Deadzone-Adapted Disturbance Suppression (DADS)

• DADS is a robust adaptive control framework that combines nonlinear damping, single-gain adjustment, and a deadzone mechanism to suppress matched uncertainties in nonlinear systems.
• It achieves attenuation of the regulated output to a preassignable level without requiring prior bounds on disturbances or parameters, ensuring practical IOS and zero output asymptotic gain.
• Extensions of DADS have addressed strict-feedback designs, partial-state feedback, ODE–PDE interconnections, and grid-forming converters, demonstrating its versatility across various uncertain environments.

Deadzone-Adapted Disturbance Suppression (DADS) is a Lyapunov-based adaptive control framework for nonlinear systems with unknown disturbances and parameters of arbitrary and unknown bounds, originally developed for matched uncertainties and later extended to strict-feedback systems, unknown input coefficients, partial-state feedback, infinite-dimensional interconnections, and grid-forming converters. Its defining architecture combines three elements—nonlinear damping, single-gain adjustment, and a deadzone in the update law—and was introduced as the first scheme to combine these three tools in one controller (Karafyllis et al., 2023). Across the cited developments, DADS is designed to achieve attenuation of the regulated state or output to an assignably small level, while preventing gain and state drift, and in several formulations it establishes practical IOS, p-UBIBS, and zero practical output asymptotic gain under matched uncertainty structures (Karafyllis et al., 2023, Karafyllis et al., 2024, Karafyllis et al., 24 Jul 2025, Karafyllis et al., 5 Oct 2025, Karafyllis et al., 20 May 2026, Rathnayake et al., 3 Mar 2026).

1. Origins, scope, and research trajectory

The initial DADS formulation addressed time-invariant nonlinear systems satisfying the matching condition, with full-state feedback and no known bounds on disturbances or unknown parameters (Karafyllis et al., 2023). The 2024 extension generalized the method to parametric strict-feedback systems via recursive backstepping (Karafyllis et al., 2024). In 2025, two distinct directions were developed: partial-state adaptive feedback with unmeasured matched dynamic uncertainty, including an infinite-dimensional reaction-diffusion PDE case (Karafyllis et al., 24 Jul 2025), and matched control-affine systems with unknown input coefficients, where only the sign of each input coefficient is assumed known (Karafyllis et al., 5 Oct 2025). In 2026, DADS was placed explicitly in the IOS/OAG tradition associated with Eduardo Sontag and extended to ODE–PDE interconnections with heat, transport, and viscously damped wave dynamics (Karafyllis et al., 20 May 2026), while a separate work incorporated DADS into nonlinear grid-forming control with safety-critical current limiting (Rathnayake et al., 3 Mar 2026).

Year	Main setting	Reference
2023	Matched uncertainties, full-state feedback	(Karafyllis et al., 2023)
2024	Parametric strict-feedback systems	(Karafyllis et al., 2024)
2025	Partial-state matched unmodeled dynamics	(Karafyllis et al., 24 Jul 2025)
2025	Unknown input coefficients	(Karafyllis et al., 5 Oct 2025)
2026	ODE–PDE partial-state feedback beyond small-gain	(Karafyllis et al., 20 May 2026)
2026	Grid-forming control with DADS and CBF safety filter	(Rathnayake et al., 3 Mar 2026)

This progression shows that DADS is not a single controller formula but a design pattern. The recurring elements are a Lyapunov function or CLF, matched-channel nonlinear damping aligned with the control direction, one adaptive gain state, and a deadzone that freezes adaptation in a prescribed residual set. A plausible implication is that the framework is best understood as a robust adaptive methodology rather than as a narrowly specialized regulator.

2. Canonical matched-uncertainty formulation

The canonical DADS problem is posed for systems of the form

$$\dot{x} = f(x) + g(x)\,\big( u + \varphi^\top(x)\,\theta + a(x)\, d\big),$$

where $x\in\mathbb{R}^n$, $u\in\mathbb{R}$, $d\in\mathbb{R}$, and $\theta\in\mathbb{R}^p$, with smooth $f,g,\varphi,a$, $f(0)=0$, and $\varphi(0)=0$ (Karafyllis et al., 2023). The matching condition means that both the parametric uncertainty $\varphi^\top(x)\theta$ and the disturbance $a(x)d$ enter through the same channel as the control, namely $x\in\mathbb{R}^n$0. This condition is structurally central in the DADS literature: it permits the Lyapunov derivative to be shaped by damping terms proportional to $x\in\mathbb{R}^n$1, and several later extensions preserve this same matched-channel logic (Karafyllis et al., 24 Jul 2025, Karafyllis et al., 5 Oct 2025).

