Abstract
We study control problems for linear systems in the behavioral framework. Our focus is a class of regular controllers that are equivalent to the canonical controller. The canonical controller is a particular controller that is guaranteed to be a solution whenever a solution exists. However, it has been shown that in most cases, the canonical controller is not regular. The main result of the paper is a parametrization of all regular controllers that are equivalent to the canonical controller. The parametrization is then used to solve two control problems. The first problem is related to designing a regular controller that uses as few control channels as possible. The second problem is to design a regular controller that satisfies a predefined input-output partitioning constraint. In both problems, based on the parametrization, we present algorithms that does the controller design.
Introduction
In the behavioral approach to control theory, the focus is on the "behavior"—the set of all possible trajectories a system can take—rather than predefined input-output labels. Within this framework, the canonical controller serves as a theoretical gold standard: if a control problem is solvable, the canonical controller is guaranteed to provide a solution. However, a significant practical hurdle exists: the canonical controller is often "irregular."
An irregular interconnection can over-constrain a system, leading to physically unrealistic scenarios where the controller "fights" the plant's natural autonomous dynamics. This paper bridges the gap between theory and practice by providing a complete mathematical parametrization of all regular controllers that are equivalent to the canonical one, ensuring that the resulting systems are both effective and physically realizable.
The Behavioral Framework
The behavioral framework treats the plant and controller as interconnected systems where the goal is to restrict the plant's behavior to a desired subset. Unlike traditional control, which pre-assigns variables as inputs or outputs, this approach allows for a more flexible description of system dynamics.
As shown in Figure 1, the controlled system consists of a plant and a controller interacting through control variables to influence to-be-controlled variables. Regularity is a crucial property in this context; it ensures that the interaction is well-posed and does not impose redundant constraints on the system's autonomous behavior. The authors demonstrate that while regularity may appear to be a property of the equations chosen, it is actually a fundamental property of the system's outputs.
Main Result: Parametrization of Regular Controllers
The authors utilize the language of polynomial matrices, specifically kernel representations, to describe system behaviors. Their core contribution is a parametrization theorem (Theorem 11). They show that if one regular controller ($C$) equivalent to the canonical one is known, any other equivalent regular controller ($C'$) can be expressed as:
$$C' = V P_c + C$$
where $P_c$ represents the manifest behavior of the plant and $V$ is a polynomial matrix. This formula defines a complete "search space," allowing engineers to navigate the set of all valid regular controllers to find one that meets specific secondary criteria.
Algorithmic Applications
Using this parametrization, the authors develop two specific algorithmic solutions for practical controller design:
- Minimal Interaction: Finding the controller with the fewest active channels is a combinatorial problem. The authors map this onto a tree-search problem solvable via a depth-first search algorithm. This "Nullify" algorithm successfully reduces the number of physical communication links needed between the plant and the controller. The paper establishes an upper bound on how many control variables can be made irrelevant: at most $c - p(C)$, where $c$ is the number of variables and $p(C)$ is the number of outputs.
- I/O Partitioning: In many real-world systems, variables are naturally constrained as inputs or outputs (e.g., a sensor's reading is an input to the controller, not something the controller should attempt to force). The authors provide a rank-based algebraic check (Lemma 24) to ensure the controller respects these physical sensor/actuator boundaries.
Conclusion and Future Work
This work provides a mathematical bridge between the theoretical "least restrictive" canonical controller and the practical requirement of regularity. By baking physical constraints like I/O partitioning and minimal interaction into the behavioral framework, the authors offer a robust toolkit for modern system design.
However, limitations remain. The search for minimal interaction is computationally expensive for systems with a very high number of control variables. Furthermore, while the paper optimizes for the number of channels, it does not yet optimize for the "MacMillan degree" (the dynamic complexity) of the controller. Future research may use this parametrization as a starting point for finding the simplest possible regular controllers in terms of dynamic order.
Summary updated