ACOUSTIC ECHO CANCELER

摘要

An echo canceler in which effective transfer function coefficients of an adaptive echo-canceling filter are adjusted periodically to track changing system conditions. Periodic adjustment of the effective coefficients is accomplished using a set of more frequently adjusted update coefficients and an update gain which is varied based on a strategic and novel combination of system parameter measurements. By updating filtering coefficients periodically rather than continuously, and by making soft, fuzzy-logic-type decisions with respect to the magnitude of the update gain, exemplary embodiments optimize system speed and adaptability without sacrificing stability. Generally, relatively large update gains are applied in situations where only a far-end user is speaking and a near-end signal is relatively noise-free. However, in near-end single talk, double-talk, and high near-end noise situations, a more conservative approach is used so that the adaptive filter does not become unstable or cause distortion in the near-end speech and noise signals. In either event, the update gain is reduced as the adaptive filter converges to match a prevailing steady-state echo environment so that erroneous perturbations of an already properly-adapted filter are minimized. The decision as to whether to apply high or low update gains during periods of nonconvergence is based in part on a system status gauge comprising a ratio of a peak update coefficient value and a baseline update coefficient value. The system status gauge indicates, among other things, a level of system convergence. In exemplary embodiments, relatively high variable update gains are applied in unconverged, far-end single-talk situations by employing a modified normalized least-mean-squares (NLMS) algorithm. In other unconverged situations, the update gain is adjusted more conservatively based on a normalized version of the system status gauge. By dynamically providing relatively large update gains while being careful not to create system instabilities, the exemplary embodiments quickly and robustly adapt to successfully cancel echoes in a wide variety of system environments and conditions.