Machine Learning-Based Hybrid Precoding with Robust Error for UAV mmWave Massive MIMO

Unmanned aerial vehicles (UAVs) can now be considered as aerial base stations (BSs) to support ultra-reliable and low-latency communications by establishing line-of-sight (LoS) connections to ground users. Moreover, combining UAVs with millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) will be a promissing solution. It can provide potentially high capacity wireless services due to their aerial positions and their ability to deploy on demand at specific locations. In this paper, we propose a low-cost and energy-efficient hybrid precoding architecture for UAVs, where the antenna part is realized by lens array. We investigate an efficient and energy-saving hybrid precoding scheme with robustness, which is inspired by the cross-entropy (CE) optimization in machine learning and the relative error estimation optimization. As for each selection of the hybrid precoders for obtaining the optimized precoder, we regarded it as a training process in machine learning, in which the training target is the CE-loss function between the predicted precoders and the target precoders. It aims to minimize the relative error between the predicted and actual values for optimizing the probability distributions of the elements in the analog hybrid precoder. Simulation results show that our proposed scheme can achieve higher sum rate and energy efficiency.