THE third Generation Partnership Project (3GPP) has finalized

Transcription

1 3 8th Internatona Conference on Communcatons and Networkng n Chna (CHINACOM MU-MIMO User Parng Agorthm to Acheve Overhead-Throughput Tradeoff n LTE-A Systems Yng Wang, Fe Peng, Wedong Zhang, Yuan Yuan State Key Laboratory of Networkng and Swtchng Technoogy, Beng Unversty of Posts and Teecommuncatons Ema: pengfe65@gma.com Abstract Ths paper nvestgates user parng agorthm n mutpe-user mut-nput mut-output (MU-MIMO systems. Consderng that users report Precodng Matrx Index (PMI and Rank Indcator (RI to ndcate precodng matrx, mutpe orthogona PMI&RI custers are desgned to reduce the MU-channe state ndcator (MU-CSI feedback overhead. On the bass of such predefned orthogona custers, a nove user parng agorthm s proposed to make the tradeoff between ce throughput and MU-CSI feedback overhead. To further smpfy computatona compexty, we propose a suboptma agorthm. Unke prevous researches, the transmsson data streams of pared users are adaptvey adusted accordng to the channe condton. Smuaton resuts show that the ce throughput gan acheved by our proposed agorthms s cose to that acheved by other user parng agorthms, wth ess feedback overhead n Long Term Evouton-Advanced (LTE-A systems. I. INTRODUCTION THE thrd Generaton Partnershp Proect (3GPP has fnazed the rado nterface specfcatons for the next generaton mobe communcaton system caed LTE-A (Long Term Evouton-Advanced systems []. To mprove spata mutpexng, mut-user MIMO (MU-MIMO has been adopted n LTE-A systems to meet the IMT-Advanced spectra effcency targets []. From the perspectve of commerca depoyment, the mpementaton of mut-antenna n evoved Node B (enb and user equpment (UE makes the MU- MIMO feasbe. Based on the channe correaton, a transmtter serves mutpe users on the same tme and frequency resource n MU-MIMO systems [3] [4]. In downnk MU-MIMO systems, nter-user nterference (IUI from pared user s data stream(s s the botteneck of the system throughput [5]. Therefore, user parng agorthm, whch chooses a set of approprate users to mtgate IUI, s regarded as a key factor to MU-MIMO systems. Usuay, the user parng agorthm needs accurate channe state nformaton of a servng users. Wth fu channe matrx nformaton at enb, [6] proposes a user parng scheme that acheves the maxma capacty wth drty paper codng (DPC to emnate IUI. But the hgh compexty makes t unreastc. [7] proposes a sem-orthogona user schedung method wth zero forcng beamformng. [8] proposes a sgna to eakage pus nose rato (SLNR based parng method. [9] proposes an orthogona parng to make fu use of spata separaton. However, most of these works need the perfect channe matrx nformaton at enbs sde, whch makes t unacceptabe n practce. Therefore, user parng agorthms wth mted feedback are necessary. In LTE-A systems, UE measures the downnk (DL channe and feeds back the channe state ndcator (CSI wth estmaton error n the form of recommed transmsson formats. On the bass of mted feedback, [] proposes the best companon parng agorthm. Each UE feeds back the preferred companon, whch s actuay the precodng matrx ndex (PMI of to-be-pared user wth east nterference to the target user. The best companon ony ndcate one preferred PMI, such that t s hard for prmary user to fnd the best companon. Thus, an suboptma agorthm s proposed n []. The best companon s exted to a best companon custer such that user has a wder range of parng companon. More optons may deterorate system throughput due to the suboptma companon. Each UE feeds back the custer ndex rather than one preferred PMI. Accordngy, the feedback overhead s decreasng at the cost of the system throughput. Moreover, for each parng user, there are two key factors nfuencng system performance: If pared users are n ower SINR regon, the ce average spectrum effcency may be deterorated. Adaptve rank coud make more use of spata separaton n parng user. Constrcted rank ndcator (RI to one s not effectve n MU-MIMO. However, the agorthms proposed n [] and [] ddn t