Modulation classification based on cyclic spectral features and neural network

There is a need, for example in cognitive radio (CR), to determine the modulation type of an incoming signal. In this paper, an approach to classify modulated signals has been proposed. Firstly, extracting features from the spectral correlation function. Values of these features locate in different ranges, so they are suitable for classification. Since the spectral correlation function (SCF) is insensitive to noise, features obtained from it have good classifying performance even in low SNR. Subsequently the BP (Back Propagation) neural network was designed for pattern recognition. By combining the features extracted from spectral correlation function and using neural network for recognition, the classifier achieved excellent results for the AM, DSB, FM, 2FSK, 4FSK, BPSK, QPSK schemes. However, it didn't work well for 16QAM and 64QAM because they are quite similar. All the simulating results are presented in this paper. Finally, we give the conclusion, and other ways for separating 16QAM from 64QAM are also discussed.