4568
9, SEPTEMBER 2014
Accurate Capacitor Voltage Ripple Estimation and
Current Control Considerations for Grid-Connected
Modular Multilevel Converters
Michail Vasiladiotis, Student Member, IEEE, Nicolas Cherix, Student Member, IEEE, and Alfred Rufer, Fellow, IEEE
Abstract—In this paper, the analytical solution for the submod-
ule voltage ripple equations of a modular multilevel converter
(MMC) is derived, based on the knowledge of the external volt-
age/current magnitudes, and enhancing a concept previously pre-
sented in the literature. In order to achieve high accuracy, all pas-
sive elements of the converter, common-mode voltage injection as
well as intentionally imposed circulating current harmonics are
taken into consideration. The natural charge level mechanism of
the capacitor voltages is also explained. As application examples,
the three- as well as the two-phase grid-connected MMC cases are
chosen. The control of line and circulating currents is also dis-
cussed and two respective independent feedback loops are formed.
The concept of fictive-axis emulation is tailored for the two-phase
MMC case, in order to achieve vector control of the line current
and therefore straightforward desired injection of active and reac-
tive power. Finally, the development of a reduced-scale laboratory
prototype is presented and a full set of experimental results are
provided, verifying the aforementioned concepts.
Index Terms—Decoupled current control, fictive axis emulator
(FAE), modular multilevel converter (MMC), prototype, voltage
ripple estimation.
I. INTRODUCTION
HE modular multilevel converter (MMC) has attracted a lot
T of industrial and academic interest during the last decade,
due to the superior properties over conventional topologies for
high-voltage and high-power applications [1]. Several applica-
tion fields have been considered, such as HVDC systems as well
as medium voltage electric drives [2]–[4]. However, the unique
features of this converter type, such as the series-connection
of capacitors as well as the existence of branch inductors and
the topology-specific circulating currents, require special atten-
tion for the comprehension of its physical behavior. Therefore,
intensive research has been conducted on the dynamic model-
ing and the development of control solutions of the converter
inner dynamics [5]–[8], as well as the tailoring of several well-
established modulation methods for such a system [9]–[11].
Manuscript received March 25, 2013; revised August 2, 2013 and September
18, 2013; accepted October 8, 2013. Date of current version April 30, 2014. This
paper has been presented in part at the 15th International Power Electronics and
Motion Control Conference and Exposition (EPE-PEMC ECCE Europe), Novi
Sad, Serbia, September 4–6, 2012. Recommended for publication by Associate
Editor J. R. Rodriguez.
The authors are with the Laboratory of Industrial Electronics, École
Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland (e-mail:
michail.vasiladiotis@epfl.ch; nicolas.cherix@epfl.ch; alfred.rufer@epfl.ch).
Digital Object Identifier 10.1109/TPEL.2013.2286293
0885-8993 © 2013 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.
IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO.
In general, it can be argued that the choice for control hard-
ware configuration and hierarchy for an MMC is closely related
to the respective choice for control system design and imple-
mentation. If the balancing of the individual submodule voltages
is guaranteed by a localized mechanism, such as the one de-
scribed in [1], it is preferable that the global MMC control will
not need to acquire their measured values. Toward this direc-
tion, the submodule capacitor voltage ripple estimation is a very
interesting feature. It implies the simplification of the commu-
nication scheme between the different hardware components, in
the sense of exchanged data amount reduction. This is beneficial
especially for a high number of submodules [12]. The authors
of [13] have extended this concept theoretically showing that
when the estimated values are utilized in the selection of the
modulation references, no additional balancing-cascaded loops
are necessary even if undesirable unbalances occur between the
converter branches. In addition, the knowledge of the accurate
submodule voltage variation is also significant for converter de-
sign purposes. Indeed, the resulted analytical expressions can be
utilized for the dimensioning of the required MMC capacitive
storage.
This paper comprises two main parts. In a first step, it aims
at the generalization and enhancement of the voltage ripple es-
timation concept presented in [12] and [14], using an accurate
model and making it possible to be extended to any MMC con-
figuration. Therefore, any strong assumptions are avoided and
all passive elements as well as intentionally injected common-
mode voltage and circulating current harmonics are taken into
consideration. In a second step, current control considerations
for grid-connected MMCs are made. More specifically, two in-
dependent feedback loops are formulated for the line and cir-
culating currents, respectively, based on the fact that there is no
coupling between these two quantities. It is noted that for the
two-phase case,1 the concept of fictive axis emulation (FAE)
is chosen to achieve line current vector control. However, an
appropriate modification of the algorithm is necessary, in order
to account for the branch inductance/resistance existence. As
a remark, the branch resistances are considered to be constant
and symmetric in the frame of this paper, although it has been
pointed out that their actual dependence on the time-varying
modulation references can affect significantly the equal split-
ting of the line current among the converter branches [15].
1 This refers to a single-phase structure with two parallel-connected legs and
is used to distinguish it from the single-phase-leg inverter considered by many
authors.
VASILADIOTIS et al.: ACCURATE CAPACITOR VOLTAGE RIPPLE ESTIMATION AND CURRENT CONTROL CONSIDERATIONS 4569

ent multilevel pulse width modulation (PWM) frequency. The

latter leads to an accurate average approximation of the branch
quantities, motivating the modeling of the converter utilizing a
unique equivalent submodule per branch. The midpoint of the
common link is grounded for facilitation of the analysis. When
the MMC is connected to a dc line, LS and RS represent the
filter or line inductance and uS is a dc voltage source. In the
case of a transformer-less direct ac/ac converter they denote the
grid choke that handles short-circuit currents [16], and uS is the
single-phase ac grid voltage. On the three-phase side, LT and
RT represent the transformer or machine leakage inductance and
cable resistance. In addition, uT is the grid or motor back emf,
respectively. Finally, the branch inductances and resistances are
symboled with L and R accordingly.

