Adaptive filtering algorithm based on a wavelet packet tree for heart sound signal analysis

Abstract

In order to further highlight heart sound signals analysis, we developed an algorithm based on a wavelet packet tree; for possible discrimination depending on the severity of pathological cases for different heart sound signals. The algorithm functions select the most informative nodes combination of a wavelet packet tree as a basis for feature extraction. To generate this adaptive filter, we need to compute the best sub tree of an initial wavelet packet tree with respect for entropy type yardstick, the node combination with the lowest total cost is selected. The decomposition into wavelet packet offers a wavelet library organised according to their time-frequency analysis and location properties and therefore of pass-band filtering, according to a binary tree architecture. This architecture makes it possible to implement algorithms for searching for adapted bases to both the desired time-frequency properties and the analysed signal, which are conventionally called better bases.