Transform domain conjugate gradient algorithm for adaptive filtering

Abstract

This paper proposed a new normalized transform domain conjugate gradient algorithm (NT-CGA), which applies the data independent normalized orthogonal transform technique to approximately whiten the input signal and utilises the modified conjugate gradient method to perform sample-by-sample updating of the filter weights more efficiently. Simulation results illustrated that the proposed algorithm has the ability to provide a fast convergence speed and lower steady-error compared to that of traditional least mean square algorithm (LMSA), normalized transform domain least mean square algorithm (NT-LMSA), Quasi-Newton least mean square algorithm (Q-LMSA) and time domain conjugate gradient algorithm (TD-CGA) when the input signal is heavily coloured.