© 2010 benjamin arthur niemoeller - ideals
TRANSCRIPT
© 2010 Benjamin Arthur Niemoeller
AN APPROACH TO CONTROLLING AN ACTIVE PARALLEL INTERFACE BETWEEN A TRANSIENT ENERGY SOURCE AND A
DC VOLTAGE BUS
BY
BENJAMIN ARTHUR NIEMOELLER
THESIS
Submitted in partial fulfillment of the requirements for the degree of Master of Science in Electrical and Computer Engineering
in the Graduate College of the University of Illinois at Urbana-Champaign, 2010
Urbana, Illinois
Adviser: Professor Philip T. Krein
ii
ABSTRACT
This thesis investigates controlling a dc-dc converter connecting a capacitor or
ultracapacitor (UC) to a dc voltage bus by regulating the series impedance of the con-
verter. Decoupling the capacitance from the voltage bus is shown to reduce the size
and/or weight of the power source. The impedance of the dc-dc converter is shaped by a
high-pass filter so that it is low at high frequencies, but infinite at dc. This shaped-
impedance controller allows load current to be shared between the dc source and decoup-
led capacitor, even when the dc source is uncontrolled, as long as the dc source has non-
zero source impedance. Charge transfer to and from the capacitor is limited to a finite
value proportional to a step change in bus voltage.
The same control approach was applied to the converter which added decoupled
bus capacitance to an electric vehicle power source, where the dc source is a battery con-
nected directly to the output voltage bus, and to a voltage regulator module (VRM) for a
desktop computer processor. Design equations and small- and large-signal models are
developed in detail for the vehicle power source. A second loop was added to the control-
ler to help regulate UC charge level. Small-signal system stability is verified across the
range of expected operating points. Experimental results of a scaled-down combined
source verify theoretical predictions. For the VRM, a multiphase buck converter is aug-
mented by a second set of phases which carry only fast current transients. Large-signal
models and switch dissipation analyses predict that the proposed topology eliminates the
need for bulk electrolytic capacitors on the voltage bus yet has efficiency comparable to a
conventional multiphase VRM.
iii
To Parents and Family
iv
ACKNOWLEDGMENTS
I am grateful for the love, support, understanding, and motivation provided by
Mom, Dad, Grammy, Grandpa, Gramma and all of my extended family on this long and
arduous journey. It made all the difference in terms of going back to school and complet-
ing the degree. I want to thank Professor Philip T. Krein, my adviser, for giving me the
opportunity to conduct this research project. His timely wisdom and preternatural pa-
tience are deeply appreciated. I also thank Joyce Mast for writing assistance and Josh
McNattin and Kevin Colravy for assistance in ordering and finding parts. I would like to
thank Dr. Santanu Kapat for insights into converter control early in the research process.
Thanks also to Sammy Nammari for his assembly help and for his resourcefulness in the
lab. Last but not least, I want to thank Ed Nykaza at CERL for funding my first year of
grad school and for general encouragement along the way.
Work on this thesis was funded by the Grainger Center for Electric Machinery
and Electromechanics.
v
TABLE OF CONTENTS
CHAPTER 1 INTRODUCTION.................................................................................................1 1.1 Combined Battery and Ultracapacitor Source for a Vehicle .......................3 1.2 VRM with Unbalanced Phases ....................................................................8
CHAPTER 2 DESIGN AND IMPLEMENTATION CONSIDERATIONS .........................................13 2.1 Vehicle Application ...................................................................................13 2.2 VRM Application.......................................................................................33
CHAPTER 3 SMALL-SIGNAL MODELS ................................................................................41 3.1 Small-Signal System Model with Three-State Plant .................................42 3.2 Small-Signal Model with Two-State Plant ................................................63 3.3 Stability Check...........................................................................................74 3.4 Summary ....................................................................................................74
CHAPTER 4 LARGE-SIGNAL MODELS ................................................................................76 4.1 Switched Model for Vehicle System .........................................................76 4.2 Nonlinear Averaged Model for Vehicle System........................................83 4.3 Switched Model for VRM .........................................................................88
CHAPTER 5 EXPERIMENTAL RESULTS ...............................................................................95 5.1 Scaling the Design .....................................................................................95 5.2 Hardware and Setup...................................................................................97 5.3 Battery Resistance Measurement.............................................................100 5.4 Step Response ..........................................................................................101 5.5 Repeated Draw Test.................................................................................104 5.6 Summary ..................................................................................................109
CHAPTER 6 CONCLUSIONS ..............................................................................................110
APPENDIX A SCRIPTS TO EVALUATE SMALL-SIGNAL TRANSFER FUNCTIONS ...............112 A.1 Main Script...............................................................................................112 A.2 Scripts for Two-State Transfer Functions................................................113 A.3 Scripts for Three-State Transfer Functions..............................................116 A.4 Helper Functions......................................................................................119
APPENDIX B EXPANDED LARGE-SIGNAL SIMULINK BLOCKS AND SET-UP CODE ...........120 B.1 Switched Nonlinear Model for Vehicle ...................................................120 B.2 Averaged Nonlinear Model for Vehicle ..................................................125 B.3 Switched Model for VRM .......................................................................130
APPENDIX C SCHEMATICS FOR BIDIRECTIONAL PCM BOOST CONVERTER...................143
vi
REFERENCES .....................................................................................................................150
1
CHAPTER 1
INTRODUCTION
A theoretically ideal voltage source has no series (source) impedance; it can sup-
ply all possible demanded currents, at any slew rate, and the voltage across source (and
load) terminals will never change. Real dc power sources cannot do this. They have a
limit on current slew rate which causes source impedance to rise with frequency. In addi-
tion, practical dc power sources have a current magnitude limit, independent of the con-
verter’s source impedance, beyond which the source will sustain damage. Adding a ca-
pacitor in parallel with the dc source raises the current slew rate limit and will also allow
the combined source to deliver more current to the load for a period of time. When the
capacitor is tied to bus voltage, though, only a small fraction of the capacitor’s stored
energy is available to the load. Decoupling the capacitor voltage from the bus voltage
makes more of this energy available, reducing the needed capacitance value. The decoup-
ling is performed by a switching dc-dc power converter that allows current to flow
through it in both directions. This connection is shown in Figure 1.1 and is called an ac-
tive parallel connection. In this figure the dc source is represented by an ideal voltage
source ocV with series resistance thR . The combination capacitor and dc-dc converter is
termed a decoupled capacitor.
A control topology which regulates the source impedance of the dc-dc converter
is applied to the decoupled capacitor. The source impedance is commanded to be low at
higher frequencies and infinite at dc, injecting transient current into the bus but limiting
the charge that is transferred from the capacitor in response to a line or load step. The
controller will inject power to the bus so long as the dc source supports bus voltage and
2
has a nonzero source impedance. Because these requirements are general to practical dc
sources, this control topology can integrate a decoupled capacitor into a dc voltage bus
without regard to application.
Rth
voc
vc
vterm
iload
Load
dc-dc
converter
Dc Source
Decoupled Capacitor
Figure 1.1. Active parallel connection of the capacitor to the voltage bus.
Combined power sources for two applications are explored in this thesis: a power
source for an electric vehicle, where the dc source is an uncontrolled battery, and a com-
puter VRM, where the dc source is a dc-dc converter with controlled nonzero source im-
pedance. Very different requirements are imposed on the power sources for these applica-
tions, and at a glance the plant and controller topologies chosen look very different. What
these applications have in common, though, is a load with a large transient energy re-
quirement and a bus voltage that is constrained to change relatively little. Each applica-
3
tion benefits from a decoupled capacitor. The features of the controller which program
the series impedance of the decoupled capacitor are the same for both proposed systems,
though other features differ to meet application requirements.
Detailed design analyses for each plant and controller combination are performed.
Small-signal models of the combined source for the vehicle are derived to determine sys-
tem stability and gauge the controller’s effectiveness at rejecting low-frequency bus volt-
age disturbances. If left unchecked, these disturbances would deplete the capacitor.
Large-signal models are derived for both systems to verify that the system stays within its
voltage, current and energy bounds for a worst-case loading. A scaled-down version of
the combined source for the vehicle was constructed and tested. Measurements closely
matched those predicted by the large-signal models.
1.1 Combined Battery and Ultracapacitor Source for a Vehicle
Energy needed to make a vehicle move must be stored on board in some form.
While most vehicles obtain their energy from hydrocarbon fuels, all-electric vehicles run
on stored electricity. The two most widely used electrical storage devices are electro-
chemical cells and capacitors. A group of electrochemical cells is called a battery. Typi-
cally energy-dense batteries are needed to provide adequate energy storage [1] in an all-
electric vehicle. However, such a battery may not provide sufficient peak power [1]. One
solution, upsizing the battery, may be unacceptably large, heavy, and expensive. Rather,
one can use the fact that the peak transient power requirement of a light passenger vehicle
is typically several times higher than the steady-state power requirement [2]. A second
source which supplies just the transient power peaks can be added to the voltage bus.
This combined source may eventually weigh and cost less than a single battery sized to
4
meet all system requirements [1]. The system described in this thesis uses ultracapacitors
(UCs) to provide the transient energy storage. The UCs are decoupled from the bus volt-
age using a dc-dc converter, and the battery is connected directly to the voltage bus.
Some background on batteries and UCs is needed to explain the choice of sources.
Charge is transported to and from the battery by means of a reversible chemical reaction
[3]. In contrast, capacitor charge is stored physically in an electric field established be-
tween two conducting sheets of material. The charge stored in a physical storage device
is directly proportional to the device terminal voltage, related by the capacitance value in
Farads. When no charge is stored in the device, the terminal voltage drops to zero. An
ultracapacitor uses the double-layer effect [4],[5] to store physical charge more densely
than a capacitor can.
Because batteries use chemical storage, they store far more energy per unit weight
than UCs [6],[5]. Battery voltage does not drop to zero when the cells are considered de-
pleted, and changes relatively little with charge level [3]. These properties make a battery
a good choice to provide steady-state energy and to support bus voltage. However, the
chemical reaction is not 100% efficient, and the active materials in the cell degrade over
time [5]. For example, energy-dense LiCoO2, or lithium-ion, cells have an input-output
efficiency of 94% when new and an efficiency of 85% after several hundred full
charge/discharge cycles [7]. Other cell chemistries are worse [7]. In addition, high current
draws substantially reduce the input-output efficiency [8],[9], apparent cell capacity [10],
and cycle life [11] of a battery. This suggests that the battery should not be exposed to
large currents. As peak current in a vehicle is transient in nature [2], a second source
which injects transient power should shield the battery from the high current peaks.
5
Two candidates to provide this transient power are a second battery and an ultra-
capacitor. Batteries in general will last for more cycles when they are shallow-cycled, or
have only a fraction of their stored energy used each cycle [11]. Some power-dense
chemistries, when shallow-cycled, have a cycle life rivaling that of a UC [12]. Table 1.1
compares properties of a commercially available “nanophosphate” LiFePO4 cell, repre-
sentative of power-dense cell chemistries, with those of a commercially available UC.
Energy and power densities are calculated from parameters supplied by the manufacturer
of each cell [14],[15], with the assumption that the battery will be discharged 10% each
cycle (shallow-cycled) and the UC will be discharged 75% each cycle. The table shows
that the technologies are roughly comparable in all areas except one: power density when
the device is being charged. Here the UC is over an order of magnitude better. This is an
important difference, as current is expected to flow through the transient energy source in
both directions at equal magnitude. For this reason a UC is chosen over a shallow-cycled
battery to serve as the transient energy source. Table 1.1. Comparison of a power-dense shallow-cycled (10% depth-of-discharge) battery to a UC.
Parameter “Nanophosphate” LiFePO4 10% DoD
UC 75% DoD
Energy density (W-h/kg) 10.8 4.5 Power density, discharging (kW/kg) 3.6 5.9 Power density, charging (kW/kg) 0.46 5.9 Input-output efficiency 94% [13] 96% [16] to
98% [17] Cycle life (number of charge/discharge cycles)
240,000 [12] 500,000 [6][5] to 1,000,000 [15]
Calendar life at room temperature 10 years [12] 10 years [15]
Strategies to incorporate a transient energy source into a vehicle traction bus fall
into two main categories: those which tie the UC directly to the bus and those which tie
the battery to the bus. Methods which make UC voltage equal to bus voltage include us-
6
ing a UC as the sole energy source [18], selecting between the battery and UC depending
on load conditions [19], and bucking the battery down to bus voltage [20]. The bus volt-
age varies by 2:1 or more in these schemes; power delivery to the load will be compro-
mised. Of the schemes which tie the battery to the bus, most use a current-mode dc-dc
converter. There is disagreement on how to set the current command, because the dc-dc
converter must be controlled not just to provide power to the bus, but also to make sure
the capacitor charge level stays within acceptable bounds. These are competing control
objectives. When only bus voltage is regulated, the capacitor is quickly depleted [21].
One approach to resolve the problem is to regulate system power and energy directly
[21]–[23]. These schemes have good design flexibility, but the resulting controllers are
nonlinear and very complicated. It is difficult to analyze the complete system for stability
and performance, and detailed analyses were not attempted in the above works.
Drolia et al. [24] and Awerbuch and Sullivan [25] propose a linear topology
which maintains much of the flexibility of the power and energy controllers but is simpler
to analyze and implement. This scheme generates an error signal by measuring bus volt-
age and high-pass filtering the measurement. The reference current is the error signal
multiplied by a finite gain value pK . In addition, a second proportional loop is used to
drive the capacitor to some nominal charge level. Since the controller is linear and time-
invariant, the full system can be analyzed using linear analysis tools when the dc-dc con-
verter is linearized.
In the above controller, the loop from bus voltage to reference current directly
programs the series impedance of the decoupled capacitor. For a buck converter, series
impedance magnitude is the reciprocal of pK [25] within the passband of the high-pass
filter (HPF). For a boost converter, series impedance magnitude is the reciprocal of pK
7
scaled by the boost ratio ctermV V . The HPF in this path gives the converter infinite im-
pedance at dc. Viewed another way, the HPF blocks the dc portion of termv ; reference
current will eventually return to zero after a step change in termv , limiting the charge
transferred to or from the decoupled capacitor. The topology seems promising, yet it has
been explored little. It was abandoned by Awerbuch and Sullivan [25] in favor of a dif-
ferent scheme. Drolia et al. [24] outlined the controller’s structure and provided a single
time-domain simulation of system response to a load step but did not follow up with
more analysis or experimental results. Lastly, these authors only investigated controlling
a UC bucked down to the bus voltage, skipping boost converters and other plant topolo-
gies.
This thesis will use a shaped-impedance controller to control a decoupled UC that
is boosted to bus voltage. The exact controller topology is different from that used in ei-
ther [24] or [25] and is shown in Figure 1.2. The bus voltage measurement is low-pass
filtered to keep noise and voltage ripple out of the current command. The loop which
regulates UC charge level is placed in parallel with the first loop and an integrator is
added to it. The integrator improves rejection of very low-frequency disturbances and
supports the offset between reference current refi and average inductor current Li . This
offset is inherent to peak current mode (PCM) control [26] and must be addressed by any
outer loop control strategy. Because the controller does not regulate power directly,
power distribution between the sources must be calculated in the design stage. Design
equations and models are presented in Section 2.1. Chapter 3 develops linearized small-
signal models of the complete system. These models are used to find system response to
disturbances in the frequency domain. As the system has different dominant dynamics at
different frequencies, frequency-domain analysis is useful for visualizing system behav-
8
ior. Large-signal models are developed and simulated results are shown in Chapter 4.
Experimental results using scaled-down power sources are presented in Chapter 5. Large-
signal models verify the design equations, and the experimental results closely follow
large-signal predictions.
UC voltage
UC reference
Bus voltage
Kp-1HPF LPF iref
PI
Figure 1.2. Outer loop controller for electric vehicle application.
1.2 VRM with Unbalanced Phases
A voltage regulator module (VRM) is a dc-dc converter which regulates the proc-
essor bus voltage busv . Full VRM design specifications for present generation Intel desk-
top microprocessors are published in a technical report [27] written by the manufacturer.
These processors draw up to 150 A of current from the VRM, with 120 A transients at
frequencies up to 2 MHz [27]. The VRM is also required to drop the 12 V rail from the
computer’s main power supply to a voltage as low as 0.8 V [27]. VRM output voltage is
allowed to droop with load current, but must stay within 50 mV of the target load line
under all operating conditions [27]. This means that the VRM needs to have a constant,
but nonzero, source impedance at all frequencies.
Typical VRMs use multiple interleaved [28] buck phases feeding into a common
bus capacitor to supply current to the processor. The term phase in this context refers to
the inductor and switch pair portion of a power converter; one buck phase is shown in
9
Figure 1.3. A typical switching frequency is 300 kHz which yields a control bandwidth
(BW) of 30-40 kHz [27], above which the series impedance of the phases rises due to a
constrained current slew rate. An array of electrolytic and ceramic capacitors near the
processor chip is employed as a second energy source to supply currents outside the con-
verter’s control BW [27]. As the current demand of processors continues to increase, and
supply voltage regulation tightens ([27] compared to [29]), more and more energy storage
available from a source with lower and lower series impedance is needed to supply these
transients.
vin vbus
Figure 1.3. One converter phase.
Adding more bulk capacitors to the processor voltage bus is one way to meet
these requirements. This solution presents packaging challenges; in order to have a suffi-
ciently low series impedance, the capacitors must be located right next to the processor
socket. Another solution is to raise the VRM’s switching frequency in order to increase
its control BW and thus reduce the required bus capacitance. Simply raising the switching
frequency, however, increases commutation loss in the switches and thus reduces effi-
ciency [30]. A third alternative is to add another active power source to the processor
voltage bus to supply current at frequencies between that of the dc source and the capaci-
tor bank. This third source carries no dc current, and therefore carries less RMS current
than the dc-carrying converter phases. The lower current requirement makes it practical
for the third source to have a high control BW and thus reduce bus capacitance. Previous
10
attempts achieve the needed BW using switched resistors [31] or linear regulators [32] to
source and sink transient current. Dissipative devices such as these unnecessarily harm
efficiency, though, if switching devices and reactive (energy-storing) elements can be
used instead.
This thesis investigates using a decoupled capacitor as the third power source.
The decoupled bulk capacitor need not be independent of other energy sources, as it was
drawn in Figure 1.1; a bulk capacitor tied to a higher voltage, more loosely regulated rail
(voltage bus) will be used more efficiently than if it were tied to busv . It can also be lo-
cated further from the processor socket as its impedance is actively controlled. As the
main power supply for the computer is capable of supplying full power to the processor,
the input of the decoupled capacitor can be connected to the same voltage rail as the dc-
carrying VRM phases. This structure is shown in Figure 1.4 and resembles a multiphase
dc-dc converter, with one set of phases carrying net dc current and another set carrying
transient current only. Section 2.2.2 shows that, through prudent choice of components
and system parameters, a VRM with fast transient-carrying phases can have substantially
less bus capacitance than, but comparable efficiency to, a typical multiphase VRM.
Cdecoup
5 mΩ
vin vbusC
Resr
transient-
carrying
phases
dc-
carrying
phases
Figure 1.4. VRM with transient-carrying phases in parallel with the dc-carrying phases.
Mossoba and Krein [33] previously modeled a VRM with a structure similar to
Figure 1.4, except a single transient-carrying phase was used. The fast transient-carrying
11
phase was controlled using sensorless current mode (SCM) control described in [34],
while the slow dc-carrying phases used PCM control. While a single phase was adequate
for processor loads in the past, multiple transient-carrying phases are needed to supply
today’s processors. Series input SCM phases share current [35] but there is no inherent
sharing between SCM-controlled phases when the inputs are in parallel. Thus the topol-
ogy studied in this thesis uses PCM control for all phases.
The current command for each phase is generated by the outer loop controller
shown in Figure 1.5. Filters split the current command cmdi by frequency, sending a high-
passed cmdi to the transient-carrying phases and a low-passed cmdi to the dc-carrying
phases. One Offset correction block makes the average of the sum of the current through
each transient-carrying phase track ,cmd fi , while the other performs an analogous function
for the dc-carrying phases. The particular correction scheme used was developed by
Huang et al. [36] and is called Native AVP in their paper. Since the decoupled capacitor
decoupC in Figure 1.4 obtains net energy from the input voltage inv , the outer loop control-
ler is not responsible for maintaining the charge level of this capacitor. The controller is
discussed in more detail in Section 2.2.1. Large-signal models and simulated results of
the proposed VRM and controller are given in Section 4.3. Simulation results show that
the converter is well within voltage tolerance limits specified in [27] when subjected to
substantial line, load and reference steps.
12
Referencevoltage
Bus voltage
Kf
HPF
LPF
iref,f
iref,s
∑iL,s
icmd
icmd,f
icmd,s
∑iL,f
Offsetcorrection
Offsetcorrection
1N
1M
to M transient-carrying phases
to N dc-carrying phases
Figure 1.5. Outer loop controller for VRM application.
For the overall thesis, Chapter 6 contains conclusions and ideas for future work.
Expanded block diagrams for the Simulink models, MATLAB code, and schematics for
the prototype converter are listed in the appendices. References to a label within a block
diagram or figure, such as Offset correction in Figure 1.5, are denoted with a sans-serif
font.
13
CHAPTER 2
DESIGN AND IMPLEMENTATION CONSIDERATIONS
This chapter discusses the plant and controller topologies in depth for each appli-
cation. Both applications use a current-mode dc-dc converter to decouple a capacitor or
UC from a voltage bus. The current command is set using high-pass filtered feedback of
bus voltage. Other sections of each controller modify the current command to meet plant
constraints or application requirements. Design equations for preliminary controller tun-
ing and part selection are presented.
2.1 Vehicle Application
Figure 2.1 is a detailed version of Figure 1.1, showing the combined power source
with its controller. Power-handling elements are in the dashed box marked Plant. The
battery is modeled by the voltage source ocv in series with resistance battR . Ultracapacitor
ucC is decoupled from the bus via a boost converter. Exact reasons for choosing a boost
topology over a buck are given in Section 2.1.1.
The Controller block includes the outer loop controller described in Figure 1.2
along with a comparator and S-R latch to implement PCM control. As inductor current Li
is fed back through the comparator, the portion of the controller between refi and the
switch pair , 'q q is called the PCM inner loop, inner current loop, or else just inner
loop. Blocks between input termv and refi compose the power-distributing loop which sets
the series impedance of the decoupled capacitor and thus influences how load power is
divided between the battery and UC. The PI controller and reference input ,uc refv between
inv and refi make up the charging loop, which is responsible for rejecting low-frequency
14
variations in termv and losses which would otherwise drain the UC. Lowercase variables
in Figure 2.1 are instantaneous quantities. The quantity outi is equal to the current through
the top FET averaged over one switching cycle [37]. Other lowercase variables in this
thesis with overhead bars represent that quantity averaged over one switching cycle.
vocLoad
Rbatt
ibatt
iL
iL
load i
iciout
LC
vin
vin
vuc,ref
vterm
vterm
QS
R Q
Kp-1HPF LPF
ramp
clock
iref
PI
Cuc
Controller
Plant
Figure 2.1. Active-parallel interface for the vehicle application.
The plant in Figure 2.1 presents several challenges for the controller to address.
The first challenge is that the plant provides no means to directly control the battery cur-
rent batti . Instead, current flow is determined passively by the value of battR and the dif-
ference between ocv and termv . The power-distributing loop commands the dc-dc con-
verter to inject current into the dc bus in response to a change in termv , with the balance of
the load current drawn from the battery. It is desired to transform knowledge of the im-
15
pedance of each source into an estimate of how power is apportioned between the sources
at any point in time. This estimate is needed to choose controller gains which keep batti
below its limit in response to a particular loading, and also to determine the power and
energy demanded from the UC bank. Equations and a Simulink model which perform this
estimate are discussed in Section 2.1.2.
The second challenge is that UC is an independent source. The dc-dc converter
must be controlled to keep the UC from becoming depleted or overcharged in the face of
UC loss, converter loss, and low-frequency variations in inv that are not completely re-
jected by the HPF. The charging loop addresses this problem, and its topology is intro-
duced in Section 2.1.3.
The third challenge is that the relatively large gain pK can cause refi to slew
faster than Li is capable of slewing. This will make the plant states overshoot their in-
tended targets, and can make the converter oscillate in extreme cases. The problem is
exacerbated in a boost converter. Section 2.1.4 analyzes two methods which guarantee
that refi will not slew faster than Li , and chooses the method appropriate for this system.
With the loops defined as such, the outer loop controller is attempting to regulate
two plant states (bus voltage termv and UC voltage inv ) using only one control input (the
switch pair , 'q q in Figure 2.1, transformed to reference current refi by the PCM inner
loop). While control of both states is possible because the two states are coupled via the
inductor, the control input cannot raise one state without driving the other state lower in
turn. Thus the power-distributing and charging loops will necessarily fight one another;
simply cranking up the gains of both will not improve regulation. The solution is to make
the power-distributing and charging loops dominant over different frequency ranges. As
the power-distributing loop must be tuned to balance battery power against UC energy,
16
the charging loop is tuned to not interfere with the power-distributing loop. This is dis-
cussed in more detail in Section 2.1.3 and in Chapter 3.
The significant power-handling components are sized in Section 2.1.5.
2.1.1 Boost vs. Buck Topology
In Figure 2.1 the UC voltage is boosted up to bus voltage. Why is a boost topol-
ogy chosen instead of a buck? For a vehicle application both ultracapacitors and batteries
have a cell voltage that is two orders of magnitude below what is needed for a traction
system. Making the cell size larger increases energy capacity and current capability but
not the potential across its terminals, as this voltage is intrinsic to the cell chemistry [3].
