IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 5, NO.

4, DECEMBER 2017 1513

Stability Assessment of Power-Converter-Based

AC systems by LTP Theory: Eigenvalue Analysis
and Harmonic Impedance Estimation
Valerio Salis, Student Member, IEEE, Alessandro Costabeber, Member, IEEE, Stephen M. Cox,
and Pericle Zanchetta, Senior Member, IEEE

Abstract— Stability analysis of power-converter-based ac sys- to the harmonic couplings at the common connection points
tems poses serious challenges not only because of the nonlinear between sources and loads.
nature of power converters, but also because linearization is not In general, there are two main possible approaches for
generally applied around a steady-state operating point, as in the
dc case, but around a time-periodic operating trajectory. Typical stability analysis of nonlinear systems: linearize the system
examples are single-phase and unbalanced three-phase systems. in order to use well known linear analysis techniques, or
In this paper, two general methods to assess stability of the afore- use the Lyapunov stability criterion directly on the nonlinear
mentioned systems are presented. Both are based on the linear model of the system. The application of the latter in practical
time periodic (LTP) systems theory. The first is model-based and applications is usually complicated, so most stability analysis
relies on the evaluation of the eigenvalues of the linearized model,
assuming a complete knowledge of the parameters. By contrast, techniques rely on the linearized model of the system. One of
the second proposes a set of small-signal current injections to the first investigations of these instabilities was presented by
measure the harmonic impedances and applies the LTP Nyquist Middlebrook [1], where the interaction between closed loop
criterion, so that the stability of the system can be assessed with controlled dc–dc converters and their input filters was the
a black-box approach, without relying on the knowledge of the cause of instability. This cascaded system was found to be
system parameters. The basic LTP systems theory is reviewed
in order to provide a mathematical justification for the second stable only if the ratio of the source output impedance to the
method. As case study, a simple network, consisting of a source load input impedance satisfies the Nyquist stability criterion,
full-bridge converter in ac voltage-control mode and a load full- with analysis based on the linearized model of the converter.
bridge converter in ac current-control mode including phase Other scenarios, where similar instability issues arise, are
locked loop, is considered. Analytical results based on average reported in [2] and [3] for different single-phase railway dis-
modeling and simulations based on both average and switching
models are presented, showing good accuracy in the identification tribution systems. In [4], the unstable behavior of line commu-
of the stability thresholds for both the proposed methods. tated converters High-Voltage DC converters is discussed. The
analysis [5] shows that renewable energy systems involving
Index Terms— Harmonic state space (HSS) model, impedance
measurement, linear time periodic (LTP) systems, power convert- grid-parallel inverters are affected by instability issues, and
ers, stability analysis. subsequently, in [6] and [7], a report is provided of ongoing
research, current harmonic-related issues, and future expected
I. I NTRODUCTION problems.
In the case of ac systems, the Middlebrook approach is still
I N RECENT decades, improvements in power electronics
have led to a significant spread of grid-connected convert-
ers, both for ac and dc applications. These systems, which
applicable, but must be adapted slightly. In [8], an equivalent
input–output impedance approach is proposed for the analysis
are intrinsically nonlinear due to the presence of switching of balanced three-phase ac power systems, in dualism with
elements, have been a challenging field of investigation for the dc case, based on the study of the synchronous dq frame
−1
researchers. Power electronics guarantees high conversion return ratio matrix Z s,dq (s)Z l,dq (s), exploiting multivariable
efficiency and high controllability of loads and sources. In fact, linear control theory. Later, Belkhayat [9] proposed a further
it is possible to achieve high dynamic performances using improvement to this method by employing the generalized
sophisticated controls and modulation schemes. However, Nyquist criterion (GNC), which was originally introduced by
these systems often show unstable behavior, especially due MacFarlane and Postlethwaite [10], and which extends the
single-input single-output frequency-response control methods
Manuscript received January 31, 2017; revised April 3, 2017 and to the multiple-input multiple-output (MIMO) case. Other
May 10, 2017; accepted May 17, 2017. Date of publication June 9, 2017;
date of current version October 30, 2017. Recommended for publication by examples of stability analysis of three-phase balanced ac sys-
Associate Editor M. Molinas. (Corresponding author: Valerio Salis.) tems are provided in [11], based on dq measurements of small-
V. Salis, A. Costabeber, and P. Zanchetta are with the Department of signal impedances. In fact, in the dc case, the nonlinear system
Electrical and Electronic Engineering, University of Nottingham, Nottingham
NG7 2RD, U.K. (e-mail: [email protected]). is linearized about a fixed operating point, leading to an Linear
S. M. Cox is with the School of Mathematical Sciences, University of Time Invariant (LTI) linearized system whose stability is
Nottingham, Nottingham NG7 2RD, U.K. assessed with LTI techniques. For the ac balanced three-phase
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. case, the system is first transformed into an MIMO system in
Digital Object Identifier 10.1109/JESTPE.2017.2714026 the dq reference frame, and then linearization is performed
2168-6777 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1514 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 5, NO. 4, DECEMBER 2017

as in the dc case and MIMO LTI techniques are exploited although the model does not include important elements,
for the stability analysis. However, for single-phase and for such as digital computation delay, zero-order hold (ZOH),
unbalanced three-phase ac systems, the methods described and pulsewidth modulator (PWM) dynamics.
previously are not suitable. The main reason is the linearization It is worth pointing out that the design of the fast control
process that in this case must be performed around a steady- loops, i.e., inner current control, is normally carried out on the
state time-periodic trajectory rather than around a fixed steady- assumption that the system is linear, and ignoring potential
state operating point. cross-converter interactions caused by the nonlinear loops,
There are several techniques available in the literature to such as PLLs, power controllers, and dc-link controllers.
analyze single phase and generic unbalanced three phase On the other hand, when modeling the impact of these nonlin-
systems. One of the most widely used is harmonic lineariza- ear loops on an interconnection of converters, the effect of the
tion, where small-signal perturbations are used to determine fast dynamics could often be neglected. To keep the analysis
the input–output small-signal impedances [12]–[14], and the as general as possible, in this paper, stability assessment
Nyquist criterion is applied to their ratio. However, this method will be performed including the effect of all the dynamics,
might be inaccurate in determining instabilities below the fun- with the only approximation being that PWM and digital
damental grid line frequency, as shown in [15]. An extension implementation delays will be modeled with their continuous-
of this method is provided in [16], where a 2-D admittance for time estimate.
single-phase voltage source converters application is defined, In this paper, LTP theory is exploited in an innovative
able to capture the cross-coupling effects (up to twice the way: based on the HSS model of the system, rewritten as
line frequency) and possibly overcome the aforementioned an interconnection between a load and source system, the
limitation. harmonic impedances are calculated for both subsystems using
Another method is the dynamic phasor approach, where a small-perturbation current injection method. Thus, a closed-
the analyzed system is described through a small-signal loop transfer function is defined and the LTP Nyquist criterion
impedance-based model. In [17], an application example to a is used to assess the stability of the overall system. The
single-phase system is provided, with a Fourier series analysis method is shown to be equivalent to the eigenvalue analysis,
truncated at the one-phasor, which keeps the mathematical and good accuracy is obtained in the identification of the
formulation closer to the dq transformation approach and, stability boundaries. The theoretical foundations of a method
therefore, uses the GNC to assess stability. to measure these harmonic impedances are also provided,
The method used in this paper is based on the ensuring the practically relevant feature that this method is
linear time periodic (LTP) systems theory, reported no longer model-based but rather can be applied to any
by Wereley and Hall [18] and Hall and Wereley [19]. ac network, provided that the harmonic impedances can be
Single-phase and unbalanced three-phase systems are measured. Analytical results, as well as simulations based
described by average models that are nonlinear time periodic on both average and switching models of the two converter
(NLTP), where all the state-space variables are periodic systems are given for both the eigenvalue analysis and the
with frequency equal to the fundamental line frequency or a harmonic impedances method, showing good agreement and
multiple. Thus, linearizing around the steady-state operating confirming the effectiveness of the proposed analysis. This
trajectories, the time-domain LTP model is derived, as well paper is organized as follows. Section II provides a review
as the frequency-domain harmonic state-space (HSS) model. of LTP theory. Section III describes the system used as case
Stability analysis can then be performed on the latter, either study and derives the nonlinear average model. In Section IV,
evaluating the eigenvalue loci plot or exploiting the LTP the steady-state time-periodic solution of the system is derived,
Nyquist criterion. The harmonic transfer function (HTF) and in Section V, the LTP model is calculated. In Section VI,
matrix-operator describes the interaction of the harmonic analytical and simulation results based on the average model
coefficients of the input and output signals, giving a precise are given, and in Section VII, simulation results based on a
description of the cross-coupling effects. The formulation of complete switching model are provided.
the LTP analysis permits easy inclusion of an arbitrary number
of harmonic coefficients in the stability assessment, and this II. R EVIEW OF L INEAR T IME P ERIODIC S YSTEMS T HEORY
is an advantage compared with harmonic linearization and
Given a general Non-Linear Time Periodic (NLTP) system,
dynamic phasor approaches, where truncations are usually
with T -periodic state-space variables, T = 2π/ωT = 1/ f T
introduced to deal with the more complicated mathematical
formulation. In [20] and [21], the HSS model and stability ẋ(t) = f (x(t)) + g(x(t))u(t), y(t) = h(x(t)) + l(x(t))u(t)
analysis are addressed for a single-phase and a three-phase (1)
grid connected converter, but without considering the and given a steady-state T -periodic input, ū(t), the system
dynamics due to the phase locked loop (PLL) and the impact equations can be solved either applying harmonic balance [18],
of digital-computational delays. In [22], an impedance-based or numerically (in MATLAB for example) and the periodic
stability method is derived and a single-phase grid-connected steady-state solutions, x̄(t), ȳ(t), can be obtained. To proceed
converter with PLL is used as a test case; however, precise with stability analysis, the LTP model is derived by linearizing
stability boundaries are not provided. In [23], the same circuit the NLTP system around the calculated steady-state solutions.
is analyzed using an eigenvalue-based approach, with the use In contrast to dc systems, where the steady-state operating
of LTP theory. Precise stability boundaries can be evaluated, point is constant, ac systems have a time-periodic steady-state
SALIS et al.: STABILITY ASSESSMENT OF POWER-CONVERTER-BASED AC SYSTEMS BY LTP THEORY 1515

