A bi-iteration instrumental variable noise-subspace tracking algorithm

In this paper, we propose a sequential bi-iteration singular value decompos ition (Bi-SVD) noise-subspace-tracking algorithm for adaptive singular valu e decomposition of the cross-covariance matrix in the recursive instrumenta l variable (RIV) methods of system identification. This algorithm can be us ed for the RIV subspace processing of signals in unknown, correlated Gaussi an noise, The algorithm is based on the pseudo-inverse lemma and standard m atrix-vector multiplication. The application and performance of the algorit hm are demonstrated by tracking the noise subspace of the cross-covariance matrix between two data vectors and finding recursively the total least-squ ares solution of the over-determined instrumental-variable normal equations . (C) 2001 Elsevier Science BY. All rights reserved.