Energy-Efficiency Optimization of Active Flying-RIS-Assisted Mobile-Edge Computing Networks: A Deep-Reinforcement-Learning Approach

This article explores the energy efficiency (EE) of a mobile-edge computing (MEC) architecture for Internet of Things (IoT) networks, in which multiple IoT devices (IoTDs) perform local computations while further offloading part of their tasks to the base station (BS)-enabled MEC server utilizing the time division multiple access (TDMA) protocol. To address the radio propagation issues caused by obstructions, we utilize an uncrewed aerial vehicle (UAV)-mounted active reconfigurable intelligent surface (RIS), referred to as a flying-RIS (FRIS), to reflect and even amplify the incident signal from IoTDs to the BS. Since tasks can be executed in parallel on both IoTDs and the BS, the assignment of service timeslots, the UAV trajectory, and the resource allocation for both IoTDs and FRIS should be jointly optimized to maximize system EE. To this end, we reformulate the optimization problem as a markov decision process (MDP) and introduce a deep reinforcement learning (DRL) approach for addressing the formulated problem, called the proximal policy optimization (PPO) based resource allocation with trajectory design and FRIS reflection matrix optimization (PPO-RATDFRO) algorithm. By continuously interacting within the constructed environment, the proposed system iteratively refines its policy to determine the FRIS trajectory and reflection matrix, along with the service timeslots and computation resources for each IoTD. Finally, simulation results demonstrate that the proposed PPO-RATDFRO algorithm significantly enhances EE for all served IoTDs, compared to various benchmark algorithms.