www.ietdl.

org
Published in IET Power Electronics
Received on 29th September 2009
Revised on 9th December 2010
doi: 10.1049/iet-pel.2010.0131

ISSN 1755-4535

Passivity-based control of an asymmetric nine-level

inverter for harmonic current mitigation
G. Sandoval1 H. Miranda2 G. Espinosa– Pérez1 V. Cárdenas2
1
DEPFI–UNAM, A.P. 70-256, 04510 México D.F., Mexico
2
CIEP-FI-UASLP, Dr. Manuel Nava 8, 72490 San Luis Potosi, Mexico
E-mail: [email protected]

Abstract: In this study is approached the design problem of a control law for an asymmetric nine-level multilevel cascade inverter
when it is implemented as active filter for harmonic current mitigation purposes. Two are the main features of the proposed
controller, namely, its structure is defined considering at a fundamental level the precise establishment, in terms of
mathematical expressions, of the steady-state converter behaviour required to achieve the control objective and its stabilisation
properties, that render this desired steady-state behaviour attractive (asymptotically stable), are obtained by exploiting the
energy-dissipation (passivity) properties of the circuit. These features lead to a simple controller structure that is easy to tune.
The stability, high performance and robustness properties of the presented control scheme are experimentally evaluated.

1 Introduction Owing to the fact that the different cells are fed with
different DC voltages, it is possible to generate more
It is a well-known fact that power electronics has become into different output voltage levels with the same number of
a fundamental element for many industrial devices because of components than their symmetric counterparts, establishing
its ability for delivering and processing power in a more a quite promising solution to the aforementioned reliability
efficient way. Among all these applications one that has problem [5 – 8].
attracted a lot of attention is related with what is called This distinctive feature of asymmetric topologies implies,
power quality, that is, the branch of electronics and in its turn, higher output quality and reduced output
electrical power systems concerned in achieving a proper filter size (than single converter topologies), among other
operation of electrical systems and equipment. In this advantages. Nevertheless, operation of the cells under
context, active filters have proven to be very useful for different voltages also generates several implementation
harmonic filtering [1]; in particular, for solving the current problems, for instance, more complex control schemes or
harmonics mitigation problem, because of its ability for the necessity for operating the sub-circuits at different
compensating several harmonic components without switching frequencies. Fortunately, diverse solutions to
producing resonances. these disadvantages can be found in the literature, providing
The remarkable success obtained by introducing power switching policies [9] or stabilising controllers [10, 11] that
converters in low/medium power applications together with achieve very good results. However, all of these solutions
the necessity for dealing with medium/high-power are based in a deep conceptual knowledge (and its clever
problems, motivated for searching mechanisms to overcome exploitation) of the device operation, without taking into
the actual semiconductor voltage and current ratings, account the particular (non-linear) structure exhibited by the
leading to the class of circuits called Multilevel converters. circuit, producing relatively complex layouts and limiting,
This topology is based on an imbricated cells structure as a consequence, a more exhaustive use of this kind of
which allows it to share the overall applied voltage among converters.
several sub-circuits (cells) leading to the possibility of The main purpose of this paper is to illustrate that it
managing higher power (without using bulky and heavy is possible to provide simple controller structures for
transformers) while generating signals with higher quality asymmetric inverters exploiting in a proper way the
(since they are produced in a stepped up form) and, at the information obtained from a detailed analysis of the non-
same time, allowing the power semiconductors withstanding linear dynamic behaviour exhibited by the converter. This is
low voltages, reducing therefore cost and power ratings done by considering the current harmonics mitigation
[2 – 4]. Evidently, the main drawback of this topology problem when an asymmetric nine-level multilevel cascade
appears with the increased number of semiconductors inverter (MCI) is implemented as an active filter (see
which reduces the overall reliability of the converter. Fig. 1). The contribution is a novel control law that solves
Although it is difficult to assess which is the best solution the voltage regulation and current tracking control problems
among the several existing multilevel topologies, multilevel imposed by the application exhibiting a high performance
asymmetric inverters exhibit a fundamental advantage. and providing, at the same time and in a quite natural way,

IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237–247 237

doi: 10.1049/iet-pel.2010.0131 & The Institution of Engineering and Technology 2012
www.ietdl.org

