Professional Documents
Culture Documents
1, JANUARY 2014
A Space-Vector Modulation Method
for Common-Mode Voltage Reduction in
Current-Source Converters
Jian Shang, Student Member, IEEE, and Yun Wei Li, Senior Member, IEEE
AbstractThe common-mode voltage (CMV) produced from a
converter system is a source of many problems, e.g., in the mo-
tor drive system, CMV might appear at the neutral point of the
motor stator windings with respect to the ground and induce de-
structive bearing current. Reduced CMVspace-vector modulation
(RCMVSVM) methods have been proposed in both voltage-source
converter (VSC) and current-source converter (CSC) systems. The
available RCMV SVMs reduce the CMV by avoiding using zero-
state vectors. However, this will lead to some negative effects, such
as shrink of modulation index range, increase of switching fre-
quencies, bipolar line-to-line voltage pulse patterns in VSCs, and
power quality performance deterioration. In this paper, a RCMV
SVM method for CSCs is proposed. By allowing the use of zero-
state vectors, the proposed RCMVSVMstill produces much lower
CMV. However, its other performance indices, such as switching
frequency and harmonic performance, are unaffected and com-
parable to the conventional SVMs. The effectiveness of the pro-
posed RCMV SVM for CSCs is veried in the simulations and
experiments.
Index TermsCurrent-source converter (CSC), power quality,
reduced common-mode voltage (CMV) SVM, zero-state vector.
I. INTRODUCTION
A
LTHOUGH voltage-source converter (VSC) has been
the preferred solution in many industrial applications,
pulsewidth modulated (PWM) CSCs could be a good alterna-
tive to VSC in the wind energy power generation, superconduc-
tor magnetic energy storage (SEMS), and HVdc transmission
systems. Currently, in the medium-voltage drive systems, trans-
formerless PWM CSCs have already become a very popular in-
dustrial solution [1]. The conguration of such a systemis shown
in Fig. 1. The transformerless CSC drive system has advantages
over the conventional drive system, such as higher power den-
sity, lower cost, and higher power efciency. However, due to
the absence of the input isolation transformer, common-mode
voltage (CMV) at the neutral point of the motor stator windings
Manuscript received December 4, 2012; revised February 10, 2013; accepted
February 11, 2013. Date of current version July 18, 2013. This work was pre-
sented at the IEEE Applied Power Electronics Conference and Exposition, Long
Beach, CA, USA, March 1721, 2013. Recommended for publication by Asso-
ciate Editor B. Wu.
The authors are with the Department of Electrical and Computer Engi-
neering, University of Alberta, Edmonton, AB T6G 2V4, Canada (e-mail:
jian.shang@ualberta.ca; yunwei.li@ualberta.ca).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TPEL.2013.2248025
with respect to the ground can induce bearing current and lead
to the bearing damage of the motor [2][4].
To mitigate CMV in CSCs, the neutral points of the input and
output capacitors are connected to clamp the CMV to nearly
zero. To limit the large circulating common-mode current caused
by this connection, an integrated dc choke with both differential
and common-mode windings is used [5]. The size of the dc
choke can be reduced if the CMV produced by the rectier and
inverter is minimized. One of the most effective methods for
CMV reduction is to modify the PWM patterns.
Most available RCMV PWM schemes are developed for
VSCs and they reduce CMV by avoiding the use of zero-state
vectors [6]. These RCMV space-vector modulations (SVMs)
can be categorized as active zero-state modulation (AZSM)
[7][9], remote-state modulation (RSM) [10], and near-state
modulation (NSM) [11], [12]. Although those nonzero-state
modulation methods can effectively reduce CMV, they are all
subject to some problems, such as the shrink of modulation in-
dex range, bipolar line-to-line voltage pulse patterns, increased
switching frequencies, higher dc-link ripples, and power qual-
ity performance deterioration. Those drawbacks make RCMV-
PWM for VSCs difcult to be applied in the industry.
