Communication Signal Modulation Recognition Algorithm Based on Soft Threshold and Hole Convolution

The current communication signal modulation recognition based on end-to-end deep learning methods is susceptible to signal noise, resulting in poor recognition performance under low signal-to-noise ratio(SNR) conditions. To address this issue, this paper presents a modulation recognition convolutional neural network that combines soft threshold and hole convolution(STCNN). Firstly, adaptive filtering of useless features such as noise is achieved through a soft threshold layer and attention mechanism. Secondly, by using dilated convolutional networks to extract multi-scale time-frequency domain features of signals, the number of model parameters is reduced while the accuracy of signal classification is improved. Finally, simulation analysis was conducted based on the RML2016.10a dataset, and the highest recognition rate of 62.7% was achieved for 11 modulation signals under different SNR conditions, which is 10.9%, 4.5%, and 1.6% higher than that of CNN, LSTM, and SCRNN algorithms, respectively. Furthermore, the number of parameters is reduced by 98.8%, 82.5%, and 91.2%, respectively, which is simulated to verify the effectiveness and recognition performance of the proposed method.