Professional Documents
Culture Documents
Multilevel Converters
by
Doctor of Philosophy
(Electrical Engineering)
at
The work presented in this thesis is my own work and to the best of my knowledge
and belief original, except as acknowledged in the text. The material contained in
this thesis has not been submitted, either in whole or in part, for a degree at The
University of Queensland or any other University.
Geoff Walker
iii
iv
Acknowledgements
There are many people who deserve my sincerest thanks. I can only name some of
those very many people here. If you are one of the many more anonymous friends,
you will know that I have valued your friendship and support during the course of
this study. Thank you.
I would like to thank Professor Gerard Ledwich for his inspiration and direction
as my supervisor and friend. Professor Matt Darveniza, Dr. Ashwin Kambadkone
and Dr. Tapan Saha have also assisted in my supervision during the course of my
candidature.
Gordon Wyeth, Pat Bellett, Ed LoGuidice and Keith Hoffman have been my
closest friends at University. Thanks for all the ideas and laughter shared over a
cup of coffee. This isn’t to dismiss the wonderful friendships I have also formed with
the office staff, technical staff, postgraduates, undergraduates and other academics.
You have together been the reason I have always preferred to work at University
rather than from home.
My family (Mum, Dad, Lloyd, Peter, Lorraine) and all my church friends who
are as close as family have been a constant source of encouragement and love. Of
course most of all I must thank my wife Maria for her support, love, and above all
patience.
Above all I thank God. My life has always been full of abundant blessing. I can
see His guiding hand clearly at work as look back across my life. His love, forgiveness
and peace are real to me everyday. It is my prayer that God will be seen in me and
my work, including this thesis, and the glory will be given to God.
v
vi
Abstract
This thesis aims to extend the body of knowledge about the modulation and control
of converters to cover multilevel converters.
Natural and uniform sampled sine-triangle modulation are examined and con-
trasted in detail. The modulators’ transfer function and spectral properties are
examined mathematically, and then confirmed both with simulations and software
and hardware implementations. The natural sampling process is shown to have a
flat, distortionless transfer function, which uniform sampling cannot achieve. Uni-
form sampling is invariably implemented digitally, either in software or hardware,
with many associated advantages, while natural sampling does not lend itself to a
digital implementation. The advantages and short-comings of each technique are
vii
generally exacerbated by multilevel converters. A digital implementation of natu-
rally sampled PWM would seem the perfect solution.
In a three phase converter with no neutral connection, any voltage which is com-
mon to all three phases does not cause any current to flow. Such a zero sequence
component may be deliberately introduced to increase the maximum possible mod-
ulation depth and hence converter utilisation, lower the switching rate, and possibly
lower distortion. A number of possible zero sequence components for multilevel
modulation are tested using simulations. Only the increase in maximum modula-
tion depth is a significant advantage. The distortion (spectral) benefits are minimal,
and discontinuous switching cannot be implemented in some topologies.
viii
Original contributions
The many possible pulse width modulation (PWM) strategies are reviewed with
particular regard to their comparative suitability for the modulation of a general
multilevel converter. Carrier based techniques are shown to be the clear choice for
both technical and pedagogical reasons.
Natural and uniform sampled PWM are compared again with a specific focus
on their relative suitability for multilevel modulation. The new demands placed on
these PWM techniques by the new opportunities opened by multilevel converters
are examined. The mathematical expressions for these PWM techniques, extended
to describe the multilevel waveforms, are examined at the very low (even fractional)
pulse numbers which only become relevant for multilevel converters. The need for
a digital implementation of natural sampling is demonstrated.
ix
ative merits are contrasted. Many examples of the flexibility of the EPROM based
hardware approach are offered.
Zero sequence vectors have been applied to conventional three phase converters,
but not previously to multilevel converters. Multilevel converters offer new degrees
of freedom to the creation of zero sequence vectors. The application of zero sequence
vectors to multilevel converters is simulated and the results analysed.
x
Publications
xi
xii
Contents
Acknowledgements v
Abstract vii
Original contributions ix
Publications xi
1 Introduction 1
2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
xiii
2.3 Three Phase Multilevel Voltage Source Converters . . . . . . . . . . . 11
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
xiv
4.2.1 Constant (DC) Modulation . . . . . . . . . . . . . . . . . . . 47
4.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.5.5 Overmodulation . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
xv
5.5.1 Carrier Variations . . . . . . . . . . . . . . . . . . . . . . . . . 89
7 Conclusion 129
xvi
7.7 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
C Circuits 163
xvii
C.2 DSP-Flex FLEX8820 configuration . . . . . . . . . . . . . . . . . . . 167
Bibliography 172
xviii
List of Figures
2.3 Voltage source, current source and three level matrix converters . . . 9
xix
4.2 A selection of linear carrier waveforms . . . . . . . . . . . . . . . . . 46
xx
5.6 The operation of the programmable gate array implementation of re-
sampled uniform. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
xxi
5.23 EPROM re-sampled uniform — transfer function vs. modulation
depth, rsr = 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.3 Multilevel spectra — Two identical random carrier pairs, one delayed;
Synchronous and Asynchronous . . . . . . . . . . . . . . . . . . . . . 107
xxii
6.18 PWM spectra — maximum modulation depth . . . . . . . . . . . . . 123
xxiii
xxiv
Chapter 1
Introduction
Conventional power electronic converters can switch each input or output connection
between two possible voltage (or three possible current) levels, namely those of the
internal DC voltage (or current) link. Multilevel converters can switch their outputs
between many voltage or current levels, and have multiple voltage or current sources
(or simply capacitors or inductors) as part of their structure.
The most common initial application of multilevel converters has been in trac-
tion, both in locomotives and track-side static converters [60]. More recent appli-
cations have been for power system converters for VAR compensation and stability
enhancement [21], active filtering [71], and most recently for medium voltage induc-
tion motor variable speed drives [25].
With multiple switches and hence degrees of freedom, the harmonic performance
2 Chapter 1. Introduction
of the converter can also be improved. Even with low pulse numbers (switching
rates), the performance of the multilevel converter can be useful, even impressive.
The modulation and control of multilevel converters is the focus of this work.
This thesis aims to extend the body of knowledge about the modulation and con-
trol of converters (fairly complete) to cover multilevel converters (still developing).
This first chapter explains the motivation and content of this thesis and presents a
summary of the content of each of the chapters.
multilevel converters. The advantages and short-comings of each are given as the
conclusion, to lead into the next chapter.
In a three phase three wire system where the converter has no neutral connection,
any voltage which is common to all three phases does not cause any current to flow.
Such a zero sequence component, or zero vector, introduces a degree of freedom
which may be used to improve the performance of the converter. The techniques of
applying zero sequence components to two level converters reported in the literature
are extended to the multilevel case.
were born from a need created by multilevel converters. However these techniques
can equally be re-applied to two level converters.
Chapter 2
2.1 Definitions
In the preface of his book Switching Power Converters [70], Wood introduces the
concept of a unified converter theory. There he states: “Most traditional views
of the field have seemed somewhat disjointed; converters were largely regarded as
related only because they all use semiconductor switches and have certain topological
similarities. . . . the view expounded herein (is that) switching power converters are
related by function and behaviour; their basic characteristics do not in any way
depend on the types of switches used, nor on the applications to which they are put,
nor on the topologies in which they are realized.”.
According to this unified theory, any power electronic converter can be viewed
as a matrix of switches which connects its input nodes to its output nodes. These
6 Chapter 2. The Multilevel Converter
nodes may be either DC or AC, and either inductive or capacitive; and the power
flow may be in either direction.
• If one set of nodes (input or output) is inductive, the other set must be ca-
pacitive, so as not to create a cut set of voltage or current sources when the
switches are closed.
• The combination of open and closed switches should never open circuit an
inductor, or short circuit a capacitor.
The term rectifier is used when the power flow is predominately from the AC
port to the DC port and the term inverter is used when power flow is predominately
from the DC port to the AC port. The term converter is used either when there
is no predominant direction of power flow or as a general term to encompass both
rectifiers and inverters.
In a Voltage Source Converter (VSC), the DC port is the capacitive port and
is voltage stiff (i.e. a large DC bus capacitor). The voltages in such a converter are
well defined by this port and are generally considered independent of the converter’s
operation. The value of the AC side inductance is comparatively small and modu-
lation of the converter controls these AC side inductor currents. Should the voltage
source converter be responsible for the control of the DC bus capacitor voltage, then
this voltage is indirectly controlled by controlling the net current flow in the capac-
itor. The switches in such a converter must block a unidirectional voltage, but be
able to conduct current in either direction if bidirectional power flow is desired.
is well defined and slow to change. The voltage (particularly at the AC port) is
considered the variable directly controlled by the converter modulation. Since the
AC port usually has significant line or load inductance, line to line capacitors must
be placed on the AC port. The switches must block either voltage polarity, but are
only required to conduct current in one direction. This naturally suits thyristors
and symmetrical GTOs.
Figure 2.1: A voltage source rectifier - inverter cascade (top) and a current source
rectifier - inverter cascade (bottom).
Since the AC line and AC motor loads are both inductive, Voltage Source Rec-
tifier – Inverter cascades (Fig. 2.1) are usually used for small and now increasingly
for large motor drives and similar applications, as GTOs and IGBTs have matured.
Larger converters have traditionally been current source converters, both because
this best suits the characteristics of the thyristors and because it requires a large
DC bus inductor, which was preferred to a large capacitor.
Some converters do not easily fall, or cannot be placed into either category. The
matrix or Venturini converter [1] is one example (Fig. 2.2). Both input and output
ports are AC, and the definition of voltage stiff or current stiff (and hence voltage
or current source) becomes somewhat arbitrary. Both input and output ports are
8 Chapter 2. The Multilevel Converter
Figure 2.2: The matrix converter, with one possible implementation of the bidi-
rectional switches.
AC, and neither port can be considered as a steady dc source, whether voltage or
current.
The next refinement is to define the meaning of multilevel. The following definition
of a multilevel converter is offered:
A multilevel converter can switch either its input or output nodes (or
both) between multiple (more than two) levels of voltage or current.
This simple definition is deliberately quite broad and inclusive, in keeping with
the spirit of the unified converter theory. For example, the multi-phase matrix
converter (Fig. 2.2) is, strictly speaking, a multilevel converter, according to this
definition. Consider the three phase to three phase matrix converter, with voltage
source inputs and an inductive load. Any single output can be switched to one of
three different voltage levels (the voltages of the three input phases) and similarly,
any input can be switched to one of four current levels (including zero).
2.1. Definitions 9
In this preceding example, both the input and the output nodes are AC peri-
odic varying quantities and so these levels can only be considered stationary for an
interval much shorter than their AC period.
Figure 2.3: The current source converter (top right), voltage source converter
(bottom left) and a simple three level voltage source converter (bottom right)
can all be derived from the general topology of the matrix converter (top left)
Both the voltage source and current source converters can be derived from the
general matrix converter by setting one port to be either a two terminal DC voltage
stiff or DC current stiff port [70, 30]. Retaining the third terminal leads to a simple
and more conventional multilevel converter (Fig. 2.3).
10 Chapter 2. The Multilevel Converter
Note that now one of the ports has been made DC and voltage or current stiff,
only one port will experience the multilevel stepped waveforms. The other will still
have a continuous waveform similar to that of an equivalent two level converter.
Most multilevel converters discussed in the literature step between multiple volt-
age levels. This is usually the most useful configuration for a high power converter,
as reducing conduction losses in both converter and machines will always favour in-
creasing the voltage rating rather than the current rating of the converter. Also as
power levels increase, the input and output voltage levels presented to the converter
increase. The structure of these multilevel converters place the switches in series to
share the duty of blocking these higher voltages.
Equally however, for high current applications, many switches can be placed
in parallel, with their current summed by inductors. When switched separately,
multilevel current waveforms result [2].
Generally multilevel topologies can be divided into two groups, although in some
cases the dividing line is indistinct. The first approach relies on summing the outputs
of a number of conventional two-level converters, to produce a resultant multilevel
output. The second group replaces the two-level switch structure with a multi-
level switch topology within an otherwise conventional converter. These two groups
will be distinguished by the terms multi-bridge converter and multilevel converter
respectively.
Any of the basic DC-DC converters (buck, boost, buck-boost, Cuk) can be
extended to a multilevel topology. Often these are not called or perhaps even recog-
nised as multilevel converters, but rather simply described as, for example, par-
allelled converters with interleaved switching instants. Two recent examples cited
are multilevel boost converters used for power factor correction [48, 34]. In both of
these examples, the switches are effectively placed in parallel and their contributions
summed by separate boost inductors. They present multilevel current waveforms to
the input and reduced voltage ripple at the output.
Multilevel DC-AC converters range from the simplest single phase, full bridge
driven with unipolar voltage switching [49] to complex multi-phase converters. These
are the most commonly recognised and reported multilevel converters and will be
further categorised and referenced in the next section.
Even multilevel AC-AC matrix converters have been shown to be at least theo-
retically possible [45].
verters
At this point in the chapter, we will narrow the focus to that of three phase volt-
age source multilevel converters. Although this may seem somewhat limiting, it
12 Chapter 2. The Multilevel Converter
encompasses most of the higher power multilevel converters both in the published
literature and in actual use. There are some examples of single phase converters
functioning as AC-DC switching rectifiers, either in traction, computer or telecom-
munications power supplies. These Power Factor Correction rectifiers have lower
inherent distortion and require less filtering because of their multilevel topology.
There are four main voltage source DC-AC multilevel topologies which have
been distinguished here and in the literature. These are:
Each of these will be examined in turn. Each of the diagrams presented are of a
five-level converter, which can produce a nine-level phase to phase voltage waveform.
verter
As the title suggests, these multilevel converters are simply a number of conven-
tional two-level bridges, whose inputs or outputs are summed using transformers or
inductors. The multiple transformer secondaries force voltage sharing between the
switches (Fig. 2.4).
The most common and well known example of a multi-bridge converter is the
twelve pulse thyristor converter, well covered in most power electronic textbooks [49].
Figure 2.4: A five-level Transformer coupled multiple bridge, which produces nine
level phase-phase waveforms on the transformer primary.
30◦ phase shift between transformer secondaries allows identical secondary switching
instants and current waveforms to appear interleaved on the transformer primary.
A series connection is used for HVDC; a parallel connection for high current ap-
plications such as electrolysis and electro-plating. The technique can and is extended
to many bridges each with a transformer secondary connection of the appropriate
phase shift to achieve cancellation of the further low order harmonics in the pri-
mary [53, 70]. By clever connection of the transformer primaries, current as well as
voltage sharing can be ensured [52].
48 pulse operation [27] is achieved by eight GTO bridges operating in square wave
mode, with reliance on the transformer for harmonic cancellation.
Research on a five level three-phase motor drive which used this technique was
2.3. Three Phase Multilevel Voltage Source Converters 15
conducted by Matsui et al [43]. The outputs of two half bridge legs were summed
with a current sharing reactor to form a three level intermediate output. This and
another similarly formed three level output were summed by a third reactor to form
the final five level phase output.
One further solution is to sum the outputs of two converters across a bridge
connected source or load. Both ends of the transformer or motor winding are brought
out and the winding must be fully floating. One converter is driven with a phase
inverted signal, so that twice the desired converter output is impressed across the
floating load. If the carriers are appropriately phased, part of the undesired carrier
component will appear as a common mode component to the load.
• The voltages within the individual converters and thus across the switches are
well defined by the stiff voltage source output of the transformer secondaries.
• Should a converter module fail, or be removed for service, the converter may
continue operating at full voltage, but at reduced current.
• Other than the transformer (inductors), the structure is modular, which allows
easier maintenance and reduced spares.
• Its mode of operation is easily understood and, again because of its modular
structure, control is more easily applied.
• The transformer itself, if not needed for isolation, adds significantly to the cost
of the converter and is one more item to maintain and potentially, to fail.
from one another and from ground. This is a significant problem at high
voltages. This also increases the cost of the transformer.
A second topology, which is really only a variation on the first, is that of series con-
nected bridge converters (Fig. 2.5). Each phase leg consists of series connected single
phase full bridges, the series connection being made directly (not by transformer as
in the first case) on the AC side. A three phase converter can be constructed by
connecting three of these single phase series strings to form a star or delta [57, 56].
Since this topology requires each full bridge to have an isolated DC bus, this
connection has not been considered useful until recently re-examined. Now this
topology is being considered for applications where no real power transfer is involved,
such as for active power filtering and VAR correction. Then only a floating DC bus
capacitor is required on each floating DC bus.
Some other sources of power which could easily be made modular and floating
are batteries for battery energy storage systems (BESS) used for load levelling, or
alternative energy sources such as solar panels.
It is of course possible to power the isolated bridges from multiple isolated trans-
former secondaries, each with their own rectifier [25]. By appropriate phase shifting
of the transformer secondary windings, harmonic cancellation can be achieved on
the primary side, as described previously, as well as at the multilevel output of the
multi-bridge converter. However the disadvantages of a transformer with multiple
isolated secondaries returns.
This multilevel converter structure has some very significant advantages, if its
limitations are acceptable. Its advantages:
• It has perhaps the simplest architecture and the lowest component count. No
transformer is needed, so capital costs are low.
2.3. Three Phase Multilevel Voltage Source Converters 17
• Again, the converter is very modular and easy to understand. This applies
not only to its structure, but also to its control.
• There is only limited access to DC bus capacitors, which limits its area of
application to either those with only reactive power flow, or those where the
power source or load can be both modular and isolated.
• Should a module fail (or be removed), it must fail short circuit, or be bypassed.
The converter can continue to operate, at full current capacity, but at reduced
voltage rating. This will in practice mean that if fault tolerance is required,
the converter will need a more conservative voltage rating — a potential cost
penalty.
The topologies discussed so far have generated multilevel waveforms by the series
or parallel connection of two level converters. These have been labelled multiple
bridge converters. Another group of converters achieve multilevel waveforms by
the series or parallel connection of switches within the converter bridge. These are
distinguished as multilevel bridge converters.
18 Chapter 2. The Multilevel Converter
The published work on these multilevel bridge converters has mostly concen-
trated on three level Neutral Point Clamped (NPC) converters [51], although more
recently four [59] and five level [44] NPC converters have been investigated (Fig. 2.6).
In a three phase N + 1 level NPC converter, each of the six valve legs is composed
of N separate series connected switches, each of which may be controlled indepen-
dently. The voltage of the non-conducting switches is clamped by diodes and a
string of series connected capacitors. This series capacitor string is shared by all the
phase legs and usually serves as the DC bus or bulk capacitor.
One major disadvantage of the NPC approach is the unequal load distribution
among the switches and capacitors, especially for higher level converters [56]. For
high modulation depths using sinusoidal modulation, outer devices conduct only
for a small part of the cycle, while the inner devices conduct for a larger fraction.
For lower modulation depths, for example below M = 0.5, the outer devices will
not conduct at all in a five level converter. Under these conditions, the capacitor
voltages will become extremely unbalanced unless advanced modulation strategies,
or additional circuitry is used to redistribute the charge in the capacitors. This
problem prevents the technique being practically extended beyond five level.
2.3. Three Phase Multilevel Voltage Source Converters 19
Figure 2.7: A diode clamped three-level voltage source converter. This topology
does not guarantee voltage sharing between the open switches, as can be seen in
the right most phase leg.
Although this topology may remove the need for a transformer, it requires a
number of auxiliary components, namely the clamping diodes. For converters larger
than three level, although often not drawn as such in the circuit diagram, these
diodes may in fact be a series string of diodes to achieve the required voltage rat-
ing [56]. Figure 2.6 has been drawn to highlight the need for these multiple diodes.
This is another reason why this topology is seldom extended beyond five level.
A further problem is that the voltage across some switches is not strictly guar-
anteed to be less than the voltage of one level [46, 33]. For example, in a three level
converter, only the voltages across the outer switches are clamped by the protection
diodes (Fig. 2.7). One solution to this problem is to place a capacitor across the two
inner switches in each phase leg, in parallel with the clamping diodes [33]. This is
a hybrid of the diode clamped and flying capacitor topologies and has most of the
advantages of both approaches, at the expense of a high component count.
In NPC converters, not all switch states are usable as some do not guarantee a
continuous path for load current to flow. Should the modulator produce these states,
they must be re-mapped to the equivalent valid state. For this reason, space vector
modulators with this knowledge of allowed states are usually used. Further, of these
allowed states, different switch states slew the voltages on the capacitors in different
directions. This fact should be used to control the capacitor voltages. Koyama et al
20 Chapter 2. The Multilevel Converter
present a voltage space vector based PWM method for modulation of a three level
GTO converter which addresses these problems as well as that of minimum pulse
width [35]. Velaerts et al present an alternative carrier based modulation strategy
for a three level NPC converter [64, 63].
Despite these many difficulties, the diode clamped converter, particularly in its
three level form, has received much attention and use. As a three level converter,
it is relatively simple, and can remove the need for a transformer where one would
otherwise exist.
• These extra components do not necessarily ensure equal voltage sharing for
all switches.
• Switch utilisation is not equal, outer switches receiving a lower average load.
This becomes particularly apparent as the number of levels increase and the
modulation depth is small.
• Similarly, the power flows to and from the different capacitors in the capacitor
string are not balanced.
• Not all switch states are allowed. The disallowed states must be re-mapped
to their equivalent allowed states.
must be used, which has been specifically tailored to the topology of the con-
verter.
Probably the most important multilevel topology to appear recently is the flying
capacitor converter, or imbricated cells multilevel converter, proposed by Meynard
and Foch [46, 47, 45].
Like the transformer summed bridges, this topology can be extended easily to an
arbitrary number of levels and to different applications such as DC-DC, DC-AC and
even AC-AC (matrix) converters, both for unidirectional and bidirectional power
flow [45]. Since both the voltage source and current source converter topologies are
derived from the matrix converter, it is also possible to develop a current source
flying capacitor converter. This will be discussed further in a later section of this
chapter.
