In this letter, we propose a novel deep learning based robust automatic modulation classification (AMC) method for cognitive radio networks. For the deep learning technique a robust convolutional neural network (CNN) is considered. The conventional methods for a input size of CNN use 2¡¿1024, which represents in-phase and quadrature-phase (IQ) components. Otherwise, in this letter 4¡¿1024 matrix is considered to improve the classification accuracy, which is combined with the input IQ components and the copied IQ components that are flipped and concatenated. Since the increase in the amount of computation due to the input size, the size from 4¡¿1024 to 2¡¿1024 is reduced by an average pooling layer in the proposed CNN architecture.
The simulation results show that the classification accuracy of the proposed model is higher than the lightweight model and MCNet in the overall signal-to-noise ratio (SNR) range.