DYNAMIC SYSTEM LEVEL PERFORMANCE FOR MC-CDMA SCHEME J. Rodriguez, X.Yang, D. Mavrakis, R. Tafazolli* D.T. Phan Huy**

Transcription

1 DYNAMC SYSEM EVE PEFOMANCE FO MC-CDMA SCHEME J. odrguez, X.Yang, D. Mavraks,. afazoll* D.. Phan Huy** *Centre for Communcaton Systems esearch, Un. of Surrey, uldford, Surrey. UK e-mal: **France élécom &D, rue du énéral eclerc ssy-es-moulneaux cedex 9, France. e-mal: Abstract: UMS s expected to provde an ntal attempt to support multmeda servces to the end user, but stll system throughput s lmted n hgh moblty, and large coverage areas. MACE (Multcarrer cdma ansmsson echnques for ntegrated broadband CEllular systems) ams to reach the next step n cellular communcatons by supersedng 3 to provde hgh bt rate packet based servces n moderate to fast moblty envronments n a cost effectve manner. he ar nterface s based on DD MC-CDMA, and we nvestgate here the mpact on system level performance. Furthermore, we employ a dynamc system level evaluaton tool, so that global performance statstcs also take nto account user moblty, and the defned MACE scenaros. 1. ntroducton he MACE proect ams to provde a feasblty study; to nvestgate an MC-CDMA ar-nterface as a potental canddate for beyond3 cellular systems. he motvaton for MC-CDMA leads to ts ablty to support hgh speed, delay strngent servces to the end user, n a cost-effectve manner. t promses to fulfll these expectatons, by ntegratng state-of-the-art sgnal processng schemes to maxmse channel throughput such as MMO technology, advanced Mult-User Detecton, and beamformng [1],[2]. n the ntal stage, the deployment wll target solated areas, operatng n the 5Hz band. Wthn the MACE framework les some challengng requrements at system level [3][4]: to provde 0Mbps n ndoor envronments, up to 20 Mbps n urban envronments, and up to Mbps at 300km/h. n ths paper, we nvestgate the mpact of MC-CDMA chan on system level performance, for the urban operatng envronment n the presence of real-tme dedcated traffc channels. We provde some ntal fndngs n terms of system capacty based on the MACE downlnk physcal layer reference chan operatng n DD mode. We am to evaluate dynamc system level performance, so that the effects of user moblty, and handover are taken nto account to attan a complete and realstc assessment of global system performance. hs paper has the followng layout: secton descrbes the system level evaluaton tool that supports the evaluaton of rado resource management enttes, desgned to support real tme servces, and dynamc smulatons for the defned test envronment; secton consders the lnk level nterface, and the specfcaton of the MACE reference lnk level platform; the smulaton results are gven n secton V; and fnally the concluson n secton V. 2. System evel Evaluaton Envronment n ths secton we recall the smulaton envronment whch ncludes a model descrpton of the test scenaro and the rado resource management enttes. 2.1 est Envronment n ths paper, we model an urban moblty envronment, whch s a subset of the MACE scenaros. he test scenaro s defned by the followng models: Deployment Model he cell radus s 300m. he deployment scheme s assumed to be a hexagonal cell layout. Omn-drectonal antennas are assumed Moblty Model he moblty model for the Vehcular est envronment s a pseudo random moblty model wth sem-drected traectores. he Moble s poston s updated accordng to the de-correlaton length, and drecton can be changed at each poston update wth probablty 0.4, and wth a maxmal angle for drecton update gven by 45. n addton, we assume the mobles are unformly dstrbuted on the map and ther drecton s randomly chosen at ntalzaton. o mnmse smulaton tme, and to beneft from accurate nterference modellng, we use a wrap around effect to model user moblty at the cell boundares. he users may enter or arrve at dfferent cells departng from a sngle cell accordng to ther drecton. Fgure 1 llustrates the results from a test smulaton wth a user movng around a 4-ter cell envronment wth a 300 m cell radus. Fgure 1: Sngle user traectory for urban moblty model

