Deep learning-based phase demodulation for distributed acoustic sensor.

Area of Science:

• Optical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Deep learning is increasingly applied to fiber optic sensing, yet raw data demodulation remains a challenge.
• Accurate demodulation is vital for understanding physical processes and optimizing fiber optic sensing systems.
• Traditional demodulation methods, like the Hilbert transform, are computationally intensive.

Purpose of the Study:

• To develop a deep learning-based phase demodulation algorithm for distributed acoustic sensor (DAS) data.
• To replace computationally expensive traditional demodulation techniques with a faster, deep learning approach.
• To enable intuitive, real-time graphical display of demodulated results.

Main Methods:

• A novel phase demodulation algorithm was designed utilizing a deep learning framework.
• The algorithm processes raw data from distributed acoustic sensing (DAS) systems.
• The approach was tested on a dataset of 2000 rows and 4500 columns at a 20 kHz pulse frequency over 1.8 km.

Main Results:

• The deep learning method significantly reduced processing time from 2.62 seconds to 0.1 seconds for the test dataset.
• The enhanced processing speed was achieved while maintaining accuracy comparable to traditional algorithms.
• The algorithm provides faster acquisition of detailed demodulated data and real-time visualization.

Conclusions:

• The proposed deep learning algorithm offers a substantial improvement in processing speed for DAS data demodulation.
• This advancement facilitates more efficient and real-time analysis of fiber optic sensing data.
• The method holds potential for enhancing measurement accuracy and system performance in various fiber optic sensing applications.