ABSTRACT. 1. Introduction

Transcription

1 Communatons an Network, 3, 5, Publshe Onlne September 3 ( Resoure Alloaton an Beamformng Algorthm Base on Interferene Avoane Approah for Deve-to-Deve Communaton Unerlayng LTE Cellular Network Huy-Dung Han, Chenx Zhu, Dorn Vorel, Akra Ito Fujtsu Laboratores of Amera In. Sunnyvale, CA, USA Emal: hhan@us.fujtsu.om, zhu@us.fujtsu.om,vorel@us.fujtsu.om, ato@us.fujtsu.om Reeve July, 3 ABSTRACT In ths work, we onser eve-to-eve (DD) ret ommunaton unerlayng a 3GPP LTE-A network. DD ommunaton enables new serve opportuntes, proves hgh throughput an relable ommunaton whle reung the base staton loa. For better total performane, DD lnks an ellular lnks share the same rao resoure an the management of nterferene beomes a rual task. We propose a rao resoure alloaton for DD lnks base on nterferene avoane approah. For system wth multple transmt antennas, we apply beamformng tehnque base on sgnal to leakage rteron to reue the o-hannel nterferene. The results show that, DD transmsson wth the resoure alloaton an beamformng tehnque proves sgnfant gan ompare to that of the regular ellular network. Keywors: Deve-to-Deve; LTE; Resoure Alloaton; Beamformng. Introuton In reent years, 3GPP Long Term Evoluton (LTE) tehnology has beome the key stanar for IMT-Avane. LTE an ts extene verson LTE-Avane (LTE-A) have alreay aopte many avane tehnologes suh as Multple-Input Multple-Output (MIMO) antennas, Coornate Multple Ponts (CoMP), enhane Inter-ell Interferene Canelaton (eicic) to meet the nreasng eman for hgh throughput an Qualty of Serve (QoS) []. The next generaton LTE network s expete to aopt tehnologes that not only nrease system performane but also bul up founaton for new type of serves. Short range ret ommunaton or eve-to-eve (DD) ommunaton s a anate to satsfy ths requrement an has beome one of the stuy tems uner nvestgaton n release of 3GPP LTE. Unlke the exstng DD ommunaton solutons, suh as Bluetooth [] or ret WIFI, DD ommunaton n LTE shares the lense rao resoure wth regular LTE lnks. It has sgnfant avantages on power onsumpton an spetral effeny. It allows eves to transmt ata wthout passng t through base statons (BS or enb) an the bakbone. Beause of the proxmty, the ret lnks often have goo hannel qualty,.e., low pathloss. Thus, even wth low transmt power, hgh throughput transmsson an be mae. Furthermore, t s possble to gan hgher spetrum usage by lettng the DD lnks utlze the spetrum resoure of other lnks wthout ausng substantal performane egraaton to eah other. Suh way of ommunaton requres a areful esgn on rao traff management. In DD ommunaton unerlayng ellular network, nterferene management s one of rual task. As DD ommunaton s an atonal feature bese the regular ellular ommunaton, DD moe UEs (DUE) shoul not ause sgnfant nterferene to ellular UEs (CUE) reevers when they are sharng the same rao resoure. Smlar to the ogntve rao ontext, the DUEs an be onsere as seonary users an aess the lense hannel n an opportunst way [3]. However, there woul be no QoS guarantee for them. Therefore, t s expete that the enb ontrols an performs nterferene management to mprove the QoS of DD lnk as well as that of the other legay eves sharng the same banwth. Gven a group of eves sharng the same rao resoure, nterferene management usually beome the problem of resoure alloaton n whh eah eve transmts usng a tme an frequeny slot, aka, rao resoure blok (RB) wth ertan power. In [4], the maxmum transmt power DD lnks s set by enb suh that the nterferene to the o-hannel CUE lnk s lmte. Power ajustment for DD lnks s also propose n [5] Copyrght 3 SRes.