Fig. 1 Hybrid asymmetric nine-level MCI for current harmonics compensation

a very simple mechanism for defining the switching this context the originality of the contribution is clear since
frequencies of both the high- and low-voltage sub-circuits. converters like the approached in this paper have not been
In addition, the structure of the controller is remarkable previously studied.
simple (basically a non-linear proportional controller) and The rest of the paper is organised in the following way: The
easy to tune. problem approached in the paper is presented in Section 2,
As it will be clear later, the aim of the paper is not to find while its solvability analysis is introduced in Section
new properties of this well-known kind of converters but to 3. With the capability of the circuit for achieving the
exploit them in a way that, to the best of the author’s desired behaviour completely characterised, in Section 4 the
knowledge, has not been done before and that leads to the proposed control law is presented carrying its experimental
new control scheme. In fact, these properties have been evaluation out in Section 5. The paper ends by stating some
recognised at some extent and are used, for instance, for concluding remarks in Section 6.
designing controllers using well-known classical linear
control tools, for example, PID compensators and repetitive 2 Problem formulation
control [12 – 14], together with an also classical two-loop
cascade control structure. Unfortunately, the limited In this section, after discussing the structure and operation of
exploitation of the acquired knowledge makes the efficient the converter considered in this paper, it is presented its
utilisation of these alternatives strongly dependent of the mathematical model and later on is formulated the control
high expertise required to tune the controllers, specially the problem that must be solved in order to achieve the desired
complex fast controller for the inner current loop, situation closed-loop system behaviour.
that is overcome under the proposed approach as a result of
the better understanding of the converter dynamics. 2.1 Inverter structure and operation
The approach for obtaining the proposed control scheme is
of the model-based type and uses at a fundamental level The circuit of Fig. 1 shows an asymmetric MCI implemented
some concepts obtained from the control theory for non- as an active filter for mitigating the current harmonics
linear systems. In particular, it is based in the very well- introduced by a non-linear load. It is composed by two H-
known concept of passivity [15], strongly related to the bridge sub-circuits, each one equipped with a capacitor as a
energy-dissipation properties of the system and that currently voltage source, C1 for converter 1 and C2 for converter
is exploited in a very successful way from a control theory 2. Resistors R1 and R2 represent their equivalent (switching)
perspective [16, 17]. The reason for this choice comes from losses while Rs is included to model the conduction losses
the fact that under the passivity approach it is possible to and the resistance of the line reactance, whose inductance is
match in a very useful way the physical system operation represented by L. In the point of common coupling, iL
with the stability and performance properties of the closed stands for the current drawn by the inverter while iNL
loop composed by a given passive plant and a proposed corresponds to the non-linear device and iv to the utility,
passivity-based controller [18]. Roughly speaking, the this one applying a voltage denoted by v.
controller design is developed in two steps, namely, it is first The advantage of using this asymmetric topology in the
recognised under what operation conditions the control considered application, lies in its ability for generating
objective can be achieved and, later on, an stabilising a nine level stepped voltage waveform in comparison to
mechanism, that renders this behaviour attractive, is proposed. the five levels that can be achieved with its symmetric
The idea of applying the passivity concept in power counterpart. This results is obtained if a 1:3 voltage ratio,
electronics is not new. It has been shown under different that is, 3vc1 ¼ vc2 , is maintained, as can be verified by
scenarios its usefulness in the field [19 – 24], but even in noticing that if each H-bridge is equipped with a single

238 IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237 –247
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-pel.2010.0131
 www.ietdl.org
switching function sw ¼ TA 2 TB , voltages and currents can the device and will be included, in a natural way, in the
be expressed as vAC ¼ vCsw and iDC ¼ iLsw and therefore proposed passivity-based control (PBC) law.
since sw1 [ {21, 0, 1} and sw2 [ {21, 0, 1}, the complete
set of achievable output voltage values is as the listed in 2.2 Inverter model
Table 1.
Unfortunately, this unbalanced operation also introduces Once the structure and operation of the converter have been
several implementation problems. On the one hand, from recognised, in this section its dynamic behaviour is stated by
the operation point of view, it is necessary to consider a means of a mathematical model. In this sense, considering
different switching frequency for the two sub-converters in the circuit of Fig. 1 and a straightforward application of
order to minimise the losses. Indeed, if the low-voltage Kirchhoff’s laws, carrying out a loop analysis for the output
converter is operated at a higher frequency with respect to mesh and a nodal one for the DC side, it is possible to
the high-voltage one, then low conduction, in the former formulate the following set of differential equations
and low switching losses, in the latter, can be achieved. On
the other hand, from the control perspective, the challenge diL
is interesting since it is desired to achieve three control L + Rs iL − sw1 vC1 − sw2 vC2 = −v (3)
objectives, regulation of the two capacitor voltages and dt
tracking of the inductor current, with only two control dvC1
inputs, the switching functions sw1 and sw2 . C1 + R−1
1 vC1 + sw1 iL = 0 (4)
To close this sub-section, it is convenient to recognise an dt
additional feature of the converter operation that in fact dvC2
makes feasible to achieve the desired behaviour. Notice that C2 + R−1
2 vC2 + sw2 iL = 0 (5)
dt
under the desired operation what is necessary to guarantee
is that the utility generates only fundamental current, that is, which describes the dynamic evolution of the converter and
current of the type iv ¼ I sin(vt + a), where I is a given where all the variables and constants have been previously
magnitude. Then, if the load current is of the form defined in Section 2.1. At this point it is useful to identify,
for reasons that will be clear later, how the switching losses
iNL = iNL1 + iNLh appear in the model. Notice that as long as the value of

N resistor R1 (R2) increases, C1 (resp., C2) capacitor losses are
= INL1 sin(v + u1 ) + INLn sin (nvt + un ) (1) smaller.
n=2 In order to exhibit in an explicit way the structural
properties of the converter and later on to simplify the
the inverter is forced to drawn currents like iL ¼ iNL 2 iv controller design, the above set of differential equations can
yielding be written in a compact matrix form as


N
Dż + Rz + sw1 M1 z + sw2 M2 z = 1 (6)
iL = INL1 sin(v + u1 ) − I sin(vt + a) + INLn sin(nvt + un )
n=2
where z ¼ [iL , vC1 , vC2]T, D ¼ diag{L, C1 , C2}, R = diag{Rs ,
(2) R−1 −1
1 , R2 }, 1 ¼ [2v, 0, 0] and
T

⎡ ⎤ ⎡ ⎤
where the important issue that must be recovered is that the 0 −1 0 0 0 −1
magnitude of the utility current I is not fixed a priori and M1 = ⎣ 1 0 0 ⎦, M2 = ⎣ 0 0 0 ⎦
can be viewed as an additional degree of freedom in the 0 0 0 1 0 0
controller design.
The advantage of representing the system as in (6), comes from
Remark 1: It is important to point out that the conceptual the facility to clearly identify the total stored energy of the
analysis presented in this section is very well known and, in converter
fact, has served (at a large extent) as the basis to propose
several control schemes for the device. From the
1
perspective of the contribution, its importance lies in the V = zT Dz, D = DT . 0 (7)
fact that all the revised features will be recovered from a 2
mathematical analysis of the dynamical non-linear model of
its dissipated energy
Table 1 Switching combinations and output voltage levels
1
sw1 sw2 Level sw1 Level sw2 Output voltage
F = zT Rz, R = RT . 0 (8)
2
21 21 –vdc1 – vdc2 – 4vdc
0 21 0 – vdc2 – 3vdc
and the internal interconnection terms included in the matrices
1 21 vdc1 – vdc2 – 2vdc
M1 and M2 , which are skew-symmetric, i.e. z TMiz ¼ 0, for all
21 0 –vdc1 0 – vdc
vector z and i ¼ 1, 2.
0 0 0 0 0
As usual, the inconvenience of having discontinuous
1 0 vdc1 0 vdc
control signals in (6), can be overcome assuming a high
21 1 –vdc1 vdc2 2vdc
switching frequency operation, for example, considering a
0 1 0 vdc2 3vdc
pulse width modulation (PWM) policy and using an
1 1 vdc1 vdc2 4vdc
averaging technique in order to obtain a continuous model.
For instance, it is possible to write a differential equation of

IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237–247 239

doi: 10.1049/iet-pel.2010.0131 & The Institution of Engineering and Technology 2012
www.ietdl.org
the form ż = A(u)z + B as the following integral equation with xi and xid, i ¼ 1, 2, 3, the actual state and its corresponding
desired behaviour, respectively, satisfying:
t1
z(t1 ) − z(t0 ) = (A(u)z + B) dt P.1 x1d is a time variant periodic function that guarantees that
t0
the utility only delivers fundamental current.
P.2 x2d and x3d are constant positive values such that
and, if T is defined as the sampling period and t0 and t1 are 3x2d ¼ x3d .
the starting and ending time instants of this period, P.3 The control law m2 commutates at the lowest possible
respectively, obtain an average state model by dividing the switching frequency with respect the control law m1 .
state difference into T and computing the limit as T  0.
This procedure yields to the averaged state model given by Regarding the stated problem and the aforementioned
assumptions, it is convenient to formulate the following
Dẋ + Rx + m1 M1 x + m2 M2 x = 1 (9) remarks.

where x ¼ [x1 , x2 , x3]T is the averaged state, while m1 and m2 Remark 4: While assumptions A.1 – A.3 are reasonable,
appear now as continuous control inputs each one belonging assumption A.4 imposes a practical limitation since it is not
to the continuous set [ – 1, 1]. easy to guarantee knowledge of some parameters, for
example, resistors R1 and R2 . Fortunately, this assumption
Remark 2: It is interesting to point out that the modeling can be relaxed to a large extent by a slight modification on
methodology and therefore the proposed controller (as will be the converter structure, as it will be clear later.
clear later), is not restricted neither by the number of converters
involved in the multi-level structure nor by the assumed Remark 5: As it will be proved in the next section, problem
asymmetric structure. Actually, the followed procedure can be P.2 is not achievable in its current presentation. Actually,
easily extended to the case of having n sub-circuits, possibly this problem is only solvable in the sense that the average
under a balanced operation. In this case, in the voltage equation value of x2d and x3d coincides with the desired constant
of the model (3) will appear one term of the form swivCi for values. Nevertheless, it is presented as above with the aim
each of the additional sub-circuits while, instead of having only to formulate the problem in a clearer way.
two current equations of the form (4–5), it would appear n.
3 Solvability analysis
Remark 3: One attractive feature of model (9) comes from its
dynamic nature, since it captures both the transient and In this section the first part of the contribution is presented,
steady-state response of the device. In this paper, the namely, the co-relation between the conceptual operation
desired behaviour of the system will be stated in terms of of the converter (presented in Section 2.1) and explicit
the steady-state operation but, since the controller design is mathematical expressions derived from the analysis of the
developed using this dynamic model, the properties of the model (9). Carrying this task out is important because it is
closed-loop system will be guaranteed even when transient possible to immerse the conceptual knowledge about the
response changes were exhibited. device operation in a more precise context where limitations
and advantages of the converter can be better identified and
2.3 Control problem in consequence better exploited, for instance, in the design of
control strategies. In this paper, in particular, this information
With the physical operation and the mathematical model of will be used later to conceive a stabilisation mechanism that
the asymmetric multilevel inverter at hand, it is possible to once identified renders the desired converter operation
formally state the problem approached in this paper. For attractive with respect to different initial conditions, changes
doing this, the following assumptions are considered: in the operation regime and/or perturbations.
In order to develop the analysis, assume that the converter
A.1 All the state variables are measurable. has achieved a steady-state operation. For reasons that will be
A.2 The utility voltage is known and represented by v ¼ V clear in the controller design, denote this operation regime as
sin(vt).
A.3 The current iNL of the non-linear load is known and (in Lẋ1d + Rs x1d − m1 x2d − m2 x3d = −v (10)
steady-state) satisfies that iNL(t) ¼ iNL(t + T ) ¼ 2iNL(t +
(T/2)), that is, this current is a periodic function with C1 ẋ2d + R−1
1 x2d + m1 x1d = 0 (11)
half-wave symmetry and therefore can be expressed by its
Fourier series. C2 ẋ3d + R−1
2 x3d + m2 x1d = 0 (12)
A.4 All the circuit parameters are known.
The first part of the analysis is related with the fact of
Under these conditions the control problem can be stated as recognising that achieving constant values for the voltage
the design of two control laws that allow to mitigate the capacitors is not feasible. This can be easily proved by
harmonic content of iNL while the voltages in the DC noting from (11) and (12) that under the assumption that
capacitors reach the desired values. Thus, the formulation of x2d and x3d are constant values, it must be satisfied that
the control problem can be expressed as:
Consider the asymmetric nine-level inverter model (9) and R−1
1 x2d + m1 x1d = 0
assume A.1 – A.4 hold. Design two control laws, m1 ¼ m(x1 ,
x2 , x3) and m2 ¼ m(x1 , x2 , x3) such that R−1
2 x3d + m2 x1d = 0

lim x − xid = 0 that is the products m1x1d , m2x1d must be constant values.
t1 i However, this condition is not achievable since in order to

