Automatic digital modulation classification algorithm based on novel combined feature vector
基于联合特征向量的自动数字调制识别算法

In the high SNR processing domain, proposed novel high order statistic amplitude features and optimization method to preserve more classification information for various modulation types. The method based on the combined feature vector improved the algorithm performance compared to conventional features. In addition, adopted linear smoothing of the intercepted signal and normalization of input feature vector to restrain the noise and reduce the training time. Based on the Euclidean distance classification method and modified neural network recognizer, the simulation results verify the novel feature vector and optimization improve the average probability of correct classification by about 30% for more modulation types (MASK, MPSK, MFSK, MQAM) at low SNR with greater interference. The algorithm efficiency is also improved markedly.