Research On Radar Signal Little Sample Recognition Method Based On Deep Learning

Radar modulation recognition technology plays a large part in electronic reconnaissance,electronic support and other traditional areas.Generally,there are two kinds of methods in traditional radar modulation recognition area.One is decision theory,the other is feature analysis.The former is based on the test statistics and compares the decision thresholds of various radar signals.Through numerical comparison to complete the radar modulation signal recognition,but this kind of method has too many parameters,and the expression is too complex.The latter requires complex prior knowledge to extract intra pulse features.In the past two years,due to the improvement of hardware computing power and data storage capacity,deep learning technology has become a hot spot again,and has made rapid development,and has more application scenarios.In the field of radar signal recognition,deep learning technology also gets more attention.However,deep learning technology is data-driven,which needs massive data to support its training,and it is very difficult to obtain radar signal data.It is difficult to build a complex and huge data set to represent the model.Therefore,the research of little sample radar signal recognition method based on deep learning technology becomes imminent.In this thesis,considering the frequency offset,phase offset and dynamic range of the receiver in the process of radar signal monitoring,the mixed method of MATLAB programming and instrument construction is used to generate the data set.And using this data set,the method to solve the problem of radar signal recognition under the condition of little sample is deeply studied.The following research is done in this thesis:(1)In order to solve the problem of lack of data,this thesis innovatively proposes the enhanced deep convolution generative adversarial networks(SDCGAN),which can generate radar modulated signal samples according to the existing real samples,so that the generated data of many kinds of radar modulated signals have a closer distribution with the real data.(2)In order to solve the problem of little sample radar signal recognition,a little sample deep learning radar signal recognition method based on SDCGAN network as generating network and convolutional neural network(CNN)as discriminant network is proposed innovatively.Under the condition of signal-to-noise ratio of 8dB,the recognition accuracy can reach more than 90%with only 40 samples.(3)The SDCGAN-CNN network is lightweight,and a lightweight network composed of CLDNN(convolutional,long short term memory,fully connected deep neural network)and with Attention mechanism,which can have high erefficiency,better robustness and requires less radar signal samples.