Blind Nonlinear Channel Equalization Using Kernel Processing

Abstract

Blind nonlinear channel equalization using kernel processing is proposed, which transforms blind equalization of nonlinear channel to formulate as a convex quadratic programming using kernel processing. The novel method acquires the optimal solution by solving a set of linear equations instead of solving a convex quadratic programming problem. It is shown the kernel processing equalization by adopting Gaussian cost function has several merit, such as: 1) The quadratic programming problem solved at each iteration is convex and has a globally optimal solution. 2) It avoids the difficulty of choosing the suitable parameters of the kernel function to obtain the satisfied blind equalization performance. 3) It need only 20% data samples of support vector machines (SVM) method to obtain the same blind equalization performance. 4) It is more robust for more nonlinear channels.