Location-Aware Coordinated Multipoint Transmission in OFDMA Networks

Transcription

1 Locatio-Aware Coordiated Multipoit Trasmissio i OFDMA Networks Ahmed Hamdi Sakr, Hesham ElSawy, ad Ekram Hossai Abstract We propose a ovel Locatio-Aware multicell Cooperatio (LAC) scheme for dowlik trasmissio i OFDMA-based etworks. Compared to the traditioal multicell cooperatio, the proposed scheme oly uses coordiated multipoit (CoMP) trasmissio to serve users with poor SINR. O the other had, users with good SINR coditios are served via multiuser MIMO by a sigle base statio (BS). The proposed scheme uses a joit zero-forcig beamformig with semi-orthogoal user selectio (ZFBF-SUS) trasmissio alog with optimized power allocatio i a semi-distributed maer to maximize the overall system eergy efficiecy (i.e., the average data rate per uit power [bps/watt], or equivaletly, average umber of successfully trasmitted bits per eergy uit [bit/joule]). Numerical results show that the proposed scheme outperforms the scheme that uses cooperatio to serve all users, i terms of eergy efficiecy as well as system capacity ad fairess. I. INTRODUCTION Multiple-iput multiple-output (MIMO) has bee of a great iterest as a key techology for wireless system to roughly icrease the capacity liearly with the miimum umber of trasmit ad receive ateas [1]. Ufortuately, the complexity at mobile equipmets limits the gai of this techique. Multiuser MIMO (MU-MIMO) ad coordiated multipoit (CoMP) trasmissio (also referred as etwork-mimo) are two forms of MIMO i which multiple users with a sigle receive atea ca be served simultaeously i the same subcarrier []. Both schemes boost the capacity by exploitig the spatial multiplexig gai, while shiftig the processig burde ad complexity to the trasmitter side. MU-MIMO is maily used to mitigate the itra-cell iterferece, while CoMP is used to alleviate the iter-cell iterferece especially for cell-edge users. I CoMP, multiple BSs share users data, via backhaul etwork, to form a sigle virtual BS with a large atea array to serve these users. I both schemes, beamformig such as ZFBF is used to cacel out the iterferece resultig from servig multiple users o the same subcarrier [3]. Ufortuately, the deploymet of these techologies to satisfy the ever-icreasig user demad, icreases the eergy cosumptio which will sigificatly cotribute to the global greehouse gas emissios [4]. Therefore, eergy efficiecy is as importat as system capacity i the desig of resource allocatio schemes for cellular etworks. I the cotext of eergyefficiet MIMO systems, [5] proposes a multicell cooperative ZFBF scheme i which all BSs use CoMP trasmissio to maximize the eergy efficiecy while cosiderig the backhaul lik capacity limitatios. I [6], a MU-MIMO resource allocatio scheme i a sigle-cell with multiple-atea BS sceario is discussed to miimize the trasmit power that satisfies give data rate requiremets. The authors i [7] cosider a two-tier sceario i which cooperatig macro BSs are overlaid with clusters of cooperatig femtocells where the eergy efficiecy is maximized uder cross-tier iterferece costraits. Motivated by the performace gais achieved by multicell 1 spectrum partitioig Fig. 1. OFDMA dowlik etwork with M = 3 BSs where each BS has A = 3 ateas. cooperatio ad i order to icrease the eergy efficiecy, we propose a Locatio-Aware multicell Cooperatio (LAC) scheme for dowlik OFDMA etworks. Ulike covetioal CoMP systems i which all users are served via cooperatio, the mai idea of LAC is to alleviate the burde of usig cooperatio to serve cell-cetre users who already receive a high sigal-to-iterferece-plus-oise ratio (SINR). For dowlik trasmissio, this scheme exploits the depedecy of SINR o users locatios to cotrol the tradeoff amog capacity, fairess, ad eergy efficiecy via selectig the appropriate mode of operatio for the users (i.e., CoMP mode or o- CoMP mode). Based o the locatio of a user, the system chooses to serve this user via CoMP usig all the BSs or via MU-MIMO by the earest BS. Icorporatig mode selectio to the system desig gives higher degrees of freedom i the resource allocatio which improves the system performace. We preset a desig paradigm for LAC scheme with the objective of maximizig the system eergy efficiecy. We perform mode selectio, subcarrier allocatio, precodig weight allocatio, as well as power allocatio. The results show that our scheme outperforms the covetioal CoMP systems i terms of system capacity, fairess, ad eergy efficiecy. The cotributios of this work are summarized as follows: We propose a ovel LAC scheme for dowlik MIMO- OFDMA etworks. I the proposed scheme, cell-edge users are served via CoMP trasmissios by multiple BSs, while cell-cetre users are served via MU-MIMO trasmissios by a sigle BS, cf. Fig. 1. We provide a semi-distributed resource allocatio scheme that maximizes the system eergy efficiecy while cosiderig both the total circuit power cosumptio ad the limited capacity of the backhaul etwork. We show that LAC scheme is promisig for improvig spectral efficiecy, eergy efficiecy, ad fairess of MIMO- OFDMA systems compared to covetioal schemes that use CoMP to serve all users i a cluster of cells.

