Research On Communication Signals Modulation Recognition Method Based On Machine Learning

The signal modulation recognition technology is in the middle position of the communication system model.It is located after the detection of the communication signal and before the demodulation of the signal.Signal modulation recognition in a non-cooperative communication environment is an extremely important key technology.And because the electromagnetic environment is more complicated and the channel environment is complex and variable,two key modules in the automatic signal recognition technology based on statistical mode are essential,that is,the extraction of stable characteristics of communication signals capable of representing different modulation modes,and the design of more accurate classifiers to adapt to more complex environments is an urgent problem to be solved.This research is mainly aimed at the detection of enemy signals during non-cooperative communication in order to determine the modulation mode of enemy communication signals in a complex noise environment.Therefore,the method of machine learning is studied and applied to the modulation pattern recognition of signals.In the process of identification,the two steps of feature extraction and classifier design are included.Because the quality of the extracted features needs to be quantitatively analyzed by the classifier,this paper firstly constructs an Extreme Learning Machine network based on particle swarm optimization and principal component analysis as a classifier and applies it to the modulation recognition of communication signals.On this basis,the two-dimensional Holder coefficient of the signal is extracted.The performance of several Extreme Learning Machine classifiers in different signal-to-noise ratios(SNR)is compared.The simulation results show that the Extreme Learning Machine network constructed in this paper has the highest classification and recognition accuracy.Secondly,for the problem that the recognition rate of the four types of digital signals is low when the signal-to-noise ratio is low,the Stacked Autoencoder is applied to the feature extraction of communication signals,and the PCA-ELM classifier is connected to construct a deep learning network based on Stacked-Auto-encoder and Extreme Learning Machine.The network is used to extract the characteristics of the timefrequency image of the signal and at the same time achieve the purpose of identifying the modulation pattern of the signal.The simulation results show that the proposed method can effectively identify the signal with a SNR of-2?2d B.Finally,for the problem that timefrequency analysis and deep learning networks are not enough to identify FM,PM and 16 QAM,in this paper,the Shannon entropy,exponential entropy and norm entropy of the seven types of signals(FM,PM,2ASK,2FSK,BPSK,QPSK,16QAM)are extracted to form a threedimensional entropy feature vector.On this basis,the three-dimensional entropy feature is combined with the cloud model theory to extract the improved entropy cloud characteristics of the signal,further improve the inter-class separation of various signals.The improved entropy cloud features are classified by PSO-ELM-PCA classifier.The simulation results show that the ideal recognition results of the signals can be obtained when the signal-to-noise ratio is-7?-3dB.