Simultaneous Design of Communication Strategies and Control Policies in Stochastic Systems

摘要

This document considers the problem of simultaneously designing communication strategies and control policies in decentralized stochastic systems. Such problems are difficult to solve, mainly because of the nonclassical nature of their information structure. We have identified classes of such problems with linear dynamics, quadratic loss functionals and Gaussian statistics for which the optimality of linear strategies can be established. The general approach used consists of first finding a lower bound on the cost, and then constructing joint strategies that attain this lower bound. For some instances of the cases where linear strategies fail to provide globally optimal solutions, explicit nonlinear strategies are obtained that demonstrate the inferiority of linear designs. The problems studied in this thesis can be viewed as important prototype problems, which are essential building blocks for a general theory of multistage distributed decision making under nonclassical information, and possibly partial statistical description. Keywords: Stochastic control, Information theory. (kr)