Dissertation > Excellent graduate degree dissertation topics show

Neural Network Modulation Recognition Classifier Based on Genetic Algorithm

Author: XieJiuNan
Tutor: WuZhiLu
School: Harbin Institute of Technology
Course: Information and Communication Engineering
Keywords: Digital Modulation Recognition Characteristics Extraction Genetic Algorithm Wavelet Neural Network
CLC: TP183
Type: Master's thesis
Year: 2007
Downloads: 155
Quote: 2
Read: Download Dissertation

Abstract

Digital modulation recognition has found important applications both in military and civil, while it is the required function of Software Defined Radio Receiver. Automatic recognition of modulation is the method of distilling proper feature parameters and then recognizes the modulation manner by proper sorting arithmetic. The main contribution of this dissertation includes three aspects. They are feature extraction, feature selection and classifier design. This dissertation presents features that have excellent quality, the method of feature selection that can gain optimal result, and Neural Network classifier.It researches four ways of characteristics extraction of signals including the method based on zero-center instantaneous feature, the method based on wavelet analysis feature, the method based on high order cumulants feature and the method based on fractal feature dimension. The experiment shows that these features are effective for classification. And we extract signal features by these ways separately, which construct an original feature set. The feature set contains the distinct information of each signal mode, but the degree of the assembling of the same species and the separating of different species in these features differs greatly, so they need to be selected in the next step.In the part of feature selection, we present a method based on GA (Genetic Algorithm). Different from traditional searching and optimizing methods, it does not evaluate one result of solution space of concrete parameters, but collaterally evaluates a lot of feasible results of entire solution space simultaneously, so it conquer the shortcoming of easily falling into local optimum. This dissertation researches the theory, basic manipulations, operation flow and mostly characteristics of GA. Then we detailedly design the arithmetic of GA for feature selection, and select the most optimal features from the original feature set to depress the dimension of feature space.DWNN (Discrete Wavelet Neural Network) is adopted to validate the validity of GA used in digital modulation recognition. To design the classifier, it centers on the research of DWNN, and gives a detailed analysis of the arithmetic of DWNN. Then it constructs a digital modulation recognition classifier, and defines the parameters of DWNN through experiments. In the end, the dissertation analyzes the capability of the classifier that uses the features selected by GA while comparing with the DWNN classifier that uses the five classical features. Simulation results demonstrate that the DWNN classifier that uses the features selected by GA gets better recognition effect than the DWNN classifier that uses the five classical features in the speed and the stability of the convergence, strong recognition ability, and the anti-noise ability. Besides, the recognition ability of the DWNN classifier that uses the five classical features changes obviously with different combinations of signal modes, but the DWNN classifier that uses the features selected by GA don’t have this problem. The forenamed results prove the superiority of GA in the domain of digital modulation recognition by comparing the upper two ways.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Automated basic theory > Artificial intelligence theory > Artificial Neural Networks and Computing
© 2012 www.DissertationTopic.Net Mobile