A New Dataset and Benchmark for Cross-Domain Automatic Modulation Classification

Automatic Modulation Classification (AMC) is a promising technique for automatically analyzing the modulation type of the received signals, and it plays an important role in both military and civilian fields. Recently, Deep Learning (DL) has been introduced into AMC for better performance. However, these previously proposed methods have not been tested for generalization and robustness, due to the lack of crossdomain datasets, such as those varying in environment and time. To tackle this challenge, we developed a new dataset for AMC, dubbed “CD2024”. This dataset comprises modulation signal samples from four distinct scenarios and encompasses 25 modulation types, totaling 50,000 samples. Utilizing this dataset, we evaluate the robustness of existing DL-based AMC methods. The experimental results indicate that these methods still exhibit significant potential for enhancement when dealing with crossdomain data.