A Modulation Recognition Method Based on Bispectrum and Ensemble Learning

Modulation recognition of emitters is of great significance for reconnaissance signal analysis. In this paper, a method based on bispectrum features and ensemble learning algorithm is proposed for modulation recognition. First, according to the application of signal modulation patterns in radar and communication fields, a set of signal samples including four modulation patterns is simulated. Then, statistical features such as variance, kurtosis and kurtosis are extracted from the bispectrum diagonal slices of signals to construct a bispectrum feature set, which used as the input feature of the recognition model. Finally, an identifier based on extreme gradient boosting (XGBoost) algorithm is constructed. Experimental results show that the bispectrum statistical features have satisfied feature expression and discrimination ability for signals with different modulation patterns, and the recognition model based on XGBoost algorithm has better recognition performance than the other seven recognition algorithms.