Towards continuous and accurate demodulation of dual AWG-based FBG array sensors with LSTM neural network

This paper proposes a continuous demodulation algorithm for fiber Bragg grating (FBG) arrays, which utilizes long short-term memory (LSTM) neural networks and dual-array waveguide grating (AWG) for synchronized multi-channel spectral acquisition. This design achieves ultra-high demodulation accuracy and enhanced feature representation. By interleaving dual AWG channels, the effective spectral period is shortened, enabling each FBG reflection wavelength to be co-sampled across multiple channels within its modulation range, thereby enhancing sensitivity to minute spectral shifts. The acquired multidimensional spectral data serves as input features for the LSTM model, effectively overcoming the limitations of traditional linear and shallow models in high-dimensional spaces. Experimental results demonstrate that this method achieves end-to-end integration of feature extraction and wavelength prediction, enabling continuous high-precision demodulation. In temperature monitoring, the mean maximum prediction error of the four FBG sensors in the FBG array sensor is ±12.125, pm, and the mean average prediction error is only ±1.672, pm. This method provides a reliable solution for efficient demodulation of high-density FBG arrays, offering a technical pathway that combines high precision and robustness for applications such as smart sensing.