As for the PWM of CSCs, selective harmonic elimination
(SHE) PWMis popular in the medium-voltage high-power drive
systems due to SHE PWMs excellent harmonic performance
with low switching frequencies. However, SVM, as a kind of
online modulation method, has been used to damp LCresonance
[13], [14], minimize dc-link current [15], control input power
factor [16], [17], etc. In [18], the authors introduce the nonzero-
state modulation concept for VSCs into CSCs. However, the
problems observed in RCMV SVMs for VSCs still exist in
those nonzero-state RCMV SVMs for CSCs.
Two most important performance indices for PWMpatterns in
the medium-voltage drives and grid-tied converters are switch-
ing frequency [19][21] and harmonic performance [22][24],
which are also two important considerations in the design of
RCMV SVM for CSCs. A novel RCMV SVM method for CSR
and CSI is proposed in this paper to reduce the CMV by se-
lecting proper zero-state vectors instead of completely avoiding
using them. To better illustrate the working principles of the pro-
posed RCMV SVM for CSCs, Section II presents the denition
of CMV in a CSC drive system and the CMV values associated
with all the space vectors. Section III describes the working prin-
ciples of the proposed RCMV SVM in detail, including zero-
state vectors selection, sequences selection for switching fre-
quency minimization, and single-sequence rule for harmonic
0885-8993/$31.00 2013 IEEE
SHANGANDLI: SPACE-VECTORMODULATIONMETHODFORCOMMON-MODEVOLTAGEREDUCTIONINCURRENT-SOURCECONVERTERS 375
Fig. 1. Conguration of transformerless current-source drive system.
performance optimization. To verify the effectiveness of the
proposed RCMV SVM for CSCs, Sections IV and V present
the simulation and experimental results for the RCMV SVM
and the comparison results with conventional three-segment and
ve-segment SVMs. The results show that the CMV can be re-
duced up to 50% compared to the conventional SVM methods.
The switching frequency and harmonic performance of the pro-
posed RCMV SVM are comparable to the conventional SVMs.
II. CMV IN CURRENT-SOURCE CONVERTERS
A. CMV in Current-Source Converters
The CMV produced in the CSC system, as shown in Fig. 1,
consists of CMVin the current-source rectier (CSR) side v
cm r
and CMV in the current-source inverter (CSI) side v
cm i
, which
can be dened, respectively, by
v
cm r
=
v
p1g
+ v
n1g
2
(1)
v
cm i
=
v
p2o
+ v
n2o
2
(2)
where v
p1g
and v
n1g
are the voltages at points p1 and n1,
respectively, with respect to the ground, and v
p2o
and v
n2o
are the voltages at points p2 and n2, respectively, with respect
to the neutral of the induction motor.
If the differential inductance in the positive dc rail is equal to
that in the negative dc rail, the CMV in the whole drive system
v
og
can be given by
v
og
= v
cm r
v
cm i
=
v
p1g
+ v
n1g
2
v
p2o
+ v
n2o
2
.
(3)
When PWM control is applied to the CSC drive systems,
v
cm r
, v
cm i
, and v
og
can be expressed as
v
cm r
= [
S
1
+ S
4
S
3
+ S
6
S
5
+ S
2 ].
0.5v
a
0.5v
b
0.5v
c
(4)
v
cm i
= [
S
1
+ S
4
S
3
+ S
6
S
5
+ S
2 ].
0.5v
u
0.5v
v
0.5v
w
(5)
Fig. 2. Operating principle of SVM for CSCs and space vectors denition.
v
og
= [
S
1
+ S
4
S
3
+ S
6
S
5
+ S
2 ].
0.5v
a
0.5v
b
0.5v
c
[
S
1
+ S
4
S
3
+ S
6
S
5
+ S
2 ].
0.5v
u
0.5v
v
0.5v
w
(6)
where S
1
S
6
are the switching states of CSRs switching de-
vices, S
1
S
6
the switching states of CSIs switching devices,
v
a
, v
b
and v
c
the phase voltages of the grid, and v
u
, v
v
, and v
w
are the phase voltages of the motor stator [25], [26].
