Relay Selection and Resource Allocation for D2D-Relaying under Uplink Cellular Power Control

Transcription

1 Relay Selectio ad Resource Allocatio for DD-Relayig uder Upli Cellular Power Cotrol Juqua Deg, Alexis A. Dowhuszo, Ragar Freij, Olav Tiroe Departmet of Commuicatios ad Networig, Aalto Uiversity, Espoo, Filad {juqua.deg, alexis.dowhuszo, ragar.freij, Abstract Device-to-Device (DD) commuicatio uderlyig 5G cellular etwors eable usig ect commuicatio betwee devices as a relay strategy for coverage extesio. We cosider such a DD-relayig approach i a sceario where there are multiple cadidate devices to be selected as relays. Both bachaul ad DD trasmissios are performed i upli cellular resources, ad are subject to cellular upli power cotrol. We ivestigate the relay selectio ad resource allocatio problem for DD-relayig i a multi-user, multi-carrier ad multi-cellular etwor. For this purpose, we formulate a joit optimizatio problem ad propose a simplified relay selectio ad resource allocatio scheme. Usig this scheme i a system simulatio, we demostrate that DD-relayig uder upli power cotrol icreases the throughput of cell-edge users sigificatly. I. INTRODUCTION Cosistet user experiece is oe of the most challegig objectives of 5G cellular etwors []. To achieve this goal, the user experiece should be idepedet from the locatio of the user i a cell. From this perspective, oe of the most importat improvemets of 5G as compared to 4G techologies should be i the throughput of cell-edge users. To achieve such improvemets, ovel commuicatio ad etworig techologies are eeded. For example, distributed atea systems ad Ultra Dese Networs (UDN) of small cells have bee cosidered. By shorteig the commuicatio distace betwee the User Equipmet (UE) ad the ifrastructure elemet, ad usig wired/wireless bachaul, these techiques have the potetial to boost cell-edge throughput [] [3]. However, as ifrastructure etwors become deser, the deploymet ad maiteace costs become larger as well. Accordigly, usig Device-to- Device (DD) relayig to improve system capacity ad coverage comes ito cosideratio for future 5G wireless etwors [4] [7]. The uderlyig idea is that there is a large umber of devices that may act as relays, either ifrastructure UDN access poits without wired bachaul, or other devices such as user-deployed devices, omadic odes, or mobile statios, which may act as relay statios to help to covey user traffic to or from the etwor. I two-hop upli commuicatio, the access hop is based o DD commuicatio, whereas the relay hop is a self-bachaul coectio based o traditioal cellular techologies. Cooperative commuicatios with relay selectio has bee widely studied i literature [8] [], where a fixed trasmissio power for both source ad relay ode is traditioally assumed. The DD-relayig problem i this paper differs from a covetioal relay selectio problem due to the Trasmissio Power Cotrol (TPC) applied by the cellular etwor [] [4]. I a multi-cellular etwor, upli TPC is used for two reasos. O oe had, upli TPC maages the iterferece betwee cells, by applyig a strict cotrol o the power radiated ito eighborig cells. O the other had, upli TPC is eeded to mitigate the ear-far effect. I 4G cellular etwors, orthogoality of frequecy domai Resource Blocs (RB) is oly approximate, ad the received powers at the Base Statio (BS) o eighborig RBs have to be ept similar to eep itra-cell i-bad emissio iterferece low. I [], [4], a model for upli DD relayig is discussed, where cellular power cotrol is used i the bachaul lis, whereas the DD lis apply a fixed trasmit power. I [], all DD lis reuse the same resources, whereas i [4], orthogoal resources are used for each DD trasmissio. However, the allocatio of resources to DD access lis