• Any switch combination is valid and ensures voltage sharing, so long as switch
pairs receive complementary drive signals. Most modulation strategies are
easily applied to this topology simply by phase shifting the drive signals.
There are some significant disadvantages, which are not at first apparent:
• The topology requires a lot of high voltage capacitors — many more than
other topologies. These capacitors need to conduct the full load current for at
least part of the switching cycle. Fortunately, if the switch frequency is high,
these capacitors can usually be relatively small in capacitance value.
• Since these capacitors initially have zero voltage across them, starting the
converter safely may be a non-trivial task.
2.4. Current Source Converters 23
The converters which were focused upon in the previous section were voltage source
converters, with multilevel voltage waveforms. These converters divide the total
input voltage among a number switches, and allow a reduction of the voltage har-
monics. As mentioned, these are the most commonly used and best understood
multilevel converters.
If the capacitive port were an AC port and the inductive port current stiff and
DC, then this would be classified as a current source, multilevel current converter.
24 Chapter 2. The Multilevel Converter
+ I
V
Figure 2.9: The flying capacitor converter (left) — a multilevel voltage converter;
and a dual derived from it, the “flying inductor” converter (right) — a multilevel
current converter.
Two of the previously presented multilevel voltage converters are shown alongside
their “duals” in figures 2.9 and 2.10. In both of these examples the duals are
multilevel current converters — they present a multilevel current waveform to the
capacitive voltage port. The voltage rating of the converter is limited to that of the
switches, however, the input current is divided between the multiple switches.
Figure 2.10: The isolated flying bridge converter is usually a series connection of
voltage source full bridges (right). These present a five level voltage waveform to
the source via the smoothing inductance. A dual of this converter is formed from
isolated current source bridges placed in parallel (left). These present a five level
current waveform which is smoothed by a parallel capacitor.
Figure 2.11: The current source circuit topology derived using duality techniques
(refer to Fig. 2.10) may be used as derived. However, for a high power converter,
the overall voltage rating of the converter needs to be increased by placing the
switches in series rather than in parallel. The final current source circuit (bottom)
bears more similarity to the voltage source converter it was originally conceived
from (top), than to its strict dual.
26 Chapter 2. The Multilevel Converter
• Current source converters are suited to the high power devices such as thyris-
tors and GTOs, which can block voltage in either direction, but conduct cur-
rent only in the forward direction.
• Switch voltages are poorly defined. Most semiconductor switches tolerate tran-
sient over-current better than transient over-voltages.
• There are more potential resonance problems in input and output filters than
for voltage source.
• When used as a back to back rectifier inverter pair, a more complex and
expensive controlled rectifier with closed loop control is required as a current
source. An uncontrolled diode rectifier is often suitable for voltage source
converters.
Voltage source converters have become the dominant configuration, even at high
power levels. The direction of semiconductor technology has reflected this, with
asymmetrical or reverse conducting thyristors (RCTs) and GTOs being developed
specifically for VSC rather than CSC applications.
At this point it should be clear that one of the major advantages of a multilevel
converter, regardless of topology, is increased power rating. A converter need not
be limited in size by the prevailing semiconductor technology, since a multilevel
converter allows the voltage and/or the current to be shared among a number of
switches. This advantage has traditionally justified the extra complexity of multi-
level converters only at very high power levels, for large motor drives and utility
applications.
The more stringent harmonic standards now being legislated also advantage
multilevel converters, since they produce lower switching harmonic spectral compo-
nents for a given switching frequency limit. This will be explained further in later
chapters.
2.6 Conclusion
The aim of this chapter has been to demonstrate the diversity of possible multilevel
converter topologies. Each has its own mixture of advantages and disadvantages
and for any one particular application, one topology will be more appropriate than
the others. Often, topologies are chosen based on what has gone before, even if that
topology may not be the best choice for the application. The advantages of the
body of research and familiarity within the engineering community may outweigh
other technical disadvantages.
Despite the diversity, these different topologies contain common underlying links.
28 Chapter 2. The Multilevel Converter
Usually the modulation and, to a lesser extent, control strategies can be developed
independently of the converter’s topology and then subsequently applied with little
or no modification. In subsequent chapters, the simplest case of the transformer
connected multi-bridge converter will be used as the implied default multilevel con-
verter topology. Required variations on modulation and control strategies will be
explained after the general technique has been presented.
Chapter 3
Almost all power electronic converters are operated in the “switched mode”. This
means the switches within the converter are always in either one of two states —
turned off (so no current flows), or saturated (turned on completely, with only a
small voltage drop across the switch). Any operation in the linear region, other
than for the unavoidable transition from conducting to non-conducting, incurs an
undesirable loss of efficiency and an unbearable rise in switch power dissipation.
To control the flow of power in the converter, the switches alternate between
these two states. This happens rapidly enough that the inductors and capacitors at
the input and output nodes of the converter average or filter the switched signal.
The switched component is attenuated and the desired DC or low frequency AC
component is retained. This process is called Pulse Width Modulation (PWM),
since the desired average value is controlled by modulating the width of the pulses.
These low pulse numbers place the greatest demands on effective modulation to
reduce the distortion as much as possible.
In this chapter, the various forms of modulation which are used for two-level
converters are considered, with emphasis on those suitable for a low pulse number.
Some will be shown to be better candidates for extension to multilevel converters
than others.
Two requirements which all low pulse number PWM candidates should observe are
synchronism with the fundamental frequency and quarter and half wave symmetry.
Synchronism with the fundamental frequency means ensuring the switching fre-
quency fc is an integer multiple of the synthesised fundamental frequency f1 . That
3.2. Review of PWM Techniques 31
is, the pulse number N = fc /f1 must be an exact integer. The frequency spectrum
of the PWM waveform will then consist of discrete frequencies at multiples of the
fundamental frequency nf1 , where n is an integer.
Quarter and half wave symmetry ensures that no even harmonics will exist in
the output spectrum [32]. This can be achieved by choosing N odd. An important
even harmonic which is eliminated is the DC component.
Beyond these basic requirements, there are many different ways of generating
PWM switching edges. Any technique can probably be placed into one of the fol-
lowing three categories:
• Hysteresis control
One further modulation technique, delta modulation, doesn’t fall into the cat-
egory of pulse width modulation, as the pulse widths aren’t modulated, but rather
are of constant width. The correct terminology for such techniques is Pulse Density
Modulation, PDM, or Pulse Frequency Modulation, PFM. Delta modulation finds
practical use where this constant pulse width is a given (such as for resonant con-
verters [74]), or an advantage (1-bit a/d and d/a converters [13, 4]). To achieve low
distortion, such converters require the switching frequency to be much higher than
the frequency content of the modulating waveform and won’t be considered further
here.
32 Chapter 3. Pulse Width Modulation Candidates for Multilevel
These calculations are slow and are done off-line. A look up table of edge angles
is created which the on-line controller uses to set the edge times.
Setting the lowest harmonics to zero can cause the first uncancelled harmonics
to be large. SHM minimises these harmonics, rather than eliminating them, accord-
ing to some cost function (such as harmonic loss in an induction motor) to give a
better overall result [22]. A summary of Harmonic elimination PWM techniques is
presented by Enjeti et al. [16].
200 200
100 100
-0 -0
-100 -100
-200 -200
0s 5ms 10ms 15ms 20ms 25ms 30ms 0s 5ms 10ms 15ms 20ms 25ms 30ms
Figure 3.1: Harmonic elimination PWM. The phase current in a three phase
inductive load is plotted along with the phase leg voltage (left) and load phase
neutral waveform (right).
PWM modulators to remove errors when they occur [32]. These techniques cannot
compensate for distortions due to DC bus ripple, or switching imperfections.
Recent work has been done on implementing regular sampled or on-line SHE-
PWM [8, 10, 38]. This work has shown that the positions of edges can be ap-
proximated relatively simply online given the modulation depth. The calculation
gives their displacement from the “sampling points”, the positions of the edges for
a modulation depth of zero.
Other work has presented alternative methods to solving the optimisation prob-
lem using for example linear block codes [3] and Walsh functions [61].
A hysteresis band modulator calculates the error between the desired output and
the measured output. The state of the switches is changed when this error exceeds a
certain bound (leaves the hysteresis band) so as to drive the error back within that
bound.
This method requires that the controlled output quantity of the converter is
integrated either by the load, or as part of the controller. For example, in a voltage
34 Chapter 3. Pulse Width Modulation Candidates for Multilevel
200
200
100
100
0 -0
-100
-100
-200
-200
0s 5ms 10ms 15ms 20ms 25ms 30ms 0s 5ms 10ms 15ms 20ms 25ms 30ms
Figure 3.2: Hysteretic PWM. The phase current in a three phase inductive load
is plotted along with the phase leg voltage (left) and load phase neutral waveform
(right). Note the lack of co-ordination between phases.
source hysteretic converter, the output current (the measured and subsequently
controlled quantity) will be integrated by an inductive load (Fig.3.2).
This technique has the advantage of bounded, predictable error and fast tran-
sient response to changes at either the input or the output. It is closed loop by
nature and demonstrates low distortion. It is simple to implement in its simplest
form. It has however a number of disadvantages, which limit its usefulness to low
power, high switching frequency applications.
One disadvantage is the variable nature of the switch period. Because of this,
the output spectra is continuous and spread to an extent, rather than discrete and
grouped as with carrier based techniques. Further, the switching instants are not
necessarily synchronous or cyclic and so sub-harmonics may be present. For these
reasons, hysteresis control is not applied for low switching frequencies.
For three phase modulators with an unconnected neutral, the independent hys-
teresis phase controllers ignore the interaction of the phases. There is no strategy to
generate zero voltage vectors, so at low modulation levels, the switching frequency
rises unnecessarily. Further, the modulator may lock into high frequency limit cycles
comprising only active vectors [12]. It is also possible for the line current error to
be double that of the hysteresis band.
By including extra comparators and/or logic, and knowledge of the load back
EMF, the modulator can make more intelligent switching decisions. Another alterna-
tive solution is to translate the three independent phase voltages into a synchronous
reference frame before being acted upon by the hysteresis controllers.
Another paper demonstrates two parallel connected boost converters with cou-
pled hysteresis control producing input ripple current cancellation [34]. With simple
cross coupling of the hysteresis feedback, the converters naturally synchronise and
operate out of phase as required.
sine-triangle PWM
fundamental fs
Hysteretic PWM
carrier fc
For the purposes of defining this broad category, carrier based PWM methods are
those where the switching decisions of the converter are made for each switching cycle
either at the beginning or during that switch cycle. That is, the PWM waveform
is calculated on a cycle by cycle basis, either pulse by pulse, or edge by edge.
This distinguishes it from SHE and SHM PWM, where multiple switching edges
are mapped out for the entire fundamental period or some fraction therein; and
hysteretic PWM, where neither edges nor switch period are defined, calculated or
even known in advance.
The carrier frequency is the same as the switch frequency. If the modulation were
reduced to zero or a DC quantity, then the PWM spectrum would consist of the
carrier and its harmonics alone and the component at zero frequency (DC) if present.
As the amplitude of the modulating waveform is increased, sidebands appear and
increase in amplitude either side of the carrier and its harmonics. As the frequency
of the modulating waveform is increased, the sidebands spread away from the central
carrier frequency. A more detailed analysis is undertaken in the next chapter.
If the same carrier signal is used to generate all three phase leg PWM signals
in a three phase inverter, the carrier spectral terms in the phase leg signals will
also be identical. Thus the carrier spectral terms (but not the carrier sidebands or
38 Chapter 3. Pulse Width Modulation Candidates for Multilevel
modulating terms) will be cancelled in the phase to phase waveforms. This is true
regardless of the pulse number N .
Va Vb Vc
Vtri
Va
Vb
Vc
Vab
Figure 3.4: Natural PWM in three phase inverter; phase leg and line to line
waveforms.
Although the phase relationship between the modulating and carrier waveforms
can be arbitary, it is suggested that the slopes of the triangular carrier and modu-
lating waveform, if sinusoidal in character, should be of the opposite polarity at the
coincident zero crossings, especially for low N [49]. This has practical implementa-
tion advantages of preserving the accuracy of the edges in analog implementations,
and easing the transistion between different pulse numbers in systems where this
may change during operation. Additionally this 180 degree phase difference (phase
relative to the carrier period) results in the minimisation of the harmonic losses in
an inductive load [11].
Clearly, this 180 degree out-of-phase relationship can only exist for odd N .
Further, the reduction in harmonic losses due to a specific phase relationship between
modulating function and carrier is only significant for odd N . To achieve this phase
3.3. Carrier Based PWM Techniques 39
relationship in a three phase inverter for all three phases requires N to be an odd
multiple of three (N = 3, 9, 15, 21, . . . ), if the same carrier is to be used for all
three phases to achieve carrier cancellation in the phase-phase output [49].
In a multi-level converter with an integer pulse number, only one carrier can
ever meet this requirement, as the other carriers are usually phase shifted relative
to it. However, if a non-integer synchronous pulse number (that is a mixed fraction
such as N = 21/4) is used in a multilevel converter, this phase relationship once
again becomes valid.
suitably scaled modulating signal is compared with this triangular carrier to produce
the switching edges.
The technique is so named because the signal is naturally sampled by the carrier
triangle waveform at the instant of the switching edge. Naturally sampled PWM can
react instantly to changes in input modulating command and produces no distortion
of the synthesised waveform.
Uniform
Natural
Figure 3.5: Uniform PWM generation. Note the phase delay in the PWM wave-
forms, when compared to the Natural PWM.
Uniform sampled PWM is a sampled data system. The modulating input signal
is sampled at the beginning of the switch cycle, before the actual switching edge
reflects this value later in the cycle. While this need for a sample and hold would
complicate an analog modulator implementation, sampling and holding the input
value is often intrinsic in digital implementations using timers and counters.
Space vector modulation is a form of uniform PWM used for the modulation
of three phase converters. A three phase system with no neutral connection only
has two independent controllable quantities, as the third is defined by the other two
according to Kirchoff’s laws. This simplification can be advantageously used in the
modulation strategy.
Instead of modulating each half bridge phase leg between its two states inde-
pendently of the others, the entire three phase bridge is treated as a whole. The
converter is switched between three or four of its eight (VSI) or nine (CSI) possible
states, chosen such that each phase leg switches once in each sub-cycle (half a switch
cycle). Using a three to two dimensional transformation, the desired output aver-
aged over the switch period and the converter states are represented as vectors. The
visualisation and calculation of switching periods is then performed using simple
vector math.
Space vector modulation can equally be applied to current source and matrix
converters [30], although voltage source is the most commonly seen. It can also be
extended to multilevel converters easily.
For synchronous PWM, the period of the carrier cycle should be a constant; an
exact sub-multiple of the fundamental synthesised waveform’s period. This may be
impossible to do in a microcontroller based system, where the triangular carrier is
generated by a digital counter. Should the counter be clocked by a fixed period
clock, the period of the carrier can only be integer multiples of this.
These errors could lead to non synchronous operation and thus sub-harmonics,
and non exact pulse edge positions too and so waveform distortion. One technique
to mitigate this problem is to accumulate this time quantisation error cycle by cycle
42 Chapter 3. Pulse Width Modulation Candidates for Multilevel
and use this error integral to create a correct average carrier period and pulse period
over a number of cycles [26]. Note that this may mean the carrier period alternates
between the quantised values either side of the ideal value. This somewhat “blurs”
the spectral term associated with the carrier frequency.
This concept can be deliberately extended and exploited. If the carrier frequency
is varied randomly about the mean carrier frequency, then the spectral terms associ-
ated with the carrier become continuous rather than discrete. The desired signal is
still faithfully reproduced, as the pulse width is always calculated to be the correct
fraction of the carrier period, even if that period is now itself varying, cycle by cycle.
A variation of this concept is to make the carrier period a variable for optimisa-
tion over the modulation period [31]. Within each carrier period, the pulse widths
are calculated as for normal space vector modulation, cycle by cycle, which ensures
the good dynamic response of carrier PWM is retained. The optimisation seeks to
produce a constant minimum ripple in the filtered output, to minimise the distortion.
This groups the switching edges towards the zero crossings, away from the crests —
a result similar to both hysteretic and SHE modulation. This modulation can be
constrained to be synchronous over the fundamental period, by pre-calculating the
desired period widths. However, this approach retains the good dynamic response of
a carrier technique, as each edge position is still determined online at the beginning
of each switch cycle.
3.4. Summary: The Best Modulation Option for Multilevel 43
Multilevel
Of all these different modulation techniques, this thesis will now focus on carrier
based PWM techniques and their application to multilevel converters. This is be-
cause carrier based PWM:
This chapter examines in detail naturally and uniformly sampled carrier pulse width
modulation (SPWM). These techniques are easily implemented, produce good re-
sults for moderate switch frequencies and are easily extended to multi-bridge con-
verters. In particular, the bandwidth and harmonics of these two techniques are
examined, both for single and multi-modulator converters.
As explained last chapter, for a carrier based PWM technique, the switching de-
cisions of the converter are made for each switching cycle either at the beginning
or during that switch cycle. That is, the PWM waveform is calculated on a cy-
cle by cycle basis, either pulse by pulse, or edge by edge. Using the terminology
of modulation theory, the carrier waveform, which has the same frequency as the
switch frequency, is modulated by the input waveform, to produce the desired PWM
waveform.
The aim of any pulse width modulator is to ensure that the averaged value over
the switch period of the output pulses is proportional to the value of the input sig-
nal at that time. This ensures a linear, distortion free transfer function. Firstly
considering a constant input signal, switching instants which satisfy this goal can
46 Chapter 4. Multilevel Carrier Based PWM
Figure 4.1: One switch cycle of carrier based pulse width modulation with single
trailing edge modulation (left) and double edge modulation (right).
a b c d
e f
Figure 4.2: All of the above carrier waveforms will produce a linear, distortion free
transfer function in a natural sampled pulse width modulator; because they are
constructed only with straight line segments. These different carriers do however
have different spectral qualities.
If the input signal is not constant or very nearly constant during a single carrier
(switch) period, then what value should the modulator assign to the pulse width?
In theory, the average value of the output pulse should be made equal to the average
value of the input signal during the total switch period; that is the areas under both
input and output signals should be equal over one switch period. Unfortunately this
requires the input signal to be known for the whole of the period ahead of time —
which may not always be possible. Distortion free reproduction is possible if a delay
is acceptable, however this is counter productive in some applications such as active
4.2. Naturally Sampled PWM 47
0 kπ π 2π ωt
0 kπ π 2π ωt
Figure 4.3: Natural sampled pulse width modulation, of pulse or carrier frequency
ω, shown for constant DC modulation k, 0 ≤ k ≤ 1 (top), and a fixed, sinusoidal
modulation k = 1/2 + M/2 cos(ω1 t) (bottom).
power filtering. In practice the technique known as natural sampling achieves the
same goal of a distortion-less transfer function, without any group delay.
In naturally sampled PWM (natural PWM), the switching instants are generated
by the intersection of the carrier waveform with the input waveform (Fig. 4.3).
The input waveform is naturally sampled by the carrier wave, the sampling instant
occurring at the same instant as the output edge (Fig. 4.4).
For a constant (DC) modulation input signal, the mathematical description based
on the Fourier decomposition of the PWM waveform is easily derived [70]. The DC
component is equal to the constant modulation input k, 0 ≤ k ≤ 1, and can be
included in the summation to give an alternative form.
m=∞
Vnat (ωc t) = k + (2/π) sin(k mπ)/m cos(mωc t)
m=1
m=∞
= (1/π) sin(k mπ)/m cos(mωc t) (4.1)
m=−∞
48 Chapter 4. Multilevel Carrier Based PWM
Equation 4.1 is expressed as a sum of sines and cosines; here the sines are zero as
the PWM waveform as shown is an even function. If the PWM waveform is not
even, but has some arbitrary phase displacement φc (in radians), then the Fourier
expansion is more clearly and compactly expressed as a sum of complex exponentials.
m=∞
Vnat (ωc t) = sin(k mπ)/mπ ejmφc ejmωc t (4.2)
m=−∞
The output PWM waveform is presented in this equation as the sum of a number
of Fourier terms. The first line is the desired output term of frequency ω1 ; the
second line, the carrier (ωc ) and its harmonics (mωc ); and the third and fourth lines
are the sidebands of the carrier and its harmonics (mωc ± nω1 ). Note that there
are no fundamental harmonic distortion terms of frequency nω1 . In the preceding
discussion, n and m are positive integers.
• the third term (second line) is simply a special case (q = 0) of the fourth term
(third line), and can be included by starting the summation in n (2q) from
zero.
• the fourth and fifth terms (third and fourth lines) can be combined, if it is
noted that when m is even (m = 2p), n is odd (n = 2q − 1), and visa versa.
The terms 2p, 2p − 1 and 2q, 2q − 1 can be replaced by m and n respectively,
if an extra term cos(m + n + 1)π/2 = − sin(m + n)π/2 is introduced to ensure
only terms for which m + n are odd are non-zero.
A more general representation which includes the phase of the carrier and fun-
damental waves is again better presented using the sum of exponential terms. A
derivation of this result can be found in Appendix A.
Vnat (ωc t, ω1 t) =
∞
∞
Jn (mM π/2) j(mφc +nφ1 ) j(mωc +nω1 )t
sin((m + n)π/2) e e (4.5)
m=−∞ n=−∞ mπ
50 Chapter 4. Multilevel Carrier Based PWM
Figure 4.4: The process of Natural (left) and Uniform (right) sampling. Uni-
form sampling introduces a delay of on average Tsam /2 which causes a sinc like
frequency response.
The DC and modulating terms are both included implicitly in the summation.
Their values are found by taking the limit as m = 0, n = 0 and m = 0, n = 1
respectively.
