This paper presents a novel switching voltage model and an offsetbased pulse width modulation (PWM) scheme for multilevel inverters with unbalanced DC sources. The switching voltage model under a DC voltage imbalance will be formulated in general form for multilevel neutralpointclamped topologies. Analysis of the reference switching voltages from active and nonactive switching voltage components in
abc
coordinates can enable voltage implementation for an unbalanced DCsource condition. Offset voltage is introduced as an indispensable variable in the switching voltage model for multilevel voltagesource inverters. The PWM performance is controlled through the design of two offset components in a subsequence. One main offset may refer to the common mode voltage, and the other offset restricts its effect on the quality of PWM control in related DC levels. The PWM quality can be improved as the switching loss is reduced in a discontinuous PWM mode by setting the local offset, which is related to the load currents. The validity of the proposed algorithm is verified by experimental results.
I. INTRODUCTION
Multilevel inverters play an important role in current highperformance applications. Two topologies have become popular in practice, namely, multilevel neutralpointclamped (NPC) inverter and multilevel cascaded inverter, as shown in
Fig. 1
. Wellknown pulse width modulation (PWM) methods include carrierbased PWM (CPWM) and space vector PWM (SVPWM) techniques
[1]

[8]
.
Multilevel inverter circuits. (a) Fivelevel diodeclamped inverter. (b) Fivelevel cascaded inverter.
In the SVPWM method, the switching states in the switching sequences are selected after determining the three pivot vectors in the space vector diagram. The control process of the SVPWM is relatively complicated in a multilevel inverter
[14]
,
[15]
. When the inverter level is high, the two main disadvantages of this PWM method are burden calculation and lookup table, both of which require large memory storage.
CPWM techniques are commonly used in practical applications because of their simple implementation. Algorithms of a SVPWM scheme with the nearest three voltage vectors can be completely realized by a corresponding CPWM method.
The offset feature in a twolevel inverter has been extensively investigated in previous studies
[9]

[11]
. In multilevel inverters, a CPWM method can offer flexible control of vector redundancies through proper offset regulation
[4]
,
[20]
.
Offset voltage significantly affects converter performance. Offset adjustments can generate different PWM modes
[1]
,
[3]
,
[26]

[31]
. Proper selection of offset voltage can reduce switching amount and the harmonic distortion factor in different PWM techniques
[4]
,
[24]
, and its regulation with load currents may minimize switching losses
[25]
. Therefore, offset voltage is practically indispensable and is an important variable in the control model of a multilevel voltagesource inverter.
Offset control is also available in space vector modulation by redistributing the switching time duties of two redundant states
[10]

[13]
. However, in multilevel inverters, an algorithm based on the SVPWM method may generally be difficult to apply if the offset is used as a variable to control DC and neutral currents, and thus to balance DC voltages and reduce the effect of common mode voltages (CMVs) on electrical drives
[16]

[18]
.
The unbalanced conditions of DClink voltages that are fed to an inverter have a negative effect on the output quality. The imbalance, which is represented by the ripple on the DC link, causes distorted output waveforms with lowfrequency harmonics. This phenomenon can lead to degradation in the load performance
[38]
. Solutions to cope with this problem can be found in
[19]
,
[23]
,
[35]

[39]
. The works in
[35]

[37]
focused on a twolevel inverter and other classic topologies. In
[38]
, a feedforward SVPWM method was proposed to obtain balanced output voltages in a threelevel NPC inverter. Other than the limitation represented by the complication of the control process in this method, the flexibility of the offset voltage has not been fully explored.
In
[19]
, a simple carrierbased method with DClink ripple feedforward compensation in a multilevel inverter was introduced. The idea of this work was to modify the carriers according to the DClink fluctuation. However, given that a sinusoidal reference was assumed and the offset voltage was constant, the maximum modulation index could not be greater than 0.866, which corresponded to the maximum modulation index of the conventional sine PWM method in a balanced condition.
In
[23]
and
[39]
, a modified SVPWM method was applied to multilevel converters to solve the problem of unbalanced DC conditions. Under balanced and unbalanced conditions, the methods analyzed the control voltage in 3D coordinates and offered a simpler control process than the 2DSVM in
[38]
. Nevertheless, the characteristics of the offset were not analyzed. In addition, given that no harmonic injection was considered, the modulation index limitation of 0.866 was unavoidable, as in
[19]
.
The present study proposes a new CPWM technique for multilevel diodeclamped inverters under balanced and unbalanced conditions. The proposed algorithm is applicable to NPC inverters with arbitrary numbers of levels. With the actual conditions of the DC links considered, a novel approach to modeling switching voltages and an offsetbased PWM scheme using
abc
coordinates are systematically analyzed. The switching voltages characterize the control characteristics of multilevel inverters, while the offsetbased PWM quality control improves the output quality following some particular demands. In the proposed PWM scheme, the global offset component enables control of the draft CMV, and the local offset component sets the PWM modes. A currentbased discontinuous PWM (DPWM) method that utilizes these two offset components to reduce switching loss is also proposed.
The modulation index control has been improved as compared with those in
[19]
,
[23]
,
[39]
. Therefore, the attainable maximum linear range of the output voltage control (under unbalanced input conditions) is always achieved.
The validity of the proposed feedforward PWM method will be demonstrated through simulation and experimental results.
