摘要
针对复杂电磁环境下无线电信号调制识别精度低的问题,提出了一种基于可学习小波变换和Transformer融合的调制识别方法。首先,通过可学习小波变换模块将信号进行奇偶分解,利用强化的预测、更新算子和注意力机制自适应提取多分辨率特征,同时引入正则化约束确保小波分解的稳定性;其次,构建双分支特征增强架构,通过挤压和激励(SE)注意力对小波特征进行自适应加权,利用Transformer捕获全局依赖关系;最后,将两个分支输出的特征在特征维度拼接后输入到全连接分类器中,以进行调制类型识别。实验结果表明,所提出的模型具有优异的调制识别精度。相较于其他深度学习方法,所提方法的整体识别精度提升了3%~10%,在不同信噪比的条件下均具有更强的特征学习能力和更好的鲁棒性。
To address the issue of low accuracy in radio signal modulation recognition under complex electromagnetic environments,a modulation recognition method based on learnable wavelet transform and Transformer fusion is proposed.First,the signal is subjected to odd-even decomposition via the learnable wavelet transform module.The enhanced prediction and update operators,along with the attention mechanism,are employed to adaptively extract multi-resolution features.Meanwhile,regularization constraints are introduced to guarantee the stability of wavelet decomposition.Then,a dual-branch feature enhancement architecture is established.The SE attention mechanism adaptively weights the wavelet features,and the Transformer is utilized to capture global dependency relationships.Finally,the features output from the two branches are concatenated in the feature dimension and then fed into a fully connected classifier for modulation type recognition.Experimental results demonstrate that the proposed model exhibits excellent modulation recognition accuracy.Compared with other deep learning methods,the proposed method achieves an overall recognition accuracy improvement of 3%~10%,demonstrating stronger feature learning capabilities and better robustness under various signal-to-noise ratio conditions.
作者
田明浩
杨盼云
姚沐汐
TIAN Minghao;YANG Panyun;YAO Muxi(School of Information Science and Engineering,Shenyang Ligong University,Shenyang Liaoning 110159,China)
出处
《通信技术》
2026年第1期31-37,共7页
Communications Technology
基金
辽宁省教育厅高等学校基本科研项目面上项目(JYTMS20230222)。
