Time-frequency modelling as a mixture of X/sup 2/ distributions

This communication presents a new method for non-stationary signal interpretation, which consists of extracting and characterizing time-frequency structures. It is based on a modelling of a time-frequency representation probability density function as a mixture of central and noncentral chi-square distributions. The problem of interpretation is reduced to a problem of estimation and classification solved by a specific formulation of the EM algorithm for such mixture.