A Recognition Method for Complex Modulated Radar Signals Based on Enhanced Generative Adversarial Network Data Augmentation in Small-Sample Conditions

This study presents a method for recognizing small sample sets of complex modulated radar signals, which relies on data enhancement using an improved Generative Adversarial Network (GAN). This technique addresses the low recognition and imbalanced recognition rate problems often encountered with small samples of complex modulated radar signals. Initially, we propose the concept of Ambiguity Function Efficient Contour Lines (AFECL) and its resolution constant. Based on this, we introduce the Ambiguity Function Contour Lines Generative Adversarial Network (AFCLGAN), a model specifically designed for the data enhancement of complexly modulated radar signals. In simulation experiments, six different types of radar signals were successfully classified and identified, proving the efficacy of our proposed model. The comparative experimental results show that the overall recognition rate increased by 80%, while the unbalance in class recognition decreased from 41 % to 1%. Notably, the AFCLGAN model significantly outperforms traditional GANs and general WGAN-GP models in terms of data enhancement. The effectiveness of the AFCLGAN-enhanced data approach is corroborated via comparative experiments.