UNIT VI

Fast Fourier Transform

CHAPTER 6

FFT-Efficient Computation of OFT, Goertzel Algorithm, radix2 and radix Decimation-in-Time and Decimation-in-
frequency FFT Algorithms.

✓ Syllabus Topic: FFT-Efficient Computation of OFT ....................................................................................................... 6-2

6.1 FFT-Efficient Computation of OFT ....................................................... ..... ...... ....................... ......................................... 6-2
✓ Syllabus Topic : Radix-2 Decimation In Time FFT Algorithm ......................................................................................... 6-3
6.2 Radix-2 Decimation In Time (DIT) Algorithm (DIT FFT) ................................................................................................. 6-3
UQ. Explain the Decimation in Time FFT algorithm. ............................................ 6-3

UQ. Explain how the decimation in time technique works with necessary equations and diagrams.
GTU - a. S(c), S-19, 7 Marks ··············· ··································· ·· ················································ ··················· 6-3

UEx. 6.2.1 GTU - S-14, 7 Marks ······························································································································· 6-11

UEx. 6.2.3 GTU - S-14, 3 Marks, 0. 4 c OR, W-17, 7 Marks, 0. 5 b, S-19. 4 Marks ..... .................................... 6-14
6.2.1 Comparison of Computational Complexity with Direct Computation ............................................................. 6-16
6.2.1.1 For Direct Computation (Computational Complexity of Direct Computation of OFT) .................................... 6-16
6.2.1.2 Computational Complexity using FFT Algorithm ........ •·········· .............................. ... ....................................... 6-16

6.2.1 .3 Solved Examples ....................... ••·•••··••···································································••••·••••·•• ........................... 6-17

UEx. 6.2.5 ... ····· ··· ····· ···· ····· ·· ······ ···· ···· ····· ·· ·· ··· ·· ·· · ··· ·· · ·· ·· ·· ·· ·· · ·.................................... 6-17
✓ Syllabus Topic : Radix-2 Decimation In Frequency FFT Algorithm ·········· ········· ·· •· •· ••···· ............................................... 6-18
6.3 Radix-2 Decimation in Frequency (DIF) FFT Algorithm ............................................................................................... 6-18
UQ. Explain Radix-2 decimation in frequency FFT algorithm with neat diagram.

GTU - a. 4 c OR , W-18, 7 Marks ············································································································ 6-18

UQ. Discuss in brief: Radix-2 Decimation-in-Time FFT algorithms. .. ............... 6_18
✓
Syllabus Topic: Goertzel Algorithm .............................................................................................................................. 6-23
6.4 Goertzel Algorithm ........................................................................................................................................................ 6-23
UQ. Explain in brief : The Goertzel Algori th m· .... ....... ....... ...... ....... ... 6-23

• Chapter End ......................... •••·····... ··· ·· ······.. ·· ····· ····· .. · ····· ·· ··· ····· ··· · ··· ·· ·· ··· ·· · ·· ··· · ·· · ··· · .. ·· · ·· · ·· · ·· · ·· · · ·•••••............ 6-24

MordenTimeTech.com
 (Fast Fourier Transform) ... Page no .. . (6-
Digital Signal Processing (GTU - Sem. 7 - ECE)
2)
<a
. 2nK
k+N - J--
Syllabus Topic: FFT-Efflclent Computation of WN = e N
DFT
Wk+
N
=
(
e
-l~)k
~ 6.1 FFT-EFFICIENT COMPUTATION OF N · · .(6. 1.6)
DFT In Equation (6.1.6), the bracket term is WN

. Wk+N = WNk
.• N ... (6.1.7)
While calculating OFT, we always calculate 'N' -point
OFT. The number N can be factored as, Equation (6.1.7) indicates that twiddle factor is
N = r 1 , r 2 , r 3 ... rv . .. (6.1.1) periodic.
Here every ' r' is a prime. k+N/2 _ Wk
2. WN -- N
Now ifr 1 = r 2 = r3 = . . .. = rv = r
Using Equation (6.1.6) we can write,
then we can write,
_j ~1t] ( k +N/2)
N =/ ... (6.1.2) Wk+N/2 = [e
N
Here ' r' is called as radix (base) of FFr algorithm and j 2nk j 2n N
' v' indicates number of stages in FFr algorithm. = e N ·e- N · 2
Now radix means base and if its value is '2' then it is j 2nk .
called as radix-2 FFf algorithm. Thus when r = 2; k+N/2 - N -Jn
Equation (6.1.2) becomes,
WN =e ·e ... (6.1.8)
-jn . .
... (6.1.3) But e = cos n - J srn n = - 1 - 0 = - 1
Thus if we are computing 8 point DFf then N = 8 . 2nk
k + N/2 _ - JN
8 = 2v WN --e
V =3 ... (6.1.4)
So for 8 point DFf, there are three stages of FFf
k+N/2 _ ( _j:-\k
WN -- e ) ... (6.1.9)
algorithm. While computing FFf, divide number of input
In Equation (6.1.9), the bracket term is, WN
samples by 2, till you reach minimum two samples. Based
on this division there are two algorithms as follows : . Wk+N/2 - -Wk
.. N - N ... (6.1.10)
1. Radix-2 Decimatio n in Time (DIT) algorithm. Equation (6.1.10) indicates that twiddle factor is
2. Radix-2 Decimatio n in Frequency (DIF) algorithm. symmetric.
Before studying these algorithms we will derive some 2
3. WN= WN/2
important properties of twiddle factor WN. The twiddle
factor WN is given by, From Equation (6.1.5) we can write,
j2n j 2n j 27t
WN = e N - -2
. .. (6.1.5) WN/2 =e-N/2 =e - N .
wk-wk+ N 2
1. N- N _ j 2n]
WN/2 = [ e N
Using Equation (6.1.5) we can write,
2
j2n]k+N j2nk WN/2 =WN
~+N =[ e- N =e N . e- j 2n
N
-j21t
But e = cos 2n - j sin 2n = 1 - jO = 1

(GTU-New Syllabus w.e.f academic year 21-22) (G7-14) ~ Tech-Neo Publications...A SACHIN SHAH Venture

MordenTimeTech.com
 . nal processing (GTU - Sem. 7 _ ECE)
Digital 5 ,g
(Fast Fourlar Transform) ... Paga no ... (6-3)
~ a,os foplc : Radix-2 Decimation In Time FFT N
S1'i' Algorithm and r2 ( m) =x (2n+l), m - O, l , ... , 2 - l ... (6.2.3)

Input sequence x ( n ) has 'N' samples. So after

~--:::~;;:;"nu=ii;;-:;:;~::::-=-=~----.:..
~ RADIX-2 DECIMATION IN TIME {DIT) decimation; f 1 ( m) and f 2 ( m) will contain~ samples.
~ ALGORITHM {DIT FFT)
-------------------------------- ---------- - Now according to the definition of DFT,
r~.--EJCPlain the Decimation in Time FFT ~~~~;i~~~------ N- 1

GTU - Q. S(c), S-18, 7 Marks X(k) = L x(n)Wkn

N
... (6.2.4)
' n=O
: '"' E,cplain how the decimation in time tech • :
: "". nique ,
works with necessary equations and diagrams. ! Since we have divided x ( n ) into two parts; we can

GTU - Q. ______________________
S(c), S-19, 7 Marks _I .:
I
I write separate summation for even and odd sequences as
follows:
-------------------
To decimate means to break into parts. Thus DIT
·
indicateS dividing (splitting) the sequence in time do mam.
The different stages of decimation are as follows :
X(k) = L
n even
x(n) w: + L
n odd
x (n ) w1;;
... (6.2.5)
!Ji' first stage of decimation
The first summation represents even sequence. So we
Let x ( n ) be the given input sequence containing 'N'
will put n = 2m in first summation. While the second
samples. Now for decimation in time we will divide x ( n)
summation represents odd sequence, so we will put
into even and odd sequences.
n = (2m + 1) in second summation. Since even and odd
:. x ( n) = f 1 ( m) + f 2 ( m) .. . (6.2.1)
sequences contain N, 2 samples each; the limits of
Here f 1 (m) is even sequence and f2 (m) is odd
summation will be from m = O to ~ - 1.
sequence
N
:. f 1( m ) = x ( 2n ), m = 0, 1, ...
2 -1 ... (6.2.2)

N N
2- l 2-1
:. X ( k ) = L x ( 2m ) w~lcm + L 2
x ( 2m + 1 ) w~ < m + l) ... (6.2.6)
m=O m=O
But x ( 2m) is even sequence, so it is f 1 ( m) and x ( 2m + I ) is odd sequence, so it is f 2 ( m ).
N N
2- l 2-1
X ( k) = L f1 ( m ) W~km + L f2 ( m ) ~km w~ :. X ( k)
m=O m=O
N
2 -l
= I: ... (6.2.7)
m=O

