Robust performance optimization of open loop type problems using models from standard identification

Abstract

This paper discusses the problem of finding the controller that optimizes the expected H 2-norm for an uncertain system. The paper gives a unification of many similar results. The approach is motivated by the so called stochastic embedding approach. A closed form solution is given using a minimum of calculations for a class of problems including interesting signal processing applications such as feedforward design, channel equalization, noise cancellation and signal filtering. The method uses covariance information on model uncertainty and can therefore be used together with standard identification methods. By using the probability distribution of model error we avoid the conservativeness related to designing for worst cases. We then obtain robust designs with soft bounds. It is shown how the optimal controller can be found by rewriting the problem as a standard H 2-problem for an extended system. The solution can hence be obtained using standard methods and software. The paper uses restrictions on where uncertain parameters enter into the system. Such restrictions are inevitable if soft bounds on parameters are used. The method has direct applications in adaptive signal processing and adaptive feedforward control.