Professional Documents
Culture Documents
SVPWM Obrad PDF
SVPWM Obrad PDF
Abstract—This paper presents a novel modulation strategy windings. A modulation strategy, applicable in conjunction with
for a seven-phase three-level inverter, aimed at variable-speed this machine type, is required to produce only the fundamental
drive applications. Developed space vector pulsewidth modulation (sinusoidal) component in the output phase voltage, without any
(SVPWM) strategy is based on vector space decomposition (VSD)
approach and an analysis of multidimensional space vector projec- low-order harmonics. Two main modulation concepts are in gen-
tions in mutually orthogonal planes. The algorithm represents an eral carrier-based and space vector techniques. The extension
extension of a recently developed five-phase three-level SVPWM of carrier-based strategies from two-level to higher number of
algorithm and is considerably more complex, due to the existence levels is well established for three-phase case [2]. Since carrier-
of three mutually orthogonal planes in a seven-phase system. The based techniques are independent of the number of phases, the
difficulties of generalization of space vector strategies, which stem
from the nature of the multiphase systems, are highlighted. In- same carrier-based techniques as for three-phase case can be
creasing the number of phases from five to seven increases the directly applied to multilevel multiphase systems. Due to the
number of switching states from 35 = 243 to 37 = 2187 and the simplicity, they are usually the preferred choice. The existing
number of orthogonal planes from two to three. This considerably discussions of carrier-based strategies for control of multilevel
affects the complexity of the offline calculations. However, the final multiphase systems are predominantly related to the capacitor
implementation of the algorithm is simple. Developed SVPWM al-
gorithm is compared to an equivalent carrier-based strategy and voltage balancing of the neutral-point clamped (NPC) voltage
it is shown that they yield identical performance. The complex- source inverter (VSI) for a five-phase passive load [3], [4].
ity of the algorithms for real-time implementation is compared. As far as the space vector approach is concerned, the algo-
Simulation and experimental results, obtained using neutral-point rithm of [5] is widely accepted as general for the three-phase
clamped inverter, verify the theoretical developments. multilevel inverter. However, it leaves 1 degree of freedom un-
Index Terms—Carrier-based pulsewidth modulation (PWM), used, which enables a further optimization of the algorithm, as
seven-phase, space vector pulsewidth modulation (SVPWM), demonstrated in [6]. A simple extension from three-phase to
three-level, vector space decomposition (VSD). multiphase space vector pulsewidth modulation (SVPWM) is
not possible, in contrast to the carrier-based approach. With an
increase in the number of phases, the number of space vectors
I. INTRODUCTION
increases and, even more importantly, the analysis cannot be
NVESTIGATION of multiphase machines (with more than
I three phases), supplied from multilevel inverters, has re-
cently started attracting more attention, due to numerous good
conducted in a single (α–β) plane, as for a three-phase sys-
tem. Assuming that the phase number n is odd, the number of
planes that has to be considered simultaneously is (n − 1)/2
features offered by such topologies [1]. Higher number of phases (star-connected load with isolated neutral point is assumed).
and more levels in the inverter output leg voltage overcome The existing SVPWM strategies for multilevel multiphase VSIs
component voltage and current limits. Hence, multilevel multi- apply to particular cases. The most frequently analyzed topol-
phase structures represent an attractive solution for high power ogy is the five-phase three-level system. In the initial attempts of
applications. using space vector approach to control this topology, the prin-
The assumed structure of an ac machine in this paper is a ciple of three-phase multilevel SVPWM was followed. Thus,
symmetrical seven-phase machine with sinusoidally distributed in [7], the first sector in the α–β plane was divided into three
triangular subsectors and only three space vectors per switching
period were used. This concept, however, does not consider the
Manuscript received December 22, 2011; revised April 11, 2012; accepted second plane (x–y) and leads to the large amount of low-order
May 24, 2012. Date of current version September 27, 2012. This work was harmonics in this plane. This is so since the number of applied
supported by the National Priorities Research Program (NPRP) under Grant
4-152-02-053 from the Qatar National Research Fund (a member of Qatar
space vectors in a switching period must equal the number of
Foundation). The statements made herein are solely the responsibilities of the phases. The same algorithm was applied for direct torque control
authors. Recommended for publication by Associate Editor J. R. Rodriguez. of two series-connected five-phase induction machines in [8].
