Wavelet-packet basis selection for abrupt changes detection in multicomponent signals

We propose a "best basis" selection method, to detect abrupt changes in noisy multicomponent signals. The basis, selected from a wavelet-packet library with an energetic criterion, separates the different frequency components of the signal while keeping the best possible time localization. From the obtained basis, we reconstruct monocomponent signals in each band. The detector is based on the adequation of this set of signals with an abrupt change model. A performance analysis is realized from synthetic signals, in terms of selected basis and detection results.