2 368 H.-D. HAN ET AL. to ynamally reue the nterferene. In [6], the power level s set to optmze the transmsson sum rate of both CUEs an DUEs. Nevertheless, the power ontrol shoul be jontly onsere wth resoure alloaton/sheulng to mnmze the potental rtal nterferene [7]. Beses, uner LTE senaro, the spetrum resoure alloaton/sheulng rate s faster than that of the power ajustment; thereby, the nstantaneous sgnal to nterferene an nose (SINR) for eah UE epens more on sheulng. Therefore, hannel/nterferene aware sheulng algorthms an be more benefal for the system performane. In [8], the area of possble DD transmsson s alulate by pathloss threshol. However, the area moelng s over smplfe leang to the sub-optmal performane. In ths work, we nvestgate an nterferene avoane sheulng algorthm for DD lnks. Gven a sheule for CUEs, eah DD lnk s pare up wth a CUE lnk an sharng the sheule RBG by a parng algorthm. The DD pars are alloate suh that the nterferene at eah nvolve eve s below a threshol. For the DD pars volatng the requrement, a efault spetrum resoure s prove so that they stll an transmt wth best effort bass. For multple transmt antennas system, we apply a beamformng base on sgnal to nterferene an nose (SLNR) metr [9,] to further reue the nterferene. The results show that, wth the parng algorthm, the o-hannel nterferene between DD pars an the CUE uplnk s mtgate an the system throughput s optmze.. System Moel We onser DD lnks sharng uplnk banwth wth the regular CUEs. The total uplnk frequeny resoure for ata s assume to be ve nto N resoure blok groups (RBG). At a gven tme slot t, the RBG an be nexe as (, t f ) where f {,.., N}. There are L uplnk CUEs an L DD pars sharng ths rao resoure. If DD lnks an a CUE uplnk sharng RBG (, t f), o-hannel nterferene egraes the performane of both lnks. Fgure represents hannel moel where DD lnks j, k an an uplnk CUE transmt n the same RBG. Here, DD par k onssts of a DUE transmtter (DUE-T) k an a DUE reever (DUE-R) k. For the sake of smplty, the RBG nex nator s omtte. The hannel an nterferene nlue the followng omponents: H () : the hannel matrx from CUE to the enb. H ( k ): the hannel matrx from DUE-T k to the enb. H ( k, j ): the hannel matrx from DUE-T j to The term resoure blok group s taken from LTE spefaton [] sne we are usng LTE senaros for smulaton. DUE-R k. H ( k ) : the hannel matrx from DUE-T k to DUE-R k. H ( k, ): the hannel matrx from the CUE to DUE-R k. If multple antennas are equppe at the transmtter, beamformng tehnque an be use to mprove the SINR. We efne the preong matres for CUE an for DUE-T k as W () an W ( k ), respetvely. In prate, the preong matres belong to a oebook. The performane of the system epens on the ohannel nterferene ausng by the eves. The ntraell o-hannel nterferene measure at enb s I () H W () kd where s the nex set of the DD lnks sharng the banwth wth CUE. The SINR for uplnk CUE an be alulate as H() W() () () N () I (), where N, () are the nose power plus the nterferene from other ells for the lnk. Smlarly, the nterferene measure at DUE-R k an ts SINR are alulate as, respetvely: jd, jk I H W ( ) H ( k, j) W ( j), (3) H ( kw ) N I, where N, are the nose power plus the nterferene from other ells for the DD lnk k. The problem of rao resoure management at enb s to ee the sets for all together wth preong matres suh that the optmum sum throughput s aheve over a long tme pero as: Fgure. Channel moel for DD pars sharng banwth wth an uplnk CUE. The sol lnes represent the ret lnks. The ot lnes represent o-hannel nterferene. (4) Copyrght 3 SRes.

