Wavelength demodulation method for FBG overlapping spectrum utilizing bidirectional long short-term memory neural network

•Enhanced wavelength demodulation techniques for overlapping spectra of FBGs.•Robust to loss-induced fluctuations in FBG reflectivity and FWHM.•Higher demodulation accuracy and stability when compared to LSTM and GRU models. Wavelength demodulation of overlapping spectrum can boost the multiplexing capacity of FBGs, thereby increasing the measurement distance and density. Yet, many current methods ignore variations in FBGs’ reflectivity and FWHM caused by insertion and splice losses, which hinders their practical applicability. This study introduces a novel demodulation method using deep learning neural network to handle the overlapping spectra of FBGs. By generating training data through a fitting model and a small amount of experiment, a multi-output regression model is trained for wavelength demodulation. Experiments were conducted to test the different model’s performance. Results indicate that the proposed method exhibits good robustness to fluctuations in reflectivity and FWHM. In the scenario with maximum reflectivity and FWHM fluctuations of 0.71 % and 8.2 pm respectively, the root mean square error of the Bi-LSTM model is less than 1.95 pm. This result surpasses the performance of both the LSTM and GRU models.