ASA 126th Meeting Denver 1993 October 4-8

3aSA13. A regression-based approach for simulating feedforward active noise control, with regression diagnostics for assessing the impact of measurement noise.

C. E. Ruckman

Acoust. Technol. Dept., David Taylor Model Basin, Bethesda, MD 20084-5000

Chris R. Fuller

Virginia Polytech. Inst. and State Univ., Blacksburg, VA 24061

Current numerical techniques for simulating feedforward active noise control in the frequency domain are mathematically equivalent to the techniques required for linear least-squares regression. Measurement error in the control system error sensors plays a role analogous to that of observation error in a statistical regression. In the statistics literature, regressions are always accompanied by regression diagnostics, i.e., computed statistics that help assess the impact of observation error. However, discussions in the acoustics literature typically fail to address this issue. The present workmodels feedforward active control as a regression of complex-valued variables, and shows how to apply two basic regression diagnostics: the F-test and the t-test. Also discussed are the physical significant of the ``cost function'' being minimized by the regression, and the assumptions that must be made regarding the variance of the measurement error. The techniques are demonstrated by numerically simulating a simple system in which radiation from an axisymmetric cylindrical shell is controlled by oscillating forces applied on the shell surface. [Work supported by David Taylor Model Basin and ONR.]