Estimation of the power spectral density in nonstationary cardiovascular time series: assessing the role of the time-frequency representations (TFR).

Abstract

Spectral analysis of cardiovascular series has been proposed as a noninvasive tool for investigating the autonomic control of the cardiovascular system. The analysis of such series during autonomic tests requires high resolution estimators that are capable to track the transients of the tests. A comparative evaluation has been made among classical (FFT based), autoregressive (both block and sequential mode) and time-frequency representation (TFR) based power spectral estimators. The evaluation has been performed on artificial data that have typical patterns of the nonstationary series. The results documented the superiority of the TFR approach when a sharp time resolution is required. Moreover, the test on a RR-like series has shown that the smoothing operation is effective for rejecting TFR cross-terms when a simple, two-three components series is concerned. Finally, the preliminary application of the selected methods to real RR interval time series obtained during some autonomic tests has shown that the TFR are capable to correctly represent the transient of the series in the joint time-frequency domain.