A. Derivation of Total Submodule Voltage Ripples

In order to obtain the analytical expression for the submod-
ule capacitor voltages of the MMC, the fundamental capacitor
current equation is considered, taking as an example the upper
branch of a converter phase leg. This relates the current icu that
flows through the equivalent capacitance Cbranch with the sum
of all the submodule voltages Σuusm in one converter branch
dΣuusm Csm
Cbranch = icu ⇔ dΣuusm = mu iu dt
dt N
Csm
⇔ dΣuusm Σuusm = uiu iu dt. (1)
N
In (1), N denotes the number of submodules in one converter
Fig. 1. (a) Multiphase MMC and (b) Detailed representation of a converter branch, each one of which having a capacitance of Csm . The
branch with one equivalent submodule.
quantity mu represents the normalized modulation reference for
the upper branch and uiu is the voltage applied by the series con-
The main contributions of this paper can be therefore sum- nection of the branch submodules, which are PWM controlled.
marized as follows: 1) documentation and comprehensive il- By integrating (1), the following expression is derived in (2),
lustration of the natural charge level mechanism of the branch where the appearing integration constant ΣE0sm refers to the dc
capacitor voltages; 2) augmentation of the branch power equa- value of the total submodule stored energy
tions in the sense of considering the consumed active/reactive  
Csm Csm
power on the respective resistance and inductance along with (Σuusm )2 − (Σuusm 0 )2 = uiu iu dt = puc dt.
2N 2N  
their experimental verification; 3) adaptation of the FAE con-
ΣE 0s m
cept for two-phase grid-connected MMCs and its experimental
(2)
evaluation; and 4) experimental demonstration of a decoupled
It is therefore clear that the product uiu iu corresponds to the
line/circulating current control for the cases of grid-connected
power contribution of the submodules to the branch, and its
three- and two-phase MMCs.
integral forms the total stored capacitive energy. This equation
This paper is organized as follows: Section II briefly sum-
is the basis for the voltage estimation of the capacitor voltages.
marizes the modeling principles of an MMC and derives the
Taking into consideration Fig. 1(a), the following relation (3)
equations that form the basis for the generalized submodule
is formed for the capacitive instantaneous power
voltage ripple expressions. Section III gives the analytical so-
lutions for the example cases of a three- and two-phase MMC. puc = uiu iu = (uU − uA − uz u ) iu =
Section IV discusses the decoupled line and circulating current  

diu
control. Finally, Section V describes the laboratory prototype = uU − (uv + ucm ) − iu R + L iu
dt
and presents several experimental results, leading to Section VI,
which concludes this paper. diu
= [uU − (uv + ucm )] iu − i2u R − L iu (3)
     dt  
p p
II. MODELING PRINCIPLES OF THE MMC b ra n ch R
pL

In Fig. 1(a), a structural representation of a three-phase MMC where the phase voltage uA comprises the fundamental fre-
is illustrated. As expected in a practical application, a large quency (uv ) plus a common-mode (ucm ) components. The
number of series-connected submodules result in a high appar- terms pbranch , pR , and pL correspond to the total branch power,
4570 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 9, SEPTEMBER 2014

the losses in the branch resistance as well as the reactive power

consumed by the inductor, respectively. For the lower branch,
the power and energy equations become as follows:

plc = uil il = (uA − uL − uz l ) il

 

dil
= −uL + (uv + ucm ) − il R + L il
dt
dil
= [−uL + (uv + ucm )] il − i2l R − L il . (4)
      dt 
pb ra n ch pR
pL

The integration finally gives the respective capacitive energy

variation in upper and lower branches. Therefore, it becomes
evident that (2) can be used for the accurate estimation of the
submodule voltage ripples. It is noted that these equations are
valid for any given dc/ac or ac/ac MMC configuration. The
difference lies in the desirable shapes of the voltage uS , current
iS (dc or ac, respectively) as well as the intentional imposition Fig. 2. Illustrative demonstration of the branch capacitor voltage charge level
of so-called circulating currents within the converter phase legs, control mechanism.
which will cause different frequency components in the branch
currents.
In the next section, two specific examples are chosen in or- a direct modulation reference scheme without further control
der to apply this theory, namely the three-phase and two-phase actions, justifying the attributed title of a natural mechanism.
dc/ac MMC cases. As a remark and for the simplification of the
analysis, the voltage drop on RS and LS is disregarded in the
III. APPLICATION ON GRID-CONNECTED MMCS
modeling procedure. This implies that the voltage uU L is con-
sidered to be known. The latter forms a fair assumption, since The application areas of the three-phase grid-connected
this magnitude can be either actively adjusted (in the case of MMC are wide and include HVDC systems, active front ends, as
a current-controlled single-phase ac grid), or directly measured well as medium voltage electric drives. As far as the two-phase
(in typical dc/ac systems). case is concerned, it can be found as an alternative proposal
Remark 1: The integration constant ΣE0sm of (2) is of signifi- for back-to-back interties (e.g., the interconnection of the three-
cant importance and is not to be confused with the inner current phase industrial grid with the single-phase railway supply).
control of an MMC, since their action mechanisms are differ-
ent. As depicted in Fig. 1(b), the equivalent submodule should
A. Analytical Solution of the Power Equations
provide a specific voltage uiu . Once the voltage reference u∗iu
is defined as a result of the two current loops described in the By exploring Fig. 1(a), it is clear that each phase leg forms
following sections, the choice of mu can be considered further- two internal ac loops and one external loop, respectively. In the
more as a degree of freedom for the capacitor charge level, since latter loop, which in the studied case can be considered as an
its input/output power equilibrium is dictated by the physics of independent “dc Loop” for each phase, the contribution of iS to
the system. Indeed, the following relation (5) holds: each branch current would ideally be iS /m. However, additional
undesirable currents are also flowing within a converter phase
u∗iu leg, if the capacitor voltage ripples are not taken into account in
mu = (5)
Σuusm 0+ Σũusm the control and modulation scheme. These currents are known
as circulating currents and their mathematical definition varies
where Σũusm denotes the sum of the branch capacitor voltage
in the literature. In this paper, it is considered that the current
ripples. The normalization constant Σuusm 0 can be freely cho-
due to the dc terminals is part of the circulating current without
sen to charge/discharge the capacitor voltages as intended. This
any loss of generality. This means that the latter will essentially
is better illustrated in the example of Fig. 2. By increasing the
comprise dc and ac components and can therefore be defined as
normalization constant at t = 40 ms, the modulation reference
(6) in its general form
mu obviously decreases. The instantaneous difference of the
voltage sum (uiu + uil ) from the dc-link voltage uU L is capa- ∞

ble of inducing a specific amount of current is /m in the branch icirc = Icirc,0 + îcirc,n cos(nωt + θn ). (6)
(m being the number of parallel-connected phase legs), which in    n =1
i S /m
turn charges the capacitors at a higher level. When the new equi-
librium is reached, the quantities mu and Σuusm 0 are changed, in
order to maintain unaltered voltages uU O and uA , respectively. Furthermore, the common-mode voltage of the three-phase
It should be noted that this concept is valid even in the case of MMC can also bear dc and ac components and can be defined
VASILADIOTIS et al.: ACCURATE CAPACITOR VOLTAGE RIPPLE ESTIMATION AND CURRENT CONTROL CONSIDERATIONS 4571

similarly as follows: Accordingly, the voltage drop on the branch inductances and
resistances will finally become as follows:
∞
 diu
ucm = Ucm ,0 + ûcm ,n cos(nωt + ξn ). (7) uz u = Riu + L =
dt
n =1 

