NONLINEAR DETECTION OF WEAK SIGNALS IN CLUTTER USING GENERALIZED NEURAL NETWORKS

THE MAJOR OBJECTIVE OF THE PROPOSED R&D EFFORT IS TO INVESTIGATE THE PERFORMANCE OF SEVERAL INNOVATIVE NEURAL NETWORK APPROACHES FOR THE DETECTION OF WEAK SIGNALS IN CLUTTER. DURING PHASE I, DIFFERENT MODELS FOR SIGNAL AND CLUTTER WILL BE USED TO GENERATE SIMULATED DATA AND TO COMPARE THE PERFORMANCE OF STANDARD METHOD WITH THAT OF GENERALIZED NEURAL NETWORKS (GNN). SIGNALS WILL BE MODELED BOTH AS NARROW BAND AND AS CHAOTIC TIME SERIES, WHEREAS CLUTTER WILL BE MODELED AS BOTH NARROW BAND AND WIDE BAND. IN ADDITION TO ONE DIMENSIONAL AMPLITUDE VERSUS DOPPLER FREQUENCY SIGNALS, TWO DIMENSIONAL SIGNALS INVOLVING RANGE AND DOPPLER FREQUENCY WILL ALSO BE TESTED. THE FOLLOWING GNN APPROACHES WILL BE CONSIDERED: (I) INTELAHEDRON BASED ON STATISTICAL PATTERN RECOGNITION (II) RADIAL BASIS FUNCTIONS (III) PROJECTION PURSUIT (IV) GROUP METHOD OF DATA HANDLING (V) STATE SPACE MARKOV MODELS. TWO OF THE ABOVE APPROACHES (INTELAHEDRON AND PROJECTION PURSUIT), WILL BE DEVELOPED AND COMPARED TO STANDARD METHODS USING MONTE CARLO SIMULATIONS ON THE ABOVE CASES. THE PROPOSED R&D EFFORT WILL BE SUPPORTED BY RAYTHEON COMPANY'S RADAR EQUIPMENT DIVISION. ANTICIPATED BENEFITS/POTENTIAL COMMERCIAL APPLICATIONS - THE SUCCESS OF PHASE I R&D WILL LEAD TO A MORE ROBUST DETECTION CAPABILITY UNDER COMPLEX CLUTTER AND ECM ENVIRONMENTS. THE BENEFITS WILL ACCRUE IN TERMS OF TRACKING MORE TARGETS AT LONGER RANGES USING LESS RADAR ENERGY. THE EVENTUAL GOAL OF THIS EFFORT IS THE DEVELOPMENT OF UNIVERSALLY APPLICABLE PLUG-IN, TRAINABLE MODULE OF TARGET DETECTION. COMMERCIAL APPLICATIONS WILL BE IN FIELDS SUCH AS AIR TRAFFIC CONTROL, COMMERCIAL SONAR AND FAILURE DETECTION IN COMPLEX MECHANICAL AND ELECTRICAL EQUIPMENT USING VIBRATION DATA.