Deep Learning-Based Two-Stage Channel Tracking for Ground-to-Air Communication Systems

In current wireless communication systems, the channel state information (CSI) is acquired from the received pilot sequence, whose accuracy is degraded due to the channel estimation error and the channel feedback delay. These adverse effects are even worse in ground-to-air (G2A) communications, because of high mobility of the unmanned aerial vehicle (UAV) nodes. In addition, the CSI inaccuracy cannot be improved via multi-point cooperation, since the CSI is locally associated with each communication link. In this paper, we consider a UAV communication system in urban environments and propose an angular domain channel fingerprint (CF)-based two-stage channel estimation and tracking framework. The proposed approach acquires the CSI of the G2A channel by first localizing the position of the UAV and then mapping to the CSI. As the UAV position is global information to the network, the acquisition of the UAV position can be improved via multi-point joint localization. In specific, we first construct a mixture-of-expert neural network (MoENN) to accurately locate the UAV by multiple ground base stations (BSs), leveraging the angular domain CFs extracted from the pilot signals. Next, based on the localization results, we utilize the extended Kalman filtering technique to predict the trajectory of UAV and reconstruct the real-time channel by mapping the UAV location back to the angular domain CFs. Simulation results demonstrate that the proposed framework exhibits excellent localization performance as well as channel tracking performance, without expense of additional overhead of the pilot training.