240 IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237 –247
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-pel.2010.0131
 www.ietdl.org
satisfy (10), considering that x1d is a periodic function while where it has been defined z = x22d . This expression is a first-
x2d , x3d are constants, both m1 and m2 must have the same order linear differential equation on z with a periodic input
harmonics of x1d , producing therefore that the products whose average value depends on m3 . Therefore the control
m1x1d , m2x1d contain also some harmonics, violating the objective can be completely solved by choosing this
required equality. parameter in such a way that the average value of z tends to
Once the impossibility of generating constant capacitor the desired value for x22d .
voltages has been recognised, in this second part of the As a result of the developed analysis it can be concluded
analysis, admitting that the capacitor voltages will exhibit in that the best approach that can be achieved of the desired
addition of a constant some harmonic components, it is steady-state closed-loop behaviour is given by the dynamic
recognised that if priority is given to the regulation control law (13), (15), with controller states (11), (12),
objectives, the tracking one cannot be accomplished. To desired inverter current (14) and a proper selection of the
this end, assume that both control laws m1 and m2 include variables M and m3 . Under these conditions current
(at least) one harmonic component exhibited by x1d . Under harmonic mitigation is assured with capacitor voltages
these conditions the products m1x1d , m2x1d will reveal two tending (in average) to their desired values.
important features. On the one hand, both of them will On the other hand, it is quite interesting to notice how, also
include a DC (average) term whose amplitude can be as a result of the developed analysis, the major concerns about
chosen in such a way that the average value of x2d and x3d the operation of circuit appear solved in a straightforward
reach their corresponding desired values, but on the other way, namely:
hand, these products will also include some harmonics
components that are not contained in x1d and that, as a † The switching separation requirement is achieved since the
consequence of the linear structure of (11) and (12), will definition of m2 satisfies the requirement of operating the
appear in the x2d and x3d variables. high-voltage converter at a low (fundamental) switching
If these control laws that solve (in average) the regulation frequency while the low-voltage circuit commutates at high
objectives are substituted into (10), then, in order to solve frequency.
in a simultaneous way the tracking objective, they must † The three control objectives are reached by mean of the
hold with the expression two physical control inputs. Moreover, the usual practice of
using the magnitude of current delivered by the utility as a
Lẋ1d + Rs x1d + v = m1 x2d + m2 x3d third degree of freedom, appear also in a very precise
expression.
The problem for satisfying this equality is that, because of the
obtained structure for x2d and x3d , the harmonic components Evidently, it must be considered that the expressions that
of the left-hand term will be different to the components have been obtained express the device operation only in
included in the right-hand term, leading to the conclusion steady-state and that is necessary to introduce a stabilisation
that this expression and therefore the tracking objective, mechanism, but the advantage of having the precise
cannot be satisfied. behaviour to be stabilised gives to the designer a great
The third part of the analysis establishes the only feasible advantage since it is possible to profit of this knowledge, as
solution to the control problem. In this case, priority is will be illustrated in the next section.
given to the tracking objective by letting one of the control
laws, for instance m1 , to satisfy Remark 6: Special attention must be given to recognise the fact
that if it is desired to exploit at maximum the available
1 knowledge about the circuit operation, then the obtained
m1 = [Lẋ1d + Rs x1d + v − m2 x3d ] (13) expressions must be satisfied. Evidently, different stabilising
x2d
alternatives can be conceived without using this information,
for a given x1d and assuming x2d = 0. but the price to be paid could be lower performances and
Taking into account the required structure for the inverter (as it is very usual) complex control schemes.
current given in (2), consider now that
4 Controller design
x1d = m3 sin(vt) + iNLh (14)
The second part of the contribution is presented in this section.
with m3 a free parameter. If the second control law is designed as It consists in the proposition of a stabilisation mechanism that
renders the identified desired steady-state operation attractive
m2 = M sin(vt) (15) by using the PBC design methodology. As it will be clear
below, the main advantage of approaching the control design
under this perspective is the quite natural way in which the
then it is possible, by choosing in a proper way the parameter M,
stated objective is achieved, leading to a very simple control
to assign to the capacitor voltage x3d its corresponding desired
scheme.
(average) value leaving m3 free.
With the aim of having a self-content document, the section
The final part of the analysis comes from considering (11)
begins with a quick review of the passivity concept to later on
under the control policy (13) which leads to
introduce the passivity-based design and the specific
proposed controller.
C1 ẋ2d x2d + R−1
1 x2d = −(Lẋ1d x1d + Rs x1d + vx1d − m2 x1d x3d )
2 2

4.1 Passivity analysis

or equivalently
Consider the dynamical model of the asymmetric converter
1
C ż + R−1
1 z = −(Lẋ1d x1d + Rs x1d + vx1d − m2 x1d x3d ) (16)
2 given in (9), with the definition of variables and parameters
2 1 presented in Section 2.2 and its total (positive definite)

IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237–247 241

doi: 10.1049/iet-pel.2010.0131 & The Institution of Engineering and Technology 2012
www.ietdl.org
stored energy function given by reached faster with respect to the open-loop system. This
action is the so-called damping injection step.
1
V = xT Dx, D = DT . 0 (17) In order to apply the PBC ideas exposed above, first the
2
related with the energy-shaping step design, define the
where x is the averaged state of the converter. The time control error x̃ = x − xd as the difference between the actual
derivative of this function along the trajectories of (9) is x and a desired xd state. In terms of this variable, the system
dynamics (9) can be written as
V̇ = xT Dẋ
Dx̃˙ + Rx̃ + m1 M1 x̃ + m2 M2 x̃ = F (20)
= x [−Rx − m1 M1 x − m2 M2 x + 1]
T
where the right-hand side term is given by
= −xT Rx + xT 1 (18)
F = 1 − {Dẋd + Rxd + m1 M1 xd + m2 M2 xd } (21)
where the last equality has been obtained by considering that
both M1 and M2 are skew-symmetric matrices. The importance of this representation lies in the fact that it
The importance of (18) lies in the fact that it establishes the exhibits exactly the same structure of the open-loop
energy balance exhibited by the converter. Indeed, if this converter, now in terms of the error variable and therefore
expression is re-written in integral form as its passivity properties. This claim can be proved if it is
considered as storage energy function for the error
t1 t1 dynamics (20) the strictly convex function
V (t1 ) − V (t0 ) = − xT Rx dt + xT 1 dt (19)
t0 t0 1
Vd = x̃T Dx̃ (22)
2
and the product x T1 is identified as the input power of the
circuit, then the left-hand term in (19) stands for the stored which enjoys the characteristic of having an isolated
energy while the quadratic term in the right-hand side minimum at x̃ = 0, that is, corresponding to the operation
represents the dissipated energy, that is, it is established that when the actual is equal to desired state.
the device is not able to internally generate energy and then The time derivative of (22) along the trajectories of (20) is
it defines a passive system [16] from the input 1 to the
output x. V̇ d = x̃T F − x̃T Rx̃ (23)
From the control perspective, this energy balance becomes
important since it can be observed that if the input 1 is zero, Hence, if F is considered as input while x̃ is defined as output
then of (20), the error dynamics defines a passive system and
therefore from the analysis presented Section 4.1, if F is
V̇ = −xT Rx , 0 equal to zero, the trajectories will tend to the minimum
argument of (22), that is, the control objective will be
that is, the energy behaviour tends to its minimum value achieved.
x ¼ 0 with a rate of change determined by the matrix R. In At this point it is very important to notice that, from (21),
terms of Lyapunov stability theory [25], this means that the satisfying the equality F ¼ 0 is equivalent to find proper
equilibrium point given by x ¼ 0 is asymptotically stable in definitions for the control inputs m1 , m2 and the desired
a natural way and the trajectories of the system will state xd that holds with this constraint. However, what
converge to it with a speed that is proportional to the makes very attractive the passivity approach is to recognise
capacity for dissipating energy. that this equality is exactly the same as the introduced in
(10 – 12). Thus, the solution found from the solvability
analysis developed in Section 3, exactly states a solution for
4.2 PBC design
satisfying the conditions under which the PBC approach
The aim of the PBC methodology design is to take advantage guarantees the achievement of the control objective, that is,
of the natural properties of the converter for dissipating definition of the control laws m1 , m2 under the PBC
energy in such a way that a desired behaviour is achieved. perspective is carried out in a natural way taking into
In this sense, the technique exploits two fundamental account the capability of the converter for accomplishing
properties [16], namely the desired operation.
Concerning the stabilisation mechanism that renders the
† If a given control law preserves the open-loop passivity desired behaviour attractive, it can be obtained in a very
properties but now with respect to a closed-loop energy simple way under the PBC perspective, since instead of
function that has as a minimum one that corresponds to a satisfying the constraint F ¼ 0, it is only necessary to
desired behaviour, then the closed-loop system will require that F = −K x̃, yielding to the alternative form for
accomplish in a natural way the control objectives since the the time derivative of Vd given by
trajectories will tend to this minimum point. This step in
the controller design is known as the energy-shaping step V̇ d = −x̃T (K + R)x̃ (24)
since the objective is to impose to the closed-loop system
an appropriate energy function. Under these conditions, if matrix (K + R) is symmetric and
† If in addition of carrying out the energy-shaping step the positive definite, that is, satisfies that (K + R) ¼ (K + R)T . 0,
control law increases the capability of the system for then it can be observed that asymptotic stability of the
dissipating energy, then the desired behaviour will be equilibrium point x̃ = 0 is ensured, but now with an increased

242 IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237 –247
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-pel.2010.0131
 www.ietdl.org
rate of convergence of the trajectories to the minimum value Once M has been defined, the attention must be concentrated
x̃ = 0, since it depends on the tuning matrix K. in (16) since the average value of its right-hand side must be
Putting all the elements together, assuming for simplicity equal to the square of the desired average value of x2d . Thus,
that K ¼ diag{k1 , k2 , k3}, the complete control scheme is under the assumed periodicity of x1d , this condition reads as
given by

N
1 x22d = −2−1 R1 [Rs m23 + cos(a)(V − Mx3d )m3 ] −2−1 R1 Rs In2
m1 = (Lẋ1d + Rs x1d + v − m2 x3d − k1 x̃1 ) (25) n=2
x2d
(29)
and m2 ¼ M sin(vt), with x1d ¼ m3 sin(vt) + iNLh , while the
desired capacitor voltages are obtained as solutions of which, considering (28), can be expressed only in terms of the
(still free) parameter m3 . Hence, after some algebraic
C1 ẋ2d + R−1
1 x2d = −m1 x1d − k2 x̃2 (26) manipulations, it is possible to obtain a second degree
algebraic equation on this free variable leading to a solution
C2 ẋ3d + R−1
2 x3d = −m2 x1d − k3 x̃3 (27) for (29) given by
√
Remark 7: It must be noticed that convergence to zero of the −k + k2 − l
errors is guaranteed for any value of m3 and M. However, m3 = (30)
2
since the definition of these variables determines the steady-
state behaviour of the system a proper selection must be where
carried out. In this sense, it is important to recognise that
the conditions found in Section 3 for defining them are still
k = VR−1
s cos(a)
valid since the scenario under which they were obtained is
recovered when x̃ = 0.
and
Remark 8: It is interesting to mention that the condition

