On the relation between blind system identification and subspace tracking and associated generalizations

Blind system identification and subspace tracking represent two important classes of signal processing problems with a variety of applications. Although originally seemingly independent from each other, the related algorithms exhibit various commonalities. In this paper, we present a novel unified derivation of the corresponding classes of adaptation algorithms. This top-down approach both clarifies the algorithmic relations and also leads to various powerful generalizations of the algorithms. Due to the rigorous approach, we obtain important practical design rules for an efficient system design. By exploiting multiple stochastic signal properties, the treatment also includes practically useful relations to blind signal extraction and blind source separation algorithms.