Machine learning assisted vector atomic magnetometry.

Area of Science:

• Atomic physics and quantum sensing
• Machine learning applications in metrology

Background:

• Traditional vector magnetometry requires complex setups with external fields for parameter mapping.
• Existing methods face challenges in simplifying architecture while maintaining high sensitivity.

Purpose of the Study:

• To propose and demonstrate a novel indirect encoding paradigm for vector atomic magnetometry using machine learning.
• To simplify the architectural complexity of vector magnetometers.
• To explore the potential for general multiparameter sensing designs.

Main Methods:

• Encoding three-dimensional magnetic field information into four optical rotation signals.
• Utilizing a pre-trained deep neural network to establish the mapping between signals and magnetic field information.
• Experimental demonstration using a single-shot, all-optical vector atomic magnetometer with a simple design.

Main Results:

• Achieved magnetic field amplitude sensitivities of approximately 100 [Formula: see text].
• Demonstrated angular sensitivities of about [Formula: see text] for a 140 nT magnetic field.
• Successfully implemented a simplified vector magnetometer design with a single elliptically-polarized laser beam and no additional coils.

Conclusions:

• The proposed machine learning-based indirect encoding significantly reduces the complexity of vector magnetometer architectures.
• This approach offers a promising pathway for developing more accessible and efficient multiparameter sensing technologies.
• The findings may inspire new designs for various sensing applications beyond magnetometry.