Performance and Analyss of CDM-FH-OFDMA for Broadband Wreless Systems Kan Zheng, Lu Han, Janfeng Wang, Wenbo Wang Wreless Sgnal Processng Lab Bejng Unversty of Posts & Telecomms,Bejng, Chna Emal: zkan@buptneteducn February 20, 2006 Abstract Frequency-hoppng (FH) methods n the Orthogonal frequency dvson multplexng access(ofdma) system,whch are to assgn user-specfc subcarrer to the actve users, have been pad much attenton to n the broadband wreless communcaton system In ths paper, we present a novel multple access scheme, referred to as CDM-FH-OFDMA, whch s the extenson of FH-OFDMA wth code dvson multplexng (CDM) Ths scheme can explot the frequency dversty gan wthout the ad of channel codng And t also can be employed n the mult-cell envronment wth one frequency reuse factor Computer smulaton demonstrates effectveness of CDM-FH-OFDMA and the concluson s followed Keywords: OFDM; FH; CDM; 1 Introducton Orthogonal frequency dvson multplexng (OFDM) as a modulaton technque s beng appled extensvely to future wreless broadband systems due to ts effcent usage of the avalable frequency bandwdth and robustness to frequency selectve fadng envronments[1][2] Meanwhle, code dvson multple access (CDMA) has already shown qute a bt of promse n ts spectral effcency through the flexble frequency reuse and multple access technque for cellular systems [3] So OFDM combned wth code dvson multple access (CDMA) has drawn a lot of nterests n the research of future moble communcaton systems[4][5] These OFDM-CDMA schemes are derved from the classc DS-CDMA approach, employng mutually orthogonal spreadng codes for user separaton wthn one cell, and scramblng codes for dstngushng dfferent cells An alternatve to usng CDMA for multple-access s to assgn user-specfc subcarrer to the actve users These schemes are known as OFDMA or frequency-hoppng OFDMA[2] These schemes mantan the user orthogonalty also on frequency-selectve fadng channels, 1
Fgure 1: Block dagram of CDM-FH-OFDMA system but rely solely on channel codng and nterleavng for obtanng the dversty gan However, t s clamed that these scheme wll be more senstve to nter-cell-nterference and therefore not sutable n frequency-reuse-one systems[6] To fnd out the most approprate multple access schemes, t s necessary to nvestgate the performances of dfferent multple access schemes not only n the one-cell scenaros but also n the mult-cell scenaros In ths paper, we propose a novel multple access scheme, referred to as CDM-FH- OFDMA, whch s the extenson of FH-OFDMA wth code dvson multplexng (CDM) Smlar to FH-OFDMA, t apples OFDMA for user separaton but addtonally uses CDM on the data symbols belongng to the same user The CDM component s ntroduced n order to acheve addtonal frequency dversty gan and average the the nter-cell-nterference Lke OFDM-CDMA, ths CDM-FH-OFDMA explots the advantages gven by the combnaton of the spread spectrum technque and mult-carrer modulaton Snce one user exclusvely uses each subset of subcarrers, there s no multple access nterference between dfferent users n the same cell And the self-nterference of one user can be easly decreased by nterference cancellaton snce all supermposed modulated spreadng codes of ts subcarrer subset are affected by the same channel fadng When consderng the cellular system, frequency reuse factor of one can be realzed by usng dfferent scramble codes n the neghbor cells and nter-cell nterference can be avod by selectng dfferent subcarrer set for the users n the neghborng cells f the system load s not heavy Ths paper s organzed as follows Secton 2 gves the bref descrpton of CDM-FH- OFDMA system The detector structure s descrbed n Secton 3 Secton 4 descrbes the smulaton confguratons, and n Secton 5 the smulaton results are presented and dscussed Fnally, Secton 6 gves the concluson Notatons: Throughout ths paper, matrces and vectors are set n boldface () T, () and () + denote transpose, conjugate transpose and Moore-Penrose pseudo-nverse, respectvely 2 System Model Fg1 shows the block dagram of a CDM-FH-OFDMA system As the requrement of frequency-hoppng method, each frame, whch the basc process unt n the system, conssts of the N t OFDM symbolsafter the nformaton bts of user m are encoded and nterleaved, 2
