Direct Minimization Methods for Adaptive Control of Non-Minimum Phase Systems

Abstract Adaptive control of non-minimum phase systems is not straightforward. Direct pole-placement methods fail, since they involve cancellation of an unstable model. Indirect methods, based on explicit identification, have the problem that the model may occasionally be non-stabilizable. In this paper another approach is studied. The control objective is formulated as an optimization criterion. It thus contains minimum-variance and dead-beat control as special cases. The criterion is minimized on-line with respect to the regulator parameters. Auxiliary parameters, necessary to determine the gradients, are estimated using an Instrumental Variable Technique. The results are compared with other approaches in simulations.