Blind equalization of multilevel signals using support vector machines

The support vector machine (SVM) has been recently proposed for blind equalization of constant modulus signals. In this paper we extend this previous work in two directions: first, the high computational cost of the original procedure is significantly reduced by transforming the original quadratic programming (QP) problem into an equivalent least squares problem. Secondly, the penalty term of the SVM is now a Godard-like error function; therefore, the proposed procedure allows the equalization of multilevel signals. A dual mode algorithm is also proposed: once convergence is achieved, the Godard-like penalty term is switched to a radius directed-like error function, which reduces the final intersymbol interference (ISI) level. Simulation experiments show that the proposed SVM equalization method performs better than cumulant-based methods: it requires a lower number of data samples to achieve the same equalization level and convergence ratio.