High-fidelity acoustic signal enhancement for phase-OTDR using supervised learning

Area of Science:

• Optoelectronics
• Signal Processing
• Machine Learning

Background:

• Phase-sensitive optical time-domain reflectometry (OTDR) is crucial for distributed acoustic sensing.
• Demodulated phase signals in OTDR suffer from noise due to laser frequency drift, phase noise, and interference fading.
• Existing noise reduction methods often address these issues individually.

Purpose of the Study:

• To develop a simultaneous noise and distortion suppression method for OTDR acoustic signals.
• To improve the fidelity of recovered acoustic signals using supervised learning.
• To overcome limitations of individual noise reduction techniques.

Main Methods:

• Utilizing supervised learning with an end-to-end convolutional neural network (CNN).
• Generating synthetic noisy differential phase signals via numerical simulations.
• Training the CNN with simulated and real-world noise data for acoustic signal enhancement.

Main Results:

• Significant improvement in signal-to-noise ratio (SNR) for PZT vibration signals (from 13.4 dB to 42.8 dB).
• Enhanced scale-invariant signal-to-distortion ratio (SI-SDR) for speech signals by an average of 7.7 dB.
• Effective suppression of noise and distortion from laser drift, phase noise, and fading.

Conclusions:

• The proposed CNN-based method effectively enhances acoustic signals in phase-sensitive OTDR.
• The approach successfully mitigates multiple noise sources simultaneously, improving signal fidelity.
• This method offers a robust solution for high-quality distributed acoustic sensing.