Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
OF MULTILEVEL CONVERTERS
Nikola Celanovic
Dissertation submitted to the Faculty of the
Virginia Polytechnic Institute and State University
in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
in
Electrical Engineering and Computer Engineering
Dushan Boroyevich, Chair
William T. Baumann
Jason Lai
Fred C. Lee
Douglas J. Nelson
September 20, 2000
Blacksburg, Virginia
Keywords: Superconductive Magnetic Energy Storage, Neutral Point Balancing Problem,
FeedForward Control
Copyright 2000, Nikola Celanovic
ii
SPACE VECTOR MODULATION AND CONTROL
OF MULTILEVEL CONVERTERS
Nikola Celanovic
(ABSTRACT)
This dissertation is the result of the research and development of a power conditioning system
for superconductive magnetic energy storage systems (SMES). The dominant challenge of this
research was to develop a power conditioning system that can match slowly varying dc voltages
and dc currents on the superconductive magnet side with the ac voltages and ac currents on the
utility side. At the same time, the power conditioning system was required to provide a bi
directional power flow to the superconductive magnet.
The focus of this dissertation is a threelevel diodeclamped dc/ac converter, which is a
principle part of the power conditioning system. Accordingly, this dissertation deals with the
space vector modulation (SVM) of threelevel converters and introduces a computationally very
efficient threelevel SVM algorithm that is experimentally verified.
Furthermore, the proposed SVM algorithm is successfully generalized to allow equally
efficient, realtime implementation of SVM to dc/ac converters with virtually any number of
levels. The most important advantage of the proposed concept is that the number of instructions
required to implement the algorithm is almost independent from the number of levels in a
multilevel converter.
More on the side of the control of multilevel converters, particular attention in this
dissertation is paid to the problem of charge balance in the split dclink capacitors of threelevel
neutralpointclamped converters. It is a known fact that although the charge balance in the
neutral point (NP) can be maintained on a line cycle level, a significant third harmonic current
iii
flows into the NP for certain loading conditions, causing the neutralpoint voltage ripple. The
logical consequence of that ripple is the deteriorated quality of the output voltage waveforms as
well as the increased voltage stress on the switching devices.
This was the motivation to more carefully explore the loading conditions that cause the
imbalance, as well as to study the fundamental limitations of dclink capacitor charge balancing
algorithms. As part of this work, a new model of the neutralpoint current in the rotating
coordinate frame is developed as a tool for the investigation of theoretical limitations and in
order to provide some insight into the problem. Additionally, the lowfrequency ripple is
quantified and guidelines are offered that can help determine the correct size for the dclink
capacitors.
Because the study of the neutralpoint balance identified the loading conditions that (under
some possible system constraints) cause an unavoidable neutralpoint voltage ripple, a feed
forward type of control method is developed next. The proposed feedforward algorithm can
effectively prevent the neutralpoint voltage ripple from creating distortions in the converter
output voltage under all loading conditions and without causing additional disturbance in the
neutralpoint voltage. The feedforward method is developed for a sine triangle as well as for the
SVM type PWM algorithm.
The simulation results that include the full dynamic model of the converter and load system
validate the feedforward approach and prove that the feedforward algorithm can effectively
compensate for the effect of the neutralpoint voltage ripple. The simulation results are then
experimentally verified.
iv
ACKNOWLEDGMENTS
I wish to express my sincere gratitude to my advisor, Dr. Dushan Boroyevich, for his
guidance, encouragement and support during my graduate studies. His impressive knowledge,
technical skills and human qualities have been a source of inspiration and a model for me to
follow.
I’m also grateful to my professor, Dr. Fred. C. Lee, for his help and guidance during my
graduate studies at Virginia Tech. I’m thankful for the opportunity to be a member of the
Virginia Power Electronics Center and to watch it grow into the Center for Power Electronics
Systems. It has been a memorable experience, both professionally and personally.
I wish to thank my committee members, Dr. William Baumann, Dr. Jason Lai and Dr.
Douglas Nelson, for inspiring classes, encouragement in difficult moments and for serving on
my advisory committee. At this point, I wish to specifically emphasize my gratiude for all the
help and encouragement I received from Dr. Vlatko Vlatkovic and Dr. Jim Lyons from GE
CR&D Schenectady, NY. Their guidance and insight gave me encouragement to proceed with
confidence towards the completion of this work.
I would also like to acknowledge CPES students and staff for providing an enjoyable
educational atmosphere. Especially, I would like to thank my brother Ivan Celanovic for endless
discussions, Dr. Dong Ho Lee for sharing with me his rich practical experience, Dengming Peng
for working with me on the project, and all others who made my work and my stay in CPES
more enjoyable.
With much love, I thank my parents Marija and Lazar for their loving care and their
unconditional support. This dissertation is their accomplishment as much as it is mine.
Finally, I would like to thank my good old friend, Dr. Vera Vlahovic, for her support,
encouragement in times of need, and her insight.
v
To my parents
vi
Table of Contents
1 INTRODUCTION.............................................................................................................. 1
1.1 Motivation and Objectives............................................................................................... 1
1.2 Review of Previous Research........................................................................................ 3
1.3 Dissertation Outline and Major Results......................................................................... 9
2 MULTILEVEL CONVERTERS’ MODULATION AND CONTROL........................... 11
2.1 Multilevel Topologies ..................................................................................................... 13
2.1.1 DiodeClamped Multilevel Topology....................................................................... 13
2.1.2 Flying Capacitor Multilevel Topology...................................................................... 14
2.1.3 Multilevel Configurations with Cascaded TwoLevel FullBridge Inverters ........... 16
2.1.4 Additional FullBridge Topologies ........................................................................... 17
2.2 Multilevel Digital Control Hardware ............................................................................. 19
2.2.1 Overview of the Architecture of Digital Control Hardware ..................................... 19
2.2.2 Components of the Digital Controller for SMES Power Conditioning System........ 21
2.2.3 Controller Architecture ............................................................................................. 22
2.2.4 Control Algorithm.................................................................................................... 25
2.2.5 Functions of the EPLD.............................................................................................. 26
2.3 Multilevel Power Conditioning System as a Power Controller for a
Superconducting Magnetic Energy Storage System.......................................................... 30
2.3.1 Power Conditioning System Design approach.......................................................... 31
2.3.2 Modeling and Control of the Chopper ...................................................................... 33
2.3.3 Modeling of a ThreeLevel Voltage Source Inverter................................................ 38
2.3.4 Control of the Inverter in Rotating Coordinates........................................................ 41
2.3.5 Simulation and Experimental Results ....................................................................... 46
2.4 Modeling and Control Approach and Problem Formulation..................................... 49
2.5 Conclusions ..................................................................................................................... 50
3 A FAST SPACE VECTOR MODULATION ALGORITHM FOR MULTILEVEL
CONVERTERS............................................................................................................... 51
3.1 Switching Vectors, Switching States and Reference Vector................................... 52
3.2 Space Vector Modulation of a ThreeLevel Converter............................................. 55
3.3 The New Space Vector Modulation Algorithm........................................................... 61
3.3.1 Coordinate Transformation....................................................................................... 63
3.3.2 Detection of Nearest Three Vectors.......................................................................... 64
3.3.3 Duty Cycles Computation......................................................................................... 65
3.3.4 SwitchingState Selection......................................................................................... 67
3.4 Practical Implementation of The New SVM Algorithm.............................................. 68
3.5 Conclusions ..................................................................................................................... 77
4 NP BALANCING PROBLEM IN THREELEVEL DIODECLAMPED
CONVERTERS............................................................................................................... 78
4.1 The Effect of Different Switching States on the Current in the NP......................... 80
4.2 Neutral Point Current Modulation................................................................................. 82
4.3 Analysis and Control of the Neutral Point Voltage Error .......................................... 85
4.4 Limitations of the NeutralPoint Control and Some Design Recommendations... 93
4.5 Experimental Results..................................................................................................... 96
vii
4.6 Conclusions ..................................................................................................................... 99
5 FEEDFORWARD CONTROL OF THREELEVEL NEUTRALPOINT CLAMPED
CONVERTERS............................................................................................................. 100
5.1 The Effect of the Capacitor Charge Imbalance on the Voltage Space Vectors . 101
5.2 Sine Triangle Modulation for ThreeLevel Converters............................................ 106
5.2.1 PWM Generation with Sine Triangle Modulation.................................................. 106
5.2.2 NP Control With Sine Triangle Modulation........................................................... 107
5.2.3 The Average Model of a Sine Triangle Modulator................................................. 108
5.3 Modeling the FeedForward Control.......................................................................... 113
5.3.1 The FeedForward Algorithm................................................................................. 114
5.3.2 Mathematical Basis of the FeedForward Algorithm.............................................. 115
5.3.3 Average Model of the ZeroSequence NP Control................................................. 117
5.3.4 A SmallSignal Model of the ZeroSequence NP Control...................................... 120
5.3.5 The Stability Analysis of the Feed Forward Algorithm.......................................... 121
5.4 Implementation of FeedForward Algorithm for Space Vector Modulation......... 124
5.5 Conclusions ................................................................................................................... 127
6 FEEDFORWARD CONTROL OF THREELEVEL NEUTRALPOINTCLAMPED
CONVERTERS—SIMULATION AND EXPERIMENTAL RESULTS ................... 128
6.1 Analysis and Experimental Verification of the Line Cycle Average Model of the
FeedForward Algorithm.......................................................................................................129
6.1.1 Control Authority Over the NP Current, with and without FeedForward Terms.. 131
6.2 Analysis of the SwitchingCycle Average Model of the Sine Triangle Modulator
with the FeedForward Algorithm........................................................................................ 135
6.2.1 Average Model of ThreeLevel Converter with Sine Triangle Modulator ............. 135
6.2.2 Performance of FeedForward Algorithm, Simulation Results .............................. 137
6.3 Average Model of Fast Space Vector Modulator, Experimental Verification of the
Model ....................................................................................................................................... 141
6.3.1 Average Model of the ThreeLevel Converter with Fast Space Vector Modulation
Algorithm............................................................................................................................ 142
6.3.2 The Fast Space Vector Modulation Algorithm, Experimental Results................... 143
6.4 Conclusions ................................................................................................................... 146
7 CONCLUSIONS ........................................................................................................... 148
REFERENCES......................................................................................................................... 152
VITA…………............................................................................................................................ 159
viii
List of Figures
Figure 2.1. Multilevel converter topologies................................................................................. 12
Figure 2.2. Multilevel modulation techniques. ............................................................................. 12
Figure 2.3. The three and fourlevel neutralpointclamped phase leg. ...................................... 13
Figure 2.4. Three and fourlevel flying capacitor phase leg. ...................................................... 15
Figure 2.5. Phase leg of a cascaded fivelevel fullbridge inverter.............................................. 16
Figure 2.6. Phase leg of a cascaded full bridge sevenlevel converter......................................... 17
Figure 2.7. Diode clamped full bridge fivelevel topology.......................................................... 18
Figure 2.8. Flying capacitor fullbridge fivelevel topology........................................................ 18
Figure 2.9. Power electronics system overview. .......................................................................... 20
Figure 2.10. Architecture of the digital controller for power electronics systems....................... 20
Figure 2.11. Architecture of the SMES digital controller board.................................................. 23
Figure 2.12. Digital interface board. ............................................................................................ 24
Figure 2.13. Analog interface board............................................................................................. 25
Figure 2.14. Control overview. .................................................................................................... 26
Figure 2.15. PWM signal generation logic within the EPLD. ..................................................... 27
Figure 2.16. A doublebuffered register bank inside the EPLD................................................... 27
Figure 2.17. Inverted gate drive signals with deadtime............................................................. 29
Figure 2.18. Power conditioning system for SMES..................................................................... 32
Figure 2.19. Equivalent switching model of threelevel chopper. ............................................... 34
Fig. 2.20. Space voltage vectors of the chopper........................................................................... 34
Figure 2.21. Average model of the chopper................................................................................. 36
Figure 2.22. Buck mode control of the chopper (charging the SC). ............................................ 37
Figure 2.23. Boost mode control of the chopper (discharging the SC)........................................ 37
Figure 2.24. Switching network of a threelevel VSI .................................................................. 38
Figure 2.25. Inverter average model in rotating dq coordinate frame. ....................................... 42
Figure 2.26. Inverter control in the utility mode (charging the SC). ........................................... 43
Figure 2.27. Inverter control in rectifier mode (discharging the SC)........................................... 43
Figure 2.28. Voltage space vectors of a threelevel inverter........................................................ 45
Figure 2.29. Experimental SMES PCS system. ........................................................................... 46
Figure 2.30. Simulated and measured rectifier current loop gain. ............................................... 47
Figure 2.31. Simulated and measured rectifier voltage loopgain................................................. 47
Figure 2.32. Rectifier response to change in current command……………………................... 48
Figure 3.1. Architecture of digital controller for threephase converters..................................... 51
Figure 3.2. Functional diagram of a threeleg nlevel converter.................................................. 52
Figure 3.3. Switching states of a threelevel converter................................................................ 54
Figure 3.4. Voltage space vectors of a threelevel converter....................................................... 56
Figure 3.5. Synthesis of V
REF
in outer small triangle................................................................... 59
Figure 3.6. Synthesis of V
REF
in middle small triangle ............................................................... 60
Fig. 3.7. Synthesis of V
REF
in inner small triangle. ..................................................................... 61
Figure 3.8. The voltage space vectors of a fourlevel converter.................................................. 62
Figure 3.9. Switchingstate vectors of a threelevel converter in a hexagonal coordinate system.
............................................................................................................................................... 64
ix
Figure 3.10. Normalized linetoline PWM voltage waveforms for three, four and fivelevel
converters. ............................................................................................................................. 68
Figure 3.11. Overview of the new SVM algorithm...................................................................... 69
Figure 3.12. Computing the nearest four vectors......................................................................... 70
Figure 3.13. Computation of duty cycles and selection of switching states. ............................... 71
Figure 3.14. Switchingstate selection with dclink capacitor charge control. ............................ 74
Figure 4.1. Circuit schematic of a threelevel VSI....................................................................... 79
Figure 4.2. Switchingstate vectors of the threelevel VSI. ......................................................... 80
Figure 4.3. The position of equivalent switches for the switching state pon............................... 81
Figure 4.4. Outer small triangle. .................................................................................................. 83
Figure 4.5. Duty cycles of SVM for modulation index m=0.8. ................................................... 85
Figure 4.6. Switching functions of the medium vectors. ............................................................. 86
Figure 4.7. Weighing factors for medium vectors with m=0.8.................................................... 90
Figure 4.8. Weighing factors for small vectors with m=0.8. ....................................................... 91
Figure 4.9. Region where LF ripple can be suppressed. .............................................................. 94
Figure 4.10. Normalized amplitude of the LF charge ripple........................................................ 95
Figure 4.11. Capacitor sizes for specified NP ripple, with and without NP control.................... 96
Figure 4.12. Experimental verification of NP voltage ripple properties...................................... 97
Figure 4.13. Normalized NP voltage for power factor angle of 85 degrees. ............................... 98
Figure 4.14. Theoretical and experimental normalized NP voltage ripple. ................................. 98
Figure 5.1. Functional diagram of a threelevel converter......................................................... 102
Figure 5.2. Voltage space vectors of a threelevel converter..................................................... 102
Figure 5.3. The definitions of the dclink voltages and the voltage error in the NP.................. 103
Figure 5.4. Shift of the medium voltage vectors due to the ∆V>0 error in NP voltage............. 104
Figure 5.5. “Splitting” of the small voltage vectors due to the ∆V>0 error in NP voltage........ 105
Figure 5.6. Principle of sine triangle modulation for a threelevel converter. ........................... 107
Figure 5.7. Principle of NP control for the threelevel converter. ............................................. 108
Figure 5.8. Duty cycles for the sine triangle threelevel modulation. ........................................ 110
Figure 5.9. Duty cycle of one phase equivalent switch for sine triangle modulation. ............... 111
Figure 5.10. Duty cycles of one phase with the zero sequence present. .................................... 113
Figure 5.11. Duty cycles for the sine triangle threelevel modulation. ...................................... 115
Figure 5.12. The model of the modulator with respect to the NP current.................................. 117
Figure 5.13. The duty cycles of the sine triangle modulator with zero sequence. ..................... 119
Figure 5.14. The PWM signal with strong feedforward component. ....................................... 122
Figure 5.15. Power flow in the inverter mode of operation with large imbalance in NP. ......... 123
Figure 5.16. The effect of NP unbalance on the boundaries of small triangles. ........................ 124
Figure 5.17. The active sector computation based on the home vertex algorithm...................... 126
Figure 6.1. Experimental measurement of the loop gain in the NP controller........................... 130
Figure 6.2. Measurement of the NP control loop gain without normalization........................... 131
Figure 6.3. The linecycle average NP current as a function of the zerosequence component δ.
............................................................................................................................................. 133
Figure 6.4. The linecycle average NP current with feedforward terms included.................... 134
Figure 6.5. Average model of the sine triangle modulator with feed forward........................... 136
Figure 6.6. Average model of one phase leg sine triangle modulator........................................ 137
Figure 6.7. Model of NP bangbang controller. ......................................................................... 137
Figure 6.8. Rectifier mode of operation with low power factor................................................. 139
x
Figure 6.9. Rectifier mode of operation for high power factor.................................................. 140
Figure 6.10. The inverter mode of operation with low power factor load. ................................ 141
Figure 6.11. Average model of the fast SVM algorithm............................................................ 142
Figure 6.12. Fast threelevel space vector modulator. ............................................................... 143
Figure 6.13. Measured and simulated NP ripple of a threelevel converter............................... 144
Figure 6.14. Measured and simulated line voltage in the presence of NP ripple....................... 145
Figure 6.15. Simulated linetoline currents for balanced and symmetric load. ........................ 146
xi
List of Tables
Table 2I Neutralpoint current i
NP
for different small vectors. ................................................... 36
Table 3I The switchingstate lookup table. ............................................................................... 75
Table 3II The performance of the algorithm implementation for a threelevel converter.......... 76
Table 4I NP current i
NP
for different space vectors. ................................................................... 81
Table 4II Currentswitching function for medium vectors......................................................... 89
Table 4III Currentswitching function for S0 small vectors....................................................... 89
Table 4V Currentswitching function for S1 small vectors........................................................ 89
Table 5I Effect of the error in NP voltage on the medium voltage vectors.............................. 104
Table 5II Effect of the error in NP voltage on the small voltage vectors................................. 105
1 INTRODUCTION
1.1 Motivation and Objectives
In recent years, there has been increased interest in renewable energy sources on one side,
while on the other, the initiative for deregulation of the utilities increased the interest in systems
that can increase their flexibility. Both these developments recognize power electronics systems
as a core enabling technology.
For example, renewable and/or clean energy sources such as wind turbine, solar power, fuel
cells and micro turbines all require powerconditioning systems. While these
powerconditioning systems provide the necessary interface to the utility, they also match the
electrical characteristics of the sources with the electrical requirements of the loads and/or the
utility. Similar to the renewable sources, energy storage systems (viewed as key components of
the envisioned flexible utilities) also require power conditioning systems that can match the
electrical characteristics of the energy storage systems with those of the utility.
What makes power electronics solutions unique in both these cases is that they are
fundamentally multifunctional and can provide not only their principle interfacing function but
various utility functions as well. This is particularly true of energy storage systems, which are
potentially well suited to provide a host of additional functions on the utility level. Some of
these functions are load leveling, utility stabilization, static var compensation, harmonic
compensation and active filtering. In addition, they can also be used as uninteruptible power
2
supply (UPS) systems at various power levels, and can be designed to provide a pulse power
capability that is of interest for certain military applications as well as for some highenergy
physics experiments.
One of the practical energy storage technologies is superconductive magnetic energy storage
(SMES). The design of a power conditioning system for a SMES system was one of the primary
objectives of the work that will be presented in this dissertation. However, some of the insight
gained from the development of a SMES power conditioning system is readily applicable to
power conditioning systems for batteries, fuel cells, flywheels, super capacitors and other energy
storage technologies.
What makes this power conditioning system (PCS) design even more interesting is that it was
built using multilevel converters, which allowed a research into the issues, and tradeoffs
involved with multilevel power converter technology in general. Therefore this work followed a
trend in the power electronics community to shift towards multilevel power electronics converter
topologies that seem to offer significant advantages for utility applications over the more
conventional twolevel topologies. This is the principle reason why this work concentrates on
some important aspects of multilevel topologies that are directly applicable to a wide range of
multilevel converters.
For example, together with the wider introduction of multilevel converter topologies, there is a
strong tendency to modularize the hardware design in order to reduce the prices, simplify
converter design and potentially increase the availability of multilevel converters. However,
with the increased number of levels, the number of allowable switching states in the converters is
also rapidly increasing. This situation places significant computational challenges in front of the
PWM modulator, which is typically implemented in software and operating in real time and for
high switching speed.
This problem is addressed with a new and general multilevel algorithm. The algorithm very
efficiently performs a switching vector identification and duty cycle computation; even more
3
importantly, the number of steps involved in the computation is almost independent of the
number of levels in the converter.
Secondly, with respect to the voltage balancing problem of the threelevel neutralpoint
clamped converter, which is currently perhaps the most widespread multilevel topology, a
problem of split dclink capacitors’ charge balancing is explored in a comprehensive way. This
is accomplished using the new model of the neutral point (NP) current in the rotating dq
reference frame. The study resulted in the identification the loading conditions that result in the
lowfrequency ripple, the quantification of the size of the ripple, and the ability to make
recommendations for the size of the dclink capacitors for any given operating point as well as
the NP ripple specification.
And thirdly, because the study of the NP balance identified the loading conditions under
which certain system constraints cause an unavoidable NP voltage ripple, a feedforward control
method is developed. It is the objective of this dissertation to investigate the effects of the feed
forward algorithm on the quality of the output waveforms as well as on the control authority over
the charge balance in the split dclink capacitors.
1.2 Review of Previous Research
When designing a system, particularly a relatively complex system such as the SMES power
conditioning system, there are many important considerations, in areas ranging from the device’s
physics and soft switching all the way to the mathematical abstractions used for modeling and
control. Because of the overwhelming breadth of published material on these topics, this
literature review will be limited to multilevel topologies, multilevel modulations and the control
of multilevel converters, because these topics are closely related to the work in this dissertation.
At this time, the area of multilevel power conversion can still be considered young. In 1980,
early interest in multilevel power conversion technology was triggered by the work of Nabae, et
al., who introduced the neutralpointclamped (NPC) inverter topology [B1]. It was immediately
4
realized that this new converter had many advantages over the more conventional twolevel
inverter. Subsequently, in the early nineties the concept of the threelevel converter was
extended further and some new multilevel topologies were proposed.
At the present time, the majority of research and development effort seems to concentrate on
the development of three classes of converters: the diodeclamped multilevel converter, the
multilevel converter with cascaded singlephase Hbridge inverters, and the multilevel converter
known as the flying capacitor converter, or sometimes as the imbricated cells multilevel
converter. The two remaining configurations, namely the multilevel converter with bidirectional
switch interconnections [A1] and the multilevel converter with multiple threephase inverters are
receiving less attention at the present time and will not be discussed further.
As mentioned, the threelevel NPC converter was the first widely popular multilevel
topology. A significant amount of research effort was invested in the development of this family
of converters [B], which resulted in the fact that today this topology probably represents the
stateoftheart multilevel converter technology. This is in part because the converters of this
type are currently in the phase of rapid market introduction by several worldrenowned
manufacturers [B20, B22].
Soon after the introduction of the NPC converter, the original topology was extended to a
higher number of levels using the same principle of diodeclamped intermediate voltage levels
[C1]. The backtoback connection of two converters of this type [A7, A8, C4, C5] enables bi
directional power flow, facilitates balancing of the intermediate capacitor levels and has a very
“clean” utility interface. The main disadvantage of this topology is that the required blocking
voltage of the clamping diodes is proportional to the level for which they are used to employ
clamping action. Consequently, series connection of the diodes may be required. In addition,
because of the high switching speed of today’s switching devices, mainly IGBTs and IGCTs, the
clamping diodes can be subject to severe reverserecovery stress, which challenges this
technology.
5
An alternative to the diodeclamped converter, the flying capacitor topology does not have
issues with clamping diodes. First proposed in 1992 [E1], this approach has the advantage of a
larger number of redundant switching states, which allows more freedom in balancing the
clamping capacitors’ voltages. The main disadvantage is the potential for parasitic resonance
between the decoupling capacitors; this is made even worse by the high number of capacitors,
which complicates packaging for small inductance. In addition, there are issues with voltage
redistribution in the case of voltage surges. Nevertheless, the flying capacitor topology seems
very promising.
Finally, the multilevel configuration with cascaded Hbridge inverters presents another
alternative in the design of multilevel converters. One of the earlier applications of this topology
was for plasma stabilization [D1]; it was then extended for threephase applications [D3]. A
primary advantage of this topology is that it provides the flexibility to increase the number of
levels without introducing complexity into the power stage. Also, this topology requires the
same number of primary switches as the diodeclamped topology, but does not require the
clamping diode. However, this configuration uses multiple dedicated dcbusses and often a
complicated and expensive line transformer, which makes this a rather expensive solution. In
addition, bidirectional operation is somewhat difficult (although not impossible) to achieve
[D10].
Perhaps the most important improvement in cascaded converter topologies is the hybrid
multilevel topology [D7]. The main strength of this approach is its combination of the high
voltage capacity of the relatively slow GTO devices with the high switching frequency of the
lower voltage capacity IGBT devices. At the same time, the different voltage levels of the IGBT
and GTO bridges create an additional voltage level without any additional complexity.
Modulation techniques in multilevel converters for utility applications were originally
restricted to the synthesis of stepped waveforms [A4]. This strategy can be interpreted as a
quantization process in which the reference sinusoidal voltage is approximated by discrete levels
of voltage available at the dcbus. The implementation of this type of modulation is relatively
simple and does not require high switching speed, which makes it suitable for converters realized
6
with GTO thyristors. However, the spectral performance is not very good and a variable dcbus
voltage is required in order to obtain variable voltage at the output.
Far more popular, especially for the converters with faster switching devices, are the socalled
sinusoidal PWM techniques. These methods have been extensively studied and are among the
most popular in industrial applications. These methods involve a comparison of the reference
signal with a triangular carrier waveform and the detection of crossover instances to determine
switching events. The variations of these methods are essentially in the polarity and shape of the
carrier waveforms. Although there are slight differences in spectral performance [C16], for the
highenough frequency ratio between the carrier and the reference waveforms, there is no
substantial difference between them.
Another very interesting method is based on multiple reference waveforms. It is also
sometimes referred to as dipolar modulation [C17]. This method uses a single triangular carrier
and two sinusoidal reference waves to modulate threelevel inverters. The sinusoidal reference
waves are usually obtained by adding positive and negative offsets to the conventional reference
signal. The principal reason to introduce the offset is to always interrupt the switching between
the outermost levels and the middle level, as well as to avoid any minimum pulse width
problems for low modulation indexes.
Perhaps one of the most popular modulation approaches for twolevel converters is space
vector modulation (SVM), which is now being used more and more in the control of multilevel
converters. The concept of space voltage vectors corresponding to various switching states has
been applied to study the impact of various switching states on the capacitor charge balancing in
almost every paper discussing the SVM approach [B6, B1011]. An advantage of the SVM is
the instantaneous control of switching states and the freedom to select vectors in order to balance
the NP. Additionally, one can realize output voltages with almost any average value by using the
nearest three vectors, which is the method that results in the best spectral performance.
The primary constraint in devising SVM (as is the case with in most multilevel modulation
techniques) involves minimizing the harmonic content of the output PWM waveform while at
7
the same time maintaining the capacitor charge balance. This is the basic tradeoff of threelevel
SVM, because the tight regulation of the capacitor charge balance can always be achieved at the
expense of increased switching losses and poorer harmonic performance.
Another typical design problem particularly important for the converters built with the
slowerswitching GTO thyristor devices is the minimum pulse width. This problem is present
whenever the reference vector is close in phase and magnitude to any of the switching state
vectors. One way of dealing with this problem is to change from the nearest three vector
approach (NTV) to nearest four vector approach. Thus a narrow pulse problem can be solved at
the expense of increased switching stress and somewhat increased harmonic content. Perhaps a
better (but not necessarily more practical) alternative involves the division of the sixtydegree
regions into fourteen subregions with their switching sequence optimized in order to avoid the
narrow pulses [B9].
Finally, it is accepted as conventional wisdom that SVM methods (although very powerful in
reducing harmonics) become computationally very demanding and all but impractical as the
number of levels increases.
With respect to modeling and control of three and higher level converters, the dominant
approach is to divide and conquer. What this means in practice is that phase legs are treated
essentially as voltage sources (as is the case in twolevel converters), while the charge balance
becomes part of the modulator, which selects between the redundant switching states in order to
maintain the charge balance in the split dclink capacitors.
This approach makes the charge balance controller relatively simple. For sinetriangle
modulated converters, this amounts to control of the modulating signal’s zero sequence for the
threelevel converter [B2, B3], or the manipulation of the redundant small vectors [B4].
Building a model that can describe the behavior of the NP voltage balance is quite difficult
because the NP current as well as NP “duty cycles” are piecewise nonlinear functions. In
addition, the steadystate response usually has a strong third harmonic component. Taking all
8
this into account, perhaps the best approach is to average the NP current across the whole line
cycle [B14]. Although this approach can not determine either the size or the shape of the voltage
ripple, it does show the trends and the overall stability of the dclink voltage, and can be
linearized and used to design a NP voltage controller.
Perhaps the first comprehensive approach to modeling threephase PWM converters was
presented by Ngo, et al. [F14]. The modeling was performed in four steps: development of the
switching model, averaging, transformation in the rotating coordinates, and linearization. The
converter models published thereafter [F2, F4, F15, F16] mainly followed the same steps and
resulted converter models that describe the lowfrequency dynamic behavior of currents and
voltages of multilevel converters as well. Naturally, they do not describe the effects of the split
dclink capacitors, but under the usually valid assumption that the NP voltages can be controlled
by the modulator and that the ripple is sufficiently small, they do accurately describe the low
frequency dynamic behavior.
Finally, it is important to mention an attempt to make a unified dynamic model [F8, F13] of a
threelevel converter that included the NP voltage dynamics together with all the other voltages
and currents. The model was developed using the usual steps of developing the switching model
first, than averaging, transforming into the rotating frame, and linearizing.
Unfortunately, in order to include the dynamics of the split dclink capacitors, the switching
model had to be developed using the phase instead of the linetoline switching functions. This
resulted in a large number of duty cycles to be controlled, altogether six for a threephase
converter. The linear model was than obtained by linearizing the average model in a rotating
coordinate frame around the operating point. Clearly, the steadystate solutions of the linear
model are constant duty cycles, which is only one of the infinite number of functions that steady
state duty cycles of the average (nonlinear) model can have. Essentially, in this case, the
coordinate transformation and linearization reduces the number of possible solutions from
infinity to one suboptimal solution, thus taking away most of the freedom to optimize the
switching patterns.
9
Therefore, although interesting from the theoretical point of view, by controlling good part of
the modulation itself, this method effectively limits the freedom to optimize the switching
patterns without offering additional insight into converter operation.
1.3 Dissertation Outline and Major Results
The work in this dissertation is divided into the following major parts.
ThreeLevel DiodeClamped Converter. This chapter is, in a sense, an extended introduction
that discusses the requirements that an effective power conditioning system for SMES needs to
satisfy. At the same time, this chapter covers the topological, control design and modeling issues
unique to the approach that was taken in the design of the prototype power processing converter
that was designed as a part of this dissertation work. In addition, there is a discussion of how this
approach compares with other available converter technologies, as well as the problems that
were encountered and the original solutions that were proposed.
Implementation of a ThreeLevel Converter in a SMES Power Conditioning System.
Finally, an approach to modeling and implementing a control system for a threelevel NP
clamped converter for the application in the SMES system is presented. The control design
enables the dc/ac converter of the power processing system to operate as a rectifier when the
magnet is being charged, and as a utility inverter when the magnet is being discharged. The
control design is based on a twolevel dc/ac converter model that is experimentally proven to
closely match the dynamic behavior of a threelevel converter. Finally, specific requirements for
the control of a dc/ac converter operating in an environment with frequent demands for changes
in operating mode and power flow direction are discussed.
A fast SVM Algorithm for Multilevel Converters. This chapter introduces an original SVM
algorithm for nlevel threephase converters. This algorithm is computationally extremely
efficient and is almost independent of the number of levels of the converter. At the same time, it
provides excellent insight into the operation of multilevel converters. In addition, the
10
experimental results from the operation of the prototype power conditioning converter are
included as is the discussion of the algorithm’s strengths compared with SVM algorithms. This
chapter also covers the details of the realtime algorithm’s implementation on the DSP processor
using the example of a threelevel space vector modulator.
Neutral Point Balancing Problem in ThreeLevel NPClamped Converters. It is known
that under certain loading conditions and for certain output voltages a charge balance in the
threelevel converter can not be maintained on the switching cycle level. This results in the low
frequency voltage ripple in the NP, which is the point between the two dclink capacitors. This
chapter explores the fundamental limitations of the NP voltage balancing problem for different
loading conditions of threelevel voltage source converters. A new model in dq coordinate frame
(utilizing current switching functions) is developed as a means to investigate theoretical
limitations and lend to more insight into the problem. The lowfrequency ripple of the NP
(caused by certain loading conditions) is observed and quantified. The theoretical results are
then experimentally verified and guidelines for choosing the dclink capacitor size are given.
FeedForward Control of ThreeLevel DiodeClampedConverters. The results from the
previous chapter indicate that under certain loading conditions, the lowfrequency voltage ripple
in the NP of the dclink capacitors is to be expected. In this chapter, a feedforward control
strategy to compensate this ripple is presented for two modulation algorithms: a sine triangle and
a space vector PWM algorithm. A full dynamic model is developed to help better understand the
effect of the feedforward algorithm on the quality of output waveforms as well as its effect on
the control authority over the charge balance of the dclink capacitors. Interestingly, the
proposed algorithm seems to effectively compensate for the disturbance in the output voltage
created by the voltage ripple in the NP, while simultaneously improving the control authority
over the capacitor charge balance. Some of the theoretical results are experimentally verified.
11
2 MULTILEVEL CONVERTERS’ MODULATION AND CONTROL
Multilevel power conversion technology is a very rapidly growing area of power electronics
with good potential for further development. The most attractive applications of this technology
are in the medium to highvoltage range (213 kV), and include motor drives, power
distribution, power quality and power conditioning applications.
In general, multilevel power converters can be viewed as voltage synthesizers, in which the
high output voltage is synthesized from many discrete smaller voltage levels. The main
advantages of this approach are summarized as follows:
• The voltage capacity of the existing devices can be increased many times without the
complications of static and dynamic voltage sharing that occur in seriesconnected devices.
• Spectral performance of multilevel waveforms is superior to that of their twolevel
counterparts.
• Multilevel waveforms naturally limit the problems of large voltage transients that occur due
to the reflections on cables, which can damage the motor windings and cause other problems.
The selection of the best multilevel topology and the best control strategy for each given
application is often not clear and is subject to various engineering tradeoffs. By narrowing this
12
study to the DC/AC multilevel power conversion technologies that do not require power
regeneration, several attractive topological, modulation and power semiconductor device choices
present themselves. The most actively developed of these multilevel topologies are listed in
Figure 2.1.
MULTILEVEL
CONVERTERS
DIODE
CLAMPED
CONVERTERS
FLYING
CAPACITOR
CONVERTERS
CASCADED
FULL BRIDGE
CONVERTERS
Figure 2.1. Multilevel converter topologies.
Similarly, the modulation strategies, with their rich variety of techniques, can be roughly
divided, as shown in Figure 2.2.
MULTILEVEL
MODULATIONS
FUNDAMENTAL
FREQUENCY
SWITCHING
PWM
SINUSOIDAL
PWM
SPACE
VECTOR
PWM
Figure 2.2. Multilevel modulation techniques.
13
2.1 Multilevel Topologies
Although definitions may vary, in this document the multilevel converter has three or more
levels. The intention is to give a brief review of the three basic topologies and their most
promising derivatives. Although this section is by no means a complete review of all the
multilevel topologies, it does cover the ones that are presently receiving the most attention.
2.1.1 DiodeClamped Multilevel Topology
The first practical (and still widely studied) multilevel topology is the neutralpointclamped
(NPC) PWM topology first introduced by Nabae, et al., in 1980 [B1]. The threelevel version of
this topology, shown in Figure 2.3, has several distinct advantages over the twolevel topology.
C
dc
p
n
C
dc
NP
V
pn
A
S
1
S
2
S
3
S
4
D
1
D
2
p
n
C
dc
NP
1
V
pn
A
C
dc
C
dc
S
1
S
2
S
3
S
4
S
5
S
6
D
1
D
3
D
2
D
4
NP
2
Figure 2.3. The three and fourlevel neutralpointclamped phase leg.
The advantages are:
• Voltages across the switches are only half of the dclink voltage.
• The first group of voltage harmonics is centered around twice the switching frequency.
14
• This topology can be generalized, and the principles used in the basic threelevel topology
can be extended for use in topologies with any number of levels.
However, practical experience with this topology revealed several technical difficulties that
complicate its application forhigh power converters. These are as follows:
• This topology requires high speed clamping diodes that must be able to carry full load
current and are subject to severe reverse recovery stress. Although measures to alleviate this
problem can be applied, this remains a serious consideration.
• For topologies with more than three levels the clamping diodes are subject to increased
voltage stress equal to V
pn
.
(n1)/n. Therefore, series connection of diodes might be required.
This complicates the design and raises reliability and cost concerns.
• The issue of maintaining the charge balance of the capacitors is still an open issue for NPC
topologies with more than threelevels. Although the threelevel NPC topology works well
with high power factor loads, NPC topologies with more than three levels are mostly used for
static var compensation circuits. This may be due to the capacitor balancing issues if the
nearest three vector (NTV) modulation is used (which is the modulation scheme with the
least amount of stress and superior spectral performance, and which is therefore preferred
over other possibilities).
2.1.2 Flying Capacitor Multilevel Topology
The flying capacitor multilevel topology from Figure 2.4, was first proposed in 1992, and is
considered to be the most serious alternative to the diodeclamped topology.
15
p
n
C
dc
V
pn
A
S
1
S
2
S
3
S
4
C
dc
p
n
C
dc
V
pn
A C
dc
C
dc
S
1
S
2
S
3
S
4
S
5
S
6
Figure 2.4. Three and fourlevel flying capacitor phase leg.
The significant advantage of this topology is that it eliminates the clamping diode problems
present in the diodeclamped multilevel topologies. Additionally, this topology naturally limits
the dV/dt stress across the devices and introduces additional switching states that can be used to
help maintain the charge balance in the capacitors. Unlike the diodeclamped converter, the
flying capacitor topology has enough switching states to control the charge balance in the single
isolated leg with converters having any number of levels, even if the phase current is
unidirectional. This makes this topology attractive even for the dc/dc converters.
At the present time it seems that this topology has few disadvantages. Nevertheless, some
possible “weak points” that still need to be explored are:
• The dclink capacitor charge controller adds complexity to the control of the whole circuit.
• The flying capacitor topology might require more capacitance than the equivalent diode
clamped topology. In addition, it is obvious that rather large rms currents will flow through
these capacitors. A study of the tradeoffs involved is not described in the available literature.
• There is a potential for parasitic resonance between decoupling capacitors.
16
2.1.3 Multilevel Configurations with Cascaded TwoLevel FullBridge Inverters
One of the early applications of the series connection of singlephase fullbridge inverter
topology was for plasma stabilization in 1988 [13]. This modular approach has since been
extended to include threephase systems as well, and is probably most successfully applied by
Robicon, with their line of mediumvoltage drives, which reportedly have an excellent field
record.
Arguably, the overall complications and cost of the isolated sources for each full bridge is
overall not too serious a drawback and is offset by the advantages of modular construction. The
modularity of this structure allows easier maintenance and provides a very convenient way to
add redundancy into the system [20]. The fivelevel phase leg of the cascaded twolevel full
bridge converter is given in Figure 2.5.
C
dc
V
dc
S
1
S
2
C
dc
V
dc
S
3
S
4
S
5
S
6
S
7
S
8
n
A
Figure 2.5. Phase leg of a cascaded fivelevel fullbridge inverter.
One major advantage of this hybrid approach is that the number of output can be further
increased without addition of any new components, requiring only the dc sources with different
voltage levels [D9]. Probably the most advantageous uses the dcsources with two different
voltage levels, V
dc
and 2
.
V
dc
, as shown in Figure 2.6.
17
C
dc
2
.
V
dc
S
1
S
2
C
dc
V
dc
S
3
S
4
S
5
S
6
S
7
S
8
n
A
Figure 2.6. Phase leg of a cascaded full bridge sevenlevel converter.
This approach can then be implemented using hybrid device technology [D9]. In that
arrangement, the slower, highervoltage devices can be used to change the output voltage level
(S
1
, S
2
, S
3
, S
4
), while the faster, lowervoltage devices can provide the full PWM capability
between levels (S
5
, S
6
, S
7
, S
8
). This arrangement can generate sevenlevel (0,+/V
dc
,+/2
.
V
dc
,+/
3
.
V
dc
) fully pulse width modulated output voltage.
2.1.4 Additional FullBridge Topologies
Compared to cascaded fullbridge inverters, some savings on isolation of dc power supplies
can potentially be achieved using fullbridge multilevel topologies. GE already successfully
applied this approach in their mediumvoltage (4160V) drive product. They relied on well
understood threelevel NPC topology to build the fullbridge fivelevel phase leg. This topology
has enough redundant switching states to allow the charge balancing of the dclink capacitors on
a switching cycle level for all line/load conditions. However the difficulties with the clamping
diodes still remain, as does the need for isolated dcvoltage source for every phase.
18
C
dc
p
n
C
dc
NP
V
pn
D
1
D
2
S
1
S
2
S
3
S
4
D
1
D
2
S
1
S
2
S
3
S
4
Figure 2.7. Diode clamped full bridge fivelevel topology.
Using the same philosophy, a fullbridge fivelevel phase leg can be built, based on the flying
capacitor topology, as shown in Figure 2.8. Very similar to the flying capacitor halfbridge
topology, this approach alleviates the clamping diode issues, but introduces the issues of
clamping capacitors instead. Both these multilevel design approaches have yet to be proven in
reallife applications.
C
dc
p
n
V
pn
S
1
S
2
S
3
S
4
S
1
S
2
S
3
S
4
C
dc
Figure 2.8. Flying capacitor fullbridge fivelevel topology.
19
2.2 Multilevel Digital Control Hardware
The complexities of the multilevel converters’ systems, together with the increase in power,
performance and availability (as well as the decrease in price) of digital control hardware, result
in almost exclusive application of digital control hardware for the majority of modern multilevel
converters and converter systems. In addition, the versatility of modern microprocessors allows
them to be used for the control of any type of converter, and allows implementation of any type
of modulation algorithm. At the same time, the everincreasing processing power of
microprocessors allows for much more userfriendly development tools that can accept input
from higher programming languages such as C and (increasingly) Matlab.
2.2.1 Overview of the Architecture of Digital Control Hardware
Regardless of the topological differences, the direction of power flow, or its application in
almost every power electronics system, has few main components, such as the switching network
with filter components on both input and output, the sensors subsystem, the power terminals and
a (digital) control subsystem.
Similarly, the architecture of modern digital controllers can be subdivided into several major
functional blocks from Figure 2.10. The central element is the processor (or a combination of
several processors) usually accompanied by some kind of programmable logic devices to
implement the PWM (if this capability is not already provided on the processor). In addition,
programmable logic devices provide the simplest way to implement some specific protection
functions that require very fast response speed as well as any “glue” logic required to interface
the processor with other components on the control board.
20
SENSORS
GATE DRIVES
SWITCHING
NETWORK
FILTER
DIGITAL CONTROLLER
FILTER
SENSORS
Figure 2.9. Power electronics system overview.
PROCESSOR
COUNTERS AND OTHER AUXILIARY
DIGITAL LOGIC
A/D CONVERTERS
SENSORS
GATE DRIVES
DIGITAL COMMUNICATION SUBSYSTEM
...
Figure 2.10. Architecture of the digital controller for power electronics systems.
21
This traditional structure of the controller is being challenged as part of the development of
the power electronic building block concept and the introduction of the digital communication
subsystem. Such a subsystem (although still in the development phase, and with uncertain
future), has the potential to standardize the signal distribution network across the whole range of
power electronics products from different manufacturers. Although different implementations
have been proposed the main idea is to establish a standardized communication protocol, most
likely through a serial fiber optic link [F17, F18]. If accepted the standardized protocol would
allow any processor to control and communicate with all elements of the power electronic
system, regardless of their manufacturer, as long as they support the same protocol. The
elements considered for communication upgrade range from single switches or phase legs with
integrated sensors, remote sensors, or the whole converters. In addition, such a system would be
easy to service/upgrade by adding or replacing various modules.
2.2.2 Components of the Digital Controller for SMES Power Conditioning System
The digital controller hardware partially developed for the control of the SMES PCS consists
of the digital signal processor, boot memory, erasable programmable logic device (EPLD),
analogtodigital (A/D) converter, and digitaltoanalog (D/A) converter. Since most of the
feedback transducers generate analog signals, A/D converters are required to convert those
signals to the digital format, which is suitable for the processor. The EPLD can be programmed
to perform any digital operation. This allows a part of the control logic to be implemented in
hardware. The D/A converters are very useful for system debugging and measurement of the
variables internal to the processor. The design and choice of components for the SMES digital
controller are listed as follows.
Processor
The present system is based on the Analog Devices ADSP 21062 (SHARC) processor. It is a
32bit floating point processor with a throughput of 40 MIPS. It has a 1M bit of onchip
program memory and a 1M bit of on chip data memory. This powerful processor is a perforce
22
requirement for power electronics control systems that execute sophisticated algorithms at high
sampling rates.
Programmable Logic Device
The EPLD can be electrically programmed to implement any logic functions. The most
important selection criteria for the EPLD are the speed, number of flipflops and usable gates as
well as the number of I/O pins that can be configured. This limits the maximum size of the logic
circuitry that can be implemented with any given EPLD. At the time of the board development
two, FLEX 8820 chips from ALTERA looked like a very good choice. Each chip has 8000
usable gates, 820 flipflops and 152 pins that can be user defined as input, output or input/output.
Analog to Digital Converter
The processor acquires the analog feedback signals from the plant through A/D converters.
The most important selection criteria for the A/D converters are the conversion time, and the
resolution. In the present design, the AD876, 10bit 20 MSPS A/D converters from Analog
Devices were used.
Digital to Analog Converter
In PWMbased systems, D/A converters are mostly used for debugging and transfer function
measurement, so their performance is not critical. We have chosen Quad, 12bit DAC8412
converters from Analog Devices, with 180ns write cycle time.
2.2.3 Controller Architecture
Figure 2.11 shows the block diagram of the DSPbased controller. It can be divided into three
functional subsystems, implemented on separate printed circuit boards:
23
• DSP subsystem
• Digital interface subsystem with flex EPLDs
• Analog interface subsystem with A/D and D/A converters
DSP
(ADSP21062)
DIGITAL INTERFACE BOARD
( 2 FLEX 8820 )
ANALOG INTERFACE
BOARD
DIGITAL BUFFER
BOARD
SENSORS GATE DRIVES
Figure 2.11. Architecture of the SMES digital controller board.
DSP Processor Board
The ADSP21062 EZLAB from Analog Devices, used in this project is a complete
development system based on the ADSP 21062 (SHARC) processor. It allows the use of EZICE
InCircuit Emulator and facilitates communication with additional modules through the
SHARCPAC expansion connectors, that let the user add additional memory and/or SHARC
processors. In the present design the SHARCPAC expansion connectors are used to interface
with digital interface board.
24
Digital Interface Board
The digital interface board from Figure 2.12 consists of two backtoback connected EPLDs.
These EPLDs share the address/data and control lines with the processor. The EPLDs can be
programmed using a serial cable from a personal computer (PC) or using the programmable read
only memories (PROMs). This board performs the majority of the PWM generation and system
protection tasks such as shutdown at repeated fault signal from gate drives, watch dog timer
function and shoot through protection.
CLK1
CLK2
CLK3
DIGITAL BUFFER BOARD
A N A L O G I N T E R F A C E B O A R D
E P L D
E P L D
D
S
P
B
O
A
R
D
ADDRESS
DATA
CONTROL
PC
P
R
O
M
P
R
O
M
SERIAL SERIAL 50PINCONNECTOR
60 PIN CONNECTOR
Figure 2.12. Digital interface board.
Analog Interface Board
Figure 2.13 shows the functional block diagram of the analog interface board. Sixteen analog
input channels are being levelshifted, inverted and filtered, and then fed to two analog
multiplexers (both 2 x 4to1). The outputs of the multiplexers are routed to four A/D
converters, which are all connected to the digital interface board via common data lines. The
main consideration in the choice of architecture was not so much the throughput as it was the
25
nearly instantaneous sampling, which can not be achieved using a single converter with
multiplexed input (the limitation primarily lies in the speed of the multiplexers). The analog
interface board also has 8 analog outputs, all of which are buffered using voltage followers, and
two of which drive the current transmitters that can be used to feed long cables.
SENSORS
(Voltage/Current)
FILTERING &
LEVEL SHIFTING
ANALOG
MULTIPLEXER
A/D
TRISTATE BUFFER
DIGITAL INTERFACE BOARD
D/A
B U F F E R S
VOLTAGE/CURRENT
DATA DATA CONTROL CONTROL
CLK
Figure 2.13. Analog interface board.
2.2.4 Control Algorithm
The primary tasks of any controller are regulation and pulse width modulation. The block
diagram of the control structure is shown in Figure 2.14. The process begins with the sensing of
output voltages and line currents (INPUT). These are then transformed from stationary three
phase (abc) to rotating threephase (dqo) coordinates and compared with reference values to
generate errors. The error signals are processed in a compensator and the result is transformed
back to abc coordinates (T1). Then the modulator takes over and selects the switching states and
duty cycles that implement the commanded reference vector from the compensators. Up to this
point the complete algorithm is implemented in software.
