Identification of multivariate FIR systems using higher-order statistics
Lang Tong · 2002
The identification of multichannel moving average (MA) parameter matrices {H(k)} using fourth-order output cumulants is considered. By analysing the eigenstructures of the cumulant matrices, it is shown that the MA parameter matrices can be identified uniquely up to a post multiplication of monomial matrices if H(0) does not have columns that are pairwise colinear and H=[H/sup t/(0),...,H/sup t/(L)]/sup t/ has full column rank. The constructive proof of this condition leads to a closed-form identification algorithm.