Interference-Aware Subcarrier Allocation in a Correlated MIMO Downlink Transmission

Transcription

1 Rodiadee Nordin Interference-Aware Sbcarrier Allocation in a Correlated MIMO Downlink Tranmiion ROSDIADEE NORDIN Department of Electrical, Electronic and Sytem Engineering National Univerity of Malayia 43600, UKM Bangi, Selangor MALAYSIA adee@eng.km.my Abtract: - The demand for high data rate in wirele commnication ha increaed dramatically over the lat two decade. Mltiple-Inpt Mltiple-Otpt (MIMO) tranmiion i one of the efficient way to tranmit high data rate i by employing mltiple antenna configration at both the tranmitter and the receiver. One of the MIMO cheme, known a Spatial Mltiplexing (SM), relie on the linear independence data tream from different tranmit antenna to exploit the capacity from the fading channel. Coneently, SM ffer coniderably from the effect of patial correlation, th become the limiting factor in achieving the capacity benefit that SM offer. In an attempt to increae the robtne of the SM tranmiion for deployment in a wide range of correlated channel, the e of interference-aware bcarrier allocation i propoed. The Signalto-Interference-and-Noie Ratio (SINR) metric i ed a the performance metric to determine bcarrier ality which can then be tilized in the allocation proce in order to exploit the mltier diverity gain that can be offered in an Orthogonal Freency Diviion Mltiple Acce (OFDMA) ytem. From the imlation relt, it i hown that the propoed bcarrier allocation ha improved the BER performance of SM tranmiion in highly correlated channel environment. Key-Word: - Correlated channel, SINR, Interference, MIMO, OFDMA, Spatial mltiplexing, Spatial correlation, Sbcarrier allocation 1 Introdction Tranmiion over the wirele medim i the fndamental challenge. The tranmiion medim can be impaired by many factor, ch a bilding obtacle, noie and interference. OFDM i one of the effective mitigation technie to combat the channel impairment by converting the ingle carrier tranmiion to mlti-carrier tranmiion. The mltier verion of OFDM, known a OFDMA conit of mltiplexing different er in the time and freency domain by aigning bet of bcarrier to individal er, th allowing efficient and flexible reorce allocation acro freency band. OFDMA tranmiion can be frther enhanced with the addition of the mltiantenna technie, commonly referred to a MIMO. One of the MIMO cheme, known a Spatial Mltiplexing (SM), introdced by Fochini et al. [], aim to increae the pectral efficiency by tranmitting independent parallel data tream over mltiple antenna. SM i commonly implemented in the form of the V-BLAST architectre [3]. However, the mltiplexing gain i dependent on the nmber of tranmit-receive antenna, which are bjected to ncorrelated fading. In other word, the pectral efficiency that can be exploited in a MIMO cheme depend trongly on the tatitical behavior of the patial fading correlation, alo decribed by Gebert et al. [4] a the effect of elf-interference. SM cheme rely on the linear independence between the channel repone correponding to each pair of tranmit-receive antenna. A the patial correlation increae, cro-correlation will occr between the patial mltiplexing data tream. Coneently, SM cheme ffer coniderably from patial correlation, relting in ill-conditioned matrice, which can cae degradation of ytem capacity. In practice, the patial bchannel between different antenna are often correlated and therefore the fll potential mlti-antenna gain may not alway be obtainable. In rban wirele network, there are normally non-line-of-ight (NLOS) mltipath propagation bchannel between the bae tation and the er, a a relt of ignal reflection, diffraction and cattering from different obtacle in the propagation channel. However, the received ignal may till have a trong patial ignatre in the ene that tronger average ignal gain are received E-ISSN: Ie 4, Volme 11, April 01

2 Rodiadee Nordin from certain patial direction. Several factor determine the degree of patial correlation, ch a antenna element pacing [5], eparation between tranmitter and receiver [6], [7], array orientation [8] and mltipath anglar pread [9]. A the channel achieve fll correlation, the effective capacity gain of a MIMO channel i imilar to a SISO ytem, th diminihed the capacity benefit offered by MIMO ytem if the effect of elf-interference i not properly mitigated. Thi paper preent a novel bcarrier allocation cheme, where the allocation cheme take advantage of independent channel variation acro er to improve the network performance throgh freency and patial diverity. The performance of the propoed algorithm i analyzed when operating in different correlated channel environment. The ret of the paper i organied a follow. The effect of elf-interference and related work to redce the effect are preented in Section II. The problem formlation for interference-aware bcarrier allocation in OFDMA network i preented in Section III. In thi ection, an interference-aware bcarrier allocation cheme that conider the effect of patial correlation in a MIMO channel i propoed. In