take these nto consderaton. As a resut, the MU-MIMO parng may not guarantee hgher throughput compared to SU-MIMO n the best companon parng agorthm. Therefore, ths paper proposes a user parng agorthm to acheve feedback overhead and throughput tradeoff. The orthogonaty of precodng matrx/vector are nvestgated and mutpe orthogona PMI&RI custers are desgned to reduce MU-CSI feedback overhead. The predefned orthogona custers gude enb to choose the potenta to-be-pared user. Moreover, unke [] [], adaptve RI of each user s adopted. Furthermore, to smpfy computatona compexty, a suboptma agorthm s proposed wthout the exhaustvey search of a possbe users. Smuaton resuts show that, the proposed agorthm can make a reasonabe tradeoff among ce throughput, feedback overhead, computatona compexty and feasbty of mpementng n both outdoor and ndoor scenaros. The rest of ths paper s organzed as foows. Secton II gves system mode. In Secton III, the parng agorthms are descrbed. Secton IV presents the system smuaton IEEE

2 assumptons and resuts. Dfferent parng agorthms, correspondng feedback overhead, and computatona compexty are compared and anayzed. Concuson s gven n Secton V. enb sde User Parng PHAES II Schedung CSI Fg.. II. SYSTEM MODEL Channe Feedback Channe UE sde Perodc CSI Cacuaton Transmsson prncpes n LTE-A systems PHAES I We consder a downnk MU-MIMO system wth tota L servng enbs. The set of a servng enbs s denoted as L. Each enb has K users unformy dstrbuted. The system s equpped N t transmt antennas and N r receve antennas at each actve user. A maxmum of two users w be pared together n the same resource bock (RB to make the MU- MIMO transmsson and each user s assumed to be aocated up to two streams. The MIMO channe s typcay modeed by H of user n the th enb. The receved sgna CNr Nt vector can be wrtten as: Y = H W, X, + K s L,s u= H u s Ws u,v X u,v s + n, ( n whch the second term represents co-channe nterference caused by other enbs, and n s the addtve whte Gaussan nose. X, denotes the desred transmt vector of pared user, (,, {,,, k,, K}: [ (x X, ( ] T = T T x, ( where x k C νk, (k =,,, K s the desred transmsson vector of user k n the th enb and ν k s the number of transmsson spata ayers of user k. ( T denotes the transpose operator. Cose-oop MU-MIMO system s consdered n ths paper, thus X, before enterng nto the MIMO channe. W, s pre-mutped by a precodng matrx W, s the combned precodng matrx of the pared users: W, = [ w w ], (3 where w k C Nt ν k, (k =,,, K s the precodng matrx of user k n the th enb. Here, w k s seected from W based on codebook mappng n LTE-A systems []. The transmsson prncpe used n LTE-A systems s depcted n Fg. and can be carred out n the foowng two phases: PHASE I: UE cacuates the preferred precodng matrx, preferred number of transmtted spata ayers, and postprocessng SINR perodcay, then reports PMI, RI, CQI to ts servng enb. Based on Equaton (, usng the maxmum channe capacty (Max-CP crteron based PMI, RI seecton agorthm, UE perodcay seects the PMI and RI such that the nstantaneous system capacty can be maxmzed [3] [4]. In case the mth RB s aocated to UE k n th enb, the MIMO channe H, transmt vector x k and post-processng SINR of νth data stream can be respectvey expressed as: H (k,m, x (k,m and SINR (k,m,ν. The RB s defned as transmt tme-frequency channe n LTE-A systems []. The channe capacty of UE k n the mth RB s gven as: ν k ( C (k,m = og + SINR (k,m,ν, (4 ν= where the cacuaton of SINR (k,m,ν can be seen n [5]. Consequenty, the RI vaue for UE k can be seected as: RI k = arg max v k {,} C(k,m (v k. (5 For UE k, to acheve hgher SINR, desred receved sgna power shoud be maxmzed. Thus, the prncpe of PMI seecton for UE k s gven as: P MI k = arg max w k W ( H (k,m w k. (6 PHASE II: Based on the reported PMI and RI, enb seects the precodng matrx and number of transmsson spata ayers to transmt the desred sgnas for UE. Accordng to certan schedung and user parng agorthms, enb schedues a prmary UE and a parng UE for each RB. III. ALGORITHM DESCRIPTION