Blind equalization of multilevel signals via support vector regression

In this paper, two families of batch algorithms are constructed in the support vector regression (SVR) framework for blind equalization of multilevel signals. Specifically, the error functions of constant modulus algorithm CMA(p, 2) and multimodulus algorithm MMA(p, 2) are contained in the penalty term of the SVR. Simulation results show that the proposed MMA(p, 2)-based algorithms perform better than the CMA(p, 2)-based ones, which exhibit lower residual intersymbol interference (ISI) and higher probability of convergence. With respect to conventional dual-mode scheme, the MMA(p, 2)-based algorithms show better performance in the case of higher noise or smaller data block, therefore they are robust and more suitable for multilevel signals. In addition, they avoid tedious switching mechanism of dual-mode scheme and overcome phase rotation.