應用於第五代非獨立地鐵通訊網路之基於換手預測的 QUIC 強化設計

Abstract

對於基於 B5G 或 6G 的自動化鐵路通訊系統而言，同時滿足乘客的安全需求與個人偏好至關重要，而這通常涉及大量資料檔案的傳輸。然而，當列車高速行駛時，5G 非獨立組網中的頻繁換手事件可能嚴重影響傳輸效能，導致封包遺失率與延遲增加。

為了解決此挑戰，我們首先在台北捷運的 5G 網路環境中進行了大量實地測量，分析各類換手事件對資料傳輸的影響。透過這些分析，我們找出幾個在換手期間對傳輸效能影響最顯著的關鍵成功因子。

基於這些洞察，我們設計了一套邊緣運算系統，整合無線電鏈路失效（Radio Link Failure, RLF）預測模型以及增強型壅塞控制演算法 Reno-HO。該系統能根據預測到的 RLF 動態調整壅塞視窗，以維持傳輸穩定性，並提升吞吐量與實際有效傳輸率。

我們透過實際場域實驗與網路模擬驗證了所提出的方法。結果顯示，此方法能有效減緩換手事件所造成的負面影響，並顯著提升高速鐵路環境中的資料傳輸效能。

For an automatic railway communication system based on Beyond 5G (B5G) or 6G, satisfying passengers safety issues and preferences are both crucial, which involves large data file transmission. However, when trains operate at high speed, frequent handover events in 5G NSA networks can severely degrade transmission performance, increasing both packet loss and latency. To address this challenge, we first conducted extensive measurements on the Taipei Metro 5G network to analyze the impact of various handover events on data transmission. Through this analysis, we identified key success factors (KSFs) that most strongly influence performance during handovers. Based on these insights, we designed an edge computing system that incorporates a Radio Link Failure (RLF) prediction model and an enhanced congestion control algorithm, Reno-HO. By dynamically adjusting the congestion window (CWND) in response to predicted RLFs, our system aims to maintain transmission stability and improve throughput and goodput. We validated the proposed approach through both real-world experiments and network emulation. The results demonstrate that our method can effectively mitigate the negative effects of handovers and significantly enhance data transmission performance in high-speed railway environments.