Improving Prediction Accuracy and Extraction Precision of Frequency Shift from Low-SNR Brillouin Gain Spectra in...

Area of Science:

• Optoelectronics and Photonics
• Signal Processing
• Machine Learning Applications

Background:

• Distributed fiber-optic sensors rely on Brillouin frequency shift (BFS) for measurements.
• Accurate BFS detection is crucial for sensor performance, especially under noisy conditions.
• Existing spectral analysis methods face challenges with signal-to-noise ratio (SNR) degradation.

Purpose of the Study:

• To enhance the accuracy of BFS detection in distributed fiber-optic sensors.
• To evaluate the effectiveness of different spectral analysis algorithms, including Lorentzian curve fitting (LCF), backward correlation (BWC), and generalized linear models (GLM).
• To investigate the combined and sequential application of these algorithms for improved precision.

Main Methods:

• Utilized Lorentzian curve fitting (LCF) with the Levenberg-Marquardt (LM) method.
• Employed the backward correlation technique (BWC).
• Integrated a machine learning algorithm, the generalized linear model (GLM), into the processing pipeline.

Main Results:

• Sequential application of algorithms (e.g., BWC + GLM) improved temperature sensing accuracy by 0.4 °C (428 kHz) for positive SNRs.
• Double processing (BWC + GLM) showed superior performance across all tested SNRs, particularly below 2.6 dB, reducing precision errors by approximately 1.5 °C.
• Significant accuracy gains were observed for SNRs ranging from 0 to 20 dB.

Conclusions:

• Combined signal processing techniques offer a robust method for increasing BFS detection accuracy in fiber-optic sensors.
• The BWC + GLM approach provides a substantial improvement in measurement precision, especially in low SNR environments.
• This technique has potential applications in structural health monitoring and metrology.