Variable-step LMS adaptive filtering method

Abstract

The invention discloses a variable-step LMS adaptive filtering method. The variable-step LMS adaptive filtering method comprises the following steps that (1) input signals are generated by means of a two-order AR model; (2) the products of input signals generated at all moments and the corresponding two-order prediction coefficients of a two-order linear prediction filter are added, so that output signals at the moments are obtained; (3) the differences between the input signals and the output signals are obtained, so that an error value is obtained; (4) the product of the error value, a step value and the input signals is taken as the two-order prediction coefficient of the two-order linear prediction filter with the updated transient variation; (5) the steps from the step (2) to the step (4) are iterated, so that w [1] (n) and w [2] (n) are made to converge, and an output signal is what is needed. According to the variable-step LMS adaptive filtering method, both the errors with the index being first power and errors with the index being second power are considered, the rate is convergence is maintained, and better steady state errors are obtained.

Description

technical field [0001] The invention relates to a variable step length LMS self-adaptive filtering method, which belongs to the technical field of digital information processing. Background technique [0002] The adaptive filtering algorithm has a strong self-tracking and self-learning ability, and the algorithm is simple and easy to implement. It has been widely used in signal processing, system identification, echo cancellation, radar array processing, and adaptive control of communication systems. Application. Especially in the 1960s, Windrow and Hoff proposed the minimum mean square error (LMS) algorithm based on the minimum mean square error as a criterion. It uses the idea of ​​the steepest descent method and replaces the exact gradient with the estimated value of the gradient Value, iterate along the negative direction of gradient estimation, continuously automatically adjust the tap coefficient of its own filter, and finally converge to the Wiener solution. Compared ...

AI Technical Summary

Problems solved by technology

As in the above theory, considering the two mutual factors of convergence speed and steady-state error, the choice of step size needs to be considered in a compromise. If it is too large or too small, the desired effect will not be obtained.

Images