26
+

Compensator
with fault
protection
Space Vector Modulation
Selection of switching states
Calculation of dutycycles
Output Pwm Signals
INPUT
in abc
Reference
T
T T1
Frequency Reference
Figure 2.14. Control overview.
Finally the PWM generation is implemented in hardware requires many timers to keep track
of the time elapsed (duty cycles) and some combinatorial logic to change between the switching
states.
2.2.5 Functions of the EPLD
The main function of the EPLD is to generate the PWM signals that are fed into the inverter
gate drives, based on the data received from the DSP. A functional block diagram of the EPLD
logic responsible for generating the PWM signals is shown in Figure 2.15.
The DSP receives periodic interrupt requests (IRQ) from the EPLD at switching frequency (1 T
S
),
which is equal to the control sampling frequency. After receiving an IRQ, the processor reads the
feedback variables from the A/D converters and calculates the switching states and their duty
cycles according to the algorithm described in the previous section. These are then loaded into
the doublebuffered register bank in the EPLD to be used in the following switching cycle. The
processor then enters the idle state until receiving new IRQ.
27
T
0
T
1
T2
T3
T4
T5
TS
Digital
Comparator
Logic
State1
State2
State3
State4
State5
State6
State0
MUX
Timer
CLR CLK
Dead
Time
Fault
Prot.
B
u
s
I
n
t
e
r
f
a
c
e
L
o
g
i
c
CLKx
CLK
T
o
D
S
P
T
o
G
a
t
e
D
r
i
v
e
s
PWM
FAULT
IRQ
Timer=T
S
DOUBLE
BUFFERED
REGISTER
BANK
Figure 2.15. PWM signal generation logic within the EPLD.
C L K CLK
BUFFER
1
BUFFER
2
DSP Write IRQ
DATA
Figure 2.16. A doublebuffered register bank inside the EPLD.
The doublebuffered register bank consists of timer registers and state registers, the numbers
of which vary for different applications. Each register is doublebuffered, as shown in Figure
2.16, so that the PWM data used in the current switching cycle (BUFFER 2) is not affected by
28
the asynchronous DSP write into BUFFER 1. The fresh data is loaded into BUFFER 2 by an IRQ
signal at the beginning of each switching cycle.
The state registers contain those switchingstate codes which are used for PWM in the current
switching cycle, ordered in the sequence in which they will be applied. The timer registers
contain the dutycycle information, i.e., the times from the beginning of the switching cycle
(measured in the units of EPLD CLK period) when the corresponding switching state ends.
Generation of PWM Signals
The timer in Figure 2.15 is a freerunning counter with the same length as the timer registers
in the register bank. It can be fed by any of the three clock signals available on the digital
interface board (CLK1, CLK2, CLK3 in Figure 2.12). The counter is cleared by the IRQ signal at
the beginning of each switching cycle. The Digital Comparator Logic and MUX in Figure 2.15
sequentially output the switching states using the current data in the doublebuffered register
bank. After the timer reaches T
S
, digital comparator logic generates the IRQ signal which starts a
new switching cycle by interrupting the processor, loading the new data into the doublebuffered
register bank, and clearing the timer counter.
DeadTime Protection
Deadtime protection is provided to avoid any shoot through of the inverter legs. This is done
by delaying the rising edges of all the signals by a particular number of clock cycles, as shown in
Figure 2.17.
29
Figure 2.17. Inverted gate drive signals with deadtime.
Fault Protection
In the event of a fault, the PWM signals are shut down to save the power stage. Two kinds of
fault protection have been provided: internal and external. If for some reason the PWM signals
received from the deadtime block are such that they turn on the top and bottom switches of a
given phase leg, then a fault signal is generated inside the EPLD (internal fault) and the PWM
pulses are shut down. The internal fault can also be generated by the watchdog timer. These
conditions should not occur under normal operating conditions. They are a protection against
software failure.
External fault protection is activated when a gate driver senses an overcurrent through the
power device and sends a fault signal to the digital buffer board. This fault signal is passed to the
EPLD where it enables an IRQ counter. If the fault persists for a particular number of switching
cycles (this number can be programmed into the IRQ counter), the PWM signals are shut down.
In this way, the complete inverter is not shut down if the overcurrent condition happens in only
30
a couple switching cycles, which is acceptable because the gate drives automatically turn off the
switch whenever overcurrent is detected.
Other EPLD Functions
The analog interface board has four A/D converters and two D/A converters that all
communicate through the same 12bit data bus. In addition, the processor and A/D converters
have different frequency clocks that are not synchronized. Therefore, to insure proper operation
of the board, the timing of the control signals is critical. The logic that receives signals from the
analog interface board and takes care of synchronization is implemented inside the EPLD. In
addition, the system watchdog timer and the address decoding of the input and output ports are
also implemented in the EPLD.
2.3 Multilevel Power Conditioning System as a Power Controller for a
Superconducting Magnetic Energy Storage System
Historically, most of the interest in SMES systems seems to have been shown for utility type
applications. The main motive was to combine the flexibility of an energy storage system with
the ease of control provided by power electronics. In fact, by providing a buffer between the
power generation and load consumption, the ease of control and operational flexibility of power
systems could be greatly increased. There are certainly other utilitytype applications to which
SMES systems could be applied, such as active filtering, supplying pulse power load and UPS
applications.
The main challenge in developing such a PCS is to match the constant ac voltage and variable
current operation of the utility with the slowly varying current and varying voltage of the
magnet, while at the same time providing bidirectional power flow.
31
In addition, there is one rather unique characteristic of SMES systems: both dc voltage and dc
current vary over a large range. This feature implies that the PCS must be overrated relative to
the designed power transfer levels by a factor of three to four, or even more. This feature is one
of the main reasons why the current source inverter is not a preferred way to design a SMES PCS
while the series connection of a dc/dc chopper and ac/dc voltage source inverter (VSI) is.
Clearly the later approach can have significant cost advantage because the whole system does not
have to be overrated (only dc/dc converter).
2.3.1 Power Conditioning System Design approach
Consistent with the preceding discussion, the PCS that was studied in this dissertation is
implemented using a combined chopper/VSI approach. Considering the various tradeoffs
(already mentioned), the threelevel diodeclamped topology was chosen. Figure 2.18 shows the
topology of the prototype PCS. Some of the advanced features of the system include the zero
currenttransition (ZCT), soft switching technique, which allows the system to be rated at
250kVA, while switching with the 20kHz switching frequency. The details of the power stage
and the intricacies of the softswitching design and control are the part of the doctoral research of
Dr. DongHo Lee, and are described in detail in his work [G9].
32
S
c
4
S
b
1
S
b
3
S
b
4
S
b
2
S
a
3
S
a
4
C
d
c
C
d
c
b
c
R
L
C
a
c A
B
C
i
p
S
a
1
S
a
2 a
S
c
1
S
c
2
S
c
3
i
N
P
i
n
i
a
i
b
i
c
R R
C
a
c
C
a
c
p n
L L
N
P
L
S
c
h
4
S
c
h
1
S
c
h
2
S
c
h
3
L
s
c
C
Figure 2.18. Power conditioning system for SMES.
33
2.3.2 Modeling and Control of the Chopper
A functional diagram of the threelevel chopper is given in Figure 2.19. The task of the
chopper is to control the power flow to and from the superconductive coil (SC). The circuit is
bidirectional and, to avoid confusion, the SC side of the circuit is defined as the output side,
while the side interfacing with the inverter is defined as the input side. Current direction in the
SC is always the same so that the voltage on the filter capacitor, V
sc
, determines the power flow.
Fig. 2.20 shows the output voltages given in the form of space voltage vectors that are
produced with different switching combinations of the Chopper. Although the space vector
notation is not common for the dc/dc converters, such as in this case chopper, it is adopted here
because it is consistent with the space vector notation introduced for the VSI in Section 2.3.4
The voltage vectors in Fig. 2.20 are named according to the switching combinations used to
produce them. For example, if in Figure 2.19, switch S
x
is connected to the prail and switch S
y
is
connected to the NP, the output voltage vector is po. There are nine vectors in all: two large
vectors (pn, np) that produce output voltages of opposite polarity with the magnitude equal to
the dclink voltage, V
pn
; four small vectors (po, on, no, op) with magnitude equal to V
pn
/2; and
three zerovoltage vectors (pp, oo, nn).
The task of the chopper modulator is to synthesize the voltage reference vector, V
xyREF
, which
is a dc voltage and is always located on the V
DC
axis shown in Fig. 2.20. The modulation scheme
in which the reference vector V
xyREF
is synthesized from the two closest vectors is adopted and
will be discussed here. The advantage of this scheme is that it minimizes the voltage ripple on
the SC and allows for effective control of the NP voltage balance.
34
p
i
SC
V
SC
i
o
o
n
L
F
C
F
i
LF
L
SC
C
DC
C
DC
s
x
s
y
NP
y
x
Figure 2.19. Equivalent switching model of threelevel chopper.
pn po
on
no
op
np
Zero vectors
pp
oo
nn
V
xyREF V
DC
Fig. 2.20. Space voltage vectors of the chopper.
35
If we define the duty cycle as
pn xyREF
V V d · , where [ ] 1 , 0 d ∈ , then for [ ] 5 . 0 , 0 d ∈ , the duty
cycles of large, small and zero vectors are given by
(21) 0 d
L
· ,
(22) d 2 d
S
⋅ · and
(23) d 2 1 d
0
⋅ − · ,
respectively, where small vectors po or on should be used for V
xyREF
> 0, and no or op for
V
xyREF
< 0. For ] 1 , 5 . 0 ( d ∈ duty cycles are given by
(24) 1 d 2 d
L
− · ,
(25) d 2 2 d
S
− · and
(26) 0 d
0
· ,
where vectors pn, po and on are used for V
xyREF
> 0, while np, no and op are used for V
xyREF
< 0.
Large vectors and zero vectors do not produce any NP current, i
NP
, and hence do not affect the
NP voltage. Small vectors come in pairs. Both vectors in a pair generate the same output voltage,
but they produce the current i
NP
in opposite directions, as shown inTable 2I. The vectors
resulting in i
NP
= i
LF
are referred to as positive small vectors, while the vectors resulting in i
NP
= 
i
LF
are called negative small vectors. The task of the NP voltage controller is to select from the
redundant small vectors in
, with the objective of minimizing the NP current and maintaining the NP voltage as close as
possible to onehalf of the dclink voltage.
36
Table 2I Neutralpoint current i
NP
for different small vectors.
V
xy
Positive
Small
Vectors
i
o
Negative
Small
Vectors
i
o
V
np
/2 po i
LF
on i
LF
V
np
/2 no i
LF
op i
LF
Neutralpoint control issues notwithstanding, the average model of the rectifier is not different
from the average model of a bidirectional twolevel converter from Figure 2.21. The
modulation index is defined as
pn xyREF
V V m · . Clearly, the duty cycle is then given by d = m
and the switch duty cycles can be easily calculated from (21)(26).
2
C
DC
L
F
i
SC C
F
pn
V m
×
LF
i m
×
i
LF
V
SC
V
pn
Figure 2.21. Average model of the chopper.
The average model is similar for both modes of operation; there is only a slight difference in
the control design. In the charging mode, the chopper operates as a buck dc/dc converter with
the inner current loop and the outer voltage loop, Figure 2.22. In the discharging mode, the
chopper operates as a boost dc/dc converter, Figure 2.23.
37
L
F
C
F
i
LF
V
SC
H
i
H
v
m
power
controller
Z
Cin
i
SC
SC
Figure 2.22. Buck mode control of the chopper (charging the SC).
2
C
DC
L
F
i
SC
C
F
i
LF
H
i
m
H
v
Z
C o u t
V
pn
V
DC_REF
S C
Figure 2.23. Boost mode control of the chopper (discharging the SC).
38
2.3.3 Modeling of a ThreeLevel Voltage Source Inverter
A functional diagram of a threelevel VSI switching network is given in Figure 2.24. The
subsequent analysis will reveal a few important properties of this type of converter. It will show
that the error in the NP, ∆v
pn
, does not dynamically influence the output voltages V
ab
, V
bc
, V
ca
,
but rather in a static way determined by the type of modulation and the NPbalancing method
applied. This result justifies application of the static method for the study of the NPbalancing
problem, such as the one described in [B20]. Additionally, it will be shown that the error in
NP, ∆v
pn
, depends only on the NP current i
o
and is not directly influenced by any other circuit
parameters.
L
a
L
b
L
c
p
n
o
i
o
C
DC
s
a
s
c
s
b
C
DC
i
a
i
b
i
c
i
DC
+
+
v
p
v
n


v
A
v
B
v
C
i
p
i
n
a
b
c
i
DC
NP
Figure 2.24. Switching network of a threelevel VSI
Equation (27) shows the definition of the switching functions that are used in (28) to write
the switching network voltages referenced to the NP.
(27)
¦ {
¦ {
¹
'
¹
· ·
∑
∈
∈
otherwise 0
j to connected is i if 1
s where , 1 s
j , i
n , o , p j
c , b , a i
j , i
39
(28)
]
]
]
⋅
]
]
]
]
]
·
]
]
]
]
]
n
p
cn cp
bn bp
an ap
co
bo
ao
v
v
s s
s s
s s
v
v
v
From the definitions in (27) and (28) and by inspection of the switching network in Figure
2.24, linetoline voltages can be written (29) as follows:
(29)
... s s s s s s : where
v
v
s s
s s
s s
v
v
v
bn an abn bp ap abp
n
p
can cap
bcn bcp
abn abp
ca
bc
ab
− · − ·
]
]
]
⋅
]
]
]
]
]
·
]
]
]
]
]
·
sn
v
By rewriting the voltages of the top and bottom dclink capacitor in terms of the full dc link
voltage and the NP error voltage (210), additional insight can be gained into the effect of the NP
voltage error, ∆v
pn
.
(210)
pn
pn
p
v
2
v
v ∆ + · ,
pn
pn
n
v
2
v
v ∆ + − ·
By substituting (210) into (29), as in (211), it becomes clear that the linetoline voltage
depends on the DClink voltage and the error of the NP, and there is no dynamic coupling
between the two.
(211)
pn
pn
pn
can cap
bcn bcp
abn abp
pn
can cap
bcn bcp
abn abp
v
2
v
v
s s
s s
s s
2
v
s s
s s
s s
∆
∆
⋅ + ⋅ ·
· ⋅
]
]
]
]
]
+
+
+
+ ⋅
]
]
]
]
]
−
−
−
·
np
sn
S S
v
40
Defining the current vector as in (211), (212) and (213) can be written. Note that the vector
i
sn
influences the NP current i
o
through the same matrix of switching functions as the error of the
NP voltage, and that there is no dynamic relationship between these two either.
(212)
]
]
]
]
]
−
−
−
⋅ ·
a c
c b
b a
i i
i i
i i
3
1
sn
i
(213)
sn
i ⋅
]
]
]
·
]
]
]
can bcn abn
cap bcp abp
n
p
s s s
s s s
i
i
(214) [ ]
sn np
i S ⋅ · + ·
T
n p o
i i i
At last we can derive equation, (215), which describes the dynamic relationship between the
NP current i
o
and the voltage error of the NP, ∆v
pn
. This equation completely describes the
dynamics of the voltage error in the NP and will be included in the complete statespace model
of the switching network.
(215)
( )
pn DC
pn
pn
DC pn
pn
DC
n
DC DC
p
DC DC o
v
dt
d
C 2
v
2
v
dt
d
C v
2
v
dt
d
C
dt
dv
C i
dt
dv
C i i
∆
∆ ∆
⋅ ⋅ ·
·
,
`
.

+ − ⋅ +
,
`
.

+ ⋅ ·
·
,
`
.

⋅ + − +
,
`
.

⋅ + ·
Finally, from Kirchof’s current and voltage laws and the definitions given, a dynamic
relationship between input and output quantities of the threelevel switching network can be
obtained (216), (217). Combining these equations with the result of the derivation in (215), a
full set of dynamic equations describing the threelevel switching network is developed. This
model is general, complete and makes no assumptions.
41
(216)
( ) [ ]
[ ]
pn
T
pn T
v
L 3
1
2
v
L 3
1
L 3
1
dt
d
∆ ⋅ ⋅
⋅
+
+ ⋅ ⋅
⋅
− ⋅
⋅
·
np
u sn
S
S v i
(217) ( ) [ ]
DC
DC
T
DC
pn
i
C
2
C
1
v
dt
d
⋅ − ⋅ ·
sn
i S
(218) ( ) [ ]
sn np
i S ⋅ ⋅
⋅
·
T
DC
pn
C 2
1
v
dt
d
∆
Different switching matrices S and S
np
correspond to each position of the switches s
a
, s
b
and
s
c
. These allowable combinations of the switches are called voltage space vectors. The switching
matrices are changed as the modulator selects new voltage space vectors. Averaging the
switching matrices using the averaging operator (219), the small signal model is obtained.
(219)
∫
−
⋅ ⋅ ·
s
s
T
T t s
d ) ( z
T
1
) t ( z τ τ
Under the assumption that ∆v
pn
is small, Equation (218) can be neglected as, can the last term
in the equation (216). What remains is the wellknown, average statespace model of a two
level inverter.
2.3.4 Control of the Inverter in Rotating Coordinates
In order to facilitate control design, threephase variables can be transformed into a rotating
dq reference frame. By transforming Equations (210) and (211), a new model in rotating
refernce frame is obtained, Figure 2.25. The control inputs into the system are d
d
and d
q
. They
are computed by the d and q channel controller.
42
2
C
DC
d d i d
2
3
× ×
q q i d
2
3
× ×
pn d V d
×
pn q V d ×
L 3×
L 3×
R 3×
R 3×
q i L
3 × w × ×
d i L
3 × w × ×
D V
0
d i
q i
pn V
Figure 2.25. Inverter average model in rotating dq coordinate frame.
The daxis of the rotating reference frame is aligned with the utility line voltage space vector,
V
ABC
, as shown in Figure 2.26. The control inputs are the linetoline duty cycles, d
d
and d
q
,
transformed into the synchronous frame. Their relationship to the voltage reference vector in
Figure 2.28 is
(220) θ θ sin
V
d , cos
V
d
pn
q
pn
d
⋅ · ⋅ ·
REF REF
V V
.
The control inputs are obtained from a closed loopcontroller, and the reference voltage vector
is then calculated from (220). Finally, the modulator generates the corresponding duty cycles.
During the period of SC’s discharging, the inverter operates in the utility inverter mode. The
task of the inverter is to control the magnitude and phase angle of the threephase output
currents, thereby controlling the amount of active and reactive power supplied to the utility, as
shown in Figure 2.26. The design of the current controllers can follow a standard procedure for
VSIs; conventional PI regulators were used in the implementation described here.
43
L 3⋅
L 3 ⋅
R 3
⋅
R 3
⋅
q
i L
3 ⋅
w
⋅ ⋅
d
i L
3
⋅
w
⋅ ⋅
D
V
0
d
i
q
i
power
controller
dref i
qref i
d
d
q
d
H
id
H
iq
Ιιν
Z
Figure 2.26. Inverter control in the utility mode (charging the SC).
2
C
DC
L 3
×
L 3
×
R 3
×
R 3
×
q
L
3
× w × ×
d
i L
3
× w × ×
D
V
0
d
i
q
i
power
controller
dref
i
qref
i
d d q d
H
id
H
iq
Iout
Z
V
∆Χ_ΡΕΦ
H
DC
ς
πν
ι
Figure 2.27. Inverter control in rectifier mode (discharging the SC).
44
In the SC charging mode, the inverter is operates as a threephase boost rectifier with the
objective of controlling the voltage on the dclink, as shown in Figure 2.27. The active power
flow is controlled by the chopper, while the inverter still independently controls the reactive ac
power depending on the utility requirements.
The outputs of the controllers can not be implemented directly by the power stage. Instead,
the voltage reference vector V
REF
(as computed by the controller in the rotating reference frame)
needs to be converted back to the stationary reference frame and implemented by the modulator.
The inverter output linetoline voltages, V
ab
, V
bc
and V
ca
(produced by available switching
combinations) can be represented as space voltage vectors in one plane, as shown in Figure 2.26.
The vector names represent the allowable states of the switches; e.g., pon represents the state in
which switch S
a
is connected to the positive dc rail, S
b
to the NP, and S
c
to the negative dc rail. .
There are 27 switchingstate vectors: six large vectors with the magnitudes equal to 3 / 2
pn
V ⋅ ;
six medium vectors with magnitudes equal to V
pn
; twelve small vectors with magnitudes equal to
3 /
pn
V , and three zerovoltage vectors.
The task of the VSI modulator is to synthesize the reference line voltage space vector, V
REF
,
by PWM of the available voltage space vectors. The space vector modulation (SVM)
implemented in this system switches the three vectors nearest to the voltage reference vector.
The duty cycles of the vectors are found as a vector decomposition of the reference vector to the
nearest three vectors. Details of the modulation process will be studied in Chapter 3.
Large vectors and zero vectors do not produce any NP current, i
NP
and hence do not affect the
NP voltage. Medium and small vectors affect the NP balance, but only the small vectors come in
pairs. Both vectors in a pair produce the same linetoline voltage, but produce the NP current in
opposite directions. Consequently, the effect of the small vectors alone could be balanced in a
way similar to that used in the chopper case, but the effect of the middle vectors cannot be
directly controlled. This may produce a lowfrequency voltage ripple of the NP, the extent of
which depends on the magnitude of the voltage reference vector, the magnitude and phase angle
of the output phase current, the NP control method, and the size of the dclink capacitors. This
45
issue will be addressed in Chapter 4, where it will be shown that the worst case scenario is when
the VSI supplies a purely reactive power at the full current capacity. In such a case, large a NP
ripple can be expected and active compensation with the chopper may be essential.
V
REF
V
ABC
w t
δ
θ q
pnn
poo
onn
pon
ppn
npp
npo
npn opn
nop
nnp onp pnp
pno
ppo
oon
opo
non
opp
noo
oop
nno
pop
ono
Zero vectors:
ppp
ooo
nnn
V
ab
V
ca
V
bc
Figure 2.28. Voltage space vectors of a threelevel inverter.
46
2.3.5 Simulation and Experimental Results
The average model simulation results were validated on the experimental setup shown in
Figure 2.29. The system in the picture is rated at 250kW, with a dclink voltage of 1800V, the
dclink current of 150A, and operates with 20kHz switching frequency. Simulations and the
experiments were performed with the inverter operating alone in the rectifier mode at 28kW
power level. The threephase utility line voltage was 208V with the 77A phase rms current. This
produced 600V on the dc link that was loaded with 13Ω resistive load.
CHOPPER
INVERTER
Figure 2.29. Experimental SMES PCS system.
The simulated and measured current loop gain in the d channel is given in Figure 2.30. The
PI regulators H
id
and H
aq
were designed using the model in Figure 2.27 for the current loopgain
crossover frequency of 1kHz. The results in Figure 2.30 show good agreement between the
design predictions, simulations and experiments. The achieved wide currentloop bandwidth
confirms the feasibility of using the system for power quality applications, including the active
filtering of loads with high harmonic contents.
47
20
0
20
40
[dB]
10 100 1 10
3
1 10
180
60
60
180
[Hz]
[
d
e
g
]
Figure 2.30. Simulated and measured rectifier current loop gain.
The voltage regulator, H
DC
, was designed for the loop gain crossover at 20Hz. The simulated
and experimental measurements of the voltage loopgain are shown in Figure 2.31. Again, the
results demonstrate the good agreement between simulations and measurement.
3 4
60
40
20
0
20
[dB]
10 100 1 10
3
1 10
4
180
60
60
180
[Hz]
[
d
e
g
]
Figure 2.31. Simulated and measured rectifier voltage loopgain.
48
The response of the rectifier to the step change in current command with the voltage loop
disabled is shown in Figure 2.32; this shows the performance of the rectifier inner current loop.
1
2
3
Figure 2.32. Rectifier response to change in current command: 1) current command (20ms/div);
2) dchannel current, i
d
; and 3) rectifier phase current (100 A/div).
49
2.4 Modeling and Control Approach and Problem Formulation
The principle research and development effort in this dissertation was to better understand
how to control and model three and more level, threephase converters for use in SMES power
conditioning systems as well as in other applications. The approach was really bottom up.
Important modeling and control concepts were first understood, then analyzed, and finally
generalized when possible. Only than was the understanding of the fundamental concepts
applied in the design of the system as a whole.
It was already mentioned that one of the important problems of multilevel converters is the
modulator design (implementation). The SVM approach is perhaps the most powerful, because
it allows more freedom to control and optimize the switching patterns than any other modulation
approach; at the same time, for converters with higher number of levels it becomes too
cumbersome for realtime implementation. Therefore, the first big task in this work was to
simplify and generalize the implementation of a SVM for converters with any number of levels.
This goal was successfully achieved through the geometric analysis of the space vector
diagrams, and through implementation of linear transformations that transform the switching
vectors from a space where the algorithm is difficult to implement to a space where
implementation of the algorithm becomes almost trivial.
The second problem that was analyzed is perhaps less general, in the sense that it directly
applies to the problem of charge balancing in the dclink capacitors of a threelevel NPC
converter. Still, it is currently one of the most popular topologies and was used for the prototype
PCS. The idea behind the research was to try to find a mathematical model that could describe
the effect of the load on the charge balance as well as to find guidelines for sizing the dclink
capacitors in order to achieve the optimal voltage ripple amplitude for any operating condition.
The modeling approach was geared towards the SVM and is general in the sense that it can
model almost any type of SVM. In addition, it offers some intuitive insight into the operation
50
and the limitations of the chargebalancing algorithms. Finally the theoretical results were
verified on the prototype PCS.
Further thinking along the lines of the charge ripple leads almost naturally to the following
question: If there are certain operating conditions that cause a significant charge ripple in the dc
link capacitors, is there a way to implement a controller such that it can cancel the effect of that
ripple in the output waveform, and therefore allow savings in the size of the dclink capacitors?
The answer to the above question is positive. By applying a feedforward controller
(proposed in this work), it is possible to compensate for any size of the imbalance in the dclink
capacitors. The properties of the feedforward algorithm were studied on two different time
scales. The initial study included the development line cycle averaging model and its
linearization for the purpose of feedback. Than a switching cycle average model was built and
simulated in the computer in order to study the system behavior in more detail. Some of the
results were verified through the experiment.
2.5 Conclusions
This is an overview chapter that covers a wide range of issues important in the design of a
generic threephase, multilevel power electronics system. The issues range from the selection of
the topology and the requirements of the digital control hardware all the way to an overview of
the PCS for the SMES application and the formulation of the research problems studied in this
dissertation. At the same time, this chapter is an overview of the issues that had to be resolved
during the course of the design and implementation of the prototype SMES PCS.
51
3 A FAST SPACE VECTOR MODULATION ALGORITHM FOR MULTILEVEL
CONVERTERS
With the increase of the number of levels, the number of allowable switching states in the
converters is also rapidly increasing. This situation places significant computational difficulties
in front of the space vector PWM modulator, which is typically implemented in software and
operating in real time and for high switching speeds. Figure 3.1 shows the structure of a generic
threephase converter digital control system.
EPLD
NEAREST
THREE
VECTORS
DETECTION
MODULATOR SOFTWARE
DSP
SENSORS
POWER
STAGE
REGULATOR
DUTY CYCLE
COMPUTATION
VECTOR
SELECTION
DIGITAL CONTROLLER
REF V
→
Figure 3.1. Architecture of digital controller for threephase converters.
52
3.1 Switching Vectors, Switching States and Reference Vector
In essence, a multilevel converter is a voltage synthesizer that generates its output voltage
from many discrete voltage levels. Therefore, without a loss of generality, most recently studied
multilevel topologies can be well represented (for the purpose studying SVM), by the functional
diagram in Figure 3.2.
.
.
.
.
...
0
1
n2
n1
a
b
c
a
b
c
1
.
.
.
0
n1
n2
.
.
1
.
.
.
0
n1
n2
.
.
1
.
.
.
0
n1
n2
.
.
+
+
+
+
V
dc
V
dc
V
dc
V
dc
Figure 3.2. Functional diagram of a threeleg nlevel converter.
Because the functionality of each phase leg can be represented by a singlepole nthrow
switch, it becomes clear that each switching state (combination of phaseleg switches) produces
uniquely defined threephase line voltages (31), which can be represented as vectors in a three
dimensional Euclidean vector space
1
:
(31) [ ]
T
ca bc ab
v v v V ·
→
.
1
Threedimensional, Euclidean coordinate space is chosen to present vectors, primarily for the elegance of mathematical description, even
though it might not be the most widely accepted way of representing vectors in threephase systems.
53
For example, if the location of the equivalent phase switches of the phases a, b and c were in
the positions i, j and k, respectively, where [ ] 1 , 0 , , − ∈ n k j i , then that switching state could be
represented by the switching vector
(32) [ ]
T
dc
ijk i k k j j i V V − − − ⋅ ·
→
) ( .
It is somewhat important to make the distinction between switching states and switching
vectors, in the sense that different switching vectors can be implemented with a various number
of switching states. There are
3
states
n N · switching states, which make
∑
−
·
+ ·
1
1
6 1
n
i
vectors
i N
switching vectors. Clearly the number of allowed combinations increases rapidly as the number
of levels increases.
On the other hand, the existence of various switching states of some of the switching vectors
has some merit. Having a variety of switching states from which to choose provides the
important degree of freedom to balance the charge in the split dclink capacitors and/or optimize
the switching pattern. Shown in Figure 3.3 are all the switching vectors of a threelevel
converter, with their corresponding switching states. The reason to naming the vectors with their
corresponding switching states is purely practical, because the converter switches are controlled
on a perphase basis, and a decision must be as to which vector to switch, as well as which
switching state to select. On the other hand, representing vectors in the line voltage coordinate
space is a nice mathematical abstraction to consistently explain the spatial relations between the
switching vectors.
Finally, because of Kirchof’s voltage law, the sum of the linetoline voltages always adds up
to zero, which is really an equation of the plane in the linetoline coordinate system. This
means that all the switching vectors of a threelevel converter lie in the plane, and that is how
they are usually represented.
54
2
1
0
1
2
2
1
0
1
2
2
 1 . 5
1
 0 . 5
0
0 . 5
1
1 . 5
2
v
a
b
/V
d
c
v
b
c
/
V
d
c
v
c
a
/
V
d
c
v
ab
v
ca
v
bc
200 220
020
022 002
202
210
120
021
012
102
201
121
010
101
212
122
011
112
001
211
100
221
110
REF
V
→
Figure 3.3. Switching states of a threelevel converter.
Coming back to the same threedimensional representation, a voltage reference vector that is
to be synthesized with the help of switching vectors can also be represented in vector form, as
follows:
(33)
]
]
]
]
]
+ ⋅ − ⋅
+ ⋅ − ⋅
+ ⋅
⋅ ·
−
→
) 3 / 4 cos(
) 3 / 2 cos(
) cos(
ϕ π ω
ϕ π ω
ϕ ω
t
t
t
V V
l l
REF ,
and also has only two degrees of freedom. Using the definition of vector norm, the length of the
reference vector is
(34)
2
3
V V V V
l l
REF REF REF ⋅ · ⋅ ·
−
→ → →
,
55
while by the same definition, the length of the longest switching vector is
(35)
dc
V n V ⋅ − ⋅ ·
→
) 1 ( 2 max .
Because the maximum length of the reference vector (33) that can be synthesized equals the
radius of the largest circle that can be inscribed in the outermost hexagon, the maximum length
of the reference vector for a general multilevel converter is
(36) ) 6 / cos( max
max
π ⋅ ·
→ →
V V REF .
By substituting (34) and (35) into (36), the maximum amplitude of the undistorted linetoline
voltage that a multilevel converter can synthesize is
(37)
dc l l
V n V ⋅ − ·
−
) 1 (
max
.
3.2 Space Vector Modulation of a ThreeLevel Converter
It is the task of the modulator to decide which position the switches should assume (switching
state), and the duration needed (duty cycle) in order to synthesize the reference voltage vector.
In other words, it is the task of the modulator to approximate the reference vector, computed by
the controller, using the PWM of several switching vectors. Arguably, the best way to
synthesize the voltage reference vector is by using the nearest three vectors
(38)
,
`
.

⋅ + ⋅ + ⋅ ·
→ → → →
3
3
2
2
1
1
V d V d V d V REF ,
with the following additional constraint on the duty cycles:
56
(39) 1
3 2 1
· + + d d d .
It was shown in the previous paragraph that the voltage space vectors of a threephase
converter are always located in the plane, and that is how they are represented in Figure 3.4. In
addition, although naming the switching states with the numbers is much more general (and
applicable for any number of levels converter), for the threelevel converter, equivalent
nomenclature in which 2, 1, 0 are named p, o, n is perhaps a little more common; the two will be
used interchangeably.
p n n
n p p
o p n
o n p
p o n
n o p
n n p p n p
p p n n p n
n p o p p o
o o n
p o o
o n n
o n o
p o p
o o p
n n o
o p p
n o o
o p o
n o n
p n o
REF
M
→
α
β
Figure 3.4. Voltage space vectors of a threelevel converter.
So far, it has been shown that all the switching state vectors and the reference vector can be
represented in a twodimensional space, and that the goal is to approximate the reference vector
using the nearest three vectors (NTVs) that satisfy (38) and (39). Clearly, there is a certain
degree of freedom in choosing the coordinate system in which to represent those vectors. This
decision will have important consequences for the reatime implementation of the modulation
algorithm. The two coordinate frames that first come to mind are the rectangular, often called
( β α, ) and the polar ( θ ρ, ).
57
Because the control loops are implemented in the rotating reference frame, and because the
conversion from rotating reference frame to the stationary polar reference frame requires
computation of the arctangent function to determine the angle θ of the reference vector, the
β α, reference frame is perhaps a better choice for the realtime implementation
1
, if the
modulator will operate within the feedback system.
The first step the modulator needs to perform is to identify the NTVs, which is really to
determine the small triangle in which the tip of the reference vector, REF V
→
, is located. The inner
triangle is always within a certain sixtydegree region. The small triangle sides are described
with simple linear equations in the β α, coordinate system. The small triangle (NTVs) are than
found by evaluating the sign of some linear expressions with reference vector coordinates
substituted. Finally, when the NTVs are located, then solving the system of three linear
equations with the three unknowns given in (38) and (39) results in the duty cycles.
However, because the derivation of the model describing the charge balance in the NP,
(which will be developed in the next chapter) turns out to be easier to develop in polar
coordinates, the SVM algorithm in polar coordinates will be presented in more detail. The
implementation of the algorithm in polar coordinates follows the same steps already outlined in
the previous paragraph.
The first step in detecting the NTVs is to find the sixtydegree region, which in polar
coordinates comes naturally, because the angle of the reference vector is known. Then the small
triangle has to be located, which requires evaluation of the sign of some nonlinear expressions.
For example, if the expression
(310) 0
) sin( ) cos( 3
1
m >
− ⋅
−
θ θ
58
is true (which is really the equation of the line in polar coordinates system that connects the
small and the medium vector from Figure 3.5), then the NTVs are located in the outer triangle in
Figure 3.5. The three duty cycles can be found by solving the system for one vector (311) and
one scalar (312) equation:
(311) V
REF
=d
S0
V
S0
+ d
M
V
M
+ d
L
V
L
and
(312) d
S0
+d
M
+d
L
=1.
The solution for duty cycles d
S0
, d
M
and d
L
can be shown to be
(313) ( ) ) sin( ) cos( 3 2
0
θ θ + ⋅ ⋅ − · m d
S
,
(314) ) sin( 2 θ ⋅ ⋅ · m d
M
and
(315) )) sin( ) cos( 3 ( 1 θ θ + ⋅ + − · m d
L
.
1
In case of the modulator is running open loop there is no need to implement arctangent function to compute the
θ coordinate since it can be freely incremented depending on the desired line frequency.
59
V
M
pn
V
3
1
⋅
p o o
o n n
θ θ
V
S0
V
REF
p o n
pn
V
3
1
⋅
pn
V
3
2
⋅
V
L
p n n
Figure 3.5. Synthesis of V
REF
in outer small triangle.
It was already mentioned that the equations for every triangular region are somewhat
different; Equations (316)(320) give the duty cycles for the middle triangular region from
Figure 3.6:
(316) V
REF
=d
S0
V
S0
+ d
M
V
M
+ d
L
V
L
,
(317) d
S0
+d
S1
+d
M
=1,
(318) ) sin( 2 m 1 d
0 S
θ ⋅ ⋅ − · ,
(319) )) cos( 3 ) (sin( m 1 d
1 S
θ θ ⋅ − ⋅ + · and
(320) )) cos( 3 ) (sin( m 1 d
M
θ θ ⋅ + ⋅ + − · .
60
p o o
o n n
o o n
p p o
p o n
V
M
V
S1
V
S0
θ θ
V
REF
Figure 3.6. Synthesis of V
REF
in middle small triangle
Finally, the sign of the left hand side of the Equation (321) determines whether the reference
vector is located in the inner smaller triangle.
(321) 0
) sin( ) cos( 3
1
m <
+ ⋅
−
θ θ
.
If the Equation (321) is true the vector is in the inner small triangle. Then the duty cycles are
found as follows:
(322) V
REF
=d
S0
V
S0
+ d
S1
V
S1
,
(323) d
S0
+d
S1
+d
Z
=1,
(324) )) sin( ) cos( 3 ( m d
0 S
θ θ − ⋅ · and
(325) ) sin( m 2 d
1 S
θ ⋅ ⋅ · .
61
p o o
o n n
o o n
p p o
V
L
V
S0
V
S1
θ θ
V
REF
Fig. 3.7. Synthesis of V
REF
in inner small triangle.
Finally, it is clear that whether implementing the SVM in the rectangular or the polar
coordinate system, the implementation becomes increasingly more complex as the number of
levels increases. This is especially true for the implementation of the algorithm in polar
coordinates, which requires the division, sine and cosine functions (or corresponding lookup
tables), and even the arctangent function for the implementation of the closedloop system.
Fortunately, most of these implementation problems (and more) can be avoided with the new
SVM algorithm described next.
3.3 The New Space Vector Modulation Algorithm
Comparing the arrangement of voltage space vectors of a fourlevel converter (from Figure
3.8) to that of the voltage space vectors from Figure 3.4, a visible pattern can be noticed in their
hexagonal structures. In fact, starting with the innermost hexagon that corresponds to the two
level converter, each new level adds one more ring of equilateral triangles to form a new
hexagon of voltage space vectors.
62
311
200
133
022
231
120
213
102
210
321
123
012
113
002
313
202
331
220
131
020
021
132
332
221
110
322
211
100
323
212
101
223
112
001
233
122
011
232
121
010
201
312
REF
M
→
0
1 2
1
2
α
β
030
031
130 230 330
320
310
300
032
033
203 103
023
013
003 303
302
301
Figure 3.8. The voltage space vectors of a fourlevel converter.
Fortunately, it turns out that this regularity in the structure of the hexagons can be used to
more efficiently represent the vectors in the plane, and can be used to generalize the SVM
algorithm. The trick is to use a set of nonorthogonal vectors as a new basis in which to
represent the switching vectors and the reference vector. One such basis set is
(326)
¹
¹
¹
'
¹
¹
¹
¹
'
¹
]
]
]
]
]
−
]
]
]
]
]
−
·
¹
'
¹
¹
'
¹
→ →
dc
dc
dc
dc
) v , v , v ( ) v , v , v (
V
V
0
,
V
0
V
h , g
ca bc ab ca bc ab
.
.
63
3.3.1 Coordinate Transformation
The first step in the algorithm is to transform the reference vector REF V
→
into a
twodimensional coordinate system. This can be achieved using a changeofbasis linear
transformation. Naturally the transformation matrix differs depending on the source coordinate
system. For example, Matrix (328) transforms the reference vector given in the linetoline
coordinate system (33) into the (g,h) coordinate system, and at the same time normalizes the
reference vector with the length of the basis vector.
(327) ) v , v , v ( REF ) h , g ( REF
ca bc ab
V T V
→ →
⋅ ·
(328)
]
]
]
− −
− −
⋅
⋅
·
1 2 1
1 1 2
V 3
1
T
dc
Because of the normalization, all the switchingstate vectors in the } h , g {
→ →
coordinate system
have only integer coordinates.
64
(2,0) (2,0)
(0,2) (2,2)
(0,2) (2,2)
(2,1)
(2,1)
(1,2)
(1,1)
(1,1)
(1,0)
(1,0)
(0,1)
(0,1) (1,1)
(1,1)
(1,2)
→
g
→
h REF M
→
α
g
β
h
Figure 3.9. Switchingstate vectors of a threelevel converter in a hexagonal coordinate system.
3.3.2 Detection of Nearest Three Vectors
The fact that all the switching vectors have integer coordinates is advantageous because the
four vectors nearest to the reference vector can be identified simply as vectors the coordinates of
which are combinations of upper and lower rounded integer values of the reference vector
coordinates, as follows:
(329)
]
]
]
]
]
]
]
]
]
]
]
·
]
]
]
·
]
]
]
·
]
]
]
·
→ →
→ →
REFh
REFg
ll
REFh
REFg
uu
REFh
REFg
lu
REFh
REFg
1 u
V
V
V ,
V
V
V
V
V
V ,
V
V
V
.
65
The endpoints of the nearest four vectors form the equalsided parallelogram that is divided into
two equilateral triangles by the diagonal connecting vectors 1 u V
→
and lu V
→
. The vectors 1 u V
→
and
lu V
→
are always two of the NTVs. The third nearest vector is one of the remaining two vectors
located on the same side of the diagonal,
(330)
ulh lg u
V V h g + · + ,
as the reference vector. Therefore, the third nearest vector can be found by evaluating the sign of
the expression
(331) ) V V ( V V
ulh g 1 u REFh REFg
+ − + .
If the sign is positive, then the vector uu V
→
is the third nearest vector; otherwise, vector ll V
→
is the
third nearest vector. This completes the identification of the NTVs for the nlevel converter.
3.3.3 Duty Cycles Computation
Once the NTVs are identified, their corresponding duty cycles can be found by solving (38)
and (39), where ul 1 V V
→ →
· , lu 2 V V
→ →
· , and ll 3 V V
→ →
· or uu 3 V V
→ →
· . Fortunately, however, because all
switchingstate vectors always have integer coordinates, the solutions are essentially the
fractional parts of the REF V
→
coordinates, or more precisely
(332)
. d d 1 d
V V d
, V V d
lu ul ll
llh REFh lu
lg l REFg ul
− − ·
− ·
− ·
and
if ll 3 V V
→ →
· , or
66
(333)
. d d 1 d
) V V ( d
), V V ( d
lu ul uu
uug REFg lu
uuh REFh ul
− − ·
− − ·
− − ·
and
if uu 3 V V
→ →
· .
Consider, for example, a threelevel converter and reference vector with amplitude
dc l l
V 8 . 1 V ⋅ ·
−
at the instant
180
50 t
π
ω ⋅ · ⋅ , which can be represented as
(334)
]
]
]
]
]
−
⋅ ⋅ ·
→
985 . 0
342 . 0
643 . 0
V 8 . 1 V
dc
REF .
Using the changeofbasis transformation (328), the reference vector in the proposed two
dimensional coordinate system is
(335)
h , g
REFca
REFbc
REFab
REFh
REFg
616 . 0
157 . 1
V
V
V
T
V
V
]
]
]
·
]
]
]
]
]
⋅ ·
]
]
]
.
From there, the nearest four vectors can be found simply as vectors the coordinates of which are
the combinations of upper and lower rounded values of the voltage reference vector:
(336)
]
]
]
·
]
]
]
·
]
]
]
·
]
]
]
·
→ → → →
0
1
V ,
1
2
V ,
1
1
V ,
0
2
V ll uu lu ul .
In order to find the third nearest vector, all that needs to be done is to evaluate the sign of the
expression:
(337) 0 16 . 0 ) 0 2 ( 616 . 0 157 . 1 < − · + − + .
67
Because it is negative, in this example, vector ll V
→
is the third nearest vector. Finally, computing
the expressions in (16) gives the solution for the duty cycles
(338)
. 227 . 0 d
616 . 0 d
, 157 . 0 d
ll
lu
ul
·
·
·
and
This completes the selection of the vectors and the computation of their corresponding duty
cycles. Nevertheless, one more step remains if the modulator is to generate the commands for
the equivalent switches from Figure 3.2.
3.3.4 SwitchingState Selection
This last step requires a transformation from a twodimensional coordinate space back to the
threedimensional space of switching combinations. In general, there is no unique solution for
this problem, which allows some freedom in optimizing the switching strategy. The first step in
choosing the best switching state or states at any given moment is to find all the switching states
that make a switching vector in question [ ]
T
h , g V ·
→
. Mathematically, all the switching states of a
switching vector satisfy a relationship
(339) [ ] 1 n , 0 h g k , g k , k where ,
h g k
g k
k
− ∈ − − −
]
]
]
]
]
− −
− .
For example, in the case of a threelevel converter, a vector [ ]
T
0 , 1 transforms into two
switching combinations: [ ]
T
0 , 0 , 1 and [ ]
T
1 , 1 , 2 . Which of the valid switching states to switch, and
for what part of their corresponding switching vector duty cycle (in case the balancing algorithm
requires use of several small vectors during one switching cycle), usually requires some
68
additional information. In fact, it depends on the strategy for balancing the dc sources, which
usually requires some knowledge of the phase currents and/or dclink voltages.
Finally, simulated switching waveforms for three, four and fivelevel converters are shown
in Figure 3.10.
V
ab
/V
dc
V
bc
/V
dc
V
ca
/V
dc
3level 4level 5level
Figure 3.10. Normalized linetoline PWM voltage waveforms for three, four and fivelevel
converters.
3.4 Practical Implementation of The New SVM Algorithm
It was already mentioned that the principal strength of the proposed algorithm is in its ease of
implementation. The ease of implementation in this case is augmented by the very small number
of instructions and with the suitability for implementation that can run in real time, for high
switching frequency, and for converters with any number of levels. The properties of the
algorithm can best be explained by presenting the flow chart of the algorithm implemented on
69
the DSP that controls the prototype SMES PCS. The algorithm can be divided in three steps as
shown in Figure 3.11.
The first step in the algorithm is to change into the (g,h) coordinate system which is in fact the
linear transformation as in (328). Other changeofbasis transformations can be constructed to
transform the reference vector from the (d,q), the (α,β) or any other basis into the (g,h) basis.
Start
Change V
REF
into (g,h) coordinate
system
Compute:
g
u
,g
l
,h
u
,h
l
End
Compute the duty cycles and
read the switching
combinations
Figure 3.11. Overview of the new SVM algorithm.
The next step in the algorithm is to find the nearest four vectors by computing the upper and
lower rounded values of the reference vector’s coordinates in the (g,h) coordinate system. Figure
3.12 shows how this operation can be conveniently implemented using a truncation function,
usually available in most floatingpoint DSPs.
70
Start
V
REFg
>0
g
u
=trunc(V
REFg
)+1
g
l
=trunc(V
REFg
)
g
u
=trunc(V
REFg
)
g
l
=trunc(V
REFg
)1
V
REFh
>0
h
u
=trunc(V
REFh
)+1
h
l
=trunc(V
REFh
)
h
u
=trunc(V
REFh
)
h
l
=trunc(V
REFh
)1
End
Yes No
Yes No
Figure 3.12. Computing the nearest four vectors.
The set of nearest four vectors needs to be reduced to the nearest three vectors, the duty cycles
need to be computed and the switching states selected. Figure 3.13 shows one possible
implementation approach in which, based on the sign of the expression (331), the algorithm
branches into two branches. One branch treats the case in which the third nearest vector is V
ll
;
the other branch treats the case in which the V
uu
is the third nearest vector. In both cases, the
duty cycles for all three vectors are computed based on the simple linear expressions from (332)
and (333). So far, the algorithm is unique and works for any number of levels converter.
In real implementation, however, one last coordinate transformation still remains. The
remaining transformation converts the vectors in the (g,h) coordinate system into the switching
functions. This transformation is not unique, because it essentially transforms a vector existing
71
in a twodimensional space into a switching combination that exists in threedimensional space
(all three singlepole triplethrow switches from Figure 3.2 can be chosen independently).
At the same time, selection of switching combinations affects the charge balance of the
dclink capacitors. A consistent choice of “wrong” switching states leads inevitably to the
charge imbalance, loss of converter levels and the overvoltage condition in the devices. In fact,
very often, control of the charge balance is performed in this part of the modulator algorithm by
selecting the switching states based on the current direction information and based on the error in
the dclink capacitors’ charge (voltage) imbalance.
Start
(V
REFg
+V
REFh
) 
(g
u
+h
l
) > 0
End
d
ul
=h
u
V
REFh
Write time to altera
switcing_state_selection
for the vector V
ul
d
lu
=g
u
VREF
g
Write time to altera
switcing_state_selection
for the vector V
lu
switcing_state_selection
for the vector V
uu
Yes
d
ul
=V
REFg
g
l
Write time to altera
switcing_state_selection
for the vector V
ul
d
lu
=V
REFh
h
l
Write time to altera
switcing_state_selection
for the vector V
lu
switcing_state_selection
for the vector V
ll
No
Figure 3.13. Computation of duty cycles and selection of switching states.
72
Figure 3.14 shows one implementation of the subroutine that selects the switching states based
on the (g,h) coordinates of the threelevel converter. Unfortunately, at this point there does not
seem to be a universal algorithm available that can select the appropriate switching states for
converters with any number of levels. That is the reason that the threelevel modulator for the
SMES system was implemented based on the lookup table (
Table 3I). The switching states are stored as hexadecimal numbers, in which the rightmost
hexadecimal number is reserved for the location of the rightmost phase, or according to
representation from Figure 3.2, phase c. The 0 equivalent switch location is coded with #3, the 1
location with #6, and finally the 2 switch location is coded with #C. The fourth hexadecimal
number from right to left is used only for the switching state of the small vectors. That number
tells the NP balancing algorithm which phase leg current is connected to the NP. The switching
states and how they affect the dclink capacitors’ charge balance in case of threelevel converters
is studied in detail in Chapter 4.
