A REGRESSION APPROACH FOR SIMULATING FEEDFORWARD ACTIVE NOISE-CONTROL

Regression analysis is used to examine feedforward active noise control from a statistical point of view. Since numerical techniques for simulating feedforward active noise control in the frequency domain are mathematically similar to linear least-squares regression, two regression-based numerical methods can be applied to control problems. The first uses regression diagnostics such as the F-test, the t-test, and confidence intervals to model the effects of error sensor measurement noise. The second uses collinearity diagnostics to address a form of numerical ill conditioning that can corrupt the results. The regression diagnostics allow realistic modeling of random measurement error; the collinearity diagnostics help avoid numerical difficulties that might otherwise go undetected. Numerical results are given for a structural-acoustic control problem involving a fluid-loaded cylindrical shell. © 1995, Acoustical Society of America. All rights reserved.