In uniformly or regularly sampled PWM (uniform PWM), the input signal is regu-
larly sampled and held constant at the beginning of each switch cycle before being
compared with the triangular carrier (Fig. 4.4). While this would complicate an
analog implementation, this sample and hold process occurs intrinsically in a digital
implementation, such as found in many microcontrollers. A complete discussion of
the implementation of natural and uniform sampled modulators is discussed in the
next chapter.
4.3. Uniformly Sampled PWM 51
The spectra of uniform sampled, double edge PWM is similar to that of natural
PWM, and is determined using the same techniques [6, 67].
∞
∞
Jn ((m + nω1 /ωc )M π/2)
Vuni (ωc t, ω1 t) = sin((m + n)π/2)
m=−∞ n=−∞ (m + nω1 /ωc )π
In the remainder of this chapter, comparisons and conclusions about the performance
of natural and uniform sampled PWM will be made based on an examination of these
equations for the output spectra of natural and uniform PWM, equations 4.5 & 4.6.
The reader is encouraged to refer back to these equations at these points.
The modulation of the transformer summed multiple bridge converter (Fig. 2.4)
is easily understood and a good starting point. Each bridge can be modulated
independently; its output will be described by the equations presented previously if
carrier PWM is used. The action of the transformer is simply to sum the outputs of
the bridges (Fig. 4.5). This is true not only of the time domain PWM waveforms,
but equally of the equivalent frequency domain spectral description.
The phase of the spectral components of natural and uniform sampled PWM
are nφ1 + mφc and nφ1 + mφc + (nw1 /wc )π/2 respectively. For both natural
52 Chapter 4. Multilevel Carrier Based PWM
Va
Figure 4.5: By using triangular carriers with specific phase relationships, a mul-
tilevel waveform is formed when the resultant PWM waveforms are summed.
and uniform PWM, changing the carrier phase φc does not affect the phase of the
modulator input signal term f1 , as m = 0. Regardless of the phase relationship of
the carriers used for the separate bridges, the input signal frequency components
will sum.
Changing the carrier phase does affect the phase of the carrier terms and their
sidebands since m = 0. In multi-modulator converters, choosing φci for each modu-
lator such that mφci = 0 will cause all carrier and sideband terms of order m to
sum to zero.
The time domain plots of four such carriers for a five level (four bridge) converter
are shown in Fig. 4.5. The initial triangular carrier is shown in bold, along with
the resultant natural PWM waveform when modulated by the sine wave shown
(N = 9,M = 0.9). Three additional carriers are generated by delaying this first
carrier by φc = π/2, π, and 3π/2 respectively, where 2π is the carrier period. Using
these carriers, three additional natural PWM waveforms can be generated, which
when summed by the multilevel converter structure result in the five level waveform
shown. In this multilevel waveform, the first three carrier terms and their sidebands
4.3. Uniformly Sampled PWM 53
48 Bridge Natural PWM, M = 0.9, N = fc/f1 = 2 48 Bridge Natural PWM, M = 0.9, N = fc/f1 = 2
0
25 10
1st, 0.900
20
15
10 −1
10
5
−5
−2
10
−10
31st, 0.0044
−15
29th, 0.0035
−3
−25 10
0.02 0.022 0.024 0.026 0.028 0.03 0.032 0.034 0.036 0.038 0.04 0 10 20 30 40 50 60 70 80 90 100
Figure 4.6: A simulation of the time domain (left) and frequency domain (right)
response of a 49 level (48 bridge) multilevel converter, modulated using natural
modulation, modulation depth M = 0.9.
have been cancelled. The first carrier-sideband group which appears in the output
spectra is grouped around 4 ∗ fc .
Many of the early simulations and practical results which are gathered in this
chapter apply to a three bridge (four level) converter. The phases of the triangular
carriers were −2π/3, 0 and 2π/3, so the phasor sum of the harmonic terms will
be zero for m = 1, 2, 4, 5, 7, 8, . . .. Only spectral components associated with m =
0, 3, 6, . . . appear in the PWM output. Although the switch frequency fc equals
450 Hz in these results, the lowest carrier and sidebands are centred around 3 * 450
= 1350 Hz (see Section 4.4.1).
It is possible to use carrier sets with relative phase shifts not equal to 2π/k, where
k is the number of carriers (converters). This will lead to incomplete cancellation
of carrier and sideband terms, but may be advantageous for other reasons, usually
related to the multilevel converter topology (see section 5.5.1) [24].
do extend down to approximately one quarter of that frequency, still greater than
1500 Hz. In any case, the amplitude of this carrier and these harmonics is also
very small, all less than one percent of the maximum modulation index. A time
domain and frequency domain plot for this example 48 bridge converter, modulated
using natural sampling with a modulation depth of M = 0.9, is shown in figure 4.6.
The frequency axis has been normalised relative to f1 (= 50/60 Hz), so that they
represent harmonics of the line frequency. The amplitude axis is relative to the
maximum modulation depth of the modulator M = 1.0. The time domain plot
amplitude is relative to the voltage seen by each bridge E; the maximum output of
the entire converter is ±24E.
However such low pulse numbers can cause some unexpected difficulties, as
explained in subsequent sections, if inappropriate modulation techniques are applied.
48 Bridge Natural PWM, M = 0.9, N = fc/f1 = 2 48 Bridge Natural PWM, M = 0.9, N = fc/f1 = 2
0
25 10
20
15
10 −1
10
5
−5
−2
10
−10
−15
−20
−3
−25 10
0.02 0.022 0.024 0.026 0.028 0.03 0.032 0.034 0.036 0.038 0.04 0 10 20 30 40 50 60 70 80 90 100
Figure 4.7: The simulated time domain (left) and frequency domain (right) re-
sponse of a 49 level (48 bridge) multilevel converter, modulated using natural
modulation, modulation depth M = 0.9, as before. One percent uniformly dis-
tributed noise has been added to the switching edges, to simulate the inaccuracies
inherent in an analog implementation.
Consider for example the effect of adding noise to the PWM edges in the previous
naturally sampled, 49 level converter simulation (Fig. 4.7). The noise is uniformly
distributed with a maximum deviation that is one percent of the carrier period.
This is a fair approximation of the error caused by triangular carriers which have
a slightly incorrect amplitude, offset, or phase shift. The overall amplitude of the
harmonics has not increased dramatically, however a number of extra harmonics
have appeared, including some at lower frequencies. Later results gathered from
actual hardware circuits also show incomplete cancellation of carriers and sidebands
(Fig. 4.9).
• accuracy
• stability
• easy integration
56 Chapter 4. Multilevel Carrier Based PWM
8 bit B0..7
Binary 8
Counter B0..7
osc
B7 A0..7
= R Q
D0..7 8
A0..7 S
D Q D Q
load
Q
A four level naturally sampled modulator was prototyped using analog ICs.
Three triangle reference waves were generated by integrating square waves and then
these were compared with the modulating wave to generate the PWM signals. These
three PWM logic level signals were summed to give a four level output waveform.
Stable and accurate phase shifted clock signals were provided by an 80C196 Intel
microcontroller rather than discrete logic, for simplicity and flexibility. This circuit
was later extended to five and then seven levels. The natural sampling analog PWM
modulator’s circuit appears in appendix C.
A number of Matlab routines (‘.m’ files) have been written to calculate and
display the spectral components generated by single and multiple converters, both
naturally and uniformly sampled. The switching instants of the edges are calculated
directly for the uniformly sampled case and iteratively for the naturally sampled
case. Non ideal effects such as time quantisation, switching delays and dead-times
can also be then easily included at this point. Jump function analysis [5], a variation
of the Discrete Fourier Transform (DFT), is then used to calculate the exact Fourier
spectrum of these PWM waveforms. This avoids the problems of the FFT (Fast
58 Chapter 4. Multilevel Carrier Based PWM
0 0
10 10
-1 -1
10 10
Vm
Vm
-2 -2
10 10
-3 -3
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
freq freq
Figure 4.9: These two spectra compare the predicted and measured spectra of
naturally sampled 2 (left) and 4 (right) level converters. M = 0.9, N = 9 (f1
= 50 Hz, fc = 450 Hz)
0 0
10 10
-1 -1
10 10
Vm
Vm
-2 -2
10 10
-3 -3
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
freq freq
Figure 4.10: These two spectra compare the predicted and measured spectra of
uniformly sampled 2 (left) and 4 (right) level converters. M = 0.9, N = 9 (f1
= 50 Hz, fc = 450 Hz)
Fourier Transform) which result from the necessary padding and windowing of data.
Some of these Matlab scripts and a summary of jump function analysis appear in
Appendix A.
4.4.2 Results
The spectral results of the two hardware approaches were collected and calculated
with the Tektronix spectrum analyser. This spectral data was imported into Matlab
and plotted along with predicted values for comparison. In figures 4.9 & 4.10 ,
the measured data appears as a continuous (1024 point) spectrum. The harmonic
4.4. Modulator Implementation 59
terms appear as “spikes” in this continuous spectra, peaking at the predicted values,
indicated by horizontal bars.
The spectra of the conventional two level PWM waveforms are presented along-
side the four level PWM spectra, for both natural and uniform PWM. The cancella-
tion of the first and second carrier terms and their associated sidebands is apparent
as predicted.
(These two features refered to in the above paragraph are not actually visible
in the plot in figure 4.10, since the lower bound of the y-axis is 1e-3. The choice
was made here and elsewhere in this document to use identical axes where direct
comparisons are desirable. The y-axis range (here three orders of magnitude) was
chosen to show all significant information, and yet maintain resolution.)
The natural converter did not achieve the same order of cancellation (Vm ≈ 3e-3
at 450 Hz, M = 0.9 ). This is simply because of the analog nature of the circuitry
and the consequent inequalities between each of the three modulators. The results
shown were achieved after trimming the amplitudes of each integrator, as standard
tolerance components led to poor cancellation. In terms of precision, the digital
modulator has a distinct advantage over the analog modulator.
60 Chapter 4. Multilevel Carrier Based PWM
1
0
10
−1
10
0.5
−2
10
−3
10
−4
−0.5 10
0 1 2 3 4 5 6 7 8 9 10 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Figure 4.11: The first five Bessel functions of the first kind, J0 (x) . . . J4 (x) plotted
vs x. On the right, a logarithmic scale is used to clearly show their behaviour
for small values of x (the usual situation for PWM modulation). As x → 0,
J0 (x) → 1, and Jn=0 (x) → 0.
delay
High power converters, both conventional and multilevel, are being considered and
used for more demanding applications such as active power filters and servo motor
drives. The design of the control system for such applications is simplified if the
PWM modulator and power conversion stage together can simply be considered as
a linear amplifier with some fixed gain. This assumption that the PWM modulator
has a flat, linear, distortion free transfer function can be validated by examining the
previously derived mathematical descriptions of the PWM waveform.
The magnitude of the harmonic terms in the Fourier summation are described
by Bessel functions of the first kind; for natural sampling:
Jn (M mπ/2)
sin((m + n)π/2)
mπ/2
The function Jn (x) is the n-th order Bessel function. Both n and x can be
real quantities, however here n is always an integer. The first five Bessel functions
J0 (x) . . . J4 (x) are plotted against x in figure 4.11.
For x > n, the Bessel functions look qualitatively like sinusoidal waves whose
amplitudes decay as x−1/2 [18]. For x < n, the Bessel functions look qualitatively
like simple power laws, with the asymptotic form for 0 < x
n
1 1
Vn (x) ∼ ( x)n n ≥ 0
Γ(n + 1) 2
1 1 n
∼ ( x) n ≥ 0, n an integer (4.7)
n! 2
These Bessel functions describe the amplitude of the carrier and sidebands
(m = 1, 2, 3, . . .). They also define the magnitude, and hence attenuation, of
the modulating signal (m = 0, n = 1) and generation of harmonics of that signal
(m = 0, n = 3, 5, 7, . . .).
For natural sampling, the argument of the Bessel function is only a function of
m. For m = 0, the Bessel function becomes Jn (0), which is either one (n = 0) or
zero (n = 0). This means that the original modulating signal appears in the PWM
output spectra unattenuated, without any accompanying distortion components,
and without phase delay, regardless of the frequency or modulation depth of that
signal. The gain of the modulator / converter cascade is equal to the peak amplitude
of the output waveform divided by the peak amplitude of the triangular carrier.
Although not shown here, a further analysis with a pair of modulating signals
demonstrates that natural sampling produces no intermodulation distortion prod-
ucts either [67]. Intermodulation products are output distortion components which
62 Chapter 4. Multilevel Carrier Based PWM
0.9
0.8
0.7
0.6
1
0.8 0.5
200
0.6 0.4
150
0.4
0.3
0.2 100
0.2
0 50
0 0.1
500 fc
1000 0
1500 0
0 500 1000 1500
frequency frequency
Figure 4.12: This waterfall plot and its front elevation show the spectra of a nat-
urally sampled 4 level converter for different input frequencies. M = 0.9, fc =
450 Hz
0.9
0.8
0.7
0.6
1
0.5
0.8
300 0.4
0.6
250
0.3
0.4 200
are harmonically related to signals in the input signal, for example, sums and dif-
ferences. This is an important result if the converter’s intended application is as an
active power filter, cancelling a number of closely spaced harmonic terms.
For uniform sampling, an extra term nw1 /wc makes the argument for the Bessel
function non-zero even when m = 0. For the pulse numbers generally encountered
(N > 5), this additional term is small, and the uniform PWM spectra is very similar
to natural PWM. However as demonstrated, it is possible to effectively use low
pulse numbers (even N < 2) in multilevel converters. Under these circumstances,
previously insignificant differences between uniform and natural sampling become
apparent.
4.5. Input signal Attenuation, Distortion and Phase delay 63
48 Bridge Uniform PWM, M = 0.9, N = fc/f1 = 2 48 Bridge Uniform PWM, M = 0.9, N = fc/f1 = 2
0
25 10
1st, 0.8449
20
15
10 −1 3rd, 0.1262
10
5
5th, 0.0425
0
−5 7th, 0.0184
−2
10
−10
57th, 0.0078
−15
55th, 0.0029
−20
−3
−25 10
0.02 0.022 0.024 0.026 0.028 0.03 0.032 0.034 0.036 0.038 0.04 0 10 20 30 40 50 60 70 80 90 100
The transfer function of the modulating input signal starts at unity gain for DC
frequency, as for natural PWM, but rolls off as the ratio f1 /fc increases (Figs. 4.13
& 4.17). The effect is more pronounced for larger modulation depths. Accompany-
ing the roll off is a constant phase group delay, equal to one half of the sampling
period. This delay also becomes significant as the ratio f1 /fc increases, and should
be considered in wide bandwidth closed loop stability calculations. Odd harmonic
distortion terms of the fundamental modulating signal, nf1 are now also non-zero
for uniform modulation.
The waterfall plot of uniform PWM spectra at different pulse numbers shows
that the amplitude of all harmonic terms are dependant on the pulse number, fc /f1
(Fig. 4.13).
These three effects are generally insignificant in normal two level converters
where, except in very high power ratings, the pulse number is at least larger than
nine [49] . Indeed, they are rarely mentioned. However in multilevel converters
where the pulse number can be made very low, they can be not only significant, but
the limiting factor.
demonstrates all the previously mentioned problems (Fig. 4.14). The first uncan-
celled carrier group would appear centred around 48 ∗ 120 = 5760 Hz, with the
significant lower sidebands extending down to approximately half that frequency,
greater than 3000 Hz. The amplitude of these carrier harmonics is also very small.
This very low level of residual switching energy is very pleasing, and justifies the use
of multilevel modulation. However with uniform PWM, the amplitude of the desired
sinusoidal output is not the 0.9 as expected, but rather 0.84; and odd harmonic dis-
tortion is clearly evident. Also obvious in the time domain plot is the group delay
due to the sampling process of uniform PWM.
tions
The transfer functions of naturally and uniformly sampled PWM were experimen-
tally verified using the previously discussed analog and microcontroller based multi-
level modulators. These modulators were extended to generate seven-level signals to
achieve greater carrier cancellation, which aided in the gathering of these particular
results. The input signal to the modulator was a swept sine-wave (“chirp”) gener-
ated by the Tektronix 2630 spectrum analyser (Fig. 4.15). The transfer functions
were calculated by the spectrum analyser based on the ratio of the corresponding
frequency components in the FFTs of the input and output waveforms. This essen-
tially isolated the desired output signal from the surrounding carrier and sideband
harmonic terms in the PWM signal, and so no separate output filter was required.
Normally the output filter or load would determine the overall response of the con-
verter, however the filter is generally included as a separate block in control system
calculations.
Figure 4.15: The transfer function of the modulators was automatically evaluated
by the Tektronix spectrum analyser. It calculated the ratio of input and output
spectra of the system, when stimulated using a swept sine wave chirp. The time
domain plots are shown here.
gain is still unity. Usually a filter placed between the converter and the load or the
load itself, will place an upper limit on the frequency response of the system.
In the seven level modulator implementations used, the first uncancelled carrier
group was at six times the switching or carrier frequency (m = 6). The first signifi-
cant lower sideband term (m = 6, n = −1) and the modulating signal (m = 0, n = 1)
coincide when the modulating frequency is three times the carrier frequency (imply-
ing a fractional pulse number 1/N = f1 /fc = 3). This is the cause of the distortion
of the transfer function at this point.
As explained, natural sampling has a linear transfer function, that is, a flat fre-
quency response (Fig. 4.16, Fig. 4.12). The small signal bandwidth is theoretically
66 Chapter 4. Multilevel Carrier Based PWM
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
N = f1/fc
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 4.16: The transfer function (for the modulating component f1 ) of a natu-
ral PWM modulator plotted against the ratio of modulating signal frequency to
carrier (switch) frequency (f1 /fc = 1/N ); shown in theory (left), and as gathered
from a seven level analog implementation (right). Natural modulation introduces
no attenuation or phase delay, but is subject to a slew rate limit (dotted) if mul-
tiple pulses per switch cycle are to be avoided.
The distortion of the transfer function at 1/N = f1 /f c = 3 is due to the coinci-
dence of the modulating term and the first significant decending lower sideband
at the modulator output. These output terms constructively and destructively
interfere as the modulator input chirp sweeps past this specific frequency.
1 1
M = 0.1
0.2
0.8 0.8
0.3
0.6 0.6
0.4 0.4
1.0
0.2 0.2
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 4.17: The transfer function (as above) of uniform PWM modulation;
shown in theory (left), and as gathered from a seven level, microcontroller based
implementation (right). For Uniform modulation, the input-output attenuation
is a function of both modulation depth M and the ratio of modulating signal
frequency to carrier (switch) frequency (f1 /fc ).
unbounded. However, a limit is imposed on the input signal’s slew rate if multiple
switching edges per switch cycle are to be avoided. The slope of the input signal
4.5. Input signal Attenuation, Distortion and Phase delay 67
should not exceed the slope of the triangular carrier wave, by remaining within the
boundary M f1 < 2/π fc .
It should be noted that a limit on slew rate does not limit the bandwidth as
such, but rather the available bandwidth at a given amplitude. If higher frequency
components in the input signal occur at lower amplitudes – as is often the case
in active filters – then these can be reproduced without attenuation and without
violating the slew rate limit.
Further, if multiple switching edges and the resultant rise in switch frequency
can be tolerated, then high frequency input signals will continue to be faithfully syn-
thesised, without distortion or intermodulation products. For example, the resonant
circuit of a radio transmitter could be excited by a pulse width modulated switching
power amplifier rather than a linear amplifier. The PWM carrier frequency could
be lower than the modulating signal — the broadcast signal carrier. The PWM
carrier and associated harmonics and sidebands would be removed by the high Q of
the transmitter’s resonant circuits (Fig. 4.18). Of course, a multilevel converter will
considerably ease the demands on the resonant filter.
The harmonic terms in the output waveform which may interfere with the original
modulating signal are the lower sidebands of the first uncancelled carrier group,
found at frequencies mfc − nf1 , m + n odd. These may be difficult or impossible
to remove by filtering when their frequency is comparable or lower than that of the
original signal. This happens for those terms for which n is close in value to mN .
These unwanted harmonic terms are usually referred to as “sub-harmonics”, since
their frequency is less than that of the desired signal.
The impact of these terms depends upon their amplitude and frequency, and the
nature of the application. For loads with an inductive component, such as induction
motors and transformers, sub-harmonics of low (near zero) frequency can cause large
68 Chapter 4. Multilevel Carrier Based PWM
0.5
0
10
0.4
0.3
0.2
−1
0.1 10
−0.1
−0.2 −2
10
−0.3
−0.4
−0.5
−3
10
20 22 24 26 28 30 32 34 36 38 40 0 200 400 600 800 1000 1200 1400 1600 1800 2000
2
0
10
1.5
0.5 −1
10
−0.5
−2
10
−1
−1.5
−2
−3
10
20 22 24 26 28 30 32 34 36 38 40 0 200 400 600 800 1000 1200 1400 1600 1800 2000
8
0
10
6
2 −1
10
−2
−2
10
−4
−6
−8
−3
10
20 22 24 26 28 30 32 34 36 38 40 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 4.18: In theory, a pulse width modulated converter using natural sampling
could be used to excite a radio transmitter’s resonant circuits, even if the PWM
switching frequency was less than the transmitter’s carrier frequency. The high-Q
resonant circuits would filter the undistorted desired frequency component from
the other switching frequency terms.
As an example, the time and frequency domain plots of a two level, five level
and seventeen level multilevel converter are shown. The naturally sampled PWM
waveforms of switch frequency fc = 450kHz are modulated by a desired waveform
of frequency f1 = 590kHz. The maximum modulation depth chosen, M = 0.4,
would avoid the possibility of multiple switch edges per switch period at the
expense of underutilisation of the power converter.
4.5. Input signal Attenuation, Distortion and Phase delay 69
currents to flow, even if their amplitude is only a fraction of the rated voltage at
rated frequency. Unwanted harmonic terms may also excite harmful mechanical and
electrical resonances in connected loads.
