Grey neural network channel estimation and RBFNN hybrid precoding schemes for the multi user millimeter wave massive MIMO

Abstract

AbstractThe millimeter wave (mm‐wave) multi‐input multi‐output (MIMO) communications are the most promising technology specially developed for next‐generation wireless networks so that the throughput and available spectrum of the network has been massively increased. At base station, it has a large number of antenna arrays with coherent transceiver processing, which helps to increase spectral efficiency. Each antenna in conventional MIMO systems makes use of a single radio‐frequency (RF) chain to boost multiplexing gain. Additionally, due to broader bandwidth, channel frequency is selective and signal‐to‐noise (SNR) is low. A successful channel estimation training phase is thus required. However, the utilization of a huge number of antennas leads to unaffordable hardware costs and excessive power usage. In order to deliver high array gain at a more affordable price, hybrid precoding techniques are applied. Both baseband combinational matrix and baseband precoding matrix must be optimised when transmitter and receiver sides of mm‐wave MIMO systems adopt hybrid precoding architecture. In addition, the combinational matrix and the precoding matrix have to be simulated to obtain the maximum system sum rate. This requires an efficient hybrid precoding approach for generating improved array gain. Therefore, this work uses a novel grey neural network approach for channel estimation and radial basis function neural network (RBFNN) for hybrid precoding. For effective channel estimation, grey wolf optimization (GWO) works in conjunction with artificial neural networks (ANN) to produce optimum results. In mm‐wave MIMO systems, the hybrid precoding RBFNN offers reduced complexity and increased efficiency. As a result, the proposed work adopts neural networks to provide effective channel estimation and hybrid precoding.