AdaPTwin digital twin boosts vehicular network speed by 90%

AdaPTwin, developed by Armin Makvandi and colleagues, introduces an adaptive multi-fidelity predictive digital twin (NDT) for proactive radio resource management (RRM) in vehicular networks. Unlike existing digital twins that use fixed fidelity levels, AdaPTwin dynamically adjusts its simulation complexity based on current network conditions and latency constraints. The system employs a hierarchical cloud-edge architecture: the cloud periodically performs computationally intensive fidelity selection, while the edge handles real-time proactive RRM. This edge loop combines vehicle trajectory prediction—using a transformer model enhanced with continual and transfer learning to adapt to new environments—with look-ahead ray tracing executed via NVIDIA Sionna on a dynamically updated virtual environment. The framework then solves a joint RSU beamforming and vehicle-RSU association problem using a scalable multi-start iterative coordinate descent algorithm to maximize proportionally fair sum-rate.

AdaPTwin was rigorously tested against reactive, single-fidelity, and non-adaptive predictive NDTs under realistic vehicular conditions. Results show it adapts successfully in diverse scenarios where other frameworks fail. Specifically, AdaPTwin achieves up to a 90% sum-rate gain and an 80% reduction in outage probability compared to non-adaptive NDTs, all while maintaining real-time performance. The work has been submitted to IEEE for publication and is currently available on arXiv. By eliminating the latency penalty of traditional ray-tracing-based digital twins, AdaPTwin opens the door to truly proactive, site-specific RRM that can keep pace with the highly dynamic nature of connected and autonomous vehicles.