Multi-radar Signal Recognition Based On Adaptive Blind Source Separation

Electronic reconnaissance is an important module of modern electronic warfare system.It can sense the battlefield situation and recognize the radar information,which provides a strong guarantee for the synthesis of countermeasures and the evaluation of effectiveness.The increasing complexity of electromagnetic environment and the emergence of new system radar bring severe challenges to electronic reconnaissance,especially radar signal recognition.Most algorithms focus on multi-signal separation or single signal recognition,which has poor compatibility when recognizing multi-radar signal.How to realize the separation,feature extraction and modulation recognition of multi-radar signal with little prior information has become an urgent technical problem in radar signal recognition field.In view of many technical bottlenecks in radar signal recognition in the field of electronic reconnaissance,this paper focuses on time-frequency analysis,feature extraction,signal separation and modulation recognition.The main work is as follows:Firstly,the mathematical model of radar signal in electromagnetic environment is constructed,and the time-frequency analysis method suitable for radar signal processing is discussed.The common time-frequency image preprocessing method and time-frequency image characteristic parameters are described.On this basis,aiming at the problem of incomplete information of traditional single characteristic parameter and unknown optimal parameters of Support Vector Machine,a radar signal recognition algorithm based on Support Vector Machine is proposed,which enhances the reliability of radar signal recognition technology.Then,the propagation algorithm,basic structure,mathematical operation and activation function of Convolutional Neural Network are analyzed,and the pre-training Goog Le Net and pre-training Res Net neural network models are analyzed.Aiming at the problem that the actual radar signal samples are few and the features extracted manually are not comprehensive,a radar signal recognition algorithm based on transfer learning is proposed,and then the Non-negative Matrix Factorization is applied to reduce the dimensions of the features and further improve the recognition speed of radar signals.Finally,the problem of Single Channel Underdetermined Blind Source Separation is discussed,focusing on the modal decomposition of Virtual Multi-channel.Aiming at the problem that the number of signals intercepted by the receiver at the same time is large,the priori information is little and the separation effect of each radar source signal is different,a multi-signal recognition algorithm based on Adaptive Variational Mode Decomposition is proposed.The Improved Particle Swarm Optimization algorithm,Support Vector Regression and Variational Mode Decomposition are combined organically to realize the adaptive separation of multi-radar signal.In addition,a combined classifier is proposed to recognize the modulation mode of multi-radar signal,which provides a new way to deal with the separation and recognition of multiple signals.