BIOLOGICAL SIGNAL PROCESSING: BIOLOGICAL SIGNAL TRANSFORMS FOR NEURAL NETWORKS PERFORMING SONAR SIGNAL CLASSIFICATION...

TIME-FREQUENCY REPRESENTATIONS OF ACOUSTIC TRANSIENT EVENTS ARE OFTEN REQUIRED IN ORDER TO DISTINGUISH SUBTLE LOCALIZED TIME AND FREQUENCY FEATURES IN THE TRANSIENT SIGNAL THAT CAN BE MISSED BY CLASSICAL SHORT-TIME FOURIER TRANSFORM ANALYSIS. TWO TIMEFREQUENCY REPRESENTATIONS APPROPRIATE FOR THIS SIGNAL CLASS ARE THE WIGNER-VILLE TRANSFORM AND THE GABOR REPRESENTATION. THE WIGNER TENDS TO HAVE SOMEWHAT HIGHER RESOLUTION CAPABILITY THAN GABOR, BUT ITS QUADRATIC FUNCTION OF THE SIGNAL LEADS TO CROSS-TERMS IN THE TRANSFORM THAT ARE DIFFICULT TO CLASSIFY IN THE PRESENCE OF NOISE. THE GABOR HAS SOMEWHAT LESS RESOLUTION, BUT ITS LINEAR FUNCTION OF THE SIGNAL LEADS TO MORE UNDERSTANDABLE EFFECTS IN THE PRESENCE OF NOISE. THIS PROPOSAL SUGGESTS TWO APPROACHES TO UTILIZE THE GABOR BASIS FUNCTIONS FROM THE GABOR REPRESENTATION THAT CAN BE USED TO ENHANCE THE PERFORMANCE OF THE WIGNER TRANSFORM IN THE PRESENCE OF NOISE. THE PRIMARY BENEFIT WILL BE AN IMPROVED WIGNER TECHNIQUE FOR ANALYZING ACOUSTIC TRANSIENT SIGNALS IN NOISE FOR A KEY DOD PROGRAM TO EXPLOIT THESE TRANSIENTS AS A MEANS TO FIND QUIET SUBMARINES.