Modulation Recognition Of Communication Signals Based On Information Theory And Machine Learning

The main function of modulation recognition of communication signal is to recognize the modulation type of received signal blindly,so it plays a key role in non-cooperative wireless communication.It is often used for spectrum supervision in civil field and communication reconnaissance,communication confrontation and so on in military field.Aiming at the problems of low recognition rate under low signal-to-noise ratio,poor applicability of manual feature extraction and few modulation types identified by the existing modulation recognition algorithms,the modulation recognition algorithm based on feature extraction and machine learning methods is studied.The research includes two aspects:Firstly,based on the artificial feature extraction,we studied two kinds of artificial feature extraction methods: new information entropy feature and traditional cyclic spectrum feature.Four kinds of information entropy feature and one kind of cyclic spectrum feature are extracted.Then the neural network method is used to classify and recognize eleven kinds of modulation signals.In view of the shortcomings of LM-BP algorithm,such as slow training speed and easy convergence to local optimum,a more advanced method called GA-ELM is introduced.Then,A new modulation recognition algorithm based on information theory and GA-ELM is proposed.By using the advantages of ELM,such as strong generalization and no need of iterative training,the problems of slow training speed and low recognition rate in low SNR of LM-BP algorithm are solved.The experimental results show that the recognition rate of the algorithm can achieve 94% at 4d B in the simulation data set as well as 95% in the actual data set,which has better practical value.Secondly,based on the method of automatic feature extraction using machine,a modulation recognition method based on feature extraction by deep learning is studied,and a modulation recognition algorithm based on cyclic spectrum and ELM-LRF is proposed.By using the advantages of better anti-noise performance of cyclic spectrum and that ELM-LRF does not need iterative training,the problem of slow training speed and low recognition rate under low SNR of modulation recognition algorithm based on deep learning is solved.The results show that the recognition rate of the algorithm can achieve 96% at 4d B in the simulation data set as well as 90% in the actual data set.