An Improved Downlink MC-CDMA System for Efficient Image Transmission

Transcription

1 Paper An Improved Downlnk MC-CDMA System for Effcent Image Transmsson Mohammed Sofane Bendelhoum,2, Al Djebbar 2, Ismal Boukl-Hacene 3, and Abdelmalk Taleb-Ahmed 4 Department of Technologe, Scences Insttute, Unversty Center Nour Bachr of El-Bayadh, Algera 2 Telecommuncatons and Dgtal Sgnal Processng Laboratory, Djllal Labes Unversty, Algera 3 Bomedcal Laboratory, Abou Bekr Belkad Unversty, Tlemcen, Algera 4 Laboratore LAMIH UMR CNRS 8530, Valencennes et du Hanaut Cambréss Unversty, Valencennes, France Abstract Image compresson s an essental stage of the data archvng and transmttng process, as t reduces the number of bts and the tme requred to complete the transmsson. In ths artcle, a study of mage transmsson over the Mult-Carrer Code Dvson Multple Access (MC-CDMA) downlnk system s presented. The soluton proposed reles on source codng combned wth channel codng. The Dscrete Wavelet Transform (DWT) method s used n conjuncton wth the SPIHT coder to compress the mage, then the data generated s transmtted wth the MC-CDMA technque over a nosy channel. The results show that mage transmssons performed over MC-CDMA usng the SPIHT model are better than the tradtonal approach lke MC-CDMA n the AWGN channel. Keywords data compresson, MC-CDMA, MSSIM, performance evaluaton, PSNR, SNR, SPIHT.. Introducton One of the most mportant results of Shannon s work [] was that source codng and channel codng can be treated separately wthout sacrfcng fdelty [2]. Tradtonally, the source and channel codes are desgned separately and then cascaded together [3]. Mult-carrer modulaton such as Orthogonal Frequency Dvson Multplexng (OFDM) has a promsng future as a new technology n next-generaton wreless communcaton systems [4], [5]. Combnng OFDM wth Code Dvson Multple Access (CDMA) results n the Multcarrer CDMA system [6], [7]. Ths system has receved much attenton among researchers. However, t suffers from Multple Access Interference (MAI) n a mult-user settng, whch decreases the overall Bt Error Rate (BER) performance [5]. Mult-user Detecton (MUD) technques have been ntroduced to mtgate MAI to mprove the system performance [8], [9]. Several studes n transmsson of mages and multmeda usng MC-CDMA technology are consdered [5], [0], []. In [0], a study of real-tme mage traffc over a rado lnk usng the wavelet transform (WT) to compress mages and a CDMA lnk to transfer these mages over wreless communcaton networks was presented. Images compressed wth JPEG2000 are transmtted at low SNRs wth the MC-CDMA technque and are consdered n []. An effcent approach for transmttng mages over MC-CDMA systems was examned n [5]. Furthermore, several codng technques have been nvestgated for effcent transmsson of mages wth MC-CDMA over wreless channels [0], [2], [3]. Despte the effcency of these codng technques, they add much redundancy to the transmtted data, whch ncreases the bandwdth and reduces the channel utlzaton [5]. Ths paper studes an unage transmsson effcency scheme relyng on the MC-CDMA system usng dscrete wavelet transforms (DWT) and set parttonng n herarchcal trees (SPIHT) codng n the Addtve Whte Gaussan Nose (AWGN) channel. The results are compared wth the tradtonal MC-CDMA technque. In the smulaton conducted, the mages are ssued wth the use of both technques, and ther recepton s objectvely evaluated by means of PSNR and MSSIM parameters. The transmsson of mages s done wth a dfferent SNR. The remander of the paper s organzed as follows: n Secton 2, a descrpton of decomposton of the mage usng DWT s presented. Secton 3 exposes the SPIHT encoder. The parameters used for judgng the qualty of mages are ntroduced n Secton 4. In Secton 5 the MC-CDMA model used for transmsson and recepton s gven. Secton 6 presents expermental and performance results. Fnally, Secton 7 offers conclusons and remarks. 2. Wavelet Transform The emergence of wavelets has led to the convergence of lnear expanson methods used n sgnal processng and appled mathematcs. Subband codng methods and ther assocated flters are closely related to the wavelet concept [3]. 5