B. CMV Associated With Space Vectors
Fig. 2 illustrates the operating principle of SVM for CSCs
and the space vectors denition. According to (4) and (5), the
instantaneous value of CMV associated with the space vectors
can be obtained and summarized in Table I using CSR as an
example. The CMV peak value is determined by many factors,
such as delay angle for CSR (in a current-source drive, the dc-
link current is usually regulated by the delay angle control in the
CSRside [1]), LClter capacitance, dc-link current, modulation
index, and CSI load power factor. In the conventional SVMs,
the maximum CMV peak value produced by zero-state vectors
in the SVM for CSR or CSI can be as high as the peak value of
376 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014
TABLE I
CMV ASSOCIATED WITH SPACE VECTORS IN SVM FOR CSCS
phase voltage of grid side |V
g
| or that of motor stator side |V
m
|,
while the maximum CMV peak value produced by active-state
vectors is equal to |V
g
| /2 or |V
m
| /2. Thus, the maximum CMV
peak value in the drive system |V
og
| is |V
g
| + |V
m
|.
The recently proposed RCMV SVMs for CSCs in [18] re-
duce CMV by avoiding the use of zero-state vectors, resulting
in a maximum CMV peak value as high as |V
g
| /2 or |V
m
| /2.
However, instead of completely avoiding the use of zero-state
vectors, another effective method for CMV reduction is to se-
lect the zero-state vector producing the lowest CMV from the
three ones in each SVM sample. In this way, the maximum
CMV peak value produced by the proposed RCMV SVM can
be lower than |V
g
| /2 or |V
m
| /2 as well, but it will not cause the
aforementioned negative effects in those nonzero-state RCMV
SVMs.
III. PROPOSED RCMV SVM FOR CSCS
The design of the proposed RCMV SVM is discussed in
this section. The zero-state vectors selection rule is proposed to
reduce CMV and a sequence selection rule is then applied to
minimize switching frequencies. The single-sequence rule is
also applied in some cases to realize the harmonic performance
optimization.
A. Zero-State Vectors Selection Rule for CMV Reduction
The zero-state vectors selection rule is that the zero-state vec-
tor producing the lowest CMV among the three zero-state vec-
tors at any sampling period would be selected. For example, in
Fig. 3, the absolute value of phase Bvoltage is the lowest among
the three phase voltages during T
a
, so the CMV produced by
I
0b
(S
3
, S
6
) would be the lowest among the three zero-state vec-
tors. Therefore,
I
0b
(S
3
, S
6
) should be used to minimize CMV
during T
a
. According to this zero-state vectors selection rule,
the selected zero-state vectors and the corresponding phase volt-
ages in one fundamental period are shown in Fig. 3. Using this
zero-state vectors selection method, the maximum CMV peak
value produced by zero-state vectors would be half of the peak
value of ac-side phase voltages, which is same as that produced
by active-state vectors.
Fig. 3. Selected zero-state vectors in RCMV SVM in each instant of one
fundamental period.
Fig. 4. Conventional SVM sequences for CSCs: (a) three-segment sequence
and (b) ve-segment sequence.
B. Sequence Selection Rule for Switching Frequency
Minimization
Six types of SVM sequences for CSCs have been compared
in [27], where it shows the ve-segment SVM sequence has
the best harmonic performance in high modulation index range,
while the three-segment SVM sequence has the best in low
modulation index range. The conventional three-segment and
ve-segment SVM sequences are shown in Fig. 4. According
to the analysis in [27], the sampling frequencies for SVM of
CSCs have to be set as an integral multiple of 6f
1
(where f
1
is
the fundamental frequency) in order to synchronize the PWM
waveform with the fundamental frequency and eliminate even
and triplen harmonics. In order to obtain lowswitching frequen-
cies t for high-power application, the sampling frequencies for
conventional three-segment and ve-segment SVMs are 18f
1
and 24f
1
, respectively.
To avoid the increase of switching frequencies compared to
the conventional SVMs, the sequence for RCMV SVM should
be redesigned since the zero-state vectors selection rule has
changed as compared to the conventional SVMs. For instance,
both
I
0b
and
I
0c
(instead of
I
0a
) could be possibly selected in
sector I according to the zero-state vectors selection rule for
CMV reduction. In this case, there will be extra switching if the
three-segment sequence is applied in RCMV SVM. To avoid
the increase of switching frequency, two types of ve-segment
sequences, as shown in Fig. 5, should be applied according to
the selected zero-state vectors. Specically, if
I
0
has one com-
mon on-state switch with
I
n+1
, sequence (a) should be selected.
