithm

STEP1:
From basic definition of N-point DFT of a sequence x(n)
N−1
nk
X(k) = ∑ x(n) W
n =0 N
Separate even n and odd n by replacing n by 2n and n by 2n + 1
N−1
nk N−1 nk
= ∑ x(n) W + ∑ x(n) W
n =Even N n=Odd N
N− 2
nk N−1
nk
= ∑ x(n) W + ∑ x(n) W
n =0,2, 4.. N n=1,3,5.. N
N− 2
2nk N−1
(2n + 1)k
= ∑
2n =0,2, 4..
x(2n) W + ∑ x(2n + 1) W
N 2n +1=1,3,5.. N
N N
−1 −1
2 2nk 2 2nk k
= ∑
n =0,1,2..
x(2n) W
N
+ ∑
n =1,2,3..
x(2n + 1) W
N
W
N
N N
−1 −1
2 nk k 2 nk
=∑ x(2n) W +W ∑ x(2n + 1) W N/2
n =0 N/2 N n=0..

N k N
= − point DFT [ x(2n) ] + W − point DFT [ x(2n + 1) ]
2 N 2
k
X(k) = I1 (k) + W I2 (k)
N
Where I1 (k) and I2 k) are periodic w ith a period of N/2 samples
N
−1
2 nk
I1 (k) = ∑ x(2n) W Equation − 1, I1 (N/2 + k) = I1 (k)
n =0 N/2
N
−1
2 nk
I2 (k) = ∑ x(2n + 1) W N/2
n =0..
Equation − 2, I2 (N/2 + k) = I2 (k)

## Here we have to compute N samples of X(k)

EX: Take N=8, for this we have to compute 8 samples of X(k), are
{X(0), X(1), X(2), X(3), X(4), X(5), X(6), X(7) } over the range 0 ≤ k ≤ 7 and use
I1(4+k) = I1(k) ⇒ I1(4) = I1(0)
⇒ I1(5) = I1(1)
⇒ I1(6) = I1(2)
⇒ I1(7) = I1(3)
I2(4+k) = I2(k) ⇒ I2(4) = I2(0)
⇒ I2(5) = I2(1)
⇒ I2(6) = I2(2)
⇒ I2(7) = I2(3)

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 14 -

0
X(0) =I1 (0) +W I 2 (0)
8
1
X(1) =I1 (1) +W I 2 (1)
8
2
X(2) =I1 (2) +W I 2 (2)
8
3
X(3) =I1 (3) +W I 2 (3)
8
4 0
X(4) =I1 (4) +W I2 (4) =I1 (0) −W I2 (0)
8 8
5 1
X(5) =I1 (5) +W I 2 (5) =I1 (1) −W I2 (1)
8 8
6 2
X(6) =I1 (6) +W I 2 (6) =I1 (2) −W I2 (2)
8 8
7 3
X(7) =I1 (7) +W I 2 (7) =I1 (3) −W I2 (3)
8 8
Signal flow graph of the stage (Butterfly Structure):

I1(0) • • X(0)

I1(1) • • X(1)

I1(2) • • X(2)

I1(3) • • X(3)

I2(0) • • X(4)
0 –1
W 8
I2(1) • • X(5)
W18 –1

I2(2) • • X(6)
W28 –1

I2(3) • • X(7)
W38 –1

STEP2:
From equation number 1 or N/2-point DFT of a sequence x(2n), i.e
N
−1
2 nk
I1 (k) = ∑ x(2n) W
n =0 N/2
Replace n by 2n and 2n + 1
N N
−1 −1
4 2nk 4 (2n + 1)k
I1 (k) = ∑ x(2(2n)) W + ∑ x(2(2n + 1)) W
n =0 N/2 n =0 N/2
N N
−1 −1
4 2nk 4 2nk k
= ∑ x(4n) W + ∑ x(4n + 2)) W W
n =0 N/2 n =0 N/2 N/2

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 15 -

N N
−1 −1
4 nk k 4 nk
I1 (k) = ∑ x(4n) W +W ∑ x(4n + 2) W N/4
n =0 N/4 N/2 n=0..

N k N
= − point DFT [ x(4n) ] + W − point DFT [ x(4n + 2) ]
4 N/2 4
k
I1 (k) = I3 (k) + W I 4 (k)
N/2
Where I3 (k) and I4 (k) are periodic w ith a period of N/4 samples
N
−1
4 nk
I3 (k) = ∑ x(4n) W Equation − 3, I3 (N/4 + k) = I3 (k)
n =0 N/4
N
−1
4 nk
I4 (k) = ∑ x(4n + 2) W N/4
n=0..
Equation − 4, I 4 (N/4 + k) = I 4 (k)

## From equation number 2 or N/2-point DFT of a sequence x(2n+1), i.e

N
−1
2 nk
I2 (k) = ∑ x(2n + 1) W
n =0 N/2
Replace n by 2n and 2n + 1
N N
−1 −1
4 2nk 4 (2n + 1)k
I2 (k) = ∑ x(2(2n) + 1) W + ∑ x(2(2n + 1) + 1) W
n =0 N/2 n =0 N/2
N N
−1 −1
4 2nk 4 2nk k
= ∑ x(4n + 1) W + ∑ x(4n + 3) W W
n =0 N/2 n =0 N/2 N/2

N N
−1 −1
4 nk k 4 nk
I2 (k) = ∑ x(4n + 1) W +W ∑ x(4n + 3) W N/4
n =0 N/4 N/2 n =0..

