Artificial Intelligence-Assisted Accurate Spectrum Prediction in Design of Terahertz Fiber Operating in 6G Communication Window

Accurate spectrum prediction in design of terahertz (THz) devices remains challenging, especially for THz fibers. In this article, we propose an approach based on artificial intelligence (AI) assisted finite element method (FEM) to achieve accurate spectrum prediction for the design of THz fiber operating in 6G communication window. The antiresonant THz fiber has been selected to verify the effectiveness of this method. Initially, the principle and physical modeling of antiresonant THz fiber are analyzed. The spectra of THz fibers with different structural parameters are designed and predicted by FEM simulation. Then, the THz fibers are fabricated by 3D printing technology and the spectra are measured by a THz time domain spectroscopy system. The spectral dataset obtained by both forms are prepared for the modeling of AI-assisted FEM methods. Different AI algorithms are induced in FEM to predict experimental spectrum, including elman neural network (Elman), support vector machines (SVM), and general regression neutral network (GRNN). The prediction performance obtained by different methods are compared and analyzed comprehensively to confirm the effectiveness of proposed methods. The AI-assisted FEM methods show great improvement of prediction accuracy in the design of THz fibers. It provides accurate data support for THz device modeling assisted by machine learning.