Small Sample Signal Modulation Recognition based on Higher-order Cumulants and CatBoost

It is difficult and costly to obtain sufficient label samples in an open environment, and the study of small sample problem has become an important direction in the field of signal modulation recognition. For the first time, we innovatively propose to use CatBoost to solve this. First, we extract high-order cumulants features from the I/Q signal, and then add slicing operations to make this feature suitable for algorithm training, and finally take advantage of CatBoost's high classification accuracy under small sample condition, to achieve effective recognition of small sample signals. The experiment obtained the results of comprehensive recognition accuracy of 93.3% and 95.1% when there are 20 and 200 samples in each type of 9 types of signals from 0 to 8dB, respectively. Compared with other traditional machine learning algorithms and deep learning algorithms, our method is more efficient.