II. SWITCHING VOLTAGE FORMULATION FOR MULTILEVEL INVERTERS UNDER INPUT VOLTAGE IMBALANCE
 A. Modeling of the Switching Voltage for a Multilevel NPC Inverter
In the fivelevel NPC topology in
Fig. 1
(a), the DClinkfed inverter voltages are assumed to have the values
V_{1}
,
V_{2}
,
V_{3}
, and
V_{4}
. With a selected neutral point “O” and the designated switches in the Xphase (X = A,B,C) represented as
SW_{1X}
,
SW_{2X}
,
SW_{3X}
, and
SW_{4X}
, respectively, as in
Fig. 1
(a), the pole (leg) voltage V
_{XO}
can be determined as
where
s
_{1}
_{x}
,
s
_{2}
_{x}
,
s
_{3}
_{x}
, and
s
_{4}
_{x}
represent the switching states of
SW_{1X}
,
SW_{2X}
,
SW_{3X}
, and
SW_{4X}
, respectively; s
_{1A}
is “1” if
SW_{1A}
is ON; otherwise, its value is “0.”
We define the
Xphase switching voltage, V_{SX}
, (X = A,B,C), which is controlled by switches, as follows:
The switching voltage presents a switchingcontrolled voltage source to supply the load. Eq. (2) shows that the switching voltage consists of (n − 1) switching voltage components. j
^{th}
switching voltage component is defined as a product of j
^{th}
switching state and its corresponding DC voltage cell, that is,
The constraint between switch states for the fivelevel NPC inverter shown in
Fig. 1
(a) is simply expressed as
During a sampling period for an Xphase inverter leg, X = A,B,C, only one switch exists, called the active switch
s_{X}
; this switch is actively turned on and off. The states of the remaining (n − 2) switches, called the nonactive switches, are unchanged (
s_{jX}
= or 0
s_{jX}
= 0). Each nonactive switch provides full voltage
V_{SXj}
=
V_{j}
= for
s_{jX}
= 1 and zero voltage
V_{SXj}
= 0 for
s_{jX}
= 0. Among (n − 2) nonactive switches, L switches are supposed to hold the ON state, and (n − 2 − L) switches hold the OFF state. We define
V_{LX}
as the component of the nonactive switching voltage provided by the nonactive switches. For an active switch
s_{X}
, if
ξ_{X}
is defined as its average value in a sampling period and
V_{DX}
as its corresponding DClinkfed inverter voltage, the instantaneous and average values of the switching voltages can be expressed in terms of two components, namely, nonactive switching voltage and active switching voltage. These components are expressed as follows:
The relation among average switching voltage
nonactive switching voltage
V_{LX}
and DC voltage linked to active switch
V_{DX}
can be described as follows:
The instantaneous switching voltage model for a single leg relative to the pole voltage is presented in
Fig. 2
(a).
Fig. 3
(b) provides an example of the Aphase switching voltage model in
Fig. 2
(a) for the case in which switch
s_{2A}
is active and has a corresponding DClinkfed inverter voltage
V
_{2}
, and the other nonactive switches
s
_{1}
_{A}
,
s
_{3}
_{A}
,
s
_{4}
_{A}
have respective DClinkfed inverter voltages of
V
_{1}
,
V
_{3}
,
V
_{4}
, which receive values of 0, 1, and 1. The Aphase average voltage model of the example given in
Fig. 2
(b) is illustrated in
Fig. 2
(c).
(a) Relation between Aphase switching voltage and pole voltage. (b) Aphase switching voltage for nonactive switches s_{1}_{A} = 0, s_{3}_{A} = 1, s_{4}_{A} = 1 and active switch s_{2}_{A}. (c) Average voltage model of the Aphase switching voltage.
(a) Average voltage model of the threephase pole voltages with active switches s_{2A}, s_{1B}, s_{3C} and nonactive switches (s_{1A},s_{3A},s_{4A}) = (0,1,1), (s_{2B},s_{3B},s_{4B}) = (1,1,1), and (s_{1C},s_{2C},s_{4C}) = (0,0,1). (b) Average voltage model of the threephase pole voltages expressed in terms of the fundamental
Eq. (1) indicates that the sum of two DClinkfed inverter voltages
V
_{3}
,
V
_{4}
exists in the pole leg voltage formulation. This component presents an offset voltage
V_{offDCLink}
, which is illustrated in the switching voltage model of the fivelevel NPC inverter in
Fig. 2
. In topologies with arbitrary numbers of levels,
V_{offDCLink}
an be calculated as the voltage between the neutral point O and the lowest voltage point of the inverter.
The pole voltage
V_{XO}
can be analyzed in relation to the fundamental output phase voltage as follows:
where
Using Eq. (8) and in consideration of Eq. (1), the instantaneous switching voltage and average switching voltage in a sampling period can be expressed in other forms as follows:
where
the reference fundamental output phase voltage and reference CMV respectively. They satisfy
From the onephase average pole voltage model expressed in terms of average switching voltage and
V_{offDCLink}
as in
Fig. 2
(c), as well as the average pole voltage obtained in Eq. (11), the threephase average pole voltage models that use the two analytical approaches are described in
Fig. 3
. The voltage model in
Fig. 3
(a) corresponds to a specific case in which the active switches for the A, B, and C phases are
s
_{2}
_{A}
,
s
_{1}
_{B}
, and
s
_{3}
_{C}
, respectively; the sets of nonactive switches are (
s
_{1}
_{A}
,
s
_{3}
_{A}
,
s
_{4}
_{A}
) = (0,1,1), (
s
_{2}
_{B}
,
s
_{3}
_{B}
,
s
_{4}
_{B}
) = (1,1,1), and (
s
_{1}
_{C}
,
s
_{2}
_{C}
,
s
_{4}
_{C}
) = (0,0,1).
