2D-SenseNet: A Simultaneous Demodulation and Denoising Network for DAS
Yongxin Liang, Shibo Zhang, Jialei Zhang, Zhenyu Ye, Anchi Wan, C. G. Liu, Jianhui Sun, Zinan Wang · IEEE Transactions on Instrumentation and Measurement · 2024
Phase-sensitive optical time-domain reflectometry (Φ-OTDR) has revolutionized geophysical research by providing a method for large-scale distributed acoustic sensing (DAS), especially with the assistance of artificial intelligence (AI). However, demodulation and AI-based denoising are separate processes, ultimately impacting the unified integration of signal processing. Here, we developed an integrated Demodulation and Denoising (2D) SenseNet, a supervised deep learning network designed to sense the state transition of the fiber Rayleigh scattering responses. Unlike all AI-based denoising schemes whose training datasets include signals or noise captured during experiments, 2D-SenseNet was trained by purely numerical simulation datasets, showing strong generalization capability. By performing a field experiment with a dark fiber deployed on campus, 2D-SenseNet can extract phase information more accurately than the demodulation method, and eliminate the need for complex demodulation processes. It establishes a foundational framework for developing an integrated demodulation-denoising-decision (3D) SenseNet, initiating a new direction in developing intelligent Φ-OTDR.