The core of the algorithm is a simple formula that computes the address of the switching state in
73
Table 3I, based on the (g,h) coordinate of the switching vector
(340) M8=7*(g+h+2)+(h+2).
In case of small vectors that can be implemented with two switching combinations, the other
switching combination is located at the address M8+4.
After the address is found, based on the measurement of the voltage imbalance and the
direction of the phase currents, the chargebalancing algorithm selects the switching state of the
small vector that will minimize the charge imbalance. If the variable NP from the algorithm is 0,
there is only one choice of the switching state for the given switching vector and that is the
switching state from the switching vector selection subroutine. If the NP is not equal to 0, one of
the two switching combinations is selected to minimize the charge imbalance.
Finally, a much more indepth treatment of the NP balancing problem is located in the next
three chapters. The main purpose of presenting the new SVM algorithm in this section was to
illustrate the simplicity of its implementation in a reallife example
.
74
Start
I8=sw_state_table
M8=7*(g+h+2)+(h+2)
M9=M8+4
sw_state=PM(M8,I8)
sw_state1=PM(M9,I8)
NP=shift right sw_state by 12
NP
not equal
0
Write
sw_state
to Altera
NP=1
Yes No
ph_a_svs=0
Yes
Write
sw_state
to Altera
Write
sw_state1
to Altera
Yes No
NP=2
Write
sw_state
to Altera
Write
sw_state1
to Altera
No
ph_b_svs=0 Yes No
ph_c_svs=0
Write
sw_state
to Altera
Write
sw_state1
to Altera
No
Yes No
Figure 3.14. Switchingstate selection with dclink capacitor charge control.
75
Table 3I The switchingstate lookup table.
Switching
State
Table
Location
Switching
Vectors, Large
& Medium
V
g,h
Small
Switching
Vectors,
V
g,h
Switching
States
(per Phase)
Hexadecima
l
Switching
Number
Small
Vectors
NP
Currents
0. (0,2) 002 #33C
1. (1,1) 012 #36C
2. (2,0) 022 #3CC
3.
4.
5.
6.
7. (1,2) 102 #63C
8. (0,1) 001 #3336 ic
9. (1,0) 122 #16CC ia
10. (2,1) 021 #3C6
11.
12. (0,1) 112 #66C ic
13. (1,0) 011 #366 ia
14. (2,2) 202 #C3C
15. (1,1) 212 #2C6C ib
16.
17. (1,1) 010 #2363 ib
18. (2,2) 020 #3C3
19. (1,1) 101 #636 ib
20.
21. (1,1) 121 #6C6 ib
22. (2,1) 201 #C36
23. (1,0) 100 #1633 ia
24. (0,1) 221 #3CC6 ic
25. (1,2) 120 #6C3
26.
27. (1,0) 211 #C66 ia
28. (0,1) 110 #663 ic
29.
30. (2,0) 200 #C33
31. (1,1) 210 #C63
32. (0,2) 220 #CC3
76
It is difficult to come up with one number that will demonstrate the superiority of this
approach, in part because due to its complexity SVM is generally not used for the converters
with more than three levels. The other difficulty is the fact that implementation of the algorithm
depends heavily on the platform. Thus, it is very hard to compare algorithms implemented on
different platforms.
Nevertheless, as an illustration of algorithm’s performance, its, reallife implementation on
the Analog Devices Shark floating point processor will be discussed next. Table 3III shows the
rough number of instructions for each step of the algorithm. It is a rough number because the
implementation does not make use of all the processor resources that can speed up the algorithm
execution, while it does include the “house keeping” instructions that communicate with PWM
generators. Therefore, substantial savings in the number of instructions can be made with an
implementation that makes use of all the available processor resources, such as the use of two
instructions per switching cycle and the use of shadow registers to pass the variables. Perhaps as
much as twothirds of the execution time can be saved in an optimized implementation,
compared to the execution time in this implementation.
Table 3III The performance of the algorithm implementation for a threelevel converter.
Number of Instructions
Change of Basis 9
Finding Nearest Four Vectors 25
Duty Cycle Computation 29
Switching State Selection (Worst Case) 3*28
TOTAL NUMBER OF INSTRUCTIONS 147
Even this suboptimal implementation (listed in the Apendix) executes on the Shark DSP
operating at 33 MHz clock speed, in less than 5µs. Therefore, the application of this algorithm
we can think of multilevel converters operating at the switching frequencies in excess of 100
kHz, where by increasing the number of levels the algorithm essentially does not change. The
switching state table becomes larger, but the speed of execution remains essentially unaffected.
77
3.5 Conclusions
This chapter introduced the fast new space vector algorithm for multilevel threephase
converters. The algorithm is general and applicable to converters with any number of levels,
since the number of steps involved in computation is always the same regardless of the location
of the reference voltage vector. In addition, the efficiency and the ease of implementation of this
algorithm makes it well suited for simulation on digital computers; it can become a very useful
tool in further exploration of the properties of multilevel power converters. The algorithmic
description of the reallife implementation shows its simplicity.
78
4 NP BALANCING PROBLEM IN THREELEVEL DIODECLAMPED
CONVERTERS
Figure 4.1 shows one of the most widespread realizations of a threelevel converter, a diode
clamped converter. At this point, it is probably clear that regardless of the implementation, the
benefits of using the multilevel topology come with a price. The price, almost as a rule, must be
paid in the form of the more complicated, more difficult to make reliable, hardware and software
realizations, and in a struggle with multiple dc sources. Threelevel diode clamped topology is
no exception to this rule.
Since one of the main advantages of this technology is its ability to utilize the switching
devices rated for onehalf of the dclink voltage, the NP must be maintained at (or very close to)
half of the dclink voltage. This is one of the critical requirements of the control system. In
addition, any disturbance in the dc link will propagate into the output of the converter, and
although the disturbance in the dclink voltage can be compensated for by the control system, a
disturbance in the charge balance (NP voltage) can not. At least not without introducing some of
the techniques that will be, perhaps for the first time, introduced in the next chapter.
79
V
pn
Sc4
Sb1
Sb3
Sb4
Sb2
Sa3
Sa4
C
dc
C
dc
b
c
R L
C
ac
A
B
C
i
p
Sa1
Sa2
a
Sc1
Sc2
Sc3
i
NP
i
n
i
a
i
b
i
c
R
R
C
ac
C
ac
p
n
L
L
NP
Figure 4.1. Circuit schematic of a threelevel VSI.
In this chapter, certain assumptions, are made (that will be relaxed later), in order to help
study the properties of NP balancing. Perhaps the most important assumption is the infinite size
of the dclink capacitors. Under this assumption the voltage in the NP will not change from its
initial condition, set at onehalf the dclink voltage. In addition, because of this assumption, the
output voltage can be assumed ideal in the average sense, which allows the use of ideal current in
place of the load.
With all these assumptions in place, this study can concentrate on modeling the properties of
SVMs with respect to the current in the NP, which will result in a more intuitive understanding
of the NP balancing problem in threelevel diodeclamped converters.
80
4.1 The Effect of Different Switching States on the Current in the NP
Figure 4.2 shows all the available voltage space vectors, and their corresponding switching
combinations. As the reference vector rotates 360 degrees, it is continuously being synthesized
by applying different switching vectors for the various length of time (duty cycles), as explained
in Chapter 3. However, this chapter develops a mathematical apparatus that is general enough to
describe the average current in the NP, and to study the contributions of different vectors to the
NP current. In addition, the model will try to explain the theoretical limitations of the control
authority over the dclink capacitors charge balance, and include guidelines for sizing the dclink
capacitors.
200
022
120
102
210
012
002 202
220 020
021 221
110
211
100
101
212
112
001
122
011
121
010
201
REF
M
→
0
1 2
1
2
α
β
Figure 4.2. Switchingstate vectors of the threelevel VSI.
81
p
n
NP
i
NP
C
DC
S
a
S
c
S
b
C
DC
i
a
i
b
i
c
o
Figure 4.3. The position of equivalent switches for the switching state pon.
Table 4I NP current i
NP
for different space vectors.
Small
Vectors
(positive
switching
state)
i
NP
Small
Vectors
(negative
switching
state)
i
NP
Medium
Vectors
i
NP
onn i
a
poo i
a
pon i
b
ppo i
c
oon i
c
opn i
a
non i
b
opo i
b
npo i
c
opp i
a
noo i
a
nop i
b
nno i
c
oop i
c
onp i
a
pop i
b
ono i
b
pno i
c
82
For the purpose of model development, it is convenient to divide the switching vectors into
three categories according to their lengths: large, medium and small. This categorization can
also be used to explain the effects of these vectors on the charge balance, because vectors from
each of the three categories affect the charge balance in a similar way. For example, the large
vectors do not affect the NP at all. The medium vectors connect one of the phase currents for the
duration of their duty cycles thus causing a charge imbalance. The effect of the medium vectors
is not controllable because it depends only on the duty cycle and the load power factor. Figure
4.3 shows how one of the medium vectors, vector pon directly connects the phase b to the NP,
and as a result, a fullphase current flows into the NP for the duration of its duty cycle.
Finally, every small vector has two realizations: one positive and one negative switching
function. Choosing one of the two switching combinations in a pair changes the sign of the phase
current that vector connects to the NP. That is the source of the positive and negative atributes.
Figure 4.3 shows the NP current for switching combinations of different small and medium
vectors.
4.2 Neutral Point Current Modulation
The previous chapter discussed the methods of synthesizing the reference vector. The
discussion will now turn to the attempt model the NP current and show its dependence on the
load current. The modeling approach to be presented uses the switching functions to model the
effect of different small and medium vectors connecting different phase currents to the NP. In
order to effectively use the switching functions, it is beneficial to have the reference vector
represented in polar coordinates.
83
In this case, the division into sixtydegree segments with inner, middle and outer small triangles
still holds. Accordingly, as explained in
Table 4I, different NP current flows for different switching combinations. For example, the
outer small triangle in Figure 4.4 consists of two components: the noncontrollable component
from the medium vector and the controllable component from the small vectors.
A controlling effect of the small vectors can be described through the introduction of the new
variable called the NP currentmodulation index. This index, [ ] 1 , 1 m
0 S
− ∈ , in the case of this
particular small triangle, represents the relative duration of the positive (onn) and negative (poo)
small vectors within V
S0
.
V
M
pn
V
3
1
⋅
p o o
o n n
θ θ
V
S0
V
REF
p o n
pn
V
3
1
⋅
pn
V
3
2
⋅
V
L
p n n
Figure 4.4. Outer small triangle.
In other words, the effective duty cycle of the vector V
S0
can be expressed, using the current
modulation index, with two duty cycles. For example in the outer small triangle of the Figure
4.4, the duty cycle of the positive switching combination (onn) is
(41) 2 / d ) m 1 ( d
0 S 0 S pos _ 0 S
⋅ + · ,
84
while the duty cycle of the negative switching combination (poo) of the same vector is
(42) 2 / d ) m 1 ( d
0 S 0 S neg _ 0 S
⋅ − · .
Then in the case of this particular outer triangle the current flowing into the NP is
(43)
a 0 S 0 S b M NP
i d m i d i ⋅ ⋅ + ⋅ · .
Using the same approach, the NP current can be expressed in the other small triangles as well.
If the reference vector is within the middle small triangle, there are two small vectors available
to aid in the NP balancing, and accordingly, two small vectormodulation indices. The NP
current is, in this case:
(44) ( ), i d m i d m i d i
c 1 S 1 S a 0 S 0 S b M NP
⋅ ⋅ + ⋅ ⋅ + ⋅ ·
Finally, as far as maintaining charge in the NP, the most beneficial is the inner triangle region
because in the inner triangle region, the reference vector is synthesized from the small vectors
only, and is therefore completely controllable:
(45)
c 1 S 1 S a 0 S 0 S NP
i d m i d m i ⋅ ⋅ + ⋅ ⋅ · .
85
4.3 Analysis and Control of the Neutral Point Voltage Error
A steady state lowfrequency ripple in the NP current is caused by a periodic variation of the
components in (43), (44) and (45) over the output voltage line cycle. In the steady state, the
voltage reference vector has a constant amplitude and rotates with constant angular speed,
dt / dθ ω · . As the reference vector sweeps through successive 60degree sectors, the duty
cycles d
S0
, d
S1
, d
M
, and d
L
become periodic functions of time. Due to the selected SVM strategy,
these functions have discontinuous first derivatives at the boundaries of the small triangles and
their wave shape dependent on the modulation index, m, as shown in Figure 4.5.
Figure 4.5. Duty cycles of SVM for modulation index m=0.8.
Although the duty cycle functions in Figure 4.5 are continuous functions, they map to
different switchingstate vectors in different 60degree sectors. For example, it can be seen from
Figure 4.2 that for
o o
60 0 < ≤θ ,
M
d is the duty cycle of the medium vector pon, and for
o o
120 60 < ≤θ ,
M
d is the duty cycle of the medium vector opn. Hence, the contribution of the
medium vector to the NP current is
b M
i d ⋅ for
o o
60 0 < ≤θ , and
a M
i d ⋅ for
o o
120 60 < ≤θ .
86
Using the switching functions (such as those in Figure 4.6), it is possible to generalize the
contribution of the medium vectors to the current in the NP for the full line cycle. In this case,
the NP current can be written in the matrix form, as in (46). It is important to notice that the
SVM algorithm completely determines the switching functions. In other words the switching
functions depend solely on the choice of modulation algorithm.
On the contrary, the duty cycle of the medium vector (also shown in Figure 4.5, for one
specific modulation index), is determined by the SVM algorithm as well as by the voltage
reference vector’s angle and magnitude. And finally, there are timevarying load currents (with
their phase angle with respect to the voltage reference vector) that also affect the current flowing
into the NP as a consequence of switching the medium vector switching states.
M
a
M
b
M
c
angle [deg] angle [deg] angle [deg]
Figure 4.6. Switching functions of the medium vectors.
(46) [ ]
.
c
b
a
c b a M vector _ mediu _ NP
i
i
i
M M M d i
]
]
]
]
]
⋅ ⋅ ·
Similar to the method used to select the switching functions of medium vectors (drawn in Figure
4.6 and represented in tabular form in
87
Table 4II, mapping functions can also be derived for the small vectors. Using the small vector
switching functions in
Table 4III and
Table 4IV, a NP current from the small vectors is expressed as
(47) [ ]
]
]
]
]
]
⋅
]
]
]
⋅
]
]
]
⋅ ·
c
b
a
c b a
c b a
1 S
0 S
1 S 0 S vector _ small _ NP
i
i
i
1 S 1 S 1 S
0 S 0 S 0 S
d 0
0 d
m m i .
Finally, combining Equations (46) and (47) results in a single expression for the NP current
over the full line cycle:
(48) [ ] [ ]
.
c
b
a
c b a
c b a
1 S
0 S
1 S 0 S
c
b
a
c b a M NP
i
i
i
1 S 1 S 1 S
0 S 0 S 0 S
d 0
0 d
m m
i
i
i
M M M d i
]
]
]
]
]
⋅
]
]
]
⋅
]
]
]
⋅ +
]
]
]
]
]
⋅ ⋅ ·
Expression (48), although valid over the full line cycle, does not offer significant insight into
the properties of the modulator with respect to the current flowing into the NP. Therefore, in
order to facilitate the analysis and control of the NP current for different loading conditions, it is
perhaps a good idea to transform Equation (48) into the rotating dq reference frame.
Substituting the phase currents with the right hand side of the expression makes the
transformation:
(49)
]
]
]
⋅
]
]
]
]
]
⋅ + ⋅ +
⋅ − ⋅ − ·
]
]
]
]
]
q
d
c
b
a
i
i
) 3 / 2 sin( ) 3 / 2 cos(
) 3 / 2 sin( ) 3 / 2 cos(
) sin( ) cos(
i
i
i
π θ π θ
π θ π θ
θ θ
,
88
After multiplying the currentmodulation indexes, the matrices of switching functions and the
transformation matrix results in a compact expression, as follows:
(410) [ ]
]
]
]
⋅
]
]
]
]
⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅
+
]
]
]
⋅ ⋅ ⋅ ·
q
d
q 1 S 1 S d 1 S 1 S
q 0 S 0 S d 0 S 0 S
q
d
q M d M NP
i
i
1 S d m 1 S d m
0 S d m 0 S d m
i
i
M d M d i .
89
Table 4II Currentswitching function for medium vectors.
Switching Function Angle
θ [deg]
Medium
Vector
M
a
M
b
M
c
i
NP
060 pon 0 1 0 i
b
60120 opn 1 0 0 i
a
120180 npo 0 0 1 i
c
180240 noip 0 1 0 i
b
240300 onp 1 0 0 i
a
300360 Pno 0 0 1 i
c
Table 4III Currentswitching function for S0 small vectors.
Current
Switching
Angle
θ [deg]
S0 Small
Vector
S0
a
S0
b
S0
c
i
NP
30060 onn/poo 1 0 0 ia
60180 non/opo 0 1 0 i
b
180300 oop/nno 0 0 1 i
c
Table 4IV Currentswitching function for S1 small vectors.
Current
Switching
Angle
θ [deg]
S1 Small
Vector
S1
a
S1
b
S1
c
i
NP
0120 ppo/oon 0 0 1 i
c
120240 opp/noo 1 0 0 i
a
240360 pop/ono 0 1 0 i
b
90
This expression is a little more readable because in steady state, i
d
and i
q
are constants and
represent active and reactive components of the load current, respectively. Note that (410) is
essentially a composite expression that combines (43), (44) and (45) into one matrix equation
that is valid over the full line cycle of the output voltage. The NP current in this formulation still
contains noncontrollable current produced by the application of the medium vector and the
controllable current produced by the small vectors.
NP current resulting from the application of medium vectors can be found by multiplying the
direct i
d
and quadrature i
q
current by the direct,
d M
M d ⋅ , and quadrature,
q M
M d ⋅ , weighing
factors, respectively. The wave shape of factors is given in Figure 4.7 for the case in which the
modulation index m=0.8. It is apparent that the quadrature component of the current will be
weighed much more heavily, and will produce much larger lowfrequency (LF) ripple than does
the direct component of current.
Figure 4.7. Weighing factors for medium vectors with m=0.8.
Similarly, NP current resulting from application of small vectors, depends on direct, and
quadrature weighing factors multiplying the direct, i
d
, and quadrature, i
q
,
load currents. Figure
4.8 shows the wave shape of the small vectors weighing factors for modulation index m=0.8.
These four weighing factors depend not only on the small vectors’ duty cycles and the current
switching functions that are determined by the type of SVM used, but also on the control inputs
m
S0
and m
S1
, as defined earlier. Two distinct sets of weighing factors are given in Figure 4.8.
91
One set of weighing factors used when only positive small vectors are involved (i.e., m
S0
=m
S1
=1), is given by the solid line; the other set when only negative small vectors are used (i.e.
m
S0
=m
S1
=1), is given by the dashed line. Between these two extreme cases, the weighing
factors can be controlled by adjusting currentmodulation indices.
Figure 4.8. Weighing factors for small vectors with m=0.8.
The weighing factors for medium vectors are periodic functions with zero average value over
a line cycle. This means that in steady state, at which i
d
and i
q
currents are constant, the NP
current from medium vectors will naturally balance over a line cycle, and the only concern is the
size of the LF ripple.
Note that the ratio of active and reactive weighing factor is opposite for medium and small
vectors. Large i
d
and small i
q
means large control authority on the NP current through the
manipulation of currentmodulation indices of small vectors, and small disturbance from middle
vectors. On the other hand, large i
q
and small i
d
means large disturbance from middle vectors,
92
with small control authority from small vectors. This means that it will be much easier to
suppress the LF ripple from direct current, than it is to suppress them from quadrature current.
These results confirm the fact already proven in practice that accomplishing the NP balance is
not very difficult for loads with high power factor. This requirement is usually met for acdrive
applications (except at noload conditions, when the currents are generally small anyway). It is
also true for threelevel rectifiers operating at or close to unity power factor. However, for static
VAR compensation, this requirement will not be met, meaning that this application will require
different design tradeoffs.
Regarding NP balancing control for SVM, with the restriction to use of the NTV method,
three distinctive approaches to the control of NP might be:
• Passive control, in which the positive and negative small vectors are selected alternately in
each new switching cycle. This method can work only in the case of a perfectly balanced
load and a perfectly balanced PWM scheme, which is unlikely to happen in practice. In
addition, this method would have difficulties to recovering from line or load transients. Still,
this control method can be used to establish a benchmark for NP controller performance.
This benchmark can then be used to evaluate the performance of other NP control methods.
• Hysteresistype control is perhaps the simplest and most popular closedloop NP control
scheme. This method requires knowledge of the current direction in each phase. Based on
that information, the small vectors’ switching combination that will move the NP voltage in
the direction opposite from the direction of imbalance can be selected. The downside of this
method is that the current ripple has a strong harmonic component at one half of the
switching frequency. One implementation of this method is described in Figure 3.14.
• Active control schemes control the current modulation indices m
S0
and m
S1
. In general, these
schemes require measurement of the voltage imbalance in the NP, and often require
measurement of the amplitudes of the phase currents as well. The benefits of these schemes
are that they do not have the ripple at half the switching frequency, and some variations of
93
these control schemes can balance the NP exactly. Unfortunately, they all increase the
switching losses (due to the introduction of additional switching states), and may be less
robust than the hysteresistype control schemes.
Because the resulting NP ripple is lowfrequency, the selection of the balancing algorithm
(which affects the highfrequency content of the NP current) is not important in determining the
size of the dclink capacitors, and will not be investigated further.
4.4 Limitations of the NeutralPoint Control and Some Design
Recommendations
From the analysis in previous sections, it is clear that regardless of control scheme the control
authority over the NP current is limited, but that the region in which exact balancing can be
achieved in each switching cycle must exist. This region is given as a shaded area in Figure 2.9.
Note that the graph is symmetrical, and that the unity power factor load represents the most
favorable case, in which the NP voltage can be balanced in every switching cycle for a
modulation index as high as m=0.96.
If the operating point exits the shaded region, lowfrequency, mainly thirdharmonic ripple
current flows into the NP. This is a highly undesirable effect because for some converter designs
it significantly increases the size of the dclink capacitors.
Figure 4.10 shows the peaktopeak value of the lowfrequency charge ripple in the NP
divided by the amplitude of output phase current. The first graph shows the normalized NP
charge ripple for the passive control of NP voltage balance. The second graph shows the best
that can be done using feedback control of NP voltage balance. The shaded region represents the
same ripple free area as the one in Figure 4.9.
94
m
o
d
u
l
a
t
i
o
n
i
n
d
e
x
m
power factor angle (deg)
Figure 4.9. Region where LF ripple can be suppressed.
The actual dclink capacitor size for the specified NP voltage ripple and for any power factor
angle and voltage modulation index m, can be computed by multiplying the normalized charge
ripple, ) , m ( Q Q
LOAD NORM NORM
φ · , from Figure 4.10, by the current amplitude, then dividing that
by two times the desired size of the NP voltage ripple:
(411)
max _ DC
max NORM
i
U 2
I Q
c
∆ ⋅
⋅
· .
This should provide sufficient guidelines to size the dclink capacitors for any expected
operation mode and desired NP voltage ripple.
95
Passive NP control Feedback NP Control
N
o
r
m
a
l
i
z
e
d
N
P
c
h
a
r
g
e
N
o
r
m
a
l
i
z
e
d
N
P
c
h
a
r
g
e
m
o
d
i
n
d
e
x
m
o
d
i
n
d
e
x
p
o
w
e
r
f
a
c
t
o
r
a
n
g
l
e
p
o
w
e
r
f
a
c
t
o
r
a
n
g
l
e
Figure 4.10. Normalized amplitude of the LF charge ripple.
For example, consider the prototype SMES PCS with 1800V dclink voltage and Imax=200A
phase current amplitude, and allow 1% voltage ripple (∆Udc_max=18 V) in the NP. For
modulation index m=0.9, the comparison of capacitor sizes for the feedback NP control and
passive NP control is summarized in Figure 4.11. It is clear that the greatest savings in the size
of capacitor can be achieved when the inverter is predominantly supplying active power; while
for the operation with purely reactive power, the benefits of feedback NP control diminish.
96
0
2000
4000
6000
8000
10000
12000
14000
1 0.8 0
µF m=0.9
cos(φ)
Figure 4.11. Capacitor sizes for specified NP ripple, with and without NP control.
4.5 Experimental Results
A full experimental verification of the results from this chapter requires the converter to
operate in all four quadrants and with all possible power factors. In addition, to investigate the
situation with the converter as a load (rectifier mode of operation) would require a highpower
threephase current source, and synchronization of that current source with the voltage reference
vector. In short, this is a rather difficult requirement, which does not hold the promise of
introducing new, and unexpected results
1
. Instead, Figure 4.12 shows the experimental setup
used to verify the theoretical results for the inductive loads of three different power factors,
namely, 85, 45 and 12 degrees.
1
Operating a converter in a rectifier mode with the current loop closed and with a high voltage ripple in the NP (due
to the low power factor) seemed like a risky experiment.
97
p
n
i
NP
C
DC
C
DC
i
a
i
b
i
c
o
L
a
L
b
L
c
Digital Modulator
V
NP
R
a
R
b
R
c
ThreeLevel
Converter
Switching
Network
Figure 4.12. Experimental verification of NP voltage ripple properties.
The dclink voltage was set to 80V. There were three different load conditions. Three, 1.8
mH, inductors in series with three .06Ω resistors were used as 85 degrees reactive power factor
load. Three .72Ω resistors in series with the same inductors for the 45 degrees reactive power
factor and finally three1Ω resistors in series with three .6mH inductors for 12 degrees inductive
load. The experimental results in Figure 4.13 show the ripple in the NP voltage normalized with
the amplitude of the phase current, when the load with the power factor angle of 85 degrees is
connected to the ac terminals. Finally, the collection of experimental results for different
operating points (Figure 4.13) show excellent agreement with the theoretical result from this
Chapter.
98
0 0.005 0.01 0.015 0.02 0.025
1
0
1
x 10
3
NP Voltage Ripple cos(fi)=.087
m
=
.
4
0 0.005 0.01 0.015 0.02 0.025
1
0
1
x 10
3
m
=
.
5
0 0.005 0.01 0.015 0.02 0.025
1
0
1
x 10
3
m
=
.
6
0 0.005 0.01 0.015 0.02 0.025
1
0
1
x 10
3
m
=
.
7
0 0.005 0.01 0.015 0.02 0.025
1
0
1
x 10
3
m
=
.
8
0 0.005 0.01 0.015 0.02 0.025
1
0
1
x 10
3
m
=
.
9
0 0.005 0.01 0.015 0.02 0.025
1
0
1
x 10
3
m
=
1
.
0
t [s]
Figure 4.13. Normalized NP voltage for power factor angle of 85 degrees.
Theoretical Results
N
o
r
m
a
l
i
z
e
d
N
P
c
h
a
r
g
e
m
o
d
i
n
d
e
x
p
o
w
e
r
f
a
c
t
o
r
a
n
g
l
e
100
0
100
0.4
0.6
0.8
1
0
0.5
1
1.5
x 10
3
p
o
w
e
r
f
a
c
t
o
r
a
n
g
l
e
m
o
d
i
n
d
e
x
N
o
r
m
a
l
i
z
e
d
N
P
c
h
a
r
g
e
Experimental Results
Ripple free area
Figure 4.14. Theoretical and experimental normalized NP voltage ripple.
99
4.6 Conclusions
This chapter introduces a novel approach to modeling the threelevel space vector modulator
for the purpose of investigating the characteristics of NP balancing. The model is general and
can be used to describe any SVM and its effect on the NP.
Using the modeling approach developed in the beginning of the chapter, properties of NP
control were studied for one particular SVM. The study identified the region in which the NP
can be balanced on a switching cycle level, and offered an intuitive explanation of the origin of
the ripple. In addition, a quantitative study of NP balancing offered a normalized size of the dc
link voltage ripple for all possible operating conditions, which is essential information for the
design of dclink capacitors.
Finally, from this study of NP balancing (and from available literature), it is clear that the NP
can be balanced in a threelevel converter for all possible operating conditions. However, the
question of tradeoff between the size of the ripple and the size of the dclink capacitors remains.
Furthermore, through the use of the feedforward algorithm, introduced in the next chapter, this
tradeoff can be extended to include the trade off between the voltage capacity of the switches and
the size of the dclink capacitors.
100
5 FEEDFORWARD CONTROL OF THREELEVEL NEUTRALPOINT CLAMPED
CONVERTERS
From the analysis in the previous chapter, as well as that shown in the many cited references,
it becomes clear that the load with the low power factor creates a ripple current in the NP of the
threelevel converter. Because the majority of the threelevel converters used for the drive
applications need a significant energy storage in order to compensate for the possible
disturbances in the utility line voltage, the lowfrequency current ripple often does not determine
the size of the dclink capacitors.
However, if the goal is to design an optimized converter, one that does not have the need for
excessive energy storage, then the amplitude of the ripple current flowing into the NP becomes a
primary factor in determining the size of the dclink capacitors. This is the situation for which
the results from the previous chapter become important.
Further thinking about capacitorsize optimization leads to the question of whether the size of
the dclink capacitors can be further reduced if a certain amount of the NP ripple can be
tolerated. The answer to that question is positive, if two basic requirements are satisfied. The
first requirement is that the switches must be able to handle the extra voltage stress, and the
second requirement is that an effective feedforward algorithm must be constructed in order to
101
compensate for the distortion in the output ac voltage waveforms that occur due to the ripple in
the NP voltage.
Therefore if an effective stable algorithm can be found, then allowing a certain amount of
voltage ripple can be a good way to save on the capacitor size. Furthermore, the only hardware
requirement for the implementation of this algorithm is the dclink sensors, which are needed for
the NP controller as well, and are therefore available almost by default. In addition, such an
approach can, with few modifications, be applied to other converter topologies such as the flying
capacitor topology or even the cascaded fullbridge converter. For these reasons, this chapter
will first propose and than make a detailed study of a feedforward algorithm that can effectively
compensate for the ripple in the NP.
5.1 The Effect of the Capacitor Charge Imbalance on the Voltage Space Vectors
The task of the modulator is to determine the switching sequence of the equivalent switches
from Figure 5.1 that will result (in the average sense) in the synthesis of the voltage reference
vector commanded by the system controllers. Also (as was already mentioned) every switching
state produces welldefined threephase line voltages, which can be represented as two
dimensional voltage space vectors (as shown in Figure 5.2). The goal at this point is to examine
how the error in the NP affects the voltage space vectors. It is convenient to use the definition of
the error in the NP voltage given in Figure 5.3.
102
a
b
c
.
.
c
1
.
.
0
2
.
a
1
.
.
0
2
.
b
1
.
.
0
2
0
1
.
.
2
.
.
.
V
p
V
n
NP
i
NP
Figure 5.1. Functional diagram of a threelevel converter.
200
022
120
102
210
012
002 202
220 020
021 221
110
211
100
101
212
112
001
122
011
121
010
201
REF
M
→
0
1 2
1
2
α
β
Figure 5.2. Voltage space vectors of a threelevel converter.
103
0
1
2
V
p
=V
dc
/2+∆V
0
V
dc
C
dc
C
dc
+
+
i
NP
V
n
=V
dc
/2∆V
Figure 5.3. The definitions of the dclink voltages and the voltage error in the NP.
Because the large voltage space vectors do not connect any of the phases to the NP, these
vectors remain unaffected by the voltage error in the NP. On the other hand, voltage error in the
NP does affect both medium and small voltage vectors. For example, if the NP voltage error,
∆V, is smaller than 0, the medium vectors shift like as shown in Figure 5.4. From the same
figure, it is clear that one of the line voltages remains unaffected, which explains why the tip of
the medium vectors slides along the outer border of the hexagon. Table 5I shows how the line
toline voltages of each medium vector are affected.
104
120
102
210
012
021
201
Vca
Vbc
V ∆
V
3
2
∆
Figure 5.4. Shift of the medium voltage vectors due to the ∆V>0 error in NP voltage.
Table 5I Effect of the error in NP voltage on the medium voltage vectors.
210 120 021 012 102 201
V
ab V
dc
/2∆V V
dc
/2+∆V V
dc V
dc
/2∆V V
dc
/2+∆V V
dc
V
bc V
dc
/2+∆V V
dc V
dc
/2∆V V
dc
/2∆V V
dc V
dc
/2∆V
V
ca
V
dc V
dc
/2∆V V
dc
/2+∆V V
dc V
dc
/2∆V V
dc
/2+∆V
Small vectors are affected by the error in NP in a somewhat different way. While there is no
error in the NP voltage, each small vector switching combination has the same magnitude and
phase. On the other hand, when there is an error in the NP voltage, each of the switching
combinations of a small vector turns into a different small vector. Therefore, the effective
“splitting” of the small vectors (shown in Figure 5.5) occurs, and only the magnitude changes
while the angle remains unaffected. Again, Table 5II shows a summary of the small vectors’
line voltages.
105
221
100
1
0
1
112
011
201
V
ab
V
ca
V
bc
211 122
1
1
0
0
1
0
0
0
1
212
121
V
3
2
∆
Figure 5.5. “Splitting” of the small voltage vectors due to the ∆V>0 error in NP voltage.
Table 5II Effect of the error in NP voltage on the small voltage vectors.
211 221 121 122 112 212
V
ab V
dc
/2∆V 0 V
dc
/2+∆V V
dc
/2+∆V 0 V
dc
/2∆V
V
bc
0 V
dc
/2∆V V
dc
/2∆V 0 V
dc
/2+∆V V
dc
/2+∆V
V
ca V
dc
/2+∆V V
dc
/2+∆V 0 V
dc
/2∆V V
dc
/2∆V 0
100 110 010 011 001 101
V
ab
V
dc
/2+∆V 0 V
dc
/2∆V V
dc
/2∆V 0 V
dc
/2+∆V
V
bc
0 V
dc
/2+∆V V
dc
/2+∆V 0 V
dc
/2∆V V
dc
/2∆V
V
ca V
dc
/2∆V V
dc
/2∆V 0 V
dc
/2+∆V V
dc
/2+∆V 0
It is important to note that because the outer boundaries of the hexagon remain unaffected by
the error in the NP voltage, the maximum amplitude of the reference vector that can be
synthesized remains unaffected by the size of the NP voltage error. This should indicate the
feasibility of finding a correction algorithm, and that it should be possible to compensate for the
effect of the NP voltage ripple without sacrificing the converter voltage gain. However, a
106
significant change in the small vectors’ amplitude will change their corresponding duty cycles
and will ultimately affect their control authority over the NP current. This is only one of the
issues discussed in thiscChapter.
5.2 Sine Triangle Modulation for ThreeLevel Converters
Although SVM is perhaps the most general and unrestricted approach to the generation of
PWM signals, and although the sine triangle modulation is basically a subset of SVM
1
, it might
be the subset that is significantly simpler to model and easier to understand. That is the principle
reason why in this chapter it is the sine triangle modulator that will be predominantly studied.
Then, with few restrictions, the conclusions can also be used to explain the behavior of the space
vector modulator.
5.2.1 PWM Generation with Sine Triangle Modulation
All sine triangle modulations are the same in the sense that that the intersection between the
carrier triangular waveform and the modulation signal determines the switching instances. The
main variations are in the exact shape of the carrier waveform and/or the modulating signal that
can be augmented with the third harmonic for increased voltage gain, and so on.
Then, using the same basic principle, a multilevel modulator can be constructed that will have
n1 carrier waveforms, where n is the number of levels. As an example, Figure 5.6 shows a basic
threelevel sine triangle modulator.
1
Historically, sine triangle modulation was the only option when implementing the modulator with discrete
components. However when implemented with the processor, the distinction between the two implementations (in
terms of versatility and freedom to optimize the PWM signals) diminishes.
107
θ·ωt
v
x0
v
x_mod
θ·ωt
V
dc
0
1
2
Figure 5.6. Principle of sine triangle modulation for a threelevel converter.
5.2.2 NP Control With Sine Triangle Modulation
Adding the zero sequence to the modulating sinusoidal waveform controls the charge balance
in the NP. This is really the method used so far throughout this dissertation, because
manipulating the switching combinations of small vectors actually manipulates zero sequence.
(51)
1 ) 3 / 2 t sin( A v
1 ) 3 / 2 t sin( A v
1 ) t sin( A v
mod _ c
mod _ b
mod _ a
+ + ⋅ + + ⋅ ⋅ ·
+ + ⋅ − + ⋅ ⋅ ·
+ + + ⋅ ⋅ ·
δ π ϕ ω
δ π ϕ ω
δ ϕ ω
Figure 5.7 shows how a constant zero sequence shifts the power transfer to the upper of the
split dclink capacitors, which then (depending on the direction of the power flow) becomes
charged or discharged more than the lower dclink capacitor. Therefore, depending on the
108
direction of the power flow, a specific amount of zero sequence can be added to correct for the
disturbance.
It is important to note, however, that similar to the space vector modulator, this type of control
has its limitations. For example, if the modulating sinusoid is too large, there might be little or
no room to add the zero sequence. In that case, the size of the ripple current will depend solely
on the load impedance and load power factor, which is a limitation that applies to the SVM as
well. Despite these limitations, from the modeling point of view, sine triangle modulation has
one big advantage. The NP controller needs to supply only one control variable, and does not
really require information on the direction of phase current, but only on the direction of power
flow. This greatly facilitates writing equations and developing the model.
v
x0
δ
θ
θ
v
x_mod
0
1
2
Figure 5.7. Principle of NP control for the threelevel converter.
5.2.3 The Average Model of a Sine Triangle Modulator
By magnifying the time scale and looking into the process of computing the duty cycles, there
are really two cases to consider. Figure 5.8 shows both cases: one in which the modulating signal
109
is compared with the upper carrier waveform, and the other in which it is compared to the lower
modulating waveform. Because it is advantageous for the utilization of multiple dc voltage
levels and because it reduces the switching frequency ripple, it almost goes without saying that
all the switching should occur between the two closest levels. Accordingly, from the lefthand
side of the illustration in Figure 5.8, the modulator will connect the phase output to the positive
rail for time T d
p
⋅ , where
(52) 1 v d
x p
− · } c , b , a { x ∈
and v
x
is the modulating signal for phase x, } c , b , a { x ∈ . The duty cycle for the NP is then
,
(53)
x p 0
v 2 d 1 d − · − · .
When the modulating signal is compared with the bottom carrier waveform the duty cycles for
NP and for the negative dc rail are
(54)
x 0
v d ·
and
(55)
x n
v 1 d − · .
110
v
x
T
0
2
d
p
.
T
d
0
.
T
1
v
x
T
0
2
d
0
.
T
d
n
.
T
1
Figure 5.8. Duty cycles for the sine triangle threelevel modulation.
Perhaps the most intuitive (and although approximate, quite accurate), is the modeling
approach published by Newton and Sumner [B14]. The modeling approach averaged the NP
current during the line cycle, which results in the expression that describes the relationship
between the zero sequence command, the amplitude of the modulating signal, the load power
factor and the NP current that is to be controlled. That expression can then be linearized around
an operating point and that model can be used for control design. This chapter will first briefly
repeat Newton and Sumner’s methodology and main resultsthen extend those results in order to
model the sine triangle modulator with feedforward control.
To find the line cycle average of the current flowing through the NP, it is necessary to first
study the duty cycles of one phase leg, as shown in Figure 5.9. Clearly, the phase flows through
the NP for the duration of the NP duty cycle d
1
. From (52) and (53), and after substituting the
modulating v
a_mod
from (51), the duty cycle of the NP can be expressed as follows:
(56)
¹
'
¹
⋅ < ≤ ⋅ +
< ≤ ⋅ −
·
π θ π θ
π θ θ
2 ) sin( A 1
0 ) sin( A 1
d
1
.
Hence the average current flowing in the NP during one line cycle is
111
(57)
]
]
]
⋅ ⋅ + + ⋅ +
+ ⋅ ⋅ − + ⋅ ⋅
⋅
·
∫
∫
⋅π
π
π
θ θ ϕ θ
θ θ ϕ θ
π
2
0
NP
d )) sin( A 1 )( sin( I
d )) sin( A 1 )( sin( I
2
1
i
,
which quite expectedly, turns out to be zero ( 0 i
NP
· ) for any power factor angle ϕ. It might be
clear (but it is still important to point out) that this analysis does not study the lowfrequency,
mainly thirdharmonic ripple that is present in the NP. All it says is that under ideal conditions,
regardless of the power factor angle, the line cycle average of the NP current will be zero.
θ
θ
d
x2
θ
θ
d
x0
d
x1
v
x_mod
0
1
2
Figure 5.9. Duty cycle of one phase equivalent switch for sine triangle modulation.
112
However, even though in ideal conditions the NP current is zero, the feedback implementation
is required to insure that the capacitor charge can balance during the transients or in the presence
of disturbances. In that case, it is important to analyze the average current in the NP for different
phase angles in order to determine the control authority of the zero sequence over the NP. The
integration is a little more involved in that case because the integration limits that correspond to
the zero sequence δ need to bedetermined. From Figure 5.10, it is clear that the first
discontinuity in the duty cycles occurs for
(58) 1 ) sin( 1 · + + δ α ,
which sets the first integration limit at
( 59)
,
`
.

− ·
−
A
sin
1
δ
α .
From there, the remaining two integration limits are simply
(510) α π β − · and α π γ + ⋅ · 2 .
The line cycyle average NP current can be computed from
(511)
]
]
]
]
⋅ + ⋅ + + ⋅ +
+ ⋅ + ⋅ − + ⋅ ⋅
⋅
·
∫
∫
γ
β
β
α
θ δ θ ϕ θ
θ δ θ ϕ θ
π
d )) ) sin( A ( 1 )( sin( I
d )) ) sin( A ( 1 )( sin( I
2
3
i
NP
,
resulting in the following expression for the line cycle average of the NP current:
(512)
]
]
]
⋅ + − ⋅ ⋅
⋅
⋅ −
·
−
A
sin A A ) cos(
A
I 3
i
1 2 2 2
NP
δ
δ δ φ
π
,
113
valid only for
(513) 2 1 A < + < δ .
θ
θ
d
2
θ
δ
α β
γ
d
0
0
v
x_mod
0
1
2
θ
0
0
0
d
1
Figure 5.10. Duty cycles of one phase with the zero sequence present.
5.3 Modeling the FeedForward Control
If the modulator computes the duty cycles and selects the vectors under the assumption that
the charge in the dclink capacitors is balanced, then the modulator would make an error
114
whenever the capacitor charge was imbalanced. Most of the modulation algorithm does not take
this possibility into account, perhaps because it is assumed that the dclink capacitors will be able
to sufficiently minimize the voltage ripple in the NP.
However, as already mentioned, if a certain amount of ripple in the NP can be tolerated, then
that is one way to make significant savings on the size of the dclink capacitors. In that case, a
feedforward control algorithm that allows the modulator to take into account the ripple in the
NP and still produce highquality sinusoidal output waveforms becomes significant.
5.3.1 The FeedForward Algorithm
The idea behind the feedforward algorithm is to modify the amplitude of the triangular
carrier waveforms so that the ratio of upper and lower carrier waveform corresponds to the ratio
of upper and lower dclink capacitor voltage. One way to satisfy this requirement is to control
the amplitude of the triangular modulating waveform to make sure that the following condition is
always satisfied:
(514)
2 / V
V
K
dc
p
p
· and
2 / V
V
K
dc
n
n
· ,
where K
p
and K
n
are the amplitudes of the top and bottom carrier waveforms, respectively. In
that case,
(515)
n
p
n
p
V
V
K
K
· and 2 K K
n p
· +
also holds true.
From Figure 5.11 (and with the help of a little trigonometry), it turns out that when the
modulating signal is larger than K
n
, the phase duty cycles are
115
(516)
p
n x
2
K
K v
d
−
· and
p
x
1
K
v 2
d
−
· ,
and when the modulating signal is lower than K
n
, the duty cycles are
(517)
n
x
1
K
v
d · and
n
x n
0
K
v K
d
−
· .
K
p
K
n
v
x
T
0
2
d
p
.
T
d
0
.
T
K
p
K
n
v
x
T
0
2
d
0
.
T
d
n
.
T
Figure 5.11. Duty cycles for the sine triangle threelevel modulation.
5.3.2 Mathematical Basis of the FeedForward Algorithm
Even though the development of the feedforward algorithm is quite intuitive, it might not be
at all obvious whether there will be a discontinuity when the modulating signal crosses the
“moving” boundary between the upper and lower carrier signals. In order to examine the effects
of crossing the boundary, it is probably easiest to write equations for the average output voltage
for the case in which the modulating signal v
x
(t)>K
n
, and when the v
x
(t)<K
n
, where
116
(518) δ θ + + ⋅ ·
n mod _ x
K ) sin( A v ,
and where
(519) ] 2 , 0 [ v
mod _ x
∈ ] 1 , 0 [ A∈ , A 2 K A
n
− < + < δ .
The output voltage for the case when v
x_mod
(θ)>K
n
is
(520) V ) ( d v
p p 0 x
⋅ · θ .
After substituting equation (516), the output voltage can be written as
(521)
p
p
n x
0 x
V
K
K ) ( v
v ⋅
−
·
θ
,
and finally, after substituting the relation from (514), the final form of the output voltgae is
(522)
2
V
) K ) ( v ( v
dc
n x 0 x
⋅ − · θ .
Using the same steps, the output voltage of the modulator, for the case when the v
x
(θ)<K
n
,
turns out to be
(523)
2
V
) K ) ( v ( V
K
) ( v K
V ) ( d ) t ( v
dc
n x n
n
x n
n n 0 x
⋅ − · ⋅
−
· ⋅ − · θ
θ
θ ,
which is exactly the same expression as (522). Therefore, there are no discontinuities at the
boundary of the lower and upper carrier waveform for the feedforward algorithm (514). This
ends the proof.
117
5.3.3 Average Model of the ZeroSequence NP Control
In terms of NP the switching network in Figure 5.1 (together with any source or the load that
might be connected to its ac side) is just a current source like the one in Figure 5.12. What
complicates this otherwise simple modeling approach is that there are multiple dependencies of
the source that are influenced not only by the phase angle of the load and the feedforward terms,
but also by the zero sequence of the modulating signal. Still, as the experimental results in the
next chapter will prove, because of the filtering properties of the dclink capacitors most of the
unmodeled dynamics from the load and the output filters will appear only as a nonlinear gain in
the NP controller loop. That is the principle argument for the application of the linecycle
averaging method in the study of NP voltage dynamics.
C/2
C/2
i
0
i
p
i
n
V
DC
v
p
+
+
v
n
+
Figure 5.12. The model of the modulator with respect to the NP current.
As shown in Figure 5.12, the voltage in the NP depends on the equivalent NP current, which
again depends on the load currents and the NP duty cycle of each phase,
118
(524)
c c b b a a
p
i d i d i d
dt
dv
C ⋅ + ⋅ + ⋅ ·
0 0 0
.
In order to perform linecycle averaging, the boundary between the piecewise linear NP duty
cycles needs to be determined. As shown in Figure 5.13, and similar to the equation (58), the
first discontinuity in the duty cycle comes from the following condition:
(525)
n n
K ) sin( A K · ⋅ + + α δ .
From this equation and Figure 5.13, it becomes clear that the integration boundaries are in fact
the same as those in (511),
(526)
A
sin
1
δ
α
−
− · , α π β − · and α π γ + ⋅ · 2 .
Substituting the modulating signal, defined in Figure 5.13 as ) sin( A K v
n x
θ δ ⋅ + + · , into
the expressions for NP duty cycles, d
0
in (516) and (517), the linecycle average of the NP
current can be found by integrating
(527)
]
]
]
]
⋅
,
`
.
 + +
⋅ + +
+
⋅
,
`
.

+ + −
⋅ + ⋅ ·
∫
∫
θ
θ δ
φ θ
θ
θ δ
φ θ
γ
β
β
α
d
K
) sin( A K
) sin( I
d
K
)) sin( A K ( 2
) sin( I 3 i
n
n
p
n
NP
,
which results in the expression
(528)
[
] )
A
( sin ) K K ( A 2
) K K ( A A ) 4 2 K 4 ( 2
K K A 4
) cos( I 3
i
1
n p
2
p n
2 2 2
n
p n
NP
δ
π δ δ
π
φ
−
⋅ + ⋅ ⋅ +
+ − ⋅ ⋅ − − ⋅ − ⋅ + ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅
⋅ ⋅ −
·
,
119
which is valid for
(529) A 2 K A
n
− < + < δ .
K
p
K
n
δ
0 α β
γ
2
0
θ
d
2
θ
d
0
θ
0
0
d
1
0
θ
v
x_mod
Figure 5.13. The duty cycles of the sine triangle modulator with zero sequence.
120
Outside the region defined in (529), the modulator works in overmodulation mode, which is
not covered in this analysis. Otherwise, if the dclink capacitor charges are balanced (K
P
=
K
n
=1), Equation (528) simplifies to the following expression for the average NP current with no
feedforward terms:
(530)
]
]
]
⋅ + − ⋅
⋅
⋅ −
·
−
A
sin A A ) cos(
A
I 3
i
1 2 2 2
NP
δ
δ δ φ
π
5.3.4 A SmallSignal Model of the ZeroSequence NP Control
The average model of the zerosequence control in the NP can be derived by developing the
expression for the average current (528) into Taylor’s series around the operating point 0 · δ ,
K
p
=K
n
=1, and then by neglecting the higherorder terms as follows:
(531) K + ⋅
∂
∂
+ ⋅
∂
∂
+ ·
2
2
NP
2
NP
NP NP
) 0 (
i
) 0 (
i
) 0 ( i ) ( i δ
δ
δ
δ
δ
The first derivative of the average NP current (with respect to δ ) is
(532)
[
]
n
2 2
2 2
p n
2 2
n
p n
2 2
p n
p n
NP
K
A A
8
A ) K K (
A
2
A A
) 2 K 2 ( 2
) K K (
A
A 2
) K K (
K K 4
) cos( I 3 i
⋅
− ⋅
⋅
− − ⋅ + +
+
−
⋅ + ⋅ ⋅ ⋅
⋅ + −
−
⋅
⋅ + ⋅
⋅ ⋅ ⋅
⋅ ⋅
− ·
∂
∂
δ
δ
δ
δ
δ δ
δ
π
φ
δ
.
