Multi-Kernel equalization for non-linear channels

Abstract

Nonlinear channel equalization using kernel equalizers is a method that has attracted lots of attention today due to its ability to solve nonlinear equalization problems effectively. Kernel equalizers based on Recursive Least Squared, K-RLS, are successful methods with high convergent rate and overcome the local optimization problem of RBF neural equalizers. In recent years, some simple K-LMS algorithms are used in nonlinear equalizers to further enhance the flexibility with the adaptive capability of equalizers and reduce the computational complexity. This paper proposes a new approach to combine the convex of two single-kernel
adaptive equalizers with different convergent rates and different efficiencies in order to get the best kernel equalizer. This is the Gaussian multikernel equalizer.