As modulation recognition based on deep neural networks has a problem of model mismatch,which leads to a significant decrease in recognition rate,this paper proposes a modulation recognition algorithm based on domain adaptation.When there is a large distribution difference between test and training samples,it still has a high recog-nition rate.First,the algorithm processes the constellation diagram of modulated signals under pulse noise to recov-er geometric features.It then enhances the features and generates images,which are input into a deep domain ad-aptation network to complete modulation recognition.This algorithm can realize modulation recognition under both generalized signal-to-noise ratio with a large dynamic range and pulse intensity with a large dynamic range.Simula-tion results show that this algorithm can solve the model mismatch problem effectively,having good recognition per-formance.
维普
万方数据