Rank-deficient prewhitening with quotientSVD andULV decompositions

This paper deals with certain theoretical and numerical aspects of prewhitening, which is a technique frequently used in signal processing when dealing with signals degraded by colored noise. In particular, we demonstrate how to prewhiten a signal contaminated by an interfering noisy signal whose covariance matrix is rank deficient. The formulation of our technique is based on the quotient (or generalized) singular value decomposition, and we also show that a quotient-version of theULV decomposition can be used to provide an efficient updatable implementation.