Gaussian Process Based Channel Prediction for Communication-Relay UAV in Urban Environments

This paper presents a learning approach to predict air-to-ground communication channel strength to support the communication-relay mission using the unmanned aerial vehicle (UAV) in complex urban environments. The knowledge of the air-to-ground communication channel quality between the UAV and ground nodes is essential for optimal relay trajectory planning. However, because of the obstruction by buildings and interferences in the urban environment, modeling and predicting the communication channel strength is a challenging task. We address this issue by leveraging the Gaussian process (GP) method to learn the communication shadow fading in a given environment and then employing the optimization-based relay trajectory planning by using learned communication properties. The key advantage of this learning method over fixed communication model based approaches is that it can keep refining channel prediction and trajectory planning as more channel measurement data are obtained. Two schemes incorporating GP-based channel prediction into trajectory planning are proposed. Monte Carlo simulations demonstrate the performance gain and robustness of the proposed approaches over the existing methods.