N
imposed to matrix (K + R) to be symmetric and positive l = 8R−1 −1 2 −1 2
s [R1 x2d + R2 x3d ] + 4 In2
definite can be satisfied in several ways. However, the n=2
diagonal one leads to the simpler structure since under this
definition k1 . 0, k2 . 0 and k3 . 0 is enough for From these expressions it is clear that the feasibility for finding
guaranteeing the desired behaviour, simplifying at the same an implementable value for m3 is restricted to satisfying the
time tuning of the controller. condition k2 2 l . 0. However, this expression imposes a
double problem, namely:
Remark 9: It must be noticed that the damping injection
mechanism, given in terms of matrix K in (24), can be † The high value of R1 and R2 leads to solutions for m3 of
interpreted as the inclusion of virtual resistors that modify the small magnitude. Hence, under a practical scenario where
value of the original ones. Indeed, as long as each entry of noise is always present, the current that will be injected to
the matrix take a higher value, faster will be the convergence the DC side of the converter to charge the capacitors could
of the corresponding error to zero. Unfortunately, for the be absorbed by this noise leading to an undesirable operation.
second and third equations, the usual very high value of R1 † There exists a great uncertainty in the values of R1 and R2 .
and R2 will demand also very high values of the gains K2 and
K3 in order to achieve faster dynamic responses increasing, Under these conditions and recalling the fact that higher
in its turn, the magnitude of the control inputs. As will be values for R1 and R2 imposes the necessity of implementing
explained in the next section, this drawback can be reduced a high gain controller, one way that results attractive for
with a slight modification of the converter topology. making the implementation of m3 feasible, is the inclusion
in the circuit topology of a small resistance parallel
4.3 Implementation issues connected with capacitors C1 and C2 . Under this
modification, it is reasonable to consider that the equivalent
In this last part of the paper, the issues that are required to take value of R1 and R2 parallel connected with the small
into account for implementing the proposed controller are resistance, denoted by R, are equal and, more important,
analysed. First, the explicit computation of m3 and M is can be approximated by the value of the introduced
approached to continue with a discussion on the resistance. Hence, the feasibility condition can be stated as
programming complications that arise when a digital
implementation is carried out. k2 − 4SNn=2 In2
In order to obtain the value of M, it must be considered that . R−1 . 0
the average value of the solution of the linear differential (12) 8[x22d + x23d ]
must be equal to the constant component of the product
m2x1d . Thus, from the assumed structure of x1d and m2 with
given in (14) and (15), respectively, the value of M is 
determined as N
k.2 In2
M = −2x3d (m3 R2 cos(a))−1 (28) n=2

with x3d the average value of x3d and whenever 0 ≤ a , 908 solving at the same time the uncertainty and the high gain
and m3 = 0. problems.

IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237–247 243

doi: 10.1049/iet-pel.2010.0131 & The Institution of Engineering and Technology 2012
www.ietdl.org
Remark 10: Note that the effect of introducing R is to increase The system under evaluation was one with nominal
the current flowing through C1 and C2 . In this sense, the capacity of 1.2 kVA connected to a non-linear load given
selection of the value for m3 in (30) must be carried by a single-phase diode rectifier circuit, that in its turn feds
carefully in order to reduce as much as possible both the a resistive element RL . The converter was built using
effort developed by the utility and the losses in the converter. MOSFETs IRF540 (100 V/32 A) each one equipped with a
driver PC923. The PWM scheme was of the sine-type with
Concerning the computational aspects for implementing the phase shifting of multiple carrier signals operating at a
proposed controller, the first point to be considered is its 9.6 KHz switching frequency implemented in a FPGA
dynamic nature, that is, the necessity of solving two Spartan 3E of the Xilinx company. The control law was
differential equations on line. Although it is possible to use programmed in DSP model TMS320F28335. The desired
simple forward difference approximations for both of them, capacitor voltages for both the high- and low-frequency
special attention must be delivered to (26) since after converters were 33 and 99 V, respectively, while the
substitution of the control law (25) it becomes into a non- considered system parameters are shown in Table 2. A
linear differential equation. In these sense, there are two complete picture of the system is included in Fig. 1, while
alternatives for dealing with its implementation: To directly the load parameters are listed in Table 3.
solve the non-linear differential equation obtaining in a The experiment is divided into two steps. In the first, it was
direct way the variable x2d , or simplify the computations by assumed that the capacitors were operating under nominal
solving the equivalent first-order differential (16) at the voltages and at time t ¼ 0 the aforementioned load was
expense of introducing the operator square root in order to applied. To illustrate the mitigation properties of the
recover the value of x2d . In the experience of the authors, the controller, in Fig. 2 the behaviour of the different currents
second is a better alternative since the computational effort involved in the circuit are shown. In the top of the figure it
demanded for dealing with the non-linear equation is higher is included the current drawn by the non-linear device while
than the required by computing the square root of the variable. the compensation current generated by the asymmetric nine-
Another important point to be considered in the controller level inverter is shown in the middle of the figure. Finally,
implementation is related with the possibility of having a the compensated current drawn into the utility is presented
singular operation when x2d ¼ 0 in (25). Fortunately, this in the bottom of the figure where it must be noted its
condition is not achievable under normal operation of the closely sinusoidal behaviour.
converter and can be avoided with a proper selection of the
initial conditions when the differential equation is solved
Table 2 Inverter and AC side parameters and damping values
and assuming that the capacitors have been charged
previously to start the operation of the converter, operation L 6.4 mH Rs 1.5 V
that can be carried out by one of the several techniques V 75 Vpeak v 377 rad/s
available in the literature. R 150 V C1 470 mf
Owing to the current availability of experimental platforms C2 470 mF k1 1.4
that allows the designer to obtain a machine code by k2 20.4 × 1026 k3 20.5 × 1026
programming in high-level languages, providing an easy
way for carrying out products and division; for instance, the
final issue that deserves special attention is the measuring
one. In this sense, the main problem for implementing Table 3 Non-linear (single-phase diode rectifier) parameters
the controller is related to the necessity of computing the Pmax 254 W
desired value x1d , namely, obtaining the magnitude of Cr 2200 mF
the harmonic components. However, dealing with this Lr 4 mH
complication can be done in a relatively simple way since it
is possible to show that this term can be equivalently
expressed as SNn=2 In2 = 2i2NLh , where iNLh can be obtained by
transforming the measured current using a single-phase dq
transformation [26] and eliminating, by mean of a low-pass
filter, its fundamental component.

