Unsupervised Representation Learning-Based Spectrum Reconstruction for Demodulation of Fabry–Perot Interferometer Sensor
Sufen Ren, Shengchao Chen, Haoyang Xu, Xuan Hou, Qian Yang, Guanjun Wang, Chong Shen
Abstract
This article presents a novel unsupervised representation learning-based demodulation framework for Fabry–Perot interferometer (FPI) sensors, which is a straightforward and effective solution for obtaining an interferometric spectrum without any optical spectrum analyzers (OSAs). The proposed framework utilizes a simple spectrum reconstruction method to reconstruct the FPI sensor’s spectrum using relatively low-scale sample points, requiring less manual effort than conventional approaches. The proposed approach involves two steps: first, an optical system converts the FPI sensing signal to transmitted intensity, and second, the unsupervised representation learning-based reconstruction framework establishes a nonlinear relationship between the intensity signal and the actual changing spectrum. The proposed approach is validated using real-world datasets generated from pressure performance tests, achieving excellent performance with a reconstruction error of 0.039 nm and a range of 73 nm. The results demonstrate the practical potential viability of the proposed framework for large-scale remote monitoring systems.