Deep Reinforcement Learning-Based Resource Allocation in Cooperative UAV-Assisted Wireless Networks

We consider the downlink of an unmanned aerial vehicle (UAV) assisted cellular network consisting multiple cooperative UAVs, whose operations are coordinated by a central ground controller using wireless fronthaul links, to serve user equipments (UEs). A problem jointly designing UAVs’ positions, transmit beamforming, as well UAV-UE association is formulated in form mixed integer nonlinear programming (MINLP) maximize sum UEs’ achievable rate subject limited capacity constraints. Solving considered hard owing its non-convexity and unavailability channel state information (CSI) due movement UAVs. To tackle these effects, we propose novel algorithm comprising two distinguishing features: (i) exploiting deep Q-learning approach issue CSI for determining (ii) developing difference convex (DCA) efficiently solve UAV’s beamforming association. The proposed recursively solves interest until convergence, where each recursion executes steps. In first step, (DQL) allows UAVs learn overall account joint all adapt their locations. second given determined positions from DQL algorithm, DCA iteratively approximate subproblem original non-convex MINLP with updated parameters, problem’s variables Numerical results show that our design outperforms existing algorithms terms algorithmic convergence performance gain up 70%.