Filtering time series using point-wise dimension: application to the study of electrocardiograms

This paper deals with dimension estimation from time series. We propose a procedure based on the pointwise dimension for selecting subsets of the time series, and of the corresponding trajectory in the phase space, depending on the choice of reference point. For a given reference point, this filtering action is obtained by considering points within the scaling region, and results vary with the embedding dimension. The procedure is applied to the analysis of a few electrocardiograms and preliminary results show that a suitable choice of reference point allows investigation of the various underlying mechanisms involved in heart activity. Although in the early stages of development, the proposed method may be a useful tool for the study of the underlying mechanisms which can be observed in nonlinear dynamical systems.