Research On The Recognition Algorithm Of Radar Signal Modulation Type Based On Deep Learning

With the coming of the technological information age,electronic science and technology continue to advance,the radar electromagnetic environment tends to be complex and diversified,and the parameter characteristics and modulation types of the signals have changed a lot.However,the classification performance of radar signal recognition method using traditional information features is difficult to meet the actual needs.The combination of deep learning technology and signal features into the field of radar signal recognition has become the mainstream,however,there are also suffers from weak identification effect and poor generalization performance at low SNR.In view of the above matter,constructs the radar original IQ data set and the time-frequency image data set,and proposes two kinds of radar signal modulation type recognition algorithms,and designs and develops the radar signal modulation type recognition system.The main work is as follows:(1)Eight modulation types of common radar signals are introduced,and their time-domain and frequency-domain characteristics are analyzed.In addition,three time-frequency analysis methods STFT transform,WVD distribution,and SPWVD distribution are used to perform time-frequency conversion on eight signal modulation types.Laying the foundation for subsequent accurate identification.(2)A radar signal modulation type recognition algorithm based on one-dimensional feature sequence and improved CNN is presented.Network architecture of the selected CNN,a convolutional layer module with a convolution kernel of 5×1 and a convolution kernel of 3×1 is designed,and the CBAM attention module is used to deepen the network’s attention to the IQ signal features in the channel and spatial dimensions.Through experimental verification,it is confirmed that the algorithm can reduce the time required to generate time-frequency images by simulation,can accurately identify the type of signal modulation,and can meet some real-time requirements.(3)We proposed a radar signal modulation type identification algorithm based on time-frequency analysis and improved Res Net-50 residual network.Firstly,the algorithm converts one-dimensional time-domain IQ signal data into two-dimensional time-frequency images as the input of the network through three time-frequency analysis methods: STFT transformation,WVD distribution,and SPWVD distribution;Secondly,in order to reduce the affect of noise and retain the time-frequency features as far as possible,the time-frequency images are preprocessed;then,Res Net-50 is selected as the network framework.In an attempt to preserve the data feature information to the best of one’s ability,the volume is increased in the residual module.At the same time,for the purpose of speed up the network convergence,the loss function adopts the cross-entropy loss function and the center loss function;It can be obtained from the experimental results that the three time-frequency analysis methods can obviously retain the time-frequency features of the SPWVD time-frequency image under the low SNR,the recognition effectiveness is the most effective,which verifies that the algorithm has excellent performance and good generalization ability under low signal-to-noise ratio.(4)The radar signal modulation type identification system is designed,which realizes the functions of signal preprocessing,identification and storage of results.The trained model was deployed in Py Qt5,and the recognition system was developed and implemented.What’s more,the modulation type recognition based on IQ signal and time-frequency image was completed.The system was verified to have good stability and recognition performance.