Otherwise, sequence (b) should be selected. In some cases,
I
0
SHANGANDLI: SPACE-VECTORMODULATIONMETHODFORCOMMON-MODEVOLTAGEREDUCTIONINCURRENT-SOURCECONVERTERS 377
Fig. 5. Two sequences for RCMV SVM. Sequence (a)applied when
I
0
has
one common on-state switch with
I
n+1
. Sequence (b)applied when
I
0
has
one common on-state switch with
I
n
.
Fig. 6. Selected zero-state vectors and sequences in RCMV SVM for CSR
( = 0
).
may have one common on-state switch with both active-state
vectors in that sector (such as
I
0a
in sector I), so both sequences
(a) and (b) are eligible in terms of switching frequency mini-
mization. To optimize the harmonic performance, the sequence
selection for this type of zero-state vector should abide by the
single-sequence rule (only one type of sequence is applied
during one fundamental period), which would be explained in
detail later. Note that the ve-segment sequence with the active-
state vector positioned in the center, as shown in Fig. 4(b), can
be also applied in RCMV SVM. It has a similar performance
with those sequences in Fig. 5, where the zero-state vector is
positioned in the center.
C. Selected Zero-State Vectors and Sequences for SVM in CSR
Control
According to the aforementioned zero-state vectors and se-
quence selection rules, the selected zero-state vectors and se-
quences in RCMV SVM in CSR control are shown in Figs. 68
when the delay angle is 0
, 30
, and 60
, respectively.
Fig. 6 shows the selected zero-state vectors and sequences
in all the sectors when the delay angle is 0
. The phase A
voltage is around the peak value in sector I when the delay
angle is 0
).
Fig. 8. Selected zero-state vectors and sequences in RCMV SVM for CSR
( = 60
).
conventional SVMs would be as high as the peak value of phase
voltages. According to the proposed zero-state vectors selection
method for CMV reduction,
I
0b
and
I
0c
(instead of
I
0a
in the
conventional SVMs) should be applied successively to reduce
the CMV in sector I. Both types of ve-segment sequence in
Fig. 5 are used according to the sequence selection rule for
switching frequency minimization.
Fig. 7 shows the selected zero-state vectors and sequences in
all the sectors when the delay angle is 30
. According to the
zero-state vector and sequence selection rules discussed previ-
ously, the sequence (b) of RCMV SVM is always used in one
fundamental period.
Fig. 8 shows the selected zero-state vectors and sequences in
all the sectors when the delay angle is 60
I
oa
(S
1
, S
4
)
378 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014
Fig. 9. 5th and 7th harmonics comparison of the switching current I
w
and THD comparison of line current I
s
when Sequence (a) and (b) and Single-sequence
(b) is applied, respectively ( = 60
I
1
(S
1
, S
6
) and
I
2
(S
1
, S
2
). As a
result, this zero-state vector can be used to build either sequence
(a) or sequence (b). On the other hand, the other selected zero-
state vector
I
ob
(S
2
, S
5
) in sector I will require sequence (b)
to avoid additional switching. Therefore, there are two possible
sequence combinations in this case: 1) only sequence (b) is used
and 2) both sequences (a) and (b) are used. Here, we name these
two possibilities as Single-sequence (b) and Sequence (a)
and (b), respectively.
Fig. 9(a) and (b) shows the 5th and 7th harmonics of the
switching current I
w
, when Sequence (a) and (b) and Single-
sequence (b) are used, respectively. Fig. 9(c) shows the THD
of line currents I
s
with an LC cutoff frequency of 3.33 p.u.
(This cutoff frequency of CSRs LC lter is selected to avoid
amplication of the 5th harmonic current [27].) As shown, 5th
and 7th harmonics of I
w
are lower and THD of I
s
is better if
Single-sequence (b) is applied. It means that rotational use of
sequences (a) and (b) in one fundamental period will deteriorate
the harmonic performance. Moreover, the switching frequency
is lower if Single-sequence (b)is applied, because switching
between sequences (a) and (b) involves more devices. As a
consequence, Single-sequence (b) will be a better option in
this case. Here, the rule that only one type of sequence in one
fundamental period is selected to maintain a better harmonic
performance is named as single-sequence rule.