The relation of the two models in
Figs. 3
(a) and
3
(b) shows that to obtain threephase output voltages with a predefined CMV, the average switching voltages have to be the sum of the fundamental voltage
and a zero sequence voltage injected from the DClinkfed inverter voltages
V_{offDCLink}
. As a result, a new reference average switching voltage model of the Aphase is derived, as shown in
Fig. 4
(b). The complete PWM algorithm for multilevel NPC inverters under balanced and unbalanced conditions will be realized by using this switching voltage model and the model illustrated in relation to the active and nonactive switching voltages in
Fig. 4
(a).
Multilevel NPC inverter: A brief description of the Aphase switching voltage from (a) the active and nonactive switching voltage components and (b) the zero/nonzero sequence voltages.
 B. Fundamental Voltage Limit and the Global Reference CMV Design
The control limits of two parameters, namely, reference fundamental voltage and CMV, should be determined to completely master the switching voltage model in
Fig. 4
.
The maximum (minimum) value of the switching voltage will be obtained if all switches are turned “ON” (“OFF”). In a fivelevel NPC inverter, for example, these values are determined as follows:
As a result, the possible maximum fundamental voltage
of the undermodulation limit is given as follows:
The reference CMV
can generally be expressed as a function of the variable
η
_{1}
, that is,
where
V_{offMX}
and off
V_{offMn}
are the maximum and minimum values of
respectively. These values can be deduced from the switching voltage model in
Fig. 4
and Eqs. (12) and (13) as follows:
The “Min()” function returns the smallest value of the three given values inside the parentheses .
In practical applications, two popular reference average CMV designs are as follows:
Medium CMV
[9]
–
[12]
: This CMV approach is often used in twolevel and multilevel inverters to maximize the linear PWM control range. In this case, the global offset is obtained by setting η
_{1}
= 0.5 in Eq. (15) as follows:
Minimum CMV
[25]
: The reference offset voltage is selected such that the absolute value of the obtained average CMV is at the minimum. The sinusoidal PWM method is a particular case of this PWM technique for a modulation index range lower than 0.866, that is,
III. SWITCHING VOLTAGE IMPLEMENTATION
 A. Switching Voltage PWM Control
To implement the reference switching voltages, the active and nonactive switching voltage components, as illustrated in
Fig. 5
(a), have to be calculated. On the basis of Eqs. (4)(7), the nonactive switching voltage components
V_{LX}
can be simply determined as follows:
where
V_{SXk}
is the k
^{th}
discrete switching voltage, and
n
is the number of levels. For example, in a fivelevel NPC inverter, five discrete switching voltages are computed in increasing order as
V
_{SX}
_{1}
= 0,
V
_{SX}
_{2}
=
V
_{4}
,
V
_{SX}
_{3}
=
V
_{3}
+
V
_{4}
,
V
_{SX}
_{4}
=
V
_{2}
+
V
_{3}
+
V
_{4}
, and
V
_{SX}
_{5}
=
V
_{1}
+
V
_{2}
+
V
_{3}
+
V
_{4}
.
Implementation of the active switching voltage switching sequence and the switching time diagram.
The DClinkfed inverter voltages
V_{DX}
(X = A,B,C), which correspond to the active switches on the three phases, can be determined as the DC cell between the two nearest discrete switching voltages of
For the active switches, their switching state sequence can be realized by the CPWM approach illustrated in
Fig. 5
with the normalized modulating signals
ξ_{X}
, (X = A,B,C), which are calculated as follows:
The switching pattern in
Fig. 5
, from which the active switching states are directly derived, is similar to that of a twolevel inverter. Supposing that in a sampling period the threephase switching sequence in the first half of the pattern in
Fig. 5
is (S
_{A}
,S
_{B}
,S
_{C}
) = (0,0,0) → (1,0,0) → (1,1,0) → (1,1,1), then the corresponding active switching sequence is derived as (0,0,0 → (
V_{DA}
,0,0) → (
V_{DA}
,
V_{DB}
,0) → (
V_{DA}
,
V_{DB}
,
V_{DC}
). The switching voltage sequence is completely realized by using the information of the threephase nonactive voltage and active switching sequence. The switching time diagram of a virtual two level in
Fig. 5
is directly imposed onto the switching voltage sequence to complete the switching voltage pattern. As a result, the characteristics of PWM control for a multilevel NPC inverter under unbalanced conditions are fully explored.
When analyzing in the αβ domain of the SVM method, eight switching states of the active switches will establish a voltage vector diagram (
Fig. 6
). Each discrete voltage vector is calculated as follows:
Vector diagram of the active switching voltages for (a) balanced DC sources and (b) unbalanced DC sources (V_{DA}>V_{DB}>V_{DC}).
In
Fig. 6
, the discrete vector denoted as [S
_{A}
,S
_{B}
,S
_{C}
] represents the active switching voltage state (S
_{A}
.V
_{DA}
,S
_{B}
.V
_{DB}
,S
_{C}
.V
_{DC}
). Under an unbalanced condition, the active switching voltage vector diagram forms an asymmetrical hexagon. The zero vector of [000] remains at the center of the old symmetrical hexagon in
Fig. 6
(a), whereas the zero vector of [111] deviates from its former position. As a result, for a reference vector, the PWM control with the three nearest vector principles cannot be implemented in some regions with minimum switching numbers. For example, when the reference vector has its tip located at point A, as illustrated in
Fig. 6
(b), the use of the three nearest vectors of [000], [100], and [111] will always require considerable switching. The switch from the continuous PWM mode to DPWM mode by simply removing the vector of [000] or [111] is also not allowed. For example, the active switching voltage vectors of [000][100][101][111] in continuous PWM after removing the vector [111] cannot implement the reference voltage vector in
Fig. 6
(b) because the area formed by the tips of the remaining vectors of [000][100][101] does not contain the A point.
