⚡ electrical engineering

Steady-state response assignment for a given disturbance and reference: Sylvester equation rather than regulator equations

This paper proposes a novel design framework for active steady-state control that extends moment matching to moment assignment by decomposing closed-loop moments and utilizing the Sylvester equation, thereby providing necessary and sufficient conditions for assigning desired responses without relying on traditional regulator equations.

Original authors: Hyeonyeong Jang, Jin Gyu Lee

Published 2026-04-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hyeonyeong Jang, Jin Gyu Lee

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are driving a car (the System) on a bumpy road. The bumps represent disturbances (like wind gusts or potholes), and your destination is a specific path you want to follow (the Reference).

Traditionally, engineers have tried to fix this car in two ways:

Model Reduction: They look at the car's complex engine and try to build a simpler, cheaper toy version that looks and feels the same when you drive it over a specific set of bumps. They call this "matching the moment." It's like making a scale model that bounces exactly like the real car on a specific test track.
Output Regulation: They build a very complex, rigid suspension system designed to force the car to ignore the bumps entirely and stay perfectly flat. This often requires solving incredibly difficult math puzzles (called "regulator equations") to make sure the car doesn't crash while doing it.

This paper proposes a third, smarter way: "Moment Assignment."

Instead of just copying the car's behavior or fighting the bumps, the authors suggest we treat the car's steady-state reaction like a dial or a knob that we can turn to any setting we want.

Here is the breakdown of their idea using simple analogies:

1. The "Moment" is the Car's "Personality"

In this paper, a "Moment" isn't a unit of time. Think of it as the car's default personality when it hits a specific type of bump.

If you hit a pothole, does the car bounce up 1 inch? 2 inches? Does it wobble left or right?
That specific reaction (the bounce height and direction) is the "Moment."
Usually, this personality is fixed by the car's design (the engine and suspension).

2. The Old Way vs. The New Way

Old Way (Moment Matching): "Let's build a fake car that acts exactly like the real one when it hits a pothole." (Good for simulation, bad for changing the outcome).
Old Way (Regulation): "Let's build a magical suspension that makes the car bounce 0 inches, no matter what." (Hard to design, requires solving a massive, scary math equation).
New Way (Moment Assignment): "Let's install a smart controller that says, 'When I hit a pothole, I want the car to bounce exactly 0.5 inches to the right.' And then we build a suspension that makes that happen."

The authors realized that the car's reaction isn't just one big mystery. It's actually two things added together:

The car's natural reaction to the bump.
The reaction caused by the new controller (the "compensator").

Because these two add up, you can calculate exactly what the controller needs to do to cancel out the bad parts and add the good parts you want.

3. The "Virtual Signal Generator" (The Magic Mirror)

This is the most creative part of the paper.

Imagine the car and the controller are two dancers.

The Car thinks the Controller is sending it a signal.
The Controller thinks the Car is sending it a signal.

The authors realized you can pretend the controller is a "Virtual Signal Generator." Instead of thinking about the complex wires and gears inside the controller, you just ask: "What kind of signal would the controller need to send to the car to get the result we want?"

Once you know that "Virtual Signal," you don't need to solve the scary "Regulator Equations" anymore. You just need to solve a much simpler math problem called the Sylvester Equation.

Analogy: It's like trying to tune a radio. The old way was trying to rebuild the radio's internal circuits to get a clear station. The new way is just turning the tuning dial (the Sylvester equation) until the static clears up and you hear the song you want.

4. The "Recipe" for Success

The paper gives a step-by-step recipe to build this new controller:

Define the Goal: Decide exactly how you want the car to react to the bumps (e.g., "Bounce 0.5 inches right"). This is your Desired Moment.
Check the Math: Ask, "Is it physically possible for this car to do that?" (This is checking if the "Moment Transfer Operator" can reach that setting).
Build the Controller:
- Part A (The Tuner): A part of the controller that forces the car to hit that exact "Desired Moment."
- Part B (The Stabilizer): A part that makes sure the car doesn't spin out of control while trying to hit that target.

5. Why This Matters

Simplicity: It replaces a nightmare of complex equations with a straightforward, linear calculation (Sylvester equation).
Flexibility: You aren't limited to "zero error" (perfect tracking). You can choose any steady-state behavior you want. Maybe you don't want to eliminate the vibration entirely; maybe you just want to dampen it to a comfortable level. This framework lets you design that.
Data-Driven: Because "Moments" can be measured from input/output data (like watching how the car actually bounces), you don't even need to know the exact engine specs to design the controller. You can learn it just by observing the car.

Summary

Think of this paper as a new GPS for Control Systems.
Instead of telling the system, "Go to the destination and ignore all obstacles," which is hard and rigid, the authors say: "Here is the exact path you should take, including how you should sway around the obstacles. Here is the math to build the car that does exactly that."

They turned the art of "fighting disturbances" into the science of "assigning a desired reaction," making it easier, more flexible, and more intuitive to design smart control systems.

1. Problem Formulation

The paper addresses the problem of steady-state response assignment for Multi-Input Multi-Output (MIMO) Linear Time-Invariant (LTI) systems subject to exogenous signals (disturbances and references) generated by a signal generator.

System Model: The plant is described by $\dot{x} = Ax + Bu + P\mu$ , $y = Cx + Du + Q\mu$ , where $\mu$ is the exogenous signal.
Signal Generator: The exogenous signal is generated by $\dot{\omega} = S\omega$ , $v = L\omega$ , which is injected into the system.
Objective: The goal is to design a dynamic compensator such that the closed-loop system achieves a predefined steady-state response (moment) $M_{des}$ for the given signal generator $(S, L)$ .
Context: Traditionally, "moment" has been used for model order reduction (matching moments to approximate a system). The authors propose shifting this concept from passive approximation to active control (assigning moments to dictate steady-state behavior). This unifies output regulation (where $M_{des}=0$ ) and closed-loop interpolation.

