betaFeedback

WorksEngineering

Decentralized Impedance Specifications for Grid-Connected Converters: a Sequence-Domain Gershgorin Method, and a Frequency-Wise Framework for Combining Different Criteria

Abstract

In complex grid scenarios, with many power electronics (PE) converters connected, having specifications for converters can ease the design process. A practical approach is to impose the specifications on the converter's small-signal impedance in the frequency domain. This is advantageous since impedance is a physical property that can be measured. This paper focuses on providing decentralized impedance specifications for each converter connected to an ac bus. With decentralized specifications, each converter has its own specifications, which are independent of other converters' impedance. This enables a competitive multi-vendor environment while guaranteeing system stability. Existing methods of imposing decentralized specifications, based on the small-gain, passivity and Gershgorin theorems, are formulated within a common frequency-domain framework. Building on this, two new solutions that reduce the conservativeness of the specifications are proposed. First, Gershgorin-based decentralized specifications are formulated in the modified sequence domain. Second, a framework for combining different specification types across frequency ranges is developed and mathematically proven. The methods are compared on examples covering sub-synchronous and harmonic-range interactions of grid-following converters, grid-forming converters, and a mixed grid-forming/grid-following system, using analytical models, black-box impedance measurements, and time-domain responses obtained from simulations and an experimental setup.

Introduction

The global transition toward renewable energy sources has significantly increased the penetration of power electronics (PE) converters into electrical grids, introducing challenges such as converter interactions and small-signal stability issues. Traditional stability assessment techniques—state-space pole analysis and frequency-domain control loop-gain methods—are effective for individual converter design, but become complicated when multiple converters are interconnected, even for simple practical cases.

Impedance-based stability assessment offers an effective framework for studying converter interactions. First applied by Middlebrook for input filter design [1] and later extended to grid-connected converters [2, 3], the method partitions the system into source and load subsystems, each represented by its Thévenin or Norton equivalent. Stability is then assessed by applying the Nyquist criterion to the source-to-load impedance ratio, known as the minor loop gain.

While a single-input single-output (SISO) representation is sufficient to capture the dynamics at the dc terminal of PE converters, the ac-side typically requires a multiple-input multiple-output (MIMO) representation, usually in the dqdq frame [4, 5, 6] to capture all relevant dynamics. Consequently, the ac-side impedance is a transfer function matrix, which necessitates the use of the generalized Nyquist criterion.

Since applying the Nyquist criterion offers limited insight into shaping the desired individual impedances, alternative impedance-based stability specifications were first proposed by Middlebrook [1] and later extended by Wildrick et al. [7]. For MIMO systems, several approaches to defining such specifications have been introduced, including methods based on matrix norms [8], on the diagonal dd- and qq-axis elements of the impedance matrix (applicable when the impedances are diagonally dominant) [9], passivity theory requiring both source and load subsystems to be passive [10], net damping criteria requiring the combined damping of both subsystems to be positive [11, 12, 13, 14], and, more recently, a rotated passivity index [15].

In the case where specifications are needed for multiple converters, the decentralized impedance specifications [16] were proposed. Here, the goal is to define individual conditions that each converter should satisfy independently of the others

This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.

The research leading to these results has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement no 101034319 and from the European Union — NextGenerationEU. This work was also partially supported by the project WIND-DIGIPOWER “Large-scale WIND integration for the future DIGItal POWER grid using innovative power electronics control and communication-based estimations” funded within the call “CETPartnership Joint Call 2022”.

L. Stojanovic and P. Mattavelli are with the Department of Management and Engineering, University of Padova, 36100 Vicenza, Italy (e-mail: [email protected]; [email protected]).

R. Cvetanovic is with Innovation Center of the School of Electrical Engineering in Belgrade, 11000 Belgrade, Serbia (e-mail: [email protected]).

S. Buso is with the Department of Information Engineering, University of Padova, 35131 Padova, Italy (e-mail: [email protected]). so that, once all converters are connected together, the system remains stable. This can be of great interest when many converters are planned for connection to the grid since the specifications for each converter can be given without knowing the impedance of the others up front.

Although impedance specifications for a single converter are well established in the literature for both dc systems [1], [7] and ac systems [8], [13], [15], decentralized specifications for multiple converters have predominantly been developed for dc systems—including a single dc bus [16], [17], series-type dc microgrids [18], multi-bus dc systems [19], [20], and dc distribution systems connected to an ac grid [21]—and only recently extended to converters connected to a common ac bus [22], [23], which is also the focus of this paper. For converters connected to different ac buses, decentralized conditions have been derived based on passivity of large-signal grid-forming models [24], on combining the small-gain and small-phase theorems [25], and, for multi-vendor grids, on robust stability analysis of grey-box converter models [26]. Related control-theoretic conditions have also been proposed for dc microgrids [27] and for generic networked systems [28], [29]. However, these conditions are formulated on control-loop transfer functions rather than on the terminal impedance [26], address large-signal rather than small-signal stability [24], or rely on properties, such as sectoriality, that converter impedances often do not satisfy [23], [25]; imposing them directly on measured frequency data is therefore not straightforward. In contrast, for converters connected to a common ac bus, the specifications reduce to simple per-converter conditions on the measured impedance.

In our previous work [22], we proposed a method that formulates decentralized specifications for the dqdq-frame impedance based on the Gershgorin theorem, previously mainly employed for stability analysis [30]–[34]. It is recognized in [22] that the approach introduces a certain degree of conservativeness; however, alternative decentralized-specification methods and a systematic comparison of their conservativeness are not addressed. More recently, [23] proposed decentralized specifications based on a combination of the net-damping criterion and a rotated passivity index. Similarly to [22], it is demonstrated that assigning individual specifications to each converter can effectively prevent instability in multi-parallel converter systems.

Despite these advances, several gaps remain for converters connected to a common ac bus. First, comparisons between decentralized-specification methods are limited: [22] does not compare against alternatives, while [23] compares its extended-passivity condition only with conventional passivity and small-gain conditions for a grid-forming case, without considering Gershgorin-based specifications. Second, both [22], [23] primarily address low-frequency (near-synchronous) interactions, leaving the harmonic frequency range largely unexplored. Finally, combinations of methods across frequency ranges have so far been limited to passivity- and small-gain-type conditions: [15], [23] combine the rotated passivity index at low frequencies with the net-damping criterion at high frequencies, and [35] unifies small-gain, passivity, and rotated-passivity conditions through frequency-dependent multipliers.

However, Gershgorin-based specifications are not covered by this framework, and [35] is validated only on a single-converter system, without deriving per-converter specifications for multi-parallel converters.

To fill in these gaps, this paper provides the following novel contributions related to imposing decentralized impedance specifications for converters connected to the same ac bus:

  • Formulating the existing decentralized specification methods within a common frequency-domain framework, with easily-applicable expressions for each method, which is later used for their combination and comparison.

  • Proposing the decentralized specifications in the modified sequence domain based on the Gershgorin theorem, extending the work in [22].

  • Developing a general framework for combining different decentralized specification methods across frequency ranges, with a mathematical proof — based on the forbidden region principle — that the combination remains a sufficient condition for system stability.

  • Comparing the conservativeness and applicability of all methods on examples covering both sub-synchronous and harmonic interactions, together with practical guidelines on which specification type is most appropriate for each frequency range.

  • Verifying the specifications applied directly to black-box impedance measurements, without relying on analytical converter models.

Impedance-Based Stability Assessment

To perform impedance-based stability assessment, the system is partitioned into source and load subsystems [1], each represented by its Thévenin or Norton equivalent with terminated small-signal impedances Zs(s)Z_s(s) and Zl(s)Z_l(s), as shown in Figure 1. One of the key advantages of this approach is that impedances are physical quantities that can be measured directly.

Figure 1. System partitioned to a source and load equivalent representation.

Throughout this paper, dc-system impedances are represented as SISO transfer functions, while ac-system impedances are modeled in the dqdq frame as transfer function matrices [4], [5], denoted in bold (e.g., Zg\mathbf{Z}_g), with elements:

Zg(s)=[Zgdd(s)Zgdq(s)Zgqd(s)Zgqq(s)].(1)\mathbf{Z}_g(s) = \begin{bmatrix} Z_g^{dd}(s) & Z_g^{dq}(s) \\ Z_g^{qd}(s) & Z_g^{qq}(s) \end{bmatrix}. \tag*{(1)}

Applying the Nyquist criterion

For a dc system, stability is assessed by applying the Nyquist criterion to the minor loop gain L(s)=Zs(s)/Zl(s)L(s)=Z_s(s)/Z_l(s). Following [2], [3], the source is assumed stable under ideal load (constant current) and the load stable under ideal source (constant voltage), so that L(s)L(s) has no RHP poles. The system is then stable if and only if the Nyquist plot of L(jω)L(j\omega) does not encircle (−1,0)(-1,0). This assumption of no RHP poles in the minor loop gain is maintained throughout the paper.

For ac systems, the generalized Nyquist criterion is applied to the matrix minor loop gain L(s)\mathbf{L}(s), under the same assumption of no RHP poles. The system is stable if the eigenloci of L(jω)=Zl−1(jω)Zs(jω)\mathbf{L}(j\omega)=\mathbf{Z}_{l}^{-1}(j\omega)\mathbf{Z}_{s}(j\omega) do not encircle (−1,0)(-1,0).

Forbidden region principle

Instead of judging the stability by counting the encirclements of the (−1,0)(-1,0) point, an alternative way is to guarantee the stability by using the forbidden region principle. The forbidden region is defined as the area in the complex plane where the frequency response of the minor loop gain L(jω)L(j\omega) (or the eigenloci of L(jω)\mathbf{L}(j\omega) in MIMO case) is not allowed to enter. The forbidden region is constructed such that staying outside of it means that no encirclement of the (−1,0)(-1,0) point is possible, which ensures stability under no RHP poles assumption.

Decentralized specifications for converters connected to a single bus

Stability analysis of the system gives a conclusion about the stability of the interconnected system, but it offers limited insight into the design of individual converters. This was addressed by introducing impedance specifications as guidelines that designers should follow [1], [7], [8], [10], [15].

However, impedance specifications presented in [1], [7], [11], [13], like stability analysis, require complete knowledge of the system, and specifications can be given for only one converter while keeping the other ones fixed. This is impractical, as the specifications given to one converter depend on the rest of the grid, and changing it will result in a change of the specifications.

This is especially problematic in a multi-vendor environment, where the converters need to be designed independently, but in such a way that the system stability is not compromised [20], [22], [24]–[26]. This is a motivation for developing decentralized impedance specifications. In this section we will focus on multiple converters connected in parallel to a single bus, as in typical grid-tied applications (Figure 2).1 To emphasize the fact that we are focusing on grid-connected converters in this paper, the source impedance is considered to be known and referred to as grid impedance ZgZ_g, while the load admittance is considered to be the admittance of the converter Yk\mathbf{Y}_k, as shown in Figure 2.

