부산대학교 도서관

As an essential candidate frequency band for next-generation communications, the terahertz band has gradually become a research hotspot. Due to wavelength comparable to surface undulation, scattering modeling becomes the core of characterizing the propagation mechanism. This paper proposes a deep learning model to learn the generation paradigm of surface currents (SC) under multiple input factors. Then, based on the generated SC, we use the dyadic Green's function to reconstruct the scattered electric field in the incidence plane and scattering hemisphere space. Experiments, among which the root mean square error of SC and electric field amplitude draws to 1.33~4.41 dB and 1.55~6.10 dB respectively, demonstrate the excellent generation ability of the proposed model and the feasibility of scattering reconstruction based on induced currents. This deterministic modeling approach can generate SC distribution and scattering beams of a specific surface more efficiently than full-wave simulation, providing a novel direction for scattering modeling.