3 H.-D. HAN ET AL. 369 L L () t () t Maxmzng T ( ) T () t () t () t, W ( ), W ( ) t k k where () t t T (), T( ) () are the throughput of CUE an DUE k at tme t. The throughputs an be efne usng Shannon apaty formula [] as Blog wth B s the banwth of a RBG an s the SINRs efne n (), (4). The tme nex t s ae to, W( ), W to represent ther hanges over tme. The problem n (5) an be lassfe as a NP-har as t an be reue to a smpler, but stll NP-har, sheulng algorthm []. In [], a proportonal far algorthm an be performe to optmze the total throughput over tme whle allowng all the UEs reeve at least a mnmal QoS. Suh algorthms usually epens on the assumpton that the nterferene s small an relatve unhange,.e., I () an I () are neglgble. When there are multple eves sheule on one RBG, ths assumpton no longer hols. Therefore, new esgns onserng nterferene after resoure alloaton are requre. The nstantaneous rate of a UE epens on ts relatve loaton wth enb, the varyng hannel an the ohannel nterferene. Compare to the transmsson rate of CUEs n tratonal ellular system, the nstantaneous rate of CUEs n the presene of DD ommunaton epens more on the ntra-ell o-hannel nterferene. Hene, for a group of a CUE an DUEs sharng the same RBG, the nstantaneous nterferene epens on the resoure assgnment. Therefore, the task of resoure alloaton for UEs sharng resoure s very fferent from the one for uplnk only. In seton 3, we propose a resoure alloaton mehansm for DD par sharng banwth wth an uplnk CUE, alle parng algorthm. Base on nterferene avoane approah, the parng algorthm allows many DD lnks reuse the uplnk rao resoure whle mantanng the uplnk performane. The problem n (5) onsers resoure alloaton an preong matres to optmze the performane an, thereby, har to solve. Therefore, we separate the two proesses by applyng the parng algorthm frst; then, gven the resoure assgnment, we etermne the optmum beamformng. In seton 5, we ntroue a beamformng tehnque to reue the n-ell o-hannel nterferene an mprove throughput performane. 3. Parng Algorthm Base on Interferene Threshol an As the general problem s NP-har, we propose a heurst algorthm whose assumpton s that the CUE has been sheule wthout knowlege of DUE hannels. The assumpton s base on two fols: CUEs requres better QoS proteton as the DUEs always have better hannel ue to ther proxmty; CUEs shoul have hgher prortyas DD ommunaton s an atonal feature n (5) ellular network. Wth ths assumpton, we an reue the general problem to the parng algorthm whose task s to assgn a group of DUEs to par wth a CUE an share ts sheule resoure. The parng algorthm also assumes that the sgnal strength of DD lnks s often very hgh an elmnatng the o-hannel nterferene s essental. Therefore, the parng proess s base on the threshols of nterferene. A CUE lnk an a DD lnk are pare to transmt on the same RBG only when the nterferene satsfes threshols ontons for both reevers. There are several exstng nterferene avoane algorthms [3] that work well when the number of DD pars s not too large. However, f the number of DD pars s large, only few DD pars an share resoure ths way. In our parng algorthm, we propose that f the DD par annot share the resoure wth the CUE, they stll an transmt on the efault resoure wth lmte rate. Wth the efault resoure, all of DD pars an transmt at the same tme an the total throughput s nrease. In the rest of the paper, wthout loss of generalty, we assume that the RBG s use as the efault RBG. The anates, nlung CUE an DUE k are pare up only the nterferene at all nvolve reevers are below the threshols: () for CUEs an ( k ) for DUEs. Otherwse, the DUE k wll be assgne to transmt on the efault RBG. The omplete parng algorthm nlung the nterferene alulaton s esrbe n the followng pseuo oe: Inputs: : set of N CUEs nex that alreay sheule on N RBG. : set of DUEs nee to be sheule. RBG : efault RBG. CUE : CUE that sheule on RBG. Outputs: : set of DUE nes pare wth CUE. Begn: for all. For eah k For eah Calulate nterferene levels I ( ) an I usng equatons (),(3). If I ( ) ( an I ) ( then ) ( ) Par up DUE lnk an CUE lnk : {} k En f En For If DD lnk k s not pare wth any CUE n Par up DD lnk k an CUE : {} k En If En For En. The parng algorthm alulates the nterferene by measurng the sgnals power from all other nterferer. Copyrght 3 SRes.

