Dilated convolutional neural networks for fiber Bragg grating signal demodulation

Area of Science:

• Optical Sensing
• Machine Learning Applications
• Signal Processing

Background:

• Quasi-distributed fiber Bragg grating (FBG) sensor networks face challenges with signal overlap, leading to demodulation errors.
• Accurate signal separation is crucial for reliable data acquisition in FBG sensor systems.

Purpose of the Study:

• To develop a novel deep learning model for accurate signal demodulation in FBG sensor networks.
• To overcome limitations in separating highly overlapped FBG sensor signals.

Main Methods:

• Implementation of a multi-peak detection deep learning model utilizing a dilated convolutional neural network (CNN).
• Evaluation of the model's performance in terms of peak wavelength determination error and demodulation time.
• Testing the model's robustness against noise and comparison with existing demodulation techniques.

Main Results:

• Achieved extremely low error (< 0.05 pm RMS) in peak wavelength determination for highly overlapped signals.
• Demonstrated a demodulation time of 15 ms for two signals, enhancing network multiplexing and detection accuracy.
• Maintained low RMS error (< 0.47 pm) even at a signal-to-noise ratio of 15 dB, showcasing noise robustness.

Conclusions:

• The proposed dilated CNN model significantly enhances the accuracy and speed of signal demodulation in FBG sensor networks.
• This deep learning approach offers a promising solution for overcoming challenges associated with signal overlap and noise in optical sensing.
• The findings highlight the potential of neural network algorithms for advancing signal demodulation techniques.