The authors are with the School of Engineering, Technology and Mar-
itime Operations, Liverpool John Moores University, Liverpool, L3 3AF, U.K.
Similarly, only the first plane was also considered in [9] and [10]
(e-mail: o.dordevic@2009.ljmu.ac.uk; e.levi@ljmu.ac.uk; m.jones2@ljmu. and less than five space vectors were used per switching period,
ac.uk). with the same consequences.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
The requirement that the number of space vectors per switch-
Digital Object Identifier 10.1109/TPEL.2012.2203148 ing period must equal the number of phases was respected
in [11]–[14]. However, although the algorithm was presented as and prove identical operation of the presented space vector mod-
the space vector technique, it does not use vector space decom- ulation and carrier-based modulation with “double injection,”
position (VSD) that is a foundation for this approach. Instead, are presented in Section III. The particular realization that re-
the approach is based on the consideration of the multidimen- duces the execution time of the SVPWM algorithm is elaborated
sional space. The algorithm of [11] is in essence equivalent to and the computation time of the algorithms is compared. Con-
the carrier-based approach with level-shifted carriers, written in clusions are given in Section IV.
the matrix form. The application of a single-phase modulator to
five-phase three-level system was discussed in [15]. However, II. SPACE VECTOR ALGORITHM
the presented concept of single-phase modulator is also equiva-
The analyzed topology is the seven-phase three-level NPC
lent to the carrier-based approach with level-shifted carriers.
VSI, shown in Fig. 1. The discrete output levels of the inverter
The algorithm in [16] for five-phase three-level VSI is the first
legs can be 0, Vdc /2 and Vdc , when referenced to the dc-bus
one that develops SVPWM for multilevel multiphase VSIs using
negative rail. Using normalization, these levels are further de-
VSD approach and, therefore, considers space vector projections
noted as 0, 1, and 2, respectively. As already noted, the SVPWM
in all planes (two in five-phase case). A suitable modification of
algorithm is a modified version of the algorithm of [18]. The
the algorithm of [16] was further suggested in [17], where the
algorithm is divided into steps and each of the steps is described
comparison of the SVPWM algorithms with some carrier-based
in what follows.
techniques was presented for a five-phase three-level VSI. An
SVPWM algorithm for a three-level seven-phase VSI, based on
A. Three-Level Seven-Phase VSD
the VSD approach, was for the first time introduced in [18], by
extending the algorithm of [16]. For the space vector analysis, the space in which variables
This paper builds on the previous work reported in [17] map has to be defined in the first place. Commonly used trans-
and [18]. To start with, the algorithm of [18] is revisited and formation in this field [19] that defines space vectors in complex
it is then modified in the same manner as the three-level five- form for a seven-phase system is
phase SVPWM in [17]. Since the results in [17] show the ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
X α −β 1 a a2 · · · a6 xa
equivalence between the modified SVPWM algorithm and a
⎢X ⎥ 2 ⎢ ⎢ 1 a a · · · a ⎥ ⎢ xb ⎥
2 4 12 ⎥ ⎢ ⎥
specific carrier-based algorithm with “double injection” for the ⎢ x 1 −y 1 ⎥
⎢ ⎥= ·⎢ ⎥ · ⎢ . ⎥ (1)
⎣ X x 2 −y 2 ⎦ 7 ⎢ 1 a a ··· a ⎥ ⎢ . ⎥
3 6 18
five-phase VSI, the same principle is applied to the carrier- ⎣ ⎦ ⎣ . ⎦
based three-level seven-phase pulsewidth modulation (PWM). 1 1 1 1
X0 ··· xg
Detailed simulation and experimental works using NPC inverter 2 2 2 2
and star-connected R–L load confirm the validity of the theo- where a = ej ·2π / 7 , [xa xb . . . xg ]T represents vector of cer-
retical considerations and prove that the two considered PWM tain variables of the system in terms of phase values, while
techniques yield the same performance. Finally, the complexity [X α −β X x1−y 1 X x2−y 2 X0 ]T represents vector of the
of the algorithms for online implementation is addressed. It is variables in the transformed space (i.e., multidimensional space
important to note that the algorithms are applied here without vector components in the 2-D planes). This transformation can
any additional means for capacitor voltage balancing. be applied to any set of symmetrical phase variables, such as
This paper is organized as follows. In Section II, the SVPWM reference phase voltages vk∗ , reference leg voltages vK ∗
, output
algorithm of [18] is summarized and the modification, based phase voltages vk , output leg voltages vK , currents, etc. The
on [17], is introduced. The difficulties encountered in the pre- axes of the new space (α, β, x1 , y1 , x2 , y2 and 0) are defined
liminary offline calculations are highlighted. Simulation and ex- with (1), and space vectors are determined with the projections
perimental results, which validate the theoretical considerations, onto each of these axes. Nature of the transformation is such
DORDEVIC et al.: VECTOR SPACE DECOMPOSITION BASED SPACE VECTOR PWM ALGORITHM 639
Fig. 4. (a) Graphical representation of single-level increasing transition in the α–β plane, for each leg. (b) The first sector in α–β plane with reduced number of
switching states and with all possible single-level increasing transitions.
