On-line Sorting Of Radar Emiter Signals

Radar emitter signals sorting is an important part of the electronic surveillancesystems and one of the key areas of electronic countermeasures. Now, with theincreasing number of radar applications, the density of electromagnetic signals isvery big now, and the radar receiver may receive a large number of signals to besorting within a short period of time, so the computation of the signal sorting have arapid increase and is difficult to meet the requirements of real‐time applications. Inaddition, A large number of new kind of radars also make the radar signalenvironment very complex, and the parameters of different signals may be very close,which make the accuracy of signal sorting decreased rapidly. The emergence of thesenew conditions proposes a severe challenge to the radar emitter signal sorting.To address these new challenges, this paper did the work as follows:First, based on the analysis to the current sorting methods, this paper proposedthe radar emitter signals online sorting method based on the online kernel clusteringalgorithm. This algorithm uses the kernel trick to map the data into thehigh‐dimensional linear space, and cluster the data in this space. There is no need toknow the number and centers of all classes in advance. This method has higheraccuracy and speed than other methods, and can handle data online. Simulationresults validate this method.Second, the parameter of penalty term is optimized. The parameter of penaltyterm measures the weight of the penalty term in the instantaneous risk function.After the analysis of the role of penalty term, this paper proposed that the parameterof penalty term should be set to decline according to certain rules. Simulation resultsvalidate this optimization strategy.Third, the step size, which is an important parameter in this method, is alsooptimized. This paper proposed two strategies, which make the step size decliningand adaptive, to optimize the step size. The first strategy set the step size to declineaccording to certain rules, and the second strategy adjust the step size adaptively.Simulation results validate these strategies.Finally, based on the optimization strategies mentioned before, this paperproposed a new method whose name was online kernel clustering with parametersadaptation. In this method, the parameter of penalty term was set to decline according to certain rules, at the same time the step size is adjusted by the stochasticmeta‐descent strategy. Then the simulation results were compared with that ofonline kernel clustering method. These results showed that the new method is betterin accuracy and speed than the old method.