operating trajectory. Thus, linearization is applied by adding

small-signal perturbation terms around the periodic steady-
state operating trajectory
u(t) = ū(t)+ ũ(t), y(t) = ȳ(t)+ ỹ(t), x(t) = x̄(t)+ x̃(t) (2)
with |ũ(t)|  |ū(t)|, | ỹ(t)|  | ȳ(t)|, and |x̃(t)|  |x̄(t)|.
Then, (2) is substituted into (1)
x̄(t) ˙
˙ + x̃(t) = f (x̄(t) + x̃(t)) + g(x̄(t) + x̃(t))[ū(t) + ũ(t)]
ȳ(t) + ỹ(t) = h(x̄(t) + x̃(t)) + l(x̄(t) + x̃(t))[ū(t) + ũ(t)]
(3) Fig. 1. General eigenvalue loci of an LTP system. Red: significant
eigenvalues. Black: translated copies.
and the LTP system is then obtained by taking into account
only the first-order terms, ignoring the steady-state and second-
order terms with N = diag(. . . , N−n , . . . , N−1 , N0 , N1 , . . . , Nn , . . . ) and
Nn being a square diagonal matrix of the same dimensions as
˙ = A(t)x̃(t)+ B(t)ũ(t),
x̃(t) ỹ(t) = C(t)x̃ (t)+ D(t)ũ(t) (4) An with diagonal coefficients equal to j nωT . Rearranging (8),
with the matrices A(t), B(t), C(t), and D(t) being T - the relationship between the harmonic coefficients of the input
periodic. The test input signal for the frequency analysis of and output signals is represented by the HTF operator of the
LTP systems is the exponentially modulated periodic (EMP) system, defined as
signal [18], [19], which is defined as Y = H(s)U, H(s) = C[sI − (A − N )]−1 B + D (9)
∞

ũ(t) = ũ n est e j nωT t , s = j ω. (5) where I is the Toeplitz form of the identity matrix. Stability
n=−∞ analysis can now be performed in two equivalent ways.
When such a signal is given as input to the LTP system, all the
state-space variables and the output are EMP signals. Thus, A. Eigenvalue Loci
input and output spaces are consistent and it is possible to This method requires a full model of the system, i.e., knowl-
define a transfer function operator. Substituting (5) into (4) and edge of all the component values and controller parameters.
applying the Fourier matrix expansion to the matrices A(t), Stability analysis is performed by evaluating the eigenvalues
B(t), C(t), and D(t) allows the system to be rearranged as of the matrix (A − N ). If all the eigenvalues have Re[λi ] ≤ 0,
∞
 ∞
 where those with Re[λi ] = 0 have algebraic multiplicity equal
( j ω + j nωT )x̃ n = An−m x̃ m + Bn−m ũ m to 1, then the system is stable, otherwise the system is unstable.
m=−∞ m=−∞ In real applications, only a finite number of harmonic
∞
 ∞
 coefficients can be considered, for computational reasons.
ỹn = Cn−m x̃ m + Dn−m ũ m . (6) A truncation of the Fourier series in (6) is applied, defining the
m=−∞ m=−∞ truncation order M, which refers to the number of harmonics
This set of equations is not convenient for analytical manipula- taken into account. If M = 2, for example, the dc-component
tion because of the convolution products. A simpler form can and the first and second harmonics are considered. The cor-
be derived by applying the Toeplitz transform to the periodic responding truncated Fourier expansion involves the Fourier
matrices, defined as follows for the generic matrix P(t): coefficients n = −2, −1, 0, 1, 2 and yields the approximation
⎡ ⎤
.. .. .. .. 
M 
2
⎢ . . . . ⎥
⎢· · · P0 P−1 P−2 · · ·⎥ A(t) = An e j nωT t = An e j nωT t (10)
⎢ ⎥
T [P(t)] = P = ⎢ P−1 · · ·⎥
n=−M n=−2
⎢· · · P1 P0 ⎥ (7)
⎢· · · P2 P P · · · ⎥ and the following associated truncated Toeplitz form:
⎣ 1 0 ⎦ ⎡ ⎤
.. .. .. ..
. . . . A0 A−1 A−2 Z A ZA
⎢ A1 A0 A−1 A−2 Z A ⎥
⎢ ⎥
which is a doubly infinite block Toeplitz matrix and the blocks T [ A(t)] = A = ⎢
⎢ A 2 A 1 A0 A−1 A−2 ⎥ ⎥ (11)
Pi are the Fourier matrix coefficients of the T -periodic matrix ⎣Z A A2 A1 A0 A−1 ⎦
P(t). This definition applies to the matrices T [ A(t)] = A, ZA ZA A2 A1 A0
T [B(t)] = B, T [C(t)] = C, and T [D(t)] = D and to the
vectors T [x̃(t)] = X , T [ũ(t)] = U, and T [ ỹ(t)] = Y. The with Z A being a zero matrix of the same dimension as
Toeplitz form of (6) provides a more compact description An . For a general LTP system of order k and truncation
of the system, enabling the definition of the HSS model order M, where k refers to the total number of state-space
associated with the LTP system (4) variables, A − N is a square (2M + 1)k × (2M + 1)k matrix,
with the total number of associated eigenvalues equal to
sX = (A − N )X + BU (2M+1)k. Among these, the number of significant eigenvalues
Y = CX + DU (8) equals the order k, while the remaining 2Mk are translated
1516 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 5, NO. 4, DECEMBER 2017

Fig. 2. Grid-connected converter: parallel current injection. Fig. 3. HTF block diagram.

