Federated learning based energy efficient scheme for MEC with NOMA underlaying UAV

Abstract

Unmanned Aerial Vehicle (UAV) enabled Mobile Edge Computing (MEC) brings the on-demand task computation services close to the user equipment (UE) by reducing the latency and enhancing the quality-of-service (QoS). However, the energy consumption remains a major issue in the system, since both mobile devices (MDs) and UAVs have limited power battery storage. Also in 5G and beyond 5G (B5G) networks, in which UEs' task requests and positions change frequently, stationary edge network implementation may increase the overall energy consumption. This article aims to minimize the overall energy consumption for MEC with Non-Orthogonal Multiple Access (NOMA) underlaying UAV systems. We have used Markov decision process (MDP) to convert the optimization problem into multi-agent reinforcement learning (MARL) problem. Then to achieve optimal policy and reduce the overall energy consumption of the system, we propose a multi-agent federated reinforcement learning (MAFRL) scheme. Simulation results show the effectiveness of the proposed scheme in reducing the overall energy consumption with respect to other state-of-art schemes.