A new scheme for parallel analog and digital implementations of model reference adaptive control

The authors present a novel architecture for model reference adaptive control that is suitable for highly parallel analog and digital hardware implementation. The structure presented here is cleanly scalable to any size model reference adaptive system, and has good numerical properties. Results from real-time experiments performed on a 2*2 multivariable plant are presented. The experiments showed good performance in the presence of nonlinearities, unmodeled dynamics, and process noise. At convergence, the controllers that were obtained were at least as good as, and occasionally better than, conventional fixed controllers such as those obtained from linear quadratic Gaussian optimization.<
>