Abstract

Much recent work has applied machine learning to nonlinear fiber optics in areas such as fiber laser control [1] and non-linear Schrödinger equation (NLSE) emulation [2]. Other studies have shown how neural networks can detect temporal extreme events based only on analyzing patterns in the corresponding spectral intensity. This has been demonstrated for modulation instability (MI) breakup of picosecond pulses, Raman soliton ejection in supercontinuum generation, and random solitons in lasers [3]. These previous works, however, have all been performed on systems including higher order effects where one might expect highly distinct spectral and temporal features. Here we report that such correlations can also be identified by neural networks in ideal continuous wave MI, the canonical case of unstable nonlinear fiber propagation [4].

© 2023 IEEE