Electromagnetic Environment Knowledge Construction and Channel Prediction: A Model-Data Dual-Driven Approach Toward 6G Digital Twin Channel

Abstract

To meet the urgent demand for high-fidelity and interpretable channel modeling in sixth-generation (6G) wireless communication networks, digital twin channel (DTC) has emerged as a key enabling technology for achieving intelligent environmental perception and dynamic interaction. However, existing DTC implementation methods mostly rely on pure data-driven approaches, facing challenges such as unclear correlation between environmental features and channel parameters, redundant environmental information input, and limited scene generalization capabilities. To address these issues, this paper proposes a novel model-data dual-driven framework for 6G DTC, which consists of model-driven electromagnetic environment knowledge (EEK) construction and data-driven channel prediction. For the former, we first detect effective scatterers from the three-dimensional environmental data by the designed lightweight Transformer-based network. Based on electromagnetic wave propagation characteristics, we then capture reflection and diffraction points in single-bounce propagation paths, as well as representative multi-bounce propagation paths. By quantifying their channel contributions, an interpretable EEK that reveals the intrinsic environment-channel correlation is constructed. For the latter, a novel physics-enhanced graph attention network (PE-GAT) is designed for channel prediction. Specifically, it takes the EEK graph as input, explicitly injecting its physical connotation as learnable attention biases into the message-passing mechanism, thereby guiding the model to reason based on physical principles. Simulation results validate the proposed methods’ effectiveness in enhancing environment–channel relationship modeling and prediction accuracy, providing a methodological foundation for further validation and development toward adaptive and intelligent DTC modeling.