PRINCIPAL DECOMPOSITION OF TIME-FREQUENCY DISTRIBUTIONS

We present a novel time-frequency analysis technique which uses princi pal components analysis to map any given time-frequency distribution ( TFD) of a signal into a set of three 1-D principal decomposition funct ions. These three functions may then be considered to be ''separable'' components of a time-frequency function which we refer to as the Prin cipal Approximation Function for the original TFD. We show how princip al decomposition analysis is useful for the enhancement and frequency- tracking of nonstationary harmonic signals.