copies of the significant ones, with translation equal to j nωT ,

n = ±1, . . . , ±M. Fig. 1 reports an example of LTP eigen-
value loci with k = 4 and M = 2. There are four vertical
lines of eigenvalues and a total number of eigenvalues equal Fig. 4. Closed loop block diagram: U = I x . Case 1: Y = Ig and
to (2 × 2 + 1) × 4 = 20. The significant ones are shown G = Zg−1 (s)Zc (s). Case 2: Y = Ic and G = Zc−1 (s)Zg (s).
in red, while their translated copies are in black. Increasing
the truncation order M results in longer vertical lines of
eigenvalues. For example, incrementing the truncation order to where the outputs are ỹ(t) = [ ỹ1 (t); ỹ2 (t); ỹ3 (t)] =
M = 3 will result in eight more eigenvalues, since now there [ĩ g (t); ĩ c (t); ṽ o (t)] and the only input is ũ(t) = ĩ x (t). The
is a total number of (2 × 3 + 1) × 4 = 28 eigenvalues. Thus, associated HTF satisfies
⎡ ⎤
each line will be incremented with two more eigenvalues, Ig
one located on the top and the other on the bottom of the ⎣ Ic ⎦ = [C[ j ωI−(A − N )]−1 B + D]Ix ≡ H( j ω)Ix (13)
eigenvalue line, giving rise to two additional translated copies Vo
of the significant eigenvalue. The system associated with the
where the Toeplitz transform has been applied also to the
eigenvalue loci plot in Fig. 1 is clearly unstable, since there
input and output signals: T [ĩ g (t)] = Ig , T [ĩ c (t)] = Ic ,
is one significant eigenvalue in the right-half plane.
T [ṽ o (t)] = Vo , and T [ĩ x (t)] = Ix . The individual HTFs,
Remark: The locations of the eigenvalues depend on the
obtained from (13), are Ig = H1 ( j ω)Ix , Ic = H2 ( j ω)Ix ,
chosen truncation order. HSS analysis is defined on the
and Vo = H3 ( j ω)Ix . Rearranging these equations gives us
assumption that an infinite truncation order is used, so that
the harmonic impedances
all the significant eigenvalues (plus their translated copies)
are correctly located in the complex plane. When a finite Vo = H3 ( j ω)H1−1( j ω)Ig = Zg ( j ω)Ig
truncation order is used, the eigenvalues are shifted, and to
Vo = H3 ( j ω)H2−1( j ω)Ic = Zc ( j ω)Ic . (14)
ensure the robustness of the numerical results, a truncation
order M = M ∗ is determined, such that for M > M ∗ , the The equivalent block diagram is shown in Fig. 3, referring
eigenvalue loci plot exhibits no noticeable shift compared with to which we have Vo = Zg ( j ω)Ig = Zc ( j ω)Ic and
the ones evaluated for M  M ∗ , which provide the correct Ix = Ig + Ic , thus
eigenvalue locations; for M < M ∗ , there is a noticeable
−1
shift of the eigenvalue locations in the complex plane. The Ig = I + Zg−1 ( j ω)Zc ( j ω) Zg−1 ( j ω)Zc ( j ω)Ix
evaluation of M ∗ has been addressed in the literature [24],
−1
Ic = I + Zc−1 ( j ω)Zg ( j ω) Zc−1 ( j ω)Zg ( j ω)Ix . (15)
but an explicit formula has not yet been provided. Thus, M ∗
is usually evaluated iteratively. In the work presented in the The equations in (15) are analogous to the input–output
following, M ∗ = 23 has been found. Hence, to guarantee an closed-loop transfer functions of an MIMO system; in this
accurate analysis, in Sections VI and VII, an order M = 40 case, the inputs are the harmonics associated with the injected
is used. perturbation current and the outputs are the harmonics associ-
ated with the grid and converter perturbation currents, respec-
B. LTP Nyquist Criterion tively. These equations can be represented by the simple block
LTP Nyquist criterion does not necessarily require full diagram shown in Fig. 4. The LTP Nyquist criterion, described
knowledge of the model; this can be extracted through small- by Hall and Wereley [19], is here exploited to assess the
signal current injections, thus making the method suitable for a stability of the system. For a detailed derivation of the LTP
generic “black-box” system, provided that the ac terminals are Nyquist Criterion, the reader is encouraged to refer to the
accessible. Fig. 2 shows the case of a grid-connected power original work. The main result useful for our purposes can
converter, where a small-current injection is performed at the be stated as follows.
interface between the ac grid and the power converter. Such Theorem: Consider the open-loop HTF Zg−1 (s)Zc (s)
an injection is small enough to not affect the steady-state [or equivalently Zc−1 (s)Zg (s)] and assume that it has no
operation of the system. The system in Fig. 2 is linearized unobservable or uncontrollable right-half plane poles. Then,
and the LTP model is derived as in (4), and for convenience the LTP system (4) is stable if and only if the LTP Nyquist
is reproduced here plot of the eigenvalues of Zg−1 (s)Zc (s) [or Zc−1 (s)Zg (s)], for
s = j ω and ω ∈ [−ωT /2, ωT /2], has a number of counter-
˙
x̃(t) = A(t)x̃(t) + B(t)ĩ x (t) clockwise encirclements of the critical point (−1, 0) in the
ỹ(t) = C(t)x̃(t) + D(t)ĩ x (t) (12) complex plane equal to the number of poles of Zg−1 (s)Zc (s)
SALIS et al.: STABILITY ASSESSMENT OF POWER-CONVERTER-BASED AC SYSTEMS BY LTP THEORY 1517

one provided by the PLL, the system does not show unstable
behavior, confirming that the instability is in fact caused by
the PLL and not by the other controllers or by limitations in
the power transfer capability [22].
The analysis of the network is based on the average model
of the system, considering that PLL instability is expected
at relatively low frequencies, where the switching behavior
does not have significant impact on the stability boundaries.
The average model is given in (20), which is a 14th-order
NLTP system, as in (1), with all the state-space variables
being Tg -periodic. To maximize the accuracy of the model
Fig. 5. LTP Nyquist contour in the complex plane. Red crosses: poles of while retaining low complexity and to generalize the analysis,
the open-loop HTF, either Zg−1 (s)Zc (s) or Zc−1 (s)Zg (s). the elements modeling the digital control, such as the com-
putational delay (Ts = 1/ f s ), along with the ZOH delay of
the PWM, 0.5Ts , are included in the analysis through their
continuous-time equivalents [25], [26]. In fact, these digital
[or Zc−1 (s)Zg (s)], included by the LTP Nyquist contour plot,
blocks are represented in the continuous-time domain by the
as shown in Fig. 5 [19].
following transfer function:
Remark:
1) It is worth noting that the LTP Nyquist contour com- F(s) = e−sTs (sTs )−1 [1 − e−sTs ]. (17)
prises the four segments N A , N B , NC , and N D , as shown
in Fig. 5, and it is used to determine the number of The complex exponential is replaced with a first-order Padé
right-half plane poles of the open-loop HTF. Considering approximation of the form
the example in Fig. 5, there are three right-half plane
poles, and thus, the LTP Nyquist plot should have e−sTs → (n 1 s + n 0 )(d1 s + d0 )−1 (18)
three counterclockwise encirclements around the (−1, 0) where the coefficients n 1 , n 0 , d1 , and d0 have been cal-
point, for a stable system. In practice, however, the culated with the MATLAB command pade, with specified
Nyquist plot is evaluated only for s ∈ N A , since the other time delay equal to Ts and order equal to one. Substituting
segments of the LTP Nyquist contour do not contribute (18) into (17) gives an approximated transfer function of the
to generate encirclements around the critical point [3]. form
2) It follows by calculation that: F(s) = (γ1 s + γ0 )(s 2 + σ1 s + σ0 )−1 . (19)
Zc−1 (s)Zg (s) = H2 (s)H1−1 (s) The state-space variables of the nonlinear model of the system
Zg−1 (s)Zc (s) = H1 (s)H2−1 (s). (16) in Fig. 6 have the following physical meaning: x 1 − x 4
describe the internal dynamics of the PLL; x 5 is associated
III. C ASE S TUDY AND AVERAGE M ODEL with the PI1 current controller; x 6 and x 7 are related to the
The system considered as case study is a single-phase voltage controller PI3,4 ; x 8 and x 9 and x 10 and x 11 represent
network made up of a voltage-controlled full-bridge inverter the internal dynamics of the computational delay, ZOH and
(source converter) and a current-controlled full-bridge inverter PWM for the current and voltage inverters, respectively; x 12
(load converter), connected through an L 2 C1 L 1 low-pass filter. represents the inductor current, i L1 ; x 13 the inductor current,
The voltage control of the source converter ensures that the i L2 ; x 14 the voltage across the capacitor. The outputs are: y1
voltage v o (t) follows the reference v g (t) = Vg sin(ωg t), while the current toward the current-source inverter; y2 the current
the load converter transfers power from the dc source Vdc2 toward the voltage-source inverter, and y3 the voltage at the
toward the source converter. common connection point, v o . Please refer to Fig. 6 to identify
To obtain unity-power-factor operation mode, the current- the location of the different state variables. Explicit indication
controlled inverter is synchronized with the source voltage- of the time dependence of the variables x 1 − x 14 , y1 − y3 ,
converter through a PLL, where the in-quadrature component i L1 , i L2 , v g , v o , v conv1 , v conv2 , and ĩ x has been suppressed in
of the input voltage v o (t) is estimated using a linear filter that the following, in order to reduce the quantity of mathematical
introduces a Tg /4 delay at ωg : D(s) = ωg2 /(s 2 +sωg +ωg2 ). The notation:
amount of power transferred by the load converter is controlled
v o = RC x 12 + RC x 13 + x 14 + RC ĩ x
through the parameter Iref , which sets the reference amplitude
for the current controller. It will be shown that there is a v conv1 = Vdc1 γ0 x 8 + Vdc1 γ1 x 9
∗ , below which the system
threshold value for this parameter, Iref v conv2 = Vdc2 γ0 x 10 + Vdc2 γ1 x 11
operates stably, and above which the system shows unstable ẋ 1 = x 2 , ẋ 2 = −ωg2 x 1 − ωg x 2 + ωg2 v o
operation. Such instability is caused by the fact that the PLL
ẋ 3 = x 4 − k p2 sin(x 3 )v o + k p2 cos(x 3 )x 1
is no longer able to generate the correct phase reference. It is
worth noting that, in the unstable operation mode, if the current ẋ 4 = −ki2 sin(x 3 )v o + ki2 cos(x 3 )x 1
controller is driven with the correct phase rather than the ẋ 5 = Iref cos(x 3 ) − x 13 , ẋ 6 = v g − v o
1518 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 5, NO. 4, DECEMBER 2017