The magnitude of the n-th sideband (lower or upper) of the first carrier term
Jn (M π/2)
equals 2 sin((n + 1)π/2) π/2
. The sidebands have their largest values for M =
1.0, and decrease rapidly as M decreases for increasing n (Fig. 4.11). Although the
sidebands decay quickly as n increases even for M = 1.0, it is generally recommended
that a reasonably large pulse number (say N > 21) should be chosen to ensure the
magnitude of the sub-harmonics is negligible [49]. This guideline does seem overly
conservative. For example, the sideband term which approaches zero frequency (the
subharmonic) as N drops below nine has an normalised amplitude of approximately
5e-5.
Two of the lower sidebands will actually coincide with the modulating sine-wave,
when n = mN − 1 and n = mN + 1, and cause an error in its desired amplitude
and possibly phase. For N = 5, M = 1.0, this is at worst less than two percent, and
improves rapidly for increasing N and decreasing M .
4.5.5 Overmodulation
If the modulating input signal exceeds the amplitude of the carrier waveform, the
output saturates, or clips, as a conventional linear amplifier would. If the converter’s
application relies on a linear transfer function (such as for active filtering), this
situation should be avoided due to the rapid rise in distortion.
In a three phase converter, any component which is a multiple of three times the
modulating frequency will appear in-phase in all three phases, and so cancel. Such a
component is called a zero sequence term. The addition of appropriate zero sequence
waveforms to the phase leg voltages can increase the phase to phase fundamental
component before overmodulation by 1.15, which is a worthwhile improvement [7].
Another important advantage of this technique is the potential for the reduction
of switching losses. Every π/6 sector, one of the three phase legs can be forced to not
switch by the correct choice of triplen waveform. This lowers the average switching
rate of the device to two thirds of the original value. However, the harmonic content
increases significantly, which may offset the advantages of this gain [28].
The addition of zero vector waveforms and the advantages to multilevel modulation
are discussed in section 6.2.
4.6 Conclusions
Multi modulator converters can achieve an effective increase in overall switch fre-
quency through the cancellation of the lowest order switch frequency terms. This
chapter has examined natural and uniform PWM modulation techniques to deter-
mine if this allows a similar effective bandwidth increase. The PWM output spectra
were calculated mathematically, simulated using Matlab, and collected from hard-
ware implementations of the modulators, and the results evaluated and compared.
Naturally sampled PWM does not attenuate or delay the synthesised fundamen-
tal, nor generate any distortion products of the fundamental. Restated another way,
the transfer function of a naturally sampled PWM modulator is a pure gain, and
places no limitations on the achievable bandwidth of the overall converter. If enough
harmonic cancellation is achieved using multilevel techniques and output filtering,
the modulating signal can even be higher than the switching frequency.
Lower sideband harmonics of the first non-zero carrier can also limit the large
signal bandwidth of both techniques. The most significant harmonics generally meet
the input signal when f1 /fc = 1, however their amplitude decreases as the number
of converters increase.
72 Chapter 4. Multilevel Carrier Based PWM
Chapter 5
5.1 Introduction
As discussed in the previous chapter, uniform sampled and space vector PWM are
sampled data systems. They sample the signal input at the beginning of the switch
cycle, before the actual switching edge reflects this value later in the cycle. This
delay in response is significant when the ratio of switch frequency to carrier frequency
(the pulse number) is small. It leads to a frequency response roll-off which obeys
a Bessel function, similar to the familiar sinc function roll-off for Pulse Amplitude
Modulation (PAM). Another unwanted effect of uniform PWM is odd harmonic
distortion of the synthesised waveform. The severity of these effects is a function of
the ratio of the modulating and carrier frequencies, f1 /fc . This ratio may approach
and pass unity in high power active filters (high f1 , low fc ), by which point these
effects have become significant and limiting.
• Switching instants are crystal accurate, at least at the signal level. It is possible
to compensate for the switching delays in the power stages. More importantly,
the switching instants are repeatable from module to module. This ensures
good cancellation of the switching frequency terms in the combined multilevel
output.
• A digital system can be distributed master slave style more easily than an
analog implementation, which is an advantage for a modular approach. Digital
signals are more easily shared among isolated modules.
5.2. Re-sampled Uniform 75
• Digital signals are more immune to noise than analog signals in a noisy high
power environment.
The re-sampling technique allows fast control loops to benefit large, low switch-
ing frequency converters. For example, Tzou [62] has implemented a fully digital
controller for DC-AC inversion using a DSP. The sampling rate of the current loop
is 15.36kHz (256 ∗ 50Hz), and the switching frequency is twice this, as the converter
is a small UPS. A larger converter with a switching frequency many times lower
could still use this high sample rate to advantage using digital natural PWM.
Figure 5.1: From left to right, the existing techniques of symmetric and asym-
metric uniform sampling; and an extension of the concept, re-sampled uniform
sampling with re-sampling ratios (rsr) of 2 and 4.
switch edge for a single modulator/converter, rather than per pulse; hence the factor
of two.
Uniform sampling already comes in two variants — symmetric, where the sample
is held for the complete carrier period (rsr = 1/2); and asymmetric, where a second
sample is taken halfway through the carrier period for the second switch edge (rsr
= 1). Asymmetric uniform sampling is the preferred method, since each switching
edge is the result of a new sample and leads to better performance [9]. For both of
these cases, the reference is held constant throughout the switch period.
However, for a re-sampling ratio greater that one, one or more of the samples
will occur part way through the switch sub-period (Fig. 5.1). If according to a mid
switch period sample, the PWM edge is still to occur, then its position is recalculated
based on this more recent sample (Fig. 5.2A). This algorithm will react to transients
at the sampling rate rather than at the switching rate and so exhibit lower delay
and wider bandwidth. On average, the group delay of the input signal is now 1/2fs
rather than 1/2fc .
The second problem is the possibility of generating more than one edge per
switching cycle. This happens when multiple intersections of the samples and the
5.3. Microcontroller Software Implementation 77
Vout
Vctrl
B D
Figure 5.2: Re-sampled Uniform PWM — by re-sampling the input during the
switch cycle instead of only at the beginning, a more accurate switching edge
position can be calculated (A). However it is now possible to miss an edge (B) or
generate multiple edges (C&D).
triangular carrier occur, either erroneously, because of the stepped nature of the
sampled waveform (Fig. 5.2C), or quite legitimately — true natural sampling would
have done the same (Fig. 5.2D).
The design of many microcontroller timer and PWM peripherals complicates or even
prevents the implementation of re-sampled uniform PWM. Double buffering and
FIFO (first-in, first-out) buffers, usually considered a feature, prevent the reloading
of the pulse width value. In these situations, the software must calculate and decide
whether the edge will occur in the current sample period, and only then load the edge
command and time. Further, an equality comparison between the pulse width value
78 Chapter 5. Digital Natural — Re-sampled Uniform
and the timer (rather than a greater-than comparison) also requires the software
to ensure edges are not missed. These complications incur a considerable software
overhead and limit the useful possible re-sampling ratio.
Figures 5.3, 5.4 & 5.5 were collected from the five level 80C196 microcontroller im-
plementation, operating with a carrier frequency of 450Hz. The five level waveforms
were created by the summation of the four two level PWM waveforms at the logic
level. The circuit configuration is shown in appendix C.
Figure 5.3 shows 50Hz and 250Hz sine waves of modulation depth M = 0.9 and
the resulting five level waveforms. Although the synthesised fundamental frequency
f1 = 250Hz is a significant fraction of the carrier and hence switch frequency fc =
5.3. Microcontroller Software Implementation 79
450Hz, a five level converter can produce a good approximation of the original. Note
that both of these waveforms were produced with an oversampling ratio of four.
The attenuation, distortion and delay inherent in uniform sampled PWM, and
then the improvement which can be achieved through re-sampling, are shown in
figure 5.4. The double edge (triangular carrier) uniform sampling converter sam-
ples at twice the switching frequency, 900Hz, which agrees with the average phase
delay seen in the fundamental of approximately 600µs (Ts /2). The odd harmonic
distortion is clearly visible.
The re-sampling ratio of four (fs = 3.6kHz) can be seen to reduce both delay
and distortion. The apparent delay of 200µs can be accounted for as 20µs for a/d
conversion, 44µs of delay deliberately introduced in the calculation of the edge times
to allow for computation time, and the average delay due to the sampling process of
140µs (Ts /2). The distortion is greatly reduced, but more interestingly, changes in
character. For simple uniform sampled, the distortion component is predominantly
third harmonic, the same as for two-level uniform sampled. The frequency of the
distortion component of the re-sampled uniform waveform is roughly 15 times f1 .
80 Chapter 5. Digital Natural — Re-sampled Uniform
The response to the sawtooth waveform shows the reduced delay, and hence
following error, and improved transient response due to re-sampling (Fig. 5.5). The
kink in the transient response is due to software latches (flags) which enforce only
one pulse per period. This was necessary due to the risk of overflowing the 80C196
timer FIFO. This demonstrates that strictly enforcing the switching frequency only
compromises the large signal transient response, or more correctly, the large signal
slew rate.
Because the PWM modulator was clocked from a crystal oscillator and not
phase locked to the input signal as it should be for truly synchronous PWM, the
PWM waveform slowly slipped with respect to the input waveform. This allowed
the oscilloscope to be used to average the waveform over 64 successive triggers.
The next approach documented here was a hardware implementation using a field
programmable gate array (FPGA) — the Altera FLEX series of SRAM based pro-
grammable logic.
A three phase, five level (twelve output) modulator was implemented in an Altera
5.4. Gate Array Implementation 81
FLEX 8820, initially with seven bit resolution, which was then extended to eleven bit
resolution (Fig. 5.7). The DSP interface to this PWM peripheral was through four
16 bit memory mapped registers, three for each of the PWM phase inputs, and the
fourth for configuration. A tap from the timer chain updated (latched) the new data
to the comparators synchronously. This tap position was programmable to enable
different re-sampling ratios to be evaluated. Similar circuitry (not shown) generated
the trigger signal for the ADC and subsequent DSP interrupt. Programmable dead
time (also omitted) was later implemented for each switch pair. A full set of circuits
for the seven bit implementation is given in appendix C.2.
The FPGA was part of a DSP (the TMS320C31) based controller board. As
well as the FPGA, other key components were external RAM and boot EPROM, a
serial port, four 12 bit analog to digital converters and a quad 8 bit digital to analog
converter.
Leading or trailing single edge pulse width modulation can be easily generated
digitally with an up or down counter, and a digital comparator. Double edge modu-
lation has previously been implemented with up-down counters. For this multilevel
implementation, a novel technique has avoided the need for multiple, synchronised
up-down counters. All the phase shifted double edge PWM outputs are generated
from a single up counter reference (Fig. 5.6).
Double edge PWM can be generated with a sawtooth carrier if the leading and
trailing edges are generated by two separate comparators. A sawtooth rather than
a triangular carrier only requires a unidirectional counter rather than the more
complex up-down counter. This sawtooth carrier can then be phase shifted trivially
by simply manipulating the most significant bits.
Because the two comparators are created using combinatorial logic, many of the
terms are common to both and so they do not consume many more resources than
a single comparator would. An up-down counter however is generally more complex
and resource hungry. The one counter is also common to all modulation and carrier
phases, which ensures that the entire modulator is synchronised at all times.
82 Chapter 5. Digital Natural — Re-sampled Uniform
F 0 1 2 3 4 5 6 7 8 9 A B C D E F 0 C
5 4 (B) 3 (C) 2 D (D)
(D<C) . (D>C)
C D
A B
5 4
3 2
2 12
12b Phase
Synch Shifter
Counter 12 2 X<Y
T10,11 DFF
T9-0 10 Clk
Clk
8
Mux Double edge
X>Y
T11-3 1of 8 digital comparator
3
D2-0
Double edge
A11-0
digital comparator
DFF
DFF
B11-0
C11-0
Double edge
digital comparator
Data15-0 16
D15-0
DFF
/wr 5-level PWM Double edge
/cs Generator. digital comparator
2-4 One of Three
Addr1,0 2
Figure 5.7: The block diagram of the programmable gate array implementation
of re-sampled uniform.
5.4. Gate Array Implementation 83
1 1
rsr=8
0.8 0.8
rsr=4
0.6 0.6
rsr=2
0.4 0.4
0.2 0.2
Uniform
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 5.8: The transfer functions for different re-sampling ratios are shown
for the 9 level EPROM re-sampled uniform modulator (left) and Flex digital
natural implementation (right) for modulation depth M = 1.0, plotted against
1/N = f1 /fc . The transfer function of the previously analysed 80C196 based
uniform modulator (conventional uniform - no re-sampling) is shown dashed for
comparison.
For the evaluation of the system, the DSP simply read the ADC result, scaled it, and
then wrote the result to the PWM peripheral. The Tektronix spectrum analyser was
again used to provide the stimulus and evaluate the transfer function. The transfer
function of this modulator was captured and plotted for a number of re-sampling
ratios for the modulation depth M = 1.0 (Fig. 5.8).
Using re-sampling is seen to improve the frequency response of the modulator, pro-
ducing a family of curves similar to those produced for uniform modulation for
different modulation depths (Fig. 4.17). As expected, less roll-off occurs for larger
re-sampling ratios. Specifically, figure 5.22 shows that every doubling of the re-
sampling ratio is equivalent to the improvement in bandwidth that occurs when the
modulation depth is halved for uniform sampling.
84 Chapter 5. Digital Natural — Re-sampled Uniform
1.4
1 0.033
1.2 0.067
0.10
1
0.8
0.20
0.8
0.6
0.6
0.4 0.30
0.2
0.4
0
0
0.2 0.2
0.4
0.6
0.5 0
0.8 1.5 1
2.5 2
1 3.5 3 0
4.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 5.9: These three dimensional waterfall plots show the transfer function
of the re-sampled uniform modulator (DSP-Flex) plotted against frequency (0 <
f1 /fc < 5) for different modulation depths (0 < M < 1). These are for a re-
sampling ratio of four.
1.4
1 0.03
1.2 0.07
0.10
0.13
1
0.8 0.17
0.8
0.20
0.6
0.6
0.4
0.2 0.4
0
0
0.2 0.2
0.4
0.6
0.5 0
0.8 1.5 1
2.5 2
1 3.5 3 0
4.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 5.10: Transfer function for differing modulation depths as before, but for
a re-sampling ratio of eight.
1.4
1 0.03
1.2
0.07
1
0.10
0.8
0.8
0.6 0.13
0.6
0.4
0.17
0.2 0.4
0 0.20
0
0.2 0.2
0.4
0.6
0
0.8 2
4
6
1 10
8 0
12 0 1 2 3 4 5 6 7 8 9 10
Figure 5.11: Transfer function for differing modulation depths, re-sampling ratio
of eight (as above). Note the different x-axis scale for 1/N
5.4. Gate Array Implementation 85
It was also expected that there would be less roll-off for lower modulation depths
for a given re-sampling ratio, as was the case for uniform sampling. To test this,
a family of 30 transfer function curves was gathered for this DSP-Flex modulator.
The modulation depth was varied from 0.033 to 1.0 while the re-sampling ratio was
held constant, first at four, then eight. It can be seen from the plots (Fig. 5.9,
Fig. 5.10) that the transfer function is relatively independent of modulation depth
until the modulating signal’s amplitude falls below 1/rsr.
At this point the roll-off improves for smaller values of M . The attenuation
curves then closely match the uniform sampled case. However the delay will be
greatly reduced. This is the expected result.
Here the equivalent carrier frequency wc has been increased five times, however, the
equivalent modulation depth M has also increased by the same amount so that the
argument of the Bessel function is the same for both. Thus, the transfer functions
for the cases M = 1.0, rsr = 1 (asymmetric uniform sampling) and M = 0.2, rsr
= 5 should be the same.
The delay and hence phase shift of the re-sampled modulator will however be
86 Chapter 5. Digital Natural — Re-sampled Uniform
Original
Alternative
P.O.V.
Figure 5.12: This alternative point of view of the creation of the multilevel wave-
form, suggests an alternative derivation of the mathematical description of the
multilevel PWM waveform.
approximately one fifth of the uniform sampled modulator, which will improve the
transient response and stability under feedback.
Further results
Later programming implemented a simple three phase five level induction motor
drive, which used sinusoidal plus third harmonic modulation and followed a constant
Volts per Hertz modulation strategy. A five level flying capacitor converter based
on MOSFETs was built and the complete system used to drive a small 0.5 kW three
phase induction motor. Some results collected from this converter are presented in
the next chapter.
5.5. EPROM Based Implementation 87
An EPROM look up table performs the digital comparison and also effectively
holds the triangular carrier waveform (Fig. 5.14). It is addressed by two values — the
modulating input data value, and a clock value generated by a digital synchronous
binary counter. This counter steps the EPROM through the comparison with the
triangular carrier waveform, overflowing at the carrier frequency (Fig. 5.13). This
allows a simple implementation of multiple comparators each with an appropriate
offset (phase shift). The EPROM also easily encodes both slopes of vtri , without
the need for an up-down counter.
The PWM input data value may change at any time, asynchronously if desired.
The output will immediately reflect this change because the EPROM performs a
greater than comparison combinatorially. Although asynchronous sampling is pos-
sible, it will lead to distortion, unless the sampling rate is many times higher than
the switch frequency.
Three large (217 ∗ 8 = 1Mbit) EPROMs can implement three phase nine level
natural PWM with 8 bit resolution (Fig. 5.13). For higher carrier frequencies, this is
an acceptable resolution, especially when combined with a resolution enhancement
technique which accumulates and corrects for the truncation error [26].
ADC 8
D Q A9-16
go
ROM
8
D0-7 D Q
9
A0-8
Binary
Counter
(PLD)
osc
Counter input
012 ... DEF 012 ... DEF 012 ... DEF
D0 D1 etc... (D3:D2:D1:D0)
Figure 5.14: The structure of the contents of the PWM EPROM. This example
is compares a three bit PWM value with a 4bit counter value. Four outputs
suitable for a five level converter have been programmed.
which compares the upper bytes of the count and data values must generate an
equality output as well as a greater than comparison, to enable cascading.
The eight outputs of the EPROM can control eight pairs of switches to create
nine level PWM. Several EPROMs can also be placed in parallel to create more
than eight outputs for large converters, or for generating other synchronised control
signals such as commutation pulses.
It may at first seem possible to control eight switches individually (five level
PWM) to allow dead time to be programmed into the EPROM. However dead
time cannot be created purely combinatorially, but requires state information. For
example, an update of the input sample just after one output has switched could
5.5. EPROM Based Implementation 89
cause the second output to switch instantaneously, before the required dead time
has elapsed.
The contents of the EPROM were created using a MATLAB script (see sec-
tion B.1.3). This vector was saved as a .mat binary data file, with one byte per
element. The header was then removed, leaving a raw binary file suitable for an
EPROM programmer. A simple C program could have equally been used to create
the required EPROM file.
To date, only the case of equally phase shifted triangular carriers has been considered
for carrier based multilevel PWM. This in theory makes the best use of the multilevel
converter, achieving complete cancellation of a number of lower carrier groups. This
achieves the minimum possible ripple current and waveform distortion.
As explained in the previous chapter (see fig 4.2), any carrier composed of purely
straight line segments will produce a linear, distortion free transfer function. The
slopes of these lines, and hence the pulse period, may however vary, spreading the
spectral power of the carrier tones.
should be updated at the beginning of each of these periods to ensure the modulating
signal is reproduced undistorted. One of EPROM output bits can be used to latch
the new value at the (now random) beginning of the period.
A clipped carrier waveform will ensure that the duty cycle can never reach 0 or
100%. This guarantees a switch pulse every cycle, which is important if
• the power supplies for the switch drive circuits are bootstrapped.
• the rise and fall times are significant, causing unacceptable non-linearities in
the transfer function as the pulses are dropped at the extremes of modulation.
The full modulation depth can be used while still observing minimum on and
off times by pulse dropping a proportion of pulses at the extremes of modulation.
Carriers which can implement non-linear transfer functions such as signal compres-
sion at high amplitudes to give “soft clipping”, or to give higher gain to smaller
signals are easily and accurately encoded in an EPROM implementation. Most of
these steps could be easily performed in software or analog hardware, however as
explained, a digital implementation can have some advantages.
5.5. EPROM Based Implementation 91
APO PO PH
AA’
BB’
Vo
Vd
In some multilevel converter topologies, the individual PWM waveforms are summed
with the aid of passive components, for example inter-phase reactors or flying ca-
pacitors. The difference in phase between the PWM waveforms is carried by these
components. To reduce the size or ratings of these passive components, it may be
better to sacrifice the perfect spectral cancellation of carriers that can be achieved
with equal phase shift between triangular carriers, and instead lower the phase dif-
ference between PWM waveforms.
For example, reducing the phase shift to a fraction of its ideal value lowers the
required Volt-Second rating (i.e. iron content) of an inter-phase reactor in a parallel
connected multiple bridge converter. Similarly, the capacitance value of the flying
92 Chapter 5. Digital Natural — Re-sampled Uniform
capacitors in a flying capacitor converter can be reduced for a given voltage ripple
and load current by reducing the phase shift between carriers.
AA’
BB’
Vo
Vd
The actual circuitry used for gathering results consisted of a 4Mbit EPROM which
allowed 9bit PWM (see Appendix C.1). A 12bit, 8µs analog to digital converter
5.5. EPROM Based Implementation 93
(AD7870) was used to supply the digital input value. Only the 9 most significant
bits were used. The binary counter and decoding logic which starts the ADC and
then latches the result is implemented in an EPLD (erasable programmable logic
device). This allowed a number of re-sampling ratios to be evaluated.
All eight phase shifted outputs of the EPROM based hardware modulator were
used to create nine level waveforms in figures 5.15–5.20. The great advantage of
multilevel conversion can be seen from figure 5.15 where the nine level waveforms
and the two level PWM signals which form it can be compared. Again the switch
frequency fc is 450Hz.
