from the conferences organized by TANGER Ltd.
The ever-growing need for increasing bandwidth of telecommunication networks forces engineers and scientists to develop new solutions, concepts and devices for data transfer. This project aims to determine the best possible structures of few-mode optical fibers for Mode-Division Multiplexing using neural networks of different kinds. Applying the Generative Inverse Design Networks (GIDNs) approach, we studied optimization of the three-mode fibers composed of two types of glass rods: pure SiO2 and SiO2 doped with 5% of GeO2. We compared principles of operation and efficiency of the dense neural network (DNN) and the convolutional neural network (CNN) within two cycles of GIDNs. We observed that convolutional network outperformed the dense one just after first cycle achieving higher average as well as maximal values of minimal separations between modes. The best structures obtained with the DNN were characterized by minimal separations exceeding 1.98×10-3, while the best fiber optimized with the CNN had a minimal separation of ca. 2.11×10-3. Moreover, the latter was applied for the first time for the free-form fibers optimization and obtained results encourage to continue research using the CNN as a main numerical tool.
Keywords: Few-Mode Fibers, Convolutional neural network, Dense neural network, Free-Form Nanostructured optical fibers© This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.