Modulation Classification of Active Attack Signals for Internet of Things Using GP-CNN Network

Abstract

The traditional modulation classification method is difficult to cope with the changing wireless electromagnetic environment and the complex signal model. On this basis, this paper proposes a data-driven automatic modulation classification (AMC) method using a global pooling-based convolutional neural network (GP-CNN). Stepping convolution is used to replace the pooling layer to avoid loss of signal details and global pooling (GP) is utilized to replace the fully-connected for a lower computational complexity. Simulations verify the superiority of the proposed method, which outperforms other deep neural network methods and approaches the optimal bound of the maximum likelihood method. Moreover, the influence of the network parameters on performance is also explored.