The only way to raise terminal voltage is to string cells together in series. As large-format
batteries and ultracapacitors mature, their cost per watt-hour and kW drop below that of
smaller cells. In future applications it will be more cost-effective to meet a power and
energy requirement with fewer large cells (boost) than with many smaller cells (buck);
therefore, a boost topology is studied.
One disadvantage to using the boost is that the PCM process regulates converter
input current, instead of output current as is the case with a buck. For a fixed pK the
converter’s source impedance (in the filter passband) will be the reciprocal of ' pD K ,
where 'D is one minus the steady-state switch duty ratio. So long as the converter’s
source impedance is much less than battR over the frequency band where the UC is ex-
pected to supply current, this variation will have little effect on power division. The boost
topology allows UC current to be controlled directly, which makes regulating its charge
level more straightforward.
17
2.1.2 Power and Energy Apportioning between Dc and Second Sources
An examination of how the power magnitude is divided between the sources is
motivated by the fact that a battery sized to meet a long-term energy requirement may not
be capable of supplying the peak load power by itself. Consider an electric vehicle with
the following traction power requirement:
1) Peak power: 100 kW for 10 seconds, both directions
2) Average power: 25 kW long-term
3) Load slew: zero to peak power in 0.2 s, peak to peak power in 0.4 s
4) 400 V nominal traction bus [38]
5) 59 kWh energy requirement for acceptable range
To minimize loss, the battery pack (henceforth called the battery) is connected directly to
the voltage bus. The 59 kWh energy requirement and 400 V nominal bus voltage trans-
lates to a battery with a charge capacity of 147 A-h. A battery consisting of 67 strings of
100 lithium-ion cells, each size 18650, will meet these requirements. Datasheets [10],[39]
for commercial lithium-ion cells suggest that charging and discharging rates should be
limited to no higher than one times the A-h rating of the battery, or 147 A. As the battery
discharges, its open-circuit terminal voltage drops. If the battery is considered fully dis-
charged when its open-circuit voltage reaches 340 V, then the worst-case power the bat-
tery can provide to the bus without exceeding 1 C discharge rate is 50 kW. Using the
datasheets and [40] the battery resistance is estimated to be 0.15 Ω at low frequencies.
Looking at Figure 2.1, the HPF and gain pK determine how load power is divided
between the battery and ultracapacitor. To better see this relationship, first define battery
power as battp , second source power as ucp , and load power as loadp . Load power is al-
ways equal to the sum of battp and ucp . For convenience, losses will be neglected from
18
the second source and power converter, and battp will be defined as power available from
the battery. Powers battp and loadp can be split into dc and ac portions, in anticipation of
establishing a relationship in the Laplace domain (where the dc values of variables are
ignored). The following relationships are defined:
, , , ,
, ,
,
ˆ
batt batt dc batt ac load load dc load ac
batt ac term batt ac
oc termbatt ac
bat
c in L
t
u
p p p p p p
p v i
v vi
R
p v i
(2.1)
Next, the ac powers and currents (including ucp , which has no dc component when losses
are excluded) are placed into the Laplace domain. In the Laplace domain, the term ˆocv is
zero, and so the relationships are
, ,
,
( ) ( ) ( )
( )( )
( ) ( ) ( )
batt ac term batt ac
termbatt ac
batt
uc i Ln
P s V s I s
V sI s
R
P s V s I s
(2.2)
As the second source and power converter are assumed to have no loss, the following
relationship holds:
in term o tL uv i v i (2.3)
The expression in the Laplace domain is
( ) ( ) ( ) ( )in tL erm outV s I s V s I s . (2.4)
A block diagram of the HPF used in the feedback path (HPF block in Figure 2.1)
is presented in Figure 2.2. The expression for this HPF in the Laplace domain is
19
High Pass Filter
Low Pass Filter
Out1
2
Out
1
Integrator1
1s
Integrator
1s
Gain1
1
Gain
1
In1
2
In
1
τ1
τ2
Figure 2.2. High-pass and low-pass filter prototypes.
1
1
( )1
sHPF s
s
. (2.5)
Further assuming that Lrefi I , the expression for the termV -to- LI control loop is
1
1
( ) 1 ( )1pL term
sI s K V s
s
. (2.6)
Next, define parameter K as
in batt
termp
v RK K
v . (2.7)
The relationship is the same in the Laplace domain:
( )
( )in batt
termp
V s RK K
V s . (2.8)
Multiplying both sides of (2.6) by ( )inV s and substituting in (2.4) and the middle line of
(2.2) gives
20
1
1,( ) ( ) ( ) ( ) 1 ( ) ( )
1in term out p in battL ac batt
sV s I s V s I s K V s I s R
s
,
1
1
( ) ( )
( ) ( ) 1out term
pbatt ac in batt
I s V s sK
I s V s R s
. (2.9)
Further substitution and simplification yields
, ,
( ) ( )
( ) ( )uc out
batt ac batt ac
P s I s
P s I s
1
1,
( )
( ) 1uc
batt ac
P s sK
P s s
. (2.10)
With losses neglected and , ( )batt acP s defined as in (2.2),
, ,load dc batt dcp p (2.11)
and
, ,
1
1,
( ) ( ) ( )
1 ( )1
load ac uc batt ac
batt ac
P s P s P s
sK P s
s
. (2.12)
The transfer function relating , ( )batt acP s to , ( )load acP s is
1
1 1
1
1
,
1
,
1
( ) 11
( ) 1 111
11 11 1 1 1
1 1
1 1 1
batt ac
load ac
P s sKsP s s K
s
s K
Ks K s K
s
s K s
(2.13)
where
11K . (2.14)
21
Likewise,
,
( )
( ) 1 1uc
load ac
P s K s
P s K s
. (2.15)
Equations (2.13) and (2.15) demonstrate how load power is apportioned to the dc and
second sources, respectively. The dc source sees the dc and low-passed portions of loadp
(per equation (2.11)), plus a portion of the high-passed portion of loadp . The second
source sees the remainder of the high-passed portion of loadp . The parameter K deter-
mines how the high-passed portion of the signal is divided between the dc and second
sources, while parameter determines the location of the frequency corner. Tuning of K
and may be done by creating a stateful model (time-domain model with state variables)
based on (2.11), (2.13) and (2.15) in a numeric simulation program and viewing the re-
sponse of ( )battp t and ( )ucp t to ( )loadp t . Figure 2.3 shows this model created in Simulink.
HPF'dPload
filter time constant
1/tau
To Workspace2
uc_energy
Puc(t)
uc_power
Pload(t)
roadpower
Pbatt(t)
batt_power
Integrator1
1s
Integrator
1s
Gain1
1/(K+1)
Gain
K/(K+1)
Figure 2.3. Simulink model to determine dynamic power and energy distribution between the battery
and ultracapacitor.
A load power waveform which will put maximum transient and steady-state
stresses on the combined source is a pulse train with 100 kW magnitude, 40 s period, and
25% duty cycle. This waveform draws full load power from the combined source for 10
22
seconds, followed by no power for 30 seconds, resulting in an average draw of 25 kW.
Figure 2.4 shows this waveform and its Fourier series expansion. Power splitting is found
by loading this waveform into the Pload(t) block in Figure 2.3 and running the simulation
for various values of K and . The Integrator blocks in Figure 2.3 have an initial condi-
tion of zero, representing a condition where neither source is transferring power.
0 20 40 60 80 100 120
0
2
4
6
8
10
x 104
Time (s)
Pow
er (
W)
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4-2
0
2
4
6x 10
4
Frequency (Hz)
Pow
er (
W)
Figure 2.4. (Top) Pulse train with 100 kW peak power and 25 kW average power. (Bottom) Fourier
series expansion of this waveform.
Three combinations of K and which keep battery power just under 50 kW are
listed in Table 2.1 along with resulting peak UC power and energy draw. Figure 2.5
draws battery power, UC power and UC energy as functions of time. The solid lines in
Figure 2.5 are the simulation output for the 100 kW pulse train input. The trend here is
that the energy drawn from the UC rises as is decreased, even while peak UC power
and K grow larger. To see why this happens, the simulation was run again with zero ini-
tial condition with Pload(t) replaced by a 25 kW step. The dashed lines in Figure 2.5
show the result. Figure 2.6 (on page 24) compares the UC power and energy response for
23
each K and more clearly. With 70 the energy drawn in response to the 25 kW step
is over twice what is drawn when is smaller. Although peak UC power is less due to
smaller K, the larger makes the filter take longer to drive UC power to zero and thus
significantly increases the area under the power curve. Figures 2.5 and 2.6 show that an
HPF alone is enough to limit net UC charge transfer when a load step is applied.
Table 2.1. Component power and energy for three combinations of K and τ. τ K Peak Battery Power Peak UC Power Peak UC Energy
70 2.8 48.8 kW 72.8 kW 1570 kJ 25 6.5 48.9 kW 83.6 kW 890 kJ 20 15 48.7 kW 89.7 kW 849 kJ
0 100 200 300 400 5000
10
20
30
40
50
Bat
tery
Pow
er (
kW)
0 100 200 300 400 500-50
0
50
100
UC
Pow
er (
kW)
= 70 K = 2.8
0 100 200 300 400 5000
500
1000
1500
UC
Ene
rgy
(kJ)
0 100 200 300 400 5000
10
20
30
40
50
Bat
tery
Pow
er (
kW)
0 100 200 300 400 500-50
0
50
100
UC
Pow
er (
kW)
= 25 K = 6.5
0 100 200 300 400 5000
500
1000
1500
UC
Ene
rgy
(kJ)
0 100 200 300 400 5000
10
20
30
40
50
Time (sec)
Bat
tery
Pow
er (
kW)
0 100 200 300 400 500-50
0
50
100
Time (sec)
UC
Pow
er (
kW)
= 20 K = 15
0 100 200 300 400 5000
500
1000
1500
Time (sec)
UC
Ene
rgy
(kJ)
100 kW 10s pulsed
Constant 25 kW
Figure 2.5. Simulation output for 100 kW pulse train and 25 kW step inputs.
24
0 100 200 300 400 5000
5
10
15
20
25
UC
Pow
er (
kW)
0 100 200 300 400 5000
500
1000
1500
UC
Ene
rgy
(kJ)
Time (s)
= 70, K = 2.8
= 25, K = 6.5
= 20, K = 15
Figure 2.6. UC power and energy response to 25 kW step for various K and τ.
2.1.3 Charging Loop
In Section 2.1.2, it was shown that the HPF prevents UC energy from exceeding
a certain threshold in response to a load step. The threshold is directly proportional to the
step magnitude; if the peak step is known, the energy can be calculated and the UC sized
to supply it. In practice, losses and ultracapacitor leakage current will consume net en-
ergy, and the controller may be reset (e.g., by the operator turning off the device) before
the power storage system reaches an equilibrium value. Some compensation is needed to
ensure the ultracapacitor does not become over- or undercharged as a result of these dis-
turbances. A feedback path from inv to refi provides this compensation and is termed the
charging loop.
The charging loop uses a proportional-integral (PI) controller to bring inv up to
(or down to) a reference level ,uc refv by manipulating refi . In steady state, refi is related to
25
Li by a dc offset. Due to the boost topology, Li has the same magnitude and opposite sign
as UC current uci . As discussed in the introduction to Section 2.1, the charging loop
needs to have low gain at frequencies where the power-distributing loop has high gain,
and may have high gain at frequencies below this. The controller discriminates between
legitimate load transients and line disturbances (or net power losses) by frequency, reject-
ing only the very long-term disturbances to inv which will drain or overcharge the UC.
A block diagram of just the second source (an ultracapacitor in this case) and the
charge controller is shown in Figure 2.7. This figure uses the assumption that Lrefi i ,
valid when the dc offset is small. The transfer function from reference ,uc refv to output inv
in the Laplace domain is
,
2
( )
( )chg ichgin
uc ref uc chg ichg
sK Kv s
v s s C sK K
. (2.16)
This equation provides a starting point for setting controller gains chgK and ichgK . Good
candidate values of chgK and ichgK will make the poles of (2.16) well-damped and give
(2.16) little gain at frequencies at and above 1 . The variable is the coefficient for the
power-apportioning filter discussed in Section 2.1.2; frequency 1 will be several times
lower than HPF angular corner frequency 11/ . Small-signal transfer functions which
relate inv to load and battery disturbances are derived in Chapter 3.
26
Charging Controller Ultracapacitor
~ i_L v_in
1
Kichg
Kichg
Kchg
Kchg
Integrator
1s
Int
1s
Gain
-1/Cuc
v_uc,ref
1i_ref v_in
Figure 2.7. Block diagram of feedback loop containing ultracapacitor and charge controller.
2.1.4 Limiting iref Slew
PCM control commands a state variable, ( )Li t , to follow a signal, ( )refi t , gener-
ated from two other state variables, termv and inv , each switching cycle. As state variables
in general represent energy stored in a device or circuit element, they may not change
value, or slew, arbitrarily quickly. When refi slews faster than Li , both Li and termv tend
to overshoot their target values, causing device stress. In extreme cases the power con-
verter can oscillate, a condition which may destroy the converter or devices connected to
it. Therefore it is desired to find ways to make sure refi cannot slew faster than Li .
The inductor current ( )Li t in a switching power converter is often analyzed by
separating it into a “ripple” portion Li plus a “fast averaged” portion Li [37]. The fast
averaged portion is the current transferred to the dc bus. A dc-dc converter is said to be
boosting when Li is flowing from a lower-voltage node to a higher-voltage node and
bucking when the opposite occurs. Looking at the power-handling components of Figure
2.1, the switching converter in this diagram is boosting when 0Li and is bucking when
0Li .
27
Practical boost converters have a hard duty ratio limit programmed into the device
which generates the switching signal [41]. As a consequence, if Li is positive and needs
to increase further (due to an increase in refi ), the rate at which it may increase is consid-
erably less than the positive-going slope of the inductor current ( )Li t when the active
switch is on. Figure 2.8 shows trajectories of ( )Li t at some operating point , in termV V .
The inductor current slopes are
21in c termV V V
m mL L
(2.17)
using the nomenclature from Figure 2.1. The solid line shows inductor current for the
case when (0) ( )L Li i T . The active switch turns off at time DT, where D is defined as
1 in
term
VD
V . (2.18)
-m2
-m2
m1
m1
iL(t)
ΔIL
0 TDT dmaxT t
Figure 2.8. Trajectory of iL(t) at an operating point (Vin,Vterm) in steady state (solid line) and at
maximum duty ratio (dashed line).
28
The dashed line is the current trajectory under the same operating condition, except the
active switch is turned off at time maxd T , where maxd is the maximum duty ratio of the
boost converter. Over the course of one switching cycle, the net change in inductor cur-
rent is
1 2
( ) (0)
1max max
L L LI i T i
m d T m d T
. (2.19)
Substituting (2.17) into (2.19) gives
1in max terL m
TI V d V
L .
Finally, substituting (2.18) for inV gives
term ma
LxV d D T
IL
. (2.20)
The peak slew rate of Li at operating point ,in termV V is just
L term mL
LaxV d DI I
it T L
. (2.21)
Notice that as D approaches maxd , the peak positive slew on Li rapidly shrinks.
This has two implications for the converter design:
1) The largest steady-state D should be kept well away from maxd . For a maxd of
0.85, D should be no higher than 0.6 or 0.65 for a practical system. This corre-
sponds to a boost ratio no higher than 2 or 2.5:1 once losses are accounted for.
2) Attention needs to be paid to refi so that the slew rate of refi should not exceed the
slew rate of Li at worst-case D.
On fast time scales, the slew rate of refi is the slew rate of termV multiplied by gain
pK . One way to limit refi slew rate is to size the bus capacitor so that termV changes suffi-
29
ciently slowly. Here the bus capacitor provides energy to the bus, so there are now three
sources on the bus which provide significant energy to the load, and is appropriate when
the load current’s slew rate exceeds the maximum possible slew rate of the decoupled
capacitor. Another method is to tune the low-pass filter LPF in Figure 2.1 to slow the
slew rate of refi . This method is appropriate when the load current’s slew rate is less than
that of the decoupled capacitor. Here the bus capacitor does not provide appreciable en-
ergy to the load, and so only needs to be sized to meet the bus voltage ripple requirement.
Equations which give an estimate of capacitance C or low-pass filter (LPF) coefficient 2
are presented next.
The required bus capacitance C is estimated by looking at capacitor current just
after a load transient occurs. When the converter is in steady state, the fast average of the
capacitor current is zero. Suppose a step change in load current
,0
loadload
c
PI
V
(2.22)
is applied, where ,0c
V represents fast-averaged capacitor voltage just before the step is
applied. At time 0t , just after the step is applied, the capacitor current is equal to
loadI since no other source has yet adjusted its current delivery. The fast average slew
rate of cV at this time is
0
0
slew cc
loaddV IV
dt C
. (2.23)
Next, cV slew is related to refi slew by the constant parameter pK . It is desired to con-
strain refi slew to be less than or equal to that of LI slew. Plugging (2.21) and (2.22) into
(2.23) produces
2
,0
load
a c
p
m x
P K LC
d D V
. (2.24)
30
This capacitance value guarantees that Li slew will not exceed the value given in (2.21)
for a load step loadP . Setting D, loadP and ,0c
V to their worst-case values produces a C
that will enforce (2.21) under all operating conditions. When using a large bus capacitor,
coefficient 2 should be set to a small value. This allows the LPF to filter out switching
noise but not degrade transient response of refi .
To set the LPF coefficient, a few assumptions must be made:
1) The bus capacitor is small enough that a step change in load current very nearly
produces a step change in termV .
2) The dc source is capable of supplying power at high frequencies. A battery fits
this description, although drawing high currents will deplete the cell’s energy [40]
and shorten its cycle life [11].
3) The system is in steady state just before the transient is applied, and thus at time
0t load current is equal to battery current.
Using the first assumption, at time 0t the battery is again supplying virtually all the
load current, as the inductor has not yet had time to respond. Thus load battI I and
,0
loadterm batt load batt
c
PV R I R
V
(2.25)
assuming ,0term c
V V . A prototype of the first-order LPF was given in Figure 2.2. Fol-
lowing the termV -to- refi feedback path in Figure 2.1, the step change incident on the LPF
integrator at time 0 is
,2 0
loadbat
pt
c
K PR
V
. (2.26)
The output of the integrator sets refi ; thus the slope of refi at time 0t is the expression
given in (2.26). Setting (2.26) less than or equal to (2.21) and solving for 2 yields
31
,
2 2
0
load batt
max c
pP K R L
d D V
. (2.27)
Notice that pK appears in the numerator of both (2.24) and (2.27). If pK is very
large, refi slew will only be held less than Li slew if the converter has either a very large
bus capacitor or else a small control bandwidth. These factors need to be traded off
against the desired UC energy storage and power apportioning when selecting pK .
2.1.5 Choosing the Power-Handling Components
The inductor L, bus capacitor C, and UC ucC are sized in this section, and
switches are chosen. The inductor needs to be large enough to keep current ripple reason-
able, but small enough to allow its current (and UC current) to slew quickly enough to
meet load requirements. Bus capacitor size is set according to the required bus voltage
ripple. As the battery is connected directly to the bus, the voltage ripple requirement is
determined by the current ripple the battery can tolerate. The UC bank is sized for energy
and voltage requirements.
The current slew rate requirement places a maximum value on inductor L. Output
current from the converter must slew from 0 to 250 A in 0.2 s. This means Li must slew
625 A in 0.2 s when steady-state duty ratio is 0.6. A switching frequency of 10 kHz
means that Li has 2000 cycles to move 625 A, or 0.31 A per cycle. The linear outer loop
controller will make Li follow either a decaying exponential or a damped second-order
trajectory. Thus, Li may slew at up to 0.62 A per cycle, or twice the above prediction.
With cV = 400 V, D = 0.6, and maxd = 0.85, rearranging (2.20) gives a maximum permis-
sible inductance value of 16.1 mH. A 500 µH inductor is chosen, which will yield an
acceptable 20 A steady-state ripple current and good current slew rate.
32
As inductor current can slew fast enough to meet load requirements, the bus ca-
pacitor does not need to provide energy to the load and so is chosen to meet the voltage
ripple requirement. The voltage ripple is set by the battery current ripple requirement. For
the 400 V, 147 A-h battery with battR = 0.15 Ω, a voltage ripple of 3.82 V limits RMS
current ripple to 10% of battery A-h capacity. At peak load this voltage ripple require-
ment is met by an 8000 µF capacitor with an equivalent series resistance (ESR) of 3.1
mΩ. Because a boost topology is used, the bus capacitor will need to supply load current
during the time the active switch is off, and absorb the difference between inductor cur-
rent and load current when the switch is on. Eighteen 1000 µF, 450 V electrolytic capaci-
tors are needed in order to reliable source and sink these large currents each switching
cycle [42]. The capacitance bank has an ESR of about 3.14 mΩ [42]; it will meet the
voltage ripple requirement.
Insulated-gate bipolar transistors (IGBTs) are selected for switches q and 'q be-
cause these devices will block the 400 V bus voltage. Both switches must be transistors in
order to allow current to flow in both directions between the UC and dc bus. The typical
on-state voltage drop of an IGBT is 2 V, or 0.5% of bus voltage. This is small enough to
be neglected from the dynamic models. Switch loss will be modeled by a small series
resistance. A 10 kHz switching frequency is an appropriate choice for IGBTs with the
required voltage and current rating.
The maximum and minimum UC voltages are chosen to be 320 and 180 V, re-
spectively, to allow for duty ratio headroom. This allows 68% of the stored UC energy to
be accessed, compared to less than 10% when the UCs are passively connected to the
bus. The UC reference voltage is chosen to correspond to 55% [43] of the usable UC
charge capacity, or 266 V. Considerations from Section 2.1.2, combined with modeling
33
to be described in the next two chapters, yield a pK of 65 and 1 of 3.71. With these pa-
rameters the UC must provide 859 kJ of energy as it discharges from 266 to 180 V. The
capacitance needed to make this happen is
2 21
2
859 kJ44.16 F
266 180ucC
. (2.28)
Two strings of 119 large ultracapacitors, each rated for 2.7 V and 3000 F, are needed to
meet this capacitance requirement. This is a physically large UC bank but still several
times smaller than what would be required if the UCs were connected directly to bus
voltage. The importance of decoupling UC voltage from bus voltage is made evident
here.
2.2 VRM Application
In Section 2.2.1 the plant and controller for the three-source VRM are described
in detail. The outer loop controller is shown to make the source impedance of the two
active sources constant over their combined control BW. In Section 2.2.2 power-handling
components for a three-source converter are chosen and compared to those used for a
typical VRM design. Required bus capacitance and power loss are calculated for both
designs and compared.
2.2.1 Plant and Controller in Detail
Figure 2.9 (on page 35) shows the VRM plant and controller in detail. The input
voltage rail is modeled by the voltage source inv with series resistance srcR . The N dc-
carrying phases have a relatively low switching frequency, a larger inductor bigL and are
programmed by the current reference ,ref si . The M transient-carrying phases have a high
34
switching frequency, a small inductor smallL and are programmed by ,ref fi . All phases
feed into a common bus capacitor C. The slow clock signals are N-way interleaved, while
the fast clock signals are M-way interleaved independently of the slow clock. The com-
parator and S-R latch in the PCM & PWM block are applied to each phase, and implement
PCM control of each switch pair. Since there are only two reference currents, an offset
correction block is used for each set of phases, rather than for each individual phase. The
offset correction acts to slowly vary an offset between ,cmd ji and ,ref ji using fed-back cur-
rent information; at high frequencies the block behaves like a constant additive term, pre-
serving the good dynamic performance of PCM.
The entire VRM, with all sources combined, needs to take on a constant nonzero
output impedance [27]. Ideally the VRM will behave like a voltage source refv in series
with a resistance thR , and busv will droop with load current. A drooping bus voltage is
straightforward to implement with PCM-controlled phases; the outer voltage loop simply
needs to have a fixed finite gain and compensated so that the voltage loop programs aver-
age current instead of peak current. In Figure 2.9, gain fK is the reciprocal of thR so
long as the average current through all phases tracks cmdi .
Representing the currents in the Laplace domain,
,
,
1( ) ( )
1
( ) ( )1
cmd s cmd
cmd f cmd
i s i ss
si s i s
s
(2.29)
where
( ) ( ) ( )f bucmd rs efi s K v s v s (2.30)
35
vin
Load
Rsrc
iL,s1iL,s1
iloadLbig
C
qs1(t)
1-qs1(t)
vref
vbus
Resr
vbus
iref,f
iref,f
iref,s
iref,s
PCM &
PWM
iL,sN
Lbig
qsN(t)
1-qsN(t)
iL,f1
Lsmall
qf1(t)
1-qf1(t)
iL,fM
Lsmall
qfM(t)
1-qfM(t)
slow clock
iL,sN
PCM &
PWM
slow clock
iL,f1
PCM &
PWM
fast clock
iL,fM
PCM &
PWM
fast clock
iL
q(t)QS
R Q
ramp
clock
iref
PCM & PWM Block
Outer Loop Controller
Kf
HPF
LPF
iref,f
iref,s
∑iL,s
icmd
icmd,f
icmd,s
∑iL,f
Offsetcorrection
Offsetcorrection
1N
1M
Figure 2.9. Topology of proposed VRM with M transient-carrying and N dc-carrying phases.
36
and is the coefficient for the high-pass and low-pass filters. For small ripple current the
approximations
,
1
,
,
1,
( ) ( )
( ) ( )
N
cmd sL sp
L fq
p
M
cmd fq
i s i s
i s i s
(2.31)
hold for 0s , and the approximation is an equality when 0s . Using (2.29) and (2.31)
the sum of the current through all phases is
1
, ,1
1( ) ( ) ( )
1 1
N M
cmdp q
L sp L fq
si s i s i s
s s
1
( ) ( )1 cmd cmd
si s i s
s
.
