ADAPTIVE SIGNAL PROCESSING BASED ON HIGH-ORDER STATISTICS

ADAPTIVE SIGNAL PROCESSING HAS FOUND MANY IMPORTANT APPLICATIONS IN COMMUNICATIONS, SURVEILLANCE, SPEECH PROCESSING, AND SEISMIC PROCESSING. ALL OF THE CURRENT ADAPTIVE PROCESSING TECHNIQUES USE ONLY SECOND-ORDER STATISTICS, MAKING THE IMPLICIT ASSUMPTION THAT THE UNDERLYING PROCESSES ARE GAUSSIAN. THE RESEARCH PROGRAM PROPOSED HERE IS AIMED AT THE DEVELOPMENT OF A NEW FAMILY OF ADAPTIVE SIGNAL PROCESSING TECHNIQUES BASED ON HIGHER-ORDER STATISTICS. THESE TECHNIQUES ARE BASED ON SOME RECENT RESULTS ON THE ESTIMATION OF ARMA PARAMETERS FROM HIGH-ORDER CUMULANTS. THE NEW ALGORITHMS ARE EXPECTED TO PROVIDE SIGNIFICANT PERFORMANCE IMPROVEMENTS FOR PROCESSING NON-GAUSSIAN SIGNALS. IN PARTICULAR, WE EXPECT THEM TO HAVE FASTER CONVERGENCE RATES THAN THEIR SECOND-ORDER COUNTERPARTS AND BE ABLE TO HANDLE NON-MINIMUM PHASE SYSTEMS. WE HAVE REASONTO BELIEVE THAT THESE ALGORITHMS CAN BE IMPLEMENTED IN LATTICE FORM WHICH IS IDEALLY SUITED FOR VLSI IMPLEMENTATION. THE RESULTS OF THIS RESEARCH CAN BE USED FOR A VARIETY OF APPLICATIONS SUCH AS ADAPTIVE CHANNEL EQUALIZATION, ADAPTIVE ECHO CANCELLATION AND ADAPTIVE ARRAY PROCESSING. IN PHASE II WE WILL APPLY THESE NEW ALGORITHMS TO A SPECIFIC COMMUNICATION APPLICATION WHICH CAN BE TURNED INTO A COMMERCIAL PRODUCT.