Training Images Generation for CNN Based Automatic Modulation Classification

Convolutional neural network (CNN) models have recently demonstrated impressive classification and recognition performance on image video processing scope. In this paper, we investigate the application of CNN to identifying modulation classes for digitally modulated signals. First, received baseband data samples signal are gathered up transformed generate constellation-like training images convolutional networks. Among resulting images, proposed gray is preferred inference because lower computational burden. Second, propose use a multiple-scale (MSCNN) as classifier. The skip-connection technique deployed mitigating negative effect vanishing gradients overfitting during process. Numerical simulations been carried out validate effectiveness scheme, results show that scheme outperforms traditional algorithms in terms accuracy.