| Radar radiator signal sorting is the primary problem that needs to be solved in the field of electronic countermeasures today.The effect of the sorting will directly affect the follow-up radar radiator signal work pattern recognition and subsequent battlefield threat level assessment.This paper has carried out a systematic study on the signal sorting of radar emitters,mainly involving the evaluation of the characteristics of radar emitter signals,the sorting of radar emitter signals in a high pulse density environment,and the identification of unknown radar emitters.The main research results obtained are:Aiming at the problem of imperfect evaluation system for the characteristics of radar emitter signals,the existing evaluation system has been improved from four perspectives,which can more comprehensively evaluate signal characteristics.Aiming at the problem of the single evaluation standard for the characteristics of the radar emitter signal,a subjective and objective decision fusion algorithm is proposed.This method uses the prior knowledge of experts to determine the index weight interval,then uses an improved projection pursuit algorithm to establish an evaluation model,and uses a non-monotonic projection spectral gradient algorithm to achieve subjective and objective decision fusion.Aiming at the problem of high pulse density faced by radar source signal sorting at this stage,a multi-parameter DBSCAN pre-sorting method is proposed.This method does not need to input the number of clustering categories in advance,and can remove interference pulses according to the characteristics of the data.Aiming at the high time complexity of the radar source signal sorting algorithm,a single-parameter PRI-based radar source signal main sorting method is proposed.This method draws on the idea of the traditional SDIF algorithm and proposes improvements to it from three aspects,which greatly improves the main sorting speed and more accurately suppresses the generation of sub-harmonics.Aiming at the problem of "increasing batches" in the sorting process of radar emitters,a multi-parameter GRU network batching method combined with deep neural networks is proposed.This method uses segmented random sampling to sample PDW sequences of different lengths and input them into the GRU network for batch judgment.Simulation experiments show that the sorting algorithm can effectively solve the problem of "increasing batches" that traditional sorting algorithms are easy to produce.Aiming at the problem of slow recognition of unknown radar emitters and low accuracy,a distributed individual identification method of unknown radar emitters combined with deep learning is proposed.This method intelligently characterizes individual radar emitters,uses a deep learning network to extract deep features,and combines support vector data description methods to identify unknown radar emitters,and through decision fusion,the recognition results of multiple sub-recognizers are fused to obtain the final recognition result.The simulation results show that this method has a greater improvement in recognition accuracy than a single recognizer. |