Fig. 6. Single-phase test case system: switching model.

TABLE I and stability analysis. Since the converter is designed for ac

S YSTEM PARAMETERS systems, the solutions take the form

x̄ i = |x̄ i | cos(ωg t + arg(x̄ i )) = [x̄ i e j ωg t + c.c.]/2

for i = 1, 2, 5, ..., 14; x̄ 3 = ωg t + x̄ 03 ; x̄ 4 = x̄ 04
(21)
j ωg t
v̄ q = |v̄ q | cos(ωg t + arg(v̄ q )) = [v̄ q e + c.c.]/2
for q = o, conv, g (22)

where c.c. stands for complex conjugate of the term preceding

it within the square brackets. Substituting these expressions in
ẋ 7 = ki3 x 6 + k p3 v g − k p3 v o − x 12 − x 13, ẋ 8 = x 9 the eighth equation of model (20) gives
ẋ 9 = −σ0 x 8 − σ1 x 9 + ki4 x 7 + k p4 ki3 x 6 + k p4 k p3 v g
x̄ 5 = [Iref e j x̄03 − x̄ 13]/( j ωg ). (23)
− k p4 k p3 v o − k p4 x 12 − k p4 x 13 + v o /Vdc1
ẋ 10 = x 11, ẋ 11 = −σ0 x 10 − σ1 x 11 + ki1 x 5 A similar approach is applied to the other equations
+ k p1 Iref cos(x 3 ) − k p1 x 13 + v o /Vdc2 x̄ 6 = [− j Vg − v̄ o ]/( j ωg ) (24)

ẋ 12 = − (R L1 + RC )x 12 − RC x 13 − x 14 + v conv1 x̄ 7 = [ki3 x̄ 6 − j k p3 Vg − k p3 v̄ o − x̄ 12 − x̄ 13]/( j ωg ) (25)

− RC ĩ x /L 1 x̄ 8 = x̄ 9 /( j ωg ) (26)

ẋ 13 = − RC x 12 − (R L2 + RC )x 13 − x 14 + v conv2 x̄ 9 = [−σ0 x̄ 8 − σ1 x̄ 9 + ki4 x̄ 7 + k p4 ki3 x̄ 6 − j k p4k p3 Vg

− RC ĩ x /L 2 − k p4 k p3 v̄ o − k p4 x̄ 12 −k p4 x̄ 13 + v̄ o /Vdc1 ]/( j ωg )

ẋ 14 = x 12 + x 13 + ĩ x /C1 (27)
y1 = −x 12 , y2 = x 12 + ĩ x x̄ 10 = x̄ 11 /( j ωg ) (28)
y3 = RC x 12 + RC x 13 + x 14 + RC ĩ x . (20) x̄ 11 = [−σ0 x̄ 10 − σ1 x̄ 11 + ki1 x̄ 5 + k p1 Iref e j x̄03
− k p1 x̄ 13 + v̄ o /Vdc2]/( j ωg ) (29)
All the theoretical analysis is based on the parameters
summarized in Table I. Based on (20), in Sections IV–VI, x̄ 12 = [−(R L1 + RC )x̄ 12 − RC x̄ 13 − x̄ 14
the steady-state solution is evaluated, linearization is applied, + v̄ conv1 ]/( j ωg L 1 ) (30)
and stability analysis is performed on the linearized model. x̄ 13 = [−RC x̄ 12 − (R L2 + RC )x̄ 13 − x̄ 14
+ v̄ conv2 ]/( j ωg L 2 ) (31)
IV. S TEADY-S TATE S OLUTION x̄ 14 = [x̄ 12 + x̄ 13 ]/( j ωg C1 ) (32)
v̄ o = RC x̄ 12 + RC x̄ 13 + x̄ 14 (33)
In this section, the steady-state solutions of the NLTP sys-
tem (20) are mathematically evaluated by applying harmonic v̄ conv1 = Vdc1 γ0 x̄ 8 + Vdc1 γ1 x̄ 9 (34)
balance. The results are then used to perform linearization v̄ conv2 = Vdc2 γ0 x̄ 10 + Vdc2 γ1 x̄ 11. (35)
SALIS et al.: STABILITY ASSESSMENT OF POWER-CONVERTER-BASED AC SYSTEMS BY LTP THEORY 1519