îT
The line-side fundamental harmonic magnitudes are given by = R Icirc,0 + cos(ωt + φ) + îcirc,2 cos(2ωt + θ2 )
2
(8), where a phase angle of φ between the converter voltage and
the line current is assumed. In addition, the branch currents are
defined as (9), assuming that the branch impedances are sym-
metrical, and therefore, the line current splits equally between
the upper and lower branches of the same phase leg
uv = ûv cos(ωt), iT = îT cos(ωt + φ) (8)
iT iT
iu = + icirc , il = − + icirc (9)
2 2
At this point, it is also assumed that the circulating current
can be explicitly controlled. This is verified in the next section.
1) Three-Phase MMC: As stated in [17], if the circulating
current is chosen to bear even harmonics, there will be no unbal-
ance between the branches of the same phase leg. Other works
have also shown that it is beneficial to intentionally inject a
second-order harmonic, in terms of capacitor voltage ripple re-
duction [14], [16], [17], which in turn leads to decreased needs
for embedded capacitive energy. Hence, the desirable circulating
current becomes as follows:
icirc = Icirc,0 + îcirc,2 cos(2ωt + θ2 ). (10)
In such a three-phase application, the common-mode voltage

îT
+ ωL − sin(ωt + φ) − 2îcirc,2 sin(2ωt + θ2 ) . (12)
2
The instantaneous branch power is finally modified as in
(13a), whereas the power terms referring to the respective branch
resistance and inductance are obtained as in (13b) and (13c) as
shown at the bottom of the page.
2) Two-Phase MMC: The two-phase MMC can be treated
similarly. It is obvious that adding a second-order harmonic
component to the circulating current would affect the dc-link
current [16], [19], although this might not always be a disadvan-
tage depending on the application. On the other hand, choosing
a plain dc circulating current leads to constant power transfer to
the dc link, canceling therefore the intrinsic pulsating behavior
of any single-phase system. This is useful when choosing to
avoid placing passive resonant filters, which are widely used
in railway interties, adding cost, volume, and hardware com-
plexity. Furthermore and taking into account that no third-order
common-mode voltage injection can be added to a single-phase
system for increase of the modulation index, (10)–(12) become
as follows:
icirc = Icirc,0 , ucm = 0 (14)
can be chosen to correspond to a typical third-harmonic injection 

in (11), which results in the increase of the maximum modulation diu îT
uz u = Riu + L = R Icirc,0 + cos(ωt + φ)
index dt 2

1 îT
ucm = ûcm ,3 cos(3ωt + ξ3 ) = − ûv cos(3ωt). (11) − ωL sin(ωt + φ) (15)
6 2

ûV îT uU îT ûV îcirc,2

pbranch = uU Icirc,0 − cos φ + cos(ωt + φ) − ûV Icirc,0 cos(ωt) − cos(ωt + θ2 )
4 2 2
ûV îcirc,2 ûV îT ûV îT
+ cos(ωt − θ2 ) + uU îcirc,2 cos(2ωt + θ2 ) − cos(2ωt + φ) + cos(−2ωt + φ)
12 4 24
ûV îcirc,2 ûV Icirc,0 ûV îT ûV îcirc,2
− cos(3ωt + θ2 ) + cos(3ωt) + cos(4ωt + φ) + cos(5ωt + θ2 ) (13a)
2 6 24 12

î2circ,2 î2 îT îcirc,2 î2
2
pR = R Icirc,0 + + T + îT Icirc,0 cos(ωt + φ) + cos(−ωt + φ − θ2 ) + T cos(2ωt + 2φ)
2 8 2 8

îT îcirc2 î2circ,2
+2Icirc,0 îcirc,2 cos(2ωt + θ2 ) + cos(3ωt + φ + θ2 ) + cos(4ωt + 2θ2 ) (13b)
2 2

îT Icirc,0 îT îcirc,2
pL = ωL − sin(ωt + φ) + sin(−ωt + φ − θ2 ) − 2îcirc,2 sin(2ωt + θ2 )
2 4

î2T 3îT îcirc,2
− sin(2ωt + 2φ) − sin(3ωt + φ + θ2 ) − Icirc,2 sin(4ωt + 2θ2 )
2
(13c)
8 4
4572 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 9, SEPTEMBER 2014