2 Mohammed Sofane Bendelhoum, Al Djebbar, Ismal Boukl-Hacene, and Abdelmalk Taleb-Ahmed A wavelet s a foundaton for representng mages n varous degrees of resoluton. Wavelet transform s a mathematcal functon dvdng an mage nto two varous frequency components matchng the resoluton. The wavelet transformaton methodology has been used because of the dsadvantages of Fourer transformaton [4]. Wavelet transformaton has been classfed, as mentoned earler, as DWT and contnuous wavelet transforms (CWT). A wavelet s represented as a mult-resoluton level, where each analyss s mplemented through hgh-pass and low-pass flters, where each hgh-pass flter s based on wavelets and low-pass flter s based on scalng functons. The wavelet transform functon s based on the converson of a one-dmensonal functon nto two-dmensonal spaces nvolvng translaton and dlaton parameters related to tme and scale factors [5]. 3. SPIHT Codng Scheme In 996, Sad and Pearlman ntroduced the SPIHT codng scheme [7]. SPIHT s vewed as one of the most effcent wavelet-based mage compresson algorthms avalable n the lterature [7], [8]. It can be a progressve transmsson and t may be relevant for the lossless compresson area of nterest. It takes advantage of the spatal orentaton of the zerotree codng that was proposed by Shapro [9] and has been shown to provde better performance [20]. In addton, SPIHT can compress mages n a lossless manner, as well as wth a loss [2]. SPIHT [22] s one of the most advanced schemes avalable, even outperformng the state-of-the-art JPEG 2000 n some stuatons. The basc prncple s the same progressve codng s appled, processng the mage respectvely to a lowerng threshold. The dfference s n the concept of zerotree (spatal orentaton trees n SPIHT). Ths s an dea that takes nto consderaton relatons between coeffcents across subbands at dfferent levels [7]. The frst dea s always the same. If there s a coeffcent at the hghest level of the transform n a partcular subband whch s consdered nsgnfcant aganst a partcular threshold, t s very probable that ts descendants n lower levels wll be nsgnfcant too. Therefore, we can code qute a large group of coeffcents wth one symbol. Fgure 2 shows how a spatal orentaton tree s defned n a pyramd constructed wth recursve four subbands splttng. The coeffcents are ordered herarchcally. Accordng to ths relatonshp, the SPIHT algorthm saves many bts that specfy nsgnfcant coeffcents [23]. Fg.. Pyramdal structure of 3-level wavelet decomposton. Wavelet analyss of an mage can be vewed n the frequency doman as parttonng t nto a set of subbands, where each parttonng step s obtaned by applyng the 2D wavelet transform. One level of 2D wavelet transforms results n four sets of data (wavelet coeffcents) that correspond to four 2D frequency subbands. For these four subbands, f the orgnal mage data s on the zerodecomposton level (scale), we use the followng notaton on the k-th decomposton level: HHk (hgh hgh or dagonal detals), HLk (hgh low or horzontal detals), LHk (low hgh or vertcal detals), LLk (low low or approxmaton). The LLk subband s also referred to as mage approxmaton, as t represents the mage on a lower scale, whle to other subbands are referred as to mage detals. Wavelet decomposton s dyadc when only the LLk subband s further transformed. It results n a new set of subbands: HHk+, HLk+, LHk+, LLk+. Dyadc decomposton used n mage compresson wll thus generate a herarchcal pyramdal structure, as shown n Fg.. If the dyadc decomposton of N levels s performed (low low subband transformaton performed N tmes), the result wll be 3 N+ subbands [6]. Fg. 2. Parent-chld relatonshp. The SPIHT flowchart s presented n Fg. 3. In the frst step, the orgnal mage s decomposed nto subbands. Then, the method fnds the maxmum and the teraton number. In the second step, the method puts the wavelet coeffcents nto a sortng pass that fnds all sgnfcant values and encodes 6