Table II shows the selected sequences for all the possible
delay angles under this sequence selection rule.
D. Single-Sequence Rule for Harmonic Performance
Optimization
As discussed in the previous section, the use of sequences
(a) and (b) in combination would deteriorate the harmonic per-
formance. From Table II, we can nd that the single-sequence
rule is violated in the delay angle ranges of 30
< 30
and 150
< 210
. Fig. 10
shows the selected zero-state vectors and sequences of RCMV
SVM using single-sequence rule, where single sequence (b)
is applied during one fundamental period. From Fig. 10, we
can see that the peak value of CMV produced by RCMV SVM
using Single-sequence (b) is slightly higher than half of the
peak value of phase voltages, but still much lower than the peak
value of phase voltages (the maximum peak value of CMV pro-
duced by the conventional SVMs). The darker shadow areas in
Fig. 10 show the difference of CMV produced by RCMV SVM
with and without using the single-sequence rule. If single-
sequence rule is applied, the worst case of CMV increasing is
when the delay angle is 0
).
TABLE III
SELECTED SEQUENCES FOR ALL THE POSSIBLE DELAY ANGLES IN RCMV
SVM USING SINGLE-SEQUENCE RULE
sequence is applied in one fundamental period, when the delay
angle is in those two ranges. With this method, the switching fre-
quency would increase, but CMV would be kept lower than half
the peak value of ac-side phase voltage and harmonic perfor-
mance would not be deteriorated. Since the increased switching
frequency is not favorable in high power CSC drives, this alter-
native method will not be discussed in detail.
Table III shows the selected sequences for all possible delay
angles in RCMV SVM using single-sequence rule.
E. Switching Frequency Analysis of the Proposed RCMV SVM
for CSCs
According to the previous analysis, the proposed RCMV
SVM, compared with the conventional ve-segment SVM, has
no extra switching during switching states transitions in one
sector if their sampling frequencies are the same. The extra
switching can only happen during the sector crossing. Fig. 11
illustrates the possible device switching of the proposed RCMV
SVM in the sector crossing from sector I to sector II. If the
sampling frequency of RCMV SVM is an even multiple of 6f
1
(just like conventional ve-segment SVM), there are three pos-
sible cases during the sector crossing as shown in Fig. 11(a)(c),
respectively, depending on the selected zero-state vectors in the
Fig. 11. Possible device switching of the proposed RCMV SVM in the sector
crossing from sector I to sector II.
sector crossing and the moment of the sector crossing. The sec-
tor crossing in Fig. 11(a) and (b) happens in the center of the
sequence, while that in Fig. 11(c) happens in the end of the
sequence.
Fig. 11(a) shows that the selected zero-state vectors for the
last sample in sector I and the rst sample in sector II are both
I
0c
. This zero-state vector selection method, which has been
applied in the conventional ve-segment SVMs in [27], leads
to minimum device switching. This scenario in Fig. 11(a) could
possibly happen in the proposed RCMV SVM if
I
0c
is the
zero-state vector needed for CMV reduction during the sector
crossing. The switching frequency in this case is 8f
1
if the
sampling frequency is 24f
1
.
Since the freedom of the zero-state vectors selection has been
used for CMVminimization in RCMVSVM, the device switch-
ing minimization, like in the case shown in Fig 11(a), cannot
be guaranteed all the time. However, the increase of switching
frequency, which is one fundamental frequency, is not signi-
cant. Other zero-state vectors, like
I
0a
in Fig. 11(b), could be
selected to reduce CMV during the sector crossing. Compared
to the case in Fig. 11(a), one extra device switching could hap-
pen in Fig. 11(b) between
I
0
and
I
n+1
in the rst sample of
sector II. As a result, the switching frequency in this case is 9f
1
if the sampling frequency is 24f
1
.