The impact of re-sampling is again evident in figures 5.16 & 5.17 where the
improvement as the re-sampling ratio is increased from 4 to 64 is clearly seen.
The re-sampling ratio of 4 is the minimum sensible value for this hardware imple-
mentation, as all phase shifted carriers share the same sampled input value and
sampling instant. For ratios less than 4 for 8 carriers, some carriers will be up-
dated midway through their cycle, but not at the beginning. For example, for a
re-sampling ratio of two, four of the carriers will be updated at 0, 90 180 & 270
degrees and the other four at 45, 135, 225 & 315 degrees. This can lead to a rapid
deterioration in performance for some input signals. It also spoils the cancellation
of the carriers and sidebands in the frequency domain. For the above example, one
would expect the first carrier and sidebands to be centred around 8 ∗ fc , however
those around 4 ∗ fc are not properly cancelled.
Compare this to the microcontroller implementation for five level, uniform with
no re-sampling. Here the waveform is sampled only once at the beginning of each
switch period for each of the phase shifted carriers. So the sampling rate is doubled,
but the re-sampling ratio remains one.
In general, the re-sampling ratio should ideally always be chosen so that each
94 Chapter 5. Digital Natural — Re-sampled Uniform
rsr
1/
2
Figure 5.18: For an five level converter, only a re-sampling ratio of two (or
greater) makes sense, since this is the lowest value which updates the sample at
the beginning of every carrier waveform. The PWM waveforms which result from
re-sampling ratios of a half, one, and two are shown, for the EPROM implemen-
tation (simultaneous sampling).
carrier triangle edge is presented with a new sample at the beginning of its sub-cycle.
For an even number of carriers, nc (an odd number of levels), carriers always come
as nc /2 inverted pairs, with their vertices (beginning of their switch cycles) at the
same instants. The required re-sampling ratio is nc /2, and so the required sampling
frequency is nc /2 ∗ 2fc = nc ∗ fc . For an odd number of carriers, the re-sampling
ratio must be at least nc to satisfy this condition.
The excellent transient response which results from re-sampling is particularly evi-
dent in figure 5.17. Also evident is the possibility of producing multiple edges per
switch cycle, which may be unacceptable in some converters, even on a transient
96 Chapter 5. Digital Natural — Re-sampled Uniform
0
10
−1
10
−2
10
−3
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
10
−1
10
−2
10
−3
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
10
−1
10
−2
10
−3
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
10
−1
10
−2
10
−3
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 5.19: The PWM time domain waveforms (left) and frequency domain
spectra (right) for a five level EPROM based PWM modulator, shown for the
re-sampling ratios of 0.5, 1, 2, and 4. To ensure the complete cancellation of the
carrier and sidebands which are not multiples of the effective carrier frequency,
each pulse must be the result of a new sample - the equivalent of symmetric
sampling (fs = fc ). For a five level converter, this requires a re-sampling ratio of
at least two. A new sample for every edge (the multilevel equivalent of asymmetric
sampling) will of course produce a better result.
5.5. EPROM Based Implementation 97
basis. This may be avoided by latching the first edge for each cycle and ignoring any
subsequent edges in that switch cycle, as was done in software in the microcontroller
implementation. The ultimate solution is to limit the slew rate of the input signal,
which limits the power bandwidth, but not the bandwidth as such. For example, a
950Hz sine wave (the 19th harmonic in a 50Hz system, and over twice the 450Hz
switch frequency) can be faithfully reproduced by a nine level converter at 10%
modulation depth without producing multiple edges per switch cycle (Fig. 5.20).
To avoid multiple switching edges in each cycle, the analog or digital circuitry
preceding the modulator could include slew rate limiting. However this will com-
promise the transient response, one of the advantages of digital natural. A better
approach would be to have the digital or analog signal processing tuned so that in
the steady state, the modulating signal remains within the slew rate limit. Then
multiple edges are only generated in transient conditions. Alternatively for GTO
applications, monostables and latches could be placed after the ROM to ensure
minimum pulse widths.
In order to quantify the improvement in bandwidth achieved through the use of re-
sampling, the transfer function of the 9 level EPROM based re-sampled modulator
was measured using the spectrum analyser. The results were measured at two
specific modulation depths, M = 1.0 and M = 0.2, for several different re-sampling
ratios (Fig. 5.21).
98 Chapter 5. Digital Natural — Re-sampled Uniform
1 1
rsr=8
rsr=8
0.8 0.8
Uniform
rsr=4
0.6 0.6
rsr=4
rsr=2
0.4 0.4
rsr=2
0.2 0.2
Uniform
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 5.21: The transfer functions for different re-sampling ratios are shown
for the 9 level EPROM re-sampled uniform modulator, for modulation depths
M = 1.0 (left) and M = 0.2 (right). The transfer function of the previously anal-
ysed 80C196 based uniform modulator (conventional uniform - no re-sampling)
is shown dashed for comparison.
0.075
rsr=16
0.8
0.15
rsr=8
0.6
0.30
rsr=4
0.4
0.60
rsr=2
0.2
Uniform
0
0 1 2 3 4 5 6 7 8 9 10
Figure 5.22: The transfer functions for different re-sampling ratios are shown
for the 9 level EPROM re-sampled uniform modulator, for modulation depths
M = 1.0. The fractional numbers indicate the equivalent 2 level modulation
transfer function.
5.5. EPROM Based Implementation 99
For the 9 level converter shown, the first uncancelled carrier and sidebands are
centred around eight times the switch frequency. The usable bandwidth in theory
extends to four times the switch frequency, at which point the modulating term
meets the first significant lower sideband harmonic. This point of conflict can be
seen as a burst of noise in the transfer function.
The lowest re-sampling ratio shown is two and is not a good choice for the 9
level converter assessed. As previously explained, four of the eight carriers are not
updated with new samples at the beginning of their switch cycle. This leads to poor
performance at higher frequencies and incomplete cancellation of the fourth carrier
harmonic and its sidebands. These sidebands can be seen to upset the transfer
function measurement at twice the carrier frequency.
For the modulating depth M = 1.0, using re-sampling is seen to improve the
frequency response of the modulator, producing a family of curves similar to those
produced for uniform modulation for different modulation depths (Fig. 4.17). As
expected, less roll-off occurs for larger re-sampling ratios. Specifically, figure 5.22
shows that every doubling of the re-sampling ratio is equivalent to the improvement
in bandwidth that occurs when the modulation depth is halved for uniform sampling.
It was also expected that there would be less roll-off for lower modulation depths
for a given re-sampling ratio, as was the case for uniform sampling. However, the
family of curves for modulation depth M = 0.2 is almost identical to those for
M = 1.0. Indeed, uniform modulation (dotted) performs better than re-sampled
uniform at low re-sampling ratios.
To confirm this, another family of 30 transfer function curves was gathered. This
time the modulation depth was varied from 0.033 to 1.0 while the re-sampling ratio
was held constant, first at four, then eight. It can be seen from the plots (Fig. 5.23,
Fig. 5.24) that the transfer function is relatively independent of modulation depth
until the modulating signal’s amplitude falls below 1/rsr.
1.2
1
1
0.8
0.6 0.8
0.4
0.6
0.2
0 0.4
1
0.8 5
0.6 4 0.2
0.4 3
2
0.2 1
0
0 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 5.23: These three dimensional waterfall plots show the transfer function
of the re-sampled uniform modulator plotted against frequency (0 < f1 /fc < 5)
for different modulation depths (0 < M < 1). These are for a re-sampling ratio
of four.
1.2
1
1
0.8
0.6 0.8
0.4
0.6
0.2
0 0.4
1
0.8 6
0.6 5 0.2
4
0.4 3
0.2 2
1 0
0 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 5.24: Transfer function for differing modulation depths as above, but for
a re-sampling ratio of eight.
5.6. Conclusions and Applications 101
depths as seen in the gathered experimental data is the opposite of what this expla-
nation suggests. It appears these results are incorrect, as the identical set of results
gathered from the gate array implementation performed as the theory suggested.
A reason for the generation of these apparently incorrect results at low modulation
depths is yet to be established.
Natural sampled PWM is the best choice for applications which require closed loop,
wide bandwidth modulation such as active power filtering. It does not attenuate
or distort the modulating signal, even when the frequency of that signal is similar
to the switch frequency. Carrier based PWM is also easily adapted to multilevel
converter modulation by phase shifting the carriers.
Multilevel
Carrier pulse width modulation produces a pulse each switch period with a width
proportional to the desired modulating value. The actual switch period may vary
from cycle to cycle without distorting the reproduction of the modulating waveform
so long as this proportionality is maintained.
Now that the period is no longer constant, it would seem difficult to achieve
multilevel cancellation, as this relies on being able to create “harmonic cancelling”
carriers by phase shifting. The term “harmonic cancelling” carrier is used here
to mean a carrier which, as part of a set of carriers, causes groups of carrier and
104 Chapter 6. Random Carriers and Zero Vectors
Given a single triangular carrier whose period has been perturbed either side of
its average period, a second harmonic cancelling carrier can be created by simply
inverting this carrier. So generation of random PWM for a three-level converter is
trivial.
However, with the carrier period no longer constant, the meaning of carrier phase
is no longer clear. How can further harmonic cancelling carriers be generated? Is
there a simple technique for the creation of random PWM for the general N-level
converter?
First, a vector of random periods that had a uniform random distribution be-
tween 90 and 110% of the average period (256 counts) was created. This vector was
then rounded so that each period was an integer. A single triangular carrier cycle
was created for each of the periods in this vector, to form a triangular carrier with
Figure 6.1: The creation of two random period vectors by Matlab. In the case on
the left, every nine (since N = 9) carrier periods sum exactly to the modulating
period to ensure the zero crossings of the carrier and modulating waveform co-
incide. This “synchronous” carrier was compared to the purely random case on
the right to examine whether any additional sub-harmonics were suppressed.
6.1. Extending Random Modulation Techniques to Multilevel 105
constant amplitude, but with each cycle slightly different in period. This carrier
was compared with the modulating sine wave using naturally sampled PWM. Nat-
ural PWM was used to avoid clouding the results with the distortions attributable
to uniform PWM. It was reasoned that natural PWM could be approximated in
practice with the new re-sampled uniform techniques.
The carrier was asynchronous — no attempt was made to align the beginning
of the (N+1)th carrier period with the zero crossings of the modulating sine wave.
However the total period of the vector was corrected to avoid any problems in the
FFT. A “synchronous random carrier” (a contradiction in terms?!) was also tried.
Here the sum of each N carrier periods was adjusted to be equal to the modulation
waveforms period to ensure synchronous operation, if this has any meaning here.
That is, each time the modulating sine passed through zero, the carrier was also at
the beginning of a cycle. An example of two random period vectors created with
Matlab are shown in figure 6.1.
The second carrier was easily formed as explained earlier, by simply inverting
this carrier. This technique was valid and all simulations had the odd carrier groups
cancelled as expected. However, the goal was to create four harmonic cancelling
carriers for a five level converter, using a technique which could be easily generalised.
A method for creating further pairs of carriers which would cancel in a multilevel
converter was not obvious.
The first attempt was to form another random sequence, phase shift it by π/2
of the average carrier period, then form its inverted pair. This was unsuccessful
(Fig. 6.2). The first and all odd carriers and sidebands were cancelled as expected,
but the second, sixth, etc. were only slightly attenuated. The pulse polarity rule was
not always observed either. On examining the individual carriers it was noted that
although the periods only varied between 90 and 110 percent of the nominal value,
several consecutive shorter or longer periods caused the two carrier pairs to move in
phase with each other. It was concluded all random carriers should be related.
Based on this assumption, the second approach tried was simply to produce a
106 Chapter 6. Random Carriers and Zero Vectors
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.2: Two separate random carrier pairs only cancelled the odd carrier
groups; as either pair would have done on its own. (left: two-level spectra, and
right: five-level spectra). This is with pulse number N = 9 (i.e. fc = 450Hz if f1
= 50Hz), modulation depth M = 0.9, and a carrier period uniform deviation of
± 10%. The carrier and modulating waveforms were asynchronous.
second pair of harmonic cancelling carriers by shifting the first pair by π/2, based
on the average period. This produced the expected result of attenuating the second,
sixth, etc. carrier groups significantly, but not completely (Fig. 6.3).
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.3: Here the second pair of random carriers were generated simply by
delaying the first pair by π/2. This is partially successful. (Two and five level
spectra overlayed). Also shown is the difference between ensuring the PWM
is synchronous each fundamental period (left), and not bothering (right). Sub-
harmonics seem unaffected, as does the multilevel spectra, although the difference
between the single level spectra is clear. When overlayed, the shape and ampli-
tude of the curves are almost identical. (N = 9, M = 0.9, dev = ±10%)
The PWM output of the comparison of this waveform and the modulating signal is
correct in its own right and can be used to modulate a single pair of switches in a
phase-leg of a two level converter. A further m − 1 carriers may be generated each
separated by an equal 2π/m phase shift. The resulting PWM signals can be used to
switch further switch pairs. When summed by the multilevel converter structure, a
multilevel waveform is formed.
Holtz defines one carrier half period as a sub-cycle. Each of the three half
108 Chapter 6. Random Carriers and Zero Vectors
Original
Tc = 1 / fc
Random
A
(failed)
To = 1 / 8 fc
Alternative
P.O.V.
Random
B
(success)
bridges in a three phase converter will make one transition in a sub-cycle [7]. For
a five level converter, in one carrier period or cycle each of the four switch pairs
(PWM waveforms) in a phase leg will switch on once and off once, so in fact there
will be eight transitions instead of just two at that converter phase output. We can
generalise the definition of sub cycle to cover the multilevel case as being the period
during which each of the phase outputs makes one transition. Using this definition,
a sub cycle T0 = 1/8fc for a five level converter.
All that is necessary for distortion free modulation is linearity over each single
switch sub-cycle. Since the load averages or integrates the stepped output of the
converter, if the area under the converter’s output equals the area under the input
modulating signal over a given period, then the modulating signal will be reproduced
faithfully (Fig. 6.4). This is true for naturally sampled PWM over a sub-cycle, even
if the period of that sub-cycle is varied.
0
10
−1
10
−2
10
−3
10
−4
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.5: A successful demonstration of the cancellation of the first three carri-
ers and their sidebands achieved by varying the sub-cycle period T0 for all carriers
at the beginning of each sub-cycle. (N = 9, M = 0.9, dev = ± 10%, two and
five-level spectra overlayed)
110 Chapter 6. Random Carriers and Zero Vectors
The optimum randomisation of the period is the amount which just causes the
uncancelled spread carrier harmonic terms to just touch one another. For a nc + 1
level converter, the first uncancelled carriers will be at frequencies nc wc ± nw1 where
n is odd if nc is even and visa versa. The separation of these terms will thus be 2w1 ,
so the spread bands of energy will just touch one another when their bandwidth
is ±w1 , i.e. 2w1 . The bandwidth of these carrier terms is ±δ ∗ nc wc = 2δnc N w1 .
Thus the optimum uniform deviation δ equals 1/nc N , or in this case 1/(4 ∗ 9) or
approximately ±3%. The amount of randomisation of the period to completely blur
the spectrum is thus generally small and falls with both rising pulse number or
number of levels.
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.6: The two level (left) and five level (right) spectra for different mod-
ulation depths M and pulse numbers N , with a uniformly distributed random
period deviation of ±10%. From top to bottom,
N = 9 (i.e. fc /f1 =450/50Hz), M = 0.9 and M = 0.2;
N = 1.8 (i.e. fc /f1 =450/250Hz), M = 0.9 and M = 0.2.
112 Chapter 6. Random Carriers and Zero Vectors
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.7: The two level (left) and five level (right) spectra for modulation depth
M = 0.9 and pulse number N = 9 (i.e. fc /f1 =450/50Hz) , for different amounts
of period deviation. The switch periods averaged Tc = 1/fc = 2.22ms, with a
uniformly distributed random deviation of (from top to bottom) ± 2, 5, 10, and
20%.
6.1. Extending Random Modulation Techniques to Multilevel 113
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
−0.2 −0.2
−0.4 −0.4
−0.6 −0.6
−0.8 −0.8
−1 −1
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
−4 −4
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.8: The two level (left) and five level (right) spectra for modulation depth
M = 0.01 and pulse number N = 9 (i.e. fc /f1 =450/50Hz), with a uniformly
distributed random deviation of ± 10%. The advantages of a multilevel converter
for very small modulation depths is clear. However the usefulness of random
modulation at these low modulation depths is dubious.
zero. However, for very low modulation depths, the signal to noise ratio of the mod-
ulating signal is compromised by the noise floor of random modulation. Random
modulation techniques are possibly not suitable for applications such as audio am-
plification.
In the previous digital hardware implementation, the carrier triangles were generated
by a counter within the gate array. The carrier frequency is determined by the clock
frequency of this counter, which is generated by one of the DSP’s timers. At the
beginning of each sub-cycle, the gate array’s carrier counter generates an interrupt
to allow the modulation input to be updated synchronously by the DSP. During
this interrupt, the DSP’s timer period can also be updated, changing the PWM
114 Chapter 6. Random Carriers and Zero Vectors
clock frequency, so that the period of each sub-cycle is varied slightly. The clock
frequency remains constant during the sub-cycle (between interrupts), to ensure the
PWM output reflects the input without distortion, as discussed.
These results were again gathered using the Tektronix spectrum analyser. The
analyser performs a 4096 point FFT on the data with no (boxcar) windowing, and
averages eight successive FFTs.
The plots in figure 6.10 show the variation in response to different pulse numbers
and modulation depth. They are consistently lower in amplitude than their simu-
lation counterparts (Fig. 6.6), particularly the low pulse number, high modulation
depth example.
The hardware did show some harmonic distortion of the modulating term, par-
ticularly for large modulation depths. This appears unrelated to the random modu-
lation technique and is most likely due to the low re-sampling ratio of two. Similar
harmonic distortion is evident in figure 5.19, for a re-sampling ratio of two.
The effect of varying the carrier’s randomness is shown in figure 6.9. The largest
collected value of ± 10% appears to be the best choice, in agreement with the
simulations (Fig. 6.7). A larger deviation could not be tried, because the minimum
carrier sub-period (and hence period between interrupts) became shorter than the
required background processing time, causing the DSP to crash.
The five level flying capacitor MOSFET inverter was driven by the DSP-Flex con-
troller board, with random modulation signals.
6.1. Extending Random Modulation Techniques to Multilevel 115
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0
10 10
−1 −1
10 10
−2 −2
10 10
−3 −3
10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
10
Figure 6.10: (left column) Spectra of random period multilevel PWM gathered
from the DSP-FLEX hardware, for different pulse numbers N and modulation
depths M . Compare these to figure 6.6. From top to bottom,
N = 9 (i.e. fc /f1 =450/50Hz), M = 0.9 and M = 0.2;
N = 1.8 (i.e. fc /f1 =450/250Hz), M = 0.9 and M = 0.2.
116 Chapter 6. Random Carriers and Zero Vectors
0
10
−1
10
−2
10
−3
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
10
−1
10
−2
10
−3
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
10
−1
10
−2
10
−3
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
10
−1
10
−2
10
−3
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.11: Voltage and current waveforms (left) and voltage spectra (right)
gathered from one phase leg of the flying capacitor inverter for different modu-
lation depths M and pulse numbers N . The re-sampling ratio was four, and the
random period deviation ±5%. From top to bottom,
N = 9 (ie fc /f1 =450/50Hz), M = 0.83 and M = 0.2;
N = 1.8 (ie fc /f1 =450/250Hz), M = 0.83 and M = 0.2.
6.2. Multilevel Zero Sequence Waveforms 117
The voltage and current waveforms of one phase half bridge are shown for dif-
ferent modulation depths and pulse numbers in figure 6.11. For these tests, the DC
buss voltage was limited to 10V, and a 10Ω + 12mH (10+j4Ω at 50Hz) series RL
load was placed from the phase output to the DC buss midpoint.
The five level flying capacitor half bridge phase leg consisted of eight 60V 25A
rated Nch MOSFETs (ST P25N06) connected in series, each with its own isolated
gate driver [65]. Four 470µF 385V low ESR capacitors served as the three flying
capacitors and the DC bus capacitor. Two 3300µF split the DC bus, and created a
mid-point return for the RL load. A laboratory bench dual supply was used for the
power supply.
It can be seen that the small ripple present on the flying capacitors’ voltages does
not appreciably affect the voltage spectra. The levels of all unexpected harmonics
are below 1% for all cases.
By allowing the neutral point voltage to vary, one phase leg can be held contin-
uously high or low for a 60 degree interval while the other two switch. The correct
phase to phase waveforms are still formed. This has two significant advantages.
Firstly, the converter’s full potential modulation depth can be used since the phase
to phase voltages are maximised. Secondly, the switching losses are lowered, since
the average switching frequency falls to two thirds of its original value.
The zero sequence component present in space vector modulation can be anal-
118 Chapter 6. Random Carriers and Zero Vectors
For a two level converter, the zero sequence waveform is generated by finding the
phase which lies closest to the minimum or maximum modulation depth. The zero
sequence waveform takes a value which, when added to the phases, forces this phase
to the minimum or maximum modulation depth. Each 60 degrees (for sinewave
modulation), a different phase has the largest magnitude and is forced to the min-
imum or maximum modulation depth; the other phases stepping with it so that
there are no discontinuities in the phase-phase quantities. By holding one phase
at a modulation depth extreme, the maximum modulation depth without introduc-
ing distortion is assured. This technique will work regardless of the shape of the
modulating waveform.
The polarity consistency rule states that a given desired phase output value
should be generated by switching between only the two output levels either side of
that value. Restated for space vector modulation, a desired output vector should be
formed by switching between the nearest three adjacent vectors. A PWM waveform
which switches to a level or vector which is not adjacent is considered sub-optimal.
