Energy Utilization Maximization for RSMA-Assisted FANETs: A Hierarchical Deep Reinforcement Learning Approach

Abstract

Flying ad-hoc networks (FANETs) offer great potential in agriculture, logistics, and surveillance due to their flexible deployment. However, high mobility poses significant challenges in energy management. To address this, this study focuses on FANETs using rate-splitting multiple access (RSMA) and proposes a multi-agent cooperative approach to maximize system energy utilization (EU) through joint optimization of spectrum allocation and power control under quality-of-service (QoS) and energy constraints. Since the optimization problem is non-convex, it is decomposed into two interrelated subproblems within a hierarchical deep reinforcement learning (DRL) framework: the upper layer handles resource block (RB) scheduling, and the lower layer performs fine-grained power control. Centralized training with decentralized execution enables efficient multi-agent collaboration. Simulation results validate the superiority of the proposed method in EU.