. <. < Performnce of Adptive ultiuser Receivers for the WCDA Uplink. Surendr Ru, GVL Bbu nd A. Chocklingm Wireless nd Brodbnd Communictions Synopsys (Indi) Pvt. Ltd, Bnglore 5005 Deprtment of lectricl Communiction ngineering Indin Institute of Science, Bnglore 50012 Abstrct In this pper, we investigte the performnce of dptive interference cncelltion receivers for the widebnd code division multiple ccess (WCDA) uplink physicl dt chnnel. For the WCDA uplink wveform, we derive blind dptive receiver (BAR) bsed on the constnt modulus lgorithm (CA), nd n multistge dptive prllel interference cncelltion (APIC) receiver. In order to improve the performnce of the APIC receiver, we propose hybrid APIC (H-APIC) receiver structure which uses the CA bsed BAR s the first stge of the APIC. We evlute nd compre the performnce of the bove receivers in nerfr scenrio nd show tht the proposed H-APIC receiver performs better thn the APIC receiver. I. ITRODUCTIO Third genertion (3G) systems re envisged to provide host of communiction services, including voice, dt, high qulity imges, nd video to mobile users. Widebnd code division multiple ccess (WCDA), one of the ir-interfce stndrds for the 3G systems, enbles high speed rdio ccess up to 2 bps, nd supports multiple services with different qulity of service requirements [1. One of the issues with CDA trnsmissions on the uplink (mobile-to-bse sttion link) is the nerfr effect. Receivers using multiuser detection [2 cn llevite the ner-fr effect nd improve system cpcity significntly, t the expense of incresed receiver complexity. Severl studies hve investigted the performnce nd complexity of vrious multiuser detectors in vriety of scenrios [2-[5. ost of these studies consider generic system models, which re not specific to ny stndrds-defined ir-interfce. It is noted tht multiuser detection cn be optionlly employed t the bse sttion on the WCDA uplink to improve system performnce. Our contribution in this pper is the derivtion nd performnce evlution of multiuser receiver structures for the WCDA uplink wveform. Specificlly, we consider multistge dptive prllel interference cncelltion receivers for the WCDA uplink nd evlute their performnces in ner-fr scenrio. On the WCDA uplink, ech ctive user trnsmits one or more (up to six) dedicted physicl dt chnnels (DPDCH) nd dedicted physicl control chnnel (DPCCH). While DPDCHs crry the user dt trffic, DPCCH crries control informtion. The DPDCHs nd DPCCH re orthogonl code multiplexed using orthogonl vrible spreding fctor (OVSF) codes. The dt rte for ech user cn be vried by vrying the spreding fctor on DPDCH nd/or by using multiple DPDCHs. The orthogonl code multiplexed DPDCHs nd DPCCH re further multiplied by the user-specific complex scrmbling codes [,[7. A conventionl receiver for the bove multiuser system will be bnk of mtched filters, ech mtched to userspecific complex scrmbling code nd the corresponding dedicted chnnel specific OVSF code. Here, we re interested in multiuser receivers for the WCDA uplink trnsmission. Prticulrly, we develop dptive multiuser receivers for detecting dt on the DPDCH. Since DPDCHs do not crry known symbols for trining, we derive blind dptive multiuser receiver (BAR) bsed on the constnt modulus lgorithm (CA) [8. We show tht this blind dptive receiver performs better thn the conventionl mtched filter receiver (CFR) in nerfr scenrios. We lso derive multistge dptive prllel interference cncelltion (APIC) receiver, which is