D. Dwell Time Calculation and Sector Partitioning in the x–y plane where the reference voltage is zero. If the x–y
pattern is placed at the same side of the line that passes through
Dwell times are determined by space vector projections rather
than by particular switching state values. So, all redundant the origin, then this pattern cannot lead to the desired zero av-
switching sequences that correspond to the same pattern have erage value in this plane, and can be omitted from the further
analysis. In seven-phase case this method is not sufficient for the
the same dwell times for the space vectors at appropriate posi-
tions. The dwell times can be calculated in the standard way, reduction of the possible switching pattern number. To be more
from the volt-second balance equations in space vector domain. precise, only 18 out of 64 patterns can be eliminated using this
graphical method. The problem is that there are patterns with
For the seven-phase case, this equation is given as follows:
which cancellation in both of x–y planes can be achieved inde-
⎡ ⎤ ⎡ ⎤ ⎡⎤ pendently, but the solutions do no not overlap, meaning that x–y
T1 Vα 1 Vα 2 Vα 3 ··· Vα 7 −1 Vα∗ components cannot be cancelled in both planes at the same time.
⎢ T2 ⎥ ⎢ Vβ 1 Vβ 2 Vβ 3 ··· Vβ 7 ⎥ ⎢ V∗ ⎥ This is indeed one of the major additional difficulties, brought
⎢ ⎥ ⎢ ⎢ β ⎥
⎥
⎢ ⎥ ⎢ ⎢ ∗ ⎥
⎥
⎢ T3 ⎥ ⎢ Vx 1 1 Vx 1 2 Vx 1 3 · · · ⎢ Vx 1 ⎥
Vx 1 7 ⎥ in by the higher number of phases, compared to [16]. The com-
⎢ ⎥ ⎢ ⎢ ∗ ⎥
⎥ plexity of the algorithms for five and seven phase three-level
⎢ T4 ⎥ = ⎢ Vy 1 Vy 1 2 Vy 1 3 · · · ·⎢
Vy 1 7 ⎥ ⎥
⎢ ⎥ ⎢ 1 ⎢ Vy 1 ⎥ · Ts
⎥ case is highlighted in Table I (the table also includes informa-
⎢ ⎥ ⎢ ⎢ ∗ ⎥
⎥
⎢ T5 ⎥ ⎢ Vx 2 1 Vx 2 2 Vx 2 3 · · · ⎢ Vx 2 ⎥
Vx 2 7 ⎥ tion on relevant algorithm aspects that are yet to be addressed).
