Proactive URLLC Adaptation for Connected Vehicles Through ML-Based Channel Prediction

DNN and LSTM models forecast vehicular channel quality to enable proactive 5G/6G URLLC service adaptation, approaching the ideal perfect-knowledge bound in realistic urban ray-tracing simulations.

Andrea Giovannini; Lorenzo Mario Amorosa; Vittorio Todisco; Claudia Campolo; Antonella Molinaro; Su Hongjia; Alessandro Bazzi· 2026· DOI 10.48550/arXiv.2607.13692

The core problem

Connected and automated vehicles (CAVs) are expected to increasingly depend on 5G and future 6G ultra-reliable and low-latency communication (URLLC) services to support safety-critical and time-sensitive applications. In urban vehicular environments, wireless link conditions can vary rapidly, making it challenging to maintain service continuity and reliability. The paper argues that proactively adapting service parameters based on future channel conditions is essential to avoid performance degradation. To this end, the authors investigate machine learning (ML) techniques for channel quality prediction in vehicular URLLC scenarios. Specifically, they evaluate deep neural network (DNN) and long short-term memory (LSTM) models to forecast future channel conditions and enable proactive service adaptation with minimized performance degradation. The analysis is conducted using realistic simulations that combine the SUMO traffic simulator and the Sionna-RT ray-tracing framework in a real urban environment reconstructed from OpenStreetMap data. The central hypothesis is that ML-based prediction can significantly outperform approaches relying solely on past channel measurements and achieve

Innovation

The simulation results demonstrate that ML-based prediction significantly outperforms the reactive baseline that relies solely on past channel measurements. Both DNN and LSTM models achieve substantial reductions in prediction error compared to the reactive approach. More importantly, the ML-based proactive adaptation yields URLLC performance that is close to the ideal case where future channel conditions are perfectly known in advance. Key quantitative findings include: - The LSTM model generally outperforms the DNN, particularly for longer prediction horizons, due to its ability to capture temporal dependencies in the channel evolution. - The performance gap between ML-based prediction and the ideal oracle is small, indicating that the proposed approach can effectively mitigate the impact of channel variability. - Proactive adaptation based on ML predictions reduces the number of URLLC violations (e.g., latency or reliability outages) compared to reactive schemes. The results are consistent across different vehicle speeds and urban scenarios, highlighting the robustness of the approach. The authors also analyze the impact of the prediction horizon Δ and the input window

Connected and automated vehicles (CAVs) are expected to increasingly depend on 5G and future 6G ultra-reliable and low-latency communication (URLLC) services to support safety-critical and time-sensitive applications. In urban vehicular environments, wireless link conditions can vary rapidly, making it challenging to maintain service continuity and reliability. The paper argues that proactively adapting service parameters based on future channel conditions is essential to avoid performance degradation. To this end, the authors investigate machine learning (ML) techniques for channel quality prediction in vehicular URLLC scenarios. Specifically, they evaluate deep neural network (DNN) and long short-term memory (LSTM) models to forecast future channel conditions and enable proactive service adaptation with minimized performance degradation. The analysis is conducted using realistic simulations that combine the SUMO traffic simulator and the Sionna-RT ray-tracing framework in a real urban environment reconstructed from OpenStreetMap data. The central hypothesis is that ML-based prediction can significantly outperform approaches relying solely on past channel measurements and achieve performance close to the ideal case in which future channel conditions are perfectly known in advance.

The study employs a simulation-based methodology to evaluate ML-based channel prediction for vehicular URLLC. The simulation environment integrates three key components:

Why it matters

The paper provides a compelling case for integrating ML-based channel prediction into vehicular URLLC systems. The use of realistic ray-tracing simulations in a real urban environment strengthens the validity of the findings. The comparison against both a reactive baseline and an ideal oracle offers a clear benchmark for assessing the benefits of prediction. Several implications emerge: - **Proactive resource management**: By anticipating channel degradation, the network can pre-emptively adjust modulation, coding, or redundancy to maintain URLLC requirements. - **Model complexity vs. accuracy**: While LSTM offers better accuracy, DNN may be preferable in scenarios with limited computational resources. The trade-off between prediction horizon and accuracy must be carefully tuned. - **Generalization**: The models are trained and evaluated in a specific urban environment; further work is needed to assess their transferability to other cities and frequency bands. Limitations include the reliance on simulated data and the assumption of perfect knowledge of vehicle trajectories. Future research could explore online learning and integration with real-world testbeds. Overall, the study demonstrates the potential of ML-driven prediction techniques to enhance the reliability and robustness of URLLC services for connected vehicular systems.

Who should read this

CS practitioners and researchers

Opening member content…