New methods for bounding adaptive filter poles

A method of bounding the coefficients of autoregressive moving average (ARMA) filters which guarantees strict sense stability (poles remain inside the unit circle) is presented. The method is general, and can be used to stabilize any filter which has a proper rational transfer function. Applications of Kharitonov´s theorem or the more general D-stability theory, with additional constraints, yield optimal coefficient bounds which can be computed offline. In contrast to previous bounding approaches, these methods apply to filters of arbitrary order. Poles can also be restricted to lie within any specified simply-connected regions of the z-plane. Significant improvements in convergence behavior and stability are noted