(GTU-New 4
Syllabus w .e.f academic year 21-22) (G7-l ) ~ Tech -Neo Publications ...A SACHIN. SHAH Venture

MordenTimeTech.com
 (Fast Fourier Transform) ...Page no ... ( ~
Digital Signal Processing (GTU - Sem. 7 - ECE)

F 1(0) = x(O) F 1(0)

- - X(O)
Even
F1(1)-:= x(2) This block
computes
F 1(1)
- X(1)

Inputs
x(2n) = f 1(m)
F1(2) = x(4) 4-point
DFT
F 1(2)
- X(2)

F1(3) = x(6) F 1(3) X(3)

F 1(k)

Adder
Block

F2(0) = x(1) F2(0) W'N ~ X(4)

~

F2(1) = x(3) This block F2'1) W'N X(S)

computes
Odd
inputs
x(2n+1) = f2(m)
F2(2) = x(S) 4-point
DFT
F 2(2) W'N
- X(6)
F2'3) WkN
F2(3) = x(7)
.
F2(k)
- . X(7)

t
Multiplication by
Wk
N

k
Fig. 6.2.1 : Graphical representation of X ( k ) = F 1 ( k ) + W N F 2 ( k )

2
Now we have WN = WN/2 Now F 1 ( k) and F 2 ( k) are 4-point (~) DFTs. They

N N
2- 1 .!:i2-l are periodic with period
2 . Using periodicity property of
:. X (k) = i.J
" f 1 { m) WN/2 2 k'°'
km + WN i.J f ( m) WN
Ian
12 DFr we can write,
m=O m=O
F 1 (k+~) =F 1 (k) , ... (6.2.10)
... (6.2.8)
Comparing each summation with the definition ofDFr, and F2 ( k + ~) = F2 ( k ) ... (6.2.11)
k
X(k)=F1(k)+WN F2(k), k=0,1, ... ,N-1
Replacing k by k + ~ in Equation (6.2.9) we get,
... (6.2.9)
We will consider an example of 8 point DFr. That
means N=8.
X (k+ 2N) =F (k+ 2N) + WN+ N/2 F ( k + 2N)
1
k
2

(6.2.12)
Here F 1 ( k) is ~ point DFr of f 1 ( m) and F2 ( k) is k+N/2 k
Now we have, WN = - WN
~ point DFr of f 2 (m). That means F 1 (k) and F2 ( k ) are
:. X ( k + ~) = F 1 ( k + ~ ) -W~ Fi( k + ~)
4-point DFrs.
Equation (6.2.9) indicates that F2 ( k ) is multiplied by ... (6.2.13)

W~ and it is added with F1 ( k ), to obtain (4 + 4) Using Equations (6.2.10) and (6.2.11) we get,

i.e. 8-point DFr. Graphically Equation (6.2.9) represented X ( k + ~) = F 1 ( k )- W~ F 2 ( k) ... (6.2.14)

as shown in Fig. 6.2.1.
Here X (k) is 'N' point DFr. We can take k = 0 to
Remember that in Equation (6.2.9), k varies from O to N
N - I (i.e. 0 to 7 for 8 point DFr). 2- l then, by using Equations (6.2.9) and (6.2.14) we can

obtain combined N-point DFr.

(GTU-New Syllabus w.e.f academic year 21-22) (G7-14) li1 Tech-Neo Publications...A SACHIN SHAH Venture

MordenTimeTech.com
____________________________ ., _...,....---

. s·gnal Processing (GTU - Sem. 7 _ ECE) (Fast Fourier Transform) ... Page no ... (&-S)
Digit8I I
~ k N 0
X(k)== F1(k)+WN F2(k),k=0,l ... 2-l X ( 0) = P 1( 0) + WN F2 ( 0)
I ... (6.2.17)
... (6.2.15) X (I) = F 1( I ) + W ~ F2 ( I ) }
and X ( k + ~) = FI (k) - w~ F2 (k), k = 0, 1, ... ~_ l X ( 2) = F 1( 2 ) + W N P2 ( 2 )
3
... (6.2.16) X ( 3) = F 1( 3 ) + W N F2 ( 3 )
We are considering an example of 8 point DFT (N == 8). 0
So in Equations (6.2.15) and (6.2.16), k varies from o to 3 _ X(0+4) = X ( 4 ) =F 1( 0 ) - W N F2 ( 0 )
l
;,loW putting k = 0 to 3 in Equations (6.2.15) and (6.2.l6) X ( 1 + 4 ) = X ( 5 ) = F 1( 1 ) - WN F2 ( 1 ) .(6.2.18)
we get. 2
and, X ( 2 + 4 ) =X ( 6 ) = F 1( 2 ) - W N F2 ( 2 )
3
X(3+4) =X ( 7 ) =F 1( 3 ) - W N F2 ( 3 )
The graphical representation of first stage of
decimation for 8 point DFT is as shown in Fig. 6.2.2.

I
X[Oj
f1(1} = x[2]:
Even X[1] From
samples N/2 point , 1
Equation
OFT F1[2) 1 1
x (2n) = f 1(m) X(2] (6.2.17)
I I

__
I
f (3)::: x{6] I F1[3] I :

, .._
I
I .
L-------------~•
,,
I I
I I
X[3)

I
I
I
I
I WO
f 0) = x(1] 1-F...!:if::...O.:..)o-'_...,.N_-41---+--+~'<-+-~M' X[4)
F 1}
I w,
f (1) = x[3] 1 N
Odd X(S] From
samples
N/2 point
Fif2]
I
I
w2 Equation
f (2) = x[5] OFT N
x(2n + 1) = f 2 (m) X[6) (6.2.18)
F2(3J
I w3
f (3) = x[7] 1 N
X[7)

Fig. 6.2.2 : First stage of decimation

N
In Fig. 6.2.2, input sequences are sequences of length 2. That means for 8-point
f1 ( m ) = x ( 2n ) DFT (N = 8); the length of each sequence is '4' as
= {x ( 0), X ( 2), X ( 4), X ( 6 )} ... (6.2.19) given by Equations (6.2.19) and (6.2.20). We discussed
and f 2 ( m ) = x ( 2n + 1) that we have to continue this process till we get '2 •
point sequence.
= { X ( 1 ) , X ( 3 ) , X ( 5 ) , X ( 7 ) } ... (6.2.20)
We can further decimate f 1 (m) into even and odd
N
That means each sequence contains 2 samples. samples. Let gll (n) = f 1 ( 2m ), which contains even
samples and let g 12 (n) = f 1 ( 2m + 1 ), which contains
Q' Second stage of decimation odd samples of f 1 (m ).
- In the first stage of decimation; we obtained th e
Note that here range of 'n' and 'm' is from o to~ _ 1.

(GTU-New Syllabus w.e.f academic year 21-22) (G 7-l 4) ~ Tech-Neo Publications...A SACHIN SHAH Venture

MordenTimeTech.com
 Digital Signal Processing (GTU - Sem. 7 - ECE) (Fast Fourier Transform) ... Page no ... (6-6)

Now recall Equations (6.2.15) and (6.2.16). We Thus, for N = 8 we have the range of k, from k = o to
obtained sequences X (k) and X ( k + ~ ) from F 1 (k) k = 1. Here G 11 ( k) is DFT of gl 1 ( n) and G12 ( k) is
DFT of g 12 ( n ).
N
and F2 (k). The length of each sequence was
2 . Here Now putting the values of 'k' in Equation (6.2.21) we
in the second stage of decimation; we are further get,
dividing the sequences into even and odd parts. So 0
similar to Equations (6.2.15) and (6.2.16) we can write; F 1( 0) = G 11 ( 0) + WN/Z G12 ( 0)
For F 1 (k), } .. . (6.2.23)

F1 ( k) = Gu ( k) + w~,2 G12 ( k) k = 0, 1, ... ~ - 1

Similarly from Equation (6.2.22) we get,
. .. (6.2.21)

and F 1 ( k +~)=Gu ( k ) - w~/2 G12 ( k )

F1 (o +¾) = F1(2) = Gu ( 0 )- WN~Z G12 (O)}
... (6.2.24)

k = 0, 1, ... 4N - 1
F 1(1 +¾) = F1(3) = Gu ( 1 )- WN~ G12 (1)
... (6.2.22)

x[0] = f1[0] = 911[0] G 11 [0]