Thus in the time domain
cL mdi i (2.32)
within the control bandwidth spanned by both the slow and fast phases. Substituting
(2.30) into (2.32) and transforming back into the time domain gives the equivalent source
resistance
1bus ref
L fth
v vR
i K
. (2.33)
The outer loop controller should keep thR nearly constant across the control bandwidth of
both the fast and slow phases. Deviations from thR will be due to a change in offset (due
to change in line or bus voltage, or duty ratio) and not from current transitioning from
fast to slow phases.
Recall from Section 1.2 that the VRM is responsible for dropping the 12 V supply
voltage to 0.8 V. A parallel-input parallel-output (PIPO) buck converter would need to
37
operate at a very low steady-state duty ratio in order to do this. Phase current would be
able to slew up (more positive) quickly, but the limited duty ratio headroom constrains
the current to slew down (more negative) slowly. As the transient-carrying phases need to
slew in both directions quickly and with equal speed, they will operate best at a steady-
state duty ratio near 0.5. Thus an additional dc-dc converter is introduced to drop the 12
V supply rail to an intermediate 2 V rail, and the transient-carrying phases drop this in-
termediate voltage to busv . Previous research [44] shows that a two-stage VRM (a VRM
with two cascaded dc-dc converters) can be more efficient than a single-stage VRM when
the voltage transformation ratio is high, and so the dc-carrying phases are connected to
the intermediate voltage rail as well. The entire connection is shown in Figure 2.10. The
second stage, labeled VRM, is identical to Figure 1.4 and is the only stage that is dis-
cussed in detail in this thesis.
12 V Cdecoup
5 mΩ
2 V 0.8 VC
Resr
First stage
transient-
carrying
phases
dc-
carrying
phases
Second stage
(VRM)
dc-dc Load
Figure 2.10. Power converters needed to drop 12 V to 0.8 V. Second stage of this structure is the
VRM studied in this thesis.
2.2.2 Impact of Fast Phases on Loss and Bus Capacitor Size
The natural design strategy is to optimize the dc-carrying phases for steady-state
efficiency and the transient-carrying phases for fast dynamic response. The dc-carrying
phases may use a lower-than-normal switching frequency, larger-than-normal inductors,
and low-Rds,on switches to minimize loss. The transient-carrying phases should use an
38
extremely high switching frequency and very small inductors in order to maximize tran-
sient response and minimize the needed bus capacitance. Since RMS current through the
transient-carrying phases is expected to be low, switches with high conduction loss but
low gate charge will be the most efficient.
Two converters will be examined for loss and bus capacitor size. The design
specifications are:
2 V input, 0.8 V output (2-stage design for both)
140 A dc load current
Desired source impedance of 0.8 mΩ.
Sample component values are compiled in Table 2.2. Values for the slow and fast phases
are listed in the middle table columns. The sample converter has four slow phases sharing
dc current and two fast phases sharing the fast transient current. Values for a typical four-
phase VRM with matched phases are listed in the right-hand column for comparison. The
FETs in this table are chosen from the same product line and represent state-of-the-art
technology.
Table 2.2. Sample component values.
Parameter Value for Slow Phases
Value for Fast Phases
Value for Typical VRM Phase
L 500 µH 10 µH 150 µH
swf 150 kHz 5 MHz 300 kHz
swR 1.4 mΩ [45] 5.2 mΩ [46] 1.4 mΩ [45]
Vgs 10 V 4.5 V 10 V
Gate Charge 35 nC [45] 13 nC [46] 35 nC [45]
Rise / Fall Time 27 / 11 ns [45] 13 / 5.4 ns [46] 27 / 11 ns [45]
Per-phase current 35 A 5 Arms 35 A
39
Table 2.3 lists expected per-phase current and loss for each design, along with
total loss and required bus capacitance. Ripple current is found using the straight-line
approximation for buck converters [47] and converted to RMS under the assumption that
the ripple current approximates a triangle wave. Commutation loss is calculated assuming
linear dsV and dsI trajectories [30]. Bus capacitance is estimated using a form of (2.24)
modified for the buck converter and for multiple interleaved [28] phases. The equation is
,
2
0min( ,1 )
pload
in
I K LC
D D V M
(2.34)
Table 2.3. Comparison of per-phase and overall switching losses and bus capacitance of proposed
VRM compared to typical VRM.
Per-phase Parameters
Parameter Value for Slow
Phases Value for Fast
Phases Value for Typical
VRM Phase RMS ripple current 1.85 Arms 2.77 Arms 3.08 Arms
RMS switch current 36.85 Arms 7.77 Arms 38.08 Arms
Gate Charge Loss 0.055 W 0.292 W 0.105 W
Commutation Loss 0.281 W 0.958 W 0.582 W
Conduction Loss 1.90 W 0.314 W 2.03 W
Whole Converter Parameters
Parameter VRM with 4 Slow, 2 Fast Phases VRM with
Balanced Phases Total Switch Loss 12.1 W 10.9 W
Bus Capacitance 670 µF 2500 µF
where M is the number of fast phases in the converter (or number of total phases for the
VRM with all phases matched). It keeps refi slew from exceeding Li slew when the cur-
rent command is not low-pass filtered. The M term is squared because both 1 tp hK R
and L are divided by the number of phases present. It is assumed for now that C in (2.34)
is big enough to meet the voltage ripple specification; voltage ripple with ESR included
will be verified by the dynamic switched model presented in Section 4.3. The VRM with
40
additional fast phases requires less than one-third the bus capacitance of the typical VRM
design, yet consumes only 1.2 W more power. The higher efficiency of the dc-carrying
phases, made possible through use of a decoupled capacitor, nearly offset the additional
power consumed by the transient-carrying phases. The overall efficiency of the VRM at
112 W load is reduced less than one percent.
41
CHAPTER 3
SMALL-SIGNAL MODELS
The active-parallel combined source for the vehicle application, shown in Figure
2.1, has different dominant behavior at different time scales. These time scales are
stretched across many orders of magnitude. It is desired to find the plant states’ sensitiv-
ity, particularly UC voltage, to changes in battery voltage and load current. Frequency-
domain analysis gives a concise representation of sensitivity at all frequencies, but only
linear time-invariant systems may be analyzed using this technique.
Section 3.1 develops a linear model of the system in Figure 2.1, comprised of the
three-state plant and its associated controller. To do this, the switched power converter
plant is first averaged over one switching cycle [37]. The averaged model is then lin-
earized using small-signal perturbation analysis [48],[41]. Inner- and outer-loop control-
lers are wrapped around this linearized plant model to make the complete closed-loop
system. The resulting small-signal model will have a cluster of poles and zeros at low
frequencies and another cluster at higher frequencies (but still well below the switching
frequency). Lower-frequency dynamics are attributed to the ultracapacitor, inv -to- refi
charging loop, and HPF on the power-distributing cv -to- refi loop. Higher-frequency dy-
namics are attributed to the current loop, the passed portion of the power-distributing
loop, and boost converter L-C dynamics. As discussed in Section 2.1.4, the outer loop
controller must be set so that per-cycle inductor current does not change too quickly un-
der worst-case operating conditions. This means that the bandwidth of the controller will
necessarily be well below the converter’s switching frequency. An averaged, linearized
42
plant should be sufficiently accurate to develop a whole system model at all frequencies
where the controller has appreciable open loop gain.
3.1 Small-Signal System Model with Three-State Plant
3.1.1 Model of Plant
Figure 3.1 presents the plant portion of the active-parallel combined source with
significant parasitic resistances added. The bus voltage termV has been labeled ( )cv t to
show that bus voltage is equal to the voltage across the bus capacitor. The circuit has
three states: capacitor voltage ( )cv t , inductor current ( )Li t and ultracapacitor voltage
( )inv t . The energy-storing circuit elements are linear. Switches q and 'q reconfigure the
interconnection between the circuit elements; they are the control input to the system.
The switches are controlled as in a boost converter: when active switch q is on, inductor
current rises; when it is off, inductor current falls. As both switches allow current to flow
in either direction, continuous-conduction mode [41] is assumed. Other plant inputs are
open-circuit voltage ( )ocv t and current source ( )loadi t . This current source is a distur-
bance input, not part of the physical system, to inject small load disturbances into the
model.
43
voc(t)
Rbatt
ibatt(t)
iL(t)iuc(t)
ic(t)
L C
vin(t)
vL(t)
vc(t)
q
q’
Ruc
RLR
Rsw
Rsw
Cuc
iload(t)
(distur-
bance
input)
(load)
Figure 3.1. Power-handling components of the combined source for an electric vehicle.
When the active switch is ON, inductor current flows through ucC , L , and para-
sitic resistances , , and uc L swR R R . When the active switch is OFF, inductor current flows
through these same circuit elements and enters the circuit node where
, , and ( )batt loadC R R i t meet. Since inductor current flows through , , and uc L swR R R regard-
less of switch position, these resistances may be replaced by an equivalent resistance inR .
State equations can be written for each linear subsystem. The state equations with the
active switch ON are
( )( ) ( ) ( )
( ) ( ) ( ) ( )( ) ( )
( )( ) ( )
in in
ocload
batt
inuc
LL L
c c c
c L
c
u
di tv t L v t R i t
dtdv t v t v t v t
i t C i tdt R R
dv ti t C i t
dt
(3.1)
When the active switch is OFF the equations are
44
( )( ) ( ) ( ) ( )
( ) ( ) ( ) ( )( ) ( ) ( )
( )( ) ( )
inL
L L c
c c cc L
in
ocload
batt
inuc uc L
di tv t L v t R i t v t
dtdv t v t v t v t
i t C i t i tdt R R
dv ti t C i t
dt
(3.2)
In (3.1) and (3.2), the lowercase variables represent instantaneous voltage or current. Up-
percase variables are constants.
The rationale and procedure for using time averaging (or simply averaging) is
given in [41] and also in [48]. As switching frequency is fixed, averaging may be used.
The time average of (3.1) and (3.2) is
(1 )
(1 )
(1 )
in in in in
oc ocload load
batt batt
i
LL L c
c c c c cL
nuc L L
diL d v R i d v R i v
dt
dv v v v v v vC d i d i i
dt R R R R
dvC d i d i
dt
(3.3)
where d is the switch duty ratio. In (3.3) the instantaneous state and input variables are
replaced by averaged quantities, denoted by an overhead bar. The averaged quantities are
allowed to “move” [41] and so are functions of time. Duty ratio d represents the fraction
of time the active switch is ON during each cycle; thus it is inherently an averaged quan-
tity. Here d is also allowed to vary with time. The bars over the variables in (3.3) will be
neglected for clarity; lowercase variables are assumed to be averaged values that vary
with time. With like terms canceled and 'd substituted for 1 d , (3.3) becomes
'
'
LL cin in
ocload
batt
inuc
c c cL
L
diL v R i d v
dtdv v v v
C d i idt R R
dvC i
dt
(3.4)
45
As the variable 'd is a control input [41], equation (3.4) contains multiplicative nonlin-
earities between a state variable and an input. It must be linearized in order to use linear
analysis tools.
Equation (3.4) is linearized by replacing each averaged state variable,
, , and c L inv i v , and input, ', , and oc loadd v i , with a constant dc value plus a time-varying
perturbed quantity [41],[48]. The perturbed quantity is assumed to be small compared to
the dc value; thus products of perturbed terms are neglected [41]. Introducing constant
and perturbed variables is functionally equivalent to performing a Taylor series expan-
sion of (3.4) and neglecting the higher-order terms [48]. The substitutions are
' 1 ( ) '
in
oc oc oc
load load l
c c c
L L L
i
oad
n in
d D d D d
v V v
i I i
v V v
v V v
i I i
(3.5)
Define loadI to be zero; no dc load current will be applied through the disturbance input.
Substituting (3.5) into (3.4) and neglecting products of perturbed quantities gives
' '
' '
L LL Lin in in
ocload
b
c c c
c c oc c c c cL L L
i
att
inuc
nL L
dI diL V v R I i D V dV D v
dt dt
d dv V v V v V vC D I dI D i i
dt dt
V
V
R R
d dvC I i
dt dt
(3.6)
46
The linearized plant model (3.6) is valid for small perturbations of the state and input
variables about an equilibrium operating point. In equilibrium, the following dc relation-
ships apply:
0 0 0
'
' 0
' 0
L
L
c
in c
c in
in in
in
oc
batt
L
c cL
dV dVdI
dt dt dt
V R ID
V
V D V R I
V V VD I
R R
(3.7)
The relationships in (3.7) cancel all of the dc terms in (3.6) except for LI in the state
equation for inv . As inv is in equilibrium only when the ultracapacitor current is zero, and
the UC current is equal to the inductor current, the dc quantity of the inductor current
must always be zero. At higher frequencies, it is possible for the converter to be in quasi-
steady-state with a nonzero LI . A two-state model presented in Section 3.2 handles this
case and discusses the result. With the dc cancellations made and LI set to zero, the re-
sulting averaged small-signal ac model of the plant in Figure 3.1 is
'
'
in in
ocload
b
Lc c
att eq
i
L
c cL
Ln
uc
diL v D v dV R i
dtdv v v
C D i idt R R
dvC i
dt
(3.8)
where eqR is the parallel combination of resistances R and battR .
The next step is to put (3.8) into the Laplace domain and obtain transfer functions
relating plant inputs to plant states. The plant states are chosen as “outputs” because each
47
is measured and fed into some portion of the controller. The Laplace transforms of the
perturbed variables are
( )
( )
( )
( )
( )
( )
in in
oc oc
load load
L L
c c
i i s
v v s
v v s
d d s
v v s
i i s
(3.9)
With three inputs and three outputs, there will be nine transfer functions representing the
plant. To facilitate production of the transfer functions, the lines of (3.8) are put into the
Laplace domain and solved for each of the state variables:
( ) ( ) ' ( )( ) cin
in
cL
v s d s V D v si s
sL R
(3.10)
( )' ( ) ( )
( )1
ocload
battL
c
eq
v sD i s i s
Rv s
sCR
(3.11)
( )
( )inuc
Li sv s
sC
(3.12)
The input-to- ( )cv s transfer functions are found by eliminating ( )Li s and ( )inv s from
(3.10)–(3.12), leaving ( )cv s in terms of the input variables. First, ( )inv s is eliminated by
substituting (3.12) into (3.10). The resulting expression is solved for ( )Li s and plugged
into (3.11). Solving this last expression for ( )cv s gives the transfer functions
48
13
1
13
2
2 2
23
2
( ) '( )
( ) '
( )( )
( ) '
( ) 1( )
( )
ucvd LC Ruc in
uc uc in uc ucR R Req eq eq
Rbattvo LC Ruc inoc
uc uc in uc ucR R Req eq eq
vl LCucloaduc uc inReq
c c
c
c
v s sC D VG s
d s s LCC s CC R s C C D C
v sG s
v s s LCC s CC R s C C D C
v sG s
i s s LCC s CC R s
2 1'Rin
uc uc R Req eqC C D C
(3.13)
The input-to- ( )Li s transfer functions are found by plugging (3.11) and (3.12) into (3.10)
and solving for ( )Li s :
2
2 2
2 2
13
'
13
3
( )( )
( ) '
( )( )
( ) '
'( )( )
( )
V
uc uc Reqid LC Ruc in
uc uc in uc ucR R Req eq eq
Duc Rbatt
io LC Ruc inocuc uc in uc ucR R Req eq eq
ucil
lo
cc
L
L
L
aduc
s C CV sCi sG s
d s s LCC s CC R s C C D C
sCi sG s
v s s LCC s CC R s C C D C
sC Di sG s
i s s LCC
2 12 '
LC Ruc inuc in uc ucR R Req eq eq
s CC R s C C D C
(3.14)
Lastly, the input-to- ( )inv s transfer functions are found by substituting (3.11) into (3.10)
to eliminate ( )cv s , solving for ( )Li s , plugging the result into (3.12), and solving for
( )inv s :
49
13
'
13 2
3
2 2
2
2
( )( )
( ) '
( )( )
( ) '
( ) '( )
( )
V
Reqinud LC Ruc in
uc uc in uc ucR R Req eq eq
D
Rin battuo LC Ruc inoc
uc uc in uc ucR R Req eq eq
inul LCucload
uc Re
cc
q
sCVv sG s
d s s LCC s CC R s C C D C
v sG s
v s s LCC s CC R s C C D C
v s DG s
i s s LCC s C
2 1'Rin
uc in uc uc R Req eqC R s C C D C
(3.15)
For the 3-state model, steady-state cV is calculated from
oca t
cb t
RV V
R R
(3.16)
since 0LI in steady state.
Each of the nine transfer functions has the same denominator. The ( )ocv s -to-
output transfer functions are equal to the ( )loadi s -to-output transfer functions scaled by
1 battR . Looking back at Figure 3.1, cv moves in response not just to changes in load
current, but also to changes in battery current. A change in ocv makes the voltage across
battR change, with batti and cv following. From an observability standpoint, the result here
says that a controller using a measurement of cv as an input cannot distinguish between a
change in load and a change in open-circuit voltage. This controller cannot inject current
to the bus in response to increased load current without also injecting current when open-
circuit voltage drops, without using an additional measurement.
Figure 3.2 shows ( )vdG s , ( )idG s and ( )udG s using parameters given in Table 3.1
(on page 51). Parameters are chosen with regard to the design considerations laid out for
the 100 kW example given in Chapter 2. The operating point is at full load with a fairly
high boost ratio; curve magnitudes will represent a worst-case operating condition but the
curve shapes are general to all stable operating points. Two of the poles are near the L-C
50
resonance and one is at a lower frequency near 'eq ucR D C . From (3.13)–(3.15) all are in
the left half plane (LHP). The L-C resonance is heavily damped by the battery, as battR
(and thus eqR ) are expected to be relatively small. Functions vdG and idG have transmis-
sion zeros attributed to the UC. This is also expected, as the UC is being modeled as a
capacitor, with associated finite energy storage.
-100
-50
0
50
100
Mag
nitu
de (
dB)
10-3
10-2
10-1
100
101
102
103
-180
-90
0
90
180
Pha
se (
deg)
Bode diagram of 3-state Gvd, Gid, Gud
Frequency (Hz)
GvdGidGud
Figure 3.2. Plot of three-state Gvd, Gid, Gud.
51
Table 3.1. Parameters used to generate small-signal model results.
Circuit Parameters Controller Gains Operating Point Parameter Value Parameter Value Parameter Value
L 500 µH pK 65
ocV 400 V
C 18,000 µF 1 3.71
inV 180 V
ucC 44.16 F R 1.6 Ω
inR 21.1 mΩ chgK 0.4
battR 0.15 Ω
T 1*10-4 s ichgK 0.00015
battI (2-state) 30 A
aM 0.7*M2 2 0.0291
2M (400-180)/L cV (3-state) 366 V
cV (2-state) 396 V
All plotted transfer functions have a very high gain due to the high nominal bus
voltage cV . The phase of ( )vdG s and also of ( )udG s wraps to almost 180° at the gain
crossover points; introducing linear, unity feedback from either cV or inV directly to the
duty ratio input would likely result in an unstable system. The phase of ( )idG s , in con-
trast, wraps from 90° to –90° everywhere; feedback of ( )Li s to the duty ratio input
should present no problems.
3.1.2 Model of PCM Process
From an averaged model perspective, PCM control commands the averaged in-
ductor current, Li , to track the averaged reference current, refi , with some offset [49]. An
LTI model of the relationship between Li and refi was originally given by Verghese et al.
in [50]. Results are given for a boost converter in [49]. This model is reasonably accurate
at frequencies up to one-tenth the switching frequency [50] even though the boost con-
verter model is known to have dc error. A skeletal derivation follows.
A graphical relationship between instantaneous ( )Li t and ( )refi t over one switch-
ing cycle is presented in Figure 3.3. The relationship is general to dc-dc converters. The
slope of the positive-going inductor current is 1m , 2m of the negative-going inductor
52
current, and am of the compensation ramp. Duty ratio is presented as the steady-state
duty ratio, D, plus a time perturbation, d . As the converter is not in steady state, ( )Li T is
not equal to (0)Li . This is reflected in the line segment 0
T
L
D di
being unequal to dL
T
D Ti
,
where
0( ) (0)
( ) ( )
L L L
L L
D d T
T
D Ld T
i i D d T i
i i T i D d T
-ma
ma
-m2m1
m1
iref
iL(t)
0 T(D+d)T
iL
(D+d)T2
(D+d)T
m2 (D+d)T2
t
(D+d)T
0
iL (D+d)TT
Figure 3.3. Relationship of iL to iref.
Looking at the plot, the distance between averaged Li and averaged refi can be written as
the distance ( )am D d T plus the weighted sum of the current ripple averaged over each
subinterval:
1 2( ) ( ' )( ) ( ) ( ' )
2 2ref L a
m D d T m D d Ti i m D d T D d D d
. (3.17)
53
Next, the variables Li , refi , 1m and 2m are linearized and perturbed. Ramp slope am is
not perturbed because its slope is programmed and thus constant; it will be replaced by
the dc value aM . Neglecting the dc and higher-order terms, noting that 1 2'DM D M , and
solving for d , (3.17) becomes
2 2
1 2
1 '
2 2ref La
D T D Td i i m m
M T
. (3.18)
The next step is to put the perturbed inductor current slopes 1m and 2m in terms of per-
turbations in the voltage across the inductor. For a boost converter, the relationships are
21in c inv v v
m mL L
.
Substituting these into (3.18) and grouping terms gives
21 (2 1) '
2 2ref L in ca
D T D Td i i v v
M T L L
. (3.19)
The model (3.19) allows d to be written in terms of perturbed state variables
, , and c L inv i v and a new input variable refi . Using notation in (3.9), (3.19) in the Laplace
domain is
( ) ( ) ( ) ( ) ( )im ref L inn t coud s F i s i s F v s F v s (3.20)
where
21 (2 1) '
2 2m in ta
ou
D T D TF F F
M T L L
. (3.21)
This controller equation is “wrapped” around the plant transfer functions (3.13)–(3.15) to
create closed-inner-loop transfer functions. The exact steps follow.
54
3.1.3 Current Loop
The controller equation (3.20) is closed around the plant by use of algebraic sub-
stitution guided by the block diagram shown in Figure 3.4. The substitution procedure
follows that in [49]; the differences are that ( )inv s is a state variable and there are addi-
tional inputs ( )ocv s and ( )loadi s . The closed-loop transfer functions of interest are the
input-to- ( )cv s , input-to- ( )inv s , and ( )refi s -to- ( )Li s functions. The former functions pro-
vide inputs to the voltage feedback loops. The ( )refi s -to- ( )Li s function is used to check if
a chosen ramp compensation slope aM allows for adequate current tracking. It is as-
sumed that all variables are in the Laplace domain; the (s) notation is dropped.
Open-loop Plant
-+
-
+
- +
++
+
++
+
++
Figure 3.4. Block diagram of inner current loop.
55
From Figure 3.4, superposition gives
id iu oc il loaL di G d G v G i (3.22)
vd vo oc vl loac dv G d G v G i (3.23)
in ud uo oc ul loadv G d G v G i (3.24)
The open-loop refi -to- Li transfer function will be an expression with the Li feedback path
open and the cv and inv paths closed. To close the cv and inv paths, plug (3.23) and
(3.24) into (3.19) and solve for d :
1
ref in uo oc ul load out vo oc vl load
m in ud m o t vd
m L
u
F i i F G v G i F G v G id
F F G F F G
Substituting into (3.22) and solving for Li gives
1
m idLir
ref im d m in ud m out vd
F GiH
i F G F F G F F G
. (3.25)
Figure 3.5 plots irH for the vehicle application. Current Li closely tracks refi between
0.001 Hz and 1 kHz. In a real system Li will still track refi at the lowest frequencies pro-
vided that the UC has not run out of energy. The plot reinforces the notion that the UC
has a charge limit.
The input-to- cv transfer functions are obtained by eliminating Li and inv from
(3.19), plugging this into (3.23), then solving for cv . Solving for cv closes this last loop.
The expression for d with (3.22) and (3.24) plugged in is
1ref io in uo oc il in ul lm coad out
m in ud m out vd
F i G F G v G F G i F vd
F F G F F G
after grouping the input terms together. The resulting input-to- cv transfer functions are
56
-60
-50
-40
-30
-20
-10
0
10
Mag
nitu
de (
dB)
10-6
10-4
10-2
100
102
104
106
-90
-45
0
45
90
Pha
se (
deg)
Bode diagram of 3-state Hir
Frequency (Hz)
Figure 3.5. Plot of three-state Hir.
1
1
1
vdvr
ref id m in ud out vd
vo vd io in uovo
oc id m in ud out vd
vl vd il in ulvl
load id m in ud out v
c m
m m
mc
m m
mc
m m d
v F GH
i F G F F G F F G
G F G G F GvH
v F G F F G F F G
G F G G F GvH
i F G F F G F F G
(3.26)
The input-to- inv transfer functions are obtained in a similar fashion. The results are
1
1
1
in udur
ref id m in ud out vd
uo ud io out voinuo
oc id m in ud out vd
ul ud il out vlinul
load id m in ud out vd
m
m m
m
m m
m
m m
v F GH
i F G F F G F F G
G F G G F GvH
v F G F F G F F G
G F G G F GvH
i F G F F G F F G
(3.27)
Again, all lowercase variables are functions of s in the Laplace domain.