3 An Improved Downlnk MC-CDMA System for Effcent Image Transmsson ther sgn. In the thrd step, the sgnfcant coeffcents that were found n the sortng pass are put nto the refnement pass that uses two bts to exactly reconstruct the value for closng to the real value [24]. where the structural smlarty ndex (SSIM) between two sgnals x and y s gven by [27], [28]: SSIM(x,y) = l(x,y)c(x,y)s(x,y). (3) Fnally, the qualty measurement can provde a spatal map of the local mage qualty, whch gves more nformaton on the mage qualty degradaton. For applcaton, we requre a sngle overall measurement of the whole mage qualty that s gven by the followng formula: MSSIM(I,Î) = M M SSIM(I,Î), (4) = where I and Î are respectvely the reference and degraded mages respectvely, I and Î are the contents of mages at the -th local wndow. M s the total number of local wndows n the mage. The MSSIM values exhbt greater consstency wth the vsual qualty. 5. Model MC-CDMA System Fg. 3. Flowchart of SPIHT wavelet transform. 4. Qualty Evaluaton The peak sgnal to nose rato (PSNR) s the most commonly used reconstructon qualty measure n mage compresson: (Dynamcs of mage)2 PSNR = 0 log. () MSE Mean square error (MSE) whch requres two M N gray scale mages I and Î where one of the mages s consdered as a compresson of the MSE 2 = M N N = M j= ( I(, j) Î(, j) ) 2. (2) Usually an mage s encoded by 8 bts. It s represented by 256 gray levels, whch vary between 0 and 255, wth the extent or dynamcs of the mage beng 255. The PSNR measurement gves a numercal value on the damage, but t does not descrbe ts type. Moreover, as s often noted n [25], [26], t does not qute represent the qualty perceved by human observers. For dfferent applcatons where mages that are degraded must eventually be examned by experts, tradtonal evaluaton remans nsuffcent. For ths reason, objectve approaches are needed to assess the mage qualty. Then, we evaluate a new paradgm to estmate the qualty of mages, specfcally the ones compressed by wavelet transform based on the assumpton that the human vsual system (HVS) s hghly adapted to extract structural nformaton. The smlarty compares the brghtness, contrast and structure between each par of vectors, In general, there are three types of hybrd schemes, e.g. Mult-Carrer Drect Sequence CDMA (MC/DS-CDMA), Multcarrer-CDMA (MC-CDMA) and Orthogonal Frequency Code Dvson Multplexng (OFCDM) [29]. The basc MC-CDMA sgnal s generated by a seral concatenaton of classcal Drect-Sequence CDMA (DS-CDMA) and Orthogonal Frequency Dvson Multplexng (OFDM). Each chp of the drect sequence spread data symbol s mapped to a dfferent subcarrer. Thus, wth MC-CDMA, the chps of a spread data symbol are transmtted n parallel on dfferent sub-carrers, n contrast to seral transmsson wth DS-CDMA [5], [30]. The MC-CDMA system model usng for transmsson data mage s dvded nto four stages, ncludng the mage data compresson stage usng DWT and SPHIT, the BPSK modulaton stage, the spreadng stage usng the Walsh code, and the OFDM modulaton stage, as shown n Fg. 4. At the transmtter, the bt stream generated after compresson step b (k) s modulated by BPSK B (k). The BPSK modulated nformaton bts of each user k = 0,..., k, are spread by the correspondng spreadng code, where Walsh- Hadamard codes are appled. In the downlnk system, the spread data of all actve users are added synchronously and mapped on to the L c sub-carrers [7]. The sequence, B (k) s appled to a seral-to-parallel converter and multpled by the k wth the user specfc spreadng code of Walsh C (k) of length L [5], [7], In [3], [7] confrmed that the Walsh- Hadamard codes are the best choce for the downlnk, also for the mnmzaton of multple access nterference. C (k) = ( C (k) 0, C(k) T,..., C(k) L ). (5) The chp rate of the seral spreadng code of Walsh C (k) before seral-to-parallel converson s [7]: T c = L T B. (6) 7

