Low-Complexity MIMO Signal Detection Employing Multistream Constrained Search

Abstract

This study proposes the multi-stream constrained search (MSCS) as a low-complexity signal detection method for multiple-input multiple-output (MIMO) communications. MSCS achieves an excellent trade-off between computational complexity and bit error rate (BER). Based on the maximum $a posteriori$ probability (MAP) estimation, MSCS applies discrete optimization to some streams and continuous optimization to others. First, the discrete optimization constrains some streams to a certain set of modulation symbols. The continuous optimization is then used to identify the optimal continuous values of the other streams under this constrained condition and quantizes the results. A signal candidate comprises the quantized optimal values and constrained streams. Multiple constraint patterns result in multiple signal candidates, and sphere decoding (SD) identifies the detected signal as the one that maximizes the likelihood function. Limiting the number of constrained streams and applying SD can reduce computational complexity. In addition, MSCS selects constrained streams that have large variances of Gaussian distributions obtained from the MAP estimation. As a result of this stream selection method, MSCS can maintain an excellent BER performance even when the constrained streams are few in number. Computer simulations in 8-by-8 MIMO channels with modulation schemes of 16- and 64-QAM demonstrate that MSCS suffers degradation of merely 0.2 dB in average BER performance compared to the maximum likelihood detection (MLD). We also show that MSCS can achieve a similar average BER performance as that of the QR decomposition and M algorithm (QRM-MLD) while requiring less computational complexity. Under uncorrelated channels, the complexity of MSCS is less than 1/2 and 1/4 that of QRM-MLD in the case of 16- and 64-QAM, respectively. By contrast, simulations under channels with transmit-side spatial correlation show that when 16-QAM is used and average $E_{b}/N_{0}$ is equal to 18 dB, the complexity of MSCS ranges from 3/10 to 2/5 that of QRM-MLD. We also show that the complexity of MSCS ranges from 1/10 to 1/6 that of QRM-MLD when 64-QAM is used and the average $E_{b}/N_{0}$ is 23 dB.