A Modulation Recognition Method Based on Adaptive Feature Fusion
Guohao Liu, Bo Qian, Zongnan Liu, Chunshan Hao · 2025
To address the issue of decreased modulation recognition accuracy caused by the insufficient feature extraction capability of traditional convolutional neural networks under low signal-to-noise ratio (SNR) conditions, this paper establishes a dual-channel convolutional long short-term memory network based on multi-dimensional signal features to achieve signal modulation recognition. Attention mechanism module and feature fusion module are introduced for suppressing redundant information and extracting deep features. This enhances the modulation recognition capability under low signal-to-noise ratio conditions, achieving high-accuracy recognition. Simulation results show that the proposed algorithm in this paper achieves $73 \%$ recognition accuracy at signal-to-noise ratios of -20 dB to 10dB. Recognition accuracy reaches $90.5 \%$ at signal-to-noise ratios higher than 2 dB.