Frequency Nf Tme Fgure 2: Multple access method of CDM-FH-OFDMA system they are modulated to the complex-value symbols wth the rate of 1/T d The vector d (m) = [d (m),0 d (m),1 d (m),q 1 ]T (21) represents one block of Q parallel modulated data symbols of user m n the th OFDM symboleach data symbol s multplexed wth another orthogonal spreadng code of length L The L Q matrx C = [c 0 c 1 c Q 1 ] (22) represents the Q dfferent spreadng codes c q = [c q,0 c q,1 c q,l 1 ] T C L 1, 0 q Q 1, used by user m, whch are the combnaton of the orthogonal Walsh codes and the basestaton-specfc scramblng codes The spreadng matrx C can be same for all the users The modulated spreadng sgnals are synchronously added, resultng n the transmsson vector n the th OFDM symbol S (m) = Cd (m) = [S (m) 0 S (m) 1 S (m) L 1 ]T C L 1 (23) To obtan OFDMA scheme, the orthogonal spreadng matrx s replaced by the dentty matrx As shown n Fg2, n a CDM-FH-OFDMA system, each user may be assgned a specfc frequency-hoppng sequence that ndcates the specfc subcarrer subset to use for data transmsson n each OFDM block nterval,e specfc Tme-Frequency(T-F) mappng pattern Multple data transmssons for multple users may be sent smultaneously usng dfferent frequency-hoppng sequences that are orthogonal to one another, so that only one data transmsson uses each subcarrer subset n each OFDM block nterval Usng orthogonal frequency-hoppng sequences, the multple data transmssons do not nterfere wth one 3
another whle enjoyng the benefts of frequency dversty For brevty, but wthout loss of generalty, the sze of subcarrer subset N f s assumed to be same as the length of the spreadng code,en f = L Then, accordng to the frequency-hoppng sequence of user m, the spreadng sgnal of user m s modulated by IDFT onto dfferent subsets of L subcarrers wthn the N total avalable subcarrers n one frame,and the resultant sgnal n the th OFDM block nterval can be expressed as s (m) (n) = 1 L 1 N l=0 S (m) l e (m) j2π(l+p L)n/N (24) where P (m) s a hoppng ndex whch ndcates subcarrer subset ndces for the th OFDM symbol of user m and the T-F mappng pattern for user m s P (m) = [P (m) 0 P (m) 1 P (m) N ] t 1 C 1 Nt Fnally the cyclc extenson of an OFDM block s added as guard nterval before transmsson n order to avod the nter-symbol-nterference The transmtted sgnals of dfferent users propagate through ndependent frequencyselectve fadng channels And the channel s modelled as a wde sense statonary, uncorrelated scatterng (WSSUS), Raylegh fadng channel wth L t paths and t s assumed that the channel state remans unchanged durng at least one OFDM block Then the channel mpulse response durng the th OFDM symbol can be expressed as h (m) (; n) = L t 1 l=0 α (m) l ()δ(n τ (m) l ) (25) where the lth tap gan α (m) l () wth propagaton delay τ (m) l of the mth user s ndependent complex Gaussan random varance wth zero mean and varable of σm,l, 2 If the cyclc prefx accommodates the channel delay spread between base staton and termnals, t s assumed that the narrowband sgnal that s transmtted through each subcarrer experences flat Raylegh fadng channel The flat fadng coeffcent at the kth subcarrer of the mth user durng the th OFDM block nterval can be expressed as L t 1 H (m),k = α (m) (m) j2πτ l e l k/n, (26) l=0 0 N t 1, 0 k N 1 After cyclc prefx removal and DFT, the receved sgnals of the desred users can be easly separated accordng to the hoppng pattern and no nterference between users exsts under the assumpton of deal synchronzaton snce the users are dstngushed by an FDMA scheme The sgnal vector of user m n the th OFDM symbol can be wrtten as Y (m) = H (m) S (m) + W = [Y (m) Y (m),p (m) L,P (m) L+1 Y (m),(p (m) +1)L 1 ]T C L 1 (27) where the L L channel fadng dagonal matrx for the desred user, AWGN term wth zero mean and varance of σ 2 n are gven respectvely by H (m) = dag{h (m),,p (m) L H(m),,,P (m) L+1 H(m) },(P (m) +1)L 1 W = [W,0 W,1 W,L 1 ] T C L 1 (28) 4
