Prior knowledge in model reference adaptive control of multiinput multioutput systems

This note shows how the prior knowledge necessary for model reference adaptive control of multiinput multioutput systems can be weakened. In particular, it is shown that the real variables appearing in the system interactor matrix can be estimated along with the other parameters appearing in the system model. This represents a significant reduction in the amount of prior knowledge required relative to previous work which assumed that the full interactor matrix was available. A theoretical convergence analysis of the resulting algorithm is presented together with simulation results.