Remark 11: After considering the main complications that

arise in the implementation of the proposed controller, it
can be concluded that the required computational
complexity can be approached with the currently available
experimental platforms with the additional advantage that
tuning the controller is quite simple.

5 Experimental results
The results obtained from the experimental evaluation of the
proposed controller are presented in this section. The purpose
of this evaluation was twofold, namely, to illustrate the ability
of the controller for regulating the capacitor voltages while
accomplishing the mitigation of the harmonic current drawn
for a fixed non-linear load and to illustrate its ability for
rejecting disturbances generated by changing the structure Fig. 2 From top to down: current drawn by the non-linear load,
of the load. the multilevel inverter and the utility

244 IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237 –247
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-pel.2010.0131
 www.ietdl.org
To illustrate in a better way the compensation capability of the
proposed control scheme, in Fig. 3, Figs. 4 and 5 the spectral
content of the three currents displayed in Fig. 2 are presented.
From here, it can be noticed that while the total harmonic
distortion (THD) of non-linear load takes a value about 67%,
the corresponding to the compensated current is reduced to
about 17%.
Concerning the regulation objectives, in Fig. 6 are shown
the voltages of both capacitors together with the modulated
voltage applied in terminals of the load, the latter to
illustrate the multilevel behaviour of the converter. The
voltage corresponding to the high-frequency converter is
included in the top of the figure while the exhibited by the
low-frequency converter is shown in the middle of the
figure. As can be noticed, there exists a slight difference
between the desired (average) values of these signals and
the actual ones. However, this discrepancies are usual in an
experimental setting because of the unmodelled dynamics
that are present under this kind of evaluation. Actually, the
1.2% error for the high-frequency converter and the 0.7%
error for the low frequency converter can be considered Fig. 5 Spectral content of the utility current
reasonable.

Fig. 6 From top to down: capacitor voltage of high-frequency

Fig. 3 Spectral content of the non-linear load current converter, capacitor voltage of low-frequency converter and
modulated voltage in load terminals

The second part of the experiment was designed to

illustrate the robustness properties of the control scheme, in
particular with respect to step changes in the resistive
element of the load. In this sense, in Fig. 7 it is presented
the behaviour of the capacitors voltages when the load
resistance changes from RL ¼ 26 V to RL ¼ 52 V while in
Fig. 8 the response of these variables are included when the
inverse load change, i.e. from RL ¼ 52 V to RL ¼ 26 V, is
applied. In both experiments it can be observed that the
capacitors voltages remain in acceptable values even when
the load current behaves as illustrated in the bottom of the
corresponding figures.
To complement the evaluation of the proposed controller
and with the purpose to show its usefulness in a more
general setting, in Figs. 9 and 10 the behaviour of the
different currents and voltages involved in the operation of
the converter are included when instead of considering a
1:3 voltage ratio, it is considered a 1:2 relationship. From
Fig. 4 Spectral content of the multilevel inverter current these figures it is clear that also under this operation

IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237–247 245

doi: 10.1049/iet-pel.2010.0131 & The Institution of Engineering and Technology 2012
www.ietdl.org

Fig. 7 From top to down: capacitor voltage of high-frequency

converter, capacitor voltage of low-frequency converter and multilevel Fig. 10 From top to down: capacitor voltage of high-frequency
inverter current under a load resistance change from 26 to 52 V converter, capacitor voltage of low-frequency converter and
multilevel inverter current with a 1:2 voltage ratio

characteristic the performance achieved under the proposed

controller is remarkable. In this case, the desired capacitors
voltages for the (seven-level) converter were 50 and 100 V.

6 Conclusions
In this paper, it has been illustrated the advantage of including in
the controller design of an asymmetric nine-level inverter, when
harmonic current mitigation objectives are pursued, a detailed
analysis of its non-linear dynamic behaviour. The contribution
can be divided into three parts, namely, statement of explicit
mathematical expression for the control law that is required to
apply to the converter in order to achieve the control
objectives in steady-state operation, exploitation of the
acquired knowledge in order to design an stabilisation
mechanism that renders the desired behaviour attractive and
illustration (via an experimental evaluation) of the stability,
Fig. 8 From top to down: capacitor voltage of high-frequency performance and robustness properties of the proposed
converter, capacitor voltage of low-frequency converter and multilevel scheme. Regarding the first part of the contribution, an
inverter current under a load resistance change from 52 to 26 V interesting byproduct is the clear establishment, in terms of
very precise mathematical expressions, of several well-known
facts recognised in a practical setting about the device
operation, while in the second part, the attractiveness of the
result lies in the exploitation of the passivity (energy-
dissipation) properties of the converter to obtain a relatively
simple and easy-to-tune control scheme.

7 Acknowledgments
Part of the work of the first author was supported by DGEP –
UNAM, while part of the work of the second author was
supported by DGAPA– UNAM (IN111211) and CONACYT
(51050).
The work of G. Espinosa–Pérez was done during a sabbatical
stay at LSS–SUPELEC, France, supported by SUPELEC,
DGAPA–UNAM, CONACYT and Université Paris–Sud XI.