The first derivative of the linecycle average in the operating point 1 K K , 0
n p
· · · δ , is
(533) ) cos(
I 6
) 1 K K , 0 (
i
n p
NP
φ
π
δ
δ
⋅
− · · · ·
∂
∂
,
121
while the average value of the linecycle average of the NP current is zero around the operating
point
(534) 0 ) 1 K K , 0 ( i
n p NP
· · · · δ .
Finally, at the operating point,
(535) δ φ
π
⋅ ⋅ ⋅ − · ) cos( I
6
i
NP
.
Therefore, around the operating point, which assumes the balanced charge in the dclink
capacitors, and when the zerosequence voltage equals zero, the smallsignal average current
model is the same as the smallsignal model for the modulator without the feedforward control
[B14].
5.3.5 The Stability Analysis of the Feed Forward Algorithm
So far, the line cycle average dynamic model of the sinetriangle modulator with and without
the feedforward control has been derived. In addition, through linearization around the
operating point, the smallsignal dynamic model of the NP control has been derived as the
function of the load power factor, the zerosequence balancing component and the feedforward
term. The feed forward algorithm has also been proven to not introduce any discontinuities in
the output waveform and has shown that it can effectively compensate for any size of imbalance.
However, the feedforward control does change the ratio between the two triangular carrier
signals and, by doing so, it effectively introduces a zerosequence component which influences
the NP control. For example, Figure 5.14 shows the situation in which the feedforward
controller compensates for the significantly higher voltage of the upper dclink capacitor. At the
same time, the size of the modulating signal does not allow the freedom to control the zero
sequence for the purpose of NP control.
122
As clearly illustrated in the situation shown in Figure 5.14, the output is switched more often
between the positive rail and the NP than between the NP and the negative rail. In addition, in
an exaggerated situation with the bottom dclink capacitor voltage equal to zero, the output of the
converter would never be connected to the bottom dc rail; therefore, all the power would flow
between the upper capacitor and the load, and there will be no power flow between the load and
the bottom capacitor.
θ
θ V
an
Figure 5.14. The PWM signal with strong feedforward component.
Evidently, depending on the direction of the power flow, the top capacitor will be either
charged or discharged. For example, Figure 5.15 shows the power flow in the inverter mode of
operation, with the upper dclink capacitor at much higher voltage than the bottom and with a
high modulation index, as in Figure 5.14. The feedforward influence will always tend to
discharge the capacitor with the higher state of charge, and in that way it exercises a stabilizing
influence. Conversely, in the rectifier mode, with the power flowing from the ac to the dc side,
the feedforward algorithm introduces a destabilizing influence by always further charging the
capacitor with the higher voltage.
123
2
0
NP
C
DC
S
a
S
c
S
b
C
DC
1
V
DC
P
Z
L
Z
L
Z
L
Figure 5.15. Power flow in the inverter mode of operation with large imbalance in NP.
In other words, in the inverter mode of operation, feedforward control reduces the size of the
NP voltage ripple by inherently providing a stabilizing influence. In the rectifier mode of
operation however, the feedforward control still compensates for the disturbance in the line
voltage caused by the dclink capacitor charge imbalance, but it also introduces a destabilizing
effect into the NP control. This destabilizing effect can be compensated for with the addition of
zerosequence voltage. Unfortunately, the compensation of the feed forward by the zero
sequence limits the maximum amplitude of the modulating signal to
]] K min[ ], K min[ [
p n
− ∈ δ . Therefore, for the rectifier operation a tradeoff exists between the
influence of the feed forward algorithm, the maximum amplitude of the modulating sequence
and the maximum ripple that can be allowed in the NP. Still, it is important to point out that this
is only a qualitative analysis. In reality, in steady state, the NP voltage has mainly a
thirdharmonic ripple component, and no fixed dc offset. To exactly determine the wave shape
of the voltage in the NP (in order to find the stability boundaries), it would be necessary to solve
(524) analytically, which is very hard if not impossible, even if the form of the solution is
known in advance.
124
5.4 Implementation of FeedForward Algorithm for Space Vector Modulation
The effects of the shift in the NP on the location of the voltage space vectors were discussed
earlier in this chapter. The modifications in SVM algorithm that can help compensate for the
imbalance will now be discussed. First consider how to synthesize the reference vector located
in the sixtydegree region from Figure 5.16.
T
5
T
7
T
6
T
4
T
2
T
10
T
1
d
L
d
S1
d
S0
d
M
d’
S1
d’
S0
d’
L
T
3
T
9
T
8
Figure 5.16. The effect of NP unbalance on the boundaries of small triangles.
There are several issues that complicate the implementation of the SVM feedforward control:
• The boundaries between the small triangles that change as the vector locations change
(due to NP ripple). In fact, because of the splitting of the small vectors, there are now ten
inner triangular regions within the sixtydegree region, instead of only four.
• Because the vectors change their locations, the fast SVM algorithm can not be used since
it relies on the integer values of the voltage space vectors’ coordinates. Likewise, the
125
other algorithms can not rely on the fixed locations of the vectors to simplify the duty
cycle calculation.
Perhaps the most troublesome aspect is the identification of the current sector because of the
splitting of the small vectors. One way to overcome this problem is based on the virtual small
vector approach shown in Figure 5.17. The basic idea of the virtual small vector approach is that
the NP controller supplies the ratio between the positive and negative small vectors’ switching
combinations, therefore determining the amplitude of the equivalent small vector amplitude
(536) 1 S 0 S V ) c 1 ( V c H
→ → →
⋅ − + ⋅ · where [ ] 1 , 0 c ∈ .
This method is approximate, because it uses the NP controller command from the previous
switching cycle, but from the implementation point of view it might be the most practical
implementation. In addition, because the switching frequency is usually very high for the slow
dynamics of dclink capacitors, it is probably more than acceptable to use the previous switching
cycle value of the NP controller.
But, for the sake of argument, consider the approach that can both compensate for the
disturbance in the NP and maintain the NP current at zero on the switchingcycle level. It is
possible to meet both these requirements in the operating region we have identified in the
previous chapter. Therefore, in order to both maintain NP current at zero during each switching
cycle and to compensate for the disturbance, an iterative approach might be required, due to the
troublesome “splitting” of the small vectors.
In fact, in order to select the small triangle, its boundary needs to be known, but the boundary
is not known before the duty cycles of the small vectors are selected. However, the duty cycles
of the small vectors depend on the boundary as well, and are computed from a different set of
equations for each small triangle. This means that the triangular region must be guessed initially,
and than all the equations are solved and then checked to be sure the boundary conditions are
satisfied. This is hardly a practical approach.
126
→
H
L
V
→
M V
→
REF
V
→
1 S
V
→
0 S
V
→
Figure 5.17. The active sector computation based on the home vertex algorithm.
Returning to the approach using the virtual small vector, it is clear that when the location of
the small vector is known (or better to say predetermined by the NP controller), then all the small
triangle boundaries are known, which allows the duty cycles to be computed, as follows:
(537) M
M
L
L H
REF V d V d H d V
→ → → →
⋅ + ⋅ + ⋅ · , where
(538) 1 d d d
M L H
· + + .
Finally, the small vectors’ duty cycles are then easy to find, as follows:
(539)
H 0 S
d c d ⋅ · ,
H 1 S
d ) c 1 ( d ⋅ − · .
127
This concludes the implementation of the proposed feedforward algorithm, for the case of the
outer triangular region. If vector
→
*
V is located in one of the inner triangle region the algorithm
remains the same. The only difference is that there are two home vertices to take into account in
determining the boundary regions and in dutycycle computations.
5.5 Conclusions
This chapter introduces and analyzes the feed forward algorithm as a method to successfully
compensate for NP voltage ripple of any size and thus allow the tradeoff between the voltage
capacity of the switching devices and the size of the dclink capacitors.
The influence of the new algorithm on the control of the NP is studied using the example of
the sine triangle modulator. The influence of the modulator on the NP controller is shown by the
average models developed at two different time scales: first at the fast scale (the time scale of the
switching frequency); then those results were used to develop an average model on the slow
scale (the time scale of the linecycle average). The linecycle average model of the sine triangle
modulator’s influence on the control of the NP is then linearized around the operating point and
can be used to design on NP voltage controller.
In addition, the qualitative analysis shows that the feedforward algorithm introduces an
additional stabilizing effect into the NP control in the inverter mode of operation. In the rectifier
mode of operation, the same analysis shows that the feedforward control introduces a
destabilizing effect, which can essentially limit the voltage gain of the system.
128
6 FEEDFORWARD CONTROL OF THREELEVEL NEUTRALPOINTCLAMPED
CONVERTERS—SIMULATION AND EXPERIMENTAL RESULTS
The goal in this chapter is to continue the analysis of the threelevel converter modulator with
feedforward control using numeric simulation tools to gain some further insight. As in the
previous chapter, the system is analyzed on two time scales by using the linecycle average and
the switchingperiod average. The general approach is to develop models, verify the models
with the experimental results, and then use the models to verify the system operation outside the
region of the experimental results.
The full dynamic simulation confirms the qualitative stability analysis from the previous
chapter, and proves that the influence of the feedforward control the rectifier mode causes
instability only in extreme cases with unrealistically large voltage ripple and high power factor
correction (PFC mode of operation). Furthermore, the danger of instability is lessened by the
fact that the higher the power factor, the smaller the ripple in the NP, therefore, the smaller the
chance of instability. In addition, in the inverter mode of operation, the feedforward control has
a slight stabilizing effect, and under no circumstances does the loss of NP control occur.
129
6.1 Analysis and Experimental Verification of the Line Cycle Average Model of
the FeedForward Algorithm
The linecycle average method of analysis leaves the impression that the modulator does
nothing other than provide a current gain. This might not be at all obvious when the nonlinear
and heavily discontinuous NP current is considered. For that reason, it is perhaps best to verify
this underlying assumption of the linecycle averaging method by measuring the loop gain of the
NP voltage controller. The experimental hardware is the SMES hardware with the NP controller
modified, as shown in Figure 6.1, in order to allow the loopgain measurements with the network
analyzer. The modulator is implemented using the fast SVM algorithm. The converter operates
in the inverter mode with 80V dclink voltage and with a 45degree power factor inductive load
(R=.72Ω, L=1.8mH).
Figure 6.2 shows the controller loop gain for different modulation indexes (output voltage and
output current) of the converter. The loop gain measurement results H(s)=A(s)/B(s) show only
the firstorder response, with the phase slowly increasing with the increase of frequency, which
occurs as a consequence of the delay in the control loop. The firstorder response can be
observed for other power factor loads as well. This proves that the assumption from the previous
chapter, that the modulator together with the load is essentially a nonlinear gain (512), is correct
and can be used for the closedloop NP controller design.
130
p
n
i
NP
C
DC
C
DC
i
a
i
b
i
c
o
L
a
L
b
L
c
Digital
Modulator
V
NP
R
a
R
b
R
c
ThreeLevel
Converter
Switching
Network
V
pn
/2
D/A D/A
perturbation
A B
Digital Controller
A/D
+ +

Figure 6.1. Experimental measurement of the loop gain in the NP controller.
131
10
1
10
2
10
3
40
20
0
20
40
f[Hz]
G
a
i
n
[
d
B
]
m=.5
10
1
10
2
10
3
160
140
120
100
80
f[Hz]
P
h
a
s
e
[
d
e
g
]
10
1
10
2
10
3
40
20
0
20
40
f[Hz]
G
a
i
n
[
d
B
]
m=.7
10
1
10
2
10
3
200
100
0
100
200
f[Hz]
P
h
a
s
e
[
d
e
g
]
10
1
10
2
10
3
40
20
0
20
40
f[Hz]
G
a
i
n
[
d
B
]
m=.9
10
1
10
2
10
3
200
100
0
100
200
f[Hz]
P
h
a
s
e
[
d
e
g
]
Figure 6.2. Measurement of the NP control loop gain without normalization.
6.1.1 Control Authority Over the NP Current, with and without FeedForward Terms
Control authority of the zerosequence term for the sine triangle modulated converters can best
be observed by studying the expressions for linecycle average currents without (61) and with
(62) feedforward correction. Expressions (61) and
(62) give the average current that flows into the NP under the control of the modulating
signal’s zero sequence δ. Both expressions assume there is either a very small ripple or none at
all in the NP. Figure 6.3 shows the linecycle average of the current flowing into the NP when
the dclink capacitors are balanced. The figure shows the symmety around the zero value of the
132
zero sequence δ. For the small, and large modulation index, A, the current that can be
commanded into or out of the NP is rather small, because the difference between the upper and
lower duty cycle is small. As the modulation index approaches the point A=.5, the control
authority reaches its maximum.
(61)
]
]
]
⋅ + − ⋅ ⋅
⋅
⋅ −
·
−
A
sin A A ) cos(
A
I 3
i
1 2 2 2
NP
δ
δ δ φ
π
The zero sequence’s control authority over the NP in the presence of the constant
feedforward terms is as expected not symmetric around zero value of the zero sequence because
of the presence of the feed forward terms. The expression for the line cycle average current
(62) is plotted for two different feed forward terms in Figure 6.4. Again, the zero sequence
exhibits the most control authority over the duty cycle around A=0.5.
(62)
[
] )
A
( sin ) K K ( A 2
) K K ( A A ) 4 2 K 4 ( 2
K K A 4
) cos( I 3
i
1
n p
2
p n
2 2 2
n
p n
NP
δ
π δ δ
π
φ
−
⋅ + ⋅ ⋅ +
+ − ⋅ ⋅ − − ⋅ − ⋅ + ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅
⋅ ⋅ −
·
133
 1  0 . 8  0 . 6  0 . 4  0 . 2 0 0 . 2 0 . 4 0 . 6 0 . 8 1
 0 . 8
 0 . 6
 0 . 4
 0 . 2
0
0 . 2
0 . 4
0 . 6
0 . 8
del t a
A=0.5
A=0.7
A=0.3
A=0.1
A=0.9
Figure 6.3. The linecycle average NP current as a function of the zerosequence component δ.
134
1  0 . 8  0 . 6  0 . 4  0 . 2 0 0 . 2 0 . 4 0 . 6 0 . 8 1
1
 0 . 8
 0 . 6
 0 . 4
 0 . 2
0
0 . 2
0 . 4
0 . 6
0 . 8
de l t a
A=0.5
A=0.7
A=0.3
A=0.1
A=0.9
Kp=1.1
Kn=0.9
1  0 . 8  0 . 6  0 . 4  0 . 2 0 0 . 2 0 . 4 0 . 6 0 . 8
 0 . 8
 0 . 6
 0 . 4
 0 . 2
0
0 . 2
0 . 4
0 . 6
0 . 8
1
de l t a
1  0 . 8  0 . 6  0 . 4  0 . 2 0 0 . 2 0 . 4 0 . 6 0 . 8 1
 0 . 8
 0 . 6
 0 . 4
 0 . 2
0
0 . 2
0 . 4
0 . 6
0 . 8
1
de l t a
A=0.5
A=0.7
A=0.3
A=0.1
A=0.9
Kp=0.9
Kn=1.1
Figure 6.4. The linecycle average NP current with feedforward terms included.
135
6.2 Analysis of the SwitchingCycle Average Model of the Sine Triangle
Modulator with the FeedForward Algorithm
All the analysis performed so far included one key assumption: that the capacitance of the dc
link capacitors is sufficient to prevent a significant ripple. Thus way, the influence of the
possible ripple voltage could be neglected in order to study the limitations of the modulation in
the sense of controlling the NP current on the switchingcycle level and on the linecycle level.
The “no ripple” approach was chosen because of the complexity of the modulator and the
inherent discontinuities in modulation, which were not only difficult to solve analytically, but
were also very hard to explain intuitively. In addition, anything other than a constant zero
sequence is just too difficult to include in the analytical expressions for the linecycle average
current.
However, the simplified model has its own limitations, which present themselves mainly in
the conservative stability estimate in the case of the analysis of the feedforward algorithm as
well as in the inability to predict the NP ripple’s effect on the output line voltage. The full
dynamic simulation of the converter and the load will overcome these limitations. Finally, some
of the simulation results will be verified experimentally.
6.2.1 Average Model of ThreeLevel Converter with Sine Triangle Modulator
Figure 6.5 shows the Matlab Simulink average model of the threephase converter with the R
L load. The only assumption in this model is that the there is an ideal dcvoltage source at the dc
link. The model consists of three phase leg average models, a model of load, and an NP
controller, which is essentially a comparator that applies maximum positive or maximum
negative zero sequence (maximum, so that the modulator is not saturated at any given moment),
depending on the sign of the NP voltage error. There are two load models that were used for the
simulations: The first model used for the simulations of the inverter mode of operation was a
136
dynamic model of the balanced threephase RL; the second load model used for the rectifier
mode of operation was three sinusoidal phaseshifted current sources.
Vdc
delta_Vnp
mod_index
np_control
feed_forward_enable
Vout
Vref
d_np
d_pos
d_neg
phase_C
Vdc
delta_Vnp
mod_index
np_control
feed_forward_enable
Vout
Vref
d_np
d_pos
d_neg
phase_B
Vdc
delta_Vnp
mod_index
np_control
feed_forward_enable
Vout
Vref
d_np
d_pos
d_neg
phase_A
delta_Vnp_error
modulation_index
np_control
np_controller
Sum5
va
vb
vc
d_np_A
d_np_B
d_np_C
delta_Vnp
RL load and
np_capacitance
.9
Mdulation Index
80
Input Voltage
Vdc
Feed Forward Control 0
Constant8
[np_voltage]
3
[mod_index]
2
[mod_index]
[np_voltage]
Zero Sequence
Figure 6.5. Average model of the sine triangle modulator with feed forward.
Figure 6.6 shows the implementation of the average model of the phase leg. Essentially, the
modulating sinusoid is scaled with the modulating index, added with the zero sequence, and then
the duty cycles and the output voltage are computed from the average equations, as in (516) and
(517).
137
5
d_neg
4
d_pos
3
d_np
2
Vref
1
Vout
Sum
Sine Wave
Saturation
Product
Mux
Mux
Demux
Demux
MATLAB
Function
Average model of
phase leg modulator
5
feed_forward_enable
4
np_control
3
mod_index
2
delta_Vnp
1
Vdc
Figure 6.6. Average model of one phase leg sine triangle modulator.
Finally, the bangbang NP controller is shown in Figure 6.7. This implementation of the
controller is probably not optimal; nevertheless, in the case of the average model, the harmonic
performance of the converter is not of interest.
1
np_cont rol
MATLAB
Funct i on
bang bang
np_cont rol l er
Mux
Mux4
2
modul at i on_i ndex
1
del t a_Vnp_error
Figure 6.7. Model of NP bangbang controller.
6.2.2 Performance of FeedForward Algorithm, Simulation Results
The qualitative analysis and the expressions for the NP current obtained through the linecycle
averaging indicate that there is a potential to lose control of the NP in the presence of the feed
forward control. This danger exists only in the rectifier mode of operation, in which the
converter has the tendency to further charge the capacitor that is already charged to a higher
138
potential. For this reason the system is simulated in the rectifier mode of operation for different
power factors’ ac currents.
Figure 6.8 shows the rectifier mode of operation with the 85degree power factor angle. The
amplitude of the phase current is 80A, the dclink voltage is 80V dc, and the modulation index is
unfavorably high at m=0.95. All the parameters closely resemble the system parameters from
practical experiments. The first graph is line voltage, the second is the line current, the third the
NP voltage and the last is the feedforward enable command. The effectiveness of the feed
forward algorithm is clear. Not only was the linevoltage waveform dramatically improved and
completely cleaned of harmonic components, but the NP ripple size was significantly reduced for
this operating point. Furthermore, it is important to notice that the operating conditions are
chosen to have an unrealistically large ripple (dc capacitors have a 75% voltage swing) in the NP
in order to more clearly demonstrate the effectiveness of the feedforward algorithm.
As many simulations show, the feedforward algorithm does not critically affect the stability
of the NP. There is a very narrow region with a very high NP ripple, high power factor (mostly
active power) and high modulation index (very little zerosequence control) when the instability
may occur. Figure 6.9 shows one boundary for which the feedforward algorithm causes the loss
of NP control. The operating conditions are the same as in previous simulations, except that the
power factor angle is 0. The simulation shows that the converter operates with no distortion in
the average line voltage, even when the dc link collapses completely.
139
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
100
50
0
50
100
V
a
b
[
V
]
t[s]
NP Control in Rectifier Mode of Operation
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
100
0
100
i
a
b
[
A
]
t[s]
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
40
20
0
20
40
V
n
p
[
V
]
t[s]
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
0.5
0
0.5
1
1.5
F
e
e
d
F
o
r
w
a
r
d
t[s]
Figure 6.8. Rectifier mode of operation with low power factor.
140
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
100
50
0
50
100
V
a
b
[
V
]
t[s]
NP Control in Rectifier Mode of Operation
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
100
0
100
i
a
b
[
V
]
t[s]
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
40
20
0
20
40
V
n
p
[
V
]
t[s]
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
0.5
0
0.5
1
1.5
F
e
e
d
F
o
r
w
a
r
d
t[s]
Figure 6.9. Rectifier mode of operation for high power factor.
Finally, the algorithm is simulated in the inverter mode of operation with the dynamic model
of RL load, where R= 0.06Ω, L=1.8 mH (85degree power factor angle), and with a high
modulation index, m=0.95. The simulation results show an improvement in the quality of the
line voltage, while the NP voltage remains unaltered.
141
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
100
50
0
50
100
V
a
b
[
V
]
t[s]
NP Control in Inverter Mode of Operation
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
100
0
100
i
a
b
[
V
]
t[s]
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
40
20
0
20
40
V
n
p
[
V
]
t[s]
0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.2
0.5
0
0.5
1
1.5
F
e
e
d
F
o
r
w
a
r
d
t[s]
Figure 6.10. The inverter mode of operation with low power factor load.
6.3 Average Model of Fast Space Vector Modulator, Experimental Verification of
the Model
Because the simulation results of the average model of threelevel converter with the sine
triangle modulator did not exactly match the experimental results, the average model of the fast
space vector modulator is developed next. It is important to mention that the discrepancy was
only to be expected, because the modulator of the experimental system was implemented using
SVM with the NP control based on the selection of redundant small vectors and not on control of
the zero sequence. This leads to the discrepancy in the zero sequence of the space vector
modulated experimental hardware and the sine triangle modulated simulation results.
142
6.3.1 Average Model of the ThreeLevel Converter with Fast Space Vector Modulation
Algorithm
Figure 6.11 shows the Matlab Simulink model of the average model of the threephase
converter with the RL load. The difference from the previous model is the modulating
technique. In this case, the fast SVM algorithm is implemented. As in the previous model, the
main assumption is the constant dclink voltage. The model implementation is relatively
straightforward: there are two essential blocks, the modulator and the load. There is still a
coordinate transformation block, which converts the rotating reference vector from the
orthogonal to the nonorthogonal (g, h) coordinate system, and the integrator that uses the NP
current to compute the NP voltage ripple.
ZeroOrder
Hold
0
Vdc2
80
Vdc
Sum3
1/3300e6
SMES_capacitance
va
vb
vc
ia
ib
ic
RL load
Mux
Mux
.9
Modulation Index
s
1
Integrator4
[A]
Goto
[A]
From
g
h
ia
ib
ic
Vdc
Vnp_error
Va
Vb
Vc
Inp
Fast 3level modulator
Demux
Demux
mod_index
g
h
Coordinate Transformation
Figure 6.11. Average model of the fast SVM algorithm.
Finally, Figure 6.12 shows some details of the implementation of the fast space vector
modulator. There are three functions implemented. The first function finds the nearest three
vectors and their corresponding duty cycles, then a switchingstate selection function converts
143
the vectors given in (g, h) coordinates into the switching combination and selects between the
redundant vectors in order to balance the NP. Finally, the third and last function adds together
the products of duty cycles and voltages to find the average phase voltage and adds the NP
currents from the three vectors in order to find the resulting average NP current.
4
Inp
3
Vc
2
Vb
1
Va
Mux
third_vector
MATLAB
Function
sw_state_selection2
MATLAB
Function
sw_state_selection1
MATLAB
Function
sw_state_selection
Mux
second_vector
MATLAB
Function
phase voltages
Mux
first_vector
MATLAB
Function
find_NTV
Mux
Mux2
Mux
Mux1
Mux
Mux
Demux
Demux2
Demux
Demux1
7
Vnp_error
6
Vdc
5
ic
4
ib
3
ia
2
h
1
g
Figure 6.12. Fast threelevel space vector modulator.
6.3.2 The Fast Space Vector Modulation Algorithm, Experimental Results
Figure 6.13 shows the NP voltage for several different modulation indexes for the 80V dclink
voltage and for the RL load, where R=0.06 Ω and L=1.8 mH (making a power factor angle of
85 degrees). The figures in the left column are the measurement results; the figures in the right
column are the simulation results with the average model of the space vector modulator. The
two columns seem to be in excellent agreement.
144
0.01 0.005 0 0.005 0.01 0.015 0.02 0.025
30
20
10
0
10
20
30
t[s]
V
n
p
[
V
]
m=.5
0.01 0.005 0 0.005 0.01 0.015 0.02 0.025
30
20
10
0
10
20
30
t[s]
V
n
p
[
V
]
m=.7
0.01 0.005 0 0.005 0.01 0.015 0.02 0.025
30
20
10
0
10
20
30
t[s]
V
n
p
[
V
]
m=.9
0.1 0.11 0.12 0.13 0.14 0.15
30
20
10
0
10
20
30
t[s]
V
n
p
[
V
]
m=.5
0.1 0.11 0.12 0.13 0.14 0.15
30
20
10
0
10
20
30
t[s]
V
n
p
[
V
]
m=.7
0.1 0.11 0.12 0.13 0.14 0.15
30
20
10
0
10
20
30
t[s]
V
n
p
[
V
]
m=.9
Figure 6.13. Measured and simulated NP ripple of a threelevel converter.
Figure 6.14 shows the measured and simulated linetoline voltage for the same operating
point as in the previous figure. It is interesting to observe the strong thirdharmonic component
in the linetoline voltage. The strong third harmonic in the linetoline voltage gets canceled for
the balanced load, resulting in nearly perfect phase currents. Figure 6.15 shows simulated line
toline currents for the same operating condition as in the previous two figures. The
measurement results are not shown because they were also nearly perfect sinusoids. This fact
points out an interesting new aspect of the design optimization of threelevel converters; It
might be possible to design threelevel converters for certain with requirement for a very low
dclink capacitance applications and very simple controller, which might even operate without
the feedforward algorithm and with only a basic NPbalancing scheme.
145
0.01 0.005 0 0.005 0.01 0.015 0.02 0.025
100
50
0
50
100
t[s]
V
a
b
[
V
]
m=.5
0.01 0.005 0 0.005 0.01 0.015 0.02 0.025
100
50
0
50
100
t[s]
V
a
b
[
V
]
m=.7
0.01 0.005 0 0.005 0.01 0.015 0.02 0.025
100
50
0
50
100
t[s]
V
a
b
[
V
]
m=.9
0.1 0.11 0.12 0.13 0.14 0.15
100
50
0
50
100
t[s]
V
a
b
[
V
]
m=.5
0.1 0.11 0.12 0.13 0.14 0.15
100
50
0
50
100
t[s]
V
a
b
[
V
]
m=.7
0.1 0.11 0.12 0.13 0.14 0.15
100
50
0
50
100
t[s]
V
a
b
[
V
]
m=.9
Figure 6.14. Measured and simulated line voltage in the presence of NP ripple.
146
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14 0.145 0.15
100
50
0
50
100
t[s]
i
a
[
A
]
m=.5
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14 0.145 0.15
100
50
0
50
100
t[s]
i
a
[
A
]
m=.7
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14 0.145 0.15
100
50
0
50
100
t[s]
i
a
[
A
]
m=.9
Figure 6.15. Simulated linetoline currents for balanced and symmetric load.
6.4 Conclusions
This chapter verifies the essential assumption of the linecycle averaging method through
experiment. The experiment proves that for the purpose of NP control the inverter together with
the load can be considered a variablegain current source with no dynamics. With the linecycle
averaging technique experimentally proven, the next step is the stability study based on the
expressions for NP current obtained through linecycle averaging. The expressions for the NP
current in the presence of the feedforward control show that the stability is compromised only in
the rectifier mode with a certain amount of voltage error in the NP and for operation with either
very low or very high modulation indexes.
147
Because the expression for linecurrent averaging assumes a constant dclink imbalance and a
constant zero sequence, it seemed likely that the danger of NP control stability problems was
exaggerated. That is the reason why a switchingcycle average model of the converter was
implemented and used to study the stability in the presence of a significant voltage ripple in the
NP. The full dynamic simulations show not only the excellent performance of the feedforward
algorithm, but also that it is unlikely to have the stability problems in any but the most extreme
conditions. The NP control stability problems were observed only with the NP ripple higher than
50% of the nominal split dclink capacitor voltage, while at the same time the converter operated
at high modulation index and in PFC (or near PFC operation).
Finally, although it was expected that the dynamic model of the converter with sine triangle
modulator was not going to exactly match the experimental waveforms of the converter with the
fast SVM algorithm, a dynamic model of the converter with the space vector modulator was
developed as well. The results of the simulations using the space vector modulator almost
exactly match the experimental results, and in that way demonstrate the validity of the switching
cycle averaging method.
148
7 CONCLUSIONS
This work discusses space vector modulation and the control of multilevel power electronics
converters, and is really the result of generalizations and insights gained from the theoretical
study of a practical problem, such as the design of a power conditioning system for
superconductive magnetic energy storage. The engineering challenge in designing the power
conditioning system (or more precisely, in the design of the digital controller for the power
conditioning system) was to enable the power conditioning hardware to control the bidirectional
power flow between the magnet and the utility line. In fact, the challenge was to control the
power conditioning system in order to provides stable, controllable and highquality power on
both the magnet and utility sides.
The first engineering part of this research shows that a power conditioner based on the
threelevel converter topology is feasible, and that its control design can provide a stable,
highbandwidth, bidirectional control of the power flow. In addition, the control design
approach in which the NP controller was designed separately from the main power stage was
theoretically and practically justified.
During the modeling and hardware development phase of the digital controller, two main
problems presented themselves as worthwhile research topics. One problem is the excessive
complexity of the implementation of the SVM algorithm, particularly for converters with a
higher number of levels. The other one is more specific and deals with the issue of controlling
149
the NP of the threelevel converters; This is a very important aspects of the threelevel converter
design.
Accordingly, Chapter 3 introduces the new SVM algorithm that is computationally very
efficient and applicable to converters with any number of levels. The computational efficiency
of the algorithm comes from its simplicity, and it is important to mention that not only can the
same algorithm used for converters with any number of levels, but the number of steps in the
algorithm remains the same regardless of the number of levels in the converter. These features
make the new algorithm not only a good design and implementation choice, but also a very good
tool for teaching the mathematical basis of multilevel SVM.
The research on NP control and the issues of the imbalance in the NP control are approached
from several different angles. The initial approach taken was to develop the model of the space
vector modulator for the purpose of investigating the NP current ripple. The developed model is
general in the sense that it can describe any SVM, and its effects on the NP. Using the
developed model (under the assumption of the infinite dclink capacitance), the model presents
the properties of the NP control in an intuitive way. In addition, a region in which the dc link
can be balanced on the switchingcycle level was identified, and a quantitative study of the NP
balancing offered as an essential tool in designing the size of the dclink capacitors. Finally,
some of these results were verified experimentally.
The study of the properties of the NP and the ripple in the NP led to the following idea: If
there is a way to compensate for the effect of the NP ripple in the control, than it might be
possible to allow designers the freedom for a new kind of design tradeoff between the size of the
dclink capacitors and the increased voltage capacity of the switching devices.
The result of this study is a feedforward algorithm, which can actively compensate for the
ripple in the NP so that it does not propagate and affect the output of the converter. The
algorithm is proposed for both the carrierbased modulation and for SVM. Subsequent analysis
was performed with multiple tools, including linecycle average, switchingcycle average, and
finally, with the dynamic simulation. The study shows that the feedforward algorithm (although
150
it affects the NP control) does not require redesign of the NP controllers; it compensates for the
effects of the imbalance in the NP, and in some cases reduces the NP ripple itself.
Unfortunately, in the rectifier mode of operation and under the most extreme conditions (high
modulation index, higher than 50% ripple and high power factor current), there exists the
possibility that the NP control algorithm can cause the loss of NP control. In that case the
amount of feedforward action can be adjusted in order to avoid stability problems.
Finally, there are several interesting topics for further research:
• The digital controller for the SMES power conditioning system was built as a centralized
controller, while using the power electronics building block (PEBB) approach. In the future,
it would be interesting to implement a distributed controller approach into the design of the
digital controller like in [F17, F18], thus standardizing the signal distribution network
throughout the converter.
• The research in the application of the new fast SVM algorithm still has potentially very
interesting research opportunities. Perhaps the most immediate research topic can be found
in the attempt to generalize the algorithm to converters with more than three phases by
exploiting the hidden symmetries in the higherdimensional spaces of the more these
converters. Another interesting opportunity can be found in the challenge to generalize and
organize the conversion of the switching vectors back into the switching states.
• With respect to modeling the space vector modulator for the purpose of NP balancing, the
model (although general) does not allow the ripple in the NP. It might be interesting to
generalize the modeling approach to include the NP ripple if the dclink capacitors can not be
assumed to be infinitely large.
Finally, with respect to the feedforward algorithm, it might be interesting to determine how
much can really be saved by trading off the increased voltage rating of the switches for the
decreased capacitance of the dclink capacitors. In addition, for the more confident application
151
of the algorithm, it is important to make a more detailed study of the conditions that cause
instability in the rectifier mode of operation.
152
REFERENCES
A. Multilevel Power Converters: Issues and Tradeoffs
[A1] Pradeep M. Bhagwat, V. R. Stefanovic, “Generalized Structure of a Muiltilevel PWM
Inverter,” IEEE Transactions on Industry Applications, pp. 10571069, vol. IA19, no. 6,
November/December 1983.
[A2] M. Marchesoni, M. Mazzucchelli, “Multilevel Converters for High Power AC Drives: A
Review,” IEEEISIE International Symposium, pp.3843, 1993.
[A3] C. Hochgraf, R. Lasseter, D. Divan, T. A. Lipo, “Comparison of Multilevel Inverters for
Static Var Compensation,” IEEEIAS Conference Record, pp. 921 –928, vol.23, 1994.
[A4] JihSheng Lai, Fang Zheng Peng, “Multilevel Converters  A New Breed of Power
Converters,” IEEE Transactions on Industry Applications, pp.509517, vol. 32, no. 3,
May/June 1996.
[A5] T. A. Meynard, H. Foch, “Multilevel Converters and Derived Topologies for High Power
Conversion,” IEEEIECON Conference Record, pp. 2126, vol. 1, 1995.
[A6] B. S. Suh, G. Sinha, M. D. Manjrekar, T. A. Lipo, “Multilevel Power Conversion – An
Overview of Topologies and Modulation Strategies,” IEEEOPTIM Conference Record,
pp. 1124, vol. 2, 1998.
[A7] M. Marchesoni, M. Mazzzucchelli, P. Tenca, “About the DCLink Capacitors Voltage
Balance in MultiPoint Clamped Converters,” IEEEIECON Conference Record, vol. 1,
pp.548553, 1998.
[A8] Leon M. Tolbert, Fang Z. Peng, “Multilevel Converters for Large Electric Drives,” IEEE
APEC Conference Record, pp. 530536, 1998.
[A9] C. Newton, M. Summer, “Multilevel converters, a real solution to medium/highvoltage
drives?,” Power Engineering Journal, pp. 2126, vol. 12.1, 1998.
[A10] M. Manjrekar, G. Venkataramanan, “Advanced topologies and modulation strategies for
multilevel inverters,” IEEEPESC Conference Record, vol. 2, pp. 10131018, 1996
153
[A11] N.P. Schibli, T. Nguyen, A. C. Rufer, “A threephase multilevel converter for highpower
induction motors,” IEEE Transactions on Power Electronics, vol. 13, pp. 978 –986, Sept.
1998.
[A12] N. Celanovic, D. Boroyevich, “A Fast Space Vector Modulation Algorithm for
Multilevel ThreePhase Converters,” IEEEIAS, 1999.
[A13] Y. Shakweh, E. A. Lewis, “Assessment of medium voltage PWM VSI topologies for
multimegawatt variable speed drive applications,” IEEEPESC Conference Record, vol.
2, pp. 965 –971,1999.
B. ThreeLevel NeutralPointClamped Converters
[B1] A. Nabae, I. Takahashi, H. Akagi, “A New NeutralPoint Clamped PWM Inverter,” IEEE
Transactions on Industry Applications, pp. 518523, vol. IA17, no.5, September/October
1981.
[B2] S. Ogasawara, H. Akagi “A Vector Control System Using a NeutralPointClamped
Voltage Source PWM Inverter,” IEEEIAS Conference Record, pp. 422427, 1991.
[B3] J. Steinke, “Switching Frequency Optimal PWM Control of a ThreeLevel Inverter,”
IEEE Transactions on Power Electronics, vol. 7, no. 3, July 1992.
[B4] S. Ogasawara, H. Akagi, “Analysis of Variation of Neutral point Potential in Neutral
PointClamped Voltage Source PWM Inverters,” IEEEIAS Conference Record, pp. 965
970, 1993.
[B5] G. Lipphardt, “Using a ThreeLevel GTO Voltage Source Inverter in a HVDC
Transmission System,” EPE Conference Record, pp.151155, September 1993.
[B6] H. L. Liu, G. H. Cho, “ThreeLevel Space Vector PWM in Low Index Modulation
Region Avoiding Narrow Pulse Problem,” IEEE Transactions on Power Electronics, pp.
481486, vol. 9, no. 5, September 1994.
[B7] M.C. Klabunde, Yifan Zhao, T.A. Lipo, “Current control of a 3level rectifier/inverter
drive system,” IEEE Transactions on Power Electronics, pp. 5765, vol. 11, no. 1,
January 1996.
[B8] J. Zhang, “High Performance Control of a ThreeLevel IGBT Inverter fed AC Drive,”
IEEE Industry Applications Society, pp. 2228, vol. 1, 1995.
[B9] R. Rojas, T. Ohnishi, T. Suzuki, “PWM Control Method for NPC Inverters with Very
Small DCLink Capacitors,” IPECYokohama, pp. 494499, 1995.
[B10] Roberto Rojas, Tokuo Ohnishi, Takayuki Suzuki “An Improved Voltage Vector Control
Method for NeutralPointClamped Inverters,” IEEE Transactions on Power Electronics,
pp. 666672, vol.10, no. 6, November 1995.
[B11] M. Cosan, H. Mao, D. Borojevic, F.C. Lee “Space Vector Modulation of ThreeLevel
Voltage Source Inverter,” VPEC Seminar Proceedings, pp. 123128, 1996.
[B12] G. C. Cho, G. H. Jung, N. S. Choi, G. H. Cho, “Analysis and Controller Design of Static
Var Compensator Using ThreeLevel GTO Inverter,” IEEE Transactions on Power
Electronics, pp. 5765, vol. 11, no. 1, January 1996.
[B13] Y.H. Lee, B.S. Suh, D.S. Hyun, “A Novel PWM Scheme for a ThreeLevel Voltage
Source Inverter with GTO Thyristors,” IEEE Transactions on Industry Applications, pp.
260268, vol. 32, no. 2, March/April 1996.
154
[B14] C. Newton, M. Sumner, “Neutral Point Control for MultiLevel Inverters: Theory, Design
and Operational Limitations,” IEEEIAS Conference Record, pp. 13361343, 1997.
[B15] B. Kaku, I. Miyashita, S. Sone “Switching Loss Minimized Space Vector PWM Method
for IGBT threelevel inverter,” IEE Proceedings on Electric Power Applications, pp. 182
190, Vol. 144, No. 3., May 1997.
[B16] H. Mao, D. H. Lee, H. Dai, F.C. Lee, D. Borojevic “Evaluation and Development of New
Power Electronic Technologies foe Superconductive Magnetic Energy Storage (SMES)
using PEBB”, VPEC Seminar Proceedings, pp. 129134, 1997.
[B17] S. Fukuda, K. Suzuki, ”Harmonic Evaluation of CarrierBased PWM Methods Using
Harmonic Distortion Determining Factor,” IEEEPCC Conference Record, pp.259264,
1997.
[B18] D. H. Lee, S. R. Lee, F. C. Lee, “An Analysis of Midpoint Balance for the NeutralPoint
Clamped ThreeLevel VSI,” IEEEPESC Conference Record, 1998.
[B19] Y.H. Lee, B.S. Suh, C.H. Choi, D.S. Hyun, “A New Neutral Point Current Control for
a 3level Converter/Inverter Pair System,” IEEEIAS Conference Record, 1999.
[B20] N. Celanovic, D. Boroyevich, “ A Comprehensive Study of NeutralPoint Voltage
Balancing Problem in Three Level NeutralPointClamped Voltage Source PWM
Inverters”, IEEEAPEC Conference Record, 1999.
[B21] J. P. Lyons, V. Vlatkovic, P. M. Espelage, F. H. Boettner, E. Larsen, “Innovation IGCT
Main Drives,” IEEEIAS Conference Record, 1999.
[B22] R. Sommer, A. Martens, M. Griggs, H.J. Conraths, M. Bruckmann, T. Greif, “New
Medium Voltage Drive Systems using ThreeLevel Neutral Point Clamped Inverter with
High Voltage IGBT,” IEEEIAS Conference Record, 1999.
[B23] P. K. Steimer, J. K. Steinke, H. E. Gruning, S. Conner, “A Reliable, Interface Friendly
Medium Voltage Drive Based on the Robust IGCT and DTC Technologies,” IEEEIAS
Conference Record, 1999.
C. NeutralPointClamped Converters with more than ThreeLevels
[C1] N. S. Choe, J. G. Cho, G. H. Cho, “A General Circuit Topology of Multilevel Inverter,”
IEEEPESC Conference Record, pp. 96103, 1991.
[C2] R. W. Menzies, P. Steimer, J. K. Steinke, “FiveLevel GTO Inverters for Large Induction
Motor Drives,” IEEE Transactions on Industry Applications, pp. 938944, vol. 30, no. 4,
July/August 1994.
[C3] R. Rojas, T. Ohnishi, T. Suzuki, “PWM Control Method for a FourLevel Inverter,” IEE
Proceedings, pp. 390396, 1995.
[C4] F. Z. Peng, J. S. Lai, J. McKeever and J. VanCoevering, “A Multilevel VoltageSource
Converter System with Balanced DC Voltages,” IEEE 1995.
[C5] G. Sinha, T. A. Lipo, “A Four Level RectifierInverter System for Drive Applications,”
IEEEIAS Conference Record, pp. 980987, 1996.
[C6] G. Sinha, T. A. Lipo, “A Four Level Inverter Based Drive With a Passive Front End,”
IEEEPESC Conference Record, pp. 590596, vol. 1, 1997.
[C7] Y. Chen, B. Mwinyiwiwa, Z. Wolanski, B.T. Ooi, “Regulating and Equalizing DC
Capacitance Voltages in Multilevel Statcom,” IEEE Transactions on Power Delivery, pp.
901907, vol. 12, no. 2, April 1997.
155
[C8] C. Hocgraf, R. H. Lasseter, “A TransformerLess Static Synchronous Compensator
Employing a MultiLevel Inverter,” IEEE Transaction on Power Delivery, vol. 12, no.
2, pp. 881887, April 1997.
[C9] B. Mwinyiwiwa, Z. Wolanski, BoonTeck Ooi, “Multilevel STATCOM with Third
Harmonic Elimination on the DC Link Capacitor Voltages,” IEEEPESC Conference
Proceedings, vol. 1, pp. 317 322, 1997.
[C10] C. Newton, M. Sumner, “A novel Arrangement for the Balancing of the Capacitor
Voltages of a Five Level Diode Clamped Inverter,” IEE Power Electronics and Variable
Speed Drives Conference Record, pp. 465470, 1998.
[C11] W. A. Hill, C. D. Harbourt, “Performance of Medium Voltage MultiLevel Inverters,”
IEEEIAS Conference Record, 1999.
[C12] C. Newton, M. Sumner, “Novel Technique for Maintaining Balanced Internal DC Link
Voltages in Diode Clamped FiveLevel Inverters,” IEE Proc. Electr. Power Appl, vol.
146, no. 3, May 1999.
[C13] L. M. Tolbert, F. Z. Peng, T. G. Habetler, “A Multilevel ConverterBased Universal
Power Conditioner,” Power Electronics Specialists Conference,” IEEEPESC Conference
Record, vol. 1, pp. 393 –399, 1999.
[C14] Y. Chen, B. Mwinyiwiwa, Z. Wolanski, BoonTeck Ooi, “Unified power flow controller
(UPFC) based on chopper stabilized diodeclamped multilevel converters,” IEEE
Transactions on Power Electronics, vol. 15, pp. 258 –267, March 2000.
[C15] L. M. Tolbert, F. Z. Peng, T. G. Habetler, “A multilevel converterbased universal power
conditioner,” IEEE Transactions on Industry Applications, vol. 36, pp. 596 –603, March
April 2000.
[C16] G. Carrara, S. Garadella, M. Marchesoni, R. Salutari, G. Sciutto, “A New Multilevel
PWM Method: A Theoretical Analysis,” IEEE Transactions on Power Electronics, vol.
7, pp. 497505, July, 1992.
[C17] B. Valaerts, P. Mathaus. D. Tatakis, “A Novel Approach to the Generation and
Optimization of ThreeLevel PWM Waveforms,” IEEEPESC Conference Record, pp.
12551262, 1988.
D. Multilevel Cascaded Inverters
[D1] M. Marchesoni, M. Mazzzucchelli, S. Tenconi, “A Non conventional Power Converter
for plasma Stabilization,” IEEE Transactions on Power Electronics, vol. 5, no. 2, April
1990.
[D2] M. Marchesoni, “HighPerformance Current Control Techniques for Applications
Multilevel HighPower Voltage Source Inverters,” IEEE Transactions on Power
electronics, pp. 189204, vol. 7, no. 1, January 1992.
[D3] F. Z. Peng, J. S. Lai, J. McKeever, J. Van Coevering, “A Multilevel VoltageSource
Inverter with Separate DC Sources for Static Var Generation,” IEEEIAS Conference
Record, pp.25412548, 1995.
[D4] P. W. Hammond, “A New Approach to Enhance Power Quality for Medium Voltage AC
Drives,” IEEE Transactions on Industry Applications, pp. 202208, vol. 33, no. 1,
January/February 1997.
156
[D5] Q. Jiang, T. A. Lipo, “Switching Angles and DC Link Voltages Optimization for
Multilevel Cascaded Inverters,” 1998.
[D6] F. Z. Peng, J. W. McKeever, D. J. Adams, “A Power Line Conditioner Using Cascade
Multilevel Inverters for Distribution Systems,” IEEE Transactions on Industry
Applications, vol. 34, no. 6, November/December 1998.
[D7] M. D. Manjrekar, T. A. Lipo, “A Hybrid Multilevel Inverter Topology for Drive
Applications,” IEEEAPEC Conference Record, pp. 523529, vol. 2, 1998.
[D8] M. D. Manjrekar, T. A. Lipo, “A Generalized Structure of Multilevel Power Converter,”
PEDES Conference Record, vol. 1, pp. 6267, 1998.
[D9] M. D. Manjrekar, P. Steimer, T. A. Lipo, “Hybrid Multilevel Power Conversion System:
A Competitive Solution for High Power Applications,” IEEEIAS Conference Record,
1999.
[D10] J. Rodriguez, L. Moran, A, Gonzales, C. Silva, “High Voltage Converter with
Regeneration Capability,” IEEEPESC Conference Proceedings, vol. 2, pp. 10771082,
1999.
[D11] R. H. Osman, “A MediumVoltage Drive Utilizing SeriesCell Multilevel Topology for
outstanding Power Quality,” IEEEIAS Conference Record, 1999.
[D12] BorRen Lin; YuanPo Chien; HsinHung Lu, “Multilevel inverter with series connection
of Hbridge cells,” IEEEPEDS Conference Record, vol. 2, pp. 859 –864, 1999.
[D13] F. Iturriz, P. Ladoux, “Phasecontrolled multilevel converters based on dual structure
associations,” IEEE Transactions on Power Electronics, vol. 15, pp. 92 –102, Jan. 2000.
E. Flying Capacitor Multilevel Converters
[E1] T. A. Meynard, H. Foch, “MultilevelLevel Conversion: High Voltage Choppers and
Voltage SourceSource Inverters,” IEEEPESC Conference Record, pp.397403, 1992.
[E2] BumSeok Suh, DongSeok Hyun, “A Novel Nlevel High Voltage Inversion System,”
IEEE Transaction on Industrial Electronics, vol. 44, no. 1, pp. 107115, February 1997.
[E3] F.R. Dijkhuizen. J. L. Duarte, “Proper Choice of Flying Capacitors Based on Distinct
Power Dissipation Models,” IEEEIAS Conference Record, vol. 2, pp. 11741180, 1998.
[E4] Y. Liang; C. O. Nwankpa,“A Power Line Conditioner Based on Flying Capacitor
Multilevel Voltage Source Converter with Phase Shift SPWM,” IEEEIAS Conference
Record, vol. 4, pp. 2337 2343, 1999.
F. Modeling and Control of ThreePhase Converters
[F1] S. Hiti, V. Vlatkovic, D. Boroyevich, “A new control algorithm for threephase PWM
buck rectifier with input displacement factor compensation,” IEEE Transactions on
Power Electronics, pp. 927933, vol. 92, March 1994.
[F2] S. Hiti, D. Boroyevich, “Control of FrontEnd ThreePhase Boost Rectifier,” IEEE
APEC, pp. 927933, vol. 2, 1994.
