Classifier Design Of Automatic Recognitiong Of Modulation Of Communication Signal

Modulation format is one of the most important characteristics used to distinguish communication signals. In many applications, it is required to monitor the activities of these signals, identify their characteristics, even to intercept the signal information content. So modulation identification for communication signals becomes a still important problem in the intercepted signal processing.The objective of modulation identification is to decide the modulation format and estimate the modulation parameters of the communication signals without any priori knowledge about the signal information content in the complicated signal environment with noise after analyzing the received signal, and to provide reference for farther analysis and processing. With the development of communication technology, the spatial signals are more and more complicated and dense. As results, there comes more demands for the research of modulation identification of communication signals.In near these decades, researchers explored many methods to solve the problem of the modulation identification for different modulation signals. Based on the research before, this dissertation has a study on the modulation identification of communication signals. The main contributions are listed as follows:Firstly illustrates kinds of daily modulation technology, gives the mathematical model of modulation signals, and observing the instantaneous amplitude, phase and frequency through emulation, which are the basic features of the signals and usually used for signal classification and recognition.For processing the received communication signal, extracts several kinds of characteristic of the signal, these are center instantaneous characteristic and wavelet transform characteristic, and has a emulation research on the arithmetic of each approach of feature extraction.Then we research on the topological structure and training algorithm of the MLP and RBF classifier. For the objection of the general BP neural network, adopt two kinds of method for improvement, which can efficiently increase the convergence speed of the network, these are center instantaneous characteristic, Based on the key features of the communication signals, these are center instantaneous characteristic and the wavelet transform characteristic presented in this paper, using MLP and RBF classifier, the simulation of modulation recognition successfully realizes the automatic recognition of various signals in a large SNR range, and obtains a satisfied outcome especially in low signal noise ratio condition.In order to solve the problem of conversion of multi-class modulation types which exists in software radio system, an algorithm based on support vector machine(SVM) for recognition of digital modulation signals was presentedo By analyzing the modulation signals, a set of key features for identifying different types of digital modulation were extracted, and were mapped into the high dimension space .The classification was carried out in the high dimension space based on SVM and decision tree, so the problem of nonlinear noirseparable classification in low dimension was resolved and the decision threshold became unnecessary.Better generalization ability was also acquired comparing with traditional neural networks o Least squares support vector machines(LS-SVM) is a new support vector machine. Due to equality type constraints in the formulation.the solution follows from solving a set of linear equations, instead of quadratic programming for classical SVM. This paper presents a multiclass least squares machines forclassification while LS-SVM is only for the case of two classes in the past. Furthermore it inducts new regularization and capacity control for it which improves accuracy and convergence of classification. The results show that the multiclass LS-SVM is an efficient classifier for solving pattern recognition.