8 References
Fig. 9 From top to down: current drawn by the non-linear load, 1 Akagi, H.: ‘New trends in active filters for power conditioning’, IEEE
the multilevel inverter and the utility with a 1:2 voltage ratio Trans. Ind. Appl., 1996, 32, pp. 1312–1322

246 IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237 –247
& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-pel.2010.0131
 www.ietdl.org
2 Rodriguez, J., Lai, J.-S., Peng, F.Z.: ‘Multilevel inverters: a survey of 14 Costa-Castell, R., Gri, R., Cardoner, R., Fossas, E.: ‘High performance
topologies, controls and applications’, IEEE Trans. Ind. Electron., control of a single-phase shunt active filter’. Proc. IEEE Int. Symp. on
2002, 49, pp. 724–738 Industrial Electronics, pp. 3350– 3355
3 Rodriguez, J., Bernet, S., Wu, B., Pontt, J.O., Kouro, S.: ‘Multilevel 15 Desoer, C.A., Khu, E.S.: ‘Basic circuit theory’ (International Student
voltage-source-converter topologies for industrial medium-voltage Series, McGraw-Hill, 1969)
drives’, IEEE Trans. Ind. Electron., 2007, 54, (6), pp. 2930–2945 16 Ortega, R., Loria, A., Nicklasson, P.J., Sira-Ramirez, H.: ‘Passivity-
4 Lai, J.-S., Peng, F.Z.: ‘Multilevel converters: a new breed of power based control of Euler– Lagrange systems: mechanical, electrical and
converters’, IEEE Trans. Ind. Appl., 1996, 32, pp. 509– 517 electromechanical applications’ (Springer-Verlag, 1998)
5 Marietoz, S., Rufer, A.: ‘New configurations for the three-phase 17 van der Schaft, A.: ‘L2-gain and passivity techniques in nonlinear
asymmetrical multilevel inverter’. 39th IAS Annnual Meeting of the control’ (Springer Verlag, London, 2000, 2nd edn. edn.)
Industry Applications Conf., 2004, vol. 2, pp. 828– 835 18 Ortega, R., Garcia-Canseco, E.: ‘Interconnection and damping assignment
6 Abdul, M.N., Hussien, Z.F.: ‘Asymmetrical multilevel inverter passivity-based control: a survey’, Eur. J. Control, 2004, 10, pp. 432–450
using ratio-three based sources’. Power and Energy Conf., 2004, 19 Noriega, D., Espinosa-Perez, G., Cardenas, V., Alvarez-Ramirez, J.:
pp. 137–142 ‘Passivity-based control of multilevel cascade inverters: high-
7 Lai, Y.-S., Shyu, F.-S.: ‘Topology for hybrid asymmetric performance with reduced switching frequency’, Int. J. Robust
multilevel inverter’, IEE Proc. -Electrc. Power Appl., 2002, 149, Nonlinear Control, 2010, 20, (9), pp. 961–974
pp. 449–458 20 Chan, C.Y.: ‘Parallel damped passivity-based control of quasi-resonant
8 Varschavsky, A., Dixon, J., Rotella, M., Moran, L.: ‘Cascaded nine- converters’, IEE Proc. Electric Power Appl., 2005, 152, pp. 1093– 1100
level inverter for hybrid series active power filter using industrial 21 Leyva, R., Cid-Pastor, A., Alonso, C., Queinnec, I., Tarbouriech, S.,
controller’, IEEE Trans. Ind. Electron., 2010, 57, (8), pp. 2761–2767 Martínez-Salamero, L.: ‘Passivity-based integral control of a boost
9 Manjrekar, M.D., Steimer, P.K., Lipo, T.A.: ‘Hybrid multilevel power converter for large-signal stability’, IEE Proc. Control Theory Appl.,
conversion system: a competitive solution for high-power 0205, 50, pp. 1025– 1031
applications’, IEEE Trans. Ind. Appl., 2000, 36, pp. 834–841 22 Prez, M., Ortega, R., Espinoza, J.R.: ‘Passivity-based PI control of
10 Veenstra, M., Rufer, A.: ‘Control of a hybrid asymmetric multilevel switched power converters’, IEEE Trans. Control Syst. Technol.,
inverter for competitive medium-voltage industrial drives’, IEEE 2004, 12, pp. 881–890
Trans. Ind. Appl., 2005, 41, (2), pp. 655– 664 23 Dell’ Aquila, A., Liserre, M., Giuseppe Monopoli, V., Rotondo, P.: ‘An
11 Gonzalez, S.A., Valla, M.I., Christiansen, C.F.: ‘Five-level cascade energy-based control for an n-H-bridges multilevel active rectifier’,
asymmetric multilevel converter’, IET Power Electron, 2010, 3, (1), IEEE Trans. Ind. Electron., 2005, 52, pp. 670– 678
pp. 120–128 24 Jeltsema, D., Scherpen, J.M.A.: ‘A power-based perspective in
12 Costa-Castelló, R., Griñó, R., Fossas, E.: ‘Odd-harmonic digital modeling and control of switched power converters’, IEEE Ind.
repetitive control of a single-phase current active filter’, IEEE Trans. Electron. Mag., 2007, 1, pp. 7 –9 and 54
Power Electron., 2004, 19, pp. 1060– 1068 25 Khalil, H.K.: ‘Nonlinear systems’ (Prentice-Hall, 2002, 3rd edn.)
13 Gri, R., Cardoner, R., Costa-Castell, R., Rossas, E.: ‘Digital repetitive 26 Miranda, H., Cardenas, V., Sandoval, G., Espinosa–Perez, G.: ‘Hybrid
control of a three-phase four-wire shunt active filter’, IEEE Trans. control scheme for a single-phase shunt active power filter’. 38th IEEE
Ind. Electron., 2007, 54, pp. 1495–1503 Power Electronics Specialist Conf., Orlando, FL, USA, 2007

IET Power Electron., 2012, Vol. 5, Iss. 2, pp. 237–247 247

doi: 10.1049/iet-pel.2010.0131 & The Institution of Engineering and Technology 2012