N k N
= − point DFT [ x(4n + 1) ] + W − point DFT [ x(4n + 3) ]
4 N/2 4
k
I2 (k) = I5 (k) + W I6 (k)
N/2
Where I5 (k) and I6 (k) are periodic w ith a period of N/4 samples
N
−1
4 nk
I5 (k) = ∑ x(4n + 1) W Equation − 5, I5 (N/4 + k) = I5 (k)
n =0 N/4
N
−1
4 nk
I6 (k) = ∑ x(4n + 3) W N/4
n =0..
Equation − 6, I6 (N/4 + k) = I6 (k)

## Here we have to compute N/2 samples of I2 (k)

EX: Take N=8, for this we have to compute 4 samples of I 1(k) and 4 samples of I2(k) are {I1(0),
I1(1), I1(2), I1(3), I2(0), I2(1), I2(2), I2(3) } over the range 0 ≤ k ≤ 3 and use
I3(2+k) = I3(k) ⇒ I3(2) = I3(0)

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 16 -

⇒ I3(3) = I3(1)
I4(2+k) = I4(k) ⇒ I4(2) = I4(0)
⇒ I4(3) = I4(1)
I5(2+k) = I5(k) ⇒ I5(2) = I5(0)
⇒ I5(3) = I5(1)
I6(2+k) = I6(k) ⇒ I6(2) = I6(0)
⇒ I6(3) = I6(1)

0
I1 (0) =I3 (0) +W I 4 (0)
4
1
I1 (1) =I3 (1) +W I 4 (1)
4
2 0
I1 (2) =I3 (2) +W I 4 (2) =I3 (0) −W I 4 (0)
4 4
3 1
I1 (3) =I3 (3) +W I 4 (3) =I3 (1) −W I 4 (1)
4 4
0
I2 (0) =I5 (0) +W I6 (0)
4
1
I2 (1) =I5 (1) +W I6 (1)
4
2 0
I2 (2) =I5 (2) +W I6 (2) =I5 (0) −W I6 (0)
4 4
3 1
I2 (3) =I5 (3) +W I6 (3) =I5 (1) −W I6 (1)
4 4
Signal flow graph of the stage (Butterfly Structure):

I3(0) I1(0)

I3(1) I1(1)

I4(0) I1(2)
W40 –1

I4(1) I1(3)
W41 –1

I5(0) I2(0)

I5(1) I2(1)
I6(2) I2(2)
W40 –1

I6(3) I2(3)
W41 –1

STEP 3:
From equations 3, 4, 5, 6 and take N = 8

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 17 -

N
−1
4 nk
I3 (k) = ∑ x(4n) W
n =0 N/4
1
nk k
= ∑ x(4n) W = x (0) + W x ( 4)
n =0 2 2
0 1 0
I3 (0) = x (0) + W x ( 4) and I3 (1) = x (0) + W x ( 4) = x (0) − W x ( 4)
2 2 2
N
−1
4 nk
I4 (k) = ∑ x(4n + 2) W N/4
n =0..
1
nk k
= ∑ x(4n + 2) W = x ( 2) + W
x( 6)
n =0 2 2
0 1 0
I 4 (0) = x ( 2) + W x (6) and I4 (1) = x (2) + W x (6) = x ( 2) − W x ( 6)
2 2 2
N
−1
4 nk
I5 (k) = ∑ x(4n + 1) W
n =0 N/4
1
nk k
= ∑ x(4n + 1) W = x(1) + W x (5)
n =0 2 2
0 1 0
I5 (0) = x (1) + W x (5) and I5 (1) = x (1) + W x(5) = x (1) − W x (5)
2 2 2
N
−1
4 nk
I6 (k) = ∑ x(4n + 3) W N/4
n =0..
1
nk k
= ∑ x(4n + 3) W = x (3 ) + W x (7)
n =0 2 2
0 1 0
I6 (0) = x (3) + W x (7) and I6 (1) = x(3) + W x (7) = x (3) − W x (7)
2 2 2
Signal flow graph of the stage (Butterfly Structure):

x(0) I1(0)

x(4) I1(1)
0 –1
W 2
x(2) I1(2)

x(6) I1(3)
0 –1
W 2
x(1) I2(0)

## x(5) Combine I2(1)

2-Point
W20 –1
two
x(3) DFT 2-Point I2(2)
DFT's,
x(7) I2(3)
2-Point
W20 –1 to get Combine
DFT 4-Point DFT two
Three Stage Computational Structure for Radix 2 DIT – FFT:4-Point
Combine DFT's,
2-Point two
x(0) DFT 2-Point X(0)
DFT's, to get
8-Point DFT
B.Ramesh2-Point to getDept. of ECE, LBRCE, Mylavaram. Page No
Reddy, Associate professor, - 18 -
DFT 4-Point DFT
x(4) X(1)

x(2) X(2)

x(6) X(3)

x(1) X(4)

x(5) X(5)

x(3) X(6)

x(7) X(7)

## Normal Order & Bit Reversed Order:

Given sequence x(n) contains samples x(0), x(1), x(2), x(3), x(4), x(5), x(6) and x(7), but here
x(n) is decimated into four 2-point DFT's consists of x(0) & x(4), x(2) & x (6), x(1) & x(5), x(3) &
x(7). This order is different from actual order is known as bit reversed order, the order of X(k) is
known as normal order.