ad the DD trasmissio power is fixed i the whole etwor, to eep multi-cell iterferece uder cotrol. I [3], all trasmissios are power cotrolled by the servig BS, ad multiple relays are selected to create a virtual MIMO system for the bachaul li. The total trasmit power radiated to a cell is ot uder cotrol, because each of the virtual-mimo trasmitters is separately power cotrolled, ad the iterferece caused to eighborig cells depeds o the selected umber of relayig devices. Furthermore, itercell iterferece is modeled by a aggressor etwor, which is idepedet from the system. I this paper, we cocetrate o DD relayig i a sceario where a large umber of cadidate relays is available. Strict cellular power cotrol is applied for all trasmissios, icludig bachaul ad DD. Upli TPC is based o full compesatio of the pathloss betwee the devices ad the servig BS, ad received power is restricted per used RB. There is a maximum trasmissio power, which prevets celledge users to achieve high data rates usig the ect li to the BS, because power budget limits the umber of RBs that ca be used. For each UE, either the best two-hop relay path or the ect trasmissio is selected. Moreover, relays operate i orthogoal resources. This maes both iter- ad itracell iterferece predictable, ad eables a flexible resource allocatio withi a cell. We schedule resources to each flow of the users, ad the resources scheduled to a relayig flow ca be flexibly allocated betwee the two hops, idepedetly i each flow. We first show that with a fixed resource allocatio to a flow, where the source has sufficiet trasmit power to reach the destiatio BS, relayig does ot provide ay gai uder strict cellular power cotrol. The gais i this sceario with strict power cotrol ad orthogoal resources comes from flexible resource schedulig. The bachaul hop i a twohop flow ca use a wider badwidth tha the correspodig ect li trasmissio ad, accordigly, a higher ed-to-ed /5/$3. 5 IEEE

2 Fig.. Illustratio of system model, where there are multiple relay cadidates, active UEs iside a cell. A UE ca choose to use ect li (sigle hop) or DD relayig (two hop) for upli commuicatio. Iter-cell iterferece is itroduced because frequecy reuse factor is. rate ca be reached. We show that by proper allocatio of badwidth resources amog DD lis ad bachaul lis for DD relayig, we ca capitalize o DD relayig to icrease fairess amog users ad provide more cosistet coverage of upli data rates to cell-edge users. II. SYSTEM MODEL AND ASSUMPTIONS We cosider the upli trasmissio i a multi-cell cellular etwor cosistig of N bs BSs, idexed by i =,,..., N bs. Upli trasmissios are based o SC-FDMA, ad the miimum uit of spectrum resources is oe RB with fixed badwidth B rb ad time duratio. The total umber of RBs i frequecy domai is M for each cell. O average there are K active UEs ad N Relay Node (RN) cadidates per cell, locatios of UEs ad RNs are modeled by a Poisso Poit Process. A RN may be either a idle UE or some other relay device. Each UE or RN is associated to the BS that has the smallest pathloss. We use U i ad R i to deote the set of UEs ad RNs associated to BS i. To simplify the aalysis, we oly cosider RN cadidates that are associated to the same BS as the cell-edge UE. The ect pathloss betwee UE ad its servig BS is L, the bachaul pathloss betwee RN ad the servig BS is L, ad the DD pathloss betwee UE ad RN is L,. Iside a cell, we assume that badwidth resources are orthogoally used by the trasmittig devices, so that there is o itra-cell iterferece. However, because SC- FDMA is sesitive to the i-bad emissio iterferece, the received powers of differet devices must be ept uder cotrol to eep the itra-cell i-bad emissio iterferece tolerable. Iter-cell iterferece is itroduced because