Spectrum-Aware Mobile Edge Computing for UAVs Using Reinforcement Learning

We consider the problem of task offloading by unmanned aerial vehicles (UAV) using mobile edge computing (MEC). In this context, each UAV makes a decision to offload the computation task to a more powerful MEC server (e.g., base station), or to perform the task locally. In this paper, we propose a spectrum-aware decision-making framework such that each agent can dynamically select one of the available channels for offloading. To this end, we develop a deep reinforcement learning (DRL) framework for the UAVs to select the channel for task offloading or perform the computation locally. In the numerical results based on deep Q-network, we con-sider a combination of energy consumption and task completion time as the reward. Simulation results based on low-band, mid-band, and high-band channels demonstrate that the DQN agents efficiently learn the environment and dynamically adjust their actions to maximize the long-term reward.