For a multilevel converter, holding one phase at this extreme value means that
that phase does not switch. The two phase-phase quantities associated with this
phase can then only be two level, and obey the polarity consistency rule. This
is not guaranteed for the third phase-phase quantity since it is the difference of
two two-level PWM waveforms. However in practice, it seems to be generally well
behaved.
The discontinuities in the zero sequence vector waveform introduce large triplen
6.2. Multilevel Zero Sequence Waveforms 119
2
1 1
0.6 0.6
1
0.4 0.4
0.5
0.2 0.2
0 0 0
−0.2 −0.2
−0.5
−0.4 −0.4
−1
−0.6 −0.6
−1 −1
−2
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Figure 6.12: Three phase sinusoidal modulation, M = 0.9, with a zero order
component added to force a duty cycle of 1.0 for 60 degree sectors. Shown are
the three phase waveforms and the zero sequence waveform before (left) and
after (centre) addition. The phase-phase quantities are given by the difference of
the three modulating waveforms (right). It can be seen that the zero sequence
waveform is cancelled.
2 2
1
0.6
1 1
0.4
0.5 0.5
0.2
0 0 0
−0.2
−0.5 −0.5
−0.4
−1 −1
−0.6
−1
−2 −2
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Figure 6.13: The five-level phase neutral PWM waveform (left) appears to show
some poor behaviour, but this is only a reflection of the discontinuities in the
modulating waveform. The phase-phase waveforms (centre and right) all obey
the pulse polarity consistency rule as a result of forcing one phase to remain at a
fixed level at any one time.
An asynchronous random carrier (dev = +- 10%) is used, with pulse number
N = 9. Natural sampling is used.
0 0 0
10 10 10
−1 −1 −1
10 10 10
−2 −2 −2
10 10 10
−3 −3 −3
10 10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.14: The desired fundamental component, as well as the harmonics due
to the triplen waveform are prominent in the frequency spectrum of the five-
level phase neutral PWM waveform (centre). The triplen frequency components
are not present in the phase-phase waveform as expected (right). The two-level
spectrum is shown (left) for comparison.
120 Chapter 6. Random Carriers and Zero Vectors
For a five-level converter, there are five such levels — +E, +E/2, 0, -E/2, -E.
The choices of level to hold at for sinusoidal modulation which minimise the step
discontinuities are:
M Action Comments
1.1547 2 tan 30 Force to +E, -E Maximum modulation depth
continuous zero seq waveform
0.8660 cos 30 Force to +E,-E/2 or +E/2,-E continuous zero seq waveform
0.7215 swap over point discontinuity in zero seq equal
0.5774 tan 30 Force to +E/2,-E/2 or equiv. continuous zero seq waveform
0.4330 cos 30/2 swap over point discontinuity in zero seq equal
0.2887 cos 30/3 Force to +E/2,0 or 0,-E/2 continuous zero seq waveform
0.1443 cos 30/4 swap over point discontinuity in zero seq equal
0.0 Force to 0 continuous zero seq waveform
Note that there is no swap over point between 1.15 and 0.866 — to swap before
reaching 0.866 would force the duty cycle of one of the other phases to exceed one.
An example of the three possible zero sequence vectors for a modulation depth
of M = 0.8 are shown in figure 6.15. For the modulation depth M = 0.8 for a
five level converter, it is possible to force the phase output to the extreme levels
E, −E (left), the midpoints E/2, −E/2 (centre), or a combination E/2, −E (right).
The combination E/2, −E produces the least discontinuities in the phase-neutral
waveforms, and so the least distortion in the phase-phase waveforms.
Five equally spaced levels are assumed for this analysis. Equally spaced lev-
els leads to equal voltage sharing between switches in most converter structures.
6.2. Multilevel Zero Sequence Waveforms 121
Switching instants can be evenly and naturally distributed using phase shifted car-
rier techniques such as those presented, since all phase leg voltage steps will be
equal. It is possible to have five levels which are not equally spaced. This would be
an unusual case. It would generally require the use of a space vector or otherwise
intellegent controller.
A continuous zero sequence waveform places less stress on the converter and
load and causes less high frequency electrical interference. This continuous triplen
component can either be maintained at a constant amplitude or reduced along with
the modulating waveform. Both techniques are assessed here.
A triplen waveform such as this can only be generated for a known sinewave modu-
lating signal. When the modulating signal is a complex signal, such as as found in
an active filter or a converter with closed loop control, a different real-time approach
is required. One successful approach is to “centre” the three phase switching tran-
sitions in the switch sub-period. Centring the pulses in this way not only ensures
the maximum possible modulation depth of the converter is achievable, but it also
results in superior harmonic performance of the converter [28].
Centring is achieved by finding the average of the most positive and most neg-
122 Chapter 6. Random Carriers and Zero Vectors
1 1 1
0 0 0
−1 −1 −1
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
1 1 1
0 0 0
−1 −1 −1
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
1 1 1
0 0 0
−1 −1 −1
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
1 1 1
0 0 0
−1 −1 −1
0 0 0
10 10 10
−1 −1 −1
10 10 10
−2 −2 −2
10 10 10
−3 −3 −3
10 10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 0 0
10 10 10
−1 −1 −1
10 10 10
−2 −2 −2
10 10 10
−3 −3 −3
10 10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.15: Three possible zero sequence waveforms for the modulation depth
M = 0.8. From the top, the modulating signals and zero sequence vector, the
modulating signals after the addition of the zero sequence vector, the phase neu-
tral waveforms, the phase phase waveforms, the spectra of the phase neutral, and
phase phase waveforms.
6.2. Multilevel Zero Sequence Waveforms 123
1 1
1.5
0.8 0.8
0.6 0.6 1
0.4 0.4
0.5
0.2 0.2
0 0 0
−0.2 −0.2
−0.5
−0.4 −0.4
−0.6 −0.6 −1
−0.8 −0.8
−1.5
−1 −1
−2
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Figure 6.16: The maximum possible distortionless three phase modulation depth,
M = 1.15, achieved by the addition of a zero order component to force a duty cy-
cle of 1.0 for 60 degree sectors. The three phase waveforms and the zero sequence
waveforms are shown before (left) and after (centre) addition. The phase-phase
quantities are given by the difference of the three modulating waveforms (right).
It can be seen that the zero sequence waveform is cancelled.
2 2
1
0.6
1 1
0.4
0.5 0.5
0.2
0 0 0
−0.2
−0.5 −0.5
−0.4
−1 −1
−0.6
−1
−2 −2
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Figure 6.17: The phase neutral PWM waveform (left) and the phase-phase wave-
forms (centre and right) for M = 1.15, N = 9. Natural sampling and random
modulation (uniform distribution, deviation of +- 10%) are used for this five level
example.
0 0 0
10 10 10
−1 −1 −1
10 10 10
−2 −2 −2
10 10 10
−3 −3 −3
10 10 10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 6.18: The desired fundamental component, as well as the harmonics due
to the triplen waveform are prominent in the frequency spectrum of the five-
level phase neutral PWM waveform (left). The triplen frequency components
are not present in the phase-phase waveform as expected (centre). The two-level
spectrum is shown (right) for comparison.
124 Chapter 6. Random Carriers and Zero Vectors
ative phase and using this signal as the zero sequence vector. When subtracted
from all three phases, it ensures the two outer phases are an equal distance from
the maximum possible modulation depth; or put another way, that the first and last
phase switch transitions are equal times from the beginning and end of the switch
cycle. This is easily done in software for a space vector modulator [15]. It has also
been successfully implemented in analog hardware [28].
All harmonics contribute equally to the THD figure regardless of their frequency.
A fairer measure would take into account the frequency of the harmonic as well as
amplitude, since many power system converters are connected via an LC filter or a
similar filter network to the power network. The Distortion Factor (DF) is a measure
of the harmonic distortion which remains after the harmonics have been subjected
to a second-order low-pass attenuation [58].
2
1 n
Vi
DF =
V1 i=2 i2
Weighted THD is used in this chapter for the analysis and comparison of spectral
performance of the different modulation techniques.
All of these formulae have been written with synchronous modulation in mind,
where all harmonics are multiples of the modulating term. For asynchronous or
random modulation, where this is not the case, the definition of weighted THD is
extended by using a continuous integral in place of the discrete summation.
2
1 f <f1 f =∞
f1
THD(weighted) = (Vi )2 + V1
V1 f =0 f >f1 f
To demonstrate this, the FFT of a square-wave voltage waveform and the result-
ing weighted THD for different numbers of FFT bins is tabulated below. Inclusion
of the first 15 harmonics (not including f1 , so f3 to f31 ) gave less than one percent
error. An accurate value of 12.1153 % was calculated with a 65536 point FFT.
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
0 0.2 0.4 0.6 0.8 1
Figure 6.19: Plots of weighted total harmonic distortion plotted against modu-
lation depth M , for different zero sequence waveforms. Four different cases are
shown:
– discontinuous modulation with five or three phase neutral levels (· · · ◦ · · ·),
– discontinuous modulation with four or two phase neutral levels (· · · ∗ · · ·),
– a continuous triplen component which allows maximum modulation depth and
which varies with M along with the sines (− ◦ − ),
– or remains fixed regardless of M (− ∗ − ).
The case of no zero sequence component is shown for comparison (solid).
57.6kHz, over 1000 times f1 , so there will be negligible error in the calculation of
the weighted THD. Random modulation with ±10% uniform carrier variation has
again been used. Natural sampling is used to ensure the modulation process itself
adds no additional distortion.
Referring to figure 6.19, the general trend for all plots is very similar. The
power of the carrier and sideband components remains relatively constant. Hence
the lowest distortion levels occur at high modulation depths, where the ratio of the
fundamental component to carrier and sideband components is largest.
The lowest overall average distortion is achieved with no zero sequence compo-
6.2. Multilevel Zero Sequence Waveforms 127
nent at all (Fig. 6.19, solid line). The other techniques do have features that may
make their implementation worthwhile despite their somewhat higher distortion.
All techniques show a minimum around M = 0.6 for this simulated five level
converter. At M = 0.5, the phase neutral PWM waveform will be a three level
waveform with the crests of the waveforms at the midpoint voltage levels of ±E/2.
The ripple current in an inductive load (and hence weighted THD) will be minimised
around this level [48].
The discontinuous zero sequence waveforms which form four and two level phase
neutral waveforms (Fig. 6.19, · · ·∗· · ·, Fig. 6.15, right column) will similarly produce
minima around M = 0.3 and M = 0.9 where the two and four level waveforms
respectively are near their relative maximum duty cycle, and hence producing the
minimum ripple.
A discontinuous technique can lower switching losses since the switches in each
phase leg need not switch for 120 degrees of each 360 degree cycle. However many
multilevel topologies cannot take advantage of this, the main feature of discontinuous
modulation, as they rely on continual modulation of all switches to maintain the
correct internal state of the converter. For example, the voltage balance on the
flying capacitors of the flying capacitor converter is inherently maintained by the
constant switching of the devices in each leg.
The use of a continuous triplen zero sequence vector component (Figs. 6.16 –
6.18), whether constant or varying in amplitude as M varies, leads to essentially
the same level of distortion at moderate modulation depths (Fig. 6.19, −−). They
extend the useful upper modulation range from M = 1.0 to M = 1.15, since plain
sinusoidal modulation begins to clip and distort beyond M = 1.0.
6.3 Conclusions
Random modulation and the addition of a three phase zero sequence vector are
two extensions of carrier modulation which have been already demonstrated for two
level modulation. This chapter has shown that both techniques can be extended to
multilevel modulation, and are equally applicable to space vector modulation.
Conclusion
Multilevel converter structures have been known for many years, although perhaps
not by that name. They were implemented using the series and parallel connection
of individual bridges by way of multiple secondary transformers and interphase reac-
tors [70]. The first modern multilevel converter was the neutral point clamped con-
verter, introduced in the early 1980s [51]. This topology created significant research
interest, particularly in the fields of traction and static power system converters. It
remains the most immediately identified multilevel converter. This decade, other
new multilevel converters topologies have been presented, such as the flying capac-
itor or imbricated cells converter [45, 46] and the cascaded inverters with separate
DC sources [56, 25].
Each of these different multilevel converter topologies has its own mixture of
advantages and disadvantages. For any one particular application, one topology
will be more appropriate than the others. Often topologies are chosen based on
historical precedent, or the designer’s preference, even if that topology may not be
the best choice for the application. A summary of the structure and operation of
each of these different multilevel topologies is given in chapter two, along with their
advantages and disadvantages.
130 Chapter 7. Conclusion
Usually the modulation strategy for each of these converters was developed in
conjunction with or subsequently to the converter. For some structures — most
notably neutral point clamped — not all switch states are allowed. The modulation
for the converter is treated as a whole, usually using space vector modulation. Other
techniques such as hysteretic or sine-triangle cannot be easily applied.
Despite the diversity, these different topologies contain common underlying links.
Usually the modulation and, to a lesser extent, control strategies can be developed
independently of the converter’s topology and then subsequently applied with little
or no modification.
In light of this, the simplest case of the transformer connected multi-bridge con-
verter was used as the implied default multilevel converter topology for the purposes
of developing good multilevel modulation and control techniques.
All of the well know modulation techniques for power converters such as se-
lective harmonic elimination, hysteretic and carrier based PWM modulation have
been extended to multilevel modulation. Of these different modulation techniques,
the focus was placed on carrier based PWM techniques and their application to
multilevel converters. This is because carrier based PWM
7.3. Natural and Uniform Sampled Carrier PWM 131
Multilevel converters place new demands on these techniques. Since carrier and
sideband groups and their multiples associated with the switch frequency can be
cancelled and removed from the modulation spectrum, a low switching frequency
can be chosen without compromising the output waveform quality. The multiple
switch edges due to the many switch pairs still promise a potentially wide modulator
bandwidth.
To ensure every switching edge influences the output with a minimum delay,
triangular carrier (double edge) asymmetric uniformly sampled or naturally sampled
PWM are the best choices. Digital or software modulator implementations are by
nature uniformly sampled. The accuracy of placement of the switch edges by a
digital implementation ensures an exact phase relationship between the switch pair
signals. This results in excellent carrier cancellation in the output spectrum.
However a low switch frequency leads to a small pulse number N , the ratio of
switch frequency to modulating frequency fc /f1 . Under these conditions, uniform
sampling exhibits the problems of modulating input signal attenuation, phase lag
and significant odd harmonic distortion. Multiple modulator (multilevel) converters
cannot hide these problems.
132 Chapter 7. Conclusion
Naturally sampled PWM does not attenuate the synthesised fundamental, nor
generate any distortion products of the fundamental. A naturally sampled modula-
tor does not affect the gain or phase response of the converter. Hence the converter
may be modelled and controlled as a linear amplifier. The passive output filter can
still of course have a significant effect on the closed loop response. The control loop
must often include knowledge of the filter states for stability.
ers
A carrier PWM technique can have its carrier frequency randomly perturbed about a
nominal switch frequency without disturbing the desired linear input output transfer
function. This moves the switching energy from discrete spectral lines and spreads it
into a continuous band of frequencies. This reduces the production of tonal acoustic
noise and reduces the problems of exciting electrical or mechanical resonances.
To extend this technique to multiple phase shifted carriers which are used to
form the multilevel waveform requires the carriers to all have equal random sub-
periods. Here a sub-period is defined as the period of one multilevel switch edge,
a submultiple of the two level PWM period. This ensures complete cancellation of
the lowest carrier and sideband spectral terms as before.
The techniques were again tested using simulations and confirmed in practice
with the FPGA re-sampled modulator. Only a relatively small period perturba-
tion is required to effectively spread the discrete spectral lines of the carrier and
sideband terms that remain, since this perturbation is large in comparison with the
frequency separation of these terms. For this small perturbation, sub-harmonics are
not generated, other than a very low level noise floor which exists from DC to the
first uncancelled sidebands.
134 Chapter 7. Conclusion
The benefits of the addition of a common mode “zero sequence vector” to the three
phase modulating waveforms was also assessed using simulations. The phase to
phase PWM waveform of a multilevel three phase converter switches between three
levels at any one instant rather than only two, since it is the difference of two two-
level phase neutral PWM waveforms at that moment. This was suggested to be
suboptimal according to the polarity consistency rule.
This thesis has concentrated on the theory of Pulse Width Modulation as applied
to multilevel converters. Evaluation of theories were carried out using simulations
and verified with analog, digital or microcontroller based modulator circuits. The
PWM control signals were analysed before their application to a power converter,
so the effects of finite power switch saturation, switching delay, and slew rate were
7.7. Future Research 135
not measured. Initial simulations showed that these effects should under normal cir-
cumstances, cause negligible distortion. An assessment of the importance of these
effects and the success of any countermeasures on the discussed multilevel modula-
tion techniques would be worthwhile.
Spectral Analysis
Assume that a waveform f (θ) is repetative or periodic over the interval 2π, so that
f (θ + 2π) = f (θ), and is defined between the values of −π and π. This waveform
can be written as the sum of a number of orthogonal sine and cosine components.
This series expansion is called the Fourier Series [36, 20].
∞
1
f (θ) = a0 + (an cos nθ + bn sin nθ) (A.1)
2 0
1π
an = f (θ) cos nθ dθ (n ≥ 0) (A.2)
π −π
1π
bn = f (θ) sin nθ dθ (n ≥ 1) (A.3)
π −π
0 kπ π 2π ωt
0 kπ π 2π ωt
Figure A.1: Natural sampled pulse width modulation, of pulse or carrier fre-
quency ω, shown for constant DC modulation k, 0 ≤ k ≤ 1 (top), and a fixed,
sinusoidal modulation k = 1/2 + M/2 sin(ω1 t) (bottom).
1
c0 = a0 ,
2
1
cn = (an − jbn )
2
1
c−n = (an + jbn ) n = 1, 2, 3, . . . (A.9)
2
a0 = 2c0 ,
an = cn + c−n
bn = j(cn + c−n ) n = 1, 2, 3, . . . (A.10)
Consider the double edge modulated naturally sampled PWM waveform with fixed
pulse width as shown in the top of figure A.1. The waveform steps between 0 and
1 in amplitude. The pulses are 2π periodic, with a rising edge at −kπ and a falling
A.2. Fourier analysis of double edge pulse width modulation 139
edge at kπ. The Fourier coefficients calculated below are for a more general case,
for a pulse 2kπ wide centred about a phase shift of φ radians.
1π
an = f (θ) cos nθ dθ (n ≥ 0)
π −π
1 kπ+φ
= 1 cos nθ dθ
π −kπ+φ
kπ+φ
1 1
= sin nθ
π n −kπ+φ
1
= (sin n(kπ + φ) − sin n(−kπ + φ))
nπ
2
= sin nk cos nφ (A.11)
nπ
Similarly,
2
bn = − sin nk sin nφ (A.12)
nπ
Hence
1
cn = (an − jbn )
2
1
= sin nk ejnφ (A.13)
nπ
∞
1
f (ωt) = sin nk ejnφ ejnωt (A.14)
n=−∞ nπ
Now assume that the width of the pulses is no longer constanst, but is allowed to
vary. The variable k becomes a function of time, k = k + M/2 sin(ω1 t + φ1 ). The
constant M is the modulation depth, and varies between 0 and 1.
1 π
cm = f (θ)e−jmθ dθ (n ≥ 0)
2π −π
1 k π+φc
= 1 e−jmθ dθ (A.15)
2π −k π+φc
140 Appendix A. Spectral Analysis
c
k π+φ
1 1 −jmθ
= − e
2π jm −k π+φc
j
= ejmk π − e−jmk π ejmφc (A.16)
2mπ
but
hence
j
cm = ejmkπ+jmM π/2 sin(ω1 t+φ1 ) − e−jmkπ−jmM π/2 sin(ω1 t+φ1 ) ejmφc
2mπ
j
= ejmkπ ejmM π/2 sin(ω1 t+φ1 ) − e−jmkπ e−jmM π/2 sin(ω1 t+φ1 ) ejmφc
2mπ ∞
j
= ejmφc ejmkπ Jn (mM π/2) ejn(ω1 t+φ1 )
2mπ n=−∞
∞
−jmkπ −jn(ω1 t+φ1 )
−e Jn (mM π/2) e (A.18)
n=−∞
where the following specific generating function of the Bessel function Jn (x) is sub-
stitued, [20]
∞
jx sin θ
e = Jn (x) ejnθ (A.19)
n=−∞
jn(ω1 t+φ1 ) −jmkπ
− J−n (mM π/2) e e
n=−∞
∞
j
= ejmφc Jn (mM π/2) ejn(ω1 t+φ1 ) ejmkπ
2mπ n=−∞
∞
− (−1)n Jn (mM π/2) ejn(ω1 t+φ1 ) e−jmkπ
n=−∞
j ∞
= ejmφc Jn (mM π/2) ejn(ω1 t+φ1 ) ejmkπ − (−1)n e−jmkπ (A.20)
2mπ n=−∞
∞
j
= ejmφc Jn (mM π/2) ejn(ω1 t+φ1 ) e−jnπ/2
2mπ n=−∞
× ejmkπ ejnπ/2 − e−jnπ/2 e−jmkπ
A.2. Fourier analysis of double edge pulse width modulation 141
∞
= Jn (mM π/2)/(mπ) ejn(ω1 t+φ1 ) e−jnπ/2 ejmφc
n=−∞
× − 1/2j ej(mk+n/2)π − e−j(mk+n/2)π
∞
= −Jn (mM π/2)/(mπ) ejn(ω1 t+φ1 ) e−jnπ/2 ejmφc sin(mk + n/2)π
n=−∞
∞
= − sin(mk + n/2)π Jn (mM π/2)/(mπ) ej(mφc −nπ/2) ejn(ω1 t+φ1 )
n=−∞
(A.21)
∞
∞
Jn (mM π/2) −jnπ/2 jn(ω1 t+φ1 ) jm(ωc t+φc )
f (t) = − sin((m + n)π/2) e e e
m=−∞ n=−∞ mπ
∞
∞
Jn (mM π/2) j(mφc +nφ1 −nπ/2) j(mωc +nω1 )t
= − sin((m + n)π/2) e e
m=−∞ n=−∞ mπ
(A.23)
Note that this result is for a modulating Sine wave. If a cosine wave is used
instead, the equation matches up with that derived by Wood.
