Deep Reinforcement Learning for Congestion Control and Routing Optimization in LEO Satellite Networks

This paper provides a exploration of the application of low earth orbit satellite networks (LSN) in the context of sixth-generation(6G) mobile communication technology. The primary focus of this research is to address issues related to congestion control and routing within satellite networks. We introduces ECSAC-DRO, a novel algorithm that leverages deep reinforcement learning (DRL) to optimize routing and congestion control. Experimental results demonstrate that the ECSAC-DRO algorithm, taking into consideration the high-dimensional characteristics of LSN and incorporating elements from the DRL-based Soft Actor-Critic (SAC) algorithm, excels in providing flexible control over LSN congestion levels and bandwidth utilization. It outperforms traditional methods like Dijkstra, making it a promising and forward-looking solution that lends support to the future development of 6G communication technology.