Energy-Efficient Deep Reinforcement Learning RSMA in Multi-UAV-Assisted Wireless-Powered Communication Network

2025-12-12

PRE

王

王克浩 (Kehao Wang)

Y. J. Sun

刘

刘佩 (Pei Liu)

Yang Zhang

Zhichao Shao

DOI：10.1109/TNSE.2025.3616379

Original

Original request for help

Save

Abstract

En 中文

This paper investigates the rate-splitting multiple access (RSMA) in a multi-unmanned aerial vehicle wireless-powered communication network (MUAVWPCN), where task UAVs (TUAVs) serve as aerial mobile base stations to wirelessly power signal acquisition devices (SADs) deployed in hazardous regions. These SADs subsequently transmit data to ground users (GUs) located outside the affected areas. Aiming to maximize system energy efficiency (EE), a non-convex joint optimization problem is formulated, involving UAV trajectory design, task scheduling, beamforming, and common rate allocation. To address this problem, a energy-efficient multi-agent collaborative optimization (EEMACO) algorithm is proposed, integrating multi-agent proximal policy optimization (MAPPO) with both recurrent neural network (RNN) modules and policy entropy regularization to enhance agent exploration capabilities and extract historical action-state value features, respectively. Simulation results validate the feasibility of the proposed scheduling and trajectory strategies, and comparative analysis confirms that EEMACO-RSMA achieves superior EE performance compared to baseline reinforcement learning algorithms and conventional multiple access schemes.

Keywords:

Rate-splitting multiple access (RSMA)

wireless-powered communication network

energy efficiency (EE)

multi-agent proximal policy optimization (MAPPO)

Energy-Efficient Deep Reinforcement Learning RSMA in Multi-UAV-Assisted Wireless-Powered Communication Network

Abstract

Journal

Organization