In ths secton, the orthogona PMI&RI custers are desgned and a nove user parng agorthm together wth a suboptma agorthm s proposed. A. Desgn Orthogona PMI&RI Custers In genera, the more uncorreated the transmssons channes of the pared users, the ower the ntra-par s nterference. In MU-MIMO systems, one way to mtgate the ntra-par s nterference s choosng uncorreated transmsson channe. Due to the practca depoyment, perfect channe nformaton s not avaabe at enb sde. Consequenty, enb needs some ndcaton to conduct user parng. Based on Equaton (6, PMI s the ndex of the recommed SU-MIMO precodng matrx n the precodng codebook. By pre-mutpyng the precodng matrx, each user coud acheve ts maxma receved power and mtgate other users nterference to the maxmum extent. Therefore, n the cose-oop MU-MIMO system, choosng the orthogona PMI of potenta parng user s an effectve way to par user. In the condton that w coud maxmze the receved sgna power H (,m of user, the ntra-par nterference w 348

3 woud be mtgate f the precodng matrx w of to-be-pared user s totay orthogona to the user s precodng matrx w. The orthogonaty between two precodng matrces s nversey proportona to the Chorda dstance. The Chorda dstance between two matrces w and w s gven n [6] and represented by d chord (w, w = w (w H ( H w w, (7 wth F beng the Frobenus norm of the matrx and ( H beng the Hermtan operator. Wth the maxmzed Chorda dstance, w stays n the antpoar poston of w and hence H (,m w s mnmzed yedng the mnmzed nter-user nterference. The foowng anayzes the orthogonay of precodng matrx/vector based on Chorda dstance. Fxed Data Stream: Through the anayss of predefned codebook n LTE-A systems [], we defne w {ν} W {ν} as the precodng matrx, where s the ndex, ν ndcate the number of data streams and W {ν} denotes the whoe set of precodng matrx. For fxed data stream, the orthogonaty of precodng vector s based on the foowng rues: ( orth w {ν}, w {ν} = ( d chord w {ν} F. (8, w {ν} In practca scenaros, the fxed data stream coud mt the system performance n some degree. Hence, the adaptve data streams shoud be anayzed to meet the parng demand. Mxed data stream: Accordng to the predefned codebook n LTE-A systems [], the precodng vector of one data stream s equa to the frst coumn of precodng matrx of two data streams. Wth out oss of generaty, assumng the data stream of w and w are one and two, respectvey. Then, the orthogonaty of mxed data streams can be defned as: orth(w {} = ( orth, (w {} + orth, (w {}. Based on Equaton (8, orth, (w {} s the orthogonaty between the precodng vector w {} and the frst coumn of precodng matrx w {}, whe orth, (w {} s the orthogonaty between the precodng vector w {} and the second coumn of precodng matrx w {}. The constructon of pared-user specfc orthogona PMI&RI custer for mxed data stream can be desgned as Agorthm I. Based on the sorted resuts n Agorthm I, each w {ν} corresponds to a orthogonaty custer C {ν}. The eements n ths custer are a the other precodng matrces, whch are sorted n the ascng order of orthogonaty wth w {ν}. From the top to the n ths custer, the PMI of w {ν} ndcates the arger orthogonaty and the arger correaton between the two precodng matrces/vectors. Ths custer s predefned and conserved both at enb and UE sde. For the predefned codebook n LTE-A systems [], the tota number (9 Agorthm I: Desgn Orthogona PMI&RI Custer for Mxed Data Streams fora the w {ν} W {ν} do f ν = then fora the w {} W {} do Cacuate orth(w {} based on Equaton (8; fora the w {}, w {} W {} do Cacuate orth(w {} based on Equaton (9; Sort a the orth(w {}, w {} and orth(w {} n the ascng order; ese n = repeat fora the w {n} W {n} do Cacuate orth(w {}, w {n} based on Equaton (8 or Equaton (9; n = n + ; unt n > ; sort a the orth(w {}, w {} and orth(w {} n the ascng order; of PMI s 6, that s to say, [, 6]. The number of data stream s up to two, such that ν [, ]. Therefore, the maxma number of eements n ths custer s 3. However, the number of eements n the custer ays great mpact on system