Research On Variable Step Size Adaptive Algorithm For Echo Cancellation

In voice communication devices,the presence of echoes severely degrades the quality of voice communications.Tracking and estimating unknown echo channels in an adaptive manner using adaptive filters is the most widely used echo cancellation method.The normalized least mean square algorithm,affine projection algorithm and normalized subband adaptive algorithm are widely used in echo cancellation because of their simple structure and good robustness.However,these algorithms need to make a trade-off between convergence speed and steady-state offset according to the specific performance requirements of the system.The parameterless variable step size adaptive algorithm solves the contradictory problem between high convergence speed and low steady state offset.At the same time,the series of algorithms do not require any prior information of the acoustic environment,and have lower practical application difficulty.In the background of echo cancellation,this paper deeply studies the parameterless variable step size adaptive algorithm.For the different problems and application scenarios,three improved parameterless variable step size algorithms are proposed.Aiming at reducing the resource occupancy rate of the algorithm in the implementation of digital logic system,a new parameterless variable step size affine projection algorithm without root-square constraint is proposed.The constraint method of hardware-easy implementation is used to replace the root-square constraint of the parameterless variablestep affine projection algorithm,so that the algorithm can keep the convergence performance unchanged while reducing the occupancy rate of digital resources.Aiming at improving the convergence performance of adaptive algorithm under high background noise,a constrained variable step size matrix normalized subband adaptive algorithm is proposed.By introducing time-varying constraints on the time-varying step sizes of each sub-band,the algorithm achieves higher convergence speed and lower steadystate offset under high background noise interference,and has higher robustness.In order to solve the contradiction problem of tap convergence speed and stability of gradient descent tap length adaptive algorithm,combined with the parameterless step-size normalized least mean square algorithm,a variable tap fit is proposed.The parameter gradient gradient is changed to the tap length adaptive algorithm.By evaluating the convergence of the adaptive filter,the tap fitting parameters are changed in real time,so that the algorithm has stronger stability while obtaining faster tap convergence speed.The experimental results show that the convergence effect and echo cancellation of the new parameterless affine projection algorithm without parameter-dependent variable-step projection are the same as those of the traditional parameterless variable-step affine projection algorithm.In the case of background noise interference,The convergence performance and filtering effect of the constrained variable step size matrix normalized subband adaptive filter is better than the traditional normalized subband adaptive filter;the gradient descent variable tap length adaptive algorithm with variable tap fitting parameters It has a higher stability when it has faster tap convergence speed.