Unsupervised domain adaptation for variable-SNR signal modulation recognition

When the source domain and target domain differ in Signal-to-Noise Ratio (SNR), models trained on source data tend to perform poorly on target data. Given a labeled source dataset, unsupervised domain adaptation (UDA) aims to predict the labels of the target domain dataset by aligning the features of the source and target domains. Motivated by UDA, we propose a method that aligns the feature representations of the source and target domains to enable modulation classification under variable SNRs. Our experiments demonstrate that this approach results in a performance improvement of 3.9% to 15.39% compared to a model trained only on the source domain.