4 Mohammed Sofane Bendelhoum, Al Djebbar, Ismal Boukl-Hacene, and Abdelmalk Taleb-Ahmed Fg. 4. The MC-CDMA system model for mage transmsson: (a) transmtter, (b) recever. The chp rate s L tmes hgher than the data symbol rate T B. The complex valued sequence obtaned after spreadng can be expressed n a vector notaton as [7]: s (k) = B (k) C (k) = ( s (k) 0, s(k),..., S(k) L ) T. (7) Then s (k) s modulated by the OFDM modulator, the nput sgnal of the OFDM operaton s statstcally ndependent [7]. The result s S (k). The advantage of the frequency doman channel model s that the IFFT and FFT operaton for OFDM and nverse OFDM can be avoded, and the fadng operaton results n one complex-valued multplcaton per subcarrer [7]. The spectral contanment of the channels s better snce they are not usng cyclc prefx [32] and t reduces the Inter-Symbol Interference (ISI) effect. Sgnal H s obtaned by addng the S (k) from K users: H = k =0 S (k). (8) The result H s appled to a parallel-to-seral converter. The transmtted vector T X s gven by: T X (t) = H c + n, (9) where H c s the data channel vector and n s the nose AWGN vector. The steps used to transmt data mages usng the MC- CDMA model can be summarzed as follows:. The data mage s transformed by DWT n the same dmenson as the orgnal mage. 2. The result of the step s assocated wth SPIHT codng to generate the bts stream to be transmtted also at a dfferent bt rate of R c. 3. It s assumed that all subchannels are dentcally modulated usng bnary phase-shft keyng (BPSK) [33]. 4. The most mportant characterstc of the sequence n the MC-CDMA system s the zero-cross correlaton property. The most wdely used orthogonal code s the Walsh-Hadamard code [2], whch s generated by usng the Hadamard transform defned as: C 0 = [0], (0) [ ] Cm C C 2m = m. C m C m () Ths transform gves matrx, C m, for m = 2 matrx, where s an nteger. For example, consder a case of n = 4: [ C2m C C 4 = 2m C 2m C 2m ] = (2) 5. An OFDM modulaton step s performed on the bnary data. 6. At the recever, after OFDM demodulaton step. The vector s multpled by the C (k) code Walsh for k-th user, and we appled the SPIHT decodng method. The nverse dscrete wavelet transform (IDWT) s used to receve the data mage. Fnally, the decompressed mage s gven at the transmtted bt rate of R c. 6. Results and Dscusson After varous applcatons ntended to show the actual performance of the method proposed, we wll make a comparson between the followng, dfferent types of wavelet trans- Fg. 5. Typcal test pctures: (a) Lena, (b) cameraman. 8

5 An Improved Downlnk MC-CDMA System for Effcent Image Transmsson forms: Daubeches (db5), symelets (sym9), Coflets (cof5), B-orthogonal (bor4.4) and Cohen-Daubeches-Feauveau (CDF9/7). They are assocated wth an SPIHT coder n the Subsecton 6.. The Lena and Cameraman mages (Fg. 5) wth pxels (grayscale) are encoded (8 bts per pxel) for each applcaton. In the step compresson step, we vary the btrate from to and transmt them va the MC-CDMA model, at dfferent SNRs from 5 to 20 db n 5 steps. The qualty of the mages transmtted and receved s evaluated objectvely wth PSNR and MSSIM parameters for dfferent bt rates R c. Subsecton 6.3 gves the performance of the MC-CDMA system n the AWGN channel wth multple users. The smulaton values are depcted n a graph/plot of BER vs. Eb/N0 (n db), whch s the energy per bt per nose spectral densty [2], [3]. It defnes the SNR per bt and t s an mportant measurement to evaluate and compare dfferent dgtal communcaton systems. The CDF9/7 wavelet s mplemented wth a SPIHT coder. Tree mages are used: IRM, IRM2 and IRM3. The orgnal mage s compressed wth dfferent bt rates R c and transmtted at dfferent SNRs for 0 to 7 db n db steps. The mage transmtted and reconstructed at dfferent SNRs s estmated wth PSNR and MSSIM factors. In calculatons and numercal smulatons, the same mages as those treated n [5] are used to evaluate the performance of the smulaton model presented. 6.. Wavelets Assocaton wth SPIHT n MC-CDMA The average PSNR and MSSIM values for the Lena mage at dfferent R c are shown n Tables 5, enablng the judgment of qualty of the SPIHT compressed mage after transmsson, as compared to ts qualty before transmsson. A plot of the varaton of PSNR and MSSIM for the Lena mage wth dfferent SNRs at R c constant s shown n Fgs Table PSNR and MSSIM values for the Lena mage at dfferent R c and SNR wth bor4.4 and SPIHT Parameter Before Transmsson wth SNR R of c trans- [db] [bpp] evaluaton msson MSSIM MSSIM MSSIM MSSIM MSSIM Table 2 PSNR and MSSIM values for the Lena mage at dfferent R c and SNR wth cof2 and SPIHT Parameter Before Transmsson wth SNR R of c trans- [db] [bpp] evaluaton msson MSSIM MSSIM MSSIM MSSIM MSSIM Table 3 PSNR and MSSIM values for the Lena mage at dfferent R c and SNR wth wth sym9 and SPIHT Parameter Before Transmsson wth SNR R of c trans- [db] [bpp] evaluaton msson MSSIM MSSIM MSSIM MSSIM MSSIM Table 4 PSNR and MSSIM values for the Lena mage at dfferent R c and SNR wth bor4.4 and SPIHT Parameter Before Transmsson wth SNR R of c trans- [db] [bpp] evaluaton msson MSSIM MSSIM MSSIM MSSIM MSSIM

