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

With the rapid development of wireless communication technology,the modulation mode and channel environment of communication signals are increasingly complicated.The determination of the modulation mode of the received signal is the basis of signal demodulation,and plays an important role in the field of military communication and civil communication.At present,most modulation recognition techniques use signal characteristics for classification.With the rise of deep learning technology,experts have gradually combined deep learning with modulation recognition,and proposed modulation recognition based on deep learning.The representative one is the modulation recognition under the CNN architecture proposed by Timothy J'O'Shea.The simulation results show that the scheme can achieve better recognition performance under low SNR.Based on the research,the paper uses the dataset published by Professor Timothy J'O'Shea to define the signal characteristic parameters of different modulation modes in the dataset,including the transient characteristics of the communication signal and the high-order cumulant.The simulation results of the characteristic parameters defined under different signal-to-noise ratios are given.On this basis,the signals are identified by different classifiers.The classifiers include support vector machine,K-nearest neighbor algorithm and neural network classifier.The paper discusses the factors affecting the classification performance of various classifiers,and chooses the best one.The experimental results show that under the Rayleigh fading channel,the recognition rate of the signal feature-based classifier exceeds the CNN-based modulation recognition when the signal-to-noise ratio is high.The modulation recognition accuracy of the CNN architecture is higher when the signal-to-noise ratio is lower.Both the high signal-to-noise ratio and the low signal-to-noise ratio CNN architecture have better recognition effects under the Gaussian channel.In order to study the influence of network architecture on modulation identification,this paper simulates and analyzes the modulation recognition performance of CLDNN architecture on the same dataset.The paper gives the network structure and training process,and discusses the modulation recognition accuracy under different channel environments.The experimental results show that compared with the CNN architecture,the recognition effect of the two network architectures under low SNR is not much different,but at high SNR,CLDNN performance is improved whether it is Gaussian channel or fadingchannel.Compared with the classifier identification scheme based on signal characteristics,the recognition effect of CLDNN under Gaussian channel is obviously improved.However,when the signal-to-noise ratio is low under fading channel,the recognition accuracy of classifier based on signal feature still dominates.Finally,the paper combines the signal features with CNN features and CLDNN features,and proposes a modulation recognition scheme with fusion features,and the training process is given.The recognition performance under the fusion scheme is simulated and analyzed.The experimental results show that the modulation identification scheme combining CLDNN features and signal features is a good choice in the case of unknown channel type and signal-to-noise ratio without considering the complexity of the model.