Section IV, ytem etp and imlation parameter are preented. In Section V, the performance of the propoed allocation cheme i frther invetigated and analyed in different correlation environment. Section VI provide the conclion drawn from thi paper, whereby important analyi and reearch finding will be mmarized and preented. MIMO and Self-Interference.1 MIMO for Capacity Increae MIMO ytem tranmit different ignal from each tranmit antenna o that the receiving antenna array receive a perpoition of all the tranmitted ignal. Theoretically, increaing the nmber of tranmit antenna allow more data to be tranmitted. Capacity in MIMO i baed on a aitatic analyi where the channel varie randomly from brt to brt. Althogh the capacity i derived for ingle er cae, in mot relt, it can be applied to mltier ytem with receiver diverity: C MIMO = log det I N SNR + HH * (1) r N t where H i the normalied channel gain matrix, SNR i the average SNR at any Rx antenna, (*) i the tranpoe-conjgate. It amed that N t tranmiion come from eal power ncorrelated orce in (1), which repreent the ideal cae where N t ignal component are received by a eparate et of N r in a manner where each received ignal component ha no patial interference to each other. From (1), it can be een that the capacity increae linearly with min(n t, N r ). Other than that, larger MIMO capacity can be obtained from larger rank and eigenvale of HH*. Spatial correlation propertie of the radio channel are the key to MIMO capacity. Theoretically, an SM cheme can achieve it maximm capacity, provided with proper detection and error correction coding, however, the capacity can be redced when the patial bchannel of the MIMO channel give rie to a high correlation cenario. The impact of patial correlation on the capacity of MIMO ytem can be analyed throgh the eigenvale of HH * from the MIMO capacity, decribed mathematically in (1). The eigenvale reveal the degree of independence between channel or the effective nmber of patial bchannel. Alternatively, the capacity eation can be preented in following form [10]: C SNR log det Ι N + R () r N t MIMO = MIMO where R MIMO i the normalied channel correlation matrix, r 1, whoe component are given by r ij = k ik ij h h * where i i for receive antenna and j i jk for tranmit antenna. The capacity eation take into accont the effect of patial correlation at both tranmit and receive end. The effect of capacity lo de to patial correlation can be illtrated in Fig. 1, whereby the channel capacity decreae btantially a the correlation coefficient increae. It can be een that the decreae in capacity alo can occr even thogh the correlation i high at either the bae tation (BS) or the mobile tation (MS). The effect i more evere when both of the element hare the ame correlation coefficient. It i alo oberved that the channel capacity decreae btantially a R MIMO > 0.5, which agree well with the MIMO channel mearement performed by Martin et al. [11] and known relt on the patial diverity technie by Jake [1]. E-ISSN: Ie 4, Volme 11, April 01

3 Rodiadee Nordin capacity (bp/hz) =0.0,R MS =0.0 =0.5,R MS =0.5 =0.0,R MS =0.9 =0.9,R MS =0.0 =0.9,R MS =0.9 =1.0,R MS = Signal to Noie Ratio (SNR), in db Fig. 1: Capacity lo de to elf-interference in a MIMO. Correlation and Self-Interference Previo work [4], [8], [13], and [14] ha ggeted that the exitence of correlation among antenna element i one of the main challenge in MIMO implementation. The potential capacity gain i highly dependent on the mltipath richne, ince a flly correlated MIMO channel offer only ingle effective channel over a SISO ytem,, a hown by Shi et al. [8]. While a flly decorrelated channel offer mltiple capacity benefit, which increae linearly a the nmber of antenna increae. SM cheme rely on the linear independence between the channel repone correponding to each tranmit antenna. Coneently, ch cheme ffer coniderably from fading correlation, relting in an ill-conditioned MIMO channel matrix [4]. Thi can cae degradation of ytem capacity, whereby retranmiion and chae combining do not improve the BER performance. A the correlation increae between the patial bchannel, cro-path from the adjacent tream of tranmit and/or receive antenna will occr, relting in effect known a elf-interference in Gebert et al. [4]. For conitency, the term elfinterference will be ed to decribe thi effect throghot the remainder of thi paper. In a highly correlated environment, the imltaneo tranmiion of independent data ymbol from mltiple antenna i perceived a imilar ymbol at the receiver and hence the receiver incapable of performing effective ymbol detection. Spatial correlation in a MIMO tranmiion occr de to a variety of factor, ch a infficient antenna eparation, mall cattering angle, alo known a anglar pread (AS) and angle of arrival (AoA) [4]. Lee [5] ggeted deigning the reired antenna pacing baed on the degree of correlation. Hi experiment ha hown that a BS with two tranmit antenna reire interelement pacing of 100λ to achieve correlation coefficient of zero (ncorrelated), provided the incoming wave angle i more