Automatic Modulation Classification Based on Novel Feature Extraction Algorithms

• Published in: IEEE access 2020, Vol.8, p.16362-16371
• Main Authors: Zhang, Xiaolin, Sun, Jianting, Zhang, Xiaotong
• Format: Article
• Language: English
• Subjects: Algorithms; Automatic modulation classification; Classification algorithms; Classifiers; Feature extraction; feature selection; Modulation; Multilayers; novel feature extraction; Phase shift keying; Random noise; Signal classification; Signal processing; Signal to noise ratio; smooth support vector machine; Support vector machines
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2020.2966019

In order to solve the problems that the accuracy of modulation recognition algorithms of MSK and MQAM signals is not ideal under the condition of low signal-to-noise ratio (SNR) in Additional White Gaussian Noise (AWGN) environment, two novel features, the differential nonlinear phase Peak Factor (PF) and the reciprocal of amplitude envelope variance of cyclic spectrum at zero frequency cross section after difference and forth power processing, are constructed, which can complete the recognition of MSK signal and the classification of MPSK and MQAM signals respectively. This paper proposes a mixed recognition algorithm based on the two new features and other classical features, and design a tree shaped multi-layer smooth support vector machine classifier based on feature selection algorithm (FS_DT-SSVM) to recognize eleven kinds of digital modulation signals. The simulation results illustrate that the algorithm can achieve the classification of the modulation signals {2FSK, MSK, 2ASK, 4ASK, 8ASK, 2PSK, 4PSK, 8PSK, 16QAM, 32QAM, 64QAM} with small SNR. When the SNR is not less than -1dB, the recognition rate of the classifier for all signals exceed 97%, which validates the effectiveness of the proposed modulation recognition algorithm.