| With the rapid development of communication technology,the communication environment becomes more and more complex,and the modulation patterns of signals become more and more diversified.In the field of un-cooperative communication,the basic premise for the receiver to demodulate the signal is to know the modulation mode and related parameters of the received signal.And the traditional hardware circuit has been unable to meet the diversity of signal modulation style,so the SDR(software Defined radio)technology which combining microelectronic technology and computer technology appears.The idea of SDR is to use a general digital signal processing platform,so that the signal modulation and demodulation processing can be designed in the software module,and the automatic modulation recognition of signal as one of the core technologies of software radio has become the current research hotspot.This thesis studies and analyzes the modulation recognition of digital signal under the SDR.Based on the study of instantaneous features and convolutional networks,a multi-channel feature fusion method is proposed to realize modulation recognition of digital signals,and the experimental results of different methods studied in this thesis are compared and analyzed.In this thesis,the main technology of software radio in signal processing is briefly described,and introduce the 6 kinds of commonly used digital modulation signal models and their modulation characteristics.Then in the real electromagnetic environment,this thesis built a data transmission acquisition platform by two software radio hardware devices.The receiver collects the modulation signal of the transmitter through the wireless channel,which is used as the input data set of the research method in this thesis.The modulation recognition method of radio signal is realized according to the instantaneous characteristic parameters of digital modulation signal,and the threshold value of instantaneous characteristic is calculated,adj usted and optimized through experimental simulation.Then the application of Convolution Neural Network(CNN)in modulation recognition is studied.In view of the shortcoming of artificial classification threshold in the recognition method based on instantaneous features,a CNN model suitable for modulated signals in this thesis is built on the basis of previous studies on CNN network.According to the characteristics of the time domain data,the data set is sliced in time domain to form the input data set suitable for the CNN network.The optimal structural parameters are selected by comparing the CNN networks with different structures,so as to realize the design of the CNN model in this thesis.Since it can automatically learn classification features and classification rules,there is no need to manually divide signal features.In addition,the filtering characteristics of the convolution kernel can reduce the influence of noise on the signals.The CNN model can solve the problem that the recognition method based on instantaneous features has low recognition rate under low SNR.Finally,this thesis designs a method based on multi-path feature fusion.The instantaneous features extracted manually,the time domain features extracted by CNN network and the time-frequency graph features extracted by CNN are fused.The new feature vectors are identified and classified by a fully connected neural network classifier.For the problem of dimensionality inconsistency of three-way features,by changing the convolutional layer structure of the model,the features output by different networks are all one-dimensional feature vectors,and then fusion is carried out in the form of feature series.Finally,classification calculation is carried out through classifier.In this model,the advantages of different features are complemented by feature fusion,which can further improve the modulation recognition performance and solve the identification confusion between classes in the process of CNN network recognition. |
PDF Full Text Request