⎢ ⎥ ⎢ ⎢ ∗ ⎥
⎥
⎣ T6 ⎦ ⎣ Vy 2 1 Vy 2 2 Vy 2 3 · · · ⎣ Vy ⎦
Vy 2 7 ⎦ From Table I and from the explained algorithm steps, one can
2
T7 1 1 1 ··· 1 1 conclude that the algorithm complexity will further increase for
(5) higher number of phases and higher number of levels.
where Vγ i is the projection of the ith leg voltage space vector The problem of sector partitioning is solved using MATLAB.
of the selected pattern onto the axis γ, where γ = α, β, x1 , The program solves (5) for each of 64 (i.e., 64 −18 = 46)
y1 , x2 , y2 , and i = 1 to 7, while Vγ∗ is the projection of the switching patterns. The column matrix with reference values
reference voltage space vector onto γ-axis. Time intervals Ti , i in (5) is [V·cos(ϕ) V·sin(ϕ) 0 0 0 0 1]T , as already stated. The
= 1 to 7, are times of application of the selected space vectors in square matrix in (5) contains projections onto α, β, x1 , y1 , x2 , and
the pattern. Since generation of the sinusoidal reference voltage y2 axes of all the seven space vectors that form a selected pattern.
is desired, one has in (5) Vα∗ = V cos(ωt), Vβ∗ = V sin(ωt), For the selected pattern, program increases in incremental steps
∗
Vx1 = 0, Vy∗1 = 0, Vx2∗
= 0, Vy∗2 = 0. values of V and ϕ gradually covering full region of the α–β
An idea, related to the sector partitioning, has been introduced plane. If the solution of (5) exists for temporary pair of V and
in [16]. If the set of vectors Vγ i in (5) is already chosen, i.e., ϕ, and if all the dwell times T1 to T7 are in the range from 0
if the pattern is selected, the equation will not have physically to Ts , then the program plots a dot at V·ej ϕ position in the α–β
acceptable solution for arbitrary values of the reference, since plane. This means that the current pair of V and ϕ (plotted dot)
the dwell times of each switching state Ti must be in the range belongs to the region of applicability for the selected pattern.
from 0 to Ts . This means that each pattern has its region of All pairs of V and ϕ (all plotted dots) represent the region
application; division of the sectors into subsectors can be done of application of the selected pattern. As an example of the
in this manner. All 64 available patterns in the first sector must output of this code, Fig. 5 shows the area that is covered by
be analyzed in this way. the pattern shown with oversized arrows in Fig. 4. Note that
In the five-phase case one intermediate step is sufficient to the graphical method for the elimination of those patterns that
omit some patterns from the analysis in advance [16]. This pro- cannot provide cancellation in the second or in the third plane
cess includes graphical analysis of the projections of the patterns is not necessary because these patterns will be recognized and
642 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 2, FEBRUARY 2013
TABLE I
COMPARISON OF THE OFFLINE COMPLEXITY FOR THE FIVE-PHASE AND SEVEN-PHASE THREE-LEVEL SVPWM ALGORITHMS
TABLE II
SELECTED SWITCHING SEQUENCES FOR SUBSECTORS OF FIG. 6
when the reference leg voltage transits from the higher to the
lower carrier zone, the leg voltage that was switching between
levels 1 and 2 will start switching between levels 0 and 1, and
thus, the starting switching state will be changed. In seven-
phase case with pure sinusoidal references as in (2), change of
the carrier zone occurs at ωt = 1π/14, 3π/14, 5π/14, etc. Also,
certain injections (e.g., min–max injection, or those that contain
only integer odd multiples of the seventh harmonic that are in
phase with the reference signals) will not change the positions
of these crossing points. Thus, points where the reference leg
voltages change carrier zone determine additional 14 switchings
per fundamental period.