Figure 2. Different converters connected to the single bus - small-signal impedance representation.

The main idea of decentralized specifications is to impose individual conditions on each converter admittance, dependent only on Yk(s)\mathbf{Y}_k(s) and the grid impedance Zg(s)Z_g(s), that ensure system stability.

To formulate the decentralized specifications mathematically, we start from the sufficient stability condition S\mathcal{S} of the system. Then, the decentralized specifications are expressed as nn individual conditions, where each condition Qk\mathcal{Q}_k corresponds to a converter kk. The satisfaction of all individual conditions Qk\mathcal{Q}_k guarantees the stability of the system:

Q1∧Q2∧⋯∧Qk∧⋯∧Qn→S→Stability.(2)\mathcal{Q}_1 \land\mathcal{Q}_2 \land\cdots\land\mathcal{Q}_k \land\cdots\land\mathcal{Q}_n \to\mathcal{S} \to\text{Stability}. \tag*{(2)}

The condition for an individual converter Qk\mathcal{Q}_k is dependent only on the admittance of that converter Yk(s)\mathbf{Y}_k(s) and the grid impedance Zg(s)Z_g(s), and independent of the admittance of other converters. Based on the condition’s Qk\mathcal{Q}_k structure, we differentiate between two types of decentralized specifications:

  • Explicit decentralized specifications, where specifications are given directly to converter’s admittance Yk(s)\mathbf{Y}_k(s).

Qk: f(Yk)<g(Mk,Zg).(3)\mathcal{Q}_k:\ f(\mathbf{Y}_k)<g(M_k,Z_g). \tag*{(3)}
  • Implicit decentralized specifications, where specifications are not given directly to the converter’s admittance, but nevertheless they depend only on one converter admittance at a time.

Qk: f(Yk,Zg)<g(Mk).(4)\mathcal{Q}_k:\ f(\mathbf{Y}_k,Z_g)<g(M_k). \tag*{(4)}

The functions ff and gg, as well as parameter MkM_k, hereafter referred to as scaling factors, must be chosen in a way that guarantees (2) is satisfied. The scaling factor MkM_k can be chosen based on the properties of the converter that are known upfront. A reasonable choice, proposed in [16], is to scale the converters based on their rated power SkS_k, Mk=Sk∑k=1nSkM_k=\frac{S_k}{\sum_{k=1}^{n}S_k}, such that converters rated for more power are allocated a larger “budget”, and this scaling will be used throughout the paper.

The explicit decentralized specifications give more insight into the design of the converter, however they tend to be more conservative. On the other hand, the implicit decentralized specifications are not direct specifications on the converter’s admittance. Nevertheless, they are a valuable tool for systems with a large number of converters, especially when those converters come from multiple vendors. By providing an independent condition for each converter (whether explicit or implicit) decentralized specifications foster a competitive multi-vendor environment for converter designers, while guaranteeing system stability.

Decentralized specifications based on Middlebrook criterion

By utilizing the small-gain theorem, the sufficient stability condition S\mathcal{S} for the whole system can be expressed as:

S: ∀ω, σ‾[Yeq(jω)]<σ‾[Zg(jω)]−1.(5)\mathcal{S}:\ \forall\omega,\ \overline{\sigma}[\mathbf{Y}_{\mathrm{eq}}(j\omega)]<\overline{\sigma}[\mathbf{Z}_g(j\omega)]^{-1}. \tag*{(5)}

By using the fact that

σ‾(Yeq(jω))=σ‾[∑k=1nYk(jω)]≤∑k=1nσ‾[Yk(jω)],(6)\overline{\sigma}(\mathbf{Y}_{\mathrm{eq}}(j\omega))=\overline{\sigma}\left[\sum_{k=1}^{n}\mathbf{Y}_k(j\omega)\right]\leq\sum_{k=1}^{n}\overline{\sigma}[\mathbf{Y}_k(j\omega)], \tag*{(6)}

decentralized specifications can be given in the form of:

Qk: ∀ω, σ‾[Yk(jω)]<Mkσ‾[Zg(jω)].(7)\mathcal{Q}_k:\ \forall\omega,\ \overline{\sigma}[\mathbf{Y}_k(j\omega)]<\frac{M_k}{\overline{\sigma}[\mathbf{Z}_g(j\omega)]}. \tag*{(7)}

where MkM_k is the scaling factor, unique for each converter, such that ∑k=1nMk≤1\sum_{k=1}^{n} M_k \le1, which is an assumed property of MkM_k throughout the paper.

Decentralized specifications based on Passivity criterion

A stable transfer function matrix G(s)\mathbf{G}(s) is passive if its Hermitian part is positive definite at all frequencies, G(jω)+GH(jω)>0\mathbf{G}(j\omega) + \mathbf{G}^{H}(j\omega) > 0, where (⋅)H(\cdot)^{H} is the conjugate transpose [36]. The degree of passivity at frequency ω\omega is quantified by the input feedforward passivity index [13]:

νF[G,ω]=12min⁡λ[G(jω)+GH(jω)].(8)\nu_{F}[\mathbf{G},\omega] = \frac{1}{2}\min\lambda\left[\mathbf{G}(j\omega) + \mathbf{G}^{H}(j\omega)\right]. \tag*{(8)}

A feedback connection of two passive subsystems is stable [36]. The net damping criterion [11], extended to transfer function matrices in [13], relaxes this requirement: a passivity deficit of the converters is allowed as long as it is compensated by the passivity surplus of the grid. With Yg=Zg−1\mathbf{Y}_{g} = \mathbf{Z}_{g}^{-1}, the sufficient stability condition S\mathcal{S} for the whole system is:

S: ∀ω, νF[Yeq,ω]>−νF[Yg,ω]∧νF[Yg,ω]>0.(9)\mathcal{S} : \ \forall\omega,\ \nu_{F}[\mathbf{Y}_{eq},\omega] > -\nu_{F}[\mathbf{Y}_{g},\omega] \land\nu_{F}[\mathbf{Y}_{g},\omega] > 0. \tag*{(9)}

To split this condition into individual conditions for each converter, Qk\mathcal{Q}_k we use the property:

νF[Yeq,ω]=νF[∑k=1nYk,ω]≥∑k=1nνF[Yk,ω],(10)\nu_{F}\left[\mathbf{Y}_{eq},\omega\right] = \nu_{F}\left[\sum_{k=1}^{n}\mathbf{Y}_{k},\omega\right] \ge\sum_{k=1}^{n}\nu_{F}[\mathbf{Y}_{k},\omega], \tag*{(10)}

that follows from Weyl’s inequality for Hermitian matrices, λmin⁡(A+B)≥λmin⁡(A)+λmin⁡(B)\lambda_{\min}(\mathbf{A}+\mathbf{B}) \ge\lambda_{\min}(\mathbf{A})+\lambda_{\min}(\mathbf{B}) [37]. Then, assuming that the grid is passive, the decentralized specifications based on the passivity criterion can be expressed as:

Qk: ∀ω, νF[Yk,ω]>−MkνF[Yg,ω].(11)\mathcal{Q}_k : \ \forall\omega,\ \nu_{F}[\mathbf{Y}_{k},\omega] > -M_k\nu_{F}[\mathbf{Y}_{g},\omega]. \tag*{(11)}

The special case Mk=0M_k = 0 corresponds to requiring every converter to be passive, νF[Yk,ω]>0\nu_{F}[\mathbf{Y}_{k},\omega] > 0.

Decentralized specifications based on Rotated passivity

A further extension is the rotated passivity approach [15], which applies a rotation matrix R\mathbf{R} to the admittances before evaluating the passivity index. While R\mathbf{R} can in principle be chosen arbitrarily, the choice R=[0−110]\mathbf{R} = \begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix} is used in [15] because it maps the inductive grid impedance to a resistive form in the rotated representation, making the criterion applicable in the low-frequency range. This R\mathbf{R} is adopted throughout the rest of this paper. Analogously to (11), the decentralized specifications based on the rotated passivity approach (also studied in [23]) can be expressed as:

Qk: ∀ω, νF[RYk,ω]>−MkνF[RYg,ω]∧νF[RYg,ω]>0.(12)\mathcal{Q}_k : \ \forall\omega,\ \nu_{F}[\mathbf{R}\mathbf{Y}_{k},\omega] > -M_k\nu_{F}[\mathbf{R}\mathbf{Y}_{g},\omega] \land\nu_{F}[\mathbf{R}\mathbf{Y}_{g},\omega] > 0. \tag*{(12)}

Note that a passive grid does not imply that νF[RYg,ω]>0\nu_{F}[\mathbf{R}\mathbf{Y}_{g},\omega] > 0, which should be checked separately when using the rotated passivity approach.

Decentralized specifications based on minor loop gain

The above-mentioned solutions for imposing decentralized specifications in the explicit form, as in (3). As mentioned before, these specifications give more information to converter designers, but they tend to be more conservative. To overcome this problem the decentralized specifications given directly to the minor loop are proposed for a dc bus in [16]. Instead of giving a specification to each converter admittance, the specifications are given for the minor loop gain, defined for each converter kk as:

Lk(s)=Yk(s)Zg(s).(13)L_k(s) = Y_k(s)Z_g(s). \tag*{(13)}

For the dc case, the sufficient stability condition is:

S: ∀ω, Re⁡{L(jω)}=Re⁡{∑k=1nLk(jω)}>−1.(14)\mathcal{S} : \ \forall\omega,\ \operatorname{Re}\{L(j\omega)\} = \operatorname{Re}\left\{\sum_{k=1}^{n}L_k(j\omega)\right\} > -1. \tag*{(14)}

Then, by using Re⁡{z1+z2}=Re⁡{z1}+Re⁡{z2}\operatorname{Re}\{z_1+z_2\} = \operatorname{Re}\{z_1\}+\operatorname{Re}\{z_2\}, the decentralized specifications can be expressed as:

Qk: ∀ω, Re⁡{Yk(jω)Zg(jω)}>−Mk.(15)\mathcal{Q}_k : \ \forall\omega,\ \operatorname{Re}\{Y_k(j\omega)Z_g(j\omega)\} > -M_k. \tag*{(15)}

The form of these specifications is implicit, as in (4).