than 60 o. Other than that, the deigner ha to enre that the angle a wide a poible, whereby AoA and AS of 90 O i deired to achieve decorrelation. However thi i not poible de to environmental concern, afety ie and limited pace for antenna intallation at the BS, which i crcial in rban development where the potential of MIMO capacity can be maximized. In addition to antenna deign, the height of the bae tation antenna, the condition of ignal path clearance and catterer rronding the bae tation have alo been identified a key factor contribting toward the effect of elf-interference a identified by Catrex et al. [13]. Thee factor can cae the channel matrix to be near rank one, i.e. non-invertible, th caing the abence of mltipath, which i an eential criterion to make patial mltiplexing work. Any correlation preent at the tranmitter effectively increae the linear dependence of the inpt tream repone and make tream detection and decoding a difficlt tak for the receiver. A a relt, SM failed to adapt and extract the nonzero capacity that i preent in the highly correlated channel. Therefore, deigning an appropriate receiver that can adjt moothly to any level of correlation i deired..3 Previo Work The impact of high patial correlation can be avoided by implementing effective MIMO antenna deign technie. Antenna eparation i one of the way to achieve decorrelation by placing the antenna element far away from each other. However, thi i not alway effective de to environmental effect. For example, the keyhole effect may limit the effectivene of thi method, a ggeted by Chizhik et al. in [14]. Other than antenna eparation, antenna array with orthogonal radiation polarization or pattern can ignificantly redce the patial correlation. Orthogonality can be achieved throgh angle, pace or polarization diverity. Thi allow the antenna element to ndergo independent cattering, relting in lower patial correlation a proven from the relt pblihed by Ramirez and De Flavii in [15] even when the channel patial correlation i high. However, thi method can cae array blindne and redction of antenna effective gain, E-ISSN: Ie 4, Volme 11, April 01

4 Rodiadee Nordin which can ltimately redce the ytem capacity. Hence, hardware deign become a challenging and complicated proce while efficient adaptive algorithm applied in oftware become a preferable choice to develop effective elf-interference mitigation technie. A power allocation trategy i an example of ing adaptive algorithm to mitigate the effect of elf-interference. In optimm power allocation, the CSI i known at the tranmitter. The CSI information can be ed to obtain the inglar vale decompoition (SVD) to identify effective independent channel. The inglar vale are ed a deciion criteria to determine the power allocation ch that parallel patial channel correponding to higher inglar vale are aigned more power, while no power i allocated to the patial interferer relting from the low inglar vale de to high correlation. Ivral et al. [16] introdced an adaptive algorithm, by combining antenna election and power allocation trategy in highly correlated channel. Intead of ing the CSI a the feedback information, the correlation matrix i ed a the performance metric a it i hown that correlation matrix overhead i mch le compared to eigenvale from the SVD method, relting in lower feedback reirement and th a fater allocation proce. From the correlation matrix, the tranmitter applie tochatic water-filling power allocation to get the maximm capacity from the correlated channel. However, achieving optimal bit aignment become the main challenge to realie the water-filling allocation, ince modlation technie are dicrete and water-filling reire contino bit allocation to btream, with more bit tranmitted to the dominant eigenmode. Thi can be overcome by implementing adaptive modlation and coding cheme. Another antenna election trategy i propoed by Gore et al. [17]. The algorithm chooe the optimm bet of tranmit and receive antenna baed on minimm ymbol error rate. It wa proven that the election algorithm enhanced the SM performance in correlated environment, bt at the expene of a lo in data rate. Akhtar and Gebert [18] propoed the e of contellation mltiplexing (CM) to minimie patial interference. The main key in CM tranmiion i from the e of power caling, which i achieved by caling down the deired M-QAM contellation ize by the caling factor from the original M-QAM ize. Spatial bchannel are differentiated throgh power caling rather than throgh patial ignatre which i ed in conventional SM tranmiion. For example in a flly correlated channel, a 16-QAM ignal i the perpoition of two 4-QAM contellation ignal, with the econd 4-QAM contellation point being centred at the contellation point of the firt 4-QAM point, a hown in Fig.. CM reire only a ingle tranmit antenna to end everal independent data tream from a given modlation cheme, provided it i power-caled to form a higher-order modlation contellation. By tranmitting in ch manner, CM doe not rely on a fll rank channel matrix to fnction properly. Hence, ch cheme yield a ratepreerving MIMO mltiplexing cheme that can operate robtly at any degree of patial correlation. Fig. : Two perpoed 4-QAM contellation, 1 and that yield the eivalent 16-QAM contellation after caled down by power of ¼ [18]. 