E=}x(n)
n=0
=1+1+1-3
And if coinputed from frequency domain, then:
N-1
E=NX«(k)|=6[0.25 +(1/9)+ (1/36) +(1/9)1=3=Efrom time domain
k=0
N-I
N-1
Or we say in general tha.: Xx*(n) =NE|X()|
n=0 k=0
Fast Forier Transforn (FFT)
This is a fast method to find the DFT.For FFT base-2 (radix-2), then N must
3e a power of 2, i.e. N=2' (4,8,16,32,64,128,256,512,....). If N¥2, then 0's
added te compleie tùe sequ:nce te the nearest 2 value.
Algorithn1 for FFT radix-2 decimation in time:
N-1
1
First, we know that X(*) =xn)e
N n=0
N
where the term (1/N) is a
N-1
N
scale fuctor that can be omitted, then X(0)=x(n)e z
n=0
2T
If we detine WeN, to ease notation, then:
W
n even n odd
ior n even, nern n=2r and for n odd, then n2r+1, v:here r=0.!2.3....
X4)=>x(2r)r+ x(2r +1jwGr)k
Note tlha: N
'e34= =WN, then:
X(k)=x(2r) +Wx(2r + 1) i
r=0
X(k)-Gk)+WNHK).. ..(1)
where GK)=DFT of the even numbered samples (N/2 samples)
H(k)=DFTof the odd numbered samples (N/2 samples).
Equation(1) states that the DFT ofN Samples can be obtained in terms of the
DFTof the even and odd numbered N/2 samples. And if eq(1) agan used to
iind the DFT'sof the N/2 samples in terms of the DFT's of N/4 samples. This
process is repeated many times (exactly r=log,N times) until we end up
wiih the DFT of 2 samples.
DFT of2-samples:
2r nk
Ifx(n)-{x(0), x(1)},then: X() =x(n)e
2
from which:
n=0
X(0)-x(0)+x(1)
X(1)Fx(0)-x(1)
Using signal flow representation, then:
x(0) X(0)
X
4
X(1)
Time domain freq domain
where the unity path gain is not marked.
The 2-point (samples) DFT is called "Butterfly".
Note: for N=2'point FFT, then segmenting into even and odd is done for r
times and the number of complex multiplications in FFT will be r.N or:
Number of complex multiplications in FFT=N log,N.
25a0
Note that, the great advantage of using FFT is when N is
larger
Theprocess of finding say 64-Doint FFT is started with 2-point
PFTButterfly). This is then used usinc eg(1)to find the 4-point FF!, ana
again, this 4-point FFT is used to find the 8-point FFT and so O.
Ex: Draw the signal flow graph of the 4-0oint FFT, then use to it to find the
spectrum of the sequence x(n)={1,-1,2,3}.
Solution:
X(K)=G)+W. HK) 3>k0.
The everi numbered samples are x(0) and x(2)
Tie ci nunbered samples are x(!) and x(3)
X(0)=G(0,+W H0)
X(l-G(I+W H(1)
)=3:2-W?H2)
since ((2)=GO)and H(2)=H(O) where 2-point DFT has aperiod of 2samples,then:
X(2)}=G(0)+W, H(0)
Z(3-G(3)+W.3H)
Ard agaiu G(3)=G(1}and H(3)-H(1), then:
3(3)=G(! j+W. H(
A
x(0) 2-point X(0j
DFT
x(2) Buterfly Ge) X(1)
x1) 2-point X(2):
DFT
Butterfiy X(3)
x(°) K(0)=5
x(2) X(1)=-1+j4
.
Sorjugate.
x(1) X(2)=3-2=1
x(3) X(3)=-1-j4.
3
Check: x´(n)1+4+1+9=15 =energy from time domaln
2X(*) '=(25+17+1+17)/4=15 =energy from frequency domain
Ex:Dravw the signal flow graph of
8-point FFT:
SolutiYr:
The 8=point FFT uses the 4-point FFT obtained in
previous example. The
sequence of sample decimation(splitting into even and
samples) is obtained using mirror image of 3-bit data as:odd numbered
00 0
0 x(0) X(0)
1 00
x(4). 4-point
0 1
x(2). FFT X1)
1 1 x(6). X2)
1
1
00
01 1
0
0
1
1
x(1).
x(5) 4-point x3)
X(4)
1 1
1 0 lo FFT X(5)
1 x(3).
1 11 1 1 1 x(7) X(6)
Time
x(7)
Mirror domain
Frequency
domain
6
3-07t+jotu
o70tja407
Eo:lo} -jo.?o7
X(0)=G(0)+W;° H(0)
(0)
X(1)-G(1)+W;' H(1)
x(4) G)
2) X(2)=G(2)+W H(2)
x(2)
2
3) X(3)-G(3)+W: H3)
x(6)
HO) X(4}-G(4)+W;*H(4)
(1) =Go)+ Heo)
X(5)-G(5)+W: H(S)
x()
=G4)+W H)
X(C)=G(E)+W,° H(6)
X(7)=G(7)+W:' H(7)
x(7)
1443) - G3)w H(3)
