A DRL Empowered Multipath Cooperative Routing for Ultra-Dense LEO Satellite Networks

Xiaoyu Liu,Ting Ma,Xiaohan Qin,Haibo Zhou,Lian Zhao

DOI: https://doi.org/10.1109/globecom54140.2023.10436959

2023-01-01

Abstract:Nowadays, the ultra-dense low earth orbit (LEO) satellite network has become an attractive solution for providing global Internet coverage and services. With the ever-increasing demand for higher transmission performance, multipath also attracts much attention due to its great potential. In this paper, we consider the multipath cooperative routing in the ultra-dense LEO satellite network. To cope with the high dynamics of the network environment, a deep reinforcement learning (DRL) empowered intelligent routing algorithm is proposed, where each satellite only observes the local network state and independently makes the next-hop forwarding decision. Meanwhile, the perceived conditions of each path are recorded hop by hop in a format-specific packet. In this way, multiple available paths can be found for cooperative transmission. To balance multipath load, an adaptive traffic scheduling scheme is further developed on the sender to adjust traffic distribution according to the varying path conditions, so that multiple sub-flows can be efficiently maintained. Simulation results show the superiority and the effectiveness of the proposed multipath cooperative routing scheme compared with other baseline schemes.