 B. Active Switching Voltage Control – Design of the Local Offset for Reduced Switching Loss
From the switching voltage model of the multilevel inverter shown in
Fig. 4
(a), the PWM performance can be improved by modifying the active switching voltages. A new equivalent circuit of the active switching voltages is obtained, as shown in
Fig. 7
, by adding an offset e
_{0}
, which is the local offset voltage, to the threephase average active switching voltages.
Equivalent circuit of the modified average active switching voltages.
Given that
the modulating signals in Eq. (20) will be modified as follows:
1) Reduced Switching Loss PWM under DC Voltage Imbalance
: The control range of the local offset
e
_{0}
depends on the reference leg voltages of the virtual twolevel inverter and the active DC cells of the DClinkfed inverter voltages. The range can be as follows:
where
The control limits of
e
_{0}
corresponding to a specific case of
V_{DX}
and
e_{X}
(X = A,B,C) are illustrated in
Fig. 8
.
Determination of local offset limits from the active voltages and related DClink voltages.
From Eqs. (24)–(26), the local offset can be expressed as a function of the variable
η
_{2}
as follows:
The DPWM mode can be achieved if the value of the parameter η
_{2}
in Eq. (27) is set to 1 or 0. We define
i_{X}
= as the threephase output currents, and
i
_{1}
and
i
_{2}
as the absolute values of the two currents, the respective phases of which can achieve the DPWM by setting the local offset voltage to
e
_{0}
_{MX}
and
e
_{0}
_{MN}
respectively.
I_{MX}
and
I_{MD}
are also defined as the maximum and medium absolute values of the threephase currents, respectively. The selection rule of
e
_{0}
for reduced switching loss can then be proposed as follows:
Applying the local offset defined in Eq. (28) allows only the output phases of maximum or medium absolute currents for DPWM control. This condition helps to avoid commutation on the phase of a large current, thus reducing the switching loss of the power converter.
 C. Proposed PWM Control Scheme
The proposed PWM control scheme for multilevel NPC inverters under a DC voltage imbalance is described in
Fig. 9
on the basis of the theoretical analysis. A global offset voltage is designed by using the information of the feedback DClinkfed inverter voltages and the reference fundamental voltages. The reference switching voltages and the respective active and nonactive components can be calculated afterwards. (
V_{DX}
, e
_{X}
) helps to determine the local offset limits and simplifies the PWM control of nlevel to that of a virtual twolevel inverter. The resultant switching sequences and switching time diagrams of the active voltages are directly derived from those of the twolevel inverter PWM. The final PWM patterns of nlevel NPC are completely realized using this information and in consideration of
V_{LX}
.
Proposed feedforward PWM control scheme for nlevel NPC inverter under DC voltage imbalance conditions.
IV. SIMULATION AND EXPERIMENTAL RESULTS
 A. Simulation Results
Simulations are conducted for a fivelevel NPC inverter with unbalanced DClink voltages and a threephase load R–L in series, where R = 40 Ω and L = 85 mH. The switching frequency is set to
f_{s}
= 2
kHz
, and the output frequency is set to
f
_{0}
= 50
Hz
. Under the conditions of
V_{1}
= 55
V
,
V_{2}
= 45
V
,
V_{3}
= 45
V
, AND
V_{4}
= 55
V
, two modulation indices of
m
=
0.3
and
m
=
0.75
are selected to analyze the typical conventional sinusoidal PWM method without compensation and with feedforward compensation. For the uncompensated algorithm, the waveforms, including output line voltage and current (a) as well as the harmonic spectra of the current (b), are shown in
Figs. 10
and
12
. The obtained fundamental currents for the given loads are not correct, and the harmonic spectra in
Figs. 10
(b) and
12
(b) show that the currents are influenced by loworder harmonics with a large magnitude. For comparison, the same quantities are given for the sinusoidal PWM method with feedforward compensation, as in
Figs. 11
(a) (m = 0.3) and
13
(a) (m = 0.75). Improvements are obtained when the compensated algorithm yields output waveforms in
Figs. 11
(a) and
13
(a) with correct fundamental components and loworder harmonics, as depicted in
Figs. 11
(b) and
13
(b). The obtained total harmonic distortion (THD) values of the output currents are 1.09% (m = 0.3) and 0.52% (m = 0.75), whereas they are 1.2% and 0.58% in the uncompensated system.
(a) Output line voltage and current waveforms. (b) Harmonic spectrum of output currents (Fundamental = 0.63 A, THD = 1.2%) when using conventional sinusoidal PWM without feedforward compensation (m = 0.3).
(a) Output line voltage and current waveforms. (b) Harmonic spectrum of output currents (Fundamental = 0.7018 A, THD = 1.09%) when using conventional sinusoidal PWM with feedforward compensation (m = 0.3).
(a) Output line voltage and current waveforms; (b) harmonic spectrum of output currents (Fundamental = 1.721 A, THD = 0.58%) when using conventional sinusoidal PWM without feedforward compensation (m = 0.75).
(a) Output line voltage and current waveforms; (b) harmonic spectrum of output currents (Fundamental = 1.79 A, THD = 0.52%) when using conventional sinusoidal PWM with feedforward compensation (m = 0.75).
A medium CMV PWM method and a proposed PWM method with reduced switching loss are applied to a fivelevel NPC inverter with the previous configuration unchanged. The medium CMV PWM is achieved by using Equ. (17) and setting the local offset to zero. In the proposed PWM with reduced switching loss, the local offset is selected as the minimum CMV [Eq. (18)], and the local offset is selected as in Eq. (28).
