The Research On Communication Emitters Identification Technology

It is very important to apply the radiometric identification techniques to improve the security of wireless communication systems and the ability of military communication countermeasure. By extracting small differences of physical layer information of radio device, which is generated during the design and production process, it is possible to discriminate radio emitters. However, it is significantly challenging to discriminate different radio emitters through radio frequency (RF)fingerprinting, because extracting these subtle features is very difficult.The theoretical framework of radiometric identification is presented in this thesis.Based on analysis of the procedure of radiometric identification, several algorithms are proposed and evaluated with experiments on real data sets. And parts of studied algorithms have been used in communication transmitter identification system.The main areas in this thesis are as following:1. The basic theoretics of radiometric identification is investigated. The transmission procedure of RF fingerprint feature is analyzed and the framework of radiometric identification is presented. Based on the coding and decoding model of RF fingerprint features, the theoretic performance of the radiometric identification system is analyzed. To minimize the error probability of identification, there should be no RF fingerprint information loss in the procedure of radiometric identification.2. Radiometric identification methods based on nonlinearity of transmitter are presented. Based on analyses of RF fingerprint mechanism, an identification method based on characterization of nonlinear behavior of power amplifier (PA) is studied. The model is simple and effective for describing the nonlinearity of the PA with weak-memory effects. To overcome that nonlinear behavior of one component is not enough to describe the subtle features, a radiometric identification method based on combined modeling of mismatches of modulator and nonlinear behavior of PA is studied. Further more, it is difficult to model the nonlinear behavior of all analog components in the transmitter. To address this problem, a novel radiometric identification method based on nature measure is proposed. The transmitter is considered as a nonlinearity system, and the features of system nonlinearity are directly extracted to effectively discriminate different devices.3. Radiometric identification methods based on statistical learning are presented. To avoid the loss of RF fingerprint information caused by the inexperience of individual in the procedure of radiometric identification, three identification methods based on statistical learning are proposed. Firstly, a radiometric identification method based on sparse representation of transient is proposed. It constructs an objective function by simultaneously maximizing reconstructed error of inter-class features and minimizing reconstructed error of intra-class features. The criterion of minimum reconstruction error of sparse representation is applied to predict the classification label of a test sample. Secondly, another radiometric identification method based on characterizing the intra-class compactness and the interclass separability by preserving both local manifold structure and the global discriminant information of signal data set is proposed. Lastly, a novel RF fingerprint learning algorithm based on information-theoretic criterion is proposed. To minimize the classification error, a new objective function is generated which simultaneously maximizing mutual information and minimizing error entropy.The fingerprint features extracted under this information-theoretic criterion are more discriminative.4, Radiometric identification methods based on combined optimization are presented. When the extracted RF fingerprint features are not relevant to the classifier,the performance of radiometric identification may deteriorate. To address this problem,two radiometric identification methods based on combined optimization are proposed.Firstly, a method based on combined optimization of dimensionality reduction and fingerprint classification is presented. It attempts to find an optimal dimension-reducing projection matrix by minimizing the classification error and maximizing the quadratic mutual information between the reduced low-dimensional features and the class label simultaneously. Then, a novel radiometric identification method is proposed. It attempts to find a low-rank representation matrix of original data and the optimal classifier parameter simultaneously by imposing a regularization item of minimum prediction error. The recovered radio frequency fingerprint features are discriminative and suitable for designated classifier.