Application of kernel principal component analysis for optical vector atomic magnetometry

Area of Science:

• Physics
• Spectroscopy
• Machine Learning

Background:

• Vector atomic magnetometers utilizing electromagnetically induced transparency (EIT) offer high precision magnetic field measurements.
• Determining the precise longitudinal angle of a magnetic field from EIT spectra remains a challenge.

Purpose of the Study:

• To develop a practical methodology for accurately recovering the longitudinal angle of a local magnetic field using EIT spectra.
• To enhance the capabilities of EIT-based atomic rubidium magnetometers for vector magnetic field measurements.

Main Methods:

• An unsupervised machine learning algorithm employing nonlinear dimensionality reduction (kernel principal component analysis - KPCA) was developed.
• KPCA was used for feature extraction from EIT spectra, reducing data to a single coordinate in a lower-dimensional space.
• A supervised support vector regression (SVR) machine modeled the relationship between KPCA features and magnetic field direction.

Main Results:

• The KPCA-SVR algorithm achieved an accuracy of within 1 degree for predicting the longitudinal angle of the magnetic field.
• The method demonstrated a resolution of 70 nT for the magnitude of the absolute magnetic field.
• The algorithm effectively streamlined the angle determination process from EIT spectroscopic measurements.

Conclusions:

• The developed KPCA-SVR algorithm provides an accurate and efficient method for vector magnetic field determination using EIT magnetometers.
• This approach enhances the competitiveness of EIT magnetometers compared to conventional vector magnetometry techniques.
• The combination of scalar and angular sensitivity makes this method highly valuable for precision magnetic field measurements.