Single-Sensor Magnetic Interference Mitigation using Singular Spectrum Analysis and Machine Learning

Accurate in-situ magnetic field measurements are necessary to understand and analyze the various space physics phenomena that couple energy and mass throughout near-Earth space. However, in most in-situ application the geophysical magnetic field of interest is contaminated with magnetic noise from the host spacecraft. One method of achieving accurate measurements is by employing a pair of magnetometers on a common boom and using the magnetic gradient between the two measurements to infer and remove the local stray field. However, this presentation demonstrates how to achieve similar results, for the example case of magnetic noise from reaction wheels, using data from a single magnetometer. This is achieved by using Singular Spectrum Analysis (SSA) to decompose magnetic field measurements into physically meaningful components. Further, a Fully Convolutional Neural Network (FCN) can be trained and applied to these decomposed signals, labeling them as interference or residual components. The interference components can be subsequently removed from the reconstructed magnetic field data, enabling the mitigation of local magnetic interference caused by spacecraft subsystems using only a single magnetometer. For example, a 57% reduction in local interference was demonstrated when this algorithm was applied to magnetic field data from a quiet interval of time captured by the CASSIOPE/Swarm-Echo Magnetic Field Instrument's outboard magnetometer (i.e., from 1.77 nT to 0.76 nT). This technique could be applied to missions designed with only a single magnetometer or as a contingency option for dual-sensor missions to allow some noise mitigation in the case of sensor failure.