3 Self-Interference Mitigation Strategy 3. SINR Metric In thi ection, a mathematical definition of the SINR metric i preented. SINR metric ha knowledge of the channel ality at every bcarrier level, whereby it indicate the patial information and elf-interference caed by the mimatch between the patial bchannel. The mathematical model for the received ignal in a SM-OFDMA ytem, after FFT and gard removal i decribed a follow: t Y = H X + N (3) where the bcript denote the MS index, denote the bcarrier index, H i the channel matrix containing MS freency repone of the channel between N t tranmit and N r receive E-ISSN: Ie 4, Volme 11, April 01

5 Rodiadee Nordin antenna at bcarrier and applied to the bcarrier of the OFDMA ignal on a clter bai, while N denote AWGN noie and X t denote N t 1 matrix containing tranmit ignal. At the receiver, the OFDMA ytem adopt a linear MMSE configration: H ( ) ( ) 1 H H SNR I H 1 G = + (4) The MMSE filter ha the ability to mitigate elfinterference whilt not adverely amplifying the received noie. The MMSE filter i alo able to eparate the patial bchannel of the SM trctre [19]. A a relt, different patial bchannel can be allocated to different er to achieve additional patial mltier diverity gain. On the other hand, thi typically increae the amont of feedback by the nmber of patial bchannel. The received ignal i mltiplied by an MMSE filter given by E.(4): Y Y t G = G X + G N (5) The receiver then calclate the SINR per bcarrier and feed back that information to the tranmitter. The SINR metric captre the variation of the elfinterference within the freency domain. The MS compte the SINR at every patial bchannel for every bcarrier [19] (the bcarrier index,, i omitted for implicity of preentation): ESINR = ( G H ) ( G H ) ε + G + G N j, j ε j j (6) where i the patial bchannel conidered in the allocation algorithm. In the cae of SISO ytem, = Q = N r = 1. ε denote the average ymbol energy and. j denote the element located in row and colmn j. The SINR metric aim to compte elfinterference with knowledge from the data tream component,. and elf-interference component,. j,j from the adjacent tranmitted data tream, alo known a patial interferer. 3.3 Interference-Aware Sbcarrier Allocation Scheme In OFDMA ytem, with er fading independently, there i likely to be a er with a good channel at any given time lot or b-carrier. By chedling tranmiion o that er tranmit at time and freency reorce when their channel condition are favorable, ignificant gain from the effective channel can be achieved. If a determinitic rather than random allocation of bcarrier i employed, the mltier diverity can be exploited to enre mot er can be allocated the bet bcarrier for them with minimal clahe. A a relt, ytem capacity can be maximized by alway erving the er with the tronget channel and the capacity can be increaed a the nmber of er i increaed. Initially, DSA work ha foced on SISO ytem [0]. Sbeently, the reearch foc ha been expanded to MIMO cheme by Y. Peng [1], whereby everal variation of the algorithm (referred to a cheme ) were introdced. Each cheme correpond to different patial correlation condition. Channel gain wa ed a a metric to determine the bcarrier allocation proce. The firt cheme, referred to here a DSA-Scheme 1, performed bcarrier allocation eparately for each patial bchannel and it take no conideration of elf-interference and relation of one patial bchannel to the other adjacent bchannel. An extenion to DSA-Scheme 1, known a DSA- Scheme 5 aim to combat the effect of elfinterference in a correlated environment. Thi cheme conider the previoly allocated bcarrier from the previo patial bchannel by mean of iteration. The nmber of near adjacent bcarrier avoided i meared by parameter l. From the imlation relt, it i hown that when l=10, ignificantly enhanced BER gain can be obtained in extremely correlated environment. The propoed interference-aware bcarrier allocation cheme can be illtrated in Fig. 3. Intead of ing the channel gain a the performance metric, the allocation cheme e two different performance metric to determine the bcarrier allocation at each of the patial bchannel: (i) at the parallel bchannel, ESINR metric i ed to identify the bcarrier that are affected by elfinterference, and (ii) the patial interferer e the channel gain a a metric to determine the bcarrier allocation. The bcarrier allocation i applied independently acro patial bchannel. By doing the allocation in thi manner, er are prevented from haring the bcarrier with the patial interferer, th helping to avoid the e of an extra coding cheme and redcing the complexity of the algorithm. The combination of different performance metric in the propoed bcarrier allocation i expected to redce the effect of elfinterference, whilt till eeking a maximal allocation of channel energy. E-ISSN: Ie 4, Volme 11, April 01