2 A. Network Model II. SYSTEM MODEL We cosider a dowlik multiuser MIMO-OFDMA etwork cosistig of M BSs ad K mobile users. Each BS is equipped with A ateas while each user termial has a sigle receive atea. The BSs adopt a LAC scheme i which the complete set of users is split ito two o-overlappig subsets, amely, o-comp users ad CoMP users. CoMP users are served cooperatively by M ateas where, for simplicity, the atea with the best chael coditio from each BS is selected. O the other had, the o-comp users are served by MU-MIMO trasmissios from their earest BS, cf. Fig. 1. All BSs are assumed to commuicate via a capacity-limited mesh backhaul etwork to exchage CoMP users data. For a geeric user, the mode of operatio, i.e., o-comp or CoMP mode, is selected based o the average received SINR. I geeral, the distace of the user from her BS is a good idicator of her average received SINR. Therefore, the user is classified based o her distace from the earest BS ad a threshold radius r th. I other words, a geeric user is served by o-comp trasmissios if her distace to the earest BS is less tha r th, otherwise, the user is cosidered a CoMP user. We defie α as the ratio betwee the o-comp trasmissio coverage area ad the cell coverage area, i.e., α = r th R. B. Chael Model ad Capacity For iterferece coordiatio, the total badwidth B is partitioed ito two disjoit sets of subcarriers, F c ad F, for CoMP ad o-comp trasmissios, respectively. Note that ay subcarrier i F ca be reused by ay BS to serve its o-comp users while a subcarrier F c is used by all BSs to serve the same CoMP user. This spectrum partitioig, alog with the distace-based mode selectio, provides exclusio regios to protect both CoMP ad o-comp users. Let F c = N c, F = N, ad N c +N = N, the size of each set is determied dyamically ad proportioal to the total umber of CoMP ad o-comp users such that N = K i K N ad N c = N N where K i is the total umber of o-comp users i the system ad deotes the set cardiality. Complete CSI is assumed at each BS ad that the chael coheret time is greater tha or equal to the frame duratio. For a o-comp user k served by BS m i subcarrier F, there are two sources of iterferece, amely, i 1,k ad i,k. While i 1,k results from other o-comp users who share the same subcarrier i the same cell, i,k results from o-comp trasmissios i the other BSs. Hece, A i 1,k = p j,a h j,awj,as j, (1) i,k = M j S m j k A b=1 j S b b m p j,a h j,aw j,as j. () O the other had, i 3,k is the iterferece that affects CoMP trasmissios due to sharig the same subcarrier F c by multiple CoMP users. That is, M i 3,k = p j,m h j,mwj,ms j. (3) j S j k Note that, p k,a is the trasmit power to o-comp user k by atea a from its servig BS i subcarrier, while p k,m is the trasmit power to CoMP user k by BS m i subcarrier. h k,a is the total chael gai (i.e., icludig both small-scale fadig ad path-loss) of subcarrier betwee o-comp user k ad atea a from its servig BS. h k,m is the best chael gai of subcarrier betwee CoMP user k ad BS m amog the A ateas. wk,a ad w k,m are the precodig coefficiets used to perform beamformig i subcarrier for o-comp ad CoMP users, respectively. S m is the set of o-comp users who are served by BS m ad share subcarrier, while S is the set of CoMP users who share subcarrier. s k is the trasmitted symbol for user k i subcarrier. The received SINR at a user k i subcarrier is give by SINR k = A p k,a h k,a w k,a I1,k +I,k +σ z M p k,m h k,m w k,m I3,k +σ z, F,, F c, where σz is the variace of AWGN with zero mea ad I1,k, I,k, ad I 3,k are the iterferece powers based o (1), (), ad (3), respectively. Note that the umerator represets the useful coded sigal power received at user k i subcarrier. Usig (4), we ca express the chael capacity for user k i subcarrier as (4) C k = B N log (1 + SINR k). (5) The total system capacity ca therefore be obtaied as follows C tot = Ck F c k S } {{ } = C CoMP C. Power Cosumptio Model + M F k S m C k }{{} = C m o CoMP. (6) For the system described above, the total power cosumptio is give by where P tot = P BH + M P CoMP = η P m o CoMP = η ( PC + PCoMP m + Po CoMP m ), (7) p k,m F c k S A F k S m w k,m, (8) p k,a w k,a. (9) The power cosumptio model cosists of a liear part which is proportioal to the RF trasmit power, a costat part that icludes all sigal processig power, ad aother costat part P BH which represets the total power cosumptio i the backhaul etwork [8]. η 1 is the power amplifier iefficiecy costat ad P C is the sigal processig power per BS.

