Hybrid Table-Assisted and RL-Based Dynamic Routing for NGSO Satellite Networks

This letter investigates dynamic routing in Next-Generation Satellite Orbit (NGSO) constellations and proposes a hybrid strategy that combines precomputed routing tables with a Deep Q-Learning (DQL) fallback mechanism. While fully RL-based schemes offer adaptability to topology dynamics, they often suffer from high complexity, long convergence times, and unstable performance under heavy traffic. In contrast, the proposed framework exploits deterministic table lookups under nominal conditions and selectively activates the DQL agent only when links become unavailable or congested. Simulation results in large-scale NGSO networks show that the hybrid approach consistently achieves higher packet delivery ratio, lower end-to-end delay, shorter average hop count, and improved throughput compared to a pure RL baseline. These findings highlight the effectiveness of hybrid routing as a scalable and resilient solution for delay-sensitive satellite broadband services