Optimal design of information channels in networked control
Serdar Yüksel, Queen’s University, Canada
Abstract:
We first investigate topological properties of measurement channels viewed as stochastic kernels for finite horizon optimal control problems under partial observations. In this context, quantizers are viewed as a special class of information channels. Structural results for optimal causal quantization policies are discussed and, for systems driven by additive noise, existence of optimal policies is established. Application to jointly optimal quantization and control for the Linear Quadratic Gaussian problem is presented.