Noninvasive fetal ECG extraction using doubly constrained block-term decomposition (Q2045665)

A major challenge in fetal electrocardiography (ECG) is separating the fetal ECG signal from the maternal signal. This paper proposes an algorithm based on tensor decomposition with quasi-periodicity constraints of fetal and maternal ECG signals. The method separates the subspaces of the mother, the fetus(es) and noise. The method is tested on both real and synthetic data sets. The quantitative and qualitative results of the proposed method show improved performance compared to other tensor and matrix decomposition methods in noisy conditions.