Individual Transmitter Identification Research Based On Spurious Characteristics

As a key technology in the area of communication countermeasure, individual transmitter identification has aroused widespread attention. With the increasing complexity of the composition of the modern communication devices, it has become a hot research content about how to intercept and analyze the individual characteristics reflected in the communication signals and classify them effectively. Among the current domestic research, the identification of individual transmitter is mainly focused on studying the features of different kinds of individual devices, moreover, feature classification should rely on sufficient numbers of samples, while foreign research is mainly focused on transient signals. Based on the related research results in recent years, this paper chooses the spurious features of similar individual transmitter signals as research objects and studies the extraction and classification methods for them.In terms of feature extraction, this paper researches the spurious characteristics on marginal spectrum of steady transmitter signals. The energy gravity and information entropy of signal marginal spectrum are calculated by the united extraction method of Empirical Mode Decomposition(EMD) and Wavelet Packet Reconstruction. However, the EMD method has a shortcoming of mode mixing and the standard of Wavelet Packet Reconstruction relies on experience in the original extraction method. Therefore, this paper proposes a improved scheme. The reason of mode mixing reflected in EMD method is firstly analyzed, moreover, the standard of Wavelet Packet Reconstruction and the input of the EMD method are improved. Furthermore, time-frequency spectrum superimposition is adopted to reduce the impact of mode mixing in the calculation of marginal signal spectrum. Ultimately, the poor aggregation and severe doping level of spurious characteristics are improved and the separation extent is promoted by a certain level.In the respect of classification recognition, this paper sets two groups of classification conditions for simulation. K-Nearest Neighbor algorithm, Neural Network classifier and Support Vector Machine(SVM) classifier are utilized to compare the classification performance of different modulation signals. Among three classification methods, SVM has the best performance and is adopted to compare the classification effect between the original feature extraction method and the improved one. The simulation result shows that the improved method has increased the classification performance of transmitter signal features and has a good applicability in different modulation types.