Highly Accurate Parallel AWG Interrogation System Using Neural Networks

This paper reports an interrogation scheme that combines the arrayed waveguide grating (AWG) with neural networks. The scheme uses a convolutional neural network (CNN) to implement the selection of the number and location of fiber Bragg gratings (FBGs) to be interrogated, and a back-propagation neural network (BPNN) to construct a complex nonlinear relationship between the light intensity of the AWG output channel and the FBG's reflection wavelength. In addition, we introduce a generative adversarial network (GAN) to process and tune the dataset, which can enhance the data quickly and with high performance during the training process. The experimental result shows that our proposed interrogation system increases N-channel AWG from a maximum of N/2 FBGs interrogated simultaneously to N-1, and the interrogation dynamic range covers the whole AWG spectrum. Meanwhile, the classification accuracy is 99.733%, the interrogation accuracy is better than 13 pm, and the optimal interrogation accuracy is 1.7 pm. This scheme significantly improves the channel multiplexing capability of AWG and contributes to further reducing the size of the FBG interrogator.