3 Detector Structure Any of the sngle-user or mult-user detecton technques presented for MC-CDMA systems [4] can be appled for the detecton of the data of a sngle user n CDM-FH-OFDMA systems Frstly the receved sgnal vector s equalzed by employng a bank of adaptve onetap equalzers to combat the phase and ampltude dstortons caused by the moble fadng channel on the subcarrers The one-tap equalzer s smply realzed by one complex-valued multplcaton per subcarrer The receved sequence at the output of the equalzer n the th OFDM block nterval can be expressed as The dagonal equalzer matrx Z (m) = G (m) Y (m) (31) G (m) = dag{g (m),0, G(m),1,, G(m),L 1 } CL L (32) represents the L complex-valued equalzer coeffcents of the subcarrers assgned to S (m) The complex-valued output Z (m) s despread by correlatng t wth the spreadng matrx C The complex-valued soft-decded values at the output of the despreader s d (m) = C T Z (m) (m) = [ d,0 d (m),1 d(m),q 1 ]T C Q 1 (33) The data symbol n the hard-decded detected vector C Q 1 s gven by ˆd (m) (m) = [ ˆd,0 ˆd (m),1 ˆd(m),Q 1 ]T ˆd (m),q = Γ{ (m) d,q }, 0 q Q 1 (34) where Γ{ } s the quantzaton operaton accordng to the chosen data symbol alphabet Several dfferent dversty combnng technques have been proposed n the lterature In ths paper the equalzer accordng to the mnmum mean error square(mmse) s used n the despreadng by choosng the equalzaton coeffcents as G (m),l = H (m),p (m) L+l Q H (m),p (m) L+l 2 + σ 2 n, 0 l L 1 (35) The detector descrbed above s orgnated from the prncple of sngle-user detecton (SD), whch only detects the desred sgnal wthout takng nto account any nformaton about mult-code nterference Wth the number of spreadng vectors (e Q) s ncreased, the performance of such SD detector wll be deterorated due to more serous mult-code nterference The nterference cancellaton (IC) can mprove the performance of the system wth heavy load at the expense of hgher recever complexty The prncple of parallel nterference cancellaton (PIC) s ntroduced[7] and can also be appled to CDM-FH-OFDMA systems The basc dea s to take the estmated data obtaned by the ntal SD detector and 5
Table 1: System Parameters Carrer 2GHz Frame duraton (ms) 2 DFT sze 1024 Cyclc Prefx nterval (samples/ µs ) 64/9803 Subcarrer separaton (khz) 6375 OFDM block duraton (µs) 16667 Number of OFDM symbols per frame N t 12 Number of useful data subcarrers 600 Sze of data subcarrer subset 40 Number of spreadng codes 8 Channel codng/decodng Turbo codec/ MAX LOG MAP (teraton=8) Channel model PB3, VA120 [9] Spreadng factor (Q) 8 Modulaton QPSK regenerate the transmtted sgnals to calculate the occurrng mult-code nterference of each desred data sgnal caused by all other spreadng sgnals, then to subtract t from the receved sgnals When only the qth,0 q Q 1,transmtted data symbol d (m) q of user m s desred and others are regarded as nterference,the correspondng sgnal after nterference-reduced can be wrtten as = Y (m) H (m) C q d (36) Y (m),q where C q denotes the modfed spreadng matrx obtaned by zerong the qth column of C After ths cancellaton step, the next data detecton ncludng equalzaton and despreadng/combnng are appled to the nterference-reduced sgnal Y (m),q, leadng to more relable data estmatons than the ntal SD detecton The equalzaton coeffcents after nterference-reducton are changed to G (m),l = H (m),p (m) L+l H (m),p (m) L+l 2 + σ 2 n, 0 l L 1 (37) In each stage, Q nterference cancellaton and data estmaton wll be performed teratve procedure can be contnued untl the performance s satsfed 4 System Confguraton The key smulaton parameters are summarzed n Table 1 and are kept n same wth [2] n order to be compatble wth FH-OFDMA system, whch has been wdely studed n 3GPP 6 Ths
