Graph-based Automatic Modulation Classifier for M-ary Generalized QAM Signals

• Published in: 2019 3rd International Conference on Imaging, Signal Processing and Communication (ICISPC) pp. 6 - 9
• Main Authors: Yan, Xiao, Zhang, Guoyu, Luo, Jie, Wu, Hsiao-Chun, Wang, Qian, Wu, Yiyan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2019
• Subjects: adjacency matrix; automatic modulation classification; AWGN channels; Feature extraction; feature vector angle; generalized QAM; Monte Carlo methods; Quadrature amplitude modulation; Signal to noise ratio; Training; unified grid model
• DOI: 10.1109/ICISPC.2019.8935869

This paper presents a novel automatic modulation classification (AMC) method using graph-based constellation analysis, to classify M-ary Generalized Quadrature Amplitude Modulation (GQAM) signals which employ uniform or nonuniform constellations for the first time. In our framework, a unified grid model is first built from the GQAM signal with the maximum constellation size in the modulation candidate set, and exploited to transform the received signal into graph domain. The graph representation of the received signal is established by mapping its recovered symbol points on the I/Q plane into the unified grid model, and then the eigenvalue(s) and eigenvector(s) of its corresponding adjacency matrix are computed. The modulation feature vector is constructed according to the eigenvector(s) corresponding to the maximum eigenvalue(s). The modulation type is eventually identified by searching the minimum angle between training features and test feature. Monte Carlo simulation results demonstrate that the proposed method can effectively classify the GQAM even in low signal-to-noise ratio.