For the easier comparison with carrier-based approach, and
Fig. 8. Division of the first sector into subsectors after introduced
modification. since switching sequence redundancy per pattern, i.e., per sub-
sector, exists, one additional subdivision per sector is intro-
duced. The additional subdivisions halve the main sectors, so
further, based on the corresponding modification in [17] for the they are located at ωt = 1π/14, 3π/14, 5π/14 etc. Subsec-
five-phase three-level VSI. At the moment, all subsectors from tors with this new subdivision for the first sector in the α–β
1 to 16 in each sector have the same starting switching state. The plane are shown in Fig. 8. Switching sequences that correspond
same starting switching states have subsectors 17 and 18 from to the new subdivision are chosen from redundant sequences
the next sector as well. This means that 14 additional switchings for subsectors in such a way to minimize losses when refer-
appear during the fundamental period when the starting switch- ence changes subsector. New switching sequences are given in
ing sequence changes. These additional switching transitions, Table IV. Switching sequences for subsectors denoted without
one increasing and one decreasing transition per fundamental additional index “dn” (i.e., without index, or with index “up”)
period, appear in each of seven legs. Two borders between sub- are the same as corresponding sequences in the original algo-
sectors when these additional switching transitions occur in the rithm (see Table II). Now, from Table IV, one can see that the
first and the second sectors are shown in Fig. 8. starting switching state is the same for all switching sequences
Let us analyze the appearance of additional switching transi- in the lower sector half 0 ≤ ωt < π/14 (they start with 1100001)
tions in time domain. If level-shifted carriers are assumed, they and for all in the upper sector half π/14 ≤ ωt < 2π/14 (starting
are shifted vertically, and each occupies certain vertical zone. switching state is 1110001). The position of additional switch-
The starting switching state changes every time when any of ing transitions (at new subdivisions) coincides now with those
the reference leg voltages change the carrier zone. For example, in time domain.
DORDEVIC et al.: VECTOR SPACE DECOMPOSITION BASED SPACE VECTOR PWM ALGORITHM 645
Fig. 11. Current THD comparison for presented carrier-based and space vector
modulation strategies (experiments) for the linear modulation index range of
m = 0.15 to 1.025.
Fig. 12. (a) Oscilloscope recording of the leg and phase voltage, and load current at m = 0.6, for considered carrier-based (left) and space vector (right) PWM
strategies (M-leg voltage, 260 V/div; Ch3-phase voltage, 250 V/div; Ch4-phase current, 500 mA/div; time = 10 ms/div), with associated (b) phase voltage and
(c) current spectra.
Fig. 13. (a) Oscilloscope recording of the leg and phase voltage, and load current at m = 1, for considered carrier-based (left) and space vector (right) PWM
strategies (M-leg voltage, 260 V/div; Ch3-phase voltage, 250 V/div; Ch4-phase current, 500 mA/div; time = 10 ms/div), with associated (b) phase voltage and
(c) current spectra.
DORDEVIC et al.: VECTOR SPACE DECOMPOSITION BASED SPACE VECTOR PWM ALGORITHM 647
Fig. 14. Simulation results illustrating (a) leg and (b) phase voltage at m = 1, for considered carrier-based (left) and space vector (right) PWM strategies with
associated (c) phase voltage spectrum.
any voltage-related considerations and affects only the current domain using Parseval’s theorem, as addressed in [24] and [25].
ripple, analysis of which is beyond the scope here). Two custom- Simulation and experimental results for obtained phase voltage
made three-level six-phase NPC inverters with common dc link THD are shown in Fig. 10. One can see that phase voltage THD
have been used to form a seven-phase NPC inverter. For dc-bus values for both strategies are identical, and the same values are
supply, an external dc source Sorensen SGI 600/25 has been obtained by simulations and by experiments. Experimentally
employed, and the dc-bus voltage value has been set to Vdc = obtained current THD is shown in Fig. 11. It can be seen that
600 V. Inverter dead time is 6 μs, and switching frequency has the two PWM methods yield in essence identical current rip-
been set to 2 kHz. The photograph of the experimental setup ple (i.e., THD) behavior, which means that the same switching
is given in Fig. 9. The reference value has been changed using harmonic-related side-bands are produced and are mapped in
V/f = const. law, where the 300 V (peak value) and 50 Hz have the three planes in the same manner. Due to hugely variable
been chosen as nominal values. Parameters and various settings impedance of the load, no attempt has been made to evaluate
are the same in simulations and in experiments. the current THD by simulation. It should be emphasised that the
To validate the algorithms and prove the equivalence of the squared value of the nominator of the voltage THD is directly
presented space vector and carrier-based algorithms, total har- related to the additional eddy current iron losses, caused by the
monic distortion THD has been calculated for the full linear PWM induced voltage harmonics, when the inverter supplies
modulation index range. THD is computed, on the basis of the an electrical machine [24], [25]. On the other hand, additional
FFT results, according to copper losses, caused by the PWM operation, are directly pro-
portional to the square of the nominator in (8) when THD applies
h 2 to the current, since this is the squared value of the current ripple
i=2 Zi
THDz = (8) rms [26].
