基于无监督领域自适应的调制识别算法

Modulation recognition is a critical task in wireless communications, and deep learning methods have achieved remarkable progress in this domain. However, their generalization ability remains limited in complex and non-cooperative environments. Existing models are typically trained on specific datasets and struggle to adapt to varying channel conditions, unknown signal parameters, and unlabeled target domains. To address these challenges, this paper proposes an unsupervised adversarial domain adaptation method for cross-channel modulation recognition. By aligning feature distributions without relying on labeled data, the proposed approach enhances the model's recognition capability under diverse channel conditions. Experimental results demonstrate that the method significantly improves modulation classification performance in unlabeled scenarios, exhibiting strong cross-domain generalization.