A stochastic model for correlated signal sources based on higher-order statistics

The generalized Gaussian model for correlated signal sources is revisited. The model accuracy is discussed using adapted series expansions. The generalized Gaussian model is employed for analytical investigations of signal decomposition algorithms. The results serve as a basis for discussion of the coding gain concept. It is shown that the simplifying assumptions in the derivation of the arithmetic mean to geometric mean ratio of the subband coefficient variances are often violated and that even extended coding gain equations generally fail for highly correlated processes. For weakly correlated processes, the admissibility of the simplifications can often be verified