Z1
Since THD is a global figure of merit, the equivalence of
where z stands for voltage or current, and Zi represents the ith the presented algorithms is further confirmed by experimentally
component in the spectrum. Value of h determines the last har- recorded time-domain waveforms and spectra of the phase volt-
monic in the spectrum whose frequency is lower than 21 kHz. age and current for modulation indices m equal to 0.6 and 1.
Since the switching frequency is 2 kHz, this means that harmon- The screenshots from the oscilloscope and associated spectra
ics from the first ten side bands are included in the THD calcula- are shown in Figs. 12 and 13. In essence, the waveforms and the
tion. The voltage THD can be calculated also directly in the time spectra are identical for all practical purposes.
648 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 2, FEBRUARY 2013
To further confirm the correspondence between simulation [4] S. Karugaba, O. Ojo, and M. Abreham, “Carrier based PWM scheme for
and experimental results, Fig. 14 shows simulation results for m a three-level diode-clamped five-phase voltage source inverter ensuring
capacitor voltage balancing,” in Proc. IEEE Appl. Power Electron. Conf.
= 1 for the two modulation techniques. Comparison of the leg Exhib., Orlando, FL, Mar. 2011, pp. 1194–1201.
and phase voltage time-domain traces and phase voltage spectra [5] N. Celanovic and D. Boroyevich, “A fast space-vector modulation al-
with the experimental results in Fig. 13 confirms a very good gorithm for multilevel three-phase converters,” IEEE Trans. Ind. Appl.,
vol. 37, no. 2, pp. 637–641, Mar./Apr. 2001.
agreement indeed. [6] B. P. McGrath, D. G. Holmes, and T. Lipo, “Optimized space vector
Finally, the complexity of the algorithms has been compared switching sequences for multilevel inverters,” IEEE Trans. Power Elec-
by measuring the algorithm execution time on dSpace using the tron., vol. 18, no. 6, pp. 1293–1301, Nov. 2003.
[7] Q. Song, X. Zhang, F. Yu, and C. Zhang, “Research on PWM techniques
dSpace Profiler tool. The execution time of the applied carrier- of five-phase three-level inverter,” in Proc. Int. Symp. Power Electron.,
based method is around 1.06 μs, while the execution time of Electr. Drives, Autom. Motion, Taormina, Italy, May 2006, pp. 561–565.
the space vector algorithm is around 2.25 μs. A relatively small [8] J. Soltani, N. R. Abjadi, J. Askari, and G. R. A. Markadeh, “Direct torque
control of a two five-phase series connected induction machine drive using
difference in the processor time consumption of the compared a three-level five-phase space vector PWM inverter,” in Proc. IEEE Int.
algorithms is due to the particular realization of the space vec- Conf. Ind. Technol., Chengdu, China [CD-ROM], 2008, pp. 1–6.
tor algorithm, which requires a lot of memory. The consumed [9] J. Huang and K. A. Corzine, “A new walking pattern SVM technique
for five-phase motor drives,” in Proc. IEEE Ind. Appl. Soc. Annu. Meet.,
memory by the space vector algorithm is 2688 integer and 2352 Edmonton, Alberta, Canada [CD-ROM], 2008, pp. 1–8.
real variables, while carrier-based method does not require any [10] C. M. Hutson, G. K. Venayagamoorthy, and K. A. Corzine, “Optimal SVM
memory. Also, carrier-based strategy does not need any offline switching for a multilevel multi-phase machine using modified discrete
PSO,” in Proc. IEEE Swarm Intell. Symp., St. Louis, MO [CD-ROM],
calculations, while the space vector algorithm, as demonstrated, 2008, pp. 1–6.
requires a very involved offline procedure. Thus, for practical [11] O. López, J. Alvarez, J. Doval-Gandoy, and F. D. Freijedo, “Multilevel
implementation the carrier-based realization will always be the multiphase space vector PWM algorithm,” IEEE Trans. Ind. Electron.,
vol. 55, no. 5, pp. 1933–1942, May 2008.
first choice. However, the importance of the SVPWM should [12] O. López, J. Alvarez, J. Doval-Gandoy, and F. D. Freijedo, “Multilevel
not be underestimated, since it gives a much better insight into multiphase space vector PWM algorithm with switching state redun-
the switching behavior of the inverter and the mapping of the dancy,” IEEE Trans. Ind. Electron., vol. 56, no. 3, pp. 792–804, Mar.
