Robust Vector Decoupling of SERF Magnetometers Using Artificial Neural Network

Résumé fourni par la source

Spin-exchange-relaxation-free (SERF) magnetometers, configurable for three-vector operation, have been deployed in many fields, such as magnetoencephalography (MEG), magnetocardiography (MCG), and precision tests of fundamental physics owing to their ultrahigh sensitivity and low size-weight-power (SWaP) characteristics. SERF magnetometers typically employ multi-axis magnetic field modulation to decouple the vector information. However, the inter-axis crosstalk brought by intrinsic spin dynamics coupling reduces measurement accuracy, compromising source localization and inflating measurement error. To suppress this coupling, a strategy of a wide-band random signal employed as training inputs, paired with photodiode (PD) readings from the SERF magnetometer to construct a neural-network model was utilized for the first time. This approach enabled the high-precision decoupling of the vector magnetic-field components by suppressing the maximum inter-axis crosstalk from 10% to <3%. Beyond reducing crosstalk, the utilization of the wide-band random signal during field training enhanced the model’s generalization capability across varying magnetic environments, offering a novel strategy for the construction of the robust vector magnetometry.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.