[F3] S. Hiti, D. Boroyevich, R. Ambatipudi, R. Zhang and Y. Jiang, “Average Current Control
of ThreePhase PWM Boost Rectifier,” IEEEPESC, pp. 131137, vol. 1, 1995.
157
[F4] S. Hiti, “Modeling and Control of ThreePhase PWM Converters,” Ph.D. Dissertation,
Virginia Polytechnic Institute and State University, Blacksburg, VA, 1995.
[F5] S. Fukuda, A. Sagawa, “Modeling and Control of a Neutral–Point Clamped Voltage
Source Converter,” IPECYokohama, pp. 470475, 1995.
[F6] M. Matsui, “Static VAR Compensator Using Neutral PointClamped PWM Inverter and
its Control Scheme,” IPECYokohama, pp. 488493, 1995.
[F7] G. H. Jung, G. C. Cho, S. W. Hong, Gyu H. Cho “DSP Based Control of High Power
Static VAR Compensator Using Novel Vector Product Phase Locked Loop”, IEEEPESC
Conference Record, pp. 238243, vol. 1, 1996.
[F8] J. Bordonau, M. Cosan, D. Borojevic, H. Mao and F. C. Lee, “A StateSpace Model for
the Comprehensive Dynamic Analysis of ThreeLevel VoltageSource Inverters,” IEEE
PESC, 1997.
[F9] V. H. Prasad, S. Dubovsky, N. Celanovic, R. Zhang and D. Boroyevich, “ DSP Based
Implementation of a Power Electronics Control System,” VPEC Seminar Proceedings,
pp. 6167, 1997.
[F10] F. Z. Peng, J.S. Lai, “Dynamic Performance and Control of a Static Var Generator Using
Cascade Multilevel Inverters,” IEEE Transactions on Industry Applications, pp. 748755,
vol. 33, no. 3, May/June 1997.
[F11] H. Mao, D. Boroyevich, F. C. Lee, “Novel ReducedOrder SmallSignal Model of a
ThreePhase PWM Rectifier and Its Application in Control Design and System
Analysis,” IEEE Transactions on Power Electronics, pp. 511521, vol. 13, no. 3, May
1998.
[F12] R. Zhang, “High Performance Power Converter System for Nonlinear and Unbalanced
Load/Source,” Ph.D. Dissertation, Virginia Polytechnic Institute and State University,
Blacksburg, VA, 1998.
[F13] S. Alepuz, J. Bordonau, J. Peracaula, “Dynamic Analysis of ThreeLevel VoltageSource
Inverters Applied to Power Regulation,” IEEEPESC, vol. 2, pp. 721726, 1999.
[F14] K. D. T. Ngo, “Topology and Analysis in PWM Inversion, Rectification, and
Cycloconversion,” Ph.D. Dissertation, California Institute of Technology, Pasadena, CA,
1984.
[F15] K. D. T. Ngo, “Low Frequency Characterization of PWM Converters,” IEEE
Transactions on Power Electronics, pp. 223230, vol. PE1, 1996.
[F16] P. Bauer, J. B. Klassens, “Dynamic Modeling of AC Power Converters,” IEEE PCC
Conference Record, pp. 502507, 1993.
[F17] I. Celanovic, I. Milosavljevic, D. Boroyevich, R. Cooley, J. Guo, “A New Distributed
Digital Controller for the Next Generation of Power Electronics Building Blocks,” IEEE
APEC Conference Record, 2000.
[F18] I. Celanovic, N. Celanovic, I. Milosavljevic, D. Boroyevich, R. Cooley, “A New Control
Architecture for Future Distributed Power Electronics Systems,” IEEE PESC Conference
Record, 2000.
G. SMES References
158
[G1] H. Okada, T. Ezaki, K. Ogawa, H. Koba, M. Takeo, K. Funaki, S. Sato, “Experimental
Study of SMES System With DC Intertie for Power Line Stabilization,” IEEE PESC, pp.
326333, April 1988.
[G2] R. H. Lasseter, S. G. Jalali, “Power Conditioning Systems for Superconductive Magnetic
Energy Storage,” IEEE Transactions on Energy Conversion, vol. 6, no. 3, pp. 381387,
September 1991.
[G3] R. H. Lasseter, S. G. Jalali, “Dynamic Response of Power Conditioning Systems for
Superconductive Magnetic Energy Storage,” IEEE Transactions on Energy Conversion,
vol. 6, no. 3, pp. 388393, September 1991.
[G4] I. D. Hassan, R. M. Bucci, K. T. Swe, “400 MW SMES Power Conditioning System
Development and Simulation,” IEEE Transactions on Energy Conversion, vol. 8, no. 3,
pp. 237249, July 1993.
[G5] KwaSur Tam, A. Yarali, “Operation Principle and Applications of Multiterminal
Superconductive Magnetic Energy Storage System,” IEEE Transactions on Energy
Conversion, vol. 8, no. 1, pp. 5462, March 1993.
[G6] S. M. Schoenung, W. R. Meier, R. L. Bieri, “Small SMES Technology and Cost
Reduction Estimates,” IEEE Transactions on Energy Conversion, vol. 9, no. 2, pp. 231
237, June 1994.
[G7] H. Mao, D. H. Lee, H. Dai, F. C. Lee, D. Borojevic, “Evaluation and Development of
New Power Electronic Technologies for Superconductive Magnetic Energy Storage
(SMES) using PEBB,” 15
th
VPEC Seminar Proceedings, pp. 129134, Sept. 1997.
[G8] N. Celanovic D. H. Lee, D. Peng, D. Borojevic, F. C. Lee, “Control Design for
Superconducting Magnetic Energy Storage Power Conditioning System,” IEEE PESC
Conference Record, 1999.
[G9] D. H. Lee, “A Power Conditioning System for Superconductive Magnetic Energy Storage
Based on Multilevel Voltage Source Converter,” Ph.D. Dissertation, Virginia Polytechnic
Institute and State University, Blacksburg, VA, 1999.
159
VITA
The author was born in Novi Sad, Yugoslavia, in 1970. He received his B.S. degree in
Electrical Engineering from the University of Novi Sad, Yugoslavia, and his M.S. degree in
Mechanical Engineering from Vanderbilt University in Nashville TN in 1994 and 1996
respectively.
In the Summer of 1996, the author enrolled as a doctoral student at VPI&SU. His research
centered on the modeling and control of multilevel threephase converter systems.
The author is a member of IEEE. He also served as president of the CPES student council
(Virginia Tech campus) and as a member of the joint CPES student council (all five campuses).
SPACE VECTOR MODULATION AND CONTROL OF MULTILEVEL C ONVERTERS
Nikola Celanovic
(ABSTRACT)
This dissertation is the result of the research and development of a power conditioning system for superconductive magnetic energy storage systems (SMES). The dominant challenge of this research was to develop a power conditioning system that can match slowly varying dc voltages and dc currents on the superconductive magnet side with the ac voltages and ac currents on the utility side. At the same time, the power conditioning system was required to provide a bidirectional power flow to the superconductive magnet.
The focus of this dissertation is a threelevel diodeclamped dc/ac converter, which is a principle part of the power conditioning system. Accordingly, this dissertation deals with the space vector modulation (SVM) of threelevel converters and introduces a computationally very efficient threelevel SVM algorithm that is experimentally verified.
Furthermore, the proposed SVM algorithm is successfully generalized to allow equally efficient, realtime implementation of SVM to dc/ac converters with virtually any number of levels. The most important advantage of the proposed concept is that the number of instructions required to implement the algorithm is almost independent from the number of levels in a multilevel converter.
More on the side of the control of multilevel converters, particular attention in this dissertation is paid to the problem of charge balance in the split dclink capacitors of threelevel neutralpointclamped converters. It is a known fact that although the charge balance in the neutral point (NP) can be maintained on a line cycle level, a significant third harmonic current
ii
flows into the NP for certain loading conditions, causing the neutralpoint voltage ripple. The logical consequence of that ripple is the deteriorated quality of the output voltage waveforms as well as the increased voltage stress on the switching devices.
This was the motivation to more carefully explore the loading conditions that cause the imbalance, as well as to study the fundamental limitations of dclink capacitor charge balancing algorithms. As part of this work, a new model of the neutralpoint current in the rotating coordinate frame is developed as a tool for the investigation of theoretical limitations and in order to provide some insight into the problem. Additionally, the lowfrequency ripple is
quantified and guidelines are offered that can help determine the correct size for the dclink capacitors.
Because the study of the neutralpoint balance identified the loading conditions that (under some possible system constraints) cause an unavoidable neutralpoint voltage ripple, a feedforward type of control method is developed next. The proposed feedforward algorithm can effectively prevent the neutralpoint voltage ripple from creating distortions in the converter output voltage under all loading conditions and without causing additional disturbance in the neutralpoint voltage. The feedforward method is developed for a sine triangle as well as for the SVM type PWM algorithm.
The simulation results that include the full dynamic model of the converter and load system validate the feedforward approach and prove that the feedforward algorithm can effectively compensate for the effect of the neutralpoint voltage ripple. The simulation results are then experimentally verified.
iii
Dr. for inspiring classes. encouragement in times of need. I’m thankful for the opportunity to be a member of the Virginia Power Electronics Center and to watch it grow into the Center for Power Electronics Systems. I wish to specifically emphasize my gratiude for all the help and encouragement I received from Dr. With much love. This dissertation is their accomplishment as much as it is mine. Dr. William Baumann. and all others who made my work and my stay in CPES more enjoyable. Dr. Jason Lai and Dr. C. His impressive knowledge. Jim Lyons from GE CR&D Schenectady. Dushan Boroyevich. iv . At this point. and her insight. Dengming Peng for working with me on the project. I would also like to acknowledge CPES students and staff for providing an enjoyable educational atmosphere. Especially. NY. It has been a memorable experience. I’m also grateful to my professor. for her support. I would like to thank my good old friend. Lee. Vera Vlahovic. technical skills and human qualities have been a source of inspiration and a model for me to follow. I wish to thank my committee members. I would like to thank my brother Ivan Celanovic for endless discussions. Their guidance and insight gave me encouragement to proceed with confidence towards the completion of this work. for his guidance. encouragement and support during my graduate studies. Dr. encouragement in difficult moments and for serving on my advisory committee. Vlatko Vlatkovic and Dr. Fred. for his help and guidance during my graduate studies at Virginia Tech. Dr.ACKNOWLEDGMENTS I wish to express my sincere gratitude to my advisor. both professionally and personally. Dong Ho Lee for sharing with me his rich practical experience. Douglas Nelson. I thank my parents Marija and Lazar for their loving care and their unconditional support. Finally. Dr.
To my parents v .
.1 Switching Vectors...................................3 Multilevel Power Conditioning System as a Power Controller for a Superconducting Magnetic Energy Storage System.....................4 Modeling and Control Approach and Problem Formulation..................... 46 2.....................................................3........................2..............................................................................1 Power Conditioning System Design approach....................................................................4 Additional FullBridge Topologies .. 61 3.............................................2 Flying Capacitor Multilevel Topology................3 The New Space Vector Modulation Algorithm ........3........................................................ 19 2........ 22 2..... 65 3........................................ Switching States and Reference Vector........... 67 3.....2 Components of the Digital Controller for SMES Power Conditioning System....................... 13 2...............1.............. 64 3........ 26 2..............3 Duty Cycles Computation........................................ 21 2.................................1......................3............................ 52 3................................1 Multilevel Topologies ..................................4 Control of the Inverter in Rotating Coordinates.............. 16 2............... 93 4............................................. 17 2........2 1........................5 Experimental Results ..............1 Coordinate Transformation............................................ 33 2............................................ 55 3.............................................................................................................................................................2 Neutral Point Current Modulation........... 77 4 NP BALANCING PROBLEM IN THREELEVEL DIODECLAMPED CONVERTERS ......2....................... 1 Motivation and Objectives .............................. 63 3............................ 82 4............................................................................................................3 Analysis and Control of the Neutral Point Voltage Error . 38 2...............2 Space Vector Modulation of a ThreeLevel Converter............1 The Effect of Different Switching States on the Current in the NP ..3 vi ..3 Modeling of a ThreeLevel Voltage Source Inverter...................3.................... 49 2.......................4 Control Algorithm........................................ 80 4........................ 14 2.................................. 50 3 A FAST SPACE VECTOR MODULATION ALGORITHM FOR MULTILEVEL CONVERTERS ........ 41 2.........................................3................................................1................................2......................................................................1 1..........5 Conclusions ..1 Overview of the Architecture of Digital Control Hardware ..5 Functions of the EPLD..................................2.3 Controller Architecture ....................................... 85 4............................................ 13 2............. 25 2..4 Practical Implementation of The New SVM Algorithm.....1..............................2.........................................3......... 68 3..... 1 Review of Previous Research................................................ 3 Dissertation Outline and Major Results..............................1 DiodeClamped Multilevel Topology.................................................................................................. 30 2..........................2 Modeling and Control of the Chopper ............ 96 1.....................Table of Contents 1 INTRODUCTION..................... 51 3....4 SwitchingState Selection...........................3. 31 2.......................................... 19 2..........................................................................................3...................................................................................2 Detection of Nearest Three Vectors................................3...........................5 Conclusions ......... 11 2................4 Limitations of the NeutralPoint Control and Some Design Recommendations............................................................. 78 4..................................................................2 Multilevel Digital Control Hardware ....3 Multilevel Configurations with Cascaded TwoLevel FullBridge Inverters .........................................................................5 Simulation and Experimental Re sults .......................................................... 9 2 MULTILEVEL CONVERTERS’ MODULATION AND CONTROL........
........ Experimental Verification of the Model .............................2 NP Control With Sine Triangle Modulation......................................................... 143 6.......... 106 5................3................... with and without FeedForward Terms.........2......3..........1 Average Model of ThreeLevel Converter with Sine Triangle Modulator .......................................................................Conclusions ... 131 6.............2 Performance of FeedForward Algorithm.................................... 113 5........................3 The Average Model of a Sine Triangle Modulator...................................... 152 VITA…………..............................................................1............2.......................................................... 99 5 FEEDFORWARD CONTROL OF THREELEVEL NEUTRALPOINT CLAMPED CONVERTERS .......................... 141 6......................2 Analysis of the SwitchingCycle Average Model of the Sine Triangle Modulator with the FeedForward Algorithm .........129 6.................................3.... 107 5....................................................3 Average Model of the ZeroSequence NP Control........ 127 6 FEEDFORWARD CONTROL OF THREELEVEL NEUTRALPOINTCLAMPED CONVERTERS—SIMULATION AND EXPERIMENTAL RESULTS ............................ 100 5...................3.................................................2...................1 Average Model of the ThreeLevel Converter with Fast Space Vector Modulation Algorithm ............3 Modeling the FeedForward Control.......................................................................................2............................ 114 5.....................................................5 Conclusions ...3 Average Model of Fast Space Vector Modulator....................................... 128 6....... 146 7 CONCLUSIONS ....1 Analysis and Experimental Verification of the Line Cycle Average Model of the FeedForward Algorithm ...... 108 5......................... 101 5................................................................................3... Experimental Results............................................2 The Fast Space Vector Modulation Algorithm.......................................................... Simulation Results ... 121 5..........................3......................... 137 6. 135 6...........................................................................................1 The Effect of the Capacitor Charge Imbalance on the Voltage Space Vectors ....2..... 135 6........ 148 REFERENCES .........4 A SmallSignal Model of the ZeroSequence NP Control................................................................................................1 PWM Generation with Sine Triangle Modulation....... 117 5..................... 142 6............................................................ 115 5...............4 Conclusions ...........6 vii ..... 124 5.........3................2 Sine Triangle Modulation for ThreeLevel Converters............................................................................................ 159 4..................................................................................... 120 5.............................................. 106 5........2 Mathematical Basis of the FeedForward Algorithm....................4 Implementation of FeedForward Algorithm for Space Vector Modulation...................1 The FeedForward Algorithm ...........5 The Stability Analysis of the Feed Forward Algorithm..........1 Control Authority Over the NP Current......................
............ 20 Figure 2............................................................ 3.............................. 20 Figure 2.......................................................... Architecture of the digital controller for power electronics systems............ .................. Three.................. 38 Figure 2.. Switching states of a threelevel converter.................................5....... 12 Figure 2.....5....................................... Synthesis of VREF in middle small triangle . 43 Figure 2............................... Synthesis of VREF in inner small triangle... Rectifier response to change in current command…………………….21...............4...and fourlevel flying capacitor phase leg......... Experimental SMES PCS system................................................................... 61 Figure 3........ Simulated and measured rectifier voltage loopgain......... Simulated and measured rectifier current loop gain... 27 Figure 2....... 23 Figure 2........... The three..... 29 Figure 2....18................................ 36 Figure 2...... 26 Figure 2... .................. 13 Figure 2.......7.............................. Architecture of the SMES digital controller board.... .....................8........ . 47 Figure 2......................... 24 Figure 2.......... 46 Figure 2................................. 47 Figure 2.................... Power conditioning system for SMES................... Inverter control in the utility mode (charging the SC). 37 Figure 2..........................10.................. 42 Figure 2..... Multilevel modulation techniques..................................2.......................16.... 64 viii .......... Digital interface board. 12 Figure 2. .... ..................2. Diode clamped full bridge fivelevel topology.........32.................... 48 Figure 3...... Average model of the chopper................ Phase leg of a cascaded fivelevel fullbridge inverter............... 34 Figure 2................ 18 Figure 2.....................................31.. 43 Figure 2......................................................... 2.. .............................6......... ........................................................3. ...............8........................... Control overview...................................... ...................................... ......... Space voltage vectors of the chopper...................................and fourlevel neutralpointclamped phase leg......27....... 16 Figure 2...................... Buck mode control of the chopper (charging the SC)...........30...... ...............9..............................11.25....................................................................................................................................... 25 Figure 2.............14.............................. ............... Functional diagram of a threeleg nlevel converter.........................1...................................17............List of Figures Figure 2.................................................................. 45 Figure 2. Inverter average model in rotating dq coordinate frame.....4........................ 54 Figure 3...... Flying capacitor fullbridge fivelevel topology............................ ....................13........................9............. 37 Figure 2... 62 Figure 3....... The voltage space vectors of a fourlevel converter. Architecture of digital controller for threephase converters............................................... ................................. Inverted gate drive signals with deadtime...23...... 27 Figure 2...... Phase leg of a cascaded full bridge sevenlevel converter............ 59 Figure 3.................................................7........... Voltage space vectors of a threelevel inverter... A doublebuffered register bank inside the EPLD..................................................................... 18 Figure 2..................3......28..........29..............................................12. Switching network of a threelevel VSI ...................... Boost mode control of the chopper (discharging the SC).... Power electronics system overview.. 17 Figure 2.......... Multilevel converter topologies...... 15 Figure 2................15.......................... 32 Figure 2........... 60 Fig....................24............................................19....... Synthesis of VREF in outer small triangle...... 51 Figure 3...................................6...... 56 Figure 3................... 52 Figure 3..........22.......... Switchingstate vectors of a threelevel converter in a hexagonal coordinate system........................ PWM signal generation logic within the EPLD............................... Inverter control in rectifier mode (discharging the SC)...........1.........20......26... Voltage space vectors of a threelevel converter............. Analog interface board............. 34 Fig.... Equivalent switching model of threelevel chopper.
..... ......... 130 Figure 6............ 95 Figure 4............. The linecycle average NP current as a function of the zerosequence component δ...12...1.............................. ......... 70 Figure 3........7......... 115 Figure 5........................................ .....................8........... The model of the modulator with respect to the NP current.....11.. four.. 79 Figure 4......8.. The definitions of the dclink voltages and the voltage error in the NP......3.......4... .................... Normalized linetoline PWM voltage waveforms for three....................... Average model of the sine triangle modulator with feed forward.7............... 113 Figure 5... 102 Figure 5. .......8........ 98 Figure 5.................. 83 Figure 4......... 122 Figure 5............. Capacitor sizes for specified NP ripple....4........... Model of NP bangbang controller.... Overview of the new SVM algorithm..........................................................6.......... 96 Figure 4................ Measurement of the NP control loop gain without normalization. Circuit schematic of a threelevel VSI....... 111 Figure 5..1................................. Principle of NP control for the threelevel converter....Figure 3......... Switchingstate selection with dclink capacitor charge control.... ...................... 94 Figure 4. Region where LF ripple can be suppressed............. 80 Figure 4.. 136 Figure 6.......1... Duty cycle of one phase equivalent switch for sine triangle modulation................... Power flow in the inverter mode of operation with large imbalance in NP......................... 123 Figure 5.. 81 Figure 4.. Outer small triangle................9...14...................13.......... 108 Figure 5.... 74 Figure 4......4... 107 Figure 5....3.......... ............................8....................................................... The active sector computation based on the home vertex algorithm.. 91 Figure 4.... ....... 98 Figure 4............... ...... ...... ...........10....... 97 Figure 4.................. 69 Figure 3. 124 Figure 5............ . Computation of duty cycles and selection of switching states......................... Duty cycles for the sine triangle threelevel modulation.... The duty cycles of the sine triangle modulator with zero sequence.....7............ 137 Figure 6..... 133 Figure 6..... Normalized NP voltage for power factor angle of 85 degrees................. Duty cycles of SVM for modulation index m=0....12................2....................12... 68 Figure 3................ The position of equivalent switches for the switching state pon........ ...............15..........17.. The PWM signal with strong feedforward component....... Shift of the medium voltage vectors due to the ∆V>0 error in NP voltage.. ............................................... 139 ix ... ..........5............. 137 Figure 6......and fivelevel converters............... 102 Figure 5.................... 86 Figure 4........................ Normalized amplitude of the LF charge ripple............ .. Switchingstate vectors of the threelevel VSI..... 126 Figure 6................5.........2.........................10.......... 119 Figure 5................................. ............................... ........ Average model of one phase leg sine triangle modulator.............................................16.......... Voltage space vectors of a threelevel converter.............................11....... 131 Figure 6............ Principle of sine triangle modulation for a threelevel converter................. Theoretical and experimental normalized NP voltage ripple..................... Experimental verification of NP voltage ripple properties........ Computing the nearest four vectors........... “Splitting” of the small voltage vectors due to the ∆V>0 error in NP voltage..6.........................................................10.. Duty cycles of one phase with the zero sequence present................................................................................9.....5.............13......... Functional diagram of a threelevel converter.... 90 Figure 4...13.................................... ..........11.............. 71 Figure 3........... 110 Figure 5.. Experimental measurement of the loop gain in the NP controller........... 85 Figure 4.......... 103 Figure 5...........3... Rectifier mode of operation with low power factor............. The effect of NP unbalance on the boundaries of small triangles.............8...................................................... Switching functions of the medium vectors.. Duty cycles for the sine triangle threelevel modulation..........6....... with and without NP control.... The linecycle average NP current with feedforward terms included..............14.............. .14.....................................2................. 117 Figure 5...... .........8.. Weighing factors for small vectors with m=0... 105 Figure 5............ Weighing factors for medium vectors with m=0............... ...... ................................................................ 134 Figure 6......... 104 Figure 5.............
.10............. . Measured and simulated line voltage in the presence of NP ripple............ The inverter mode of operation with low power factor load....15.... Measured and simulated NP ripple of a threelevel converter........................13....................9..... 144 Figure 6...........11.............................. 140 Figure 6................................. 145 Figure 6. Rectifier mode of operation for high power factor......................... 143 Figure 6...... Average model of the fast SVM algorithm..... 146 x ................................14...Figure 6.......... Simulated linetoline currents for balanced and symmetric load.....12....................................... ..... 141 Figure 6. Fast threelevel space vector modulator... 142 Figure 6....... .
.................. 89 Table 4V Currentswitching function for S1 small vectors... ............................................................................................................................................................. 105 xi ................................................................................ 89 Table 5I Effect of the error in NP voltage on the medium voltage vectors.............. ............ 104 Table 5II Effect of the error in NP voltage on the small voltage vectors.................... 76 Table 4I NP current iNP for different space vectors.... 36 Table 3I The switchingstate lookup table............................ 81 Table 4II Currentswitching function for medium vectors....................List of Tables Table 2I Neutralpoint current iNP for different small vectors........................ 75 Table 3II The performance of the algorithm implementation for a threelevel converter......... .............. 89 Table 4III Currentswitching function for S0 small vectors.......................................
What makes power electronics solutions unique in both these cases is that they are fundamentally multifunctional and can provide not only their principle interfacing function but various utility functions as well.1 INTRODUCTION 1. harmonic compensation and active filtering. fuel cells and micro turbines all require powerconditioning systems. Some of these functions are load leveling. Similar to the renewable sources. static var compensation. renewable and/or clean energy sources such as wind turbine. This is particularly true of energy storage systems. In addition. they also match the electrical characteristics of the sources with the electrical requirements of the loads and/or the utility. they can also be used as uninteruptible power . solar power. utility stabilization. there has been increased interest in renewable energy sources on one side.1 Motivation and Objectives In recent years. While these powerconditioning systems provide the necessary interface to the utility. energy storage systems (viewed as key components of the envisioned flexible utilities) also require power conditioning systems that can match the electrical characteristics of the energy storage systems with those of the utility. For example. Both these developments recognize power electronics systems as a core enabling technology. while on the other. which are potentially well suited to provide a host of additional functions on the utility level. the initiative for deregulation of the utilities increased the interest in systems that can increase their flexibility.
flywheels. One of the practical energy storage technologies is superconductive magnetic energy storage (SMES). This is the principle reason why this work concentrates on some important aspects of multilevel topologies that are directly applicable to a wide range of multilevel converters. with the increased number of levels. simplify converter design and potentially increase the availability of multilevel converters. super capacitors and other energy storage technologies.supply (UPS) systems at various power levels. Therefore this work followed a trend in the power electronics community to shift towards multilevel power electronics converter topologies that seem to offer significant advantages for utility applications over the more conventional twolevel topologies. However. However. even more 2 . the number of allowable switching states in the converters is also rapidly increasing. The algorithm very efficiently performs a switching vector identification and duty cycle computation. together with the wider introduction of multilevel converter topologies. What makes this power conditioning system (PCS) design even more interesting is that it was built using multilevel converters. and tradeoffs involved with multilevel power converter technology in general. which is typically implemented in software and operating in real time and for high switching speed. there is a strong tendency to modularize the hardware design in order to reduce the prices. The design of a power conditioning system for a SMES system was one of the primary objectives of the work that will be presented in this dissertation. This problem is addressed with a new and general multilevel algorithm. and can be designed to provide a pulse power capability that is of interest for certain military applications as well as for some highenergy physics experiments. This situation places significant computational challenges in front of the PWM modulator. some of the insight gained from the development of a SMES power conditioning system is readily applicable to power conditioning systems for batteries. which allowed a research into the issues. For example. fuel cells.
In 1980. with respect to the voltage balancing problem of the threelevel neutralpointclamped converter. a feedforward control method is developed. the number of steps involved in the computation is almost independent of the number of levels in the converter. multilevel modulations and the control of multilevel converters. which is currently perhaps the most widespread multilevel topology. Secondly. early interest in multilevel power conversion technology was triggered by the work of Nabae. and the ability to make recommendations for the size of the dclink capacitors for any given operating point as well as the NP ripple specification. et al. a problem of split dclink capacitors’ charge balancing is explored in a comprehensive way. 1. It is the objective of this dissertation to investigate the effects of the feedforward algorithm on the quality of the output waveforms as well as on the control authority over the charge balance in the split dclink capacitors.. because these topics are closely related to the work in this dissertation. The study resulted in the identification the loading conditions that result in the lowfrequency ripple. who introduced the neutralpointclamped (NPC) inverter topology [B1]. Because of the overwhelming breadth of published material on these topics. This is accomplished using the new model of the neutral point (NP) current in the rotating dq reference frame.2 Review of Previous Research When designing a system. particularly a relatively complex system such as the SMES power conditioning system. because the study of the NP balance identified the loading conditions under which certain system constraints cause an unavoidable NP voltage ripple. in areas ranging from the device’s physics and soft switching all the way to the mathematical abstractions used for modeling and control. And thirdly. the quantification of the size of the ripple. there are many important considerations. At this time. It was immediately 3 . the area of multilevel power conversion can still be considered young. this literature review will be limited to multilevel topologies.importantly.
the clamping diodes can be subject to severe reverserecovery stress. C4. or sometimes as the imbricated cells multilevel converter. which challenges this technology. A8. facilitates balancing of the intermediate capacitor levels and has a very “clean” utility interface. the multilevel converter with cascaded singlephase Hbridge inverters. 4 . mainly IGBTs and IGCTs. The main disadvantage of this topology is that the required blocking voltage of the clamping diodes is proportional to the level for which they are used to employ clamping action. the threelevel NPC converter was the first widely popular multilevel topology. The two remaining configurations. because of the high switching speed of today’s switching devices. In addition. the majority of research and development effort seems to concentrate on the development of three classes of converters: the diodeclamped multilevel converter. At the present time. and the multilevel converter known as the flying capacitor converter. C5] enables bidirectional power flow. This is in part because the converters of this type are currently in the phase of rapid market introduction by several worldrenowned manufacturers [B20. the original topology was extended to a higher number of levels using the same principle of diodeclamped intermediate voltage levels [C1]. As mentioned. which resulted in the fact that today this topology probably represents the stateoftheart multilevel converter technology. in the early nineties the concept of the threelevel converter was extended further and some new multilevel topologies were proposed.realized that this new converter had many advantages over the more conventional twolevel inverter. Consequently. The backtoback connection of two converters of this type [A7. Subsequently. namely the multilevel converter with bidirectional switch interconnections [A1] and the multilevel converter with multiple threephase inverters are receiving less attention at the present time and will not be discussed further. Soon after the introduction of the NPC converter. B22]. series connection of the diodes may be required. A significant amount of research effort was invested in the development of this family of converters [B].
An alternative to the diodeclamped converter, the flying capacitor topology does not have issues with clamping diodes. First proposed in 1992 [E1], this approach has the advantage of a larger number of redundant switching states, which allows more freedom in balancing the clamping capacitors’ voltages. The main disadvantage is the potential for parasitic resonance between the decoupling capacitors; this is made even worse by the high number of capacitors, which complicates packaging for small inductance. In addition, there are issues with voltage redistribution in the case of voltage surges. Nevertheless, the flying capacitor topology seems very promising.
Finally, the multilevel configuration with cascaded Hbridge inverters presents another alternative in the design of multilevel converters. One of the earlier applications of this topology was for plasma stabilization [D1]; it was then extended for threephase applications [D3]. A primary advantage of this topology is that it provides the flexibility to increase the number of levels without introducing complexity into the power stage. Also, this topology requires the same number of primary switches as the diodeclamped topology, but does not require the clamping diode. However, this configuration uses multiple dedicated dcbusses and often a complicated and expensive line transformer, which makes this a rather expensive solution. In addition, bidirectional operation is somewhat difficult (although not impossible) to achieve [D10].
Perhaps the most important improvement in cascaded converter topologies is the hybrid multilevel topology [D7]. The main strength of this approach is its combination of the high voltage capacity of the relatively slow GTO devices with the high switching frequency of the lower voltage capacity IGBT devices. At the same time, the different voltage levels of the IGBT and GTO bridges create an additional voltage level without any additional complexity.
Modulation techniques in multilevel converters for utility applications were originally restricted to the synthesis of stepped waveforms [A4]. This strategy can be interpreted as a quantization process in which the reference sinusoidal voltage is approximated by discrete levels of voltage available at the dcbus. The implementation of this type of modulation is relatively simple and does not require high switching speed, which makes it suitable for converters realized
5
with GTO thyristors. However, the spectral performance is not very good and a variable dcbus voltage is required in order to obtain variable voltage at the output.
Far more popular, especially for the converters with faster switching devices, are the socalled sinusoidal PWM techniques. These methods have been extensively studied and are among the most popular in industrial applications. These methods involve a comparison of the reference signal with a triangular carrier waveform and the detection of crossover instances to determine switching events. The variations of these methods are essentially in the polarity and shape of the carrier waveforms. Although there are slight differences in spectral performance [C16], for the highenough frequency ratio between the carrier and the reference waveforms, there is no substantial difference between them.
Another very interesting method is based on multiple reference waveforms.
It is also
sometimes referred to as dipolar modulation [C17]. This method uses a single triangular carrier and two sinusoidal reference waves to modulate threelevel inverters. The sinusoidal reference waves are usually obtained by adding positive and negative offsets to the conventional reference signal. The principal reason to introduce the offset is to always interrupt the switching between the outermost levels and the middle level, as well as to avoid any minimum pulse width problems for low modulation indexes.
Perhaps one of the most popular modulation approaches for twolevel converters is space vector modulation (SVM), which is now being used more and more in the control of multilevel converters. The concept of space voltage vectors corresponding to various switching states has been applied to study the impact of various switching states on the capacitor charge balancing in almost every paper discussing the SVM approach [B6, B1011]. An advantage of the SVM is the instantaneous control of switching states and the freedom to select vectors in order to balance the NP. Additionally, one can realize output voltages with almost any average value by using the nearest three vectors, which is the method that results in the best spectral performance.
The primary constraint in devising SVM (as is the case with in most multilevel modulation techniques) involves minimizing the harmonic content of the output PWM waveform while at
6
the same time maintaining the capacitor charge balance. This is the basic tradeoff of threelevel SVM, because the tight regulation of the capacitor charge balance can always be achieved at the expense of increased switching losses and poorer harmonic performance.
Another typical design problem particularly important for the converters built with the slowerswitching GTO thyristor devices is the minimum pulse width. This problem is present whenever the reference vector is close in phase and magnitude to any of the switching state vectors. One way of dealing with this problem is to change from the nearest three vector approach (NTV) to nearest four vector approach. Thus a narrow pulse problem can be solved at the expense of increased switching stress and somewhat increased harmonic content. Perhaps a better (but not necessarily more practical) alternative involves the division of the sixtydegree regions into fourteen subregions with their switching sequence optimized in order to avoid the narrow pulses [B9].
Finally, it is accepted as conventional wisdom that SVM methods (although very powerful in reducing harmonics) become computationally very demanding and all but impractical as the number of levels increases.
With respect to modeling and control of three and higher level converters, the dominant approach is to divide and conquer. What this means in practice is that phase legs are treated essentially as voltage sources (as is the case in twolevel converters), while the charge balance becomes part of the modulator, which selects between the redundant switching states in order to maintain the charge balance in the split dclink capacitors.
This approach makes the charge balance controller relatively simple.
For sinetriangle
modulated converters, this amounts to control of the modulating signal’s zero sequence for the threelevel converter [B2, B3], or the manipulation of the redundant small vectors [B4].
Building a model that can describe the behavior of the NP voltage balance is quite difficult because the NP current as well as NP “duty cycles” are piecewise nonlinear functions. In addition, the steadystate response usually has a strong third harmonic component. Taking all
7
et al. and linearizing. Perhaps the first comprehensive approach to modeling threephase PWM converters was presented by Ngo. in order to include the dynamics of the split dclink capacitors. 8 . the steadystate solutions of the linear model are constant duty cycles. perhaps the best approach is to average the NP current across the whole line cycle [B14]. but under the usually valid assumption that the NP voltages can be controlled by the modulator and that the ripple is sufficiently small. Clearly. transformation in the rotating coordinates. and can be linearized and used to design a NP voltage controller. F16] mainly followed the same steps and resulted converter models that describe the lowfrequency dynamic behavior of currents and voltages of multilevel converters as well. Unfortunately. Essentially. in this case. they do accurately describe the lowfrequency dynamic behavior. Finally. F15.this into account. The converter models published thereafter [F2. than averaging. F4. it is important to mention an attempt to make a unified dynamic model [F8. altogether six for a threephase converter. averaging. thus taking away most of the freedom to optimize the switching patterns. the coordinate transformation and linearization reduces the number of possible solutions from infinity to one suboptimal solution. the switching model had to be developed using the phase instead of the linetoline switching functions. The model was developed using the usual steps of developing the switching model first. and linearization. F13] of a threelevel converter that included the NP voltage dynamics together with all the other voltages and currents. they do not describe the effects of the split dclink capacitors. [F14]. Although this approach can not determine either the size or the shape of the voltage ripple. The modeling was performed in four steps: development of the switching model. The linear model was than obtained by linearizing the average model in a rotating coordinate frame around the operating point. it does show the trends and the overall stability of the dclink voltage. Naturally. which is only one of the infinite number of functions that steadystate duty cycles of the average (nonlinear) model can have. transforming into the rotating frame. This resulted in a large number of duty cycles to be controlled.
this method effectively limits the freedom to optimize the switching patterns without offering additional insight into converter operation. control design and modeling issues unique to the approach that was taken in the design of the prototype power processing converter that was designed as a part of this dissertation work. this chapter covers the topological. specific requirements for the control of a dc/ac converter operating in an environment with frequent demands for changes in operating mode and power flow direction are discussed. This chapter is. Implementation of a ThreeLevel Converter in a SMES Power Conditioning System. an approach to modeling and implementing a control system for a threelevel NPclamped converter for the application in the SMES system is presented. and as a utility inverter when the magnet is being discharged. it provides excellent insight into the operation of multilevel converters. 1.3 Dissertation Outline and Major Results The work in this dissertation is divided into the following major parts. as well as the problems that were encountered and the original solutions that were proposed. In addition. At the same time. In addition. The control design is based on a twolevel dc/ac converter model that is experimentally proven to closely match the dynamic behavior of a threelevel converter. the 9 .Therefore. This algorithm is computationally extremely efficient and is almost independent of the number of levels of the converter. there is a discussion of how this approach compares with other available converter technologies. The control design enables the dc/ac converter of the power processing system to operate as a rectifier when the magnet is being charged. At the same time. although interesting from the theoretical point of view. Finally. This chapter introduces an original SVM algorithm for nlevel threephase converters. ThreeLevel DiodeClamped Converter. an extended introduction that discusses the requirements that an effective power conditioning system for SMES needs to satisfy. by controlling good part of the modulation itself. Finally. A fast SVM Algorithm for Multilevel Converters. in a sense.
This chapter also covers the details of the realtime algorithm’s implementation on the DSP processor using the example of a threelevel space vector modulator. The lowfrequency ripple of the NP (caused by certain loading conditions) is observed and quantified. while simultaneously improving the control authority over the capacitor charge balance. In this chapter. a feedforward control strategy to compensate this ripple is presented for two modulation algorithms: a sine triangle and a space vector PWM algorithm. The theoretical results are then experimentally verified and guidelines for choosing the dclink capacitor size are given. which is the point between the two dclink capacitors. This chapter explores the fundamental limitations of the NP voltage balancing problem for different loading conditions of threelevel voltage source converters. A full dynamic model is developed to help better understand the effect of the feedforward algorithm on the quality of output waveforms as well as its effect on the control authority over the charge balance of the dclink capacitors. The results from the previous chapter indicate that under certain loading conditions. This results in the lowfrequency voltage ripple in the NP. the lowfrequency voltage ripple in the NP of the dclink capacitors is to be expected. Some of the theoretical results are experimentally verified.experimental results from the operation of the prototype power conditioning converter are included as is the discussion of the algorithm’s strengths compared with SVM algorithms. FeedForward Control of ThreeLevel DiodeClampedConverters. 10 . the proposed algorithm seems to effectively compensate for the disturbance in the output voltage created by the voltage ripple in the NP. Interestingly. Neutral Point Balancing Problem in ThreeLevel NPClamped Converters. It is known that under certain loading conditions and for certain output voltages a charge balance in the threelevel converter can not be maintained on the switching cycle level. A new model in dq coordinate frame (utilizing current switching functions) is developed as a means to investigate theoretical limitations and lend to more insight into the problem.
power distribution. • Spectral performance of multilevel waveforms is superior to that of their twolevel counterparts. power quality and power conditioning applications. multilevel power converters can be viewed as voltage synthesizers. The selection of the best multilevel topology and the best control strategy for each given application is often not clear and is subject to various engineering tradeoffs. which can damage the motor windings and cause other problems. • Multilevel waveforms naturally limit the problems of large voltage transients that occur due to the reflections on cables. and include motor drives.2 MULTILEVEL CONVERTERS’ MODULATION AND CONTROL Multilevel power conversion technology is a very rapidly growing area of power electronics with good potential for further development. The most attractive applications of this technology are in the medium. By narrowing this 11 . advantages of this approach are summarized as follows: The main • The voltage capacity of the existing devices can be increased many times without the complications of static and dynamic voltage sharing that occur in seriesconnected devices. in which the high output voltage is synthesized from many discrete smaller voltage levels. In general.to highvoltage range (213 kV).
1.2. can be roughly divided.2. Multilevel modulation techniques. 12 . MULTILEVEL MODULATIONS FUNDAMENTAL FREQUENCY SWITCHING PWM SINUSOIDAL PWM SPACE VECTOR PWM Figure 2. Multilevel converter topologies. with their rich variety of techniques.study to the DC/AC multilevel power conversion technologies that do not require power regeneration. Similarly. several attractive topological. as shown in Figure 2.1. The most actively developed of these multilevel topologies are listed in Figure 2. the modulation strategies. modulation and power semiconductor device choices present themselves. MULTILEVEL CONVERTERS DIODE CLAMPED CONVERTERS FLYING CASCADED CAPACITOR FULL BRIDGE CONVERTERS CONVERTERS Figure 2.
13 . shown in Figure 2. 2. The first group of voltage harmonics is centered around twice the switching frequency.1 DiodeClamped Multilevel Topology The first practical (and still widely studied) multilevel topology is the neutralpointclamped (NPC) PWM topology first introduced by Nabae.3. The three. it does cover the ones that are presently receiving the most attention. in 1980 [B1]. et al.and fourlevel neutralpointclamped phase leg. p Cdc p S1 Cdc Vpn NP D2 Cdc n S3 S4 n Cdc D1 S2 A Vpn NP1 D3 D Cdc 2 NP2 D4 S5 S6 S3 S4 A S1 D1 S2 Figure 2. Although this section is by no means a complete review of all the multilevel topologies.3. in this document the multilevel converter has three. The threelevel version of this topology. The intention is to give a brief review of the three basic topologies and their most promising derivatives. The advantages are: • • Voltages across the switches are only half of the dclink voltage.or more levels.1 Multilevel Topologies Although definitions may vary..1.2. has several distinct advantages over the twolevel topology.
• For topologies with more than three levels the clamping diodes are subject to increased voltage stress equal to Vpn . Therefore.1. This complicates the design and raises reliability and cost concerns. and is considered to be the most serious alternative to the diodeclamped topology. • The issue of maintaining the charge balance of the capacitors is still an open issue for NPC topologies with more than threelevels. and which is therefore preferred over other possibilities). Although measures to alleviate this problem can be applied. However. was first proposed in 1992. practical experience with this topology revealed several technical difficulties that complicate its application forhigh power converters. 14 . series connection of diodes might be required.• This topology can be generalized. and the principles used in the basic threelevel topology can be extended for use in topologies with any number of levels.2 Flying Capacitor Multilevel Topology The flying capacitor multilevel topology from Figure 2. this remains a serious consideration.4. Although the threelevel NPC topology works well with high power factor loads. These are as follows: • This topology requires high speed clamping diodes that must be able to carry full load current and are subject to severe reverse recovery stress. This may be due to the capacitor balancing issues if the nearest three vector (NTV) modulation is used (which is the modulation scheme with the least amount of stress and superior spectral performance. NPC topologies with more than three levels are mostly used for static var compensation circuits. 2.(n1)/n.
The significant advantage of this topology is that it eliminates the clamping diode problems present in the diodeclamped multilevel topologies. This makes this topology attractive even for the dc/dc converters. Three. 15 . • There is a potential for parasitic resonance between decoupling capacitors. some possible “weak points” that still need to be explored are: • • The dclink capacitor charge controller adds complexity to the control of the whole circuit. Additionally. it is obvious that rather large rms currents will flow through these capacitors. At the present time it seems that this topology has few disadvantages.4. even if the phase current is unidirectional. A study of the tradeoffs involved is not described in the available literature. the flying capacitor topology has enough switching states to control the charge balance in the single isolated leg with converters having any number of levels. Nevertheless. this topology naturally limits the dV/dt stress across the devices and introduces additional switching states that can be used to help maintain the charge balance in the capacitors. Unlike the diodeclamped converter. In addition.and fourlevel flying capacitor phase leg.p S1 p S1 S2 Vpn Cdc Cdc S3 S4 n n A Vpn Cdc Cdc Cdc S2 S3 A S4 S5 S6 Figure 2. The flying capacitor topology might require more capacitance than the equivalent diode clamped topology.
16 .3 Multilevel Configurations with Cascaded TwoLevel FullBridge Inverters One of the early applications of the series connection of singlephase fullbridge inverter topology was for plasma stabilization in 1988 [13]. The modularity of this structure allows easier maintenance and provides a very convenient way to add redundancy into the system [20]. Probably the most advantageous uses the dcsources with two different voltage levels.2. Vdc and 2. One major advantage of this hybrid approach is that the number of output can be further increased without addition of any new components. the overall complications and cost of the isolated sources for each full bridge is overall not too serious a drawback and is offset by the advantages of modular construction. requiring only the dc sources with different voltage levels [D9]. Arguably. Vdc Cdc S1 S2 S3 A S4 Vdc Cdc S5 S6 S7 S8 n Figure 2.5.1. as shown in Figure 2.5. with their line of mediumvoltage drives.6. and is probably most successfully applied by Robicon. Phase leg of a cascaded fivelevel fullbridge inverter. which reportedly have an excellent field record. This modular approach has since been extended to include threephase systems as well.Vdc. The fivelevel phase leg of the cascaded twolevel fullbridge converter is given in Figure 2.
as does the need for isolated dcvoltage source for every phase. In that arrangement. This topology has enough redundant switching states to allow the charge balancing of the dclink capacitors on a switching cycle level for all line/load conditions. However the difficulties with the clamping diodes still remain.Vdc Cdc S1 S2 S3 A S4 Vdc Cdc S5 S6 S7 S8 n Figure 2. This approach can then be implemented using hybrid device technology [D9]. the slower.+/2.1.+/3.Vdc) fully pulse width modulated output voltage. some savings on isolation of dc power supplies can potentially be achieved using fullbridge multilevel topologies. GE already successfully applied this approach in their mediumvoltage (4160V) drive product. S4 ). 2. S8 ).6. while the faster. S2 .4 Additional FullBridge Topologies Compared to cascaded fullbridge inverters. They relied on wellunderstood threelevel NPC topology to build the fullbridge fivelevel phase leg. S6 .Vdc. S3 . This arrangement can generate sevenlevel (0. Phase leg of a cascaded full bridge sevenlevel converter. highervoltage devices can be used to change the output voltage level (S1 .+/Vdc. S7 . lowervoltage devices can provide the full PWM capability between levels (S5 .2. 17 .
8. based on the flying capacitor topology. Diode clamped full bridge fivelevel topology. but introduces the issues of clamping capacitors instead.7. Flying capacitor fullbridge fivelevel topology. p S1 S2 Vpn Cdc Cdc S3 S4 n Figure 2. S1 S2 S3 S4 18 . as shown in Figure 2. S1 S2 D1 D2 D1 S 2 S3 S4 S3 S4 Using the same philosophy.p S1 Cdc Vpn NP D2 Cdc n Figure 2.8. a fullbridge fivelevel phase leg can be built. this approach alleviates the clamping diode issues. Very similar to the flying capacitor halfbridge topology. Both these multilevel design approaches have yet to be proven in reallife applications.
At the same time. Similarly.1 Overview of the Architecture of Digital Control Hardware Regardless of the topological differences. In addition. and allows implementation of any type of modulation algorithm. the everincreasing processing power of microprocessors allows for much more userfriendly development tools that can accept input from higher programming languages such as C and (increasingly) Matlab. the direction of power flow. has few main components.2. the architecture of modern digital controllers can be subdivided into several major functional blocks from Figure 2. 19 . the versatility of modern microprocessors allows them to be used for the control of any type of converter. or its application in almost every power electronics system.2. The central element is the processor (or a combination of several processors) usually accompanied by some kind of programmable logic devices to implement the PWM (if this capability is not already provided on the processor). such as the switching network with filter components on both input and output. 2.2 Multilevel Digital Control Hardware The complexities of the multilevel converters’ systems. the power terminals and a (digital) control subsystem. performance and availability (as well as the decrease in price) of digital control hardware. together with the increase in power. the sensors subsystem. In addition. result in almost exclusive application of digital control hardware for the majority of modern multilevel converters and converter systems.10. programmable logic devices provide the simplest way to implement some specific protection functions that require very fast response speed as well as any “glue” logic required to interface the processor with other components on the control board.
FILTER SWITCHING NETWORK FILTER GATE DRIVES SENSORS SENSORS DIGITAL CONTROLLER Figure 2.. Power electronics system overview. 20 . Architecture of the digital controller for power electronics systems..9.10. SENSORS Figure 2. PROCESSOR COUNTERS AND OTHER AUXILIARY DIGITAL LOGIC DIGITAL COMMUNICATION SUBSYSTEM A/D CONVERTERS GATE DRIVES .
This traditional structure of the controller is being challenged as part of the development of the power electronic building block concept and the introduction of the digital communication subsystem. The EPLD can be programmed to perform any digital operation. The elements considered for communication upgrade range from single switches or phase legs with integrated sensors. remote sensors. If accepted the standardized protocol would allow any processor to control and communicate with all elements of the power electronic system. The D/A converters are very useful for system debugging and measurement of the variables internal to the processor. boot memory. and digitaltoanalog (D/A) converter. This powerful processor is a perforce 21 . has the potential to standardize the signal distribution network across the whole range of power electronics products from different manufacturers. Since most of the feedback transducers generate analog signals. The design and choice of components for the SMES digital controller are listed as follows. erasable programmable logic device (EPLD). and with uncertain future).2. as long as they support the same protocol. It is a 32bit floating point processor with a throughput of 40 MIPS. 2. or the whole converters. F18]. Although different implementations have been proposed the main idea is to establish a standardized communication protocol. A/D converters are required to convert those signals to the digital format. Such a subsystem (although still in the development phase. most likely through a serial fiber optic link [F17. Processor The present system is based on the Analog Devices ADSP 21062 (SHARC) processor. This allows a part of the control logic to be implemented in hardware. It has a 1M bit of onchip program memory and a 1M bit of on chip data memory. such a system would be easy to service/upgrade by adding or replacing various modules. analogtodigital (A/D) converter. regardless of their manufacturer.2 Components of the Digital Controller for SMES Power Conditioning System The digital controller hardware partially developed for the control of the SMES PCS consists of the digital signal processor. In addition. which is suitable for the processor.