similr to the one proposed in [5, but modified for the WCDA uplink wveform. In this APIC receiver, the interference is estimted t every stge nd cncelled from the received signl so s to provide n lmost interference-free signl for dt estimtion. Since the performnce of the APIC receiver t given stge is dependent on the relibility of the dt estimtes from the previous stge, we propose hybrid APIC (H-APIC) receiver structure which uses the CA bsed BAR s the first stge. Our performnce results show tht the H-APIC performs better thn the APIC receiver with moderte increse in receiver complexity. The rest of this pper is orgnized s follows. In Section II, we present the WCDA uplink system model considered. The receiver structures for the WCDA uplink, including CFR, BAR, APIC nd H-APIC receivers re derived in Section III. Performnce results re presented in Section IV, nd conclusions re given in Section V. II. SYST ODL We consider the WCDA uplink trnsmission from ctive users. ch user is ssumed to be trnsmitting one DPDCH (on I-brnch) nd DPCCH (on Q-brnch). In WCDA, the userspecific complex scrmbling codes cn be either short codes (of length 25) or long codes (of length 38400). Here, we ssume tht ll users use short scrmbling codes. The bsebnd trnsmitted signl from the user,, is given by!#"%$ ' 0 1( '32!4!578$ :0 =(7 2 > ' )( :)( $*-,,/. $*-;,/<?1BA C3D57 A CGF (1) where IK HG is the power of the trnsmitted signl of the user, L/ IK, O1QPSRKTUKV-WXU=Y represent the modulting DPDCH nd Z\[ DPCCH dt strems, respectively, for the user, nd nd Z re the gin fctors for the DPDCH nd the DPCCH, respectively. The DPDCH nd DPCCH dt strems re spred
š ^ ^ W T  _ using the wveforms ^ be written s nd `_b `_y 4 f1gihk f1gih O nd ^ l#:mnogtolip Ol#:mnOGTzlip O, respectively, which cn qvsrutvwtxp d (2) qv{rutxwtvpi =V (3) where Ol#, Ol#}P~RKTUKV-WXU1Y represent the li chip of the ) user s OVSF spreding codes for the DPDCH nd the DPCCH, respectively, nd mno represents the chip wveform of durtion p nd unit energy. d nd re the spred fctor vlues for the DPDCH nd the DPCCH, respectively, nd p d nd p re the symbol durtions of the DPDCH nd the DPCCH dt strems, respectively. Hence, p d _ d p nd p _ p. Also, ƒwy O represents the complex scrmbling wveform for the user, where nd `_ g c f=giĥ Ol#/m TzL:Qp Tzlip (4) `_ g c f=gih l#:mgtol:qp Tzlip!V (5) such tht Ol#)WŠ Ol#wPzRB X o Y represents the l complex chip of the user s scrmbling code. is the periodicity of the scrmbling code. For the short scrmbling codes considered here, Œ_b KŽ1. Although the uplink is synchronous, in order to focus minly on the reltive performnce of different multiuser receiver structures, we ssume tht ll the users trnsmissions rrive t the bse sttion synchronously. The received signl t the bse sttion due to ll the ctive users is then given by where )Oˆ_ " )O exp3 1 g d #W 11 > W l OqV () is the crrier frequency, is the rndom crrier phse of the user, which is ssumed to be uniformly distributed in rv 1œ#, nd l O is the dditive white Gussin noise with zero men nd two sided power spectrl density, X ž= W/Hz. The received signl is down converted nd the resulting sequence fter chip mtched filtering is given by Ÿ ewz K Ÿ, where q 3 3 ª - i 3 3 q 3 3 ª - i q D B±³ D B±³ 3ÓDµ 3ÓDµ 3 3¹ 3 3 Bº» Bº)! µ! µ 3 B 3 In the bove eqution, L is the integer prt of Ÿ ž! d nd is the integer prt of Ÿ ž!, nd l Ÿ W `l Ÿ is complex Gussin noise smple of zero men nd vrince of À ž1 W/Hz. (7) III. RCIVR