Nam H., Suh S., Yoon K., Ahn D.

Active Noise Control (ANC) with additional noise sources to cancel noise is effective at low frequency. Most adaptive control filters have used FIR structures based on filtered-x LMS algorithms. But IIR structures are more desirable for the active control of duct noise in order to remove the poles introduced by the acoustic feedback. If the FIR structure is used in these cases, the length of the FIR filter should be very large when the poles of that should be removed are near the unit circle. Even though the IIR structure is more efficient than the FIR structure, the FIR filters have been widely used for the control filter due to the complexity of the coefficient fitting algorithm of the IIR filter and the stability problems especially when the adaptive algorithm for adaptive filters is not yet converged. In this paper, a stabilizing procedure of RLMS(Recursive LMS) algorithms for adaptive multi-channel IIR filters is proposed. Recursive LMS algorithms usually diverge before the algorithm is not yet converged. So, in the beginning of the ANC system, the stability of the RLMS algorithms could be improved by pulling the poles of the IIR filter to the center of the unit circle, and returning the poles to their original positions after the filter converges. Experimental ducts for monopole and dipole ANC systems are modeled under the assumption that only plane waves propagate in the duct and also that the sources and the sensor are sufficiently far apart that near field effects can be ignored. It is shown by computer simulations and experiments in experimental ducts that the proposed algorithm is more stable even if the convergence parameters are larger than those of conventional RLMS algorithms.