Adaptive IIR filtering algorithms for enhanced CMUT performance

The use of adaptive filtering as a means of signal processing in sensorapplications provides stability and accuracy when operating with sensors thathave slowly varying coefficients in their transfer function. This work wasconducted using a nonlinear state space model of a capacitive micromachinedultrasonic transducer (CMUT) based on FEM data to analyze the simulated effectsof adaptive infinite impulse response (IIR) filtering on a through transmissionCMUT system. A number of different IIR filter algorithms were investigated andthe convergence rates, final mean squared error (MSE) and filter stability amongother parameters were analyzed. Included in these algorithms were the fullgradient descent method, simplified gradient method, Feintuchs' method,recursive predictor error (RPE) method, orthogonal triangular (QR) decompositionand pseudo linear regression recursive least squares (PLR-RLS). The adaptiveIIR filters were applied for system identification, equalization and activenoise cancellation (ANC) operations for the study. Exponential convergentapproximation time coefficient, a measure of the adaptive filter's ability totrack changes, for the ANC case has been shown to vary by more than 20%. MSEvariations for the differing algorithms of greater than 10dB have been obtainedand filter stability was found to be dependant on a number of internal algorithmparameters, such as the numerator/denominator adaptation ratio, as well as thechoice of algorithm.