Radar Signal Sorting Method Based On Neural Network

As an important part of modern electronic countermeasure technology,radar reconnaissance is the basis of electronic interference and defense.Radar signal sorting is the key step of radar reconnaissance,which is also the premise and guarantee for information processing analysis of the reconnaissance system.With the advancement of technology,the electromagnetic environment in which the reconnaissance system located is becoming more and more complex.Therefore,how to effectively and efficiently perform radar signal sorting in interference and noise has become a concern.In this thesis,the traditional signal sorting method and deep learning algorithm are studied.The methods of image and speech processing in deep learning are used to reduce the number of parameters that need to be set in signal sorting and improve signal sorting accuracy,which makes the signal sorting process more automated and intelligent.In this thesis,the Faster RCNN network is used for target detection,and the receiver pulse stream is generated by simulation to train the network,the pulse width(PW)and carrier frequency(RF)parameters of the radiation source contained in the received signal is estimated.In this way,the presorting of the radar singnal is achieved.In addition,the long-short-term memory(LSTM)network is used to predict the time-of-arrival(TOA)of the radiation source,and the pulse of the radiation source in the received signal is separated according to the prediction result.The main work and research results are as follows: 1.The traditional radar signal sorting algorithm is studied firstly,and the histogram method of signal main-sorting is analyzed.The situation of deinterleaving two signal is discussed.The Cumulative Difference Histogram(CDIF)algorithm and Sequential Difference Histogram(SDIF)are compared and analyzed.The SDIF algorithm is less computationally intensive than the CDIF algorithm,but the threshold parameter setting is more complicated.2.A pre-sorting algorithm based on the Faster RCNN network is proposed,considering that the traditional signal sorting algorithm need to preset the number of clusters,clustering results are susceptible to interference signals.The method predicts the PW and RF parameters of the radiation source contained in the received pulse stream.Firstly,a certain number of received pulse streams are generated based on the existing radar radiation source information.Then the pulse stream is mapped to a picture which length and width are both 256 pixels according to the PW and RF two-dimensional distribution of the received pulse stream.The length and width of the picture represent the PW and RF parameters respectively.The label of the picture is set according to the radiation source number and the picture area corresponding to the PW and RF parameters contained in the received pulse stream.Finally,The PW and RF parameters range of the radiation source contained in sample set receiving pulse stream is obtained by the Faster RCNN target detection method.and the received pulse stream is divided according to the range of PW and RF parameters.the pre-sorting of the radar singnal is achieved.The algorithm has achieved good results in matching the known radar pulse from the radar source library and sorting the unknown radar,and achieves the purpose of blind source sorting.3.A pulse sequence retrieval algorithm based on LSTM network is proposed,considering that the traditional direct sequence search method is simple,the probability of mis-selection and missed selection is high,and it is difficult to retrieve complex system radar pulses.The LSTM network is used to model the pulse law of the radar radiation source.By extracting the PRI information from the time series of the radiation source arrival time sequence that has been successfully sorted and use the information to train the network,the PRI sequence of the radiation source is predicted.Retrieval pulse sequence according to the TOA information which is calculated by PRI information.Experiments show that compare to the traditional direct sequence search algorithm,the pulse sequence retrieval algorithm based on LSTM network can achieve higher accuracy,which can effectively suppress the misselection of interference pulses and the missed selection of target pulses.The method of predicting TOA through the LSTM network can also effectively sort the staggered PRI signal and the jittered PRI signal.