Automatic modulation classification in [formula omitted]-stable noise using graph-based generalized second-order cyclic spectrum analysis

Abstract

An innovative automatic modulation classification using graph-based generalized second-order cyclic spectrum analysis in the background of α-stable noise is presented in this paper. In our proposed method, the three-dimensional generalized second-order cyclic spectrum of the modulated signal is first constructed from the nonlinear Hilbert transform. According to the generalized second-order cyclic spectrum, the signal is mapped onto a set of direct weighted rings in the graph domain. Then, the graph features of the modulated signal are extracted from the lower-upper decomposition of the corresponding adjacency matrices. The Hamming distance is employed to measure the distinctions between the feature parameters derived from the training and test data, and the modulation scheme can be identified thereby. Monte Carlo simulation results demonstrate that the proposed approach can achieve better classification accuracy than other existing methods for the α-stable noise scenario.