The original controller is

$x\in\mathbb{R}^n$2

with a single adaptive state $x\in\mathbb{R}^n$3 evolving according to

$x\in\mathbb{R}^n$4

where $x\in\mathbb{R}^n$5, $x\in\mathbb{R}^n$6, $x\in\mathbb{R}^n$7, and $x\in\mathbb{R}^n$8 is smooth (Karafyllis et al., 2023). Here $x\in\mathbb{R}^n$9 is a Lyapunov function for the nominal plant, $u\in\mathbb{R}$0 is a nominal stabilizer, and $u\in\mathbb{R}$1 is the positive-part operator. The deadzone is encoded by $u\in\mathbb{R}$2: once $u\in\mathbb{R}$3, the adaptive state stops evolving.

This architecture makes precise the three ingredients from which the acronym derives. The term involving $u\in\mathbb{R}$4 is the nonlinear damping; the single dynamic state $u\in\mathbb{R}$5 implements single-gain adjustment; and the positive-part term supplies the deadzone. The disturbance and parameter bounds are not required a priori, and no identification law for $u\in\mathbb{R}$6 is introduced (Karafyllis et al., 2023).

3. Lyapunov mechanism, performance properties, and the role of the deadzone

The original matched-uncertainty analysis assumes smooth $u\in\mathbb{R}$7 with $u\in\mathbb{R}$8 positive definite and radially unbounded and

$u\in\mathbb{R}$9

together with a regressor-growth condition

$d\in\mathbb{R}$0

and a local exponential-type condition $d\in\mathbb{R}$1 in a neighborhood of the origin (Karafyllis et al., 2023). Under these assumptions, the closed-loop Lyapunov derivative is reduced to an estimate of the form

$d\in\mathbb{R}$2

for some $d\in\mathbb{R}$3 (Karafyllis et al., 2023). The technical significance is that the uncertainty contribution appears divided by $d\in\mathbb{R}$4, so growth of $d\in\mathbb{R}$5 directly attenuates the effect of unknown matched terms.

The main consequences are practical IOS, boundedness of the adaptive state, and zero practical output asymptotic gain. In the 2023 formulation, Theorem 1 yields bounded solutions, an estimate

$d\in\mathbb{R}$6

a bound

$d\in\mathbb{R}$7

and the assignable residual property

$d\in\mathbb{R}$8

(Karafyllis et al., 2023). In the strict-feedback extension, the same logic is expressed through p-UBIBS, p-IOS, and zero p-OAG for the output $d\in\mathbb{R}$9, with

$\theta\in\mathbb{R}^p$0

after recursive backstepping construction (Karafyllis et al., 2024).

A recurring point in the DADS literature is that the deadzone is not merely a convenience. A common misconception is that a leakage-type gain update is an interchangeable substitute. The 2023 paper gives a scalar counterexample showing that replacing the deadzone law with a leakage-type update may ensure p-IOS and boundedness of $\theta\in\mathbb{R}^p$1, but cannot achieve the zero p-OAG property with a bound independent of $\theta\in\mathbb{R}^p$2 (Karafyllis et al., 2023). The deadzone therefore has a specific structural role: it freezes adaptation once the regulated variable enters the prescribed tube, which prevents gain drift and underpins assignable residual regulation.

Another important clarification is that DADS is not an identification method. The strict-feedback paper states explicitly that the framework targets output attenuation and robustness to unknown and arbitrarily large parameters and disturbances, and does not aim at parameter identification or convergence (Karafyllis et al., 2024).

4. Recursive designs and unknown input coefficients

The strict-feedback extension studies systems with an integrator chain in $\theta\in\mathbb{R}^p$3 and a triangular chain in $\theta\in\mathbb{R}^p$4,

$\theta\in\mathbb{R}^p$5

with $\theta\in\mathbb{R}^p$6 and $\theta\in\mathbb{R}^p$7 (Karafyllis et al., 2024). The output to be attenuated is

$\theta\in\mathbb{R}^p$8

The controller is obtained through a step-by-step backstepping procedure, introducing virtual controls $\theta\in\mathbb{R}^p$9 and error coordinates $f,g,\varphi,a$0, and using a single gain integrator

$f,g,\varphi,a$1

The resulting composite Lyapunov function satisfies a derivative estimate with a disturbance term scaled by $f,g,\varphi,a$2 and a parameter term involving $f,g,\varphi,a$3, so that sufficiently large $f,g,\varphi,a$4 suppresses unknown parameter magnitudes without requiring prior bounds (Karafyllis et al., 2024). The paper contrasts this explicitly with $f,g,\varphi,a$5- or $f,g,\varphi,a$6-modification: unlike those methods, DADS does not make the residual set grow with $f,g,\varphi,a$7 (Karafyllis et al., 2024).