of frequecy reuse factor is oe. The cells are assumed to be fully loaded such that all users eed a as high throughput as possible. All UEs ad RNs are uder trasmissio power cotrol by the servig BS. The system model is depicted i Fig.. There are three ids of lis. The ect li is used whe a active UE has a good ect chael to BS. The DD li is the first hop whe a active UE uses RN for relayig, ad the bachaul li is the secod hop betwee the RN ad BS. The relays are halfduplex, which is tae ito accout whe schedulig resources to the bachaul ad DD lis. Distace-depedet path loss ad shadow fadig are cosidered for all commuicatio lis. The pathloss is measured ad estimated by the UE or RN ad reported to the servig BS. A. Upli Trasmissio Power Cotrol The trasmissio power of a UE or RN is cotrolled by the servig BS to esure that the received power per RB at servig BS is o the right level. This trasmissio power is determied by several parameters give to the UE or RN via cotrol chaels [5]. The overall trasmissio power i the data chael is defied as P tx (L, N rb )=mi { P max,n rb P o L β}, () where P max is the maximum trasmit power, which is assumed to be the same for UE ad RN, N rb is the umber of RBs allocated for UE or RN, P o is the target received power per RB at servig BS, β is the cell-specific fractioal compesatio factor, ad L is the pathloss from device to BS. We assume L = L d α f s, where L is average pathloss o cell border icludig atea gais, feedlie ad miscellaeous losses, d is the ormalized distace with respect to cell radius R, α is the pathloss expoet, f s is the log-ormal distributed shadow fadig. I this paper we use a β =, which correspods to full pathloss compesatio. We assume that pathloss estimatio is accurate ad the dowli ad upli average pathloss are the same. The actual received power per RB at the BS is P rx,rb (L, N rb )= P { } tx (L, N rb ) Pmax =mi N rb L N rb L,P o, () ad the received power per RB at RN for DD li is ( ) Prx,rb ( L ) P tx L,N rb,n rb = N rb L. (3) The relayig problem uder cellular upli power cotrol Num of used RBs Received Power per RB i dbm ( P max = 3 dbm ) Distace from BS (m) Fig.. Upli power cotrol with differet values of P o whe oly distacedepedet pathloss is cosidered ad P max =3dBm. Each cotour curve shows how may RBs a ect or bachaul li ca use to guaratee a specified P o value at BS. differs from the traditioal relayig problems because the trasmissio power is ot fixed ad chages accordig to the differig pathloss to BS ad the umber of allocated RBs accordig to (). If perfect pathloss compesatio is performed (with P max = ad β =), relayig is couterproductive. A ect li has sufficiet trasmit power to meet the power cotrol target irrespectively of the resources allocated. I reality, a UE has a maximum trasmissio power P max. Whe

3 the pathloss betwee UE ad BS is too large, ad UE has bee grated a umber of RBs, it is possible that eve usig the maximum trasmissio power o these resources caot fulfill the target received power at BS. Istead of usig all the allocated resources for a ect li trasmissio, the UE ca choose to cocetrate its power o fewer resources, so that BS should get right received power. To guaratee P o for the ect li, the maximum umber of RBs UE ca use is N max = Pmax P o L, (4) where represets the ceilig fuctio. Whe oly distacedepedet pathloss is cosidered, we see from Fig. that i a cell of radius R =m, to guaratee P o = 9 dbm per RB at BS, a UE at cell-edge ca oly use oe RB for ect li, while a UE or RN at d = R ca use up to RBs. Limited by the used umber of RBs ad P max, the throughput for celledge users is very low whe usig the ect li. If cell-edge user ca exploit the availability of a large umber of relay cadidates i the etwor