F 1[0]
Even N/4 point From
samples OFT Equation
of f1 [m] x[4] = f 1[2] = 9 11 [1] (2-point) G 11 [1] (6.2.23)
F1 [1)

0
x[2l= t1[1l = 9 12[01 G12[0] W N/2
F 1[2]
Odd N/4 point From
samples OFT Equation
1
of f 1[m] X[6] = f1[3) = 912[1) (2-point) G12[1) w N/2 (6.2 .24)
F 1[3]
-1

Fig. 6.2.3 : F 1 ( k ), ~ point DFT

Here the values of k are 'O' and I. That means it is 2-point DFT. Thus Equations (6.2.23) and (6.2.24) shows that we
can obtain 4- point DFT by combining two 2-point DFTs. The graphical representation is shown in Fig. 6.2.3.
Note that here,
gll ( n)::;; f1 ( 2m) = x ( 4n) = {x ( 0 ), x ( 4) } }
and g 12 ( n) = f 1 ( 2m + 1) = x ( 4n + 2) = {x ( 2 ), x ( 6)) ... (6.2.25)
Now similar to Equations (6.2.21) and (6.2.22) we can write equations for F 2 ( k) as follows:
k N
F2 ( k ) = G 21 ( k) + W N/2 G 22 ( k ), k = 0, 1, . .. 4 - 1 ... (6.2.26)

... (6.2.27)

Here G 21 (k) is DFT of g 21 (n) and G2 2 (k) is DFT of g22 (n). The values of k are o and 1. Putting these values in
Equation (6.2.26),
(GTU-New Syllabus w.e.f academic year 21-22) (G7-14) li1 Tech-Neo Publications ... A SACHIN SHAH Venture

MordenTimeTech.com
 (
) Page no .. • (6-7 )
. 1
~~ ~Signrn~a~I
GTU S
P~r~oc~e~ss~in~g~~~ ~~====-=
-0~e~m;_;;,~?~-~E~C~E~)~================~(~F:a:st~F~o~u~ri:e~rrT~ra:n:s:fo!:;rm~_;,;..~·

~ ' ( o) e G 21 ( 0 ) + W ~/2 G 22 ( O ) }
... (6.2.28)
f2 ( 1 ) = G21 ( 1 ) - WN/2 G22 ( 1 )

similarly from Equation (6.2.27) we get,

0
=F2 ( 2 ) = G21 ( 0 )
}
F2(0+¾) - W N/Z G 22 ( O )
... (6.2.29)
I
F2(1+¾) = F2 ( 3 ) = G21 ( 1) - W N/ G ( 1 )
2 22

Tbe graphical representation of Equations (6.2.28) and (6.2.29) is shown in Fig. 6.2.4.

Note that here,

g21 ( n ) = f2 ( 2n ) = x ( 4n + 1 ) = {x ( 1 ) , x ( 5 ) } }
... (6.2.30)
g22 ( n) = f2 ( 2n + 1 ) = x ( 4n + 3 ) = {x ( 3 ) , x ( 7 ) }

x( 1) = t2(0) = 921 (0) G 21 (0)

F2(0)
Even N/4 point From
samples Equation
OFT
of f2(m) x(S) = f2(2) = 921 (1) G21 (1) (6.2.28)
(2-point) F2(1)

x(3) = f2(1) = 922(0) G22 (0)

Odd N/4 point From

samples Equation
OFT
of f2(m) x(S) = f 2(3) = 922(1) G22 (1) (6.2.29)
(2-point)
-1

Fig. 6 .2.4 : F 2 ( k ), ~ point DFT

~ Tech-Neo Publications ... A $ACHIN SHAH Ventu re

0

(GTU-N .
ew Syllabus w.e.f academic year 21- 22) (G7-14)

MordenTimeTech.com
 01g1tal Signal Pnx-e~ing (Gl U • Sorn, 7 • 1:CL) f Ml f ourlor Trnnsform) . Pag9 no S.8

011(0)
"[O'
N4 JX)inl
DFT
0 11 (1)
xf4]

G12(0)
l,f2]
N 4 point
OFT
G12(1)
xf6]

G21(0)
>-f1J
N,4 point
OFT
G21(1)
~51

x[3~ 0------1
Ni4 point
OFT
xf7] 0 - - - - - 1

Stage 2 Stage 1

Fig. 6.2.5 : Combinatio n of first and second stage of decimation

le' Combination of first and second stage of DFf. It is separately drawn as shown in Fig. 6.2.6.
cledmadon x(O) =f 1(0) =9 11 (0) G11CO)
G11Ck)
Combming Fig. 6.2.3 and Fig. 6.2.4 in Fig. 6.2.2 we 911Cn)
where n Is
{ 2polnl where k IS
OFT
get the combination of first and second stage of 0 and 1 xj4) =f1(2) =911(1) G 11 (1)
-}
O and 1
- .
decimation. It is shown in Fig. 6.2.5.

At this stage we have ~ that means 2 point sequences. Fig. 6.2.6 : Block of 2-point DFf

So further decimation is not possible. As shown in Here input sequences are g 11 ( O) and gll (I). We can
Fig. 6.2.5; we have to compute 2-point DFf. denote it by g 11 ( n ); where n varies from Oto 1. Now
the output sequences are Gu ( O ) and G 11 ( 1 ). We
rs Computation of 2-polnt DFT
can denote it by G 11 ( k ); where 'k' varies from Oto I.
According to the basic definition of DPT, IJereG 11 (k)isDFfo fg 11 (n).
N- I kn Thus for G 11 ( k) we can write Equation (6.2.31) os.
X(k)= I x(nJWN; k=O,l, ... ,N-1 ... (6.2.31)
1
n=O
We will use Equation (6.2.31) to compute 2 point DFT.
GJ J ( k) = I gl I ( n) w;n .k =0, I ... (6.2.32)
n =O
From Fig. 6.2.6, consider the first block of 2 point 2
Nole that this is 2 point D1-<T, so we have put N = ·

(GTU-New SVflabus w.e.f academic year 21-22) (G7-14) [iJ Tech Neo Publications...A SACHIN SHAH Venture

MordenTimeTech.com
 . s·gnal processing (GTU - Sem. 7 - ECE) (Fast Fourier Transform) ... Page no ... (6--9 )
Digit81 I •

rJoW Putting values

of k rn Equation (6·2 ·32) we get, we can modify
Now we know that W~ = 1. Thus
1
fork== O ==> Gn <o) = L gll ( n) W~ Equations (6.2.33) and (6.2.35) as follows :
0 ... (6.2.36)
n=O GJl(O) =g 11 (0)+W2 gll(l)
_.o == I 0
and Gu(l) =g 11 (0)-W2 g 11 (1)
... (6.2.37)
Wz
13111
I This modified butterfly structure is shown in Fig. 6.2.8.
;. G1 I ( 0 ) == L g 11 ( n )
Similarly for other 2-point DFTs we can draw the
n=O
butterfly structure.
Expanding the summation we get,
Fig. 6 .2.8 modified butterfly structure Similarly for
Gu ( 0) = gll ( O) + gll ( l ) ... (6.2.33)
other 2-point DFfs we can draw the butterfly structure.
1
911(0) 1 G 11 (0)
fork== I ==> Gu ( 1 ) = L g{1 ( n) W~ x(O)
n=O
Expanding the summation we get,
Gu (I ) = g 11 ( o) w~ + g 11 (1 ) w~ ... (6.2.34) O
911(1 ) _ W
_..,2~_ _ _ _......,_ _ _ _---4
j 27t x(4) -1

We have WN = e- N s Fig. 6.2.8 : Modified butterfly structure

-
j 2rc)l
N
.
- J1t ~ Total signal flow-graph for 8-point DIT FFT
WI
2
=( e =e
The total signal flow graph is obtained by
= cos 7t - j sin 1t = - 1 - j O
interconnecting all stages of decimation. In this case, it
1 0
:. W2 = - 1 And W2 =I is obtained by interconnecting first and second stage of
Putting these values in Equation (6.2.34) we get, decimation. But the starting block is the block used to
G11 (I) = gll ( 0) - g 11 ( I ) ... (6.2.35) compute 2-point DFf (butterfly structure).
The total signal flow graph is shown in Fig. 6 .2.9.
Using Equations (6.2.33) and (6.2.35), we can represent
the computation of2-point DFf as shown in Fig. 6.2.7. This
structure looks like a butterfly. So it is called as FFf
butterfly structure.
911{0) 1
x(O) ~ - - - -.....-----c