These equations, (23)–(35), can now be solved numerically of the system. In Section VII, time-domain simulations are
in MATLAB and the steady-state solutions x̄ 5 , . . . , x̄ 14 , v̄ o , performed based on the switching model, showing that the
v̄ conv1 , v̄ conv2 obtained. Next, x̄ 3 is first written in the form stability boundaries are still well predicted by the LTP analy-
x̄ 3 = ωg t + x̄ 03 , then sis. Subsequently, a possible method to measure the harmonic
impedances from the switching model is presented, allowing
x̄˙3 = ωg = x̄ 4 − k p2 sin(ωg t + x̄ 03 )|v̄ o | cos(ωg t + v̄ o )
one to treat the circuit as a “black-box,” providing a practical
+ k p2 cos(ωg t + x̄ 03 )|v̄ o | sin(ωg t + v̄ o ). (36) solution to assess stability using the LTP Nyquist criterion. It is
Applying trigonometric simplification gives worth pointing out that in this case, the full analysis discussed
previously is not required, as stability assessment would be
x̄ 4 = ωg − k p2 sin( v̄ o − x̄ 03 ) (37) purely based on measurements
so it follows that x̄˙4 (t) = 0. But from the state-space model, x̃˙1 = x̃ 2 , x̃˙2 = −ωg2 x̃ 1
again using trigonometric simplifications, it follows that:
− ωg x̃ 2 +ωg2 RC x̃ 12 +ωg2 RC x̃ 13 +ωg2 x̃ 14 + ωg2 RC ĩ x
x̄˙4 = ki2 [− sin(x̄ 3 )v̄ o + cos(x̄ 3 )x̄ 1 ] (38)
x̃˙3 = x˜4 − k p2 sin(x̄ 3 )RC x̃ 12 − k p2 sin(x̄ 3 )RC x̃ 13
and so − k p2 sin(x̄ 3 )x̃ 14 −k p2 cos(x̄ 3 )v̄ o x̃ 3 +k p2 cos(x̄ 3 )x̃ 1
x̄˙4 = ki2 |v̄ o | sin( v̄ o − x̄ 03 ) (39)
− k p2 sin(x̄ 3 )x̄ 1 x̃ 3 −k p2 sin(x̄ 3 )RC ĩ x
which implies v̄ o − x̄ 03 = 0 or ±π. In our case, v̄ o = x̄ 03, ˙x̃ 4 = −ki2 sin(x̄ 3 )RC x̃ 12 − ki2 sin(x̄ 3 )RC x̃ 13
which gives the solutions
− ki2 sin(x̄ 3 )x̃ 14 −ki2 cos(x̄ 3 )v̄ o x̃ 3 +ki2 cos(x̄ 3 )x̃ 1
x̄ 3 = ωg t + v̄ o , x̄ 4 = ωg . (40) − ki2 sin(x̄ 3 )x̄ 1 x̃ 3 − ki2 sin(x̄ 3 )RC ĩ x
Finally, the last two equations are x̄ 2 = j ωg x̄ 1 and x̄ 1 = − j v̄ o , x̃˙5 = −Iref sin(x̄ 3 )x̃ 3 − x̃ 13
for which the solutions x̄ 1 and x̄ 2 are given by x̃˙6 = −RC x̃ 12 − RC x̃ 13 − x̃ 14 − RC ĩ x
x̄ 1 = |v̄ o | cos(ωg t + v̄ o − π/2) (41) x̃˙7 = ki3 x̃ 6 − k p3 RC x̃ 12 − k p3 RC x̃ 13 − k p3 x̃ 14 − x̃ 12
x̄ 2 = ωg |v̄ o | cos(ωg t + v̄ o ). (42) − x̃ 13 − k p3 RC ĩ x
The steady-state solution has been calculated by applying x̃˙8 = x̃ 9 , x̃˙9 = −σ0 x̃ 8 − σ1 x̃ 9 + ki4 x̃ 7
harmonic balance. Please note that in [27], a similar method + k p4 ki3 x̃ 6 − k p4 k p3 RC x̃ 12 − k p4 k p3 RC x̃ 13
is developed for steady-state analysis of a modular multilevel − k p4 k p3 x̃ 14 − k p4 x̃ 12 − k p4 x̃ 13 − k p4 k p3 RC ĩ x
converter. Regardless of the specific application, harmonic
+ [RC x̃ 12 + RC x̃ 13 + x̃ 14 ]/Vdc1 + RC /Vdc1ĩ x
balance is a convenient technique that can be exploited when
a steady-state solution of a system of nonlinear differential x̃˙ 10 = x̃ 11 , x̃˙11 = −σ0 x̃ 10
equations is needed. − σ1 x̃ 11 + ki1 x̃ 5 − k p1 Iref sin(x̄ 3 )x̃ 3 − k p1 x̃ 13
+ RC /Vdc2ĩ x + [RC x̃ 12 + RC x̃ 13 + x̃ 14 ]/Vdc2
V. L INEARIZED M ODEL
˙x̃ 12 = − (R L1 + RC )x̃ 12 − RC x̃ 13 − x̃ 14 + Vdc1 γ0 x̃ 8
Linearization around the steady-state solution calculated in

+ Vdc1 γ1 x̃ 9 − RC ĩ x /L 1
Section IV is applied to the model (20), according to (3).
Each variable in (20) is replaced by its steady-state solution x̃˙13 = RC x̃ 12 − (R L2 + RC )x̃ 13 − x̃ 14 + Vdc2 γ0 x̃ 10

plus a small-signal perturbation. After substitution, the product + Vdc2 γ1 x̃ 11 − RC ĩ x /L 2

of perturbations is neglected, as well as the purely steady-

x̃˙14 = x̃ 12 + x̃ 13 + ĩ x /C1
state terms (which balance with one another). Doing so,
the LTP small-signal model (43) is derived, in the form ỹ1 = −x̃ 12 , ỹ2 = x̃ 12 + ĩ x
˙ = A(t)x̃(t) + B(t)ũ(t) and ỹ(t) = C(t)x̃(t) + D(t)ũ(t),
x̃(t) ỹ3 = RC x̃ 12 + RC x̃ 13 + x̃ 14 + RC ĩ x . (43)
with A(t), B(t), C(t), and D(t) being Tg -periodic matrices,
ũ(t) = ĩ x (t) being the injected current perturbation and ỹ(t)
the output vector. Applying Fourier series expansion to these
VI. AVERAGE M ODEL R ESULTS
matrices gives the result that only Ai , Bi for i = −1, 0, 1,
and C j , D j for j = 0 are different from zero, with A−1 = A∗1 To limit the complexity of the results, and to focus on
and B−1 = B1∗ (complex conjugate). All the other coefficients the effect of a single parameter, the current reference for the
of the Fourier series are zero. The Toeplitz transform is load converter, Iref , has been selected as the only variable
applied and the HSS model is calculated. In the following, quantity, while components and controllers are assumed to be
in Section VI, the stability of the system is assessed based on constant. In the system under study, the stability boundary,
∗ , lies between 11.3 and 11.4 A. For I ∗
the average model (20), comparing time-domain simulations of Iref ref < Iref , the system
the average model with the analytical results from the two LTP ∗
operates stably, while for Iref > Iref , the PLL is no longer
analysis methods, i.e., eigenvalue and LTP Nyquist based on able to generate the correct phase reference and the system
the return-ratio of the harmonic impedances, showing a very exhibits unstable operation. These results are supported by the
good accuracy in the identification of the stability boundaries following analysis.
1520 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 5, NO. 4, DECEMBER 2017

re f
Fig. 7. Average model. Currents: blue—i L1,2 (t) and red—i L1,2 (t). Voltages: blue—v g (t) and red—v o (t). Phase: blue—θ (t). (a)–(d) Transient analysis.
(e)–(h) Stable system with Iref = 11.3 A. (i)–(l) Unstable system with Iref = 11.4 A.

A. Time-Domain Simulation
Time-domain simulations have been performed in the
MATLAB-Simulink environment, based on the NLTP sys-
tem (20). In Fig. 7, the time-domain evolution of currents,
i L1 (t), i L2 (t), voltage, v o (t), and phase, θ (t) is shown, with
the reference current being equal to Iref = 11.3 A for t < 0.2 s
and to Iref = 11.4 A for t ≥ 0.2 s. It can be observed that
the system starts to strongly oscillate for t > 0.5 s, which
reflects the fact that the two unstable significant eigenvalues,
as reported in Section VI-(b), are very close to the imaginary
axis, providing a long transient. In Fig. 7(e)–(l), a zoomed-in
view of both stable and unstable operation regions is reported.
For Iref = 11.3 A, the system is stable [Fig. 7(e)–(h)], whereas
for Iref = 11.4 A, the system starts to oscillate until currents
and voltages exceed the rated values and the control switches
OFF [Fig. 7(i)–(l)].

B. Eigenvalue Analysis
Based on the HSS model, obtained from the LTP sys-
tem (43), the eigenvalue loci plot of the matrix A − N is
evaluated with Iref = 11.3 and 11.4 A. A truncation order
M = 40 has been chosen, in order to guarantee a good
Fig. 8. Eigenvalue loci plot. (a) Stable system with Iref = 11.3 A.
accuracy of the results, and at the same time, to keep low (b) Unstable system with Iref = 11.4 A.
the computational complexity. It can be seen from Fig. 8
that for Iref = 11.3 A, all the significant eigenvalues lie in right-half plane, confirming that in this case, the system is
the left-half plane, as required by a stable system, whereas unstable. The significant eigenvalues that cause instability are
for Iref = 11.4 A, some significant eigenvalues lie in the located at sunst = 1.175 ± j 5238. Thus, the natural frequency
SALIS et al.: STABILITY ASSESSMENT OF POWER-CONVERTER-BASED AC SYSTEMS BY LTP THEORY 1521

associated with these modes is ωn = 5238 rad/s, which

corresponds to an oscillating frequency fn = 830 Hz, which
is consistent with the time-domain simulations, where the
frequency of the oscillations is measured to be fn = 820 Hz
[see Fig. 7(g)]. Some spurious eigenvalues arise because of the
applied truncation, but they do not have a physical meaning
and are therefore neglected.