3 D. Eergy Efficiecy Metric We defie the system eergy efficiecy as the average umber of successfully trasmitted bits per eergy uit (bit/joule) or, equivaletly, the average data rate per power uit (bps/watt), i.e., EE = C tot P tot, (10) where C tot, ad P tot are give by (6) ad (7), respectively. III. PROBLEM FORMULATION The objective is to perform resource allocatio to maximize the eergy efficiecy of the etwork subject to backhaul capacity limitatios ad a total power budget. I this cotext, resource allocatio (RA) meas power allocatio P, precodig coefficiet allocatio W, ad subcarrier allocatio S. A. Problem Decompositio We propose a semi-distributed eergy-efficiet RA algorithm i which the problem is decomposed ito two phases. I the first phase, each BS performs RA idividually for its o-comp users, the it passes the values of Co CoMP m ad Po CoMP m to a cetral uit to proceed with the secod phase. I the secod phase, the cetral uit allocates resources for all the CoMP users. 1) RA for No-CoMP Users: I order to solve the RA problem for the o-comp users distributively, we igore the effect of i,k, i.e., I,k 0. This assumptio is reasoable due to the exclusio regio that separates the o-comp coverage regios, cf. Fig. 1. The, we have M differet optimizatio problems, oe problem per each BS. Our objective is to maximize the eergy efficiecy, however, it was show i [5] that whe the trasmit power budget is low, maximizig the spectral efficiecy leads to maximizig the eergy efficiecy. Therefore, we maximize Co CoMP m while limitig the total power available for the o-comp trasmissios. Limitig the trasmit power also reduces the effect of igorig i,k. For ay BS m, the RA problem for the o-comp users ca be writte as max C m P m,w m,s m o CoMP (11) s.t. A w k,a Pth, (1) F k S m p k,a S m A, F, (13) p k 0,, k, (14) where P th is a threshold to limit the total trasmit power used by BS m for o-comp trasmissios. Note that P th also cotrols the amout of iterferece itroduced to the other o-comp trasmissios at eighborig BSs. (13) guaratees that each subcarrier is reused by a maximum of A users for proper MU-MIMO trasmissios. ) RA for CoMP Users: For give Co CoMP m ad Po CoMP m, RA for CoMP users is performed i order to maximize the overall system eergy efficiecy defied i (10), ad hece the problem ca be writte as max EE (15) P,W,S s.t. w k,m PT P th, m, (16) p k,m F c k S Ck R max, m, (17) F c k S U m S M, F c, (18) Pk 0,, k, (19) where (16) is the total power budget costrait ad P T is the maximum allowable trasmit power by ay BS. (18) is to esure that o more tha M users are simultaeously selected to use ay subcarrier. R max is the maximum backhaul lik capacity ad U m is the set of users assiged to BS m. IV. SOLUTION METHODOLOGY The problems i (11) ad (15) are mixed-combiatorial o-covex which are computatioally huge ad it is ifeasible to obtai the optimal solutio. Therefore, to solve these problems sub-optimally, we adopt a three-step approach. I the first two steps, based o [9], a semi-orthogoal user selectio (SUS) scheme ad zero-forcig beamformig (ZFBF) are used to obtai S ad W, respectively. I the third step, for a give S ad W, the problem reduces to a power allocatio problem which ca be solved by differet methods. A. Zero-forcig Beamformig with Semi-orthogoal User Selectio (ZFBF-SUS) Scheme This scheme performs precodig coefficiet ad subcarrier allocatio at the same time. The mai idea of the scheme is to achieve the capacity of CoMP ad MU-MIMO by groupig the semi-orthogoal users to be simultaeously served o the same subcarrier; i.e., S, while cacelig out the iterferece by usig ZFBF; i.e., W. This scheme imposes a additioal costrait o the problems, (11) ad (15), such that the resultig set of users selected to trasmit o ay subcarrier are distict. I additio, this scheme satisfies (13) ad (18) where o more tha