Analysis and classification of time-varying signals with multiple time-frequency structures

We propose a time-frequency (TF) technique designed to match signals with multiple and different characteristics for successful analysis and classification. The method uses a modified matching pursuit signal decomposition incorporating signal-matched dictionaries. For analysis, it uses a combination of TF representations chosen adaptively to provide a concentrated representation for each selected signal component. Thus, it exhibits maximum concentration while reducing cross terms for the difficult analysis case of multicomponent signals of dissimilar linear and nonlinear TF structures. For classification, this technique may provide the instantaneous frequency of signal components as well as estimates of their relevant parameters.