AN ITERATIVE REWEIGHTED LEAST-SQUARES ALGORITHM FOR THE DESIGN OF 2D IIR FILTERS (TueAmPO1)

Author(s) : Bogdan Dumitrescu (Tampere University of Technology, Finland) Riitta Niemistö (Tampere University of Technology, Finland)

Abstract : We present a new method for optimizing two-dimensional IIR filters with respect to a $p$-norm error criterion, especially for large $p$, when quasi-equiripple filters are obtained. We combine a Gauss-Newton convexification of the criterion with the iterative reweighted least-squares (IRLS) algorithm. Stability is obtained by using 1-D or 2-D convex stability domains based on a positive realness description. A final optimization of the numerator, keeping the denominator fixed, allows more freedom in choosing the value of a key parameter.

Menu