University of Bristol - Explore Bristol Research. Peer reviewed version. Link to published version (if available): /VETECS.2009.

Transcription

1 Nordin, R., Armour, S. M. D., & McGeehan, J. P. (9). Overcoming elfinterference in SM-OFDMA with and dynamic ubcarrier allocation. In IEEE 69th Vehicular Technology Conference 9 (VTC Spring 9), Barcelona (pp. 1-5). Intitute of Electrical and Electronic Engineer (IEEE). DOI: 1.119/VETECS Peer reviewed verion Lin to publihed verion (if available): 1.119/VETECS Lin to publication record in Explore Britol Reearch PDF-document Univerity of Britol - Explore Britol Reearch General right Thi document i made available in accordance with publiher policie. Pleae cite only the publihed verion uing the reference above. Full term of ue are available:

2 Overcoming Self- Interference in SM-OFDMA with and Dynamic Subcarrier Allocation R. Nordin, S. Armour, and J.P. McGeehan Centre for Communication Reearch Univerity of Britol Britol, United Kingdom Abtract For a ingle uer with multiple receive antenna, coantenna interference can be the dominant ource of impairment in a Spatial Multiplex Orthogonal Freuency Diviion Multiple Acce (SM-OFDMA) ytem, particularly when correlation exit between patial ub-channel. By exploiting nowledge of the channel repone and combining it with Dynamic Sub- Carrier Allocation (DSA), the algorithm propoed in thi paper aim to reduce the effect of interference, while providing fair gain and maximizing the SINR acro all uer. Simulation reult reveal that the propoed algorithm conider elf-interference that exit between multiple antenna and i capable of achieving a capacity gain for every uer with improved BER performance compared to previou wor. Keyword- MIMO, OFDMA,, DSA I. INTRODUCTION Information theoretic reult have hown the remarable capacity potential of wirele communication ytem uing multiple antenna array at both tranmitter and receiver. Thi ytem, nown a Multiple Input Multiple Output (MIMO) ha been hown by Fochini and Gan [1] to give ignificant capacity increae, which grow linearly with the minimum number of tranmit and receive antenna when operating on a ingle lin with Gauian noie. Performance can be further improved in wideband cenario by combining the MIMO architecture with OFDM(A). OFDM i well uited in an environment where multipath i a major ource of ditortion and offer further advantage uch a high pectral efficiency, better reitance againt interference and implified implementation. OFDMA further offer enhanced upport for multi-uer tranmiion through the ability to hare channel reource with fine granularity in the freuency domain. A a reult of the combined benefit of MIMO with OFDMA, thi technology ha been elected a the PHY Layer air-interface for WiMAX (8.16) whilt Third Generation Partnerhip Project (3GPP) ha aumed OFDMA will be ued a the downlin interface for the Long Term Evolution (LTE) networ - an enhancement to exiting 3G tandard. The IEEE 8. Woring Group ha alo nominated OFDMA a it pacet-baed air interface for future Mobile Broadband Wirele Acce (MBWA) []. The capacity of a ytem employing a MIMO architecture ha been hown by Gebert et al. [3] and Andrew et al. [4] to decreae a the number of tranmit antenna increae if the patial interference i not uitably addreed. Thi paper will invetigate the capacity performance of a SM cheme ince the main challenge in interference-limited ituation decribed in thi SM cheme i to maximize the capacity. SM aim to achieve maximum capacity by achieving higher pectral efficiencie, i.e., more bit//hz of bandwidth. Thi i achieved by dividing the high rate data tream input into parallel independent data tream with each ub-tream being tranmitted over different antenna, thu increaing the viability of high data rate by a factor of min(n t, N r ) [3]. There are three type of interference in MIMO ytem. Conider a cellular networ with M Bae Station (BS) adjacent to each other with N t antenna. The ource of interference for a ingle MS can be determined by M N t interfering ignal if there are M non-negligible neighbouring bae tation. Thi effect i nown a intra-cell interference, or other cell interference (OCI) [4]. A oppoed to OCI, interference that come from other uer within a ingle cell i nown a elf-cell interference. The third ource of interference i called co-antenna interference, alo nown a elfinterference reulting between the patial multiplexing data tream of a ingle uer. Self-interference can be meaured uing an effective ignal-to-interference-and-noie-ratio () metric [5]. Thi indicate the channel uality from the nowledge of elf-interference caued by the mimatch between the patial ub-channel. In thi paper the metric i ued in combination with the dynamic ubcarrier allocation algorithm propoed in [6] to allocate the bet ubcarrier to different uer in a multiuer cenario where the metric i ued to determine the ubcarrier uality; thereby taing into account both patial ubchannel uality and elf interference. The propoed algorithm thu attempt to (i) utilie the multi-uer channel to mitigate channel fading and exploit it a a ource of diverity, (ii) conider the interference that exit between patial ub-channel, and (iii) provide fair benefit to all uer. Thi paper i organized a follow. Section II decribe the PHY Layer model of the OFDMA ytem and the propoed DSA- algorithm for the MIMO-SM architecture. In Section III, the imulation etup and ytem parameter are explained and in Section IV the reult from the numerical imulation and information theoretic analyi are preented. Section V conclude the paper /9/$. 9 IEEE

