Research On Modulation Recognition Technology Of Communication Signal Based On Deep Learning

In the field of cognitive radio,the most important thing is modulation recognition.Whether it is wired communication or wireless communication,it is necessary to determine the modulation mode of the signal to e nsure the establishment of communication,and AMC is a solution that does not affect the spectrum efficiency.Due to the limitations of traditional modulation recognition,the advantages of deep learning algorithms have been highlighted,bringing new progress to the AMC domain.However,modulation recognition is becoming more and more diversified,and with the shortage of channel resources,multiple types of signals coexist under the same channel,and aliasing occurs in bot h frequency and time domains.The purpose of this paper is to solve the problem of signal modulation pattern recognition and the problem of aliasing multi-signal modulation pattern recognition based on the existing neural network,and propose different solutions for different signal environments.This paper first proposes the AMC method for a known modulated single signal.Due to the limitation of traditional feature extraction methods,a convolutional neural network is proposed to classify and recognize th e modulated signal.Using CNN to directly learn features suitable for recognition in a large dynamic signal-to-noise ratio environment from massive samples through feature learning can improve efficiency and liberate manpower.And through testing and setti ng,it is verified that CNN can achieve high-accuracy recognition in different environments.Secondly,for the AMC method proposed for a single signal with an unknown modulation method,since the neural network can only achieve the closed set classification of the modulation method,this paper proposes a recombination judgment network,which is more suitable for one-dimensional modulation signals and realizes the unknown modulation method.Signal judgment.The network fully trained using diversified training tests is horizontally compared with the convolutional neural network proposed above,and it has good adaptability in open set classification.It has also had considerable generalization in different signal-to-noise ratio environments.Capacity and adaptability.Finally,the MAMC proposed for co-channel aliasing of multiple signals,in non-cooperative communications,electromagnetic spectrum supervision and other systems,when multiple narrowband signals appear in the same frequency band,the frequency domain is aliased from time to time.The existing AMC method cannot handle the blind recognition of aliased signals.Aiming at this problem,a modulation pattern recognition method based on one-dimensional capsule network for aliasing multiple signals is designed.The structure model of the capsule network is determined,and then the number of signals contained in the aliased signal is automatically recognized by the method of automatically identifying the target number based on the threshold judgment.The trained capsule network can realize the recognition of aliasing multiple signals.Compared with the traditional method and CNN method,this method has better recognition effect and also has good generalization ability and adaptability under large dynamic SNR.