6 Rodiadee Nordin Fig. 1: Performance metric of the propoed interference-aware bcarrier allocation Thi alo enre that patial diverity cannot be lot de to highly correlated patial bchannel ince each of the patial bchannel will be allocated with different et of bcarrier. Before the allocation take place, the -th MS compte the MMSE filter, (4), and mltiplie the received ignal by G k, (5). The -th mobile tation then compte the SINR, of the -th patial bchannel, (6). The -th MS calclate each SINR metric for each patial bchannel and then end them back throgh the feedback channel to the bae tation. The bae tation e the feedback information a an inpt for the allocation algorithm. The nomenclatre are et firt a reference. In the following algorithm, ={1,,Q} repreent patial bchannel conidered for the allocation algorithm. P and E repreent the average received power gain and ESINR metric for er at th patial bchannel, U i the total nmber of er, S i an Q by N matrix where each row i a vector containing the indice of the eable bcarrier for the particlar patial bchannel, i.e. N ={1,,N b } where N b i the total nmber of eable h, k bcarrier. The and SINR, i the channel repone and ESINR metric for er at bcarrier and patial bchannel and C, i a matrix to tore the bcarrier indice (bcarrier location) of the allocated bcarrier for er and bcarrier. The interference-aware bcarrier allocation algorithm can be decribed a follow: (1) Initialization Set P, = 0 for all er, =1,,U; Set C,, =0 for all er =1,,U and patial bchannel ={1,,,Q} ; Set =1 () Main proce While N 0 Nb (a.) Generate hort lit of er tart with er with mall SINR metric 1 and channel gain. Find er atifying: Main: E E i for all i, 1 i U Interferer: P P i (b.) For the er in (a), find bcarrier atifying: Main: SINR, SINR, j for all j S Interferer: h, h, (c.) Update SINR,, j P, N and C, with and in (b) according to: E = E + SINR, (for main patial bchannel) P = P + h, (for interferer channel) N = N - n C, = n = +1 where C, i the allocation matrix to record the allocated bcarrier, for er. (d.) Go to the next er in the hort lit in (a), repeat (b) to (c) ntil all er are allocated another bcarrier, N 0. } The algorithm rank er from the lowet to highet SINR metric at each main patial bchannel. Coneently, the next bet bcarrier are allocated to er in rank order, allowing er with lowet SINR, i.e. er that ffer from evere elf-interference effect or poor received ignal at that particlar patial bchannel to have the next bet bcarrier with highet SINR metric that i available for the next tranmiion. 4 Simlation Environment 4.1 SM-OFDMA Deign Parameter The imlation i performed in an SM-OFDMA downlink environment. Modlation type affect both the data capacity in a given channel a well a the robtne with regard to noie and interference. The OFDM parameter and the conidered Qadratre Phae Shift Keying (QPSK), ½ rate modlation and coding cheme (MCS) are mmaried in Table 1 and Table and ed throghot thi paper. 1 For the firt iteration, ame all er have eal SINR metric a no bcarrier have been allocated; hence, the lit may be entirely arbitrary. E-ISSN: Ie 4, Volme 11, April 01

7 Rodiadee Nordin Table 1: Phyical layer parameter for OFDM ytem Parameter Vale Operating Freency 5 GHz Available Bandwidth (BW) 100 MHz Tranmit Information 10 n Dration (T tx ) FFT ize (N FFT ) 104 Ueable bcarrier (N b ) 768 Sbcarrier pacing ( f ) khz Uefl Symbol dration (T) 10.4 µ Gard Interval (GI) 176 Total Symbol Dration (T ) 1.00 µ Table : QPSK, ½ rate modlation cheme for the conidered SM-OFDMA ytem Modlation QPSK Coding Rate ½ Data bit per OFDM ymbol 1536 Coded bit per OFDM ymbol 307 Total Bit per OFDM ymbol 048 Nominal Bit Rate [Mbp] 18 Total Bit Rate [Mbp] The ideal wirele network are characteried by their ability to operate in low SNR condition bt till offering high capacity with efficient pectrm tiliation. Therefore, itable channel model are propoed to invetigate the appropriate ytem. Both of the channel are adopted from two tandard: (1) ETSI BRAN Channel E [] and () SCM Urban Micro [3]. Channel E correpond to a pico-cell type large open pace environment with NLOS condition, while Urban Micro environment repreent a very mall cell in an ltra high denity rban area with cell radi of approximately le than 500m and BS antenna located at rooftop level. The MIMO channel model are imlated in a fixed channel matrix o that both receiver and tranmitter are tatic with repect to each other and the path lo remain approximately contant dring the mearement dration. Other than that, both channel model repreent different mltipath environment with different tranmiion bandwidth, which i relevant for comparion prpoe in the imlation work throghot thi paper. The key parameter for both channel model are mmaried in Table 3. A packet ize of 54 byte i conidered throghot thi paper. In addition to that, 000 i.i.d ai-tatic Rayleigh ditribted channel ample per OFDM ymbol are ed in each imlation to Table 3: Channel Model Parameter Parameter HIPERLAN SCM Urban E Micro Environment Large open Otdoor rban pace NLOS NLOS Bandwidth 100 MHz 5 MHz Exce Delay Spread 1760 n 93 n Mean Delay Spread 50 n 51 n Carrier Freency 5 GHz GHz achieve table averaging over wide fading channel. 