STRUCTURS In this section, we derive vrious receiver structures, including CFR, BAR, APIC nd H-APIC receivers, for the WCDA uplink signl model described in the previous section. A. Conventionl F Receiver (CFR) The conventionl mtched filter receiver is essentilly bnk of mtched filters with ech filter mtched to the user-specific complex scrmbling code nd the corresponding dedicted chnnel specific OVSF code. This is low complexity receiver which needs informtion of only the scrmbling nd OVSF codes nd the timing of ll the ctive users. Using the CFR, I» soft estimtes of the DPDCH nd DPCCH dt symbols, IK I OL/%_ IK O1È_ OL/ nd O1 for the user re given by O 4:à d/ä OL: d WÆl# f=g 4% Ol# L d W l# OL: d W l# Ol# L: d W l#d à dä f=g W l# B W l# l# Ol# W l# OB W l#dëê (8) The DPDCH nd DPCCH bit estimtes re obtined to be the sign of the bove soft estimtes. B. Blind Adptive Receiver (BAR) Since the DPDCH on the uplink do not crry known symbols for trining, we derive blind dptive receiver (BAR). The BAR consists of bnk of dptive receivers, one for ech ctive user, which estimte the DPDCH nd DPCCH symbols from the received signl. Like the CFR, the BAR requires the knowledge of scrmbling nd OVSF codes nd the timing of ll the ctive users. The dptive receivers considered here re liner trnsversl filters whose coefficients re updted ccording to the constnt modulus Í lgorithm [8,[, so s to minimize the cost function,, given by Í U Î ƒïð TÑU Ò Ê () In the context of the system model considered, Ð represents the estimtes of DPDCH or DPCCH symbols. The dptive receiver for the user cn be implemented using four liner trnsversl filters, the impulse responses of which re given s R Ó Ol#qY, RBÓ l#-y, RÔ Ol#qY, RDÔ define the following vectors: Ø Gß ã Ø ã ß Ú _ Ol#-Y, lõ_yrvdd!vñt U. We Û Ó rk!v Ó /U qvddv Ó ÜT}U:ÝÞ àoó rk!v Ó /U!VDV-Ó átxu â Þ ÛÔ à Ô r»qvô rk!v/ô /UqVDDV/Ô /U!VDVÔ 8TxU Ý Þ átxu â Þ Ê (10)
Û Û Û Û ÿ ì ô ô Þ T ì ü Ò Ò ü Þ ÿ ÿ î î î î The elements of the vectors defined bove re the coefficients of the trnsversl I» filters. The estimtes of the DPDCH nd DPCCH symbols, IK OL/ nd O1 re given, respectively, s I š OL/%_ Ø»ä š å Ø Þ ä š W ß ä š å ß Þ ä I š O1È_ ã ß ä š å ß ä š ã»ä Ø š å Ø ä V (11) where š å š Ø the vectors ä š Gß, ä š, ã Ø ä š, ã ß ä š å Ø, ä š å ß, ä š å ß, ä ß nd ä 3Óî re defined Dïð î:ò/òò/ 3 óî s î õdïnð Û æ ç è Ýéë Ûíì 3 óî ïð în ò/òò Óî î ķïð Ýö Û æ) è Ý é Ûíì!î üqdïð î:ò/òò ì!î ü î ü ķïð ÛCø èç Ýù ÛCû 3qî ü ïð în òò/ò!î ü î ü ķïð ÛCø» è Ý ù ÛC 3Óî òò/ò/ Óî î ķ ÛCý ç Ý é Óî :/òòò/ Óî î ķ Û ý Ýé q3óî üq òò/ò/ q!î ü î ü ķ Û ý ç Ý ù Óî ü!:/òòò/!î ü î`ü ķ Û ý Ýù nþ (12) We define cost functions, Í nd Í ü 3 ó xô, corresponding to DPDCH where nd DPCCH symbols, respectively, s IK º ü initil estimtes, Ó Ï }ô i.e., ÿ º Ï The obective is to minimize the bove cost functions by dpting the filter coefficients. ï The filter 3ïn¹ coefficients re updted ccording to the following reltions: Û æ è ï ç Ý é Û æ è 3ïn¹ ç Ý é º ï 3ïn¹ º Û æ) è Ý é Û ø è ï ç Ý ù ÛCø è Ýù D Û æ) è Ý é Û ø èç Ý ù ÛCø è Ýù 3ïn¹ ÿ þ (13) º ÿ º (14) where is the grdient of w.r.t. The constnt refers to the step size nd refers to the itertion index. Substituting the expressions for Í Í, Í Í nd, in ï 3ïn¹ Ó ü õô Ó (14), we obtin the following updte equtions: Û æ ç è ï Ýé ï Û æ) è Ýé Ûø è ï ç Ý é Û ø» è Ý é 3ïn¹ Û æ ç è Ýé 3ïn Û æ) è Ýé ÛCø è ï ç Ý é Û ø» è Ý é Ó ü õô Ò ÛCý-ç Ýé 3Ó Ó ü ķô Ó Ò ÛCý Ýé Ó ü õô Ò Ûíý ç Ý é 3Ó Ò Û ý Ý é Ï Ï Ï Ï (15) It is noted tht the cost functions Í nd Í re multi-modl, s result of which there cn be more thn one chievble minim [8. Hence, "!