Tracking of frequency in a time-frequency representation

For non-stationary signals, the evolution of frequency characteristics with time may bring useful information to approach the underlying physical process. Time-frequency representations may facilitate such an interpretation. Here, we consider a representation obtained by the ARCAP method, which is adapted to narrow band signals. At the end of the analysis, at each sampling date, the signal is characterized by a set of two component vectors: a characteristic frequency and the signal power at that frequency. The problem is to track automatically these sparse points to obtain the evolution along time for each modulation. The originality of the proposed method is to track the points of the ARCAP representation thanks to a Kalman filter, based on a frequency modulation model. After a brief presentation of the theoretical methods, we show the results obtained on various signals.