911(1)
X{4) -----+-_1- - - - - t
Fig. 6.2.7: FFf butterfly structure

(GTU-New Syllabus w.e.f academic year 21- 22) (G7-l4)

l'1 Tech- Neo Publications.·· A SACHIN SHAH V enture

MordenTimeTech.com
.... CJ:L.cnnc:.
 G) 0
4 (C.
C
z ~
t1)
~
en
(C.
Vl ::,
'S.. ~
ii, x[O)
CT
X[O) ""C
C:
V, g
(D
~ (/)
(1) (/)
:... 3·
Cl) WO <C

MordenTimeTech.com
n
Cl)
8
a. x[4) G)
t1) X[1] -I
3 C
n·
'<
t1)
en
(D
~ ~
N
I;"' x[2)
......
N X[2]
~ m
()
G)
-;J !!!
~
~ I WO
8
x[6) • . .c::::: . . _, . ~ +}-- X[3]
>(D:I 0
-1 I -1
\ X 7\. .. X

x[1] )lo ,, I
• )lo ll
I ':::::: _;A!)! I II~ Xl 4]
I
. 59 • X I\ 1\--1
~
4
t1)
n WO
::r 8
z
t1)
0
x[5)
' xtSJ
,, a>
~
C: .,,
0
"~
n·
Cl)
~
.-+ ~
cs-
::, x[3) -I
r)> X{6] in
::,
Vl a-
►
(")
I WO -3
z 8 VY 8
V'I x[7) vv 8 :
I • . ...:::::::::: . ~ l ,, / )lo I I ,, X[7]
-1 I
►
. -0
-1
I 0----
-1
<
(1)
:::, Fig. 6.2.9 : Total signal flow graph for 8-point DIT FFT
'c
~

0
I!-
 . al processing (GTU - Sem. 7 - ECE)
. :.al 5,gn
(Fast Fourier Transform) ... Page no ... (6- 11 )
o,g,,
~ GTU S
~.&.2,1 .
11 te the eigbt-pomt DFf of a sequence.
colllPu{1 . I ,11
, ( )~
} . . .
11 2 2 2 • 2 • O, O, O, O USmg m-place radix-2 decimation in time FFT algorithm.
&'.1 soln, : 'Ibis flow graph is shown in Fig. Ex. 6.2.1. .
11ere s1 (n) represents output of stage - 1 and s2 (n) represents output of :,Lage - 2. The different values of twiddle factor are

w~ =eo =1
-j~
4
Wl = e
8
= 0.707 - j 0.707

'

...-4
N
a)
N
Cl) ~
~ ;,,<
~
bi>
ri:;

'i

Os:'° Os:'°

E
)(

(GTU
·NewsYilabus
w.e.f academic year 21-22) (G7-1 4) ~ Tech- Neo Publications...A SACHIN SHAH Venture
MordenTimeTech.com
 Digital Signal Processing (GTU - Sem. 7 - ECE) (Fast Fourier Transform) ... Page no ... (6·1 )
2
.n 3 ==--
2 -JI . x (3 ) = S2 ( 3 ) + W 8 S2 ( 7 )
W8 =e =-J

Wg = - 0.707 - j 0.707 =( ½+ j ½) + ( - 0.707 - j 0.707 ) ( ½+ j ½)

Output of stage - 1 = ( ½ + j ½) + (0 - j 0.707) = 0.5 - j 0.207
o I I
0
s1 ( 0) = X ( 0) + Ws X ( 4) =2 + I ( 0) =2
X ( 4 ) = S2 ( 0 ) - W 8 S2 ( 4 ) = 1 - 1.1 = 0
-~
St ( 1 ) = X ( 0) - W g X ( 4) = 21 I
- 1 ( 0) = 2 1
X ( 5 ) = S2 ( 1 ) - W 8 S2 ( 5 )
o 1 1
Sl ( 2) = X ( 2) + W S X ( 6) =
2
+ 1 ( 0) = 1
= ( ½- j ½ ) - ( 0.707 - j 0. 707 ) ( ½- j ½)
o 1 I
St(3) =x(2)-W8 x(6)= -1(0)=
2 2 =(½-j½)-(-0.707j)
o 1 1
st(4) =x(l)+W8 x(5)= +1·(0)= :. X ( 5) = 0.5 + j 0.207
2 2
o 1 1 X ( 6 ) = S2 ( 2 ) - ~ S2 ( 6 ) = 0 + j · (0) = 0
St ( 5) = X ( 1 ) - WS X ( 5) =
2- 1 • ( 0) =
2
3
-~ I 1 X(7) =s2(3)-W8 s2(7)
=x(3)+w 8 x(7)= +1 ·(0)=
2 2
o 1 1 = (½+ j ½) - <-0.707 -j 0.707) [½+ j ½]
St ( 7) =x(3)-W8 x(7)= -1·(0)=
2 2
= (½+ j ½) + 0.707j = 0.5 + j 1.21
Output of stage - 2
Thus, X ( k) = {X ( 0 ), X ( 1 ), X ( 2 ), X ( 3 ), X ( 4 ),
S2 ( 0) = si( O ) + W~ si( 2 ) = ½ + 1 . ( ½) = 1 X( 5 ), X ( 6 ), X ( 7)}
:. X ( k) = {2, 0.5 - j 1.207, 0, 0.5 -j 0.207, 0, 0.5
S2 ( 1 ) = St ( 1) + ~ St ( 3) = ½ -j ½ + j 0.207, 0, 0.5 + j 1.21}
0 1 1 n
S2 ( 2) =s 1 (0)-W 8 s 1 (2)= 2 -2 =0 Ex. 6.2.2 : Given x (n) = 2 and N = 8 Find X (k) using DIT

_ _2 I .I
FFT algorithms.
S2 ( 3) =s 1 ( 1 )-W8 _SJ (3)= 2 +J2 0 Soln.:
Here, N= 8 n =0 to N - 1 = 0 to 7
0
For n = 0 ⇒ x(0)=2 =1
2 1 .I t
S2 ( 5) = St ( 5) + W 8 SJ ( 7) = 2 - J2 n= 1⇒ x(1)=2=2
2
0 1 1 n=2 ⇒ x(2) =2 =4
S2 ( 6) =si (4)-W8 SJ (6)= 2 -2 =0 3
n=3 ⇒ x(3) =2 =8
2 1 .1 4
s 2 ( 7 ) = St ( 5 ) - W S Si{ 7) = 2 + J 2 n=4 ⇒ X ( 4) = 2 = 16
5
Final output
n=5 ⇒ X ( 5) = 2 =32
6
n=6 ⇒ x(6) =2 = 64
0
X(0) =s 2 (0)+W s (4)=1+1=2 7
8 2 n=7 ⇒ X( 7) = 2 = 128
1 The flow graph is shown in Fig. Ex. 6.2.2.
X ( 1) =s2( 1 )+W 8 s2(5)

= ( ½- j ½) + ( 0. 707 - j 0.707 ) ( ½- j ½)
= 0.5 - j 1.207

X(2) =s 2 (2)+~ si(6)=0+(-j){0)=0

(GTU-New Syllabus w.e.f academic year 21-22) (G7-14) ~ Tech- Neo Publications ...A SACHIN SHAH Venture
MordenTimeTech.com
 !:.~~ ~:l.i rm) ...!.~-1
ii~i! !;~! - 2.
·:n~a~Ip=roc~e;;;ss=in;aig~(G;;;;;;;;T;;;;U;;;;-;;;;S;;;;e;;;m~.~7~-E;;;C~E~)----,-----J!::!.::i:(Fast Fourier Transfo Page no.. . (6 13
.d(81
l/l!l..o1111 5,g

~ere• w~ = 1 _jtt SI ( 1) a X( 0 ) - Wo
g X( 4) s: 1 - 16 =- 15

w! : : e4 = 0.707 - j 0.707
81 ( 2) = X ( 2) + W~
0
X ( 6) = 4 + 64 = 68
s1(3) =x(2) - Wg x(6) = 4 - 64 =- 60

Vi's
2 :::e 2-
- jtt
=- j 81 ( 4) = X ( 1) + Wg
0

0
X ( 5) = 2 + 32 = 34
-j6lt St ( 5) = X ( 1 ) - Wg X ( 5 ) = 2 - 32 = - 30
3 ~ = - 0 .707-j 0.707
v-, :::e 0
8 s1(6) =x(3) +W8 x(7)= 8+128 =136
0
ut of stage 1 S1 ( 7) = X ( 3 ) - Wg X ( 7 ) = 8 - 128 = - 120
o~P o
s1(0) = x( 0 )+ W g x ( 4 ) =1+16 =17

0 ., 0.,
~ ;:

,.
§: ~ c,j' ~ ~ " '
~>< ~
-:=~==-==::--;,~~~::----1~~=-=-=-~---------
--- ->< x ><
x x
( -------

