Optimal Power Masking In Soft Frequency Reuse based OFDMA Networks

Transcription

1 Eupean Wieless 009 Optimal Pwe Masking In Sft Fequency Reuse based OFDMA Netwks Mathias Bhge* James Gss", Adam Wlisz*+ *Technische Univesitat Belin, TKN Gup, Einsteinufe, 1087 Belin, Gemany bhgc Otkn.tu-bclin.dc, Tel.: +9 (0) 1-8 trwth Aachen Univesity, UMIC Reseach Cente, Mies-van-de-Rhe-St. 1, 07 Aachen, Gemany gss Ounic.wth-aachen.de, Tel.: +9 (1) Univesity f Califnia, Bekeley, BWRC, 108 Allstn Way, Suite 00, Bekeley, CA , USA aw@ieee.g Abstact-Sft euse is a stng tl f c-channel intefeence mitigatin in cellula OFDMAILTE netwks. The pefmance f such netwks signifintly depends n the cnfiguatin f the pwe masks that implement the sft euse pattens. In this pape, we investigate the pefmance f diffeent pwe mask cnfiguatins against the ptimal se, in which a cental entity ptimally distibutes pwe and esuce blcks amng the uses f the netwk. It is shwn that lage diffeences exist between the pefmance f diffeent mask types and the ptimal se in bth, the veall cell thughput, as well as the cell-edge use pefmance. 1 Index Tems-LTE, OFDMA, sft euse, pwe masks, ptimizatin I. INTRODUCTION The Lng Tem Evlutin (LTE) mbile badband system [1] uses Othgnal Fequency Divisin Multiple Access (OFDMA) as cmbined tansmissin and multiple access technique in the dwnlink. With OFDMA, the system bandwidth is split int a numbe f sub-ies, each featuing a bandwidth smalle than the system's cheence bandwidth, n which data f diffeent uses is tansmitted in paallel. While the sub-ie thinness and the esulting lage OFDM symbl time educes the effect f inte-symbl intefeence (lsi), the thgnality amng them mitigates inte-ie intefeence (ICI). By using apppiate cyclic pefixes, lei and lsi n almst cmpletely be avided. OFDMA theefe is a pmising technique f use in vaius systems and scenais. When applied t mbile cellula systems, a key issue with OFDMA is c-channel intefeence (CCI): Especially teminals lted at the cell bde lagely suffe fm the pwe adiated by the base statin f neighbing cells in thei cmmunitin band. Thee ae thee maj altenatives f mitigating CCI in cellula OFDMA systems: had euse (HFR), factinal euse (FFR), and sft euse (SFR). 1This wk has been suppted by the Geman Ministy f Edutin and Science (BMBF) and Eicssn Reseach, Gemany, in the cntext f the pject SleNet. Had euse splits the system bandwidth int a numbe f distinct sub-bands accding t a chsen euse fact and lets neighbing cells tansmit n diffeent subbands. Factinal euse and sft euse bth apply a euse fact f ne t teminals lted in the cell's cente. F teminals clse t the cell edge, hweve, a euse fact geate than ne applies. Factinal euse [] splits the given bandwidth int an inne and an ute pat. The inne pat is cmpletely eused by all base statins; the ute pat is divided amng the base statins with a euse fact geate than ne. With sft euse [], the veall bandwidth is shaed by all base statins (i. e., a euse fact f ne is applied), but f the tansmissin n each sub-ie the base statins ae esticted t a cetain pwe bund. All these appaches t mitigating CCI n be descibed in tems f cell-specific pwe masks ve the system bandwidth. A pwe mask pescibes the factin f the maximum tansmit pwe that the base statin may use depending at the pat f the spectum. In Fig. 1 we assume a scenai f thee neighbing cells. In the se f had euse with a euse fact f thee (Fig. 1b), the pwe masks blck all but ne thid f the spectum. In u example f factinal euse (Fig. 1c), the pwe mask f the fist half f the spectum is unifm, and the secnd half cespnds t a cndensed vesin f the had euse se. Fig. 1d illustates the sft euse se. The pwe masks have a signifint impact n the system's pefmance. Pevius wk [] shws that sft euse has a pacity advantage ve the plain had euse. Futheme, adapting the pwe mask used f sft euse t the cuent taffic situatin has ecently been shwn t be a geat leve n pacity []. Hweve, the questin emains, hw clse this and simila adaptive schemes get t ptimality. In this pape, we pesent a means t evaluate pwe mask pefmance in cellula OFDMA systems. Ou cental cntibutin is the fmulatin f a glbal knwledge expliting 16

