alAssMate

chap Fast Fouhien Taansfonm (FFT Date

Page

Mutivation
a dataBeq ofize N a clisee t comput
ti'en of the N Þoint DETAeguhey Qn the
LNN-L) complez additons onter
N( N) N Com þlex wwtthblicationy N= N
Onder of N auithmetic opehatios
FF T aly0 itth ms 3 foa the ef ient
Comput a in f the DEL

in the DET Computa tiog byexplaiting lcestainy

y mmetn iea) of the ous ies
and only he
heswu'he on the OAdes ef Nlu,
as th metic opehatons
i NJog N complex mult plicating
adoi tions )
EET algonith m pub li's hed in
Lis heol ia 1965 by couley
euke
Numenous afplicatie ng.
EET& hauld not be thought af a neatasnfen
mehely algonithm
the
Re lies 0n the conCept of divide f cenguen
t is obtained by bh eak ing the DE I of
Amalleh A'e DETA
fAs hul e no. tht cante
Two esgeutial in gheolit
() N must be 4 Compsite nm be N=b =2x3)
Not pime na
() periedicity audAynsetay phapeatio
the tac tuh W must be
K - W KtN e
N
W
LK K
W,

--

W
e
 cAssmAte.
Date

Ttes of FPTA Page

N2

Decinatim in time (DLT).

Decination n fheauemcy (DF)
(i)
e9-Y 3,4 that isdivisilk
and 1
b itef
where Pi e r I Me num bes

Mxed-Y4dixEEs
Radix-2 FETs
a) DETA Is delombosed into a Cas code of
stages
(6Each stage is made-up of 2 2+eit
DEI (DET,)
(DET,)
Radi- 2 FET via D1
Mun ptes of DIT
N-1
K
K-0,1, --, N1
(O Split the ite neve tZott
(i Exþhess tthe suma and oodt
Paopey Cembining)
Else
Aepeat the aboe Ateps fu
Ik dividual Npo int DEIa
 alassMate
Date
Page
2-po t'nt DET
te)
N= 2.220 a
XLK]
Since Wg =
X1]= xlo] -xL 1] =Xlo]+( L

KY

4 -Point DET
NE 4 = 22 -2 , = 2 (-2stages)
step 1 2K
3K
+IL3] W4
t
( Lo].tz [27W W L 1 )3]w)
Step 2
N
4 W4p w

CLK]+
whens
HIK] DeT
DEI,f oddsamples'
2- þero ic
Nobe
G[fn] xfo]t(h]w*
H[K+2] : H k
Step Sinc e N 2. We sto p That
GLKI R HCKI C
2 put DET CLKI Cannatbe fuatha
simplífied Uia DIT.
 clAssMAte
Date
Page

XCK] = GIK] + Wa HCKl, K=o, 1,2, 3

t ey ony need t be omputed foa Keo,4

Hence,

GC11 + W HL11 2

CTo]-wHLo] = X;IRI = Gl2]+ W¢ H[21 =(lol+W H)

aL1]-WHU] = GL31 twH3] =411] +WH
Note tthe &peeial ordering 0f lp data
i/p
W4 1 -W W R
Kt2
Fun K=1
Wa=-L 2
J = -Wi

DFT
WA

|DPT, X[2]
HL1]
(DET, bloc Ko e
 classMate
Date
Page
Geneal Case
step 1 : Even and oddsamples of x Ln? Can
hethesented Aestectively by the seauen ces
cL2Y] L2r t 1], where rY = 0,L, --,N
N1.
X LK] - Kn

put n2Y N-L N-1

N
25L21] W ial2r+ 11,

step 2
t271

whene
CLK] DETI. [2Yl,-0 1,
HEK] DFTM L2r+41,r=0,1, - -
N XKY-XK-1
ony be lo mputed foa K=0 to
Thus, fh K NG
XLK] =4LK-N
step 3
N
i3e Til , 2?)
 WT wf
XL?
zl2
point
2peint>o
[3] DE
W
>Xl0]
staseg
acrdinglystaget
odifed be
mustI/pse. the Ohden
of
FFT. DIT int po
Replacon
4
pess Second
W
L314pointDPT A5)
XtsT
X4]
X[3]
2
X1] X{41
Tfo
Pass. Enst
EX
Poge
clAssMAte Date
 elassMate.
Date

Combutat'onal Co mplexi t Page

A
N=2 : lo1, N steg es
Eacu attye 's made «p of basiC binaty flow
grsfhs called buttef lie.
Theneis y omexmttiplica tion.
a ddition pebutterfly
Awote
staye2

2 buttefly
1 red +1 geen.

N2xo =1
2
2

W
6 e
Verfy NOte
fpmabove
Staye2
staye3
W -1
WN
w 1
W
W
W

w,
 2ompormpCatl classute
Date
latpa. additln Poge
Alctool s.
2Ny los, N Nl09, N
2Nog, N -N9,N
me,

Howeve: Onde Ss n log, N whicth is sam

in too Cases
Proaki Modified buttefly with ody orOne mlti lcation
N
For any butt
Can 5e drawn
figure n tHS
-1
mutti pliction
Ohdening of iP dat a with W
IIp data is anhanged. in bit-ewehsed onde.
Bit n'
BingyeleA
Bina

10 D 4
2 2

4 10D
01
6 o1 1
111

tukA w

A1 As +tmp
 neAement
Storase ..Atbave ip an
need stoaeAassMAte
pain mY be

hcgite ffneteheudast
becs
buttfbyCo miputatin,
P rogham
( phethocesAing to addhes the dta in bit
Aevensed Oh den.
stepu) An owten loo ovey the stage numben,
fam
step fii) An innen loop over the buttety nmben,
Lsay fhon K=1to Nstet het the ste cifc
indi'ces of the buttefly ie depend onth stge
mum beh l

