A Gradient-Based Variable Step-Size Scheme for Kurtosis of Estimated Error

This paper proposes a new variable step-size LMS (VSLMS) algorithm with an approach in which a gradient-based weighted average of a kurtosis of an estimated error signal is used to improve the drawback of a previous algorithm for application to an unknown channel estimation. The proposed scheme leads not only to the enhancement of the convergence rate, but also to robustness in terms of low-SNR environments. It could also lead to obtaining a lower misadjustment error.