The waveforms obtained using the two PWM methods are shown in
Figs. 14
and
15
. The resulting THDs of the output currents when using the feedforward PWM with the medium CMV are 0.99%, 0.56%, and 0.38%, which correspond to modulation index m values of 0.3, 0.75, and 0.95, respectively. By contrast, the THDs are 1.46%, 0.66%, and 0.59% with the proposed PWM method with reduced switching loss.
Output line voltage and current waveforms when using PWM method with medium CMV and feedforward compensation.
Waveforms of output line voltage and output pole voltage with its corresponding current when using the proposed feedforward PWM method with reduced switching loss.
In the PWM with reduced switching loss, the waveforms of the pole voltage and its respective current for the three different modulation indices are illustrated in
Fig. 15
. No commutation on one phase occurs at some intervals of its maximum or medium absolute current. As a result of this proposed currentcontrolled DPWM mode, the switching loss can be significantly reduced, such as in the balanced input voltage case
[25]
.
Figs. 16
(a)–
16
(c) provide a detailed comparison of the line voltage THD when using the three feedforward PWM control methods: the proposed PWM with reduced switching loss, the conventional sine PWM, and the medium CMV PWM. The THD is calculated up to the 100th harmonics of the output frequency. Each PWM method is analyzed to its maximum modulation index. For the sine PWM method, the maximum value of m is 0.866, whereas it is 1 for the other two methods. The DClink voltages (
V
_{1}
,
V
_{2}
,
V
_{3}
,
V
_{4}
) for the THD analysis as in
Figs. 16
(a)–
16
(c) are set to (50 V, 50 V, 50 V, 50 V), (55 V, 45 V, 45 V, 55 V), and (45 V, 55 V, 55 V, 45 V), respectively.
THD comparison of the three feedforward PWM methods (f_{S} = 5 kHz, R = 40Ω, L = 80 mH).
 B. Experimental Results
Experimental hardware is built for the fivelevel NPC inverter to validate the proposed theory. The algorithm is implemented using the eZdsp TMS320F28335 control kit. Four DC voltages are measured with LEM LV25 NP Hall sensors. For the switching loss PWM algorithm, three additional Hall LEM LA25NP current sensors are used to measure the phase load currents. The frequency of the triangle carrier waveform is 2 kHz, and the desired output frequency is 50 Hz. The experimental load parameters are set to
R
= 40Ω and
L
= 85
mH
.
In the first experiment, two DClink voltages
V
_{1}
,
V
_{4}
are held constant, and the others have large ripples (
Fig. 17
). Each DClink voltage of
V
_{1}
,
V
_{4}
is created by a threephase ullwave diode bridge rectifier and a 6800
µ
F capacitor. Two middle DClink voltages
V
_{2}
,
V
_{3}
, each of which is created by one singlephase fullwave diode bridge rectifier and a 680
µ
F capacitor, have a measured ripple in the range from 42 V to 60 V] when the inverter is in operation at the analyzed modulation index of 0.7.
Setup of the DClink fed inverter voltages (X: 10 ms/div; Y: 50 V/div).
The different capacitors produced different DC sources, which, in turn, caused asymmetrical output voltages. As shown in
Fig. 18
(a), without compensation, the distortion of the output current leads to significant loworder harmonics, as depicted in the harmonic spectrum in
Fig. 18
(b). The sinusoidal current is obtained as in
Fig. 19
(a) by using the compensated algorithm. The harmonic spectrum of the current in
Fig. 19
(b) also demonstrates a great reduction in loworder harmonics. The measured THD of the current with the compensated algorithm is 1.72%, whereas it is 2.725% in the uncompensated system.
(a) Waveforms include output line voltage V_{AB}, pole voltage V_{AO} and current i_{A} (Xaxis:10 ms/div). (b) Harmonics spectrum of output current when using conventional sinusoidal PWM without feedforward compensation (m = 0.7).
(a) Waveforms including output line voltage V_{AB}, pole voltage V_{AO}, and current i_{A} (Xaxis:10 ms/div). (b) Harmonic spectrum of output current when using conventional sinusoidal PWM with feedforward compensation (m = 0.7).
In the second experiment, the DClink voltages are set to
V
_{1}
= 55
V
,
V
_{2}
= 45
V
,
V
_{3}
= 45
V
,
V
_{4}
= 55
V
.
Figs. 20

22
depict the experimental waveforms of the output line voltage and the output current with its respective pole voltage when the feedforward PWM methods of the sine PWM, the PWM with medium CMV, and the proposed DPWM are applied. The pole voltage waveform in
Fig. 22
shows that similar to the simulation results, the implemented DPWM mode that depends on the phase load currents sets the noncommutation phase at intervals, in which its current attains the maximum or medium absolute value.
Waveforms of output line voltage and pole voltage with its corresponding current when using sine PWM method with feedforward compensation (Xaxis:5ms/div).
Waveforms of output line voltage and pole voltage with its corresponding current when using PWM method with medium CMV and feedforward compensation (Xaxis: 5 ms/div).
Waveforms of output line voltage and pole voltage with its corresponding current when using feedforward PWM method with reduced switching loss (Xaxis: 5 ms/div).
A comparison of the experimental line voltage THD (calculated up to the 100th harmonics) of the proposed feedforward DPWM method with the feedforward sine PWM and medium CMV PWM is illustrated in
Figs. 23
(a)
23
(c). modulation index. For the sine PWM method, the maximum value of m is 0.866, whereas it is 1 for the other two methods. The DClink voltages (
V
_{1}
,
V
_{2}
,
V
_{3}
,
V
_{4}
) for the THD analysis as in
Figs. 16
(a)–
16
(c) are set to (50 V, 50 V, 50 V, 50 V), (55 V, 45 V, 45 V, 55 V), and (45 V, 55 V, 55 V, 45 V), respectively.