GTu-N ew SyllabLus w.e.f academic year 21-22) (G7-14) Tech-Ne o Publicat ions ... A SACHIN SHAH Venture

MordenTimeTech.com
 Digital Signal Processing (GTU - Sem. 7 - ECE) (Fast Fourier Transform) ... Page no ... (6·14)

Output of stage 2 = 78.6 - j 46.08 -

2
0 X(6) =-S2 (2) - Ws s2 (6) =-5 J +j(-102)
S2 ( 0 ) = Sl ( 0 ) + W 8 SI ( 2 ) = 17 + 68 = 85
2 =- 51 - j 102
S2 ( 1 ) = SJ ( 1 ) + W 8 SJ ( 3 ) = - 15 - j ( - 60) 3
X (7) =S2 ( 3 ) - W 8 S2 ( 7 )
= -15 + j 60
0 = (- 15 - j 60) - ( - 0.707 - j 0.707) ( - 30- j 120)
s2 (2) =si(0)- W 8 si(2)= 17-68 = -51
2 = - 15 - j 60- ( 63.6 + j 106.08)
S2 ( 3 ) = SI ( 1 ) - W S Sl ( 3 ) = - 15 + j ( - 60 )
X ( 7) = 48.6 - j 166.08
= -15- j 60
X ( k) = { 255, ( 48.68 + j 166.04 ), ( - 51 + j 102),
0
S2(4) =s1(4)+Ws S1(6)=34+136 =170 ( - 78.67 + j 46.05 ) , ( - 85 ), ( - 78.6 - j 46.08 ),
2
S2 ( 5 ) = Si( 5 ) + W S Sl ( 7 ) = - 30 - j( - 120 ) ( - 51 - j 102), ( 48.6 - j 166.08) }

= -30+ j 120
0
UEx. 6.2.3 GTU - S-14, 3 Marks, a. 4 c OR. W-17.
S2 ( 6) = S1 ( 4 )- W 8 SJ ( 6) = 34 - 136 = - 102 7 Marks, a. 5 b. S-19. 4 Marks
2
Sz ( 7) = S1 ( 5 ) - W8 S1 ( 7) = - 30 + j ( - 120) Derive DIT FFf flow graph for N = 4 hence find DFf of
=-30-j 120 x (n) = {1, 2, 3, 4}.

Final output
0 Soln. : First stage of decimation

0
We have the equations for first stage of decimation.
X ( 0) = s2 ( o) + W8 s2 ( 4) = 85 + 170 = 255 k N
X ( k) = F 1 ( k) + WN F 2 ( k), k = 0, 1 ... -1
l
X ( 1 ) = Sz ( 1 ) + W 8 s2 ( 5 ) 2
... (1)
= (- 15 + j 60) + (0.707 - j 0.707) (- 30 + j 120)
= ( - 15 + j 60) + ( 63.68 + j 106.04) and X (k + ~) = F1 ( k )- w~ F2 ( k ), k = 0, I, ... ~ -1
.·. X ( 1) = 48.68 + j 166.04 ... (2)
2
X(2) = S2(2)+Ws s2(6) HereN=4

= - 51 - j ( - 102) = - 51 + j 102 X ( k) = F 1 ( k) + W~ F 2 ( k ), k = 0, 1 ... (3)

3
X ( 3) = s2 ( 3) + W 8 s 2 ( 7) = ( - 15 - j 60) and X(k+2)=F1(k)-w1 Fz(k), k=0,l ... (4)
+ ( - 0.707 - j 0.707 )( - 30- j 120) Putting values of kin Equation (3) we get,
= ( - 15 - j 60) + ( - 63.67 + j 106.05 ) 0
X ( 0 ) = F 1( 0 ) + W ~ F2 ( 0 ) } ... (5)
= - 78.67 + j 46.05
0 and X ( 1 ) = F 1( 1 ) + W F2 ( 1 )
X ( 4 ) = S2 ( 0 ) - W 8 Sz ( 4 ) = 85 - 170 = - 85 4
l Similarly putting values of kin Equation (4) we get,
X (5) = Sz ( 0 ) - W8 Sz ( 5 )
0
= (- 15 + j 60) - (0.707- j 0.707) ( - 30 + j 120) X (2) = F 1( 0 ) - W ~ F2 ( 0 ) } ... (6)

=- 15 + j 60-( 63.6 + j 106.08) and X ( 3 ) = F 1( l ) - W F2 ( 1 )

4

(GTU-New Syllabus w.e.f academic year 21-22) (G7-14)

~ Tech-Neo Publications...A SACHIN SHAH Venture
MordenTimeTech.com
~ I Signal processing (GTU • Sem. 7 • ECE) (Fast Fourier Transform) ... Page no ... {6-1S)
_...ari,is signal flow graph is shown in Fig. Ex. 6.2.3(a).

X[O]
Even
samples 2-polnt
of x(n) OFT
X[1]

X[2]
Odd
samples 2-polnt
of x(n) OFT
X[3]

Fig. Ex. 6.2.3(a)

Now we will replace each 2-point DFf by butterfly structure as shown in Fig. Ex. 6.2.3(b).
X[O]

X[1]

X[2]

X[3]

Fig. Ex. 6.2.3(b)

The output s ( n ) is,

The given sequence is,
So = X (Q) + X ( 2 ) = 1 + 3 = 4
x(n ) = {l , 2, 3, 4}
SI =X (Q) - X ( 2 ) =1 - 3 = - 2
The different values of twiddle factor are as follows :
s2 = [ x (l) + x ( 3)] w~ = 2 + 4 = 6
~ =1
I
j 27t j 7t s3 =[x(l) - x(3)]W4 =(2 - 4)•(- j)=2j
Wl
=e
-7 · 1 = e
- 2
The final output is,
4
X ( O) = s0 + s2 = 4 + 6 = 10
7t . . 7t .
= cos
2 - J sin 2 = -' J
X ( l) =S1 + S3 =- 2 + j2
X( 2 ) =s0 - s2 =4 - 6= - 2

(GTU-New Syllabus w.e.f academic year 21-22) (G?·l 4)

~ Tech-Neo Publications...A SACHIN SHAH Venture

MordenTimeTech.com
 Digital Signal Proces sing (GTU • Sem. 7 • ECE)
(Fast Fourie r Transf orm) ... Page no ... (6-1 S)
X (3) = s, - S3 = - 2-j2 In Equati on (6.2.40 ); four complex multiplications are
Thus, X ( k ) = { X ( 0 ), X ( I ), X ( 2 ), X ( 3 ) } require d and three compl ex additio ns are required. Here We
.·. X(k) = (10,- 2+j2 ,-2,- 2-j2} have consid ered N = 4. Thus for each value of 'k'; N
compl ex multip licatio ns are requir ed and 'N - l ' comple
x
~ 6.2. t Computson of Computadonal additio ns are require d. Now the total values of k and 'N'.
Complexity with Direct Computadon :. Comp lex Additi ons = N ( N - l) = N 2 - N

Fin,t we will calcul ate the compu tationa l compl exity

for direct DFf calcul ation.
a. 6.2. 1.2 Computadonal ComplexJty using
FFT Algorithm
a. 6.2.1 .1 For Direct Computadon Firstly we will calcul ate the compu tation complexity
(Computadonal Complexity of requir ed for one butterf ly. Consi der the genera l structure of
Direct Computation of DFT) butterf ly as shown in Fig. 6.2.10.