## Frequency Domain [ X(k) ] Time Domain [ x(n) ]

Normal Order Bit Reversed Order
X(0) = X(000) x(000) = x(0)
X(1) = X(001) x(100) = x(4)
X(2) = X(010) x(010) = x(2)
X(3) = X(011) x(110) = x(6)
X(4) = X(100) x(001) = x(1)
X(5) = X(101) x(101) = x(5)
X(6) = X(110) x(011) = x(3)
X(7) = X(111) x(111) = x(7)

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 19 -

Signal Flow Graph of 8-point Radix-2 DIT-FFT (Butterfly Structure)

x(0) X(0)

x(4) X(1)
W20 –1

x(2) X(2)
0 –1
W 4
x(6) X(3)
0 –1 1 –1
W2 W 4
x(1) X(4)
0 –1
W8
x(5) X(5)
0 –1 1 –1
W 2 W8
x(3) X(6)
0 –1 2 –1
W4 W8
x(7) X(7)
W20 –1
W41 –1
W83 –1

## Stage 1 Stage 2 Stage 3 Total

Number of Complex Adders 8 8 8 24
Number of Complex Multipliers 4 4 4 16
Number of Butterflies 4 2 1 7
Different Twiddle Factors W20 W40, W41 W80, W81, W82, W83 7

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 20 -

DIF RADIX – 2 FFT ALGORITHM:
In DIF Radix – 2 FFT, the frequency domain N-point sequence X(k) is decimated into possible
number of 2-point sequences. In this algorithm N-point time domain sequence is converted into
two N/2 point sequences. Then each N/2 point sequence is converted into two N/4 point
sequences. This process is continued until we get required number of 2-point sequences.
Following steps were required to implement the DIF-FFT algorithm

STEP1:
From basic definition of N-point DFT of a sequence x(n)
N −1
nk
X(k) = ∑ x(n) W
n=0 N
Separatethe totalrange of N samplesinto 0 to N/2 - 1 and N/2 to N - 1.
N/2− 1
nk N −1 nk
= ∑ x(n) W + ∑ x(n) W
n=0 N n = N/2 N
N/2− 1
nk N −1 nk
= ∑ x(n) W + ∑ x(n) W
n=0 N n = N/2 N
N/2− 1 nk N − 1 nk
= ∑ x(n) W + ∑ x(n) W , and Let n − N/2 = m
n=0 N n-N/2= 0 N
N/2− 1 nk N/2−1 (N/2 + m)k
= ∑ x(n) W + ∑ x(N/2+ m) W
n=0 N m=0 N
N/2− 1
nk N/2− 1
Nk/2 mk
= ∑ x(n) W + ∑ x(N/2+ m) W W
n=0 N m=0 N N
N/2− 1
nk N/2−1 k mk
= ∑ x(n) W + ∑ x(N/2+ m) W W
n=0 N m=0 2 N
Replacethe dummy variablem by n.
N/2− 1
nk N/2−1 k nk
= ∑ x(n) W + ∑ x(N/2+ n ) W W
n=0 N n= 0 2 N
N/2− 1
k nk
X(k) = ∑ [ x(n)+ W x(N/2+ n) ] W
n=0 2 N

## Replace k by 2k for even location

N/2− 1 2k n2k
X(2k)= ∑
n= 0
[ x(n)+ W
2
x(N/2+ n) ] W
N
N/2− 1
nk
= ∑
n= 0
[ x(n)+ x(N/2+ n) ] W
N/2
N/2− 1
nk
X(2k)= ∑ g1(n) W - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -- → ( 1)
n= 0 N/2
N
= − pointDFT of g1(n)
2
Where g1(n) = x(n) + x(N/2 + n)

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 21 -

Replace k by 2k +1 for odd location
N/2− 1
2k + 1 n(2k + 1)
X(2k+ 1) = ∑
n= 0
[ x(n)+ W
2
x(N/2+ n) ] W
N
N/2− 1
2k 1 2nk n
= ∑
n= 0
[ x(n)+ W
2
W
2
x(N/2+ n) ] W
N
W
N
N/2− 1
n nk
= ∑
n= 0
[ x(n)− x(N/2+ n) ] W
N
W
N/2
N/2− 1
nk
X(2k+ 1) = ∑ g2 (n) W - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -- → ( 2)
n= 0 N/2
N
= − pointDFT of g2 (n)
2
Where g2(n) = [ x(n) - x(N/2 + n) ] WNn

EX: Take N =8
For this example, we have to compute 8 samples from g1(n) and g2(n).
g1(n) = x(n) + x(4 + n) ⇒ g1(0) = x(0) + x(4)
⇒ g1(1) = x(1) + x(5)
⇒ g1(2) = x(2) + x(6)
⇒ g1(3) = x(1) + x(7)
g2(n) = [ x(n) - x(4 + n) ] W8 ⇒ g2(0) = [ x(0) - x(4) ] W80
n

## ⇒ g2(1) = [ x(1) - x(5) ] W81

⇒ g2(2) = [ x(2) - x(6) ] W82
⇒ g2(3) = [ x(3) - x(7) ] W83

x(0) g1(0)

x(1) g1(1)

x(2) g1(2)

x(3) g1(3)

x(4) g2(0)
0
-1 W 8
x(5) g2(1)
1
-1 W 8
x(6) g2(2)
2
-1 W 8
x(7) g2(3)
3
-1 W 8

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 22 -

STEP2: From equation 1
N−1
nk
X(2k)= ∑ g1(n) W
n= 0 N/2
Separatethe totalrangeof N/2 samplesinto 0 to N/4 - 1 andN/4 to N - 1.
N/4− 1
nk N/2− 1 nk
= ∑ g1(n) W + ∑ g1(n) W
n= 0 N/2 n = N/4 N/2
N/4− 1
nk N/2− 1
nk
= ∑ g1(n) W + ∑ g1(n) W
n= 0 N/2 n = N/4 N/2
N/4− 1 nk N/2−1 nk
= ∑ g1(n) W + ∑ g1(n) W , andLet n − N/4 = m
n= 0 N/2 n -N/4= 0 N/2
N/4− 1 nk N/4−1 (N/4 + m)k
= ∑ g1(n) W + ∑ g1(N/4 + m) W
n= 0 N/2 m = 0 N/2
N/4− 1
nk N/4−1 Nk/4 mk
= ∑ g1(n) W + ∑ g1(N/4 + m) W W
n= 0 N/2 m = 0 N/2 N/2
N/4− 1
nk N/4− 1
k mk
= ∑ g1(n) W + ∑ g1(N/4 + m) W W
n= 0 N/2 m = 0 2 N/2
Replacethe dummyvariablem by n.
N/4− 1
nk N/4−1 k nk
= ∑ g1(n) W + ∑ g1(N/4 + n) W W
n= 0 N/2 n = 0 2 N/2
N/4− 1
k nk
X(2k)= ∑ [ g1(n) + W g1(N/4 + n) ] W
n= 0 2 N/2
Replace k by 2k for even location
N/4 −1
2k n.2k
X(2.2k) = ∑
n=0
[ g1(n) + W
2
g1(N/4 + n) ] W
N/2
N/4 −1
nk
X(4k) = ∑
n=0
[ g1(n) + g1(N/4 + n) ] W
N/4
N/4 −1
nk
= ∑
n=0
g3 (n) W
N/4
− − − − − − − − − − − − − − − − − − − − − − − −− → (3)

