A class of subspace tracking algorithms based on approximation of the noise-subspace

This correspondence introduces a novel class of so-called subspace tracking algorithms applicable to, for example, sensor array signal processing. The basic idea pursued in this correspondence is to reduce the amount of computations required for an exact SVD update, applying a perturbation-like strategy, which is interpreted as an approximation of a noise subspace. An interesting property of the derived algorithms is that they can be applied to SVD updating of both auto- and cross-covariance matrices.