1
I. INTRODUCTION
urrent energy arena is changing. The feeling of dependence on fossil fuels and the progressive increase of its cost is leading
to the investment of huge amount of resources, economical and human, to develop new cheaper and cleaner energy resources
not related to fossil fuels. In fact since decades, renewable energy resources have been the focus for researchers and different
families of power converters have been designed to make the integration of this type of systems into the distribution grid a current
reality. Besides, in the transmission lines, high power electronic systems are needed to assure the power distribution and the
energy quality. Therefore, power electronic converters have the responsibility to carry out these tasks with high efficiency.
The increase of the world energy demand has entailed the apparition of new power converter topologies and new
semiconductors technology capable to drive all needed power. A continuous race to develop higher voltage and current power
semiconductors to drive high power systems still goes on. In this way, the last generation devices are suitable to support high
voltages and currents (around 6.5 kV and 2.5 kA).However, currently there is a tough competition between the use of classic
power converter topologies using high voltage semiconductors and new converter topologies using medium voltage devices. This
idea is shown in Fig. 1, where multilevel converters built using mature medium power semiconductors are fighting in a
development race with classic power converters using high power semiconductors which are under continuous development and
are not mature. Nowadays, multilevel converters are a good solution for power applications due to the fact that they can achieve
high power using mature medium power semiconductors technology [1][2].
Multilevel converters present great advantages compared with typical and very well known two-level converters [1],[3]. These
advantages are fundamentally focused on improvements in the output signals quality and a nominal power increase in the
converter. In order to show the improved quality of the output voltages of a multilevel converter, the output voltages of a single-
phase two level converter is compared to a 3 and 9 level multilevel waveform in Fig. 2. The power converter output voltage
improves its quality as the number of levels increases reducing the Total Harmonic Distortion of the system.
These properties make multilevel converters very attractive to the industry and nowadays, researchers all over the world are
spending great efforts trying to improve multilevel converter performances such as the control simplification [4][5] and the
performance of different optimization algorithms in order to enhance the Total Harmonic Distortion (THD) of the output signals
[6][7], the balancing of the DC capacitors voltage [8][9], the ripple of the currents [10][11]. For instance, nowadays researchers
are focused on the harmonic elimination using pre-calculated switching functions [12], harmonic mitigation to fulfill specific grid
codes [13], the development of new multilevel converter topologies (hybrid or new ones) [14] and new control strategies
The Age of Multilevel Converters Arrives
Leopoldo G. Franquelo, IEEE Fellow, Jose Rodríguez, IEEE Senior Member, Jose I. Leon, IEEE
Member, Samir Kouro, IEEE Student Member, Ramon Portillo, IEEE Student Member, Maria M.
Prats, IEEE Member
C
2
[15][16].
The most common multilevel converter topologies are: Neutral-Point-Clamped Converter (NPC) [17], Flying Capacitor
Converter (FC) [18], and Cascaded H-Bridge Converter (CHB) [19]. These converters can be classified among the power
converters for high power applications according to Fig. 3. Several surveys on multilevel converters have been published to
introduce these topologies [1][2]. In 1980s, power electronics concerns were focused on the converters power increase
(increasing voltage or current). In fact, current source inverters were the main focus for researchers in order to increase the
current. However, other authors began to work on the idea of increasing the voltage instead the current. In order to achieve this
objective, authors were developing new converter topologies and in 1981, A. Nabae, I. Takahashi and H. Akagi presented the
first Neutral-Point-Clamped PWM converter (NPC) also named diode-clamped converter [17]. This converter was based on a
modification of the classic two-level converter topology adding two new power semiconductors per phase (see Fig. 1). Using this
new topology, each power device has to tolerate at the most half voltage compared with the two-level case with the same DC-
Link voltage. So, if these power semiconductors have the same characteristics than the two-level case, the voltage can be
doubled. The NPC converter was generalized in [21],[22] in order to increase the number of output levels, and was referred as
multi-point clamped converter (MPC), although it has not reached the medium voltage market jet.
