A symmetric adaptive visibility graph classification method of orthogonal signals for automatic modulation classification

Visibility graph methods allow time series to mine non-Euclidean spatial features of sequences by using neural network algorithms. Unlike the traditional fixed-rule-based univariate visibility methods, a symmetric adaptive method is proposed orthogonal signals, applicable in-phase and quadrature (I/Q) signals for mapping radio modulated in automatic modulation classification tasks. The directly models intra-channel inter-channel relations I/Q two different types convolutional kernels. It captures feature information combining sampling aggregation differentiable pooling as extractor. Extensive experimental results on benchmark datasets simulated dataset containing channel fading show that Quadrature Signal Symmetric Adaptive Graph (QSSAVG) this paper outperforms terms accuracy also more robust against noise variations.