B. Choice of Circulating Current Components sate for this frequency, in order to suppress it significantly [16]
The explicit relations for the circulating current compo- ûV îT
nents can now be derived by taking into account the fact that uU îcirc,2 cos(2ωt + θ2 ) − cos(2ωt + φ) = 0
4
puc = pbranch − pR − pL . The capacitor power should be of zero
average, in order to maintain a stable charge. Therefore, the con- θ 2 =φ ûV îT
⇒ îcirc,2 = . (17)
stant terms in the respective mathematical expression need to 4uU
compensate for each other. By choosing the angle of the second-order harmonic compo-
1) Three-Phase MMC: The aforementioned considerations nent as θ2 = φ, a constant value for îcirc,2 is achieved in (17)
can be formulated as follows in (16) for the case of the three- and the steady-state analytical expressions are simplified signif-
phase MMC icantly. The remaining power terms of second-harmonic order
in (13a)–(13c) will then be significantly smaller, due to their
  scaling factors (1/24, R and ωL).
ûV îT î2circ,2 î2T After the elimination of the terms associated with (17) the
uU Icirc,0 − cos φ − R Icirc,0 −
2
− = 0.
4 2 8 power equations can be finally integrated, in order to acquire
(16) the explicit relations for the total capacitive energy fluctuation,
In the above relation, the first two terms denote the input and which is stored in the upper branch of each phase leg. For
output powers. The three last terms refer to the resistive losses the sake of completeness, this long equation is provided in the
within the branch, which are caused by the squared dc (Icirc,0 ) bottom of the page (18), taking into account the integration
and ac rms values of the circulating current (icirc,2 ), as well as constant corresponding to the amount of dc capacitive stored
half of the respective output current value (iT /2). This means energy in steady-state operation. The respective formulas for
that if Icirc,0 is chosen according to power equilibrium, the total the lower branches and other phase legs can be derived in a
branch submodule capacitor voltages will be stable without the similar manner.
need for additional actions. In order to compensate for the losses 2) Two-Phase MMC: The two-phase case can be obviously
within the converter branch, Icirc,0 is evaluated by solving the considered as a respective three phase without imposition
previous second-order equation (16). of second-order harmonic circulating current component and
Regarding the values of îcirc,2 and θ2 , different possibilities common-mode voltage injection. Therefore, the equations can
exist. By examining the power equations (13a)–(13c), it can be be modified by setting the respective parameters îcirc,2 and θ2
observed that the second harmonic (2ω) is the dominant compo- to zero as well as reintroducing the compensated term of (17).
nent of the submodule voltages after the one of the fundamental This results eventually in (19) and (20) provided at the bottom
frequency (ω). Thus, a reasonable choice would be to compen- of the page.
  uU îT ûV Icirc,0 RîT Icirc,0 LîT Icirc,0
Ecu = ΣEsm
0
+ sin(ωt + φ) − sin(ωt) − sin(ωt + φ) − cos(ωt + φ)
2ω ω ω 2
RîT îcirc,2 ûV îcirc,2 LîT îcirc,2 uˆV îcirc,2
+ sin(−ωt + φ − θ2 ) − sin(ωt + θ2 ) − cos(−ωt + φ − θ2 ) + sin(ωt − θ2 )
2ω 2ω 4 12ω
RIcirc,0 îcirc,2 Rî2 Lî2
− sin(2ωt + θ2 ) − T sin(2ωt + 2φ) − T cos(2ωt + 2φ)
ω 16ω 16
ûV îT
− LIcirc,0 îcirc,2 cos(2ωt + θ2 ) − sin(−2ωt + φ)
48ω
RîT îcirc,2 ûV îcirc,2 LîT îcirc,2 ûV Icirc,0
− sin(3ωt + φ + θ2 ) − sin(3ωt + θ2 ) − cos(3ωt + φ + θ2 ) + sin(3ωt)
6ω 6ω 4 18ω
Lî2circ,2 Rî2circ,2 ûV îT ûV îcirc,2
− cos(4ωt + 2θ2 ) − sin(4ωt + 2θ2 ) + sin(4ωt + φ) + sin(5ωt + θ2 ) (18)
4 8ω 96ω 60ω
uU îT îT Icirc,0
pcu = cos(ωt + φ) − ûV Icirc,0 cos ωt + ωL sin(ωt + φ) − RîT Icirc,0 cos(ωt + φ)
2 2
ûV îT ωLî2T Rî2
− cos(2ωt + φ) + sin(2ωt + 2φ) − T cos(2ωt + 2φ) (19)
4 8 8
  uU îT ûV Icirc,0 RîT Icirc,0 LîT Icirc,0
Ecu = ΣEsm0
+ sin(ωt + φ) − sin ωt − sin(ωt + φ) − cos(ωt + φ)
2ω ω ω 2
ûV îT Lî2 Rî2
− sin(2ωt + φ) − T cos(2ωt + 2φ) − T sin(2ωt + 2φ) (20)
8ω 16 16ω
VASILADIOTIS et al.: ACCURATE CAPACITOR VOLTAGE RIPPLE ESTIMATION AND CURRENT CONTROL CONSIDERATIONS
TABLE I
STUDIED SYSTEM PARAMETERS
Fig. 3. Three-phase MMC: Numerically calculated sum of the submodule
voltage ripples in upper branch of phase leg A in regard with the estimated
value (upper figure). Middle figure shows the sum of submodule voltages in
upper and lower branches. The circulating current components are shown in the
lower figure.
C. Simulation Results
Both the three- and the two-phase systems have been simu-
lated using the parameters summarized in Table I, which regard
the nominal values of a reduced-scale implemented laboratory
setup. The converters are current controlled in a closed-loop
manner according to the considerations and implementation,
which is developed in the next section. The results are given
in Figs. 3 and 4, respectively. More specifically, the upper fig-
ures show the estimated as well as the numerically calculated
values of the total submodule capacitor voltage ripples, which
correspond to the upper branch of phase leg A for both cases.
It can be observed that the two curves almost coincide. The
middle figures illustrate that both upper and lower branches are
balanced in terms of voltage level and shifted by π. Finally,
the lower figures depict the respective values of the circulat-
ing current components. The latter comprises a dc as well as
a superimposed ac components with the same amplitude at the
4573
Fig. 4. Two-phase MMC: Numerically calculated sum of the submodule volt-
age ripples in upper branch of phase leg A in regard with the estimated value
(upper figure). Middle figure shows the sum of submodule voltages in upper
and lower branches. The circulating current components are shown in the lower
figure.
second-harmonic frequency for the three-phase MMC. On the
contrary, only a dc value is considered for the two-phase case.
In order to demonstrate the suppression of the second-order har-
monic power oscillation, which also leads to a branch energy
variation reduction, the same capacitance values were utilized
for both simulations. It becomes therefore obvious that in the
case of the pure dc circulating current and for the same in-
put power, a larger voltage ripple can be observed in the total
capacitor voltages.
D. Capacitive Storage Requirements
The solution of the branch capacitor energy fluctuation in
(18) presents another significant contribution. It can be used
to accurately identify the capacitive storage requirements for
a specific converter design, extending the results introduced
by [1]. The analytic calculation of the capacitor energy variation
ΔEsm would lead to complicated expressions that cannot be
easily handled. Instead, the global minimum and maximum of
(18) within one period of the grid frequency can be evaluated
numerically for any given operating point. For an allowable
voltage deviation kr from its average value (ripple factor), the
following equation can be utilized for the calculation of the
4574 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 9, SEPTEMBER 2014

