ADAPTIVE NONLINEAR POLYNOMIAL NETWORKS FOR ROTORCRAFT CABIN NOISE REDUCTION

TO REDUCE ROTORCRAFT-CABIN NOISE AND VIBRATION VIA ACTIVE ANTI-SOUND TECHNIQUES THIS PROJECT INVESTIGATES ADAPTIVE, NONLINEAR, POLYNOMIAL NETWORK-PREDICTION AND NOISE-CANCELLATION ALGORITHMS. IN RECENT YEARS, THERE HAS BEEN A CONSIDERABLE AMOUNT OF WORK ON THE USE OF STATISTICA SIGNAL-PROCESSING TECHNIQUES TO PRODUCE ANTI-SOUND ACOUSTIC FIELDS FOR REDUCING VIBRATION AND NOISE IN ENCLOSED SPACES. THESE BASIC TECHNIQUES EMPLOY ADAPTIVE LINEAR PREDICTION ALGORITHMS TO THE NOISE FIELD, AS SENSED FROM AN ARRAY OF SPEAKERS, TO CANCEL THE PREDICTABLE PART OF THE NOISE IN THE ARRAY. THE ADAPTIVE LINEAR PREDICTION AND CONTROL METHODS USED IN THE REDUCTION OF CABIN NOISE FOR PROPELLER-DRIVEN AIRCRAFT, APPEARS TO BE QUITE SUCCESSFUL. ALTHOUGH THE NOISE FIELD PRODUCED BY A PROPELLER CAN BE WELL-MODELED AS A GAUSSIAN PROCESS (AND THUS LINEAR PREDICTION IS OPTIMAL), ROTOR NOISE TYPICALLY HAS SIGNIFICANT IMPULSIVE COMPONENTS AND CANNOT BE WELL-MODELED AS A GAUSSIAN PROCESS. THEREFORE, NONLINEAR PREDICTORS ARE NEEDED TO ACHIEVE THE DESIRED CABIN-NOISE REDUCTION. THE PHASE I WORK EMPHASIZES THE DEVELOPMENT OF POLYNOMIAL NETWORK PREDICTOR ARCHITECTURES TO MINIMIZE THE NOISE RESIDUALS FROM AN ARRAY OF SENSORS WITHIN THE CABIN. IN PHASE II, A MORE DETAILED MULTIPLE-ERROR, NOISE-ABATEMENT SYSTEM WILL BE DESIGNED AND EVALUATED USING ACTUAL ROTORCRAFT DATA. TO REDUCE ROTORCRAFT-CABIN NOISE AND VIBRATION VIA ACTIVE ANTI-SOUND TECHNIQUES THIS PROJECT INVESTIGATES ADAPTIVE, NONLINEAR, POLYNOMIAL NETWORK-PREDICTION AND NOISE-CANCELLATION ALGORITHMS. IN RECENT YEARS, THERE HAS BEEN A CONSIDERABLE AMOUNT OF WORK ON THE USE OF STATISTICA SIGNAL-PROCESSING TECHNIQUES TO PRODUCE ANTI-SOUND ACOUSTIC FIELDS FOR REDUCING VIBRATION AND NOISE IN ENCLOSED SPACES. THESE BASIC TECHNIQUES EMPLOY ADAPTIVE LINEAR PREDICTION ALGORITHMS TO THE NOISE FIELD, AS SENSED FROM AN ARRAY OF SPEAKERS, TO CANCEL THE PREDICTABLE PART OF THE NOISE IN THE ARRAY. THE ADAPTIVE LINEAR PREDICTION AND CONTROL METHODS USED IN THE REDUCTION OF CABIN NOISE FOR PROPELLER-DRIVEN AIRCRAFT, APPEARS TO BE QUITE SUCCESSFUL. ALTHOUGH THE NOISE FIELD PRODUCED BY A PROPELLER CAN BE WELL-MODELED AS A GAUSSIAN PROCESS (AND THUS LINEAR PREDICTION IS OPTIMAL), ROTOR NOISE TYPICALLY HAS SIGNIFICANT IMPULSIVE COMPONENTS AND CANNOT BE WELL-MODELED AS A GAUSSIAN PROCESS. THEREFORE, NONLINEAR PREDICTORS ARE NEEDED TO ACHIEVE THE DESIRED CABIN-NOISE REDUCTION. THE PHASE I WORK EMPHASIZES THE DEVELOPMENT OF POLYNOMIAL NETWORK PREDICTOR ARCHITECTURES TO MINIMIZE THE NOISE RESIDUALS FROM AN ARRAY OF SENSORS WITHIN THE CABIN. IN PHASE II, A MORE DETAILED MULTIPLE-ERROR, NOISE-ABATEMENT SYSTEM WILL BE DESIGNED AND EVALUATED USING ACTUAL ROTORCRAFT DATA.