  1. Decentralized specifications based on Gershgorin theorem:

Extending this idea to the ac case, where the total loop gain L(s)=∑kLk(s)\mathbf{L}(s) = \sum_k L_k(s) is a transfer function matrix with Lk=YkZgL_k = Y_kZ_g, is not straightforward. The challenge is that, for non-Hermitian matrices, the eigenvalues of the sum are not directly related to those of the individual terms:

min⁡Re⁡{λ[∑k=1nLk(jω)]}>−1 ⇏\min\operatorname{Re}\left\{\lambda\left[\sum_{k=1}^{n}\mathbf{L}_k(j\omega)\right]\right\} > -1 \ \nRightarrow
⇏∑k=1nmin⁡Re⁡{λ[Lk(jω)]}>−1.(16)\nRightarrow\sum_{k=1}^{n}\min\operatorname{Re}\left\{\lambda\left[\mathbf{L}_k(j\omega)\right]\right\} > -1. \tag*{(16)}

To deal with this problem, the solution initially proposed in [22], and further explored in this article, uses the Gershgorin theorem. This theorem involves estimating the position of eigenvalues based on individual matrix elements, which enables the definition of decentralized specifications. According to [22], by applying the Gershgorin theorem, the sufficient stability conditions are:

S: ∀ω, Re⁡{Ldd(jω)}−∣Ldq(jω)∣>−1 ∧ Re⁡{Lqq(jω)}−∣Lqd(jω)∣>−1.(17)\mathcal{S}: \ \forall\omega,\ \operatorname{Re}\left\{L^{dd}(j\omega)\right\}-\left|L^{dq}(j\omega)\right|>-1 \ \land\ \operatorname{Re}\left\{L^{qq}(j\omega)\right\}-\left|L^{qd}(j\omega)\right|>-1. \tag*{(17)}

The graphical illustration of the application of the Gershgorin theorem, for imposing specifications at frequency ωc\omega_c is given in Figure 3.

Graphical illustration of Gershgorin theorem at frequency $\omega_c$

Figure 3. Graphical illustration of Gershgorin theorem at frequency ωc\omega_c. According to the Gershgorin theorem, the eigenvalues (λ1\lambda_1, λ2\lambda_2) of L(jωc)\mathbf{L}(j\omega_c) must lie within the two circles [22].

For decentralized assessment, the loop gain is partitioned as:

L(jω)=∑k=1nLk(jω)=[∑Lkdd(jω)∑Lkdq(jω)∑Lkqd(jω)∑Lkqq(jω)].(18)\mathbf{L}(j\omega)=\sum_{k=1}^{n}\mathbf{L}_k(j\omega)=\begin{bmatrix}\sum L_k^{dd}(j\omega)&\sum L_k^{dq}(j\omega)\\\sum L_k^{qd}(j\omega)&\sum L_k^{qq}(j\omega)\end{bmatrix}. \tag*{(18)}

Then according to [22] the decentralized specifications are:

Qk: ∀ω, Re⁡{Lkdd(jω)}−∣Lkdq(jω)∣>−Mkd ∧ Re⁡{Lkqq(jω)}−∣Lkqd(jω)∣>−Mkq,(19)\mathcal{Q}_k:\ \forall\omega,\ \operatorname{Re}\left\{L_k^{dd}(j\omega)\right\}-\left|L_k^{dq}(j\omega)\right|>-M_k^d \ \land\ \operatorname{Re}\left\{L_k^{qq}(j\omega)\right\}-\left|L_k^{qd}(j\omega)\right|>-M_k^q, \tag*{(19)}

where the scaling factors MkdM_k^d and MkqM_k^q must satisfy:

∑k=1nMkd≤1 ∧ ∑k=1nMkq≤1.(20)\sum_{k=1}^{n}M_k^d\leq1 \ \land\ \sum_{k=1}^{n}M_k^q\leq1. \tag*{(20)}

The proposed criterion enables the selection of the scaling factors for the dd and qq channels independently, as long as (20) is satisfied.

Reducing the conservativeness of the Gershgorin based decentralized specifications

The Gershgorin theorem enables us to define well-posed decentralized specifications, but, at the same time, increases the conservativeness of the assessment. This is due to the fact that we are defining the circle where eigenvalues can be located, and considering the worst-case scenario [22, 34]. The conservativeness is especially pronounced when the loop gain has a strong coupling between dd and qq channels (non-diagonally dominant) [22, 34].

A possible way to reduce conservativeness is to apply a similarity transformation Az\mathbf{A}_z (with inverse Az−1\mathbf{A}_z^{-1}) to the loop gain [38]:

Lt(jω)=AzL(jω)Az−1,(21)\mathbf{L}^{t}(j\omega)=\mathbf{A}_z\mathbf{L}(j\omega)\mathbf{A}_z^{-1}, \tag*{(21)}

which preserves eigenvalues [39, 40].

In the context of decentralized specifications, according to (18), the transformed loop gain is given by:

Lt(jω)=∑k=1nAzLk(jω)Az−1.(22)\mathbf{L}^{t}(j\omega)=\sum_{k=1}^{n}\mathbf{A}_z\mathbf{L}_k(j\omega)\mathbf{A}_z^{-1}. \tag*{(22)}

This means that if the same arbitrary transformation Az\mathbf{A}_z is applied to each converter’s minor loop gain Lk(jω)\mathbf{L}_k(j\omega), applying (19) to the newly transformed quantities still guarantees stability. The transformation of the system ensures that eigenvalues are conserved, but at the same time, it changes the evaluation based on Gershgorin theorem since individual elements of L(jω)\mathbf{L}(j\omega) are now changed.

The main challenge is selecting the right transformation that minimizes the conservativeness of the Gershgorin theorem assessment. This is not an easy task, since the transformation must be predetermined and the same for all converters.

A special case of transformation called the modified sequence domain is presented in [39]. This transformation is defined by:

Az=12[1j1−j].(23)\mathbf{A}_z=\frac{1}{\sqrt{2}}\begin{bmatrix}1&j\\1&-j\end{bmatrix}. \tag*{(23)}

One of the notable properties of this transformation is that it maps the symmetric dqdq-frame impedance of balanced three-phase elements to diagonal form in the modified sequence domain. For example, a three-phase inductance LiL_i whose dqdq impedance is

ZLdq(s)=[sLi−ωsLiωsLisLi],(24)\mathbf{Z}_{L}^{dq}(s)=\begin{bmatrix}sL_i&-\omega_sL_i\\\omega_sL_i&sL_i\end{bmatrix}, \tag*{(24)}

becomes diagonal in the modified sequence domain:

ZLpn(s)=AzZLdq(s)Az−1=[(s+jωs)Li00(s−jωs)Li].(25)\mathbf{Z}_{L}^{pn}(s)=\mathbf{A}_z\mathbf{Z}_{L}^{dq}(s)\mathbf{A}_z^{-1}=\begin{bmatrix}(s+j\omega_s)L_i&0\\0&(s-j\omega_s)L_i\end{bmatrix}. \tag*{(25)}

This diagonality reduces the off-diagonal (coupling) terms of the loop gain, which are directly responsible for the conservativeness of the Gershgorin assessment.

In this paper, to distinguish between the analyses in the dqdq domain and the modified sequence domain, the notation for quantities in the modified sequence domain will use the superscript pnpn instead of dqdq. For example, loop gain will be written as:

Lpn(jω)=AzLdq(jω)Az−1=[Lpp(jω)Lpn(jω)Lnp(jω)Lnn(jω)].(26)\mathbf{L}^{pn}(j\omega)=\mathbf{A}_z\mathbf{L}^{dq}(j\omega)\mathbf{A}_z^{-1}=\begin{bmatrix}L^{pp}(j\omega)&L^{pn}(j\omega)\\L^{np}(j\omega)&L^{nn}(j\omega)\end{bmatrix}. \tag*{(26)}

From here, the dual expression to (19) for decentralized specifications in the modified sequence domain is obtained:

Qk: ∀ω, Re⁡{Lkpp(jω)}−∣Lkpn(jω)∣>−Mkp ∧ Re⁡{Lknn(jω)}−∣Lknp(jω)∣>−Mkn,(27)\mathcal{Q}_k:\ \forall\omega,\ \operatorname{Re}\left\{L_k^{pp}(j\omega)\right\}-\left|L_k^{pn}(j\omega)\right|>-M_k^p \ \land\ \operatorname{Re}\left\{L_k^{nn}(j\omega)\right\}-\left|L_k^{np}(j\omega)\right|>-M_k^n, \tag*{(27)}

where the MkpM_k^p and MknM_k^n can be selected independently in the same way as MkdM_k^d and MkqM_k^q with conditions: ∑k=1nMkp≤1 ∧ ∑k=1nMkn≤1\sum_{k=1}^{n}M_k^p\leq1 \ \land\ \sum_{k=1}^{n}M_k^n\leq1.

For a system consisting only of symmetric subsystems, such as RLC branches, each local loop gain in the modified sequence domain is diagonal, i.e., Lkpn=Lknp=0L_k^{pn}=L_k^{np}=0. The Gershgorin circles then have zero radius, the eigenvalue locations are known exactly, and (27) is never more conservative than (19). Converters are generally not symmetric, particularly at low frequencies; nevertheless, the case studies in Section IV show that the conservativeness can still be reduced by using the modified sequence domain. In the high-frequency range, the influence of asymmetric control loops (e.g., PLL and active and reactive power control) diminishes, so the modified sequence domain is generally preferred there.

Combination of different decentralized specifications

Although satisfying one type of decentralized specifications is sufficient for the stability of the system, it is not necessary, and can lead to limited applicability of the specifications. For example, it is well known that the passivity-based specifications are almost impossible to satisfy for low frequencies in realistic scenarios [15]. Because of this, using different types of specifications for different frequency ranges is of interest.

To justify combining different specification types across frequency ranges, we must verify that the combination still constitutes a sufficient condition for stability.

To formalize this, instead of defining the conditions as QkQ_k for the whole frequency range, we need them to be defined for each frequency ω\omega, as Qk(ω)Q_k(\omega). The type of specification is denoted by the superscript θ∈Θ={M,P,RP,dq,pn}\theta\in\Theta= \{\mathrm{M},\mathrm{P},\mathrm{RP},dq,pn\}, so that QkM(ω)Q_k^{\mathrm{M}}(\omega), QkP(ω)Q_k^{\mathrm{P}}(\omega), QkRP(ω)Q_k^{\mathrm{RP}}(\omega), Qkdq(ω)Q_k^{dq}(\omega), and Qkpn(ω)Q_k^{pn}(\omega) are the conditions (7), (11), (12), (19), and (27), respectively, evaluated at the single frequency ω\omega. For example, QkP(ω)Q_k^{\mathrm{P}}(\omega) is defined on frequency ω\omega, based on (11), as:

QkP(ω): νF[Yk,ω]>−MkνF[Yg,ω].(28)Q_k^{\mathrm{P}}(\omega):\ \nu_F[\mathbf{Y}_k,\omega] > -M_k\nu_F[\mathbf{Y}_g,\omega]. \tag*{(28)}

The other frequency dependent conditions are defined in a similar way.

Theorem 1. Assume that the minor loop gain L(s)=Yeq(s)Zg(s)L(s)=Y_{\mathrm{eq}}(s)Z_g(s) has no RHP poles and that ∑k=1nMk≤1\sum_{k=1}^{n}M_k\leq1 holds for every type of specifications. If at each frequency at least one specification type is satisfied by all converters, i.e.,

∀ω, ∃θ∈Θ: Q1θ(ω)∧Q2θ(ω)∧⋯∧Qnθ(ω),(29)\forall\omega,\ \exists\theta\in\Theta:\ Q_1^\theta(\omega)\land Q_2^\theta(\omega)\land\cdots\land Q_n^\theta(\omega), \tag*{(29)}

then the system is stable.