57
Figure 3.6 plots vrH and urH . As function vrH has a low gain, feeding the bus
voltage back to refi should be stable. Also, the higher-frequency poles are spread apart in
frequency, as is typical for current-mode control [49],[51]. The urH plot shows the UC’s
sensitivity to low-frequency current commands. Because sensitivity is high at low fre-
quencies, only a small amount of charging loop gain is needed to drive the UC to the de-
sired charge level. Furthermore, the slope of urH is first order between 0.001 Hz and 1
kHz and constant below 0.001 Hz. Placing an HPF on the function which sets the domi-
nant portion of refi (in this case the power-splitting cv -to- refi loop) is sufficient to enforce
a charge limit on the UC in response to a change in cv . That is, if the load current
changes by X amount, the UC charge level will change no more than Y. The tools in
Chapter 2 and the averaged nonlinear model presented in Chapter 4 provide a way to find
what the UC discharge will be in response to a particular load. This approach is different
from that taken by Awerbuch and Sullivan [25],[52] and others [22],[23] where the UC
charge is held to a hard limit Y regardless of load.
-250
-200
-150
-100
-50
0
50
Mag
nitu
de (
dB)
10-6
10-4
10-2
100
102
104
106
-180
-90
0
90
180
Pha
se (
deg)
Bode diagram of 3-state Hvr, Hur
Frequency (Hz)
HvrHur
Figure 3.6. Plot of three-state Hvr, Hur.
58
3.1.4 Outer Loop
Equations (3.26) and (3.27) describe Figure 3.1 with the current loop closed
around it. The power-distributing cv -to- refi and charging inv -to- refi voltage loops can
then be wrapped around this model. Figure 3.7 gives the block diagram for this connec-
tion. The high-pass and low-pass filters are included in the block labeled BPF(s), while
the PI controller for the charging loop is in the block labeled Kc(s). The final closed-loop
system will have disturbance inputs ocv and loadi , UC voltage reference ,uc refv , and out-
puts inv and cv .
+
+
+
+
-+
+
+
Plant with CurrentLoop Closed
+
+
Figure 3.7. Block diagram of outer voltage loops.
59
From Figure 3.7, the input-to-output relationship of the controller is
,( ) ( )p cref c uc fn reii HPF s K v K s v v . (3.28)
Equations of the plant with current loop closed are
vr ref vo oc loac vl dv H i H v H i (3.29)
in ur ref uo oc ul loadv H i H v H i (3.30)
Closed-loop input-to- cv transfer functions are obtained by plugging (3.30) into (3.28),
solving for refi , then plugging this result into (3.29) and solving for cv . The closed-loop
input-to- inv transfer functions are found in a similar fashion. The results are
,
( )
1 ( ) ( )
1 ( ) ( )
1 ( ) ( )
1 ( ) ( )
1 ( ) ( )
vrvr
uc ref ur vr
vo ur c uo vrvo
oc ur vr
vl ur c ul vrvl
load ur
c c
c p
cc
c p
cc
c p vr
v K s HJ
v K s H BPF s K H
H K s H K s H HvJ
v K s H BPF s K H
H K s H K s H HvJ
i K s H BPF s K H
(3.31)
,
( )
1 ( ) ( )
1 ( ) ( )
1 ( ) ( )
1 ( ) ( )
1 ( ) ( )
urur
uc ref ur vr
uo vr vo urinuo
oc ur vr
ul vr vl urinul
load ur
in c
c p
p p
c p
p
c v
p
p r
v K s HJ
v K s H BPF s K H
H BPF s K H BPF s K H HvJ
v K s H BPF s K H
H BPF s K H BPF s K H HvJ
i K s H BPF s K H
(3.32)
The transfer functions of blocks ( )BPF s and ( )cK s are
60
1 12
1 2 1 2 1 2
1( )
1 1 1
( ) ichg chg ichgc chg
s sBPF s
s s s s
K sK KK s K
s s
(3.33)
where gains 1 2, , and chg ichgK K are those defined in Section 2.1.
At first the ( ) uc rK s H term in the denominator of (3.31)–(3.32) looks problem-
atic. However, the numerator of urH is um dF G , and from (3.15) udG has a negative nu-
merator. Thus the negative signs cancel and (3.31)–(3.32) will be stable whenever the H’s
are stable.
Figure 3.8 plots uoJ for various loop configurations; Figure 3.9 does the same for
ulJ . Figure 3.10 plots vlJ , the magnitude of which is the dynamic output impedance (the
impedance of the combined source in parallel with the load). In Figures 3.8 through 3.10
the frequency range where the UC is being controlled to inject the most power is marked
2nd src contrib. The cutoff points are those of ( )BPF s in (3.33) scaled by 1K , K de-
fined in (2.7). The output impedance rises at higher frequencies due to the LPF attenuat-
ing refi at frequencies below which the bus capacitor may provide energy. At the lowest
frequencies output impedance is equal to eqR .
61
-150
-100
-50
0
50
100
Mag
nitu
de (
dB)
2nd src contrib
10-6
10-5
10-4
10-3
10-2
10-1
100
101
102
103
-180
-90
0
90
180
Pha
se (
deg) 2nd src contrib
Bode diagram of Juo with various outer loop configurations
Frequency (Hz)
No HPF, No Kc(s)With HPF, No Kc(s)With HPF and Kc(s)
Figure 3.8. Plots of three-state Juo.
-200
-150
-100
-50
0
50
Mag
nitu
de (
dB) 2nd src contrib
10-6
10-5
10-4
10-3
10-2
10-1
100
101
102
103
-180
-90
0
90
180
Pha
se (
deg)
2nd src contrib
Bode diagram of Jul with various outer loop configurations
Frequency (Hz)
No HPF, No Kc(s)With HPF, No Kc(s)With HPF and Kc(s)
Figure 3.9. Plots of three-state Jul.
62
-80
-60
-40
-20
0
Mag
nitu
de (
dB) 2nd src contrib
10-6
10-5
10-4
10-3
10-2
10-1
100
101
102
103
-180
-90
0
90
180
Pha
se (
deg)
2nd src contrib
Bode diagram of Jvl with various outer loop configurations
Frequency (Hz)
No HPF, No Kc(s)With HPF, No Kc(s)With HPF and Kc(s)
Figure 3.10. Plots of three-state Jvl.
For battR that is much less than load resistance R (true for all practical batteries),
changes in ocV have a strong effect on cV . This in turn will move the UC charge level via
the power-distributing loop. With no HPF, inv (and UC charge level) is very sensitive to
ocv and loadi disturbances. Adding the HPF with 1 3.71 reduces sensitivity at the low-
est frequencies (over a timespan of hours or days) but leaves a peak at around 0.5 mHz.
Introducing the charging loop with 0.4chgK and 0.00015ichgK attenuates the peak to
just below 0 dB without disturbing output impedance in and around the passband. In-
creasing the charging loop gain moves the peak lower in magnitude and up in frequency
but starts to impact vlJ within the 2nd src contrib region.
At low and very high frequencies, the phase of vlJ , uoJ and ulJ wraps considera-
bly. It was noted in [52] that a first-order HPF causes the UC phase in response to load
disturbances to wrap beyond where it should be, resulting in the UC contributing nega-
63
tively to load current at very low frequencies. The uoJ and ulJ plots support this observa-
tion; the phase of both transfer functions is advanced 90° near 0.001 Hz with no filter,
compared to phase in the “2nd src contrib” region. Phase is advanced about 180° with the
HPF installed. However, the magnitude of any current command at 0.001 Hz will be very
small with 1 3.71 . In addition, adding the HPF flattens the magnitude and phase of
vlJ , uoJ and ulJ across the current command passband. This is a performance improve-
ment, since the current delivered from the UC is more in phase with the load current in
this region.
3.2 Small-Signal Model with Two-State Plant
In Section 3.1.1, it was seen that the three-state plant shown in Figure 3.1 was in
equilibrium only when steady-state inductor current LI was equal to zero. On a fast time
scale, the fast dynamics of the time-averaged plant may be very close to equilibrium
when inductor current is nonzero. The small-signal averaged model of a typical boost
converter exhibits a right-half-plane (RHP) zero in the d -to- cv transfer function that is
attributed in part to dc inductor current [53]. This RHP zero carries through to the closed-
current-loop refi -to- cv transfer function, limiting feedback gain pK compared to what is
possible with a buck topology.
Figure 3.11 shows the system in Figure 3.1 with the ultracapacitor replaced by a
voltage source. The state equations with active switch q ON are
( )( ) ( ) ( )
( ) ( ) ( ) ( )( ) ( )
LL L
c
in in
oc cload
b
cc
att
di tv t L v t i t R
dtdv t v t v t v t
i t C i tdt R R
(3.34)
With the active switch OFF the equations are
64
( )( ) ( ) ( ) ( )
( ) ( ) ( ) ( )( ) ( ) ( )
LL L c
c
in in
oc cload
ba
cc
ttL
di tv t L v t i t R v t
dtdv t v t v t v t
i t C i t i tdt R R
(3.35)
voc(t)
ibatt(t)
Rbatt
RiL(t)
iload(t)
ic(t)
LC
vin(t)
vc(t)vL(t)
q
q’
RL
(distur-
bance
input)
(load)
Figure 3.11. Plant with voltage source in place of ultracapacitor.
The time-average of (3.34) and (3.35) is
'
'
in in
ocl
LL L c
c cc oad
batt eqL
div L v i R d v
dtdv v v
i C d i idt R R
(3.36)
where eqR is the parallel combination of R and battR . The variables to be perturbed are
the same as those for the three-state plant. Substituting (3.5) into (3.36), with 0loadI ,
gives
' '
' '
in in L in inL L
c L c c
c c c cL L L
oc ocload
batt eq batt eq
dI diL V R I D V v R i D v dV
dt dt
d dv V V v vC D I D i dI i
dt dt R R R R
V
(3.37)
65
With the converter in equilibrium, the dc relationships listed in (3.7) hold for the
two-state model. Thus all pure dc terms in (3.37) are canceled. The averaged ac small-
signal system has four inputs, , , , and in oc loadd v v i , and two state variables, cv and Li .
Transfer functions are found by placing the remainder of (3.37) into the Laplace domain
and substituting one equation into the other to get the appropriate output variable in terms
of the input variables. The open-loop plant transfer functions are
2
2
1
2
2
2
2
2
( ) '( )
( ) '
( ) '( )
( ) '
( )( )
( ) '
( )( )
( )
invd RL in
inR Req eq
vu RL inininR Req eq
inRbattvo RL inoc
inR Req eq
invl
Lload
Re
c L c L
c
c
c
v s sLI V D R IG s
d s s LC s CR D
v s DG s
v s s LC s CR D
sL Rv sG s
v s s LC s CR D
sL Rv sG s
i s s LC s
2'Rin
in Rq eqCR D
(3.38)
66
2
1
2
'
2
2
2
2
2
'( )( )
( ) '
( )( )
( ) '
( )( )
( ) '
( ) '( )
( )
V
Reqid RL in
inR Req eq
Reqiu RL inin
inR Req eq
D
Rbattio RL inoc
inR Req eq
ilL
loadiRe
cc L
L
L
q
L
L
sCV D Ii sG s
d s s LC s CR D
sCi sG s
v s s LC s CR D
i sG s
v s s LC s CR D
i s DG s
i s s LC s CR
2'
Rinn Req
D
(3.39)
At a particular operating point , , , ,oc in batt battV V I R R , cV , LI and 'D are defined by
'
c
c cL
in L
oc batt batt
battin
in in L
c
V V I R
V VI I
V R I R
V R ID
V
(3.40)
where LI may be solved implicitly or approximated by setting inR to zero.
The transfer function vdG has an RHP zero for all practical values of inR . Figure
3.12 plots vdG with a high LI and other parameters typical to the vehicle application.
Here it is seen that even a small feedback from ( )cv s to ( )d s will destabilize the plant if
no other compensation or feedback is introduced.
67
0
10
20
30
40
50
60
70
80
Mag
nitu
de (
dB)
10-1
100
101
102
103
-270
-180
-90
0
Pha
se (
deg)
Bode diagram of 2-state Gvd, Gid
Frequency (Hz)
GvdGid
Figure 3.12. Plot of two-state Gvd, Gid.
The linearized model of the PCM process derived in Section 3.1.2 will be applied
to the small-signal model of the two-state plant. The block diagram of the resulting sys-
tem is shown in Figure 3.13. Variables and blocks are in the Laplace domain; the (s) no-
tation is dropped. Either block diagram manipulation or algebraic substitution gives the
closed current loop refi -to- Li transfer function
( )1
idir
id out vd
m
m m
F GH s
F G F F G
. (3.41)
The closed-current-loop input-to- cv transfer functions are
68
Open-loop Plant
-+
-
+
-++
+
+
+
+
+
+
Figure 3.13. Block diagram of inner current loop for two-state plant.
1
1
1
1
1
1
1
c m
m m
m mc
m m
m mc
vdvr
ref id out vd
vu id vd iu invu
in id out vd
vo id vd iovo
oc id out vd
vl id vd i
m m
m mc lvl
load id out vdm m
v F GH
i F G F F G
G F G F G G FvH
v F G F F G
G F G F G GvH
v F G F F G
G F G F G GvH
i F G F F G
(3.42)
69
With vdG in the numerator of vrH , vrH also has an RHP zero. Root locus plots show the
effect of the zero when feedback from cv to refi is added. Figure 3.14 is the root locus of
vrH (using vdG plotted in Figure 3.12) with no filtering in the feedback path. A pK
greater than about 28 makes the closed-loop system unstable. This is a rather small pK ; it
corresponds to a K of about 1.9 per (2.7). Adding an LPF to the voltage feedback path is
one solution to the problem. Figure 3.15 plots the root locus of vrH with such a filter in
place. The peak allowed pK is now 377, and a pK of 65 results in a well-damped re-
sponse to a load step. The prospect that voltage feedback into the current command of a
current-mode converter could be unstable was brought up in [25]; models here verify that
this is the case and the problem is addressed by adding an LPF and choosing appropri-
ate pK .
-3 -2 -1 0 1 2 3 4
x 104
-6000
-4000
-2000
0
2000
4000
6000
System: Hvr_no_LPFGain: 28.6Pole: 31.2 + 5.25e+003iDamping: -0.00594Overshoot (%): 102Frequency (rad/sec): 5.25e+003
Root Locus of Hvr without LPF in feedback path
Real Axis
Imag
inar
y A
xis
Figure 3.14. Root locus of two-state Hvr, no filters in feedback path.
70
-1000 -500 0 500 1000 1500-1000
-500
0
500
1000System: Hvr_with_LPFGain: 377Pole: 0.433 + 546iDamping: -0.000793Overshoot (%): 100Frequency (rad/sec): 546
System: Hvr_with_LPFGain: 65Pole: -195 - 164iDamping: 0.766Overshoot (%): 2.38Frequency (rad/sec): 254
Root Locus of Hvr with LPF; 2 = 0.0291
Real Axis
Imag
inar
y A
xis
Figure 3.15. Root locus of two-state Hvr, LPF in feedback path.
The last step is to obtain the complete system sensitivity functions, with the volt-
age loop closed. Because the HPF typically has a large time constant, it will have little
effect at higher frequencies. More importantly, the open-loop pole in the HPF is no
longer counteracted by the transmission zero attributed to ucC in the plant input-to- cv
equations, so including the HPF on the two-state model will actually produce an errone-
ous result. Therefore, the HPF is neglected from the expression. The charging loop is
neglected for the same reasons. Figure 3.16 shows the remaining connection.
71
Plant with Current Loop Closed
+ +
++
Figure 3.16. Block diagram of the portion of the outer voltage loop that is dominant at high
frequencies.
Following the procedure outlined in Section 3.1.4, the closed-loop input-to- cv
transfer functions are
1 ( )
1 ( )
1 ( )
vuvu
in vr
vo
c
p
c
p
c
vooc vr
vlvl
load vr p
v HJ
v H LPF s K
v HJ
v H LPF s K
v HJ
i H LPF s K
(3.43)
where
2
1( )
1LPF s
s
. (3.44)
Figure 3.17 plots voJ defined in (3.43) against its three-state equivalent; Figure 3.18 does
the same for vlJ . The point marked “LP corner” in Figure 3.17 is the corner frequency of
72
the LPF scaled by 1K , where K is defined in (2.7). Function vlJ is the dynamic output
impedance of the closed-loop system, valid at higher frequencies. These plots show that
cv is considerably more sensitive to disturbances ocv and loadi when the converter is
boosting current ( 0LI ) than when 0LI . The hump in voJ and vlJ around 40 Hz is
attributed to the “gap” where neither the bus capacitor nor the second source is providing
energy to the load. It is created by the choice of LPF coefficient 2 and bus capacitance.
The output impedance vlJ rises to eqR at the peak of the hump. Finally, Figure 3.19 plots
, and vu vu vuG H J . These functions correspond to audio susceptibility in a boost converter
[53] and follow the same basic shape as the line-to-output transfer functions [53],[49].
The double pole on vdG is damped by small eqR .
-40
-30
-20
-10
0
Mag
nitu
de (
dB)
LP corner
10-1
100
101
102
103
-180
-90
0
90
180
Pha
se (
deg)
LP corner
Bode diagram of 2-state, 3-state Jvo
Frequency (Hz)
2-state Jvo3-state Jvo
Figure 3.17. Plots of two-state and three-state Jvo.
73
-60
-50
-40
-30
-20
-10
Mag
nitu
de (
dB)
LP corner
10-1
100
101
102
103
-180
-90
0
90
180
Pha
se (
deg)
LP corner
Bode diagram of 2-state, 3-state Jvl
Frequency (Hz)
2-state Jvl3-state Jvl
Figure 3.18. Plots of two-state and three-state Jvl.
-60
-40
-20
0
20
Mag
nitu
de (
dB)
LP corner
10-1
100
101
102
103
-180
-135
-90
-45
0
45
Pha
se (
deg)
LP corner
Bode diagram of Gvu, Hvu, Jvu
Frequency (Hz)
GvuHvuJvu
Figure 3.19. Comparison plot of the two-state Gvu, Hvu, Jvu.
74
3.3 Stability Check
The small-signal models provide a simple way to see whether a particular operat-
ing point is stable — just check to see whether or not all poles are in the open LHP
(OLHP). A MATLAB script which computes the various G, H and J transfer functions
and checks their poles was written and is documented in Appendix A. To determine sys-
tem stability, the script was run over the range of operating points given in Table 3.2. The
two-state models were closed-loop stable at all tested points, with both forward and re-
verse LI . The three-state models were closed-loop stable everywhere except at the ex-
treme operating point of 140 VinV , 420 VocV , 0.1 battR and any R. System sta-
bility is maintained up to the target 2.5:1 maximum boost ratio and across all load cur-
rents with the choice of circuit parameters and controller gains.
Table 3.2. Operating points at which system stability was checked.
Parameter Values Checked
inV 140 to 320 V, 20 V steps
inV 300 to 420 V, 20 V steps
R 1.6 Ω, 16 Ω, 1 MΩ
battR 0.1, 0.15, 0.75 Ω
loadI (2-state only) 0, 14, 147 A
3.4 Summary
The small-signal models derived in this chapter are useful tools for analyzing the
system proposed for the vehicle application. The three-state model shows that an HPF in
the cv feedback path is capable of enforcing a charge limit on the UC for a given change
in cv , and that the charging loop makes UC voltage (state-of-charge) less sensitive to
very low frequency changes in cv . The two-state model shows that too high a pK can
drive the converter unstable on account of the RHP zero, and filtering the current com-
75
mand with an LPF increases the permissible pK . The controller cannot distinguish
whether changes in cv result from a change in load current or a change in ocv ; and the
models showed that small changes in ocv can have a dramatic effect on UC charge level.
System stability is validated for the choice of circuit parameters and controller gains for
this application. The small-signal plant models provide a solid platform on which to
evaluate other controller designs.
76
CHAPTER 4
LARGE-SIGNAL MODELS
4.1 Switched Model for Vehicle System
The power-handling portion of the combined source and controller is viewed as a
switched system comprised of two subsystems (3.1) and (3.2). Both (3.1) and (3.2) are
linear ordinary differential equations (ODEs). The system is solved by computing the
state trajectories according to (3.1) for a period of time, then switching over to (3.2) and
continuing the computation. Alternating between the two, the final state values of one
equation are used as initial values of the other equation. Switching instants are decided by
a switching signal created by another routine, in this case a stateful model of the Control-
ler portion of Figure 2.1. In this fashion the system may be solved out to an arbitrary
point in time given some initial condition at time zero.
The switched model is derived and documented in Section 4.1.1. Solver step size
is discussed in Section 4.1.2 and the state initialization routine is given in Section 4.1.3.
Simulation results are presented in Section 4.1.4.
4.1.1 Model Derivation
The Simulink package was used to create the switched model. Plant equations
(3.1) and (3.2) were modified to include the capacitor ESR. They are
in in
oc esr esrload
ba
LL L
c c c ccc
tt
inuc uc L
div L v R i
dtv v i R v i Rdv
i C idt R R
dvi C i
dt
(4.1)
77
when the active switch is ON and
LL L c c
c c
in in esr
oc esr esc rload
batt
inuc
cc
uc L
cL
div L v R i v i R
dtv v i R v i Rdv
i C i idt R R
dvi C i
dt
(4.2)
when the active switch is OFF. Simulink handles these implicitly-defined functions by
using the previous value of ci to compute the next value of state variables , , and c L inv i v .
An overview of the switched model is presented in Figure 4.1. The blocks on the right-
hand side of the diagram represent disturbances that may be injected into the system.
Next, the controller block is expanded as shown in Figure 4.2. The upper half of this dia-
gram contains the outer voltage loops and the PCM process is modeled in the lower half.
The active switch is turned ON by a positive clock pulse and turned OFF when ( )Li t
meets or exceeds ( )refi t . Controller sub-blocks are broken out in Appendix B.1. The
boost converter plant model is based upon work done by Logue [54],[55].
78
ucap capa-citance
-K-
Vuc, Vout
Voc step
TransportDelay
Manual Switch5Manual Switch4
Manual Switch3Manual Switch1
Manual Switch
ManualSwitch2
Loadstep
Loadramp
Iout, Ibatt, Iload
Integrator
1s
Iload step
Iload ramp
Iload pulse
IL, Iout, Ibatt, Iload, Iref Fixed iload
0
Fixed R
1e6
Divide
Controller
Vc
I_ind
Vin
Scope out
q(t)
Boost converter(Bidirectional;
Lossy switching)
vin
iout
q(t)
vout
iin
Batt open-ckt voltage
Voc
1/Rbatt
-K-
Figure 4.1. Overview of switched model of the combined source for an electric vehicle.
Iref
q(t)
2
Scope out
1
Vuc,ref
-C-
Terminator
Stabilizing Ramp
SR latch
S
R
Q
Q'
P
Kp
Manual Switch
Low corner
In1 Out1
Kichg
-K-
Kchg
-K-
Integrator1
1
s
Fixed Iref
-C-
Comparator
+
-
Out
Clock
BPF
In1 Out1
-BPF'd Vc
Vin
3
I_ind
2
Vc
1
Figure 4.2. Controller block of Figure 4.1, expanded.
79
4.1.2 Step Size Limitation
Because a numerical solver must break the continuous time domain into discrete
steps, most of the time the ON-to-OFF switching transition will happen when ( )Li t is not
exactly equal to ( )refi t . This represents an estimation error compared to what would hap-
pen in an idealized continuous-time system. Recall from the small-signal model deriva-
tion in Section 3.1.1 that there is a high gain between duty ratio and state variables cv and
Li . A small error in the estimation of a switching instant will result in large state trajec-
tory errors. When running the switched model, the maximum solver step size needed to
be set to one one-thousandth of a switching period ( 710 seconds for a 10 kHz switching
frequency) in order to get acceptable accuracy. The solver dropped its step size further at
the switching instants, where the derivatives of the state trajectories are discontinuous.
The small step size made it impractical to execute more than a few seconds’ worth of
simulation time. The switched model is very useful for verifying fast transient behavior,
but is impractical for analyzing behavior which occurs over long periods of time.
4.1.3 State Initialization Procedure
The states of the switched model are never in equilibrium. When the system is
considered to be in steady state, plant and controller states follow a limit cycle trajectory
[56],[57]. Initial state values need to be chosen such that the model starts in this limit
cycle trajectory with no disturbances. The states which most strongly determine whether
or not the model is in a limit cycle are , L ci v and the state of the integrator which follows
gain ichgK (the charging loop integrator) in Figure 4.2, and so this section focuses on
finding the initial values of these states.
80
When the UC is in steady state, it supplies no power to the load. This means that
average LI is zero and average cV is
oca t
cb t
RV V
R R
. (4.3)
If no net power is transferred to the load, after one switching cycle no net charge is trans-
ferred from the inductor to the capacitor. This means
0
0 0
( ) 0
( )
T
T T
oc ocbatt batt
L
c
i t dt
R Rv t dt V V T
R R R R
. (4.4)
A numerical solver is used to compute the trajectory of ( )Li t and ( )cv t over one
switching cycle, using model equations (4.1) and (4.2). An implicit solver is used due to
the presence of ci on the right-hand side of the equations. The switching instant occurs at
time DT, where D is the steady-state duty ratio
1 in in
esr batt lo
L
L adc
V R ID
V R I I I
. (4.5)
Here, LI , battI and loadI are ( )Li t , ( )batti t and ( )loadi t , respectively, averaged over one
switching cycle. Since 0LI and obatt l adI I in pure steady state, D reduces to
1 i
c
nVD
V . (4.6)
To get initial states (0)Li and (0)cv , the solver is executed for various initial conditions
until a set is found where both
(0) ( )
(0) ( )L L
c c
i i T
v v T
(4.7)
81
and equation (4.4) hold. The initial state of the charging loop integrator, ,ref initI , is then
given by
, ( )ref in t ai LI i DT m DT . (4.8)
Current ( )Li DT is the inductor current found by the solver at time DT using initial condi-
tions (0)Li and (0)cv , and am is the slope of the compensation ramp required to ensure
the stability of the PCM process [51].