6 Mohammed Sofane Bendelhoum, Al Djebbar, Ismal Boukl-Hacene, and Abdelmalk Taleb-Ahmed Table 5 PSNR and MSSIM values for Lena mage at dfferent R c and SNR wth CDF9/7 and SPIHT Parameter Before Transmsson wth SNR R of c trans- [db] [bpp] evaluaton msson MSSIM MSSIM MSSIM MSSIM MSSIM For all fgures the varatons of PSNR and MSSIM, ncreases proportonally wth the SNR. For the lower SNR (0 db) the mages that are receved are of very poor qualty (low MSSIM). In the case of SNR greater than 0 db, the qualty of mages that are receved ncreases untl they reach the maxmum value obtaned wthout a transmsson (SNR = 20 db). As far as the assocaton of dfferent types of wavelets (db5, cof2, sym9, bor4.4 and CDF9/7) wth the SPHIT encoder at dfferent bt rates (Rc) s concerned, let s say that the CDF9/7 wavelet provdes the most mportant PSNR and MSSIM values compared to those obtaned by other wavelets for a SNR = 20 db [20] (except for R c = bpp, where PSNR = db). The CDF9/7 wavelet wll be subsequently retaned for varous mages wth dfferent R c transmsson n a telecommuncatons MC-CDMA system. The receved mages wll be evaluated objectvely by PSNR and MSSIM Analyss of MC-CDMA Performance n a Wreless Channel Fgure 8 shows the BER performance comparson for multuser cases n the AWGN channel wth the MC-CDMA model used and tradtonal MC-CDMA. In theory, BER lowers wth ncreasng Eb/N0. It s observed that BER performance decreases as Eb/N0 ncreases. Ths s because Fg. 6. Varaton of PSNR and MSSIM for the Lena mage wth and bpp. (See color pctures onlne at publcatons/journal-jtt) 0

7 An Improved Downlnk MC-CDMA System for Effcent Image Transmsson Fg. 7. Varaton of PSNR and MSSIM for the Lena mage wth and bpp. a lower AWGN nose level can be obtaned wth a hgher Eb/N0. Ths s due to the fact that BER performance s only affected by AWGN nose [2], [3]. In the case of a low SNR, the two transmsson systems show the same performance. As SNR ncreases, as t s clear than ths transmsson scheme offers better performance n terms of BER termnology. The BER results show that the proposed transmsson scheme s based on the SPHIT coder coupled to MC-CDMA, and s superor to that obtaned from the tradtonal MC-CDMA. In the MC-CDMA model proposed the SNR ncreases to 5 db, n comparson wth the tradtonal MC-CDMA technque, where the SNR reaches 22 db, whch means a gan of about 7 db Performance Analyss of CDF9/7 wth SPIHT coder MC-CDMA System Several experments were performed to test the performance and the robustness of the medcal mages transmtted through the MC-CDMA system usng the CDF9/7 and SPHIT coder. The results of the smulaton are presented n Tables 6 8 for all mage users (IRM, IRM2 and IRM3). The IRM mage transmsson va a wreless MC-CDMA system wth the mage compressed at bpp needs the SNR = 5 db to acheve the maxmum PSNR and MSSIM values before the transmsson. On the other hand, for the R c =,, and bpp we need the SNR = 6 db for a perfect transmsson n our MC-CDMA system. The same remarks are noted n the case of IRM3 mages. For the transmsson of the IRM2 mage wth R c = bpp and, the SNR = 5 db s necessary, but for R c bpp greater than all results show that the SNR = 6 db can perfectly transmt all mages. Examples of the receved mages are shown n Fgs. 9 for SNR = 4, 5 and 6 db. Usng human eyesght only, we can see that the mages recovered are not dentcal wth the orgnals, once the SNR s lowered to 4 db. One can observe that the best qualty s obtaned at SNR = 6 db, n ths case the MC-CDMA model s consdered as a relable wreless communcatons system. The PSNR of the receved mages are exceeds 30 db, whch means that we can also transmt an mage usng R c = bpp and obtan a good level of vsual qualty (MSSIM 0.8) of the receved mage. All results obtaned by transmttng the mages over the MC-CDMA model are better than those relyng on the tradtonal MC-CDMA technque.