Proof. The proof is done by identifying a common forbidden region shared by all specification types, as illustrated in Figure 4.

Figure 4. Graphical illustration of the forbidden region principle for different types of specifications.

For small-gain specifications it is well known that fulfilling (5) keeps the eigenvalues of L(jω)L(j\omega) inside the unit circle, as illustrated in Figure 4.

Gershgorin based specifications are designed to keep the real part of eigenvalues of L(jω)L(j\omega) greater than −1-1, as discussed in Figure 3.

For passivity-based specifications, the corresponding forbidden region is less direct. It can be obtained via the sectorial decomposition and phase properties of complex matrices presented in [41]. For our case the sectorial decomposition of the Zg(jω)Z_g(j\omega) gives the two angles ϕ1(jω)\phi_1(j\omega) and ϕ2(jω)\phi_2(j\omega), and they are ordered such that ϕ‾(jω)=ϕ1(jω)≥ϕ2(jω)=ϕ‾(jω)\overline{\phi}(j\omega)=\phi_1(j\omega)\geq\phi_2(j\omega)=\underline{\phi}(j\omega) [41]. These angles form a sector in the complex plane (ϕ‾(jω),ϕ‾(jω))(\underline{\phi}(j\omega),\overline{\phi}(j\omega)), and all eigenvalues of Zg(jω)Z_g(j\omega) must lie within this sector. The sectorial decomposition exists when the numerical range of the matrix does not contain the origin, which is always the case for strictly passive matrices [41]. Or in other words, the phase of the eigenvalues of Zg(jω)Z_g(j\omega) must be in between ϕ‾(jω)\underline{\phi}(j\omega) and ϕ‾(jω)\overline{\phi}(j\omega): arg⁡{λ[Zg(jω)]}∈(ϕ‾(jω),ϕ‾(jω))\arg\{\lambda[Z_g(j\omega)]\}\in(\underline{\phi}(j\omega),\overline{\phi}(j\omega)) [41]. For our case we will use the following properties:

  • If we have two complex matrices ZgZ_g, with sector (ϕ2,ϕ1)(\phi_2,\phi_1) and Yeq\mathbf{Y}_{eq}, with sector (ψ2,ψ1)(\psi_2,\psi_1), then the phase of the eigenvalues of L=YeqZg\mathbf{L}=\mathbf{Y}_{eq}Z_g must be located in the sector (ϕ2+ψ2,ϕ1+ψ1)(\phi_2+\psi_2,\phi_1+\psi_1) [41].

  • If matrix ZgZ_g is passive (Zg+ZgH>0⟺νF[Zg]>0)(Z_g+Z_g^H>0\Longleftrightarrow\nu_F[Z_g]>0) then the sector of ZgZ_g must be fully contained in (−π/2,π/2)(-\pi/2,\pi/2) [41].

Note also that passivity is preserved under inversion: if A+AH>0A+A^H>0, then A−1+A−H=A−H(A+AH)A−H>0A^{-1}+A^{-H}=A^{-H}(A+A^H)A^{-H}>0. Hence, νF[Yg,ω]>0\nu_F[\mathbf{Y}_g,\omega]>0 implies νF[Zg,ω]>0\nu_F[\mathbf{Z}_g,\omega]>0.

We first consider the special case Mk=0M_k=0 of (11), where all converters, and therefore Yeq\mathbf{Y}_{eq}, are passive. If we assume that both Zg(jω)Z_g(j\omega) and Yeq(jω)\mathbf{Y}_{eq}(j\omega) are passive, then the eigenvalues of L(jω)=Yeq(jω)Zg(jω)\mathbf{L}(j\omega)=\mathbf{Y}_{eq}(j\omega)Z_g(j\omega) must be located in the sector (−π/2−π/2,π/2+π/2)=(−π,π)(-\pi/2-\pi/2,\pi/2+\pi/2)=(-\pi,\pi). In other words, the eigenvalues of L(jω)\mathbf{L}(j\omega) can’t enter the wedge (π−δ/2,−π+δ/2)(\pi-\delta/2,-\pi+\delta/2) with infinitely small angle δ>0\delta>0 around the negative real axis, as illustrated in Figure 4.

For the general case Mk>0M_k>0 of (11), i.e., net damping specifications, the situation is different, since νF[Yeq,ω]>0\nu_F[\mathbf{Y}_{eq},\omega]>0 is not guaranteed. Then we define:

H(jω)=I+L(jω)=[Yg(jω)+Yeq(jω)]Zg(jω).(30)\mathbf{H}(j\omega)=\mathbf{I}+\mathbf{L}(j\omega)=[\mathbf{Y}_g(j\omega)+\mathbf{Y}_{eq}(j\omega)]\mathbf{Z}_g(j\omega). \tag*{(30)}

Since νF[Yeq,ω]+νF[Yg,ω]>0\nu_F[\mathbf{Y}_{eq},\omega]+\nu_F[\mathbf{Y}_g,\omega]>0 then it is also true that νF[Yg+Yeq,ω]>0\nu_F[\mathbf{Y}_g+\mathbf{Y}_{eq},\omega]>0. And since Zg(jω)\mathbf{Z}_g(j\omega) is passive, the eigenvalues of H(jω)\mathbf{H}(j\omega) can’t be located in the wedge (π−δ/2,−π+δ/2)(\pi-\delta/2,-\pi+\delta/2). Because eigenvalues of L(jω)\mathbf{L}(j\omega) are just eigenvalues of H(jω)\mathbf{H}(j\omega) shifted by −1-1, the eigenvalues of L(jω)\mathbf{L}(j\omega) can’t be located in the wedge (π−δ/2,−π+δ/2)(\pi-\delta/2,-\pi+\delta/2) around the point −1-1 on the real axis, as illustrated in Figure 4.

Rotated passivity specifications can be analyzed in the same way. For invertible matrix R\mathbf{R}, we can transform the loop gain without changing the eigenvalues, as in (21):

L=YeqZg⟶Lt=RYeqZgR−1.(31)\mathbf{L}=\mathbf{Y}_{eq}\mathbf{Z}_g\longrightarrow\mathbf{L}^{t}=\mathbf{R}\mathbf{Y}_{eq}\mathbf{Z}_g\mathbf{R}^{-1}. \tag*{(31)}

Now we analyze:

Ht=I+Lt=[RYg+RYeq]ZgR−1.(32)\mathbf{H}^{t}=\mathbf{I}+\mathbf{L}^{t}=[\mathbf{R}\mathbf{Y}_g+\mathbf{R}\mathbf{Y}_{eq}]\mathbf{Z}_g\mathbf{R}^{-1}. \tag*{(32)}

Since νF[RYeq,ω]+νF[RYg,ω]>0\nu_F[\mathbf{R}\mathbf{Y}_{eq},\omega]+\nu_F[\mathbf{R}\mathbf{Y}_g,\omega]>0, and νF[ZgR−1,ω]>0\nu_F[\mathbf{Z}_g\mathbf{R}^{-1},\omega]>0 (which follows from νF[RYg,ω]>0\nu_F[\mathbf{R}\mathbf{Y}_g,\omega]>0 and the inversion property above, because ZgR−1=(RYg)−1\mathbf{Z}_g\mathbf{R}^{-1} = (\mathbf{R}\mathbf{Y}_g)^{-1}, the same forbidden region applies as for net damping criterion.

Since all specification types share a common forbidden region, the part of the negative real axis to the left of −1-1, (−∞,−1](-\infty,-1] (Figure 4), if (29) holds the eigenloci of L(jω)\mathbf{L}(j\omega) do not enter this region at any frequency. The eigenloci form closed curves, and the region (−∞,−1](-\infty,-1] connects the point (−1,0)(-1,0) to infinity without crossing them; therefore, the eigenloci can’t encircle (−1,0)(-1,0). Since L(s)L(s) has no RHP poles, the system is stable according to the generalized Nyquist criterion. ■

It is important to note that stability is not guaranteed if different specification types are assigned to different converters at the same frequency (e.g., converter 1 satisfies passivity while converter 2 satisfies the Gershgorin-based specification):

[Q1dq(ω)∨Q1P(ω)]∧⋯∧[Qndq(ω)∨QnP(ω)]⇏Stability.(33)\left[\mathcal{Q}^{dq}_1(\omega)\lor\mathcal{Q}^{P}_1(\omega)\right]\land\cdots\land\left[\mathcal{Q}^{dq}_n(\omega)\lor\mathcal{Q}^{P}_n(\omega)\right]\not\Rightarrow\text{Stability}. \tag*{(33)}

A counterexample for this claim is given in the Appendix. Because of this, the frequency range where a certain type of decentralized specification is satisfied should be provided upfront, as a part of the provided specifications. For example, combined decentralized specifications can be given as: all converters should satisfy Middlebrook specifications (7) for frequencies lower than ωt\omega_t, while for frequencies higher than ωt\omega_t they should satisfy passivity-based specifications (11). Other combinations will be presented in subsequent sections.

Characterizing the conservativeness of different methods for decentralized specifications

So far, several methods for imposing decentralized specifications on converters connected to the ac bus have been presented. All of these methods are frequency-domain conditions. To compare their conservativeness, for each method and converter kk a parameter Dk(ω)D_k(\omega) is defined, with Dk(ω)>0D_k(\omega)>0 indicating that the specifications are satisfied and Dk(ω)<0D_k(\omega)<0 that they are violated. Since the specifications have different mathematical forms, only the sign of Dk(ω)D_k(\omega) is meaningful, and its magnitude cannot be compared across methods; the conservativeness is therefore assessed by evaluating whether the specifications are satisfied over a range of converter parameters, as done in Section IV. For each type of specification, the parameter Dk(ω)D_k(\omega) is defined as follows.