to help to forward its traffic, the by allocatig more resources amog DD li ad bachaul li, the throughput for edge users ca be icreased. B. Iter-Cell Iterferece The achievable throughputs for the UEs ot oly deped o the received powers at BSs or RNs, but also o the overall iterferece powers comig from adjacet cells. I this paper, we are iterested i the log-term average throughputs for all UEs, so the average iterferece levels both at both BSs ad RNs are cosidered. The average SINR at the serverig base statio i for the ect li of UE is γ = P ( ) rx,rb L,N, (5) I i + P N,bs where N is the umber of RBs used by UE for ect li, P N,bs is the thermal oise power per RB at BS. For a bachaul li betwee RN ad the servig BS, γ = P ( ) rx,rb L,N, (6) I i + P N,bs I (5) ad (6), the average iterferece power per RB at BS i is I i = w P tx, + w P tx,. (7) L,i L,i j i Uj R j Where w,w are factors related to iter-cell chael radomizatio ad the activities (both o frequecy ad time domai) of UE ad RN. For example, uder idepedet ad radomized chael allocatio, for a device which is selected to trasmit o N RBs all the time, we have w = N /M. I i depeds heavily o the distributio of UEs, the power cotrol parameter β ( with fractioal power cotrol ) ad the targeted received power P o. Uder upli TPC, P rx,rb at BSs is close to P o. However, for a DD li, the received sigal power at RN depeds o the varyig DD pathloss. For DD relayig, the targeted received power for DD should be larger tha P o. Also, the iterferece powers at RNs have differet statistical characteristics compared to BSs simply because upli TPC is doe for the servig BSs, ot for DD lis, ad some eighbor cell iterferers are closer to RNs tha BSs. For a DD li betwee UE ad RN, the average SINR is ( ) P tx L γ, =,N N L, (I + P N,r ), (8) where P N,r is the received oise power per RB at the RN. I is the iterferece power per RB at the RN. The istataeous throughput for a li (with SINR γ ad N rb RBs) is approximated by Shao equatio as R = N rb B rb log ( + γ). (9) III. ANALYSIS OF DD RELAYING UNDER TPC Let us assume that each UE ca get at most N R RBs for its flow o average over time. For UEs which are closer to the BS, uder strict TPC, they ca use more RBs for ect li. For cell edge users, they ca oly mae sure that targeted received power is achieved o fewer RBs at BSs, see Fig. whe P o = 9 dbm for example. If there are RNs that are close to a cell-edge UE ad at the same time have good chaels to BS, the usig relayig would help to boost edto-ed throughput performace for cell-edge UE by usig more RBs o bachaul hop. For relayig trasmissio, due to the half-duplex costrait, RN caot receive ad trasmit at the same time, we allocate xn RBs for bachaul li ad ( x)n RBs for the DD li, where x is the fractio of time duratio allocated to bachaul li. It should be oticed that the trasmissio power of DD li is determied by the ect li pathloss, other tha DD pathloss. As a result, a cell-edge UE would use P max liely. For bachaul li, uder TPC, the received power per RB at BS would be P o. The estimated ed-to-ed throughput for DD relayig is: R, ee =mi { xr, ( x)r } () where R = N B rb log ( + γ ) is the istataeous rate for bachaul li, R = N B rb log ( + γ, ) is the istataeous rate for DD li. R icreases as N icreases, ad R also icreases as N icreases. For ect li, the ed-to-ed throughput is: R = N B rb log ( + γ ) () Whe usig DD relayig, the resource allocatio problem ca be formulated as: max x,n s.t.,n xr (a) ( x)r = xr (b) ( x)n + xn N R (c) N Pmax P o L (d) N P max P o L, (e) <x<. (f) To maximize (a), both N,N should be as large as possible. However, the badwidth resource allocatio should also meet with the costraits (b) (f). If (c) starts to tae effect, the larger badwidth should be allocated to the better hop. I this paper, a efficiet resource allocatio

