Machine Learning-based Signal Sorting And Recognition Algorithm Of Unknown Radiation Source

Under the background of increasingly complex battlefield electromagnetic environment,the laboratory simulation electromagnetic environment platform can not meet the needs of the existing signal processing algorithm for signal high density characteristics and high fidelity.For the sorting of unknown radar signal,the traditional algorithm depends on the first signal preprocessing sorting method,and the current preprocessing algorithm is too single,this leads to the problems of low efficiency of dense signal processing and lower accuracy of sorting of complex signals.In order to establish a realistic electromagnetic environment platform and study accurate radar signal processing algorithm,this paper mainly researches the following 3 aspects:1.Aiming at the problem that the current electromagnetic environment platform can not provide real signal source for electronic reconnaissance.By analyzing the complex electromagnetic environment faced by modern warfare.The electromagnetic environment platform which can provide approximate actual combat for cognitive reconnaissance is designed by using graphical interactive interface.The platform models the characteristics of the signal,using multithreads to generate data in parallel,To obtain electromagnetic environment data similar to the real battlefield.And visualize the data through the chart.So that the simulation platform can be more effectively used in the process of modern radar reconnaissance signal processing.The simulation results show that the data generated by the platform simulation has a good similarity compared with the real battlefield data,and can provide an effective data source for signal processing at the present stage.2.Aiming at the difficult problem that low precision of unknown radiation source signal sorting under the current complex electromagnetic environment,an unknown radiation source signal sorting method based on multi-domain combination is proposed,which analyzes the different working modes of radar and presents different characteristics under different working modes.The full pulse signal of the incoming radiation source is intercepted according to its overlapping degree,and the expert rules are extracted and used as the input,the classifier is designed by the machine learning method,and the optimal staggered parameters and methods are selected for the full pulse data,it can avoid the use of serious overlapping parameters for sorting,the separation of radar radiation source signal under the serious condition is realized.The simulation results show that the method can make a clear judgment on the unknown radiation source under the condition of serious noise and mixing,and through continuous testing and use,the model sorting ability is more and more powerful,and it can realize the staggered sorting under different conditions.3.Aiming at the problem the signal repetition frequency modulation type is complex and can not be automatically identified under the complex electromagnetic environment,and the complex electromagnetic environment and the potential law of radar radiation source signal are analyzed.This paper presents a deep belief network which combines unsupervised and supervised.The simulation results show that the accuracy of the recognition method has been greatly improved within the permitting time,and it can adapt the loss pulse and the false pulse,and can effectively identify the repetition frequency modulation type of radiation source signal in complex environment.