Heart sound analysis comparing wavelet and autoregressive techniques

Three different types of wavelets (Morlet, Bessel filter and Chirp) and four different types of autoregressive modelling techniques (Burg, forward, backward and modified covariance) are considered. A straightforward algorithm is suggested for all wavelet calculations. A modification is made to the standard autoregressive technique to reduce the dominance of certain peaks. The heart sounds of a set of normal patients and on a set of patients diagnosed as having aortic stenosis are investigated. After separating the sound into individual heart cycle and further separating into the four phases (S1, systolic, S2 and diastolic), the relative energy levels in different phases over a number of cycles are compared. Typical results are selected for individual cycles and show the benefit of the modified autoregressive approach.