---
title: Decentralized Virtual Oscillator Control (dVOC)
url: https://www.emergentmind.com/topics/decentralized-virtual-oscillator-control-dvoc
type: topic
---

# Decentralized Virtual Oscillator Control (dVOC)

Decentralized Virtual Oscillator Control (dVOC) is a fully decentralized grid-forming strategy for inverter-dominated power systems that combines virtual-oscillator synchronization with explicit active-power, reactive-power, and voltage-magnitude setpoints. In the recent literature it is also termed **complex droop control**, and in some analyses it is associated with the Andronov–Hopf oscillator. Its defining feature is that each converter regulates its own terminal voltage from local voltage and current measurements only, while the network itself supplies the coupling that produces synchronization, voltage regulation, power sharing, black start, and dispatchability [1811.08842][2210.16288][2310.09935].

## 1. Definition, scope, and nomenclature

The original dVOC literature presents the method as a decentralized grid-forming inverter controller that preserves the synchronization and transient advantages of Virtual Oscillator Control (VOC) while adding **dispatchability**, meaning that the inverter can be commanded by user-defined setpoints \(p_k^\*\), \(q_k^\*\), and \(v_k^\*\) [1811.08842]. In this sense, dVOC is not merely a synchronization mechanism; it is a setpoint-tracking grid-forming control law with nonlinear oscillator dynamics.

A central terminological development is the later recognition that dVOC is algebraically equivalent to **complex droop control**. In that formulation, the controller is written in terms of a complex voltage state and normalized complex power, so that active and reactive power influence phase and amplitude through a single rotated complex gain \(e^{j\varphi}\) [2210.16288]. This equivalence is important because it connects dVOC to droop-control intuition without reducing it to a small-signal approximation.

A recurrent source of confusion is that the acronym “dVOC” is not used uniformly across the literature. One 2020 paper uses “dVOC” to denote **decentralized \(\lambda\)–\(\omega\) virtual-oscillator control**, a different oscillator family inspired by reaction-diffusion models, with amplitude and angle states \((r_k,\theta_k)\) rather than the dispatchable complex-droop formulation [2012.10907]. For precision, dispatchable virtual oscillator control and decentralized \(\lambda\)–\(\omega\) VOC should be treated as distinct control constructions that share a decentralized oscillator-based philosophy but not the same governing equations.

## 2. Control law and equivalent representations

In the \(\alpha\beta\) frame, the 2018 dVOC formulation regulates each inverter voltage vector \(v_i\) through
\[
\dot v_i
=\omega_0 J v_i
+\eta\Bigl(K_i v_i - R(\kappa)i_{o,i} + \alpha\,\phi_i(v_i)\,v_i\Bigr),
\]
where \(J=R(\pi/2)\), \(R(\kappa)\) is a rotation by \(\kappa\), \(K_i\) encodes \((p_i^\*,q_i^\*,v_i^\*)\), and
\[
\phi_i(v_i)=\frac{v_i^{\star2}-\|v_i\|^2}{v_i^{\star2}}.
\]
Here \(\eta>0\) sets the overall convergence speed, \(\alpha>0\) sets the strength of voltage-magnitude regulation, and \(\kappa\approx\arctan(\omega_0L/R)\) or, equivalently in later notation, \(\varphi\), aligns the controller with the network impedance angle [1811.08842][1802.08881].

In complex notation, the same controller is written as
\[
\dot v_k
= j\omega_0 v_k
+\eta e^{j\varphi}\bigl(\varsigma_k^\* v_k - i_{o,k}\bigr)
+\eta\alpha\,\Phi_k(v_k)\,v_k,
\]
with
\[
\varsigma_k^\*=\frac{p_k^\*-j q_k^\*}{v_k^{\*2}},
\qquad
\Phi_k(v_k)=\frac{v_k^{\*2}-|v_k|^2}{v_k^{\*2}}.
\]
This form makes explicit that dVOC is a rotated feedback of normalized complex power and voltage-magnitude error [2210.16288].

Around synchronous operation, dVOC reduces to familiar programmable droop relationships. For predominantly inductive coupling, the 2018 analysis gives the small-deviation approximation
\[
\omega_i \approx \omega_0 + \frac{\eta}{v_i^{\star2}}(p_i^\star-p_i),
\qquad
\|v_i\| \approx v_i^\star + \frac{1}{\alpha v_i^\star}(q_i^\star-q_i),
\]
which recovers the usual \(P\)–\(\omega\) and \(Q\)–\(V\) interpretation while retaining a nonlinear large-signal oscillator model [1811.08842].