8 Mohammed Sofane Bendelhoum, Al Djebbar, Ismal Boukl-Hacene, and Abdelmalk Taleb-Ahmed Fg. 8. BER vs. Eb/N0 for mult-user envronment n AWGN channel wth Lena and Cameraman mage coded at: (a) bpp, (b) bpp, (c) bpp, (d) bpp, (e) bpp (MC-CDMA proposed), and a sngle user n tradtonal MC-CDMA [30]. Fg. 9. The receved IRM mages wth MC-CDMA usng R c = bpp over AWGN channel: (a) orgnal mage, (b) PSNR = db, (c) PSNR = 29.8 db, (d) PSNR = db. 2

9 An Improved Downlnk MC-CDMA System for Effcent Image Transmsson Fg. 0. The receved IRM2 mages wth MC-CDMA usng Rc = bpp over AWGN channel: (a) orgnal mage, (b) PSNR = 28.6 db, (c) PSNR = db, (d) PSNR = db. Fg.. The receved IRM3 mages wth MC-CDMA usng Rc = bpp over AWGN channel: (a) orgnal mage, (b) PSNR = db, (c) PSNR = 3.87 db, (d) PSNR = db. Table 6 Performance results of PSNR and MSSIM values for the IRM mage at dfferent R c and SNR Parameter of evaluaton Rc [bpp] Before transmsson 0 2 MSSIM MSSIM MSSIM MSSIM MSSIM 7. Conclusons In ths paper, an effcent mage transmsson system usng the downlnk MC-CDMA model s presented. The soluton proposed transmts mages over wreless AWGN channels usng DWT assocated wth the SPIHT coder. Accordng to Shannon s theorem [], the com- Transmsson wth SNR [db] bnaton of the optmum source coder and optmum channel coder can create an optmum system, therefore jont desgn of source coder and channel coder s necessary to optmze communcaton systems [0]. Therefore, the CDF9/7 wavelet s used wth the SPIHT coder to optmze data mage compresson. We can also transmt mages wth Rc = bpp, whch equals to the compresson rate 90.63% of the orgnal mage. 3

10 Mohammed Sofane Bendelhoum, Al Djebbar, Ismal Boukl-Hacene, and Abdelmalk Taleb-Ahmed Table 7 Performance results of PSNR and MSSIM values for the IRM2 mage at dfferent R c and SNR Parameter Before Transmsson wth SNR [db] R c [bpp] of evaluaton transmsson MSSIM MSSIM MSSIM MSSIM MSSIM Table 8 Performance results of PSNR and MSSIM values for the IRM3 mage at dfferent R c and SNR Parameter Before Transmsson wth SNR [db] R c [bpp] of evaluaton transmsson MSSIM MSSIM MSSIM MSSIM MSSIM For a perfect recepton of the mage, we need the SNR = 7 db (for natural mages). On the other hand, for medcal mages, the SNR = 6 db s necessary. Ths enables to use an optmzed wreless communcatons system. References [] C. E. Shannon, A mathematcal theory of communcaton, The Bell System Tech. J., vol. 27, no. 3, pp , 948. [2] C. E. Shannon, Codng theorems for a dscrete source wth a fdelty crteron, IRE Int. Conv. Rec., pp , 959. [3] H. Keang-Po and M. K. Joseph, Image transmsson over nosy channels usng multcarrer modulaton, Sgnal Process.: Image Commun., vol. 9, no. 2, pp , 997. [4] J. Latal et al., Expermental measurement of thermal turbulence effects on modulated optcal beam n the laboratory, n Proc. 9th OptoElectron. and Commun. Conf. and 39th Australan Conf. on Optcal Fbre Technol. OECC/ACOFT 204, Melbourne, Australa, 204, pp [5] E. M. El-Bakary et al., Effcent Image Transmsson wth Mult-Carrer CDMA, Wreless Pers. Commun., vol. 69, no. 2, pp , 203 (do: 0.007/s ). [6] H. Schulze and C. Luders, Theory and Applcatons of OFDM and CDMA. New York: Wley, [7] K. Fazel and S. Kaser, Mult-Carrer and Spread Spectrum Systems. Chchester: Wley, [8] S. Hara and R. Prasad, Overvew of multcarrer CDMA, IEEE Commun. Mag., vol. 