4 Algorithm BH/DD Resource Allocatio INPUT: L,,L OUTPUT: R ee : N : N,L,,N Pmax P o. L Pmax P o. L,, ad average umber N R of RBs,N 3: Calculate R ad R. 4: Calculate time partitio x R R +R 5: if ( x)n + xn N R the 6: brea; / / 7: else if R N <R N the 8: N NR xn ( x), go to 3. / / 9: else if R N R N the : N NR ( x)n x, go to 3. : ed if : R, ee = xr, retur. algorithm is proposed to solve the above optimizatio problem, see algorithm for full details. To show the gai of DD Relayig uder power cotrol, aalysis for -Dimesioal settig with distace-depedet pathloss is performed. As depicted i Fig. 3, distace betwee ( x) N d N d xn d. Fig. 4. Achievable Max Ed to Ed Throughput Data Rate (Mbps) Optimal RN positio DD distace d Best RN positio for cell-edge UE with d = R Direct li DD li Bachaul li Max Throughput of Relayig 5 5 RBs i frequecy domai Fig. 3. DD Relayig -Dimesioal example Fig. 5. Example: DD relayig gai for cell-edge UE UE ad servig BS is d, betwee UE ad RN is d, betwee RN ad servig BS is d = d d. Assume the umber of RBs used for ect li is N. For a RN at differet positios, algorithm is adopted to fid the optimal resource allocatio. We ca see from result i Fig. 4 that the optimal RN positio is d =.5 R. At this optimal positio, the istataeous throughputs for the ect li, DD li ad bachaul li are plotted i Fig. 5. IV. RELAY SELECTION AND RESOURCE ALLOCATION SCHEME I a cell with multiple users, oe has to allocate resources amog users. We assume that for each cell, the pathlosses L, L ad L, ( U i, R i ) are give. We search for a proportioally fair solutio to the joit relay selectio ad resource allocatio problem. Let us use X to idicate whether active UE uses relayig or ect trasmissio, X = meas that UE commuicates ectly with BS, X = meas that UE uses RN to relay its data. For this, we eed to search for X, N, NX ad N, so that the proportioally fair utility fuctio U = log ( R ee ) (3) U i is maximized i each cell. The costrait for this optimizatio problem is U i ( N +( x )N + x N X ) = M. (4) I geeral, the joit relay selectio ad resource allocatio problem uder upli power cotrol is hard to be solved due to its mixed iteger character ad the oliearity brought by TPC. A. Joit Relay Selectio (RS) ad Resource Allocatio (RA) Algorithm With DD relayig, we divide the resource allocatio scheme ito two step. Firstly, to guaratee the fairess amog all the active users i oe cell, the RBs are allocated to each user flow evely. Secodly, the resources available for each user is allocated to DD li ad bachaul li usig the algorithm. The relay selectio here is doe together with the resource allocatio, see algorithm. For compariso, i the sceario where DD relayig is ot adopted (oly ect lis are used), the resource allocatio i oe cell is also preseted here. Firstly, we sort the UEs i curret cell accordig the pathloss to BS, ad calculate Nrb rest rest rest /NUE, where Nrb is

5 umber of remaiig RBs ad NUE rest is the umber of UEs ot served. Secodly, startig from the UE havig the largest pathloss, N x =mi(nrb rest rest /NUE, P max/(p o L) ) RBs are allocated to this UE to mae sure that the received power per RB at BS is P o, the we update ew Nrb rest = Nrb rest N x, ew NUE rest = N UE rest. The the remaiig RBs are allocated amog the ot served UEs i the same way util all UEs i this cell are served. Algorithm Joit RS ad RA for DD Relayig : There are ˆK UEs ad ˆN RNs i curret cell, the UEs i this cell are deoted by =,,..., ˆK, RNs are deoted by =,,..., ˆN, N R = M/ ˆK. : for =to ˆK do 3: Calculate N max usig (4), N mi (N max,n R ), calculate R usig (). 