Prediction of confinement loss, mode field diameter, and effective index in anti-resonant fibers using deep learning

We use convolutional neural networks to predict the optical characteristics of anti-resonant optical fibers, including confinement loss, mode field diameter, and effective index. Configuration of anti-resonant fiber (ARF) using discretized sampling and of a dataset correlating the structure with three optical characteristics is constructed. The constructed neural network demonstrates high predictive accuracy and excellent generalization ability. Compared with finite element method calculations of these optical characteristics, the model significantly reduces computation time by nearly four orders of magnitude. This research can be used for the reverse design of the configuration of ARF and is also expected to combine with image processing technologies to assist in the rapid measurement of the configuration of ARF.