Deep Reinforcement Learning to Enhance the Energy-efficient Performance of Unmanned Aerial Vehicles-enabled Fog Radio Access Networks

Nowadays, the use of unmanned aerial vehicles (UAVs) as the fog access points (F-APs) is of high practical value to future fog radio access networks (F-RANs). Compared to the F-AP, UAV enabled F-AP possesses stronger line-of-sight links with the ground terminals (GTs) due to its high altitude as well as high and flexible mobility in 3D spaces. Nevertheless, unlike terrestrial F-AP that has reliable power supply, UAV- enabled F-AP in practice has limited on board power, but requires high propulsion power to stay air-borne and support high mobility. Motivated by the above considerations, this paper aims to explore how deep reinforcement learning (DRL) can be applied to enhance the energy-efficient performance of UAV- enabled F-RANs. Such DRL method supports UAV enabled F-AP to real-timely optimize the air-to-ground fog computing by reasoning network configuration in an online manner, which sheds new light on the existing works in UAV-enabled networks. In specific, the DRL based network configuration considers parameters including resource allocation, task offloading, cache deployment and 3D UAV trajectory, and get validated via simulation in this paper. In the end, the main design challenges and promising directions for future research are also discussed.