Dynamic Time-Expanded Topology Design for LEO Satellite Network

Abstract

The dynamic nature of Low Earth Orbit (LEO) satellite constellations, exemplified by systems like Starlink, Amazon Kuiper, and OneWeb, presents significant challenges in network topology design, impacting parameters such as latency and capacity. This research focuses on exploring Deep Reinforcement Learning (DRL) algorithms for the design of dynamic topologies in LEO satellite networks. Additionally, it investigates the use of the SpaceNet platform and Mininet for conducting LEO satellite network experiments. The proposed approach utilizes DRL algorithms to dynamically adjust satellite links, optimizing for reduced latency and minimized link churn while maximizing network capacity. These algorithms are rigorously tested through real-time simulations on the SpaceNet platform and Mininet, providing a comprehensive evaluation of their performance compared to static and traditional methods. Preliminary results indicate that DRL-based topology design algorithms offer substantial improvements in network performance, including enhanced capacity and lower latency. The study also underscores the effectiveness of the SpaceNet platform and Mininet as valuable tools for simulating and experimenting with LEO satellite networks. This research contributes to the field by providing a detailed investigation into the application of DRL for dynamic topology design and demonstrating the utility of advanced simulation platforms for experimental validation. These insights are crucial for the development of efficient and scalable network topologies in next-generation LEO satellite constellations.