The most important selection criteria for the EPLD are the speed. It can be divided into three functional subsystems. Each chip has 8000 usable gates. Programmable Logic Device The EPLD can be electrically programmed to implement any logic functions. 2. D/A converters are mostly used for debugging and transfer function measurement.11 shows the block diagram of the DSPbased controller. number of flipflops and usable gates as well as the number of I/O pins that can be configured. This limits the maximum size of the logic circuitry that can be implemented with any given EPLD. FLEX 8820 chips from ALTERA looked like a very good choice. output or input/output.3 Controller Architecture Figure 2. with 180ns write cycle time.requirement for power electronics control systems that execute sophisticated algorithms at high sampling rates. so their performance is not critical. In the present design. 10bit 20 MSPS A/D converters from Analog Devices were used. The most important selection criteria for the A/D converters are the conversion time. the AD876. 820 flipflops and 152 pins that can be user defined as input. Analog to Digital Converter The processor acquires the analog feedback signals from the plant through A/D converters. implemented on separate printed circuit boards: 22 . We have chosen Quad. 12bit DAC8412 converters from Analog Devices. Digital to Analog Converter In PWMbased systems. and the resolution.2. At the time of the board development two.
11. In the present design the SHARCPAC expansion connectors are used to interface with digital interface board. 23 . DSP Processor Board The ADSP21062 EZLAB from Analog Devices. used in this project is a complete development system based on the ADSP 21062 (SHARC) processor. It allows the use of EZICE InCircuit Emulator and facilitates communication with additional modules through the SHARCPAC expansion connectors. Architecture of the SMES digital controller board.• • • DSP subsystem Digital interface subsystem with flex EPLDs Analog interface subsystem with A/D and D/A converters DSP (ADSP21062) DIGITAL INTERFACE BOARD ( 2 FLEX 8820 ) ANALOG INTERFACE BOARD DIGITAL BUFFER BOARD SENSORS GATE DRIVES Figure 2. that let the user add additional memory and/or SHARC processors.
This board performs the majority of the PWM generation and system protection tasks such as shutdown at repeated fault signal from gate drives. Sixteen analog input channels are being levelshifted. watch dog timer function and shoot through protection. and then fed to two analog multiplexers (both 2 x 4to1).Digital Interface Board The digital interface board from Figure 2. ANALOG INTERFACE BOARD PC SERIAL SERIAL 5P C N ET R 0 I O NCO N EPLD D P R O M S P B O A R D A D ES D RS DT AA P R O M CLK1 EPLD CLK2 CLK3 C NR L OTO 60 PIN CONNECTOR DIGITAL BUFFER BOARD Figure 2. Analog Interface Board Figure 2. The outputs of the multiplexers are routed to four A/D converters.13 shows the functional block diagram of the analog interface board. inverted and filtered. which are all connected to the digital interface board via common data lines. The main consideration in the choice of architecture was not so much the throughput as it was the 24 . Digital interface board.12 consists of two backtoback connected EPLDs.12. These EPLDs share the address/data and control lines with the processor. The EPLDs can be programmed using a serial cable from a personal computer (PC) or using the programmable read only memories (PROMs).
The process begins with the sensing of output voltages and line currents (INPUT). These are then transformed from stationary threephase (abc) to rotating threephase (dqo) coordinates and compared with reference values to generate errors. and two of which drive the current transmitters that can be used to feed long cables.2. The block diagram of the control structure is shown in Figure 2.13. Then the modulator takes over and selects the switching states and duty cycles that implement the commanded reference vector from the compensators. DIGITAL INTERFACE BOARD DATA CONTROL DATA CONTROL TRISTATE BUFFER A/D ANALOG MULTIPLEXER FILTERING & LEVEL SHIFTING CLK D/A B U F F E R S VOLTAGE/CURRENT SENSORS (Voltage/Current) Figure 2. The analog interface board also has 8 analog outputs. 2. Analog interface board. all of which are buffered using voltage followers.4 Control Algorithm The primary tasks of any controller are regulation and pulse width modulation. which can not be achieved using a single converter with multiplexed input (the limitation primarily lies in the speed of the multiplexers).nearly instantaneous sampling.14. 25 . The error signals are processed in a compensator and the result is transformed back to abc coordinates (T1). Up to this point the complete algorithm is implemented in software.
After receiving an IRQ.Frequency Reference INPUT in abc T + Compensator T1 T Reference Output Pwm Signals with fault protection Space Vector Modulation Selection of switching states Calculation of dutycycles Figure 2. A functional block diagram of the EPLD logic responsible for generating the PWM signals is shown in Figure 2. 26 .5 Functions of the EPLD The main function of the EPLD is to generate the PWM signals that are fed into the inverter gate drives. based on the data received from the DSP. 2. These are then loaded into the doublebuffered register bank in the EPLD to be used in the following switching cycle. the processor reads the feedback variables from the A/D converters and calculates the switching states and their duty cycles according to the algorithm described in the previous section. The DSP receives periodic interrupt requests (IRQ) from the EPLD at switching frequency (1 TS ). The processor then enters the idle state until receiving new IRQ.14.2.15. Finally the PWM generation is implemented in hardware requires many timers to keep track of the time elapsed (duty cycles) and some combinatorial logic to change between the switching states. which is equal to the control sampling frequency. Control overview.
IRQ CLKx DOUBLE BUFFERED REGISTER BANK CLK CLR Timer Digital Comparator Logic FAULT CLK Bus Interface Logic T0 T1 T2 T3 T4 T5 TS State0 State1 State2 State3 State4 State5 State6 To DSP MUX Dead Time Fault Prot. A doublebuffered register bank inside the EPLD.15. PWM Figure 2.16.16. so that the PWM data used in the current switching cycle (BUFFER 2) is not affected by To Gate Drives Timer=TS 27 . the numbers of which vary for different applications. PWM signal generation logic within the EPLD. The doublebuffered register bank consists of timer registers and state registers. as shown in Figure 2. Each register is doublebuffered. DATA BUFFER 1 BUFFER 2 C L K CLK IRQ DSP Write Figure 2.
This is done by delaying the rising edges of all the signals by a particular number of clock cycles. 28 .17. The fresh data is loaded into BUFFER 2 by an IRQ signal at the beginning of each switching cycle. It can be fed by any of the three clock signals available on the digital interface board (CLK1. The state registers contain those switchingstate codes which are used for PWM in the current switching cycle. the times from the beginning of the switching cycle (measured in the units of EPLD CLK period) when the corresponding switching state ends.15 is a freerunning counter with the same length as the timer registers in the register bank. The timer registers contain the dutycycle information. After the timer reaches TS. DeadTime Protection Deadtime protection is provided to avoid any shoot through of the inverter legs.e. The Digital Comparator Logic and MUX in Figure 2. and clearing the timer counter. ordered in the sequence in which they will be applied. The counter is cleared by the IRQ signal at the beginning of each switching cycle.15 sequentially output the switching states using the current data in the doublebuffered register bank.12). CLK2. Generation of PWM Signals The timer in Figure 2. loading the new data into the doublebuffered register bank. CLK3 in Figure 2.. digital comparator logic generates the IRQ signal which starts a new switching cycle by interrupting the processor.the asynchronous DSP write into BUFFER 1. as shown in Figure 2. i.
Figure 2.17. Inverted gate drive signals with deadtime.
Fault Protection
In the event of a fault, the PWM signals are shut down to save the power stage. Two kinds of fault protection have been provided: internal and external. If for some reason the PWM signals received from the deadtime block are such that they turn on the top and bottom switches of a given phase leg, then a fault signal is generated inside the EPLD (internal fault) and the PWM pulses are shut down. The internal fault can also be generated by the watchdog timer. These conditions should not occur under normal operating conditions. They are a protection against software failure.
External fault protection is activated when a gate driver senses an overcurrent through the power device and sends a fault signal to the digital buffer board. This fault signal is passed to the EPLD where it enables an IRQ counter. If the fault persists for a particular number of switching cycles (this number can be programmed into the IRQ counter), the PWM signals are shut down. In this way, the complete inverter is not shut down if the overcurrent condition happens in only
29
a couple switching cycles, which is acceptable because the gate drives automatically turn off the switch whenever overcurrent is detected.
Other EPLD Functions
The analog interface board has four A/D converters and two D/A converters that all communicate through the same 12bit data bus. In addition, the processor and A/D converters have different frequency clocks that are not synchronized. Therefore, to insure proper operation of the board, the timing of the control signals is critical. The logic that receives signals from the analog interface board and takes care of synchronization is implemented inside the EPLD. In addition, the system watchdog timer and the address decoding of the input and output ports are also implemented in the EPLD.
2.3 Multilevel Power Conditioning System as a Power Controller for a Superconducting Magnetic Energy Storage System
Historically, most of the interest in SMES systems seems to have been shown for utility type applications. The main motive was to combine the flexibility of an energy storage system with the ease of control provided by power electronics. In fact, by providing a buffer between the power generation and load consumption, the ease of control and operational flexibility of power systems could be greatly increased. There are certainly other utilitytype applications to which SMES systems could be applied, such as active filtering, supplying pulse power load and UPS applications.
The main challenge in developing such a PCS is to match the constant ac voltage and variable current operation of the utility with the slowly varying current and varying voltage of the magnet, while at the same time providing bidirectional power flow.
30
In addition, there is one rather unique characteristic of SMES systems: both dc voltage and dc current vary over a large range. This feature implies that the PCS must be overrated relative to the designed power transfer levels by a factor of three to four, or even more. This feature is one of the main reasons why the current source inverter is not a preferred way to design a SMES PCS while the series connection of a dc/dc chopper and ac/dc voltage source inverter (VSI) is. Clearly the later approach can have significant cost advantage because the whole system does not have to be overrated (only dc/dc converter).
2.3.1 Power Conditioning System Design approach Consistent with the preceding discussion, the PCS that was studied in this dissertation is implemented using a combined chopper/VSI approach. Considering the various tradeoffs
(already mentioned), the threelevel diodeclamped topology was chosen. Figure 2.18 shows the topology of the prototype PCS. Some of the advanced features of the system include the zerocurrenttransition (ZCT), soft switching technique, which allows the system to be rated at 250kVA, while switching with the 20kHz switching frequency. The details of the power stage and the intricacies of the softswitching design and control are the part of the doctoral research of Dr. DongHo Lee, and are described in detail in his work [G9].
31
p ip Sa1 Cdc Sa2 a b Sa3 Sa4 n in Sb3 Sb4 c Sc3 Sc4 Sb2 NP iNP Sch3 Cdc Sch4 Sc2 Sb1 Sc1 Sch1 L C Sch2
Lsc
ia L ib L ic L
A B Cac Cac Cac
R R CR
Figure 2.18. Power conditioning system for SMES.
32
2.3.2 Modeling and Control of the Chopper A functional diagram of the threelevel chopper is given in Figure 2.19. The task of the chopper is to control the power flow to and from the superconductive coil (SC). The circuit is bidirectional and, to avoid confusion, the SC side of the circuit is defined as the output side, while the side interfacing with the inverter is defined as the input side. Current direction in the SC is always the same so that the voltage on the filter capacitor, Vsc, determines the power flow. Fig. 2.20 shows the output voltages given in the form of space voltage vectors that are produced with different switching combinations of the Chopper. Although the space vector notation is not common for the dc/dc converters, such as in this case chopper, it is adopted here because it is consistent with the space vector notation introduced for the VSI in Section 2.3.4
The voltage vectors in Fig. 2.20 are named according to the switching combinations used to produce them. For example, if in Figure 2.19, switch Sx is connected to the prail and switch Sy is connected to the NP, the output voltage vector is po. There are nine vectors in all: two large vectors (pn, np) that produce output voltages of opposite polarity with the magnitude equal to the dclink voltage, Vpn ; four small vectors (po, on, no, op) with magnitude equal to Vpn /2; and three zerovoltage vectors (pp, oo, nn).
The task of the chopper modulator is to synthesize the voltage reference vector, VxyREF, which is a dc voltage and is always located on the VDC axis shown in Fig. 2.20. The modulation scheme in which the reference vector VxyREF is synthesized from the two closest vectors is adopted and will be discussed here. The advantage of this scheme is that it minimizes the voltage ripple on the SC and allows for effective control of the NP voltage balance.
33
p iSC LSC CF y sy VSC LF iLF x sx CDC io o NP
CDC
n
Figure 2.19. Equivalent switching model of threelevel chopper.
Zero vectors pp oo nn np no op poVxyREF pn VDC on
Fig. 2.20. Space voltage vectors of the chopper.
34
For d ∈ ( 0. The task of the NP voltage controller is to select from the redundant small vectors in . small and zero vectors are given by dL = 0 . no and op are used for VxyREF < 0. po and on are used for VxyREF > 0. The vectors resulting in iNP = iLF are referred to as positive small vectors. while the vectors resulting in iNP = iLF are called negative small vectors. where d ∈ [0 . the duty cycles of large. Both vectors in a pair generate the same output voltage. as shown inTable 2I. d S = 2 ⋅ d and d0 = 1−2⋅ d . 0. (21) (22) (23) respectively. then for d ∈ [0 . where small vectors po or on should be used for VxyREF > 0.5] . 1 ] duty cycles are given by d L = 2d − 1. d S = 2 − 2 d and (24) (25) (26) d0 =0 . while np. 1] . 5. Small vectors come in pairs. and hence do not affect the NP voltage. and no or op for VxyREF < 0. where vectors pn. 35 . Large vectors and zero vectors do not produce any NP current. with the objective of minimizing the NP current and maintaining the NP voltage as close as possible to onehalf of the dclink voltage.If we define the duty cycle as d = VxyREF V pn . but they produce the current iNP in opposite directions. iNP .
Figure 2. Figure 2. the duty cycle is then given by d = m and the switch duty cycles can be easily calculated from (21)(26).23. The average model is similar for both modes of operation. The modulation index is defined as m = VxyREF Vpn . the average model of the rectifier is not different from the average model of a bidirectional twolevel converter from Figure 2. 36 . Clearly.21.21. In the discharging mode. the chopper operates as a boost dc/dc converter. there is only a slight difference in the control design.22. the chopper operates as a buck dc/dc converter with the inner current loop and the outer voltage loop. V SC iSC C F L F i LF m×i L F V pn m× pn V C DC 2 Figure 2. Average model of the chopper. In the charging mode.Table 2I Neutralpoint current iNP for different small vectors. Positive Vxy Small Vectors Vnp /2 Vnp /2 po no iLF iLF io Negative Small Vectors on op iLF iLF io Neutralpoint control issues notwithstanding.
37 .i SC SC VSC LF CF i LF m power controller Hv Hi ZCin Figure 2. i S C L i S C C F L F V p n C D C 2 F m H Z V C o u t i H v D_E CRF Figure 2.23. Boost mode control of the chopper (discharging the SC).22. Buck mode control of the chopper (charging the SC).
c j ∈{ p . The subsequent analysis will reveal a few important properties of this type of converter. where s i . Vca. but rather in a static way determined by the type of modulation and the NPbalancing method applied. depends only on the NP current io and is not directly influenced by any other circuit parameters. ∆vpn .b .o . it will be shown that the error in NP. iDC + CDC vp i NP o o . Vbc. This result justifies application of the static method for the study of the NPbalancing problem. n } ∑s { } i.CDC vn + iDC ip p sa sb sc c b a ia ib ic L a vA vB vC Lb Lc in n Figure 2. Additionally.24. It will show that the error in the NP. j = 0 otherwise (27) i ∈ a . such as the one described in [B20].j 38 .3. Switching network of a threelevel VSI Equation (27) shows the definition of the switching functions that are used in (28) to write the switching network voltages referenced to the NP. 1 if i is connected to j = 1. ∆vpn .2. does not dynamically influence the output voltages Va b.3 Modeling of a ThreeLevel Voltage Source Inverter A functional diagram of a threelevel VSI switching network is given in Figure 2.24.
it becomes clear that the linetoline voltage depends on the DClink voltage and the error of the NP. v sn (211) s abp − s abn s abp + s abn v pn = s bcp − s bcn ⋅ + s bcp + s bcn ⋅ ∆v pn = s − s 2 s + s can can cap cap v pn =S⋅ + S n p ⋅ ∆v pn 2 39 . By rewriting the voltages of the top and bottom dclink capacitor in terms of the full dc link voltage and the NP error voltage (210). v n = − v pn 2 + ∆v pn By substituting (210) into (29).(28) v ao s ap s an v = s s ⋅ v p bo bp bn v vco scp s cn n From the definitions in (27) and (28) and by inspection of the switching network in Figure 2.. and there is no dynamic coupling between the two.. linetoline voltages can be written (29) as follows: (29) v ab s abp s abn v p v sn = v bc = s bcp s bcn ⋅ v v ca s cap s can n where : s abp = s ap − s bp s abn = s an − s bn . ∆vpn . as in (211). additional insight can be gained into the effect of the NP voltage error.24. (210) vp = v pn 2 + ∆v pn .
complete and makes no assumptions. (217). from Kirchof’s current and voltage laws and the definitions given.Defining the current vector as in (211). (212) and (213) can be written. Combining these equations with the result of the derivation in (215). dv p dv + − i DC + C DC ⋅ n = io = i DC + C DC ⋅ dt dt d v pn d v pn = C DC ⋅ 2 + ∆ v pn + C DC ⋅ dt − 2 + ∆ v pn = dt d = 2 ⋅ C DC ⋅ ∆ v pn dt (215) ( ) Finally. 40 . ia − ib 1 = ⋅ ib − i c 3 ic − i a (212) i sn (213) i p s abp s bcp s cap = ⋅ i sn i n s abn sbcn s can (214) io = i p + in = S n p [ ] T ⋅ i sn At last we can derive equation. Note that the vector i sn influences the NP current io through the same matrix of switching functions as the error of the NP voltage. This model is general. This equation completely describes the dynamics of the voltage error in the NP and will be included in the complete statespace model of the switching network. ∆v pn . (215). a dynamic relationship between input and output quantities of the threelevel switching network can be obtained (216). and that there is no dynamic relationship between these two either. which describes the dynamic relationship between the NP current io and the voltage error of the NP. a full set of dynamic equations describing the threelevel switching network is developed.
3. a new model in rotating refernce frame is obtained. can the last term in the equation (216). Equation (218) can be neglected as. (219) z( t ) = 1 s ⋅ z( τ ) ⋅ dτ Ts t −∫ s T T Under the assumption that ∆vpn is small. By transforming Equations (210) and (211). threephase variables can be transformed into a rotating dq reference frame. What remains is the wellknown. The switching matrices are changed as the modulator selects new voltage space vectors. Averaging the switching matrices using the averaging operator (219). They are computed by the d and q channel controller. 2. Figure 2.25. The control inputs into the system are dd and dq. 41 .4 Control of the Inverter in Rotating Coordinates In order to facilitate control design. These allowable combinations of the switches are called voltage space vectors. average statespace model of a two level inverter. sb and sc.(216) v d (i sn ) = 1 ⋅ v u − 1 ⋅ [S ]T ⋅ pn + dt 3⋅L 3⋅L 2 1 T + ⋅ S n p ⋅ ∆v pn 3⋅ L [ ] (217) d (v pn ) = 1 [S ]T ⋅ i sn − 2 ⋅ iDC dt C DC C DC d (∆v pn ) = 1 ⋅ S n p dt 2 ⋅ C DC (218) [ ] T ⋅ i sn Different switching matrices S and Snp correspond to each position of the switches sa. the small signal model is obtained.
transformed into the synchronous frame. The control inputs are obtained from a closed loopcontroller. conventional PI regulators were used in the implementation described here. During the period of SC’s discharging. dd and dq . Inverter average model in rotating dq coordinate frame.26. as shown in Figure 2. The design of the current controllers can follow a standard procedure for VSIs. d q = VREF V pn ⋅ sin θ . the inverter operates in the utility inverter mode. 42 .id dd × pn V 3× L 3 × × ×q Lw i 3× R VD C DC 2 3 × × 2 dd id 3 × × 2 dq iq V pn iq dq × pn V 3× L 3 × × ×d Lw i 3× R 0 Figure 2. as shown in Figure 2. Their relationship to the voltage reference vector in Figure 2. and the reference voltage vector is then calculated from (220). thereby controlling the amount of active and reactive power supplied to the utility. the modulator generates the corresponding duty cycles.28 is (220) dd = V REF V pn ⋅ cos θ . The daxis of the rotating reference frame is aligned with the utility line voltage space vector. VABC . The control inputs are the linetoline duty cycles.26.25. The task of the inverter is to control the magnitude and phase angle of the threephase output currents. Finally.
Inverter control in the utility mode (charging the SC). 3L 3 × ×× ×q id Lwι 3 R × ς πν CDC 2 3 L 3 V D × ×× ×d Lwi iq 3 R × 0 Z Iout dq dd H id H iq V∆Χ_ΡΕΦ H DC i i qref power controller dref Figure 2.27.3⋅ L 3⋅ ⋅ ⋅ Lwiq id 3⋅ R V D 3 ⋅L 3⋅ ⋅ ⋅ Lwid iq 3⋅ R 0 ZΙιν dq dd H id H iq idref i qref controller power Figure 2. Inverter control in rectifier mode (discharging the SC). 43 .26.
and three zerovoltage vectors. iNP and hence do not affect the NP voltage.26. the inverter is operates as a threephase boost rectifier with the objective of controlling the voltage on the dclink.. and the size of the dclink capacitors. VREF. Large vectors and zero vectors do not produce any NP current. twelve small vectors with magnitudes equal to V pn / 3 . The duty cycles of the vectors are found as a vector decomposition of the reference vector to the nearest three vectors. The outputs of the controllers can not be implemented directly by the power stage. . There are 27 switchingstate vectors: six large vectors with the magnitudes equal to 2 ⋅ Vpn / 3 . Both vectors in a pair produce the same linetoline voltage.In the SC charging mode. the NP control method. the magnitude and phase angle of the output phase current. V . as shown in Figure 2. six medium vectors with magnitudes equal to Vpn . the effect of the small vectors alone could be balanced in a way similar to that used in the chopper case. e. pon represents the state in which switch Sa is connected to the positive dc rail. The vector names represent the allowable states of the switches. the extent of which depends on the magnitude of the voltage reference vector.27. This may produce a lowfrequency voltage ripple of the NP. but the effect of the middle vectors cannot be directly controlled. Instead. Medium and small vectors affect the NP balance. The inverter output linetoline voltages. This 44 . Details of the modulation process will be studied in Chapter 3. Sb to the NP. The task of the VSI modulator is to synthesize the reference line voltage space vector. Vbc and V (produced by available switching ab ca combinations) can be represented as space voltage vectors in one plane. Consequently. the voltage reference vector VREF (as computed by the controller in the rotating reference frame) needs to be converted back to the stationary reference frame and implemented by the modulator. The space vector modulation (SVM) implemented in this system switches the three vectors nearest to the voltage reference vector. and Sc to the negative dc rail. The active power flow is controlled by the chopper. but produce the NP current in opposite directions. by PWM of the available voltage space vectors. but only the small vectors come in pairs.g. as shown in Figure 2. while the inverter still independently controls the reactive ac power depending on the utility requirements.
where it will be shown that the worst case scenario is when the VSI supplies a purely reactive power at the full current capacity. Voltage space vectors of a threelevel inverter.issue will be addressed in Chapter 4. In such a case. V bc δ Zero vectors: ppp ppn ooo nnn pon npn opn V npo opo non R F VA C E B θ npp opp noo q wt ppo oon poo onn pnn nop V ca oop nno pop ono pno V ab nnp onp pnp Figure 2. large a NP ripple can be expected and active compensation with the chopper may be essential. 45 .28.
simulations and experiments. 46 .3.30. and operates with 20kHz switching frequency. Experimental SMES PCS system. INVERTER CHOPPER Figure 2.2. The simulated and measured current loop gain in the d channel is given in Figure 2. Simulations and the experiments were performed with the inverter operating alone in the rectifier mode at 28kW power level. The threephase utility line voltage was 208V with the 77A phase rms current.29.27 for the current loopgain crossover frequency of 1kHz. The system in the picture is rated at 250kW. This produced 600V on the dc link that was loaded with 13Ω resistive load. the dclink current of 150A. including the active filtering of loads with high harmonic contents. with a dclink voltage of 1800V.30 show good agreement between the design predictions.29. The achieved wide currentloop bandwidth confirms the feasibility of using the system for power quality applications. The results in Figure 2.5 Simulation and Experimental Results The average model simulation results were validated on the experimental setup shown in Figure 2. The PI regulators Hid and Haq were designed using the model in Figure 2.
20 0 [dB] 20 40 60 180 3 4 60 [deg] 60 180 10 100 [Hz] 3 1 10 4 1 10 Figure 2.31.31. The voltage regulator. the results demonstrate the good agreement between simulations and measurement. The simulated and experimental measurements of the voltage loopgain are shown in Figure 2. Simulated and measured rectifier voltage loopgain. HDC .40 20 [dB] 0 20 180 60 [deg] 60 180 10 100 [Hz] 3 1 10 1 10 Figure 2. was designed for the loop gain crossover at 20Hz. 47 . Again.30. Simulated and measured rectifier current loop gain.
and 3) rectifier phase current (100 A/div).32. id . 2) dchannel current. 48 .32.The response of the rectifier to the step change in current command with the voltage loop disabled is shown in Figure 2. this shows the performance of the rectifier inner current loop. Rectifier response to change in current command: 1) current command (20ms/div). 1 2 3 Figure 2.
in the sense that it directly applies to the problem of charge balancing in the dclink capacitors of a threelevel NPC converter. The idea behind the research was to try to find a mathematical model that could describe the effect of the load on the charge balance as well as to find guidelines for sizing the dclink capacitors in order to achieve the optimal voltage ripple amplitude for any operating condition. In addition. and through implementation of linear transformations that transform the switching vectors from a space where the algorithm is difficult to implement to a space where implementation of the algorithm becomes almost trivial. It was already mentioned that one of the important problems of multilevel converters is the modulator design (implementation).and more level.4 Modeling and Control Approach and Problem Formulation The principle research and development effort in this dissertation was to better understand how to control and model three. Therefore.2. it offers some intuitive insight into the operation 49 . Important modeling and control concepts were first understood. and finally generalized when possible. Still. Only than was the understanding of the fundamental concepts applied in the design of the system as a whole. The SVM approach is perhaps the most powerful. because it allows more freedom to control and optimize the switching patterns than any other modulation approach. for converters with higher number of levels it becomes too cumbersome for realtime implementation. The modeling approach was geared towards the SVM and is general in the sense that it can model almost any type of SVM. then analyzed. at the same time. This goal was successfully achieved through the geometric analysis of the space vector diagrams. it is currently one of the most popular topologies and was used for the prototype PCS. the first big task in this work was to simplify and generalize the implementation of a SVM for converters with any number of levels. threephase converters for use in SMES power conditioning systems as well as in other applications. The approach was really bottom up. The second problem that was analyzed is perhaps less general.
multilevel power electronics system. and therefore allow savings in the size of the dclink capacitors? The answer to the above question is positive. 50 . verified on the prototype PCS. 2. Finally the theoretical results were Further thinking along the lines of the charge ripple leads almost naturally to the following question: If there are certain operating conditions that cause a significant charge ripple in the dclink capacitors.and the limitations of the chargebalancing algorithms. The properties of the feedforward algorithm were studied on two different time scales. By applying a feedforward controller (proposed in this work). is there a way to implement a controller such that it can cancel the effect of that ripple in the output waveform. At the same time. Some of the results were verified through the experiment.5 Conclusions This is an overview chapter that covers a wide range of issues important in the design of a generic threephase. The initial study included the development line cycle averaging model and its linearization for the purpose of feedback. The issues range from the selection of the topology and the requirements of the digital control hardware all the way to an overview of the PCS for the SMES application and the formulation of the research problems studied in this dissertation. it is possible to compensate for any size of the imbalance in the dclink capacitors. Than a switching cycle average model was built and simulated in the computer in order to study the system behavior in more detail. this chapter is an overview of the issues that had to be resolved during the course of the design and implementation of the prototype SMES PCS.
DIGITAL CONTROLLER DSP MODULATOR SOFTWARE NEAREST THREE VECTORS DETECTION → V REF EPLD POWER STAGE DUTY CYCLE COMPUTATION VECTOR SELECTION REGULATOR SENSORS Figure 3. This situation places significant computational difficulties in front of the space vector PWM modulator.1 shows the structure of a generic threephase converter digital control system. 51 .3 A FAST SPACE VECTOR MODULATION ALGORITHM FOR MULTILEVEL CONVERTERS With the increase of the number of levels. the number of allowable switching states in the converters is also rapidly increasing.1. Architecture of digital controller for threephase converters. Figure 3. which is typically implemented in software and operating in real time and for high switching speeds.
Vdc Vdc + + . 52 . . . 1 2 . + + Vdc Vdc .2..2. a multilevel converter is a voltage synthesizer that generates its output voltage from many discrete voltage levels. n1 n1 n2 1 n2 a 0 n1 n2 1 b 0 1 c 0 1 Figure 3. . without a loss of generality. .. .nn. Because the functionality of each phase leg can be represented by a singlepole nthrow switch. . it becomes clear that each switching state (combination of phaseleg switches) produces uniquely defined threephase line voltages (31). which can be represented as vectors in a threedimensional Euclidean vector space1 : (31) V = [vab → 1 Threedimensional. Switching States and Reference Vector In essence. . most recently studied multilevel topologies can be well represented (for the purpose studying SVM).. primarily for the elegance of mathematical description.. a b c vbc v ca ]T .3. Functional diagram of a threeleg nlevel converter. even though it might not be the most widely accepted way of representing vectors in threephase systems.0 . Therefore. Euclidean coordinate space is chosen to present vectors.1 Switching Vectors... . . by the functional diagram in Figure 3. . .
the existence of various switching states of some of the switching vectors has some merit. which make N vectors = 1 + 6∑ i i =1 n −1 switching vectors. representing vectors in the line voltage coordinate space is a nice mathematical abstraction to consistently explain the spatial relations between the switching vectors. because of Kirchof’s voltage law. if the location of the equivalent phase switches of the phases a. Having a variety of switching states from which to choose provides the important degree of freedom to balance the charge in the split dclink capacitors and/or optimize the switching pattern. with their corresponding switching states. There are N states = n 3 switching states. 53 . and a decision must be as to which vector to switch. j .3 are all the switching vectors of a threelevel converter. j and k. b and c were in the positions i. n − 1] . because the converter switches are controlled on a perphase basis.For example. On the other hand. respectively. The reason to naming the vectors with their corresponding switching states is purely practical. Finally. Clearly the number of allowed combinations increases rapidly as the number of levels increases. where i . This means that all the switching vectors of a threelevel converter lie in the plane. in the sense that different switching vectors can be implemented with a various number of switching states. It is somewhat important to make the distinction between switching states and switching vectors. which is really an equation of the plane in the linetoline coordinate system. the sum of the linetoline voltages always adds up to zero. and that is how they are usually represented. then that switching state could be represented by the switching vector (32) V ( ijk ) = V dc ⋅ [i − j → j − k k − i ]T . as well as which switching state to select. Shown in Figure 3. On the other hand. k ∈ [0.
Using the definition of vector norm. 54 . Coming back to the same threedimensional representation.2 002 102 112 001 vca 012 122 011 022 021 vca/Vdc 1. as follows: cos(ω ⋅ t + ϕ) cos(ω ⋅ t − 2 ⋅ π / 3 + ϕ) . ⋅ cos(ω ⋅ t − 4 ⋅ π / 3 + ϕ) (33) V REF = Vl −l → and also has only two degrees of freedom.5 1 1. the length of the reference vector is (34) → V REF = V REF ⋅ V → → REF = Vl − l ⋅ 3 2 .5 0 0.5 2 2 1 0 202 101 212 vab 201 211 100 200 vbc 221 110 121 010 120 020 → V REF 210 1 1 2 2 220 0 2 1 vbc /V d c v ab/V dc Figure 3.5 1 0.3. a voltage reference vector that is to be synthesized with the help of switching vectors can also be represented in vector form. Switching states of a threelevel converter.
computed by the controller. → → By substituting (34) and (35) into (36). In other words. using the PWM of several switching vectors. (38) with the following additional constraint on the duty cycles: 55 .while by the same definition. → Because the maximum length of the reference vector (33) that can be synthesized equals the radius of the largest circle that can be inscribed in the outermost hexagon. the maximum amplitude of the undistorted linetoline voltage that a multilevel converter can synthesize is Vl − l max = ( n − 1) ⋅ Vdc . the length of the longest switching vector is (35) V max = 2 ⋅ ( n − 1) ⋅ Vdc . it is the task of the modulator to approximate the reference vector. Arguably.2 Space Vector Modulation of a ThreeLevel Converter It is the task of the modulator to decide which position the switches should assume (switching state). the best way to synthesize the voltage reference vector is by using the nearest three vectors → → → → V REF = d1 ⋅ V 1 + d 2 ⋅ V 2 + d 3 ⋅ V 3 . (37) 3. and the duration needed (duty cycle) in order to synthesize the reference voltage vector. the maximum length of the reference vector for a general multilevel converter is (36) V REFmax = V max ⋅ cos(π / 6) .
β npn opn ppn npo opo non ppo oon pon → M REF poo onn pnn npp opp noo α nop oop nno ono pop pno nnp onp pnp Figure 3. This decision will have important consequences for the reatime implementation of the modulation algorithm. 1. 0 are named p. In addition.(39) d1 + d 2 + d3 = 1. o. θ ). 56 . The two coordinate frames that first come to mind are the rectangular. Voltage space vectors of a threelevel converter. there is a certain degree of freedom in choosing the coordinate system in which to represent those vectors. It was shown in the previous paragraph that the voltage space vectors of a threephase converter are always located in the plane. n is perhaps a little more common. β ) and the polar ( ρ. for the threelevel converter. So far. Clearly. although naming the switching states with the numbers is much more general (and applicable for any number of levels converter). equivalent nomenclature in which 2.4. and that the goal is to approximate the reference vector using the nearest three vectors (NTVs) that satisfy (38) and (39). the two will be used interchangeably.4. and that is how they are represented in Figure 3. it has been shown that all the switching state vectors and the reference vector can be represented in a twodimensional space. often called ( α.
The first step the modulator needs to perform is to identify the NTVs. However. if the expression (310) m− 1 3 ⋅ cos( θ ) − sin( θ ) >0 57 . and because the conversion from rotating reference frame to the stationary polar reference frame requires computation of the arctangent function to determine the angle θ of the reference vector.Because the control loops are implemented in the rotating reference frame. when the NTVs are located. (which will be developed in the next chapter) turns out to be easier to develop in polar coordinates. The small triangle sides are described with simple linear equations in the α. The implementation of the algorithm in polar coordinates follows the same steps already outlined in the previous paragraph. which is really to → determine the small triangle in which the tip of the reference vector. is located. β coordinate system. The small triangle (NTVs) are than found by evaluating the sign of some linear expressions with reference vector coordinates substituted. The first step in detecting the NTVs is to find the sixtydegree region. if the modulator will operate within the feedback system. then solving the system of three linear equations with the three unknowns given in (38) and (39) results in the duty cycles. because the angle of the reference vector is known. which requires evaluation of the sign of some nonlinear expressions. the SVM algorithm in polar coordinates will be presented in more detail. V REF . Finally. because the derivation of the model describing the charge balance in the NP. For example. The inner triangle is always within a certain sixtydegree region. which in polar coordinates comes naturally. Then the small triangle has to be located. the α. β reference frame is perhaps a better choice for the realtime implementation1 .
is true (which is really the equation of the line in polar coordinates system that connects the small and the medium vector from Figure 3. ) (314) d M = 2 ⋅ m ⋅ sin( θ ) and d L = −1 + m ⋅ ( 3 cos(θ ) + sin( θ )) . (315) 1 θ In case of the modulator is running open loop there is no need to implement arctangent function to compute the coordinate since it can be freely incremented depending on the desired line frequency. 58 . (312) The solution for duty cycles dS0 . then the NTVs are located in the outer triangle in Figure 3.5). dM and dL can be shown to be (313) d S0 = 2 − m ⋅ ( 3 ⋅ cos(θ ) + sin( θ ) .5. The three duty cycles can be found by solving the system for one vector (311) and one scalar (312) equation: (311) VREF=dS0 VS0 + dM VM + dL VL and dS0+dM+dL=1.
d S 0 = 1 − m ⋅ 2 ⋅ sin( θ ) . dS0+dS1 +dM=1. 59 . Equations (316)(320) give the duty cycles for the middle triangular region from Figure 3. Synthesis of VREF in outer small triangle.6: (316) (317) (318) VREF=dS0 VS0 + dM VM + dL VL.5. It was already mentioned that the equations for every triangular region are somewhat different. d S 1 = 1 + m ⋅ (sin( θ ) − 3 ⋅ cos( θ )) and (319) (320) d M = −1 + m ⋅ (sin( θ ) + 3 ⋅ cos( θ )) .1 3 ⋅ V pn 2 ⋅V pn 3 1 ⋅ V pn 3 VM pon VREF p n n VL θ o n n VS0 poo Figure 3.
If the Equation (321) is true the vector is in the inner small triangle. Then the duty cycles are found as follows: (322) VREF=dS0 VS0 + dS1 VS1 . (323) (324) (325) 60 .6.VS1 ppo oon VREF θ VS0 o n n poo VM pon Figure 3. (321) m− 1 3 ⋅ cos( θ ) + sin( θ ) < 0. dS0+dS1 +dZ=1. d S 0 = m ⋅ ( 3 cos( θ ) − sin( θ )) and d S 1 = 2 ⋅ m ⋅ sin( θ ) . the sign of the left hand side of the Equation (321) determines whether the reference vector is located in the inner smaller triangle. Synthesis of VREF in middle small triangle Finally.
each new level adds one more ring of equilateral triangles to form a new hexagon of voltage space vectors.7. Finally. it is clear that whether implementing the SVM in the rectangular or the polar coordinate system. starting with the innermost hexagon that corresponds to the twolevel converter. Fortunately. and even the arctangent function for the implementation of the closedloop system. the implementation becomes increasingly more complex as the number of levels increases.8) to that of the voltage space vectors from Figure 3. 3. most of these implementation problems (and more) can be avoided with the new SVM algorithm described next. which requires the division. In fact. 61 . sine and cosine functions (or corresponding lookup tables). 3. This is especially true for the implementation of the algorithm in polar coordinates.VS1 ppo oon VREF θ VS0 o n n poo VL Fig. a visible pattern can be noticed in their hexagonal structures. Synthesis of VREF in inner small triangle.3 The New Space Vector Modulation Algorithm Comparing the arrangement of voltage space vectors of a fourlevel converter (from Figure 3.4.
vbc . h ( vab .β 030 130 230 330 031 131 020 2 231 120 331 220 320 032 021 132 232 121 010 233 122 011 223 112 001 1 332 221 110 322 211 1 100 323 212 101 210 321 310 → M REF 033 133 022 311 2 200 300 α 0 023 123 012 201 312 301 013 113 002 213 102 313 202 302 003 103 203 303 Figure 3. − Vdc − V dc (326) . it turns out that this regularity in the structure of the hexagons can be used to more efficiently represent the vectors in the plane. 62 . One such basis set is V dc 0 → → g ( vab . The voltage space vectors of a fourlevel converter.vbc . and can be used to generalize the SVM algorithm. Fortunately. The trick is to use a set of nonorthogonal vectors as a new basis in which to represent the switching vectors and the reference vector.8. V dc .vca ) = 0 .vca ) .
(327) V REF ( g . h } coordinate system have only integer coordinates. Naturally the transformation matrix differs depending on the source coordinate system.1 Coordinate Transformation The first step in the algorithm is to transform the reference vector V REF into a twodimensional coordinate system. Matrix (328) transforms the reference vector given in the linetoline coordinate system (33) into the (g. For example. and at the same time normalizes the reference vector with the length of the basis vector. all the switchingstate vectors in the { g .vca ) 2 ⋅ − 1 − 1 − 1 2 −1 → → → → (328) T= 1 3 ⋅V dc Because of the normalization.h ) = T ⋅ V REF ( vab .h) coordinate system.vbc . This can be achieved using a changeofbasis linear → transformation. 63 .3.3.
2) (2.1) (2.2) h (1.3.0) α g (2.1) → M REF g (1.1) (1.2 Detection of Nearest Three Vectors The fact that all the switching vectors have integer coordinates is advantageous because the four vectors nearest to the reference vector can be identified simply as vectors the coordinates of which are combinations of upper and lower rounded integer values of the reference vector coordinates.1) (0. Switchingstate vectors of a threelevel converter in a hexagonal coordinate system.2) Figure 3.2) (1.2) (1.V lu = V REFh V REFh .0) → (0. as follows: (329) → V REFg → V REFg V u1 = . 3. V ll = V REFh V REFh 64 .β (2.0) (1.1) (1.2) (0.1) h → (1. → V REFg → V REFg V uu = .1) (2.9.1) (0.0) (2.
their corresponding duty cycles can be found by solving (38) and (39). → → → → → → → → → → if V 3 = V ll . where V 1 = V ul . because all switchingstate vectors always have integer coordinates. The vectors V u 1 and V lu are always two of the NTVs. as the reference vector. (331) If the sign is positive. and V 3 = V ll or V 3 = V uu . (330) g + h = Vu lg + Vulh . Fortunately. → → 3. This completes the identification of the NTVs for the nlevel converter.3 Duty Cycles Computation Once the NTVs are identified. V 2 = V lu .3. or → 65 . then the vector V uu is the third nearest vector. otherwise. The third nearest vector is one of the remaining two vectors → → → → located on the same side of the diagonal. (332) d lu = VREFh − Vllh and d ll = 1 − d ul − d lu . Therefore. the third nearest vector can be found by evaluating the sign of the expression VREFg + VREFh − ( Vu 1 g + Vulh ) . the solutions are essentially the fractional parts of the V REF coordinates.The endpoints of the nearest four vectors form the equalsided parallelogram that is divided into two equilateral triangles by the diagonal connecting vectors V u 1 and V lu . vector V ll is the third nearest vector. or more precisely d ul = VREFg − Vl lg . however.
a threelevel converter and reference vector with amplitude Vl −l = 1.8 ⋅ Vdc at the instant ω ⋅ t = 50 ⋅ π . V uu = .643 = 1.8 ⋅ Vdc ⋅ 0.h VREFca (335) VREFg =T VREFh From there. V lu = . which can be represented as 180 → → → (334) V REF 0. d lu = −( VREFg − Vuug ) and d uu = 1 − d ul − d lu . the nearest four vectors can be found simply as vectors the coordinates of which are the combinations of upper and lower rounded values of the voltage reference vector: → 2 → 1 → 2 → 1 V ul = .157 ⋅ VREFbc = 0. 0 1 1 0 (336) In order to find the third nearest vector. (337) 66 . if V 3 = V uu .342 .616 − ( 2 + 0 ) = −0. Consider. the reference vector in the proposed twodimensional coordinate system is VREFab 1. for example.(333) d ul = −( VREFh −V uuh). − 0.157 + 0. all that needs to be done is to evaluate the sign of the expression: 1.16 < 0 .V ll = . g .616 .985 Using the changeofbasis transformation (328).
0 .4 SwitchingState Selection This last step requires a transformation from a twodimensional coordinate space back to the threedimensional space of switching combinations.227. where k . one more step remains if the modulator is to generate the commands for the equivalent switches from Figure 3.157 . 3. a vector [1.Because it is negative. Finally.3. 0 ]T transforms into two switching combinations: [1. which allows some freedom in optimizing the switching strategy. Nevertheless. In general. → (338) This completes the selection of the vectors and the computation of their corresponding duty cycles. 1 . k − g − h → (339) For example. k − g . there is no unique solution for this problem. d lu = 0. Which of the valid switching states to switch. n − 1] . all the switching states of a switching vector satisfy a relationship k k − g . computing the expressions in (16) gives the solution for the duty cycles d ul = 0.616 and d ll = 0. The first step in choosing the best switching state or states at any given moment is to find all the switching states that make a switching vector in question V = [g . vector V ll is the third nearest vector. in the case of a threelevel converter. 1]T . usually requires some 67 . h ]T . 0 ]T and [2 . and for what part of their corresponding switching vector duty cycle (in case the balancing algorithm requires use of several small vectors during one switching cycle).2. in this example. Mathematically. k − g − h ∈ [0 .
3level Vab/Vdc 4level 5level Vbc/Vdc Vca/Vdc Figure 3. simulated switching waveforms for three. which usually requires some knowledge of the phase currents and/or dclink voltages. The properties of the algorithm can best be explained by presenting the flow chart of the algorithm implemented on 68 .4 Practical Implementation of The New SVM Algorithm It was already mentioned that the principal strength of the proposed algorithm is in its ease of implementation.additional information. The ease of implementation in this case is augmented by the very small number of instructions and with the suitability for implementation that can run in real time.and fivelevel converters are shown in Figure 3. 3. it depends on the strategy for balancing the dc sources. Normalized linetoline PWM voltage waveforms for three.and fivelevel converters. and for converters with any number of levels.10. four. for high switching frequency. In fact. four. Finally.10.
11.h) coordinate system. The next step in the algorithm is to find the nearest four vectors by computing the upper and lower rounded values of the reference vector’s coordinates in the (g. The algorithm can be divided in three steps as shown in Figure 3. Figure 3.gl.h) coordinate system Compute: gu .hu. the (α. The first step in the algorithm is to change into the (g.h) basis.the DSP that controls the prototype SMES PCS.11. Overview of the new SVM algorithm. 69 . Other changeofbasis transformations can be constructed to transform the reference vector from the (d.12 shows how this operation can be conveniently implemented using a truncation function.β) or any other basis into the (g. usually available in most floatingpoint DSPs.h) coordinate system which is in fact the linear transformation as in (328).q). Start Change VREF into (g.hl Compute the duty cycles and read the switching combinations End Figure 3.
the algorithm is unique and works for any number of levels converter. one last coordinate transformation still remains.Start Yes VREFg>0 No gu=trunc(V )+1 REFg gl=trunc(V ) REFg gu=trunc(V ) REFg gl=trunc(V )1 REFg Yes VREFh>0 No hu=trunc(V )+1 REFh hl=trunc(V ) REFh hu=trunc(V ) REFh hl=trunc(V )1 REFh End Figure 3. the duty cycles for all three vectors are computed based on the simple linear expressions from (332) and (333). In real implementation. So far. because it essentially transforms a vector existing 70 . This transformation is not unique. based on the sign of the expression (331). the algorithm branches into two branches. One branch treats the case in which the third nearest vector is Vll. the duty cycles need to be computed and the switching states selected. Computing the nearest four vectors.13 shows one possible implementation approach in which.12. In both cases. The set of nearest four vectors needs to be reduced to the nearest three vectors. the other branch treats the case in which the Vuu is the third nearest vector.h) coordinate system into the switching functions. The remaining transformation converts the vectors in the (g. Figure 3. however.
13. loss of converter levels and the overvoltage condition in the devices. control of the charge balance is performed in this part of the modulator algorithm by selecting the switching states based on the current direction information and based on the error in the dclink capacitors’ charge (voltage) imbalance. At the same time. 71 . Computation of duty cycles and selection of switching states. very often.in a twodimensional space into a switching combination that exists in threedimensional space (all three singlepole triplethrow switches from Figure 3. selection of switching combinations affects the charge balance of the dclink capacitors. A consistent choice of “wrong” switching states leads inevitably to the charge imbalance.2 can be chosen independently). In fact. Start Yes (VREFg+VREFh) (gu+hl) > 0 No dul =huVREFh Write time to altera dul =VREFggl Write time to altera switcing_state_selection for the vector Vul dlu=guVREF g Write time to altera switcing_state_selection for the vector Vul dlu=VREFhhl Write time to altera switcing_state_selection for the vector Vlu switcing_state_selection for the vector V uu switcing_state_selection for the vector Vlu switcing_state_selection for the vector V ll End Figure 3.
The 0 equivalent switch location is coded with #3. Unfortunately. The core of the algorithm is a simple formula that computes the address of the switching state in 72 . or according to representation from Figure 3.Figure 3.h) coordinates of the threelevel converter.14 shows one implementation of the subroutine that selects the switching states based on the (g. That number tells the NP balancing algorithm which phase leg current is connected to the NP. The fourth hexadecimal number from right to left is used only for the switching state of the small vectors. and finally the 2 switch location is coded with #C. The switching states and how they affect the dclink capacitors’ charge balance in case of threelevel converters is studied in detail in Chapter 4.2. the 1 location with #6. phase c. in which the rightmost hexadecimal number is reserved for the location of the rightmost phase. at this point there does not seem to be a universal algorithm available that can select the appropriate switching states for converters with any number of levels. That is the reason that the threelevel modulator for the SMES system was implemented based on the lookup table ( Table 3I). The switching states are stored as hexadecimal numbers.
the other switching combination is located at the address M8+4. the chargebalancing algorithm selects the switching state of the small vector that will minimize the charge imbalance. After the address is found. In case of small vectors that can be implemented with two switching combinations. based on the (g.Table 3I. If the NP is not equal to 0.h) coordinate of the switching vector (340) M8=7*(g+h+2)+(h+2). a much more indepth treatment of the NP balancing problem is located in the next three chapters. Finally. there is only one choice of the switching state for the given switching vector and that is the switching state from the switching vector selection subroutine. one of the two switching combinations is selected to minimize the charge imbalance. If the variable NP from the algorithm is 0. based on the measurement of the voltage imbalance and the direction of the phase currents. The main purpose of presenting the new SVM algorithm in this section was to illustrate the simplicity of its implementation in a reallife example . 73 .