A, or M, users are allowed to share the same subcarrier. Aother beefit of ZFBF-SUS scheme is that it cacels out the iterferece betwee the CoMP users who share the same subcarrier, as well as, the o-comp users who share the same subcarrier at the same cell. Hece, I 1,k = I 3,k = 0. The algorithm ca be foud i [9] where it also has bee show that its performace is asymptotically optimal with low complexity. To decouple the power allocatio from the calculatio of ZFBF coefficiets, a additioal costrait is added to the problems, (11) ad (15). That is, o ay subcarrier, the trasmit powers to a certai CoMP (o-comp) user from the M BSs (A ateas) are the same, i.e., { p k,a = p k,a, a {1,,..., A} p k,m = p k,m, m {1,,..., M}. (0) Now, we ca rewrite the SINR as SINR k = A h k,a w k,a p k,a σ z M h k,m w k,m p k,m σz, F, F c. (1)

4 TABLE I. GREEDY ALGORITHM FOR NON-COMP TRANSMISSIONS TABLE II. DINKELBACH ITERATIVE ALGORITHM Power Allocatio for No-CoMP Trasmissios ( ) Defie: f(p, k, ) = B log N 1 + p γ k σz, p = P th L Iitializatio: Co CoMP m = 0, p k,a = 0, F c, k S m Repeat L times c k = f(p k,a+ p, k, ) f(p k,a, k, ), F, k S m {, k } = arg k, max c k p k,a = p k,a + p Co CoMP m = Co CoMP m + c k ed B. Power Allocatio Schemes 1) No-CoMP Trasmissios: We use Lagragia relaxatio L 1 (P, λ) of the primal problem (11) to obtai the dual problem as give by mi max λ P L 1 (P, λ), () where λ is the Lagragia multiplier of (1). KKT coditios are ecessary ad sufficiet to get the optimal solutio of the dual problem (). Therefore, we ca obtai a closed-form solutio for the power allocatio for give S m ad W m as p k,a = B Nλ l() σ z A h k,a w k,a +, (3) The solutio i (3) is i the form of multi-level water-fillig ad ca be solved iteratively usig the subgradiet method [10]. However, we propose a less complex greedy algorithm to solve the problem. The algorithm is summarized i Table I, where L is the umber of power levels which cotrols the precisio ad speed of the solutio. c k deotes the data rate gai achieved by assigig oe additioal power level to user k i subcarrier. The algorithm maximizes the capacity by addig oe more power level, p, to user k i subcarrier that has the maximum data rate gai c k. ) CoMP Trasmissios: For give W ad S, the objective fuctio i (15) is i a fractioal form ad o-covex i geeral; however, accordig to [11], it ca be trasformed ito a equivalet form with the same optimal decisio parameters, i.e., P, as stated i the followig theorem. Theorem 1. EE = max P C tot(p) P tot(p) if, ad oly if, C tot (P ) EE P tot (P ) = max P C tot(p) EE P tot (P) = 0. The, we ca rewrite the optimizatio problem (15) as max C tot EE P tot, (4) P s.t. (16), (17), (19). To obtai P, we use Dikelbach iterative method which is proved to coverge to the optimal value of EE whe the umerator is cocave ad the deomiator is covex. The proposed algorithm is summarized i Table II. To solve optimizatio problem (4) for a give EE, we also use Lagragia relaxatio L (P, λ, β) to obtai the dual problem as give by mi max λ,β P L (P, λ, β), (5) Resource Allocatio Based o Dikelbach Method Iitializatio: EE = 0, tolerace ɛ, maximum iteratios I max, iteratio idex i = 0, covergece = false while covergece = false ad i < I max solve optimizatio problem (4) for a give EE if C tot(p ) EE P tot(p ) < ɛ the covergece = true retur P P ad EE EE else EE C tot(p ) ad i i + 1 P tot (P ) ed if ed while TABLE III. SIMULATION PARAMETERS Parameter Value Iter-BS distace 500 m Total badwidth, B 1.5 MHz Total umber of subcarriers, N 18 Noise variace, σz 110 dbm Power amplifier iefficiecy costat, η 5 Sigal processig power per BS, P C 10 W Backhaul liks power cosumptio, P BH 6 15 W Backhaul lik maximum capacity, R max 40 Mbps where λ, β are the Lagragia multipliers of (16) ad (17), respectively. The closed-form solutio for the power allocatio for give S ad W is