Automatic Modulation Classification Based on Improved R-Transformer

Cognitive radio technology is an essential branch in wireless communication, and automatic modulation classification (AMC) is an essential component of intelligent communication systems. As the spectrum traffic becoming congested, the fast modulation classification becomes challenging. While modulation classification has been a well-studied problem, achieving high accuracy from a small number of samples is challenging. This paper builds on previous work by using an R-transformer-based network to identify a complex open-source RadioML dataset. We obtain several network structures suitable for AMC. The network consisting of CNN and attention mechanism can rely on minor parameters and minimal model size to obtain the best performance. Still, the disadvantage is the requirement of the input sequence length. In contrast, an improved-RT model can solve this problem. Compared to existing results, networks proposed in this paper can achieve better accuracy with very few required parameters.