Research On Deep Learning Anti-radar Main Lobe Repeater Jamming Method

Radar electronic countermeasures are an indispensable part of modern military warfare.The enemy creates a complex and ever-changing electromagnetic jamming environment,which makes our radar unable to detect and track targets normally.At present,with the continuous development of DRFM(Digital Radio Frequency Memory)technology,radar main lobe repeater jamming is highly correlated with radar transmitted signals and enters the radar receiver from the antenna main lobe,making most traditional radar anti-jamming strategies ineffective,and the radar target detection performance has dramatically decreased.Therefore,research on suppression methods for radar main lobe repeater jamming is an important research direction to ensure radar detection ability,and has important research significance.Aiming at these problems,this thesis has carried out research on radar main lobe repeater jamming analysis and intelligent suppression of main lobe repeater jamming,and verified the research methods through experiments.The specific research content is as follows:(1)This thesis analyzes the jamming strategies and characteristics of various radar main lobe repeater jamming and establishes a rich jamming generation model.This thesis also simulates the echo data of various repeater jammings when the radar transmits a linear frequency modulation signal,and studies their characteristic distributions in the time domain,frequency domain,and time-frequency domain,laying a foundation for intelligent suppression of the main lobe repeater jammings and accurate detection of radar targets.(2)This thesis proposes a multi-stages multi-domains joint jamming suppression network to gradually recover the true echo of the target from both time-frequency-domain and time-domain aspects in view of the jamming environment of a single radar target.In the first stage,we use Transformer to extract features in the time-frequency domain of the signal,suppress jamming signals,and combine the original phase information to preliminarily reconstruct the signal.In the second stage,the U-Net network with complex1D-CNN is used to locally repair the signal details in the time domain of the output signal in the first stage.(3)In view of the complex multi-target and multi-jammer environment,this thesis designs a new signal feature transformation domain by using the non-negative matrix decomposition(NMF)method and combining the prior information of the transmitted signal,and proposes a multi-target and multi-jammer jamming suppression network based on the NMF decomposition of the time-domain signal.In order to ensure the complete recovery of multi-target signals,we also introduce the reversible residual structure into the suppression network to achieve the approximate lossless transmission of signal information.(4)Further considering the possible new type of jamming in the actual environment,the Prompt learning module is used to realize the fast learning of the new type of jamming without changing the parameters of the original jamming suppression network,thus preserving the effective suppression of the old type of jamming by the network.In addition,in order to ensure the optimal performance of the model on all data sets,we add the knowledge distillation method to the network update,use the new type of jamming to slowly update our jamming suppression module,and fine-tune the model parameters.The effectiveness of the above methods has been verified by simulation and actual radar data.The results show that the suppression method against radar main lobe repeater jamming based on deep learning proposed in this thesis can effectively suppress the jamming while preserving the complete information of the target,which greatly facilitates the subsequent radar target detection.