{LE TRANSACTIONS ON CIRCUTTS AND SYSTEMS, VOL. 36, NO. 9, SEPTEMBER 1989 (6). Nac, PB Jana, and R. Nan, “Nove ating ial tunable FDNR Sionolation using cvtent conveyors.” IEEE Travs. Cirts Spt. v0 CAS SI, pp. 08-403. ps. 18 (7) YSACN? Abdalla, "Comment "Novel Noaig ies! tunable FDNR Simlation using curent conveyer” TEE Trans, Crs Sit, ¥0 (CAS, p. 308. Mar 1985. (8) R Seman, “On the resloation of foting ate elements IEEE Tras ‘Cieit Shr v0l CASS, pp. 32-326, Mar, 186 ‘Computation of Discrete Hilbert Transform ‘Through Fast Hartley Transform 'SO0-CHANG PEI axp SY-BEEN JAW Absirct —X new fast lgorthm is proposed to compute the discrete Hilbert trandorm vin the fast Harley transform (FHT). lastead of the conventional fast Fourier transform (FFT) approach, the processing is ‘ried out entirely in the real doin, ako many efficient fst algoithins exist for computing the disrete Hartley transform, the computation com plexity cam be grealy reduced from 2 complex FFTs into 2 real FHT 1. IwrRopuciox Hilbert transforms have played a useful role in signal and network theory, and have also been of practical importance in various signal processing systems; analyte signal, bandpass sam: pling, minimum phase networks, and much of specteal analysis theory are based on Hilbert transform relations. Systems for performing Hilbert transform operations have been proven useful in such diverse fields as radar moving target indicators, analytic signal rooting, envelope detection, and generation of the phase of 2 spectrum given its amplitude. To be able to use digital signal processing techniques in these applications, Kak, Cizek, and ‘Oppenheim [1]-{3] have developed the one-dimensional discrete ‘Hilbert transform, analytic signal, minimum phase sequence, ete, Tor the time series of discrete samples. ‘The conventional approach for computing the discrete Hilbert transform is through the fast Fourier transform (FFT), It is performed in complex domain and needs many time consuming complex multiplications, Recently, the discrete Hartley transform (DHT [4] has been considered as an interesting alternative to the FFT for spectral analysis and fast convolution of real data [5 and many efficient fast algorithms have been developed for ‘computing the DHT [6)-(}. In this paper, a new fast algorithm is proposed to compute the discrete Hilbert transform through the fast Hartley transform (FHT), Since the processing is performed entirely in real domain, the computation complexity can be greatly reduced from 2 com plex FET's into 2 real FHT's UL, DiscReve Hiner TRaNsronM By FFT. In the continuous case, the Fourier transform of the Hilbert transform f(r) of f(a) is FLA] = FLA) H(w)-FOw) = pose (we) Foe) a) Manosciptsecived October 3, 198 This eter was eecommended by ‘Tawar Univers, Taper Tana, Repubic of China ao 1281 ade [fore] 08 peer te] [ser] oo essa tides spf swt] 8, Fig. 1. Implementation of dicot Hert tansform by FFT where - wd Hw) =jsen(w)=) 0, wa @ lei weo For the discrete case, Cizek 1] derived a discrete Hilbert trans form fin) for a periodic sequence f(n) for which the above relationship stil holds: pet[f(»)] = DFT (DHTLA]) H(k)-F(K) = j-se0(k) FUR) (3) where (4) So itis very easy to obtain the discrete Hilbert transform f, of the sequence f, via the FFT shown in Fig. L INL Disceere Hagrury TRANSFORM (DHT) ‘AND CONVOLUTION THEOREM ‘The DHT of f(n) is defined as [4] Hk E Andes[ © where c25(2) = sn(2)+608(2), the inverse relation is Ho) EM dea| er) © ‘The comvoaton theorem obeyed by the DHT is a follows it (a) is the convolution of 3) with 39(0), ke, x(n)em(mer(m = Es(Koa(n&) (2) then Hk) H.