Potential for improving the local realization of coordinated universal time with a convolutional neural network

Area of Science:

• Metrology
• Artificial Intelligence
• Signal Processing

Background:

• Accurate timekeeping is crucial for global navigation and communication systems.
• The National Metrology Institute of Japan (NMIJ) realizes Coordinated Universal Time (UTC) using a hydrogen maser as a master oscillator.
• Predicting the time difference between UTC and local oscillators is essential for maintaining synchronization.

Purpose of the Study:

• To evaluate the efficacy of a one-dimensional convolutional neural network (1D-CNN) for predicting the time difference between UTC and a hydrogen maser at NMIJ.
• To compare the predictive accuracy of the 1D-CNN with the traditional Kalman filter method.

Main Methods:

• Implementation of a one-dimensional convolutional neural network (1D-CNN) for time series prediction.
• Utilizing deep learning techniques for analyzing time differences in atomic clock measurements.
• Comparative analysis against the Kalman filter algorithm for time prediction accuracy.

Main Results:

• The 1D-CNN demonstrated improved accuracy in predicting the time difference between UTC and the NMIJ's hydrogen maser compared to the Kalman filter.
• Initial results indicate the potential of deep learning for precise time synchronization.

Conclusions:

• Deep learning, specifically 1D-CNN, shows promise as a powerful tool for enhancing the accuracy of UTC realization.
• Further research is warranted to fully establish the 1D-CNN as a reliable predictor for timekeeping applications.
• Computational approaches leveraging deep learning may offer versatile solutions for improving the synchronous accuracy of national metrology institute time scales.