Mitigating Noise Amplification in High Gain State Observers

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 221 papers31 conceptsfrozen 2026-10-04

In control theory, the "high-gain" approach is a powerful lever. By increasing the gain of an observer or controller, engineers can force a system to converge to its target state rapidly, even when the underlying dynamics are uncertain or plagued by external disturbances. This philosophy underpins foundational techniques like Model Reference Adaptive Control (MRAC) and Active Disturbance Rejection Control (ADRC).

However, this power comes with a steep price: the "noise explosion." As researchers like Masayoshi Tomizuka noted as early as 1985, high adaptation gains intended to minimize tracking errors simultaneously amplify the microscopic fluctuations of sensor noise. This creates a fundamental tension that defines modern research in state estimation: the faster you try to "catch" a disturbance, the more likely you are to be tripped up by the noise of your own sensors.

The High-Gain Dilemma

To understand why this happens, consider the Extended State Observer (ESO), a centerpiece of ADRC. An ESO treats everything it doesn't know about a system—friction, external wind, or unmodeled physics—as a "total disturbance" to be estimated and canceled in real-time.

To capture fast-moving disturbances, the observer needs a high bandwidth. But as Krzysztof Łakomy and Rafał Madoński explain in Active Disturbance Rejection Control with Sensor Noise Suppressing Observer, increasing this bandwidth allows high-frequency measurement noise to enter the control loop. This noise is amplified by the high gain, leading to "jitter" or "buffeting" in the actuators. In precision applications like DC-DC power converters or high-speed motion stages, this jitter can degrade performance or even cause hardware failure.

This conflict is often framed as a rigid trade-off between tracking bandwidth and noise suppression. If the gain is too low, the system is sluggish and fails to reject disturbances; if it is too high, the system becomes hypersensitive to sensor errors.

Peaking and Instability

Beyond simple jitter, high-gain observers (HGOs) suffer from a phenomenon known as "peaking." When an observer is first turned on, the initial discrepancy between the estimated state and the true state can be large. A high-gain mechanism reacts to this discrepancy violently, producing a massive, transient spike in the control signal.

Mona Buisson-Fenet and Sebastian Trimpe highlight that these peaking effects can destabilize the entire control loop before the observer has a chance to converge. This is particularly dangerous in safety-critical systems. As Kendra Lesser argues in Safety verification of output feedback controllers for nonlinear systems, designers must often "shrink" their safety constraints to account for these initial estimation errors, leading to overly conservative system behavior.

Decoupling Noise from Estimation

The frontier of this research area lies in breaking the coupling between high-speed estimation and noise sensitivity. Rather than accepting the trade-off, recent work focuses on architectural changes to the observer itself.

Cascaded and Multi-Level Observers

One prominent strategy is the Cascade Extended State Observer (CESO). Developed by Łakomy and colleagues, the CESO decomposes the disturbance estimation into multiple levels. By using a multi-level structure, the observer can effectively "silence" sensor noise while maintaining the high-gain necessary for accurate voltage tracking. A similar "ensemble" approach is seen in the work of Guojie Tang, who proposed Parallel Multi-Extended State Observers to balance the complex disturbance capture of high-order observers with the robustness of lower-order ones.

Dual-Loop Architectures

In ultra-precise applications like laser power stabilization, researchers have moved toward decoupling the sources of error. Yanpei Shi and Zhuo Shi introduced a Dual-Loop ADRC strategy. In this framework, an inner loop handles model uncertainty while a cascaded outer loop specifically manages sensor noise decoupling. This separation allowed for a tenfold improvement in precision by preventing the high-frequency noise from "polluting" the disturbance rejection mechanism.

Nonlinear and Adaptive Filtering

Other researchers focus on the "differentiator" problem—the fact that observers essentially act as high-pass filters that naturally amplify noise. Xinyu An and Haoda Li addressed this in underwater robotics by integrating nonlinear tracking differentiators. These components allow the system to follow a path accurately even with low-bandwidth sensors, reducing the "buffeting" effect on thrusters that traditional high-gain methods would cause.

Conclusion

The evolution of high-gain control is a transition from "brute force" convergence to "surgical" estimation. While the foundational work of Tomizuka identified the fundamental trade-off, modern researchers like Łakomy, Tang, and Shi are finding ways to bypass it. By using cascaded, parallel, and dual-loop architectures, it is becoming possible to achieve the rapid response of high-gain systems without the catastrophic "noise explosion" that once limited their practical use.

Papers in this version

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 16 papers
  2. Version 1 · 2026-10-04 · 21 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_high_gain_noise_sensitivity_1a26f2fa,
  title = {high-gain noise sensitivity},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/high-gain-noise-sensitivity/rdr_1a26f2faed3e4ebea1a770a8027efc3b},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_mitigating_noise_amplification_in_high_g_1a26f2fa_v1,
  title = {Mitigating Noise Amplification in High Gain State Observers},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/mitigating-noise-amplification-in-high-gain-state-observers/rdr_1a26f2faed3e4ebea1a770a8027efc3b/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}