Multiagent Collaborative Learning for UAV Enabled Wireless Networks

The unmanned aerial vehicle (UAV) technique provides a potential solution to scalable wireless edge networks. This paper uses two UAVs, with accelerated motions and fixed altitudes, to realize a wireless edge network, where one UAV forwards downlink signals to user terminals (UTs) distributed over an area while the other one collects uplink data. The conditional average achievable rates, as well as their lower bounds, of both the uplink and downlink transmission are derived considering the active probability of UTs and the service queues of two UAVs. In addition, a problem aiming to maximize the energy efficiency of the whole system is formulated, which takes into account communication related energy and propulsion energy consumption. Then, we develop a novel multi-agent Q-learning (MA-QL) algorithm to maximize the energy efficiency, through optimizing the trajectory and transmit power of the UAVs. Finally, simulation results are conducted to verify our analysis and examine the impact of different parameters on the downlink and uplink achievable rates, UAV energy consumption, and system energy efficiency. It is demonstrated that the proposed algorithm achieves much higher energy efficiency than other benchmark schemes.