Middlebrook (small-gain) based specifications (7):

DkM(ω)=Mkσ‾[Zg(jω)]−σ‾[Yk(jω)].(34)D_k^{M}(\omega)=\frac{M_k}{\overline{\sigma}[\mathbf{Z}_g(j\omega)]}-\overline{\sigma}[\mathbf{Y}_k(j\omega)]. \tag*{(34)}

Passivity based specifications (11):

DkP(ω)=νF[Yk,ω]+MkνF[Yg,ω].(35)D_k^{P}(\omega)=\nu_F[\mathbf{Y}_k,\omega]+M_k\nu_F[\mathbf{Y}_g,\omega]. \tag*{(35)}

Rotated passivity based specifications (12):

DkRP(ω)=νF[RYk,ω]+MkνF[RYg,ω].(36)D_k^{RP}(\omega)=\nu_F[\mathbf{R}\mathbf{Y}_k,\omega]+M_k\nu_F[\mathbf{R}\mathbf{Y}_g,\omega]. \tag*{(36)}

Gershgorin based specifications in dqdq (19):

Dkd(ω)=Re⁡{Lkdd(jω)}−∣Lkdq(jω)∣+MkdDkq(ω)=Re⁡{Lkqq(jω)}−∣Lkqd(jω)∣+MkqDkdq(ω)=min⁡[Dkd(ω),Dkq(ω)].(37)\begin{aligned} D_k^{d}(\omega)&=\operatorname{Re}\left\{L_k^{dd}(j\omega)\right\}-\left|L_k^{dq}(j\omega)\right|+M_k^{d}\\ D_k^{q}(\omega)&=\operatorname{Re}\left\{L_k^{qq}(j\omega)\right\}-\left|L_k^{qd}(j\omega)\right|+M_k^{q}\\ D_k^{dq}(\omega)&=\min\left[D_k^{d}(\omega),D_k^{q}(\omega)\right]. \tag*{(37)} \end{aligned}

Gershgorin based specifications in modified sequence domain pnpn (27):

Dkp(ω)=Re⁡{Lkpp(jω)}−∣Lkpn(jω)∣+MkpDkn(ω)=Re⁡{Lknn(jω)}−∣Lknp(jω)∣+MknDkpn(ω)=min⁡[Dkp(ω),Dkn(ω)].(38)\begin{aligned} D_k^{p}(\omega)&=\operatorname{Re}\left\{L_k^{pp}(j\omega)\right\}-\left|L_k^{pn}(j\omega)\right|+M_k^{p}\\ D_k^{n}(\omega)&=\operatorname{Re}\left\{L_k^{nn}(j\omega)\right\}-\left|L_k^{np}(j\omega)\right|+M_k^{n}\\ D_k^{pn}(\omega)&=\min\left[D_k^{p}(\omega),D_k^{n}(\omega)\right]. \tag*{(38)} \end{aligned}

Examples

Rectifiers connected to the inductive grid

To illustrate the proposed decentralized specifications for ac systems, we consider the symmetric ac grid in Figure 5, with per-phase impedance Lg=0.24L_g=0.24 p.u. and RLg=0.06R_{Lg}=0.06 p.u.. Two two-level, three-phase voltage-source converters are planned for connection to the grid at the PCC.

Figure 5. Schematic of the application scenario that shows the two converters planned for connection to the inductive grid (example A).

Both converters operate as rectifiers with synchronous-reference-frame control, as shown in Figure 7. The parameters for case I and case II are listed in Table I, and the analytical admittance model is taken from [5]. The power-based scaling factors are chosen as M1=M1d=M1q=M1p=M1n=2/3M_1=M_1^{d}=M_1^{q}=M_1^{p}=M_1^{n}=2/3 and M2=M2d=M2q=M2p=M2n=1/3M_2=M_2^{d}=M_2^{q}=M_2^{p}=M_2^{n}=1/3.

Figure 7. Control structure of the grid-following converters presented in Figure 5 and 6. dc-link voltage loop (light blue), current loop (tan), PLL (pink).

In Figure 8 the decentralized specifications for converters 1 and 2 are compared for case I. The Gershgorin-based criteria in the dqdq and modified sequence domains (DkdqD_k^{dq}, DkpnD_k^{pn}) are satisfied over the entire frequency range, guaranteeing stability, whereas the small-gain, passivity, and rotated-passivity specifications (DkMD_k^{M}, DkPD_k^{P}, DkRPD_k^{RP}) are violated in some frequency ranges and are therefore more conservative in this scenario. Passivity-based specifications DkPD_k^{P} are violated at low frequencies, while small-gain and rotated-passivity specifications DkMD_k^{M}, DkRPD_k^{RP} are violated at higher frequencies. Rotated passivity DkRPD_k^{RP} is satisfied below 50 Hz but exhibits a sharp drop to negative values around 50 Hz, a characteristic behavior also reported in [15]. This complementary behavior motivates combining different specifications across frequency ranges, to obtain specifications that can be satisfied over the entire frequency range. For the passivity–small-gain combination we define

Figure 8. Comparison of different decentralized specifications for converter 1 and 2 from Figure 5, example A, case I.

DkMP(ω)={DkP(ω),ω≥ωl,DkM(ω),ω<ωl,(39)D_k^{MP}(\omega)= \begin{cases} D_k^P(\omega), & \omega\ge\omega_l,\\ D_k^M(\omega), & \omega< \omega_l, \end{cases} \tag*{(39)}

and analogously for the passivity–rotated-passivity combination, DkRPP(ω)D_k^{RPP}(\omega), where DkM(ω)D_k^M(\omega) is replaced by DkRP(ω)D_k^{RP}(\omega).

The choice of ωl\omega_l differs between the two combinations. For DkRPP(ω)D_k^{RPP}(\omega), the sharp drop of DkRP(ω)D_k^{RP}(\omega) to negative values around 5050 Hz provides a natural upper bound, and we select ωl=47\omega_l=47 Hz to avoid numerical issues near 5050 Hz. This also exposes a fundamental limitation of the rotated passivity approach: it cannot be used for converters whose bandwidth renders them non-passive near 5050 Hz, where DkRP(ω)D_k^{RP}(\omega) is no longer satisfied. For DkMP(ω)D_k^{MP}(\omega), the choice is less constrained, since the passivity and small-gain specifications do not exhibit a distinctive feature around 5050 Hz. The value of ωl\omega_l significantly affects conservativeness: for converters with slow dynamics, a lower ωl\omega_l is appropriate, since passivity is then typically satisfied; for faster converters, ωl\omega_l must be higher to avoid relying on passivity at frequencies where it is violated. However, increasing ωl\omega_l becomes problematic in weak grids, where the small-gain condition is harder to satisfy. To illustrate

this sensitivity, we compare two choices: ωl=50\omega_l=50 Hz (denoted DMP50D^{MP50}) and ωl=60\omega_l=60 Hz (denoted DMP60D^{MP60}), with ωl=47\omega_l=47 Hz also used for DRPPD^{RPP}.

To compare the conservativeness of the different specifications, their satisfaction is evaluated over a parametric space in which the DVC and PLL bandwidths and the corresponding Ki/KpK_i/K_p ratios are varied, as shown in Figure 9. The compared specifications are the Gershgorin-based criteria in the dqdq and modified sequence domains (DdqdqD_{dq}^{dq}, DpnD^{pn}), the passivity–small-gain combinations with ωl=50\omega_l=50 Hz and ωl=60\omega_l=60 Hz (DMP50D^{MP50}, DMP60D^{MP60}), and the passivity–rotated-passivity combination (DRPPD^{RPP}). For both converters, the modified-sequence Gershgorin specification and DRPPD^{RPP} are the least conservative. The DRPPD^{RPP} combination is less conservative when the dominant interaction occurs at low frequencies, well below 5050 Hz, whereas DpnD^{pn} is less conservative when the interaction shifts toward 5050 Hz. For the passivity–small-gain combinations, increasing ωl\omega_l from 5050 Hz to 6060 Hz extends the feasible region toward faster dynamics in converter 1, but drastically reduces it to only small areas of the parametric space in converter 2, illustrating that an ill-chosen ωl\omega_l can severely limit the admissible controller parameter set.

Comparison of satisfaction of decentralized specifications for converter 1Comparison of satisfaction of decentralized specifications for converter 2

Figure 9. Comparison of satisfaction of decentralized specifications for converters 1 and 2, from Figure 5 example A, for different methods.

Finally, stability is verified using the generalized Nyquist criterion applied to three minor-loop gains: L1=Y1Zg\mathbf{L}_1=\mathbf{Y}_1\mathbf{Z}_g (converter 1 only), L2=Y2Zg\mathbf{L}_2=\mathbf{Y}_2\mathbf{Z}_g (converter 2 only), and L=(Y1+Y2)Zg\mathbf{L}=(\mathbf{Y}_1+\mathbf{Y}_2)\mathbf{Z}_g (both converters). Figure 10 shows the results for case I (parameters from Table I) and case II (increased DVC and PLL bandwidths and modified Ki/KpK_i/K_p ratios): the system is stable in case I and unstable in case II when both converters are connected.

Eigenloci of $\mathbf{L}(j\omega)$

Figure 10. Eigenloci of L(jω)\mathbf{L}(j\omega) (ω>0\omega>0), case I and case II, example A

To demonstrate that the specifications can be applied directly to black-box measurements, a Simulink model is built in which the converter admittances are identified by injecting small-signal perturbations at the output voltage and measuring the resulting input currents, with the converters connected to an ideal grid. The measured frequency responses are then substituted into (38) to evaluate the decentralized specifications. Figure 12 shows the modified-sequence Gershgorin specifications for cases I and II, which are satisfied over the entire frequency range in case I, and violated around 35 Hz in case II. This is consistent with the analytical results.

Decentralized specifications for converter 1 and 2

Figure 12. Decentralized specifications for converter 1 and 2, example A, based on Gershgorin theorem in modified sequence domain, for case I and II.

The time-domain simulation results are presented in Figure 11, where the controller parameters are switched from case I to case II at 0.5 s. The system becomes unstable after the switch, with oscillations at approximately 35 Hz, consistent with the frequency range of specification violation in Figure 12.

Phase current going to grid

Figure 11. Phase current going to grid iPCCi_{PCC} (p.u.), simulation, example A.

Grid-feeding inverters connected to the grid with resonance

In this example, two grid-feeding inverters are connected to a grid with resonance, as shown in Figure 6, with grid inductance Lg=0.15L_g = 0.15 p.u., capacitance Cg=0.05C_g = 0.05 p.u., and capacitor series resistance RCg=0.1R_{C_g} = 0.1 p.u.. The converter parameters are given in Table II, and the scaling factors MkM_k are power-based, identical to those in example A.

Figure 6. Schematic of the application scenario that shows the two converters planned for connection to the grid with resonance (example B).

ParameterConverter 1Converter 2
Steady-state active phase power P0P_0−2/3-2/3 p.u.−1/3-1/3 p.u.
Steady-state reactive power Q0Q_00 p.u.0 p.u.
Steady-state RMS phase current−0.38-0.38 p.u.−0.19-0.19 p.u.
Steady-state dd voltage EdE_d1 p.u.1 p.u.
Steady-state qq voltage EqE_q0 p.u.0 p.u.
Dc link capacitance CdcC_{dc}1.6 p.u.1.6 p.u.
Dc link voltage EdcE_{dc}2 p.u.2 p.u.
Filter inductance LfL_f0.03 p.u.0.06 p.u.
Filter resistance RfR_f0.005 p.u.0.005 p.u.
Current loop bandwidth1 kHz1 kHz
Zero position in current PI20 Hz20 Hz
Delay present in the control loop{100;200} μ\{100; 200\}\ \mus{100;200} μ\{100; 200\}\ \mus
DVC loop bandwidth20 Hz20 Hz
Ratio of DVC PI controller Ki/KpK_i/K_p125125
PLL loop bandwidth20 Hz20 Hz
Ratio of PLL PI controller Ki/KpK_i/K_p125125

Table II. Parameters of Example B, for the {case I; case II}.