TABLE II where
CAPACITIVE DESIGN NUMERICAL CALCULATIONS ⎛ ⎞ ⎛ ⎞
Req 1
− 0 − 1
A=⎝ Leq ⎠ , B = ⎝ Leq Le q ⎠,
0 −R
L
− 2L
1
− 2L
1

⎛ ⎞
2
− 0
⎜ Leq ⎟
D=⎝
1 ⎠
required capacitance [1] and .
0 −
2ΔEsm ΔEsm 2L
Csm = = (21)
− ucap,m in
u2cap,m ax 2 2kr u2cap,avg The fact that A is a diagonal matrix shows that a decoupled
current control of these two magnitudes is at hand. This state-
where the maximum submodule capacitor voltage is defined ment is quite important and implies that in fact the input power
as ucap,m ax = ucap,avg (1 + kr ) and the respective minimum as can be controlled independently from the output power [15]. In
ucap,m in = ucap,avg (1 − kr ). the studied case, the circulating current has been modeled to
Considering a balanced utilization of the submodules within bear a dc component associated with the output power. In the
one branch in terms of voltage sharing, the individual submod- case where the common terminals are connected to a single-
ule energy fluctuation can be expressed as ΔEsm = ΔEc /N , phase grid, such a property also holds with an appropriate and
linking therefore (21) with (18). equivalent definition of icirc .
Table II shows the results for the numerical calculations per- Fig. 5 shows the block diagrams for the proposed line and
formed for the cases of a pure dc circulating current as well as circulating current control loop implementations. The block
an imposed second-order harmonic. The simulated values are Gpe (s) represents the first-order transfer function which mod-
taken from the studied case of Table I. It is clearly observed line
els the application-specific delays Tpe circ
and Tpe , caused by the
that the deliberate imposition of the harmonic component icirc,2 modulator, measurement and computational system as well as
achieves a capacitive reduction in the range of 30%. However, sampling time [20].
the branch current value will be higher, leading to an increased The two respective system transfer functions Gline S (s) and
requirement for installed switching power [18]. Gcirc (s) are expressed as follows:
S
Remark 2: The branch capacitor energy fluctuation ΔEc can
be estimated with great accuracy, given the high apparent branch 1 1
Gline
S (s) = , Gcirc
S (s) = (25)
PWM frequency. Very low switching frequencies, however, can Req + sLeq 2 (R + sL)
lead to small inaccuracies in the individual capacitor energy vari- which shows that they are typically of first order. This is similar
ation estimations ΔEsm . In a real application, the ripple margin to [21] but considering the branch resistances as well.
kr can be set higher to account for such imperfections. Taking
into consideration the nonidealities of the balancing algorithms
A. Line Current Control
is beyond the scope of this paper.
The line current control can be performed as for any grid-
IV. LINE AND CIRCULATING CURRENT CONTROL connected voltage source converter, according to Gline
S given
by (25). Two different methods have been chosen for the cases
By formulating the upper and lower inner ac voltage loop of the three- and two-phase systems and are described in the
laws in Fig. 1(a), (22) and (23) can be obtained, respectively following.
 u   diu
1) Three-Phase MMC: A proportional-resonant (PR) con-
S
− + ud + uiu + Riu + L + uA = 0 (22) troller has been chosen for the three-phase MMC case, which is
 2   dt
expressed by (26) in a generalized manner [23]
−u U
 u   dil
 line
2Kih ωc s
−
S
+ ud + uil + Ril + L − uA = 0 (23) Gline line
P R (s) = Kp + . (26)
s2 + 2ωc s + (hω)2
 2   dt h=2n +1
uL
The transfer function of (26) comprises a proportional plus
where uA = RT iT + LT didtT
+ uT + ucm . several resonant terms. These can be tuned to control odd har-
As only the fundamental harmonic of the line current needs monic frequencies (hω), which are the most prominent in a
to be controlled, the common-mode voltage component can be typical current spectrum [22], [23]. This is especially beneficial
neglected in the disturbance vector. The continuous-time state- in cases of harmonically polluted networks, where an MMC
space representation can be formulated by considering the states might also be required to provide active filtering services. The
x = (iT icirc )T , the manipulated variables u = (uiu uil )T cutoff frequency ωc is utilized in order to avoid the practical sta-
as well as the disturbances w = (uT uU L )T . Defining fur- bility, real-time implementation, as well as parameter (such as
thermore Leq = L + 2LT and Req = R + 2RT , the following grid frequency) variation issues associated with a theoretically
relation is obtained: infinite gain at the chosen ac frequencies [22], [23]. An addi-
tional advantage of the PR-controller is the fact that grid voltage
ẋ = Ax + Bu + Dw (24) disturbance feed-forwarding is not necessary, in contrast to the
VASILADIOTIS et al.: ACCURATE CAPACITOR VOLTAGE RIPPLE ESTIMATION AND CURRENT CONTROL CONSIDERATIONS
Fig. 5. Current control loop block diagrams for: (a) line current and (b) circu-
lating current.
Fig. 6. FAE concept for line current vector control of a two-phase MMC.
Fig. 7. Time-domain MMC current control implementation with estimation
of capacitor voltage ripples: (a) three-phase and (b) two-phase converter cases.
case of a rotating reference frame current controller where such
an action is usually preferred. This is illustrated in Fig. 5(a).
2) Two-Phase MMC: A PR control would be suitable for the
two-phase MMC as well. In this paper, however, an alternative
solution is presented for the case where the rotating reference
frame implementation is preferred. In order to imitate the be-
4575
havior of a three-phase system and achieve a straightforward
active/reactive power regulation through a current vector con-
trol scheme, an FAE is utilized [24]. The general idea behind
the concept is to utilize the β-component of the phase voltage
uβT as well as the vector control output uβA B , in order to gen-
erate the imaginary component of the line current iβT by means
of the system transfer function itself. The block diagram of the
algorithm is depicted in Fig. 6. However, a modification needs
to take place, in order to account for the branch inductances and
resistances as well. Thus, the quantities LFAE and RFAE for the
two-phase structure are given as follows:
LFAE = Leq , RFAE = Req . (27)
It is noted that the imaginary component of the grid phase
voltage uβT can be generated through the use of a second-order
generalized integrator (SOGI) [25].
B. Circulating Current Control
For the circulating current control, a proportional term Kpcirc
has been implemented as the transfer function of Gcirc P (s).
Fig. 5(b) shows that in this case, a feed-forward disturbance
rejection signal of the dc-link voltage uU L is also used.
By using a proportional term for the circulating current con-
trol, a steady-state error will appear between the reference and
the measured values, especially in the case where a sinusoidal
value has to be tracked. For the elimination of this error, two
more terms could be added to the respective controller transfer
function. Therefore, it would consist of a proportional term, an
integral action for controlling the dc part of the circulating cur-
rent as well as a resonant term for the imposed second-order
harmonic component. This is shown in the following equation:
Kicirc circ
2Ki2 ωc s
Gcirc circ
PIR (s) = Kp + + 2 . (28)
s s + 2ωc s + (2ω)2
In such a case, the rejection of the disturbance uU L in the
sense of feed-forwarding would not be necessary. It is noted,
however, that such high integral actions in both line and circu-
lating current controllers might eventually threaten the system
stability and have not been considered further throughout this
paper.
C. Simulation Results
The dynamic performance of the closed-loop system has been
tested with the parameters taken from Table I, both for the cases
of the three- and two-phase MMC. The time-domain imple-
mentation of the current controllers with the associated signal
processing blocks is depicted in Fig. 7.
The control of line and circulating currents of the three-phase
MMC is shown in Fig. 8, where for reasons of graphical clarity
only the first phase is depicted. At t = 120 ms and while the line
current of phase A is at its peak (≈24 A), a power flow reversal is
executed to an active current of îTA = 22 A. This is immediately
followed by an injection of reactive power at t = 160 ms, which
corresponds to a current component with a peak value of 15 A
in rotating reference frame. It can be clearly observed that both
controllers bring the system to steady state after a few cycles,
4576 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 9, SEPTEMBER 2014

