Deep-learning-assisted on-chip Fourier transform spectrometer

We proposed and demonstrated a deep learning assisted on-chip Fourier transform spectroscopy (FTS), using an artificial neural networks (ANN) to analyze the output stationary interferogram. It is found that, compared with the conventional FTS, the resolution could be improved without increasing the maximum path length difference and the number of MZIs, thus reducing the burden of adding more power budget. This new concept of enhancing spectral resolution may hold great promise for potential applications in integrated FTS.