The objective of this research program is to investigate new approaches to the problem of modeling and tracking of a large class of time-varying signals. The class of signals under consideration includes signals relatively concentrated in time-frequency, signals with slowly varying amplitude and frequency, and some signals with rapid variation in spectral content. Both parametric and nonparametric approaches are being developed. Parsimonious models incorporating parameters that reflect non-stationary that are specific to the class of signals under consideration also being considered. Those models will provide an accurate and efficient parametrization of signals in these classes. Algorithms are being developed in this project for efficient estimation of the model parameters from noisy observations of the signal. A nonparametric approach utilizing new time-frequency representations based on determination of optimal signal-dependent kernels is also being developed to overcome several problems which have plagued current time-frequency methods.
