Characterization of Complex Signals Using Time-Frequency-Phase Concept

A real environment identification system is based on observations which are often non stationary. A mechanical machine generating acoustic signals or an underwater environment are examples of systems characterized by non-stationary signals. Their analysis in the time-frequency domain is well adapted so that it offers appropriated structures which are good candidates for the information post-processing. The system architecture is defined through three blocks (detection of regions of interest, observations segmentation and separation, analytical characterization). This architecture is mainly based on joint use of time, frequency and local phase analysis. More precisely, the phase information will be locally analyzed, using generalized instantaneous moments, on the time-frequency regions previously selected thanks to the time-frequency grouping algorithm. This architecture permits an efficient scheme to solve the constraints brought by this type of signals with a complex time-frequency behavior and by the human operator to reduce this tasks in the decision process. Examples from underwater behavior (underwater mammals vocalizations) will prove the efficiency of the proposed approach.