Double-talk insensitive NLMS algorithm
AssigneeZarlink Semiconductor Inc.
InventorsGary Q. Jin, Gordon J. Reesor, Thomas Qian
Life of this patent
Estimating the term…
Abstract
A method of canceling echo signals in a telephone network operating in a double talk mode has been developed. A system for implementing the method is also presented. In the invention the conventional normalized least mean square (NLMS) algorithm for echo cancellers is modified such that echo path changes continue to be tracked even after a double talk condition has been detected. The modified algorithm, known herein as a double-talk normalized least mean square (DNLMS) algorithm adaptively adjusts the convergence rate based on the power difference between the echo signal or the residual signal and the far end signal.
Claims10
A method of controlling an acoustical echo canceller in a telephone system to cancel echo signals from near end signals in a double talk condition wherein said echo signal comprises a reflected far end signal, the acoustical echo canceller implementing a double-talk normalized least mean square (DNLMS) algorithm and having means to converge said DNLMS algorithm, the method comprising generating a replica echo signal based on said far end signal and an estimated echo path length;
- subtracting said replica echo signal from said near end signal by adjusting convergence rate of said DNLMS algorithm;
- wherein the convergence rate is adaptively adjusted based on the relative strengths of said echo signals and said near end signals, and wherein the convergence rate is based on a variable step size in which the step size is reduced inversely proportional to the power level of the near end signal.
A method of controlling an acoustical echo canceller in a telephone system to cancel echo signals from near end signals in a double talk condition wherein said echo signal comprises a reflected far end signal, the acoustical echo canceller implementing a double-talk normalized least mean square (DNLMS) algorithm and having means to converge said DNLMS algorithm the method comprising generating a replica echo signal based on said far end signal and an estimated echo path length;
- subtracting said replica echo signal from said near end signal by adjusting convergence rate of said DNLMS algorithm;
- wherein the convergence rate is adaptively adjusted based on the relative strengths of said echo signals and said near end signals and, wherein the convergence rate is based on a variable step size in which the step size value is controlled through the energy level of the far end signal.
Dependencies are inferred from the claim text.