Deep learning enables temperature-robust spectrometer with high resolution

Abstract:

Traditional multi-mode fiber spectrometers rely on algorithms to reconstruct the transmission matrix of the fiber, facing the challenge that the same wavelength can lead to many totally de-correlated speckle patterns as the transfer matrix changes rapidly with environment fluctuations (typically temperature fluctuation). In this manuscript, we theoretically propose a multi-mode-fiber (MMF) based, artificial intelligence assisted spectrometer which is ultra-robust to temperature fluctuation. It has been demonstrated that the proposed spectrometer can reach a resolution of 0.1 pm and automatically reject the noise introduced by temperature fluctuation. The system is ultra-robust and with ultra-high spectral resolution which is beneficial for real life applications.