$#% the initiliztion µ ïð of the î liner filters is importnt. A nturl choice of initiliztion would ¹ò!$#% be to use µ ïð î ¹ò '!(#% ì µ ïð î üqò!(#% µ ïð î üqò (1) where the superscript ) r* indictes the initil stte of the filters. C. Adptive Prllel Interference Cnceller (APIC) The dptive prllel interference cnceller derived by Xue et l in [5 is n dptive multistge receiver. The obective is to minimize the ucliden distnce between the received signl nd the weighted sum of the estimtes of ech user s signl during bit intervl w.r.t the weighting fctors. These weighting fctors re updted through n dptive lgorithm for ech user. Let Ÿ nd Ÿ be the estimtes of Ÿ nd Ÿ, respectively, for the K stge.! These -.% estimtes re defined s q ª : 0/ Óµ/ 3 B - /( 4! -.% /,!µ /( 4 3 3 ª 0/ Óµ/ 3 B - /( 4 ü:ü! -.% /,!µ /( 4 (17) IK :( 4 OL/ IK :( 4 nd 1 I» re the estimtes of OL/ nd O1, respectively, t the nt U : stge for the ) user. The I» $1 OL/ I» (1, 1 re provided by the CFR, Ó Ó54 $1! 32!54w D $1 32 (18) IK where OL/ I» Â$7 Ä nd L: re given by (8). The fctors d d'8, Â(7 Ä Â(7 Ä Â(7 Ä d 8, d"8 nd 8, re the weighting coefficients corresponding to the K stge for the user. Let nd represent the error between the desired received signl nd its estimte t the stge for the Ÿ chip. 3 3 Then, 3 3 þ (1) We need to find the optiml weighting coefficients so tht the squre of the error terms, given in the bove eqution, is minimized.! -:% The optimum weights re derived vi LS lgorithm s : 0/ follows:! -.% <; : 0/ Ó µ/! -:%! -.% ; /( 4 / ü! -:% 0/ ü:ü! -:% / / 0/ ü:ü! -.% / <; ;! µ :( 4 Ó µ/ :( 4 3Ó µ :( 4 : (20) where >= is the step size used in the dpttion. The choice of the initil vlues of the coefficients is importnt to chieve fster convergence of the weighting coefficients to their optiml vlues. Generlly, if knowledge of ll user s mplitudes re vilble, the initil vlues of the weighting coefficients for ech user re set to its corresponding mplitudes. The bove updtion is done for itertions nd the resulting weights t the end of the Q itertion re used in the interference cncelltion. The interference-free estimtes for the
?? é ù é ù? 4' Î _ - ¹ ª 0/ ³! -:% / óîxķ BA user is obtined s? 3 3 ª i 0/ ü:ü! -:% / óîxķ! -.% : 0/ îvõ, 3!Dµ :( 4 3 3 0/ óî Š ¹, 3!Dµ :( 4 3 3 BA :( 4 3 :( 4 3 3 3 ó µ Ó µ 3 3 3 (21) where the nottion ËÜTÆU indictes the vlue of the weighting coefficient t the Q itertion. The bove interferencefree estimtes re used!f in forming %IH the estimtes of DPDCH nd DPCCH 3Ó symbols to be used by the Óî next stge. Thus, we hve 4:G µ B Óî DC u k 3!? Óî 4 µ!$ À %IH G C 3!î ü Dµ u 3Óî u!î ü µ u!î ü À? LK B!î ü u K þ (22) D. Hybrid-Adptive Prllel Interference Cnceller (H-APIC) In ner-fr scenrio, the performnce of APIC for wek user cn be good t high ner-fr rtios d (FR) becuse the stronger intereferers dt cn be relibly detected in the first stge by the CFR, nd hence the interference signl cn be reconstructed nd removed effectively for the next stge. For exmple, in 2-user scenrio with high FR vlue, the performnce of the wek user using APIC cn pproch singleuser performnce since the CFR in the first stge cn relibly detect the strong interferer s dt so tht the input to the second stge cn be lmost interference-free. However, t low FR scenrios, the relibility of the interferers dt my not be s good, nd hence the APIC my perform poor becuse