Title:Distributed Generative Adversarial Networks for mmWaveChannel Modeling in Wireless UAV Networks

Abstract:In this paper, a novel framework is proposed to enable air-to-ground channel modeling over millimeter wave (mmWave) frequencies in an unmanned aerial vehicle (UAV) wireless network. First, an effective channel estimation approach is developed to collect mmWave channel information allowing each UAV to train a local channel model via a generative adversarial network (GAN). Next, in order to share the channel information between UAVs in a privacy-preserving manner, a cooperative framework, based on a distributed GAN architecture, is developed to enable each UAV to learn the mmWave channel distribution from the entire dataset in a fully distributed approach. The necessary and sufficient conditions for the optimal network structure that maximizes the learning rate for information sharing in the distributed network are derived. Simulation results show that the learning rate of the proposed GAN approach will increase by sharing more generated channel samples at each learning iteration, but decrease given more UAVs in the network. The results also show that the proposed GAN method yields a higher learning accuracy, compared with a standalone GAN, and improves the average rate for UAV downlink communications by over 10%, compared with a baseline real-time channel estimation scheme.