16 er are conidered to exploit the mltier diverity gain with a total of 768 eable bcarrier to be eally hared among the er with FFT ize, N FFT = 104. A antenna configration i ed for the SM cheme. It i amed that the BS ha perfect knowledge of the channel tranfer fnction for thoe bcarrier that have been allocated to it and thi i later ed for ealiation and decoding prpoe. 4. Correlation Coefficient The patial correlation matrix of the MIMO channel can be derived from the Kronecker model [8], a well-known correlation analytical model. Spatial correlation matrice between BS and MS, are given by: R MIMO = R MS (7) where repreent the Kronecker prodct, and R MS and are N r N r and N t N t matrice correponding to tranmit and receive covariance matrice, which can be frther defined a ymmetrical complex correlation matrice a follow: ρ 11 ρ1 L ρ1nt ρ 1 ρ L ρ N t = (8) M M O M ρ Nt 1 ρ Nt L ρ Nt N t N r1 Nr Nr Nr Nt Nt ρ11 ρ1 L ρ1nr ρ 1 ρ L ρ N r R MS = (9) M ρ ρ M O L ρ M Nr Nr (z) where ρ ij i the tranmiion correlation coefficient: ij N N N N ρ = h, h (10) t r t r E-ISSN: Ie 4, Volme 11, April 01

8 Rodiadee Nordin where 0 < ρ ij < 1 and a, b compte the correlation coefficient between a and b. Eation (7) ame that the receive and tranmit antenna are correlated independently on each other and ame there are local catterer near both antenna element. The Kronecker model allow eparate optimiation at both link end, o that receive and tranmit correlation can be dealt with eparately ince the correlation propertie at both tranmit and receive antenna are independent from each other. Relt pblihed in [4] hown that Kronecker model i phyically reaonable to e when the correlation between the adjacent antenna i larger than the correlation between non-adjacent antenna. Tranmit Power (db) Sbcarrier Fig. : Example of ncorrelated channel at every patial bchannel for SCM Urban Micro 5 h 11 h 1 h 1 h Correlated Channel Model The main aim of thi paper i to invetigate the SM performance at different level of patial correlation: between ideal to extreme correlation cenario. In the ideal cae, the channel i ncorrelated and the effect of elf-interference i very minimal, while flly correlated channel repreent the wort cae cenario, whereby the effective capacity gain i eal to a SISO ytem. The propoed correlation cenario a mmaried in Table 4. The correlation cae are derived baed on practical wirele channel mearement from previo work a pblihed in [5], [6] and [7]. Table 4: Correlation Mode and It Coefficient Correlation Correlation Mode Coefficient, R MIMO R MS Fll Uncorrelated Fig. 4 how all the patial bchannel in the ncorrelated MIMO channel. It i oberved that each patial bchannel i randomly ditribted acro the bcarrier, which gget that there i negligible inflence from the other tranmitting patial bchannel (interferer) dring the tranmiion. Fig. 5 i the example of the channel gain when it ffer from flly correlated channel. It can be oberved that the pectral hape and amplitde of channel gain at every patial bchannel i almot identical. Thi i when the effect of elf-interference dominate the MIMO performance and the effective channel i imilar to SISO. Tranmit Power (db) Sbcarrier Fig. 3: Example of Flly correlated channel at every patial bchannel for SCM Urban Micro 5 Simlation Relt In Fig. 6, the performance of the propoed interference-aware bcarrier allocation i compared againt DSA-Scheme 1 and DSA-Scheme 5, decribed earlier in Section 3.3. From the relt, it i hown that the interference-aware allocation ha better advantage in term of BER gain, compared to DSA-Scheme 1 and DSA-Scheme 5 when imlated in both ncorrelated channel. Thi relt alo gget that the ncorrelated channel model carrie ome degree of elfinterference, relting in lo of BER gain. Thi can be confirmed from the mearement of correlation coefficient, R MIMO between the parallel channel and the patial interferer of the generated channel model (Table 5). The correlation coefficient for both channel model are conidered a moderate nmber, th acceptable in practical implementation a a MIMO channel i expected to generate ome amont of patial correlation, epecially in a poplated rban environment h 11 h 1 h 1 h E-ISSN: Ie 4, Volme 11, April 01

