Research on signal modulation identification method based on residual neural network

Aiming at the problems of large number of parameters, high complexity and low recognition accuracy of the current modulation recognition model based on high-performance neural network, a lightweight signal modulation recognition algorithm based on residual neural network (ResNet) is proposed. Firstly, the ResNet network model is built, which contains two residual units and three fully connected layers. Secondly, the pruning method of network slimming is introduced to compress the neural network model without affecting the recognition accuracy. Finally, the model was trained using modulated signal samples for the RaidoML2016.10a international standard dataset. The experimental results show that the recognition accuracy of the pruning ResNet modulation recognition algorithm can reach 95% under the condition of 12dB signal-to-noise ratio, and the number of model parameters is only 5.7×104.