Table 2: Informaton bt payload and code block szes for transport format Modulaton Code Informaton 24-bt-CRC Code Block R=1/3(K=4) Rate Rate bt payload addton Segmentaton Turbo-codng matchng QPSK 1/2 480 504 1 504 1524 960 QPSK 2/3 640 664 1 664 2004 960 QPSK 4/5 768 792 1 792 2388 960 long-tme evolutonthe selectve fadng channel models ncludng PB3, VA120 defned by ITU are used n the smulatons[9] Each path of the channel s modelled as a classcal Jakes Doppler spectrum Under the assumpton of a quas-statonary channel, the channel s constant durng one OFDM nterval At the recever, perfect symbol/carrer synchronzaton and channel state nformaton are assumed to be avalable 41 Interleavng The nterleavers have to be appled n the CDM-FH-OFDMA systems n order to better explore the dversty gan nherent n the tme-frequency selectve fadng channel There are two postons that the nterleavng wll be performed at the transmtter One s before modulaton module, e bt-nterleavng, the other s before the IDFT,e chp-nterleavng Snce all the data symbols are processed accordng to the frame unt n the tme doman, the nterleavng wll be appled wthn one frame ncludng N t OFDM symbols In the transmtter, the bnary nformaton s frst coded wth CRC attachment and code block segmentaton before channel codng The coded bt s punctured and bt-nterleaved accordng to [8] Table 2 provdes approprate nformaton bt payload and code block segmentaton values for the test cases n the smulatons For the sake of mplcaton,only one transport block sze for each user s assumed n ths paper Other szes may also be evaluated f necessary A block nterleaver s appled n case of chp-nterleavng, whch s a matrx wth N depth rows and N length columns The symbols are wrtten nto the matrx by rows and read out afterward by columns The denterleaver puts the symbols nto a matrx wth the same sze, but the symbols are wrtten by columns and read out by rows Usually the number of rows N depth s defned as the nterleavng depth whle the number of the column N length as the nterleavng length In the smulatons, the N depth and N length of the block chpnterleavng equal to the sze of the subcarrer subset and the number of OFDM symbol per frame,respectvely 42 Frequency-hoppng pattern and spreadng The dfferent users are dstngushed by allocatng each user wth a separate pattern for the tme-frequency(t-f) mappng of OFDM unts All the T-F mappng patterns n a cell should 7
be orthogonal to avod the cross-nterference between dfferent userseach of them should provde not only a maxmzed dversty gan wthn a cell but also a mnmzed nter-cellnterference between the neghborng cells [2] Therefore,the set of 15 orthogonal T-F pattern, one for each user, s derved from a sngle generc Costas sequence of length 15 Snce the number of OFDM symbol per frame s 12 (e N t = 12), the rght T-F pattern of length N t s obtaned by dscardng the last three symbols of the generc Costas sequence Then, the frst T-F mappng pattern s gven by P (0) = [13 5 3 9 2 14 11 15 4 12 7 10] (41) All the T-F pattern n the set are obtaned from the frst patten n the set by all the dfferent cyclc shfts n the frequency doman In that way, t s ensured that the set of pattern s orthogonal All the tme-frequency mappng patterns n a frame are cyclcally tme-shfted by a cellspecfc offset, correspondng to an nteger number of OFDM symbols The tme offset s changed for each frame, accordng to a cell-specfc scramblng sequence In that way, even f the two cells are synchronous n one frame, they wll very probably be asynchronous n the next frame, resultng n a mnmzed cross-nterference, as predcted by the correlaton propertes of tme-frequency mappng patterns In the smulated CDM-FH-OFDMA system, the data subcarrer subset sze N f s larger than the spreadng code length L So there are N f /L spreadng sgnal of length L wthn one data subcarrer subset of sze N f 5 Performance Evaluaton 51 Sngle-Cell In Fg3(a) and Fg3(b), the