Automatic Modulation Format Classification of USRP Transmitted Signals Based on SVM

Automatic modulation classification (AMC) plays a key role in the cognitive radio systems. However, Quadrature Amplitude Modulations (QAM) are hard to distinguish by the AMC technology. What’s worse, most AMC models only function well in the Additive White Gaussian Noise (AWGN) channels where frequency offset is not considered. Hence, the performance of these models in frequency selective real-world scenario needs to be double-checked. In this paper, we propose an automatic modulation classification (AMC) method based on SVM. It analyzes the baseband signal and chooses four statistical features of it as the input of the SVM model. In order to have a better result to classify the 16QAM, 64QAM and 256QAM signals, the number of peaks in the distribution of amplitude is chosen as one of the features. In the presence of frequency off-set, the proposed method can classify seven kinds of modufation formats: BPSK (Binary Phase Shift Keying), QPSK (Quadrature Phase Shift Keying), 16QAM, 64QAM, 256QAM, GMSK (Gaussian Filtered Minimum Shift Keying) and OFDM (Orthogonal Frequency Division Multiplexing). The proposed method has been tested experimentally and shows attractive prospect.