System identification from frequency samples of the bispectrum

A system identification technique for general noncausal, nonminimal-phase autoregressive moving-average (ARMA) processes is presented. The technique combines rational modeling from frequency response samples with system identification from higher-order statistics information. The formulation is based on samples of the bispectrum along a straight line on the frequency plane, referred to as a bispectrum slice. The identification is carried out in two steps; first a rational model of the bispectrum slice is obtained via solution of two uncoupled systems of equations and, subsequently, the coefficients of the underlying ARMA model are obtained by using closed-form expressions. The method is verified through simulations