Switchingstate selection with dclink capacitor charge control.I8) NP=shift right sw_state by 12 Yes NP not equal 0 No Write sw_state to Altera Yes Yes NP=1 No ph_a_svs=0 No NP=2 No Write sw_state1 to Altera Write sw_state to Altera Yes ph_b_svs=0 No Write sw_state1 to Altera Write sw_state to Altera Yes ph_c_svs=0 No Write sw_state1 to Altera Write sw_state to Altera Figure 3.Start I8=sw_state_table M8=7*(g+h+2)+(h+2) M9=M8+4 sw_state=PM(M8.14. 74 .I8) sw_state1=PM(M9.
2) (0. Switching Vectors.1) (1.2) (1. Switching State Table Location 0. 22. 30.h Switching States (per Phase) Hexadecima l Switching Number 002 #33C 012 #36C 022 #3CC Small Vectors NP Currents (1.Table 3I The switchingstate lookup table. 14.1) (2.h (0. 31. 5. 24.1) (1. Vg. 13.1) (1.1) (1. 23. 29. 21.1) (1. 17. 12. 1. 19.2) 102 001 122 021 112 011 202 212 010 020 101 121 201 100 221 120 211 110 200 210 220 #63C #3336 #16CC #3C6 #66C #366 #C3C #2C6C #2363 #3C3 #636 #6C6 #C36 #1633 #3CC6 #6C3 #C66 #663 #C33 #C63 #CC3 ic ia ic ia ib ib ib ib ia ic ia ic 75 .1) (2. Large & Medium Vg. 11.0) (0. 7.1) (0.1) (2. 10. 6.0) (2. 25.1) (0. 32. 26.0) (2. 20.2) (1.0) (1. 9.2) (1.2) (1.1) (1.1) (2.0) (0. 16. 28. 8. 2. 18. 4. 3. 15.0) Small Switching Vectors. 27.
The switching state table becomes larger. reallife implementation on the Analog Devices Shark floating point processor will be discussed next. while it does include the “house keeping” instructions that communicate with PWM generators. Perhaps as much as twothirds of the execution time can be saved in an optimized implementation. compared to the execution time in this implementation. such as the use of two instructions per switching cycle and the use of shadow registers to pass the variables. where by increasing the number of levels the algorithm essentially does not change. its. 76 . in less than 5µs. Number of Instructions Change of Basis Finding Nearest Four Vectors Duty Cycle Computation Switching State Selection (Worst Case) TOTAL NUMBER OF INSTRUCTIONS 9 25 29 3*28 147 Even this suboptimal implementation (listed in the Apendix) executes on the Shark DSP operating at 33 MHz clock speed. the application of this algorithm we can think of multilevel converters operating at the switching frequencies in excess of 100 kHz. Therefore. Thus. in part because due to its complexity SVM is generally not used for the converters with more than three levels. but the speed of execution remains essentially unaffected. Nevertheless. Table 3III shows the rough number of instructions for each step of the algorithm. The other difficulty is the fact that implementation of the algorithm depends heavily on the platform. it is very hard to compare algorithms implemented on different platforms. substantial savings in the number of instructions can be made with an implementation that makes use of all the available processor resources. as an illustration of algorithm’s performance. Therefore. Table 3III The performance of the algorithm implementation for a threelevel converter. It is a rough number because the implementation does not make use of all the processor resources that can speed up the algorithm execution.It is difficult to come up with one number that will demonstrate the superiority of this approach.
the efficiency and the ease of implementation of this algorithm makes it well suited for simulation on digital computers. since the number of steps involved in computation is always the same regardless of the location of the reference voltage vector. In addition.3. it can become a very useful tool in further exploration of the properties of multilevel power converters. The algorithm is general and applicable to converters with any number of levels.5 Conclusions This chapter introduced the fast new space vector algorithm for multilevel threephase converters. 77 . The algorithmic description of the reallife implementation shows its simplicity.
perhaps for the first time. and in a struggle with multiple dc sources. hardware and software realizations. the benefits of using the multilevel topology come with a price. any disturbance in the dc link will propagate into the output of the converter. more difficult to make reliable. introduced in the next chapter.4 NP BALANCING PROBLEM IN THREELEVEL DIODECLAMPED CONVERTERS Figure 4. it is probably clear that regardless of the implementation. Since one of the main advantages of this technology is its ability to utilize the switching devices rated for onehalf of the dclink voltage. In addition. The price. Threelevel diode clamped topology is no exception to this rule. At least not without introducing some of the techniques that will be. a diodeclamped converter. 78 . At this point. a disturbance in the charge balance (NP voltage) can not. This is one of the critical requirements of the control system. must be paid in the form of the more complicated. and although the disturbance in the dclink voltage can be compensated for by the control system. the NP must be maintained at (or very close to) half of the dclink voltage.1 shows one of the most widespread realizations of a threelevel converter. almost as a rule.
With all these assumptions in place. the output voltage can be assumed ideal in the average sense. in order to help study the properties of NP balancing. set at onehalf the dclink voltage. because of this assumption. which will result in a more intuitive understanding of the NP balancing problem in threelevel diodeclamped converters. certain assumptions.p ip Sa1 Sb1 Sb2 b Sa3 Sa4 Sb3 Sb4 c Sc3 Sc4 Sc1 Sc2 ia L ib L ic L A B Cac Cac Cac R R CR Cdc Vpn NP iNP Cdc Sa2 a n in Figure 4.1. Under this assumption the voltage in the NP will not change from its initial condition. Circuit schematic of a threelevel VSI. 79 . are made (that will be relaxed later). which allows the use of ideal current in place of the load. Perhaps the most important assumption is the infinite size of the dclink capacitors. In this chapter. In addition. this study can concentrate on modeling the properties of SVMs with respect to the current in the NP.
and include guidelines for sizing the dclink capacitors. the model will try to explain the theoretical limitations of the control authority over the dclink capacitors charge balance. it is continuously being synthesized by applying different switching vectors for the various length of time (duty cycles). this chapter develops a mathematical apparatus that is general enough to describe the average current in the NP.4. Switchingstate vectors of the threelevel VSI.2 shows all the available voltage space vectors. In addition. as explained in Chapter 3. As the reference vector rotates 360 degrees. However. and their corresponding switching combinations. 80 .1 The Effect of Different Switching States on the Current in the NP Figure 4. and to study the contributions of different vectors to the NP current.2. β 020 2 120 021 121 010 122 011 012 112 001 1 221 110 211 1 100 101 212 201 210 M REF 200 α 2 → 220 022 0 002 102 202 Figure 4.
3.p CDC Sa iNP NP ia Sb Sc ib ic o CDC n Figure 4. Table 4I NP current iNP for different space vectors. The position of equivalent switches for the switching state pon. Small Vectors (positive switching state) onn ppo non opp nno pop Small Vectors (negative switching state) poo oon opo noo oop ono iNP iNP Medium Vectors pon opn npo nop onp pno iNP ia ic ib ia ic ib ia ic ib ia ic ib ib ia ic ib ia ic 81 .
In order to effectively use the switching functions. because vectors from each of the three categories affect the charge balance in a similar way. 82 . Finally. For example. it is convenient to divide the switching vectors into three categories according to their lengths: large. Figure 4. Figure 4.3 shows the NP current for switching combinations of different small and medium vectors. That is the source of the positive and negative atributes. This categorization can also be used to explain the effects of these vectors on the charge balance. medium and small. a fullphase current flows into the NP for the duration of its duty cycle. Choosing one of the two switching combinations in a pair changes the sign of the phase current that vector connects to the NP.2 Neutral Point Current Modulation The previous chapter discussed the methods of synthesizing the reference vector. vector pon directly connects the phase b to the NP. every small vector has two realizations: one positive and one negative switching function. 4. it is beneficial to have the reference vector represented in polar coordinates.3 shows how one of the medium vectors. and as a result.For the purpose of model development. The modeling approach to be presented uses the switching functions to model the effect of different small and medium vectors connecting different phase currents to the NP. The discussion will now turn to the attempt model the NP current and show its dependence on the load current. The medium vectors connect one of the phase currents for the duration of their duty cycles thus causing a charge imbalance. the large vectors do not affect the NP at all. The effect of the medium vectors is not controllable because it depends only on the duty cycle and the load power factor.
with two duty cycles. different NP current flows for different switching combinations. the effective duty cycle of the vector VS0 can be expressed. 1] .In this case.4. middle and outer small triangles still holds. as explained in Table 4I. Outer small triangle. A controlling effect of the small vectors can be described through the introduction of the new variable called the NP currentmodulation index. This index. For example in the outer small triangle of the Figure 4. For example.4. the duty cycle of the positive switching combination (onn) is d S 0 _ pos = ( 1 + mS 0 ) ⋅ d S 0 / 2 . 1 3 ⋅V pn 2 ⋅V pn 3 1 ⋅V 3 pn VM pon VREF p n n VL θ o n n VS0 poo Figure 4. m S 0 ∈ [− 1.4 consists of two components: the noncontrollable component from the medium vector and the controllable component from the small vectors. represents the relative duration of the positive (onn) and negative (poo) small vectors within VS0 . using the current modulation index. the division into sixtydegree segments with inner. In other words. the outer small triangle in Figure 4. in the case of this particular small triangle. Accordingly. (41) 83 .
the reference vector is synthesized from the small vectors only. the most beneficial is the inner triangle region because in the inner triangle region. The NP current is. (45) 84 . (44) Finally. two small vectormodulation indices.while the duty cycle of the negative switching combination (poo) of the same vector is d S 0 _ neg = ( 1 − mS 0 ) ⋅ d S 0 / 2 . in this case: i NP = d M ⋅ i b + (mS 0 ⋅ d S 0 ⋅ i a + mS 1 ⋅ d S 1 ⋅ i c ). (42) Then in the case of this particular outer triangle the current flowing into the NP is i NP = d M ⋅ ib + mS 0 ⋅ d S 0 ⋅ ia . there are two small vectors available to aid in the NP balancing. the NP current can be expressed in the other small triangles as well. If the reference vector is within the middle small triangle. and is therefore completely controllable: i NP = mS 0 ⋅ d S 0 ⋅ ia + mS 1 ⋅ d S 1 ⋅ i c . as far as maintaining charge in the NP. and accordingly. (43) Using the same approach.
these functions have discontinuous first derivatives at the boundaries of the small triangles and their wave shape dependent on the modulation index.3 Analysis and Control of the Neutral Point Voltage Error A steady state lowfrequency ripple in the NP current is caused by a periodic variation of the components in (43). dS1 . and d M ⋅ i a for 60 o ≤ θ < 120 o .4. 85 . as shown in Figure 4. and for 60 o ≤ θ < 120 o . In the steady state. they map to different switchingstate vectors in different 60degree sectors. d M is the duty cycle of the medium vector opn. Although the duty cycle functions in Figure 4. the voltage reference vector has a constant amplitude and rotates with constant angular speed. dM. m. (44) and (45) over the output voltage line cycle.5. it can be seen from Figure 4. Figure 4.5. d M is the duty cycle of the medium vector pon. the duty cycles dS0 . Due to the selected SVM strategy.5 are continuous functions. ω = dθ / dt . the contribution of the medium vector to the NP current is d M ⋅ i b for 0 o ≤θ < 60 o .8. For example. Duty cycles of SVM for modulation index m=0. As the reference vector sweeps through successive 60degree sectors.2 that for 0 o ≤θ < 60 o . Hence. and dL become periodic functions of time.
6). the NP current can be written in the matrix form. the duty cycle of the medium vector (also shown in Figure 4. as in (46). Ma Mb Mc angle [deg] angle [deg] angle [deg] Figure 4. In this case. Similar to the method used to select the switching functions of medium vectors (drawn in Figure 4. It is important to notice that the SVM algorithm completely determines the switching functions. there are timevarying load currents (with their phase angle with respect to the voltage reference vector) that also affect the current flowing into the NP as a consequence of switching the medium vector switching states. Switching functions of the medium vectors. for one specific modulation index). In other words the switching functions depend solely on the choice of modulation algorithm. On the contrary. it is possible to generalize the contribution of the medium vectors to the current in the NP for the full line cycle. is determined by the SVM algorithm as well as by the voltage reference vector’s angle and magnitude.Using the switching functions (such as those in Figure 4.5. And finally.6 and represented in tabular form in 86 . (46) i NP _ mediu_ vector = d M ⋅ [M a Mb i a M c ] ⋅ i b i c .6.
it is perhaps a good idea to transform Equation (48) into the rotating dq reference frame. Expression (48). Using the small vector switching functions in Table 4III and Table 4IV. b i ic cos( θ + 2 ⋅ π / 3 ) sin( θ + 2 ⋅ π / 3 ) q (49) 87 .Table 4II. Substituting the phase currents with the right hand side of the expression makes the transformation: cos( θ ) sin( θ ) i a i = cos( θ − 2 ⋅ π / 3 ) sin( θ − 2 ⋅ π / 3 ) ⋅ i d . in order to facilitate the analysis and control of the NP current for different loading conditions. combining Equations (46) and (47) results in a single expression for the NP current over the full line cycle: (48) i NP = d M ⋅ [M a Mb i a i a i + [m m ] ⋅ d S 0 0 ⋅ S 0a S 0b S 0c ⋅ i M c ]⋅ b S0 S1 0 d S 1 S 1 S1 b S1 a b c i c i c . although valid over the full line cycle. 0 d S 1 S 1a S 1b S 1c i c (47) i NP _ small _ vector = [mS 0 Finally. does not offer significant insight into the properties of the modulator with respect to the current flowing into the NP. mapping functions can also be derived for the small vectors. Therefore. a NP current from the small vectors is expressed as i a d S 0 0 S 0 a S 0b S 0c mS1 ] ⋅ ⋅ ⋅ i b .
After multiplying the currentmodulation indexes. the matrices of switching functions and the transformation matrix results in a compact expression. m S 1 ⋅ d S 1 ⋅ S 1q i q (410) i NP = d M ⋅ M d [ ] 88 . as follows: m S 0 ⋅ d S 0 ⋅ S 0 d i d dM ⋅ M q ⋅ + m S 1 ⋅ d S 1 ⋅ S 1d i q m S 0 ⋅ d S 0 ⋅ S 0 q i d ⋅ .
Angle θ [deg] 060 60120 120180 180240 240300 300360 Medium Vector pon opn npo noip onp Pno Switching Function Ma Mb Mc 0 1 0 1 0 0 0 0 1 0 1 0 1 0 0 0 0 1 iNP ib ia ic ib ia ic Table 4III Currentswitching function for S0 small vectors. Angle θ [deg] 0120 120240 240360 S1 Small Vector ppo/oon opp/noo pop/ono Current Switching S1a S1b S1c 0 0 1 1 0 0 0 1 0 iNP ic ia ib 89 . Angle θ [deg] 30060 60180 180300 S0 Small Vector onn/poo non/opo oop/nno Current Switching S0a S0b S0c 1 0 0 0 1 0 0 0 1 iNP ia ib ic Table 4IV Currentswitching function for S1 small vectors.Table 4II Currentswitching function for medium vectors.
Weighing factors for medium vectors with m=0. load currents. (44) and (45) into one matrix equation that is valid over the full line cycle of the output voltage. NP current resulting from application of small vectors. and quadrature weighing factors multiplying the direct. id . It is apparent that the quadrature component of the current will be weighed much more heavily.This expression is a little more readable because in steady state. respectively. and quadrature. These four weighing factors depend not only on the small vectors’ duty cycles and the current switching functions that are determined by the type of SVM used.7.8. The NP current in this formulation still contains noncontrollable current produced by the application of the medium vector and the controllable current produced by the small vectors. weighing factors. id and iq are constants and represent active and reactive components of the load current. iq . Note that (410) is essentially a composite expression that combines (43). Two distinct sets of weighing factors are given in Figure 4. and quadrature. NP current resulting from the application of medium vectors can be found by multiplying the direct id and quadrature iq current by the direct.8 shows the wave shape of the small vectors weighing factors for modulation index m=0. but also on the control inputs mS0 and mS1 .7 for the case in which the modulation index m=0.8. and will produce much larger lowfrequency (LF) ripple than does the direct component of current.8. Similarly. depends on direct. as defined earlier. The wave shape of factors is given in Figure 4. Figure 4. Figure 4.8. d M ⋅ M d . d M ⋅ M q . respectively. 90 .
Between these two extreme cases. The weighing factors for medium vectors are periodic functions with zero average value over a line cycle. is given by the solid line. is given by the dashed line. Note that the ratio of active and reactive weighing factor is opposite for medium and small vectors.8. On the other hand.e. the NP current from medium vectors will naturally balance over a line cycle. mS0 =mS1 =1). mS0 =mS1 =1).e. large iq and small id means large disturbance from middle vectors. and small disturbance from middle vectors. at which id and iq currents are constant. the other set when only negative small vectors are used (i. Weighing factors for small vectors with m=0. Figure 4. 91 .8.One set of weighing factors used when only positive small vectors are involved (i. and the only concern is the size of the LF ripple. This means that in steady state. Large id and small iq means large control authority on the NP current through the manipulation of currentmodulation indices of small vectors. the weighing factors can be controlled by adjusting currentmodulation indices..
this requirement will not be met. The benefits of these schemes are that they do not have the ripple at half the switching frequency. This method can work only in the case of a perfectly balanced load and a perfectly balanced PWM scheme. in which the positive and negative small vectors are selected alternately in each new switching cycle. It is also true for threelevel rectifiers operating at or close to unity power factor. when the currents are generally small anyway). Still. The downside of this method is that the current ripple has a strong harmonic component at one half of the switching frequency. These results confirm the fact already proven in practice that accomplishing the NP balance is not very difficult for loads with high power factor. the small vectors’ switching combination that will move the NP voltage in the direction opposite from the direction of imbalance can be selected. and often require measurement of the amplitudes of the phase currents as well. In general. This requirement is usually met for acdrive applications (except at noload conditions. three distinctive approaches to the control of NP might be: • Passive control. meaning that this application will require different design tradeoffs. these schemes require measurement of the voltage imbalance in the NP. This means that it will be much easier to suppress the LF ripple from direct current. In addition. this control method can be used to establish a benchmark for NP controller performance. Based on that information. This method requires knowledge of the current direction in each phase. which is unlikely to happen in practice. • Hysteresistype control is perhaps the simplest and most popular closedloop NP control scheme. this method would have difficulties to recovering from line or load transients. One implementation of this method is described in Figure 3.with small control authority from small vectors. with the restriction to use of the NTV method. than it is to suppress them from quadrature current. However.14. and some variations of 92 . • Active control schemes control the current modulation indices mS0 and mS1 . for static VAR compensation. Regarding NP balancing control for SVM. This benchmark can then be used to evaluate the performance of other NP control methods.
mainly thirdharmonic ripple current flows into the NP. If the operating point exits the shaded region. it is clear that regardless of control scheme the control authority over the NP current is limited. Note that the graph is symmetrical. but that the region in which exact balancing can be achieved in each switching cycle must exist. This is a highly undesirable effect because for some converter designs it significantly increases the size of the dclink capacitors.96. The first graph shows the normalized NP charge ripple for the passive control of NP voltage balance.these control schemes can balance the NP exactly.9. The shaded region represents the same ripple free area as the one in Figure 4. 4. and that the unity power factor load represents the most favorable case. This region is given as a shaded area in Figure 2.9.4 Limitations of the NeutralPoint Control and Some Design Recommendations From the analysis in previous sections. and may be less robust than the hysteresistype control schemes. the selection of the balancing algorithm (which affects the highfrequency content of the NP current) is not important in determining the size of the dclink capacitors. Because the resulting NP ripple is lowfrequency. 93 . Figure 4. in which the NP voltage can be balanced in every switching cycle for a modulation index as high as m=0. The second graph shows the best that can be done using feedback control of NP voltage balance. lowfrequency.10 shows the peaktopeak value of the lowfrequency charge ripple in the NP divided by the amplitude of output phase current. and will not be investigated further. Unfortunately. they all increase the switching losses (due to the introduction of additional switching states).
modulation index m power factor angle (deg) Figure 4. φ LOAD ) .10. Q NORM = Q NORM ( m .9. Region where LF ripple can be suppressed. 94 . 2 ⋅ ∆U DC _ max (411) ci = This should provide sufficient guidelines to size the dclink capacitors for any expected operation mode and desired NP voltage ripple. by the current amplitude. then dividing that by two times the desired size of the NP voltage ripple: QNORM ⋅ I max . The actual dclink capacitor size for the specified NP voltage ripple and for any power factor angle and voltage modulation index m. from Figure 4. can be computed by multiplying the normalized charge ripple.
Normalized amplitude of the LF charge ripple.10.11. It is clear that the greatest savings in the size of capacitor can be achieved when the inverter is predominantly supplying active power. consider the prototype SMES PCS with 1800V dclink voltage and Imax=200A phase current amplitude. the comparison of capacitor sizes for the feedback NP control and passive NP control is summarized in Figure 4.Passive NP control Feedback NP Control Normalized NP charge mo di nd ex Normalized NP charge gle r an cto er fa pow mo di nd ex ngle or a ct er fa pow Figure 4. 95 . the benefits of feedback NP control diminish.9. while for the operation with purely reactive power. For example. and allow 1% voltage ripple (∆Udc_max=18 V) in the NP. For modulation index m=0.
85. In short. 96 . which does not hold the promise of introducing new. 45 and 12 degrees.µF 14000 12000 10000 8000 6000 4000 2000 0 1 0. and unexpected results 1 . Capacitor sizes for specified NP ripple. with and without NP control. this is a rather difficult requirement. Instead.11. and synchronization of that current source with the voltage reference vector.9 cos(φ) Figure 4.12 shows the experimental setup used to verify the theoretical results for the inductive loads of three different power factors. 1 Operating a converter in a rectifier mode with the current loop closed and with a high voltage ripple in the NP (due to the low power factor) seemed like a risky experiment.8 0 m=0. Figure 4. namely. 4.5 Experimental Results A full experimental verification of the results from this chapter requires the converter to operate in all four quadrants and with all possible power factors. In addition. to investigate the situation with the converter as a load (rectifier mode of operation) would require a highpower threephase current source.
13) show excellent agreement with the theoretical result from this Chapter.6mH inductors for 12 degrees inductive load.8 mH. Three . inductors in series with three .12. Three. when the load with the power factor angle of 85 degrees is connected to the ac terminals.13 show the ripple in the NP voltage normalized with the amplitude of the phase current.p CDC o CDC iNP ThreeLevel Converter Switching Network ia ib ic La Lb Lc Ra Rb Rc n VNP Digital Modulator Figure 4. the collection of experimental results for different operating points (Figure 4. The dclink voltage was set to 80V. The experimental results in Figure 4.72Ω resistors in series with the same inductors for the 45 degrees reactive power factor and finally three1Ω resistors in series with three . 1. 97 .06Ω resistors were used as 85 degrees reactive power factor load. Experimental verification of NP voltage ripple properties. There were three different load conditions. Finally.
7 1 0 1 x 0 10 m=. 98 .5 0 1 mo di nd ex le ang tor ac er f pow mo 0.01 0.005 0.02 0.005 0.01 0.01 0.025 3 0.005 0.025 0.025 0.015 0.01 0.015 0.005 0.14.9 1 0 1 x 0 10 m=1. Normalized NP voltage for power factor angle of 85 degrees.4 1 0 1 3 NP Voltage Ripple cos(fi)=.025 0.6 di nd ex 0.8 1 0 1 x 0 10 m=.025 0.087 0 10 x m=.02 0.3 x 10 m=.005 0.02 0.01 0.005 0.0 1 0 1 0 3 3 3 3 0.025 Figure 4.5 Experimental Results Normalized NP charge Normalized NP charge 1 0.01 t [s] 0.015 0.8 100 0.02 0.6 1 0 1 x 0 10 m=.015 0. Theoretical Results x 103 1.015 0.02 0.015 0.01 0.5 1 0 1 x 0 10 m=.025 0.13.4 gle r an acto er f pow 0 100 Ripple free area Figure 4. Theoretical and experimental normalized NP voltage ripple.02 0.02 0.015 0.005 0.
a quantitative study of NP balancing offered a normalized size of the dclink voltage ripple for all possible operating conditions.4. it is clear that the NP can be balanced in a threelevel converter for all possible operating conditions. The study identified the region in which the NP can be balanced on a switching cycle level. the question of tradeoff between the size of the ripple and the size of the dclink capacitors remains. The model is general and can be used to describe any SVM and its effect on the NP. introduced in the next chapter. Using the modeling approach developed in the beginning of the chapter. and offered an intuitive explanation of the origin of the ripple. Furthermore.6 Conclusions This chapter introduces a novel approach to modeling the threelevel space vector modulator for the purpose of investigating the characteristics of NP balancing. this tradeoff can be extended to include the trade off between the voltage capacity of the switches and the size of the dclink capacitors. from this study of NP balancing (and from available literature). Finally. which is essential information for the design of dclink capacitors. properties of NP control were studied for one particular SVM. However. through the use of the feedforward algorithm. 99 . In addition.
The first requirement is that the switches must be able to handle the extra voltage stress. the lowfrequency current ripple often does not determine the size of the dclink capacitors. However. if the goal is to design an optimized converter. This is the situation for which the results from the previous chapter become important. one that does not have the need for excessive energy storage. it becomes clear that the load with the low power factor creates a ripple current in the NP of the threelevel converter.5 FEEDFORWARD CONTROL OF THREELEVEL NEUTRALPOINT CLAMPED CONVERTERS From the analysis in the previous chapter. Because the majority of the threelevel converters used for the drive applications need a significant energy storage in order to compensate for the possible disturbances in the utility line voltage. then the amplitude of the ripple current flowing into the NP becomes a primary factor in determining the size of the dclink capacitors. The answer to that question is positive. if two basic requirements are satisfied. as well as that shown in the many cited references. and the second requirement is that an effective feedforward algorithm must be constructed in order to 100 . Further thinking about capacitorsize optimization leads to the question of whether the size of the dclink capacitors can be further reduced if a certain amount of the NP ripple can be tolerated.
Furthermore. be applied to other converter topologies such as the flying capacitor topology or even the cascaded fullbridge converter.1 The Effect of the Capacitor Charge Imbalance on the Voltage Space Vectors The task of the modulator is to determine the switching sequence of the equivalent switches from Figure 5. this chapter will first propose and than make a detailed study of a feedforward algorithm that can effectively compensate for the ripple in the NP. Also (as was already mentioned) every switching state produces welldefined threephase line voltages.3. For these reasons. and are therefore available almost by default.1 that will result (in the average sense) in the synthesis of the voltage reference vector commanded by the system controllers. which are needed for the NP controller as well. 5. 101 . with few modifications. then allowing a certain amount of voltage ripple can be a good way to save on the capacitor size.compensate for the distortion in the output ac voltage waveforms that occur due to the ripple in the NP voltage. It is convenient to use the definition of the error in the NP voltage given in Figure 5.2). Therefore if an effective stable algorithm can be found. In addition. which can be represented as twodimensional voltage space vectors (as shown in Figure 5. The goal at this point is to examine how the error in the NP affects the voltage space vectors. such an approach can. the only hardware requirement for the implementation of this algorithm is the dclink sensors.
Functional diagram of a threelevel converter. 2 a 2 0 1 b 0 2 1 0 β 2 120 220 a b c c Figure 5. . . 102 .1. . 020 1 Vp i NP 1 NP Vn 0 . . . . . 021 121 010 122 011 1 221 110 211 1 100 101 212 210 M REF 200 α 2 201 → 022 0 112 001 012 002 102 202 Figure 5. .. . . . .2.2 . Voltage space vectors of a threelevel converter.
these vectors remain unaffected by the voltage error in the NP. it is clear that one of the line voltages remains unaffected. Because the large voltage space vectors do not connect any of the phases to the NP. which explains why the tip of the medium vectors slides along the outer border of the hexagon. 103 .3. if the NP voltage error. is smaller than 0.4. voltage error in the NP does affect both medium and small voltage vectors.2 Vdc V Cdc + p=Vdc/2+∆V iNP 1 Cdc + Vn=Vdc/2∆V 0 0 Figure 5. the medium vectors shift like as shown in Figure 5. ∆V. For example. Table 5I shows how the linetoline voltages of each medium vector are affected. The definitions of the dclink voltages and the voltage error in the NP. From the same figure. On the other hand.
and only the magnitude changes while the angle remains unaffected. Table 5II shows a summary of the small vectors’ line voltages. each small vector switching combination has the same magnitude and phase. the effective “splitting” of the small vectors (shown in Figure 5. On the other hand. each of the switching combinations of a small vector turns into a different small vector. 210 Vab Vbc Vca Vdc/2∆V Vdc/2+∆V Vdc 120 Vdc/2+∆V Vdc Vdc/2∆V 021 Vdc Vdc/2∆V Vdc/2+∆V 012 Vdc/2∆V Vdc/2∆V Vdc 102 Vdc/2+∆V Vdc Vdc/2∆V 201 Vdc Vdc/2∆V Vdc/2+∆V Small vectors are affected by the error in NP in a somewhat different way. Table 5I Effect of the error in NP voltage on the medium voltage vectors. 104 . Again. While there is no error in the NP voltage. Therefore. Shift of the medium voltage vectors due to the ∆V>0 error in NP voltage.Vbc 120 2 ∆ V 3 021 210 ∆V 012 201 102 Vca Figure 5.5) occurs. when there is an error in the NP voltage.4.
a 105 . However. Table 5II Effect of the error in NP voltage on the small voltage vectors. “Splitting” of the small voltage vectors due to the ∆V>0 error in NP voltage. and that it should be possible to compensate for the effect of the NP voltage ripple without sacrificing the converter voltage gain.Vbc 110 121 011 122 010 112 212 001 2 ∆V 3 221 211 100 201 V ab Vca Figure 5. This should indicate the feasibility of finding a correction algorithm. 211 Vab Vbc Vca Vdc/2∆V 0 221 0 Vdc/2∆V 121 122 112 0 212 Vdc/2∆V Vdc/2+∆V Vdc/2+∆V Vdc/2∆V 0 010 Vdc/2∆V Vdc/2+∆V 0 0 Vdc/2∆V 011 Vdc/2∆V 0 Vdc/2+∆V 101 Vdc/2+∆V Vdc/2+∆V Vdc/2∆V 001 0 Vdc/2∆V Vdc/2+∆V 0 101 Vdc/2+∆V Vdc/2∆V 0 Vdc/2+∆V Vdc/2+∆V 100 110 0 Vdc/2+∆V Vdc/2∆V Vab Vbc Vca Vdc/2+∆V 0 Vdc/2∆V It is important to note that because the outer boundaries of the hexagon remain unaffected by the error in the NP voltage.5. the maximum amplitude of the reference vector that can be synthesized remains unaffected by the size of the NP voltage error.
sine triangle modulation was the only option when implementing the modulator with discrete components. This is only one of the issues discussed in thiscChapter.2 Sine Triangle Modulation for ThreeLevel Converters Although SVM is perhaps the most general and unrestricted approach to the generation of PWM signals. the distinction between the two implementations (in terms of versatility and freedom to optimize the PWM signals) diminishes.significant change in the small vectors’ amplitude will change their corresponding duty cycles and will ultimately affect their control authority over the NP current. and although the sine triangle modulation is basically a subset of SVM 1 .2. 5. The main variations are in the exact shape of the carrier waveform and/or the modulating signal that can be augmented with the third harmonic for increased voltage gain. it might be the subset that is significantly simpler to model and easier to understand. a multilevel modulator can be constructed that will have n1 carrier waveforms. Figure 5. However when implemented with the processor. 106 .1 PWM Generation with Sine Triangle Modulation All sine triangle modulations are the same in the sense that that the intersection between the carrier triangular waveform and the modulation signal determines the switching instances. where n is the number of levels.6 shows a basic threelevel sine triangle modulator. Then. Then. That is the principle reason why in this chapter it is the sine triangle modulator that will be predominantly studied. 5. As an example. using the same basic principle. the conclusions can also be used to explain the behavior of the space vector modulator. 1 Historically. with few restrictions. and so on.
Therefore.vx_mod 2 1 0 vx0 Vdc θ=ωt θ=ωt Figure 5. which then (depending on the direction of the power flow) becomes charged or discharged more than the lower dclink capacitor. 5. This is really the method used so far throughout this dissertation.2 NP Control With Sine Triangle Modulation Adding the zero sequence to the modulating sinusoidal waveform controls the charge balance in the NP. depending on the 107 .6. Principle of sine triangle modulation for a threelevel converter.7 shows how a constant zero sequence shifts the power transfer to the upper of the split dclink capacitors.2. because manipulating the switching combinations of small vectors actually manipulates zero sequence. v a _ mod = A ⋅ sin( ω ⋅ t + ϕ ) + δ + 1 (51) v b _ mod = A ⋅ sin( ω ⋅ t + ϕ − 2 ⋅ π / 3 ) + δ + 1 v c _ mod = A ⋅ sin( ω ⋅ t + ϕ + 2 ⋅ π / 3 ) + δ + 1 Figure 5.
from the modeling point of view. however. In that case.8 shows both cases: one in which the modulating signal 108 . if the modulating sinusoid is too large. vx_mod 2 1 δ 0 vx0 θ θ Figure 5. The NP controller needs to supply only one control variable. Despite these limitations. This greatly facilitates writing equations and developing the model. sine triangle modulation has one big advantage. Figure 5. For example. which is a limitation that applies to the SVM as well. 5. this type of control has its limitations. Principle of NP control for the threelevel converter. but only on the direction of power flow. there might be little or no room to add the zero sequence.3 The Average Model of a Sine Triangle Modulator By magnifying the time scale and looking into the process of computing the duty cycles.2.direction of the power flow. there are really two cases to consider. the size of the ripple current will depend solely on the load impedance and load power factor.7. It is important to note. a specific amount of zero sequence can be added to correct for the disturbance. that similar to the space vector modulator. and does not really require information on the direction of phase current.
(55) 109 . The duty cycle for the NP is then. the modulator will connect the phase output to the positive rail for time d p ⋅ T . Because it is advantageous for the utilization of multiple dc voltage levels and because it reduces the switching frequency ripple. where d p = vx − 1 x ∈ { a .c } (52) and v x is the modulating signal for phase x. d0 = 1 − d p = 2 − vx .8. it almost goes without saying that all the switching should occur between the two closest levels. from the lefthand side of the illustration in Figure 5.b . and the other in which it is compared to the lower modulating waveform.is compared with the upper carrier waveform.c } .b . x ∈ { a . (53) When the modulating signal is compared with the bottom carrier waveform the duty cycles for NP and for the negative dc rail are d 0 = vx (54) and d n = 1 − vx . Accordingly.
is the modeling approach published by Newton and Sumner [B14]. To find the line cycle average of the current flowing through the NP.9. 1 + A ⋅ sin( θ ) π ≤ θ < 2 ⋅ π (56) Hence the average current flowing in the NP during one line cycle is 110 . the duty cycle of the NP can be expressed as follows: 1 − A ⋅ sin( θ ) 0 ≤ θ < π d1 = . T 2 vx 1 1 d0. the phase flows through the NP for the duration of the NP duty cycle d1 . which results in the expression that describes the relationship between the zero sequence command.8. From (52) and (53). it is necessary to first study the duty cycles of one phase leg. and after substituting the modulating v a_mod from (51). The modeling approach averaged the NP current during the line cycle. the amplitude of the modulating signal. the load power factor and the NP current that is to be controlled. Perhaps the most intuitive (and although approximate. This chapter will first briefly repeat Newton and Sumner’s methodology and main resultsthen extend those results in order to model the sine triangle modulator with feedforward control. Clearly.2 d p. That expression can then be linearized around an operating point and that model can be used for control design. Duty cycles for the sine triangle threelevel modulation. quite accurate). T vx 0 T 0 T Figure 5. as shown in Figure 5. T d0.T d n.
All it says is that under ideal conditions. vx_mod 2 1 0 dx2 θ dx1 θ dx0 θ θ Figure 5. regardless of the power factor angle. mainly thirdharmonic ripple that is present in the NP. 111 . which quite expectedly.i NP = (57) 1 2 ⋅π π ⋅ ∫ I ⋅ sin( θ + ϕ )( 1 − A ⋅ sin( θ )) ⋅ dθ + 0 + ∫ I ⋅ sin( θ + ϕ )( 1 + A ⋅ sin( θ )) ⋅ dθ π 2 ⋅π . It might be clear (but it is still important to point out) that this analysis does not study the lowfrequency. Duty cycle of one phase equivalent switch for sine triangle modulation. the line cycle average of the NP current will be zero. turns out to be zero ( i NP = 0 ) for any power factor angle ϕ.9.
In that case.However. (58) which sets the first integration limit at δ α = − sin −1 . the remaining two integration limits are simply β = π − α and γ = 2 ⋅ π + α .10. the feedback implementation is required to insure that the capacitor charge can balance during the transients or in the presence of disturbances. (510) The line cycyle average NP current can be computed from i NP (511) β 3 = ⋅ I ⋅ sin( θ + ϕ )( 1 − ( A ⋅ sin( θ ) + δ )) ⋅ dθ + 2 ⋅π ∫ α + ∫ I ⋅ sin( θ + ϕ )( 1 + ( A ⋅ sin( θ ) + δ )) ⋅ dθ β γ . A ⋅π A (512) iNP = 112 . From Figure 5. The integration is a little more involved in that case because the integration limits that correspond to the zero sequence δ need to bedetermined. it is important to analyze the average current in the NP for different phase angles in order to determine the control authority of the zero sequence over the NP. it is clear that the first discontinuity in the duty cycles occurs for 1 + sin( α ) + δ = 1 . even though in ideal conditions the NP current is zero. A ( 59) From there. resulting in the following expression for the line cycle average of the NP current: − 3⋅ I δ ⋅ cos( φ )δ ⋅ A 2 − δ 2 + A 2 ⋅ sin − 1 .
valid only for A < 1+δ < 2 .10.3 Modeling the FeedForward Control If the modulator computes the duty cycles and selects the vectors under the assumption that the charge in the dclink capacitors is balanced. (513) vx_mod 2 1 0 δ d2 α 0 β γ θ d1 0 θ d0 0 θ 0 θ Figure 5. Duty cycles of one phase with the zero sequence present. then the modulator would make an error 113 . 5.
then that is one way to make significant savings on the size of the dclink capacitors. perhaps because it is assumed that the dclink capacitors will be able to sufficiently minimize the voltage ripple in the NP. if a certain amount of ripple in the NP can be tolerated. it turns out that when the modulating signal is larger than Kn . In that case. In that case. the phase duty cycles are 114 . as already mentioned.3. Vdc / 2 where Kp and Kn are the amplitudes of the top and bottom carrier waveforms. (515) Kp Kn = Vp Vn and K p + K n = 2 also holds true.1 The FeedForward Algorithm The idea behind the feedforward algorithm is to modify the amplitude of the triangular carrier waveforms so that the ratio of upper and lower carrier waveform corresponds to the ratio of upper and lower dclink capacitor voltage. a feedforward control algorithm that allows the modulator to take into account the ripple in the NP and still produce highquality sinusoidal output waveforms becomes significant.11 (and with the help of a little trigonometry). However. One way to satisfy this requirement is to control the amplitude of the triangular modulating waveform to make sure that the following condition is always satisfied: (514) Kp = Vp Vdc / 2 and K n = Vn . Most of the modulation algorithm does not take this possibility into account. From Figure 5. respectively. 5.whenever the capacitor charge was imbalanced.
In order to examine the effects of crossing the boundary.T Kn 0 T 0 Kn T Figure 5. Duty cycles for the sine triangle threelevel modulation.T Kp Kp d0. the duty cycles are vx K − vx and d 0 = n .11. where 115 .(516) d2 = vx − K n 2 − vx and d 1 = . Kp Kp and when the modulating signal is lower than Kn . it might not be at all obvious whether there will be a discontinuity when the modulating signal crosses the “moving” boundary between the upper and lower carrier signals. Kn Kn (517) d1 = 2 d p.3. it is probably easiest to write equations for the average output voltage for the case in which the modulating signal v x(t)>Kn . 5. T 2 vx d0.T vx dn.2 Mathematical Basis of the FeedForward Algorithm Even though the development of the feedforward algorithm is quite intuitive. and when the v x(t)<Kn .
Kn 2 (523) v x 0 ( t ) = −d n ( θ ) ⋅ Vn = which is exactly the same expression as (522). (519) The output voltage for the case when v x_mod (θ)>Kn is v x 0 = d p (θ ) ⋅ V p . after substituting the relation from (514). (520) After substituting equation (516). 2 Using the same steps. the final form of the output voltgae is (522) v x 0 = ( v x (θ ) − K n ) ⋅ Vdc . there are no discontinuities at the boundary of the lower and upper carrier waveform for the feedforward algorithm (514). the output v oltage of the modulator. and where v x _ mod ∈ [ 0 .2 ] A∈ [ 0 . for the case when the v x(θ)<Kn . Therefore. This ends the proof.(518) v x _ mod = A ⋅ sin( θ ) + K n + δ . 116 . A < K n + δ < 2 − A . the output voltage can be written as v x (θ ) − K n ⋅Vp . turns out to be K n − v x (θ ) V ⋅ Vn = ( v x ( θ ) − K n ) ⋅ dc .1 ] . Kp (521) vx0 = and finally.
12.3 Average Model of the ZeroSequence NP Control In terms of NP the switching network in Figure 5.5. The model of the modulator with respect to the NP current.3.12. as the experimental results in the next chapter will prove. As shown in Figure 5.12. Still. because of the filtering properties of the dclink capacitors most of the unmodeled dynamics from the load and the output filters will appear only as a nonlinear gain in the NP controller loop. What complicates this otherwise simple modeling approach is that there are multiple dependencies of the source that are influenced not only by the phase angle of the load and the feedforward terms. 117 . but also by the zero sequence of the modulating signal. which again depends on the load currents and the NP duty cycle of each phase.1 (together with any source or the load that might be connected to its ac side) is just a current source like the one in Figure 5. the voltage in the NP depends on the equivalent NP current. ip C/2 + VDC + vp + C/2 in vn i0 Figure 5. That is the principle argument for the application of the linecycle averaging method in the study of NP voltage dynamics.
β = π − α and γ = 2 ⋅ π + α . In order to perform linecycle averaging. which results in the expression − 3 ⋅ I ⋅ cos( φ ) ⋅ 2 ⋅ ( 4 ⋅ K n + 2 ⋅ δ − 4 ) ⋅ A 2 − δ 2 − A2 ⋅ π ⋅ ( K n − K p ) + 4 ⋅π ⋅ A ⋅ K n ⋅ K p δ + 2 ⋅ A ⋅ ( K p + K n ) ⋅ sin ( )] A 2 −1 i NP = (528) [ . (525) From this equation and Figure 5.(524) C dv p dt = d a 0 ⋅ i a + d b 0 ⋅ ib + d c 0 ⋅ i c . and similar to the equation (58). A (526) α = − sin −1 Substituting the modulating signal. the boundary between the piecewise linear NP duty cycles needs to be determined. defined in Figure 5. it becomes clear that the integration boundaries are in fact the same as those in (511). into the expressions for NP duty cycles. δ . As shown in Figure 5. 118 .13 as v x = K n + δ + A ⋅ sin( θ ) . the first discontinuity in the duty cycle comes from the following condition: K n + δ + A ⋅ sin( α ) = K n .13.13. d0 in (516) and (517). the linecycle average of the NP current can be found by integrating i NP (527) β 2 − ( K n + δ + A sin( θ )) ⋅ dθ + = 3 ⋅ ∫ I sin( θ + φ ) ⋅ Kp α K + δ + A sin( θ ) ⋅ dθ + ∫ I sin( θ + φ ) ⋅ n Kn β γ .
13.which is valid for A < Kn +δ < 2 − A. (529) vx_mod 2 Kp Kn 0 δ d2 α 0 β γ θ d1 0 θ d0 0 θ 0 θ Figure 5. 119 . The duty cycles of the sine triangle modulator with zero sequence.
if the dclink capacitor charges are balanced ( P= K Kn =1).K p = K n = 1 ) = − cos( φ ) .3. Kp =Kn =1. 2 8⋅δ + ( K n + K p ) ⋅ A2 − δ 2 − ⋅ Kn ] A A ⋅ A2 − δ 2 The first derivative of the linecycle average in the operating point δ = 0 . ∂δ π 120 . Equation (528) simplifies to the following expression for the average NP current with no feedforward terms: (530) iNP = − 3⋅ I δ cos( φ )δ ⋅ A 2 − δ 2 + A 2 ⋅ sin − 1 π⋅ A A 5. is (533) ∂i NP 6 ⋅I ( δ = 0. which is not covered in this analysis. and then by neglecting the higherorder terms as follows: ∂i NP ∂ 2 i NP (0 )⋅δ + ( 0 ) ⋅δ 2 + K 2 ∂δ ∂δ (531) i NP ( δ ) = i NP ( 0 ) + The first derivative of the average NP current (with respect to δ ) is ∂ i NP 2 ⋅δ ⋅ ( 2 ⋅ Kn + 2 ⋅δ ) 3 ⋅ I ⋅ cos(φ ) 2⋅ A =− ⋅ [( K n + K p ) ⋅ − ( Kn + K p ) ⋅ + 2 2 ∂δ 4 ⋅ π ⋅ Kn ⋅ K p A −δ A A2 − δ 2 (532) . Otherwise. the modulator works in overmodulation mode. K p = K n = 1 .4 A SmallSignal Model of the ZeroSequence NP Control The average model of the zerosequence control in the NP can be derived by developing the expression for the average current (528) into Taylor’s series around the operating point δ = 0 .Outside the region defined in (529).
which assumes the balanced charge in the dclink capacitors. by doing so. For example. The feed forward algorithm has also been proven to not introduce any discontinuities in the output waveform and has shown that it can effectively compensate for any size of imbalance. through linearization around the operating point. (535) i NP = − 6 ⋅ I ⋅ cos( φ ) ⋅ δ . at the operating point. π Therefore. the smallsignal dynamic model of the NP control has been derived as the function of the load power factor. it effectively introduces a zerosequence component which influences the NP control. the feedforward control does change the ratio between the two triangular carrier signals and. the smallsignal average current model is the same as the smallsignal model for the modulator without the feedforward control [B14]. At the same time. the zerosequence balancing component and the feedforward term. Finally. In addition. the size of the modulating signal does not allow the freedom to control the zero sequence for the purpose of NP control. 121 .14 shows the situation in which the feedforward controller compensates for the significantly higher voltage of the upper dclink capacitor. around the operating point.while the average value of the linecycle average of the NP current is zero around the operating point (534) i NP ( δ = 0 . K p = K n = 1 ) = 0 . Figure 5. the line cycle average dynamic model of the sinetriangle modulator with and without the feedforward control has been derived. However.5 The Stability Analysis of the Feed Forward Algorithm So far. 5. and when the zerosequence voltage equals zero.3.
14. Figure 5. with the upper dclink capacitor at much higher voltage than the bottom and with a high modulation index. The PWM signal with strong feedforward component. in an exaggerated situation with the bottom dclink capacitor voltage equal to zero. The feedforward influence will always tend to discharge the capacitor with the higher state of charge. the output is switched more often between the positive rail and the NP than between the NP and the negative rail.15 shows the power flow in the inverter mode of operation. as in Figure 5. Conversely. In addition. the top capacitor will be either charged or discharged. in the rectifier mode. all the power would flow between the upper capacitor and the load.14. and in that way it exercises a stabilizing influence.As clearly illustrated in the situation shown in Figure 5. 122 . and there will be no power flow between the load and the bottom capacitor. depending on the direction of the power flow. Evidently. therefore. Van θ θ Figure 5. the feedforward algorithm introduces a destabilizing influence by always further charging the capacitor with the higher voltage. For example.14. the output of the converter would never be connected to the bottom dc rail. with the power flowing from the ac to the dc side.
it would be necessary to solve (524) analytically. min[ K p ]] . in steady state. the NP voltage has mainly a thirdharmonic ripple component. even if the form of the solution is known in advance. 123 . for the rectifier operation a tradeoff exists between the influence of the feed forward algorithm. it is important to point out that this is only a qualitative analysis. which is very hard if not impossible.15. In the rectifier mode of operation however. the maximum amplitude of the modulating sequence and the maximum ripple that can be allowed in the NP. To exactly determine the wave shape of the voltage in the NP (in order to find the stability boundaries).2 CDC P 1 Sa NP ZL Sb Sc ZL ZL VDC CDC 0 Figure 5. In other words. the compensation of the feed forward by the zero maximum amplitude of the modulating signal to δ ∈ [ − min[ K n ]. sequence limits the Unfortunately. Therefore. but it also introduces a destabilizing effect into the NP control. Still. Power flow in the inverter mode of operation with large imbalance in NP. and no fixed dc offset. in the inverter mode of operation. This destabilizing effect can be compensated for with the addition of zerosequence voltage. the feedforward control still compensates for the disturbance in the line voltage caused by the dclink capacitor charge imbalance. feedforward control reduces the size of the NP voltage ripple by inherently providing a stabilizing influence. In reality.
Likewise.16. the 124 . The effect of NP unbalance on the boundaries of small triangles.5. there are now ten inner triangular regions within the sixtydegree region. because of the splitting of the small vectors.16. The modifications in SVM algorithm that can help compensate for the imbalance will now be discussed. the fast SVM algorithm can not be used since it relies on the integer values of the voltage space vectors’ coordinates. First consider how to synthesize the reference vector located in the sixtydegree region from Figure 5.4 Implementation of FeedForward Algorithm for Space Vector Modulation The effects of the shift in the NP on the location of the voltage space vectors were discussed earlier in this chapter. There are several issues that complicate the implementation of the SVM feedforward control: • The boundaries between the small triangles that change as the vector locations change (due to NP ripple). In fact. instead of only four. • Because the vectors change their locations. d’L d’ S1 dM d’S0 T9 T8 T10 dS0 T4 T7 T3 T2 T5 T6 dS1 dL T1 Figure 5.