Experimental THD comparison of the three feedforward PWM methods (f_{S} = 5 kHz, R = 40 Ω, L = 80 mH).
 B. Experimental Results
Experimental hardware is built for the fivelevel NPC inverter to validate the proposed theory. The algorithm is implemented using the eZdsp TMS320F28335 control kit. Four DC voltages are measured with LEM LV25 NP Hall sensors. For the switching loss PWM algorithm, three additional Hall LEM LA25NP current sensors are used to measure the phase load currents. The frequency of the triangle carrier waveform is 2 kHz, and the desired output frequency is 50 Hz. The experimental load parameters are set to
R
= 40Ω and
L
= 85
mH
.
In the first experiment, two DClink voltages
V
_{1}
,
V
_{4}
are held constant, and the others have large ripples (
Fig. 17
). Each DClink voltage of
V
_{1}
,
V
_{4}
is created by a threephase fullwave diode bridge rectifier and a 6800
µ
F capacitor. Two middle DClink voltages
V
_{2}
,
V
_{3}
, each of which is created by one singlephase fullwave diode bridge rectifier and a 680
µ
F capacitor, have a measured ripple in the range from 42 V to 60 V] when the inverter is in operation at the analyzed modulation index of 0.7.
The different capacitors produced different DC sources, which, in turn, caused asymmetrical output voltages. As shown in
Fig. 18
(a), without compensation, the distortion of the output current leads to significant loworder harmonics, as depicted in the harmonic spectrum in
Fig. 18
(b). The sinusoidal current is obtained as in
Fig. 19
(a) by using the compensated algorithm. The harmonic spectrum of the current in
Fig. 19
(b) also demonstrates a great reduction in loworder harmonics. The measured THD of the current with the compensated algorithm is 1.72%, whereas it is 2.725% in the uncompensated system.
In the second experiment, the DClink voltages are set to
V
_{1}
= 55V,
V
_{2}
= 45V,
V
_{3}
= 45V,
V
_{4}
= 55V.
Figs. 20

22
depict the experimental waveforms of the output line voltage and the output current with its respective pole voltage when the feedforward PWM methods of the sine PWM, the PWM with medium CMV, and the proposed DPWM are applied. The pole voltage waveform in
Fig. 22
shows that similar to the simulation results, the implemented DPWM mode that depends on the phase load currents sets the noncommutation phase at intervals, in which its current attains the maximum or medium absolute value.
A comparison of the experimental line voltage THD (calculated up to the 100th harmonics) of the proposed feedforward DPWM method with the feedforward sine PWM and medium CMV PWM is illustrated in
Figs. 23
(a)
23
(c).
The characteristics of the line voltage THD using the three PWM methods under the same conditions of DC voltages as in
Figs. 16
(a)–
16
(c) yield acceptable results in comparison with those obtained by simulation (
Fig. 16
). The experimental THD characteristics indicate that the proposed DPWM method under three given DClink voltage conditions yield higher THD values than the two other PWM methods in nearly the entire region of the modulation index.
V. CONCLUSIONS
In this study, a model of switching voltages and an offsetbased PWM scheme for multilevel NPC inverters under a DC voltage imbalance are proposed. In the proposed scheme, the global offset extends the maximized output voltage range in the undermodulation range and defines a draft CMV. The local offset can be flexibly modified to establish different PWM modes related to the PWM quality, such as switching loss and current ripple. A DPWM that utilizes these flexible offsets and the feedback information of the currents is proposed to reduce switching loss. The proposed PWM for switching loss reduction can be analyzed in relation to other carrierbased methods using the same switching voltage model. Simulations and experiments are conducted to confirm the validity of the proposed switching voltage modeling and the proposed DPWM control method.
Acknowledgements
This research is funded by the Vietnam National Foundation for Science and Technology Development under grant number 103.012011.67, and partly supported by the Network Based Automation Research Center, University of Ulsan, Korea and by the VNUHCM project C20132010.
BIO
NhoVan Nguyen was born in Vietnam in 1964. He received his M.S. and PhD. degrees in Electrical Engineering from the University of West Bohemia, the Czech Republic in 1988 and 1991 respectively. Since 1992, he has been with the Department of Electrical and Electronics Engineering, Ho Chi Minh City University of Technology, Vietnam, where he is currently an Associate Professor. He was with KAIST as a Postdoc Fellow for six months in 2001 and a Visiting Professor for a year in 2003–2004. He was a visiting scholar at the Department of Electrical Engineering, University of Illinois at UrbanaChampaign for a month in 2009. His research interests include modeling and control of switching power supplies, AC motor drives, active power filters, and PWM techniques for power converters. He is a member of the Institute of Electrical and Electronics Engineers (IEEE).
TamKhanh Tu Nguyen received his B.S. and M.S. degrees in Electrical Engineering from Ho Chi Minh City University of Technology, Vietnam, in 2010 and 2012 respectively. He is currently a researcher at the Power Engineering Research Lab, Department of Electrical and Electronics Engineering, Ho Chi Minh City University of Technology, Vietnam. His current research interests include active power filters, PWM techniques for matrix converter, multilevel inverters and advanced control of AC motor drives.