4.1.4 Simulation Results
A sequence of full-scale load steps was performed on the switched model. The
resulting current trajectories are drawn in Figure 4.3. An instantaneous 250 A or 500 A
load step is an abusive condition that should never be seen under normal operation; the
test is used to see how Li responds to large changes in termv . Setting the LPF filter coeffi-
cient 2 to 0.0291 makes Li follow a damped second-order trajectory. This confirms
what was predicted in Figure 3.15 using the two-state small-signal model. The lower por-
tion of Figure 4.3 is a close-up of Li and refi trajectories during the steepest rise in induc-
tor current at 305 ms. The duty ratio is about 0.65 in this graph, near the steady-state D of
0.55 and well within the 0.85 limit. The 18,000 µF bus capacitor prevents the LPF from
seeing a true step, making (2.27) inaccurately predict the relationship between the duty
ratio swing and 2 .
82
0 50 100 150 200 250 300 350 400 450 500
-400
-200
0
200
400
600
Cur
rent
(A
)
Time (ms)
iL
ibatt
load I
304 304.2 304.4 304.6 304.8 305 305.2 305.4 305.6 305.8 306-200
-150
-100
-50
0
50
Time (ms)
Cur
rent
(A
)
iL
iref
Figure 4.3. (Top) Response of iL, ibatt to load step. (Bottom) Close-up of iL rise after 500 A step up.
Figure 4.4 shows current and voltage trajectories when a load ramp is applied.
The ramp has the peak magnitude and slew that the load is expected to demand in prac-
tice. The plot shows that batti is well-controlled; the control bandwidth allowed by 2 is
sufficient to make the second source absorb the fast components of the ramp. The trajec-
tories, especially termv in the lower plot, vary in thickness with load because the switched
model outputs both the ripple and averaged portions of the variables. Twenty thousand
switching cycles elapse over two seconds, and so the line thickness corresponds to ripple
magnitude. The termv plot in the lower portion of Figure 4.4 shows that voltage ripple is
about 2.5 V and within design specification.
83
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2-500
-375
-250
-125
0
125
250
375
500
Cur
rent
(A
)
iL
ibatt
load I
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2390
395
400
405
410
Time (s)
Vol
tage
(V
)
vterm
Figure 4.4. (Top) Response of iL, ibatt to load ramp. (Bottom) Bus voltage in response to load ramp.
4.2 Nonlinear Averaged Model for Vehicle System
As discussed in Section 4.1, it is impractical to use the switched model to observe
dynamic behavior which occurs on slow time scales. Since the step size was constrained
by switching phenomena, removing the “switching” part of the model allows the step size
to be greatly reduced. The switching information is removed by time-averaging (4.1) and
(4.2). The time-averaged dynamic equations for the plant, with capacitor ESR included,
are
84
'
'
in in out
oc out outload
batt
inuc
LL
cL
L
cu e rco t s
diL v R i d v
dt
dv v v vC d i i
dt R R
dvC i
dt
v v i R
(4.9)
Here the lowercase variables represent quantities that are time-averaged over one switch-
ing cycle. The duty ratio d is
1
' 1
L
c
in in
esr batt loaL d
v R id
v R i i i
d d
(4.10)
where
oc esr
batt
c cbatt
esrload
c c
v v i Ri
R
v i Ri
R
(4.11)
Equation (4.9) and the quantities that substitute into it are left implicitly defined. The past
value of ci is used to compute the next value of each state variable. This gives acceptable
accuracy for small esrR and allows for all non-switching loss to be accounted for.
At lower frequencies, the reference current refi is related to time-averaged Li by
'
'2 2
cref L a
ininv v d Tv dT
i i d d m dTL L
(4.12)
per [50]. Since Li is completely controlled at low frequencies, the state attributed to the
inductor is eliminated from the model. Equation (4.12) accounts for the impact of the
85
current loop on plant dynamics at lower frequencies when the voltage loops are closed
around it.
Figure 4.5 shows an overview of the complete averaged model. Individual blocks
are expanded in Appendix B.2. This model is used to verify that a choice of gains pK and
1 distributes power as predicted by the equations in Section 2.1.2. It can also show the
charging loop’s impact on the system in the time domain. Selecting initial states which
put the system in equilibrium is straightforward because the average system exhibits true
steady-state behavior.
load R
Vc
Pload_desRiref_to_iL
Vin
i_ref
v_c
dp
i_L
ibatt, il=-iuc, iref
i_uc
-u(1)
i_c
Vin
i_L
Vc
R
i_c
dp
Voc
ibatt
Voc, Vc, Vin_state,Vin_term
Vin_term
f(u)
UC energy
Terminator1
TerminatorRepeatingSequence1
Product Integrator2
1s
Integrator1
1s
Integrator
1s
FromWorkspace
sine_vec
Dp
Controller
Vin
Vciref
100 kW pulse
1/Cuc
-K-
1/C
-K-Vc
Vin
Figure 4.5. Overview of averaged nonlinear model of the combined source for an electric vehicle.
86
4.2.1 Simulation Results
Figure 4.6 shows the response of the averaged model to the 100 kW pulse train
described in Section 2.1.2. The model was started in steady state with inv equal to
, 266 Vuc refv . Battery current reaches a maximum of 143 A at the end of the third dis-
charge cycle, which is within the 147 A specification. UC voltage bottoms out at 188 V
before being pulled back toward the reference level by the charging loop. This is within
the 180 V design specification. The middle plot shows that peak inductor current reaches
320 A when the pulse is first applied. This is lower than the ~500 A inductor current seen
in Figure 4.4 because inv starts at 266 V instead of 180 V. The charging loop brings inv
close to ,uc refv after about 300 s, successfully compensating for the offset between Li and
refi .
0 100 200 300 400 500 600 700180
220
260
300
Vol
tage
(V
)
vin
0 100 200 300 400 500 600 700-200
0
200
400
Cur
rent
(A
) iL
iref
0 100 200 300 400 500 600 7000
50
100
150
Time (s)
Cur
rent
(A
) ibatt
Figure 4.6. Response to the 100 kW pulse train described in Figure 2.4.
87
Figure 4.7 shows the response of inv and batti to a 20 A, 0.5 mHz sinusoidal load
current. The gray line is the system response when charging loop gains chgK and ichgK
are both set to zero. The UC voltage varies considerably while the batti magnitude ex-
ceeds load current and lags it in phase. Awerbuch and Sullivan [52] observed that the
extra battery draw hurts system efficiency. Thus, the currents flowing through the UC in
phase with load current need to be minimized. Adding the charging loop reduces the inv
ripple magnitude by almost a factor of 3, indicating that the UC current was also reduced
by this magnitude, but does not completely eliminate the circulating current, as the mag-
nitude of batti still exceeds load current.
0 2000 4000 6000 8000 10000 12000 14000236
246
256
266
276
Vol
tage
(V
)
Time (s)
vin
no chg loop
vin
w/ chg loop
6000 6200 6400 6600 6800 70000
5
10
15
20
Cur
rent
(A
)
Time (s)
ibatt
no chg loop
ibatt
w/ chg loop
load i
Figure 4.7. (Top) vin response to 0.5 mHz sine wave load. (Bottom) Zoomed view of one half-cycle of
ibatt and load current.
88
4.3 Switched Model for VRM
A switched model is developed in Simulink for the VRM with multiple transient-
carrying phases described in Section 2.2 and shown in Figure 2.9. The model will be used
to predict the system’s response to large load, line and reference steps. The objective is to
see whether the proposed topology will regulate bus voltage to the specifications outlined
in [27] using the small 670 µF bus capacitance predicted by (2.34).
Figure 4.8 shows the major model blocks and their connections. The dc phase, ac
phase and Bus Capacitor blocks implement state equations for the plant portion of Figure
2.9 and are based upon work done by Logue [54],[55]. Here “dc” refers to the four dc-
carrying phases and “ac” refers to the two fast phases which carry transient current only.
The 2 V source block models the first converter stage in Figure 2.10 as a voltage source
with series impedance of 5 mΩ, substantially higher than the 0.8 mΩ required by the
processor. With 5 mΩ series impedance, the intermediate voltage swings between 2 V
and 1.25 V with load current, allowing the VRM’s steady-state duty ratio to be between
0.4 and 0.6 under all loads. Figure 4.9 (on page 90) details the outer loop controller. The
LPF and BPF blocks divide the main current command by frequency. They contain first-
order low-pass and bandpass filters, respectively. Offset correction for each group of
phases is provided by the Native AVP blocks [36]. The full model is expanded in Appen-
dix B.3.
89
dc pcm process
IL_1
IL_2
IL_3
IL_4
Iref/4
Scope out
q(t)_1
q(t)_2
q(t)_3
q(t)_4
ac pcm process
IL_1
IL_2
Iref/2
Scope out
q(t)_1
q(t)_2
Vzero
Voc
Vin, Vout
Outer loopcontroller
sum_IL_s
Vout
Vzero
sum IL f
Iref_s
Iref_f
Load R
Irefs, ILs
Divide
Dc phase 4
vin
iout
q(t)
vc
iL
iin
Dc phase 3
vin
iout
q(t)
vc
iL
iin
Dc phase 2
vin
iout
q(t)
vc
iL
iin
Dc phase 1
vin
iout
q(t)
vc
iL
iin
BusCapacitor
i_c
vout
vc
Ac phase 2
vin
iout
q(t)
vc
iL
iin
Ac phase 1
vin
iout
q(t)
vc
iL
iin
2 V source
iinVin
iout
Figure 4.8. Overview of switched model of the VRM.
90
Iref_f
2
Iref_s
1
iref_s/4
1/4
iref_f/2
1/2
iL/4
1/4
iL/2
1/2
Native AVP1
iL
iref
ioff
Native AVP
iL
iref
ioff
LPF
In1 Out1
Kp/4
Kp
BPF
In1 Out1
sum_IL_f
4
Vzero
3
Vout
2
sum_IL_s
1
Vzero iref_s
iref_f
Figure 4.9. VRM outer loop controller (Controller block in Figure 4.8).
Key simulation and model parameters are listed in Table 4.1 (on page 91). The
dividing point between the fast and slow phases was chosen to be 3 kHz. The corner for
the LPF in the Native AVP block is chosen to be slow so that the offset provided by this
block is essentially fixed on fast time scales. Each set of phases is interleaved with regard
to other phases in that set, but fast phases run asynchronously from the slow phases. Code
to start the model very close to steady state with no load current is listed in Appendix
B.3. At present these scripts will not start the model in steady state with nonzero load
current.
91
Table 4.1. Parameters used in the VRM model.
Per-Phase Circuit Parameters Controller Gains Common Parameters Parameter Value Parameter Value Parameter Value
bigL 500 nH pK 1250
refV 0.8 V
Slow LswR R 4.15 mΩ , ,c lpf dcf 3 kHz
inV 2 V
slowT 66.67 10 s , ,c hpf transf 3 kHz srcR 5 mΩ
,a slowM 0.7 0.8 bigL , ,c lpf transf 7.96 MHz C 670 µF
,c navpf 200 Hz
esrR 0.8 mΩ
smallL 10 nH
Fast LswR R 7.5 mΩ
fastT 72 10 s
,a fastM 0.7 0.8 smallL
4.3.1 Simulation Results
Substantial load, line and reference steps are applied to the simulated system.
Phase currents are plotted in response to each disturbance, along with bus voltage. Plots
show how current is distributed between the phases and if bus voltage meets the targets
established in [27].
Figure 4.10 shows phase currents and bus voltage when a 120 A load is applied at
t = 0.1 ms, and then removed at t = 0.5 ms. The specification in [27] calls for bus voltage
to be 0.8 V when load current is 0 A and 0.704 V when load current is 120 A, with no
more than 50 mV overshoot of these targets. Figure 4.10 shows that BPF and LPF in the
outer loop distribute current as is expected, and current trajectories are well-behaved. Bus
voltage overshoot is 15 mV or less, well within the 50 mV specification.
92
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
-60
-40
-20
0
20
40
60
Pha
se c
urre
nt (A
)
Time (ms)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.90.65
0.7
0.75
0.8
Time (ms)
Vol
tage
(V)
Vbus
Slow phase currents
Fast phase currents
Figure 4.10. Response of phase currents (top) and bus voltage (bottom) to 120 A load step.
Figure 4.11 shows the response to 25% line steps. Load current before the steps
were applied is 120 A. The simulation was started at no load, then load resistance was
changed to 5.87 mΩ and the system was allowed to settle before the line steps were ap-
plied. A 2 to 2.5 V line step up was applied 1 ms into the simulation run. The left-hand
plots in Figure 4.11 show the response of phase currents and bus voltage to the step. Plots
on the right do the same for a 2.5 to 2 V line step down occurring 1.5 ms into the run. The
bus voltage deviates less than 5 mV from nominal in response to either line step. Voltage
ripple increases because of the higher line voltage but is still only a few mV.
93
0.98 1 1.02 1.04
-10
0
10
20
30
40P
hase
cur
rent
(A
)
Time (ms)
Slow phase currents
Fast phase currents
0.98 1 1.02 1.040.68
0.69
0.7
0.71
0.72
Time (ms)
Vol
tage
(V
)
Vbus
1.48 1.5 1.52 1.54
-10
0
10
20
30
40
Pha
se c
urre
nt (
A)
Time (ms)
Slow phase currents
Fast phase currents
1.48 1.5 1.52 1.540.68
0.69
0.7
0.71
0.72
Time (ms)
Vol
tage
(V
)V
bus
Figure 4.11. (Left plots) Phase current, vbus response to 2-to-2.5 V line step up at t = 1 ms.
(Right plots) Phase current, vbus response to 2.5-to-2 V line step down at t = 1.5 ms.
Figure 4.12 shows the response to 25% open-circuit reference steps. Load resis-
tance was 8.89 mΩ, corresponding to about 80 A load when the open-circuit reference is
0.8 V. Like the line step plots, the simulation was started under no load, the load resistor
applied, and the system allowed to settle 1 ms before the reference was stepped from 0.8
to 1 V. The reference was stepped back down to 0.8 V at time 1.5 ms. Due to the small
output capacitor and fast phases, output voltage responds very quickly to a step change in
reference voltage. Again there is little voltage overshoot, and current through the fast
phases returns to zero in steady state. In all three of the figures, fast phase current appears
“noisy” because the displayed simulation output was sampled only ten times per fast
switching cycle. No subharmonic instability was observed on these phases.
94
1 1.05 1.1
-20
0
20
40
60P
hase
cur
rent
(A
)
Time (ms)
Slow phase currents
Fast phase currents
1 1.05 1.1
0.7
0.75
0.8
0.85
0.9
Time (ms)
Vol
tage
(V
)
Vbus
1.5 1.55 1.6-60
-40
-20
0
20
40
Pha
se c
urre
nt (
A)
Time (ms)
Slow phase currents
Fast phase currents
1.5 1.55 1.6
0.7
0.75
0.8
0.85
0.9
Time (ms)
Vol
tage
(V
)
Vbus
Figure 4.12. (Left plots) Phase current, vbus response to 25% reference step up at t = 1 ms.
(Right plots) Phase current, vbus response to 25% reference step down at t = 1.5 ms.
It is emphasized that simulations were run using a 670 µF bus capacitor with 0.8
mΩ ESR. This is a smaller capacitor, with larger ESR, than the decoupling capacitors
near the processor socket [27]. These results show that the fast phases can completely
replace the bulk capacitors on the board.
95
CHAPTER 5
EXPERIMENTAL RESULTS
A combined power source for an electric vehicle application was built and ex-
perimentally tested to verify predictions of the large-signal models, and by association
the design equations and small-signal models. As full-scale hardware is difficult to test,
experiments were performed on scaled-down hardware. Scaling is discussed in Section
5.1. The hardware itself is described in Section 5.2. Battery impedance was characterized
in Section 5.3 in order to supply an accurate value of battR to the models, as power appor-
tioning and thus system state values depend strongly on this parameter. Step response and
repeated full-power draw tests were performed on the hardware in Sections 5.4 and 5.5,
respectively, to verify that the system meets its design targets and to see how well its be-
havior follows that predicted by the models.
5.1 Scaling the Design
A 100 kW, 400 V power source requires special equipment to test. A low-voltage,
lower power system can be tested using bench equipment. A 12 V nominal bus is selected
to ease development and circuit measurements; this is 3% of 400 V. Scaled current is 7.5
A, or 3% of 250 A, and scaled power is 90 W. Load requirements are scaled in magnitude
but not in time:
Peak power: 90 W for 10 seconds, both directions
Average power: 22.5 W long-term
Load slew: zero to peak power in 0.2 s, peak to peak power in 0.4 s
Battery power restricted to 45 W at low cell voltage.
96
Since the time parameters are the same as before, the required UC power and energy
scales by 0.032. Maximum and minimum UC voltage is 80% and 45% of bus voltage,
respectively. Table 5.1 shows a summary of the scaling.
Table 5.1. Full and scaled system values. Parameter Full Value Scaled Value Scaling Factor
Bus voltage 400 V 12 V 0.03 Peak load current 250 A 7.5 A 0.03 Peak load power 100 kW 90 W 0.032 Battery charge capacity 147 A-h 4.4 A-h 0.03 Peak battery power 50 kW 45 W 0.032 Long-term average power 25 kW 22.5 W 0.032 Battery resistance 0.15 Ω 0.15 Ω Unchanged Load slew rate 1250 A/s 37.5 A/s 0.03 UC energy needed 859 kJ 773 J 0.032 Peak UC voltage 320 V 9.6 V 0.03 Min UC voltage 180 V 5.4 V 0.03 Required Cuc 44.16 F 44.16 F Unchanged
With system parameters scaled according to Table 5.1, battR and ucC are identical
for both systems. Other critical ratios, such as peak-to-average power, are also un-
changed. These properties give the two systems the same dynamic response at lower fre-
quencies, and therefore both systems should require the same outer-loop control tunings
to get the desired power apportioning and UC charging. Setting 2 the same for both
converters will make their control bandwidths approximately equal.
The scaled battery is made from two strings of three cells each, of the same type
and size chosen in Section 2.1.2 for the full-sized battery. Cells of this size and type are
well-characterized in [40]. The UC is a string of five 350 F ultracapacitors, resulting in
70 F capacitance. This is about 50% more than the 44.16 F scaled value. The extra ca-
pacitance will not affect how power is apportioned but it will require increasing the
charging loop gains to maintain disturbance rejection. Key converter parameters are
97
compiled in Table 5.2. These parameters will be entered into the averaged and switched
nonlinear models; the models themselves do not change as a result of scaling.
Table 5.2. Hardware and controller parameters for the scaled-down converter.
Circuit Parameters Controller Gains Parameter Value Parameter Value
L 100 µH pK 65
C 8000 µF 1 3.71
ucC 70 F
inR 22.9 mΩ chgK 1.05
T 1*10-4 s ichgK 0.002
aM 0.7*M2 2 0.0291
2M (12-5.4)/L
5.2 Hardware and Setup
Figure 5.1 is a close-up picture of the power converter hardware. The PCM proc-
ess is implemented using a UC3823A PWM IC. To make the IC support bidirectional
current flow, the protection features were disabled and the sensed inductor current signal
was level-shifted. Logic was added to allow the FETs to be shut down by an external
command from the outer loop controller in order to provide a minimum level of circuit
protection. Appendix C contains the hardware schematics.
98
Figure 5.1. Close-up of power converter hardware.
Figure 5.2 gives an overview of the controller for the power converter hardware.
This software controller was implemented using MATLAB Real-Time Workshop and is
executed on a PC running xPC Target operating system. It contains the linear outer loop
controller described in Section 4.1.1 and routines to provide circuit protection. The con-
troller takes in measurements of inv , Li , and a high-passed termv , and outputs refi and a
FET Enable signal to the board. Bus voltage termv is high-pass filtered and amplified in
hardware, on the power converter board, in order to get sufficient amplitude resolution
when this signal is passed to the A/D interface. Controller states are updated at fixed 15
µs intervals.
99
xPC Target SetOverload Counter
SetOverload Counter
Vin_gain
2
Vin
Undervoltage Hiccup Mode
Vin
Iref_in
Enable
Iref
Re-enable Delay
Enable_in Enable
Q8 DA
Q8Quanser
Analog Output
1
5Q8 AD
Q8Quanser
Analog Input
1
2
5
Overcurrent Protection
Iref_in
I_L
Iref
Enable
LogicalOperator
ANDLinear controller
Vin
Vterm
Iref
Iref_gain
0.01
Iref
I_L IL_gain
10
HPFd_Vtermgain
-0.1
HPFd_Vterm
Enable Gain
10
Enable
Data TypeConversion2
Convert
Data TypeConversion1
Convert
Data TypeConversion
Convert
Constant2
0
Constant1
-C-
Constant
0
Add1
Add
v_in
i_L
Iref
-HPFd Vterm
Figure 5.2. Overview of outer loop controller running on xPC Target.
A Quanser Q8 A/D interfaces the xPC Target machine with the power converter
hardware. The A/D was read and written to every 90 µs, with about 17 µs latency [58].
This is sufficient speed for this application. Figure 5.3 shows a picture of the complete
hardware setup with various components labeled. Load is provided by either a load resis-
tor or a programmable dc load as is appropriate for the test.
100
Figure 5.3. Test setup.
5.3 Battery Resistance Measurement
The battery resistance battR is vitally important for both power distribution and
system dynamics. A characterization of battR across frequency is needed. To measure
impedance, the battery was hooked up to a programmable dc load biased with a dc supply
so that current could both flow into and out of the battery [59]. A sinusoidal current was
101
drawn from the battery and its terminal voltage was measured. Table 5.3 lists the imped-
ance magnitudes measured at different frequencies. The battery was charged to 12 V
nominal and current magnitude was 1 A peak-to-peak (p-p).
Table 5.3. Battery impedance measurement. Frequency P-P Current
Draw P-P
Voltage Battery Impedance
Magnitude 0.1 Hz 1 A 0.58 V 0.58 Ω 0.3 Hz 1 A 0.48 V 0.48 Ω
1 Hz 1 A 0.35 V 0.35 Ω 3 Hz 1 A 0.35 V 0.35 Ω
10 Hz 1 A 0.35 V 0.35 Ω 30 Hz 1 A 0.34 V 0.34 Ω
100 Hz 1 A 0.33 V 0.33 Ω 300 Hz 1 A 0.34 V 0.34 Ω
1000 Hz 1 A 0.33 V 0.33 Ω 3000 Hz 1 A 0.43 V 0.43 Ω
10000 Hz 1 A 0.94 V 0.94 Ω
Expected cell impedance magnitude is between 0.06 Ω and 0.1 Ω depending on
frequency [40]. With three cells in series and two in parallel, impedance of the battery
pack should be 1.5 times this value, or 0.09-0.15 Ω. The impedance reported in Table 5.3
is over three times the expected value. Very high series impedance is an indicator of poor
battery health [60]; further testing revealed that the battery is indeed at the end of its life.
It could not sustain even a moderate current draw for more than a few minutes. This bat-
tery was used to complete the step response test in Section 5.4, which requires only brief
current draws, but a dc supply was used for the repeated draw test in Section 5.5.
5.4 Step Response
The purpose of the step response test is to make sure the choice of 2 is sufficient
to keep inductor current from ringing when termv changes abruptly. As the capacitor in
this system is not large enough to provide transient energy, the difference between load
102
power and the average current through the top converter FET must be supplied by the dc
source (the battery).
A load step was generated by switching a load resistor into and out of the voltage
bus. The load resistor had a measured resistance of 1.41 Ω. A FET was used as the switch
to get a sharp, clean step. Initial UC voltage was 6 V and battery voltage was 12 V. The
UC voltage is set low in order to evaluate worst-case behavior. Voltages and currents
were measured using an oscilloscope set to 2 ms per division and 25 kHz sampling rate.
The same load step was applied to a switched model of the system. The model was set up
using the initial UC and battery voltages above. Parameter battR was set to the average
battery resistance at frequencies between 100 and 1000 Hz, 0.34 Ω.
Figure 5.4 plots Li , termv and batti after the 1.41 Ω load resistor was switched into
the bus. Measurements matched the predictions well. Inductor current Li follows a sec-
ond-order trajectory with a small amount of overshoot. This overshoot was predicted by
the simulation; it can be removed if needed through further controller tuning. With the
present tuning, no load transient during normal operation will make Li ring or become
untracked from refi . All variables are close to their final values after 15 ms, adequate for
the application and similar to what was predicted for the 100 kW converter.
Figure 5.5 plots Li , termv and batti when the load resistor is abruptly removed from
the bus. The measured step down was initiated about 5 s after the load resistor was con-
nected to the bus. The switched model was run under the same conditions — load was
applied and then removed 5 s later, and responses were recorded when the load was re-
moved. The model and measured system have different state values just before the step,
but after the step their trajectories follow similar paths. Inductor current follows a second-
order trajectory with slight overshoot of the final value and no ringing.
103
0 5 10 15 20 25
0
5
10
load
i (A
)
SimMeas
0 5 10 15 20 25
0
10
20
i L (A
)
SimMeas
0 5 10 15 20 2510
11
12
v term
(V
)
SimMeas
0 5 10 15 20 25
0
2
4
i batt (
A)
Time (ms)
SimMeas
Figure 5.4. Simulated and measured response to step addition of 1.41 Ω load.
104
0 5 10 15 20 25
0
5
10
load
i (A
)SimMeas
0 5 10 15 20 25-10
0
10
20
i L (A
)
SimMeas
0 5 10 15 20 25
11
12
13
v term
(V
)
SimMeas
0 5 10 15 20 25-4
-2
0
2
i batt (
A)
Time (ms)
SimMeas
Figure 5.5. Simulated and measured response to step removal of 1.41 Ω load. Time t = 0 is set to the
instant when the load was removed.
