Research And Implementation On Jtids Signal User Sorting Classifier

Combined with JTIDS signal characteristic, this thesis studies some representative classification algorithms in the application of JTIDS signal user sorting, such as support vector machine, clustering and neural network, and then analyzes and compares their sorting performance. Based on these researches, two schemes of JTIDS signal user sorting classifier and software platforms are built, respectively aiming at whether the number of the user sorts is known or not. The primary work is summarized as follows:1. The research of JTIDS user classifier based on SVM is carried out in this thesis. Some transformative algorithms have also been investigated, the performance of these algorithms above have been analyzed and compared by means of computer simulation. And LS-SVM has been chosen as the preferred algorithm. Taking LS-SVM for example, multi-class algorithm has been studied in close integration with the multi-user characteristic of JTIDS signal.2. JTIDS user classifier based on clustering has been researched. Integrated with actual JTIDS signal, the algorithm based on K-means has been discussed. While aiming at the disadvantage of both that its clustering number and the initial clustering center can't be ascertained appropriately, an enhanced one has been proposed which can ascertain the clustering number and the initial clustering center adaptively and also can avoid the subjectivity and blindness in making the choice of the clustering number and the initial partition. Moreover, the algorithm SVC has been studied, and the better classifier based on clustering has been selected—the novel K-means algorithm, on the ground of the analysis above.3. JTIDS user classifier based on neural network has been researched. On the basis of analyzing RBFN, aiming at the disadvantage that the crytic layer number, crytic layer width and RBF center can't be ascertained suitably, a new algorithm K-RBFN which combined RBFN with K-means has been proposed, using K-means to ascertain the parameters in RBFN. In addition, SOFM algorithm has been investigated. Combined with the merit of SOFM and LS-SVM, a new integrated algorithm has been presented, making use of ameliorated SD validity function to ascertain the clustering number of SOFM whose classification performance is greatly improved.4. After analyzing the classification performance, aiming at whether the number of the user sorts is known or not, two kinds of schemes of JTIDS user sorting classifier and software platforms are respectively implemented, and tested with simulation data, and the results have validated the feasibility of the scheme and the reliability of the software platform.Finally, the conclusion is given in the last part of the thesis, which contains the shortcomings, the improvements needed, and the prospect for the future work.