(k)-even [ H,,(4)] + H,(~k)-odd | Ck) @ where H,(k), Hf,(k), and H.,(k) are the DHT’ of x(n), (7). and x3(m), respectively, “and H,,(k) = even [H,,(k)] + ‘odd H,,(4)], the sum of its even and odd pasts. The proof of this Hartley’ transform convolution theorem is shown in the Ap: pendix (0098-4094/89/0900-1251$01,00 ©1989 IEEE.feast {WEE FRANEACTIONS ON CIRCUITS AND SYSTEMS, VOL, 36, NO. 8, SEPTEMRER 1989 es eke fen ttaca) Fi, 2 Implementation of dre ier transform by FHT. IV. Discnite Hitnerr Transtona ny FHT ‘The ‘elation of the DHT and the DFT of a real sequence finyis H(k) ~even[ H,(K)] todd 1,(4)]: DET (9) F(k) =eveal H,(k)] ~ jodd[ H,(K)]: DFT. (10) rom (3), the discrete Hilbert transform of a sequence f(n) is equivalent to the convolution of f(n) with the Hilbert transform Kernel h(n) a8 follows: Aon) = sn) 0 b(n) 2 Ak) = FC) -F-sen(B). (1) Using the Hartley transform convolution theorem in (8), we obtain Hy) = Hy) even H,(K)] + Hy(— k) oda Hy()]. (2) Since the Hilbert transform kernel h(n) is ano function, then the even part ofits Harley transform sero, ic cven{ Hy(4)] = and 7(8) ~ 0d4{ 4}, the convolution theorem of (12) sm- plies into the multiplication product ofthe two separate Harley transforms Hy() = Hy(~) oa Hh (4)] = where H( =k) HC) (13) (14) wNoL and H,(—k) isthe time reversal of Hy (k), ie [As k=0 MOCO, kehae 8-1 ‘Then itis very efficient to compute the discrete Hilbert transform f, of the sequence f, through FHT shown in Fig. 2. Right now, the computational complexity is greatly reduced from 2 complex FFT's into 2 real FHT's. V. Concwustons In this paper, we have presented a new fast algorithm for ‘computing the discrete Hilbert transform through the FHT. ‘Compared with the conventional FFT approach, itis performed entirely in the real domain, also many efficient FHT algorithms ‘exist and can be applied here, and the computation complexity is ‘greatly reduced from 2 complex FFT's into 2 real FHT's. ‘ApenDix Proof of the Hartley transform convolution theorem if x(n) is the convolution of x;() wit x3(7), ie, LX x(m) (n= m). (AD) X(n) = (myer a) = ‘Taking discrete Hartley transformation on both sides: H()=S E (m)-n(n—m) ca) (ay 54) = Sag) | Ea(n—mcas{™e}]. (a3 By the Hartley transform shift theorem (4), (5) sro BE (tps 2M) 0) 2) a) $0 (A3) becomes Ak) Ei ssco[ coon 5" mk = H(-k)sin( en (as) AK) nc] Tistmeo( Ze ~ a0] 'S scons) (a8) H(k) = H.C) even] H,,(k)] ~ H,,(~ &)-odd | H,,(4)] (a “Then the proof is complet REFERENCES: UW} $. ©. Kak, “The discrete Hilbert tansorm,” Proc IBEE. vo. 58 pp. SiS- ste, pe 100. 12} VGaek, “Discrete Biber transform.” IEEE Tran, Au Eetroacont, ol AULI8 pp. 0-345, De. 1970 (01 AV Oppesici soa RW Scale, Dipl Signal Procening. Engle stood Clit NI: Premise Hall 1995 pp 937-378 la] UN. Braswell, "Disrete Harey wansoem." J Opt Soe. Amer vo Tipp, eis 85, Dee 198, (sy Poti fant Harley transona” Proc, IEEE, vol. 72, p. 1010-2018, (oy STE Per and 4. We, “Spitradix fat Harley transform” Ebon Let, ol. 22 no pp. 26-F7, Jan 1986. (7) HEA "Sovemen DE Jones C'S. Burrus, and MT. Meideman, “On omputing the dsree Harley wansorm” ZEEE Trane. Aout Sow, ‘Staal Provesng, vol ASSP pp, 51-1238, Or. 1985 [sy HS. Hoa. "The last Hares transform algortom IEEE Trane. Cone ut, vol C36, pp. 17-186, Feb 1987 Implementation of 2-D Filters for Real-Time Processing KLAUS RANK aN ROLF UNBEHAUEN Abstract—A systematic way for implementation of high speed -D linear fers for realtime image procesing 1s developed. Combining. a ‘consequent parallelization and a new RAMCbased implementation tech Manuscript rseivod November 24, 1988, This Iter was recommended by ‘Associate Elton TR Viswanathan The authors are wih the Leth fr Algemene and Thereache Elk TEEE Log Number 92720. : (0098-40489 /0900-1252501,00 ©1989 IEEE