A Reinforcement Learning-based Multipath Scheduling for Heterogeneous Wireless Networks

With the development and commercialization of new mobile network generations such as 5G and beyond, future communications are shifting from the traditional single-path paradigm to multipath transport protocols such as MPTCP and MPQUIC. One of the most critical issues in dealing with the multipath transmission is appropriately scheduling the pathways in order to guarantee QoS. Despite the fact that tremendous effort has been put into developing multipath scheduling algorithms, existing approaches suffer from several limitations when dealing with the network's dynamicity, including congestion and packet loss. In this paper, we propose a novel Reinforcement learning-based multipath transport protocol named SATO, which efficiently schedules multipath communication in heterogeneous wireless networks. By leveraging the self-learning ability of reinforcement learning, a node equipped with SATO can capture the environmental changes and select transmission paths based on an appropriate policy to optimize QoS. Our evaluation results show that SATO improves the QoS by 10%-15% in simulation and 12% in a real deployment compared to the state-of-the-art algorithm.