HongHee Lee received his B.S., M.S., and Ph.D. degrees in Electrical Engineering from Seoul National University, Seoul, Korea in 1980, 1982, and 1990, respectively. From 1994 to 1995, he was a Visiting Professor at the Texas A&M University. He has been a Professor at the School of Electrical Engineering in the Department of Electrical Engineering, University of Ulsan, Ulsan, Korea since 1985. He is also the Director of the Networkbased Automation Research Center, which is sponsored by the Ministry of Trade, Industry and Energy. His research interests include power electronics, networkbased motor control, and renewable energy. Dr. Lee is a member of the IEEE, the Korean Institute of Power Electronics (KIPE), the Korean Institute of Electrical Engineers, and the Institute of Control, Robotics, and Systems. He is currently the President of KIPE.
Carrara G.
,
Gardella S.
,
Marchesoni M.
,
Salutari R.
,
Sciutto G.
1992
“A new multilevel PWM method A theoretical analysis,”
IEEE Trans. Power Electron.
7
(3)
497 
505
DOI : 10.1109/63.145137
Rodríguez J.
,
Lai J.S.
,
Peng F. Z.
2002
“Multilevel inverters: A survey of topologies, controls, and applications,”
IEEE Trans. Ind. Electron.
49
(4)
724 
738
DOI : 10.1109/TIE.2002.801052
McGrath B. P.
,
Holmes D. G.
,
Lipo T.
2003
“Optimized space vector switching sequences for multilevel inverters,”
IEEE Trans. Power Electron.
18
(6)
1293 
1301
DOI : 10.1109/TPEL.2003.818827
Nho N.V.
,
Youn M. J.
2006
“Comprehensive study on Space Vector PWM and carrier based PWM correlation in multilevel invertors,”
IEE Proceedings Electric Power Applications
153
(1)
149 
158
DOI : 10.1049/ipepa:20050046
BusquetsMonge S.
,
Bordonau J.
,
Boroyevich D.
,
Somavilla S.
2004
“The nearest three virtual space vector PWM – A modulation for the comprehensive neutralpoint balancing in the threelevel NPC inverter,”
IEEE Power Electronics Letters
2
(1)
11 
15
DOI : 10.1109/LPEL.2004.828445
Celanovic N.
,
Boroyevich D.
2001
“A fast space vector modulation algorithm for multilevel three phase converters,”
IEEE Trans. Ind. Appl.
37
(2)
637 
641
DOI : 10.1109/28.913731
Rodriguez J.
,
Correa P.
,
Moran L.
“A vector control technique for medium voltage multilevel inverters,”
Applied Power Electronics Conference and Exposition (APEC)
2001
Vol. 1
173 
178
Wei S.
,
Wu B.
,
Li F.
,
Liu C.
“A general space vector PWM control algorithm for multilevel inverters,”
in Proc. the Applied Power Electronics Conference and Exposition (APEC '03)
2003
Hava A. M.
,
Kerkman R. J.
,
Lipo T. A.
1998
“A highperformance generalized discontinuous PWM algorithm,”
IEEE Trans. Ind. Appl.
34
(5)
1059 
1071
DOI : 10.1109/28.720446
Blasko V.
“A hybrid PWM strategy combining modified space vector and triangle comparison methods,”
in Proc. IEEE PESC Conf.
1996
1872 
1878
Jacobina C. B.
,
Lima A. M. N
,
da Silva E. R. C.
,
Alves R. N. C.
,
Seixas P. F.
2001
“Digital scalar pulsewidth modulation: A simple approach to introduce nonsinusoidal modulating waveforms,”
IEEE Trans. Power Electron.
16
(3)
351 
359
DOI : 10.1109/63.923767
Van Der Broeck H.
,
Skudelny H.
,
Stanke G.
“Analysis and realization of a pulse width modulator based on voltage space vectors,”
in IEEEIAS Conf. Rec.
1986
244 
251
Zhou K.
,
Wang D.
2002
“Relationship between spacevector modulation and threephase carrierbased PWM: A comprehensive analysis,”
IEEE Trans. Ind. Electron.
49
(1)
186 
196
DOI : 10.1109/41.982262
Pou J.
,
Boroyevich D.
,
Pindado R.
2005
“Effects of imbalances and nonlinear loads on the voltage balance of a neutralpointclamped inverter,”
IEEE Trans. Power Electron.
20
(1)
123 
131
DOI : 10.1109/TPEL.2004.839823
Pou J.
,
Pindado R.
,
Boroyevich D.
2005
“Voltagebalance limits in fourlevel diodeclamped converters with passive front ends,”
IEEE Trans. Ind. Electron.
52
(1)
190 
196
DOI : 10.1109/TIE.2004.837915
Celanovicive K.
,
Boroyevich D.
2000
”A comprehensive study of neutralpoint voltage balancing problem in threelevel voltage source PWM inverters,”
IEEE Trans. Power Electron.
15
(2)
242 
249
DOI : 10.1109/63.838096
Loh P. C.
,
Holmes D. G.
,
Fukuta Y.
,
Lipo T. A.
2003
“Reduced commonmode modulation strategies for cascaded multilevel inverters,”
IEEE Trans. Ind. Appl.
39
(5)
1386 
1395
DOI : 10.1109/TIA.2003.816547
Nho N. V.
,
Lee H. H.
“Generalized carrier PWM algorithms for multilevel inverters with unbalanced DC voltages,”
in Proc. the 37th IEEE Power Electronics Specialists Conference (PESC)
2006
Kouro S.
,
Lezana P.
,
Angulo M.
,
Rodriguez J.
2008
“Multicarrier PWM with DCLink ripple feedforward compensation for multilevel inverters,”
IEEE Trans. Power Electron.
23
(1)
52 
59
DOI : 10.1109/TPEL.2007.911834
Nho N. V.
,
Lee H. H.
