Two-stage Bayesian Radar Signal Identification with Multidimensional Mixture Distribution Model

Most of current radar systems can operate numerous parameters and a radar signal database records large radar signal items, typically for different coherent processing intervals (CPIs). In online applications however, the pulse repetition interval (PRI) parameter is often difficult to estimate when multiple radar signals are present together, making it difficult to build a maximum likelihood (ML) measure in one stage. In this paper, we study a two-stage radar signal identification method based on a radar signal database built with a multidimensional mixture distribution model of radar CPI parameters. At the first stage, raw signal association is performed without the PRI information by using multidimensional mixture distribution model, because different radar signals may be mixed together. At the second stage, the PRI informations of associated CPI items are used to build the final ML. The ML is then processed by bayesian theory to determine whether it agrees with an record of the radar signal database or not. Numerical results with a radar signal base with more than 20,000 items of 7 radar systems indicate that the accuracy of the method in this paper is improved by 4%, 12%, 23%, and 34% over the conventional algorithms of Naive Bayes Classifier (NBC), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Classification And Regression Tree (CART).