Tapered optical fiber spectrometer based on speckle pattern with deep learning

Abstract In this study, a novel tapered fiber-based spectrometer has been proposed to enhance wavelength sensitivity. The performance of the developed spectrometer has been experimentally investigated through tapered and non-tapered configurations. Speckle patterns corresponding to different wavelengths have been recorded using a CCD camera. The sensitivity of the proposed spectrometer to 1 pm wavelength variations in the 1537.995–1538.005 nm range has been evaluated using the ResNet18 architecture. Test accuracy rates reaching up to 96% for the tapered fiber and 80% for the non-tapered configuration have been observed. The increased mode complexity in the tapered region and its wavelength-dependent speckle transformations have enabled the tapered configuration to achieve highly accurate spectral predictions. Consequently, the obtained results have demonstrated that the tapered fiber provides higher accuracy compared to the non-tapered configuration.