5.5 Repeated Draw Test
In this test, the dc load is commanded to run the following program:
1) Ramp from 0 to 7.5 A in 0.2 s.
2) Hold at 7.5 A for 9.8 s.
3) Ramp from 7.5 to 0 A in 0.2 s.
105
4) Hold at 0 A for 29.8 s.
5) Repeat.
This signal has period 40 s and draws the same average current as a 7.5 A, 25% duty
pulse train. The xPC Target was also used to program the dc load and store measured
data, in addition to providing system control and hardware protection. Additional hard-
ware was used to condition the measured currents and voltages and feed them into the
A/D; schematics are in Appendix C. For these tests the UC is initialized to 8.03 V and
bus voltage is set to 12.07 V no-load. Simulation results are provided by the averaged
nonlinear model presented in Section 5.2.
Because of the battery’s high series impedance and inability to sustain a current
draw, the battery was replaced by the dc source shown in Figure 5.6. A diode prevents
reverse current from flowing into the dc supply. The voltage across the diode varies with
the current flowing through it, even with the strong bias provided by the 3 Ω ballast resis-
tor. In addition, the bench supply does not perfectly regulate output voltage as load cur-
rent varies. These nonidealities warrant characterization if the system model is to match
the physical system.
dc supply 3 Ω vtermvballast
Rbatt
ibatt
Figure 5.6. Dc source used for the repeated draw test.
106
As the series resistance battR is fixed, the dc supply can be modeled by a current-
dependent voltage source with fixed series resistance. The resistors implementing battR
measured 0.176 Ω with lead wires included. Table 5.4 lists the voltage across the ballast
resistor, ballastv , as current batti varied from 0 to 6 A. Voltage measurements were taken
when the diode was warm, as it will be warm during the repeated draw test. Figure 5.7 is
a graph of the table data. The graph generally has the nonlinear I-V characteristic of a
diode. A lookup table implements the current-dependent voltage source in the averaged
nonlinear model.
Table 5.4. Voltage across 3 Ω ballast resistor at specific ibatt current values.
batti ballastv
0 A 12.07 V 0.5 A 12.06 V 1 A 12.047 V
1.5 A 12.026 V 2 A 12.012 V 3 A 12.006 V 4 A 11.995 V 5 A 11.983 V 6 A 11.979 V
0 1 2 3 4 5 611.96
11.98
12
12.02
12.04
12.06
12.08
ibatt
(A)
v balla
st (
V)
Figure 5.7. Plot of the data in Table 6.4.
107
Figure 5.8 plots measured and predicted termv , batti and Li in response to the re-
peated load draws. Data was filtered somewhat and sampled every 0.2 s, removing the
switching ripple from the measurement. Measured data matches predictions very well
after about five loading cycles, and are nearly identical after 350 s. The active-parallel
system is effective at smoothing battery current — load current is 7.5 A peak, yet battery
current never exceeds 4 A. Figure 5.9 shows inv measured over a longer time interval,
along with the prediction. Here the measured value diverges from the prediction by about
half a volt after several loading cycles, but the two match after around 2400 s (40 min-
utes). Based on the data presented here, the likely reason for the divergence is that the
offset between reference current refi and per-cycle average inductor current Li was not
initialized correctly for the physical system. Since Li is more negative than predicted by
the model, the offset was likely too small. The more negative Li raises inv until the charg-
ing loop compensates. The charging loop integrator slowly brings inv down until it cycles
about the reference value. UC voltage stays within its physical bounds over the entire
time interval, and the charging loop ultimately rejects the disturbance without impacting
the ability of the power-distributing loop to supply legitimate load transients. Line distur-
bances from the dc source, where supply voltage varied nonlinearly due to the series di-
ode, are also rejected.
108
0 50 100 150 200 250 300 350 400 450 500
11.2
11.4
11.6
11.8
12v te
rm (
V)
Measured Simulated
0 50 100 150 200 250 300 350 400 450 5000
1
2
3
4
5
i batt (
A)
Measured Simulated
0 50 100 150 200 250 300 350 400 450 500
-5
0
5
10
15
Time (s)
i L (A
)
Measured Simulated
Figure 5.8. vterm(t), ibatt(t) and iL(t) of measured scaled-down combined source, compared to simulated.
0 500 1000 1500 2000 2500
6.5
7
7.5
8
8.5
9
Time (s)
v in (
V)
Measured Simulated vuc,ref
Figure 5.9. vin(t) measured scaled-down combined source, compared to simulated.
109
5.6 Summary
The hardware demonstration proves that the active-parallel system is capable of
reducing peak battery current by smoothing the load peaks. The outer loop controller
provides rejection of disturbances which would otherwise cause net flows of energy into
or out of the UC. Furthermore, the LPF and charging loop are effective at keeping Li
trajectory well-behaved and at compensating for the offset between refi and Li inherent to
the PCM process. In all tests the measured voltages and currents closely followed those
predicted by the switched and averaged large-signal models, proving the models are a
valid means to predict system behavior. As large-signal model results are congruent with
predictions made by the design equations small-signal models, these experimental results
indirectly reinforce the validity of these tools as well.
110
CHAPTER 6
CONCLUSIONS
This thesis presented and analyzed the use of a shaped-impedance controller to
control a dc-dc converter which decouples bulk capacitance from a dc voltage bus. Com-
plete combined power sources were developed for an electric vehicle power source and
for a desktop computer processor VRM. Employing a decoupled capacitor reduced the
size and/or weight of the power source needed to meet a given load requirement, whether
it was used in place of additional batteries or additional bus capacitance.
Design equations and large-signal dynamic models were developed for each sys-
tem. Linearized small-signal models were developed for the vehicle application to enable
this system to be analyzed in the frequency domain. These equations and models provide
a systematic way to tune the controller and select power-handling components. Together
the models give a good picture of overall system behavior. A scaled-down combined
source for a vehicle was constructed and tested. There was generally good agreement
between design equations, models, and measured results.
The controllers for both applications set the decoupled capacitor’s reference cur-
rent using a high-pass filtered and scaled measurement of bus voltage. The controller for
the vehicle application kept battery power and UC energy depletion from exceeding an
amount proportional to a step change in bus voltage. The charging loop kept UC charge
level within its bounds in the face of net loss, offset between reference and average in-
ductor currents, and slowly-varying disturbances in bus voltage. Design equations and the
switched VRM model predicted that decoupled bulk capacitance could be integrated ef-
111
fectively into a VRM design. Processor voltage bus capacitance could be substantially
reduced with little impact on voltage regulation or efficiency.
Future work can involve further optimizing the controller developed for the vehi-
cle application. Small-signal models of the system for the vehicle application revealed
that there is a large frequency region over which neither bus voltage nor UC charge is
tightly regulated. UC currents flowed opposite the direction intended within this band.
Making this crossover region smaller may allow better low-frequency disturbance rejec-
tion, further decreasing the UC size. Simply raising the filter order will decrease phase
margin appreciably, so this direction does not look promising. A nonlinear filter in the
charging loop has the potential to narrow the transition band without hurting phase mar-
gin. It would also be worth investigating the use of power-dense batteries as the transient
energy source, as these devices are more energy-dense than UCs and continue to become
more power-dense. The charging loop would need to be redesigned in order to properly
maintain the charge level of these batteries.
The VRM design looks promising as a way to reduce bus capacitor size without
harming efficiency or adding too much system complexity. Future work includes per-
forming further analysis and fabricating a prototype. It is also worth seeing if series-input
parallel-output (SIPO) SCM-controlled phases could be employed to carry the transient
current. Voltage droop would be implemented using adaptive voltage positioning to mod-
ify the SCM reference voltage. Bus voltage could be dropped from 12 V nominal to 0.8 V
nominal using five transient-carrying phases operating at a duty ratio near 0.5. The first
converter stage would no longer be needed, eliminating a point of failure and potentially
increasing efficiency.
112
APPENDIX A
SCRIPTS TO EVALUATE SMALL-SIGNAL TRANSFER FUNCTIONS
In Chapter 3, system stability was determined by evaluating small-signal models
across a range of system operating points. This section details the scripts which per-
formed this work.
A.1 Main Script
% nominal or constant parameters % These values are for the 100 kW converter; other values are substituted % in for the scaled-down converters L = 500e-6; C = 18000e-6; R = 1.6; Cuc = 44.16; Vin = 180; Voc = 400; Rbatt = 0.15; Ibatt = 30; %relevant for 2-state plant only Resr = 0.00314; Rin = (17.2+1.5+2.4)*1e-3; T = 100e-6; %switching period Ma_r = 0.7; %relationship of Ma to off-time slope of inductor current M2 = (400-180)/L; % abs. value of off-time slope of inductor current Kp = 65; Kchg = 0.4; Kichg = .00015; Kdchg = 0; tau1 = 3.71; tau2 = 0.0291; D_filter = 10; vin_corner = 200; %nominal parameters which take on a range of values R_v = [1.6 16 1e6]; Vin_v = [140:20:320]; Voc_v = [300:20:420]; Rbatt_v = [0.1 0.15 0.75]; Ibatt_v = [0 14 147]; % run the sub-scripts iL_dir = 1; iteration_script_2state_vcloop; twostate_result = PMs; iL_dir = -1; iteration_script_2state_vcloop;
113
twostate_result = [twostate_result;PMs]; iteration_script_3state_bothloops; threestate_result = PMs; [mag,idx] = max(twostate_result(:,[5 6 15 17])) [mag,idx] = max(twostate_result(:,8:14)) [mag,idx] = max(threestate_result(:,[5 6 20 22 24 26])) [mag,idx] = max(threestate_result(:,8:19))
A.2 Scripts for Two-State Transfer Functions
A.2.1 iteration_script_2state_vcloop.m
%Script to iterate through all combinations of variable parameters for %2-state plant, Rin and Resr, vc-to-iref loop with LPF only % calculated, but fixed, parameters M2 = (Voc-Vin)/L; % abs. value of off-time slope of inductor current Ma = Ma_r*M2; %initialize storage array q = 0; clear PMs; PMs = 0; clc; for m = 1:length(Vin_v) Vin = Vin_v(m); for n = 1:length(Voc_v) Voc = Voc_v(n); for o = 1:length(R_v); R = R_v(o); for p = 1:length(Rbatt_v) Rbatt = Rbatt_v(p); for w = 1:length(Ibatt_v) Ibatt = Ibatt_v(w); Vc = Voc-Ibatt*Rbatt; if Vin < Vc %calculated nominal parameters clear IL; Iout = Vc/R-Ibatt; IL_nom = Iout*Vc/Vin; IL = fzero(@(IL) ... lossy_IL_fcn2(IL,Iout,Rin,Vin,Vc),IL_nom); if iL_dir < 0 IL = -IL; end D = 1 - (Vin-IL*Rin)/(Vc); Dp = 1 - D; Req = 1/(1/R+1/Rbatt); q = q+1; %generate transfer functions tfs_2state_fast_vc_loop;
114
%find PM and crossover frequency of each tf, store in a %vector fprintf('Vin %5.3g, Voc %5.3g, R %5.3g, Rbatt %5.3g, Ibatt %5.3g, Vc %5.3g \n',Vin,Voc,R,Rbatt,Ibatt,Vc) yidx = 8; PMs(q,19) = 0; PMs(q,1) = Vin; PMs(q,2) = Voc; PMs(q,3) = R; PMs(q,4) = Rbatt; PMs(q,5) = Ibatt; PMs(q,6) = Vc; % [GM, PM, GMfreq, crossfreq] = margin(open_iref_il); rr = rlocus(open_iref_il,[0 1]); PMs(q,7) = max(real(rr(:,1))); %see if open-loop iref-to-il is unstable PMs(q,8) = max(real(rr(:,2))); %see if closed-loop iref-to-il is unstable PMs(q,yidx) = isOpenLoopUnstable(Hvr); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Hvu); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Hvo); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Hvl); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Jvu); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Jvo); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Jvl); yidx = yidx+1; [mag,freq] = get_max_mag(Jvo); PMs(q,yidx) = mag; yidx = yidx+1; PMs(q,yidx) = freq; yidx = yidx+1; [mag,freq] = get_max_mag(Jvl); PMs(q,yidx) = mag; yidx = yidx+1; PMs(q,yidx) = freq; yidx = yidx+1; end end end end end end
A.2.2 tfs_2state_fast_vc_loop.m
% Small-signal transfer functions for current-controlled converter, no % outer loop, battery connected to output, Vin treated as constant on a % fast scale, converter losses Rin and Req included. % Fixed tfs 6/11 and added input for perturbed Voc. % clear Gvd Gvv Gvl Gid Giv Gil Fm Fin Fout Hvr Hvv Hvl Fll Jvl Jvv outer_loop_gain;
115
% linearized perturbed transfer functions of plant, with no control applied Gvd = tf([-L*IL -Rin*IL+Dp*Vc],... [L*C (C*Rin+L/Req) Dp^2+Rin/Req]); %control-to-Vc tf Gvu = tf(Dp,... [L*C (C*Rin+L/Req) Dp^2+Rin/Req]); %vin-to-Vc tf Gvo = tf([L/Rbatt Rin/Rbatt],... [L*C (C*Rin+L/Req) Dp^2+Rin/Req]); %iload-to-Vc tf Gvl = tf([-L -Rin],... [L*C (C*Rin+L/Req) Dp^2+Rin/Req]); %iload-to-Vc tf Gid = tf([C*Vc Dp*IL+Vc/Req],... [L*C (C*Rin+L/Req) Dp^2+Rin/Req]); %control-to-IL tf Giu = tf([C 1/Req],... [L*C (C*Rin+L/Req) Dp^2+Rin/Req]); %vin-to-il tf Gio = tf(-Dp/Rbatt,... [L*C (C*Rin+L/Req) Dp^2+Rin/Req]); %vin-to-il tf Gil = tf(Dp,... [L*C (C*Rin+L/Req) Dp^2+Rin/Req]); %iload-to-il tf % gains and relating iref to il Fm = 1/(Ma*T); Fin = (2*D-1)*T/(2*L); Fout = Dp^2*T/(2*L); % iref-to-il gain Hir = Fm*Gid/(1+Fm*Gid+Gvd*Fm*Fout); %iref-to-il tf % closed inner loop transfer functions Hvr = Fm*Gvd/(1+Fm*Gid+Gvd*Fm*Fout); %iref-to-vc tf Hvu = (Gvu*(1+Fm*Gid) - (Giu+Fin)*Fm*Gvd)/... (1+Fm*Gid+Gvd*Fm*Fout); %vin-to-vc tf Hvo = (Gvo*(1+Fm*Gid) - Gio*Fm*Gvd)/... (1+Fm*Gid+Gvd*Fm*Fout); %voc-to-vc tf Hvl = (Gvl*(1+Fm*Gid - Gil*Fm*Gvd))/... (1+Fm*Gid+Gvd*Fm*Fout); %iload-to-vc tf % remove pole-zero pairs that cancel from the H's Hvr = cancelPZPairs(Hvr,0.05); Hvu = cancelPZPairs(Hvu,0.05); Hvo = cancelPZPairs(Hvo,0.05); Hvl = cancelPZPairs(Hvl,0.05); % outer loop controller TF Fll = Kp*tf(1,[tau2 1]); %tau1 = HPF time constant, tau2 = LPF time constant % check loop gain of outer loop - it is desirable for the outer loop to be % stable when open; at minimum this facilitates loop gain measurements outer_loop_gain = Hvr*Fll; %closed outer loop transfer functions Jvu = Hvu/(1+Hvr*Fll); %vin-to-vc tf Jvo = Hvo/(1+Hvr*Fll); %voc-to-vc tf Jvl = Hvl/(1+Hvr*Fll); %iload-to-vc tf % remove pole-zero pairs that cancel from the J's Jvu = cancelPZPairs(Jvu,0.05); Jvo = cancelPZPairs(Jvo,0.05); Jvl = cancelPZPairs(Jvl,0.05);
116
A.3 Scripts for Three-State Transfer Functions
A.3.1 iteration_script_3state_bothloops.m
%Script to iterate through all combinations of variable parameters for %3-state plant, Rin, no Resr, both vc-to-iref and vin-to-iref loops % calculated, but fixed, parameters M2 = (Voc-Vin)/L; % abs. value of off-time slope of inductor current Ma = Ma_r*M2; %initialize storage array q = 0; clear PMs; PMs = 0; clc; for m = 1:length(Vin_v) Vin = Vin_v(m); for n = 1:length(Voc_v) Voc = Voc_v(n); for o = 1:length(R_v); R = R_v(o); for p = 1:length(Rbatt_v) Rbatt = Rbatt_v(p); Vc = Voc*R/(R+Rbatt); if Vin < Vc %calculated nominal parameters clear IL; IL_nom = Vc^2/(R*Vin); IL = fzero(@(IL) ... lossy_IL_fcn(IL,0,Rbatt,Rin,R,Vin,Vc,Kp),IL_nom); D = 1 - (Vin-IL*Rin)/(Vc); Dp = 1 - D; Req = 1/(1/R+1/Rbatt); q = q+1; %generate transfer functions tfs_3state_plant_and_chg_loop; %find PM and crossover frequency of each tf, store in a %vector fprintf('Vin %5.3g, Voc %5.3g, R %5.3g, Rbatt %5.3g, Vc %5.3g \n',Vin,Voc,R,Rbatt,Vc) yidx = 8; PMs(q,27) = 0; PMs(q,1) = Vin; PMs(q,2) = Voc; PMs(q,3) = R; PMs(q,4) = Rbatt; PMs(q,5) = Vc; % [GM, PM, GMfreq, crossfreq] = margin(open_iref_il); rr = rlocus(open_iref_il,[0 1]); PMs(q,6) = max(real(rr(:,1))); %see if open-loop iref-to-il is unstable PMs(q,7) = max(real(rr(:,2))); %see if closed-loop iref-to-il is unstable PMs(q,yidx) = isOpenLoopUnstable(Hvr); yidx = yidx+1;
117
PMs(q,yidx) = isOpenLoopUnstable(Hvo); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Hvl); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Hur); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Huo); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Hul); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Jvr); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Jvo); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Jvl); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Jur); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Juo); yidx = yidx+1; PMs(q,yidx) = isOpenLoopUnstable(Jul); yidx = yidx+1; [mag,freq] = get_max_mag(Jvo); PMs(q,yidx) = mag; yidx = yidx+1; PMs(q,yidx) = freq; yidx = yidx+1; [mag,freq] = get_max_mag(Jvl); PMs(q,yidx) = mag; yidx = yidx+1; PMs(q,yidx) = freq; yidx = yidx+1; [mag,freq] = get_max_mag(Juo); PMs(q,yidx) = mag; yidx = yidx+1; PMs(q,yidx) = freq; yidx = yidx+1; [mag,freq] = get_max_mag(Jul); PMs(q,yidx) = mag; yidx = yidx+1; PMs(q,yidx) = freq; yidx = yidx+1; end end end end end
A.3.2 tfs_3state_plant_and_chg_loop.m
% Small-signal transfer functions for current-controlled converter, no % outer loop, battery connected to output, Vin treated as constant on a % fast scale. Main difference between this model and a regular boost % converter is that R is replaced by Req = R||Rbatt. % Transfer functions only; this file is meant to be run as a script. % Parameters need to be loaded in the workspace before this script is run. % % linearized perturbed transfer functions of plant, with no control applied % uc is included, il points toward c % input-to-vc tfs Gvd = tf([Dp*Cuc*Vc 0],[L*C*Cuc L*Cuc/Req+C*Cuc*Rin ... C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]); Gvo = tf([L*Cuc/Rbatt Cuc*Rin/Rbatt 1/Rbatt],[L*C*Cuc ... L*Cuc/Req+C*Cuc*Rin C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]); Gvl = tf([-L*Cuc -Cuc*Rin -1],[L*C*Cuc ... L*Cuc/Req+C*Cuc*Rin C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]); %input-to-vin tfs Gud = tf([-C*Vc -Vc/Req],[L*C*Cuc L*Cuc/Req+C*Cuc*Rin ... C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]); Guo = tf(Dp/Rbatt,[L*C*Cuc L*Cuc/Req+C*Cuc*Rin ... C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]); Gul = tf(-Dp,[L*C*Cuc L*Cuc/Req+C*Cuc*Rin ... C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]);
118
%input-to-il tfs Gid = tf([C*Cuc*Vc Cuc*Vc/Req 0],[L*C*Cuc L*Cuc/Req+C*Cuc*Rin ... C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]); Gio = tf([-Cuc*Dp/Rbatt 0],[L*C*Cuc L*Cuc/Req+C*Cuc*Rin ... C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]); Gil = tf([Cuc*Dp 0],[L*C*Cuc L*Cuc/Req+C*Cuc*Rin ... C+Cuc*Dp^2+Cuc*Rin/Req 1/Req]); % gains and relating iref to il Fm = 1/(Ma*T); Fin = (2*D-1)*T/(2*L); % Fin = (1-2*D)*T/(2*L); Fout = Dp^2*T/(2*L); % iref-to-il gain with current loop open open_iref_il = Fm*Gid/(1+Fm*Fin*Gud+Fm*Fout*Gvd); open_iref_il = cancelPZPairs(open_iref_il,.005); % with current loop closed Hir = Fm*Gid/(1+Fm*Gid+Fm*Fin*Gud+Fm*Fout*Gvd); % new closed inner-loop transfer functions with new current loop model % input-to-vc tfs Hvr = Fm*Gvd/(1+Fm*Gid+Fm*Fin*Gud+Fm*Fout*Gvd); Hvo = (Gvo-Gvd*Fm*(Gio+Fin*Guo))/(1+Fm*Gid+Fm*Fin*Gud+Fm*Fout*Gvd); Hvl = (Gvl-Gvd*Fm*(Gil+Fin*Gul))/(1+Fm*Gid+Fm*Fin*Gud+Fm*Fout*Gvd); % input-to-vin tfs Hur = Fm*Gud/(1+Fm*Gid+Fm*Fin*Gud+Fm*Fout*Gvd); Huo = (Guo-Fm*Gud*(Gio+Fout*Gvo))/(1+Fm*Gid+Fm*Fin*Gud+Fm*Fout*Gvd); Hul = (Gul-Fm*Gud*(Gil+Fout*Gvl))/(1+Fm*Gid+Fm*Fin*Gud+Fm*Fout*Gvd); % remove pole-zero pairs that cancel from the H's Hvr = cancelPZPairs(Hvr,0.05); Hvo = cancelPZPairs(Hvo,0.05); Hvl = cancelPZPairs(Hvl,0.05); Hur = cancelPZPairs(Hur,0.05); Huo = cancelPZPairs(Huo,0.05); Hul = cancelPZPairs(Hul,0.05); Hir = cancelPZPairs(Hir,0.05); % control gains for vc and vin feedback loops % COMMENT these out when running Juo_only routine BPF_Kp = Kp*tf([tau1 0],[tau1*tau2 tau1+tau2 1]); %tau1 = HPF time constant, tau2 = LPF time constant Kc = tf([Kchg Kichg],[1 0]); % check loop gain of outer loop - it is desirable for the outer loop to be % stable when open; at minimum this facilitates loop gain measurements outer_loop_gain = Hvr*BPF_Kp; % tfs with both voltage feedback loops closed - 6 total % input-to-vc tfs Jvr = -Hvr*Kc/(1-Hur*Kc+Hvr*BPF_Kp); %vuc_ref-to-vc tf Jvo = (Hvo*(1-Hur*Kc)+Hvr*Huo*Kc)/(1-Hur*Kc+Hvr*BPF_Kp); Jvl = (Hvl*(1-Hur*Kc)+Hvr*Hul*Kc)/(1-Hur*Kc+Hvr*BPF_Kp); % input-to-vin tfs Jur = -Hur*Kc/(1-Hur*Kc+Hvr*BPF_Kp); %vuc_ref-to-vin tf Juo = (Huo*(1+Hvr*BPF_Kp)-Hur*Hvo*BPF_Kp)/(1-Hur*Kc+Hvr*BPF_Kp); Jul = (Hul*(1+Hvr*BPF_Kp)-Hur*Hvl*BPF_Kp)/(1-Hur*Kc+Hvr*BPF_Kp);
119
% remove pole-zero pairs that cancel from the J's Jvr = cancelPZPairs(Jvr,0.05); Jvo = cancelPZPairs(Jvo,0.05); Jvl = cancelPZPairs(Jvl,0.05); Jur = cancelPZPairs(Jur,0.05); Juo = cancelPZPairs(Juo,0.05); Jul = cancelPZPairs(Jul,0.05);
A.4 Helper Functions
The scripts above call a few helper functions. A description of the functions is as follows:
sys2 = cancelPZPairs(sys,eps). When MATLAB divides one tf object
by another, the quotient contains poles and zeros which numerically cancel, but
are present nonetheless. These extraneous terms cause problems when the quo-
tient is further manipulated. The function cancelPZPairs removes pole/zero
pairs from sys which are within magnitude eps of each other, and returns the re-
sult in sys2.
[ff,maxpole] = isOpenLoopUnstable(sys). This function sees if any
poles of sys are not in the OLHP, and marks ff accordingly. The angular fre-
quency of largest (fastest) unstable pole is returned in maxpole.
[mag,freq] = get_max_mag(sys). This function returns the peak magnitude
of the Bode plot of sys, along with the frequency (in Hz) at which the peak oc-
curs.
120
APPENDIX B
EXPANDED LARGE-SIGNAL SIMULINK BLOCKS AND SET-UP CODE
B.1 Switched Nonlinear Model for Vehicle
B.1.1 Expanded Controller Blocks
The parent for these blocks is the Controller detailed in Section 4.1.1. The header
of each sub-figure in Figure B.1 shows the name of the expanded block, along with its
location in the model hierarchy.
