FFT-deep learning denoising method for resonant integrated optical gyroscope.

Area of Science:

• Optical Engineering
• Signal Processing
• Machine Learning

Background:

• Noise in resonant integrated optical gyroscopes (RIOGs) limits precision.
• Developing effective noise reduction techniques is crucial for advanced applications.

Purpose of the Study:

• To present a cascaded denoising framework combining physics-based and data-driven approaches.
• To significantly suppress noise in RIOGs for ultra-high precision.

Main Methods:

• Implemented a hybrid denoising framework: Fast Fourier Transform (FFT) comb filtering followed by a Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) network.
• The CNN-LSTM network utilized 1D convolutional layers for feature extraction and LSTM for temporal dependencies.
• Applied to 1 kHz-sampled, 2,000,000-point field data.

Main Results:

• Achieved a reduction in angular random walk (ARW) to 0.28 °/h.
• Reduced bias instability to 2.52 °/h.
• Demonstrated an approximate 50% improvement in both ARW and bias instability compared to previous methods.

Conclusions:

• The proposed cascaded framework offers a significant advancement in gyroscopic noise suppression.
• The combination of frequency-domain physical constraints and spatiotemporal deep learning shows strong generalization for gyroscope signal processing.