4: for =to ˆN do 5: if L L or L, L the 6: R, ee =, brea; 7: else 8: Calculate R, ee,n 9: ed if : ed for ( : R ee max R,Ree,,..., Ree, ˆN best relay X for UE. IfR ee is used for UE. : ed for,n usig Algorithm. ), ad assig the = R, ect li V. SIMULATION RESULTS The simulatio parameters is listed i table I. We assume a urba eviromet with a propagatio expoet of β =3.76, ad log-ormal shadowig. The simulatio area is square with wrap-aroud edges. The TPC parameters P o = 9 dbm ad β =are used to esure that the multi-cell system is properly worig [7]. The I i ad I are first estimated, ad the to be refied durig simulatio i a iterative process. The average iterferece power experieced at each RN or BS is computed after the relay selectio ad resource allocatio have bee doe. For each iteratio, N bs K UEs ad N bs N are dropped iside the simulatio sceario, pathlosses are calculated usig distaces ad the geerated shadow fadigs. The joit RS ad RA algorithm is used to fid the best relay for each active UE ad to determie the umber of RBs used by each li. Fially, the actual SINRs at each BS ad RN are TABLE I. SIMULATION PARAMETERS Simulatio Parameter Symbol value Number of BSs N bs Average active UEs per cell K Average relay cadidates per cell N Iter-site Distace D.73m Pathloss expoet α 3.76 Pathloss icludig atea gai at cell border L 4 db TPC parameter β Targeted received power per RB P o -9 dbm UE maximum Tx power P max 3 dbm Stadard deviatio of SF δ SF 8dB Thermal oise level at BS N,bs -7 dbm/hz Thermal oise level at RN N,r -65 dbm/hz Resource bloc badwidth B sub Hz Number of RBs for each cell M calculated usig the pathlosses, trasmissio powers ad the resource allocatio results for the whole system. A. Relay Selectio Result The relay selectio result i Fig. 6 shows that the best relay is heavily depedet o the locatio of RN ad the result is very similar to [8]. The domiat compoet i pathloss comes from the locatios of the odes. I our simulatio, we foud that the best relays for cell-edge users are maily located at half-distace locatios. This is because that RNs at these places ca use more RBs ad are simultaeously close eough to cell-edge UEs to get higher DD SINRs. Fig Distace i m Relay selectio results B. SINR ad throughput performace CDF Fig at BS at RN CDF of Sigal SINRs Active UE Selected RN 3 4 SINR i db SINR at base statios ad relay odes The SINRs i the simulatio are explicitly computed usig the pathloss values ad trasmissio powers. It ca be see from Fig. 7 that that average SINR experieced at selected RNs is better the at BSs. As TPC is performed for BSs, ot for BS RN

6 RNs, the received sigal powers per RB at RNs are larger tha P o. Fig. 8 shows the throughput performace of DD-Relayig CDF Fig CDF of Data Rate(Mbps) w/o Relayig DD Relayig Data Rate Throughput performace of DD-Relayig ad without DD-Relayig. I the sceario without DD- Relayig, UEs close to BSs use more resources tha cell-edge users. Despite attemptig proportioal fairess i schedulig, cell-edge UEs simply do ot have sufficiet power to use their fair share of resources. I the DD-Relayig sceario, UEs close to BSs share the RBs equally with cell-edge UEs. With DD-Relayig, cell-edge UEs ca leverage the availability of multiple devices to use more RBs for their flows. Usig DDrelayig, cell-edge throughput at th percetile shows a improvemet of more tha 3%, while there is a slight loss i average system throughput. I the sceario without DD relayig, all RBs are used for trasmittig data to the BS, with the target P, whereas i the DD relayig sceario, some RBs are used for DD trasmissios, which do ot cotribute to EE throughput ectly. The egative average effect of this is couteracted by the decrease i the average iterferece level due to relayig the average trasmit power i a cell becomes lower, so the