Algorithm of digital modulation recognition based on support vector machines

A new algorithm based on high order cumulants (HOC) and support vector machines (SVM) for modulation and recognition of digital communication signals is proposed. The new method can identify six digital modulation signals: 2ASK, 4ASK, 8ASK, 4PSK, 8PSK and 16QAM digital signals using fourth and sixth order cumulants of the signals as vectors and SVM based on binary tree as classifiers. The method uses new class distance as rules of constructing binary tree, which separates the furthest class from others first, so that the method can maintain high generalization ability. The computer simulation results justify the method´s validity.