Low‐complexity unsupervised learning‐based hybrid precoding for massive MIMO systems

Abstract

AbstractHybrid precoding can significantly reduce the number of required radio frequency (RF) chains in massive MIMO systems. However, due to the large number of antennas and the high‐dimensional channel, the existing hybrid precoding algorithms usually require very high complexity. In this paper, we propose a convolutional neural network (CNN)‐based hybrid precoding scheme by utilizing unsupervised learning to reduce both the computational complexity and space complexity. Specifically, we first design a low‐complexity CNN structure to achieve the analog precoder, where the size of the convolution kernel and the number of channels are reduced by the further improved inception network. Then, we obtain the digital precoder by the classical zero‐forcing (ZF) algorithm. Simulation results show that, the proposed CNN‐based hybrid precoding can approach the sum‐rate performance of the classical two‐stage hybrid precoding but has much lower complexity.