Using similar techniques to those above, the Fourier series description of Uniformly
sampled PWM is
142 Appendix A. Spectral Analysis
∞
∞
Jn ((m + nω1 /ωc )M π/2)
f (t) = sin((m + n)π/2)
m=−∞ n=−∞ (m + nω1 /ωc )π
The analysis requires that the waveform can be expressed as the summation of a
number of step or jump functions. This is a natural fit for Pulse Width Modulation
(or any form of pulse modulation for that matter), where each switching edge can be
represented as a step function of a given amplitude, at a given time. It assumes that
the waveform to be analysed is periodic over all time, past and future. This period
is the lowest common multiple of all the periodic components present, which for syn-
chronous modulation, is the modulating period. Rather than the regularly sampled
amplitude vector required by a FFT, jump function analysis uses the shorter, exact
description of a vector of step time / step amplitude pairs.
A2 A2
A1 δ2 A1
Ak-1
A0 δ1 A0 δ1
δk
0 θ1 θ2 θ3 θk 2π θ1 θ2 θ3
Finding the complex Fourier coefficient cn for the general waveform above:
1 θ1 1 θ2 1 0
cn = A0 e−jnθ dθ + A1 e−jnθ dθ + . . . + A0 e−jnθ dθ
2π 0 2π θ1 2π θk
−1
= (A0 e−jnθ1 − A0 e−jn0 + A1 e−jnθ2 − A1 e−jnθ1 + . . . + A0 e−jn0 − A0 e−jnθk )
j2nπ
j
= ((A0 − A1 )ejnθ1 + (A1 − A2 )ejnθ2 + . . . + (Ak−1 − A0 )ejnθk )
2nπ
−j
= ((A1 − A0 )ejnθ1 + (A2 − A1 )ejnθ2 + . . . + (A0 − Ak−1 )ejnθk )
2nπ
j
cn = (δ1 e−jnθ1 + δ2 e−jnθ2 + . . . + δk e−jnθk ) (A.29)
2nπ
−j
= ((δ1 cos nθ1 + δ2 cos nθ2 + . . . + δk cos nθk )
2nπ
+j(δ1 sin nθ1 + δ2 sin nθ2 + . . . + δk sin nθk )) (A.30)
−j
= (δ1 cos nθ1 + δ2 cos nθ2 + . . . + δk cos nθk )
2nπ
1
+ (δ1 sin nθ1 + δ2 sin nθ2 + . . . + δk sin nθk ) (A.31)
2nπ
1
an = (δ1 sin nθ1 + δ2 sin nθ2 + . . . + δk sin nθk ) (A.32)
nπ
1
bn = (δ1 cos nθ1 + δ2 cos nθ2 + . . . + δk cos nθk ) (A.33)
nπ
δ1 = +1 δ2 = −1
A2
θ1 = 0 θ2 = π 2π
The amplitudes of the first three cos harmonics are zero. This is expected since the
given square wave function is odd (h(t) = −h(−t)), and so is composed of only sine
terms.
1 1
a1 = (1 sin 0 + (−1) sin π) = (0 + 0) = 0
π π
1 1
a2 = (1 sin 0 + (−1) sin 2π) = (0 + 0) = 0
2π 2π
144 Appendix A. Spectral Analysis
1 1
a3 = (1 sin 0 + (−1) sin 3π) = (0 + 0) = 0
3π 3π
1 1 2
b1 = (1 cos 0 + (−1) cos π) = (1 + 1) =
π π π
1 1
b2 = (1 cos 0 + (−1) cos 2π) = (1 − 1) = 0
2π 2π
1 1 2
b3 = (1 cos 0 + (−1) cos 3π) = (1 + 1) =
3π 3π 3π
Appendix B
These first four scripts are examples of the code used to perform the jump function
frequency analysis of the separately generated PWM waveforms. t2f.m performs
the actual time to frequency transformation; tplot.m and frplot.m plotted the
time and frequency vectors for checking and subsequent printing, and initt2f.m
created some necessary constants for the above files to run.
The steps (δ1 , δ2 , . . . , δk ) and angles (θ1 , θ2 , . . . , θk ) are two vectors of arbitrary
but equal length k. Since the data input will invariably be edge times rather than
angles, and the expected output will be frequency components rather than harmonic
numbers, the angles are replaced with times by using the relationship θ = ωr t. In
hindsight, this translation between angles and times, frequencies and harmonic num-
bers should have be kept separate from the jump function processing and plotting.
The frequency of periodicity, which will also be the frequency of the first (or
fundamental) harmonic is called fres , since this frequency is also the separation, or
resolution, of the harmonic terms which make up the discrete spectra.
It is quite permissible to generate multiple periods of times and edges. This will
generate multiple spectral ‘bins’, but many will be zero. If the times and edges are
not periodic over the period used, then this technique will generate ‘leakage’ errors
in the same way as a conventional Discrete Fourier Transform (DFT) calculation
would. The main value of the Jump function technique is the ability to generate the
exact spectra using the minimum calculations, if the stepped waveform’s period is
known.
145
146 Appendix B. Matlab Scripts and C Code
t2f.m
% t2f.m - takes two vectors of arbitrary but equal length:
% tlst - a vector of edge times
% ylst - a vector of edge steps (jumps)
% and generates:
% cn - a complex vector, length equal to vector n.
% abs(cn) - gives the magnitude and
% angle(cn) - gives the phase. (see fplot.m)
%
% run initt2f.m first (once) to define n, iwr, Ai_pi
initt2f.m
% initt2f.m - initialises the constants needed by t2f.m
% Needs to be run once at beginning of session, or after a clear,
% or to change fres or ftop.
wr = 2*pi*fres;
iwr = i * wr;
Ai_pi = A*i/pi;
tplot.m
% tplot.m Takes the tlst and ylst from hdn.m and does a time plot
% of them so I can see if they are really PWM
%
% First it converts the delta form of ylst to an absolute y vector yab
% Then it doubles up tlst and yab in such a way that plot will plot
% straight edges as appropriate for PWM (txa, tya)
txa = [tlst;tlst];
tya = [yabl;yab];
txa = txa(:);
tya = tya(:);
figure(1)
plot (txa,tya);
ax = axis;
% ax(1) = 0.0; % sets t_min of plot
ax(2) = 0.04; % sets t_max of plot
ax(3) = ax(3) - 0.2; % y_min
ax(4) = ax(4) + 0.2; % y_max
axis(ax)
B.1. Matlab Scripts 147
frplot.m
% fplot.m - plots the freq mag response
% cn - generated by t2f.m
% other consts - generated by initt2f.m
The next six Matlab scripts were used to generate the random modulation, zero
vector PWM results found in chapter 6. makedist.m called the following scripts
to from a distortion vector. rmod2.m creates the three phase modulating waveform
vector, including any zero vector; and rcarr.m creates the random triangular carrier
(run once at the outset). rcomp.m compares the modulating and carrier vectors to
produce the PWM vectors; rfft.m then performs a conventional fft on these to
create a frequency spectrum. wthd.m calculates the weighted harmonic distortion
of this spectrum.
makedist.m
% makedist.m
% calls other files repetitively to form a distortion vector dist
% files called are rmod2.m, rcomp.m, rfft.m, wthd.m
% see individual files for details.
% takes a few minutes to run.
tic
dist = [];
for M = 0.025:0.025:1.2;
% M = 0.025
rmod2
rcomp
rfft
y = wthd(fv3)
dist = [ dist; M y]
end
plot(dist(:,1),dist(:,2:4))
toc
rcarr.m
% rcarr.m produces just the random carrier. Follow with
148 Appendix B. Matlab Scripts and C Code
%
% based on
% r4c3ph.m produces natural PWM time amplitude vector which can be
% directly processed by fft routines. and added. And shifted.
% based on nat.m, but has a random PWM period.
% 4 -- aims to do this for 4 level
% c -- the cleaned up version
% 2 -- makes each subcycle random (2N), not just the cycle (N)
% 3 -- another try at quadrature - Got it!
% 3ph -- 3 phase version
%
% quadrature is achieved by allowing the slope of the triangle to
% change partway through the slope - here at zero. Not a problem for
% digital natural where the sample is updated here. Shouldn’t be a
% problem for uniform either.
N = 9; % Pulse number
dev = 0.1; % PWM period random deviation +- dev.
cycles = 10; N = N*cycles; % determines the fft resolution
figure(1)
plot(1:2*N,pwmper,[1 2*N],[pwmres pwmres])
% pause
rmod2.m
% rmod.m creates three modulating waveforms.
% run with rcarr.m, rcomp.m
% plotsines.m plots the modulating waveforms from r4c3ph.m
wtindex = 0:(2*N*pwmres-1);
wt = wtindex(:) ./ (2*N*pwmres);
% sineveca = M .* sin(2*pi*wt*cycles);
sineveca = sin(2*pi*wt*cycles);
sinevecb = sin((2*pi*wt*cycles)+ 2*pi/3 );
sinevecc = sin((2*pi*wt*cycles)+ 4*pi/3 );
sinevec = [sineveca sinevecb sinevecc];
figure(3);
xti = (1:2*N*pwmres/cycles);
xt = wt(xti)*cycles/50;
% plot(xt,sinevec(xti,1)); pause;
% plot(xt,zerovec(xti,1)); pause;
% plot(xt,sinevec2(xti,1)); pause;
% plot(xt,sinevec(xti,1),xt,sinevec(xti,2),xt,sinevec(xti,3),...
% xt,zerovec(xti,1));
% ax = axis; ax(3)=-1.2; ax(4)=+1.2; axis(ax);
% pause; % print -deps abcz115a1.eps
% plot(xt,phphmod(xti,1),xt,phphmod(xti,2),xt,phphmod(xti,3));
% ax = axis; ax(3)=-1.2; ax(4)=+1.2; axis(ax);
% pause; % print -deps abct115a1.eps
plot(xt,sinevec2(xti,1),xt,sinevec2(xti,2),xt,sinevec2(xti,3))
ax = axis; ax(3)=-1.2; ax(4)=+1.2; axis(ax);
% pause; print -deps pptm115a1.eps
rcomp.m
% rcomp.m compares the carrier and modulating waves to get PWM
% run after rcarr, rmod
bothr = 2.*bothr - 1;
bothf = 2.*bothf - 1;
bothrq = 2.*bothrq - 1;
bothfq = 2.*bothfq - 1;
allv = (bothr + bothf + bothrq + bothfq)/4;
phph = [ (allv(:,1)-allv(:,2)) (allv(:,2)-allv(:,3)) ...
(allv(:,3)-allv(:,1))];
figure(1)
xti = (1:2*N*pwmres/cycles);
xt = wt(xti)*cycles/50;
% plot(xt,allv(xti,1));
% ax = axis; ax(3)=ax(3)-0.1; ax(4)=ax(4)+0.1; axis(ax);
% pause; % print -deps pwma115a1.eps
plot(xt,phph(xti,1));
ax = axis; ax(3)=ax(3)-0.1; ax(4)=ax(4)+0.1; axis(ax);
% pause; % print -deps pwmab115a1.eps
% plot(xt,phph(xti,1),xt,phph(xti,2),xt,phph(xti,3));
% ax = axis; ax(3)=ax(3)-0.1; ax(4)=ax(4)+0.1; axis(ax);
150 Appendix B. Matlab Scripts and C Code
rfft.m
% rfft.m calcs the spectrum of the PWM waveforms
% run after rcarr rmod rcomp
figure(2)
fv1 = fft(bothf) ./ (N*pwmres);
fv2 = fft(allv) ./ (N*pwmres);
fv3 = fft(phph) ./ (N*pwmres);
xi = 1:40*cycles; xp = xi*50/cycles;
% semilogy(xp,abs(fv1(xi,1)),’-’,xp,abs(fv2(xi,1)),’-’);
% semilogy(xp,abs(fv1(xi,1))); grid;
% ax = axis; ax(3)=1e-3; ax(4)=2; axis(ax);
% pause; % print -deps f1a115a1.eps
% semilogy(xp,abs(fv2(xi,1))); grid;
ax = axis; ax(3)=1e-3; ax(4)=2; axis(ax);
% pause; % print -deps f4a115a1.eps
semilogy(xp,abs(fv3(xi,1))); grid;
ax = axis; ax(3)=1e-3; ax(4)=2; axis(ax);
% pause; % print -deps f4ab115a1.eps
wthd.m
function y = wthd(x,i)
% WTHD Weighted Total Harmonic Distortion y = wthd(x,i)
% Weighted wrt x(n), if i is specified, otherwise, wrt max(x).
% assumes x is the output of fft, hence removes dc component and
% reflection of negative freqs.
% for vectors, WTHD returns a single THD figure
% for matrices, WTHD returns a row vector containing the THD
% of each column.
[m,n] = size(x);
if m == 1 % handle isolated row vector
x = x’; % (this is not how they do it)
[m,n] = size(x);
end % now m rows (!= 1) and n columns
x2(i(ii),ii) = 0;
end
y = sqrt(sum((x2./mi).^2)) ./ x(i); % weighted
% y = sqrt(sum((x2).^2)) ./ x(i); % unweighted
These last two Matlab scripts were used to generate the binary files used for the
re-sampled uniform EPROM implementation EPROMs. A simple C program was
used to strip off the .mat file header. This left a simple binary file suitable for use
by a standard EPROM programmer.
pwmrom.m
% pwmrom.m Geoff Walker Fri Sep 9 1994
% this file creates a matrix which is a rom table for natural pwm
clear
pwm_bits = 9;
pwm_res = 2^pwm_bits;
addr_a = 2*pwm_res;
addr_b = 7*addr_a/8;
all1s = ones(pwm_res);
pwm1u = triu(all1s,0);
pwm1l = tril(all1s,0);
pwm1l(:,1:1:pwm_res) = pwm1l(:,pwm_res:-1:1);
pwm_all = pwm1;
for i = 1:1:7
pwmd = pwm1(addr_b+1:1:addr_a,:);
pwma = pwm1(1:1:addr_b,:);
pwm1 = [pwmd;pwma];
pwm_all = 2*pwm_all + pwm1;
end;
pwm_all_all = pwm_all(:);
pwmrom2.m
% pwmrom2.m Geoff Walker Fri Sep 9 1994
% this file creates a matrix which is a rom table for natural pwm
% outputs 0-3 are 4 * 90 degree phase shifted "greater than" outs,
% while 4-7 are corresponding "equal" outputs to allow cascading.
clear
pwm_bits = 3; % 7;
pwm_res = 2^pwm_bits;
addr_a = 2*pwm_res;
addr_b = 6*addr_a/8; % 270 degrees thru the table
all1s = ones(pwm_res);
pwm1u = triu(all1s,0); % the greater than matrix
pwm1l = tril(all1s,0);
152 Appendix B. Matlab Scripts and C Code
pwm1l(:,1:1:pwm_res) = pwm1l(:,pwm_res:-1:1);
pwm_all = pwm1;
for i = 1:1:3
pwmd = pwm1(addr_b+1:1:addr_a,:);
pwma = pwm1(1:1:addr_b,:);
pwm1 = [pwmd;pwma];
pwm_all = 2*pwm_all + pwm1;
end;
pwm1d = eye(pwm_res);
pwm1d2 = pwm1d;
pwm1d2(:,1:1:pwm_res) = pwm1d2(:,pwm_res:-1:1);
pwm1 = [pwm1d; pwm1d2]; % the equals matrix
pwm_all = 2*pwm_all + pwm1;
for i = 1:1:3
pwmd = pwm1(addr_b+1:1:addr_a,:);
pwma = pwm1(1:1:addr_b,:);
pwm1 = [pwmd;pwma];
pwm_all = 2*pwm_all + pwm1;
end;
pwm_all = pwm_all(:);
Four representative C programs written for the 80C196KB evaluation board are in-
cluded here. The first two, tria.c and upwmd.c generated the four level natural
and uniform PWM results of section 4.4.1. The tria.c program produced square
waves with the desired frequency and phase shifts, with crystal accuracy and sta-
bility. These square waves were each integrated to produce triangular carriers used
in the analog natural PWM circuit.
The second two programs, upwm4.c and upwm4o.c are five level uniform PWM
modulators. The upwm4o.c was an extention which includes re-sampling. The re-
sults of this modulator are shown in section 5.3.1
B.2.1 tria.c
/* tria.c96 - creates three phase shifted square waves for natural
* PWM circuit integrators, which create the triangular carriers.
* Every PERIOD = 2.22ms / 6 an interrupt serviced by HSO_event occurs.
* It also alternately sets/clears one of HSO.0,1,2 in sequence.
* ie set1, clear2, set3, clear1, set2, clear3, set1, ...
* Note that this simple method only works for an odd number.
* It sets up the next interrupt (caused by the next HSO edge).
* This program writes a moving bit to ioport1 LEDs in backgrd to show
* that process is still alive.
*/
#pragma code
#include <k:80c196.h>
B.2. 80C196 micro-controller code 153
main()
{
up = 1;
leds = 0x04;
count = 0;
pin_no = 0;
cmd = 0x10; /* setting this bit causes an HSO int */
void HSO_event()
{
ioport1 |= 0x01; /* set pin on entry to interrupt - debug */
if (++pin_no > 2) pin_no = 0;
cmd = cmd ^ 0x20; /* toggle set/clear */
edge_time += PERIOD ;
hso_command = cmd + pin_no; /* cause int + set/clear pin_no */
hso_time = edge_time; /* at time edge_time */
ioport1 &= 0xfe; /* clear pin as we leave interrupt */
}
B.2.2 upwmd.c
/* upwmd.c96 - four level (three output) uniform PWM modulator
* reads the a-d every SM_PERIOD
* then sets/clears one of HSO.0,1,2
*
* Written for 16MHz xtal, 1MHz timer clock.
* uses unsigned integers for time_offset
* this program writes a moving bit to ioport1 LEDs
* in backgrd to show that process is still alive
* include files are found in k: because of Novell map
*/
#pragma code
#include <k:80c196.h>
#define DELAY 16000
#define SM_PERIOD 370 /* sampling period, ie a/d interrupt period */
#define START_DELAY 50 /* to allow for calculations in interrupt */
#define SW_PERIOD 1110 /* = 3 * 370 = edge period of PWM wave */
#define SCALE 58 /* 0x8000 / 58d = 556 */
#pragma interrupt (ad_done = 1)
main()
{
up = 1;
leds = 0x02;
count = 0;
clear = 0;
pin_no = 0;
int_mask = 0x02;
int_pending = 0x00 ;
enable();
ad_command = 0x00;
sample_time = timer1 + 1000;
hso_command = 0x0f;
hso_time = sample_time;
start_time = sample_time + START_DELAY;
end_time = start_time + SW_PERIOD;
while(1){
for (count=0;count<DELAY;count++){
}
if (leds==0x80) up=0;
if (leds==0x02) up=1;
if (up==1) leds <<= 1;
else leds = (leds>>1);
ioport1 = leds;
}
}
void ad_done()
{
ioport1 |= 0x01;
ad_in = ad_result;
time_offset = (ad_in / SCALE);
pwm_control = ad_result_hi; /* for debugging */
sample_time += SM_PERIOD;
ad_command = 0x00;
hso_command = 0x0f;
hso_time = sample_time;
B.2.3 upwm4.c
#pragma code
#include <k:80c196.h>
#define DELAY 16000
#define SM_PERIOD 556
#define MID_DELAY 600
#define SCALE 59 /* 0x8000 / 59d = 556 */
#pragma interrupt (ad_done = 1)
main()
{
up = 1;
leds = 0x02;
count = 0;
pin_to_set = 0;
pin_to_clr = 2;
sample_time = timer1 + 1000; /* start in 1ms */
ad_command = 0x00;
hso_command = 0x0f;
hso_time = sample_time;
mid_time = sample_time + MID_DELAY;
int_mask = 0x02;
int_pending = 0x00 ;
enable();
while(1){
for (count=0;count<DELAY;count++){
/* waste some time */
}
if (leds==0x80) up=0;
if (leds==0x02) up=1;
if (up==1) leds <<= 1;
else leds = (leds>>1);
ioport1 = leds;
}
}
void ad_done()
{
ioport1 |= 0x01;
ad_in = ad_result;
data_in = 0x8000 - ad_in; /* conv to twos comp */
time_offset = (data_in / SCALE); /* can be pos or neg */
A/D samples
if pin_state[pin] is SET
(to be cleared)
pwm0_t
sama_t
sami_t
hso_time
CALC_D
sam_pwm_d edge_d_sam
PWM_P_2 time_in
edge_d_0 = clr_d
ad_done()
sample_time += SM_PERIOD;
ad_command = 0x00;
hso_command = 0x0f;
hso_time = sample_time;
mid_time = sample_time + MID_DELAY;
if (++pin_to_set > 0x03) pin_to_set = 0x00;
if (++pin_to_clr > 0x03) pin_to_clr = 0x00;
ioport1 &= 0xfe;
}
B.2.4 upwm4o.c
/* upwm4o.c - uniform PWM modulator -
* 4 - four output, ie five level
* o - with oversampling
* starts the a-d every SAM_PER -- some multiple of the switch period.
* Completion of this causes an interrupt, data is fetched, and next
* a-d conversion is scheduled.
* Also decides if there should be and edge for any of the four outputs
* before the next SAM_PER value is read. If so, then sets/clears any
* of HSO.0-3 as required.
