Foetal ECG extraction using broadband signal subspace decomposition

In this paper, we consider the extraction of foetal electrocardiogram (ECG) signals from cutaneous electrode recordings. Most existing methods cannot account for the broadband nature of the ECG signals, producing foetal ECG (FECG) estimates that are not sufficiently accurate. We propose a novel way of addressing this problem: the broadband equivalent of principal component analysis (PCA) is applied through the use of an algorithm that generalises the eigenvalue decomposition (EVD) to polynomial matrices; FECG extraction is achieved by way of an orthonormal projection of the data onto the broadband foetal subspace, whereby the maternal ECG and noise components are suppressed. The algorithm is contrasted with conventional PCA and the classical multi-reference adaptive noise cancelling method. The three techniques are applied to real multi-channel ECG recordings obtained from a pregnant woman. A qualitative comparison indicates that the proposed algorithm is more robust and yields cleaner FECG estimates.