N
X(4k) = − point DFT of g3 (n)
4
Where g3(n) = g1(n) + g1(N/4 + n)
Replace k by 2k+1 for odd location
N/4− 1
2k + 1 n(2k + 1)
X(2(2k+ 1)) = ∑
n= 0
[ g1(n) + W
2
g1(N/4 + n) ] W
N/2
N/4− 1
2k 1 2nk n
X(4k + 2) = ∑
n= 0
[ g1(n) + W
2
W g1(N/4 + n) ] W
2
W
N/2 N/2
N/4− 1
n nk
= ∑
n= 0
[ g1(n) − g1(N/4 + n) ] W W
N/2 N/4
N/4− 1
nk
= ∑
n= 0
g4 (n) W
N/4
− − − − − − − − − − − − − − − − − − − − − − − − → ( 4)

N
X(4k)= − pointDFT of g4 (n)
4
Where g4(n) = [ g1(n) - g1(N/4 + n) ] WN/2 n

From equation 2

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 23 -

N− 1
nk
X(2k+ 1) = ∑ g2 (n) W
n= 0 N/2
Separatethe totalrangeof N/2 sam plesinto 0 to N/4 - 1 andN/4 to N - 1.
N/4− 1
nk N/2− 1 nk
= ∑
n= 0
g2 (n) W + ∑ g2 (n) W
N/2 n = N /4 N/2
N/4− 1
nk N/2− 1 nk
= ∑
n= 0
+ ∑ g2 (n) W
g2 (n) W
N/2 n = N /4 N/2
N/4− 1 nk N /2− 1 nk
= ∑ g2 (n) W + ∑ g2 (n) W , andLet n − N/4 = m
n= 0 N/2 n -N/4= 0 N/2
N/4− 1
nk N /4− 1 (N/4+ m )k
= ∑ g2 (n) W + ∑ g2 (N/4+ m) W
n= 0 N/2 m = 0 N/2
N/4− 1
nk N /4− 1
Nk/4 m k
= ∑ g2 (n) W + ∑ g2 (N/4+ m) W W
n= 0 N/2 m = 0 N/2 N/2
N/4− 1
nk N /4− 1 k mk
= ∑ g2 (n) W + ∑ g2 (N/4+ m) W W
n= 0 N/2 m = 0 2 N/2
Replacethe dumm yvariablem by n.
N/4− 1
nk N /4− 1 k nk
= ∑ g2 (n) W + ∑ g2 (N/4+ n) W W
n= 0 N/2 n = 0 2 N/2
N/4− 1
k nk
X(2k+ 1) = ∑ [ g2 (n) + W g2 (N/4+ n) ] W
n= 0 2 N/2
Replace k by 2k for even location
N/4−1 2k n.2k
X(2.2k + 1) = ∑
n= 0
[ g2 (n) + W
2
g2 (N/4 + n) ] W
N/2
N/4−1 nk
X(4k + 1) = ∑
n= 0
[ g2 (n) + g2 (N/4 + n) ] W
N/4
N/4−1
nk
= ∑
n= 0
g5 (n) W
N/4
− − − − − − − − − − − − − − − − − − − − − − − − − −− → (5)

N
X(4k + 1) = − point DFT of g5 (n)
4
Where g5(n) = g2(n) + g2(N/4 + n)
Replace k by 2k+1 for odd location
N/4− 1
2k + 1 n(2k+ 1)
X(2(2k+ 1) + 1) = ∑
n= 0
[ g2 (n) + W
2
g2 (N/4+ n) ] W
N/2
N/4− 1
2k 1 2nk n
X(4k+ 3) = ∑
n= 0
[ g2 (n) + W W g2 (N /4+ n) ] W
2 2
W
N/2 N/2
N/4− 1
n nk
= ∑
n= 0
[ g2 (n) − g2 (N/4+ n) ] W W
N/2 N/4
N/4− 1
nk
= ∑
n= 0
g5 (n) W
N /4
− − − − − − − − − − − − − − − − − − − − − − − − → ( 6)

N
X(4k+ 3) = − pointDFTof g6 (n)
4
Where g6(n) = [ g2(n) – g2(N/4 + n) ] WN/2 n

EX: Take N = 8
Here we have to compute 8 samples from g3(n), g4(n), g5(n) and g6(n)
B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 24 -
g3(n) = g1(n) + g1(2 + n) ⇒ g3(0) = g1(0) + g1(2)
⇒ g3(1) = g1(1) + g1(3)
g4(n) = [g1(n) - g1(2 + n)]W4n ⇒ g4(0) = [ g1(0) - g1(2) ]W40
⇒ g4(1) = [ g1(1) - g1(3) ]W41
g4(n) = g1(n) + g1(2 + n) ⇒ g5(0) = g2(0) + g2(2)
⇒ g5(1) = g2(1) + g2(3)
g5(n) = [g1(n) - g1(2 + n)]W4n ⇒ g6(0) = [ g2(0) - g2(2) ]W40
⇒ g6(1) = [ g2(1) - g2(3) ]W41

## Signal Flow graph of Step –2 (Butterfly Structure) :

g1(0) g3(0)

g1(1) g3(1)

g1(2) g4(0)
0
-1 W 4
g1(3) g4(1)
1
-1 W 4
g2(0) g5(0)

g2(1) g5(1)

g2(2) g6(0)
0
-1 W 4
g2(3) g6(1)
1
-1 W 4

STEP 3:
This process will continue until we get 2-point DFT, but for this example N = 8 decimation
process is not required because of N/4 = 2 called "radix 2". Here we have to compute 8
samples of X(k) from X(4k), X(4k+2), X(4k+1) and X(4k+3).