Fig. 8. Three-phase MMC system response to consecutive current reference

d from 24 to −18A at t = 140 ms, followed by b) iq from
step changes: a) ire f re f
0 to −15A at t = 160 ms. Only phase-leg A-related magnitudes are depicted.

Fig. 10. Multiphase MMC prototype.

Fig. 9. Two-phase MMC system response to consecutive current reference

d from 24 to −18A at t = 140 ms, followed by b) iq from
step changes: a) ire f re f
0 to -15A at t = 160 ms. Only phase-leg A-related magnitudes are depicted.

without threatening the stability. The dc-bus current iS is also

illustrated, which is clearly a continuous current, since the ac
components of the three-phase circulating currents add up to
zero.
The same test has been performed for the case of the two-
phase MMC and the results are displayed in Fig. 9. Here, the
effectiveness of the FAE is also well demonstrated. The orthog-
onal line current component iβT , acts in fast and accurate way,
following closely all the transients of the actual current iαT . The
circulating currents of both phase legs contain only a dc compo-
nent, and therefore lead to a constant power transfer, canceling
therefore the second-order harmonic oscillating behavior of the
dc-link current iS , which is also therefore a continuous quantity.
Fig. 11. Simplified scheme of the proposed implemented MMC control plat-
form functionality.
V. EXPERIMENTAL EVALUATION
In this section, a description of the implemented prototype
is presented and several experimental results are provided. The
reduced-scale laboratory setup, which has been used for the val-
idation of the proposed concepts, is shown in Fig. 10. The design
has been focused on the versatility of the implementation, which
facilitates the reconfiguration toward the examination of several
VASILADIOTIS et al.: ACCURATE CAPACITOR VOLTAGE RIPPLE ESTIMATION AND CURRENT CONTROL CONSIDERATIONS
Fig. 12. Experimental results for a three-phase MMC hardware configuration.
modular topologies, including isolated dc/dc converters [26] as
well as converters with integrated storage elements [27]. Each
rack constitutes a phase leg consisting of eight submodules, a
branch current measurement board as well as the two branch
inductances. The submodules have been designed to comprise
four power switches, giving the user the option to switch be-
tween a half- or full-bridge implementation. In addition, the
submodules provide measurements for their capacitor voltages.
The backplanes of the rack-mounted boards carry all gate sig-
nals, measurements, and power supplies.
The converter control is performed through a custom-
designed hardware platform. The communication between the
various control units is handled by a common parallel bus, un-
der the supervision of a central microcontroller which is pro-
grammed in C. The microcontroller executes all higher level
control functions by receiving the system voltage and current
4577
Fig. 13. Experimental results for a two-phase MMC hardware configuration.
measurements, which are processed in an FPGA-based board.
The latter is also controlling the relays, which are used for con-
verter precharging and discharging operations.
The localized lower level modulation and submodule volt-
age balancing algorithms are carried out in dedicated VHDL-
programmed FPGA control boards. Each of these boards corre-
sponds to one phase leg, thus following a modularized concept.
The FPGAs receive data regarding the calculated modulation
references from the microcontroller. They are also responsible
for retrieving the capacitor voltage measurements and branch
current polarities from the converter. According to a sorting al-
gorithm, an active selection process dynamically distributes the
gate control signals to the switches [1]. These FPGAs are fi-
nally in charge of ensuring converter protection, in cases where
overvalues are detected.
4578
Fig. 14. Experimental results for total estimated and measured submodule
voltage ripples.
The functionality of the control platform implementation is
illustrated in Fig. 11. The motivation for the real-time branch
capacitor voltage estimation is justified by the necessity for
data communication reduction between the master and slave
controllers, which is represented by the bold line.
Fig. 12 illustrates the results from a converter rectifying oper-
ation. The power factor is controlled to be unity; therefore, the
line currents are in perfect phase opposition with their respec-
tive grid phase voltages. The line–line converter voltages are
PWM-controlled with a triangular carrier signal of 5 kHz per
branch and can comprise up to (4N+1) distinct voltage levels
at high modulation indexes. The intentional second-order har-
monic on the circulating current of Phase A is visible, which has
been imposed according to (17). This in turn produces a respec-
tive component on the two branch currents without, however,
affecting the output current.
In a second step, the setup is reconfigured so as to have a
two phase-leg converter structure connected to a single-phase
voltage source. The synchronization with the grid is carried out
using an SOGI-phase-locked loop structure. The representative
results for a steady-state operation are shown in Fig. 13. The
dashed-lined curves in the graphs of line current and grid phase
voltage are a direct result of the FAE and SOGI utilization, re-
spectively. As the converter is operated at a high modulation
index, the measured voltage uA B between the converter ter-
minals features all the possible levels. The circulating current
is very close to being continuous, leading to almost sinusoidal
branch currents, obviously in contrast with the respective figure
