A STATISTICAL APPROACH TO SUBSPACE BASED BLIND IDENTIFICATION

Blind identification of single input multiple output systems is consid ered herein. The low-rank structure of the output signal is exploited to blindly identify the channel using a subspace fitting framework. Tw o approaches based on a minimal linear parameterization of a subspace are presented and analyzed. The asymptotically best consistent estimat e is derived for the class of blind subspace-based techniques. The asy mptotic estimation error covariance of the subspace estimates is deriv ed, and the corresponding covariance of the statistically optimal esti mates provides a lower bound on the estimation error covariance of sub space methods. A two-step procedure involving only linear systems of e quations is presented that asymptotically achieves the bound. Simulati ons and numerical examples are provided to compare the two approaches.