The Influence of Introduced Channel Attention on Automatic Modulation Recognition Technology

Automatic Modulation Recognition (AMR) is a signal processing technique for identifying the signal’s modulation type when prior information about the signal is unavailable. AMR technology has broad applications in both military and civilian fields. This study utilizes the widely validated RadioML 2018.10A dataset, which includes more than 24 modulation formats such as On-off Keying (OOK), and trains convolutional neural networks (CNNs) under 26 signal-to-noise ratio (SNR) conditions ranging from -20 dB to 30 dB. To enhance network performance, this study introduces channel attention mechanisms and compare their effectiveness against standard CNNs, residual neural networks (ResNet), and CNNs augmented with the Squeeze-and-Excitation (SE) module. In addition, the author also evaluates the performance of CNN-SE, CNN under varying SNR conditions. Experimental results show that CNNs can maintain 70% accuracy when processing datasets with a 10db signal-to-noise ratio, while RNNs show suboptimal performance across most modulation formats. However, after the introduction of the SE module in the convolutional neural network, it can be observed that the same training effect can be achieved, and CNN-SE requires fewer training rounds. Its network convergence is better, the accuracy rate is up to 80%, and the performance is better under the condition of low SNR.