Matched generalized Gabor expansion of nonstationary processes

The key motivation for performing linear signal transforms or filter bank analyses is to obtain a decorrelation of a given time series. In this paper, we generalize Gabor's classical idea such that the mutual coefficient correlation of the time-frequency parametrized signal expansion is minimized. Given a nonstationary process characterized by its second order statistic we derive the matched time-frequency sampling grid and the optimum analysis window such that the Gabor coefficients show minimum mutual correlation. The window optimization is a nonlinear problem which can be solved iteratively. We corroborate the theory by simple source coding experiments.<<ETX>>