4 37 H.-D. HAN ET AL. Therefore, only the ampltue of sgnal s neee. 4. The Choe of Interferene Threshols The nterferene threshols an be alulate from the target SINRs. As the expresson ervatons for CUEs an DUEs are smlar, we only esrbe the CUE ase. The parng algorthm shoul pk the hannel suh that the average SINR s larger than the target SINR, () as H() W() E N, () I(), (). (6) Conserng the sgnal, nterferene an nose are nepenent, we arrve at. E H() W() I() N, (). () We an set () to be the rght han se of (7). If the eves s equppe wth sngle transmt antenna W () an be omtte. Applyng the parng algorthm wth ths threshol guarantees that the target SINR s met. For multple transmt antennas ase, the threshol ontons may not be guarantee as the esre sgnal an the nterferene epen on beamformng. One an set the nterferene threshol orresponng to the mnmum sgnal strength. However, ths value may be negatve. Therefore, we an set the threshol as the average value over all possbltes of the preong matrx as: ( n) E H () W n () N, () (8). () ( ) where W n s the value of n -th entry of the oebook an s the arnalty of the oebook. Although, there s no guarantee that the nterferene s always below the threshol, we stll an hoose the preong matrx of the nterferer so that the nterferene s below the threshol most of the tme. In the next seton, we suss several preong matrx seleton algorthms. 5. Beamformng Strategy Base on SLNR Crteron (7) Gven the spetrum alloaton, the optmum beamformng shoul onser all beamformers at the same tme to maxmze the sum throughput of the CUE an DUEs as n (5). However, the omplexty of ths approah s hgh as t s proportonal to the exponental of number of eves. Furthermore, beause the SINRs of CUE an DUEs are ouple, ths optmzaton problem has no lose form soluton. Therefore, we onser an alternatve beamformng tehnque base on SLNR rteron. Usng ths rteron, the preong matrx an be etermne loally an nepenent from other lnks. The prnple s to fn the preong matrx that ahevng the balane of maxmzng the reeve sgnal strength at the ntene reever an mnmzng the nterferene at the others. As the result, the throughput an be optmze. The SLNR for CUE s efne as: an the SLNR for DUE H() W() () N H () W (), k k s H W, ( ) ( ) (, ) ( ) l, lk (9). N H k W k H l k W k () The optmum soluton an be alulate usng Raylegh-Rtz quotent[4]. Uner preefne oebook, the optmum preong matrx an be foun by tryng all possbltes as W() argmax (), W argmax () For eah transmtter, the seleton of preong matrx base on SLNR requres hannel nformaton between the transmtter an the nvolve reevers. If the preong matrx s alulate at enb, all the hannel nformaton s transmtte to the enb. On the other han, f the preong matrx s alulate at the transmtter, for example, DUE-T k, the orresponng hannels, H (, k ), H ( jk, ), are reporte to the DUE-T. Nevertheless, uner Tme-Dvson Duplex (TDD) system, the reproty of hannels an be explote suh that the hannel from a DUE-T to the orresponng DUE-R an enb s alreay avalable at the DUE-T. Therefore, n TDD system, the SLNR approah requres less hannel nformaton exhange an further reues the ontrol overhea. 6. Numeral Results In ths seton, we evaluate the performane of CUEs an DUEs usng aforementone algorthms by a system level smulator (SLS). We onser uplnk synhronze LTE ellular network wth Urban Mroell (UM) senaro wth UE movng spee of 3km/h as efne n moel of [6]. There are 7 enbs equal spae loate wth stane of 5 m an eah ves nto 3 ells usng egree vertal antennas. The network s assume to operate on a MHz ban wth GHz arrer. The ban s ve nto 9 RBGs. Eah ell s assgne CUEs an DUEs. The loaton of CUEs an DUE-T are unformly ranom generate over the area of network. The DUE-Rs Copyrght 3 SRes.

