Segmentation of Non-Stationary Signals based on the Characteristic Function of their Evolutionary Spectrum

Segmentation and separation of non-stationary signals is of great interest for many engineering fields and applications. In this paper we present a segmentation approach using the characteristic width function of the joint time-frequency evolutionary spectrum of non-stationary or multi-component signal. The proposed segmentation algorithm is based on the characteristic function of the time-frequency evolutionary spectrum which identifies the different segments of the signal due to its frequency change or any other dynamic changes that will change its statistical properties. The distribution of the energy density at the evolutionary spectrum provides the measures of these changes as a function of both time and frequency. Time-frequency representation of the signal is computed using the discrete evolutionary transform DET. At the experimental part we applied our algorithm for multi-component sinusoidal signal which has a wide range of use in many engineering applications. The obtained results at the segmentation level provide a clear and précised measures of the desired segments and their boundaries corresponding to their time domain.