Automatic Identification Method Of Digital Modulation Based On Multi-domain Joint Classifier

As time progresses,communication technology is advancing at an unprecedented pace,resulting in a proliferation of digital modulation signal that are becoming increasingly intricate and diverse.The complex signal environment makes the traditional digital modulation signal recognition method difficult in practical application.Traditional recognition methods face the following shortcomings: the types of signals that can be recognized are not enough,and the stability and accuracy of recognition can not be guaranteed.With the advancement of science and technology,artificial intelligence is more and more applied to all aspects,and it also provides a new idea for the automatic identification of digital modulation signals.In this context,the recognition of digitally modulated signal is analyzed and studied in this thesis.Thesis implements a system that can automatically identify the type of digitally modulated signal,including signal generation,feature extraction,classification identification and interface design,and finally the system a series of tests were carried out,the specific work is as follows:1.First determine the type of digitally modulated signal to be identified.The modulation types selected in thesis include 2ASK,4ASK,8ASK,2FSK,4FSK,8FSK,MSK,GMSK,2PSK,4PSK,8PSK,π/4DQPSK,8QAM,16 QAM,32QAM,64 QAM signal.Thesis first briefly introduces the above-mentioned signals,and then generates digital signal sequences for subsequent processing.2.Thesis extracts high-order cumulant features,instantaneous features,and wavelet transform features for all signals.In thesis,the high-order moments are calculated,and then the estimated values of the high-order cumulants of various signals are obtained.Thesis extracts the instantaneous parameters of various signals,including amplitude,phase,frequency,and then calculates a series of instantaneous characteristic parameters that can be classified.In thesis,wavelet transform is performed on all signals,and some detailed information can be obtained,which can be used to distinguish between phase continuous FSK and phase jump FSK.In order to simulate the presence of noise in reality,all the above-mentioned features are extracted under the signal with Gaussian white noise added,and the signal-to-noise ratio spans from-5dB to 20 dB.3.In thesis,several classifiers based on multi-domain joint feature parameters are designed,and the high-order cumulant feature parameters,instantaneous feature parameters,and wavelet transform feature parameters are mixed.Thesis proposes and evaluates three different recognition methods: decision tree-based recognition,SVMbased automatic recognition,andBP neural network-based automatic recognition.Test results indicate that at a signal-to-noise ratio of 20 dB,all three methods achieve high recognition accuracy,with overall accuracy rates of 97.68%,99.28%,and 99.47%,respectively.And at the end of thesis,an interactive interface is designed to display the recognition process.