Machine learning approach to predict the optical properties of the superposition-assisted structure of homogeneous multi-core fiber

In order to harness the full potential of the increased data carrying capacity, the concept of multi-core fibers (MCFs) was born, i.e., to have multiple cores in a single fiber. MCFs are engineered with superposition-assisted structures (SASs) that allow the control of mode coupling and interference. This development would enable high-throughput communication, high-power beam delivery, and increased efficiency and functionality in nonlinear photonics. This work applies machine learning methodologies to predict key optical parameters of solid-core SAS-MCFs, including effective refractive index, mode effective area, inter-core crosstalk, and confinement loss. Artificial neural networks (ANNs) are trained on simulation datasets spanning relevant design ranges: wavelengths from 0.6 to 1.95 μm , structural variations like relative refractive index (RI) of core-clad, Δ 1 between 0.5% and 1%, relative RI of clad-trench, Δ 2 between 0.98% and 0.34%, and fiber geometries comprising 14 cores arranged in two concentric rings. In contrast to traditional numerical solvers, our proposed direct neural network architecture enables rapid and accurate predictions for unseen fiber configurations, significantly reducing both computational time and the need for large-scale training datasets. The accuracy and efficiency of the proposed machine learning model are validated by comparing its results to those obtained from full-vector simulations using COMSOL Multiphysics. The findings confirm that the ANN-based approach offers a robust and computationally efficient tool for the optical characterization and design optimization of SAS-MCFs.