Title: Leveraging Measurements to Forecast Reliability Failures in 5G Non-Standalone Railway Networks

Abstract:

This research offers a data-centric investigation into the early detection of reliability breakdowns within 5G non-standalone (NSA) railway infrastructures. By analyzing 10 Hz measurement traces from metro trains—incorporating indicators for both serving and neighboring cells—we evaluate the performance of six prominent machine learning models: CNN, LSTM, XGBoost, Anomaly Transformer, PatchTST, and TimesNet. The assessment is conducted across various observation windows and prediction horizons.

Instead of introducing a novel predictive architecture, this work establishes a measurement-based benchmark designed to assess the viability and operational compromises associated with predicting reliability issues seconds in advance within 5G NSA railway contexts. Our findings demonstrate that these learning algorithms can successfully forecast radio link failure (RLF)-induced reliability breakdowns several seconds prior to occurrence, relying solely on lightweight radio features accessible via commercial devices. This benchmark yields valuable insights for sensing-assisted communication management and lays an empirical groundwork for embedding sensing and analytics into future mobility control systems.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC