Spectrum Efficiency Optimization in 6G Satellite Networks: An Inverse Reinforcement Learning Approach
Year: 2025
Type: conference-paper
Abstract: In the evolving landscape of 6G satellite-based non-terrestrial networking, optimizing network resources presents significant challenges, particularly due to the limitations of conventional reinforcement learning (RL) methods. Existing RL approaches often rely on manually designed reward functions, which require extensive parameter tuning that fails to adapt effectively to dynamic network conditions. To overcome this challenge, we propose a novel framework leveraging generative adversarial imitation learning (GAIL) with inverse reinforcement learning (IRL) to automate reward function design and enhance policy learning for beamforming and spectrum allocation. Our approach addresses the non-convex optimization challenge of maximizing spectrum efficiency (SE) while ensuring minimum information rate requirements for remote user equipment (RUEs). By integrating the Whale optimization algorithm (WOA) to generate expert policies, we provide high-quality training data for the GAIL framework, which enables intelligent, data-driven adaptation to network dynamics. This innovative methodology allows satellite agents to autonomously learn optimal strategies through iterative interactions with the environment that significantly advance traditional network optimization paradigms. The simulation results validate the effectiveness of the proposed approach which demonstrates a notable 14.6% improvement in convergence speed and reward values over traditional RL technique.
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4413967590