Adaptive multiresolution for wavelet analysis

We present a new method of wavelet packet decomposition to be used in gravitational wave detection. An issue in wavelet analysis is what is the time-frequency resolution which is best suited to analyze data when in quest of a signal of unknown shape, like a burst. In the other wavelet methods currently employed, like LIGO WaveBurst, the analysis is performed at some trial resolutions. We propose a decomposition which automatically selects at any frequency the best resolution. The criterion for resolution selection is based on minimization of a function of the data, named entropy in analogy with the information theory. As a qualitative application we show how a multiresolution time-frequency scalogram looks in the case of a sample signal injected over Gaussian noise. For a more quantitative application of the method we tested its efficiency as a non-linear filter of simulated data for burst searches, finding that it is able to lower the false alarm rate of the WaveBurst algorithm with negligible effects on the efficiency.