of the three-phase counterpart.
In order to validate the accuracy of the estimation equations
in the three-phase MMC studied case, the total submodule volt-
age ripples in both upper and lower branches of phase A are
illustrated in Fig. 14. The measured values are in good agree-
ment with the estimated ones, although there are some slight
differences which are mainly due to measurement and control
inaccuracies. It is obvious, however, that the second-harmonic
IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 9, SEPTEMBER 2014
Fig. 15. Experimental results for capacitor voltage ripple reduction due to
a second-order circulating current harmonic injection in converter inverting
operation. Phase A is illustrated as an example.
component on the capacitive ripples has been suppressed by
means of the respective harmonic injection (icirc,2 ).
Finally, the effect of a second-order harmonic injection in
the circulating current is validated experimentally for the three-
phase MMC case in inverting operation. The results for phase
A are shown in Fig. 15. Initially, the circulating current is con-
trolled to bear a dc component. At t ≈ 0.09 s, the imposition
of icirc,2 takes place according to (10), leading to an expected
increase of the branch current peak values. The top graph de-
picts the voltages for two submodules in the upper and lower
branches of phase A, where the respective ripple decrease from
9 to 5.5 V is visible. The capacitor voltage harmonics due to the
second-order circulating current as well as third-order common-
mode voltage harmonic injections, as presented in (13), can be
also observed. It is noted that due to some measurement noise,
the full bandwidth of the circulating current control was not
reached, which is the reason for the small oscillations until the
moment that the respective desired harmonic is imposed.
VI. CONCLUSION
In this paper, equations for the submodule voltage ripples in
an MMC have been derived in a generalized manner. These are
beneficial not only from a control point of view, but also for the
design of the capacitive energy storage requirements. The results
have been applied to typical grid-connected examples of three-
and two-phase MMCs, for which the analytical solutions have
been provided. A comprehensive documentation for the natu-
ral charge level mechanism of the branch capacitors has been
made through an illustrative example. Two independent explicit
control loops have also been formed for the line and circulat-
ing currents in a decoupled manner. In the three-phase MMC
studied application, the circulating current has been specifically
controlled in order to reduce the second-harmonic power oscil-
lation of the capacitor voltages, leading to a respective reduc-
tion of the submodule capacitance. In addition, a third-harmonic
VASILADIOTIS et al.: ACCURATE CAPACITOR VOLTAGE RIPPLE ESTIMATION AND CURRENT CONTROL CONSIDERATIONS
injection has been added in order to increase the maximum value
of the modulation index. The FAE technique has been tailored
and applied for the line current control of the two-phase MMC.
Simulation results have shown that good converter dynamic per-
formance is at hand. The validity of the aforementioned concepts
has been also demonstrated by means of experimentation on a
reduced-scale laboratory prototype.
REFERENCES
[1] A. Lesnicar and R. Marquardt, “A new modular voltage source in-
verter topology,” presented at the 10th Eur. Conf. Power Electron. Appl.,
Toulouse, France, Sep. 2–4, 2003.
[2] M. Saeedifard and R. Iravani, “Dynamic performance of a modular mul-
tilevel back-to-back HVDC system,” IEEE Trans. Power Del., vol. 25,
no. 4, pp. 2903–2912, Oct. 2010.
[3] M. Guan and Z. Xu, “Modeling and control of a modular multilevel
converter-based system under unbalanced grid conditions,” IEEE Trans.
Power Electron., vol. 27, no. 12, pp. 4858–4867, Dec. 2012.
[4] M. Hagiwara, K. Nishimura, and H. Akagi, “A medium-voltage motor
drive with a modular multilevel PWM inverter,” IEEE Trans. Power Elec-
tron., vol. 25, no. 7, pp. 1786–1799, Jul. 2010.
[5] M. Hagiwara, R. Maeda, and H. Akagi, “Control and analysis of the
modular multilevel cascade converter based on double-star chopper-cells
(MMCC-DSCC),” IEEE Trans. Power Electron., vol. 26, no. 6, pp. 1649–
1658, Jun. 2011.
[6] K. Ilves, A. Antonopoulos, S. Norrga, and H.-P. Nee, “Steady-state analy-
sis of interaction between harmonic components of arm and line quantities
of modular multilevel converters,” IEEE Trans. Power Electron., vol. 27,
no. 1, pp. 57–68, Jan. 2012.
[7] M. A. Pérez, J. R. Rodrı́guez, E. J. José, and F. Kammerer, “Predictive
Control of AC-AC Modular Multilevel Converters,” IEEE Trans. Ind.
Electron., vol. 59, no. 7, pp. 2832–2839, Jul. 2012.
[8] M. A. Pérez and J. R. Rodrı́guez, “Generalized modeling and simulation
of a modular multilevel converter,” presented at the IEEE Int. Symp. Ind.
Electron., Gdansk, Poland, Jun. 27–30, 2011.
[9] Q. Tu, Z. Xu, and L. Xu, “Reduced switching-frequency modulation and
circulating current suppression for modular multilevel converters,” IEEE
Trans. Power Del., vol. 26, no. 3, pp. 2009–2017, Jul. 2011.
[10] S. Rohner, S. Bernet, M. Hiller, and R. Sommer, “Modulation, losses,
and semiconductor requirements of modular multilevel converters,” IEEE
Trans. Ind. Electron., vol. 57, no. 8, pp. 2633–2642, Aug. 2010.
[11] K. Ilves, A. Antonopoulos, S. Norrga, and H.-P. Nee, “A new modu-
lation method for the modular multilevel converter allowing fundamen-
tal switching frequency,” IEEE Trans. Power Electron., vol. 27, no. 8,
pp. 3482–3494, Aug. 2012.
[12] L. Ängquist, A. Antonopoulos, D. Siemaszko, K. Ilves, M. Vasiladiotis,