From equation 3
1
nk
X(4k) = ∑ g3 (n) W
n=0 2
0k 1k
= g3 (0) W + g3 (1) W
2 2
k
= g3 (0) + W g3 (1)
2
⇒ X(0) = g3(0) + W20 g3(2) = g3(0) + g3(2)
⇒ X(4) = g3(0) + W21 g3(2) = [ g3(0) - g3(2) ] W20

From equation 4

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 25 -

1
nk
X(4k + 2) = ∑ g4 (n) W
n=0 2
0k 1k k
= g4 (0) W + g4 (1) W = g4 (0) + W g4 (1)
2 2 2
⇒ X(2) = g4(0) + W20 g4(2) = g4(0) + g4(2)
⇒ X(6) = g4(0) + W21 g4(2) = [ g4(0) - g4(2) ] W20
From equation 5
1
nk
X(4k + 1) = ∑ g5 (n) W
n= 0 2
0k 1k k
= g5 (0) W + g5 (1) W = g5 (0) + W g5 (1)
2 2 2
⇒ X(1) = g5(0) + W20 g5(2) = g5(0) + g5(2)
⇒ X(5) = g5(0) + W21 g5(2) = [ g5(0) - g5(2) ] W20
From equation 6
1
nk
X(4k + 3) = ∑ g6 (n) W
n=0 2
0k 1k k
= g6 (0) W + g6 (1) W = g6 (0) + W g6 (1)
2 2 2
⇒ X(3) = g6(0) + W20 g6(2) = g6(0) + g6(2)
⇒ X(7) = g6(0) + W21 g6(2) = [ g6(0) - g6(2) ] W20

g3(0) X(0)

g3(1) X(4)
0
-1 W 2
g4(0) X(2)

g4(1) X(6)
0
-1 W 2
g5(0) X(1)

g5(1) X(5)
0
-1 W 2
g6(0) X(3)

g6(1) X(7)
0
-1 W 2

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 27 -

Signal Flow Graph of 8-point Radix-2 DIF-FFT (Butterfly Structure)

x(0) X(0)

x(1) X(4)
-1
W20
x(2) X(2)
-1 0
W4
x(3) X(6)
-1 1 -1 0
W4 W2
x(4) X(1)
-1 0
W 8
x(5) X(5)
-1 1 -1 0
W 8 W2
x(6) X(3)
-1 2 -1 0
W 8 W4
x(7) X(7)
-1
W83 -1
W41 -1
W20

## Stage 1 Stage 2 Stage 3 Total

Number of Complex Adders 8 8 8 24
Number of Complex Multipliers 4 4 4 16
Number of Butterflies 1 2 4 7
Different Twiddle Factors W80, W81, W82, W83 W40, W41 W20 7

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 28 -

COMPARISON b/n DIT AND DIF METHODS:

1. The time domain sequence x(n) is The frequency domain sequence
decimated X(k) is decimated
2. The time domain sequence x(n) is The time domain sequence x(n) is
in bit reversed order in normal order order
3. The frequency domain sequence The frequency domain sequence
X(k) is in normal order X(k) is in bit reversed order
In each stage of computation, the In each stage of computation, the
4.
phase factors are multiplied before phase factors are multiplied after
5. Both the algorithms require same number of operations (Complex
additions & multiplications) to compute DFT
6. Both the algorithms require bit reversal at some place during
computation
7. For both the algorithms, the value of N should be expressed as N = 2m.
where m is no. of stages
For both the algorithms, required number of
8.
Complex additions = N Log2 N.
Complex multiplications = N/2. Log2 N.

INVERSE FFT:
Computation of IDFT of a sequence X(k) = x(n), through FFT known as inverse FFT.
From basic definition of IDFT of a sequence
1 N −1 - nk
x(n) = ∑
N k =0
X(k) W
N

1  N −1 

 - nk 
= ∑  X(k) W  
N  k =0  N  
 

1  N −1 nk 
=  ∑ X * (k) W 
N  k =0 N
1
= ( DFT (X * (k)) )∗
N
y(n)
=
N
Where y(n) = DFT[ X*(k) ]

PROCEDURE:
1. Take complex conjugate of given X(k), to get X*(k).
2. Compute N-point DFT of X*(k) using radix-2 DIT / DIF FFT method.
3. Take complex conjugate of above sequence. to get y(n).
4. Finally obtain x(n) using x(n) = y(n) / N.

USES OF FFT:
1. Single FFT algorithm is used to compute both DFT & IDFT.
2. It is used to obtain the response of discrete LTI system.

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 29 -

DESCRIPTIVE QUESTIONS:
1. Let X(k) denotes the N-point DFT of the N-point sequence x(n), show that with N even and if
x(n) = x(N – 1 – n) then X(N/2) = 0.

2. Let X(k) denotes the N-point DFT of the N-point sequence x(n), show that if x(n) satisfies the
relation x(n) = - x(N – 1 – n) then X(0) = 0.

