Modulation Recognition Algorithm Based On Partial Domain Adaptation

Aiming at the fact that the intelligent modulation recognition algorithm cannot adapt to the test data (target domain) that is inconsistent with the distribution of the train data (source domain), especially when the modulation signals types of the source domain and the target domain are different (the types of modulation signals in the target domain is a proper subset of the types of modulation signals in the source domain), the recognition performance deteriorates sharply, we propose a Class Weighted Domain adversarial Neural Network-based Partial Domain Adaptation Modulation Recognition (CWDA-MR) algorithm. We use the domain adversarial neural network as the backbone network, and applying class weights to the loss functions of the label predictor and the domain discriminator to improve the modulation recognition algorithm performance when the modulation signals source domain and target domain data distribution and modulation types are all different. Simulation experiments have verified its effectiveness and reliability. Compared with ordinary domain adversarial neural network-based modulation recognition algorithms, the accuracy of the modulation recognition algorithm is increased by 37.12%. Compared with training on the source domain and direct testing on the target domain, the accuracy is increased by 48.39%.