Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning-based Radio Link Failure Prediction Based on Measurement Dataset in Railway Environments

Po-Heng Chou, Da-Chih Lin, Hung-Yu Wei, Walid Saad, Yu Tsao

TL;DR

This work introduces a measurement-driven framework for early radio link failure prediction in 5G NSA railway environments, leveraging 10 Hz field measurements (RSRP/RSRQ and signaling) to proactively trigger reliability enhancements. By benchmarking six models—CNN, LSTM, XGBoost, Anomaly Transformer, PatchTST, and TimesNet—across multiple observation windows and prediction horizons, it demonstrates that deep temporal models can predict failures several seconds ahead with high accuracy, particularly TimesNet at $T_s=3$ s and $T_p=3$ s. The study provides practical insights into deployment points, thresholding, and trade-offs between context and latency, highlighting that RLFs are predominantly impacted by handover-related events and can be anticipated to enable redundancy and adaptive handovers. Overall, the results offer a viable path to early-warning control in 5G-based railway systems, with implications for edge-enabled, transfer-learning enabled, and multi-cell deployments in future work.

Abstract

In this paper, a measurement-driven framework is proposed for early radio link failure (RLF) prediction in 5G non-standalone (NSA) railway environments. Using 10 Hz metro-train traces with serving and neighbor-cell indicators, we benchmark six models, namely CNN, LSTM, XGBoost, Anomaly Transformer, PatchTST, and TimesNet, under varied observation windows and prediction horizons. When the observation window is three seconds, TimesNet attains the highest F1 score with a three-second prediction horizon, while CNN provides a favorable accuracy-latency tradeoff with a two-second horizon, enabling proactive actions such as redundancy and adaptive handovers. The results indicate that deep temporal models can anticipate reliability degradations several seconds in advance using lightweight features available on commercial devices, offering a practical path to early-warning control in 5G-based railway systems.

Learning-based Radio Link Failure Prediction Based on Measurement Dataset in Railway Environments

TL;DR

This work introduces a measurement-driven framework for early radio link failure prediction in 5G NSA railway environments, leveraging 10 Hz field measurements (RSRP/RSRQ and signaling) to proactively trigger reliability enhancements. By benchmarking six models—CNN, LSTM, XGBoost, Anomaly Transformer, PatchTST, and TimesNet—across multiple observation windows and prediction horizons, it demonstrates that deep temporal models can predict failures several seconds ahead with high accuracy, particularly TimesNet at s and s. The study provides practical insights into deployment points, thresholding, and trade-offs between context and latency, highlighting that RLFs are predominantly impacted by handover-related events and can be anticipated to enable redundancy and adaptive handovers. Overall, the results offer a viable path to early-warning control in 5G-based railway systems, with implications for edge-enabled, transfer-learning enabled, and multi-cell deployments in future work.

Abstract

In this paper, a measurement-driven framework is proposed for early radio link failure (RLF) prediction in 5G non-standalone (NSA) railway environments. Using 10 Hz metro-train traces with serving and neighbor-cell indicators, we benchmark six models, namely CNN, LSTM, XGBoost, Anomaly Transformer, PatchTST, and TimesNet, under varied observation windows and prediction horizons. When the observation window is three seconds, TimesNet attains the highest F1 score with a three-second prediction horizon, while CNN provides a favorable accuracy-latency tradeoff with a two-second horizon, enabling proactive actions such as redundancy and adaptive handovers. The results indicate that deep temporal models can anticipate reliability degradations several seconds in advance using lightweight features available on commercial devices, offering a practical path to early-warning control in 5G-based railway systems.

Paper Structure

This paper contains 22 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Measurement-based Reliability Studies
  4. Device-Side Reliability Enhancement
  5. Learning-Based Reliability Prediction and Anomaly Detection
  6. Measurements and Observations
  7. Experimental Environment and Tools
  8. Observations
  9. RLF Prediction Design
  10. Performance Results
  11. Implementation Environment and Model Setup
  12. Evaluation Setup
  13. Overall Trends Across Horizons
  14. Model-wise Comparison
  15. Effect of Observation Window $T_s$
  16. ...and 7 more sections

Figures (3)

  • Figure 1: Downlink packet loss distribution by event types and occurrence frequency.
  • Figure 2: RLF predictor architecture using data from observation window $T_s$ to predict events in prediction horizon $T_p$.
  • Figure 3: Prediction hits of the top-3 models based on Tables \ref{['tab:performance3']}. Profiles use $T_s = 3$ s and $I_s = 0.1$ s; CNN and XGBoost are shown with $T_p = 2$ s, and TimesNet with $T_p = 3$ s. Red markers indicate successful predictions, i.e., time slots where an RLF was predicted to occur within the next $T_p$ seconds.