AN EXPERIMENTAL STUDY OF THE EIGENDECOMPOSITION METHODS FOR BLIND SIMO SYSTEM IDENTIFICATION IN THE PRESENCE OF NOISE
Soroush Javidi, Nikolay D. Gaubitch, Patrick A. Naylor · 2006
Subspace methods for blind SIMO system identification have been proposed which rely on the null subspace of the data correlation matrix to estimate the impulse response coefficients. It is known that the performance of these algorithms degrades with increasing noise and large sys-tem orders. In this paper, we present results of an exper-imental study that links the performance of the subspace algorithm with the eigenvalues of the multichannel input correlation matrix. It is demonstrated that the eigenvec-tor corresponding to the smallest eigenvalue is not always the best solution neither in terms of normalised projection misalignment nor cross-relation error.