Deep Learning Based Robust Beamforming for UAV Communication System

摘要

Unmanned aerial vehicles (UAVs) have shown its application potentiality in communication system because of its inherent mobility and ease of organization. In this paper, we develop a robust beamforming method for an UAV communication system by deep learning (DL), in which a eavesdropping UAV attempts to wiretap the confidential transmission. Specifically, by modeling the legitimate channel and eavesdropping channel undergo Rician fading, a robust beamforming neural network (RBNN) is trained to maximize the average secrecy rate. Both perfect and imperfect channel state information (CSI) are used to train the RBNN in the off-line train stage, then all weights are fixed for on-line deployment stage with only imperfect CSI. Simulation results verify that the proposed DL based beamforming method outperforms benchmark beamforming methods with imperfect legitimate CSI and eavesdropper CSI, and generalization ability of the trained RBNN is shown. Last but not least, the proposed beamforming method is still effective while the eavesdropper CSI is completely unknown.