EM-Based Receiver Design for Uplink MIMO-OFDMA Systems Meng Wang, Graham C. Goodwin and Daniel E. Quevedo School of Electrical Engineering & Computer Science The University of Newcastle, Callaghan, NSW 2308, Australia Emails: MWang@studentmail.newcastle.edu.au, Graham.Goodwin@newcastle.edu.au, dquevedo@ieee.org Abstract— In this paper we propose an iterative receiver for uplink MIMO-OFDMA systems based on the expectation maximization (EM) algorithm. Iterating between the E-step and the M-step, the EM-based receiver updates the channel estimates and, refines data detection by increasing the likelihood function. Practical implementation issues are also considered: space-time block-coding (STBC) is incorporated to improve system perfor- mance against fading; a reduced-complexity algorithm is pro- posed, which simplifies the computation whilst not compromising performance. I. I NTRODUCTION Orthogonal frequency division multiple access (OFDMA) has been proposed as a promising multiple access scheme for broadband wireless networks, e.g., IEEE 802.16 [1]. In OFDMA systems, the multiple access is realized by assigning mutually exclusive subsets of available subcarriers to different users for simultaneous data transmission. The orthogonality of subcarriers eliminates intra-cell interference and allows simple receiver implementation. Furthermore, OFDMA systems offer high flexibility in radio resource management according to the quality of service (QoS) demands of different users. By em- ploying multiple antennas at both transmitter and receiver side, MIMO systems can increase channel capacity and mitigate the effect of multipath fading [2, 3]. The combination of OFDMA and MIMO techniques, MIMO-OFDMA, has become a strong candidate for next-generation wireless networks. In uplink MIMO-OFDMA systems, coherent signal detec- tion at the base station (BS) requires the channel state informa- tion (CSI) of all uplink channels between the BS and mobile users. To obtain the CSI, channel estimators are designed at the BS receiver using pilot symbols. Due to the time-varying nature of multipath fading channels, the CSI has to be updated continuously and promptly, e.g., [4]. Notwithstanding, pilot- based receivers have the following shortcomings: 1) Considerable data rate is potentially sacrificed for channel estimation in multipath fading environments. 2) Error floors may be introduced by CSI estimate errors, especially for fast fading channels. To deal with the above shortcomings, non pilot-aided, blind receiver designs have been reported in the literature (see [5] and references therein). However, blind receiver design suffers from several disadvantages including long data records and high complexity. Thus, there has been increasing interest in iterative receiver design, where joint channel estimation and data detection are performed iteratively. The expectation maximization (EM) algorithm [6, 7] has been employed in iter- ative receivers to perform maximum likelihood (ML) detection asymptotically with practical complexity. In the literature, EM- based receiver has been considered for single-user MIMO [8] and MIMO-OFDM systems [9, 10]. In this paper, we consider the design of an EM-based iterative receiver for uplink MIMO-OFDMA systems. The proposed receiver design can deal with flexible subcarrier allocation schemes [11] and approaches the ML receiver per- formance with limited complexity. For each user with allocated subcarriers, the proposed algorithm consists of two steps: E-step and M-step. The E-step performs channel estimation based upon hard decisions, assuming that the estimated data are known and correct. In the M-step, data detection is performed. The task can be split into several subproblems on a per-tone basis due to the orthogonality between subcarriers. To reduce the receiver complexity for implementation, we propose an inner-loop EM algorithm to avoid frequent inversion and multiplication of large matrices. II. SYSTEM MODEL We consider an uplink OFDMA system with one BS and U mobile users. The BS and each user are equipped with N R and N T antennas, respectively. We assume that N subcarriers are shared by all users and N u subcarriers are assigned exclusively to the uth user. The subcarriers are allocated dynamically through an allocation algorithm [11]. The index set of the subcarriers allocated to the uth user is given by I u = {i 1 ,i 2 , ··· ,i Nu } , u =1, ··· , U. (1) At the uth user’s transmitter, the N u -length data symbol transmitted on the tth antenna S t,u =  S t,u k  is mapped into an N -length data symbol X t,u =  X t,u k  by X t,u k =  S t,u k , k ∈ I u 0, k/ ∈ I u k =1, ··· ,N. (2) Then X t u is processed by an inverse fast Fourier transform (IFFT) unit. Before transmission, an N G -length cyclic prefix (CP) is appended to the IFFT output. The time-domain signal transmitted on the tth antenna is given by x t,u n = 1 √ N  k∈Iu X t,u k e j2πkn N , −N G ≤ n ≤ N − 1 (3) This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2008 proceedings. 978-1-4244-2075-9/08/$25.00 ©2008 IEEE 4516 Authorized licensed use limited to: University of Newcastle. Downloaded on December 15, 2008 at 23:31 from IEEE Xplore. Restrictions apply.