A study has been designed to add insight to some questions that have not been fully investigated in the heart rate variability field and the cardiovascular regulation system in normal and Chronic Obstructive Pulmonary Disease (COPD) subjects. It explores the correlations between heart rate variability and cardiovascular regulation, which interact through complex multiple feedback and control loops. This work examines the coupling between heart rate (HR), respiration (RESP), and blood pressure (BP) via closed-loop system identification techniques in order to noninvasively assess the underlying physiology.

In the first part of the study, the applications of five different bilinear time-frequency representations are evaluated on modeled HRV test signals, actual electrocardiograms (ECG), BP and RESP signals. Each distribution: the short time Fourier transform (STFT), the smoothed pseudo Wigner-Ville (SPWVD), the ChoiWilliams (CWD), the Bom-Jordan-Cohen (BJC) and wavelet distribution (WL), has unique characteristics which is shown to affect the amount of smoothing and the generation of cross-terms. The CWD and the WL are chosen for further application because of overcoming the drawbacks of other distributions by providing higher resolution in time and frequency while suppressing interferences between the signal components.

In the second part of the study, the Morlet, Meyer, Daubechies 4, Mexican Hat and Haar wavelets are used to investigate the heart rate and blood pressure variability from both COPD and normal subjects. The results of wavelet analysis give much more useful information than the Cohen's class representations. Here we are able to quantitatively assess the parasympathetic (HF) and sympatho-vagal balance (LF:HF) changes as a function of time. As a result, COPD subjects breathe faster, have higher blood pressure variability and lower HRV.

In the third part of the study, a special class of the exogenous autoregressive (ARX) model is developed as an analytical tool for uncovering the hidden autonomic control processes. Non-parametric relationships between the input and outputs of the ARX model resulting in transfer function estimations of the noise filters and the input filter were used as mechanistic cardiovascular models that have shown to have predictive capabilities for the underlying autonomic nervous system activity of COPD patients. Transfer functions of COPD cardiovascular models have similar DC gains but show a larger lag in phase as compared to the models of normal subjects.

Finally, a method of severity classification is presented. This method combines the techniques of principal component analysis (PCA) and cluster analysis (CA) and has been shown to separate the COPD from the normal population with 100% accuracy. It can also classify the COPD population into "at risk", "mild", "moderate" and "severe" stages with 100%, 90%, 88% and 100% accuracy respectively. As a result, cluster and principal component analysis can be used to separate COPD and normal subjects and can be used successfully in COPD severity classification.