Accor ding to the definit ion of DFf we have, a --11 1::- ----....1----- -::-; 1>-- -o A= a+ w'Nb
N-1
X ( k) = L x ( n) ~ , k =0, 1, 2, .... N - 1. .. (6.2.38)
n=O
Equati on (6.2.38 ) indica tes that we have to take b
w'N
multip licatio n of x ( n ) and twiddl e factor. Then we have -1
to
add all the terms. Since twiddl e factor is compl ex; we Fig. 6.2.10 : Gener al struct ure of butterfly
need
to perfor m compl ex multip licatio ns and compl ex additions.
Here ' a' and 'b' are inputs and A and B are outputs of
~ Comp lex muldp llcatio ns
butterf ly. The output s are given by,
As given by Equat ion (6.2.38 ), for one value of 'k'
A =a+W ~ b ... (6.2.41)
multip lkatio n should be perfor med for all values of 'n'. The
range of ' n' is from O to N - 1. So for one value of 'k';
N
and B = a-W~ b ... (6.2.42)
compl ex multip licatio ns are require d. Now the range of k
is 1. To calcul ate any output (A or B), we need to multiple
also from k = 0 t.o k = N - 1. The total compl ex
multip licatio ns are, input 'b' by twiddl e factor WN .
So one complex
multip licatio n is requir ed for one butterfly.
Comp lex multip licatio ns = N x N = N2 ... (6.2.39)
q» Comp lex additi ons
2. To calcul ate output 'A', one compl ex addition is
require d, while to calcul ate output 'B' one complex
Accor ding to Equati on (6.2.38), for each value of k we subtra ction is require d as given by Equation (6.2.42).
need to add the produc t terms of x ( n ) W~ . For examp le, But the compu tationa l compl exity of addition and
let us say N = 4. subtra ction is same. So we can say that for one
3 butterf ly two compl ex additio ns are required.
For k =0 ⇒ X ( 0 ) = L x ( n ) w~ x 0 3. As shown in Fig. Ex. 6.2.1 , for 8 point DFf; 4
n=O butterf lies are there at each stage. So for 'N' point
3
DFf, at each stage, 'N' butterf lies are required.
= L x(n)W ~ 2
n=O 4. As shown in Fig. Ex. 6.2.1; three stages are required to
:. X ( 0) =x(O )W04 +x(l )W0 compu te 8-poin t DFf. In general, for 'N' point DFT,
4
log2 N stages are required.
0
+x(2 ) W 4 +x(3 )W0
4 ... (6.2.40)

(GTU-N ew Syllabus w.e.f acade mic year 21-22) (G7-14

) Iii Tech-N eo Publications...A SACHIN SHAH Venture

MordenTimeTech.com
 (6-17)
_!
. I 5,g!!,81!p:roce~~ss:m~g~(G!T!.:U~
-~s:e~m~.~7~-!E~C;!E~)----~~---- ~(F~a!!s:!.t~F~ou~n~·e~r~T~ra:;n~s~fo~rm~)~.~P,;a~g,;,e,.n.._
0
...._ _

~eX mufdpOadons q- Complex aclclldons

Jf N N
stage. 2
-b
!,.tea~
stage there are 2 butterflies. Total number of Total number of stages are 10g2 N At each
Jog2 N. And for each butterfly, one complex rfly •r complex
butterflies are required. And for each butte '
~ ::Von is required. additions are required.
IJlllltIP . . . N N
-i complex mult:J.plicat:J.ons = -2 log2 N ... (6.2.43) Total complex additions = 2 X2 log2 N
•• TOIJI.I

Total complex additions= N log2 N

Table 6.2.1 shows companson - of direct DFT
computation and computation using FFf algorithms.
Table6.21 .
Direct Computation I Using FIT
~

~•N' " Complex Multiplication

Nl
Complex Additions
(N 2_N)
Complex Multiplication Compa Additioa
(Nlogl N)
(~ log2 N)
- 4 16 12 2 4
- 8 64 56 12 24
I 16 256 240 32 64
I
f 32 1024 992 40 80

64 4096 4032 96 192

128 16,384 16,256 224 448

I

Toe Table 6.2. l shows that, by the use of FFf

Using FFT
algorithms the number of complex multiplications and
complex additions are reduced. So there is tremendous N 1024
Number of multiplications = tog2 N = - -1og2 1024
improvement in the speed.
2 2
= 5120
'&. 6.2.1 .3 Solved Examples Number of additions = N log2 N = 10240

l&.&.2.4 GTU - S-13. 7 Marks

UEx. 6.2.5 GTU - 0. 4 b. W-18. 4 Marks
Find the number of complex multiplications and compl~x
Calculate number of complex additions and multiplication
additions required in the direct computation of 1024 pomt
require when computing 512 point DFf using direct
DFT. How many complex multiplications and complex
method. Also do the done if radix - 2 FFr is used.
additions will be required using FFf algorithm ?
@Soln.:
0 Soln.:
Given N = 512
Using OFT
Using OFT
Rere N = 1024
Number of complex multiplications
Complex multiplication = N2 = (512/ = 262144
2
Complex additions= N - N = (512/ - 512 = 261632
=N7 = (1024/ = 1048576
Number of complex additions= N2 - N = 1047552

(GT\.J_New Syllabus w .e.f academic year 21-22) (G7- 14) li1 Tech-Neo Publications ...A SACHIN SHAH Venture

MordenTimeTech.com
 Digital Signal Processing (GTU - Sem. 7 - ECE) (Fast Fourier Transform ) ... Page no ... (6-18)
Using FFT N
N- 1 2- I
Complex multiplic ations=
N
2 log 2N = 2
512
log 2512 = 2304
:. L x ( n) W~ = L ~ ( n + ~) w~; ( n +~)
N n-0
Complex additions = N log 2N = 512 log25 I 2 = 4608 n =2
Putting this value in Equation (6.3.2) we get.
Syllabus Topic: Radlx-2 Decimation In Frequency
N N
FFT Algorithm -2- 1 -2- 1

.., 6.3 RADIX-2 DECIMATION IN

X(k)= L x(n)W~ + L x(n+~ )-~ (n+~)
n=O n=O
FREQUENCY (DIF) FFT ALGORITHM N
2- l N 1
-------------------------------------------------------------- 2-
kn
1 UQ. Explain Radix-2 decimation in frequency FFT = L x(n)W 0 + L
1 algorithm with neat diagram. n=O n=O
'
''
GTU - Q. 4(c) (OR), W-18, 7 Marks ~2- 1 ~
2- 1
1 UQ. Discuss in brief: Radlx-2 Decimation-in-Time FFT
algorithms. GTU - Q. S(c), W-19, 7 Marl<s
llllX(k)= L x(n)W ~"+~
2
L x( n+~)~
n=O n=O
Decimati on in frequenc y stands for splitting the
... (6.3.3)
sequence s in terms of frequency . That means we have to
Now we have,
split output sequence s into smaller subseque nces.
j21t
This decimati on is done as follows :
=e N
~ Ant stage of decimation
j 21t kN
--x- . )k
Accordin g to the definition of OFf, ~/2
N
=e N 2 =e-jnk = ( e- J7t
N-1
X(k) = L x(n)W ~ ... (6.3.1) W~
12
= ( cos 1t - j sin 1t / = ( - 1 - j O )k
n=O
W~/2 = (-1 )k
We can divide the summati on into two parts as follows:
Putting this value in Equation (6.3.3) we get,
N 1
2- N-1 N N
2-1 2-1
X ( k) = L x ( n) w~ + ~ x ( n) w~ ... (6.3.2)
n=O n= 2 x ( k) = L x ( n) w~ +(-1)k L x( n +~) w~
n=O n=O
Now consider the second summati on that means,
Taking the summatio n common we get,
N-1
N
L x(n)W ~ 2-
1

n=2
N
x ( k) = L [x ( n) + ( - 1 l x ( n + ~)] W~ ... (6.3.4)
n=O
Put n = n + ~ ; the limits will change as follows : Here we have to split the sequence in terms of
N N N frequency . So we will split X ( k ) in terms of even
when n = ⇒ =n +2 :. n=0
2 2 numbered and odd numbered OFT coefficien ts. Let X ( 2r )
N represent s even numbere d OFT and X ( 2r + l ) represent
and when n = N - 1 ⇒ N - 1 = n +2 odd numbere d OFT.
N N Thus putting k = 2r in Equation (6.3.4), we will get
n =N-l-
2 =2 -1
even numbere d sequence .

(GTU-New Syllabus w.e.f academic year 21-22) (G7-14) ~ Tech-Neo Publications ... A SACHIN SHAH Venture

MordenTimeTech.com
 ... (6.3.5)
BY putting k = 2r + 1, in Equat ion (6.3.4), we will get
odd nUlllbered sequence. Now let, g ( n ) =x ( n ) + x ( n + ~) ... (6.3.1 5)
N _1

X (2r + l)= n
2
~ J x ( n) + ( - 1) 2r + 1 x ( n + ~)]
and h ( n ) = [ x ( n ) - x ( n + ~)] W~

