Radio transmitter identification based on bispectra with tensor representation
Yong Gong, Guyu Hu, Zhisong Pan · 2010
In this paper, we present a novel feature extraction and classification approach for radio transmitter recognition based on bispectra with tensor representation. Bispectra are quite effective to capture the stray features of radio transmitters in stationary state. Traditional approaches such as integral bispectra usually transform a bispectra matrix into a vector for feature extraction and classification, which fail to take into account the local geometry structure information of the bispectra matrices. In our approach, we consider a bispectra matrix as a second order tensor and extract features by tensor subspace analysis methods, including Tensor PCA and Tensor LDA. Then we use SVM and STM respectively for classification. Experiments show that our approach has good efficiency and achieves better recognition accuracy, especially when the training sample set is quiet small.