Expert Knowledge-Infused Learning for Indoor Radio Propagation Environment Digital Twins

Digital Twin (DT) technology, which enables the simulation, evaluation, and optimization of physical entities through synchronized digital replicas, has attracted increasing attention in the context of wireless networks. Among the various components involved, the radio propagation environment is fundamental to communication performance, making its accurate digital replication a critical challenge. This paper focuses on constructing a high-precision radio propagation environment DT using deep learning (DL) methods. While data-driven DL has become a mainstream solution for signal propagation prediction in DTs, its performance depends heavily on the model's ability to learn intrinsic propagation patterns from data. Owing to the complex interactions between wireless signals and environmental obstacles, conventional DL models often struggle to efficiently capture implicit propagation laws solely from raw data. To address this issue, we propose a general methodology for incorporating expert knowledge of radio propagation into DL frameworks. Building upon the widely adopted encoder-decoder architecture, the proposed approach explicitly integrates theoretical propagation knowledge to enhance learning efficiency and prediction accuracy. Ablation experiments demonstrate that the inclusion of expert knowledge significantly improves the performance of DL-based radio environment DTs. This work highlights the potential of knowledge-data dual-driven DL as a promising direction for advancing radio propagation environment DTs.