C. LTP Nyquist Criterion

Using the theory developed in Section II, harmonic
impedances Zg and Zc are evaluated using the LTP
model (43). The estimation of the LTP Nyquist criterion based
on small-signal perturbations will be discussed in Section VII.
Considering the truncation order M, these matrices will have
the form (44). It can be noted that, for s ∈ (− j ∞, + j ∞), the
elements on the same diagonal of the matrix impedance are
shifted copies of each other, with shifting proportional to the
fundamental period of the grid Tg = 2π/ωg . In fact, Zg0 (s) and
Zg0 (s + j ωg ), evaluated for s ∈ (− j ∞, + j ∞), give the same
amplitude and phase Bode plot, except for a frequency shift
of ωg between the two plots. Thus, the harmonic impedance
associated with the zero-component (which is related to the
ωg -component) can be obtained from any Zg0 (s + j kωg ),
k ∈ [−M/2, .., M/2], and the same applies to the harmonic
Fig. 9. (a)-Source and (b)-load harmonic impedances: stable system with
impedances of other orders. However, in the application of Iref = 10 A.
the LTP Nyquist criterion, the LTP Nyquist contour is defined
by s ∈ (− j ωg /2, + j ωg /2). Hence, the same range is used of the impedances provides a check on the quality of the
to derive the plots of the impedances. The Bode plot of the selected truncation order by observing the relative amplitudes
harmonic impedance Zg is obtained as follows. of the different impedances. It should also be noted that
1) Zg (s) [(44) with x = g] is evaluated for s ∈ −q
Zg,c (s) = [Zg,c (−s)]∗ .
q
(− j ωg /2, + j ωg /2), s = 0. LTP Nyquist criterion is now applied, with the return-
2) Zg0 (s) gives the Bode plot for ω ∈ (−ωg /2, +ωg /2), ratio matrix Zg−1 Zc considered in the following analysis
Zg0 (s + j ωg ) gives the Bode plot for ω ∈ (ωg /2, 3ωg /2), (note that the LTP Nyquist criterion can also be applied
and so on. by considering Zc−1 Zg instead of Zg−1 Zc ). Two cases are
3) These individual Bode plots are merged together, considered: a robustly stable system, with Iref = 10 A,
giving the overall zero-harmonic impedance Bode and a strongly unstable system, with Iref = 13 A. This
plot. choice has been made in order to obtain clearer results,
The same procedure applies to Zc and to the harmonic which are more easily read from the Nyquist plots. In any
impedances of other orders. It is worth noting that the event, the same stability boundaries discussed before can be
harmonic impedances are different from zero only for the calculated with this approach. The Nyquist plot is obtained
±q
even components, i.e., Zg,c (s) = 0 for q = 1, 3, 5.. . by evaluating the eigenvalues of this return-ratio matrix for
Fig. 9 shows these impedances for the system under study: s ∈ (− j ωg /2, + j ωg /2). Then, the system is stable if the
for the grid impedance, the only relevant component is the number of counterclockwise encirclements of the point (−1, 0)
0 one, since the amplitude of all the other components is is equal to the number of right-hand plane poles of Zg−1 Zc
significantly lower, while for the converter impedance, com- included within the LTP Nyquist contour. Using (16), the poles
ponents are relevant up to the third harmonic. The evaluation of H1 H2−1 are evaluated using a contour plot. This is because

⎡ 0 ⎤
Zx (s − j (M/2)ωg ) ···
⎢ .. .. .. .. .. ⎥
⎢ . . . . . ⎥
⎢ ⎥
⎢ Zx (s − j (1)ωg )
0 −1
Zx (s) −2
Zx (s + j (1)ωg ) ⎥
⎢ ⎥
Zx (s) = ⎢
⎢ Zx1 (s − j (1)ωg ) Zx0 (s) Zx−1 (s + j (1)ωg ) ⎥ , x = c, g.
⎥
⎢ Zx2 (s − j (1)ωg ) Zx1 (s) Zx0 (s + j (1)ωg ) ⎥
⎢ ⎥
⎢ .. .. .. .. .. ⎥
⎣ . . . . . ⎦
··· Zx0 (s + j (M/2)ωg )
(44)
1522 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 5, NO. 4, DECEMBER 2017

Fig. 11. Zoomed-in view of the contour plot reported in Fig. 10 around the
poles of the return-ratio matrix Zg−1 Zc included by the LTP Nyquist contour.
(a) Iref = 10 A—stable system. (b) Iref = 13 A—unstable system.

Fig. 10. Contour plot from which the poles with Re[s] ≥ 0 of the return-
ratio matrix Zg−1 Zc are determined. (a) Iref = 10 A—stable system. (b), poles with Re[s] ≥ 0. Hence, the system is assumed to be
Iref = 13 A—unstable system. stable if the Nyquist plot does not encircle the (−1, 0) point,
and it is unstable when there are encirclements of this point.
However, as demonstrated in this paper, this assumption is not
an explicit expression for these matrices in terms of s is not generally true, and in the system under analysis, it is actually
easily obtainable analytically, but the evaluation for single incorrect. Therefore, the identification of the number of poles
values of s is more feasible. Note that the contour plot has with Re[s] ≥ 0 becomes a crucial task to perform in the
been used here only for the sake of theoretical validation of stability analysis of a system using the impedance method.
the method, and would not be feasible in practice, as discussed From a practical perspective, when the analytical model of
in the following. In fact, the contour plot allows one to choose the system is known, one possibility is to apply the procedure
a grid of values for s, such that s = p + j q, p, q ∈ R in based on the contour plot, as discussed previously in detail,
the domain of interest and for each of these values H1 H2−1 is whereas when the analytical model is not known, it is possible
evaluated [recall (13) and (14), with j ω substituted by p+ j q], to fit the measured harmonic impedances Zg and Zc using
the modulus of each element of the matrix is calculated and one of the several methods for curve fitting from Bode plots,
the largest of these is stored and used in the contour plot. The in order to obtain an estimation of the transfer functions as
idea is that as s approaches the location of a pole, the modulus function of s.
of one of the elements of the matrix H1 H2−1 will diverge and it The Nyquist plots are reported in Fig. 12, with the sta-
will be shown in red–yellow color in the contour plot, whereas ble case shown in Fig. 12(b), where it can be seen that
if s is away from a pole, the greatest modulus will be small. there are two counterclockwise encirclements of the point
Thus, with an appropriate set of values of s, the contour plot (−1, 0), ensuring that the system is stable; by contrast, the
allows a good estimation of the pole locations, as shown in unstable case is shown in Fig. 12(c), where the external
Fig. 10. Since the return-ratio matrix Zg−1 Zc depends on the counterclockwise encirclement is compensated by the internal
steady-state solution, which is a function of Iref , the poles in clockwise encirclement, and thus, no net encirclements of
both stable and unstable cases are reported. Fig. 11 shows the point (−1, 0) are present, leading to an unstable system.
a zoomed-in view of the region of interest in the Nyquist It is worth noting that the external encirclement is obtained
LTP contour. In the stable case, two poles are included inside with an infinite-closure, which provides a counterclockwise
this contour, at s = 0 and s = 903.3, while in the unstable encirclement, in accordance with the theory for infinite-
case, again two poles are included inside the contour, now closures [19], which states that when two segments of the
at s = 0 and s = 1139.2, confirming the dependence upon Nyquist diagram are diverging toward plus and minus infinity,
Iref . Thus, in both cases, two poles are present inside the respectively, and then, these two segments must be joined
Nyquist LTP contour. This is a particularly interesting feature by n semicircles (with infinite radius) in a counterclockwise
of the system under analysis. In fact, most of the impedance- fashion, where n is the number of eigenvalues of the open-
based stability approaches take for granted that for a stable loop HTF in the origin of the complex plane. In Fig. 10,
system, the open-loop transfer function does not have any it can be seen the presence of one pole in the origin, thus
SALIS et al.: STABILITY ASSESSMENT OF POWER-CONVERTER-BASED AC SYSTEMS BY LTP THEORY 1523

the time-domain simulations and with the analytical results

based on the average model is obtained.

