Hybrid Precoding for Beamspace MIMO Systems with Sub-Connected Switches: A Machine Learning Approach

By employing lens antenna arrays, the number of radio frequency (RF) chains in millimeter-wave (mmWave) communications can be significantly reduced. However, most existing studies consider the phase shifters (PSs) as the main components of the analog beamformer, which may result in a significant loss of energy efficiency (EE). In this paper, we propose a switch selecting network to solve this issue, where the analog part of the beamspace MIMO system is realized by a sub-connected switch selecting network rather than the PS network. Based on the proposed architecture and inspired by the cross-entropy (CE) optimization developed in machine learning, an optimal hybrid cross-entropy (HCE)-based hybrid precoding scheme is designed to maximize the achievable sum rate, where the probability distribution of the hybrid precoder is updated by minimizing CE with unadjusted probabilities and smoothing constant. Simulation results show that the proposed HCE-based hybrid precoding can not only effectively achieve the satisfied sum-rate, but also outperform the PSs schemes concerning energy efficiency.