A further abstraction is the **complex-frequency** viewpoint. Defining the complex angle \(\underline{\vartheta}(t)=\ln v(t)+j\theta(t)\), the complex frequency is
\[
\underline{\mathfrak F}=\dot{\underline{\vartheta}}=\frac{d}{dt}\ln \underline v=\epsilon + j\omega,
\]
where \(\epsilon=\dot v/v\) is the normalized rate of change of voltage and \(\omega=\dot\theta\) is angular frequency. In this representation, dVOC becomes a static complex-droop relation between power deviations and the coupled outputs \((\epsilon,\omega)\), which motivates later dynamic complex-frequency generalizations [2404.10071].

## 3. Synchronous equilibria and decentralized stability

Nonlinear stability theory for dVOC has developed along several complementary lines. For static Kron-reduced converter networks, the 2022 nonlinear analysis studies **non-nominal synchronous steady states**, rather than only nominal synchronous solutions. Let
\[
\mathbf A=j\omega_0 I+\eta e^{j\varphi}(\boldsymbol\Sigma^\*-\mathbf Y),
\]
with dominant eigenvalue \(\lambda_1\) and eigenvector \(\boldsymbol\phi_1\). Under the setpoint-consistency condition
\[
v_l^\*/v_k^\*=\bigl|\phi_{l1}\bigr|/\bigl|\phi_{k1}\bigr| \quad \forall k,l,
\]
the equilibrium family is \(T=S\cap A\), where \(S\) is the synchronous phase-growth set and \(A\) is the steady-state amplitude set. The paper then proves almost global asymptotic stability of \(T\) by a Lyapunov function built from distance to the synchronized subspace and voltage-magnitude error, with the origin as the unstable zero-measure exception [2210.16288].

A different but closely related stability strand addresses the effect of **transmission-line dynamics**. When line dynamics are neglected and the network is represented algebraically, dVOC admits almost global asymptotic stability with respect to a prescribed AC power-flow solution. When line RL dynamics are included,
\[
L_T\dot i=-Z_T i+B^\top v,
\]
the electromagnetic transients can compromise stability, and explicit bounds are required on controller setpoints, branch powers, the voltage-restoration gain \(\alpha\), and the synchronizing gain \(\eta\) [1802.08881]. This establishes that algebraic-network stability does not automatically imply full-order stability in inverter-dominated systems.

The 2023 passivity result substantially strengthens the decentralized stability picture. In a large-signal dq-frame model, each dVOC-controlled converter node is shown to be **output-feedback passive** with an explicit passivity index \(\delta_k\), and the rotated network is input-feedforward passive with index
\[
\varepsilon_{\rm net}
=\lambda_{\min}\bigl(\Re\{e^{j\varphi}\mathbf Y\}\bigr).
\]
The resulting decentralized transient-stability condition is
\[
\delta_k+\varepsilon_{\rm net}>0,\qquad \forall k=1,\dots,N.
\]
If the equilibrium is unique, this yields global asymptotic stability; otherwise it yields asymptotic stability of the equilibrium under consideration. The associated composite storage function
\[
\mathcal V(\{\underline v_k\})=\sum_{k=1}^N V_k(\underline v_k)
\]
gives a Lyapunov/energy-based proof of large-signal stability without centralized eigenvalue computation or small-signal linearization [2310.09935].

The same passivity framework also gives an explicit gain-selection rule. Using a conservative bound on the node passivity index, the paper states the practical design condition
\[
\alpha_k \ge \Re\Bigl\{e^{j\varphi}(p_k^\*-j q_k^\*)/v_k^{\*2}\Bigr\}
-\varepsilon_{\rm net}+\gamma,
\]
for some small margin \(\gamma>0\). This formalizes an engineering intuition already present in earlier dVOC work: larger power setpoints or weaker networks require stronger voltage-amplitude regulation [2310.09935].

## 4. Network dynamics and reduced-order modeling

dVOC analysis often relies on time-scale separation, but later work makes that approximation explicit and systematic. In the \(\lambda\)–\(\omega\) VOC line of work, a reduced model is obtained under the fast-line assumption
\[
\frac{L_O}{\sqrt{R_O^2 + (L_O\omega^\*)^2}}\ll 1,
\]
which permits instantaneous line-current approximation and yields a cascaded reduced-order model. A two-step Lyapunov argument then proves almost-global asymptotic stability of the desired synchronous set \(\mathcal S'\) under a controller-gain condition involving \(\alpha_k\), \(\gamma_k\), and constants derived from the line-impedance matrix \(L\) [2012.10907]. Although this is a distinct controller family, it illustrates the general importance of time-scale separation in decentralized oscillator control.

