Trajectory Design and Bandwidth Allocation Considering Power-Consumption Outage for UAV Communication: A Machine Learning Approach

Recent research has demonstrated that the heat induced by high data rate transmission could cause low-temperature burns. As a promising application for future wireless networks, unmanned aerial vehicle (UAV) communication is capable of providing high data rate transmission for ground users. Inspired by this progress, this article focuses on the resource allocation for the UAV scenario and proposes a novel framework to consider the newly mentioned phenomenon named by power-consumption outage (PCO). Specifically, we give the analysis of heat transfer model in the smartphone based on which we initially integrate the influence of PCO into the optimization problem of the UAV scenario. Furthermore, to solve the problem with joint optimization of bandwidth allocation and trajectory design, we propose a machine learning model consisting of the position prediction based on echo state network and the joint optimization based on deep reinforcement learning (DRL). Due to the continuity in action space, DRL optimization is specifically implemented by the normalized advantage function algorithm. Besides, considering the restriction for the implementation of machine learning, we propose a digital twin-enabled architecture to provide a virtual environment for the training. Simulation results show the advantage of the proposed scheme in total throughput and the adaptability for trajectory design in the presence of PCO.