The 2025 note on unknown input coefficients extends DADS to multi-input control-affine systems

$f,g,\varphi,a$8

where the input coefficients $f,g,\varphi,a$9 may be time-varying and unknown, with only their sign assumed known and $f(0)=0$0 used only in the analysis, not in the control law (Karafyllis et al., 5 Oct 2025). The controller adopts the CLF-based form

$f(0)=0$1

The damping gains $f(0)=0$2 are explicit functions of $f(0)=0$3, the CLF gradient, and matched regressors. The paper emphasizes two structural consequences: the controller never divides by $f(0)=0$4, and Nussbaum-type functions are not used (Karafyllis et al., 5 Oct 2025). In this setting, DADS preserves the same qualitative guarantees—assignable attenuation, bounded $f(0)=0$5, and absence of drift of gains, states, and inputs—while relaxing the requirement of known input magnitudes.

These two extensions show that DADS is compatible with both recursive nonlinear design and CLF-based direct damping. The invariant ingredient is not the exact algebraic form of the law, but the way the adaptive gain scales matched damping until the Lyapunov residual enters the deadzone.

5. Partial-state feedback, matched unmodeled dynamics, and ODE–PDE interconnections

The partial-state DADS problem in (Karafyllis et al., 24 Jul 2025) considers

$f(0)=0$6

where $f(0)=0$7 is measured, $f(0)=0$8 is unmeasured, and the matched uncertainty enters additively in the input channel through $f(0)=0$9 (Karafyllis et al., 24 Jul 2025). The unmeasured state $\varphi(0)=0$0 models dynamic uncertainty, and the signals $\varphi(0)=0$1 are assumed only bounded in the sense of $\varphi(0)=0$2, with unknown sup norms. The controller uses only measured quantities such as $\varphi(0)=0$3, $\varphi(0)=0$4, $\varphi(0)=0$5, $\varphi(0)=0$6, $\varphi(0)=0$7, $\varphi(0)=0$8, and a single adaptive gain $\varphi(0)=0$9, with deadzone law

$\varphi^\top(x)\theta$0

The main theorem gives

$\varphi^\top(x)\theta$1

boundedness of $\varphi^\top(x)\theta$2, and an ultimate bound on the unmeasured state through $\varphi^\top(x)\theta$3 (Karafyllis et al., 24 Jul 2025). A central claim of the paper is that DADS can bypass small-gain conditions: instead of requiring known bounds on interconnection strength, it increases $\varphi^\top(x)\theta$4 until the denominators $\varphi^\top(x)\theta$5 and $\varphi^\top(x)\theta$6 render interconnection terms sufficiently small, while the deadzone prevents subsequent drift (Karafyllis et al., 24 Jul 2025).

The same paper includes an infinite-dimensional example in which the unmeasured dynamic uncertainty is governed by a reaction-diffusion PDE with unknown diffusion coefficient and unknown reaction term. Even in that case, a DADS controller can be designed and guarantees robust regulation of the plant state (Karafyllis et al., 24 Jul 2025). In the PDE controller, stronger nonlinear damping appears through high-order terms such as $\varphi^\top(x)\theta$7, reflecting the stronger coupling created by the infinite-dimensional uncertainty.

The 2026 paper "Beyond Nonlinear Small-Gain Design: DADS with Partial-State Feedback" recasts this line of work in the IOS/OAG framework associated with Eduardo Sontag and studies a scalar ODE interconnected with an almost completely unknown infinite-dimensional system (Karafyllis et al., 20 May 2026). The measured dynamics are

$\varphi^\top(x)\theta$8

with unmeasured $\varphi^\top(x)\theta$9 in a Banach space $a(x)d$0, unknown bounded inputs, and only the mild assumption $a(x)d$1 (Karafyllis et al., 20 May 2026). The controller is

$a(x)d$2

with

$a(x)d$3

Under abstract dissipation assumptions on the PDE and a bound $a(x)d$4, Theorem 1 proves p-IOS, p-UBIBS, bounded $a(x)d$5, and

$a(x)d$6

(Karafyllis et al., 20 May 2026). The same controller is shown to achieve robust regulation for three distinct interconnections: a heat PDE, a transport PDE, and a wave PDE with viscous damping (Karafyllis et al., 20 May 2026).

