Lopes P., Piedade M.

One of the major issues in broadband feedforward active noise control is secondary path inversion. The Multi Input Multi Output Inversion Theorem (MINT) states that it is possible to invert non-minimum phase transfer functions, as long as there are more input channels than output channels, and that there are no common zeros in the paths. In active noise control this means that apart from true delays in the system, non minimum phase secondary paths can be inverted as long as there are more anti-noise sources than error sensors. However, this violates a common rule of thumb that is used to guarantee a reasonably large quite zone: that the number of error sensors should be twice the number of anti-noise sources. In previous papers a new technique that limits the control bandwidth was proposed to prevent the reduction of the quiet zone size for systems with more anti-noise sources than error sensors. In this paper the new technique is analyzed for sensitivity to secondary path modeling errors. For no secondary path modeling errors the effect of the secondary path is actually removed and the algorithm has exact the same convergence properties than the underlying algorithm used to update the control filter, namely the LMS, RLS, LSL or other. It is shown that the new structure is less sensitive to secondary path modeling errors than the FX-LMS, or the MFX-LMS algorithm. In fact it is shown that the allowed secondary modeling errors increase as the control bandwidth decreases. The results are obtained using narrow band analysis which results in a big simplification but still maintains enough information about the dynamics of the system to give useful results. Finally the results are confirmed throw computer simulations.