Evaluation of FFT-Based and Moder Parametrc Methods for the Spectral Analysis of Bioprosthetic Valve Sounds

Abstract

The objective of this paper is to compare the performance of conventional FFT-based (basic periodogram and Welch's method) and modern parametric (all-pole and pole-zero modeling) methods in estimating the spectral distribution of cardiac bioprosthetic valve sounds, and for the extraction of the two most dominant frequency peaks (DFP). These methods were tested for stability by adding random noise and truncating the bioprosthetic valve closing sounds, and for reproducibility by measuring the variance of the spectra obtained from three consecutive recordings of each patient. Results from a group of 11 patients show that the basic periodogram and Steiglitz-McBride's method with maximum entropy (pole-zero modeling) provide the most consistent (minimal variance) estimates of the DFP's of the closing sounds. However, for estimating spectral distributions, the most stable methods appear to be the basic periodogram and Steiglitz-McBride's method with extrapolation to zero. The basic periodogram appears to be the best compromise to estimate both the spectral distribution and the DFP's of the bioprosthetic closing sounds.