Multi‐agent reinforcement learning based transmission scheme for IRS‐assisted multi‐UAV systems

Abstract

AbstractIn this paper, a transmission scheme based on multi‐agent reinforcement learning for intelligent reflecting surface (IRS)‐assisted multiple unmanned aerial vehicles (UAVs) systems is proposed. The proposed scheme is based on reinforcement learning and alternating optimization algorithm, which can effectively improve communication quality and ensure fairness. The scheme is divided into two parts. In the first part, the multi‐UAV cooperation problem is modeled as a markov decision process. The objective of each UAV is to maximize the minimum user channel gain. To achieve stable strategies for all agents, the Multi‐agent Deep Deterministic Policy Gradient (MADDPG) algorithm is applied to train UAVs trajectories to reach the Nash equilibrium. The MADDPG algorithm is centralized trained at the base station and executed in a distributed manner by each UAV, ensuring efficient and effective coordination among agents. In the second part, an alternating optimization algorithm is formulated to optimize active and passive beamforming. Considering the non‐convexity of the fairness objective, by using auxiliary variables and semi‐definite relaxation method, the problem of maximizing the minimum user achievable rate is transformed into a feasibility problem. Simulation results show that the proposed scheme can effectively train UAVs trajectories and improve the communication performance of all users fairly.