Output Regulation of Linear Stochastic Systems

TL;DR

This paper develops methods for controlling linear stochastic systems to achieve output regulation, introducing approximate solutions based on sampling and estimation that approach ideal regulation as sampling improves.

Contribution

It formulates and solves practical approximate output regulation strategies for stochastic systems using hybrid observers and estimators, bridging the gap to ideal solutions.

Findings

Proposed a hybrid regulator scheme employing Brownian motion estimation.

Demonstrated the approach on a circuit affected by electromagnetic noise.

Showed the approximate solution converges to the ideal as sampling frequency increases.

Abstract

We address the output regulation problem for a general class of linear stochastic systems. Specifically, we formulate and solve the ideal full-information and output-feedback problems, obtaining perfect, but non-causal, asymptotic regulation. A characterisation of the problem solvability is deduced. We point out that the ideal problems cannot be solved in practice because they unrealistically require that the Brownian motion affecting the system is available for feedback. Drawing from the ideal solution, we formulate and solve approximate versions of the full-information and output-feedback problems, which do not yield perfect asymptotic tracking but can be solved in a realistic scenario. These solutions rely on two key ideas: first we introduce a discrete-time a-posteriori estimator of the variations of the Brownian motion obtained causally by sampling the system state or output; second we introduce a hybrid state observer and a hybrid regulator scheme which employ the estimated Brownian variations. The approximate solution tends to the ideal as the sampling period tends to zero. The proposed theory is validated by the regulation of a circuit subject to electromagnetic noise.