Embedded Model Control: design separation under uncertainty

The key signal of the Embedded Model Control (EMC) is the model error, measured minus model output, which is used by Internal Model Control and in a more efficient way by Active Disturbance Rejection Control (ADRC). EMC extends ADRC along four lines explained in the paper: a more generic disturbance dynamics, design of the noise layout (where uncertainty enters the model), dynamic uncertainty estimation, a design degree of freedom similar to Youla parameterization. The latter freedom is shown to favor separation between control and estimation under uncertainty, a principle hidden in ADRC. As a result, control law can be practically designed model-based, only limited by command saturation. The main theorems are developed in some detail starting from two model classes (design and embedded model), that are essential to define uncertainty. The paper extends part of previous results from SISO to MIMO systems, but some theorems and the free feedback dynamics are original. Theory is accompanied by a study case.