You are on page 1of 32
Tord rr poTA oF DFT EFFICIENT COMPUTATION a Prowrenr STRCHENT! — Given N Compl numnbers na al Sx},.. goo” (xp ““\orr ovifut yout We know the dah we x -hn = m Yon! S72) Ww ond sty TY X00 neo Fi —w - n/a . wree Wns é ya eHivent a\goxi roms fo solve mis wisn & find we Problem it: ger {xta}e-e oars io) ae or Viceversca, We Will See Yee appvoache, fas COMPUBNY ‘DEY. 4, Direa Compuransn. Evaluating ke equation b divectly 2. DIVIDE 2 CONGVER APPROACH ~ Fast FouRIER TRANSFORM ALQORITHNS. a. GOERTEEL ALGORITHM, : May 1. “DIREET compUTaTION Given treo 9 {ws coon k, NG ha xCrrz So ™00) Way n= ip Vequire, N COmpux MULH bu cohons, N=) compux Adel hon, So total fw alt , O *THe ognenr osactl ae ee beme el oO SméMq Mabying = TEA wz Lawl. Sivnitary DET volves X Ck) can be vead from ka matnix Values X [P14] Cittey vow wise oF coluinn wise Row wise Column Wik xta > X [ha] Xue] 3 Xf) R= MPT4 ke qb + pe Now, wedevioh this abbyoatn “slog on trakmpie, Anst N-15 - 5x3 Se 15 bolnt DEF Can be Combukd Using simadew Spomt DFTs ong S boing DET, Wout x= Patan... = = fH) } Output YX TRY 2 { Xb). x¢9, wenn es Ling Y SEPA. Write XO) Ina Lyn Matrix where Le Si h=3, let us use a Column wise aPbroath, feo xa | 08 (10) «ty | aC) | aca Les gr | xO) xt) tog | Ry Yo) ash aa) 103) Meo Corn Duke N= bont DET OF each vow Yo Get Ono hey matnix A. OI \ LO | x fio} => A fo.) VP xeo Yc Pan ayn = AD Rae a 2 | XO) | xt) xf) Deets A (2) — ‘ 2 Gea porter! at sf eet | each A (3,5 ur Y [tay |rtsa (agg | roo ATa3 mM=0 y 2 b-1 my Tah, ALLA? > Lm) Wy M=0 Se Sirentel ond P= Ore 1! STEPS utnby each eament of ALLA] by 2 bhose foros WAP 5 Le oye bY ond 9 = Oty =n Meh B= WyA Point by Point rnulnpican, Wye {Waa} motaix ue obtain, .. LX) B matinix. ster 4. — Compute Kk hoo DET OL Och Column of ta 6 MON te 0 | Bloo}| fou) | Blo) 8 7 Ble] Bh) B79 > | Biv) | Bld] BO.2) 2/80) | BR] eg S| Blus) | Bin} ete mM: 0 \ 2 vb WwW wv Loont Lpond Lhons DFT Ort DFT \n woth Lb X [pa] = > Bia Wy, L=o pe Ortaer bel $ Ge ON, Mt srers Reod xX maine, TOW wise ro get X fk). [xt] ¥i0.3 [xtued| x ton | xtwen Xfr0) (Mo) | x f2,2) Xn) Reoa Row Wise Fr summarige de sibs . ALOORITH i: NeLxn ~ @® coluron. Sh ove “ke inpur sequenu Ow wise 10 aymatn'x Sz © Compure ke HM point DFT of ¢4h row fo get o LX moinin A> 3) Mutnpy ka eaon entyy of A by Phase foctows hy Wal ofhE tT ana 0 ea £ M1 YO get a new matsix 8 84 ed Compur ke L- point DFT of each column of Boinx fo get X watnix 6) Rena X matnix ow wise dw ger &X wo R&R ue g TM We angee ke mato} Algosisma 1 as input — Column Wise ke ne Lem ourput «Yow wise 2 R= MP+Y “ ne X [ek] = > Cn] Wr neo ens 49: oo Me ui La Mette X[rale S_ > Dim] Wy : m=0 120 bok M-1 LMb dq Lttmp mtg = > 2. aEim) Wy + Wy iy Ms Azo care LneNn Arwongig de terms Lot n- . SS Mt ~ 4 XT) = > fF Wy f Com wth | We ‘ Leo m=o — : y ; a hi pewe DFT bhase factor pon voutnby L bony DFT We need > Combuk ony L2T pon dere. @ Comburational Combixiy, of Algerithmd. (A 21S) ® We do five B-point DE Is. Combe wuttipi cane = IXBNT © US Complex addins = 3xarE = Bo & Matrix mattipti catins LAM = 15 compur WELHpUcanme @& Thre, S point DET. Compuxr mul pUcadoas = 3X SAS = FS Compe addivme = SANSXY = 60 TJotot compar malope cake = 45}1S475 2185 N@eu4)) Wn Qrnerat Total compe addinms = S460 = Io N (Mair) AW genera ALGOoRVTHM 2 We can get Arohey agouittin by weiking Input Yow wise in a tmatmix And — Teading our out Cet ase {nom be rnainix: Ne Mham ke gut pb. Im marr Me (ne+m) (4b 4b) XEords 2 2. xfer Wy mM=0 L=o yr Reowe ma ta bm yy PA n = So afar) Wy Lo Wn 5 lee ee a a > * fom) wi} Wn > Jug Ss tz 0 ee prase eee Focrac epihpua Mm point DET Wqoa\tinn? . Ne Lan 6 ly Ae \ , ~ SWoOve fa Signed Cn) iq a matnin vow wise. Gx) Combure Le point DET of each Column, st () Wathply Me Yesurting metab by WP, Os Pele), OWS MI . SH) ompure ee Me bomt DET of each Yow. e ) , BS) Reaa be Vesulhing matrix | column wise Ie act XIM- 4 We can continue dividing N inte smaller Value, Ne Wy XMX --- X MV and use an of ke NW algorithms successively, One PorrNewtow Important case ts when Ny sys =H y T prime nurever. Nea Jy ris Case we ned fb Cormpuse ony H- point DET, , ANd ke compuratm of ke N-point DET hos & Yequlery Vater. These ANgositins owe caled — vadix-m Alpritnm and mis cabled Ke odin 04 bee FFT a\qoxi Nn. Common Used Fer Algoriting. R= 2 Radix-2 FET Al gositnm H=3 Radix-3 FRT Al gost thre ney Radn-4 FFT Algoriim Fos each (ase, we have hwo O\goridme Sehemang Ol we Wik Input Column Wige oe TOW wie, N2nxXNn. ‘NaN = N Bexie Deetmaton ja me Decimaren jn Diy Rl goritms begun Dip Algoviin in es ALGO Riri N= ax N/a we i i 4 Fru de inbut into a x Nh matnie column Wise ie. We are dedmaring Yee 200) in dime, by 2 —— x To} 40) “Wo subsequences yf) ong a2 f} of lengh Nib N Qo) x0) AO Ge, 5 REO Nyad x [ons] We need bb Combu X Lk] Using AiCK) ono ArCk) N- Rn X(k} = > 10) Wy h-0 kn — kn = SS) xo) Ww + 2 *H) Wy = PL 2km : x Dm) W in fisr bone > xe") : > " pone 204 wo mv in sccond vpcl Lisa km k 5 = >_ alm) Wan + > Or fm) Way, ws Wee eo u k = Aik] + Wy Ry Lk). ke 01, cnepet Sina Ay Ek) ano Ar(kl ave Nj, point DET be eno wih W/.- XU) = Av Ok) F Wat Ar th) Nh a X[ke ND = AVCkY + Wy pled. 2 Oe z nie 2 ACO — Wet hfe Shmdarty We can Yeheart He peocess UnbL We get a 2 boint Sequence . by tr) = Afar) arty) eee — but: ahaa a J bat) = 2 [2n) ah NN brie) : = 4, Can4y) Niu Now » by fn) , rr By tn) bre toy 4 NIN Bn Er boy ¢ NI Bas CH) bam < pitt » Ba Ch). A Me can wrile, ACR) and Ay Ck) in teome of Bis Oo). 