Putting these values in Equat ions (6.3.13) and (6.3 .14)

we get,
. .. (6.3.1 6)

w<2r+ 1) n
N ... (6.3.6) N 1
Here 'r' is an intege r simila r to k and it varies from o to 2-
N
--1.
x ( 2r ) = I g(n) w;;12 ... (6.3.1 7)
2 n=O
:. ( - 1 )2r ~ 1 ... (6.3.7) N 1
2-
and (- 1) 2r+ 1 = (- 1 )2r (- 1 )1 = - 1 ... (6.3.8)
Putting these value s in Equat ions (6.3.5) and (6.3.6) we get,
and X ( 2r + 1 ) = I h ( n ) w;;
12 ... (6.3.1 8)
n=O
N 1 Note that at this stage we have decim ated the seque nce
2-
of 'N' point DFf into two ~ point DFfs given
X(2r ) = L [x(n )+x( n+~ )]w~ m ... (6.3.9) by
n=O - Equations (6.3.17) and (6.3.18). Let us consi der an exam
ple
N 1 of 8-poin t DFf. That means N = 8. So comb
2- ining
.
Equations (6.3.17) and (6.3.18) that mean s N = 4 we
and X ( 2r + 1 ) = L [ [x ( n ) - x ( n + ~)] wir + l) n 2 can
n=O obtain N (8 point) point DFf. This is first stage
of
... (6.3.10) decimation. Note that Equat ion (6.3.17) indica tes 4 (
~)
Now consid er the term ~rn point DFf of g (n) and Equat ion (6.3.18) indica tes 4 (~)

w~m = (w~) m point DFf of h ( n ). For 8-poi nt DFT Equat ion (6.3.1
5)
becomes,
But we have
g(n) =x(n )+x( n+4 )
... (6.3.1 9)
W~ =WN /2 Here we are computing '4' point DFT. So range of 'n'
W~m = (WN12) m = w;,2 ... (6.3.11) is n = 0 ton= 3. Putting these values in Equat ion (6.3.1
9)
we get,
Now we can write,
\JI.N2r+ l) n _ W2m w..n -Wm • W" For n = 0 ⇒g ( 0 ) = x ( 0 ) + x ( 4 ) }
- N • WN -
... (6.3.12)
N/2 N For n = 1 ⇒ g ( 1 ) = x ( 1 ) + x ( 5 )
. these values
Putting . (6 3 9) and (6.3.10)
in Equations · · For n = 2 ⇒ g ( 2 ) = x ( 2 ) + x ( 6 ) ... (6.3.2 0)
We get,
Fam =3 ⇒ g ( 3 ) =x ( 3 ) + x ( 7 )
N
2- l Similarly for 8 point DFT Equat ion (6.3.1 6) becom es,
X(2r) = L [x(n )+x( n+~ )Jw ;12 ... (6.3.13)
n=O
h ( n) = [ x ( n )- x ( n + 4)] w:, ...
(6.3.2 1)

~ - - - - - - - - - - - - ) - ( G --__ _)_
_ _ _ _1'1~T~e=c~h-~N:e:o~P~u~b~lic:a:ti:on:s~..~.A~S~A~C~H=l~N~
w Syllabus w.e.f academic year 21- 22 7 14 S~H~A~H--V
_ __
entur e

MordenTimeTech.com
 Digital Signal Processing (GTU - Sem. 7 - ECE) (Fast Fourier Transform) ... Page no ... (6-20)
u.,ing J~quat,on'> (6. 3.20) and (6.3.22), and
For n =0 ⇒ h(Q);.(x(O) x(4)J w08
J:.qualJon<; (6.3. 17) ond (6.3 18) we can draw the Oow graph
1 ol f m,t stage of dcc1mat1on a$ E.hown in Pig. 6.3. 1.
For n = I ⇒ h ( I ) = l x ( I ) - x ( 5 ) ] W
8
... (6.3.22)
2
For n =2 ⇒ h ( 2 ) = [ x ( 2 ) - x ( 6 ) ] W8

3
For n =3 ⇒ h ( 3 ) =[ x ( 3 ) - x ( 7 ) ] W8

g(O] ...__._ _-oX(O]

x(O]
g(1 l l:!. point 1----1---0 X[2) From
x[1] 2 Equation
g[2} OFT
x[2] 1---------0X[4] (6.3.17)
(41)oint)
x[3]

x[4]
g[3]

h{O] ----- 1---------0X[6J

- - - - - X [ 1}
x[5]
h[1] Ji point 1----1---0X[3} From
2 Equation
h[2] OFT
x[6] 1----1---oX[SJ (6.3 .18)
(41)olnt)
x[7]
h[3]
...,___;;-----L_ ___.._ __,.._ _ X[7]

Fig. 6.3.1 : First stage of decimation

~ Second stage of decimation

In the first stage of decimation we have used 4-point DFf. We can further decimate the sequence by using 2-point DFI'.
The second stage of decimation is shown in Fig. 6.3.2. This is similar to DIT-FFf.
Stage 1 Stage 2
x(O) X[O]
¾point
x(1) OFT X[4]

x(2) X[2]
¾point
x{3) OFT X[6]

x{4) X[1]
~ point
x[5} OFT X[5]

x[6) X[3)
~ point
x[7] OFT X[7)

Fig. 6.3.2 : Second stage of decimation

(GTU-New Syllabus w.e.f academic year 21-22) (G7-14) lil Tech-Neo Publications...A SACHIN SHAH Venture
MordenTimeTech.com
 m~.~ 7~-;_:E~C~E~)~======= ;;;;;;;
(Fast Fouri Trans::~~i
er ~~~~
;=~~ ... ~~e1n~o~(~6;;-2~1;J;)
form~P~a
. ~~S~ig~na:I:P;ro~c~ess~in:g:(:G:T~U=-~S;:;:e;:
Oig1tsl
fblrd siage of decimation
(I' Stage 1
x[O) X[O]
xr1 1-~ ..._ .--- -~- 1
v/'8 X[4)

x[2) _ __,, ..,_ _~~ -~-' X[2)

x[3) -~~~=--~"+-1 v/'8 X[6)

X[1]
x[4] ~➔~➔E--~:...::,
vi. X[5]
x[5] ~-'7 '--7 '4~~ ....,: .,
X(3]
x(6) 6--- -;,,. .__ --~- ~ WO
x[7) <S--------....._, 8 X[7)

FFT
Fig. 6.3.3 : Total flow graph for 8-po int DIF-
ble. Now we will use a
secon d stage of decim ation we have used 2-po int DFT. So further decim ation is not possi
In the flow graph for 8 poin t
rfly is same as shown in Fig. 6.2.9. Thus the total
butterfly structure to obtai n 2-po int DFT. This butte is in the oppo site direc tion.
show n in Fig. 6.3.3 . This flow graph is similar to the flow graph of DIF- FFT but it
DlF---F Ff is N = log2 8 = 3 stage s.
t is shuffled. Similar to DIT- FFT there are log2
Note that here inpu t x ( n ) is in sequ ence but outpu
requirement is same as that of DIT- FFT.
Toe computational comp lexit y and the mem ory

Ex. 6.3.1 : Obta in DFf of a sequ ence x ( n ) =

(½ ,½,½,½, 0 , O , O , O)
Using decimation in frequ ency FFf algorithm.
0 Soln.:
l.
The total flow graph is show n in Fig. Ex. 6.3.
~----------------- .~ ---- ---- ---- ---- -~
g(O): I
I h(O)
I -----------
I
I
I X(O)
-1- .;o, --+ ---- --~ +
;,._ ---- ---t l'"{ •+- I I
X(O) I
I
I O I
h(1) 1 W8 1

I -1 ____ _ ,I
I, ____ ____ ____
X(4)
X(1)
W8 o:-t:--
h(2) : :
o--4 o:;: ---- --+l ir--- 1 X(2)
t1-1 ;:\..- -+-: :0-~ --~~ -~+> --=- I
X(2) :.,__ __;: ..,... ..--¥ ---r-
I I
I I
21
W h(3) :
-_:::;. __ _ _ _--+-f
:\..-... ....:... ..::a;_ ;+L-- -~:--- i~+-} --=8- i1r--O :____ ____ __ -1 ____ _ I
X(3) 1,-_. ;¥,-- ,¥--- r:;.- tl'1 ___ :

h~) r----------------~: X(1)

<:::: ::::- ----~ ::i+
X(4) v- _ ___ :~__ ,~-- ;;i._ -~0~ ~;... .:.... ,.--- ---+ -W, --t-~ --.1 I
I
I
I
h(5) : W o I
I + 8 I X(5)
X(S)
Ws
o: h(6)
-1
r--- ---- ---- ---- -.:;
I I

....L.--~-----+-l+
I

•~_..,,i.L--~---~;.:..!:..+:::o.-r---::x:::-~+-}-.::.h(7):
X(3)
I I
X(6) .,!: 21
W5
...;..'-~ ~- --- .= r.t ~~1 X(7)
X(7) ll~ ~-- ---_ ,;,~ .:JL -~~ -'-- ---= ~+ !}-. :::. 1 : -1
1:____ ____ ____ ____ _ _! I
I
-1
I
•-- -- --- -- ---- ---- --- Seco nd stage of decim ation Third stage of decim ation -
stage of decim ation
First
Fig. Ex. 6.3.1

