Large Dynamic Range Interrogation Technique for Fiber-Optic Interferometric Sensor Based on AWG and Deep Learning Algorithm

[] Aiming at the application scenario of interferometric sensors used in large dynamic range measurement, an efficient interrogation technique combining array waveguide gratings (AWGs) and a deep learning algorithm is proposed in this article. The features of the interference spectrum are first extracted by AWG’s multiple channels, and then fed into an attention-based long short-term memory (LSTM) model to establish the relationship between the spectral intensity distribution information sampled by AWG and the target measurand. The measurand can be directly identified by the well-trained model. In a proof-of-concept experiment, a Mach-Zehnder interferometer (MZI)-based displacement sensor is constructed to verify the proposed demodulation scheme. The experimental results show that, within the displacement range of 0– $830 ~\mu \text{m}$ (the corresponding free spectrum range (FSR) various from 7.6 to 1.6 nm), the root-mean-square errors (RMSEs) of the predicted displacements using 16, 8, and 4 AWG channels are 3.78, 5.30, and $6.22 ~\mu \text{m}$ , respectively, which indicates that the proposed demodulation scheme has the ability of precise demodulation in large dynamic range. Besides, compared with other deep learning algorithms, attention-based LSTM is more resist to the influence of interference spectrum wavelength drift caused by external environment fluctuations on demodulation performance. This proposed method shows great potential in demodulating other interferometric sensors with large dynamic range in practical engineering applications.