performances of the proposed CDM-FH-OFDMA and FH- OFDMA system applyng MMSE prncple wth the dfferent code rates under PB3 channel or VA120 channel are compared, where QPSK modulaton s used In case of low or medum code rate, the BLER performances of FH-OFDMA are equvalent to or better than these of CDM-FH-OFDMA because t can well explot the frequency dversty gan through channel codng and the mult-code nterference n CDM-FH-OFDMA deterorates the BLER performance However, wth the hgh code rate, the CDM-FH-OFDMA system can make better use of the frequency dversty gan by the operatons of spreadng and combnaton than the channel codng For example, f the target BLER s assumed to be 10 2, the BER gan of CDM-FH-OFDMA compared wth FH-OFDMA s about 1 or 2dB under PB3 channel or VA120 channel n case of R c = 4/5 Fg4(a) and Fg4(b) and compare the performances of the proposed CDM-FH-OFDMA and FH-OFDMA system applyng MMSE prncple or PIC detector wth the dfferent code rates under PB3 channel or VA120 channel, where QPSK modulaton s used When the code rate s low(eg R c = 1/2), the performance of CDM-FH-OFDMA system are manly affected by the channel codng/decodng and the nterference cancellaton won t acheve much performance gan Then, the BLER performance of CDM-FH-OFDMA system wth 8
01 01 BLER 001 Channel PB3 QPSK BLER 001 Channel VA120 QPSK R c FH-OFDMA CDM-FH-OFDMA 1/2 2/3 4/5 1E-3 4 5 6 7 8 9 10 11 12 13 14 15 16 E b /N 0 (db) 1E-3 R c FH-OFDMA CDM-FH-OFDMA 1/2 2/3 4/5 4 5 6 7 8 9 10 11 12 13 14 15 16 E b /N 0 (db) (a) PB3 (b) VA120 Fgure 3: BLER performances n FH-OFDMA or CDM-FH-OFDMA usng MMSE detector Rc=1/2 Rc=4/5 Rc=1/2 Rc=4/5 01 01 BLER 001 Channel PB3 QPSK FH-OFDMA CDM-FH-OFDMA IC-CDM-FH-OFDMA 1E-3 4 5 6 7 8 9 10 11 12 13 14 15 16 E b /N 0 (db) BLER 001 Channel VA120 QPSK FH-OFDMA CDM-FH-OFDMA IC-CDM-FH-OFDMA 4 5 6 7 8 9 10 11 12 13 14 15 16 E b /N 0 (db) (a) PB3 (b) VA120 Fgure 4: BLER performances n FH-OFDMA or CDM-FH-OFDMA usng PIC detector 9
PIC stll keeps lttle worse than that of FH-OFDMA system On the other hand, wth the hgher code rate(egr c = 4/5), the PIC detector wll gve about 1dB gan f the target BLER s assumed to be 10 2 under both channel envronments 52 Mult-Cell One of the advantages of CDM-FH-OFDMA system s that the same bandwdth can be reused n each cell, whch s often refereed to a full frequency re-use, or frequency re-use factor of 1 The man beneft of such a frequency reuse s manly ease of deployment, gven that no frequency plannng s requred However, the CDM-FH-OFDMA system wth a frequency reuse of 1 becomes nterference-lmted,and the nterference perceved by the termnal from the dfferent cells mght not be perfectly whtefrst, the nterferng sgnals undergo tme dsperson, and hence, do not have a flat spectrum Furthermore, f the OFDM unts are not all beng used n the nterferng cells, the resultng spectrum from each of these partally-load nterferng cells wll contan gaps It s therefore lkely that the total nterference spectrum observed by the termnal would not be flat, and hence, t mght not be accurately modelled usng whte nose The mpact of realstc nter-cell nterference on performance has therefore been evaluated usng the mult-cell smulator wth a rado geometry concept, whch consder relatve ntercell nterference and ndependent fadng for a lmted number of chp-exact modelled termnal n a neghborng cell[10] To become ndependent from absolute path gans as well as cell layouts, we use the geometry factor whch s defned as G = I or I oc,b + I oc (51) where I or denotes the receved total power orgnated from the servng, I oc s the receved total power from the all the nterferng cells and I oc,b s the porton of the receved power from those cells modelled as AWGN Consderng the trade-off between the relablty and smulaton complexty,the nter-cell-nterference s assumed to be generated by a sngle neghborng cell It s generated by usng the same tme-frequency mappng patterns as n the servng cell Namely, n all