CFI-Avoiding Beam Hopping for LEO Satellites in Spectrum Sharing With GEO Systems: A Collaborative Dual-Agent SAC Framework

Abstract

LEO satellites play a critical role in non-terrestrial networks (NTNs), where beam hopping (BH), enabled by phased-array antennas and flexible time-frequency division multiplexing (TFDM), enhances payload utilization. However, existing BH studies for spectrum-sharing LEO–GEO systems often neglect the co-frequency interference (CFI) imposed by LEO downlink beams on GEO user terminals, and lack a coordinated optimization of the strongly coupled BH pattern and spectrum resource allocation. These limitations hinder effective interference mitigation and restrict system capacity. Therefore, this paper proposes a downlink BH method with CFI avoidance for LEO satellites in spectrum sharing with GEO systems. Specifically, a dynamic multi-objective optimization mechanism is designed to jointly optimize system throughput and delay fairness while reducing the CFI on GEO user terminals through coordinating beam pointing and frequency assignment. Furthermore, a dual-agent soft actor-critic (SAC) framework is proposed to collaboratively optimize BH pattern and resource allocation. It balances exploration and exploitation through entropy maximization, enhances training stability via soft target network updates, and mitigates estimation errors using the actor-critic architecture. Simulation results demonstrate that the proposed method reduces CFI on GEO user terminals by 31%, while ensuring that 83% of LEO cells achieve over 90% traffic satisfaction rate, thereby significantly improving system performance and resource utilization efficiency.