Abrupt changes detection in the time-frequency plane

This paper deals with a comparison between two new non-parametric methods for detecting abrupt spectral changes in non-stationary signals. The first one consists in comparing two smoothed time-frequency representations (TFRs), as well as for parametric approaches, a long-time and a short-time model are compared. In front of abrupt changes, the instantaneous distance between the two TFRs will sharply peak at the change times. The second method is based on a stationarity index which is calculated from only one TFR. Simulation tests are exhibited for each method and show the performances and the limitations of these methods for the study of mono and multicomponent signals.