2009.
harmonics into different planes. It should be noted that there [13] J. I. Leon, O. Lopez, L. G. Franquelo, J. Doval-Gandoy, S. Vazquez,
are situations where only space vector approach can be used to J. Alvarez, and F. D. Freijedo, “Multilevel multiphase feedforward space-
explain physically the working of the system: this applies to, for vector modulation technique,” IEEE Trans. Ind. Electron., vol. 57, no. 6,
pp. 2066–2075, Jun. 2010.
example, independent control of series-connected multiphase [14] J.-S. Hu, K.-Y. Chen, T.-Y. Shen, and C.-H. Tang, “Analytical solutions
multimotor drive systems [27], where no insight can be gained of multilevel space-vector PWM for multiphase voltage source inverters,”
from time (or, equivalently, carrier-based) domain, while com- IEEE Trans. Power Electron., vol. 26, no. 5, pp. 1489–1502, May 2011.
[15] J. I. Leon, S. Vazquez, J. A. Sanchez, R. Portillo, L. G. Franquelo, J.
plex space vector domain makes analysis not only clear but also M. Carrasco, and E. Dominguez, “Conventional space-vector modulation
easy to understand. techniques versus the single-phase modulator for multilevel converters,”
IEEE Trans. Ind. Electron., vol. 57, no. 7, pp. 2473–2482, Jul. 2010.
[16] L. Gao and J. E. Fletcher, “A space vector switching strategy for three-
IV. CONCLUSION level five-phase inverter drives,” IEEE Trans. Ind. Electron., vol. 57, no. 7,
pp. 2332–2343, Jul. 2010.
A space vector algorithm, based on VSD and aimed at seven- [17] O. Dordevic, M. Jones, and E. Levi, “A comparison of PWM techniques
phase motor drives supplied from the three-level NPC VSI is for three-level five-phase voltage source inverters,” in Proc. Eur. Power
Electron. Appl. Conf., Birmingham, U.K. [CD-ROM] paper 0620, 2011,
presented in this paper. The algorithm is introduced on the step- pp. 1–10.
by-step basis and the difficulties in solving some of the steps are [18] O. Dordevic, M. Jones, and E. Levi, “A space vector PWM algorithm for
emphasized. Despite the complicated offline process, once when a three-level seven-phase voltage source inverter,” in Proc. Eur. Power
Electron. Appl. Conf., Birmingham, U.K. [CD-ROM] paper 0123, 2011,
the switching sequences and matrices for dwell times are deter- pp. 1–11.
mined, the final realization of the algorithm is rather simple. The [19] E. Levi, R. Bojoi, F. Profumo, H. A. Toliyat, and S. Williamson, “Multi-
algorithm is compared by simulation and experimentally with phase induction motor drives—A technology status review,” IET Electr.
Power Appl., vol. 1, no. 4, pp. 489–516, Jul. 2007.
the carrier-based algorithm based on “double injection,” using [20] O. López, E. Levi, F. D. Freijedo, and J. Doval-Gandoy, “Number of
a passive seven-phase load. Comparison is based on phase volt- switching state vectors and space vectors in multilevel multiphase con-
age and current THD analysis, and on time-domain waveforms verters,” Electron. Lett., vol. 45, no. 10, pp. 524–525, 2009.
[21] E. Levi, D. Dujic, M. Jones, and G. Grandi, “Analytical determination of
and spectra. Identical performance is demonstrated for the two DC-bus utilization limits in multiphase VSI supplied AC drives,” IEEE
PWM algorithms. However, complexity of the space vector al- Trans. Energy Convers., vol. 23, no. 2, pp. 433–443, Jun. 2008.
gorithm is, as always the case is, much higher. [22] F. Wang, “Sine-triangle versus space-vector modulation for three-level
PWM voltage-source inverters,” IEEE Trans. Ind. Appl., vol. 38, no. 2,
pp. 500–506, Mar./Apr. 2002.
