Implementation and prevention of unstable optimal compensators

Optimal control design methods frequently lead to unstable compensators, even for stable plants. These compensators present possible implementation difficulties because they are conditionally stable. If unstable compensation is not inherently required for performance, then a stable compensator is preferable. The reasons why unstable compensators arise within the linear quadratic Gaussian framework are discussed. Examples and experimental results from the Middeck Active Control Experiment (MACE) are used to illustrate the implementation difficulties associated with unstable compensators. Several approaches to designing stable compensators are compared.