Radar Emitter Signal Deinterleaving Based On Support Vector Clustering

As an important part of the electronic intelligence system and electronic supportmeasure systems, radar emitter signal delinterleaving is a process to detect andrecognize the pulse signals interleaved in time from different radar emitters. Andseparate the signals into some indinidual emitter groups,where pulses in one groupbelong to the same radar emitter.The conventional methods of radar signals deinterleaving mainly use the pulsedescriptor word (PDW)composed of pulse signal parameters, i.e., time of arrival(TOA),radio frequency (RF), direction of arrival (DOA), pulse amplitude (PA) and pulseduration (PD). These techniques work well only when the radar signals belong tocontinuous wave radar emitters and when the number of emitters are notlarge．However,as the density of electromagnetism signals environment gets larger andthe application of new complex waveform modulations in modern radar appearscontinuely, the conventional deinterleaving techniques may not to be good enough to geta more satisfactory result in such environment.The support vector clustering (SVC) algorithm is an unsupervised learning methodinspired by the support vector machine and determines automatically the number ofclusters by a unified framework. The SVC algorithm is a good method to solve theproblem of unknown radar emitter signals. The dissertation discusses severalconventional radar pulse signals deinterleaving methods, and a brief analysis of thecumulative difference histogram (CDIF), the sequential difference histogram (SDIF)and the pulse repetition interval (PRI) transformation is presented. Computersimulations have been given to demonstrate the effectiveness of the three algorithms.Traditional unsupervised clustering algorithms will be introduced. After that, thedissertation analyses the fuzzy C-means clustering algorithm especially, and thecorresponding classification results for data set be given and compared respectively. Anovel parameter optimizing FCM algorithm is proposed to find the classfication resultof radar emmiter signals.The dissertation introduces the SVC algorithm and a clustervalidity method for SVC.And the cluster validity indexs are discussed. Computersimulation is conducted on the Iris data set to testify the validity of the algorithm.