co-chael iterferece from other cells is lower. I the simulated sceario, these two effects almost cacel. Fially, the fairess metric related to (3) shows a 9% improvemet by usig DD-Relayig. TABLE II. SIMULATION RESULTS -PERCENTILE THROUGHPUT Sceario Average th 5th w/o DD Relayig 5.5 Mbps.3 Mbps.8 Mbps with DD Relayig 4.86 Mbps.95 Mbps 5. Mbps DD Relayig Gai -3.8% 33% 78% VI. CONCLUSIONS I this paper, we studied a DD-relayig eabled cellular etwor i upli. We cosidered a multi-user, multi-carrier ad multi-cell etwor with upli power cotrol to mitigate the iter-cell iterferece ad i-bad emissio iterferece. With power cotrol, the relay selectio problem is differet from that without, due to the costat receive powers per RB at BSs. To eable DD-relayig, the relay selectio, resource allocatio ad power cotrol problem must be addressed together. We formulated this mixed problem as a optimizatio problem. We used system simulatio to study the performace of a simplified scheme. The simulatio results demostrate that via proper resource schedulig, DD-relayig uder power cotrol icreases throughput performace for cell-edge users sigificatly, which results i cosistet user experiece. REFERENCES [] NGMN Alliace, NGMN 5G White Paper, Mar. 5. [] D. Lopez-Perez, M. Dig, H. Clausse, ad A. H. Jafari, Towards Gbps/UE i cellular systems: uderstadig ultra-dese small cell deploymets, arxiv:53.39, Mar. 5. [3] X. Ge, H. Cheg, M. Guizai, ad T. Ha, 5G wireless bachaul etwors: challeges ad research advaces, IEEE Networ, vol. 8, o. 6, pp. 6-, Nov. 4. [4] Z. Li, M. Moisio, M. A. Uusitalo, et al., Overview o iitial METIS DD cocept, It. Cof. o 5G for Ubiquitous Coectivity, pp. -6, Nov. 4. [5] METIS Deliverable D6.3, Itermediate system evaluatio results, Aug. 4. [6] M. Tehrai, M. Uysal, ad H. Yaiomeroglu, Device-to-device commuicatios i 5G cellular etwors: challeges, solutios, ad future ectios, IEEE Comm. Mag., vol. 5, o. 5, pp. 86-9, May. 4. [7] J. da Silva Jr, G. Fodor, ad T. Maciel, Performace aalysis of etworassisted two-hop DD commuicatios, i Proc. IEEE Globecom Worshops, pp. 5-56, Dec. 4. [8] A. Bletsas, H. Shi, ad M. Wi, Cooperative commuicatios with outage-optimal opportuistic relayig, IEEE Tras. Wireless Commu., vol. 6, pp , Sept. 7. [9] S. H. Nam, M. Vu, ad V. Taroh, Relay selectio methods for wireless cooperative commuicatios, i Proc. Cof. o Ifo. Scieces ad Systems, pp , Mar. 8. [] Y. Jig, H. Jafarhai, Sigle ad multiple relay selectio schemes ad their achievable diversity orders, IEEE Tras. Wireless Commu., vol.8, o.3, pp , Mar. 9. [] K. Vagauru, M. Puzio, G. Sterberg, Y. Fa, S. Kaur, Upli System Capacity of a Cellular Networ with Cooperative Mobile Relay, i Proc. Wireless Telecommu. Symp., pp. -7, Apr.. [] K. Vagauru, S. Ferrate, ad G. Sterberg, System capacity ad coverage of a cellular etwor with DD mobile relays, i Proc. IEEE Military Commu. Cof., pp. -6, Oct.. [3] S. H. Seyedmehdi ad G. Boudreau, A efficiet clusterig algorithm for device-to-device assisted virtual MIMO, IEEE Tras. Wireless Commu., vol. 3, o. 3, pp , Mar. 4. [4] J. Liu, N. Kato, Device-to-Device commuicatio overlayig two-hop multi-chael upli cellular etwors, i Proc. ACM It. Symp. o Mobile Ad Hoc Networig ad Computig, pp , Jue 5. [5] 3GPP TS 36.3 V.5., Evolved Uiversal Terrestrial Radio Access, Physical layer procedures (Release ), Apr. 4. [6] U. Oruthota, O. Tiroe, P. Dharmawasa, Aalysis of upli power cotrol i cellular mobile systems, i Proc. IEEE Veh. Tech. Cof, pp. -5, Jue 3. [7] E. Tejaswi, B. Suresh, Survey of Power Cotrol Schemes for LTE Upli, It. J. Comp. Sci. ad Ifo. Tech., Vol. 4, No., 3. [8] L. Xiao, T. Fuja, ad D. Costello, Mobile relayig: coverage extesio ad throughput ehacemet,ieee Tras. Commu., vol. 58, o. 9, pp , Sept..