Years later, other multilevel converter topologies as the FC [18] or CHB [19],[20] appeared. These multilevel converters
present different characteristics compared with NPC as the number of components, modularity, control complexity, efficiency
and fault tolerance. Depending on the application, the multilevel converter topology can be chosen taking into account these
factors as it is shown in Table I.
TABLE I
COMPARISON OF MULTILEVEL CONVERTER TOPOLOGIES DEPENDING ON IMPLEMENTATION FACTORS
NPC FC CHB
Number of
components
↑switches
↑diodes
↑switches
↑capacitors
↓diodes
↓switches
↓diodes
isolated DC
sources
Modularity
Low High High
Control
complexity
Medium High High
Control concerns
Voltage
balancing
Voltage setup
Power
sharing
Fault tolerance
Difficult Easy Easy
Nowadays, there are several commercial multilevel converter topologies which are sold as industrial products for high power
applications [23]-[25]. However, although the advantages of using multilevel converters have been demonstrated, there has not
been an industrial boom in the application of these power systems in the electrical grid in spite of their demonstrated good
features to be used as medium voltage drives. Maybe, technological problems as reliability, efficiency, the increase of the control
complexity and the design of simple and fast modulation methods have been the barrier that has slowed down the application of
multilevel converters all over the world. Finally, the effort of researchers has overcome this technical barrier and it can be
affirmed that multilevel converters are prepared to be applied as a mature power system in the electric energy arena.
This work is devoted to review and analyze the most relevant characteristics of multilevel converters, to motivate possible
solutions, and to show that we are in a decisive instant in which energy companies have to bet for these converters as a good
solution compared with classic two-level converters. The paper is organized as follows. In section II, a brief overview of the
actual applications of multilevel converters is presented. An introduction of the modeling techniques and the most common
modulation strategies is respectively presented in sections III and IV. Finally, the operational and technological issues have been
addressed in section V and some conclusions are presented in last section.
3
II. MULTILEVEL CONVERTER DRIVEN APPLICATIONS
Multilevel converters are considered today as a very attractive solution for medium voltage high power applications. In fact,
several major manufacturers commercialize NPC, FC or CHB topologies with a wide variety of control methods, each one
strongly depending on the application. Particularly the NPC has found an important market in more conventional high power ac
motor drives applications like conveyors, pumps, fans and mills among others, which offer solutions for industries including oil
and gas, metals, power, mining, water, marine and chemistry [26][27].
The back to back configuration for regenerative applications has been also a major hit of this topology, used for example in
regenerative conveyors for the mining industry [28], or grid interfacing of renewable energy sources, like wind power [29][30].
On the other hand FC converters have found particular applications for high bandwidth – high switching frequency applications
such as medium voltage traction drives [31]. Finally the cascaded H-bridge has been successfully commercialized for very high
power and power quality demanding applications up to a range of 31MVA, due to its series expansion capability. This topology
has also been reported for active filter and reactive power compensation applications [32], electric and hybrid vehicles [33][34],
photovoltaic power conversion [35]-[37], uninterruptible power supplies [38], and Magnetic Resonance Imaging [39]. As an
example of a commercial multilevel power converter a 34kV-15MW three-phase six-cell CHB converter from SIEMENS for
regenerative drives is shown in Fig. 4. A summary of multilevel converter driven applications is illustrated in Fig. 5.
