MEC-Enhanced Aerial Serving Networks via HAP: A Deep Reinforcement Learning Approach

Abstract

Next-generation communication networks tend to bring global connectivity, even in rural areas, disaster areas, etc., where terrestrial base stations are difficult or impossible to develop. For this reason, the aerial platform is considered a compulsory technology for future networks, where the aerial vehicles act as access points from the sky. In this paper, we study a mobile edge computing (MEC)-enhanced aerial serving network scenario that the aerial vehicles, such as drones, unmanned aerial vehicles (UAVs), etc., are flying in the sky to serve remote areas, where have no terrestrial base station. In addition, a high-altitude platform (HAP) equipped with a computing server plays the role of mobile edge computing (MEC) that enhances the performance of the system. In this scenario, we consider a partial offloading scheme, where the aerial vehicles decide to choose the offloading destination and the offloading rate to minimize the total cost function for completing the tasks. Considering network dynamics, we use a deep reinforcement learning (DRL) framework to represent the problem, and propose a deep deterministic policy gradient (DDPG)-based algorithm, named HAMEC, to solve the problem. The experimental results demonstrate that HAMEC outperforms benchmark schemes.