TF is (F1(s)-1)(F2(s) = -s*tau1/((s*tau1+1)(s*tau2+1)) tau1 = low cutoff
tau2 = high cutoff
Out1
1
Low corner
In1 Out1High corner
In1 Out1In1
1
nonlin_switched_10kHz/Controller/BPF
Figure B.1. Expanded Controller blocks from switched model.
121
Out1
1
Integrator
1s
1/tau1
-K-
In1
1
nonlin_switched_10kHz/Controller/BPF/Low corner
Figure B.1. (Continued.)
Out1
1
Integrator
1s
1/tau2
-K-
In1
1
nonlin_switched_10kHz/Controller/BPF/High corner
Figure B.1. (Continued.)
122
Out1
1
Integrator
1s
Hz to rad
-K-
Corner freq
-K-
In1
1
nonlin_switched_10kHz/Controller/Low corner
Figure B.1. (Continued.)
Out
1Subtract1
MinMax
max
Gain
-1
CompareTo Zero
>= 0
85% Duty Ratio Limit
-
2
+
1
nonlin_switched_10kHz/Controller/Comparator
Figure B.1. (Continued.)
123
B.1.2 Set-Up Code for Switched Nonlinear Model
% nominal or constant parameters % These parameters are for the 100 kW converter. L = 500e-6; C = 18000e-6; R = 1.6; Cuc = 44.16; Vin = 180; Vc = 400; Voc = Vc; Rbatt = 0.15; Resr = 0.00314; Rin = (17.2+1.5+2.4)*1e-3; T = 100e-6; %sampling period Ma_r = 0.7; %relationship of Ma to off-time slope of inductor current Kp = 65; Kchg = 0.4; Kichg = .00015; Kdchg = 0; tau1 = 3.71; %4.167; tau2 = 0.0291; %.001478; D_filter = 10; vin_corner = 2000; % %nominal parameters which take on a range of values % R_v = [1.3 13 1000]; % Vin_v = [5 5.9:1:12.9]; % Vc_v = [9:1:13]; % Rbatt_v = [0.1 1 10000]; % simulation parameters Pload_step = 1e5; s_t = 1e-6; %rise time of step r_t = 0.2; %length of rise time of ramp %Calculated parameters % clear Vterm; % Vterm = fzero(@(Vterm) Vterm_fcn(Vterm,Vin,Voc,Rbatt,Kp,T,L,Ma_r),Voc); Vuc_ref = Vin; Vterm = Vc; Dp = Vin/Vterm; D = 1 - Dp; M1 = Vin/L; %slope of rising inductor current; is equal to (Vin-Vswitch)/L in boost converter M2 = (Vterm-Vin)/L; %in lossless boost converter Ma = M2*Ma_r; % find initial states of inductor, capacitor, chg loop integrator h0T = get_f0T_handle(Voc,Vin,1e6,Rin,Resr,Rbatt,L,C,D,T); x0 = fsolve(h0T,[-Vin*D*T/(2*L);Voc],optimset('TolFun',1e-10)); Ilmin = x0(1); Vcmax = x0(2); h0DT = get_f0DT_handle(Voc,Vin,1e6,Rin,Resr,Rbatt,L,C); [ttt,xDT] = ode15i(h0DT,[0,D*T],[Ilmin;Vcmax],[0;0]); Iref_init = xDT(end,1)+Ma*D*T; Ibatt_min = (Vterm-Vcmax)/Rbatt; Iload_nom = (Voc-Vterm)/Rbatt; Iout_max = Pload_step/Voc+20;
124
Iload_step = Pload_step/Voc;
Helper functions used by fsolve and ode15i function h = get_f0DT_handle(Voc,Vin,R,Rin,Resr,Rbatt,L,C) h = @f0DT; function err = f0DT(t,x,xp) err = [0;0]; Vout = x(2) + xp(2)*C*Resr; err(1) = L*xp(1) - (Vin-Rin*x(1)); err(2) = C*xp(2) - ((Voc-Vout)/Rbatt-Vout/R); end end
function h = get_fDT_T_handle(Voc,Vin,R,Rin,Resr,Rbatt,L,C) h = @fDT_T; function err = fDT_T(t,x,xp) err = [0;0]; Vout = x(2) + xp(2)*C*Resr; err(1) = L*xp(1) - (Vin-Rin*x(1)-Vout); err(2) = C*xp(2) - (x(1)+(Voc-Vout)/Rbatt-Vout/R); end end
function h = get_f0T_handle(Voc,Vin,R,Rin,Resr,Rbatt,L,C,D,T) h = @f0T; function err = f0T(x0) % integrates systems f0DT and FDT_T from time 0 to T % initialize vars and function handles err = [0;0]; hDT = get_f0DT_handle(Voc,Vin,R,Rin,Resr,Rbatt,L,C); hT = get_fDT_T_handle(Voc,Vin,R,Rin,Resr,Rbatt,L,C); [t,x] = ode15i(hDT,[0,D*T],x0,[0;0],odeset('MaxStep',T/100)); [t2,x2] = ode15i(hT,[D*T,T],[x(end,1);x(end,2)],[0;x(end,1)],odeset('MaxStep',T/100)); % find area under Vc(t) t_shift = [0;t;t2(1:end-1)]; ddt = [t;t2]-t_shift; vc_area = [x(:,2)' x2(:,2)']*ddt; % These metrics are valid when the converter is sourcing no power. % Note that the square of a small number is even smaller - fsolve % tolerance needs to be very small for these equations to be % useful. err(1) = 2*(x0(1)+x(end,1)-Voc/R)^2 + (x0(1)-x2(end,1))^2; err(2) = 1000*(vc_area-Voc*T*R/(R+Rbatt))^2 + (x0(2)-x2(end,2))^2; end end
125
B.2 Averaged Nonlinear Model for Vehicle
B.2.1 Expanded Model
This section contains the averaged nonlinear model with significant blocks ex-
panded. The overview drawing from Section 4.2 is reprinted here for clarity. The header
of each sub-figure in Figure B.2 shows the name of the expanded block, along with its
location in the model hierarchy. Individual function blocks implement the equations
given in Section 4.2.
126
load R
Vc
Pload_desRiref_to_iL
Vin
i_ref
v_c
dp
i_L
ibatt, il=-iuc, iref
i_uc
-u(1)
i_c
Vin
i_L
Vc
R
i_c
dp
Voc
ibatt
Voc, Vc, Vin_state,
Vin_term
Vin_term
f(u)
UC energy
Terminator1
TerminatorRepeating
Sequence1
Product Integrator2
1
s
Integrator1
1
s
Integrator
1
s
From
Workspace
sine_vec
Dp
Controller
Vin
Vciref
100 kW pulse
? ? ?
1/Cuc
-K-
1/C
-K-Vc
Vin
nonlinear_avgd_model_lossy2_100kW
Figure B.2. Averaged nonlinear model of combined power source for vehicle.
127
iref
1
Vuc,ref
-C-
P
Kp
Kichg
-K-
Kchg
-K-
Integrator1
1s
BPF
In1 Out1
Vc
2
Vin
1
nonlinear_avgd_model_lossy2_100kW/Controller
Figure B.2. (Continued.)
ibatt
4
Voc
3
dp
2
i_c
1
ic_fcn
f(u)
d'
Vin
i_L
Vc
Voc
R
i_c
Dp
ibatt
Voc1
ibatt Voc
Req
f(u)
Memory2
Memory1
Memory
R
4
Vc
3
i_L
2
Vin
1
nonlinear_avgd_model_lossy2_100kW/i_c
Figure B.2. (Continued.)
128
ibatt
2
Dp
1
iload
f(u)
ibatt_fcn
f(u)
d'
f(u)
i_c
6
R
5
Voc
4
Vc
3
i_L
2
Vin
1
nonlinear_avgd_model_lossy2_100kW/i_c/d'
Figure B.2. (Continued.)
Voc
1
Integrator
1s
Gain
-Kbatt
ibatt
1
nonlinear_avgd_model_lossy2_100kW/i_c/Voc1
Figure B.2. (Continued.)
129
i_L
1
iref-to-il
f(u)
d
1-u(1)
dp
4
v_c
3
i_ref
2
Vin
1
nonlinear_avgd_model_lossy2_100kW/iref_to_iL
Figure B.2. (Continued.)
R
1
Fcn
u(1)^2/u(2)
Pload_des
2
Vc
1
nonlinear_avgd_model_lossy2_100kW/load R
Figure B.2. (Continued.)
130
B.2.2 Set-Up Code for Averaged Nonlinear Model
% nonlinear averaged model setup % These parameters are for the 100 kW converter. % physical parameters L = 500e-6; C = 18000e-6; Cuc = 44.16; %3000*2.5/320; Voc_init = 400; Vuc_init = 266.2; Resr = 0.00314; Rin = (17.2+1.5+2.4)*1e-3; Rbatt = 0.15; batt_ah = 147; T = 1e-4; Ma_r = 0.7; Pmax = 1e5; %controller parameters tau1 = 3.71; %HPF cutoff tau2 = 0.0291; %.001478; D_filter = 10; Kp = 65; Kchg = .4; %.1; Kichg = 0.00015; %.00001; Kdchg = 0; %100; %calculated parameters Kbatt = 0; %(12.6-9)/(3600*batt_ah); uc_soc_init = .55; Vuc_ref = sqrt(abs(-(1-uc_soc_init)*(.45*Voc_init)^2-uc_soc_init*(.8*Voc_init)^2)); M2 = (400-180)/L; % abs. value of off-time slope of inductor current Ma = Ma_r*M2; Dp = Vuc_ref/Voc_init; D = 1-Dp; Iref_init = (1/2)*(D^2*Vuc_ref*T/(2*L) + Dp^2*Voc_init*T/(2*L)) + Ma*D*T % create sinusoid time_vec = [0:1:14400]; sine_vec = 20*sin(2*pi*time_vec./2000); sine_vec = [time_vec;sine_vec]';
B.3 Switched Model for VRM
B.3.1 Expanded Model
This section contains the switched model of the VRM system. The system over-
view drawing from Section 4.3 is reprinted here for clarity. The header of each sub-figure
in Figure B.3 shows the name of the expanded block, along with its location in the model
hierarchy.
131
v ref
vzero
dc pcm process
IL_1
IL_2
IL_3
IL_4
Iref/4
Scope out
q(t)_1
q(t)_2
q(t)_3
q(t)_4
ac pcm process
IL_1
IL_2
Iref/2
Scope out
q(t)_1
q(t)_2
Vin, Vout
Outer loopcontroller
sum_IL_s
Vout
Vzero
sum IL f
Iref_s
Iref_f
Load R
Out1
Irefs, ILs
Divide
Dc phase 4
vin
iout
q(t)
vc
iL
iin
Dc phase 3
vin
iout
q(t)
vc
iL
iin
Dc phase 2
vin
iout
q(t)
vc
iL
iin
Dc phase 1
vin
iout
q(t)
vc
iL
iin
BusCapacitor
i_c
vout
vc
Ac phase 2
vin
iout
q(t)
vc
iL
iin
Ac phase 1
vin
iout
q(t)
vc
iL
iin
2 V source
iinVin
iout
nonlin_switched_slowfast
Figure B.3. Switched model for VRM application with significant blocks expanded.
132
Iref_f
2
Iref_s
1
iref_s/4
1/4
iref_f/2
1/2
iL/4
1/4
iL/2
1/2
Native AVP1
iL
iref
ioff
Native AVP
iL
iref
ioff
LPF
In1 Out1
Kp/4
Kp
BPF
In1 Out1
sum_IL_f
4
Vzero
3
Vout
2
sum_IL_s
1
Vzero iref_s
iref_f
nonlin_switched_slowfast/Outer loop controller
Figure B.3. (Continued.)
133
tau1 = low cutofftau2 = high cutoff
TF is (F1(s)-1)(F2(s) = -s*tau1/((s*tau1+1)(s*tau2+1))
Out1
1
Low corner
In1 Out1High corner
In1 Out1In1
1
nonlin_switched_slowfast/Outer loop controller/BPF
Figure B.3. (Continued.)
Out1
1
Integrator
1s
1/tau1
-K-
In1
1
nonlin_switched_slowfast/Outer loop controller/BPF/Low corner
Figure B.3. (Continued.)
134
Out1
1
Integrator
1s
1/tau2
-K-
In1
1
nonlin_switched_slowfast/Outer loop controller/BPF/High corner
Figure B.3. (Continued.)
Out1
1
Integrator
1s
1/tau1
-K-
In1
1
nonlin_switched_slowfast/Outer loop controller/LPF
Figure B.3. (Continued.)
135
ioff
1
filter coef
-K-
Integrator1
1s
iref
2
iL
1
nonlin_switched_slowfast/Outer loop controller/Native AVP
Figure B.3. (Continued.)
Vin
1
line step
Rsrc
-K-
Manual SwitchConstant
2
iin
1
nonlin_switched_slowfast/2 V source
Figure B.3. (Continued.)
136
q(t)_4
5
q(t)_3
4
q(t)_2
3
q(t)_1
2
Scope out
1
Terminator3
Terminator2
Terminator1
Terminator
Stabilizing Ramp4
Stabilizing Ramp2
Stabilizing Ramp1
Stabilizing Ramp
SR latch3
S
R
Q
Q'
SR latch2
S
R
Q
Q'
SR latch1
S
R
Q
Q'
SR latch
S
R
Q
Q'
Comparator3
+
-
Out
Comparator2
+
-
Out
Comparator1
+
-
Out
Comparator
+
-
Out
Clock 4
Clock 3
Clock 2
Clock 1
Iref/4
5
IL_4
4
IL_3
3
IL_2
2
IL_1
1
nonlin_switched_slowfast/dc pcm process
Figure B.3. (Continued.)
137
q(t)_2
3
q(t)_1
2
Scope out
1
Terminator1
Terminator
Stabilizing Ramp1
Stabilizing Ramp
SR latch1
S
R
Q
Q'
SR latch
S
R
Q
Q'
Comparator1
+
-
Out
Comparator
+
-
Out
Clock 2
Clock 1
Iref/2
3
IL_2
2
IL_1
1
nonlin_switched_slowfast/ac pcm process
Figure B.3. (Continued.)
138
iin
2
iL
1
iLstate
1s
Vd
Vd
Rs
Rs
Rl+Resr
-K-
Resr1
-K-
Rd
Rd
Logic Invert Data Type Conversion3
Convert
Data Type Conversion
Convert
1/L
1/L
vc
4
q(t)
3
iout
2
vin
1
nonlin_switched_slowfast/Dc phase 1
Note: Other phases follow this topology.
Figure B.3. (Continued.)
139
vc
2
vout
1
cap
1s
TransportDelay
Resr
-K-
1/C
-K-
i_c
1
nonlin_switched_slowfast/Bus Capacitor
Figure B.3. (Continued.)
B.3.2 Set-Up Code
% Parameters for the PCM batt-ucap converter, 10kHz, 1000uF output % capacitance % This tuning is stable for new controller % Plant parameters Lfast = 10e-9; Ldc = 500e-9; C = 670e-6; R = 0.08; Vin = 2; Vc = 0.8; Voc = 0.8; Rth = 0.8e-3; Resr = Rth; Rin_s = (1.75+2.4)*1e-3; Rin_f = (1+6.5)*1e-3; Rsw_s = 2.4e-3; Rsw_f = 6.5e-3; Rsrc = 5e-3; T_fast = 1/5e6; %fast switching period T_slow = 1/150000; %slow switching period Ma_r = 0.7; %relationship of Ma to off-time slope of inductor current Kdc = 1/Rth; Kp = 1/Rth; tau1 = 1/(2*pi*3000); tau2 = T_fast*0.1; %.001478; D_filter = 10;
140
navp_corner = 200; % simulation parameters Pload_step = 160; s_t = 1e-6; %rise time of step r_t = 0.2; %length of rise time of ramp %Calculated parameters % clear Vterm; % Vterm = fzero(@(Vterm) Vterm_fcn(Vterm,Vin,Voc,Rbatt,Kp,T,L,Ma_r),Voc); Vterm = Vc; D = Vterm/Vin; Dp = 1 - D; M2dc = Vterm/Ldc; %slope of rising inductor current; is equal to (Vin-Vswitch)/L in boost converter M1dc = (Vterm-Vin)/Ldc; %in lossless boost converter Madc = M2dc*Ma_r; M2fast = Vterm/Lfast; %slope of rising inductor current; is equal to (Vin-Vswitch)/L in boost converter M1fast = (Vterm-Vin)/Lfast; %in lossless boost converter Mafast = M2fast*Ma_r; % find initial states of slow inductors h0T = get_f0T_handle(Vc,Vin,1e6,Rin_s,Resr,Ldc,C,D,T_slow); x0 = fsolve(h0T,[-(Vin-Vc)*D*T_slow/(2*Ldc);Vc],optimset('TolFun',1e-10)); Ilmin1 = x0(1); Vcmax = Vc+(x0(2)-Vc)/4; if D < 0.25 h0DT = get_f0DT_handle(Vin,1e6,Rin_s,Resr,Ldc,C); hT = get_fDT_T_handle(1e6,Rin_s,Resr,Ldc,C); [~,xDT] = ode15i(h0DT,[0,D*T_slow],[Ilmin1;Vcmax],... [(Vin-x0(2))/Ldc;x0(1)/(1e6*C)],odeset('MaxStep',T_slow/100)); [t2,x2] = ode15i(hT,[D*T_slow,0.25*T_slow],[xDT(end,1);xDT(end,2)],... [-xDT(end,2)/Ldc;(xDT(end,1)-xDT(end,2))/(1e6*C)],... odeset('MaxStep',T_slow/100)); Ilmin2 = x2(end,1); Q2init = 0; else h0DT = get_f0DT_handle(Vin,1e6,Rin_s,Resr,Ldc,C); [~,xDT] = ode15i(h0DT,[0,0.25*T_slow],[Ilmin1;Vcmax],... [(Vin-x0(2))/Ldc;x0(1)/(1e6*C)],odeset('MaxStep',T_slow/100)); Ilmin2 = xDT(end,1); Q2init = 1; end if D < 0.5 h0DT = get_f0DT_handle(Vin,1e6,Rin_s,Resr,Ldc,C); hT = get_fDT_T_handle(1e6,Rin_s,Resr,Ldc,C); [~,xDT] = ode15i(h0DT,[0,D*T_slow],[Ilmin1;Vcmax],... [(Vin-x0(2))/Ldc;x0(1)/(1e6*C)],odeset('MaxStep',T_slow/100)); [t2,x2] = ode15i(hT,[D*T_slow,0.5*T_slow],[xDT(end,1);xDT(end,2)],... [-xDT(end,2)/Ldc;(xDT(end,1)-xDT(end,2))/(1e6*C)],... odeset('MaxStep',T_slow/100)); Ilmin3 = x2(end,1); Q3init = 0; else h0DT = get_f0DT_handle(Vin,1e6,Rin_s,Resr,Ldc,C); [~,xDT] = ode15i(h0DT,[0,0.5*T_slow],[Ilmin1;Vcmax],... [(Vin-x0(2))/Ldc;x0(1)/(1e6*C)],odeset('MaxStep',T_slow/100)); Ilmin3 = xDT(end,1); Q3init = 1; end if D < 0.75
141
h0DT = get_f0DT_handle(Vin,1e6,Rin_s,Resr,Ldc,C); hT = get_fDT_T_handle(1e6,Rin_s,Resr,Ldc,C); [~,xDT] = ode15i(h0DT,[0,D*T_slow],[Ilmin1;Vcmax],... [(Vin-x0(2))/Ldc;x0(1)/(1e6*C)],odeset('MaxStep',T_slow/100)); [t2,x2] = ode15i(hT,[D*T_slow,0.75*T_slow],[xDT(end,1);xDT(end,2)],... [-xDT(end,2)/Ldc;(xDT(end,1)-xDT(end,2))/(1e6*C)],... odeset('MaxStep',T_slow/100)); Ilmin4 = x2(end,1); Q4init = 0; else h0DT = get_f0DT_handle(Vin,1e6,Rin_s,Resr,Ldc,C); [~,xDT] = ode15i(h0DT,[0,0.75*T_slow],[Ilmin1;Vcmax],... [(Vin-x0(2))/Ldc;x0(1)/(1e6*C)],odeset('MaxStep',T_slow/100)); Ilmin4 = xDT(end,1); Q4init = 1; end % get initial state of slow iref integrator h0DT = get_f0DT_handle(Vin,1e6,Rin_s,Resr,Ldc,C); [~,xDT] = ode15i(h0DT,[0,D*T_slow],[Ilmin1;Vcmax],... [(Vin-x0(2))/Ldc;x0(1)/(1e6*C)]); Iref_init_dc = xDT(end,1)+Madc*D*T_slow; % find initial states of fast inductors clear h0T h0DT hT; h0T = get_f0T_handle(Vc,Vin,1e6,Rin_f,Resr,Lfast,C,D,T_fast); x0 = fsolve(h0T,[-(Vin-Vc)*D*T_fast/(2*Lfast);Vc],... optimset('TolFun',1e-10)); Ilmin5 = x0(1); % Vcmax = Vc+(x0(2)-Vc)/4; if D < 0.5 h0DT = get_f0DT_handle(Vin,1e6,Rin_f,Resr,Lfast,C); hT = get_fDT_T_handle(1e6,Rin_f,Resr,Lfast,C); [~,xDT] = ode15i(h0DT,[0,D*T_fast],[Ilmin5;Vcmax],... [(Vin-x0(2))/Lfast;x0(1)/(1e6*C)],odeset('MaxStep',T_fast/100)); [t2,x2] = ode15i(hT,[D*T_fast,0.5*T_fast],[xDT(end,1);xDT(end,2)],... [-xDT(end,2)/Lfast;(xDT(end,1)-xDT(end,2))/(1e6*C)],... odeset('MaxStep',T_fast/100)); Ilmin6 = x2(end,1); Q6init = 0; else h0DT = get_f0DT_handle(Vin,1e6,Rin_f,Resr,Lfast,C); [~,xDT] = ode15i(h0DT,[0,0.5*T_fast],[Ilmin5;Vcmax],... [(Vin-x0(2))/Lfast;x0(1)/(1e6*C)],odeset('MaxStep',T_fast/100)); Ilmin6 = xDT(end,1); Q6init = 1; end % get initial state of fast iref integrator h0DT = get_f0DT_handle(Vin,1e6,Rin_f,Resr,Lfast,C); [ttt,xDT] = ode15i(h0DT,[0,D*T_fast],[Ilmin5;Vcmax],... [(Vin-x0(2))/Lfast;x0(1)/(1e6*C)]); Iref_init = xDT(end,1)+Mafast*D*T_fast; % find needed C value to keep iref from exceeding il slew il_step = 120; Vin0 = Vin-il_step*Rsrc; NN = 2; %# of phases Cmin = il_step*Lfast*Kp/((0.4)*Vin0*NN^2)
142
Helper functions for fsolve and ode15i function h = get_f0DT_handle(Vin,R,Rin,Resr,L,C) h = @f0DT; function err = f0DT(t,x,xp) err = [0;0]; Vout = x(2) + xp(2)*C*Resr; err(1) = L*xp(1) - (Vin-Rin*x(1)-Vout); err(2) = C*xp(2) - (x(1)-Vout/R); end end
function h = get_fDT_T_handle(R,Rin,Resr,L,C) h = @fDT_T; function err = fDT_T(t,x,xp) err = [0;0]; Vout = x(2) + xp(2)*C*Resr; err(1) = L*xp(1) + (Rin*x(1)+Vout); err(2) = C*xp(2) - (x(1)-Vout/R); end end
function h = get_f0T_handle(Vc,Vin,R,Rin,Resr,L,C,D,T) h = @f0T; function err = f0T(x0) % integrates systems f0DT and FDT_T from time 0 to T % initialize vars and function handles err = [0;0]; hDT = get_f0DT_handle(Vin,R,Rin,Resr,L,C); hT = get_fDT_T_handle(R,Rin,Resr,L,C); [t,x] = ode15i(hDT,[0,D*T],x0,[(Vin-x0(2))/L;x0(1)/(R*C)],odeset('MaxStep',T/100)); [t2,x2] = ode15i(hT,[D*T,T],[x(end,1);x(end,2)],[-x(end,2)/L;(x(end,1)-x(end,2))/(R*C)],odeset('MaxStep',T/100)); % find area under Vc(t) t_shift = [0;t;t2(1:end-1)]; ddt = [t;t2]-t_shift; vc_area = [x(:,2)' x2(:,2)']*ddt; % These metrics are valid when the converter is sourcing no power. % Note that the square of a small number is even smaller - fsolve % tolerance needs to be very small for these equations to be % useful. err(1) = 2*(x0(1)+x(end,1)-Vc/R)^2 + (x0(1)-x2(end,1))^2; err(2) = 1000*(vc_area-Vc*T)^2 + (x0(2)-x2(end,2))^2; end end
143
APPENDIX C
SCHEMATICS FOR BIDIRECTIONAL PCM BOOST CONVERTER
This section lists the schematics for the scaled-down hardware. Figure C.1 is the
schematic for parts on the power converter board, and Figure C.2 is the schematic for a
second board with signal conditioning circuitry. In Figure C.1, parts with 9000-series
reference designators were added after the PCB was made, and pads do not exist for
them. The connection of these parts to the rest of the circuit is shown with thick wires.