Figure 13 shows that only the modified-sequence Gershgorin specification is satisfied over the entire frequency range, while all other specifications are violated in some ranges. Notably, passivity is satisfied precisely where the small-gain and dqdq Gershgorin specifications fail, motivating a combination of the two to achieve full-range coverage. Rotated passivity is not considered here, as the interactions of interest occur above 50 Hz, where DkRPD_k^{RP} is not satisfied.

Comparison of different decentralized specifications for converter 1 and 2

Figure 13. Comparison of different decentralized specifications for converter 1 and 2 from Figure 6, example B, case I.

To compare the conservativeness of the different methods, specification satisfaction is evaluated in a parametric space of current-loop bandwidth and control delay, as shown in Figure 14. The compared specifications are the modified-sequence Gershgorin criterion DpnPD^{pnP}, the passivity–small-gain combination DMPD^{MP}, and the passivity–Gershgorin combination DpnPD^{pnP}, defined as:

Figure 14. Comparison of satisfaction of decentralized specifications for converter 1 and 2, from Figure 6 example B, for different methods.

DkpnP(ω)={DkP(ω),ω∈[50 Hz, 2 kHz],Dkpn(ω),ω∈(0,50 Hz)∪(2 kHz,∞),(40)D_k^{pnP}(\omega) = \begin{cases} D_k^P(\omega), & \omega\in[50\ \mathrm{Hz},\,2\ \mathrm{kHz}], \\ D_k^{pn}(\omega), & \omega\in(0,50\ \mathrm{Hz}) \cup(2\ \mathrm{kHz},\infty), \end{cases} \tag*{(40)}

where passivity governs the mid-frequency range where the small-gain and Gershgorin dqdq specifications are violated.

From Figure 14, the standalone modified-sequence Gershgorin specification DpnD^{pn} is satisfied only for a limited parameter set. When combined with passivity, both DMPD^{MP} and DpnPD^{pnP} yield comparable conservativeness, indicating that the small-gain and modified-sequence Gershgorin bases perform similarly in this scenario. Analogous results hold for the dqdq Gershgorin criterion but are omitted from the figure for readability.

Applying the generalized Nyquist criterion to the system in Figure 6 with case I parameters confirms stability, as there are no encirclements of the (−1,0)(-1,0) point (Figure 15), consistent with the satisfied specifications in Figure 14. Increasing the control delay to 200 μs\mu\mathrm{s} (case II) renders the system unstable when both converters are connected, as confirmed by the Nyquist plot. Since instability occurs almost immediately upon leaving the specification-satisfied region, the combined Gershgorin–passivity specifications introduce negligible conservativeness in this example.

Eigenloci of $L(j\omega)$ for case I and case II, example B

Figure 15. Eigenloci of L(jω)L(j\omega) for case I and case II, example B.

As in example A, a Simulink model is used for small-signal measurements of converter admittance and grid impedance. The combined Gershgorin–passivity specification (40) is then evaluated from the measurements and shown for converter 1 and 2 in Figure 17: satisfied over the entire frequency range in case I, and violated around 1.5 kHz in case II.

Decentralized specifications for converter 1 and 2, example B

Figure 17. Decentralized specifications for converter 1 and 2, example B, based on combination of Gershgorin theorem in modified sequence domain and passivity, DkpnPD_k^{pnP}, for case I and II.

The time-domain simulation results are shown in Figure 16, where the controller parameters are switched from case I to case II at 0.5 s. The system becomes unstable after the switch, with oscillations at approximately 1.5 kHz, consistent with the frequency range of specification violation in Figure 17.

Phase current going to the grid $i_{\mathrm{PCC}}$ (p.u.), simulation, example B

Figure 16. Phase current going to the grid iPCCi_{\mathrm{PCC}} (p.u.), simulation, example B.

Grid-forming converters connected to the inductive grid

Up to now, the proposed specifications have been tested on grid-following converters. In this example, two grid-forming converters with active power synchronization, voltage magnitude control, droop on reactive power for parallel operation, and virtual-admittance control structure shown in Figure 18 are connected to the PCC of an inductive grid with inductance Lg=0.1L_g = 0.1 p.u. and resistance RLg=0.02R_{L_g} = 0.02 p.u., like in Figure 5.

Control structure of the grid-forming converters, example C

Figure 18. Control structure of the grid-forming converters, example C.

The converter parameters are listed in Table III, where all per-unit values refer to the global base, i.e., the total rated power of both converters; case II triples the voltage-loop and power-loop bandwidths of both converters. The scaling factors MkM_k are power-based, identical to those in example A.

Parameter*Converter 1Converter 2
Rated power SkS_k2/3 p.u.1/3 p.u.
Steady-state active power P0P_00.56 p.u.0.28 p.u.
Steady-state reactive power Q0Q_00 p.u.0 p.u.
Filter inductance LfL_f0.075 p.u.0.15 p.u.
Filter resistance RfR_f0.0075 p.u.0.015 p.u.
Virtual inductance LvL_v0.705 p.u.1.41 p.u.
Virtual resistance RvR_v0.465 p.u.0.93 p.u.
Current loop bandwidth500 Hz500 Hz
Voltage loop bandwidth{0.7;2.1}\{0.7; 2.1\} Hz{1;3}\{1; 3\} Hz
QQ–VV droop mqm_q0.18 p.u.0.3 p.u.
Transient damping resistance Ra,vcR_{a,vc}0.63 p.u.1.5 p.u.
Transient damping high-pass filter5.3 Hz4 Hz
Power loop bandwidth{9.5;28.5}\{9.5; 28.5\} Hz{9.5;28.5}\{9.5; 28.5\} Hz
Active damping ratio Ra/Kp,pcR_a/K_{p,pc}1.11.4

Table III. Parameters of Example C, for the {Case I; Case II}

Figure 19 compares the decentralized specifications for case I. The modified-sequence Gershgorin specification DkpmD_k^{pm} is satisfied over the entire frequency range for both converters, whereas DkdqD_k^{dq} is violated for converter 2 around 8 Hz. In contrast to the grid-following converters, the grid-forming converters are non-passive at low frequencies, below approximately 9 Hz, where the power loop is active, while small gain is violated above approximately 230 Hz, and additionally around 8 Hz for converter 2. Rotated passivity is satisfied below 50 Hz, which covers the frequency range where passivity fails; hence, the combination DkRPPD_k^{RPP} is satisfied for both converters, while DkMP50D_k^{MP50} is violated for converter 2.

Figure 19. Comparison of different decentralized specifications for grid-forming converters 1 and 2, example C, case I.

The conservativeness of the specifications is compared in a parametric space of voltage-loop and power-loop bandwidths, shown in Figure 20. As in example A, DpnD^{pn} and DRPPD^{RPP} are the least conservative and yield regions of comparable size, followed by DdqD^{dq}, while DMP50D^{MP50} is the most conservative, since small gain must cover the low-frequency range where the power-loop dynamics make the converters non-passive. The case I parameters lie inside the regions where DpnD^{pn} and DRPPD^{RPP} are satisfied for both converters, whereas the case II parameters lie outside all regions.

Comparison of satisfaction of decentralized specifications for grid-forming converters 1 and 2

Figure 20. Comparison of satisfaction of decentralized specifications for grid-forming converters 1 and 2, example C, for different methods. For each map, only the loops of the corresponding converter are varied.

The generalized Nyquist criterion in Figure 21 confirms that the system is stable in case I. In case II, each converter is stable when connected alone (L1L_1 and L2L_2 do not encircle (−1,0)(-1,0)), but the system with both converters is unstable, with eigenloci of LL encircling (−1,0)(-1,0) around 23 Hz. Thus, the instability results from the interaction of the two converters, and it is captured by the decentralized specifications, even though each converter alone would pass a conventional single-converter stability assessment.

Figure 21. Eigenloci of L(jω)L(\mathrm{j}\omega) for case I and case II, example C.

As in the previous examples, the converter admittances are identified from small-signal measurements in Simulink, and the modified-sequence Gershgorin specifications are evaluated from the measurements, as shown in Figure 23. The specifications are satisfied over the entire frequency range in case I and violated approximately between 20 Hz and 30 Hz for both converters in case II, in agreement with the analytical model.

Figure 23. Decentralized specifications for grid-forming converters 1 and 2, example C, based on Gershgorin theorem in modified sequence domain, for case I and II.

The time-domain simulation results are shown in Figure 22, where the controller parameters are switched from case I to case II at 1 s. The system becomes unstable after the switch, with oscillations at approximately 23 Hz, consistent with the frequency range of specification violation in Figure 23.

Phase current going to the grid

Figure 22. Phase current going to the grid iPCCi_{PCC} (p.u.), simulation, example C.

Mixed grid-forming and grid-following converters connected to the inductive grid

Since the specifications depend only on the frequency response of each converter, they apply equally to systems combining different control strategies. To illustrate this, converter 2 in example C is replaced by the grid-following converter 2 from example A. In case II, only the grid-forming converter 1 is switched to the case II parameters from Table III, while the grid-following converter 2 is left unchanged.

As shown in Figure 24, the specifications are satisfied for both converters in case I, whereas in case II they are violated around 23 Hz by converter 1 only. This single violation is sufficient to destabilize the system, as confirmed by the eigenloci of LL encircling (−1,0)(-1,0), even though converter 1 alone remains stable.

Figure 24. Mixed grid-forming converter 1 and grid-following converter 2, cases I and II: decentralized specifications in the modified sequence domain (left) and eigenloci of L(jω)L(\mathrm{j}\omega) (right).

Experimental Results

So far, the specifications have been applied to analytical models and simulation measurements. One of the key benefits of decentralized specifications is that, since they require only the frequency response of each individual converter, they can be evaluated directly from measurements on real devices, without a converter analytical or simulation model. To demonstrate this, example A is implemented on the experimental setup shown in Figure 25, consisting of two three-phase Imperix TPI 8032 inverters connected to a Cinergia grid emulator through an inductive impedance, with each inverter dc side supplied by a Keysight power supply in constant-current mode. The system parameters match Table I, with base values of 100 V (voltage) and 3 kW (power). The inverters switch at 50 kHz, with the control executed once per switching period.

ParameterConverter 1Converter 2
Steady-state active power P0P_02/32/3 p.u.1/31/3 p.u.
Steady-state reactive power Q0Q_000 p.u.00 p.u.
Steady-state RMS phase current0.380.38 p.u.0.190.19 p.u.
Steady-state dd voltage EdE_d11 p.u.11 p.u.
Steady-state qq voltage EqE_q00 p.u.00 p.u.
Dc link capacitance CdcC_{dc}1.61.6 p.u.1.61.6 p.u.
Filter inductance LfL_f0.090.09 p.u.0.090.09 p.u.
Filter resistance RfR_f0.0360.036 p.u.0.0360.036 p.u.
Current loop bandwidth450450 Hz700700 Hz
Zero position in current PI55 Hz55 Hz
DVC loop bandwidth{5;30}\{5;30\} Hz{10;30}\{10;30\} Hz
Ki/KpK_i/K_p of DVC PI controller{31;450}\{31;450\}{63;450}\{63;450\}
PLL loop bandwidth{5;30}\{5;30\} Hz{10;30}\{10;30\} Hz
Ki/KpK_i/K_p of PLL PI controller{31;450}\{31;450\}{63;450}\{63;450\}

Table I. Parameters of example A, for the {Case I; Case II}.