performance. Because the orthogonaty between the two matrces coud be deterorated f the to-be-pared precodng matrx s at the of the custer. The arger orthogonaty of the pared users precodng matrx, the hgher the ntra-par nterference s. Smuaton resuts n Secton IV show the proper number of eement n the custer s 4 or 5. B. User Parng Agorthms In cose-oop MU-MIMO system, we propose two user parng agorthms carred out n the foowng three steps. Step : Custer Seecton Based on the preferred PMI&RI reported from user, enb seects the correspondng orthogona custer and chooses the preferrng PMI&RI of the desred par user. The user s preferred PMI&RI corresponds to a precodng matrx denoted as w {ν}, where ndcates PMI and ν ndcates RI. Snce each w {ν} hods a orthogona PMI&RI custer C {ν} and ths custer s predefned and conserved both at enb and UE sde. Therefore, enb can focus on the desred orthogona PMI&RI custer to fnd the potenta to-be-pared PMI&RI. 349

4 Step : Parng After focusng on the desred PMI&RI orthogona custer, the next step s searchng the rest of users to match the preferrng PMI&RI. To acheve optma performance, exhaustve search of parng user s adopted. However, f the number of UE s huge, the exhaustve search may ead to hgher computatona compexty. Then, the foowng gves an optma and a suboptma agorthm to execute ths step. Orthogona PMI&RI Sort Parng Agorthm (OPS: For prmary user UE, assumng the rest of potenta parng user set s U and the number of eements n the user-pared specfc orthogona PMI&RI custer s P [, 3]. To acheve optma performance, enb conducts exhaustve search of potenta parng users. The detaed procedure of ths proposed OPS s presented n Agorthm II. Agorthm II: Orthogona PMI&RI Sort Parng (OPS Intazaton U a the rest of potenta parng user; p = ; par ndcator = Fase; for p = to P do fora the UE U do f w {ν} = C {ν} {p} then Cacuate C MU C(,m (4; ( Cacuate C SU on Equaton (4; f C MU C SU then Par UE and UE ; par ndcator = True; Break ths oop; C (,m ese Contnue next oop; f par ndcator = Fase then Transmt UE under SU-MIMO mode; based on Equaton + C (,m based Orthogona PMI&RI Custer Parng Agorthm (OPC: On the condton that the number of potenta user s huge, a suboptma user parng agorthm s proposed wthout exhaustve search. The detaed procedure of ths proposed OPC s presented n Agorthm III. Step 3: Comparson Between MU and SU Based on the CQI reported from user, enb cacuates the throughput of the MU-MIMO user par and SU-MIMO snge users respectvey, then decde to conduct MU-MIMO or SU- MIMO transmsson. If each UE ony reports CQI to maxmze the throughput of one snge nk wthout consderng the Agorthm III: Orthogona PMI&RI Custer Parng (OPC Intazaton U a the rest of potenta parng user; par ndcator = Fase; fora the UE U do f w {ν} C {ν} then Cacuate C MU ( C(,m based on Equaton (4; Cacuate C SU C (,m + C (,m based on Equaton (4; f C MU C SU then Par UE and UE ; par ndcator = True; Break ths oop; ese Contnue next oop; f par ndcator = Fase then Transmt UE under SU-MIMO mode; potenta MU parng wth other actve UEs [7], t w ntroduce the msmatch of SU-CQI and MU-CQI. To mprove the CQI, we take the average CQI from a PMIs n the PMI&RI custer as MU-CQI. For some UE sufferng ower SINR, choosng another user to conduct MU-MIMO may deterorate the system throughput. Therefore, after choosng one potenta parng user, enb need cacuate the throughput of the MU-MIMO user par C MU and SU-MIMO snge users C SU respectvey. If the throughput of MU-MIMO par s hgher than that of two SU-MIMO snge users, MU- MIMO transmsson shoud be mpemented and vce versa. The detaed procedure s presented both n Agorthm II and Agorthm III. IV. SIMULATIONS AND DISCUSSIONS The system performance of the proposed agorthm s e- vauated by means of system-eve smuatons usng 3GPP LTE-Advanced evauaton methodooges [8]. We consder the typca depoyment scenaros InH and UM. Tabe I gves the man system-eve smuaton parameter confguraton. A. Performance Evauaton of OPC and OPS As mentoned above, a maxmum of two users w be pared together n the same RB to make the MU-MIMO transmsson and each UE s assumed to be aocated up to two streams. When a user s pared va MU-MIMO, the transmt power of each user s reduced by haf to provde smar Interference over Therma (IoT eve between SU-MIMO and MU-MIMO. The average spectrum effcency (SE of dfferent number of eement n to-be-pared PMI custer s compared n Fg.. We can observe that wth 4 or 5 eements n PMI&RI custer, the 35

