Research On Radar Anti-Jamming Method Based On Deep Learning

Intermittent sampling repeater jamming can accumulate energy after pulse compression,so feature classification method is needed to suppress the jamming.Deep learning is a classification method that can automatically extract features.Deep learning has been widely used in image,speech,natural language and robotics.However,there are still few researches on how to use these technologies in radar detection.Therefore,this thesis uses deep learning to study the single-pulse anti-jamming point target recognition problem under the phase code radar system.The main work includes the following contents:1.Deep learning and its application in radar detection are introduced.Firstly,the concept of deep learning is briefly introduced,and the convolutional neural network,an important part of deep learning,is emphatically introduced.Then,the model structure of convolution neural network is introduced,and the performance advantages and application scope brought by its unique connection mode are analyzed.Some convolution structures and network structures related to this paper are introduced,including one-dimensional convolution,1 × 1 convolution,dilated convolution,spatial separable convolution,concatenate structure and residual units.Finally,the feasibility of the application of deep learning in radar signal processing is analyzed and its main methods are introduced,which will prepare for the radar anti-jamming target detection based on convolution network later in this paper.2.Aiming at the anti-jamming problem of intermittent jamming background,a target detection method based on convolution network is proposed.Firstly,the different working modes of jammer are introduced.In particular,the time-sharing interference of intermittent sampling convolution modulation is modeled.By referring to the language processing system Wave Net,and using the structure of dilated convolution,residual network,skip connection and other structures to build anti-interference detection network,the complex received signal is converted into real number,and the end-to-end deep model from receiving signal to anti-interference target detection is constructed.The simulation data and the stochastic gradient descent algorithm are used to minimize the cross entropy loss between the output detection result and the real target probability,optimize the network parameters,and finally complete the anti-interference target detection.The model has a good anti-interference detection effect for a given transmit waveform and a given form of interference.3.Aiming at the problem of anti-jamming under arbitrary radar transmitting waveform,a target detection method based on separate concatenate network is proposed.For any phase code transmission waveform,compared with the fixed transmission waveform,the shortcomings and drawbacks of the previous detection model are analyzed,and the processing ability of the network for the received signal are studied.By sliding window matching of the received signal and the transmitted waveform,the sliding window matching feature is extracted,and the two-dimensional convolutional network is separated into two one-dimensional convolutional networks,and then the characteristics obtained by the two networks are concatenated.Finally,a separate concatenate network with residual connections is constructed to perform anti-interference target detection under arbitrary transmit waveform and given form interference.Compared with the convolutional network,fully connected network on the independent distance unit,and other structured networks,separate concatenate network has achieved the best results.