An new method for fast fading channel prediction using Recurrent Least Squares Support Vector Machines (RLS-SVM) combined with reconstructed embedding phase space is investigated.