Font Size: a A A

Researches On Digital Modulatin Recognition Based On Super-Feature

Posted on:2014-03-30Degree:MasterType:Thesis
Country:ChinaCandidate:B ChengFull Text:PDF
GTID:2268330401966924Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
Much attention has been paid on the modulation recognition studies, and it playsan important role in both military and civilian applications. With the development ofSDR(Software Defined Radio) and OFDM(Orthogonal Frequency DivisionMultiplexing) technologies, requirements for modulation recognition techniques alsoincrease. The algorithm is required to be implemented conveniently, meanwhile, itneeds to adapt to various modulation types and obtain high recognition accuracy.Based on the literature study, much profound research on modulation recognitionhas been done. The work includes in this thesis following:Firstly, The modulation theory was introduced, and the common feature extractionmethods(instantaneous information, spectral, wavelet transform, high order cumulants,etc.) were reviewed. And then, the distribution of classification feature was analyzedbased on the simulation results, Owing to the influence of the noise and data lengthlimitation, the classification performance on a single feature is limited, thus wesuggested the new concept of super-feature, which is capable of extracting the featurevector, and eventually recognizes modes in a simple space.Benefit from the super-feature, we combine the genetic algorithm and KNNclassifier together to study the modulation recognition method. For the basic problem,such as recognizing QPSK and16QAM,based on the7value of high order cumulant,the classification parameter T was built by c42and c60,and the decision tree wasapplied to recognize the two types; Then, all the features were put into the system, afterfinishing the evolution, we get the super-feature,in the hyperplane, the featureprovides a high intra-class variance and a low inter-class variance, indicating that theintrinsic mode extraction is effective. Then, some complicated issues in the system werediscussed, for instant, the K-value, the reference sample selection, the design of thefitness function and the strategies for multi-class classification.A case study on modulation recognition was researched under a specific scene. Thecharacters of several common TT&C signals (PCM/FM, PCM/BPSK, PCM/QPSK,PCM/UQPSK, PCM/BPSK/PM(standard TT&C)) were studied. By considering whether the signal residues carrier component, square spectrum spectral projections, thedistribution of the psd and the number of peak, four classification features wereproposed, decision tree and genetic algorithm methods were both applied forclassification. The simulation results verified that the proposed features were effective,and reach90%(accuracy) beyond10dB in the case study, consequently, we claim thatthe genetic algorithm is capable in optimize the recognition performance.
Keywords/Search Tags:modulation recognition, TT&C, super-feature, decision tree, geneticalgorithm, KNN classifier
PDF Full Text Request
Related items