TRANSIENT ACOUSTIC ANALYSIS UTILIZING QUADRATIC SIGNAL REPRESENTATIONS AND NEUTRAL NETWORKS

RECENT SOVIET ADVANCES IN ACOUSTIC QUIETING OF SUBMARINES IS PLACING INCREASED EMPHASIS ON THE EXPLOITATION OF PASSIVE INTERMITTENT AND TRANSIENT EMISSIONS. TRANSIENT SIGNALS CAN BE DESCRIBED EFFICIENTLY IN THE TIME--FREQUENCY PLANE USING THE WIGNER DISTRIBUTION OR IN THE CYCLIC SPECTRUM. BOTH OF THESE REPRESENTATIONS OFFER DISTINCT ADVANTAGES OVER SHORT-TIME FOURIER OR SPECTRAL ANALYSIS FOR TRANSIENT SIGNALS. NEURAL NETWORK TECHNOLOGY HAS EMERGED RECENTLY AS A PROMISING TOOL FOR BUILDING SIGNAL CLASSIFICATION SYSTEMS. NEURAL NETWORKS HAVE THE ABILITY TO GENERATE THE COMPLEX (AND OFTEN NONLINEAR) FEATURE TO SIGNAL CLASS MAPPINGS THAT ARE REQUIRED TO SOLVE ALL BUT THE SIMPLEST PROBLEMS. NEURAL NETWORKS ARE CAPABLE OF SELF ORGANIZATION TO CONSTRUCT DECISION FUNCTIONS, AND TO HAVE THE ABILITY TO LEARN ON-LINE. LNK PROPOSES TO USE QUADRATIC SIGNAL REPRESENTATIONS SUCH AS THE WIGNER DISTRIBUTION AND THE CYCLIC SPECTRUM TO PROVIDE INPUTS TO NEURAL NETWORKS THAT WILL CLASSIFY TRANSIENT SIGNALS TYPICAL OF THOSE ENCOUNTERED IN THE UNDERSEA ENVIRONMENT. THE OVERALL SYSTEM CONCEPTS AND TECHNIQUES WILL BE DEVELOPED AND TESTED IN PHASE I. A COMPLETE PROTOTYPE SYSTEM WILL BE DEVELOPED IN PHASE II.