A PARALLEL LINEARLY CONSTRAINED FAST RLS-ALGORITHM BASED ON INVERSE QR-DECOMPOSITION WITHOUT SQUARE ROOT OPERATIONS

摘要

A parallel version of the fast RLS-algorithm of multichannel adaptive filtering with a sliding window and linear constraints is considered. The parallel computations in this algorithm are related to the possibility of independent processing of data flows dictated by modification of the correlation matrix of the adaptive filter due to sliding window and dynamic regularization. The algorithm is derived based on the generalized lemma about matrix inversion and on inverse QR-decomposition without square root operations.