Research On Automatic Modulation Recognition Algorithms Based On Deep Learning

Automatic modulation recognition is a key technology in wireless communication.It is the basis of parameter estimation,spectrum monitoring,modulation and demodulation,and is widely used in many fields such as information interception,interference selection,type identification,and radiation source classification.However,with the development of communication technologies,the environment of wireless communication becomes more and more complex,and the types of modulation are more diverse,which bring great challenges to modulation recognition technology.Therefore,it is urgent to study efficient and intelligent automatic modulation recognition methods.Deep learning,as a new technology in the field of artificial intelligence,has broken through many scientific and technical bottlenecks and is applied to areas such as image recognition,text classification,and speech recognition.Therefore,the use of deep learning technology in the field of the automatic modulation and recognition of communication signals is the future development trend.Based on the existing deep learning technology,this paper conducts an in-depth study on the automatic modulation and recognition method of signals.The details are as follows:?1?Based on the basic VGG network structure,an improved VGG unit model suitable for modulation identification is studied,and a combined network structure of VGG-LSTM is proposed,which is for PSK2,PSK4,CPM2,CPM4,FSK2,FSK4,QAM16,etc.Seven digitally modulated signals are classified and identified.The simulation results show that compared with the modulation recognition method based on CNN network proposed by Timothy J O'Shea et al.,the improved VGG-LSTM network performance is more excellent;when the experimental environment is completely the same,and the signal-to-noise ratio is5dB,VGG-LSTM network ₁F-score is up to 90%,with the performance increasing by26.3%.especially in low SNR environment,its advantages are more obvious.?2?Based on the basic Inception network structure,the improved Inception unit model suitable for modulation recognition is studied,and a combined network structure of Inception-LSTM is proposed,which is for PSK2,PSK4,CPM2,CPM4,FSK2,FSK4,QAM16,etc.Seven digitally modulated signals are classified and identified.The simulation results show that compared with the modulation identification method of the first improved network VGG-LSTM model proposed in this paper,the improved Inception-LSTM network performance is more excellent;when the experimental environment is completely the same,and the signal-to-noise ratio is 0dB,the Inception-LSTM network ₁F-score is up to 97.4%,with the performance increasing by2.8%based on VGG-LSTM.It can be seen that the improved Inception-LSTM composite network structure has a more classification and recognition function than the VGG-LSTM network.