An Improved Encoder Network Method for Modulation Recognition

Automatic modulation recognition (AMR) is a key technology in wireless communication application systems, and the application of deep learning improves the accuracy of AMR. Existing deep learning based modulation recognition methods can only extract local features of the signal, usually by increasing the number of model layers to extract deeper features of the modulated signal, which makes the model complexity enhance the modulation recognition performance degradation. Aiming at this type of problem, an AMR method with TCB (Transformer-Context Broadcasting) model is proposed. The Transformer model is introduced into AMR to extract globally relevant features of the signal, and the unified attention is inserted manually to increase the signal denseness and obtain higher classification accuracy. Experimental results on the dataset show that the TCB model improves the average recognition accuracy by about 4.3%-10.5% compared to existing network models such as CNN2, LSTM, ResNet, CLDNN, etc., which verifies its superiority.