A New Reduced Complexity ML Detection Scheme for MIMO Systems

Abstract

For multiple-input multiple-output (MIMO) systems, the optimum maximum likelihood (ML) detection requires tremendous complexity as the number of antennas or modulation level increases. This paper proposes a new algorithm which attains the ML performance with significantly reduced complexity. Based on the minimum mean square error (MMSE) criterion, the proposed scheme reduces the search space by excluding unreliable candidate symbols in data streams. Utilizing the probability metric which evaluates the reliability with the normalized likelihood functions of each symbol candidate, near optimal ML detection is made possible. Also we derive the performance analysis which supports the validity of our proposed method. A threshold parameter is introduced to balance a tradeoff between complexity and performance. Besides, we propose an efficient way of generating the log likelihood ratio (LLR) values which can be used for coded systems. Simulation results show that the proposed scheme achieves almost the same performance as the ML detection at a bit error rate (BER) of 10(-4) with 28% and 15% of real multiplications compared to the conventional QR decomposition with M-algorithm (QRD-M) in 4-QAM and 16-QAM, respectively. Also we confirm that the proposed scheme achieves the near-optimal performance for all ranges of code rates with much reduced complexity. For instance, our scheme exhibits 74% and 46% multiplication reduction in 4-QAM and 16-QAM, respectively, compared to the sphere decoding based soft-output scheme with rate-1/2 convolutional code.