5 H.-D. HAN ET AL. 37 are unformly ranom loate aroun the orresponent DUE-Ts wth maxmum stane max 3m. The UEeNB pathloss moel n B s efne n [6] as L log ( [ km]) () UE enb an UE-UE pathloss moel s efne n [7] as L 48 4log ( [ km]). (3) UE UE We assume open loop power ontrol wth omplete pathloss ompensaton. The average reeve sgnal power s Bm for CUEs an 8 Bm for DUEs. Assumptons on the network nlues full buffer traff moel, proportonal far (PF) algorthm for sheulng the CUEs an eal hannel estmaton wth MMSE reever. The throughput s measure by the number of bts suessfully transmtte usng ertan Moulaton an Cong Sheme (MCS) efne n [5] over the smulate tme. The suess of transmsson s sgnale by ACK/NAK assumng a Blok Error Rate of %, the HARQ s nremental reunany (IR) wth maxmum 3 retransmssons. We stuy the propose algorthms uner SIMO ( ) MIMO ( ) hannel moel. To see the benefts of the algorthms, we ompare the followng setups for SLS: Only uplnk CUEs transmt (marke as CUE only ). Eah DD par s assgne wth a unformly ranom RBG (marke as Ranom RBG ). The DD pars use the efault RBG only (marke as ). The DUE-Ts s assgne usng the aforementone parng algorthm. There oul be more than one DD pars transmttng on the same RBG. The nterferene threshol for CUEs s Bm an for DUEs s 9 Bm (marke as Parng ). We frst examne the parng algorthm for system wth one transmt antenna. Fgure an Fgure 3 show the CUE throughput an DUE throughput umulatve strbuton funton (), respetvely. Comparng Default RBG an CUE only setups, we see that the use of the efault RBG for DD transmsson oes not affet CUE throughput sgnfantly. Instea, enablng of DD transmsson has mprove the total ell throughput. The transmsson wth ranom RBG alloaton oes not take nto aount the CUEs' QoS proteton; thereby, the CUE throughput egraes ompare to the Default RBG ase. On the other han, the DD throughput has mprove as the DD lnks o not nee to share only one RBG. The parng algorthm takes nto aount the possble nterferene to the other eves. Therefore, we an see that, the CUE throughput oes not egrae ompare to the ase that DD pars transmttng on the efault RBG. The DUE-Ts, whenever possble, transmt on the RBG other than the efault one. Hene, the DUE throughput mproves an s omparable wth that of ranom RBG ase. Table shows the average ell throughput of eah test ase. The ell throughput s measure by summng all the CUEs an DUEs throughput of the ell an average over ells. We an see that the DD transmsson mprove the ell throughput by 5% (Ranom RBG) an up to 9% (usng parng algorthm) Ranom RBG. Parng CUE only Fgure. CUE throughput (SIMO ase)... Ranom RBG Parng Fgure 3. DUE throughput (SIMO ase). Table. The average throughput of ells an UEs. Average ell throughput (Mbps) SIMO MIMO CUE only Ranom RBG Parng 35.9 Copyrght 3 SRes.

6 37 H.-D. HAN ET AL. In Fgure 4 an Fgure 5, we show the CUE throughput an DUE throughput, respetvely, for multple antennas system. The parng algorthm s use frst to alloate the resoure; then the preong matrx seleton soluton. We an see that the parng algorthm stll effetve n ths test although the nterferene threshol may be volate. It s beause, the beamformng algorthm wth SLNR approah helps to reue the number of volaton. From Table, we an see that the parng algorthm wth SLNR beamformng metho mproves the ell throughput by 45%, 55%, 5% ompare to that of CUE only, efault RBG an ranom RBG ase, respetvely. 