3 e - 26-2 Fis exarnpte N=-8=2 _ bnh fol, *fa) [2 poinr Aube an Sequere] . x [0J,%* (2) x{9) mt (6) bn Cay (6) an fn) 2 —»> bal 21 (15x05) HOI 03) mos) x04) bar tn) N aC a4) By: f0) X ©} x6) —f wen BuO), xis) —|_oet | x XO) xG) x tw xt) nh) Bir fel ante) —|_dFT 2M Nor Xx 6? x15} ot xCe) xl Wer ti] — aeeaee STAGE 2 STROE Mier veoching A point sequeae we netda 2 doint dE, ae A ao er 4 ——| °F 8 LU A= aoe Woh A 8 RADIX-2 BUTTER FLY. Now , enband ing OU Ke blocky, We can get O& Compu Signal Mow aroph of Neg pont DFT using Radw-2 Dir FET Algoritim, ddd aBLidony Lia w-xavd @=N ont Syst SAE wx a LC we (ee as ns . BK b : afm ‘ - j Wier - <—“ong : Che gn SK. OX - TT aN - ae * (a OFA (yg on em fA’ ~~ - oe q's nag : S (x < — tg ox _ _ Cg \- é (AX er) tog Peoarannina THE FFT basic Tre bons (Ravi? BOITERFLY- Te i ONIABRONG combutaten \equired in Vadiw =? DIT FFT at eaten shoe Is aq vadix 2 — burner (ly 0% Shown below - Jaye is a Noo Corsi Number (a,b) and output Xs 2 Commas Dupes (A,B). Each bulieeh Cotains — | Combux rwullipucatm ane 2 Compu addin Total compus MVE = “No of Staje xX Buhertty pew srage X Combux Mut hes buter hy “= (log,) x N pooh C oe ) = * a & log. Tota compur Apoition= (log ad) x Nyxr n = N log Ww IN- PLACE CompurATION— Once a buted{ly Oberorn je Verformed ana pair of Combia numbers (Oi) , We do 04 Weed BW save ke inbur pay (arb) and Crone He result Cas6) in Ka Some locanon at (ab). Fo for wadix-2 FFT Olgaeitte. Wl need & stove ony N Combis numbers oe 2N Menon values. “This is called in-plou computations ® BIT- REVERSED ORDER, WM Dit Cdedmatm in hme) FET algosinw, Atty y-) deamanen stage inpur , dora Sequim is shubled. te. input SeHLENLE Order Xf, CH), 107, 1667 0D KC, G7, x) Our but Sequmee is in Noval Order, X fo}, XC, x02), -- ~, X067, x09) We Con show thor inputss in bi} Tevenced oxger. DECIMAL Natura orver Bipaay BIT REVERSED oO oO 000 000 \ 00) too 4 2 ovo O10 2 3 oN) )\o 6 4 \oo ool \ s lol Jol 5 G lio on 2 + yt VW 4 A BIT- Reverse ORDER. So. if we Stont Wik npuagy. in bit —Yevensed creer, Wwe wi ger output Im Naluval osdew- ( trewever, we can force dnar bole Inbur | ouput con ve Wn Narwro) owaer) Fi atlON mii AvlON IN FREQUENCY RADIX-2 FFT ALeveRItHn. jee Wide EM} , TMA Mat Yow wise Tn DIF algpsittrns, we dry be dedmar XCk) by 2- Let Atk) = XBR ee Oy No? prt) = X(2k+) Now aytn) <— > All OFT art ¢——te > Arle. OFT write Ata} ond Arr) Jaterms of XE). We wort © N-V a kn XL] = 2 fod Wag ne © NE ya kn an nfo) Wy + es ne Nii j he N- Nir in second port Nh-! Nh R(AENA) RO : 2 OV eS onteetnd Was Leo Nha ole Nak 2 > nO) Wee W . D> x for Nh] N " ed neo Wha = > Sout + ee wae nee ° wh ao Wry a) Wha 2: 3 {xin _ | matax wnt} Wise neo . Now Ay Cel = x Tae) Nir4 = a Luo + C1) a fae] ee Wha . ke 7 > {xm 4 n(neenr) Way Yeo ue avi). Oya) =. fn) + xfns Nir) <—> Arfk)= Xf2e) Similaty. Ar tk) = xX [2eH) On {ny = Sato . acne d} wer <> pil) =X (peri) sively Bute) = Aif2k) 2 Bale) = As Ree Buy Te: Ar {2 % Ba CH= Ar (2a) ond Wrik byte} 2 bith) —fareemsof Atm), bay try & bar Cn) interme Of 4% (nd - 5. Buf) = ALD) Fox enampu. = N= Q. ¥ 00), X Cay fee, CR) Alle) t ' BaCk)= Ry CR+i) Xf. X09 Xt) — XW, XB) Dx £6) ——> Bay [ed = Ar [20] XG1, X Cs} ArTk) =X farts) XD), X33 a xs}, Xt) Baz Ced= Ar [arr] X B), Xf) = 2) xt) —] ¢ 7 a af} —— 6 nH [bul] ce x &) xh) — § lye ge | Be xy : M Im! 5. ow af) —— % [on ®_} ope [88 ex) Q [ors ™ Cy) —— | Q, (0) eS a SL Rn) xc § Paw = BuO x0) nes N : oFT n (6) E bn ay Bul) xa e : but | oe. xe KF a ———— Block DIAGRAM RADIK-2 DIE Naa ae Nye alveady seen de 2 bom DET ay 2 — A on b —~— 9 _q-weh. We CH also wrike Az=- aah g ” @-3) We o> WP =| A Rad 2 Die 8 Baterty FAS suey afk fom Yaaix-2 DIT butinghy Sine it is ke Some operation, tee reed to do in ohex stage. (9)X Xx (ny X (1x & | Qu @ta Coa . oven = on Qe ou wtq onean (ge i : ara oa gy Gielen (he a) IT os 54 A=gM ye uaa = ‘ og 3 : wuq toe (oe eno (tg (2) tq {30 (w]e NAT I~ i wg o"'4 Lo \ {wlo (1) * ~ (gq. lo} So {9} % ("9 WW summary. \MPorTANT FAVOTONS foR RDIX-2 HIT Qin) = xfen) ~~] Ai fh) X fwd. — > Alm) = xfrnsf T B, (kd. X Ek) = AiCky + Wy Aa Le) | heOnn , X (eeny) = Arte) — War A Ok) RADIK-2. DIF Aifele X (2k) t = Tovey ee X Bars) a re atm ah) + Coen) } N= Oy), -» he Ort) = } nih) - xf Nan) } War WOFRTZEL ALHORITHH ‘uselul When Wened W Compute Xfe) far 1 value ones can: my Chu ole 12 QL DET value sitnultaneouny » HOwewa, “en wena X [nd vale for fer by indide, ,we con ve Goerhy However Goeveyl Algo 5b algo rite ox Vivek wethod Xesulb in Some Temon SING Convolunder - We con do DFT using a }— 4%] SiO, n-N key IDEA: nt) hf = -kA uolnere hy fn) = Wy &lrd. Na ae = 2 w We cat) show trot Yel laren X fa) nt . LHS. yytrd = *Cr) * he &) t ° = k(n-n) = > afta) Ww ufn-n) _ 2-8 we ony VOW . Na . hi ona. vue | 2 > 1) wee a neo = XCk), 2d Wee: = £ hytor] _ _) -k \- Wa 2 7 . non) a ond — difnence equatm . ca YO | Ww FIRST x(%) Jawa" 24 OwER . . GOERTZEL ok ALQdRImHin 2 YR@) =8Y,@) Wy +X taking inverse E-bans Fom. ‘ -k t r . Ye = Wa Yalo-) +m) Ll. HLGoRITHM. Assume YL 20 : _ . _ Foe Re Orly-W \ Yin) = Ww Yalhy = Wat Yana +% (0) 46-0 ends = X[k). FIRST onder GOBRT IC n L Alooriran. (PRomea rns) GWEN NpuT fxa},®" dO We ON, oN ( 4nGi <0) Yated © Yutoad. Way 4% fm & Xfey = Ys = Ww yy tw). Tera no of combux Mvlbbu'carnms = N Fa each XC}, Fox N pons DET > Comply Mut= wan end (same ay Divect computa) . -k However fr cack R, We Med only Ww Phase four as (t do Mot dehend on nr, (memo vy saving) ; Appvcaim of Goeryel Plyori brn ~ Whene we Netd to compur few XCk) valves . We dar’ ates hove second ovden Choead yet ~—_ Aqor\in | cy \ | detyh et Wye) 2 ow VeWyo et be whet —— \- Wy V4 ae = afos 208) #4 a han a eat’ CoMictemt , Secause | densiminarire Somme Saving in Aurmbew of Combux youth puicalin will be Yoeve,’

You might also like