Figure 25. Experimental setup.

The converter admittance is identified using an impedance measurement unit (IMU) implemented on an additional Imperix TPI 8032 inverter, whose dc side is connected to a constant-voltage source and whose ac output modulates a small-signal perturbation on top of the steady-state fundamental, as shown in Figure 26. An EMI filter suppresses switching noise. With the system in steady state, a perturbation at frequency ωp\omega_p is injected, the phase voltages vpv_p and currents ipi_p are measured, transformed to the dqdq frame, and the admittance Y(jωp)Y(j\omega_p) is obtained via Fourier analysis. This procedure is repeated for both converters in cases I and II.

Figure 26. Schematic representing the procedure for measuring the converter admittance.

The decentralized specifications are then evaluated from the measured admittances, and the results are shown in Figure 27: the modified-sequence Gershgorin specifications are satisfied over the entire frequency range in case I and violated around 35 Hz in case II, in agreement with both the analytical model and simulation. Time-domain experimental results are presented in Figure 28 (converter 1 internal signals) and Figure 29 (oscilloscope waveforms). Switching the parameters from case I to case II causes instability with oscillations at approximately 35 Hz, confirming that specification violation in a given frequency range can lead to system instability.

Experimental results of specifications in the modified sequence domain for case I and II, example A

Figure 27. Experimental results of specifications in the modified sequence domain for case I and II, example A.

Figure 28. Experimental results, idi_d and iqi_q from converter 1, case I and II, example A.

Change of the parameters from case I to case II for phase voltages and currents of converter 1, results from oscilloscope

Figure 29. Change of the parameters from case I to case II for phase voltages and currents of converter 1, results from oscilloscope.

Discussion

Across all examples, covering GFL converters, GFM converters, and their combination, as well as sub-synchronous and harmonic-range interactions, two observations hold. First,

the specifications do not lead to an overly conservative design; some conservativeness is inherent, since each converter’s specification is independent of the others, but this is the price of decentralized design. Second, when the specifications are violated by changing the converter parameters, the system can become unstable. In the mixed example of Section IV-D, a violation by a single converter, which is itself stable when connected alone, can also lead to system instability.

The results presented in Sections IV and § 4 also show that the most suitable type of decentralized specification depends on the frequency range of interest. The frequency ranges can be broadly classified as follows.

  • Low-frequency subsynchronous range (up to tens of Hz): In this range converters fulfill their primary function of supplying active and reactive power, and are typically not passive; net-damping specifications are therefore almost inapplicable. Small-gain, Gershgorin-based (dqdq and modified sequence domain), and rotated-passivity spec-

ifications are the most suitable. In the tested scenarios, the least conservative specifications are those based on rotated passivity, followed by the Gershgorin criterion in the modified sequence domain and then in the dqdq domain. Small-gain-based specifications are also applicable, but tend to be more conservative.

  • Synchronous range (frequencies around the fundamental frequency): The dd–qq coupling is pronounced in this range. For converters with sufficiently slow dynamics, which is typical in power-system applications, passivity-based specifications are applicable; for converters with higher control bandwidths, passivity cannot be used. Rotated passivity is inapplicable in this range because DkRP(ω)D_k^{RP}(\omega) drops sharply to negative values around the fundamental frequency, as discussed in Section IV-A. Gershgorin-based specifications are well-suited here, with the modified-sequence-domain variant being less conservative owing to its inherent handling of dd–qq coupling. The small-gain theorem is also applicable but generally more conservative.

  • Harmonic range (from hundreds of Hz to a few kHz): Converters are not expected to exchange active or reactive power in this range and should be passive. Passivity-based specifications are therefore the most suitable and robust choice, as demonstrated in example B. In scenarios without undamped resonances, Gershgorin-based specifications can also be applied, though passivity-based specifications are preferable in general.

  • High-frequency range (above a few kHz): This range exceeds the control bandwidth of the converter, where modulation, delays, and other non-idealities may cause the converter to be non-passive. When EMI filters render the converter passive at these frequencies, passivity-based specifications apply; otherwise, Gershgorin-based or small-gain specifications may be used. Note that this range is not covered by the examples presented and is included here for completeness; it is generally of greater relevance to single-converter design than to multi-converter grid stability.

This classification (except for the high-frequency range, which is not of interest for this paper) is summarized in Table IV.

MethodLow-frequency subsynchronous rangeSynchronous rangeHarmonic range
Small-gain theoremUsableUsable, but conservativeNot applicable
Gershgorin theorem in dqdq domainGoodGoodApplicable only without resonances
Gersh. in modified sequence domainVery goodVery goodApplicable if resonances are damped
Passivity-basedNot applicableVery good - in most casesExcellent
Rotated passivity-basedExcellentNot applicableNot applicable

Table IV. Summary of the applicability of different methods for different frequency ranges.

Specification providers can therefore tailor the specifications to the grid the converters connect to. For mainly inductive grids without many undamped resonances, Gershgorin-based specifications in the modified sequence domain suffice over the whole frequency range. When harmonic-range issues are expected, passivity-based specifications are preferable. Since passivity must be combined with other specifications, the frequency range in which the converter must be passive should be stated. Its lower bound depends on the desired control speed and typically lies in the 10–40 Hz range; below it, rotated-passivity or Gershgorin-based specifications apply. If control delays are expected to cause active behavior at higher frequencies, an upper bound should also be set (as in example B), with small-gain or Gershgorin-based specifications above it.

Finally, since the specifications are based on small-signal impedances, they hold only at the operating point at which the impedance is obtained. Connecting new converters may shift the operating point of the existing ones (e.g., the PCC voltage or the power flow), so each converter should satisfy its specification over its whole operating range. As this range is known upfront from the converter ratings and grid-code requirements, compliance can be verified analytically or by impedance measurements at a set of representative operating points, such as the corners of the operating range.

Conclusions

This paper presents a systematic treatment of decentralized impedance specifications for power electronics converters connected to a common ac bus. The following contributions are made. First, decentralized specifications based on the small-gain theorem, passivity, and the Gershgorin theorem are unified in a single framework, with simple closed-form frequency-domain expressions provided for each method. As a novel extension of the Gershgorin-based approach, decentralized specifications formulated in the modified sequence domain are proposed, which reduce the conservativeness introduced by the Gershgorin theorem in the dqdq domain by exploiting the near-diagonal structure of symmetric impedances in the modified sequence domain. Second, a general framework for combining different types of decentralized specifications across distinct frequency ranges is introduced. Using the forbidden region principle, it is proved that the combined specifications remain a sufficient condition for system stability, provided that, at each frequency, the same type of specification is satisfied by all converters. Assigning different specification types to different converters at the same frequency does not guarantee stability, as shown by the counterexample in the Appendix. Third, the conservativeness and applicability of the different methods are systematically compared on three examples: sub-synchronous interactions of GFL rectifiers connected to an inductive grid (example A), harmonic-range interactions of GFL inverters connected to a resonant grid (example B), and sub-synchronous interactions of GFM converters connected to an inductive grid (example C); their applicability to systems combining different control strategies is further demonstrated on a mixed GFM–GFL system. Based on these results, practical guidelines are provided for selecting the most appropriate specification type depending on the frequency range and scenario: rotated-passivity and Gershgorin-based methods are preferred at sub-synchronous frequencies, Gershgorin-based methods are most suitable around the fundamental frequency, and passivity-based specifications are the natural choice in the harmonic range. Finally, the specifications are applied directly to black-box frequency-response measurements obtained from both simulation and a laboratory experimental setup, without relying on analytical converter models. It is consistently shown that the proposed specifications are a valuable tool for preventing unwanted interactions and ensuring stability in multi-converter ac systems.

Appendix

This appendix provides a counterexample showing that, at a single frequency ωc\omega_c, combining different types of decentralized specifications across converters does not necessarily guarantee stability, as discussed in (33). Since we focus on a single frequency, only the impedance and admittance values at ωc\omega_c are considered.

Consider two converters of equal nominal power, so that the power-based scaling factors are M1=M2=0.5M_1=M_2=0.5 and ∑kMk=1\sum_k M_k=1. The same values are used for all specification types, i.e., Mkd=Mkq=MkM_k^d=M_k^q=M_k. Let the admittances of the converters and the grid impedance at ωc\omega_c be given by:

Y1=[−2j001],Y2=[−1+j001],Zg=[1−j001](41)\mathbf{Y}_1= \begin{bmatrix} -2j & 0\\ 0 & 1 \end{bmatrix},\quad \mathbf{Y}_2= \begin{bmatrix} -1+j & 0\\ 0 & 1 \end{bmatrix},\quad \mathbf{Z}_g= \begin{bmatrix} 1-j & 0\\ 0 & 1 \end{bmatrix} \tag*{(41)}

The grid is passive, since Yg=Zg−1=diag⁡(0.5+0.5j, 1)\mathbf{Y}_g=\mathbf{Z}_g^{-1}=\operatorname{diag}(0.5+0.5j,\,1) gives νF[Yg,ωc]=0.5>0\nu_F[\mathbf{Y}_g,\omega_c]=0.5>0. The minor loop gains of the converters are

L1=Y1Zg=[−2−2j001],L2=Y2Zg=[2j001].(42)\mathbf{L}_1=\mathbf{Y}_1\mathbf{Z}_g= \begin{bmatrix} -2-2j & 0\\ 0 & 1 \end{bmatrix},\quad \mathbf{L}_2=\mathbf{Y}_2\mathbf{Z}_g= \begin{bmatrix} 2j & 0\\ 0 & 1 \end{bmatrix}. \tag*{(42)}

Now we can calculate the specifications based on passivity and Gershgorin theorem in dqdq for converters 1 and 2, according to (35) and (37):

D1P(ωc)=0.25,D1dq(ωc)=−1.5,D_1^P(\omega_c)=0.25,\qquad D_1^{dq}(\omega_c)=-1.5,
D2P(ωc)=−0.75,D2dq(ωc)=0.5.(43)D_2^P(\omega_c)=-0.75,\qquad D_2^{dq}(\omega_c)=0.5. \tag*{(43)}

From these values, we can see that converter 1 satisfies the passivity-based specification, while converter 2 satisfies the Gershgorin-based specification in dqdq domain. However, the loop gain L=L1+L2=(Y1+Y2)Zg=diag⁡(−2, 2)\mathbf{L}=\mathbf{L}_1+\mathbf{L}_2=(\mathbf{Y}_1+\mathbf{Y}_2)\mathbf{Z}_g=\operatorname{diag}(-2,\,2) has an eigenvalue at −2-2, which is exactly inside the forbidden region depicted in Figure 4, meaning that combinations of different specifications can’t be used at a single frequency.

