ARMA modeling and phase reconstruction of multidimensional non-Gaussian processes using cumulants

A statistical multidimensional sequence, generated by a non-Gaussian process, passes through a linear space-invariant (perhaps ARMA) model and colored Gaussian noise is added at the output. Given output cumulants, algorithms are derived for the nonparametric reconstruction of the system's phase as well as for the parametric identification of the ARMA model, which can be noncausal or have nonseparable denominator.<<ETX>>