Deep reinforcement learning-based fountain coding for concurrent multipath transfer in high-speed railway networks

Concurrent multipath transfer (CMT) has been proved to significantly improve the end-to-end throughput with its multihoming property. However, due to the extremely high unpredictability around high-speed railway (HSR) environment, the receive buffer blocking problem seriously degrades the overall transmission reliability. To address this issue, this paper proposes a learning-based fountain coding for CMT (FC-CMT) scheme to mitigate the negative influence of the path diversity of HSR networks. Specifically, we first formulate a multi-dimensional optimal problem to mitigate receive buffer blocking phenomenon and improve the transmission rate with requirement constrains. Then, we transform the data scheduling and redundancy coding rate problem into a Markov decision process, and propose a deep reinforcement learning (DRL)-based fountain coding algorithm to dynamically adjust data scheduling policy and redundancy coding rate. We conduct the extensive experiments in a P4-based programmable network platform. Experimental results indicate the proposed algorithm mitigates the packet out-of-order problem, and improves the average throughput compared with traditional multipath transmission scheme.