EXIT Chart Analysis of Low-Complexity Bayesian Turbo Multiuser Detection for Rank-Deficient Multiple Antenna Aided OFDM

This paper studies the mutual information transfer characteristics of a novel low-complexity Bayesian Multiuser Detector (MUD) proposed for employment in Space Division Multiple Access (SDMA) aided Orthogonal Frequency Division Multiplexing (OFDM) systems. The design of the Bayesian MUD advocated is based on extending the optimum single-user Bayesian design to multiuser OFDM signals modeled by a Gaussian mixture, rather than by a single Gaussian distribution, when characterizing the conditional PDF of the received signal. In order to reduce the complexity of the Bayesian MUD, we introduce an a priori information threshold and then discard the low- probability terms during the calculation of the extrinsic information generated . The achievable complexity reduction as a function of different threshold values is analyzed and the best tradeoff values are derived with the aid of simulation. Both non-systematic and recursive systematic convolutional codes are used for exchanging extrinsic information with the MUD for the sake of achieving a turbo-detection aided iteration gain. The convergence behavior of the proposed low-complexity Bayesian turbo MUD is investigated using Extrinsic Information Transfer (EXIT) chart analysis and compared to that of Soft Interference Cancellation aided Minimum Mean Square Error (SIC-MMSE) MUD schemes. As expected, the simulation results show that the proposed low-complexity Bayesian Turbo MUD outperforms the SIC-MMSE MUDs. A substantial benefit of the proposed MUD is that it is potentially capable of supporting up to three times higher number of users than the number of receiver antennas. In this challenging multiuser scenario, the resultant channel- matrix becomes rank-deficient, resulting in a linearly non-separable detector output phasor constellation, when classic linear receivers tend to exhibit a poor performance.