“Carrier PWM algorithm for multileg multilevel inverters,”
EPE 2007  12th European Conference on Power Electronics and Applications
2007
Lopez O.
,
Alvarez J.
,
Leon J. I.
,
DovalGandoy J.
,
Freijedo F. D.
2008
“Multilevel multiphase space vector PWM algorithm,”
IEEE Trans. Ind. Electron.
55
(5)
1933 
1942
DOI : 10.1109/TIE.2008.918466
Leon J. I.
,
Vazquez S.
,
Sanchez J. A.
,
Portillo R.
,
Franquelo L.G.
,
Carrasco J. M.
,
Dominguez E.
2010
“Conventional spacevector modulation techniques versus the singlephase modulator for multilevel converters,”
IEEE Trans. Ind. Electron.
57
(7)
2473 
2482
DOI : 10.1109/TIE.2009.2034674
Leon J. I.
,
Vazquez S.
,
Portillo R.
,
Franquelo L. G.
,
Carrasco J. M.
,
Wheeler P. W.
,
Watson A. J.
2009
“Threedimensional feedforward space vector modulation applied to multilevel diodeclamped converters,”
IEEE Trans. Ind. Electron.
56
(1)
101 
109
DOI : 10.1109/TIE.2008.928110
Nho N. V.
,
Hai Q. T.
,
Lee H.H.
2010
“Carrier based singlestate PWM technique of minimizing vector errors in multilevel inverter,”
Journal of Power Electronics
10
(4)
357 
364
DOI : 10.6113/JPE.2010.10.4.357
Nho N. V.
,
Bac N. X.
,
Lee H.H.
2011
“An optimized discontinuous PWM method to minimize switching loss for multilevel inverters,”
IEEE Trans. Ind. Electron.
58
(9)
3958 
3966
DOI : 10.1109/TIE.2010.2102312
Brückner T.
,
Holmes D. G.
2005
“Optimal pulsewidth modulation for threelevel inverters,”
IEEE Trans. Power Electron.
20
(1)
82 
89
DOI : 10.1109/TPEL.2004.839831
Corzine K. A.
,
Baker J. R.
2002
“Multilevel voltagesource dutycycle modulation: analysis and implementation,”
IEEE Trans. Ind. Electron.
49
(5)
1009 
1016
DOI : 10.1109/TIE.2002.803203
Wang F.
2002
“Sinetriangle versus spacevector modulation for threelevel PWM voltagesource inverters,”
IEEE Trans. Ind. Appl.
38
(2)
500 
506
DOI : 10.1109/28.993172
McGrath B. P.
,
Holmes D. G.
,
Meynard T.
2006
“Reduced PWM harmonic distortion for multilevel inverters operating over a wide modulation range,”
IEEE Trans. Power Electron.
21
(4)
941 
949
DOI : 10.1109/TPEL.2006.876864
Pou J.
,
Zaragoza J.
,
Ceballos S.
,
Saeedifard M.
,
Boroyovich D.
2012
“A carrier based PWM strategy with zero sequence voltage injection for a threelevel neutral point clamped converter,”
IEEE Trans. Power Electron.
27
(2)
642 
651
DOI : 10.1109/TPEL.2010.2050783
Chenchen W.
,
Yongdong L.
2010
“Analysis and calculation of Zero sequence voltage considering neutralpoint potential balancing in threelevel NPC converters,”
IEEE Trans. Ind. Electron.
57
(7)
2262 
2271
DOI : 10.1109/TIE.2009.2024093
Dai N. Y.
,
Wong M. C.
,
Chen Y. H.
,
Han Y. D.
2005
“A 3D generalized direct PWM algorithm for multilevel converters,”
IEEE Power Electronics Letters
3
(3)
85 
88
DOI : 10.1109/LPEL.2005.855112
Nho N. V.
,
Lee H. H.
“Analysis of carrier PWM method for common mode elimination in multilevel inverter,”
EPE 2007  12th European Conference on Power Electronics and Applications
2007
Nho N. V.
,
Lee H. H.
“Linear overmodulation control in multiphase multilevel inverters for unbalance DC voltages,”
The 7th IEEE Int'l Conference on Power Electronics and Drive Systems (PEDS 2007)
2007
Enjeti P.
,
Shireen W.
“An advanced programmed PWM modulator for inverters which simultaneously eliminates harmonics and rejects dc link voltage ripple,”
in Proc. Applied Power Electronics Conf. and Expo. (APEC’90)
1990
681 
685
Blaabjerg F.
,
Pedersen J.
,
Thoegersen P.
1997
“Improved modulation techniques for PWMVSI drives,”
IEEE Trans. Ind. Electron.
44
(1)
87 
95
DOI : 10.1109/41.557503
Sakly J.
,
Delarue P.
,
Bausiere R.
“Rejection of undesirable effects of input DCvoltage ripple in singlephase PWM inverters,”
in Proc. 5th Eur. Conf. on Power Electronics and Applications
1993
Vol. 4
65 
70
Pou J.
,
Boroyevich D.
,
Pindado R.
2002
“New feedforward spacevector PWM method to obtain balanced AC output voltages in a threelevel neutralpointclamped converter,”
IEEE Trans. Ind. Electron.
49
(5)
1026 
1034
Leon J. I.
,
Lopez O.
,
Franquelo L. G.
,
DovalGandoy J.
,
Vazquez S.
,
Alvarez J.
,
Freijedo F. D.
2010
“Multilevel multiphase feedforward space vector modulation technique,”
IEEE Trans. Ind. Electron.
57
(6)
2066 
2075
DOI : 10.1109/TIE.2009.2034171