A novel dynamic Principal Component Analysis method, applied to ECG signals

Formerly, we have established relations between the time and frequency domain characteristics of multidimensional time series, and proved asymptotic relation between the eigenvalues of the block Toeplitz matrix of autocovariances and the union of spectra of the spectral density matrices at the Fourier frequencies. On this basis, a novel dynamic Principal Component Analysis algorithm (dynPCA) is developed. The algorithm estimates the dynamic principal components of the multidimensional time series input. The objective of this paper is to describe and study the technique with some numerical considerations. Applications to multi-channel ECG data are also presented, with promising results despite the lack of stationarity.