## 3. Prove the following

(a) x*(n) → X*(( - k ))N RN(k)
(b) x*((-n))N RN(n) → ½ [ X((k))N + X*((-k))N ] RN(k)

## 4. Prove the following

(a) arg[ X(k) ] = - arg [ X(( - k ))N RN(k) ]
(b) Im[ X(k) ] = - Im [ X(( - k ))N RN(k) ]

5. Prove
(a) Re [ X(k) ] = Re [ X( - k ) ]
(b)  X(k) =  X( - k )

6. Let x(n) be a real valued sequence with N-points and let X(k) represent its DFT, with real
and imaginary parts denoted by XR(k) and XI(k) respectively. So that X(k) = XR(k) + j XI(k).
Now show that if x(n) real, XR(k) is even and XI(k) is odd.

## 7. Define DFT, IDFT. Give TOW properties of DFT.

8. Compute the DFT of the three point sequence x(n) = { 2, 1, 2}. Using the same sequence ,
compute the 6 point DFT and compare the two DFT's.

## 9. Define DFT of a sequence. Compute the N-point DFT of the sequence

 2ππr 
x(n) = Cos   , 0 ≤ n ≤ N −1 and 0 ≤ r ≤ N −1
 N 
10. Compute the Discrete Fourier Transform of each of the following finite length sequences
considered to be of length N.
(a) x(n) = δ (n)
(b) x(n) = δ (n – n0), where 0 < n0 < N
(c) x(n) = an, 0 ≤ n ≤ N – 1.

11. Compute the Discrete Fourier Transform of each of the following finite length sequences
considered to be of length N.
(a) x(n) = δ (n + n0), where 0 < n0 < N
(b) x(n) = α n, where 0 < α < 1.

12. Compute the DFT of x(n) = { - 1, 0, - 1} With T = 0.5. Plot the DFT sequence suggest a
method for improving frequency resolution.

## 13. Distinguish between DFT and DTFT.

14. Define DFT of a sequence x(n). Obtain the relationship between DFTS.

15. Consider a sequence x(n) = {2, -1, 1, 1, 1} and T = 0.5, compute its DFT and compare it with
DTFT.

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 30 -

16. Explain how DFT can be obtained by sampling DFS for a given sequence

17. State and prove circular time shifting and frequency shifting properties.

N−1 N−1
1
∑ ∑
2 2
18. Prove the Parsevalls relation for the DFT given by x(n) = X(k)
n =0 N k =0
with usual notation. Give its physical inter pretation.
19. If x(n) denotes a finite length sequence of length N. Show that x((-n))N = x((N-n))N

20. Prove that the convolution in time domain leads to multiplication in frequency domain for
discrete time signals.

21. What is " Padding with zeros " with an example, explain the effect of padding a sequence of
length N with L zeros or frequency resolution.

22. Let X(k) be the N-point DFT of the sequence x(n), for 0 ≤ n ≤ N – 1. Define a 2N-point
sequence y(n) as
x(n/2), neven
y(n) = 
 0, nodd
Express the 2N point of y(n) in terms of X(k)

23. Prove the following properties of the Discrete Fourier Series for periodic sequences.
Sequence Discrete Fourier Series
(a) x(n+m) WN-km X(k)
(b) x*(n) X*(-k)
(c) x*(-n) X*(k)
(d) Re[ x(n) ] Xe(k)
(e) j Im[ x(n) ] Xo(k)

24. Consider two periodic sequences x(n) and y(n), x(n) has period N and y(n) has period M.
The sequence w(n) is defined as w(n) = x(n) + y(n)
(a) Show that w(n) is periodic with period MN
(b) Determine W(k) in terms of X(k) and Y(k), where X(K), Y(k) and W(k) are the Discrete
Fourier Series coefficients with a period of N, M, MN respectively.

## 25. Compute the circular convolution of the sequences

x1(n) = {1, 2, 0, 1} and
x2(n) = {2, 2, 1, 1}, using DFT approach

 1, 0 ≤ n ≤ N - 1
26. Let x 1 (n) = x 2 (n) = 
0, otherwise
Find out the circular convolution of these sequences

27. If x1(n) and x2(n) are two finite sequence, derive the linear convolution of these sequence
using circular convolution.

28. Let x2(n) be a finite duration sequence of length N and x1(n) = δ (n – n0), where n0< N.
Obtain the circular convolution of two sequences.

