An attention-based algorithm for wireless signal modulation scheme recognition under low signal-to-noise ratio conditions

Abstract

Automatic modulation recognition is a key step for spectrum analysis and signal processing in cognitive radio and non-cooperative communication systems. However, traditional recognition methods lack robustness. Existing deep learning models achieve high accuracy but involve too many parameters and are hard to deploy on edge devices. To solve these problems, this paper presents a lightweight modulation recognition algorithm that integrates a one-dimensional ResNet (ResNet-1D) with the Convolutional Block Attention Module (CBAM). Experiments are carried out on the public dataset DeepSig RadioML2018.01A. The results show that the proposed algorithm obtains the best performance in the low SNR range from -10 dB to 0 dB. It effectively overcomes the drawbacks of traditional models under low-SNR conditions and shows strong robustness. Meanwhile, the lightweight structure makes the algorithm suitable for practical deployment.