Deep Reinforcement Learning-Based Routing Optimization for Software-Defined Satellite Networks

The integration of satellite internet with 6G tech-nology is propelling advancements in global communication networks, with satellite systems playing a pivotal role in achieving global seamless coverage. Software-Defined Networking (SDN) presents a promising solution for satellite routing by separating the control and data planes, thus enhancing routing flexibility and network management. This approach, combined with the capabilities of machine learning, offers greater adaptability in handling dynamic network conditions. Specifically, Deep Rein-forcement Learning (DRL) optimizes real-time routing decisions by processing real-time state data such as congestion, latency, and packet loss. Despite the potential of SDN in satellite networks, several challenges remain, including maintaining stable commu-nication links in the face of continuous satellite movement and high communication delays. To address these issues, we propose a novel SDN architecture for satellite networks, coupled with a Deep Q-Network (DQN) algorithm. Our architecture enhances routing efficiency by dynamically adjusting routing strategies based on real-time network conditions. Experimental validation demonstrates the superior performance of our method, partic-ularly in scenarios involving sudden link outages, highlighting its potential to improve the reliability and efficiency of satellite networks.