7. Conlusons an Future Works In ths work, we have shown the avantage of enablng DD ommunaton unerlayng LTE ellular network. We propose a parng algorthm for approprately assgn rao resoure for DD ommunaton an ausng only Ranom RBG. Parng+ SLNR CUE only Fgure 4. CUE throughput (MIMO ase)... Ranom RBG Parng+SLNR Fgure 5. DUE throughput (MIMO ase). mnmal nterferene to the uplnk. The results show that the DD lnks performane s mprove ompare to the ranom RBG assgnment an efault RBG assgnment whle mantanng the CUEs throughput. For multple transmt antennas senaro, the parng algorthm together wth the beamformng tehnque usng SLNR approah also nrease the ell throughput sgnfantly. The use of SLNR may reue the hannel nformaton exhange, espeally n TDD senaro. The future works nlue the examnaton of the parng algorthm n MIMO ase, where the volaton of nterferene threshol may happen. Aressng the outage problem an help to optmze the threshol parameters an further mprove the system performane. Dynam efault RBG an also be stue to optmze the performane. REFERENCES [] TS 36.-Evolve Unversal Terrestral Rao Aess (E-UTRA); Physal hannels an moulaton (Release ), 3GPP St. [] Teleommunatons an Informaton Exhange between Systems- Loal an Metropoltan Area Networks- Spef Requrements Part 5.: Wreless Meum Aess Control (MAC) an Physal Layer (PHY) Spefatons for Wreless Personal Area Networks (WPANs), IEEE St. [3] S. Huang, X. Lu an Z. Dng, Opportunst Spetrum Aess n Cogntve Rao Networks, n INFOCOM 8, The 7th Conferene on Computer Communatons, IEEE, 8, pp [4] P. Jäns, C.-H. Yu, K. Doppler, C. Rbero, C. Wjtng, K. Hugl, O. Trkkonen an V. Kovunen, Deve-to-eve Communaton Unerlayng Cellular Communatons Systems, Internatonal Journal of Communatons, Network an System Senes, Vol., No. 3, 9, pp o:.436/jns.9.39 [5] P. Lu, C. Hu, T. Peng, R. Qan an W. Wang, Amsson an Power Control for Deve-to-eve Lnks wth Qualty of Serve Proteton n Spetrum Sharng Hybr Network, n Pro. IEEE 3r Int Personal Inoor an Moble Rao Communatons (PIMRC) Symp.,, pp [6] C.-H. Yu, O. Trkkonen, K. Doppler an C. Rbero, Power Optmzaton of Deve-to-eve Communaton Unerlayng Cellular Communaton, n Pro. IEEE Int. Conf. Communatons, 9, pp.-5. [7] H. Xng an S. Hakola, The Investgaton of Power Control Shemes for a Deve-to-eve Communaton Integrate nto OFDMA Cellular System, n Pro. IEEE st Int Personal Inoor an Moble Rao Communatons (PIMRC) Symp.,, pp [8] H. Mn, J. Lee, S. Park an D. Hong, Capaty Enhanement Usng An Interferene Lmte Area for Deve-to-eve Uplnk Unerlayng Cellular Networks, IEEE Transatons on Wreless Communatons, Vol., Copyrght 3 SRes.

7 H.-D. HAN ET AL. 373 No.,, pp o:.9/twc [9] A. Targhat, M. Saek an A. H. Saye, A Mult User Beamformng Sheme for Downlnk MIMO Channels Base on Maxmzng Sgnal-to-leakage Ratos, n Pro. IEEE Int. Conf. Aousts, Speeh, an Sgnal Proessng (ICASSP 5), Vol. 3, 5. [] J.-H. Lee, S. Km, S.-R. Jn an D.-J. Park, A Mult-user Beamformng Sheme n Mmo Downlnk Channels for Mult-ell Networks, n Internatonal Conferene on Consumer Eletrons (ICCE), IEEE,, pp [] T. M. Cover an J. A. Thomas, Elements of nformaton theory. New York, Wley, 99. o:./476 [] S.-B. Lee, I. Pefkanaks, A. Meyerson, S. Xu an S. Lu, Proportonal Far Frequeny-oman Paket Sheulng for 3GPP LTE Uplnk, n INFOCOM, IEEE, 9, pp [3] M. Zulhasnne, C. Huang an A. Srnvasan, Effent Resoure Alloaton for Deve-to-eve Communaton Unerlayng LTE Network, n Pro. IEEE 6th Int Wreless an Moble Computng, Networkng an Communatons (WMob) Conf.,, pp [4] G. Golub an C. V. Loan, Matrx Computatons. The Johns Hopkns Unversty Press, 996. [5] TS 36.3-Evolve Unversal Terrestral Rao Aess (E-UTRA); Physal layer proeures (Release ), 3GPP St. [6] TS Evolve Unversal Terrestral Rao Aess (E-UTRA); further avanements for E-UTRA physal layer aspets, 3GPP St. [7] Propagaton Data an Preton Methos for the Plannng of Short-range Outoor Rao Communaton Systems an Rao Loal Area Networks n the Frequeny range 3 MHz to GHz, Reommenaton ITU-R P.4-4 St. Copyrght 3 SRes.