Low-complexity MMSE detector based on refinement Gauss-Seidel method for massive MIMO systems

Minimum mean square error (MMSE) linear detector has been considered as one of the potential detection algorithm for uplink multi-user massive MIMO systems, because it can achieve the near-optimal bit error rate (BER) performance. However, it involves matrix inversion with high complexity. Thus, the conventional Gauss-Seidel (GS) method has been applied for obtain a low-complexity MMSE detector without employing the computationally intensive matrix inversion. In this paper, we propose a refinement for the conventional GS method based on band matrix concept in order to accelerate the convergence rate, that means reduce the number of iterations and consequently the complexity, guaranteeing the near-optimal BER performance. Analytical and numerical results show the efficiency of the proposed method in comparison to the conventional GS method.