Research On Automatic Modulation Recognition Of Communication Signals In Large Dynamic SNR

With the development of software defined radio and cognitive radio technology,research on automatic modulation recognition(AMR) of communication signals basedon feature extraction and pattern recognition has made a lot of progress andachievements, but still can not meet the demand in large dynamic SNR environment. Toovercome the drawback, a new method starting from the robust feature selection withthe generalization ability in large dynamic SNR environment is explored in this article.First, the original features of the communication signals including transientfeatures, higher order cumulant features, wavelet features, fractal features and spectralcorrelation features are extracted and analyzed under different SNR. After that, noiserobust features are chosen by constructing anti-noise evaluation function.Then, in order to improve the classification capability and ruduce the informationredundancy of the noise robust features, the rough set theory is used to reduce it again.The experimental results show that the redundant features are reduced without affectingthe classification ability of the noise robust feature set, so that it can be more effectivefor classification.Finally, the recognition performance to the10kinds of communication signalsusing the reduced noise robust feature set is investigated with the application of SVM.The experimental results showed that, after training in a certain dB, the recognitionperformance of the SVM classifier using the reduced noise robust feature set is muchmore better than using the classic feature set or the initial noise robust feature set whentesting throughout the range of0~20dB. In addition, the excellent recognitionperformance of the new method in the small sample show that SVM classifier isappropriate for recognition. To sum up, the new method for the AMR of multimodulation communication signals explored in this article has the generalization abilityin large dynamic SNR environment with low computational complexity and highrecognition rate and is easy to recognition in real time.