5 Parameter Smuaton scenaro Carrer frequency System bandwdth Load Channe mode Dupex method TABLE I SYSTEM LEVEL SIMULATION PARAMETERS Number of antenna eements (enb,ue Traffc mode CQI feedback deay CSI reportng mode Lnk-to-system mappng Inter-ce nterference Schedue HARQ Soundng reference Recever agorthm Lnk adaptaton and target nta BLER Assumpton InH & UM GHz MHz Average UE per sector Spata Channe Mode(SCM FDD downnk (4, Fu buffer 5ms Mode-:Wdeband PMI,CQI, Subband CQI EESM Reastc Proportona Far HARQ-CC(Chase Combng; 8 processses; Maxmum 4 retransmsson tmes Idea MMSE FLA,% SE acheves hgher gan. In addton, because the orthogonaty between the eement rankng ast n the to-be-pared custer and the target PMI s gettng worse, such that the SE s reducng wth the ncreasng of the number of eement n to-bepared custer. Wth the resut, the adaptve user data streams outperform the n both scenaros, because the mxed data streams coud make more use of spata separaton. B. Performance Comparson and Dscusson We use Orthogona Parng (OR [9], Best Companon Parng (BCP [] and Best Companon Custer (BCC [] as the comparson, whe BCC s the basene. OR seects users wth orthogona rado channes to emnate nterference. BCP needs feedback nformaton of so-caed Best Companon ndex at UE sde, whch s actuay the PMI of to-be-pared user seected by the prmary user. Each UE shoud feed back 4bts Best Companon PMI. In BCC [], the best companon s exted to an best companon custer such that user has a wder range of parng companon. Each UE feeds back the custer ndex(bts rather than one preferred PMI. The average spectrum effcency of dfferent parng agorthms, the reatve gan, feedback overhead and compexty anayss are gven n Tabe II. Wth the resuts, the OR outperforms a the other parng agorthms because of the fu quantfed channe matrx at enb sde. The OPS and OPC can outperform BCC n both scenaros whe acheve amost dentca average SE compared to BCP. Moreover, wth the predefned PMI&RI custer conserved at enb, OPS and OPC need not to feed back companon ndex or companon Tota average SE of OPC parng (InH number of eement n to-be-pared custer Tota average SE of OPC parng (UM number of eement n to-be-pared custer Fg Tota average SE of OPS parng (InH number of eement n to-be-pared custer Tota average SE of OPS parng (UM number of eement n to-be-pared custer Tota average SE of dfferent eements n orthogona custer custer ndex. Therefore, OPC and OPS can make reasonabe tradeoff between feedback overhead and ce average throughput. Meanwhe, compared to the, the throughput gap of mxed data streams between the two proposed agorthms and BCP s smaer. The reason s that the orthogonaty dfference n s more obvous than mxed data streams. Then the nterference mtgaton of s more effectve than the mxed data streams. Furthermore, the SE gan of InH s hgher than UM due to the sowy varyng channe n InH scenaro. In terms of computatona compexty, the OPS outperforms OPC wth the exhaust search of a ce UEs. OR need to cacuate the channe matrx correaton between prmary UE and a the rest of potenta parng UE. Then to fnd the best parng user such that the computatona compexty s o (K! wth K users n each ce. Unke OPC and BCC, BCP and OPS a conduct exhaustve search to fnd the potenta parng user. The computatona compexty s ony correaton wth traversng a the K users. Accordngy, the computatona compexty of BCP and OPS s o (K. Therefore, the OPC agorthm conducts more smpe procedure wth a ower computatona compexty, whch s more practca than dea OR agorthm and more robust than BCP, BCC and OPS. V. CONCLUSION In ths paper, we desgn mutpe pared-user specfc orthogona PMI&RI custers to reduce the MU-CSI feedback overhead. An optma MU-MIMO user parng agorthm (OPS s proposed to make the tradeoff between ce throughput and MU-CSI feedback overhead. To further smpfy computatona compexty, a suboptma agorthm (OPC s proposed wthout 35

