Robustness Enhancement of Underdetermined Blind Source Separation for Magnetic Signals With Continuous Wavelet Transform

Abstract

In situ magnetic measurements on compact platforms are often degraded by stray interference from onboard electronics. Traditional magnetic cleanliness programs such as booms and shielding are often impractical for compact magnetic sensing platforms, including CubeSats and Unmanned Aerial Vehicles, highlighting the need for advanced algorithmic solutions. We therefore propose a robustness‐enhanced Underdetermined Blind Source Separation (UBSS) algorithm based on Continuous Wavelet Transform (CWT), referred to as CWT‐UBSS. CWT‐UBSS redesigns UBSS with a dual‐path architecture: one path employs CWT and Single‐Source Point (SSP) clustering to estimate the mixing matrix, while the other applies a reversible transform with compressed sensing for signal reconstruction. This redesign enhances time‐frequency sparsity, improves the availability of SSPs for mixing matrix estimation, and decouples source separation from signal reconstruction, resulting in improved robustness and practical applicability. Evaluated on real satellite magnetic data with simulated interference typical of CubeSats, CWT‐UBSS reduces the Root Mean Square Error (RMSE) from 90.5 to 4.93 nT, outperforming the unfiltered baseline of the prototype UBSS published code (RMSE 6.56 nT). Across additional datasets, it remains robust under subtle noise perturbations, with a median Mean Absolute Error variation rate of 0.44%, roughly three times lower than the 1.47% from the prototype UBSS baseline. These results demonstrate enhanced robustness in magnetic signal separation and suggest potential applicability to resource‐constrained magnetic sensing platforms under challenging conditions.