17. in order to select the small triangle. However. and than all the equations are solved and then checked to be sure the boundary conditions are satisfied. an iterative approach might be required. But. In fact. This means that the triangular region must be guessed initially. the duty cycles of the small vectors depend on the boundary as well. This is hardly a practical approach. Therefore. because it uses the NP controller command from the previous switching cycle. in order to both maintain NP current at zero during each switching cycle and to compensate for the disturbance. Perhaps the most troublesome aspect is the identification of the current sector because of the splitting of the small vectors. This method is approximate.1] . The basic idea of the virtual small vector approach is that the NP controller supplies the ratio between the positive and negative small vectors’ switching combinations. for the sake of argument. It is possible to meet both these requirements in the operating region we have identified in the previous chapter. consider the approach that can both compensate for the disturbance in the NP and maintain the NP current at zero on the switchingcycle level. therefore determining the amplitude of the equivalent small vector amplitude (536) H = c ⋅ V S0 + ( 1 − c ) ⋅ V S 1 → → → where c ∈ [0 . 125 . it is probably more than acceptable to use the previous switching cycle value of the NP controller. One way to overcome this problem is based on the virtual small vector approach shown in Figure 5. but from the implementation point of view it might be the most practical implementation.other algorithms can not rely on the fixed locations of the vectors to simplify the duty cycle calculation. and are computed from a different set of equations for each small triangle. but the boundary is not known before the duty cycles of the small vectors are selected. due to the troublesome “splitting” of the small vectors. In addition. because the switching frequency is usually very high for the slow dynamics of dclink capacitors. its boundary needs to be known.
d S 1 = ( 1 − c ) ⋅ dH . (539) 126 . Returning to the approach using the virtual small vector. the small vectors’ duty cycles are then easy to find.17. The active sector computation based on the home vertex algorithm. as follows: d S 0 = c ⋅ dH . → → → → (538) Finally.→ VM → V REF → V S0 H V S1 → → → VL Figure 5. where d H + dL + d M = 1 . which allows the duty cycles to be computed. then all the small triangle boundaries are known. as follows: (537) V REF = d H ⋅ H + d L ⋅ V L + d M ⋅ V M . it is clear that when the location of the small vector is known (or better to say predetermined by the NP controller).
The only difference is that there are two home vertices to take into account in determining the boundary regions and in dutycycle computations. If vector V * is located in one of the inner triangle region the algorithm remains the same. 127 . for the case of the → outer triangular region.This concludes the implementation of the proposed feedforward algorithm. the qualitative analysis shows that the feedforward algorithm introduces an additional stabilizing effect into the NP control in the inverter mode of operation. The influence of the modulator on the NP controller is shown by the average models developed at two different time scales: first at the fast scale (the time scale of the switching frequency).5 Conclusions This chapter introduces and analyzes the feed forward algorithm as a method to successfully compensate for NP voltage ripple of any size and thus allow the tradeoff between the voltage capacity of the switching devices and the size of the dclink capacitors. the same analysis shows that the feedforward control introduces a destabilizing effect. which can essentially limit the voltage gain of the system. The influence of the new algorithm on the control of the NP is studied using the example of the sine triangle modulator. then those results were used to develop an average model on the slow scale (the time scale of the linecycle average). 5. In the rectifier mode of operation. In addition. The linecycle average model of the sine triangle modulator’s influence on the control of the NP is then linearized around the operating point and can be used to design on NP voltage controller.
As in the previous chapter. In addition. and proves that the influence of the feedforward control the rectifier mode causes instability only in extreme cases with unrealistically large voltage ripple and high power factor correction (PFC mode of operation). the smaller the ripple in the NP. the danger of instability is lessened by the fact that the higher the power factor. The full dynamic simulation confirms the qualitative stability analysis from the previous chapter.6 FEEDFORWARD CONTROL OF THREELEVEL NEUTRALPOINTCLAMPED CONVERTERS—SIMULATION AND EXPERIMENTAL RESULTS The goal in this chapter is to continue the analysis of the threelevel converter modulator with feedforward control using numeric simulation tools to gain some further insight. and then use the models to verify the system operation outside the region of the experimental results. The general approach is to develop models. therefore. in the inverter mode of operation. the feedforward control has a slight stabilizing effect. and under no circumstances does the loss of NP control occur. verify the models with the experimental results. Furthermore. 128 . the system is analyzed on two time scales by using the linecycle average and the switchingperiod average. the smaller the chance of instability.
2 shows the controller loop gain for different modulation indexes (output voltage and output current) of the converter. L=1. is correct and can be used for the closedloop NP controller design. as shown in Figure 6.8mH).6. it is perhaps best to verify this underlying assumption of the linecycle averaging method by measuring the loop gain of the NP voltage controller.72Ω. The experimental hardware is the SMES hardware with the NP controller modified. The converter operates in the inverter mode with 80V dclink voltage and with a 45degree power factor inductive load (R=. The modulator is implemented using the fast SVM algorithm.1 Analysis and Experimental Verification of the Line Cycle Average Model of the FeedForward Algorithm The linecycle average method of analysis leaves the impression that the modulator does nothing other than provide a current gain. The loop gain measurement results H(s)=A(s)/B(s) show only the firstorder response. For that reason. This proves that the assumption from the previous chapter. 129 . The firstorder response can be observed for other power factor loads as well. that the modulator together with the load is essentially a nonlinear gain (512). This might not be at all obvious when the nonlinear and heavily discontinuous NP current is considered. Figure 6.1. with the phase slowly increasing with the increase of frequency. in order to allow the loopgain measurements with the network analyzer. which occurs as a consequence of the delay in the control loop.
p CDC V NP o CDC iNP ThreeLevel Converter Switching Network ia La Lb Lc Ra ib ic Rb Rc n Vpn/2 + + Digital Modulator Digital Controller D/A A/D D/A A B perturbation Figure 6. 130 .1. Experimental measurement of the loop gain in the NP controller.
1. 6. Figure 6.2.9 3 10 f[Hz] 2 10 3 10 f[Hz] 2 10 3 10 f[Hz] 2 10 3 Figure 6. with and without FeedForward Terms Control authority of the zerosequence term for the sine triangle modulated converters can best be observed by studying the expressions for linecycle average currents without (61) and with (62) feedforward correction.1 Control Authority Over the NP Current.m=. Expressions (61) and (62) give the average current that flows into the NP under the control of the modulating signal’s zero sequence δ.5 40 20 0 20 40 1 10 40 20 0 20 40 1 10 40 20 0 20 40 1 10 Phase[deg] Gain[dB] Phase[deg] Gain[dB] Phase[deg] 2 3 80 100 120 140 160 1 10 200 100 0 100 200 1 10 200 100 0 100 200 1 10 Gain[dB] 2 3 10 f[Hz] 10 m=.3 shows the linecycle average of the current flowing into the NP when the dclink capacitors are balanced. The figure shows the symmety around the zero value of the 131 . Both expressions assume there is either a very small ripple or none at all in the NP.7 10 f[Hz] 10 10 f[Hz] 2 10 m=. Measurement of the NP control loop gain without normalization.
As the modulation index approaches the point A=. − 3 ⋅ I ⋅ cos( φ ) ⋅ 2 ⋅ ( 4 ⋅ K n + 2 ⋅ δ − 4 ) ⋅ A 2 − δ 2 − A2 ⋅ π ⋅ ( K n − K p ) + 4 ⋅π ⋅ A ⋅ K n ⋅ K p + 2 ⋅ A 2 ⋅ ( K p + K n ) ⋅ sin −1 ( δ )] A i NP = (62) [ 132 . A.5. the current that can be commanded into or out of the NP is rather small. Again.zero sequence δ. (61) iNP = − 3⋅ I δ ⋅ cos( φ )δ ⋅ A 2 − δ 2 + A 2 ⋅ sin − 1 A ⋅π A The zero sequence’s control authority over the NP in the presence of the constant feedforward terms is as expected not symmetric around zero value of the zero sequence because of the presence of the feed forward terms. and large modulation index.4. The expression for the line cycle average current (62) is plotted for two different feed forward terms in Figure 6. the zero sequence exhibits the most control authority over the duty cycle around A=0. because the difference between the upper and lower duty cycle is small. the control authority reaches its maximum. For the small.5.
4 0. 133 . The linecycle average NP current as a function of the zerosequence component δ.2 0 A=0.4 0.6 0.2 A=0.2 0 delta 0.3.8 1 0.4 0.7 A=0.4 0.1 0.9 A=0.6 0.8 1 Figure 6.3 A=0.6 0.8 0.2 0.8 0.0.6 0.5 0.
8 0.8 0.9 Kn=1.4 0.6 0.8 1 1 0. 6 0. .6 0.4 0.1 A=0.0 .00 68 .81 0. 8 0.0.2 0 4 .7 A=0.1 A=0. 0 0..4 0.6 A=0.5 0.7 0.6 0.2 0.2 0.2 0 0.4.3 0.4 0 . 0 04 6 .6 .4 0.02 delta 00.6 Kp=1.2 0 delta 0.8 1 1 1 0. 134 .8 1 0.4 0.1 A=0. ..2 0.2 0.9 0.1 Kn=0.6 0.2 A=0..5 .2 0 0 0. The linecycle average NP current with feedforward terms included.0 .4 0.8 0.6 A=0.8 0.9 A=0.8 0.. 2 delta 00 2 4 .4 Kp=0.3 A=0.9 0.6 0.8 1 Figure 6.4 A=0.
some of the simulation results will be verified experimentally. which present themselves mainly in the conservative stability estimate in the case of the analysis of the feedforward algorithm as well as in the inability to predict the NP ripple’s effect on the output line voltage. a model of load. so that the modulator is not saturated at any given moment). anything other than a constant zero sequence is just too difficult to include in the analytical expressions for the linecycle average current. However. The full dynamic simulation of the converter and the load will overcome these limitations. depending on the sign of the NP voltage error. which were not only difficult to solve analytically.2.6.2 Analysis of the SwitchingCycle Average Model of the Sine Triangle Modulator with the FeedForward Algorithm All the analysis performed so far included one key assumption: that the capacitance of the dclink capacitors is sufficient to prevent a significant ripple. Thus way. Finally. The “no ripple” approach was chosen because of the complexity of the modulator and the inherent discontinuities in modulation. The model consists of three phase leg average models.1 Average Model of ThreeLevel Converter with Sine Triangle Modulator Figure 6. which is essentially a comparator that applies maximum positive or maximum negative zero sequence (maximum.5 shows the Matlab Simulink average model of the threephase converter with the RL load. 6. and an NP controller. The only assumption in this model is that the there is an ideal dcvoltage source at the dc link. There are two load models that were used for the simulations: The first model used for the simulations of the inverter mode of operation was a 135 . but were also very hard to explain intuitively. In addition. the simplified model has its own limitations. the influence of the possible ripple voltage could be neglected in order to study the limitations of the modulation in the sense of controlling the NP current on the switchingcycle level and on the linecycle level.
Average model of the sine triangle modulator with feed forward.6 shows the implementation of the average model of the phase leg. and then the duty cycles and the output voltage are computed from the average equations. 80 Input Voltage Vdc Vdc delta_Vnp mod_index np_control feed_forward_enable Vout Vref d_np d_pos d_neg va Vout Vref d_np d_pos d_neg vb vc d_np_A delta_Vnp d_np_B d_np_C [np_voltage] Vdc delta_Vnp phase_A . as in (516) and (517). the second load model used for the rectifier mode of operation was three sinusoidal phaseshifted current sources. the modulating sinusoid is scaled with the modulating index. Figure 6. added with the zero sequence. Essentially.9 Mdulation Index mod_index np_control feed_forward_enable phase_B 2 [mod_index] Vdc delta_Vnp mod_index np_control feed_forward_enable Vout Vref d_np d_pos d_neg RL load and np_capacitance 3 [np_voltage] Feed Forward Control phase_C 0 Constant8 delta_Vnp_error Zero Sequence np_control modulation_index Sum5 [mod_index] np_controller Figure 6.dynamic model of the balanced threephase RL. 136 .5.
2 Performance of FeedForward Algorithm. Average model of one phase leg sine triangle modulator. Simulation Results The qualitative analysis and the expressions for the NP current obtained through the linecycle averaging indicate that there is a potential to lose control of the NP in the presence of the feedforward control. 6.7.7. Model of NP bangbang controller. the harmonic performance of the converter is not of interest.6. bang bang np_controller 1 np_control MATLAB Function Mux 1 delta_Vnp_error 2 modulation_index Mux4 Figure 6.1 1 Vdc 2 delta_Vnp Mux MATLAB Function Average model of phase leg modulator Sine Wave 3 mod_index Product Sum Saturation Demux Mux Demux Vout 2 Vref 3 d_np 4 d_pos 5 d_neg 4 np_control 5 feed_forward_enable Figure 6. in which the converter has the tendency to further charge the capacitor that is already charged to a higher 137 . nevertheless. Finally. This implementation of the controller is probably not optimal. This danger exists only in the rectifier mode of operation. the bangbang NP controller is shown in Figure 6.2. in the case of the average model.
except that the power factor angle is 0. the third the NP voltage and the last is the feedforward enable command. The operating conditions are the same as in previous simulations. 138 . Figure 6. The effectiveness of the feedforward algorithm is clear.potential. The amplitude of the phase current is 80A. the second is the line current. As many simulations show. and the modulation index is unfavorably high at m=0. but the NP ripple size was significantly reduced for this operating point. All the parameters closely resemble the system parameters from practical experiments. it is important to notice that the operating conditions are chosen to have an unrealistically large ripple (dc capacitors have a 75% voltage swing) in the NP in order to more clearly demonstrate the effectiveness of the feedforward algorithm. Not only was the linevoltage waveform dramatically improved and completely cleaned of harmonic components. the feedforward algorithm does not critically affect the stability of the NP. Furthermore.8 shows the rectifier mode of operation with the 85degree power factor angle. The simulation shows that the converter operates with no distortion in the average line voltage. even when the dc link collapses completely.95. There is a very narrow region with a very high NP ripple. For this reason the system is simulated in the rectifier mode of operation for different power factors’ ac currents. The first graph is line voltage.9 shows one boundary for which the feedforward algorithm causes the loss of NP control. the dclink voltage is 80V dc. Figure 6. high power factor (mostly active power) and high modulation index (very little zerosequence control) when the instability may occur.
2 0.2 0.12 0.17 0.14 0.2 Figure 6.5 0 0.17 0.1 0.15 t[s] 0.1 40 Vnp[V] 20 0 20 40 0.1 1.15 t[s] 0. 139 .18 0.16 0.15 t[s] 0.12 0.1 0.13 0.18 0. Rectifier mode of operation with low power factor.NP Control in Rectifier Mode of Operation 100 50 Vab[V] 0 50 100 0.16 0.18 0.15 t[s] 0.11 0.14 0.17 0.11 0.19 0.14 0.8.17 0.13 0.19 0.19 0.16 0.13 0.19 0.13 0.16 0.12 0.11 0.12 0.18 0.5 0.2 100 iab[A] 0 100 0.5 Feed Forward 1 0.14 0.11 0.
16 0.11 0.5 0 0.13 0.15 t[s] 0.15 t[s] 0.13 0.1 40 20 Vnp[V] 0 20 40 0.17 0.11 0.5 0. Finally.14 0.2 100 iab[V] 0 100 0. where R= 0.17 0.13 0.18 0.16 0.11 0.12 0.14 0.14 0. m=0.2 0.11 0. The simulation results show an improvement in the quality of the line voltage.18 0.12 0.NP Control in Rectifier Mode of Operation 100 50 Vab[V] 0 50 100 0.9.16 0.15 t[s] 0.16 0.17 0.17 0.12 0.19 0.1 0.5 Feed Forward 1 0.19 0.19 0. 140 .1 1. and with a high modulation index.13 0.2 Figure 6.12 0.06Ω.19 0.2 0. Rectifier mode of operation for high power factor.14 0.95.8 mH (85degree power factor angle).1 0.18 0. the algorithm is simulated in the inverter mode of operation with the dynamic model of RL load.15 t[s] 0.18 0. while the NP voltage remains unaltered. L=1.
18 0.12 0.2 100 iab[V] 0 100 0.15 t[s] 0.17 0.17 0. Experimental Verification of the Model Because the simulation results of the average model of threelevel converter with the sine triangle modulator did not exactly match the experimental results.14 0.2 0.5 0.18 0.11 0.10.15 t[s] 0.13 0.1 1.11 0. This leads to the discrepancy in the zero sequence of the space vector modulated experimental hardware and the sine triangle modulated simulation results.1 40 20 Vnp[V] 0 20 40 0.2 0.19 0.18 0.11 0.15 t[s] 0.5 0 0.NP Control in Inverter Mode of Operation 100 50 Vab[V] 0 50 100 0.15 t[s] 0.19 0.13 0.16 0.14 0.13 0.19 0.18 0.17 0.12 0.14 0.17 0.16 0.1 0. 141 . 6.11 0.16 0.12 0.14 0. the average model of the fast space vector modulator is developed next.12 0.16 0.13 0. The inverter mode of operation with low power factor load. It is important to mention that the discrepancy was only to be expected.1 0.3 Average Model of Fast Space Vector Modulator.5 Feed Forward 1 0.19 0. because the modulator of the experimental system was implemented using SVM with the NP control based on the selection of redundant small vectors and not on control of the zero sequence.2 Figure 6.
The first function finds the nearest three vectors and their corresponding duty cycles. then a switchingstate selection function converts 142 . In this case.3. Figure 6.9 mod_index h Modulation Index Coordinate Transformation [A] From 80 Vdc Demux Demux Mux [A] Goto ib ic Vdc Inp Vnp_error Vc vc ic Mux RL load Fast 3level modulator ZeroOrder Hold 0 Vdc2 1/3300e6 Sum3 SMES_capacitance 1 s Integrator4 Figure 6. Average model of the fast SVM algorithm. Finally. h) coordinate system.12 shows some details of the implementation of the fast space vector modulator. g g Va va ia h ia Vb vb ib .1 Average Model of the ThreeLevel Converter with Fast Space Vector Modulation Algorithm Figure 6. which converts the rotating reference vector from the orthogonal to the nonorthogonal (g. the fast SVM algorithm is implemented. the main assumption is the constant dclink voltage. There are three functions implemented.11 shows the Matlab Simulink model of the average model of the threephase converter with the RL load. and the integrator that uses the NP current to compute the NP voltage ripple.11. As in the previous model. The difference from the previous model is the modulating technique. The model implementation is relatively straightforward: there are two essential blocks. There is still a coordinate transformation block. the modulator and the load.6.
the third and last function adds together the products of duty cycles and voltages to find the average phase voltage and adds the NP currents from the three vectors in order to find the resulting average NP current. The figures in the left column are the measurement results. Finally. Mux MATLAB Function sw_state_selection first_vector 1 1 g 2 h Mux find_NTV Mux MATLAB Function Demux Mux MATLAB Function Mux sw_state_selection1 second_vector MATLAB Function phase voltages Va 2 Demux Vb 3 Vc 4 Inp Demux2 Mux 3 ia 4 ib 5 ic 7 Vnp_error Mux2 Mux Demux1 third_vector MATLAB Function sw_state_selection2 Mux1 6 Vdc Figure 6.3. The two columns seem to be in excellent agreement. 6. where R=0. the figures in the right column are the simulation results with the average model of the space vector modulator.12.the vectors given in (g.8 mH (making a power factor angle of 85 degrees). 143 . Fast threelevel space vector modulator.13 shows the NP voltage for several different modulation indexes for the 80V dclink voltage and for the RL load.2 The Fast Space Vector Modulation Algorithm.06 Ω and L=1. Experimental Results Figure 6. h) coordinates into the switching combination and selects between the redundant vectors in order to balance the NP.
9 0.1 0.13 m=.11 0.01 0.9 Vnp[V] 30 20 10 0 10 20 30 0.01 0.7 0.005 0 0.13 0.12 m=.005 Vnp[V] 30 20 10 0 10 20 30 0.15 t[s] 0. It might be possible to design threelevel converters for certain with requirement for a very low dclink capacitance applications and very simple controller. which might even operate without the feedforward algorithm and with only a basic NPbalancing scheme. The strong third harmonic in the linetoline voltage gets canceled for the balanced load.02 0.025 t[s] Vnp[V] Vnp[V] 30 30 20 10 0 10 20 30 0.01 0.1 30 20 10 0 10 20 30 0.15 t[s] Figure 6.025 t[s] 0. 144 .005 0.15 shows simulated linetoline currents for the same operating condition as in the previous two figures.025 t[s] Vnp[V] 0 0.01 m=.5 0.015 0.14 shows the measured and simulated linetoline voltage for the same operating point as in the previous figure.m=.01 0.14 0. Measured and simulated NP ripple of a threelevel converter.005 0.11 0.005 0.005 0 0.1 30 20 10 0 10 20 30 0.12 0.12 0.02 0.015 0.13.5 Vnp[V] 20 10 0 10 20 30 0. It is interesting to observe the strong thirdharmonic component in the linetoline voltage.11 0. The measurement results are not shown because they were also nearly perfect sinusoids.7 0.14 0.02 0. Figure 6.13 m=.15 t[s] 0. Figure 6. resulting in nearly perfect phase currents.14 0. This fact points out an interesting new aspect of the design optimization of threelevel converters.01 m=.015 0.
145 .01 0.9 0.11 0.005 0.015 0.01 0.11 0.025 t[s] 0.15 t[s] Figure 6.7 0.1 m=. Measured and simulated line voltage in the presence of NP ripple.01 0.13 0.13 0.015 m=.m=.005 Vab[V] 100 50 0 50 100 0 0.12 m=.1 100 50 0 50 100 0.14 0.005 0 0.15 t[s] 0.11 0.1 Vab[V] 100 50 0 50 100 0.005 0.14.12 m=.7 0.15 t[s] 0.5 Vab[V] Vab[V] 100 50 0 50 100 0.025 t[s] 0.14 0.01 0.005 0 0.02 0.12 0.015 m=.005 0.02 0.01 0.14 0.13 0.9 Vab[V] 100 50 0 50 100 0.5 0.02 0.01 0.025 t[s] Vab[V] 100 50 0 50 100 0.
125 0.1 0.12 0.14 0.14 0. Simulated linetoline currents for balanced and symmetric load.5 100 50 ia[A] 0 50 100 0.105 0. 146 .135 0.135 0.105 0.11 0.135 0.9 0.145 t[s] 0.13 0.1 100 50 ia[A] 0 50 100 0.115 0.125 m=.11 0.15 Figure 6. 6. The expressions for the NP current in the presence of the feedforward control show that the stability is compromised only in the rectifier mode with a certain amount of voltage error in the NP and for operation with either very low or very high modulation indexes.m=.125 m=.115 0.115 0.14 0. With the linecycle averaging technique experimentally proven.4 Conclusions This chapter verifies the essential assumption of the linecycle averaging method through experiment.15.145 t[s] 0.145 t[s] 0. The experiment proves that for the purpose of NP control the inverter together with the load can be considered a variablegain current source with no dynamics.15 0.12 0.13 0.12 0.7 0.105 0. the next step is the stability study based on the expressions for NP current obtained through linecycle averaging.15 0.1 100 50 ia[A] 0 50 100 0.13 0.11 0.
Because the expression for linecurrent averaging assumes a constant dclink imbalance and a constant zero sequence, it seemed likely that the danger of NP control stability problems was exaggerated. That is the reason why a switchingcycle average model of the converter was implemented and used to study the stability in the presence of a significant voltage ripple in the NP. The full dynamic simulations show not only the excellent performance of the feedforward algorithm, but also that it is unlikely to have the stability problems in any but the most extreme conditions. The NP control stability problems were observed only with the NP ripple higher than 50% of the nominal split dclink capacitor voltage, while at the same time the converter operated at high modulation index and in PFC (or near PFC operation).
Finally, although it was expected that the dynamic model of the converter with sine triangle modulator was not going to exactly match the experimental waveforms of the converter with the fast SVM algorithm, a dynamic model of the converter with the space vector modulator was developed as well. The results of the simulations using the space vector modulator almost exactly match the experimental results, and in that way demonstrate the validity of the switching cycle averaging method.
147
7
CONCLUSIONS This work discusses space vector modulation and the control of multilevel power electronics
converters, and is really the result of generalizations and insights gained from the theoretical study of a practical problem, such as the design of a power conditioning system for superconductive magnetic energy storage. The engineering challenge in designing the power conditioning system (or more precisely, in the design of the digital controller for the power conditioning system) was to enable the power conditioning hardware to control the bidirectional power flow between the magnet and the utility line. In fact, the challenge was to control the power conditioning system in order to provides stable, controllable and highquality power on both the magnet and utility sides.
The first engineering part of this research shows that a power conditioner based on the threelevel converter topology is feasible, and that its control design can provide a stable, highbandwidth, bidirectional control of the power flow. In addition, the control design
approach in which the NP controller was designed separately from the main power stage was theoretically and practically justified.
During the modeling and hardware development phase of the digital controller, two main problems presented themselves as worthwhile research topics. One problem is the excessive complexity of the implementation of the SVM algorithm, particularly for converters with a higher number of levels. The other one is more specific and deals with the issue of controlling
148
the NP of the threelevel converters; This is a very important aspects of the threelevel converter design.
Accordingly, Chapter 3 introduces the new SVM algorithm that is computationally very efficient and applicable to converters with any number of levels. The computational efficiency of the algorithm comes from its simplicity, and it is important to mention that not only can the same algorithm used for converters with any number of levels, but the number of steps in the algorithm remains the same regardless of the number of levels in the converter. These features make the new algorithm not only a good design and implementation choice, but also a very good tool for teaching the mathematical basis of multilevel SVM.
The research on NP control and the issues of the imbalance in the NP control are approached from several different angles. The initial approach taken was to develop the model of the space vector modulator for the purpose of investigating the NP current ripple. The developed model is general in the sense that it can describe any SVM, and its effects on the NP. Using the
developed model (under the assumption of the infinite dclink capacitance), the model presents the properties of the NP control in an intuitive way. In addition, a region in which the dc link can be balanced on the switchingcycle level was identified, and a quantitative study of the NP balancing offered as an essential tool in designing the size of the dclink capacitors. Finally, some of these results were verified experimentally.
The study of the properties of the NP and the ripple in the NP led to the following idea: If there is a way to compensate for the effect of the NP ripple in the control, than it might be possible to allow designers the freedom for a new kind of design tradeoff between the size of the dclink capacitors and the increased voltage capacity of the switching devices.
The result of this study is a feedforward algorithm, which can actively compensate for the ripple in the NP so that it does not propagate and affect the output of the converter. The algorithm is proposed for both the carrierbased modulation and for SVM. Subsequent analysis was performed with multiple tools, including linecycle average, switchingcycle average, and finally, with the dynamic simulation. The study shows that the feedforward algorithm (although
149
It might be interesting to generalize the modeling approach to include the NP ripple if the dclink capacitors can not be assumed to be infinitely large. Unfortunately. Finally. it might be interesting to determine how much can really be saved by trading off the increased voltage rating of the switches for the decreased capacitance of the dclink capacitors. Finally. In the future. in the rectifier mode of operation and under the most extreme conditions (high modulation index. • The research in the application of the new fast SVM algorithm still has potentially very interesting research opportunities. In that case the amount of feedforward action can be adjusted in order to avoid stability problems. with respect to the feedforward algorithm. there are several interesting topics for further research: • The digital controller for the SMES power conditioning system was built as a centralized controller. thus standardizing the signal distribution network throughout the converter. the model (although general) does not allow the ripple in the NP. Another interesting opportunity can be found in the challenge to generalize and organize the conversion of the switching vectors back into the switching states. it would be interesting to implement a distributed controller approach into the design of the digital controller like in [F17. it compensates for the effects of the imbalance in the NP. while using the power electronics building block (PEBB) approach. there exists the possibility that the NP control algorithm can cause the loss of NP control. higher than 50% ripple and high power factor current). • With respect to modeling the space vector modulator for the purpose of NP balancing. Perhaps the most immediate research topic can be found in the attempt to generalize the algorithm to converters with more than three phases by exploiting the hidden symmetries in the higherdimensional spaces of the more these converters. In addition. and in some cases reduces the NP ripple itself. for the more confident application 150 .it affects the NP control) does not require redesign of the NP controllers. F18].
it is important to make a more detailed study of the conditions that cause instability in the rectifier mode of operation. 151 .of the algorithm.
921 –928.548553. M.509517. Tolbert. Mazzzucchelli. pp. 1998. vol. “Multilevel Converters for High Power AC Drives: A Review. Marchesoni.” Power Engineering Journal. pp. [A5] T. Meynard. 1996 152 . vol. May/June 1996. 2126. no.3843. 12. A. Bhagwat. Stefanovic. Tenca. Sinha. A. 32. 6. [A8] Leon M. 2. “Multilevel Converters for Large Electric Drives. [A10] M.REFERENCES A. 10131018. 1124. Divan. “About the DCLink Capacitors Voltage Balance in MultiPoint Clamped Converters. Hochgraf. G. 1995. Summer. pp. 10571069. D. R. pp.” IEEEISIE International Symposium. V. M. vol. vol. pp. Peng. [A4] JihSheng Lai. D. 1998. Fang Zheng Peng. pp. “Advanced topologies and modulation strategies for multilevel inverters. pp. T. Lipo. “Comparison of Multilevel Inverters for Static Var Compensation. November/December 1983. A. vol. “Multilevel Converters and Derived Topologies for High Power Conversion. 2126. 530536.” IEEE Transactions on Industry Applications. P.1. 1998.” IEEEIECON Conference Record. no. 1993. 1.” IEEEIAS Conference Record. “Multilevel Power Conversion – An Overview of Topologies and Modulation Strategies. Newton. Suh. Marchesoni. H. “Multilevel Converters . 1998. vol. 1994.” IEEE Transactions on Industry Applications. 1.” IEEEOPTIM Conference Record. vol. G. M. “Multilevel converters. R. “Generalized Structure of a Muiltilevel PWM Inverter. pp. IA19.” IEEEAPEC Conference Record. Mazzucchelli. [A9] C. 3.A New Breed of Power Converters. vol. Manjrekar. Foch. S. [A2] M. pp. Fang Z. M. a real solution to medium/highvoltage drives?.” IEEEIECON Conference Record. 2. T.” IEEEPESC Conference Record. Lipo. Manjrekar. pp. Lasseter. Venkataramanan. [A1] Multilevel Power Converters: Issues and Tradeoffs Pradeep M.23. [A7] M. [A6] B. [A3] C.
Choi. G. H.” IPECYokohama. vol.” IEEE Transactions on Power Electronics. 5765. 9. D. Cho. 123128. “Switching Frequency Optimal PWM Control of a ThreeLevel Inverter. Suzuki. Akagi. Rojas.C. “A Fast Space Vector Modulation Algorithm for Multilevel ThreePhase Converters.” IEEE Transactions on Power Electronics. H. “Using a ThreeLevel GTO Voltage Source Inverter in a HVDC Transmission System.10. [A12] N.” IEEE Transactions on Industry Applications. 260268.” IEEE Transactions on Power Electronics. D. “High Performance Control of a ThreeLevel IGBT Inverter fed AC Drive. G. 1993. Boroyevich. Lee. Sept. J.[A11] N. Hyun. March/April 1996. pp. September 1993. 2. G. Suh.P. 1991. Schibli. 3. 666672. H. no. Nguyen. Klabunde. 5765. G. pp. M. no.” IEEEIAS Conference Record.S. 1999.5. 1. F. “Analysis and Controller Design of Static Var Compensator Using ThreeLevel GTO Inverter. H. Akagi “A Vector Control System Using a NeutralPointClamped Voltage Source PWM Inverter. Ogasawara. 494499. B. T. Roberto Rojas.H. 11. N. 2. I. L. no. 5.” IEEE Industry Applications Society. “Analysis of Variation of Neutral point Potential in NeutralPointClamped Voltage Source PWM Inverters. [B1] ThreeLevel NeutralPointClamped Converters A. B. “Assessment of medium voltage PWM VSI topologies for multimegawatt variable speed drive applications. 422427. C. Nabae. pp. Tokuo Ohnishi.” IEEEPESC Conference Record. Zhang. no. A. 2228. Lipo. Lee “Space Vector Modulation of ThreeLevel Voltage Source Inverter. pp.1999. “ThreeLevel Space Vector PWM in Low Index Modulation Region Avoiding Narrow Pulse Problem.C. 1996. January 1996. no. Jung. Lipphardt. T. Liu.” IEEEIAS Conference Record.” IEEEIAS. 11. J. January 1996. A. Lewis. Shakweh.” IEEE Transactions on Power Electronics. pp. “Current control of a 3level rectifier/inverter drive system. no. 1.S. E.A. Y. pp. July 1992. 7.” EPE Conference Record. 1995. S. “A Novel PWM Scheme for a ThreeLevel Voltage Source Inverter with GTO Thyristors. D. vol. Cosan. [B2] [B3] [B4] [B5] [B6] [B7] [B8] [B9] [B10] [B11] [B12] [B13] 153 . pp. [A13] Y. “A threephase multilevel converter for highpower induction motors. pp. 481486. no. S. “A New NeutralPoint Clamped PWM Inverter. 1995. H. November 1995. 1.” VPEC Seminar Proceedings. September/October 1981. vol.” IEEE Transactions on Power Electronics. Rufer. Mao. Ohnishi. vol. pp. 6. “PWM Control Method for NPC Inverters with Very Small DCLink Capacitors. Celanovic. vol. pp. H. T. 1998. Yifan Zhao. pp.” IEEE Transactions on Industry Applications. September 1994. 978 –986.151155. vol. 518523. pp. Cho. Takayuki Suzuki “An Improved Voltage Vector Control Method for NeutralPointClamped Inverters. Takahashi. T. 965 –971. vol. C. vol. Akagi. 965970. 13. H. 32. pp. Cho.” IEEE Transactions on Power Electronics. Borojevic. Ogasawara. IA17. pp. S. vol. vol. M. H. Steinke. G. R.
P. Vol. Wolanski. Rojas. 13361343.” IEE Proceedings. Kaku. May 1997. K.” IEEEIAS Conference Record. 182190. M. 1999. A. Conraths. B. Steimer. pp.. Cho. K.” IEEEPESC Conference Record. E. 1999. Steimer. [B21] J. 1997. G. “A Four Level RectifierInverter System for Drive Applications. J. S. B. Peng. “A New Neutral Point Current Control for a 3level Converter/Inverter Pair System.” IEEEPCC Conference Record. 1991. “An Analysis of Midpoint Balance for the NeutralPointClamped ThreeLevel VSI. “FiveLevel GTO Inverters for Large Induction Motor Drives. Lee. 1997. pp. F. 1997.” IEEE Transactions on Power Delivery. “New Medium Voltage Drive Systems using ThreeLevel Neutral Point Clamped Inverter with High Voltage IGBT. H. J. 1999.[B14] C. K. Dai. 1. Borojevic “Evaluation and Development of New Power Electronic Technologies foe Superconductive Magnetic Energy Storage (SMES) using PEBB”. Miyashita.S. Lipo. G. Ohnishi. D.” IEEEIAS Conference Record. vol. “Regulating and Equalizing DC Capacitance Voltages in Multilevel Statcom. C. S. P. H. 901907. Lipo. 1996. 1997. IEEEAPEC Conference Record. Bruckmann. Sinha.H. K. C. 30.” IEEEIAS Conference Record. 980987. 1998. Celanovic. Lyons. Boettner. McKeever and J. H. Lee.J. A. Menzies.C. I. E. VanCoevering. [B22] R.” IEEEIAS Conference Record. Hyun. G. J. A. 1999.” IEEEPESC Conference Record. Lee. Steinke. 390396. [B20] N. Espelage. 4. Design and Operational Limitations. Steinke. H. 129134. 96103. 590596. “A Four Level Inverter Based Drive With a Passive Front End. [B23] P. pp. Lee. Choe. Ooi. C. 938944. F. Z. 1999. Y. G. pp. pp. W.S. Lai. Fukuda. VPEC Seminar Proceedings. Interface.” IEEE 1995. R. H. Conner. M. Mwinyiwiwa. pp. T. R. F.” IEEEPESC Conference Record. [B15] B. Suh. “A General Circuit Topology of Multilevel Inverter. 1995. Sone “Switching Loss Minimized Space Vector PWM Method for IGBT threelevel inverter. pp. J. [C3] [C4] [C5] [C6] [C7] 154 . pp.” IEEEIAS Conference Record. T. Lee. Larsen. [B17] S. [B18] D. T. B. April 1997. 2. “A Multilevel VoltageSource Converter System with Balanced DC Voltages. “PWM Control Method for a FourLevel Inverter. R. Mao. Martens. vol. Newton. 12. Suzuki. “Neutral Point Control for MultiLevel Inverters: Theory. M. D. no. pp. T. pp. Sinha. 3.” IEE Proceedings on Electric Power Applications. D.259264. July/August 1994. [B16] H. Boroyevich. Sumner. S. “Innovation IGCT Main Drives. Greif. [B19] Y. ”Harmonic Evaluation of CarrierBased PWM Methods Using Harmonic Distortion Determining Factor.Friendly Medium Voltage Drive Based on the Robust IGCT and DTC Technologies. “A Reliable. Chen. T. 144. S. Cho. H. Suzuki. Sommer. V. Griggs.H. H. Choi. Vlatkovic. vol. Gruning. no. [C1] [C2] NeutralPointClamped Converters with more than ThreeLevels N. J. F. D. S.” IEEE Transactions on Industry Applications. M. No.” IEEEIAS Conference Record. Lee. Z.T. P. “ A Comprehensive Study of NeutralPoint Voltage Balancing Problem in Three Level NeutralPointClamped Voltage Source PWM Inverters”.
Lasseter. vol. 1999. C. Habetler. T. “A Multilevel ConverterBased Universal Power Conditioner. 3. G. Garadella.” IEEEIAS Conference Record. Marchesoni. Carrara.” IEEE Transactions on Power Electronics. BoonTeck Ooi. D.” Power Electronics Specialists Conference. Newton. 5. P.” IEEEPESC Conference Proceedings. 1997. Lai. Tolbert. Peng. 596 –603.” IEEE Transaction on Power Delivery. pp. 1. Mwinyiwiwa. Wolanski. Valaerts. no. S. “Multilevel STATCOM with Third Harmonic Elimination on the DC Link Capacitor Voltages. “A Multilevel VoltageSource Inverter with Separate DC Sources for Static Var Generation.[C8] [C9] [C10] [C11] [C12] [C13] [C14] [C15] [C16] [C17] C. 497505. pp. 7. Wolanski. S. M. M. no. 2.” IEEEIAS Conference Record. F. W. 2. L. Mwinyiwiwa. Tatakis. C. [D1] Multilevel Cascaded Inverters M. M. 258 –267. 1.” IEEEPESC Conference Record. Newton. no. T. 1. vol. “Novel Technique for Maintaining Balanced Internal DC Link Voltages in Diode Clamped FiveLevel Inverters.” IEE Power Electronics and Variable Speed Drives Conference Record. 1. Z. Chen. 1992. J. no. Z. pp. 146. vol. April 1997. M. D. vol. pp.” IEE Proc. vol. pp.” IEEE Transactions on Power electronics. 202208. 36. pp. 1998. “A TransformerLess Static Synchronous Compensator Employing a MultiLevel Inverter.” IEEEPESC Conference Record. “A multilevel converterbased universal power conditioner. 33. B. [D2] [D3] [D4] 155 . P. M. 881887. Z. Mathaus. April 1990. Marchesoni.” IEEE Transactions on Power Electronics. Sciutto. R. Hocgraf. L. C. Peng. March 2000. Tolbert. vol. Hill. F. 7. J. Marchesoni. McKeever. Habetler. “A Non conventional Power Converter for plasma Stabilization. no. Van Coevering.” IEEE Transactions on Industry Applications. H. 15. G. pp.” IEEE Transactions on Power Electronics. MarchApril 2000.25412548. “Unified power flow controller (UPFC) based on chopper stabilized diodeclamped multilevel converters. 12551262. Harbourt. Peng. pp. “A novel Arrangement for the Balancing of the Capacitor Voltages of a Five Level Diode Clamped Inverter. 317 322.” IEEE Transactions on Industry Applications. vol. S. W. “Performance of Medium Voltage MultiLevel Inverters. Z. Tenconi. “A Novel Approach to the Generation and Optimization of ThreeLevel PWM Waveforms. Sumner. Sumner. January 1992. G. vol. 189204. Power Appl. Hammond. M. M. pp. vol. 1999. R. “HighPerformance Current Control Techniques for Applications Multilevel HighPower Voltage Source Inverters. D. Z. pp. Y. J. G. 393 –399. F. July. pp. B. 1995. “A New Approach to Enhance Power Quality for Medium Voltage AC Drives. Salutari. January/February 1997. BoonTeck Ooi. Electr. vol. “A New Multilevel PWM Method: A Theoretical Analysis. Mazzzucchelli. 12. 1988. B. 465470. May 1999. A.
6. Manjrekar. A. Z. pp. Hiti. C. vol. “MultilevelLevel Conversion: High Voltage Choppers and Voltage SourceSource Inverters. Duarte. Rodriguez. vol.[D5] [D6] [D7] [D8] [D9] [D10] [D11] [D12] [D13] Q. pp. HsinHung Lu.” PEDES Conference Record. Liang. vol. S. P. Vlatkovic. Adams. 1999. “A Generalized Structure of Multilevel Power Converter. Hiti. [E1] [E2] [E3] [E4] Flying Capacitor Multilevel Converters T. D. “Average Current Control of ThreePhase PWM Boost Rectifier.397403. H. vol.” IEEE Transactions on Power Electronics. 1995. 2. M. Lipo. 1999. F. J. “High Voltage Converter with Regeneration Capability. Silva. 92. 1998. 2. Jiang. pp. vol. H. Jan.” IEEE Transaction on Industrial Electronics.” IEEE Transactions on Power Electronics. A. V. “A Hybrid Multilevel Inverter Topology for Drive Applications. Osman.” IEEEAPEC. A. Lipo. pp. M. 1. 927933. Nwankpa. March 1994. “Control of FrontEnd ThreePhase Boost Rectifier. Manjrekar. 2. 1992. no. 1999. D. BorRen Lin. “A new control algorithm for threephase PWM buck rectifier with input displacement factor compensation.” IEEEAPEC Conference Record. pp. 92 –102. F. [F1] Modeling and Control of ThreePhase Converters S.” IEEEPESC. November/December 1998. 1998. 1999. F. “Multilevel inverter with series connection of Hbridge cells. 2000. Meynard. 4. J. Boroyevich. Lipo. 927933.” IEEE Transactions on Industry Applications. 11741180. pp. 34. “A Power Line Conditioner Using Cascade Multilevel Inverters for Distribution Systems. McKeever.” IEEEIAS Conference Record. “A MediumVoltage Drive Utilizing SeriesCell Multilevel Topology for outstanding Power Quality. 1994.R. vol. A. “Switching Angles and DC Link Voltages Optimization for Multilevel Cascaded Inverters. 859 –864. D. February 1997. no. vol. 1. vol.” IEEEIAS Conference Record. T. “Hybrid Multilevel Power Conversion System: A Competitive Solution for High Power Applications. D. O. Lipo. pp. 107115. “A Novel Nlevel High Voltage Inversion System.” IEEEPESC Conference Record. A.” IEEEIAS Conference Record. BumSeok Suh. J. R. F. Hiti. Dijkhuizen. pp. Steimer. [F2] [F3] 156 . R. Gonzales. pp. Y. 44. 1998. Boroyevich. Manjrekar. DongSeok Hyun. L. pp. vol. C. M. Moran. D. T. Ambatipudi. D. “Proper Choice of Flying Capacitors Based on Distinct Power Dissipation Models. 6267. A. L. D.“A Power Line Conditioner Based on Flying Capacitor Multilevel Voltage Source Converter with Phase Shift SPWM. P. Iturriz. T. Zhang and Y. 523529. W. Ladoux.” IEEEIAS Conference Record. Boroyevich. 2337 2343. T.” 1998. 15. 2. Foch. E. S. J. R. 2. 10771082. 131137. 1999. “Phasecontrolled multilevel converters based on dual structure associations.” IEEEPESC Conference Proceedings. YuanPo Chien. vol.” IEEEPEDS Conference Record. Jiang. vol. pp. 1. Peng. vol. pp.
A. Z.S. Lee. Blacksburg.” Ph. S. H. Alepuz. D. Lai. vol. 1995. Boroyevich. Cooley. Blacksburg. 3. Klassens. V. Bordonau. M. Zhang and D. pp. “Static VAR Compensator Using Neutral PointClamped PWM Inverter and its Control Scheme. pp. 2000. PE1. G.” Ph. Celanovic. “Modeling and Control of ThreePhase PWM Converters. Virginia Polytechnic Institute and State University. no. Cosan. 1995. S. 2. I. Lee. Hong. H. R. 223230. Dissertation. Sagawa. pp.” IEEE PESC Conference Record. 721726. J. Dissertation. Celanovic. pp.” IEEEPESC. 1998. Zhang. R.” Ph. Boroyevich. 1995. Cho.D. T. J. H. Pasadena. 1997. Cho “DSP Based Control of High Power Static VAR Compensator Using Novel Vector Product Phase Locked Loop”. Peracaula. 3. D. Mao and F. “Topology and Analysis in PWM Inversion. Matsui. “Dynamic Modeling of AC Power Converters. Gyu H. “Modeling and Control of a Neutral–Point Clamped Voltage Source Converter. K. vol. May 1998. vol. Celanovic. Cooley. pp. 1. M. Borojevic. R. J. Milosavljevic. 488493. Hiti.” VPEC Seminar Proceedings. D. Boroyevich. 748755. VA. “High Performance Power Converter System for Nonlinear and Unbalanced Load/Source.” IEEE Transactions on Industry Applications. D. C. “A New Control Architecture for Future Distributed Power Electronics Systems. “ DSP Based Implementation of a Power Electronics Control System. K. J. C. IEEEPESC Conference Record.” IEEE PCC Conference Record. 1993.” IPECYokohama.” IEEE Transactions on Power Electronics. J. Fukuda. B. Jung. F. 33. “Novel ReducedOrder SmallSignal Model of a ThreePhase PWM Rectifier and Its Application in Control Design and System Analysis. “A StateSpace Model for the Comprehensive Dynamic Analysis of ThreeLevel VoltageSource Inverters. I. I. Mao. and Cycloconversion. 511521. Prasad. Ngo. 1999. May/June 1997. Virginia Polytechnic Institute and State University. 470475. pp. 1996. “Dynamic Performance and Control of a Static Var Generator Using Cascade Multilevel Inverters. pp. C. Peng. 2000. Bauer.D. J. vol. S. Bordonau. G. D.” IEEE Transactions on Power Electronics. T. 238243.” IEEEPESC. N. CA. 1996. 1984. W. I. 13. D. Dubovsky. Ngo. H. Guo.” IPECYokohama. Celanovic. S. 6167.” IEEE APEC Conference Record. “Dynamic Analysis of ThreeLevel VoltageSource Inverters Applied to Power Regulation. vol. Dissertation. P. California Institute of Technology. Milosavljevic. G. 1997. “A New Distributed Digital Controller for the Next Generation of Power Electronics Building Blocks.D. no. Boroyevich. Rectification. F. N. VA. 502507. R. pp. SMES References 157 . pp.[F4] [F5] [F6] [F7] [F8] [F9] [F10] [F11] [F12] [F13] [F14] [F15] [F16] [F17] [F18] S. “Low Frequency Characterization of PWM Converters.
231237. 6. Lasseter. 3. M. Swe. Sept. pp. F. D. 2. 8. Dai. 237249. Funaki.D. G. D. R. Ogawa. 326333. Celanovic D. “Power Conditioning Systems for Superconductive Magnetic Energy Storage. R. H. Hassan. Bucci. L. D. Sato. A. T. pp. vol. C. M. April 1988.” IEEE Transactions on Energy Conversion. H. “Control Design for Superconducting Magnetic Energy Storage Power Conditioning System. Jalali. Meier.” IEEE Transactions on Energy Conversion. N. F. no. 6. I. Lee. Schoenung. no. “A Power Conditioning System for Superconductive Magnetic Energy Storage Based on Multilevel Voltage Source Converter. no. C. March 1993. D. Lee. September 1991. “Experimental Study of SMES System With DC Intertie for Power Line Stabilization. “Dynamic Response of Power Conditioning Systems for Superconductive Magnetic Energy Storage. 1999. S. pp. Takeo. pp. 3. 1999. 388393. H.” IEEE Transactions on Energy Conversion. 1997. T. VA. D. Lasseter.” IEEE PESC Conference Record. Ezaki. vol. 9. Blacksburg. July 1993. vol. pp. Lee. K. pp. H. 5462. H. “Operation Principle and Applications of Multiterminal Superconductive Magnetic Energy Storage System.” IEEE Transactions on Energy Conversion. S. Borojevic. KwaSur Tam.” 15th VPEC Seminar Proceedings. “Small SMES Technology and Cost Reduction Estimates. 158 . D. Lee. R. Borojevic. W. Virginia Polytechnic Institute and State University. 3. K. Bieri. Peng.” IEEE PESC. 381387. Jalali. 8. June 1994.” Ph. K. Mao. September 1991. R.[G1] [G2] [G3] [G4] [G5] [G6] [G7] [G8] [G9] H. vol. G. S.” IEEE Transactions on Energy Conversion. R. Dissertation. no. Okada. H. H. 129134. H. pp. vol. no. “Evaluation and Development of New Power Electronic Technologies for Superconductive Magnetic Energy Storage (SMES) using PEBB. Koba. S. “400 MW SMES Power Conditioning System Development and Simulation. M. Lee. Yarali. 1.
S. His research centered on the modeling and control of multilevel threephase converter systems. Yugoslavia. 159 . In the Summer of 1996. He also served as president of the CPES student council (Virginia Tech campus) and as a member of the joint CPES student council (all five campuses). and his M. degree in Electrical Engineering from the University of Novi Sad. The author is a member of IEEE. Yugoslavia.S. He received his B. degree in Mechanical Engineering from Vanderbilt University in Nashville TN in 1994 and 1996 respectively. in 1970. the author enrolled as a doctoral student at VPI&SU.VITA The author was born in Novi Sad.
This action might not be possible to undo. Are you sure you want to continue?
Use one of your book credits to continue reading from where you left off, or restart the preview.