of which more stges my be required to chieve desired performnce. To llevite the performnce of APIC in such low FR scenrios, we propose the following hybrid APIC (H-APIC) which uses the BAR in the first stge, insted of CFR. The BAR would give relible estimtes of the dt symbols of the other users in the initil stge itself, thus iding better estimtion of the interference for the weker user. This lso reduces the number of stges required to obtin relible estimtes. The initil estimtes, OL/ I» $1 I» $1 nd O1, re then given by I (1 OL/%_ Ô l I» $1 O1 _ Ô l é user-1 is tken s the desired user. When I V OL/D IK O1 O }ô (23) er-fr rtio is defined s the rtio of the received power of interfering user to the received power of the desired user, i.e., FR =, where, FR = 0 db. I where L: I» nd OL/ re given by (11). IV. RSULTS AD DISCUSSIO We evluted the performnce of the vrious multiuser receivers described bove in ner-fr scenrio through simultions. A system with four ctive users (Œ_ ) is considered nd user-1 is tken to be the user-of-interest. The ner-fr rtio, H ž1h d V³QP U of ll interfering users is ssumed to be equl. The spred fctor for the DPDCH nd the DPCCH symbols for ll the users is set to 25, i.e., d _{X n_s =Ž=. The bit error performnce of the user-of-interest (user-1) s function of RTSqž! h nd FR is evluted. Fig. 1 shows the bit error performnce s function of R S žq h for the vrious receivers including CFR, BAR nd APIC t FR vlue of 15 db. The performnce of APIC is plotted for different number of stges (1-stge APIC, 2-stge APIC nd 4-stge APIC) without nd with perfect knowledge of the mplitudes. The single user performnce is lso plotted for comprison. The following observtions cn be mde from Fig. 1. As expected, the BAR nd the APIC perform much better thn the CFR. Also, when perfect knowledge of the mplitudes is not vilble t the receiver, incresing the number stges in the APIC improves performnce. In the unknown mplitudes cse, more thn four stges re required to perform close to the single user performnce. However, with perfect knowledge of the mplitudes t the receiver, even 1-stge APIC is shown to chieve close to single user performnce. Fig. 2 shows the bit error performnce s function of FR for CFR, BAR nd APIC t R S žqà vlue of 8 db. The rnge of FR vlues considered is 0 to 15 db. The CFR performnce degrdes with incresing FR indicting its poor ner-fr resistnce. The ner-fr resistnce of BAR nd APIC re shown to be much better. In the unknown mplitudes cse, the nerfr resistnce of APIC improves s the number of stges is incresed. Two key observtions cn be mde in Fig. 2. Firstly, in the high FR region (FR U Ur db), the APIC with known mplitudes performs better thn BAR nd chieves close to single user performnce becuse of the high relibility of the interfering users dt estimtes. Secondly, in the low FR region (FR VXW db), however, even in the known mplitudes cse, the APIC performs poorer thn BAR. This implies tht t low FRs, the number of stges in the APIC hs to be incresed to chieve given performnce, even though perfect knowledge of the users mplitudes is vilble t the receiver. This is minly becuse the relibility of the dt estimtes of the interfering users in the initil CFR stge is poor t low FRs. Fig. 3 shows the bit error performnce of the APIC nd the proposed H-APIC s function of RTSqžq t low FR vlue of 5 db. It is noted tht, even with no knowledge of the users mplitudes, single stge H-APIC performs better thn single stge APIC with perfect knowledge of ll the users mplitudes. A single stge H-APIC with perfect knowledge of the users mplitudes is shown to perform the best nd its performnce is close to the single user bound. Fig. 4 shows the bit error performnce of the APIC nd H-APIC s function of FR t R S ž!x vlue of 8 db. It is observed tht t low ner-fr rtios, in the rnge 0 to 7 db, the ner-fr resistnce of single