2 pwe pwe pwe pwe U ul N N (a) Unifm (b) Had euse (c) Factinal euse (d) Sft euse Fig. 1. Diffeent pwe masks nn-linea ptimizatin pblem and the slving f seveal accding pblem instances in a basic efeence scenai. We cmpae sme existing pwe mask cnfiguatins t the ideal esults, in de t shw hw u mdel seves as a basis f the pefmance evaluatin f me sphistited static as well as adaptive pwe mask cnfiguatins. T the best f u knwledge, this is the fist wk t pesent esults f the nn-linea esuce scheduling ptimizatin pblem elated t cellula OFDMA netwks. The emainde f this pape is ganized as fllws. In the fllwing sectin, we descibe esuce scheduling in LTE systems and intduce u scheduling gal. Then, in Sec. III we intduce u scheduling ptimizatin mdel f the ll (Sec. III-A), as well as the glbal ptimizatin se (Sec. III-B). In Sec. IV, we pesent u efeence scenai and the accding efeence esults. We cnclude u wk and identify tpics f futhe study in Sec. V. II. SYSTEM MODEL In LTE, time is sltted int tansmissin time intevals (TTl) [1] f duatin TTTI (in the de f millisecnds). Duing a single TTl, dwn-link use data multiplexing is dne in divisin multiplexing (FDM) fashin, whee the smallest addessable bandwidth-unit is a esuce blck. Fllwing the llized mapping scheme, a esuce blck cnsists f adjacent sub-ies in the dmain. In the time dmain, a esuce blck spans all OFDM symbls available f use data tansmissin f the espective TTL Each esuce blck is expected t expeience mstly flat fading thughut a single TTL A. The schedule F each TTl and in each cell, a base statin schedule assigns the esuce blcks t the seved teminals. LTE uses adaptive cding and mdulatin (ACM) pe esuce blck, s the schedule detemines als the mdulatin type and cding, based n available channel state infmatin (CSI). In this wk, hweve, we use the theetil Shannn pacity f a channel [6] instead f efeing t the cding and mdulatin type cmbinatins actually cnsideed f LTE. Ou methd is nnetheless als applible t ealistic cding schemes. Shannn's theem states that f a channel with bandwidth B and a given signal-t-intefeence-and-nise ati (SINR), thee exists a cde that achieves a thughput f THR == B. lg(1 + SINR). (1) We assume that the tansmit pwe is pescibed f each esuce blck by a pwe mask. F cell i, we will dente the pwe mask by pi: sk E [0,1]. This value dentes the factin f the ttal available utput pwe p(max). On esuce blck, cell i thus tansmits with a pwe f p(max). pi: sk. We dente the channel gain in TTl t by I'Jt f use m, base statin i, and esuce blck, and lclte the cuent SINR as p(max). p:na sk. lt) SINRt) ==,,m, (),m, '" (MAX). mask. (t) + L...Jj -i P Pj, lj,m, Tl The abve denminat sums up the c-channel intefeence fm cncuently tansmitting base statins j -I- i and the nise pwe Tl. Nte that in a eal system, the TTls wuld nt be synchnized amng cells, and the teminal wuld simply measue the cuent SINR. III. SCHEDULER OPTIMIZATION Scheduling cmmnly aims at maximizing system thughput, but fainess has t be taken int accunt, t. Slely maximizing the aw system thughput n lead t stavatin f uses at the cell edge and vesupply f bandwidth t uses that ae easy t seve. Diffeent kinds f fainess cnstaints cicumvent this: guaanteeing each use a cetain minimum ate [7], multiplying each use's thughput by an individual pptinal fai fact [8], utility-based pe-use thughput ptimizatin [9]. Utility-based ptimizatin is faiest, but it is highly cmplex. A. Ll ptimizatin mdel Discussing diffeent scheduling and fainess plicies is beynd the scpe f this pape. We use a simple fainess mdel t cmpae diffeent sft euse scenais, but u appach easily adapts t the fainess ntins. We assume that f each use thee is a maximum thughput 16

