A New CAD Method and Associated Architectures for Linear Controllers

A new CAD method and associated architectures are proposed for linear controllers. The design method and architecture are based on recent results which parametrize all controllers which stabilize a given plant. With this architecture, the design of controllers is a convex programming problem which can be solved numerically. Constraints on the closed-loop system such as asymptotic tracking, decoupling, limits on peak excursions of variables, step response settling time and overshoot, as well as frequency domain inequalities are readily incorporated in the design. The minimization objective is quite general, with LQG, H¿, and new l1 types as special cases. The constraints and objective are specified in a control specification language which is natural for the control engineer, referring directly to step responses, noise powers, transfer functions, and so on. This control specification language will be the input to a compiler which will translate the specifications into a standard convex program in R^L, which is then solved by some numerical convex program solver. A small but powerful subset of the language has been specified and its associated compiler implemented. The architecture proposed not only makes design of the controller simple but also its implementation. These controllers can be built right now from off the shelf components or integrated using standard VLSI cells.