35, no. 2, pp , 997 (do: 0.09/ ). [9] S. Verdu, Multuser Detecton. Cambrdge, UK: Cambrdge Unversty Press, 998. [0] P. P. Dang and P. M. Chau, Robust mage transmsson over CDMA channels, IEEE Trans. on Consumer Electron., vol. 46, no. 3, pp , [] T. Kathyaah and T. H. Oh, Robust JPEG2000 mage transmsson through low SNR Mult-Carrer CDMA system n frequencyselectve Raylegh fadng channels: a performance study, Int. J. Wavelets Multresolut Inform. Process., vol. 9, no. 2, pp , 20. [2] T. H. Oh and G. T. Km, Image transmsson through MC-CDMA channel: an mage qualty evaluaton, Int. J. Wavelets Multresolut Inf. Process., vol. 6, no. 6, pp , [3] K. Ramchandran, M. Vetterl, and C. Herley, Wavelets subband codng and best bases, Proceedngs of the IEEE, vol. 84, no. 4, pp , 996. [4] R. C. Gonzalez and R. E. Woods, Dgtal Image Processng. Pearson- Prentce Hall,

11 An Improved Downlnk MC-CDMA System for Effcent Image Transmsson [5] A. S. Vjendran and B. Vdhya, A hybrd mage compresson technque usng wavelet transformaton MFOCPN and Interpolaton, Global J. of Com. Scence and Technol., vol., no. 3, pp , 20. [6] N. Sprljan, S. Grgc, and M. Grgc, Modfed SPIHT algorthm for wavelet packet mage codng, Real Tme Imag., vol., no. 5 6, pp , [7] A. Sad and A. P. Wllam, A new, fast, and effcent mage codec based on set parttonng n herarchcal trees, IEEE Trans. Crcuts and Syst. for Vdeo Technol., vol. 6, no. 3, pp , 996 (do: 0.09/ ). [8] A. Bourdane et al., A very low bt-rate embedded color mage codng wth SPIHT, n Proc. IEEE Int. Conf. on Acoust., Speech and Sg. Process. ICASSP 2004, Montreal, Quebec, Canada, 2004, pp [9] J. M. Shapro, Embedded mage codng usng zerotrees of wavelet coeffcents, IEEE Trans. on Sgnal Process., vol. 4, no. 2, pp , 993. [20] I. Boukl-Hacene, M. Beladghem, and A. Bessad, Lossy compresson color medcal mage usng CDF wavelet lftng scheme, I. J. Image, Graphcs and Sg. Process., vol., pp , 203 (do: 85/jgsp ). [2] T. Brahm, A. Melt, and F. Khelfb, An mproved SPIHT algorthm for lossless mage codng, Dgt. Sg. Process., vol. 9, no. 2, pp , [22] S. G. Maou, S. T. Chen, and S. N. Chao, Wavelet-based lossyto-lossless medcal mage compresson usng dynamc VQ and SPIHT codng, Bomed. Eng. Appl. Bass Commun., vol. 5, no. 6, pp , [23] Y. Y. Chen and S. C. Ta, Embedded medcal mage compresson usng DCT based subband decomposton and modfed SPIHT data organzaton, n Proc. 4th IEEE Symp. on Bonform. and Boengn. BIBE 2004, Tachung, Tawan, 2004 (do: 0.09/BIBE ). [24] M. Beladgham et al., MRI mage compresson usng Borthogonal CDF wavelet based on lftng scheme and SPIHT codng, n Proc. 4th Int. Conf. on Elec. Engn. CIGE 0, Bechar, Algera, 200, pp , 200. [25] Z. Xong, K. Ramchandran, and M. Orchard, Space frequency quantzaton for wavelet mage codng, IEEE Trans. on Image Process., vol. 6, no. 5, pp , 997. [26] K. A. Navas, M. L. Aravnd, and M. Saskumar, A novel qualty measure for nformaton hdng n mages, n Proc. IEEE Comp. Socety Conf. on Comp. Vson and Pattern Recogn. Worksh. CVPRW 08, Anchorage, AK, USA, 2008 (do: 0.09/CVPRW ). [27] M. Doaa and F. Abou-Chad, Image compresson usng block truncaton codng, J. of Selec. Areas n Telecommun. (JSAT), pp. 9 3, 20. [28] W. S. Geslerand and M. S. Banks, Vsual performance n Handbook of Optcs, vol. III, M. Bass, Ed. McGraw-Hll, 995. [29] A. B. Watson and L. Kreslake, Measurement of vsual mparment scales for dgtal vdeo, Proc. of SPIE, Human Vson and Electronc Imagng VI, 79, vol. 4299, pp , 200 (do: 0.7/ ). [30] M. S. Salh et al., A proposed mprovement model of MC- CDMA based FFT n AWGN channel, n IEEE Int. Conf. on Electro/Inform. Technol., Chcago, IL, USA, 2007, pp (do: 0.09/EIT ). [3] S. Noblet, J.