3 THR(MAX), and that any thughput beynd THR(MAX) is useless. In the wds, we ty t maximize the system thughput while assuing that nne f the uses gets me than a cetain maximum ate. This appach cespnds t a simple piecewise linea utility functin. Fmally, we n wite u scheduling gal as an ptimizatin mdel by intducing the binay use/esuce blck assignment vaiable X, which is 1 if use m btains esuce blck, and 0 thewise [10]. The sets f all active uses and f all esuce blcks ae dented by M and R, espectively. The task f the schedule then is descibed by the fllwing intege linea pgam: max L L x(t) m,. THR(t) m, (a) m s. t. LX ::; 1 V E R (b) m L x. THR ::; THR(MAX) VmEM (c) The scheduling bjective (a) is t maximize the ttal expected thughput THR f all uses m n all esuce blcks. Cnstaint (b) ensues that each esuce blck is assigned t at mst ne use at a time (i. e., m exclusively uses at TTl time t). Cnstaint (c) is the utility cnstaint that guaantees that use m des nt get me than the maximum equied thughput THR(MAX). The expected thughput THR f use m n blck depends n the expected SINR, which we will dente by SINR, t, : The expected SINR is deived fm the latest SINR measuement. Accding t Eq. (1), the expected thughput is A (t) (A (t) ) THRm == f lg 1 + SINR i m,,,, () if m is lted in cell j ut f the set f active cells :J. Nte that f is the esuce blck bandwidth. B. Glbal ptimizatin mdel The glbal ptimizatin mdel aims at scheduling the available system esuces such that the system wide thughput is maximized. We assume that thee is a cental entity equipped with all infmatin necessay t achieve this gal. The cental entity nt nly is in chage f detemining the ptimal use/esuce blck assignments, but als f finding the ptimal pwe levels f each esuce blck in each cell j ut f the set f active cells :J. Nte that in this se, the ptimizatin pblem is nt subject t pwe masking, but t an veall maximum pwe p(max) cnstaint that is allwed t be adiated by each base statin (expessed in Cnstaint c). As a cnsequence, we mdify u ll esuce scheduling ptimizatin pblem t include the thee-dimensinal cell/use/esuce blck ptimizatin vaiable x;,: max LLL (t) A Xj,m,. THRj,m, (t) (a) j m s.t. Lx\t) J,m, < - 1 Vj, (b) m LY; < p(max) Vj (c) L x\t). THR\t) < THR(MAX) J,m, J,m, - m Vj,m (d) Optimizatin vaiable Y; is the pwe assignment vaiable. The expected use thughput is cmputed fllwing Eq.. The expected SINR, hweve, is nw cmputed as fllws: (t) (t) SINR" (t ) == y.,. '",m,,m, (t) (t) Lj#i Yj,. 'j,m, + fi Cmbining Eqs. and 6, it n easily be seen, that the glbal esuce scheduling pblem is nn-linea in natue. Meve, since the use/esuce blck assignment vaiable is binay, wheeas the esuce blck/pwe assignment is a eal vaiable, the glbal esuce ptimizatin pblem is a nnlinea mixed intege pblem, which is knwn t be extemely had t slve. Still, in the next sectin we pesent a basic efeence scenai f which slving the glbal esuce alltin pblem is pssible unde cmmn ptimizatin pblem slving sftand hadwae cnditins. A. Setup IV. REFERENCE SCENARIO Ou efeence system and channel mdel paameteizatin (path lss, shadwing, and fading mdel) lagely fllws the paametes f the UTRA and EUTRA simulatin se 1 as pesented in Tables A and A f [] with an inte-site distance f 00 m and uses dpped unifmly in each hexagnal cell. Diffeing paametes ae shwn in Table I. We have cnsideed tw hexagnal cells. The easn f this is that f signifintly lage scenais, the glbal ptimizatin pblem is nt slvable within easnable time cnstaints using egula had- and sftwae equipment. Ou LTE system-level simulat is based n the fee timed discete Paamete Symbl Value Cells in the system I:JI Uses IMI 8 Resuce blcks IRI Resuce blck spacing f 180 khz Maximal tansmissin pwe pe cell pmax dbm TABLE I SIMULATION PARAMETERS. (6) 16

