Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fast frequency reconstruction using Deep Learning for event recognition in ring laser data

Giuseppe Di Somma, Giorgio Carelli, Angela D. V. Di Virgilio, Francesco Fuso, Enrico Maccioni, Paolo Marsili

TL;DR

This work addresses the need for rapid, high-precision frequency reconstruction from Ring Laser Gyroscope beat notes. It introduces a dual-CNN denoising and regression architecture to extract the Sagnac frequency from a 50-point window at 5 kHz, achieving ~10 ms latency and about 2× precision over FFT-based methods within the $[100,\;500]$ Hz range, guided by the Sagnac relation $f_s = \frac{4 \Omega A}{P \lambda} \cos \theta$. In addition, a real-time disturbance mask and a seismic-event classifier are developed; the classifier reaches 99–100% accuracy on unseen data, using a dataset of 1,164 events with 4-fold validation. The results enable real-time triggers, improved seismic monitoring, and robust operation of RLG-based geophysical systems, including potential deployment across diverse sites.

Abstract

The reconstruction of a frequency with minimal delay from a sinusoidal signal is a common task in several fields; for example Ring Laser Gyroscopes, since their output signal is a beat frequency. While conventional methods require several seconds of data, we present a neural network approach capable of reconstructing frequencies of several hundred Hertz within approximately 10 milliseconds. This enables rapid trigger generation. The method outperforms standard Fourier-based techniques, improving frequency estimation precision by a factor of 2 in the operational range of GINGERINO, our Ring Laser Gyroscope.\\ In addition to fast frequency estimation, we introduce an automated classification framework to identify physical disturbances in the signal, such as laser instabilities and seismic events, achieving accuracy rates between 99\% and 100\% on independent test datasets for the seismic class. These results mark a step forward in integrating artificial intelligence into signal analysis for geophysical applications.

Fast frequency reconstruction using Deep Learning for event recognition in ring laser data

TL;DR

This work addresses the need for rapid, high-precision frequency reconstruction from Ring Laser Gyroscope beat notes. It introduces a dual-CNN denoising and regression architecture to extract the Sagnac frequency from a 50-point window at 5 kHz, achieving ~10 ms latency and about 2× precision over FFT-based methods within the Hz range, guided by the Sagnac relation . In addition, a real-time disturbance mask and a seismic-event classifier are developed; the classifier reaches 99–100% accuracy on unseen data, using a dataset of 1,164 events with 4-fold validation. The results enable real-time triggers, improved seismic monitoring, and robust operation of RLG-based geophysical systems, including potential deployment across diverse sites.

Abstract

The reconstruction of a frequency with minimal delay from a sinusoidal signal is a common task in several fields; for example Ring Laser Gyroscopes, since their output signal is a beat frequency. While conventional methods require several seconds of data, we present a neural network approach capable of reconstructing frequencies of several hundred Hertz within approximately 10 milliseconds. This enables rapid trigger generation. The method outperforms standard Fourier-based techniques, improving frequency estimation precision by a factor of 2 in the operational range of GINGERINO, our Ring Laser Gyroscope.\\ In addition to fast frequency estimation, we introduce an automated classification framework to identify physical disturbances in the signal, such as laser instabilities and seismic events, achieving accuracy rates between 99\% and 100\% on independent test datasets for the seismic class. These results mark a step forward in integrating artificial intelligence into signal analysis for geophysical applications.

Paper Structure

This paper contains 10 sections, 9 equations, 18 figures.

Table of Contents

  1. Introduction
  2. Data delay line for seismology, triggers and feedback controls
  3. Creation of a dataset
  4. Choice of NN supporting structure
  5. Discussion of results
  6. Applications: classification of laser disturbances and seismic events
  7. Mask for Real-Time Data Classification
  8. Detection of seismic events through Artificial Intelligence
  9. Conclusions and Future Prospects
  10. Tests on Single Tone Algorithm

Figures (18)

  • Figure 1: Plot showing the behavior of simulated Gaussian noise compared to GINGERINO data (red) in terms of Amplitude Spectral Density. The red curve represents an estimate of the noise from GINGERINO data, while the cyan curve (“p_wnoise006”) corresponds to Gaussian noise with $\sigma = 0.006$, providing the best fit to the real data. The agreement is particularly evident around the GINGERINO operating frequency, near 280 Hz, which represents the main band of interest for the experiment.
  • Figure 2: Network structure for frequency reconstruction. The sinusoidal input is processed through three ConvBlocks that perform denoising, producing cleaned sinusoids. The final part of the network estimates the associated frequency, with a dense output layer.
  • Figure 3: The frequency distributions obtained from 100000 sinusoids with a base frequency of 280 Hz and added Gaussian noise. The distribution on the right is derived from the ST, while the one on the left is obtained from the NN.
  • Figure 4: Comparison of the standard deviation and spread of the frequency in the region of GINGERINO’s typical operating frequency, obtained from an NN (in red) and an ST (in blue). The superior precision of the NN across the analyzed frequency range is evident from both metrics.
  • Figure 5: Both the NN and the ST accurately reconstruct the frequency of the simulated signal over the entire frequency range analyzed, but the NN has a spread on average of $1.4$ Hz versus that of the ST of $2.4$ Hz
  • ...and 13 more figures