10 0 CFR BAR 1 stge 10 0 APIC, Unknown mplitudes 2 stge APIC, Unknown mplitudes 4 stge APIC, Unknown mplitudes Probbility of Bit rror CFR BAR 1 stge APIC, Unknown mplitudes 2 stge APIC, Unknown mplitudes 4 stge APIC, Unknown mplitudes Probbility of Bit rror 0 1 2 3 4 5 7 8 / (db) b o î # Fig. 1. Bit error performnce of CFR, BAR, nd APIC receivers s function of Y[Z5\.. FR = 15 db. î # _^ 0 5 10 15 er Fr Rtio (db) Fig. 2. Bit error performnce of CFR, BAR, nd APIC receivers s function of FR.. Y Z \ db. stge H-APIC with no knowledge of users mplitudes is better thn tht of single stge APIC with perfect knowledge of ll the users mplitudes. As the ner-fr rtio increses beyond 7 db, the APIC with perfect knowledge of the users mplitudes begins to perform better. The single stge H-APIC with perfect knowledge of ll the users mplitudes performs the best mong ll the receivers considered. V. COCLUSIOS We investigted the performnce of dptive interference cncelltion receivers for the WCDA uplink physicl dt chnnel. We derived blind dptive receiver (BAR) bsed on the constnt modulus lgorithm, nd n dptive prllel interference cncelltion (APIC) receiver. With motivtion to improve the performnce of APIC receiver under low FR conditions, we proposed hybrid APIC (H-APIC) receiver which used the BAR s the first stge. We evluted nd compred the performnce of the bove receivers in ner-fr scenrio nd showed tht the H-APIC receiver performs better thn the APIC receiver. RFRCS ± [1 H. Holm, A. Toskl, WCDA for UTS - Rdio ccess for third genertion mobile communiction, ohn Wiley, October 2000. [2 S. Verdu, ultiuser Detection, Cmbridge University Press, 18. [3 Z. Xie, R. T. Short, nd C. K. Rushforth, A fmily of suboptimum detectors for coherent multiuser communictions, I l. Sel. Ares Commun., vol. 8, pp. 83-0, y 10. [4. L. Honig, U. dhow nd S. Verdu Blind dptive multiuser detection, I Trns. Informtion Theory, vol. 41, pp. 44 0, uly 15. [5 G.Xue,. Weng, T. Le-goc nd S. Thr, Adptive multistge prllel interference cncelltion for CDA, Proc. I ICC, 1. [ 3GPP Technicl Specifiction 25.211: Physicl chnnels nd mpping of trnsport chnnels onto physicl chnnels (FDD), V4.5. [7 3GPP Technicl Specifiction 25.213: Spreding nd odultion (FDD), V4.5. [8. R. Treichler nd B. G. Agee, A new pproch to multipth correction of constnt modulus signls, I Trns. ASSP, vol. 31, pp. 45 471, April 183. [ P. He, T. T. Thung, nd L. K. Rsmussen, Constnt modulus lgorithm for CDA communiction systems, Proc. I VTC 8, pp. 4 53, 18. [10 S. Hykin, Adptive Filter Theory, ed., nglewood Cliffs, : Prentice-Hll, 11. Probbility of Bit rror 10 0 CFR 1 stge APIC, Unknown mplitudes 1 stge H APIC, Unknown mplitudes 1 stge H APIC, Known mplitudes 0 1 2 3 4 5 7 8 b / o (db) î # Fig. 3. Performnce comprison of APIC nd H-APIC receivers s function of Y Z \.. FR = 5 db. CFR 1 stge 10 0 APIC, Unknown mplitudes 1 stge H APIC, Unknown mplitudes 1 stge H APIC, Known mplitudes Probbility of Bit rror î # _^ 0 5 10 15 er Fr Rtio (db) Fig. 4. Performnce comprison of APIC nd H-APIC receivers s function of FR.. Y>Z5\ db.