*
* This program writes a moving bit to ioport1 LEDs
* in backgrd to show that process is still alive
* include files are found in k: because of Novell map
*/
#pragma code
#include <k:80c196.h>
#define TRUE 1
#define FALSE 0
#define DELAY 16000 /* for the background "I’m alive" loop */
/*
* The new figures are for 18.432M xtal, old for 16M.
* Both generate fc = 450Hz
* With 16.0MHz crystal, 450Hz = 2222.2us = 2222 counts approx.
* With 18.432M crystal, 450Hz = 2222.2us = 2560 counts exactly.
* Note that PWM_PER is the edge period, not pulse period.
* This SAM_PER (160) is the shortest power of two possible, since it must
* be greater than CALC_D to ensure the interrupt completes and exits
* before the next one is called.
*/
#define PWM_PER 1280 /* 1112 */
#define PWM_P_2 640 /* 556 */
#define SAM_PER 160 /* 139 */
/*
* This calculation delay is inserted between beginning of interrupt and
* the earliest possible edge to allow for calculation time of this
* edge position to ensure we don’t miss it.
*/
#define CALC_D 100
/*
* This is just short of 640, so a full scale a/d input causes
* an almost 100% duty cycle (630/640).
*/
/* #define SCALE 59 /* 0x8000 / 59d = 555d for 16M xtal */
#define SCALE 52 /* 0x8000 / 52d = 630d for 18.432M */
main()
{
up = 1;
leds = 0x02;
count = 0;
sam_no = 0;
pin_state[0] = SET;
pin_state[2] = CLR;
pin_state[1] = SET;
pin_state[3] = CLR;
sama_t = timer1 + 1000;
sami_t = sama_t + CALC_D;
ad_command = 0x00;
hso_command = 0x0f;
hso_time = sama_t;
pwm0_t[0] = sami_t;
pwm0_t[2] = sami_t;
pwm0_t[1] = sami_t - PWM_P_2;
158 Appendix B. Matlab Scripts and C Code
int_mask = 0x02;
int_pending = 0x00;
enable();
while(1){
for (count=0;count<DELAY;count++){
}
if (leds==0x80) up=0;
if (leds==0x02) up=1;
if (up==1) leds <<= 1;
else leds = (leds>>1);
ioport1 = leds;
}
}
void ad_done()
{
ioport1 |= 0x01; /* set pin 1.0 on entry for debugging and profiling */
ad_in = ad_result;
data_in = 0x8000 - ad_in; /* conv to twos comp */
time_in = (data_in / SCALE);
set_d = (PWM_P_2 - time_in);
clr_d = (PWM_P_2 + time_in);
for (pin=0; pin<PIN_NO; pin++) {
if (pin_done[pin] == FALSE) {
if(pin_state[pin] == CLR) /* pin is to be set */
edge_d_0 = set_d;
else
edge_d_0 = clr_d;
sam_pwm_d = (sami_t - pwm0_t[pin]);
edge_d_sam = edge_d_0 - sam_pwm_d;
/*
* This period is done. Now set up for the next one.
* schedule the next a/d conversion and subsequent interrupt
*/
sama_t += SAM_PER;
ad_command = 0x00;
hso_command = 0x0f;
hso_time = sama_t;
pin_done[0] = FALSE;
pin_done[2] = FALSE;
odd = TRUE;
}
}
ioport1 &= 0xfe; /* clear port pin 1.0 on exit */
}
B.3.1 hdn4si.c
/*
* hdn4si.c
* hdn -- Harmonics of Digital Natural
* 4 -- the five level version.
* s -- only a single edge per edge period allowed
* i -- integer based (previous versions were floating point)
*
* Calculates the time and delta vectors of a PWM waveform as
* required for jump function analysis, and saves these as a Matlab .mat
* file ready for loading into Matlab.
*
* Can have third harmonic added to f1 - comment in/out appropriate code.
*
* f1, T1 = fundamental synthesised freq, period. eg 50 Hz
* fc, Tc = _edge_ freq, per, ie 2* carrier freq. eg 900 Hz
* fs, Ts = sample freq, per, an integer multiple of fc.
* osr = oversampling ratio = fs / fc = Tc / Ts
* nfc = normalised carrier frequency = fc / f1 = T1 / Tc
*/
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <math.h>
#define TRUE 1
#define FALSE 0
#define RASIZE 20
typedef struct{
long int type;
long int mrows;
long int ncols;
long int imagf;
long int namelen;
} Fmatrix;
void write_matrix(Fmatrix *x, char *pname, double *pr, double *pi, FILE *fp);
Fmatrix tm;
mi = 9;
m = 0.9;
f1 = 50;
f1_ph_deg = 0;
osri = 12;
osr = 1;
for (i=0;i<osri;i++)
osr = osr * 2;
{
double tlst[itop];
double ylst[itop];
/*
* mtri = 4 / Tc = 2 / (Tc/2) = dy / dt
* mtri = (yx - y0) / (tx - t0) =>
* tx = t0 + (yx - y0) * one_mtri;
*/
w1 = 2.0 * M_PI * f1 ;
one_mtri = Tc / 2.0;
i = 0;
edge_delta = 1;
for (is=0; is<Tc; is+=Ts) {
/* printf("%g, %g, ",ic,is); */
/* yx = m * sin(w1*(ic+is+To)); */
/* yx = m * sin((w1*(ic+is+To))+f1_ph); */
yx = m*(1.15*sin((w1*(ic+is+To))+f1_ph)+0.21*sin(3.0*(w1*(ic+is+To)+f1_ph)));
tx = (yx + 1) * one_mtri;
if (tri_slope_pos)
tx = Tc - tx;
if (tx <= (is+Ts)) { /* edge this sample */
if (tx < is) /* edge missed */
tx = is;
tx += To + ic;
if (tx > Tres)
tx -= Tres;
tlst[i] = (double) (tx);
ylst[i] = (double) (edge_delta);
printf(".");
/* printf("%d, %g, %g, %d\n",i,tx,yx,edge_delta); */
i++;
break;
}
}
}
}
printf("\n done calculating - %d edges\n",i);
/*
* Now we write out the results in standard matlab .mat file format
*/
tm.type = 1000;
tm.mrows = 1;
tm.ncols = (itop);
tm.imagf = 0;
tm.namelen = 5;
filename = "temp";
printf (".mat filename? <%s> :",filename);
gets(rply);
if (rply[0] != NULL) {
filename = rply;
strcat (filename,".mat");
}
else
filename = "temp.mat";
/* printf("rply = %s\n",rply); */
printf(" Ok, filename = %s \n",filename);
if((fp = fopen (filename,"w")) == NULL) {
puts("error opening file");
exit;
}
fclose(fp);
}
return(0);
}/* main */
void write_matrix(Fmatrix *x, char *pname, double *pr, double *pi, FILE *fp)
{
int mn;
Circuits
The naturally sampled PWM modulator was initally wired as a four level circuit
using a slightly different circuit, with different components (most notably faster
LM319 comparitors). This circuit was used for the results gathered in figure 4.9.
The seven level circuit shown here was used later to gather the transfer function
results seen in figure 4.16. This circuit as shown, used slower LM339 comparitors to
give a more noise immune circuit, along with minor changes in topology to reduce
the circuits sensitivity to component tolerance.
The 80C196 microcontroller circuit was also tried as a four, five and seven level
modulator using different software. Only the appropriate number of outputs were
summed. Asymmetric uniform and re-sampled uniform were both evalutated using
this circuit. The master crystal frequency for the 80C196 was chosen to be 18.432
MHz as this was an exact multiple of the 450 Hz switching frequency that was used
for most simulations and experiments.
163
164 Appendix C. Circuits
C.1. Natural, 80C196 and EPROM modulators 165
166 Appendix C. Circuits
C.2. DSP-Flex FLEX8820 configuration 167
INPUT
CLK VCC
LPM_AVALUE=
LPM_SVALUE=
LPM_WIDTH=8
LPM_DFF
INPUT
TMR[7..0] VCC
INPUT data[]
CMP[7..0] VCC
INPUT
q[]
UPDATE VCC
OUTPUT
Q[3..0]
TITLE
1ph / 5lvl PWM Macro
COMPANY
UQ, UN
DESIGNER
G. Walker, E. LoGiudice
SIZE NUMBER REV
C 1.00 A
DATE SHEET OF
7:37p 9-17-1996 1 1
Name: Type: 250.0ns 500.0ns 750.0ns 1.0us 1.25us 1.5us 1.75us 2.0u
[I] TMR[11..9] INPUT 0 1 2 3 4 5 6 7
q[] OUTPUT
TMR7 p8b5ld@137
INPUT
p8b5ld@121 CLK_IN VCC
INPUT
/INT[3..1] VCC
INPUT
XF[1..0] VCC
INPUT
p8b5ld@11 /IACK VCC
INPUT
ADDR[7..2] VCC
OUTPUT
A[3..0]
OUTPUT
At[3..0]
OUTPUT
Ab[3..0]
OUTPUT
B[3..0]
INPUT OUTPUT
D[15..0] VCC Bt[3..0]
OUTPUT
Bb[3..0]
INPUT
p8b5ld@7 /FLEX VCC
INPUT
p8b5ld@8 R/W VCC
INPUT
ADDR[1..0] VCC
OUTPUT
C[3..0]
OUTPUT
Ct[3..0]
OUTPUT
Cb[3..0]
LPM_SIZE=8
LPM_WIDTH=1
LPM_WIDTHS=CEIL(LOG2(LPM_SIZE))
LPM_PIPELINE=
LPM_MUX
sel[]
LPM_SIZE=4
LPM_WIDTH=1
LPM_WIDTHS=CEIL(LOG2(LPM_SIZE))
LPM_PIPELINE=
SOFT LPM_MUX
INPUT
p8b5ld@118 EXT_CLK VCC
SOFT data[][] result[] OUTPUT
INPUT CLK_OUT p8b5ld@133
p8b5ld@12 TCLK0 VCC
SOFT
INPUT sel[]
p8b5ld@13 TCLK1 VCC
SOFT
INPUT
p8b5ld@17 H1 VCC
Appendix C. Circuits
NOT
OUTPUT
/AD_STROBE p8b5ld@194
TITLE
3Ph 5Lvl 8Bit PWM
COMPANY
UQ, UN
DESIGNER
p8b5ld@190 /AD_BUSY
INPUT
VCC
OUTPUT
/INT0 p8b5ld@205 G. Walker, E. LoGiudice
SIZE NUMBER REV
D 1.00 A
DATE SHEET OF
5:07p 3-16-1998 1 1
C.2. DSP-Flex FLEX8820 configuration 169
INPUT
CLK VCC
LPM_REPRESENTATION=
LPM_WIDTH=8
LPM_COMPARE
alb
INPUT
CMP[7..0] VCC dataa[]
INPUT F*
S0[7..0] VCC datab[]
OR2 DFF
PRN OUTPUT
D Q Q0
LPM_REPRESENTATION=
LPM_WIDTH=8
LPM_COMPARE CLRN
dataa[] agb
datab[]
LPM_REPRESENTATION=
LPM_WIDTH=8
LPM_COMPARE
alb
dataa[]
INPUT F*
S1[7..0] VCC
datab[]
OR2 DFF
PRN OUTPUT
D Q Q1
LPM_REPRESENTATION=
LPM_WIDTH=8
LPM_COMPARE CLRN
dataa[] agb
datab[]
LPM_REPRESENTATION=
LPM_WIDTH=8
LPM_COMPARE
alb
dataa[]
INPUT F*
S2[7..0] VCC
datab[]
OR2 DFF
PRN OUTPUT
D Q Q2
LPM_REPRESENTATION=
LPM_WIDTH=8
LPM_COMPARE CLRN
dataa[] agb
datab[]
LPM_REPRESENTATION=
LPM_WIDTH=8
LPM_COMPARE
alb
dataa[]
INPUT F*
S3[7..0] VCC
datab[]
OR2 DFF
PRN OUTPUT
D Q Q3
LPM_REPRESENTATION=
LPM_WIDTH=8
LPM_COMPARE CLRN
dataa[] agb
datab[]
TITLE
1ph / 3lvl PWM Macro
COMPANY
UQ, UN
DESIGNER
G. Walker, E. LoGiudice
SIZE NUMBER REV
DV 1.00 A
DATE SHEET OF
4:50p 9-30-1996 1 1
170 Appendix C. Circuits
LPM_SIZE=8
LPM_WIDTH=1
LPM_WIDTHS=CEIL(LOG2(LPM_SIZE))
LPM_MUX
AND2 DFF
data[][] result[]
PRN OUTPUT
D Q PWM_TOP
sel[]
INPUT CLRN
SEL[2..0] VCC
OR2 DFF
PRN OUTPUT
D Q PWM_BOT
CLRN
LPM_AVALUE=
LPM_SVALUE=
LPM_WIDTH=8
LPM_DFF
INPUT shiftin
PWM_IN VCC
shiften
1
GND
INPUT
q[]
CLK VCC
INPUT
CLK VCC
INPUT
SEL[2..0] VCC
INPUT
PWM[3..0] VCC
OUTPUT
PWMT[3..0]
OUTPUT
PWMB[3..0]
C.2. DSP-Flex FLEX8820 configuration 171
LPM_AVALUE=
LPM_SVALUE=
LPM_WIDTH=8
LPM_DFF
data[]
q[]
enable
LPM_AVALUE=
LPM_SVALUE=
LPM_WIDTH=8
LPM_DFF
data[]
q[]
enable
LPM_AVALUE=
LPM_SVALUE=
LPM_WIDTH=8
LPM_DFF
data[]
q[]
enable
LPM_AVALUE=
LPM_SVALUE=
LPM_WIDTH=8
LPM_DFF
data[]
q[]
enable
TITLE
DSP Interface and Buffers
COMPANY
UQ, UN
DESIGNER
G. Walker, E. LoGiudice
SIZE NUMBER REV
C 1.00 A
DATE SHEET OF
12:15p 3-14-1998 1 1
TITLE
Edge to Level Interrupt Line Conditioner
COMPANY
The University of Newcastle
DESIGNER
E. LoGiudice
SIZE NUMBER REV
C 1.00 A
DATE SHEET OF
2:58p 8-13-1996 1 1
172 Appendix C. Circuits
Bibliography
[3] Johnson A. Asumadu and Richard G. Hoft. New algorithms for generation of
notches in PWM waveforms using linear block codes. IEEE PESC’92, pages
465–472, 1992.
[6] Harold S. Black. Modulation theory. Van Nostrand, Princeton, N.J., 1953.
[7] Bimal K. Bose, editor. Power Electronics and Variable Frequency Drives. IEEE
Press, New York, 1997.
[10] Sidney R. Bowes. Advanced regular-sampled PWM control techniques for drives
and static power converters. IEEE IECON’93, pages 662–669, 1993.
[11] J. T. Boys and S. J. Walton. A loss minimised sinusoidal pwm inverter. IEE
Proc-B, 132(5):260–268, September 1985.
173
174 BIBLIOGRAPHY
[12] David M. Brod and Donald W. Novotny. Current control of VSI PWM inverters.
IEEE Trans Industry Applications, IA-21(4), May 1985.
[13] James C. Candy and Gabor C. Temes, editors. Oversampling delta-sigma data
converters : theory, design and simulation. IEEE Press, Piscataway, NJ, 1992.
[14] Giuseppe Carrara, Simone Gardella, Mario Marchesoni, Raffaele Salutari, and
Giuseppe Sciutto. A new multilevel PWM method: A theoretical analysis.
IEEE Trans. Power Electronics, 7(3):497–505, July 1992.
[15] P. Doulai and G. Ledwich. Microcontroller based strategy for reactive power
control and distortion compensation. IEE Proc B. Electric Power Applications,
137(6):364–372, November 1990.
[18] William H. Press ... [et al.]. Numerical recipes : the art of scientific computing.
Cambridge Cambridge University Press, 1986.
[19] Frank M. Flinders, Peter J. Wolfs, and Ken C. Kwong. Improved techniques
for switching power amplifiers. IEEE Trans. Power Electronics, 8(4):673–679,
October 1993.
[20] Gerald B. Folland. Fourier Analysis and its Applications. Brooks/Cole, Cali-
fornia, 1992.
[21] H. Fujita, S. Tominaga, and H. Akagi. Analysis and design of an advanced static
VAR compensator using quad-series voltage-source inverters. IEEE Industry
Apps Meeting, 3:2565–2572, 1995.
[22] G. A. Goodarzi and R. G. Hoft. GTO inverter optimal PWM waveform. 1987
IEEE Industry Applications Society Annual Meeting, 1:312–16, 1987.
[25] Peter W. Hammond. A new approach to enhance power quality for medium
voltage AC drives. IEEE Trans. Industry Applications, 33(1):202–208, January
1997.
[28] D. G. Holmes. The significance of zero space vector placement for carrier based
PWM schemes. IEEE Transactions on Industry Applications, 32(5):1122–1129,
September 1996.
[30] Donald G. Holmes. A unified modulation algorithm for voltage and current
source inverters, based on AC-AC matrix converter theory. IEEE Trans. In-
dustrial Applications, 28(1):31–40, January 1992.
[31] Joachim Holtz and Bernd Beyer. Optimal pulsewidth modulation for AC ser-
vos and low-cost industrial drives. IEEE Industry Applications Meeting 1992,
1:1010–1017, 1992.
[32] Joachim Holtz and Bernd Beyer. Optimal synchronous pulsewidth modula-
tion with a trajectory tracking scheme for high dynamic performance. IEEE
APEC’92, pages 147–154, 1992.
[33] Young-Seok Kim, Boem-Seok Seo, and Dong-Seok Hyun. A new N-level high
voltage inversion system. IEEE IECON 93, 2:1252–1257, November 1993.
[34] J. Kolar, G. Kamath, N. Mohan, and F. Zach. Self-adjusting input current rip-
ple cancellation of coupled parallel connected hysteresis-controlled boost power
factor correctors. PESC’95, 1:164–173, 1995.
[36] Erwin Kreyszig. Advanced Engineering Mathematics. John Wiley, New York,
1999.
[37] G. F. Ledwich and P. Doulai. Multiple converter performance and active filter-
ing. IEEE Trans. Power Electronics, 10(3):273–9, May 1995.
[40] L. Malesani and P. Tenti. A novel hysteresis control method for current-
controlled voltage-source PWM inverters with constant modulation frequency.
IEEE Trans. Industry Applications, 26(1):88–92, January 1990.
[41] L. Malesani, P. Tenti, E. Gaio, and R. Piovan. Improved current control tech-
nique of VSI PWM inverters with constant modulation frequency and extended
voltage range. IEEE Trans. Industry Applications, 27(2):365–369, March 1991.
[46] T. Meynard and H. Foch. Multi-level conversion: High voltage choppers and
voltage source inverters. IEEE PESC’92, pages 397–403, 1992.
[47] T. Meynard and H. Foch. Imbricated cells multi-level voltage source inverters
for high voltage applications. European Power Electronics Journal, 3(2):99–106,
June 1993.
[48] B. Miwa, D. Otten, and M. Schlecht. High efficiency power factor correction
using interleaving techniques. APEC’92, pages 557–568, 1992.
[50] B. Mwinyiwiwa, Z. Wolanski, and Boon-Teck Ooi. High power switch mode lin-
ear amplifiers for flexible AC transmission system. IEEE PES Winter Meeting,
January 1996.
[52] Guy Olivier, George-Emile April, Eloi Ngandui, and Carlos Guimaraes. Novel
transformer connection to improve current sharing in high current DC rectifiers.
IEEE Trans. Industry Applications, 13:127–133, jan 1995.
[53] Derek A. Paice. Power electronic converter harmonics : multipulse methods for
clean power. IEEE Press, Piscataway, NJ, 1996.
BIBLIOGRAPHY 177
[56] Fang Zheng Peng and Jih-Sheng Lai. Multilevel converters - A new breed of
power converters. IEEE Trans. Industry Apps, 32(3):509–517, May 1996.
[57] Fang Zheng Peng, Jih-Sheng Lai, John McKeever, and James VanCoevering.
A multilevel voltage-source inverter with separate DC sources for static VAR
generation. IEEE Industry Apps Meeting, 3:2541–2548, 1995.
[59] G. Sinha and T. A. Lipo. A four level rectifier inverter system for drive appli-
cations. IEEE Industry Apps Meeting, 2:980–7, 1996.
[61] Fred Swift and Adam Kamberis. A new walsh domain technique of harmonic
elimination and voltage control in pulse-width modulated inverters. IEEE
Trans. Power Electronics, 8(2):170–185, April 1993.
[62] Y Tzou. DSP-based fully digital control of a PWM DC-AC converter for AC
voltage regulation. PESC’95, 1:138–144, 1995.
[64] B. Velaerts, P. Mathys, E. Tatakis, and G. Bingen. A novel approach to the gen-
eration an optimisation of three-level PWM waveforms. In Power Electronics
Specialists Conference, PESC ’88, pages 1255–1262, 1988.
[66] L. Walker. 10-MW GTO converter for battery peaking service. IEEE Trans.
Industry Applications, 26(1):63–72, January 1990.
[67] P. Wolfs and G. Ledwich. New methods of spectral analysis for pulse width
modulators. ISSPA’87, pages 897–902, August 1987.
[69] Peter J Wolfs. High frequency link power conversion. Ph.D. thesis, University
of Queensland, 1992.
[70] Peter Wood. Switching Power Converters. Van Nostrand Reinhold, New York,
1981.
[72] Z. Zhang, J. Kuang, X. Wang, and B. Ooi. Force commutated HVDC and SVC
based on phase-shifted multi-converter modules. IEEE Trans. Power Delivery,
8(2):712–718, April 1993.
[73] Z. Zhang and B. Ooi. Multimodular current source SPWM converters for a
superconducting magnetic energy storage system. IEEE Trans. Power Elec-
tronics, 8(3):250–256, July 1993.
[74] P. D. Ziogas. The delta modulation technique in static PWM inverters. IEEE
Trans Industrial Applications, IA-17(2):199–204, March 1981.