Simulation and experimental validation of the JCR-CNN model for hybrid modes recognition and decomposition in few-mode fibers

Abstract

The issue of mode decomposition (MD) in optical fibers is regarded as a regression problem has been widely studied and applied, however, in contrast, the identification of different classes of hybrid modes groups when different modes are excited as a typical classification problem has rarely been studied, and the possibility of incorporating both tasks into a single neural network to be solved simultaneously is an important issue that deserves to be explored. Thus, a novel joint classification and regression task convolutional neural network (JCR-CNN) architecture is proposed, which can perform the discrimination of hybrid mode group categories in few-mode fibers (FMFs) and predict corresponding mode coefficients simultaneously. In addition, the introduction of the data augmentation (DA) technique establishes a tight link between experimental and synthetic simulation data, allowing the model trained only on simulation data to cope well with the acquisition errors in the experimental optical path configuration when making predictions based on experimental data. The quantitative results of both simulation and experimental data demonstrate the feasibility and validity of the proposed model and DA method. The mode classification accuracy and the Pearson coefficient of the image reached 99.71% and 98.81%, respectively. To the best of our knowledge, this work provides a new perspective for a more comprehensive characterization of hybrid modes in optical fibers, which is expected to have applications in various fields.