HOS-based orthogonal subspace algorithm for causal ARMA system identification. (Q1978293)

A new method, based on subspaces of a cumulant matrix, is proposed for the blind identification of an ARMA system which is driven by white non-Gaussian noise. The relationship between the cumulants and the impulse response is exploited to arrive at a relation between a cumulant matrix and a matrix consisting of impulse responses. This will lead to the formulation of a new algorithm for impulse response recovery and hence estimation of the ARMA parameters. The proposed method is shown to do well at low SNR values. When impulse response recovery is the key issue, it is shown to offer reduction in computational complexity.