3 Figure 1. Bloc Diagram of DSA- with SM-OFDMA Architecture II. SYSTEM MODEL A. OFDMA Sytem In thi paper, the downlin path of an OFDMA ytem i conidered for multiple uer to exploit diverity. Thu OCI i not conidered. In a ingle cell, one Bae Station (BS) etablihe communication with multiple Mobile Station (MS) imultaneouly and each MS experience independent fading channel. For each MS, a ingle ub-channel i allocated, coniting of a number of OFDMA ubcarrier and each ubcarrier i ubjected to a flat Rayleigh-fading narrowband channel with independent fading coefficient for each path. All uer are aigned with the ame number of ubcarrier in each ub-channel. The phyical layer of a ingle lin i preented a a bloc diagram in Figure 1. The BS tranmit data tream to all MS and applie independent bit level error control coding, ymbol mapping and erial to parallel converion. The DSA mapping proce allocate the parallel data ymbol to uitable ubcarrier from the ubcarrier indexe generated from the allocation algorithm. OFDM modulation i implemented by mean of an invere Fat Fourier Tranform (IFFT) with ubeuent guard interval (GI) inertion to combat ISI due to multipath delay. The GI i appended by mean of a cyclic extenion. When the GI i longer than the exce delay of the channel, ISI i eliminated. At the MS, the reultant coded/ multiplexed ymbol tream are aigned to appropriate ubcarrier a indicated by the DSA algorithm. The allocated ubcarrier are then extracted from the given MS ub-channel. Each MS receiver extract the GI, applie a parallel-to-erial converion and a Fat Fourier Tranform (FFT). Eualiation and oft de-mapping yield bitlevel information which i ubject to error control. Thi paper conider a MIMO architecture, but the analyi i readily extendable to higher MIMO order. B. Metric The mathematical model for a received ignal in an SM- OFDMA ytem, after FFT and guard removal i decribed a follow: t Y = H X + N (1) where the ubcript denote MS index, denote the ubcarrier index, H i the channel matrix containing MS freuency repone of the channel between N t tranmit and N r receive antenna at ubcarrier and applied to the ubcarrier of the OFDMA ignal on a cluter bai, while N denote a complex circular ymmetric colored noie with invertible covariance matrix and X t denote N t 1 matrix containing tranmit ignal. At the receiver, the propoed OFDMA ytem adopt a linear MMSE configuration: H 1 ( ) H H + SNR I ( H ) 1 G = () 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 layer of the MIMO tructure [5]. A a reult, different patial layer can be allocated to a different MS to achieve an additional patial multi-uer diverity gain. On the other hand, thi increae the amount of feedbac by the number of patial layer. The received ignal i multiplied by an MMSE filter given by (): Y Y t G = G X + G N (3)

