Neural Networks in Hybrid Precoding for Millimeter Wave Massive MIMO Systems

Abstract: Neural networks have been applied to the physical layer of wireless communication systems to solve complex problems. In millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) systems, hybrid precoding has been considered as an energy-efficient technology to replace fully-digital precoding. The way of designing hybrid precoding in mmWave massive MIMO systems by multi-layer neural networks has not been investigated. Based on further decomposing the baseband precoding matrix, an idea is proposed in this paper to map hybrid precoding structure to a multi-layer neural network. Considering the deterioration in the throughput and energy efficiency of mmWave massive MIMO systems, the feasibility of the proposed idea is analyzed. Moreover, a singular value decomposition (SVD) based decomposing (SVDDE) algorithm is proposed to evaluate the feasibility of the proposed idea. Simulation results indicate that there is an optimal number of users which can minimize the performance deterioration. Moreover, the simulation results also show that slight deterioration in the throughput and energy efficiency of mmWave massive MIMO systems is caused by further decomposing the baseband precoding matrix. In other words, further decomposing the baseband precoding matrix is a feasible way to map the hybrid precoding structure to a multi-layer neural network.