Steepest Descent Method Based Soft-Output Detection for Massive MIMO Uplink

Abstract

For linear minimum mean square error (MMSE) detection massive multiple-input multiple-output (MIMO) system uplink, the convergence rate of some iterative methods such as conjugate gradient (CG) slows down with more complicated antenna configurations. And the conventional Cholesky method relies on a higher computational complexity to achieve a better performance. In this paper, we introduce the steepest descent (SD)-based soft-output method to solve the reduction of convergence rate with a low complexity by introducing an effective preconditioner. Further more, an efficient approximated method with much lower complexity to calculate the log-likelihood ratios (LLRs) is proposed. The analytical as well as the numerical results have indicated the proposed SD-based method outperforms the recent proposed classical Conjugate Gradient (CG) method with around 1 dB performance improvement at the BER of 10-3, and also achieves 75% complexity reduction from conventional Cholesky approach.