Emitter recognition based on modified X-means clustering

This paper presents a new algorithm to divide multidimensional data into clusters. It enhances the K-means clustering algorithm (Linde-Buzo-Grey) so that the number of clusters is determined at run time. The paper uses radar classification problem as an example application. Most of naturally existing processes possess Gaussian distribution because of central limit theorem. This paper assumes that parameters of radars are Gaussian. Chi-squared test for goodness of fit is used far evaluating the hypothesized distribution from sampled data. The data is divided and output of chi-squared test is used to decide whether to carry on sub-clustering or not. Test results on simulated data are shown to demonstrate the working of algorithm.