Department of Computer Science, K.U.Leuven, Leuven, Belgium
TW Reports vol:TW430
Fast estimation and tracking of the principal subspace of a sequence of random vectors is a classic problem, widely encountered in areas such as radar, sonar and speech processing, data compression, data ﬁltering, parameter estimation, pattern recognition, neural analysis, wireless communications, to name just a few.
Among the most robust algorithms for subspace tracking there are the so called OPERA–based algorithms with computational complexity 2nr2 + O(r2), where n is the input vector dimension and r (n ≫ r) is the desired number of eigencomponents. In this paper we propose a fast and stable algorithm for subspace tracking based on the EVD–OPERA algorithm with 6nr +15r2 computational complexity.