B. Current-Injection Measurement
In the switching model, the harmonic impedances can
be obtained by injecting a small-perturbation current at the
common-connection point and measuring the relative per-
turbation currents and voltages, as shown in Fig. 2. From
Section II, it is known that the LTP Nyquist contour for
which it is relevant to calculate the LTP Nyquist plot is
s ∈ [− j ωg /2, + j ωg /2], thus for  ∈ [−ωg /2, +ωg /2].
A succession of values for  is chosen and the LTP Nyquist
plot is evaluated separately for each value. A total number of
2M + 1 separate current injections is required, which are of
the form (M is the truncation order of the system)
ĩ 1 (t) = I˜1 cos(( + (−M)ωg )t)
ĩ 2 (t) = I˜2 cos(( + (−M + 1)ωg )t)
..
.
ĩ k (t) = I˜k cos(( + (−M−1 + k)ωg )t), k ∈ [1, 2M + 1].
(45)
Each single current injection will generate the following
small-signal current and voltage perturbations in the grid and
converter systems:
+∞

p̃(t) = p̃n cos(( + nωg )t) (46)
n=0
Fig. 12. Nyquist plot. (a) One external counterclockwise encirclement with p̃(t) = ĩ g (t), ĩ c (t), ṽ o (t), respectively. Hence, when the
around the point (−1, 0). (b) Iref = 10 A—stable system, one
internal counterclockwise encirclement around the point (−1, 0). single frequency  (or  + nωg , n ∈ [−M, M]) is injected in
(c) Iref = 13 A—unstable system, one internal clockwise encirclement the system through the current injection, all the other currents
around the point (−1, 0). and voltages of the circuit will have a harmonic component at
the same injected frequency,  (or  + nωg , n ∈ [−M, M]),
plus all the other components  + kωg , k ∈ Z, k = 0 (or
the Nyquist plot has an enclosure of one counterclockwise  + kωg , k ∈ Z, k = n). Thus, for each current injection,
semicircle. the grid current will be
+∞

VII. S WITCHING M ODEL R ESULTS ĩ g1 (t) = I˜g1,n cos(( + nωg )t)
In this section, simulations are reported on the full switching n=0
model of the system. The control has been discretized and +∞

implemented in C-language and the circuit has been simu- ĩ g2 (t) = I˜g2,n cos(( + nωg )t)
lated in the PLECS-MATLAB environment, including digital n=0
computation delays as is usual in real systems. The stability ..
.
boundary is slightly different to the one evaluated based on the +∞
∗ lies in 11.4–11.5 A. Such a difference, 
average model and Iref ĩ gk (t) = I˜gk,n cos(( + nωg )t). (47)
quantified to be less than 2%, is entirely expected because of n=0
the linear modeling of the digital delays. The predicted sta- These grid small-signal currents are measured and an Fast
bility boundaries have been validated through a time-domain Fourier Transform is applied to extract the harmonic compo-
simulation and implementation of an impedance measurement nents, which are then stored in the following matrix:
technique based on current injections to practically evaluate ⎡ ˜ ⎤
the LTP Nyquist criterion with a “black-box” approach, more Ig1,(−M) I˜g2,(−M) ··· I˜gk,(−M)
⎢ I˜g1,(−M−1) I˜g2,(−M−1) · · · I˜gk,(−M−1) ⎥
oriented to practical scenarios. ⎢ ⎥
⎢ .. .. .. .. ⎥
Ig ( j ) = ⎢ . . . . ⎥
⎢ ⎥
A. Time-Domain Simulation ⎣ I˜g1,(M−1) ˜Ig2,(M−1) · · · I˜gk,(M−1) ⎦
Stable and unstable operation modes are reported in Fig. 13, I˜g1,(M) I˜g2,(M) ··· I˜gk,(M)
showing relevant currents and voltages. Good agreement with (48)
1524 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 5, NO. 4, DECEMBER 2017

re f
Fig. 13. Switching model. Currents: blue—i L1,2 (t) and red—i L1,2 (t). Voltages: blue—v g (t) and red—v o (t). Phase: blue—θ (t). (a)–(d) Stable system with
Iref = 11.4 A. (e)–(h) Unstable system with Iref = 11.5 A.

for  ∈ [1, 3, .., 49] [(15) and Theorem of Section II-B],

and is indicated by the black asterisk in Fig. 14. The num-
ber of the set of injections can be decided according to
the required resolution. In this case, two sets are enough
to describe the LTP Nyquist plot around the critical point
(−1,0). Good agreement is achieved, showing the potential
of the current-injection method as a means of estimating
harmonic impedances in practical applications.

VIII. C ONCLUSION
In this paper, two general methods are presented, based on
LTP theory, to perform stability analysis of complex NLTP
Fig. 14. Switching model. Black asterisk: calculated Nyquist plot with power systems. A case study of a single-phase network,
f  = 1, 3, 5, .., 49 Hz. Blue dots: measured Nyquist plot with f  = 1 Hz.
Red dots: measured Nyquist plot with f  = 17 Hz. Stable system with comprising a voltage-source and a current-source converter is
Iref = 10 A. considered for the development of the stability analysis. One
method is based on the eigenvalue loci of the linearized system
and requires a full knowledge of the model. The other, based
and similarly for Ic ( j ) and Vo ( j ). It then follows that: on the LTP Nyquist criterion, has the advantage that it allows
the stability analysis to be performed using small-signal cur-
Vo ( j ) = Zg ( j )Ig ( j ) rent injections: by measuring the relevant perturbed quantities,
Vo ( j ) = Zc ( j )Ic ( j ) (49) harmonic impedances can be calculated and stability assessed,
considering the system as a black-box. Simulation results are
from which the harmonic impedances are obtained by inverting provided based on both average and switching models of the
the respective current matrices in (49). This procedure must system, showing a good accuracy in the identification of the
be repeated for each value of  ∈ [−ωg /2, +ωg /2], remem- stability boundaries.
bering that the LTP Nyquist plot for  ∈ [−ωg /2, 0] is the
complex conjugate of the one for  ∈ [0, +ωg /2], so only the
evaluation of the latter is required. R EFERENCES
Fig. 14 shows the Nyquist plot obtained from the measured [1] R. D. Middlebrook, “Input filter considerations in design and application
harmonic impedances compared with the one from analytical of switching regulators,” in Proc. IEEE IAS, May 1979, pp. 366–382.
calculations (Section VI-C), for the stable operation mode at [2] E. Mollerstedt and B. Bernhardsson, “Out of control because of
Iref = 10 A. The measured LTP Nyquist plot is obtained harmonics-an analysis of the harmonic response of an inverter loco-
motive,” IEEE Control Syst., vol. 20, no. 4, pp. 70–81, Aug. 2000.
injecting two sets of perturbation currents: the first with [3] E. Mollerstedt and B. Bernhardsson, “A harmonic transfer function
f  = 1 Hz and f = /(2π) (blue dots in Fig. 14), and model for a diode converter train,” in Proc. IEEE Power Eng. Soc. Winter
the second one with f  = 17 Hz (red dots in Fig. 14). Meet., vol. 2. Jan. 2000, pp. 957–962.
[4] H. Liu and J. Sun, “Modeling and analysis of DC-link harmonic
The calculated LTP Nyquist plot is obtained by plotting the instability in LCC HVDC systems,” in Proc. IEEE 14th Workshop
eigenvalue loci of the open-loop HTF Zg ( j )−1Zc ( j ), Control Modeling Power Electron. (COMPEL), Jun. 2013, pp. 1–9.
SALIS et al.: STABILITY ASSESSMENT OF POWER-CONVERTER-BASED AC SYSTEMS BY LTP THEORY 1525