give by p k,m = B(1 β m) σ ( ) z N l() η EE+ M λ b wk,b M h k,b w k,b b=1 b=1 +, (6) Note that, (6) takes the form of multi-level water-fillig i which the Lagragia multipliers cotrols the power levels i.e., water levels, i order ot to violate ay of the costraits. I additio, EE excludes the iefficiet liks that have poor effective chael gais; i.e., M b=1 h k,b w k,b, from beig allocated power by trucatig their powers levels. V. SIMULATION RESULTS We evaluate the performace of the proposed LAC scheme via Mote Carlo simulatios i terms of eergy efficiecy (bps/watt), capacity (bps/hz/bs), i.e., spectral efficiecy, ad fairess. The proposed scheme is compared to the covetioal All-CoMP scheme i which BSs cooperate over all the subcarriers, i.e., α = 0. Uless otherwise stated, the system uder cosideratio cosists of M = 3 BSs ad serves K = 60 users uiformly distributed over D = [0, 1000]. Each BS has A = 3 ateas while each user termial has a sigle receive atea. i.i.d. Rayleigh fadig with uit variace alog with 3GPP urba path-loss model [1] are used to simulate the chaels. Table III summarizes the simulatio parameters. A. Performace of No-CoMP Trasmissio 1) Average Iterferece Power: Fig. shows the average iterferece power received by a geeric o-comp user, E[I,k ], ad the capacity of o-comp trasmissios as a fuctio of the total power budget for o-comp trasmissio P th. It is observed that P th cotrols the system behavior ad may lead to a oise-limited operatio, i.e., E[I,k ] < σ z, or a iterferece-limited operatio, i.e., E[I,k ] > σ z. Note that, the

5 Average iterferece power i o-comp trasmissio Thermal Noise Power Total Iterferece Power Foresee Average Capacity (I,k = 0) Actual Average Capacity Total power budget for o-comp trasmissio, P th Fig.. Average iterferece power at a geeric o-comp user ad capacity vs. total power budget P th for o-comp trasmissios. Eergy efficiecy of o-comp trasmissio (bps/watt).5 x Actual Eergy Efficiecy Foresee Eergy Efficiecy (I,k = 0) Total power budget for o-comp trasmissio, P th Fig. 3. Eergy efficiecy of o-comp trasmissio vs. total power budget P th for o-comp trasmissio. resource allocatio for o-comp trasmissios is performed distributively, hece, the iformatio about the iter-cell iterferece, I,k from eighborig BSs is ot available. That is, from the BS perspective, the foresee capacity icreases liearly with the power i the absece of I,k. While this is true for also the actual capacity i the oise-limited regio, the actual capacity saturates i the iterferece-limited regio ad the gap betwee the actual capacity ad the foresee capacity icreases with P th due to the liear icrease i the iterferece. ) Eergy Efficiecy: Fig. 3 illustrates the effect of icreasig P th o both the foresee eergy efficiecy ad the actual eergy efficiecy of o-comp trasmissios. Due to the uavailability of RA iformatio at other BSs, o average, the foresee eergy efficiecy is 9% higher tha the actual eergy efficiecy. Note that, this over estimatio error is sacrificed to decompose the problem ad solve it i a semi-distributed maer with less complexity. I the low power budget regime, Fig. 3 shows that the system maximizes the eergy efficiecy i additio to icreasig the capacity (Fig. ). O the other had, i the high power budget regime, maximizig the total capacity results i a reduced eergy efficiecy sice the capacity gai dimiishes with icreasig P th due to iterferece. Therefore, P th is chose to achieve maximum eergy efficiecy while limitig the error due to igorig the effect of I,k, e.g., P th = dbm i this case. Note that, icreasig the trasmit power beyod this value does ot cotribute much to the o- CoMP trasmissios capacity, cf. Fig Capacity of No-CoMP trasmissio (bps/hz/bs) Average umber of seved users Total Users, LAC No-CoMP Users, LAC CoMP Users, LAC Total Users, All-CoMP Fig. 4. Number of served users