Optical Fiber Parameters Determination Using Neural Network with Noise Analysis

The objective of this work is to develop and analyze a method based on the use of a convolutional neural network (CNN), which would be capable of inferring the most relevant physical parameters of an optical fiber—specifically, the second-order group velocity dispersion (GVD) coefficient β2 and the nonlinearity coefficient γ—from the observation of temporal profiles of propagated optical pulses. We analyze the model robustness in the presence of noise, a critical aspect for any potential experimental application. In particular, as the main result, we explore how different training schemes affect the model’s ability to determine the aforementioned parameters.