Taken together, these partial-state works establish that DADS can regulate measured finite-dimensional outputs in the presence of unmeasured finite- or infinite-dimensional matched dynamics without constructing an observer for the uncertainty state. This suggests an adaptive alternative to small-gain-based partial-state design when interconnection strengths are not known a priori, although the structural assumptions on dissipation and matching remain essential.

6. Grid-forming control, comparisons, and limitations

The most application-specific development in the provided corpus integrates DADS into nonlinear grid-forming control for a three-phase VSC connected to the grid through an LCL-like output and a series RL line (Rathnayake et al., 3 Mar 2026). The nominal architecture is droop-based with inner–outer backstepping: voltage references and frequency are generated by droop laws, an outer-loop voltage controller generates current references, and an inner-loop current controller synthesizes the terminal voltage. The grid voltage is treated as an unknown bounded disturbance, without requiring knowledge of its bound, and the controller design does not rely on network parameters beyond the point of common coupling (Rathnayake et al., 3 Mar 2026).

In this setting, DADS is applied separately to the $a(x)d$7- and $a(x)d$8-axis error energies

$a(x)d$9

with deadzone map

$x\in\mathbb{R}^n$00

and adaptive gains governed by

$x\in\mathbb{R}^n$01

(Rathnayake et al., 3 Mar 2026). The resulting closed-loop yields exponential transient decay with rate $x\in\mathbb{R}^n$02, practical voltage regulation

$x\in\mathbb{R}^n$03

and an assignable residual set

$x\in\mathbb{R}^n$04

for the PCC voltage errors (Rathnayake et al., 3 Mar 2026). A control-barrier-function safety filter is then wrapped around the nominal DADS-BS law to enforce strict current limits through a single-constraint quadratic program with a closed-form solution, guaranteeing forward invariance of the safe-current set (Rathnayake et al., 3 Mar 2026).

The comparative claims made in the papers are careful and domain-specific. In the grid-forming study, numerical results show that Safe DADS-BS and Safe PI both enforce the hard current bound, but Safe DADS-BS exhibits faster recovery during and after current-limiting events and resumes GFM behavior more rapidly than Safe PI (Rathnayake et al., 3 Mar 2026). In the strict-feedback paper, DADS achieves smaller steady-state output than a $x\in\mathbb{R}^n$05-mod adaptive controller in the reported wing-rock-inspired example, while the no-$x\in\mathbb{R}^n$06 variant exhibits parameter drift under persistent disturbance (Karafyllis et al., 2024). In the unknown-input-coefficient note, DADS is contrasted with leakage-based controllers, with the claim of less control effort and better attenuation in the reported uncertain double-integrator example (Karafyllis et al., 5 Oct 2025).

The limitations of DADS are equally explicit in the literature. Matching of uncertainties in the input channel is essential, and unmatched disturbances generally preclude the zero p-OAG property; the 2023 paper gives impossibility examples for unmatched cases (Karafyllis et al., 2023). The designs rely on Lyapunov or CLF hypotheses—Assumption (A), or related ISS/nominal GAS conditions in the partial-state case—and these assumptions are structural rather than cosmetic (Karafyllis et al., 24 Jul 2025, Karafyllis et al., 5 Oct 2025). Full-state measurement is required in some formulations, whereas partial-state versions depend on special interconnection inequalities rather than generic observability arguments (Karafyllis et al., 2024, Karafyllis et al., 24 Jul 2025). Very small deadzone thresholds tighten the residual set but may require larger adaptive gains and higher control effort, and the grid-forming paper notes sensitivity to measurement noise and performance degradation during prolonged safety-filter activation (Rathnayake et al., 3 Mar 2026). For strongly coupled PDE cases, high-order damping terms may be conservative and can demand high control effort (Karafyllis et al., 24 Jul 2025).

DADS is therefore best viewed neither as a universal adaptive controller nor as a parameter estimator. It is a matched-uncertainty regulation framework whose distinctive contribution is to use deadzone-controlled gain escalation to attain assignable practical regulation without prior disturbance or parameter bounds, while proving boundedness of the adaptive gain itself.