Transform domain approximate QR-LS adaptive filtering algorithm

Abstract

An improved approximate QR-least squares (A-QR-LS) algorithm, called the transform domain A-QR-LS (TA-QR-LS) algorithm, is introduced. The input signal vector is approximately decorrelated by some unitary transformations before applying the A-QR-LS, which is shown to improve considerably the convergence speed of the A-QR-LS algorithm recently proposed by Liu [4]. Further, it is possible to reduce the arithmetic complexities (O(N)) of the A-QR-LS, and TA-QR-LS algorithms by using Givens rotations instead of the Householder transformation. Simulation results show that the proposed TA-QR-LS algorithm is a good alternative to the conventional recursive least squares (RLS) algorithm in adaptive filtering applications involving multiple channels, acoustic modeling, and fast parameter variations.