An Implicit Representation Network-Based Method for Material Calibration in Digital Twin Channel

Ray-tracing-based digital twin channel holds great promise for future 6G communication systems. However, its accuracy depends heavily on the fidelity of electromagnetic (EM) material parameters, making their precise estimation a key challenge. This paper presents a material EM parameter calibration approach that integrates differentiable physical modeling, implicit neural representations, and high-frequency feature learning. A deep neural network based on sinusoidal activation functions is employed, combined with carefully designed normalization and initialization strategies, to enable efficient inversion and optimization of key material parameters. Simulation and experimental results demonstrate that the proposed method can effectively calibrate EM material parameters based on received signal strength, significantly enhancing the accuracy and predictive capability of digital twin channel modeling.