Epanechnikov based Correntropy Analyses in Channel Equalization
Information Theoretic Learning based channel equalization has the potential to deal with scenarios in which classical techniques fail, as in the presence of correlated signals or non-linear channels. One of the measures in this context is correntropy, a generalized correlation measure. In this work, we analyze the effect of using a non-Gaussian kernel do obtain correntropy, i.e., we use the Epanechnikov kernel. Since such kernel is given by a second order polynomial, we include the Correlation Retrieval Criterion in the analyses, discussing their similarities and performance.
In addition, kernel adaptive filters are also included in our analysis. Such filters are interesting due to their capacity of equalizing non-linear channels. Thus, we study the Kernel Maximum Correntropy with the Epanechnikov kernel and its relation to the well known Kernel Least Mean Square algorithm.