Application of time series spectral analysis theory: analysis of cardiovascular variability signals.

The paper focuses on the most important application problems commonly encountered in spectral analysis of short-term (less than 10 min) recordings of cardiovascular variability signals (CVSs), critically analysing the different approaches to these problems presented in the literature and suggesting practical solutions based on sound theoretical and empirical considerations. The Blackman-Tukey (BT) and Burg methods have been selected as the most representative of classical and AR spectral estimators, respectively. For realistic simulations, 'synthetic' CVSs are generated as AR processes whose parameters are estimated on corresponding time series of normal, post-myocardial infarction and congestive heart failure subjects. The problem of resolution of spectral estimates is addressed, and an empirical method is proposed for model order selection in AR estimation. The issue of the understandability and interpretability of spectral shapes is discussed. The problem of non-stationarity and removing trends is dealt with. The important issue of identification and estimation of spectral components is discussed, and the main advantages and drawbacks of spectral decomposition algorithms are critically evaluated.