cells, the tme-frequency mappng patterns n a frame are cyclcally tmeshfted by a cell-specfc offset, correspondng to an nteger number of OFDM symbols The tme offset s changed for each frame, accordng to a cell-specfc scramblng sequence For the sake of smplfcaton, there s only one user n the servng cell and the nter-cell nterference domnates over the nose n ths smulaton scenaro It s assumed that 100% of the addtve nterference plus nose s due to the nter-cell-nterference and 0% due to the thermal nose, e, I oc I oc,b + I oc = 1 (52) The smulaton results wth the dfferent cell load n the nterferng cell are plotted n Fg5,whch gves the performances of FH-OFDMA and CDM-FH-OFDMA wth MMSE detecton Less geometry factor, further dstance to the base staton the user s When 10
Geometry factor (BLER=001) 11 Rc=4/5 10 9 8 7 6 5 4 3 2 1 0-1 -2 Rc=1/2 FH-OFDMA CDM-FH-OFDMA PB3 VA120 QPSK 1 user n the servng cell 0% AWGN BLER = 001 2 4 8 16 Number of ntereference users n the neghbour cell Fgure 5: Requred geometry factor wth BLER=10 2 under dfferent cell load n the nterferng cells the cell load n the nterferng cell ncreases, the lnk-performance mproves Ths perhaps somewhat counterntutve result s explaned as followswhen the users n the nterferng cells ncreases, the collson probablty of the subcarrers between the servng user and the nterferng users becomes larger So the nterference becomes more lke Gaussan accordng to the center-lmted prncple Meanwhle, the sgnal power of each nterferng users decreased wth the number of users ncreases n order to keep the total nterference power constant Therefore, the requred geometry factor wth BLER=10 2 decreased wth the number of the users n the nterferng cell,whch means the servng user can get the same BLER performance n the further poston to the centerng base staton In case of low code rate, the performances of FH-OFDMA are lttle better than those of CDM-FH-OFDMA because the channel codng can explot the dversty gan well,whch s smlar to the sngle-cell case However, wth the code rate ncreases, the proposed CDM- FH-OFDMA system wth the smple MMSE detecton outperforms FH-OFDMA system 6 Concluson In ths paper,we propose a CDM-FH-OFDMA scheme for the downlnk hgh data rate transmsson n the broadband wreless communcaton system Smulaton results demonstrate that the performances of CDM-FH-OFDMA system are better than those of conventonal FH-OFDMA n case of medum or hgh code rate Also, the nterference cancellaton can be appled n CDM-FH-OFDMA systems to further mprove the performance 11
References [1] Eklund, C; Marks, RB; Stanwood, KL; Wang, S, IEEE standard 80216: a techncal overvew of the WrelessMAN ar nterface for broadband wreless access, IEEE Communcatons Magazne,vol40,no6,pp98-107,2002 [2] 3GPP TR 25892 V200 (2004-06)Feasblty Study for OFDM for UTRAN enhancement (Release 6) [3] H Holma and A Toskala, WCDMA for UMTS New York: Wley, 2000 [4] S Hara,R Prasad, Overvew of multcarrer CDMA, IEEE Communcatons Magazne,vol35,no12,pp126-133,Dec 1997 [5] K Zheng,G Zeng,W Wang, Performance Analyss for OFDM-CDMA wth Jont Frequency-tme Spreadng, IEEE Transactons on Broadcastng, vol51, no1, pp144-148,march 2005 [6] Maeda, N; Atarash, H; Abeta, S; Sawahash, M; Throughput comparson between VSF-OFCDM and OFDM consderng effect of sectorzaton n forward lnk broadband packet wreless access, n ProcIEEE Vehcular Technology Conference, vol1, pp47-51,fall 2002 [7] MKVaranas, BAazhang, Multstage detecton n asynchronous code-dvson multpleaccess communcatons, IEEE Trans Commun,vol38,no4,pp 509-519, Aprl 1990 [8] 3rd Generaton Partnershp Project (3GPP), 3G TS 25211, v350, Physcal Channels and Mappng of Transport Channels onto Physcal Channels (FDD), Dec 2000 [9] Gudelnes for Evaluaton of Rado Transmsson Technologes for IMT-2000, Rec ITU- RM1225 [10] Weber, R; Schulst, M; Schotten, H WCDMA mult-cell lnk-level performance, ProcIEEE Personal, Indoor and Moble Rado Communcatons,vol3,pp1362-1366,Sept 2002 12