Semi-supervised generative adversarial network framework for modulation recognition of communication signals

Aiming at the problem that the accuracy of the existing modulation signal recognition model was low under the condition of weak supervision with only a small amount of labeled data, a semi supervised learning framework based on generated countermeasure network was proposed. By performing a redundant spatial transformation on the communication signals, the method can adapt to the generative adversarial network model and retain rich signal adjacent features. Through the introduction of Wasserstein generative adversarial network-gradient penalty, a semi-supervised learning framework suitable for electromagnetic signal processing was constructed to realize the effective utilization of unlabeled signal samples. In order to verify the effectiveness of the proposed algorithm, sufficient experiments were conducted on the RADIOML 2016.04C dataset. Experimental results show that the proposed method can train an efficient classifier under semi-supervised conditions and obtain excellent modulation recognition results.