Signal segmentation and modelling based on EquiPartition principle

Abstract

In this paper, we propose a method for time interval segmentation of signals based on an EquiPartition principle (EP). According to EP, the signal is segmented into segments that give equal errors in reconstruction selecting the most suitable model to describe each segment. Moreover, the segments are equivalent in the content domain, since the signal is segmented into segments that are modelled by the same number of coefficients. The proposed method has been successfully applied on different types of signals like: physiologic, speech, human motion, financial time series. Finally, the proposed methodology is very flexible on changes of error criteria, signal modelling and on signal dimension yielding a robust method for segmentation and modelling of signals.