## 29. Compute the circular convolution of the sequences

x1(n) = {1, 2, 3, 4} and
x2(n) = {5, 6, 7, 8},

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 31 -

30. Compute the linear convolution of the sequences
x1(n) = {1, 2, 3} and
x2(n) = {4, 5, 6, 7, 8},
31. Compute the linear convolution of the sequences
x1(n) = {1, 2, 3} and
x2(n) = {4, 5, 6, 7, 8}, through circular convolution
32. Determine the response of discrete LTI system, if
input sequence x(n) = {1, 2, 3, 4} and
impulse response h(n) = {5, 6, 7, 8, 9}, using
a. Linear convolution.
b. Circular convolution.
c. DFT approach
33. Give the steps involved in implementing radix –2, DIT FFT algorithm.
34. Compute the 16-point DFT of x(n) = Cos(n π /2), 0 ≤ n ≤ 15 using radix-2, DIT algorithm.
35. Develop a radix –2 , DIF / FFT algorithm for evaluating the DFT for N = 16 and hence
determine the 16-point DFT of the sequence
x(n)={0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1}.
36. Develop a radix –2 , DIF / FFT algorithm for an 8-point sequence.
37. Evaluate the 8-point DFT for the following sequence using DIT-FFT algorithm.
1, -3 ≤n ≤3
x(n) = 
0, otherwise
38. What is computational efficiency of FFT Radix-2 DIT algorithm for an N-point sequence?
Compare it with that of direct DFT computation?
39. A designer has available a number of 8-point FFT chips. Show explicitly how he should
connect three such chips in order to compute 24-point DFT.
40. Explain what do you understand by 'Bit reversal' and In-place computation.
41. What are DFT properties of the phase factor that are exploited to get the FFT.
42. Given a sequence x(n) = n for 0 ≤ n ≤ 7, find its frequency spectrum via FFT. How do you
improve the spectral resolution?
43. Out line steps in implementing DIF FFT algorithm for an N-point sequence.
44. Compare Decimation in time and decimation in frequency FFT computations.
45. Compute the FFT of x(n) ={1,2,3,4,4,3,2,1}, using DIT FFT algorithm.
46. Compare the computational requirements of DFT and FFT of a given N-point sequence.
47. Compute the 8-point DFT of the sequence x(n)={½, ½, ½, ½, 0, 0, 0, 0}, using in-place radix-
2 DIT FFT algorithm.
48. Implement the decimation in time FFT algorithm for N = 16, in this case how many non –
trivial multiplications are required.
49. Compute FFT of the sequence x(n) = {1, 0, 0, 0, 0, 0, 0, 0}.
50. Consider two real valued sequences x1(n) and x2(n) with DFTs X1(k) and X2(k) respectively.
Let g(n) = x1(n) + j x2(n) and G(k) its DFT. {When GOR(k), GER(k), GOI(k), GEI(k) denote the
odd part of the real, even part of the real, odd part of imaginary and even part of imaginary
respectively}. Determine X1(k) and X2(k).

## B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 32 -

OBJECTIVE QUESTIONS:
1. Let x(n) = 1, 0 ≤ n ≤ N – 1, and DFT[ x(n) ] = X(k), then X(0) is
(A) 1 (B) N (C) N – 1 (D) 2N.
2. Let X(k) = 1, 0 ≤ k≤ N – 1, and IDFT[ X(k) ] = x(n), then x(0) is
(A) 1 (B) N (C) N – 1 (D) 2N.
3. Let x(n) be a periodic sequence with period N, then what is the period of N point DFT of
x(n)(A) 1 (B) N (C) N – 1 (D) 2N.
4. Let x(n) be a discrete time domain real and periodic and DFT[ x(n) ] = X(k) = XR(k)+jXI(k),
(i). XR(k) is even function. (ii)XI(k) is odd function.
(iii). Magnitude of X(k) is even function. (iv) Phase of X(k) is odd function.
(A) i and ii (B) i, ii and iii (C) i, ii, iii and iv (D) None
5. Let x(n) be a discrete time sequence and N point DFT of x(n) is X(k), Then which of the
following relation is “Parsevals Theorem”.
N −1 2 N −1 2 N −1 2 N −1 2

(A) ∑x(n)
n =0
= ∑X(k)
k =0
(B) ∑x(n)
n =0
= N ∑X(k)
k =0
N −1 N −1 N −1 2 N −1 2
1 1
(C) ∑ x(n)
n =0
= ∑ X(k)
N k =0
(D) ∑x(n)
n =0
= ∑X(k)
N k =0
.
6. Let x(n) be a discrete time domain periodic sequence, then the N point DFT of x*(n) is
(A) X*(k) (B) X(k) (C) X*(N – k) (D) X*(k*).
7. Let x(n) be a discrete time domain periodic sequence, then the N point DFT of x(n – a) is
ak −ak ak
(A) W N X(k) (B) W N X(k) (C) W N X(k-a) (D) X(k-a).
8. Let x(n) be a discrete time domain periodic sequence, then the N point DFT of x(N –n) is
(A) X*(k) (B) X*(N – k) (C) X(N – k) (D) X(k).
9. Let x(n) be a discrete time domain periodic real sequence and
8 point DFT of x(n) = X(k), then which are complex conjugate pairs
(i). X(0) and X(7) (ii) X(1) and X(6) (iii) X(2) and X(5) (iv) X(3) and X(4) r.
(A) i and ii (B) i, ii and iii (C) i, ii, iii and iv (D) None
10. Which of the following relation is correct with respect to DFT.
2π k 2k Nk 2π N
(A) ω = (B) ω = (C) ω = (D) ω = .
N N 2 k
11. The complex quantity W1 value is
8
1 −i 1 +i 1 +i 1 −i
(A) (B) (C) (D) .
2 2 2 2
12. Appending zeros to a sequence in order to increase the size or length of the sequence is
called

## 13. The complex quantity W7 value is

8
1 −i 1 +i 1 +i 1 −i
(A) (B) (C) (D) .
2 2 2 2
14. Magnitude of phase factor W614 is
(A) 0 (B) 1 (C) 64 (D) -64.

