A deep learning algorithm with soft threshold for automatic modulation recognition

Due to the increasing complexity of electromagnetic spectrum environment, noise has significant influence on correct modulation recognition. In this paper, a deep learning model based on soft threshold is designed under the condition of low signal-noise ratio (SNR). The model is based on the convolutional neural network(CNN), and the soft threshold function is added. The IQ data is transformed into amplitude and phase information as the input of the model. The CNN is used to extract the features from the amplitude and phase data, and the soft threshold learning network can learn different thresholds aims to different features, which filters out the sample noise and improves the recognition rate under the condition of low SNR. Validation on the open source data set RML2016.10a shows that compared with other network structures, the model proposed in this paper has higher recognition rate and lower time complexity.