2 nterference modellng for the cell of nterest whch s not the centre cell, wll take nto account the base staton, and moble transmtted powers from sources that are four ters ways, a subset of these wll be from nterferng sources that are wrapped around. Furthermore, cell statstcs n terms of total throughput, and user QoS can be taken from all cells wth confdence, per smulaton run Propagaton channel We use the path loss model that was defned for the "Vehcular est Envronment" n [5]. hs model s applcable to test scenaros n urban and sub-urban areas outsde the hgh-rse core where the buldngs are of nearly unform heght. he path loss formula s gven by: = 40(1 4 hb )]log m ( ) 18log km hb ( ) + 21log m 3 [ f ( ) + 80dB MHz where s the base staton-moble staton separaton, f s the carrer frequency and h b s the base staton antenna heght, measured from the average rooftop level. herefore, for a carrer frequency of 5000 MHz and a parameter h b set to 15 m, the path loss expresson becomes: = log ( ) (2) km Shadowng At the system level, we assume the receved power to be a slow varyng quantty due to the shadowng effects. hs phenomena, typcally known as slow fadng s taken nto account n receved power calculaton by multplyng the transmt power by a random log-normal varable. he log of the shadowng varable has aussan characterstcs wth zero mean and a standard devaton σ n db. he shadowng effect s correlated n dstance, therefore the values of the shadowng varable for two postons of the moble staton separated by x are correlated. he slow fadng process for the moble user can be descrbed by eqn(3)[4]. [ db] t = (1) 2 t t[ db] + 1 X[ db] (3) Where t [db] s the Shadowng value at tme t; t- t [db] the Shadowng value at tme t- t; X n db s an ndependent log-normally dstrbuted random value wth mean of zero and standard devaton of σ; t = d cor /v; where v s the user's speed; d cor = he de-correlaton length defned as the value of the covered dstance x, for whch the auto-correlaton s equal to ½; s the normalsed autocorrelaton functon of the shadowng. 2.2 ado esource Management We now recall the rado resource management enttes that form part of the system level archtecture Call Admsson Control he obectve of Call Admsson Control s to regulate the operaton of a network n such a way that ensures unnterrupted servce provsonng to the exstng connectons and to accommodate n an optmum way the new connecton request. CAC s performed when a moble staton requests communcatons, and s performed separately for uplnk and downlnk. hs s especally mportant f the traffc s hghly asymmetrc; however n ths paper we only consder the downlnk scenaro. Downlnk CAC s assumed based on the downlnk total transmsson power.e. the new connecton s admtted f the new total downlnk transmsson power does not exceed the predefned target value [6]: P + P < P t otal _ old total threshold where the threshold value s set by rado network plannng. he total base staton transmsson power can be presented as: = p' + P' (5) P total _ old tb, ctl tb, x x cellb where p tb,ctl represents the control channel transmsson power, and P' tb,x, s the downlnk transmsson power of cell b allocated to moble x. he load ncrease P total n the downlnk can be estmated based on the ntal power. he ntal power depends on dstance from the base staton and s determned by the open loop power control algorthm nk Adaptaton o mantan user throughput over the wreless channel, we need lnk adaptaton technques that am to mantan lnk qualty at the desred level. nk qualty s senstve to the rado condtons, and servce requrements, leadng to several alternatve solutons. For eal tme servces such as crcut-swtched servces, voce servces, and vdeo conferencng (and less strongly streamng vdeo) requre low delay, low tter, and constant qualty. We use CPC (Close-oop Power Control) that adapts the average S to the desred target level, therefore the scheme compensates for the path loss, and slow fadng varatons n the receved sgnal, whch means that these assumptons must be reflected n the lnk level scenaro; therefore the fast fadng effect s modelled, and confned to the lnk smulatons only. he CPC functon s called on every pc perod. pc s chosen to be much larger than the fast fadng coherence tme and much lower than the shadowng coherence tme. At tme k pc, the CPC functon s called for the UE (User Equpment). he CPC functon frst checks f the UE s n the recevng state. hen t updates the D nterference and the D S, usng the most recent values of the D transmsson powers, path loss and shadowng. he CPC functon deduces the necessary amount of transmsson power P new as a functon of the current S value, S target, and the prevous BS transmsson power P t accordng to [7] n lnear scale: (4)

