5G 非獨立雙無線電網路中的無線連結失效預測與自適 應頻段鎖定技術

Abstract

穩定的無線連線對於鐵路運行的安全與效率至關重要。在高移動性的 5G 非獨立組網環境中，通訊經常遭遇無線連結失敗（Radio Link Failures，RLF）與頻繁的切換（Handovers，HOs），導致傳輸層延遲增加與封包遺失。透過在地鐵與高速鐵路網路的廣泛實地測量，我們清楚觀察到這些挑戰。為了解決這些問題，我們提出 DBLC（雙頻鎖定與控制），一種結合機器學習為基礎的 RLF 預測與動態頻段切換的機制，以提升連線穩定性。我們開發了一個模擬器，透過重播真實數據機記錄並模擬傳輸層損傷，精準反映實地行為，對 DBLC 進行驗證。模擬與實際測試均證明，DBLC 相較於傳統單無線電及雙無線電方案，在縮短斷線時間與提升封包傳送率方面具有明顯優勢。本研究突顯了結合控制平面感知與數據驅動預測，提升安全關鍵行動通訊性能的優勢。

Reliable wireless connectivity is critical for safe and efficient railway operations. In high-mobility 5G Non-Standalone environments, communication often suffers from Radio Link Failures (RLFs) and frequent handovers (HOs), causing increased latency and packet loss at the transport layer. Based on extensive field measurements across metro and high-speed rail networks, we observe these challenges clearly. To tackle them, we propose DBLC (Dynamic Band Locking), a mechanism that combines machine learning-based RLF prediction with dynamic band switching to improve connection stability. We validate DBLC through an emulator that replays real modem logs and simulates transport-layer impairments, accurately reflecting field behaviors. Both emulation and real-world tests demonstrate DBLC’s superiority over traditional single- and dual-radio approaches by reducing disconnection duration and improving packet delivery. This study highlights the advantage of integrating control-plane awareness with data-driven prediction to enhance mobile communication in safety-critical contexts.