Machine Learning-based Inverse Model for Few-Mode Fiber Designs

Abstract

The medium for next-generation communication is considered as fiber for fast, secure communication and switching capability. Mode division and space division multiplexing provide an excellent switching capability with high data transmission rate. In this work, the authors have approached an inverse modeling technique using regression-based machine learning to design a weakly coupled few-mode fiber for facilitating mode division multiplexing. The technique is adapted to predict the accurate profile parameters for the proposed few-mode fiber to obtain the maximum number of modes. It is for a three-ring-core few-mode fiber for guiding five, ten, fifteen, and twenty modes. Three types of regression models namely ordinary least-square linear multi-output regression, k-nearest neighbors of multi-output regression, and ID3 algorithm-based decision trees for multi-output regression are used for predicting the multiple profile parameters. It is observed that the ID3-based decision tree for multioutput regression is the robust, highly-accurate machine learning model for fast modeling of FMFs. The proposed fiber claims to be an efficient candidate for the next-generation 5G and 6G backhaul networks using mode division multiplexing.