Learnable Wireless Digital Twins: Reconstructing Electromagnetic Field With Neural Representations

Wireless digital twins, which approximate the communication environment and signal propagation, find many applications in system design and operation, and are envisioned as key enablers for future wireless systems. Building these digital twins, however, is challenging as it requires modeling, not just the environment geometry, but also the electromagnetic (EM) property and the EM interaction effects. To address these challenges, motivated by the latest advancements in computer vision (3D reconstruction and neural radiance field), this paper develops a physics-inspired and modular deep learning framework for practically constructing learnable wireless digital twins. In particular, the proposed framework segments the environment into objects and learns their EM properties and the EM interaction effects using crowd-sourced world-locked wireless channel samples. To showcase the efficacy of the proposed learnable digital twin framework, we evaluate the digital twin by predicting wireless channels based on users’ positions. Simulation results demonstrate that the proposed learnable digital twin can learn the EM property and interaction effects, accurately predict wireless channels, and generalize to changes in the environment. This highlights the prospect of this novel direction for future generation wireless platforms.