III. MODELS: A TOOL TO ENHANCE MULTILEVEL CONVERTER POSSIBILITIES
The simulation and the determination of “Input to Output (I/O)” relations are a fundamental task in the study and design
process of the multilevel converters. These I/O relations become essential for the development of suitable models which allow to
obtain all the necessary information about the converter previously to the implementation stage. The modeling of multilevel
converters is not a trivial task since they are made up of linear and non linear components. Historically, modeling techniques
applied to DC power electronics converters have been adapted to be used in the study of AC ones, giving place to different
approximations that achieve, according to their objectives, snubber circuits design, control schemes and controllers development,
steady state study, dynamic and transient response study, stability analysis, etc. The operation of the multilevel converter is a
periodic sequencing of its possible states corresponding to discrete states of the switches. Fig. 6 shows a three-level NPC phase
has and the two possible modeling techniques. Taken these remarks into account, two types of models can be developed:
equivalent circuit simulation or state-space averaged.
A. Circuit Simulation Modeling of Multilevel Converters
A model of the converter can be obtained with the help of powerful simulation tools as SPICE-based simulators. In this case,
the modeling of the multilevel converters is reduced to the generation of an adequate electric circuit model that fully includes the
non-linearities of the switches allowing the complete characterization of the system dynamic. Considering ideal switches, a linear
description of the converter can be obtained for every switching state of the power converter. Fig. 6 shows a phase of the three-
level diode-clamped converter where the switches have been replaced by an ideal switch and can be easily seen that the phase
acts like a voltage source for every switch position so a linear equivalent circuit description of the converter phase can be
obtained for each one. With this model, a linear piecewise simulation can be carried out. If the integration method for the model
equations is properly chosen [40], the simulation time and results accuracy are good enough. However, this modeling approach
often leads to large simulation times and possible unreliable results due to convergence problems. The main drawbacks of this
modeling technique are that the integration of advanced control techniques with the model is almost impossible [40] and that the
4
model is usually complex being its use for control design often troublesome [41][42]. These models can be used in the tuning
process of the control loops and to evaluate the high order harmonics due to switching that can be easily seen on currents shown
in Fig. 6.
B. State-Space Averaged Modeling of Multilevel Converters
State-space averaged models can be easily obtained from the discrete ones when varying quantities are assumed as their
averaged value over a switching period, remaining the DC value of those quantities. Since in AC converters these quantities are
time varying even in the steady state, it is necessary to make a change of coordinates to convert AC sinusoidal quantities to DC
quantities previously to the averaging process [43][44]. Time invariant systems controller design techniques can be used with
these models when important components other than the fundamental harmonic are not present in the system. With the
transformation to this “Rotating Reference Frame” DC quantities corresponds to the fundamental harmonic of the signals, but
some multilevel converter topologies are not completely characterized by only the first harmonic and it is necessary to draw on to
“Harmonic models” where a greater number of harmonics are taken into account obtaining an adequate modeling of the converter
[41]. These “Harmonic models” are complex and only some advanced complex control techniques are suitable to be applied to
them [42].
Recently a new state-space averaging modeling technique has been introduced based on approximations over the exact
averaged linear piecewise characteristics of the converter [30]. In the phase of the three-level diode-clamped converter shown in
Fig. 6, the ideal switch will be switching between the three possible states so an average model can be deduced considering
a
d
as
the averaged value of the switch position. Fig. 6 shows the graphic representation of the exact averaged linear piecewise
approximation and the proposed quadratic approximation [29]. This technique provides simple enough models to be used in the
controller design [45] and carries out fast simulations without convergence problems due to the continuous nature of the obtained
equations. Therefore, the use of these models overcomes one of the technological handicaps in which the multilevel converters
are involved, making the design stage of multilevel power systems a more accessible task. Fig. 6 shows the currents obtained with
this kind of model and when compared with those obtained with the equivalent circuit simulation it can be seen that the results are
almost the same except for the high order harmonics.
IV. MULTILEVEL MODULATION METHODS
Multilevel inverter modulation and control methods have attracted much research and development attention over the last
decade [1][2][46][47]. Among the reasons are: the challenge to extend traditional modulation methods to the multilevel case, the
inherent additional complexity of having more power electronics devices to control, and the possibility to take advantage of the
extra degrees of freedom provided by the additional switching states generated by these topologies. As consequence, a large
number of different modulation algorithms have been developed, each one with unique features and drawbacks, depending on the
application.
