| Title |
Modified Kernel-width Adjustment for Maximum Error Entropy |
| DOI |
https://doi.org/10.5573/ieie.2022.59.2.3 |
| Keywords |
MEE; Adaptive; Kernel-width; Convergence-speed; Impulsive noise |
| Abstract |
Minimum error entropy (MEE) as a method of information theoretic learning has been effectively used in many applications such as signal processing, machine learning and automatic control in Gaussian or non-Gaussian noise environments. The choice of kernel width is very sensitive and has important effects on system performance. In many signal processing environments, the error signal is usually non-stationary so that the fixed kernel width is not appropriate for implement of the MEE algorithm. The conventional kernel width adaptation methods based on Kullback-Leibler divergence contain the cube of kernel width in the denominator of its update equation. This term is found in this paper to lead to a very slow convergence in impulsive noise situations. This paper presents a modified kernel width adaptation method in a way that removes the term having negative impact on convergence speed while preserving the same optimum kernel width. The experiment of channel equalization demonstrates the superiority of the proposed method in learning performance by showing above 3dB enhancement in steady state MSE. |