For dispatchable VOC with nested current and voltage loops, model-reduction work in 2021 extends the analysis to include **current-reference limiting**, which earlier reduced-order approaches had omitted. The full-order averaged model contains outer dVOC dynamics, voltage PI control, current PI control, and an LCL filter. A smooth approximation of the hard current limiter is introduced through
\[
\rho = -\epsilon\ln\!\Bigl[\exp(-1/\epsilon)+\exp\!\bigl(-I_{\max}/(\epsilon\|I_{idq}^\*\|)\bigr)\Bigr],
\]
so that the limited current reference becomes \(I_{idq}^z=\rho I_{idq}^\*\) [2105.09754].

That paper then applies participation-factor analysis to a \(12\times 12\) Jacobian and reports a clean eigenvalue split at about \(\omega\simeq -260\,{\rm rad/s}\). States with \(\Re(\lambda)<-260\,{\rm rad/s}\) are classified as fast, those with \(\Re(\lambda)>-260\,{\rm rad/s}\) as slow. For the inductive network case, the slow manifold is \(x=[\delta;E^\*;I_{g,dq}]\); for the resistive case, it is \(x=[\delta;E^\*]\). Singular perturbation then yields reduced-order models tailored to inductive and resistive coupling, while retaining the smooth limiter and current-limiting action [2105.09754].

These reduction results are significant because they bridge idealized dVOC theory and implementation-grade inverter models. A plausible implication is that they make decentralized stability analysis computationally tractable in settings where full averaged or switched models would be cumbersome.

## 5. Current limitation, fault behavior, and saturation-informed control

Transient stability under current saturation is a critical issue for grid-forming converters, and a 2024 study analyzes dVOC explicitly in that regime. The classical circular limiter is
\[
\bar i=
\begin{cases}
\hat i, & |\hat i|\le i_{\lim},\\[4pt]
i_{\lim}\,\hat i/|\hat i|, & |\hat i|>i_{\lim},
\end{cases}
\]
and the **Degree of Saturation (DoS)** is defined as
\[
\mu=\frac{\bar i}{\hat i},
\]
so that \(\mu=1\) when unsaturated and \(0<\mu<1\) when saturated. In a conventional virtual-admittance inner loop,
\[
\hat i=\frac{1}{z_v}(\hat v-v),
\]
saturation induces the effective impedance \(z_v/\mu\), which becomes current dependent and can grow arbitrarily large as \(\mu\to 0\) [2404.07682].

The proposed remedy is a **saturation-informed current-limiting control**. After low-pass filtering the DoS,
\[
\mu_f(s)=\frac{1}{\tau s+1}\mu,
\]
the outer dVOC loop scales the current feedback by \(\mu_f\), and the inner voltage loop scales the voltage error by \(1/\mu_f\). Defining the internal scaled voltage
\[
\hat v_\mu=\mu_f \hat v,
\]
the saturated converter becomes equivalent to a classical dVOC converter behind the **constant** virtual impedance \(z_v\), and \(\hat v_\mu\) obeys a dVOC law of exactly the same form as the unsaturated case, except with \(v_\mu^\*=\mu_f v^\*\) [2404.07682].

This equivalence permits Lyapunov proofs to be imported from the unsaturated setting. The paper states parametric decentralized stability conditions for single-converter grid-connected systems, multi-converter grid-connected systems, and islanded microgrids. In the grid-connected multi-converter case, the condition is expressed in terms of the augmented admittance \(\tilde Y_c\) and \(\lambda_{\min}(\Re\{e^{j\varphi}\tilde Y_c\})\), identified there as the generalized short-circuit ratio of the augmented network. For islanded microgrids, the condition involves the second eigenvalue \(\lambda_2(\Re\{e^{j\varphi}\tilde Y_m\})\) and steady-state bounds on angle and voltage deviations [2404.07682].

The same paper reports EMT case studies for a single converter, three converters connected to a common PCC, and an IEEE 9-bus islanded microgrid. Under the conventional circular limiter, synchronism is lost during severe faults; under the saturation-informed scheme, the terminal voltage and phase remain locked and the converters ride through the fault while the output current remains clamped at \(i_{\lim}\) [2404.07682].

