On weighted subspace estimates in system identification

Subspace based estimates, i.e. estimates obtained by exploiting the orthogonality between a set of vector statistics and a set of parameter-dependent vectors have gained much popularity in the signal processing literature. The article develops a general theory for such estimates. It discusses the generalization of the optimal weighted subspace fitting approach, introduced by Viberg (1989) in the DOA estimation context. It establishes that the optimal weighted estimate enjoy some invariance properties.