Deep learning-based phase demodulation for distributed acoustic sensor

With the rapid advancement of deep learning, its applications in fiber optic sensing are expanding significantly, particularly in recognizing diverse scenarios where substantial progress has been made. However, the crucial task of obtaining demodulation results from raw data remains relatively underexplored. Accurate demodulation is essential for a deeper understanding of the physical processes in fiber optic sensing systems, enhancing measurement accuracy, and optimizing system performance. In this paper, a phase demodulation algorithm for distributed acoustic sensor (DAS) data processing was developed using deep learning framework. The proposed approach aims to replace the computationally intensive processes of traditional demodulation algorithms, such as the Hilbert transform, in specific scenarios. The proposed method enables faster acquisition of detailed demodulated data and provides an intuitive, real-time graphical display of the results. Experimental results demonstrate that the proposed method significantly enhances processing speed while maintaining accuracy comparable to traditional demodulation algorithms. Specifically, for a dataset comprising 2000 rows and 4500 columns (approximately 9 million data points), the response time is reduced from 2.62 to 0.1 s when processing data at a pulse frequency of 20 kHz over a sensing distance of 1.8 km.