Trajectory Learning, Clustering, and User Association for Dynamically Connectable UAV Base Stations

This work examines trajectory learning, clustering, and user association policies for dynamically connectable unmanned aerial vehicle base stations (UAV-BSs). Here, UAV-BSs are allowed to dynamically form physically-connected collocated antenna arrays that enable joint transmission and cooperative energy sharing among the connected UAV-BSs. The UAVs' trajectories, their clustering decisions, and the user association are jointly determined for the case with static users by maximizing a proportional-fair objective given by the sum log-rate of all users. Then, for the case with dynamic users, the UAVs' trajectory learning, re-clustering, and user handover policies are developed to adapt to changes in the users' locations. In particular, the UAVs' locations are adjusted gradually in each time slot based on a stochastic gradient ascent algorithm, and user handover decisions are made in each time slot based on a reward function that is inspired by the solution in the static case. The clustering decisions are updated every T time slots by combining two UAV clusters or by separating a cluster into two. To determine the transmission schemes in each time slot, a semi-orthogonal user scheduling policy is first employed to determine the users that are served in each time slot. Then, joint design of the transmit beamformers and powers is proposed to maximize the sum of log signal-to-leakage-plus-noise ratio (SLNR) of scheduled users. Simulations are provided to demonstrate the effectiveness of the proposed schemes.