3 St arg et Pnew = Pt S (6) he calculated transmsson power s then bounded n such a way that P new =max(p mn, mn(p new, P max )). Before allocatng the new transmt power, the BS total transmsson power P tot s computed: f P tot +P new -P old <P maxbs (.e. the Base Staton can provde the requred amount of power) then, the new transmsson power s allocated: P t =P new.; f P tot +P new -P old P maxbs, (.e. the Base Staton cannot provde the requred amount of power), then P new = P old +P maxbs. - P tot (the requred power s truncated and then allocated). For the D CPC, P mn s chosen equal to zero n lnear value and P max s chosen equal to the maxmum BS transmt power P maxbs Handover Handover s one of the essental features of cellular moble systems used to mantan user throughput at the cell boundares. n FDD CDMA based systems, soft handover scheme s mplemented, where the new lnk between a user and the target base staton s bult before breakng the old lnk from the source base staton. Soft handover mproves the qualty of servce due to the sgnal dversty provded n both lnks[8], however, t ntroduces more nterference n the downlnk and s more complex to mplement n DD systems due to synchronsaton ssues. n hard handover systems such as SM, the user s connected only to a sngle base at any gven tme, therefore the connecton s broken, before a new lnk s establshed. We consder Hard Handover to be more approprate soluton for a DD MC- CDMA system, at the expense of no handover dversty gan. We now recall the hard handover algorthm employed n the system. he th averaged Ec/o from CPCH (Common Plot Channel) from cell can be gven by: N 1 ( Ec / o), = ( Ec / o), k (7) N k = 1 where we assume N s the flter tap length. However we assume that N s suffcently large so as to average out the fast fadng effect n the channel. When the CPCH of the best canddate cell s better than current servng cell by a quantty Hyst (hysteress value used to prevent mmature handovers), the handover wll be performed: ( Ec / o ) ( Ec / o ) > max ser Hyst (8) where ( Ec / o) max ndcates the best CPCH among canddate cells and ( Ec / o) ser corresponds to the CPCH of current servng cell. 3. nk evel nterface he lnk level nterface tres to map system level scenaros to physcal layer parameters so as to provde cross-layer coherency. A soluton can be to have a ont system-lnk level smulator to provde real-tme processng of nformaton bts to provde nstantaneous block error rate (BE) readngs, however ths would be computatonally excessve, and would result n large smulaton tmes n mult-cell, mult-user envronments. herefore typcally look-up tables are used based on the average value nterface technque, whch wll map the average sgnal-tonterference rato to the BE statstcs. n ths way, we model the effect of the physcal layer transmsson on user throughput, through statstcal tables. hs scheme s vald for real-tme crcut-swtched servces, due to followng assumptons: the nformaton bt rate s constant; the perod of actvty of a real tme connecton s very long when compared wth the fast fadng coherence tme; the Block Error ate must be constant and equal to the target BE; for the nterference computaton, we neglect nter-slot nterferences (Only neghbourng cells need to be consdered for the nterference calculaton, we can assume that the sgnal from the servng cell and the nterferng sgnals arrve quas-synchronously at the moble). 3.1 nk evel MACE V.0 chan We now recall the lnk level chan that s requred to produce the ook-up ables (U) and the Orthogonalty Factor (OF). he MACE V0 D system level smulaton chan s gven by Fgure 2. EF_X_D EN_BNAY_SOUCE_X A EN_ENCODN_X B EN_PUNCUN_X C EN_NEEAVN_X D EN_MAPPN_X E EN_USE_MUX_D F EF_SPEADN_D EF_FEQ_MUX_X H EF_OFDM_FAMN_X EN_OFDM_MOD_X J Z EN_SSO_CHANNE_D EN_EO_AES_X V EF_DECODN_X U EN_DEPUNCUN_X EN_DENEEAVN_X S EN_DEMAPPN_X EN_USE_DEMUX_D Q EF_DESPEADN_D P EF_EC_EQUA_D O N EF_OFDM_DEFAMN_X M EN_OFDM_DEMOD_X EN_AWN_X EF_CHANNE_D Fgure 2: eference lnk level smulaton chan able 1 provdes a lst of the Physcal layer smulaton parameters used to generate the U. V0 set of WP3 system parameters Channelzaton bandwdth B C 50 MHz Occuped bandwdth B O MHz Samplng frequency 57.6 MHz = f S =1/ S 15*3.84 MHz Slot duraton P ms W EF_FEQ_DEMUX_X X EF_X_D EF_FEQ_DEMUX_X K

