Variable step-size LMS algorithm using regulated least square error criterion

A new variable step-size algorithm is proposed using an evolutionary approach for the least mean square algorithm. The step-size candidates are evaluated by calculating their square errors, which are composed of a priori and a posteriori errors. The composition of the square error measure is regulated according to different properties of errors. The new algorithm always outperforms other traditional variable step-size methods to provide fast converging and good tracking capability.