4 X , - - -, , --HFR --FR1..c E. SFR b - [1 ; 0.1] OPTIMUM (J) c...c :s - - SFR a - [1 ; 0.]. c....c C) ;:] e ; " C) t:: ;:] E (j) --HFR --FR1 - - SFR a - [1 ; 0. ] SFR b - [1 ; 0.1 ] OPTIMUM (j) Ol.---..J-----'----'-----L-----'----'-----'----'.. maximum ate pe use theshld / bps.. maximum ate pe use theshld / bps (a) Mean system thughput (bps), 99 % cnfidence intevals (b) Mean thughput (bps) f weakest use, 99 % cnfidence intevals Fig.. All uses achieve thei maximum equied ate if the esuces ae distibuted ptimally. In the se f ll knwledge based esuce assignement ptimizatin, se SFR0b, which featues a pwe level diffeence between the ne f SFR0a and the had euse HFR scheme achieves the best pefmance, especially in the weakest use pefmance (b). event simulatin libay OMNeT++ [11]. The instances f nnlinea glbal ptimizatin pblem () have been slved using LINDO's LINGO nn-linea ptimizatin pblem slve [1]. The ll scheduling ptimizatin pblem instances () have been slved using ILOG's CPLEX linea pblem slve [1]. We have simulated 100 diffeent use distibutins. In each un, the instances f the espective glbal and ll scheduling ptimizatin pblem wee slved f each cell. Theeby, we have cmpaed the pefmance f the glbal ptimal scheduling decisin t the pefmance f the llly ptimal schedule featuing the fllwing pwe masks: Had Fequency Reuse (HFR). Each cell may nly use half f the ttal system bandwidth. An accding pwe mask f had euse is depicted in Fig. 1b. Unifm/Fequency Reuse 1 (FRl). The available pwe is distibuted evenly acss all esuce blcks in all cells, cf Fig. 1a. Sft Reuse (SFR). Thee ae tw diffeent pwe levels: high, and lw. Each cell uses half f the spectum with each pwe level, see Fig. 1d f the accding sft euse mask. Thee ae tw vesins f sft masks: (a) SFRa [1;0.] and (b) SFRb [1;0.1], which means that in the fist se the the lw pwe level equals half, and in the latte ne tenth f the high pwe level. B. Results The esults f u simulatins ae shwn in Fig.. Meve, in de t incease the eadability f the esults, the mean values ae als pesented in Tables II and III. Fig. a shws the mean system thughput aveaged ve all simulatin uns f the diffeent pwe mask schemes detailed abve. Fig. b shws the aveage thughput btained by the weakest use, i. e., the individual use that eceived the least thughput in each f the 100 diffeent use distibutin scenais. In mst ses, this use is clsest t the cell-edge (des csinally Appach HFR FRI SFRa SFRb Optimum Maximum Rate pe Use Theshld / Mbps TABLE II SIMULATION RESULTS: MEAN SYSTEMTHROUGHPUT/MBPS. Appach HFR FRI SFRa SFRb Optimum Maximum Rate pe Use Theshld / Mbps TABLE III SIMULATION RESULTS: WEAKEST USER THROUGHPUT/MBPS. nt apply unde cetain fading cnditins). In additin, the e bas display cnfidence intevals with a cnfidence level f 99%. Addessing fist the mean system thughput in Fig. a, the esults shw that all pwe masks shw clse t ptimal pefmance, if the maximum ate pe use is aund Mbps. With inceasing maximum ate theshld, hweve, the diffeences in pefmance becme clea. As expected, the had euse scheme achieves the wst pefmance values. This is mainly due t the fact that its bandwidth is limited and its intefeence advantage des nt payff in the highe max equied ate ses, which fav the uses clse t the base statin (that ae less susceptible against intefeence). Inteestingly, nne f the sft euse ses SFRa 16

