WirelessDT: A Digital Twin Platform for AI-Enabled Wireless Systems

Abstract

Wireless technologies play a critical role in various domains, including communication, pedestrian localization, autonomous vehicles, and healthcare. Their dependence on radio signals for transmitting information exposes them to vulnerabilities from signal fluctuations, influenced by both wireless sensing systems and environmental dynamics (like crowd densities and stochastic events). This aspect complicates the analysis of wireless signals and related applications. Moreover, integrating AI into these systems requires substantial, high-quality data for training deep learning models, but acquiring such data is often laborious and time-consuming. To overcome these challenges, WirelessDT, a groundbreaking digital twin platform, is proposed for simulating and generating wireless signals in dynamic 3D environments in real-time. It combines a high-performance 3D game engine, a GPU-based real-time ray-tracing (RTRT) rendering pipeline, and adaptive learning algorithms, enhancing data fidelity, application efficiency, and scalability. WirelessDT revolutionizes wireless signal propagation simulation using the 3D engine. Its novel application of GPU RTRT for wireless analysis integrates seamlessly with digital twin technology for a cohesive real-virtual environment blend. Furthermore, an innovative learning-based calibration technique viewing the digital twin system as a generative model is presented to fine-tune parameters using realworld data, significantly improving simulation realism and accuracy. These advancements make WirelessDT a vital tool for applications needing interactive simulation and immediate feedback. Comprehensive evaluations confirm WirelessDT's exceptional data accuracy, application efficiency, and scalability performance.