A classification of the modulation methods for multilevel inverters is presented in Fig. 7. The modulation algorithms are
divided into two main groups depending on the domain in which they operate: the state space vector domain in which the
operating principle is based on the voltage vector generation, and the time domain in which the method is based on the voltage
level generation over a time frame. In addition, in Fig. 7 the different methods are labeled depending on the switching frequency
they produce. In general, low switching frequency methods are preferred for high power applications due to the reduction of
switching losses, while the better output power quality and higher bandwidth of high switching frequency algorithms are more
5
suitable for high dynamic range applications.
A. Multilevel Converters PWM strategies
Traditional Pulse Width Modulation (PWM) techniques [48] have been successfully extended for multilevel converter
topologies, by using multiple carriers to control each power switch of the converter. Therefore, they are known as Multicarrier
PWM methods as shown in Fig. 7. For multicell topologies, like FC and CHB, each carrier can be associated to a particular
power cell to be modulated independently using sinusoidal bipolar PWM and unipolar PWM respectively, providing an even
power distribution among the cells. For a converter with m cells, a carrier phase shift of 180º/m for the CHB and of 360º/m for
the FC is introduced across the cells to generate the stepped multilevel output waveform with low distortion [23]. Therefore this
method is known as Phase Shifted PWM (PS-PWM). The difference between the phase shifts and the type of PWM (unipolar or
bipolar) is because one CHB cell generates 3-level outputs, while one FC cell generates two level outputs. This method naturally
balances the capacitor voltages for the FC, and also mitigates input current harmonics for the CHB.
The carriers can also be arranged with shifts in amplitude relating each carrier with each possible output voltage level
generated by the inverter. This strategy is known as Level Sifted PWM (LS-PWM), and depending on the disposition of the
carriers, they can be in Phase Disposition (PD-PWM), Phase Opposition Disposition (POD-PWM) and Alternate Phase
Opposition Disposition (APOD-PWM) [49], all shown in Fig. 7.
A in depth assessment between these PWM methods can be found in [50]. LS-PWM methods can be implemented for any
multilevel topology, however, they are more suited for the NPC, since each carrier signal can be easily related to each power
semiconductor. Particularly LS-PWM methods are not very attractive for CHB inverters, since the vertical shifts relate each
carrier and output level to a particular cell, producing an uneven power distribution among the cells. This power unbalance
disables the input current harmonic mitigation that can be achieved with the multipulse input isolation transformer, reducing the
power quality.
Finally, the hybrid modulation is in part a PWM based method which is specially conceived for the CHB with unequal dc
sources [14],[51]-[53]. The basic idea is to take advantage of the different power rates among the cells of the converters to reduce
switching losses and improve the converter efficiency. This is achieved by controlling the high power cells at fundamental
switching frequency by turning on and off each switch of each cell only one time per cycle, while the low power cell is controlled
using unipolar PWM. Also asymmetric or hybrid topologies have been proposed based on the MPC structure [54].
B. Space Vector Modulation techniques
Space Vector Modulation (SVM) is a technique where the reference voltage is represented as a reference vector to be
generated by the power converter. All the discrete possible switching states of the converter lead to discrete output voltages and
they can be also represented as the possible voltage vectors (usually named state vectors) that can be achieved. SVM technique
generates the voltage reference vector as a linear combination of the state vectors obtaining an averaged output voltage equal to
the reference over one switching period [55].
In the last years, several space vector algorithms extended to multilevel converters have been found in bibliography. Most of
them are particularly designed for a specific number of levels of the converter and the computational cost and the algorithm
complexity are increased with the number of levels. Besides, these general modulation techniques for multilevel converters
involve trigonometric function calculations, look-up tables or coordinated system transformations which increases the
computational load.
Recent SVM strategies have drastically reduced the computational effort and the complexity of the algorithms compared with
