Modulation signal recognition based on lightweight complex residual attention neural network

To address the problem of poor recognition performance of real-valued deep learning models in modulation recognition applications and the practical need for lightweight in real applications, this paper proposes a lightweight complex residual attention neural network model. By constructing a complex convolution module, the complex convolution operation of the convolution layer is implemented to extract features in the complex domain of the modulated signal, and a residual attention mechanism is introduced to further select key features for recognition and classification. The experimental results show that compared with the real-valued deep learning model, the proposed model extracts more spatial variability in signal classification features and is easier to recognize, and increases the recognition rate by 10% when the signal-to-noise ratio is greater than OdB. In addition, the resource consumption of the model is reduced. Compared with the basic model, the model parameters are reduced by 99%, and the floating-point operation is reduced by 56%.