Joint Trajectory Design and Radio Resource Management for UAV-aided Vehicular Networks

In the last years, the number of applications for Unmanned Aerial Vehicles (UAVs) has increased. Among them, the possibility to deploy them as flying base stations, namely Unmanned Aerial Base Stations (UABSs), has attracted the attention of industry and researchers. The unmatched mobility of UAVs, together with the unique quality of air-to-ground radio links, allow a boost in the capacity and coverage of existing mobile networks. In this paper, the use of UABSs is studied to assist a terrestrial mobile network aiming at serving moving connected vehicles, denoted as Ground User Equipments (GUEs), implementing Vehicle-To-Anything (V2X) extended sensing applications. To this aim, techniques are presented to tackle two important problems: trajectory design for the UABS allowing for tracking GUEs moving in a complex urban scenario and the scheduling of radio resources used to serve them. The former is solved by leveraging a Deep Reinforcement Learning (DRL) algorithm, Double Dueling Deep Q-Network (3DQN), whereas the latter is modelled via Integer Linear Program (ILP). Since we assume radio resources are all shared among GUEs, Macro Base Stations (MBS) and the UABS, the positioning of the UABS deeply affects interference, that is the radio resource management (RRM) algorithm; therefore, the two problems must be considered and solved jointly, choosing the reward function of the DRL algorithm properly. Two different scenarios are addressed: a coverage limited and a capacity limited one. Performance metrics shown are both machine learning related, delivering the training outcome of the agent, and network related, such as the percentage of satisfied GUEs for different application requirements.