Diseño inverso de una rejilla de fibra óptica de bragg uniforme mediante redes neuronales

Abstract

The design of photonic devices is challenging due to it is mathematical complexity, even more so when inverse designs are desired. This article proposes the inverse design of a uniform FBG with low computational load by training a Feed Forward neural network through their spectrums and predict their length and modulation index. The results show that with a single hidden layer of 80 neurons, a set of 1000 spectrums and the Fletcher-Powell Conjugate Gradient training function, it is able to correctly predict up to 99.5% of the samples.