[5] M. Cespedes and J. Sun, “Renewable energy systems instability involv- [26] S. Buso and P. Mattavelli, Digital Control in Power Electronics, 2nd ed.
ing grid-parallel inverters,” in Proc. 24th Annu. IEEE Appl. Power San Rafael, CA, USA: Morgan & Claypool, 2015.
Electron. Conf. Expo. (APEC), Feb. 2009, pp. 1971–1977. [27] K. Ilves, A. Antonopoulos, S. Norrga, and H.-P. Nee, “Steady-state
[6] M. Bollen et al., “Future work on harmonics—Some expert opinions: analysis of interaction between harmonic components of arm and
Part I—Wind and solar power,” in Proc. 16th Int. Conf. Harmonics line quantities of modular multilevel converters,” IEEE Trans. Power
Quality Power (ICHQP), May 2014, pp. 904–908. Electron., vol. 27, no. 1, pp. 57–68, Jan. 2012.
[7] J. Meyer et al., “Future work on harmonics—Some expert opinions:
Part II—supraharmonics, standards and measurements,” in Proc. 16th
Int. Conf. Harmonics Quality Power (ICHQP), May 2014, pp. 909–913. Valerio Salis (S’16) received the master’s degree
[8] S. Hiti, V. Vlatkovic, D. Borojevic, and F. C. Lee, “A new control algo- in electronic engineering from the University of
rithm for three-phase pwm buck rectifier with input displacement factor Rome Tor Vergata, Rome, Italy, in 2014. He is
compensation,” in Proc. 24th Annu. IEEE Power Electron. Specialists currently pursuing the Ph.D. degree in electrical and
Conf. Rec. (PESC), Jun. 1993, pp. 648–654. electronic engineering with the Power Electronics,
[9] M. Belkhayat, “Stability criteria for AC power systems with regulated Machines and Control Group, University of Notting-
loads,” Ph.D. dissertation, Dept. Eng. Electron. Elect., Purdue Univ., ham, Nottingham, U.K.
West Lafayette, IN, USA, Dec. 1997. His current research interests include the study of
[10] A. G. J. MacFarlane and I. Postlethwaite, “The generalized Nyquist instability issues in microgrids, linear time periodic
stability criterion and multivariable root loci,” Int. J. Control, vol. 25, system analysis, and control design.
no. 1, pp. 81–127, 1977.
[11] 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. Alessandro Costabeber (S’09–M’13) received the
[12] J. Sun and K. J. Karimi, “Small-signal input impedance modeling of master’s degree (Hons.) in electronic engineering
line-frequency rectifiers,” IEEE Trans. Aerosp. Electron. Syst., vol. 44, and the Ph.D. degree in information engineering
no. 4, pp. 1489–1497, Oct. 2008. from the University of Padova, Padua, Italy, in 2008
[13] J. Sun, “Input impedance analysis of single-phase PFC converters,” IEEE and 2012, respectively.
Trans. Power Electron., vol. 20, no. 2, pp. 308–314, Mar. 2005. In 2014, he joined the PEMC Group, Department
[14] V. Salis, A. Costabeber, P. Zanchetta, and S. Cox, “A gener- of Electrical and Electronic Engineering, University
alised harmonic linearisation method for power converters input/output of Nottingham, Nottingham, U.K., as a Lecturer
impedance calculation,” in Proc. 18th Eur. Conf. Power Electron. in power electronics. His current research interests
Appl. (EPE ECCE Europe), Sep. 2016, pp. 1–7. include HVDC converters topologies, high power
[15] J. Sun, “Small-signal methods for ac distributed power systems—A density converters for aerospace applications, control
review,” IEEE Trans. Power Electron., vol. 24, no. 11, pp. 2545–2554, solutions and stability analysis of ac and dc microgrids, control and modeling
Nov. 2009. of power converters, power electronics, and control for distributed and
[16] S. Shah and L. Parsa, “On impedance modeling of single-phase renewable energy sources.
voltage source converters,” in Proc. IEEE Energy Convers. Congr. Dr. Costabeber received the IEEE Joseph John Suozzi INTELEC Fellowship
Expo. (ECCE), Sep. 2016, pp. 1–8. Award in power electronics in 2011.
[17] S. Lissandron, L. D. Santa, P. Mattavelli, and B. Wen, “Experimental
validation for impedance-based small-signal stability analysis of single-
phase interconnected power systems with grid-feeding inverters,” IEEE Stephen M. Cox received the B.A. degree in math-
J. Emerg. Sel. Topics Power Electron., vol. 4, no. 1, pp. 103–115, ematics from the University of Oxford, Oxford,
Mar. 2016. U.K., in 1986, and the Ph.D. degree in applied
[18] N. M. Wereley and S. R. Hall, “Frequency response of linear time peri- mathematics from the University of Bristol, Bristol,
odic systems,” in Proc. 29th IEEE Conf. Decision Control, Dec. 1990, U.K., in 1989.
pp. 3650–3655. From 2004 to 2006, he was a Senior Lecturer with
[19] S. R. Hall and N. M. Wereley, “Generalized Nyquist stability crite- the School of Mathematical Sciences, The University
rion for linear time periodic systems,” in Proc. Amer. Control Conf., of Adelaide, Adelaide, Australia. He was a Lecturer
May 11990, pp. 1518–1525. and then a Senior Lecturer with the School of
[20] J. B. Kwon, X. Wang, F. Blaabjerg, C. L. Bak, A. R. Wood, and Mathematical Sciences, University of Nottingham,
N. R. Watson, “Harmonic instability analysis of a single-phase grid- Nottingham, U.K., where he is currently a Reader.
connected converter using a harmonic state-space modeling method,” His current research interests include the development of applied mathematical
IEEE Trans. Ind. Appl., vol. 52, no. 5, pp. 4188–4200, Sep. 2016. techniques for the modeling of class-D amplifiers and other power-electronic
[21] J. Kwon, X. Wang, F. Blaabjerg, C. L. Bak, V. S. Sularea and C. Busca, switching devices.
“Harmonic interaction analysis in a grid-connected converter using
harmonic state-space (HSS) modeling,” IEEE Trans. Power Electron.,
vol. 32, no. 9, pp. 6823–6835, Sep. 2017. Pericle Zanchetta (M’00–SM’15) received the
[22] C. Zhang, X. Wang, and F. Blaabjerg, “Analysis of phase-locked M.Eng. degree in electronic engineering and the
loop influence on the stability of single-phase grid-connected inverter,” Ph.D. degree in electrical engineering from the Tech-
in Proc. IEEE 6th Int. Symp. Power Electron. Distrib. Generat. nical University of Bari, Bari, Italy, in 1993 and
Syst. (PEDG), Jun. 2015, pp. 1–8. 1997, respectively.
[23] V. Salis, A. Costabeber, P. Zanchetta, and S. Cox, “Stability analysis of In 1998, he became an Assistant Professor of
single-phase grid-feeding inverters with PLL using harmonic linearisa- Power Electronics with the Technical University
tion and linear time periodic (LTP) theory,” in Proc. IEEE 17th Workshop of Bari. In 2001, he became a Lecturer in con-
Control Modeling for Power Electron. (COMPEL), Jun. 2016, pp. 1–7. trol of power electronics systems with the PEMC
[24] H. Sandberg, E. Mollerstedt, and B. Bernhardsson, “Frequency-domain Research Group, University of Nottingham, Notting-
analysis of linear time-periodic systems,” IEEE Trans. Autom. Control, ham, U.K., where he is currently a Professor in
vol. 50, no. 12, pp. 1971–1983, Dec. 2005. control of power electronics systems. He has authored or co-authored over
[25] D. M. Van de Sype, K. De Gusseme, A. P. Van den Bossche, and 260 peer reviewed papers. His current research interests include control of
J. A. Melkebeek, “Small-signal Laplace-domain analysis of uniformly- power converters and drives, matrix, and multilevel converters.
sampled pulse-width modulators,” in Proc. IEEE 35th Annu. Power Prof. Zanchetta is the Chair of the IAS Industrial Power Converter Com-
Electron. Specialists Conf. (PESC), vol. 6. Jun. 2004, pp. 4292–4298. mittee IPCC.