SIGNAL DECOMPOSITION

A SIGNAL DECOMPOSITION APPROACH BASED ON COMPLEX EXPONENTIAL MODELING OF THE SIGNAL IN NOISE IS PROOSED HERE FOR USE WITH THE AUTOMATIC TARGET RECOGNITION (ATR) PROBLEM. SUCH AN APPROACH HAS BEEN USED VERY SUCCESSFULLY TO RECOGNIZE NONSTATIONARY SIGNALS FROM UNDERWATER ACOUSTIC SENSORS AND FROM MODULATION-ON-PULSE RADAR WAVEFORMS. THE TECHNIQUE DOES NOT REQUIRE AN A PRIORI KNOWLEDGE OF THE NUMBER OF COMPLEX EXPONENTIAL COMPONENTS OR OTHER MODEL ORDER INFORMATION. FAST COMPUTATIONAL ALGORITHMS EXIST TO EXPLOIT THE STRUCTURE OF THE EQUATIONS TAT ESTIMATE THE PARAMETERS OF THE SIGNAL DECOMPOSITION. SINGULAR-VALUE DECOMPOSITION (SVD) HAS BEEN USED TO ENHANCE THE SIGNAL DECOMPOSITION IN THE PRESENCD OF NOISE. THIS PROPOSAL WILL DEVELOP FURTHER ENHANCEMENTS, INCLUDING THE USE OF TOTAL LEAST SQUARES (TLS) TECHNIQUES, GENERALIZED SINGULAR-VALUE DECOMPOSITION (GSVD) FOR COLORED NOISE SITUATIONS, AND HIGHER-ORDER CORRELATION STATISTICS TO SUPPRESS GAUSSIAN NOISE CONTRIBUTIONS THAT CAN BIAS SIGNAL DECOMPOSITION ALGORITHMS.