vs. the ratio betwee the o-comp coverage area ad the total coverage area of a cell, α (for K = 100). B. Overall System Performace 1) Fairess: As a idicator of fairess, Fig. 4 depicts the relatio betwee the ratio α ad the umber of served users for both LAC ad All-CoMP schemes. A user is said to be served if it is assiged at least oe trasmissio. The figure shows the improvemet gaied by LAC compared to All-CoMP scheme. Sice ZFBF-SUS is a greedy scheme, the umber of users served by All-CoMP scheme is relatively low. As α icreases, CoMP users who have high SINR with respect to their earest BS are offloaded to the o- CoMP trasmissios. Thus, users with less SINR have higher opportuity to be selected by ZFBF-SUS ad beig served by CoMP trasmissio; cosequetly, the fairess icreases. Note that, after a certai value of α, the umber of served CoMP users starts to fall due to the limited umber of cell-edge users. After the total umber of users reaches its maximum value, icreasig α icreases the umber of o-comp users ad decreases the umber of CoMP users by the same rate which maitais a costat total umber of served users. Based o the results i Fig. 4, α is chose such that the umber of served users is maximized while the loss i the umber of CoMP users is limited, e.g., α = 0.5. Compared to All-CoMP scheme i which 16.3% of the users are served, our proposed LAC scheme serves up to 78.43% of the users, which represets a 383% improvemet i the total umber of served users. ) System Eergy Efficiecy: Fig. 5 shows the system eergy efficiecy of both LAC ad All-CoMP schemes versus the total trasmit power budget P T. It also compares the performace of two differet objective fuctios; amely, Eergy Efficiecy Maximizatio (EEM) ad Spectral Efficiecy Maximizatio (SEM). While EEM is the scheme described above, SEM scheme maximizes the system capacity, C tot, subject to the same costraits (16)-(19). By comparig EEM to SEM, we ca observe that SEM is a greedy scheme sice its objective is to maximize the system capacity, ot the eergy efficiecy. That is, the more the available power, the more is the cosumed power regardless of the relative gai i the system capacity compared to the cost of icreasig the trasmit power. Therefore, the eergy efficiecy deteriorates i the high power budget regime. O the other had, EEM scheme teds to approach a costat performace to maitai higher eergy efficiecy sice the system does ot cosume more power after achievig the maximum eergy efficiecy.

6 System eergy efficiecy, EE (bps/watt) 4.5 x LAC, EEM LAC, SEM All-CoMP, EEM All-CoMP, SEM Total trasmit power budget, P T Fig. 5. System eergy efficiecy vs. total trasmit power budget, P T, P th = dbm ad α = 0.5. System capacity, C tot (bpb/hz/bs) LAC All-CoMP Total trasmit power budget, P T Fig. 6. System capacity vs. total trasmit power budget, P T, P th = dbm ad α = ) System Capacity: The results i Fig. 6 reiforce the results i Fig. 5 ad together explai the behavior of the system for both LAC ad All-CoMP schemes. For istace, sice the eergy efficiecy icreases with P T i the low power budget regime, cf. Fig. 5, the system teds to cosume all the available power. O the cotrary, i the high power budget regime, the system does ot cosume more power after achievig the maximum eergy efficiecy. Therefore, the system capacity icreases with the power i the low power budget regime, while it approaches a costat value i the high power budget regime as show i Fig. 6. Figs. 4-6 altogether clearly show that LAC outperforms All-CoMP i terms of eergy efficiecy, capacity, ad fairess. LAC scheme exploits the fact that users located withi a close proximity from oe of the BSs, i.e., cell-cetre users, will most likely experiece good SINR coditios. Although cacelig the low iter-cell iterferece for cell-cetre users via CoMP ca icrease the per-user performace, the system performace