Dirty derivatives for output feedback stabilization

Matteo Marchi, Lucas Fraile, Paulo TabuadaarXiv 2022 · arXiv

Abstract

While "dirty derivatives"—low-pass filtered approximations of signal derivatives—are widely used in industrial control, they have historically lacked rigorous stability guarantees for high-order systems. This paper provides the first formal Lyapunov-based stability proofs for using these approximations in output feedback stabilization of linear time-invariant (LTI) systems. Numerical simulations demonstrate that dirty derivatives offer a robust, low-complexity alternative to state-of-the-art observers, particularly in noisy environments.

Introduction

In control engineering, stabilizing a system often requires knowledge of state derivatives, such as velocity derived from position. However, implementing a "pure" derivative is practically impossible due to its non-causal nature and its tendency to amplify high-frequency measurement noise. To circumvent this, engineers frequently employ "dirty derivatives," which pass the signal through a high-pass filter that rolls off at high frequencies. Despite their ubiquity in PID controllers, these approximations have lacked a theoretical framework ensuring stability for general high-order systems. Most formal methods instead rely on complex state observers, which are prone to "peaking"—large, transient spikes in control effort—and require detailed plant models.

Methodology

The authors address Single-Input Single-Output (SISO) LTI systems in controller canonical form. Their approach centers on a recursive definition of higher-order dirty derivatives, where each derivative is a filtered version of the previous one, governed by a single tuning parameter $\sigma$ (the filter pole).

The core contribution is a stability proof using a custom Lyapunov function that integrates both the system state and the estimation errors of the dirty derivatives. The analysis demonstrates that for a sufficiently large $\sigma$, the closed-loop system achieves global asymptotic stability. Furthermore, the authors extend this framework to adaptive control scenarios where the control gain is unknown, showing that stabilization is possible without explicit gain estimation.

Performance Evaluation

The proposed method was evaluated via numerical simulations of a 5th-order nonlinear system subjected to Gaussian noise (0-200 Hz). The dirty derivative approach was compared against two state-of-the-art techniques: low-power peaking-free high-gain observers and algebraic derivative estimation.

Comparison of derivative estimation methods across different techniques.

The results indicate that dirty derivatives provide significantly smoother transient behavior than high-gain observers, effectively suppressing the peaking phenomenon. Quantitatively, the method achieved lower absolute estimation errors for both first and second derivatives compared to algebraic methods.

Absolute error of first derivative estimates, highlighting the robustness of the dirty derivative approach.

Key Insights

  • Simplicity and Robustness: Dirty derivatives require only one tuning parameter and no knowledge of internal plant dynamics, making them easier to implement than model-based observers.
  • Noise Mitigation: By design, the low-pass filtering nature of dirty derivatives makes them inherently robust to measurement noise, often outperforming "optimal" observers in real-world conditions.
  • Transient Management: Unlike high-gain observers, dirty derivatives naturally limit the magnitude of initial transients, protecting physical hardware from excessive control effort.

Conclusion and Future Work

This work bridges the gap between industrial practice and control theory by providing a formal stability foundation for dirty derivatives. While currently limited to systems in controller canonical form, the results suggest that these simple filters are a powerful tool for output feedback. Future research will focus on extending these proofs to more general observable systems and exploring the theoretical limits of the high-gain requirement for the filter parameter $\sigma$.

Summary updated

Abstract

Dirty derivatives are routinely used in industrial settings, particularly in the implementation of the derivative term in PID control, and are especially appealing due to their noise-attenuation and model-free characteristics. In this paper, we provide a Lyapunov-based proof for the stability of linear time-invariant control systems in controller canonical form when utilizing dirty derivatives in place of observers for the purpose of output feedback. This is, to the best of the authors' knowledge, the first time that stability proofs are provided for the use of dirty derivatives in lieu of derivatives of different orders. In the spirit of adaptive control, we also show how dirty derivatives can be used for output feedback control when the control gain is unknown.
BibTeX
@misc{marchi2022dirty,
  author = {Matteo Marchi and Lucas Fraile and Paulo Tabuada},
  title = {{Dirty derivatives for output feedback stabilization}},
  year = {2022},
  eprint = {2202.01941},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2202.01941},
}
Figure