Exploiting sparsity during the detection of high-order QAM signals in large dimension MIMO systems

This paper proposes a detector for large-scale multiple-input multiple-output (MIMO) systems for 16-QAM constellation and channel knowledge at the receiver. The detector is composed of multiple stages. During the first stage, linear MMSE filter is employed and nearest neighbor quantization is performed resulting in a sub-optimal estimate. In the second stage, the residual in the measurement vector is calculated and the subsequent detector works on the error vector which has additional structure. The error vector is often sparse (has few non-zero components) with the all-zero and lowest energy errors having the largest priors. Large number of antennas reduces the dependencies between error and noise vectors and allows the residual detection problem to be modeled as a linear inverse problem with sparse regularizer. The familiar sparse structure motivates the application of Sparse Bayesian Learning method in the detection. The resulting detector shows promise: the SNR gain over MMSE receiver is ~ 10 dB at a bit error rate (BER) of 10^-3 for 16-QAM 16 × 16 system.