(GTU-New Syllab us w.e.f acad emic year

7 4
21- 22 ) (G -l )
~ Tech-Neo Publications ...A SACHIN SHAH Venture
MordenTimeTech.com
 Digital Signal Processing (GTU • Sem. 7 • ECE) (Fast Fourier Transform) ... Page no (6·22)

Here g (n) is output of first stage. h (n) is output of h(5) - j 0.707

second stage. h{6) = [g(4) - g(6)]W~
The values of twiddle factor are as follows :
= [ 0.5 + j 0.5 J l = 0.5 + j 0.5
j 1t
1 - -
~ =e 0 =1 w = e 4 =0.707-j0.707 h ( 7 ) = [ g ( 5 ) - g ( 7 ) ] W~
8
j1t
= [0.3535 - j 0.3535 + 0.3535 + j 0.3535] (- j)
\~ =e-2 = -jW~ =-0.707-j0.70 7
h ( 7 ) = - j 0. 707

Output of stage - 1 Final output

1 X(O) =h(O)+h(l) =l+l=2
g ( 0 ) = X ( 0 ) + X ( 4 ) = + 0 = 0.5
2
X ( 1) = h ( 4) + h ( 5) = 0.5 - j 0.5 - j 0.707
1
g ( } ) = X {1 ) + X ( 5 ) = + 0 = 0.5
2 = 0.5 - j 1.207

g(2) =X ( 2 ) + X ( 6 ) =21 + 0 = 0.5 X(2) =h(2)+h(3) =0+0=0

X(3) =[h(6)+h(7) ]W~
1
g(3) =X ( 3)+X ( 7)=
2 + 0 = 0.5 = (0.5 + j 0.5 - j 0.707] · 1 = 0.5 -j 0.207
0
g ( 4) = [ X ( 0) - X( 4)] W~ = [½- 0] 1 = 0.5 X(4) =[h(O)-h(l) ]W8 =[l-1]·1=0

X ( 5) = [ h ( 4 )- h ( 5 )] W~
g(5) =[x(l)-x(5) ]W8l
= [( 0.5 - j 0.5) + j 0.707] · 1
= [½-o]co.101- jo.101)
= 0.5 + j 0.207
= 0.3535 - j 0.3535 X(6) = [h(2)-h(3)]W ~ =0
g ( 6) = [x (2) - x (6)] w~ ]<- j ) = - j 0.5
= [ ½- o X(7) =[h(6)-h(7) ]W~

g(7) = [ X ( 3)- X ( 7 )] W~ = [ 0.5 + j 0.5 + j 0.707] = 0.5 + j 1.21

:. X (k) = { X (0), X (1), X (2), X (3), X (4), X (5), X (6), X (7)}
= [½-o]c-0.101 -jo.101)
:. X(k) = {2, 0.5- j 1.207, 0, 0.5- j 0.207, 0, 0.5 +j 0.207, 0, 0.5 +j 1.21}
= - 0.3535 - j 0.3535
Ex. 6.3.2 : Compute the DFT of sequence
Output of stage - 2 01t .
x ( n) = cos where N = 4. Using DIF FFf algonthm.
2
h ( 0 ) = g ( 0 ) + g ( 2 ) = 0.5 + 0.5 = 1
h(l) =[g(l)+g(3) ]
0 Soln.:
Here N =4 :. n = 0, 1, 2, 3.
= (0.5 + 0.5) = 1
For n=O ⇒ X( 0 ) = COS (0) = 1
h(2) =[g(O)-g(2) ]W08
n= l⇒ X (1) =COS(;)= 0
= ( 0.5 - 0.5 ) ( + 1 ) = 0
b ( 3 ) = [ g ( I ) - g ( 3 ) ] W~ n=2 ⇒ X (2) = COS ( 1t ) = - 1

= ( 0.5 - 0.5 )( - j) = 0 3
n=3 ⇒ X (3) = COS ( ;) = 0
h ( 4 ) = g ( 4 ) + g ( 6 ) = 0.5 - j 0.5
X( n ) = { 1,0, - 1, 0.}
h(5) =g(5)+g(7) The flow graph of 4-point DIFFT is shown in
= 0.3535 - j 0.3535 - 0.3535 - j 0.3535 Fig. Ex. 6.3 .2

(GTU-New Syllabus w.e.f academic year 21-22) (G7-14) ~ Tech-Neo Publications...A SACHIN SHAH Venture
MordenTimeTech.com
 . ;tal Signal Processing (GTU - Sern. 7 • ECE)
o,g o
(Fast Fourier Transform ) · · · Pag
e no (6-23)
W4 =1
~ere, . . .
But there are certain appllcal10ns
· which selected
in .
-j27t - J7t
number of DFf values are required; but not the entICe
\'/4-
I -e-r- = e-2-= cos (21tJ-j sin (1t)- 0 .
2 - -J=-J
.
DFf. In such cases the direct computation of DFf is
more efficient than FFT algorithms. The direct
o&Stput of First stage computation of OFT for selected values can be re a1··
ized
S0 = X<) + X (2) = 1- 1=0 using linear filtering approach.
SJ = X (1) + X (3) = Q + Q = Q This type of computation can be performed using the
Sz = X (Q) - X (2) = 1 + l =2 algorithm, called as Goertzel algorithm.

S3 = X (1) - X (3) = Q In this algorithm the periodicity property of twiddle

k
factor (W N ) is used.
Final output
Now recall the definition of DFT,
x ( o) =s0 + s 1 =o N-1
0 km
X ( 4) = (So+ S1) w4 = 0 X (k) = L x (m) W N ... (6.4.1)
0 I m=0
X(2) =s 2 •W4 +srW4 =2
0 l Note that here we have used notation x (m) for input
X(6 ) = S2 • W 4 - S3 • W 4 = 2 sequence.
X ( k) = { X ( 0 ), X ( 2 ), X ( 4 ), X ( 6 ) } Now the twiddle factor is given by,
X(k ) = {O, 2, 0, 2} j21t
= e- N

_ili)-kN
= (e N = e+ j 21tk
= cos 2nk + j sin 2nk = 1 + j O = 1
-kN
Since WN = 1 we can multiply Equation (6.4.1) by
-kN
WN
N-1
X (k) = '°'
~ x (m) W N
km
·
-kN
WN
Fig. Ex. 6.3.2 m=0
N-1
Syllabus Topic : Goertzel Algorithm :. X(k) = '°'
~ x(m)WN
-k(N- m)
... (6.4.2)
m=0

-~ 6.4 GOERTZEL ALGORITHM

:··- --- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - j
Now according to the equation of linear convolution,
00

! UQ. Explain in brief : The Goertzel Algorithm.

'' GTU - Q. S(c)(OR), W - 19, 7 Marl<s
Yk (n) = L x (m) hd n - m) ... (6.4.3)
m=-oo
----------------------- ----- ------------- -------------------·
- We have studied that Radix-2 FFf algorithms are used We know that hk ( n) is an impulse response. Now let,
N - kN
to compute 'N'-point DFT. In this case, 2 log2 N
hk(n) =WN u(n) ... (6.4.4)
complex multiplications and N logz N complex Replacing n by 'n - m' we get,
additions are required to compute N-point DFT. - k(n- m)
hk ( n - m ) = WN u ( n - m)
... (6.4.5)

----- ----- ----- ----- r~-- ----= ----- -

~
(GTu.New Syllabus w.e.f academic year 21-22) (G7-14) Tech-Neo Publications ... A SACHIN SHAH Venture

MordenTimeTech.com
Digital Signal Processing (GTU - Sem. 7 - ECE)
(Fast Fourier Transform) ... Page no {6-24)
Put this value in Equation (6.4.3). Here u (n - m) is unit Equations (6.4.2) and (6.4.7) are identical. Thus we
step. So beca use can
of mul tiplication by u (n - m) the limi ts of conclude that DFf , X (k) can be obtained as outp
summation will become m = 0 to m ut of LTI
=00 •
syst em at n = N. (yk (n)). But in this case, we will get
N-1
X (k) only at one value of K. So we need to
use parallel
:. Yk ( n ) = L x (m) WN
-k( n- m)
... (6.4.6) syst
.,...
ems to obtain X (k) at different values of K.
m= O
-~~: :·Note__: Goertzel Algorithm is efficient when X (k)
Now we will calculate Yk ( n) at n = N. ,; "''. .:~ • ".~ is to
N
,, be computed at ~?ints less than log .
N-1 2
~ -k( N- m)
· Yk ( n) I = L..J x (m) WN ... (6.4.7)
n=N m= O
Chapter Ends ...
DOD

MordenTimeTech.com