4 MS then compute the at every data tream for every ubcarrier [5] (the ubcarrier index,, i omitted): ( G H ) E = (4) j, j + ( G H ) E + ( G G )N j, j where i the patial ub-channel at every ubcarrier, E denote the average ymbol energy and. j denote the element located in row and column j. The metric aim to compute elf-interference from the data tream component,. and elf-interference component,. j,j from the other tranmitted data tream within the ame channel. The data rate of each ubcarrier for MS in the freuency domain can be calculated a [5]: r = log (1 + ) (5) For comparion purpoe, the feedbac capacity for the metric i compared with effective SNR (ESNR). ESNR conider the eigenvalue of the MIMO channel to indicate the channel uality from every MS: ESNR = D D H ( V ) E H ( V ) E + N j, j where D i a diagonal matrix and V i the unitary matrix obtained by applying the Singular Value Decompoition (SVD) to H. Capacity feedbac of each ubcarrier for MS in freuency domain i: N = r (6) r log (1 + ESNR ) (7) C. DSA Algorithm Each MS provide feedbac information 1 to the BS and the DSA algorithm then allocate ubcarrier to the MS. The DSA algorithm i baed on previou wor of Peng [6], which conider five different cheme to predict the performance in correlated and uncorrelated channel. Thi paper only conider DSA-Scheme 1, whereby ubcarrier allocation at each patial layer i treated independently in an uncorrelated channel. The ue of the metric propoed in thi paper i an alternative method for mitigating interference to thoe method ued by cheme -5 in [6]. The algorithm ran uer from the lowet to highet data rate at each patial layer. Coneuently, the next bet ubcarrier are allocated to uer in ran order, allowing uer 1 Refer to (4) or r (5). Either one can be ued in thi tep with the lowet data rate at that particular patial layer to have the bet data rate that i available for the next tranmiion. The propoed Dynamic Subcarrier Allocation (DSA) with metric algorithm can be decribed a below: 1) After the bae tation tranmit data vector X t, the -th mobile tation compute the MMSE filter given by (). ) The -th mobile tation then compute the (4) and data rate, r (5), of the -th patial ub-channel. 3) With the feedbac information from the MS, the uer with the lowet allocated data rate i allowed to have the next choice of bet ubcarrier a follow: (a) Short lit of uer i generated, tarting at uer with leat data rate, Γ. Find uer atifying: i Γ Γ for all i, 1 i K (b) For the uer above, find the ubcarrier n atifying: r r j for all j N (c) Then r, N and C, are updated with and n in (b) above according to the following relationhip: Γ = Γ + r, n N = N C, = n where N i a vector containing the indice of the ueable ubcarrier, C, i the allocation matrix to record the allocated ubcarrier, n for uer and i the cluter ize. 4) Go to the next uer in the hort lit defined in (a) above, until all uer are allocated another ubcarrier, N. The algorithm can be repeated when the channel change. In thi algorithm, uer are not allowed to hare ubcarrier, thu reducing the complexity of the algorithm. In the SISO cae, Jiho and Kwang Bo [7] how that the capacity can be maximized if a ubcarrier i only aigned to one uer, a it help to reduce the interference from other uer ignal that happen to hare the ame ubcarrier. Thi paper extend the theory preented by Jiho and Kwang Bo loc. cit. [7] into the MIMO cae, where the number of uer haring the ame ubcarrier i limited to the number of available patial ubchannel (two patial layer in the reult below). III. SIMULATION ENVIRONMENT AND PARAMETERS The channel model employed for thi paper i baed on Channel E adopted by the ETSI-BRAN tandard [8]. Channel E imulate the typical large open pace outdoor environment for NLOS condition, with exce delay pread of 176n, ampling period of 1n and RMS delay pread of 5n. The ey parameter for thi paper and the tranmiion mode to imulate the OFDMA ytem are ummaried in Table I. n

5 TABLE I. PARAMETERS FOR THE PROPOSED OFDMA SYSTEM 15 Probability(Normalied perceived metric > abcia Operating freuency 5 GHz Bandwidth 1 MHz FFT Size 14 Ueful Subcarrier 768 Subcarrier pacing KHz Ueful ymbol duration 1.4 μ Total ymbol duration 1. μ Channel Coding Punctured ½ rate convolution code, contraint length 7, {133, 171} octal Normalied perceived metric (db) Figure. CCDF comparion of total metric between three different allocation trategie independent identically ditributed (i.i.d.) uai-tatic random channel ample per uer are adopted in the imulation. The calculation i performed at MS level and the feedbac information i ent to the BS. The propoed algorithm i imulated for 16 MS uer, where each of the MS i allocated a ingle ub-channel coniting of 48 ueable ubcarrier. It i aumed that the BS ha perfect nowledge of the channel gain matrix and ue thi to determine the ubcarrier allocation. IV. PERFORMANCE ANALYSIS In thi ection, reult are preented for an OFDMA ytem with different allocation trategie. Thee are: random allocation, DSA uing channel gain a the metric (from [6]), DSA uing a the metric and DSA uing ESNR a the metric. Eual power allocation i aumed for all allocation trategie. The total normalied metric for all 16 uer achieved by different algorithm are compared in Figure, which repreent complementary cumulative ditribution function (CCDF) of all channel ample for the three allocation algorithm a decribed earlier. It can be een that DSA uing the metric outperform the random allocation trategie by up to 7 db. DSA uing channel gain a the metric ha lightly lower gain; approximately db compared to DSA uing the metric. Normalied metric (db) DSA DSA Subcarrier index Figure 3. Example of allocated ubcarrier for two different allocation algorithm acro metric. Thee effect can be jutified by conidering the intantaneou wideband channel repone in the freuency domain for an arbitrary uer and the correponding ubcarrier allocation achieved by three allocation algorithm a hown in Figure 3. Thi reult illutrate the multi-uer diverity benefit that the propoed algorithm i able to achieve. It can be een that the ubcarrier allocated baed on the DSA- algorithm have conitent high metric (higher than the mean for the channel) and a flatter repone than the actual channel. For DSA baed on channel gain, it can be oberved that ome of the allocated ubcarrier have a ignificantly lower metric, which how that the DSA with channel gain doe not conider the effect of elf-interference (and it impact upon ytem capacity and BER) for the elected ubcarrier. Fairne gain acro all uer i invetigated from the mean variance and average metric acro uer. The average gain and variance for all uer i hown in Figure 4 and further implified in Table II. DSA uing the metric ha the mallet variance compared to other type of allocation algorithm. DSA uing the metric alo ha the larget average channel repone, which how that the propoed algorithm offer fairne to every uer whilt maximizing the average channel gain of any given uer without minimizing it in the other uer. Figure 5 compare the capacity between random allocation and DSA, which ue and ESNR metric acro the normalied SNR (E b /N ). Auming perfect nowledge of the channel matrix, the SVD OFDM ytem capacity bound i alo plotted for reference. At below db, both and ESNR ytem have almot identical data rate. A a coneuence of the ue of the DSA algorithm, the average data rate performance for both the and ESNR ytem increae at approximately twice the capacity of random ubcarrier allocation. The margin of improvement can be oberved to increae ignificantly for ytem employing DSA-. Thi i due to the ability of MMSE receiver to minimize elfinterference and noie. Thi reult can be aociated with obervation from Figure 4, where the propoed algorithm i able to achieve the deired balance between fairne and ytem capacity.