5 and SFRb pefm signifintly bette than the equal pwe level euse 1 (FR1) se, when it cmes t the mean system thughput. In tems f weakest use pefmance, hweve, signifint diffeences becme visible when lking at Fig. b. Hee, SFRa achieves an incease f app. % cmpaed t FR1, wheeas SFRb even has an app. -0 % gain ve FR 1. This is an immense gain, cnsideing the fact that the gain slely stems fm masking the esuce blck pwe levels. Anthe inteesting effect shws the weakest use thughput cuve f the had euse se HFR. Up t a maximum equied ate f Mbps, the pefmance f the cell edge uses is bette than in the euse 1 and sft euse SFRa se. This advantage taces back t the fact that thee is ze intefeence fm the neighb cell, and, thus, the channel states f the cell edge uses ae geneally bette than in the euse 1 the sft euse se. Due t the limited esuces in the had euse HFR se, hweve, the weakest uses nnt take advantage f the inceasing maximum ate abve that Mbps theshld. This is mainly beuse the cell edge uses hadly get any esuces at all, if the stnge uses ae allwed t cnsume esuces f such high ates. Accdingly, thei mean thughput deceases with the inceasing max ate afte that tuning pint. This is vey likely t happen t the the schemes as well at diffeent pints n the max ate theshld axis. In geneal, nne f the llly ptimized schemes gets clse t the ptimum in the highe maximum equied ate ange. Nte that all uses achieve the maximum equied ate in the cnsideed ange, if the esuces ae distibuted ptimally. Even thugh ll ptimizatin stategies ae vey unlikely t get t a pefmance simila t the glbal ptimum, thee is much space f impvements. Using u efeence mdel, pmising ndidates n be judged with espect t glbal ptimality. REFERENCES [1] GPP; Technil Specifitin Gup Radi Access Netwk, "Physil channels and mdulatin (elease 8)," TS-6.11, Jun. 007, vesin [] M. Stenad, T. Ottsn, A. Ahlen, and A. Svenssn, "Attaining bth cveage and high spectal efficiency with adaptive OFDM dwnlinks," in Pc. fthe 8th IEEE Vehicula Technlgy Cnfeence (VTC-Fall 0), vl., Oct. 00, pp [] GPP; Huawei, "Sft euse scheme f UTRAN LTE," Rl 0007, May 00. [] --, "Futhe analysis f sft euse scheme," RI-0081, Sep. 00. [] K. Dpple, X. He, C. Witjing, and A. Si, "Adaptive sft euse f elay enhanced cells," in Pc. fthe 6th IEEE Vehicula Technlgy Cnfeence (VTC-Sping 07), Ap. 007, pp [6] C. E. Shannn, "A mathematil they f cmmunitin," Bell System Tech. I., vl. 7, 198. [7] I. Kim, H. Lee, B. Kim, and Y. Lee, "On the use f linea pgamming f dynamic subchannel and bit alltin in multiuse OFDM," in Pc. f the IEEE Glbal Telecmmunitins Cnfeence (Glbecm '01), Nvembe 001, pp [8] Q. Wang, J. Xu, and Z. Bu, "Pptinal-fai bit and pwe adaptatin in multi-use OFDM systems," in Pc. f the IEEE Intenatinal Sympsium n Pesnal, Ind and Mbile Radi Cmmunitins (PIMRC), Helsinki, Finland, Sep. 006, pp. 1-. [9] G. Sng and Y. Li, "Utility-based jint physil-mac laye ptimizatin in OFDM," in Pc. fthe IEEE Glbal Telecmmunitins Cnfeence (Glbecm '0), vl. 1, Nvembe 00, pp [10] M. Bhge, 1. Gss, M. Meye, and A. Wlisz, "Dynamic esuce alltin in fdm systems: An veview f css-laye ptimizatin pinciples and techniques," IEEE Netwk Magazine, Special Issue: "Evlutin twad G wieless netwking", vl. 1, n. 1, pp. -9, JanuaylFebuay 007. [11] A. Vaga, OMNeT++ Use Manual.. [1] LINDO Systems Inc., LINGO 10 - Use's Guide, Chig, Illinis, 008. [1] S. ILOG, ILOG CPLEX Use's Manual, Pais, Fance, 00. v. CONCLUSION We have pesented a means t evaluate pwe mask pefmance in cellula OFDMA systems. It is based n slving a glbal knwledge expliting nn-linea ptimizatin pblem. We have slved seveal accding pblem instances in a basic OFDMAILTE efeence scenai. Theeby, we have shwn that thee is a signifint gap in pefmance between the applitin f simple static pwe masks in cmbinatin with a llly ptimal schedule and the glbal ptimum. This is tue especially f the cell-edge use pefmance. Accding futue wks include, thus, the develpment f me sphistited static pwe masks, as well as schemes f pwe mask adaptatin, and a cmpaisn f thei pefmance t the ptimal se indited in this pape. Cnsequently, slving the pesented nn-linea ptimizatin mdel in a lage efeence mdel is als f maj inteest. 166