-F Helard, and D. Motter, Spreadng sequences for uplnk and downlnk MC-CDMA systems: PAPR and MAI mnmzaton, European Trans. on Telecommun., vol. 3, pp , [32] K. Fazel and S. Kaser, Mult-Carrer and Spread Spectrum Systems. New York: Wley, [33] A. Gupta, P. Ngam, and V. Chaurasa, Removal of cyclc prefx n adaptve OFDM for dynamc spectrum access usng DWT and WT, Int. J. of Engn. Sc. & Res. Technol., vol. 3, no. 3, pp. 87 9, 204. Mohammed Sofane Bendelhoum receved a B.Sc. degree n Electroncs Bomedcal n 2004, and an M.Sc. n 2008, from Abou Bakr Belkad Unversty of Tlemcen, Algera. He s currently pursung hs Ph.D. degree at the Djllal Labes Unversty of Sd Bel Abbes, Algera. Hs man research nterests nclude the mage processng, medcal mage compresson, wavelets transform, wreless communcaton systems and sgnal processng. E-mal: sbendelhoum3@gmal.com Department of Technologe, Scences Insttute Unversty Center Nour Bachr of El-Bayadh El-Bayadh, 32000, Algera Al Djebbar receved a B.Sc. n 988, from Unversty of Scences and Technology of Oran (USTO, Algera), an M.Sc. n 99, from Sd Bel Abbes Unversty, and a Ph.D. n 997, from USTO. Snce 99, he works at the Department of Telecommuncatons at the Unversty of Sd Bel Abbes and performs hs research at the Telecommuncatons and Dgtal Sgnal Processng Laboratory. Hs research nterests nclude sgnal processng for telecommuncatons, communcaton over multpath and fadng channels, wreless networks, channel codng for communcaton networks. He has more than 60 publcatons n specalzed journals wth mpact factors. E-mal: adjebar2002@yahoo.fr Telecommuncatons and Dgtal Sgnal Processng Laboratory Djllal Labes Unversty Sd Bel Abbes, 22000, Algera Ismal Boukl-Hacene obtaned a B.Sc. n Electroncs, an M.Sc. n Electronc Bomedcal, and a Ph.D. degree n Electronc Bomedcal from Unversty of Tlemcen, Algera, Hs research nterests are mage processng, medcal mage compresson, wavelets transform, wreless communcaton systems and optmal encoder. He has more than 20 publcatons n specalzed journals wth mpact factors. E-mal: small80@yahoo.fr Bomedcal Laboratory Abou Bekr Belkad Unversty Tlemcen, 3000, Algera 5

12 Mohammed Sofane Bendelhoum, Al Djebbar, Ismal Boukl-Hacene, and Abdelmalk Taleb-Ahmed Abdelmalk Taleb-Ahmed receved hs Ph.D. n Electroncs and Mcrowaves from the Unversty of Llle n France n 992. From 992 to 2004, he was an Assocate Professor at the Unversty of Lttoral, Calas. He s currently a Professor at the Unversty of Valencennes n the department GE2I, and does hs research at the LAMIH FRE CNRS 3304 UVHC. Hs research nterests nclude sgnal and mage processng, mage segmentaton, pror knowledge ntegraton n mage analyss, partal dfferental equatons and varatonal methods n mage analyss, mage compresson, multmodal sgnal processng, medcal mage analyss, ncludng multmodal mage regstraton. He s an author of more than 200 publcatons. E-mal: abdelmalk.tale-ahmed@unvalencennes.fr Laboratore LAMIH UMR CNRS 8530 Valencennes et du Hanaut Cambréss Unversty Valencennes, 5933, France 6