On a frame theoretic measure of quality of LTI systems

Mohammed Rayyan Sheriff, Debasish Chatterjee

Abstract

A measure of quality of a control system, i.e., a quantitative extension of the classical notion of controllability is a practically relevant idea. In this article we demonstrate that the three standard measures of quality involving the trace, minimum eigenvalue, and the determinant of the controllability grammian achieve their optimum values when the columns of the controllability matrix form a tight frame. Motivated by this, and in view of some recent developments in frame theoretic signal processing, we propose a natural measure of quality for LTI systems based on a measure of tightness of the columns of the reachability matrix. A sufficient condition in terms of the proposed measure of quality for the classical boolean notion of controllability has been provided.

Introduction

In classical control theory, controllability is typically treated as a binary property: a system is either controllable or it is not. However, for practical engineering applications, this "yes/no" distinction is often insufficient. Some systems are "more controllable" than others in the sense that they require significantly less energy to reach desired states. This paper addresses the need for a quantitative Measure of Quality (MOQ) for Linear Time-Invariant (LTI) systems, providing a bridge between control theory and frame theory from signal processing.

The Geometry of Control

The fundamental insight of this work is that the quality of a control system depends on the geometric orientation of its reachability vectors. These vectors, which form the columns of the reachability matrix $R = [B, AB, A^2B, \dots, A^{n-1}B]$, define the directions in which control inputs can influence the state space.

Geometric comparison of reachability vectors. Configuration (a) shows well-distributed vectors, while (b) shows clustered vectors that lead to poor control quality in certain directions.

As illustrated in Figure 1, if these vectors are clustered together (as in 1b), the system becomes "fragile." Moving the system in a direction orthogonal to this cluster requires massive amounts of energy. A high-quality system should have its reachability vectors spread out as evenly as possible to ensure efficient control in all directions.

Frame Theory and Tightness

To quantify this "spread," the authors utilize frame theory. A set of vectors is called a tight frame if it behaves like an orthonormal basis but contains more vectors than the dimension of the space. Geometrically, tight frames represent the most symmetric and well-distributed arrangement of vectors possible.

Degrees of frame tightness: (a) is a perfectly tight frame with vectors at 120-degree angles; (b) and (c) show increasing levels of non-tightness and clustering.

The paper rigorously proves that three standard measures of control quality—the average control effort ($tr(G^{-1})$), the maximum control effort ($\lambda_{min}^{-1}$), and the volume of the reachable ellipsoid ($det(G)$)—all reach their optimal values if and only if the reachability vectors constitute a tight frame.

The Normalized Frame Potential (NFP)

While standard MOQs are effective, they can be computationally expensive for high-dimensional systems, often requiring matrix inversions or determinant calculations. The authors propose a new metric, the Normalized Frame Potential (NFP), denoted as $\eta(A, B, T)$.

This metric is defined as: $$\eta = \frac{tr(G^2)}{(tr G)^2}$$

where $G$ is the controllability Grammian. The NFP is significantly easier to compute as it only requires calculating the trace of the Grammian and its square. The value of $\eta$ ranges from $1/n$ (perfectly tight/optimal) to higher values. Furthermore, the authors establish a sufficient condition for classical controllability: if $\eta < \frac{1}{n-1}$, the system is guaranteed to be controllable.

Conclusion and Future Work

By framing controllability as a geometric problem of vector distribution, this research provides a robust theoretical foundation for assessing system quality. The NFP metric offers a computationally efficient tool for engineers to evaluate control system designs. Future research directions include extending these frame-theoretic measures to time-varying and non-linear systems, as well as integrating the NFP into automated hardware and controller design pipelines.

Summary updated

Abstract

A measure of quality of a control system, i.e., a quantitative extension of the classical notion of controllability is a practically relevant idea. In this article we demonstrate that the three standard measures of quality involving the trace, minimum eigenvalue, and the determinant of the controllability grammian achieve their optimum values when the columns of the controllability matrix form a tight frame. Motivated by this, and in view of some recent developments in frame theoretic signal processing, we propose a natural measure of quality for LTI systems based on a measure of tightness of the columns of the reachability matrix. A sufficient condition in terms of the proposed measure of quality for the classical boolean notion of controllability has been provided.
BibTeX
@misc{sheriff2017frame,
  author = {Mohammed Rayyan Sheriff and Debasish Chatterjee},
  title = {{On a frame theoretic measure of quality of LTI systems}},
  year = {2017},
  eprint = {1703.07539},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1703.07539},
}
Figure