A multiresolution transform for the analysis of cardiovascular time series

This paper proposes a multiresolution transform which allows to choose, at each frequency, the most appropriate balance between time and frequency resolution according to the signal characteristics and to specific goals of the analysis. For the analysis of cardiovascular time series it is proposed to use a resolution in time which changes with frequency according to a parametric sigmoid function. It results a near constant Q analysis in the central range of frequency and a near constant time-frequency resolution at the extremities. Synthetic and real data obtained from 20 patients during of autonomic tests were analyzed. The following results were obtained: at high frequency high time resolution is achieved and even short duration components are detected; at low frequency the good frequency resolution allows to discriminate among close low frequency components.