AI-assisted spectrometer based on multi-mode optical fiber speckle patterns
Fu Feng, Jiaan Gan, P. F. Chen, Wei Lin, GuangYong Chen, Changjun Min, Xiaocong Yuan, Michael G. Somekh
Abstract
Speckle patterns at the output end of a multi-mode fiber (MMF) are generated by the interference of eigenmodes propagating along the fiber. A MMF can support thousands of eigenmodes, and the amplitude and phase of each eigenmode changes rapidly with the properties of input light, thus changing the speckle that emerges at the output. Therefore, a MMF can provide ultra-high sensitivity to the changes of the properties of the input light. However, environmental fluctuations of parameters such as temperature, humidity, and vibration limit the applicability of MMF sensors. In this paper, we theoretically and experimentally demonstrate an AI-assisted MMF spectrometer, and examine the limits of the method. The resolution of the spectrometer reached 0.1 picometer (pm, the resolution of the laser source) for linearly polarized monochromatic incidence. The integration of a deep learning algorithm effectively eliminated the perturbation from environmental fluctuations and made the spectrometer ultra-robust. At the end, we have also demonstrated that by using parallel network configuration, the bandwidth of the system can be easily increased without sacrificing the resolution of the system. The AI-assisted fiber spectrometer is compact, super-robust, and low-cost and provides ultrahigh resolution.