9 Rodiadee Nordin Table 5: Correlation coefficient mearement for ncorrelated channel Correlation Channel Model Coefficient, R MIMO R MS ETSI SCM In Fig. 7, the performance of the propoed interference-aware bcarrier allocation cheme i compared againt DSA-Scheme 1 and DSA-Scheme 5 in a Flly correlated channel environment. From the relt, the benefit of interference-aware allocation i apparent in a flly correlated channel. In both channel model, the interference-aware allocation cheme hare almot imilar BER performance to that of the DSA-Scheme 5. The margin of BER difference i increaed ignificantly a the SNR increae, particlarly beyond 10 db, which implie the ability of interference-aware allocation to minimie elf-interference in a highly correlated channel, while DSA-Scheme 5 ffer from propagation error a the SNR increae. DSA- Scheme 5 performance i alo achieved an error floor a the SNR increae when imlated in ncorrelated Urban Micro channel, a hown in Fig. 7(b). Thi can be explained from the deign of the algorithm which relie on the parameter l. Peng [1] ha ggeted that l = 10 i the optimal ditance in Channel E bt for the cae of Urban Micro, the optimal length, l = 10 no longer valid ince the patial correlation reirement change according to different type of channel. Other than that, DSA- Scheme 5 acrifice greater degree from the election of the bet available bcarrier in preference for avoiding allocation of the ame or nearby bcarrier on different patial bchannel, which depend on the ize of l. In other word, there i a trade-off between mitigating elf-interference and achieving diverity gain in DSA-Scheme 5. Thi i hown in Fig. 7(b), where BER crve for DSA- Scheme 5 experience error floor, while interferenceaware allocation offered higher BER gain compared to DSA-Scheme 5 a the modlation order and patial correlation increaed for both type of channel model. The dynamic combination between SINR metric in parallel channel and channel gain metric at patial interferer make the interference-aware bcarrier allocation an attractive candidate to combat the debilitating effect of elf-interference, compared to the DSA-Scheme 1 cheme. Bit Error Rate (BER) Bit Error Rate (BER) DSA-Sch1 DSA-Sch5 SINR Signal-to-Noie Ratio (SNR) in db (a) ETSI HiperLAN E DSA-Sch1 DSA-Sch 5 SINR Signal-to-Noie Ratio (SNR) in db (b) SCM Urban Micro Fig. 6: BER comparion in ncorrelated channel Thi can be explained by oberving the allocated bcarrier for a er, at a particlar time ample in a flly correlated channel between one of the parallel channel and it adjacent patial interferer, a hown in Fig. 8. In DSA-Scheme 1, the allocated bcarrier at the patial interferer are almot the ame a to thoe allocated in the parallel channel (orce), which cae the cheme to be nable to differentiate between bcarrier that are everely affected by elf-interference and thoe bcarrier that till can offer good channel ality in a flly correlated channel. In the interference-aware allocation cheme, the SINR metric at the parallel channel avoid the allocation of imilar bcarrier that are affected by elf-interference while not compromiing the bet channel ality given from the channel gain metric at the interfering channel. Thi relt in improved BER performance, a hown in both Fig. 6 and Fig. 7. Other than that, the interference-aware allocation cheme ha lower complexity compared to the DSA-Scheme 5 cheme, ince it only e the patial parallel bchannel to determine the elfinterference caed by the patial interferer, while the DSA-Scheme 5 need to perform the iteration at every patial bchannel to conider the E-ISSN: Ie 4, Volme 11, April 01

10 Rodiadee Nordin previoly allocated bcarrier at the previo patial bchannel. In other word, interferenceaware allocation offer lower allocation complexity with extra BER gain compared to DSA-Scheme 5. Bit Error Rate (BER) DSA-Sch1 DSA-Sch5 SINR Signal-to-Noie Ratio (SNR) in db Bit Error Rate (BER) (a) ETSI HiperLAN E DSA-Sch1 DSA-Sch5 SINR Signal-to-Noie Ratio (SNR) in db (b) SCM Urban Micro Fig. 7: BER comparion in flly correlated channel Channel Gain (db) Sorce Interferer SINR Metric (Sorce) -15 SINR Metric (Interferer) Alloc Sb (SINR) Alloc Sb (Ch. Gain) Sbcarrier Fig. 8: Example of bcarrier allocation in a Flly correlated channel 6 Conclion Thi paper addree the problem of SM downlink tranmiion over patially correlated fading channel from the apect of bcarrier allocation. From the nmerical imlation and analyi, the propoed interference-aware bcarrier allocation ha hown to offer low complexity and to achieve btantial BER gain when operating nder divere patial correlation cae. A the patial correlation increaed in the channel, the allocation cheme were hown to redce the effect of elf-interference, particlarly interference-aware allocation, which had perior BER performance compared to the other conidered allocation cheme, namely DSA-Scheme 1 and DSA-Scheme 5. Error analyi alo revealed the limitation of DSA-Scheme 5 in mitigating the effect of elf-interference. The nit of eparation on the next bcarrier parameter, denoted by l, i not flexible and nable to adapt to different type correlation cae and channel model type. In general, it can be conclded that the propoed algorithm contribte toward the goal of achieving high-peed, yet reliable SM tranmiion for ftre wirele network. Acknowledgement The athor wold like to thank Univeriti Kebangaan Malayia for their financial pport, nder grant cheme nmber UKM-GGPM-ICT Reference: [1] S. Alamoti, A imple tranmit diverity technie for wirele commnication, IEEE Jornal on Selected Area in Commnication (JSAC), Vol. 16, No. 8, pp , Oct [] G.J. Fochini, G.D. Golden, R.A. Valenzela, and P.W. Wolnianky, Simplified Proceing for High Spectral Efficiency Wirele Commnication Employing Mlti-Element Array, IEEE Tranaction on Selected Area in Commnication, Vol. 17, No. 11, pp , Nov [3] G.D. Golden, G.J. Fochini, R.A. Valenzela, and P.W. Wolniaky. Detection algorithm and initial laboratory relt ing the V-BLAST pace-time commnication architectre, Electronic Letter, Vol. 35, No. 1, pp , Jan [4] D. Gebert, M. Shafi, D.S Shi, P.J. Smith, and A. Nagib. From theory to practice: an overview of MIMO pace-time coded wirele E-ISSN: Ie 4, Volme 11, April 01

