Design of the pulse compression fiber taper based on deep learning

In this study, we propose a design method for pulse compression fiber taper, enabling rapid development of compression tapers tailored to the output characteristics of front-end lasers. By utilizing a deep neural network to predict the cross-sectional characteristics of the Pb-silicate photonic crystal fiber, and globally searching based on the output of the previous fiber and the fundamental soliton condition, the structural parameters of the next fiber are calculated to achieve continuous self-similar pulse compression. In a 0.276 m fiber, we successfully compressed a 1.05 ps pulse to 49.1 fs with a compression factor of 21.55. This study provides a fast and feasible solution for the generation of femtosecond pulses, and the proposed method has important reference significance for the design of irregular optical waveguides.