Marcus J. Bronzel
Christopher R. Fuller
Vib. and Acoust. Lab., Virginia Tech., Blacksburg, VA 24061-0238
The design of an optimal feedforward controller in active noise control applications for controlling the sound radiation from a vibrating structure can be seen as a system identification problem. Most control implementations use a stochastic gradient search algorithm to adaptively identify the impulse response of the optimal control filter. The convergence properties of the widely used filtered-x LMS algorithm depend on the signal properties of the filtered reference signals. Sensing the reference signals from the vibrating structure is inevitable in most real applications and will effectively result in structural filtering of the unknown noise source. The filtered reference signals from a structural sensor generally exhibit a large eigenvalue spread resulting in poor convergence properties of the adaptive controller. A new extended-filtered-x approach has been developed which makes the implementation of fast RLS type algorithms possible. These algorithms are based on an orthogonal projection of the predictor error into the subspace of the filtered reference signals and therefore converge independent from the underlying signal statistics. Experiments for active control of noise from a vibrating plate have been conducted which demonstrate the superior control performance of fast transversal filters.