Inverse design and optimization of an aperiodic multi-notch fiber Bragg grating using neural networks

Recent developments in the application of aperiodic fiber Bragg gratings (AFBGs) in astrophotonics, such as AFBG for astronomical near-infrared OH suppression and gas detection based on cross-correlation spectroscopy, have illuminated the problem that the optimization for AFBG with certain fabrication constraints has not been fully investigated and solved. Previous solutions will either sacrifice part of the spectral features or consume a significant amount of computation resources and time. Inspired by recently successful applications of artificial neural networks (ANNs) in photonics inverse design, we develop an AFBG optimization approach employing ANNs in conjunction with genetic algorithms (GAs) for the first time, to the best of our knowledge. The approach maintains the spectral notch depths and preserves the fourth-order super-Gaussian spectral features with improvements of interline loss by ∼100 times. We also implement, to our knowledge, the first inverse scattering neural network based on a tandem architecture for AFBG, using a first-order Gaussian notch profile. The neural network successfully converges but has a poor predictive capability for the phase part of the design. We discuss possible ways to overcome these limitations.