Machine Learning-Based Hybrid Precoding With Low-Resolution Analog Phase Shifters

Abstract

In this letter, inspired by the cross-entropy (CE) optimization in machine learning, we propose a CE-based algorithm for hybrid precoding with low-resolution analog phase shifters in millimeter wave (mmWave) multi-input multi-output (MIMO) systems. The main idea is to generate some candidate analog precoders according to a series of pre-defined probability distributions and select partial analog precoders as elites to update the probability distributions. Through iteration, the probability distributions will converge to a stable state and the optimal precoders can be obtained with a sufficiently high probability. Furthermore, we extend the proposed algorithm to multi-user hybrid precoding. The simulation results demonstrate that the proposed algorithm can achieve near-optimal performance of the fully-digital precoding with lower computational complexity than other near-optimal algorithms.