2. Methodology and Theoretical Framework

The core methodology relies on decomposing the closed-loop dynamics on an invariant manifold using Sylvester equations rather than the traditional regulator equations.

A. Moment Decomposition

The authors establish that the closed-loop moment ( $M_{cl}$ ) of an interconnected system can be decomposed into two parts:

The open-loop moment ( $M_{open}$ ) of the plant.
A term linearly parameterized by the moment of the compensator ( $M_c$ ).

Mathematically, this is expressed as:
$M_{cl} = M_{open} + T_S(M_c)$
Where $T_S$ is a Moment Transfer Operator defined by the plant dynamics $(A, B, C, D)$ and the signal generator matrix $S$ . This operator maps the compensator's moment to the resulting change in the closed-loop moment.

B. The Moment Transfer Operator ( $T_S$ )

The operator $T_S$ is defined as:
$T_S(M_{in}) := C \mathcal{L}_{(S,A)}^{-1}(B M_{in}) + D M_{in}$
where $\mathcal{L}_{(S,A)}$ is the Sylvester operator ( $\Pi \mapsto \Pi S - A\Pi$ ).

Key Insight: The existence of a compensator that assigns a desired moment $M_{des}$ is equivalent to the existence of a solution $M_c$ to the linear algebraic equation:
$T_S(M_c) = M_{des} - M_{open}$
Significance: This replaces the complex, non-linear regulator equations (typically involving $\Pi S = A\Pi + B\Gamma$ and $0 = C\Pi + D\Gamma + Q$ ) with a single linear equation involving the Sylvester equation.

C. Canonical Compensator Structure

The paper proposes a canonical structure for the dynamic compensator that parameterizes all possible controllers achieving the moment assignment. The compensator state $\xi$ is split into two parts:

$\xi_a$ (Moment Matching Part): A canonical structure that ensures the compensator generates the required moment $M_c$ when driven by the virtual signal generator $(S, M_{des})$ .
$\xi_b$ (Stabilization Part): Redundant states used solely to stabilize the closed-loop system without altering the assigned moment.

The structure is given by:
$\begin{aligned} \dot{\xi}_a &= (S - G_a M_{des})\xi_a + F_a \xi_b + G_a u_\xi \\ \dot{\xi}_b &= -G_b M_{des} \xi_a + F_b \xi_b + G_b u_\xi \\ y_\xi &= M_c \xi_a + H_b \xi_b \end{aligned}$

3. Key Contributions

Shift from Matching to Assignment: The paper transitions the concept of "moment" from a tool for model reduction (matching) to a control design tool (assignment), allowing engineers to explicitly dictate steady-state behavior.
Sylvester Equation over Regulator Equations: The authors demonstrate that the solvability of the output regulation problem can be characterized entirely by the Sylvester equation. This "opens the algebraic black box" of classical regulator equations, revealing that they are simply the algebraic conditions for finding a compensator moment $M_c$ .
Necessary and Sufficient Conditions:
- Assignability: A moment $M_{des}$ is assignable if and only if $(M_{des} - M_{open})$ lies in the range of the moment transfer operator $T_S$ .
- Stabilizability: A stabilizing moment-assigning compensator exists if and only if:
  1. The plant $(A, B)$ is stabilizable.
  2. The pair $(C, A)$ is detectable.
  3. The pair $(M_{open}, S)$ is detectable.
- Physical Interpretation: The condition $(M_{open}, S)$ being detectable implies that no unstable mode of the signal generator is completely blocked by the plant's transmission zeros (which would make the disturbance invisible to the output).
Universal Parameterization: The paper provides a constructive synthesis procedure and a canonical form that captures the entire family of controllers achieving the desired steady-state response.

4. Results and Validation

Numerical Example: The authors applied the framework to a NASA HiMAT aircraft model.
- Scenario: The system was subjected to a mixture of constant wind and oscillatory gusts (frequency 3 rad/s).
- Objective: Perfectly reject the constant disturbance while attenuating (but not eliminating) the oscillatory disturbance to a specific trajectory.
- Outcome: The designed compensator successfully eliminated the constant bias and forced the oscillatory response to converge precisely to the assigned trajectory $M_{des}$ .
Simulation: The results validated that the proposed framework can enforce arbitrary steady-state behaviors determined by the designer while rejecting disturbances.

5. Significance and Impact

Unification of Control Paradigms: The framework unifies Output Regulation (setting $M_{des} = 0$ ) and Closed-Loop Interpolation (setting $M_{des} = M_{open}$ ) under a single theoretical umbrella.
Simplification of Design: By reducing the problem to linear algebraic equations (Sylvester equations) and standard output feedback stabilization, the design process is significantly simplified compared to solving non-linear regulator equations.
Data-Driven Potential: Since moments can be extracted directly from input-output data without full knowledge of state-space matrices, this work lays the groundwork for data-driven output regulation.
Generalizability: The authors note that since moment theory has been generalized to nonlinear, hybrid, and stochastic systems, this framework provides a foundation for extending steady-state assignment to these complex domains.

In summary, this paper offers a fundamental shift in how steady-state control is viewed, replacing complex regulator equation solving with a geometric and algebraic approach based on moment assignment and Sylvester equations, providing both theoretical clarity and practical synthesis tools.