Composite proportional normalized least mean squares adaptive algorithm for network echo cancellation

This paper presents a new approach for network echo cancellation, based on a combination of normalized least mean square (NLMS) and proportionate NLMS (PNLMS) adaptive filtering algorithms. The PNLMS has a good initial convergence region (that is, the first echo path cocfficients are fast adapted), but for long path echo the last coefficients are slowly updated. The PNLMS convergence speed is faster (for initial region) than NLMS convergence speed. After this initial region, the NLMS convergence speed is greater than PNLMS convergence speed. The CPNLMS algorithm estimates when the PNLMS become slow and switch the adaptation to NLMS. Computational effort and numerical aspects (floating and fixed point) are considered in order to analyze the real time implementation of CPNLMS algorithm.