References

  1. [1]R. D. Middlebrook, “Input filter consideration in design and application of switching regulators,” IEEE Ind Appl. Soc Annu Meet., pp. 366–382, 1976.
  2. [2]J. Sun, “Impedance-Based Stability Criterion for Grid-Connected Inverters,” IEEE Trans. Power Electron., vol. 26, no. 11, pp. 3075–3078, Nov. 2011.DOI
  3. [3]B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli, and Z. Shen, “Inverse Nyquist Stability Criterion for Grid-Tied Inverters,” IEEE Trans. Power Electron., vol. 32, no. 2, pp. 1548–1556, Feb. 2017.
  4. [4]X. Wang and F. Blaabjerg, “Harmonic Stability in Power Electronic-Based Power Systems: Concept, Modeling, and Analysis,” IEEE Trans. Smart Grid, vol. 10, no. 3, pp. 2858–2870, May 2019.
  5. [5]B. Wen, D. Dong, D. Boroyevich, R. Burgos, P. Mattavelli, and Z. Shen, “Impedance-Based Analysis of Grid-Synchronization Stability for Three-Phase Paralleled Converters,” IEEE Trans. Power Electron., vol. 31, no. 1, pp. 26–38, Jan. 2016.
  6. [6]B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli, and Z. Shen, “Analysis of D-Q Small-Signal Impedance of Grid-Tied Inverters,” IEEE Trans. Power Electron., vol. 31, no. 1, pp. 675–687, Jan. 2016.
  7. [7]C. Wildrick, F. Lee, B. Cho, and B. Choi, “A method of defining the load impedance specification for a stable distributed power system,” IEEE Trans. Power Electron., vol. 10, no. 3, pp. 280–285, May 1995.
  8. [8]Z. Liu, J. Liu, W. Bao, and Y. Zhao, “Infinity-Norm of Impedance-Based Stability Criterion for Three-Phase AC Distributed Power Systems With Constant Power Loads,” IEEE Trans. Power Electron., vol. 30, no. 6, pp. 3030–3043, Jun. 2015.DOI
  9. [9]B. Wen, R. Burgos, D. Boroyevich, P. Mattavelli, and Z. Shen, “AC Stability Analysis and dq Frame Impedance Specifications in Power-Electronics-Based Distributed Power Systems,” IEEE J. Emerg. Sel. Top. Power Electron., vol. 5, no. 4, pp. 1455–1465, Dec. 2017.
  10. [10]L. Hanerfors, X. Wang, A. G. Yepes, and F. Blaabjerg, “Passivity-Based Stability Assessment of Grid-Connected VSCs—An Overview,” IEEE J. Emerg. Sel. Top. Power Electron., vol. 4, no. 1, pp. 116–125, Mar. 2016.
  11. [11]L. Hanerfors, “Proof and Application of the Positive-Net-Damping Stability Criterion,” IEEE Trans. Power Syst., vol. 26, no. 1, pp. 481–482, Feb. 2011.
  12. [12]G. Stamatiou and M. Bongiorno, “Stability Analysis of Two-Terminal VSC-HVDC Systems Using the Net-Damping Criterion,” IEEE Trans. Power Delivery, vol. 31, no. 4, pp. 1748–1756, Aug. 2016.
  13. [13]F. Chen, X. Wang, L. Hanerfors, S. Z. Khong, D. Wang, L. Zhao, K. C. Sou, M. Routimo, J. Kukkola, H. Sandberg, and K. H. Johansson, “Limitations of Using Passivity Index to Analyze Grid–Inverter Interactions,” IEEE Trans. Power Electron., vol. 39, no. 11, pp. 14 465–14 477, Nov. 2024.DOI
  14. [14]J. Kowalewski, A. Lorenz, F. Thomas, R. A. A. Valenzuela, and K. Graichen, “Passivity-Based Robust Stability Analysis of the Converter–Grid Interaction and Passivity-Shaping Controller Design,” IET Gener. Transm. Distrib., vol. 19, no. 1, p. e70090, 2025.
  15. [15]F. Chen, S. Z. Khong, L. Hanerfors, X. Wang, D. Wang, H. Sandberg, L. Zhao, M. Routimo, J. Kukkola, K. C. Sou, and K. H. Johansson, “An Extended Frequency-Domain Passivity Theory for MIMO Dynamics Specifications of Voltage-Source Inverters,” IEEE Trans. Power Electron., pp. 1–15, 2024.DOI
  16. [16]X. Feng, C. Liu, and F. Lee, “Impedance specifications for stable DC distributed power systems,” IEEE Trans. Power Electron., vol. 17, no. 2, pp. 157–162, Mar. 2002.DOI
  17. [17]X. Wang, Y. Peng, J. Zhu, Y. Xia, M. Yu, H. Hu, H. Cai, and W. Wei, “Decentralized Impedance Specifications for Small-Signal Stability of DC Distributed Power Systems,” IEEE J. Emerg. Sel. Top. Power Electron., vol. 5, no. 4, pp. 1578–1588, Dec. 2017.DOI
  18. [18]X. Wang and Y. Peng, “Decentralized Impedance-Based Stability Analysis for Series-Type DC Microgrids,” in 2019 IEEE Sustain. Power Energy Conf. (iSPEC), Nov. 2019, pp. 342–346.DOI
  19. [19]C. Zhang, W. Chen, and B. He, “Decentralized Impedance Specifications for Paralleled DC Distribution Power System With Multiple-Voltage-Level Buses,” IEEE Trans. Power Electron., vol. 39, no. 1, pp. 112–124, Jan. 2024.DOI
  20. [20]L. Stojanović, R. Cvetanović, P. Mattavelli, and S. Buso, “Power Converter Decentralized Specifications in Different Grid Nodes for DC Distribution Systems,” IEEE Trans. Power Electron., pp. 1–14, 2025.DOI
  21. [21]B. He, W. Chen, C. Zhang, Y. Yuan, and C. Zhang, “Impedance Specifications for Stability Design of Grid-Connected DC Distribution Power Systems,” IEEE Trans. Ind. Electron., pp. 1–14, 2023.DOI
  22. [22]L. Stojanović, R. Cvetanović, P. Mattavelli, S. Buso, and D. Biaene, “Decentralized Impedance Specifications for Grid-Connected Converters in AC Grids,” in 2024 Energy Conversion Congress & Expo Europe (ECCE Europe), Darmstadt, Germany, Sep. 2024, pp. 1–7.DOI
  23. [23]F. Chen, X. Wang, L. Hanerfors, S. Z. Khong, J. Kukkola, M. Routimo, L. Zhao, D. Wang, H. Sandberg, K. C. Sou, and K. H. Johansson, “Assessment, Cause Localization, and Mitigation for Small-Signal Stability Risks in Multi-Parallel Grid-Forming Converter Systems,” IEEE Trans. Smart Grid, pp. 1–1, 2025.
  24. [24]X. He and F. Dörfler, “Passivity and Decentralized Stability Conditions for Grid-Forming Converters,” IEEE Trans. Power Syst., vol. 39, no. 3, pp. 5447–5450, May 2024.DOI
  25. [25]L. Huang, D. Wang, X. Wang, H. Xin, P. Ju, K. H. Johansson, and F. Dörfler, “Gain and Phase: Decentralized Stability Conditions for Power Electronics-Dominated Power Systems,” IEEE Trans. Power Syst., pp. 1–16, 2024.arxiv.org/abs/2309.08037
  26. [26]F. Cecati and M. Liserre, “Interoperability Specifications for Multi-Vendor Converter-Dominated Grid: A Robust Stability Perspective,” IEEE Trans. Smart Grid, pp. 1–1, 2025.DOI
  27. [27]K. Laib, J. Watson, Y. Ojo, and I. Lestas, “Decentralized stability conditions for DC microgrids: Beyond passivity approaches,” Automatica, vol. 149, p. 110705, Mar. 2023.arxiv.org/abs/2109.02708
  28. [28]D. Zhao, W. Chen, and L. Qiu, “When Small Gain Meets Small Phase,” Feb. 2022.arxiv.org/abs/2201.06041
  29. [29]E. Baron-Prada, A. Anta, and F. Dörfler, “On Decentralized Stability Conditions Using Scaled Relative Graphs,” IEEE Control Syst. Lett., vol. 9, pp. 691–696, 2025.
  30. [30]Z. Liu, J. Zhao, K. Ou, Z. Zeng, and L. Mao, “A Small-Signal Stability Criterion for DC Microgrid Based on Extended-Gershgorin Theorem,” IEEE Trans. Power Deliv., vol. 38, no. 3, pp. 2047–2057, Jun. 2023.DOI
  31. [34]Y. Li, Peng Zhang, Lingyu Ren, and T. Orekan, “A Gersgorin theory for robust microgrid stability analysis,” in 2016 IEEE Power Energy Soc. Gen. Meet. PESGM. Boston, MA, USA: IEEE, Jul. 2016, pp. 1–5.DOI
  32. [35]F. Chen, S. Z. Khong, X. Wang, and L. Hanerfors, “Unified and Flexible Frequency-Domain Stability Assessment Framework for Power-Electronic-Based Power Systems,” IEEE Trans. Power Electron., pp. 1–6, 2025.DOI
  33. [36]C. A. Desoer and M. Vidyasagar, Feedback Systems: Input-Output Properties. Philadelphia, PA.: Society for Industrial and Applied Mathematics (SIAM, 3600 Market Street, Floor 6, Philadelphia, PA 19104), 2009.
  34. [37]R. A. Horn and C. R. Johnson, Matrix Analysis. Cambridge: Cambridge University Press, 1985.
  35. [38]Y. Ren, R. Duan, L. Chen, W. Huang, C. Wu, and Y. Min, “Stability Assessment of Grid-Connected Converter System Based on Impedance Model and Gershgorin Theorem,” IEEE Trans. Energy Convers., vol. 35, no. 3, pp. 1559–1566, Sep. 2020.DOI
  36. [39]A. Rygg, M. Molinas, C. Zhang, and X. Cai, “A Modified Sequence-Domain Impedance Definition and Its Equivalence to the dq-Domain Impedance Definition for the Stability Analysis of AC Power Electronic Systems,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 4, no. 4, pp. 1383–1396, Dec. 2016.arxiv.org/abs/1605.00526
  37. [40]Y. Zhou, H. Hu, J. Yang, and Z. He, “A Novel Forbidden-Region-Based Stability Criterion in Modified Sequence-Domain for AC Grid-Converter System,” IEEE Trans. Power Electron., vol. 34, no. 4, pp. 2988–2995, Apr. 2019.DOI
  38. [41]D. Wang, W. Chen, S. Z. Khong, and L. Qiu, “On the phases of a complex matrix,” Linear Algebra and its Applications, vol. 593, pp. 152–179, May 2020.arxiv.org/abs/1904.07211

Paper details

vlnc:W_zsyq93q5
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Not recorded for this paper

View statement
Looking for related papers…
Contents