Fast Beam Training for mmWave UAV Communications Using Machine Learning

Establishing and training beams in unmanned aerial vehicle (UAV) assisted millimeter-wave (mmWave) communications is a challenging task. In this paper, a novel beam training method is proposed by employing the machine learning (ML) method. Firstly, we analyze the applicable conditions of ML and preplan an ideal relationship of received signals. The required beam patterns based on this ideal relationship can be obtained by using the Fourier series method (FSM). We then formulate the beam selection issue as a polynomial regression problem based on hand-crafted features. Especially, we utilize the denoising autoencoder (DAE) to modify the error caused by the channel noise. Numerical simulation results demonstrate that our proposed beam training algorithm is able to provide precise beam selection for the mmWave UAV communications.