CST: Automatic Modulation Recognition Method by Convolution Transformer on Temporal Continuity Features

With the rapid development of deep learning (DL) in recent years, automatic modulation recognition (AMR) with DL has achieved high accuracy. However, insufficient training signal data in complicated channel environments is critical factors that make DL methods difficult to deploy in practice. Aiming to these problems, we propose a novel neural network named convolution signal transformer (CST). The CST is accomplished through three primary modifications: a hierarchy of transformer containing convolution, a novel signal-specific self-attention mechanism to replace the multi-headed self-attention mechanism in Transformer, and a novel convolutional transformer block named convolution-transformer projection (CTP) to leverage a convolutional projection. The simulation results demonstrate that the CST outperforms advanced neural networks on all datasets, which is very beneficial for the deployment of AMR in complicated channel environments.