4 Number of samples per slot N P Number of OFDM symbols per slot N S 30 Symbol total duraton 21.5 µs uard nterval 3.75 µs U/D guard tme µs Codng ate c 1/2, 2/3, 3/4 able 1: V0 D smulaton Parameters We now defne the nk evel smulaton assumptons that have been used to generate the U for our system level smulatons: we assume that no Mult-user detecton s performed; all downlnk users are synchronous but experence mult-path fadng; perfect synchronzaton, and channel estmaton; mono-code, and mono slot connecton; and a sngle modulaton, and codng scheme s smulated (1/2 code rate, and QPSK modulaton s assumed). 3.2 nterference Modellng n the system level smulatons, we consder the followng defnton for the nterference modellng. Consder a moble M, camped on base staton B, and usng the downlnk transport channel C k.(.e. usng a defned set of frequency slots, tme slots and codes). he average receved power P (M,B,C k ) at the desred moble M, from BS B, for the gven downlnk transport channel C k s gven by the followng equaton n lnear scale: P ( M, B, C ) k P ( M, B, C ) k M (9) = M P M, where P M, B, C ) s the sgnal power transmtted by ( k BS B to UE M on transport channel C k averaged durng pc; B and M are the base-staton, and moble antenna gan respectvely; M s the Body oss or/and ncar oss of UE M ; s the Cable loss of BS B ; and M P, s the propagaton loss between BS B and UE M (ncludng path loss and shadowng). o compute the average receved nterference power of the UE, we need to capture the nter-cell and ntra-cell nterference n lnear scale. β () M, B, C ) = ( M, B, C ) + β P ( M, B, C ) + ( k 1 ntra k 2 k nter k he ntra-cell component s gven by s: ntra ( M, B, C ) = k M nt ra M P ( M nt ra, B, Ck ) P ( M, B, C ) M M, B B B M (11) where M ntra s a UE connected to BS B and recevng on the same set of frequency and tme slots as the transport channel C k, P M, B, C ) s the correspondng ( nt ra k nterferng transmt power (transmt power from BS B to UE M ntra on the same set of frequency and tme slots as the transport channel C k ); and β 1 s the Orthogonalty factor for ntra-cell nterference from other UEs other than M. he nter-cell nterference n lnear scale s: nter M Bnt er ( M, B, Ck ) = P B er B ( Bnt er, Ck ) (12) nt P M, Bnt where P ( B nt er, C k ) s the transmsson power of base statons B nter other than B, on the same set of tme and frequency slots than the transport channel C k ; and β 2 s the Orthogonalty factor for ntra-cell nterference from UE M. We compute the updated S value n lnear scale for the current [(k-1)pc; kpc]: P ( M, B, Ck ) S( M,, Ck ) = ( M, B, C ) + N W F k 0 Ck M er Bnt er M (13) Where F M s the Nose fgure of UE M ; N o s the hermal nose densty, and channel C k. WC k s the bandwdth occuped by the 3.3 nterface Structure hs leads us to defne the followng nterface structure gven by Fgure 3 P u Pu o CH PDU = f(mcs) β U CC DEC MC-CDMA x Orthogonalty Factor BE AV BE Fgure 3:Average Value nterface Scheme ACUA Where Pu s the desred sgnal power receved by the consdered MS; o s the total nterference consstng of or (ntra-cell nterference receved power), oc (nter-cell nterference) and nose receved power; BE (Block Error ate), whch s the rato of erroneous receved nformaton data blocks over the total number of receved nformaton bts durng the overall lnk level smulaton; the orthogonalty factor β = f (MCS, envronment); herefore, t can be sad that the average value nterface n ths nstance, s equvalent to the actual average BE determned n a practcal system that uses the cyclc redundancy codes to determne the number of bts n error. 4. Numercal esults A set of smulatons have been carred out to evaluate the mpact of MC-CDMA on system level performance. he smulaton parameters are gven by able 2 Smulaton Parameters Shadowng de-correlaton length d cor Settng 20m (Urban)

