Predicting Networks Before They Happen: Experimentation on a Real-Time V2X Digital Twin

TL;DR

This work presents a real-time V2X Digital Twin framework that combines live mobility data from a city-scale Mobility DT with a high-fidelity ray-traced full-stack simulator (VaN3Twin) to forecast network performance for near-future vehicle interactions. By formalizing a latency-aware end-to-end workflow and introducing a multi-level ray tracing fidelity scheme (RTDI), the authors quantify the trade-offs between model fidelity, computation time, and trajectory prediction horizons. Experimental validation in Tokyo demonstrates timely predictions with RSSI errors around 1 dB on average and reliable LoS transition forecasting within a bounded latency, confirming the practicality of real-time predictive NDTs for urban V2X. The work underscores how end-to-end latency constraints shape the feasible prediction horizon and channel accuracy, establishing a path toward real-time, network-aware decision making in safety-critical V2X scenarios.

Abstract

Emerging safety-critical Vehicle-to-Everything (V2X) applications require networks to proactively adapt to rapid environmental changes rather than merely reacting to them. While Network Digital Twins (NDTs) offer a pathway to such predictive capabilities, existing solutions typically struggle to reconcile high-fidelity physical modeling with strict real-time constraints. This paper presents a novel, end-to-end real-time V2X Digital Twin framework that integrates live mobility tracking with deterministic channel simulation. By coupling the Tokyo Mobility Digital Twin-which provides live sensing and trajectory forecasting-with VaN3Twin-a full-stack simulator with ray tracing-we enable the prediction of network performance before physical events occur. We validate this approach through an experimental proof-of-concept deployed in Tokyo, Japan, featuring connected vehicles operating on 60 GHz links. Our results demonstrate the system's ability to predict Received Signal Strength (RSSI) with a maximum average error of 1.01 dB and reliably forecast Line-of-Sight (LoS) transitions within a maximum average end-to-end system latency of 250 ms, depending on the ray tracing level of detail. Furthermore, we quantify the fundamental trade-offs between digital model fidelity, computational latency, and trajectory prediction horizons, proving that high-fidelity and predictive digital twins are feasible in real-world urban environments.

Figures (5)

• Figure 1: Schematic architecture of the framework.
• Figure 2: End-to-end communication workflow.
• Figure 3: Real picture (a), the corresponding VaN3Twin ray tracing representation (b), and an example of the computed propagation rays (c) of the road segment used in the experiments on the Ookayama campus of the Institute of Science Tokyo, Japan.
• Figure 4: Distribution of the measured end-to-end latency $\tau_{e2e}$ for different Ray Tracing DI configurations.
• Figure 5: Experimental RSSI RMSE (blue) and LoS Prediction Accuracy (green) trends for different trajectory prediction perturbation.