## 6. Extensions, experimental validation, and related architectures

The experimental basis for dVOC was established early. A hardware testbed at NREL used up to five identical \(1\,{\rm kVA}\) inverters, with reported results for two inverters in parallel supplying a purely resistive load. In that setup, a single dVOC inverter black-started a \(500\,{\rm W}\) resistive \(+250\,{\rm var}\) filter load from zero voltage and reached \(90\%\) of nominal in approximately \(50\,{\rm ms}\). When a second inverter was connected, the two synchronized in approximately \(150\,{\rm ms}\) (10 cycles) and split a \(500\,{\rm W}\) load equally with no current overshoot. Load steps and setpoint redispatch were tracked within one cycle, while the controller remained backward-compatible with droop concepts [1811.08842].

Several later extensions broaden the dVOC framework rather than replacing it. Dynamic complex-frequency control upgrades the static gains of complex droop to complex-valued transfer functions,
\[
\Delta\underline{\mathfrak F}(s)
=
\underline T(s)\Bigl(-\Delta\underline{\overline\varsigma}(s)-\underline T^v(s)\Delta v(s)\Bigr),
\]
so that inertia, damping, and lead/lag behavior can be shaped explicitly. The small-signal stability claim is that if \(e^{-j\varphi_z}\underline T(s)\) is strictly positive real and \(\underline T^v(s)\) is asymptotically stable, then the closed-loop converter-network system is internally stable, with simultaneous frequency synchronization and voltage stabilization. Numerical studies on the IEEE nine-bus system include a single-converter case with \(M=2\,{\rm s}\), \(D=50\), \(\alpha=5\), \(\varphi=\pi/4\), and a \(+25\,{\rm MW}\) load step, as well as a three-converter aggregation case with exact \(1{:}1{:}1\) sharing [2404.10071].

A 2025 contraction-based analysis treats dVOC as an Andronov–Hopf oscillator with auxiliary virtual impedance. The decentralized large-signal synchronization condition is
\[
\kappa\beta - 2\xi X_{\rm nom}^2 \ge c>0,
\]
which requires no bus-level or line-impedance data. Under full synchronization, proportional current sharing follows from admittance ratios \(I_i:I_j=Y_i:Y_j\). In a 33-turbine wind-plant example, the controller achieved the current-sharing ratio \(I_{19}/I_{33}\approx 20/10.5\) within \(0.63\%\) error, and remained contracting under a randomly inserted \(200\times\) impedance spike for \(t\in[0,0.4]\,{\rm s}\) [2509.09277].

The control family has also been extended toward unified grid-forming/grid-following behavior. A 2026 framework integrates dVOC with reference-following synchronization through
\[
\dot v
=j\big[\epsilon\omega_0+(1-\epsilon)\omega_u\big]v
+\mu(V_0^2-\|v\|^2)v
+\eta(i_0-i)e^{j\phi}
+\gamma(\tilde v-v),
\]
with continuous parameters \(\{\epsilon,\mu,\eta_1,\eta_2,\gamma\}\). By tuning these gains, the controller interpolates among PQ, PV, Qf, Vf, and Hybrid modes without discrete switching; pre-synchronization is achieved by setting \(\gamma\gg 0\) and \(\tilde v=u\). The paper reports EMT simulations on the IEEE 39-bus system and hardware-in-the-loop tests on a \(10\,{\rm kW}\) inverter [2607.01616].

A related development is the current-source dual of dVOC, termed **dispatchable current source virtual oscillator control (dCVOC)**. There the current frequency is generated through reactive-power control and the current magnitude reference through active-power control. The paper states that the proposed scheme always admits a steady-state equilibrium and ensures global stability under reasonable conditions on grid and converter parameters, even with LVRT and current saturation constraints. This is presented as a duality with dVOC rather than a reformulation of dVOC itself [2510.26977].

Taken together, these results position dVOC as a nonlinear, decentralized, grid-forming control architecture with multiple mathematically equivalent representations—virtual oscillator, complex droop, and complex frequency—and with a stability theory that now spans almost-global synchronization, non-nominal drooped equilibria, passivity-based transient stability, contraction-based synchronization, and saturation-informed fault behavior [2210.16288][2310.09935].

Source: https://www.emergentmind.com/topics/decentralized-virtual-oscillator-control-dvoc