gai is egligible compared to the umber of userved users, system complexity, amout of used resources, ad system eergy efficiecy loss. The proposed scheme offloads cellcetre users from CoMP service to be served by the earest BS via a MU-MIMO trasmissio. Note that, the effective chael gai betwee a cell-cetre user ad her servig BS is better tha that betwee this user ad all the cooperatig BSs. Furthermore, to protect o-comp users from excessive itercell iterferece, LAC scheme cotrols the trasmit power of o-comp trasmissios P th to chage the size of exclusio regios accordigly. The aforemetioed factors explais the higher capacity (Fig. 6) ad the higher eergy efficiecy (Fig. 5) of the proposed scheme compared to All-CoMP scheme. VI. CONCLUSION We have proposed a ovel Locatio-Aware Cooperatio (LAC) scheme alog with a methodology to choose the system parameters. Based o the users SINR, the proposed scheme uses CoMP trasmissios oly for users with poor SINR coditios i order to reduce the system complexity (i.e., both computatioal ad sigalig complexity) ad icrease the eergy efficiecy. By servig the users with high SINR coditios through the o-comp (MU-MIMO) trasmissios, the opportuities for users with poor SINR to be selected by ZFBF-SUS for CoMP trasmissios icrease. After selectio of users mode of operatio, the problem is solved i a semi-distributed maer. Compared to All-CoMP scheme, LAC scheme better utilizes the available resources to provide higher capacity ad higher eergy efficiecy. I additio, the umber of users that LAC scheme ca support is much higher compared to the umber of users supported by All-CoMP scheme. I this way, it icreases the fairess amog users. ACKNOWLEDGMENT This work was supported by a Strategic Project Grat (STPGP 43085) from the Natural Scieces ad Egieerig Research Coucil of Caada (NSERC). REFERENCES [1] A. J. Paulraj, D. A. Gore, R. U. Nabar, ad H. Bolcskei, A overview of MIMO commuicatios - a key to gigabit wireless, Proc. IEEE, vol. 9, o., pp , 004. [] S. Vekatesa, A. Lozao, ad R. Valezuela, Network MIMO: Overcomig itercell iterferece i idoor wireless systems, i Cof. Rec. of IEEE 41st Asilomar Coferece o Sigals, Systems ad Computers (ACSSC), 007, pp [3] O. Somekh, O. Simeoe, Y. Bar-Ness, A. M. Haimovich, ad S. Shamai, Cooperative multicell zero-forcig beamformig i cellular dowlik chaels, IEEE Tras. If. Theory, vol. 55, o. 7, pp , 009. [4] G. Fettweis ad E. Zimmerma, ICT eergy cosumptio-treds ad challeges, i Proc. of the 11th Iteratioal Symposium o Wireless Persoal Multimedia Commuicatios (WPMC), vol., o. 4, 008. [5] D. W. K. Ng, E. S. Lo, ad R. Schober, Eergy-efficiet resource allocatio i multi-cell OFDMA systems with limited backhaul capacity, i IEEE Tras. Wireless Commu., vol.11, o.10, pp , 01. [6] Y. Shi, T. S. Kag, ad H. M. Kim, A efficiet resource allocatio for multiuser MIMO-OFDM systems with zero-forcig beamformer, i 007 IEEE 18th Iteratioal Symposium o Persoal, Idoor ad Mobile Radio Commuicatios (PIMRC), 007, pp [7] A. H. Sakr ad E. Hossai, Eergy-Efficiet Dowlik Trasmissio i Two-Tier Network MIMO OFDMA Networks, i Proc. of IEEE Iteratioal Coferece o Commuicatios 014 (ICC 14), Sydey, Australia, 014. [8] F. Richter, A. J. Fehske, ad G. P. Fettweis, Eergy efficiecy aspects of base statio deploymet strategies for cellular etworks, i IEEE 70th Vehicular Techology Coferece Fall (VTC), 009, pp [9] T. Yoo ad A. Goldsmith, O the optimality of multiatea broadcast schedulig usig zero-forcig beamformig, IEEE J. Select. Areas Commu., vol. 4, o. 3, pp , 006. [10] S. Boyd, L. Xiao, ad A. Mutapcic, Subgradiet methods, lecture otes of EE39o, Staford Uiversity, Autum Quarter, vol. 004, 003. [11] W. Dikelbach, O oliear fractioal programmig, Maagemet Sciece, vol. 13, o. 7, pp , [1] 3GPP, Further Advacemets for E-UTRAN physical layer aspects (Release 9), 3GPP, TR , Mar. 010.