11 Rodiadee Nordin ytem, Ttorial paper. IEEE Jornal on Selected Area in Commnication (JSAC), Vol. 1, No. 3, pp , Apr [5] W. Lee, "Effect on Correlation between Two Mobile Radio Bae-Station Antenna," IEEE Tranaction on Commnication, Vol.1, No.11, pp , Nov [6] P. Kyriti, D. C. Cox, R. A. Valenzela, and P. W. Wolnianky, Correlation analyi baed on MIMO channel mearement in an indoor environment, IEEE Jornal on Selected Area in Commnication (JSAC), Vol. 1, No. 5, pp , Jne 003. [7] M. Chamchoy, S. Promwong, P. Tangtianon, J. Takada, "Spatial correlation propertie of mltiantenna UWB ytem for in-home cenario," IEEE International Sympoim on Commnication and Information Technology, 004. ISCIT 004, Vol., pp , Oct [8] D.-S. Shi, G. J. Fochini, M. J. Gan, and J. M. Kahn, Fading correlation and it effect on the capacity of mltielement antenna ytem, IEEE Tranaction on Selected Area in Commnication, Vol. 48, No. 3, Mar [9] A. Intarapanich, P. L. Kafle, R. J. Davie, A. B. Seay, and J. McRory, Spatial correlation mearement for broadband MIMO wirele channel, IEEE 60 th Vehiclar Technology Conference, 004. VTC004-Fall, Vol.1, pp. 5-56, Sept [10] G. J. Fochini, Layered Space-Time Architectre for Wirele Commnication in a Fading Environment when Uing Mltielement Antenna, Bell Lab Tech. Jornal, pp , Atmn [11] C.C. Martin, J.H. Winter, and N.S. Sollenberger, Mltiple-Inpt Mltiple-Otpt (MIMO) Radio Channel Mearement, IEEE VTC 000 Fall Conference, Sept , Boton, USA. [1] Jake, W.C. Jr., Microwave Mobile Commnication, John Wiley and Son, New York, [13] S. Catrex, P. F. Drieen, and L. J. Greentein, Attainable throghpt of an interference-limited mltiple-inpt mltipleotpt (MIMO) celllar ytem, IEEE Tranaction on Commnication, Vol. 49, No. 8, pp , Ag [14] D. Chizhik, G.J. Fochini, M.J. Gan, R.A. Valenzela, "Keyhole, correlation and capacitie of mlti-element tranmit and receive antenna", IEEE 53 rd Vehiclar Technology Conference, 001. VTC 001- Spring, Vol. 1, pp , 001. [15] R.R. Ramirez and F. De Flavii, "A mtal copling tdy of linear and circlar polarized microtrip antenna for diverity wirele ytem", IEEE Tranaction on Antenna and Propagation, Vol. 51, No., pp , Feb [16] M.T. Ivrlac, W. Utchick, J.A. Noek, "Fading correlation in wirele MIMO commnication ytem", IEEE Jornal on Selected Area in Commnication, Vol. 1, No. 5, pp , Jne 003. [17] D. Gore, R. Heath, A. Palraj, "Statitical antenna election for patial mltiplexing ytem", IEEE International Conference on Commnication, 00. ICC 00, Vol. 1, pp , May 00. [18] J. Akhtar, D. Gebert, "A cloed-form precoder for patial mltiplexing over correlated MIMO channel", IEEE Global Telecommnication Conference, 003. GLOBECOM '03, Vol. 4, pp , Dec [19] K.E. Yong and C. Joohwan, "Random beamforming in MIMO ytem exploiting efficient mltier diverity", in IEEE 61 t Vehiclar Technology Conference, 005. VTC 005-Spring. 005, Vol. 1, pp. 0-05, May 005. [0] A. Dofexi and S. Armor, "Deign Conideration and Phyical Layer Performance Relt for a 4G OFDMA Sytem Employing Dynamic Sbcarrier Allocation", IEEE 16 th International Sympoim on Peronal, Indoor and Mobile Radio Commnication, 005. PIMRC 005, Vol. 1, pp , Sept [1] Y. Peng. Dynamic Sb-carrier Allocation in MIMO-OFDMA Sytem. Ph.D. diertation, Univerity of Britol, UK, Ag [] J. Medbo and P. Schramm, "Channel Model for HIPERLAN/," ETSI/BRAN docment no. 3ERI085B, [3] 3GPP, Spatial channel model for MIMO imlation, TR V7.0.0, 3GPP, 007. [Online]. Available: [4] S.L. Loyka, Channel capacity of MIMO architectre ing the exponential correlation matrix, IEEE Commnication Letter, Vol.5, No.9, pp , Sep 001. [5] K.I. Pederen, P.E. Mogenen, B.H. Flery, Spatial Channel Characteritic in Otdoor E-ISSN: Ie 4, Volme 11, April 01

12 Rodiadee Nordin Environment and Their Impact on BS antenna Sytem Performance, IEEE Proc. Vehiclar Technology Conference. VTC 98, Vol., pp , May [6] K.I. Pederen, P.E. Mogenen, B.H. Flery, Power Azimth Spectrm in Otdoor Environment, IEEE Electronic Letter, Vol. 33, No. 18, pp, , Ag [7] U. Martin, A Directional Radio Channel Model for Denely Bilt-p Urban Area, Proc. on nd Eropean Peronal Mobile Radio Conference. EPMCC 97, pp , Oct Rodiadee Nordin received hi B.Eng. from Univeriti Kebangaan Malayia in 001 and Ph.D. from Univerity of Britol, United Kingdom in 011. He i crrently a lectrer in Department of Electrical, Electronic and Sytem Engineering in Univeriti Kebangaan Malayia. Hi reearch interet inclde Mltiple-Inpt Mltiple- Otpt (MIMO), Orthogonal Freency-Diviion Mltiple Acce (OFDMA), reorce allocation, green radio, intercell interference, cooperative diverity and indoor wirele localization. E-ISSN: Ie 4, Volme 11, April 01