An improved proportionate affine projection algorithm for network echo cancellation

Recently, a fast converging algorithm for use in network echo cancellers called proportionate normalized least mean squares (PNLMS) has been proposed (Duttweiler, 2000). In comparison to the normalized least mean square (NLMS) algorithm, PNLMS has got fast initial convergence speed and tracking when the echo path is sparse. The idea behind PNLMS was to update each coefficient of the adaptive filter independently of the others by adjusting the adaptation step size in proportion to the magnitude of the estimated filter coefficient. Unfortunately, when the echo path becomes dispersive, the PNLMS algorithm converges much slower than NLMS. In many simulations, it seems that PNLMS has an advantage only when the impulse response is close to a delta function. More recently, an improved PNLMS (IPNLMS) was proposed (Benesty and Gay, 2002). It performs better than both NLMS and PNLMS algorithms. In this paper, generalization of the reliable method from the IPNLMS algorithm to a proportionate affine projection algorithm (PAPA) (Gansler et al., 2000) is presented. Simulations with real speech signals show that proposed algorithm (improved PAPA) is better than NLMS, PNLMS and IPNLMS and not dependent on the nature of the impulse response.