MODULAR NEURAL EW SYSTEMS (MNEWS)

THE PROCESSING THROUGHPUT OF AN ESM SYSTEM IS DIRECTLY RELATED TO ITS ABILITY TO GROUP INCOMING PULSES INTO CLUSTERS, EACH OF WHICH CONTAIN A UNIQUE EMITTER. THE APPROACH TO THIS PROBLEM PRESENTED HERE PROVIDES A STRATEGY FOR PERFORMING UNIQUE SIGNAL CLUSTERING AND CLASSIFICATION BASED UPON THE USE OF NEURAL NETWORKS. PRE-SORT CLUSTERING WILL BE ACHIEVED THROUGH A SELF ORGANIZING CLUSTERED (SOC) NETWORK WHICH ADAPTIVELY DERIVES CLUSTER GEOMETRIES. THE SOC UTILIZES A MINIMUM DISTANCE CRITERION AND IT IS TOTALLY DATA ADAPTIVE IN THAT NO A PRIORI LIBRARY INFORMATION IS NEEDED TO SET UP AN INITIAL CLUSTER CONFIGURATION. CLASSIFICATION WILL BE ACCOMPLISHED THROUGH THE USE OF AN ADAPTIVE NETWORK SENSOR PROCESSOR (ANSP) WHICH CAN UTILIZE INCOMPLETE OR NOISY PULSE DESCRIPTOR WORDS, SUCH AS MIGHT OCCUR IN WARM SITUATIONS. THE TWO NETWORKS ARE CONFIGURED TO MINIMIZE THE AMOUNT OF PULSE DEINTERLEAVING REQUIRED TO OBSERVE AND IDENTIFY EACH EMITTER.