Decimation in freauency (DIE)

Anothen Way of de compos ing the DET Comþutation
thatadgonith mhas complexty ol Nlog, N).
The phincip les aue ver similan to the deimatioy
in time CDIT) alonithm
Beaic steps
ABsume N=2Vie integerly>
pantition DErÁanj les xIK] (K= 0,4, -,N-9)
two ub seGuence
"Evem indexedsamples: x[2r],Y-91, .-. -- N4
P odd inderedsamples: X[2Y t1], r= 0,4 ,N4
2 Expn es8 each ubsesuence as a

size N.
3. N 22 to else, apply steps ()-3)
to each DET Of s3

Wenote that ance the olp afa Partiatau buttey

have beon comuted, the CUn nesponin i/ps
me no lone needed Q Can Le oveAWYItn

1&3 Oe not ed otthe buttety

 A
ondehhevensed bt- X[K]
in oP
hct o(wlor,
DiT of ofComputatonal
N) complexty
Wi Remanks
-
L31 X
L1] x
+1
-x[r+),
Y=0,4 WA
i vtol, DPT X[27t1]
=
amplas DFT indezed odd for Jay, same tCLn
he
1tLY+21,
=0,1 Y ULY1
uJt uol, DER where
2
txl3]+(xC1]
) [2]) tx
(xlo] [arl- X
w Thenefone,
N,
W
Wa
Note,
4 4
W,Y=1 0, 27LnJ X[2Y]
mples 2nY saOFT
dered ewen-
jn Fon
N=4 Case
Poge
clASSMLe Daie
 alAssMute
Reyefan Date
Page

EsA Co mpute tte 4-oi t DET of the s Punce

DIT FFT aho nithm
no.0f Comlex multiplicatin
the no.
Aso, determine the
Sol N bit
Lit-newenyed onder
Olp: NOnmal onder.
4

-2.
W
2

ao)-10
NO. co mpl mutti Flicatos
X ) -2+i2
4 log4 xt)2
2
= 2X =4
Euye amp}e usws ing decin atibn sinn- fa esueny
FET lgo thm

4
2]-2
w21
-2 WA
X]-3*
2 Wy XI3272

6r W Y
 Obhenher
2n/edition

alassMAte
Coentzel tgonithm Date
THaw the Peuiolicity Fie gw,ken an
Motatior sa toheduc owutetion)
aiven aBmal n] with cohnes ponding 8-Point
DET XK], What isthe most eftiient way
i ter of totl mittip
licatin Aegned to
Compute XL5] fom XDn]?
Note thetsing a hadi-2 FET dgorthm
the
isn'test ehoice, jdicios apb icatin
claim tte
beat it

hegines 8l9,822t
complex mutt'plica tonwhen em loying
8nce X LK] fer k so, 4,4
-mwstfist be computed
Howeweh, caleulatty

eo mpled' Muttiliea tbnd andis stainKingly

Ain plecompahed to The ade ET lgo
Choentzel gathm
Tt is a etficient method Cin tem of t
K
muttiplicatios) f eomputing xlK] foa aglen K
mi'ing xtKI fon 0me K
N-1
XCKJ =

k(N-r)
1.1
lines
YorK <-o toaoin
Convoluto.)
 elAsSAte.
Date
Page

FLOm Land the foct that

CLTL fittenig ÞADceLs)

Repheseatikg thes flte by zT.
1

Astest of-o t
dela dontConfue)th
elenent 1- WAtx)-*
K. n-y =XLn]

Cinitil nest
Ftow gtsph
condition) httibiahay addit
Hl):
1-W L
1- w.R
havy
Rloson Aorovplens2X) z14z-2

Resonat

fovhealizatibn
 classMAte
w owtpex Date
Page
Comyle.
Twoveiste ddbe
Not CawmtednE?)
3

2ade.4 sel-add
4 H a e t e Venin vahesbecqus
depet tnetesttrtn
Jrn)
Atisty, mast be
c comþuted} in og cde to co mpute yK WJXLK]
s s o , 4 N)
tthan
ealmuttipliatian.
4N Nnesl addition
Res
eAPole

dd
4aslde h mutti
facton-t neet not be ¥vnPoMy
not
Atee Nth itetà =

add
TotalC)) hel muttitatoiy
4tN+1) ea additions
Aunes Namplex. matt
c DiheetDET m ethod 8 xdug, 8 3XR
aprAo mately hatf DET nethod.

MattiN 00o
 Ophenhein 3rd edition
Page

less efhiieut han tho

dinect m thod. Ho evee, it avocs th
Co mpyation o &tohage of fhe coeffitie
Rn
WasIMe ethese quantities
Computed bthe he curs l on.
(4.1
For fole
Dfthe p is complen, ondy two ncal nuttir ica bug
per sa mþle ne Aequned to in ple ent
the
þoles of t s yte,&jMc e tte eseffuenta
ane aal ? ttefacta -1 need inot be
counte
multi licatoWlj
-

Since we only neeel to bhing the ytem

to astate faom whichyLN1 can be Canpate
tte co wn plex nultaplica tibn by -WE ' eqined
ti inplement the 3eno of the system fu
need at be pefoamed al eey teatiox
of the dfeince e" but oy aftea t
Iterotio.
Thus the to tal computatn i£ 2nNAeal

The totl computation thenefonee Cw)

2(N +1)eal t, about Dhalt
the no. of el Aeqeawithdaeetiethot
ttiueut seheme, we stl have tte
adu. thet Cos('AK) 2w aette oly coetf. tht
muyt be Computed <Atoaed. the coeff w
ane again Compuotet implioty in tte iteatian.