and H. P. Nee, “Open-loop control of modular multilevel converters us-
ing estimation of stored energy,” IEEE Trans. Ind. Appl., vol. 47, no. 6,
pp. 2516–2524, Nov./Dec. 2011.
[13] L. Harnefors, A. Antonopoulos, S. Norrga, L. Ängquist, and H.-P. Nee,
“Dynamic analysis of modular multilevel converters,” IEEE Trans. Ind.
Electron., vol. 60, no. 7, pp. 2526–2537, Jul. 2013.
[14] K. Ilves, A. Antonopoulos, L. Harnefors, S. Norrga, L. Ängquist, and H.
P. Nee, “Capacitor voltage ripple shaping in modular multilevel convert-
ers allowing for operating region extension,” presented at the IEEE Ind.
Electron. Conf., Melbourne, Australia, Nov. 7–10, 2011.
[15] N. Cherix, M. Vasiladiotis, and A. Rufer, “Functional modeling and ener-
getic macroscopic representation of modular multilevel converters,” pre-
sented at the 15th Int. Power Electron. Motion Control Conf., Novi Sad,
Serbia, Sep. 4–6, 2012.
[16] M. Winkelnkemper, A. J. Korn, and P. Steimer, “A modular direct converter
for transformerless rail interties,” presented at the IEEE Int. Symp. Ind.
Electron., Bari, Italy, Jul. 4–7, 2010.
[17] S. P. Engel and R. W. De Doncker, “Control of the modular multi-level
converter for minimized cell capacitance,” presented at the 14th Eur. Conf.
Power Electron. Appl., Birmingham, U.K., Aug. 30–Sep. 1, 2011.
[18] A. Korn, M. Winkelnkemper, P. Steimer, and J. W. Kolar, “Direct modular
multi-level converter for gearless low-speed drives,” presented at the 14th
Eur. Conf. Power Electron. Appl., Birmingham, U.K., Aug. 30–Sep. 1,
2011.
[19] M. Vasiladiotis, S. Kenzelmann, N. Cherix, and A. Rufer, “Power and DC
link voltage control considerations for indirect AC/AC modular multi-
4579
level converters,” presented at the 14th Eur. Conf. Power Electron. Appl.,
Birmingham, U.K., Aug. 30–Sep. 1, 2011.
[20] H. Bühler, Réglage de Systèmes d’électronique de Puissance: Théorie, vol.
1, Lausanne, Switzerland: PPUR, Presses Polytechniques et Universitaires
Romandes, 1997.
[21] J. Kolb, F. Kammerer, and M. Braun, “A novel control scheme for the low
frequency operation of the modular multilevel converter,” presented at the
Power Convers. Intell. Motion Eur., Nuremberg, Germany, May 17–19,
2011.
[22] D. N. Zmood and D. G. Holmes, “Stationary frame current regulation of
PWM inverters with zero steady-state error,” IEEE Trans. Power Electron.,
vol. 18, no. 3, pp. 814–822, May 2003.
[23] R. Teodorescu, F. Blaabjerg, M. Liserre, and P. Loh, “Proportional-
resonant controllers and filters for grid-connected voltage source con-
verters,” IEE Proc. Electr. Power Appl., vol. 153, no. 5, pp. 750–762, Sep.
2006.
[24] B. Bahrani, A. Rufer, S. Kenzelmann, and L. A. C. Lopes, “Vector control
of single-phase voltage-source converters based on fictive-axis emula-
tion,” IEEE Trans. Ind. Appl., vol. 47, no. 2, pp. 831–840, Mar./Apr. 2011.
[25] M. Ciobotaru, R. Teodorescu, and F. Blaabjerg, “A new single-phase PLL
structure based on a second order generalized integrator,” in Proc. IEEE
37th Power Electron. Spec. Conf., Jun. 2006, pp. 1–6.
[26] S. Kenzelmann, A. Rufer, M. Vasiladiotis, D. Dujic, F. Canales, and Y.
R. de Novaes, “A versatile DC-DC converter for energy collection and
distribution using the modular multilevel converter,” presented at the 14th
Eur. Conf. Power Electron. Appl., Birmingham, U.K., Aug. 30–Sep. 1,
2011.
[27] M. Vasiladiotis, A. Rufer, and A. Béguin, “Modular converter architec-
ture for medium voltage ultra fast EV charging stations: Global system
considerations,” presented at the IEEE Int. Electr. Veh. Conf., Greenville,
SC, USA, Mar. 4–8, 2012.
Michail Vasiladiotis (S’09) was born in Athens,
Greece, in 1986. He received the Diploma degree
in electrical and computer engineering from the
National Technical University of Athens, Athens,
Greece, in 2009. Since 2010, he has been with the
Laboratory of Industrial Electronics, École Polytech-
nique Fédérale de Lausanne, Lausanne, Switzerland
working toward the Ph.D. degree.
His research interests include multilevel conver-
sion systems for high power applications, modern
control methods for power converters as well as
power electronic interfaces for ultra-fast EV charging.
Nicolas Cherix (S’11) was born in Château-d’Oex,
Switzerland, in 1986. He received the B.Sc. and M.Sc.
degrees from the Swiss Federal Institute of Technol-
ogy, Lausanne, Switzerland.
Since 2010, he has been with the Laboratory of In-
dustrial Electronics, Swiss Federal Institute of Tech-
nology. His research interests include modular energy
conversion systems and embedded control systems.
His PhD thesis will address the improvement of the
power quality of the Swiss railway electricity supply
chain, using modular multilevel converters with inte-
grated split energy storage.
Alfred Rufer (M’95–SM’01–F’06) received the
M.S. degree from the Swiss Federal Institute of Tech-
nology (EPFL), Lausanne, Switzerland, in 1976.
In 1978, he joined ABB, Turgi, Switzerland, where
he was involved in the fields of power electronics and
control, such as high-power variable frequency con-
verters for drives, and where he was a Group Leader
involved with power electronic development begin-
ning in 1985. In 1993, he became an Assistant Pro-
fessor at EPFL, where, since 1996, he has been a Full
Professor and the Head of the Industrial Electronics
Laboratory (LEI). LEI is active in power electronics used in energy conversion
and energy storage and in the modeling and simulation of systems, including
control strategies and control circuits. He has authored or coauthored many pub-
lications on power electronics and its applications, such as multilevel converters
and various energy-storage systems. He is the holder of several patents.