6 Average (db) Variance Uer number Uer number BER SNR(dB) Figure 4. Average metric and mean variance acro all uer Figure 6. BER performance between three different allocation trategie Throughput (bp/hz) TABLE II. AVERAGE NORMALISED METRIC AND VARIANCE (ACROSS ALL USERS) FOR THREE ALLOCATION ALGORITHMS Algorithm Channel gain Mean (db) Variance DSA - ESNR-DSA ESNR- Capacity bound E b /N (db) Figure 5. Feedbac capacity comparion Figure 6 how the BER performance comparion between three allocation trategie (a decribed earlier) in a SM- OFDMA ytem, uing 64-QAM modulation and ¾-rate coding, imulated in an uncorrelated channel environment. It can be een that DSA uing the metric ha better performance than DSA uing channel gain by 4 db (at 1-5 BER), implying that DSA- ha the additional benefit of minimizing the effect of elf-interference while providing fair ditribution of multiuer diverity benefit between uer. allocation degrade much more everely and almot reache an error floor, a there i no diverity exploited and elf-interference everely affect the SM-OFDMA ytem. The capacity gain and BER curve for the metric employing an MMSE receiver how that it offer better performance compared to the DSA uing channel gain and random allocation. Thi i due to the ability of MMSE receiver to minimize elf-interference and noie. CONCLUSIONS By employing the metric in combination with the DSA algorithm, initial invetigation ha revealed that the next generation of wirele ytem would be capable of improving the capacity performance while providing fair gain and improved BER performance. The propoed algorithm conider co-antenna interference within a SM-OFDMA ytem. Future wor will focu on invetigating BER and capacity performance of the ytem in correlated channel where the effect of elf-interference i more dominant. Baed on the initial reult, the propoed algorithm can be expected to provide even greater benefit. REFERENCES [1] G. J. Fochini and M. J. Gan, "On Limit of Wirele Communication in a Fading Environment when Uing Multiple Antenna," Wirele Peronal Communication., vol. 6(3), pp , [] A. Greenpan, M. Klerer, J. Tomci, R. Canchi, and J. Wilon, "IEEE 8.: Mobile Broadband Wirele Acce for the Twenty-Firt Century," Communication Magazine, IEEE, vol. 46(7), pp , 8. [3] D. Gebert, M. Shafi, S. Da-han, P. J. Smith, and A. Naguib, "From theory to practice: an overview of MIMO pace-time coded wirele ytem," Selected Area in Communication, IEEE Journal on, vol. 1(3), pp. 81-3, 3. [4] J. G. Andrew, C. Wan, and R. W. Heath, "Overcoming interference in patial multiplexing MIMO cellular networ," Wirele Communication, IEEE, vol. 14(6), pp , 7. [5] K. Eun Yong and C. Joohwan, " beamforming in MIMO ytem exploiting efficient multiuer diverity," in Vehicular Technology Conference, 5. VTC 5-Spring. 5 IEEE 61 t 5, pp. -5 Vol. 1. [6] Y. Peng,"Dynamic Sub-carrier Allocation in MIMO-OFDMA Sytem." Ph.D. diertation. Britol: Univerity of Britol, 6. [7] J. Jiho and L. Kwang Bo, "Tranmit power adaptation for multiuer OFDM ytem," Selected Area in Communication, IEEE Journal on, vol. 1(), pp , 3. [8] J. Medbo and P. Schramm, "Channel Model for HIPERLAN/," ETSI/BRAN document no. 3ERI85B, [9] A. Doufexi and S. Armour, "Deign Conideration and Phyical Layer Performance Reult for a 4G OFDMA Sytem Employing Dynamic Subcarrier Allocation," in Peronal, Indoor and Mobile Radio Communication, 5. PIMRC 5. IEEE 16th International Sympoium on 5, pp