Wavelength detection technique of overlapping spectra in the serial WDM FBGs by convolutional neural network

Abstract

Wavelength detection technique of overlapping spectra of a parallel WDM FBG sensor network by deep learning methods has been studied. But it is difficult to the situation of serial WDM FBG sensor network. For FBGs cascaded in one channel, the training data set cannot be effectively constructed. In this paper, superimposed FBGs were proposed to construct the training data set. One FBG in the superimposed FBGs is used to construct the overlapped spectral data set, and the other one is used to mark the central wavelength. It provides a reliable and sufficient training data set for the demodulation of convolutional neural network-based overlapped spectra. Then, the well-trained one-dimensional convolutional neural network is used to identify the central wavelength of the overlapped spectra. The high-precision demodulation of the central wavelength of the overlapped spectra is verified. The root-mean-square error of the model is 1.819 pm and the demodulation time is better than 53.86 ms. This fiber grating center wavelength demodulation method has good application in serial WDM FBG sensor networks.