Research On Recognition And Sorting Technology Of Passive Radar

Recognition and sorting technology of passive radar is one of the key technologies in the field of electronic reconnaissance,which provides important information for receiver to take appropriate countermeasures.However,with the rapid development of electronic information technology,the modulation of radar signal parameters is being more complex and variable,while the signal density in space is being extremely larger,and the phenomenon of parameter overlapping is serious.It brings great challenges to recognition and sorting technology of passive radar.This paper focuses on the of intra-pulse modulation identification and signal sorting of radar,mainly solving the problem of small sample identification and end-to-end signal sorting.Firstly,a small-sample intra-pulse modulation identification method based on CNN and RN network is proposed,in order to solve the problem that the radar actually cannot get enough sample data which is large and well-balanced to train a deep learning network.The method first uses a time-frequency transformation to process the signal into a time-frequency image and use the CNN network to extract its deep features.Then get the spliced feature vector pairs of the signal to be measured and the known signals in several different types.Finally,use RN network to determine the category of the signal to be measured.Research shows that this method can identify 9 different modulation types including LFM,NLFM,etc.This method shows high accuracy under the condition of small samples.Secondly,A radar signal sorting on DSP TMS320C6678 is implemented,using pulse searching and inter-pulse modulation type identification while using the improved K-means method for pre-sorting,SDIF for main sorting.At the same time,a method is proposed in which to make another inter-group discrimination on the clustering results of conventional signals,to solve the problem that traditional sorting methods always misidentify frequency diversity radar and pulse group frequency agile radar as conventional radar signals.The frequency diversity radars can be sorted by counting the carrier frequency characteristics of the simultaneously arriving signals,while the pulse group frequency agile radars can be sorted by counting the carrier frequency characteristics of the non-simultaneously arriving signals.Finally,the hardware implementation of sorting 6 kinds of radar signals including repetitive frequency staggered radar is completed,and the average sorting time is 137 ms.Finally,a new end-to-end radar signal sorting method based on deep learning image segmentation technology is proposed in this paper,aiming to solve the problem that the traditional radar signal sorting methods must rely on the set parameters or given priori information,while their structures are restricted and not flexible.The steps of this method include PDW collecting,PDW sequence image generating,U-Net deep segmentation,and PDW sequence searching.The collected multi-dimensional PDW data stream is processed into a pixel matrix,and a trained U-Net network model is used for pixel classifying.PDW searching completes the classifying of pulses.Research shows that this method can sort PDWs with variable inter-pulse modulation types and serious overlap in time-frequency domains,and performs well in case of even 20% pulse loss.It proves that this method is effective.In addition,the overall end-to-end process is easy to operate,flexible in processing,and rely less on the set parameters or given experience thresholds.