Improving the Congestion Control Performance for Mobile Networks in High-Speed Railway via Deep Reinforcement Learning

Due to the poor Transmission Control Protocol (TCP) performance in high-speed mobile scenarios, passengers have bad network experiences on High-Speed Railway (HSR). As a result, improving network performance for HSR scenarios has become urgent and widespread concerns. Some previous works quantitatively analyzed the TCP performance on HSR and proposed relevant solutions. Other works focused on the handover problem (the leading cause of poor network performance on HSR), and proposed a series of handover algorithms for HSR scenarios. However, the existing works are either limited to only measurement studies without algorithm implementation or lack of integration with real-world scenarios. In this paper, with a large amount of field measurement data in real HSR networks, we study the main reasons why traditional TCP performs poorly in HSR scenarios. To improve the TCP performance, we propose Hd-TCP, a customized Congestion Control (CC) algorithm designed to deal with frequent handover on HSR from the transport layer perspective. For the transmission characteristic on HSR, Hd-TCP can accurately evaluate the link condition and apply a Deep Reinforcement Learning (DRL) method to make a fine control. The simulation results show that Hd-TCP outperforms traditional CC algorithms in both throughput and latency by fully utilizing the transmission gap between handovers.

I. Introduction

With the rapid development of the High-Speed Railway (HSR) all over the world, HSR is becoming one of the dominant ways for people to travel. For example, the business mileage of HSR in China already reached 29,000 km by the end of 2018, and is expected to increase 8.06% every year [1]. With more and more people traveling via HSR, the amount of time they will spend on trips indicates a huge demand for network services.