THE FAST NEWTON TRANSVERSAL FILTER - AN EFFICIENT SCHEME FOR ACOUSTICECHO CANCELLATION IN MOBILE RADIO

In this paper, we investigate the recently suggested fast Newton famil y for adaptive filtering in the context of acoustic echo cancellation, with emphasis on the mobile radio case. A distinctive advantage of th e fast Newton transversal filter (FNTF) is that it can offer high perf ormance with speech inputs at low computational cost. We discuss possi ble implementations and compare the FNTF with classical schemes in ter ms of complexity. A complete numerically stabilized version is present ed, and additional features for proper real-time operation with speech are discussed. Experimental comparisons using various signals and rea l situations show that in all cases, the FNTF behaves similarly to the standard fast RLS transversal filter (FTF) algorithm, whereas its com plexity is only slightly higher than that of the normalized LMS (NLMS) . Compared with the NLMS, the experiments show that in the context inv estigated, the latter exhibits inferior performance with respect to co nvergence and tracking. Thus, we demonstrate that the FNTF is an effic ient scheme for acoustic echo cancellation in mobile radio.