Low-Delay Transmission for Non-Terrestrial Networks Based on FEC and Reinforcement Learning

Non-terrestrial networks (NTNs) which often contain lossy links with long propagation delay play an important role in the Future 6G network. The transport-layer performance on NTNs links will directly impact user experience. The Transmission Control Protocol (TCP), the dominant transport protocol of the past decades, exhibits poor performance on NTN links due to the low link utilization and the high end-to-end in-order delivery. In this paper, we propose an enhanced framework for the transport layer protocol which is based on reinforcement learning (RL) for congestion control and utilizes a proactive packet-level forward erasure correction (FEC) approach called streaming coding to provide low-delay loss recovery without retransmission at all. By validating in a popular QUIC open-source implementation, namely quic-go, we have demonstrated that the enhanced framework is feasible and it can achieve much shorter end-to-end in-order delivery delay.