5 Shadowng standard devaton 8dB (Urban) BE target -3 P max 2 w P mn 0 w P threshold 15.8 w P tot 20 w S target db Handover Hyst 3 db MCS1 QPSK 2/3 rate turbo encoder nformaton rate per user 55.2 kbt/s Orthogonalty Factor ( ) D me slots 14 DCH 30*14 resource unts Smulaton me 2 hours User Moblty 60 km/h able 2. Smulaton parameters and settngs o evaluate the system throughput, and effcency, we nvestgate the percentage of satsfed users aganst spectral effcency, where the former term defnes the average number of users that satsfy the followng condtons: Users that do not get blocked when requestng a connecton Users that do not get dropped wthn a sesson Over the call sesson, the S < S target for less than 5 % of the call sesson he spectral effcency defnes the offered load to the system normalsed by the total bandwdth.e. Kb/s/cell/MHz. Fgure 4 shows the % of satsfed users versus spectral effcency for a D MC-CDMA DD system. Fgure 4 % of satsfed users v spectral effcency Fgure 4 shows the tradtonal underlyng trend, as we ncrease the system load, the generated nterference ncreases the average power allocaton to the exstng connectons, and thus causng the system to be nterference lmted rather than havng hard capacty, whch s characterstc of CDMA systems. We defne the system capacty to be the system load that can support a 98 % user satsfacton rato. Fgure 4 show that a spectral effcency of around 0.23 can be acheved, whch s smlar to the effcency of a UMS system. hs effcency maps to an actual system throughput of 11.5 Mbps, where the theoretcal throughput s constraned to Mbps. Fgure 5 addresses the valdty of the hard handover mechansm. n the smulatons, the handover hysteress has been chosen so that on average the hard handover takes place around the cell boundary regon, whlst keepng the Handover sgnallng overhead to a mnmum, as can be seen n fgure 5. Fgure 5 co-ordnate map that shows handover postons he handover mechansm wll have an effect on system throughput, dependng on the values chosen for the hysteress value, and the S averagng flter tap length. A handover must happen when the average power requred to support the dynamc user would be lower on an adacent cell when compared to the servng cell, n order to maxmse system throughput. However, f we are to consder handover sgnallng as addtonal desgn requrement, then we would need to trade-off between throughput, and sgnallng overhead leadng to the desgn of adaptve handover algorthms; however ths s out of the scope of ths paper. 4 Concluson n ths paper, the mpact of the Downlnk MACE DD MC-CDMA reference chan on system level performance has been nvestgated. he smulator archtecture encapsulates slow closed loop power control to adapt the D average S varatons to the S target, and hard handover desgned to optmse system throughput only. he effect of call admsson control based on downlnk avalable transmsson power has also been ncluded to mnmse the call blockng rate. Moreover the effect of the MC-CDMA physcal layer has been taken nto account through an extensve look-up table, that provdes nformaton surroundng the average orthogonalty factor, and the S targets based on QPSK, and 2/3 rate turbo encoder based on frequency doman spreadng. t has been shown that the MACE system can support large system throughput due to ncreased avalable bandwdth, and thus provde at least 11.5 Mbps whch can support up to around 200 voce users per cell. hs throughput can be supported n an urban envronment at 60 km/h. However, these are

6 prelmnary fndngs, and larger throughputs are antcpated, when we take nto account the enhanced features of the physcal layer that are currently beng nvestgated n the MACE proect. Acknowledgements hs study was sponsored by the S MACE proect (Proect number S ) eferences [1] S. Hara and. Prasad, Overvew of multcarrer CDMA, EEE Communcatons Magzne, pp , December, 1997 [2] D. Motter et al. Physcal ayer Smulaton Chan Descrpton, S MACE, D3.1, Nov [3] J odrguez, D.. Phan Huy et al., ado nk ayer Desgn for S MACE proect, S Moble and Wrless Communcaton Summt 2003, vol pp847. [4] D.. Phan Huy et al., Specfcaton of the performance evaluaton methodology and the target performance, S MACE, D1.3, Jan [5] ES, "Unveral Moble elecommuncaton System (UMS); Selecton procedures for the choce of rado transmsson technologes of the UMS (UMS verson 3.2.0)" v3.2.0, Aprl [6] H. Holma and A. oskla, WCDMA for UMS- ado Access for hrd eneraton Moble Communcatons, John-Wley & Sons, 2000 [7]. J. Foschn and Z. Mlanc, A smple dstrbuted autonomous power control algorthm and ts convergence, EEE ran. Veh. ech., vol 42, pp , 1993 [8] A. J. Vterb et al., "Soft Handoff Extends CDMA Cell Coverage and ncreases everse nk Capacty," EEE J. Select. Areas Commun., vol. 12, no. 8, pp , Oct. 1994