[23] G. Carrara, S. Gardella, M. Marchesoni, R. Salutari, and G. Sciutto, “A
REFERENCES new multilevel PWM method: A theoretical analysis,” IEEE Trans. Power
Electron., vol. 7, no. 3, pp. 497–505, Jul. 1992.
[1] E. Levi, “Multiphase electric machines for variable-speed applications,”
[24] A. Ruderman and B. Reznikov, “PWM power converter voltage quality
IEEE Trans. Ind. Electron., vol. 55, no. 5, pp. 1893–1909, May 2008.
bounds and their applicability to non-PWM control schemes,” in Proc. Int.
[2] B. Wu, High-Power Converters and AC Drives. Hoboken, NJ: IEEE
Conf. Optimization Electr. Electron. Equipment, Brasov, Romania, 2010,
Press–Wiley, 2006.
pp. 58–68.
[3] B. Mwinyiwiwa, O. Ojo, and Z. Wu, “A five phase three-level neutral
[25] A. Ruderman and B. Reznikov, “Time domain evaluation of filterless grid-
point clamped converter using DSP and FPGA based control scheme,” in
connected multilevel PWM converter voltage quality,” in Proc. IEEE Int.
Proc. IEEE Power Electron. Spec. Conf., Jeju, South Korea [CD-ROM],
Symp. Ind. Electron., Bari, Italy, Jul. 2010, pp. 2940–2945.
2006, pp. 1–7.
DORDEVIC et al.: VECTOR SPACE DECOMPOSITION BASED SPACE VECTOR PWM ALGORITHM 649
[26] D. Dujic, M. Jones, and E. Levi, “Analysis of output current ripple RMS Emil Levi (S’89–M’92–SM’99–F’09) received the
in multi-phase drives using polygon approach,” IEEE Trans. Power Elec- M.Sc. and Ph.D. degrees from the University of
tron., vol. 25, no. 7, pp. 1838–1849, Jul. 2010. Belgrade, Belgrade, Serbia, in 1986 and 1990,
[27] E. Levi, M. Jones, S. N. Vukosavić, and H. A. Toliyat, “A novel concept respectively.
of a multiphase, multimotor vector controlled drive system supplied from From 1982 to 1992, he was with the Department
a single voltage source inverter,” IEEE Trans. Power Electron., vol. 19, of Electrical Engineering, University of Novi Sad. In
no. 2, pp. 320–335, Mar. 2004. May 1992, he joined Liverpool John Moores Univer-
sity, Liverpool, U.K., where since September 2000, he
has been a Professor of electric machines and drives.
Dr. Levi serves as a Co-Editor-in-Chief of the
IEEE TRANSACTION ON INDUSTRIAL ELECTRONICS,
as an Editor of the IEEE TRANSACTION ON ENERGY CONVERSION, and as an
Editor-in-Chief of the IET Electric Power Applications. He is the recipient of
the Cyril Veinott award of the IEEE Power and Energy Society for 2009.
Obrad Dordevic (S’11) received the Dipl. Ing. de- Martin Jones (M’07) received the B.Eng. degree
gree from the University of Belgrade, Belgrade, (First Class Hons.) from the Liverpool John Moores
Serbia, in 2008. Since 2009, he has been working to- University, Liverpool, U.K., in 2001. He was a Re-
ward the Ph.D. degree in the Liverpool John Moores search Student at the Liverpool John Moores Univer-
University, Liverpool, U.K. sity from September 2001 till Spring 2005, when he
From 2008 to 2009, he was with the Digital Drive received the Ph.D. degree.
Control laboratory, University of Belgrade. His main He is currently a Reader at Liverpool John Moores
research interests include the areas of power electron- University.
ics, electrostatic precipitators, and advanced variable Dr. Jones was a recipient of the IEE Robinson Re-
speed drives. search Scholarship for his PhD studies.