Multiagent Learning for Secure Wireless Access From UAVs With Limited Energy Resources

The terrestrial wireless network deployment challenges and the high associated costs encourage the exploration of aerial base stations (ABSs). An ABS carried by an unmanned aerial vehicle (UAV) can be dispatched at a relatively low cost to provide coverage on demand, such as in emergency situations and during temporary hot-spot events. While relatively inexpensive, battery-powered UAVs have a limited flight time and can only provide temporary service in practice. This article, therefore, considers and monitors the available energy of UAVs as a constraint for the proposed communication architecture consisting of dynamically dispatched ABSs that are managed by a high-altitude platform station (HAPS) performing network optimization. We consider a fleet of UAVs for providing secure wireless service to sparsely distributed users in urban areas and propose an efficient coverage strategy to satisfy the users’ data rate demands while meeting their secrecy rate requirements. Because of the complexity, dynamics, and distributed nature of the problem, we employ multiple ABSs as the agents and design a deep deterministic policy gradient (DDPG) algorithm to optimize their positions in the ABS network with time-constrained nodes. Numerical results illustrate how the DDPG-empowered HAPS is able to coordinate and leverage the ABSs fleet for wide-spread secure coverage and adjust the network deployment topology when nodes become unavailable. While the DDPG has a higher training complexity, it provides better performance over state-of-the-art solutions in terms of the number of securely served users. We discuss the practical implications of the training process and identify opportunities for research and development.