As shown in Fig. 11(c), the sector crossing can also happen in
the end of the sequence. There is one switching during the sector
crossing under this scenario (which is similar to the situation that
conventional ve-segment SVMin Fig. 4(b) with sector crossing
at the center of the sequence). The switching frequency in this
case is also 9f
1
if the sampling frequency is 24f
1
. Furthermore,
if two types of sequences in RCMV SVM are used in one
fundamental period, this extra switching would happen in the
moment of sequences (a) and (b) transition instead of sector
crossings.
380 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 29, NO. 1, JANUARY 2014
TABLE IV
MAIN CIRCUIT PARAMETERS OF CSR SYSTEM IN THE
SIMULATION AND EXPERIMENT
Therefore, the switching frequency of RCMV SVM is 8f
1
or 9f
1
depending on the selected zero-state vectors in the sector
crossing and the moment of sector crossing if the sampling fre-
quency is 24f
1
. Thus, under the same sampling frequencies, the
switching frequency of RCMV SVM is equal to or f
1
different
from conventional ve-segment SVM.
F. Implementation of RCMC SVM in the CSR and CSI control
Similar to the conventional ve-segment SVM, each ve-
segment sequence of the proposed RCMV SVM needs two
samples to complete. The sectors and active-state vectors are
updated every sample, whereas the zero-state vectors are up-
dated before the beginning of the sequence (two samples) and
kept unchanged during the two samples. In this way, good con-
trol accuracy could be obtained and the switching frequency
would not increase.
In the CSI application, the number of samples in one cycle of
SVM (or the sampling frequency) is changing with the funda-
mental frequency, which is related to the motor speed. Moreover,
there is no delay angle control in the CSI control. However, these
two differences from CSR control would not affect the RCMV
SVMs application in the CSI side. In the application in the CSI
side, phase voltages of the motor stator or CSI output capacitor
V
m
can be detected to determine which zero-state vector and
sequence should be applied.
To determine whether the single-sequence rule for har-
monic performance optimization should be applied, the equiv-
alent delay angle (or voltage current displacement angle) in the
CSI control should be known. It depends on the CSI output ca-
pacitance, motor parameters, and motor operating conditions.
This equivalent delay angle in CSI control should be dened as
the phase displacement angle of V
m
(jw)/I
mw
(jw), where V
m
is the phase voltage of motor stator and I
mw
is the CSI output
switching current.
IV. SIMULATION VERIFICATIONS
The proposed RCM SVM method is rst tested in MATLAB/
Simulink simulations. The CSR testing system parameters in
the simulation are shown in Table IV.
Figs. 12 and 13 show the comparison of the CMV, switch-
ing current, and line current under the operating conditions
of m
a
= 0.7 and = 0
and m
a
= 0.3 and = 0
, re-
spectively. According to previous analysis, the sampling fre-
quency for the conventional three-segment SVM is 1080 Hz,
while that for the conventional ve-segment and RCMV SVM
is 1440 Hz. In the gures, 3Seg (such as CMV_3Seg)
and 5Seg (such as CMV_5Seg) represent conven-
tional three-segment and ve-segment SVMs, respectively,
while RCMV (such as CMV_RCMV) and RCMV_TF
(such as CMV_RCMV_TF) represent the proposed RCMV
SVM without and with the single-sequence rule applied,
respectively.
Table V summarizes the simulation results. In Table V,
I
w5
/I
w1
and I
w7
/I
w1
represent the 5th and 7th harmon-
ics over the fundamental current in the switching current. f
s
, 30
, and 60
,
m
a
= 0.7 and = 30
, and m
a
= 0.7 and = 60
, re-
spectively. Its performance is compared with the conventional
three-segment and ve-segment SVM. The sampling frequen-
cies of the conventional three-segment and ve-segment SVMs
and RCMV SVM are same as those in the simulation. When
the delay angle is 0
and 60
AND m
a
= 0.3 AND = 0
, (b) = 30
, and (c) = 60
.
Fig. 15. CMV waveforms under the operating condition of m
a
= 0.7 and
= 0
, m
a
= 0.7 AND = 30
, AND
m
a
= 0.7 AND = 60