6 TABLE II PERFORMANCE COMPARISON Average SE Gan User parng agorthm One data stream Mxed data stream Feedback overhead Computatona compexty InH UM InH UM OR [9] 5.7% 3.3% 3.%.6% Fu quantfed channe matrx o (K! BCP [] 5.4%.% 4.3%.67% SU-CQI+MU-CQI+Best Companon PMI o (K 4bts+4bts+4bts BCC [].%.%.%.% SU-CQI+MU-CQI+Best Companon Custer Index < o (K 4bts+4bts+bts OPC 4.95%.% 4.6%.7% SU-CQI+MU-CQI < o (K 4bts+4bts OPS 8.7%.74% 6.9%.59% SU-CQI+MU-CQI o (K 4bts+4bts exhaustve search of potenta parng user. Unke prevous researches, the transmsson data streams of pared users are adaptvey adusted accordng to the channe condton. Smuaton resuts show that, n cose-oop MU-MIMO system, the proposed agorthms can outperform tradtona agorthm n InH scenaro and acheve amost same ce throughput gan wth ess feedback overhead n UM scenaro. ACKNOWLEDGMENT Ths work s supported by Chnese Natona Key Proect under Grant No.3ZX35, Natona Nature Scence Foundaton of Chna (6, Program for Changang Schoars and Innovatve Research Team n Unversty(No. IRT49. REFERENCES [] 3GPP TS 36., Evoved Unversa Terrestra Rado Access (EU- TRA; LTE physca ayer; Genera descrpton (Reease, v., Dec.; ftp://ftp.3gpp.org. [] L. Lu, R. Chen, S. Gerhofer, K. Sayana, Z. Sh, and Y. Zhou, Downnk mmo n te-advanced: Su-mmo vs. mu-mmo, Communcatons Magazne, IEEE, vo. 5, no., pp. 4 47,. [3] H. Du and P. Chung, A probabstc approach for robust eakagebased mu-mmo downnk beamformng wth mperfect channe state nformaton, Wreess Communcatons, IEEE Transactons on, vo., no. 3, pp ,. [4] N. Ravndran and N. Jnda, Mut-user dversty vs. accurate channe state nformaton n mmo downnk channes, Wreess Communcatons, IEEE Transactons on, vo., no. 9, pp , 9. [5] X. Xa, S. Fang, G. Wu, and S. L, Jont user parng and precodng n mu-mmo broadcast channe wth mted feedback, Communcatons Letters, IEEE, vo. 4, no., pp. 3 34,. [6] M. Costa, Wrtng on drty paper (corresp., Informaton Theory, IEEE Transactons on, vo. 9, no. 3, pp , 983. [7] T. Yoo and A. Godsmth, On the optmaty of mutantenna broadcast schedung usng zero-forcng beamformng, Seected Areas n Communcatons, IEEE Journa on, vo. 4, no. 3, pp , 6. [8] X. Xa, G. Wu, J. Lu, and S. L, Leakage-based user schedung n mu-mmo broadcast channe, Scence n Chna Seres F: Informaton Scences, vo. 5, no., pp , 9. [9] Norte, UL Vrtua MIMO System Leve Performance Evauaton for E-UTRA, 3GPP doc. R-54, RAN, 5; ftp://ftp.3gpp.org. [] Acate-Lucent, Best Companon Reportng for Improved Snge- Ce MU-MIMO Parng, 3GPP doc. R-996, R-937, R- 93,R-9546, R-93333, RAN, 9; ftp://ftp.3gpp.org. [] Y. Du, J. Tong, J. Zhang, and S. Lu, Evauaton of pm feedback schemes for mu-mmo parng, Systems Journa, IEEE, vo. 4, no. 4, pp. 55 5,. [] 3GPP TS 36., Physca Channes and Moduaton for E-UTRA, v.5., ; ftp://ftp.3gpp.org. [3] 3GPP TS 36., User Equpment (UE rado transmsson and recepton for E-UTRA, v.5., ; ftp://ftp.3gpp.org. [4] G. Foschn and M. Gans, On mts of wreess communcatons n a fadng envronment when usng mutpe antennas, Wreess persona communcatons, vo. 6, no. 3, pp , 998. [5] C. M. S. Schwarz and M. Rupp, Cacuaton of the spata preprocessng and nk adapton feedback for 3GPP UMTS/LTE, Wreess Advanced (WAD, 6th Conference on, June, pp. C6. [6] D. J. Love and R. W. Heath, Lmted feedback untary precodng for spata mutpexng systems, Informaton Theory, IEEE Transactons on, vo. 5, no. 8, pp , 5. [7] Y. Peng, W. Yang, Y. Zhu, and Y. Km, A nove cs feedback method for dynamc su/mu mmo adaptaton, Wreess Persona Communcatons, pp. 3,. [8] Y. Wang, T. Wang, L. Lng, and C. Sh, On the evauaton of mtadvanced canddates: Methodooges, current work and way forward, n Wreess Communcatons Networkng and Mobe Computng (WCOM, 6th Internatona Conference on. IEEE,, pp