Research on Modulation Classification Based on High-order Cumulants and Fractal Theory

The method of feature extraction based on the combination of high-order cumulants and fractal theory is presented. Cumulants with advantage of good anti-noise performance is used to classify the modulation types widely. But the high-order cumulants of 2ASK and BPSK are equal, as well as 2FSK, 4FSK, and 8FSK, which leads to insufficiency of modulation classification. In this paper we import fractal theory to solve this problem. High-order cumulants and fractal box dimension of the received signal are extracted to build joint feature parameters. Cascade neural network classifier is structured in order to improve classification efficiency in this technique. Seven types signals as 2ASK, 4ASK, BPSK, 4PSK, 2FSK, 4FSK, 16QAM are used in simulation and the results show that the joint feature extracted by this method has lower compute complexity and better anti-noise performance. Correct classification rate is more than 85% under the condition that signal-to-noise ratio was higher than 5 dB and test samples were more than 200.