## If x(n) = {1, 2, 3, 4} and DFT[ x(n) ] = X(k), then answer Q.No. 15 to 19

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 33 -
15. X(0) =
(A) 10 (B) 8 (C) 12 (D) None
16. X(4) =
(A) 10 (B) 8 (C) 12 (D) None
17. X(1) =
(A) 10 (B) 8 (C) 12 (D) None
18. X(2) =
(A) 10 (B) 8 (C) 12 (D) None
19. X(3) =
(A) 10 (B) 8 (C) 12 (D) None
20. Let X(k) denotes the N-point DFT of the N-point sequence x(n), if N even and x(n) = x(N – 1
– n) then X(N/2) is.
(A) N (B) N/2 (C) 0 (D) None
21. Let X(k) denotes the N-point DFT of the N-point sequence x(n), if x(n) satisfies the relation
x(n) = - x(N – 1 – n) then X(0) is
(A) N (B) N/2 (C) 0 (D) None
22. Convolution property is
(A) Commutative (B) Associative (C) Distributive (D) All
23. Which of the following method is used to compute response of a discrete LTI system
(A) Circular convolution method (B) Linear convolution method
(C) DFT – IDFT method (D) All
24. Response of a discrete LTI system is exactly equal to
(A). Circular convolution of input sequence and impulse response
(B). Linear convolution input sequence and impulse response
(C) Both A and B (D) None
25. Which of the following is also known as periodic convolution
(A) Circular convolution (B) Linear convolution (C) Both A and B (D) None
26. DFT support only
(A) Circular convolution (B) Linear convolution(C) Both A and B (D) None
27. What is the duration of linear convoluted sequence x(n) = x1(n) * x2(n), if
N1 is the duration of x1(n) and N2 is the duration of x2(n)
(A) N = N1 + N2 +1 (B) N = N1 + N2 - 1 (C) N = N1 + N2 (D) N = N1 + N2
28. What is the duration of linear convoluted sequence x(n) = x(n) * x(n), if N is the duration of
x(n).
(A) N (B) 2N (C) 2N – 1 (D) N – 1
29. What is the duration of circular convoluted sequence x(n) = x1(n) * x2(n), if
N is the common duration of x1(n) and x2(n).
(A) N (B) 2N (C) 2N – 1 (D) N – 1
30. What is the circular convoluted sequence x(n) = x1(n) * x2(n), if x1(n) = x2(n)=1, over the
range 0 ≤ n ≤ 3.
(A) x(n)={4,4,4,4} (B) x(n)={3,3,3,3} (C) x(n)={2,2,2,2} (D)x(n)={1,1,1,1}
31. What is the linear convoluted sequence x(n) = x1(n) * x2(n), if
x1(n) = 1, over the range 0 ≤ n ≤ 1.and x2(n) = 1, over the range 0 ≤ n ≤ 2
(A) x(n)={1,2,1,2} (B) x(n)={1,2,2,1} (C) x(n)={2,2,2,2} (D)x(n)={1,1,1,1}
32. What is the linear convoluted sequence x(n) = x1(n) * x2(n), if
x1(n) = 1, over the range 0 ≤ n ≤ 1.and x2(n) = 1, over the range 0 ≤ n ≤ 2
(A) x(n)={1,2,1,2} (B) x(n)={1,2,2,1} (C) x(n)={2,2,2,2} (D)x(n)={1,1,1,1}
33. Let y(n) be the response of discrete LTI system having input x(n) with a duration of N1
samples start at n1 and impulse response h(n) with a duration of N2 samples start at n2, then
(A) Length of y(n) is N = N1 + N2 -1 (B) y(n) will start at n1 + n2
(C) y(n) will end at n1 + n2 + N1 + N2 -2 (D) All

## 34. Which of the following method is useful to compute circular method

B.Ramesh Reddy, Associate professor, Dept. of ECE, LBRCE, Mylavaram. Page No - 34 -
(A) Circular (B) Matrix (C) Table (D) All
35. FFT is a method or algorithm is used to compute ___ with minimum number of calculations
(A) FT (B) DFT (C) DTFT (D) CTFT

36. The process of converting higher order point DFTs into lower order point DFTs called
(A) Decimation (B) Animation (C) Zero Padding (D) None.
37. How many number of stages required to compute the N-Point DFT of a sequence in radix –
r DIT / DIF FFT algorithm.
(A) logr N (B) logN r (C) N log (r ) (D) log (Nr).
38. What is the speed of FFT (Adders) over DFT for a 64-point sequence
(A) 10.5 (B) 11.5 (C) 12.5 (D) 13.5
39. What is the speed of FFT (Multipliers) over DFT for a 256-point sequence
(A) 8 (B) 16 (C) 32 (D) 64
40. How many number of stages required to compute 1024-Point DFT of a sequence in radix-2
DIT / DIF FFT algorithm.
(A) 1024 (B) 10 (C) 24 (D) 512
41. How many number of complex adders and complex multipliers required to compute 64-Point
DFT of a sequence in direct DFT
(A) 4032, 4096 (B) 4096, 4032 (C) 4032, 4032 (D) 4096, 4096
42. How many number of complex adders and complex multipliers required to compute 64-point
DFT of a sequence in radix-2 DIT / DIF FFT algorithm.
(A) 384, 192 (B) 384, 384 (C) 192, 192 (D) 192, 384
43. Which of the following radix number is possible, for DIT / DIF FFT algorithm.
(A) radix – 2 (B) radix – 3 (C) Both (D) None
44. What is the percentage saving due to additions in calculations of 64 point radix–2 FFT
(A) 90% (B) 90.5% (C) 91% (D) 91.5%
45. What is the percentage saving due to multipliers in calculations of 64 point radix–2 FFT
(A) 90.3% (B) 92.6% (C) 95.3% (D) 98.6%
46. What are the phase factors involved in the last stage computation of 8-Point DFT of a
sequence in radix-2 DIT FFT algorithm.
0 1 0 1 2 3 0
(A) W
4
,W
4
(B) W 8 , W 8 , W 8 , W 8 (C) W 2 (D) All
47. What are the phase factors involved in the first stage computation of 8-Point DFT of a
sequence in radix-2 DIT FFT algorithm.
0 1 0 1 2 3 0
(A) W ,W (B) W 8 , W 8 , W 8 , W 8 (C) W 2 (D) All
4 4
48. Response of discrete LTI system with input x(n) and impulse response h(n) is defined as
∞ ∞
(A) y(n) = ∑x(m)h( −m) (B) y(n) = ∑x(m)h(n + m)
m =−∞ m =−

∞ ∞
(C) y(n) = ∑x(m)h(n − m) (D) y(n) = ∑x(m)h(m − n) .
m =−
∞ m =−∞
49. Where the bit reversing is required for radix-2 DIT FFT computation.
(A) x(n) (B) X(k) (C) Both (D) None
50. Which of the following sequence is decimated in radix-2 DIT FFT computation.
(A) x(n) (B) X(k) (C) Both (D) None