144
Battery
-Ultra
capacitor
Inte
rface
PC
B50027
A02
UC
+
UC
-
Out +
Out -
(Vin
)
Pow
er
Convert
er
and C
onnections
Batt +
Batt -
Board
sta
ndoffs
Fo
r 3
00
kH
z v
ers
ion
, in
sta
ll 6
8u
F 1
00
V c
ap
s in
ste
ad
of 1
00
0u
F
Fo
r 3
00
kH
z v
ers
ion
, in
sta
ll 1
0u
H in
du
cto
r
10 turn
s o
n 2
x T
200-2
6 c
ore
SA
I: C
onnect D
9004 a
cro
ss Q
1
SA
I: C
onnect D
9003 a
cro
ss Q
2U
se T
O220 leaded p
ackage
Use T
O220 leaded p
ackage
E D C B A
12
34
56
E D C B A
2512
2512
2512
2512
2512
2512VB
60100C
-E3/4
WT
O-2
63
AB
VB
60100C
-E3/4
WT
O-2
63
AB
T400-6
0D
Z-P
ow
er
50 A
mp
0.0
02
CR
F2512F
VR
002E
LF
AT
C F
use H
old
er
0.0
02
CR
F2512
FV
R002E
LF
Z-P
ow
er
50 A
mp
3 P
IN H
EA
DE
R
3 P
IN H
EA
DE
R
18121812
TO
-22
0A
B
TO
-22
0A
B
3 P
IN H
EA
DE
R
3 P
IN H
EA
DE
R
1210
1210
1210
2.2
IN
CH
SP
AC
ER
2.2
IN
CH
SP
AC
ER
2.2
IN
CH
SP
AC
ER
2.2
IN
CH
SP
AC
ER
VB60100C-E3/4WTO-263AB
VB60100C-E3/4WTO-263AB
12
R80
12
R79
12
R78
12
R71
12
R72
12
R73
1 2
3 D7
1 2
3 D6
12
L500
12345678910
J9
12
R70
21
C27
21
C26
21
C32
21
C46
21
C47
21
C48
21
C49
21
C51
21
C52
21
C53
21
C54
1 2
3 4
FH
1
12
R77
12345678910
J6
11
22
33
J11 1
1
22
33
J10
12
C4012
C36
Q2 Q
1
11
22
33
J8
11
22
33
J7
12C45
12
C50
12
C33
SP
211
SP
411
SP
3
11
SP
111
1
2
3
D9003
1
2
3
D9004
TO
P_G
AT
E
BO
TT
OM
_G
AT
E
IL_IN
V
IBA
TT
_N
ON
INV
IBA
TT
_IN
V
VIN
_A
ND
_IL
_N
ON
INV
TO
P_S
RC
5.6
5.6
5.6 5
.6
5.6
5.6
100uH
330uF 35V
330uF 35V
330uF 35V
1000uF 35V
1000uF 35V
1000uF 35V
1000uF 35V
1000uF 35V
1000uF 35V
1000uF 35V
1000uF 35V
GN
D
GN
D
GN
D
GN
D
GN
D
0.1uF 100V0.1uF 100V
IRF
2903Z
PB
F
IRF
2903Z
PB
F
GN
D
10uF
10uF
10uF
GN
D
GN
D
F
igu
re C
.1. S
chem
atic
for
pow
er c
onve
rter
PC
B.
145
Battery
-Ultra
capacitor
Inte
rface
PC
B50027
A02
Gate
Drive a
nd E
nable
Circuit
LE
D S
hin
es w
hen F
ET
s D
isable
d
Layout B
ridge
110ns D
ead T
ime
150ns D
ead T
ime
40ns p
ropagation d
ela
y
Hex Invert
er
SA
I: R
em
ove R
55 a
nd D
2,
Insta
ll U
90
00
, C
90
00
, C
90
01
connecte
d to the n
odes s
how
n.
Use g
round p
ad o
f D
2 a
s g
round.
Insta
ll D
90
00
on
to D
3's
pa
ds
Insta
ll D
90
01
on
to D
4's
pa
ds
Revers
e o
rder
of C
34 a
nd C
35 -
insta
ll 0.1
uF
part
clo
ser
to U
10
than the 1
uF
part
E D C B A
12
34
56
E D C B A
MA
X5062A
AS
A+
SO
IC-8
HE
F4069U
BT
SO
IC-1
4
1206
MA
X4542E
UA
+8-u
MA
X
TLV
2371ID
RS
OIC
-8
SM
L-L
X0603IW
-TR
0603 R
ed L
ED
RC
A J
AC
K B
LA
CK
0603
0603
1206
1206
0603
0603
0603
0603 0603
0603
0603
1210
0603
1206
1206
0603
0603
0603
0603
0603
0603
0603
1206
DO
-214A
C
DO
-214A
C
DO
-214A
C
0603
0603
12.0 V Zener
SOD-123
SO
D-1
23
TO
-92
TL750L12C
LP
VD
D1
BO
OS
T2
OU
T_H
3
H_S
RC
4IN
_H
5
IN_L
6
GN
D7
OU
T_L
8
U10
1A
1
1Y
2
2A
3
2Y
4
3A
5
3Y
6
GN
D7
4Y
8
4A
9
5Y
10
5A
11
6Y
12
6A
13
VC
C14
U5
12
R75
2
1 3
SW
1
NC
11
CO
M1
2
SW
_2
3
GN
D4
NC
25
CO
M2
6
SW
_1
7V
+8
U6
NC
_1
IN-
2
IN+
3
GN
D4
NC
5
OU
T6
V+
7
_N
C8
U9
21
D1
1
2
J1
12
R8
12
R38
12
R52
12
R55
12
C16
1 2
C15
12
C19
12
R54
12
R53
12
R37
12
R36
12
C31
12
C30
12
C34
12
R69
12
R1
12
R63
12
R68
12
R64
12
R65
12
R67
12
R66
12
C35
12
D3
12
D4
12 D
5
12
C29
12
C28
1 2
D2
12
FB
1
12D9000
12
D9001
21
C9000
21
C9001
IN1
GND2
OU
T3
U9000
+11V
+11V
+11V
+12V
+12V
+12V
PW
M_O
UT
TO
P_G
AT
E
BO
TT
OM
_G
AT
E
TO
P_S
RC
0.0
MO
UN
TA
IN_108-0
051-E
VX
GN
D
27k
27k
1k
GN
D
GN
D
GN
D
GN
D
GN
D
NI
0.1uF
0.1uF
GN
D
GN
D
GN
D
0.1uF
GN
D
GN
D
100k 100k
GN
D
10
10
10uF
0.1uF
GN
D
1uF
0.0
GN
D
47k
GN
D
100k
100k
GN
D
1k
100
1.2
k220
GN
D
GN
D
0.1uF
SS
13-E
3/6
1T
SS
13-E
3/6
1T
B330LA
-E3/6
1T
100pF
100pF
MMSZ4699-V-GS08
BLM
18E
G601S
N1D
MMSD701T1G
MM
SD
701T
1G
10uF
10uF
GN
DG
ND
F
igu
re C
.1. (
Con
tinu
ed.)
146
Battery
-Ultra
capacitor
Inte
rface
PC
B50027
A02
I_batt s
ignal to
AD
C
Gain
: 100 m
V/1
A
Range +
/- 1
00 A
V_in
sig
nal to
AD
C
Gain
= 1
/2
I_L s
ignal to
AD
C
Gain
: 100 m
V/1
A
Range +
/- 1
00 A
BW
23kH
z
BW
35kH
z
BW
35kH
z
Outp
uts
to A
DC
SA
I: Insta
ll 100k trim
pots
to U
8 a
nd U
4,
connect as s
how
n
SA
I: Insta
ll 100k trim
pot to
U3,
connect w
iper
to p
in 7
of U
3
IBA
TT
_IN
V
Analo
g H
PF
of V
out
tau =
3.7
1
E D C B A
12
34
56
E D C B A
0603
0603
0603
0603
0603
0603
AD
8675A
RM
Z8-u
MA
X
0603
0603
0603
0603
0603 0603
0603
RC
A J
AC
K W
HIT
E
AD
8675A
RM
Z8-u
MA
X
0603 0603
0603
0603
0603
0603
0603
0603
RC
A J
AC
K W
HIT
E
AD
8675A
RM
Z8-u
MA
X
0603
0603
0603
0603
0603
0603
0603
0603
0603
0603
RC
A J
AC
K W
HIT
E
0603
0603
06030603
AD
8675A
RM
Z8-u
MA
X
0603
0603
0603
0603
0603
0603
0603
0603
3296W
3296W
3296W
1206
1206
1206
1206
0603
12
R58
12
R61
12
R59
12
R60
12
R47
12
R62
NU
LL
1
IN-
2
IN+
3
V-
4N
C5
OU
T6
V+
7
_N
ULL
8
U8
12
R48
12
R51
12
C23
12
C18
12
R50
12
R46
12
R49
1
2
J5
NU
LL
1
IN-
2
IN+
3
V-
4N
C5
OU
T6
V+
7
_N
ULL
8
U2
12
R3
12
R4
12
R22
12
R6
12
R24
12
R25
12
C2
12
C10
1
2
J3
NU
LL
1
IN-
2
IN+
3
V-
4N
C5
OU
T6
V+
7
_N
ULL
8
U3
12
R26
12
R27
12
R28
12
R21
12
R23
12
C11
12
C12
12
R30
12
R45
12
R29
1
2
J4
12
C9
12
C1
12
C24
12
C25
NU
LL
1
IN-
2
IN+
3
V-
4N
C5
OU
T6
V+
7
_N
ULL
8
U4
12
R33
12
R34
12
C13
12
C14
12
R35
12
R32
12
R7
1 2R31
1 3
2
VR9000
1 3
2
VR9001
13
2
VR
9002
12
C9003
12
R9001
12
R9003
12
R9002
12
R9004
12
R9005
IBA
TT
_IN
V
IBA
TT
_N
ON
INV
+12V
+12V
+12V
+12V
-12V
-12V
-12V
-12V
VIN
_A
ND
_IL
_N
ON
INV
+1V
_R
EF
IL_1S
T_A
MP
_O
UT
820
820
1.2
k
1.2
k
10k
10k
100
100
0.1uF
0.1uF
10 10
GN
D
GN
D
GN
D
GN
D
GN
D
NI
GN
D
27k 27k
100
10
10
100
0.1uF
0.1uF
GN
D
GN
D
GN
D
GN
D
GN
D
10k
82k
8.2
k
1k
1k
0.1uF
0.1uF
10
10
GN
D
GN
D
GN
D
100
GN
D
6800pF
180pF
5600pF5600pF
820
8.2
k
0.1uF
0.1uF
10
10
GN
D
GN
D
100
1k
GN
D
100k
100k
100k
3.8
5uF
2.2M
1M
11M
39k
GN
D
100
F
igu
re C
.1. (
Con
tinu
ed.)
147
Battery
-Ultra
capacitor
Inte
rface
PC
B50027
A02
PW
M C
ontr
olle
r and I_L A
mplif
ier
(Vbus)
Pow
er
for
Contr
ol C
ircuitry
BW
1.8
5 M
Hz
BW
1.8
5 M
Hz
I_re
f fr
om
AD
C
Com
pensation A
dju
st
Ma,r
= 2
at 44.4
kO
hm
s
For
300kH
z b
oard
, re
pla
ce C
17
with 2
20
0p
F 1
206 C
0G
part
.
On r
ev A
01, lif
t one leg o
f L1 a
nd
connect to
an e
xte
rnal +
15V
supply
SA
I: Insta
ll 100k trim
pot to
U1, con
nect w
iper
to p
in 7
of U
1
SA
I: R
em
ove
R4
0, in
sta
ll D
90
02
onto
R40's
pads. C
ut and b
end
leads a
s n
ecessary
.
+15V
SA
I fo
r 10kH
z b
oard
only
:
Insta
ll R
90
00
an
d C
90
02
in
pa
ralle
l w
ith
C1
7.
Locate
part
s p
hysic
ally
on top o
f C
17.
E D C B A
12
34
56
E D C B A
UC
3823A
DW
SO
IC-1
6 W
ide
TC
7662B
CO
AS
OIC
-8
SD
R0805
0603
1210
1210
12100603
0603
0603
0603
0603
0603
0603
AD
8675
AR
MZ
8-u
MA
X
0603
0603
0603
0603
0603 0603
0603
3296W0603
0603
0603
RC
A J
AC
K B
LA
CK
0603
0603
0603
0603
0603 0603
06030603
1210
06030603
0603
1210
1210
0603
0603
3296W
LM
385-1
.2T
O-9
2
06030603
INV
1
NO
NIN
V2
EA
OU
T3
CLK
/LE
B4
RT
5
CT
6
RA
MP
7
SS
8
I_LIM
9
GND10
OU
TA
11
PGND12
VC
13
OU
TB
14
VCC15
VR
EF
16
U7
BO
OS
T1
CA
P+
2
GN
D3
CA
P-
4
-VO
UT
5
LV
6
OS
C7
+V
IN8
U11
21
C42
21
C41
2 1
C37
2 1
C38
12
L1
12
R74
12
C39
12
C43
12
C44
12
R76
12
R10
12
R9
12
R13
12
R12
12
R17
12
R14
NU
LL
1
IN-
2
IN+
3
V-
4N
C5
OU
T6
V+
7
_N
ULL
8
U1
12
R18
12
R11
12
C6
12
C8
12
R19
12
R5
12
R20
13
2
VR1
12
C7
12
R42
12
R15
1
2J2
12
R41
12
R2
12
R43
1 2R16
12
R44
12
R57
12
C5
12
C4
1 2
C17
12
R56
12
R39
12
R40
12
C3
12
C20
12
C21
12
FB
2
12
C22
13
2
VR9003
2 1
NC
3
D9002
12
R9000
12
C9002
+11V
PW
M_O
UT
IBA
TT
_IN
V
-12V
-12V
+12V
+12V
+12V
IL_IN
V
VIN
_A
ND
_IL
_N
ON
INV
+1V
_R
EF
+1V
_R
EF
IL_1S
T_A
MP
_O
UT
GN
D
330uF 35V
330uF 35V
330uF 35V
330uF 35V
47uH
GN
D10
GN
D
GN
D
10uF
10uF
10uF0.0
GN
D
22
22
2.2
k
2.2
k
1k
1k
10k
10k
0.1uF
0.1uF
10 10
GN
D
GN
D
GN
D
GN
D
1.2
k
100k0.1uF
0.0
NI
GN
D
1k
47k
GN
D
GN
D
1k
1k
1M 1k
GN
D
GN
DG
ND
3900pF3900pF
0.068uF
1.5k100
NI
10uF
10uF
0.1uF
GN
D
GN
D
BLM
18E
G601S
N1D
0.1uF
GN
D
100k
0.06800pF
F
igu
re C
.1. (
Con
tinu
ed.)
148
5 5
4 4
3 3
2 2
1 1
DD
CC
BB
AA
I_L
-
I_L
+
V-
V+
V-
V+
V+
V-
To
AD
C C
h5
I_ba
tt -
I_ba
tt +
V-
V+
V-
V+
V+
V-
To
AD
C C
h6
Titl
e
Siz
eD
ocum
ent N
umbe
rR
ev
Dat
e:S
heet
of
<D
oc>
1
Aux
iliar
y S
igna
l Mea
surin
g C
ircui
ts
A
12
Tue
sday
, Aug
ust 1
0, 2
010
Titl
e
Siz
eD
ocum
ent N
umbe
rR
ev
Dat
e:S
heet
of
<D
oc>
1
Aux
iliar
y S
igna
l Mea
surin
g C
ircui
ts
A
12
Tue
sday
, Aug
ust 1
0, 2
010
Titl
e
Siz
eD
ocum
ent N
umbe
rR
ev
Dat
e:S
heet
of
<D
oc>
1
Aux
iliar
y S
igna
l Mea
surin
g C
ircui
ts
A
12
Tue
sday
, Aug
ust 1
0, 2
010
Refe
renc
e Ad
just
Refe
renc
e Ad
just
Indu
ctor
Cur
rent
Sen
se A
mp
Gain =
50
BW
~ 1
00 k
Hz
Batt
ery
Curr
ent
Sens
e Am
pGa
in =
10
BW
= 1
00 k
Hz
Cur
rent
Sen
se A
mps
All r
esis
tors
sho
uld
be 1
%.
Resi
stor
s R3
& R
4 an
d R1
6 &
R17
shou
ld b
e m
atch
ed b
y ha
nd f
or
best
gai
n ac
cura
cy a
cros
s fr
eque
ncy.
C2, C
3, C
11, C
12 a
nd C
21 s
houl
d be
50
V N
P0/C
0G c
eram
ic t
ype.
From
2 m
Ohm
I-
sens
e R
From
100
mO
hm
I-se
nse
RC
310
00p
C3
1000
p
R5
10R
510
R20
10R20
10R
2510
kR
2510
k
R4
1.5k
R4
1.5k
C14
0.1u
C14
0.1u
C6
0.1uC
60.
1u
C16
0.1u
C16
0.1u
R24
10k
R24
10k
R12
10k
R12
10k
C13
0.1u
C13
0.1u
C15
0.1u
C15
0.1u
R23
100
R23
100
C12
1000
pC
1210
00p
R22
100
R22
100
C7
0.1u
C7
0.1u
R14
4.7k
R14
4.7k
C4
0.1u
C4
0.1u
C18
0.1u
C18
0.1u
C11
1000
pC
1110
00p
C9
0.1uC
90.
1u
R2
12R
212
C17
0.1u
C17
0.1u
C8
0.1uC
80.
1u
R11
100
R11
100
R10
100
R10
100
C10
0.1u
C10
0.1u
VR
110
kV
R1
10k
R6
10R6
10
R1
1kR
11k
VR
210
kV
R2
10k
C5
0.1u
C5
0.1u
U1
LT19
20U
1LT
1920
6O
UT
+3 2
-5
RE
F
V+7
V-4
RG
+8
RG
-1
C1
0.1u
C1
0.1u
R15
820
R15
820
U2
NE
5534
U2
NE
5534
OU
T
+ -tr
imtr
im
V+ V-
U3
LT19
20U
3LT
1920
6O
UT
+3 2
-5
RE
F
V+7
V-4
RG
+8
RG
-1
R7
10R7
10
R17
1.5k
R17
1.5k
R9
10k
R9
10k
R16
1.5k
R16
1.5k
R13
100
R13
100
C2
1000
pC
210
00p
R19
10R19
10
R18
10R
1810
R3
1.5k
R3
1.5k
R8
10R8
10
R26
100
R26
100
R21
10R21
10
U4
NE
5534
U4
NE
5534
OU
T
+ -tr
imtr
im
V+ V-
F
igu
re C
.2. S
chem
atic
s fo
r au
xili
ary
mea
sure
men
t ci
rcu
it.
149
5 5
4 4
3 3
2 2
1 1
DD
CC
BB
AA
V-
V+
V_t
erm
To
AD
C C
h3
V+
V-
V+
Titl
e
Siz
eD
ocum
ent N
umbe
rR
ev
Dat
e:S
heet
of
<D
oc>
1
Aux
iliar
y S
igna
l Mea
surin
g C
ircui
ts
A
22
Sat
urda
y, A
ugus
t 21,
201
0
Titl
e
Siz
eD
ocum
ent N
umbe
rR
ev
Dat
e:S
heet
of
<D
oc>
1
Aux
iliar
y S
igna
l Mea
surin
g C
ircui
ts
A
22
Sat
urda
y, A
ugus
t 21,
201
0
Titl
e
Siz
eD
ocum
ent N
umbe
rR
ev
Dat
e:S
heet
of
<D
oc>
1
Aux
iliar
y S
igna
l Mea
surin
g C
ircui
ts
A
22
Sat
urda
y, A
ugus
t 21,
201
0
V_t
erm
Sen
se/B
uffe
r
V_te
rm V
olta
ge D
ivider
and
Buf
fer
+15
V
-15
V
Caps
to
supp
ort
Bus
Voltag
e
R29
27k
R29
27k
C26
330u
25 V
C26
330u
25 V
R31
100
R31
100
C27
10u
50 V
C27
10u
50 V
U5
NE
5534
U5
NE
5534
OU
T
+ -tr
imtr
im
V+ V-
R28
10R
2810
R27
10R27
10
C23
330u
25 V
C23
330u
25 V
C20
0.1u
C20
0.1u
R30
27k
R30
27k
VR
310
0kV
R3
100k
R32
100
R32
100
C19
0.1u
C19
0.1u
C24
10u
50 V
C24
10u
50 V
C22
330u
25 V
C22
330u
25 V
C21
180p
C21
180p
C25
330u
25 V
C25
330u
25 V
F
igu
re C
.2. (
Con
tinu
ed.)
150
REFERENCES
[1] J. M. Miller, “Energy storage system technology challenges facing strong hybrid, plug-in and battery electric vehicles,” in Proc. IEEE Vehicle Power and Propulsion Conference, 2009, pp. 4-10.
[2] P. T. Krein, “MathCad 6.0 file for hybrid vehicle traction system analysis,” April 2007. [Online]. Available: http://courses.ece.illinois.edu/ece431/FileLister/ Useful_Files/HEVALL407.htm
[3] R. A. Huggins, “Introductory material,” in Advanced Batteries: Material Science Aspects. New York, NY: Springer, 2009, pp. 1-21.
[4] R. A. Huggins, “Transient behavior of electrochemical systems,” in Advanced Batteries: Material Science Aspects. New York, NY: Springer, 2009, pp. 442-446.
[5] B. E. Conway, “Similarities and differences between supercapacitors and batteries for storing energy,” in Electrochemical Supercapacitors: Scientific Fundamentals and Technological Applications. New York, NY: Plenum Press, 1999, pp. 29-31.
[6] B. Destraz, P. Barrade, and A. Rufer, “Power assistance for diesel-electric locomotives with supercapacitive energy storage,” in Proc. IEEE Power Electronics Specialists Conference, 2004, pp. 677-682.
[7] B. T. Kuhn, G. E. Pitel, and P. T. Krein, “Electrical properties and equalization of lithium-ion cells in automotive applications,” in Proc. IEEE Vehicle Power and Propulsion Conference, 2005, pp. 55-59.
[8] M. Ortúzar, J. Moreno, and J. Dixon, “Ultracapacitor-based auxiliary energy system for an electric vehicle: Implementation and evaluation,” IEEE Trans. on Industrial Electronics, vol. 54, no. 4, pp. 2147-2156, Aug. 2007.
[9] N. Jinrui, W. Jhifu, and R. Qinglian, “Simulation and analysis of performance of a pure electric vehicle with a super-capacitor,” in Proc. IEEE Vehicle Power and Propulsion Conference, 2006, pp. 1-6.
[10] Panasonic CGR18650CG, Product datasheet, June 3, 2010. [Online]. Available: http://industrial.panasonic.com/www-data/pdf2/ACA4000/ACA4000CE234.pdf
[11] S. S. Choi and H. S. Lim, “Factors that affect cycle-life and possible degradation mechanisms of a Li-ion cell based on LiCoO2,” Journal of Power Sources, vol. 111, no. 1, pp. 130-136, Sept. 2002.
151
[12] P. G. Patil, “Developments in lithium-ion battery technology in the Peoples Republic of China,” Argonne National Laboratory, Argonne, IL, Rep. No. ANL/ESD/08-1, Jan. 2008.
[13] “Managing A123 cells with FMA cell balancing technologies,” FMA, Inc., Frederick, MD, Tech. Rep., March 2007. [Online]. Available: www.fmadirect.com/support_docs/item_1229.pdf
[14] A123 Systems ANR26650M1A, Product datasheet, June 2, 2010. [Online]. Available: http://www.a123systems.com/cms/product/pdf/1/_ANR26650M1A.pdf
[15] Maxwell Technologies K2 Series, Product datasheet, Jan. 23, 2010. [Online]. Available: http://www.maxwell.com/ultracapacitors/datasheets/DATASHEET_ K2_SERIES_1015370.pdf
[16] J. M. Miller and M. Everitt, “Ultra-capacitor augmentation of the vehicle electrical system to reset its power budget,” in Proc. IEEE Power Electronics in Transportation Conference, 2004, pp. 19-26.
[17] Maxwell Technologies BCAP0350, Product datasheet, June 20, 2010. [Online]. Available: http://www.maxwell.com/ultracapacitors/datasheets/DATASHEET_ DCell_energy_1014624.pdf
[18] K. Kawashima, T. Uchida, and Y. Hori, “Development of a novel ultracapacitor electric vehicle and methods to cope with voltage variation,” in Proc. IEEE Vehicle Power and Propulsion Conference, 2009, pp. 724-729.
[19] S. Lu, K. A. Corzine, and M. Ferdowsi, “A new battery/ultracapacitor energy storage system design and its motor drive integration for hybrid electric vehicles,” IEEE Trans. on Vehicular Technology, vol. 56, no 4, pp. 1516-1523, July 2007.
[20] L. Gao, R. A. Dougal, and S. Liu, “Active power sharing in hybrid battery/ capacitor power sources,” in Proc. IEEE Applied Power Electronics Conference, 2003, vol. 1, pp. 497-503.
[21] W. Lhomme, P. Delarue, P. Barrade, A. Bouscayrol, and A. Rufer, “Design and control of a supercapacitor storage system for traction applications,” in Conf. Record IEEE Industry Applications Conference, 2005, pp. 2013-2020.
[22] L. C. Rosario and P. C. K. Luk, “Implementation of a modular power and energy management structure for battery-ultracapacitor powered electric vehicles,” in Proc. IET (UK) Hybrid Vehicle Conference, 2006, pp. 141-156.
[23] P. Thounthong, S. Raël, and B. Davat, “Analysis of supercapacitor as second source based on fuel cell power generation,” IEEE Trans. on Energy Conversion, vol. 24, no. 1, pp. 247-255, March 2009.
152
[24] A. Drolia, P. Jose, and N. Mohan, “An approach to connect ultracapacitor to fuel cell powered electric vehicle and emulating fuel cell electrical characteristics using switched mode converter,” in Proc. Annual Conference of the IEEE Industrial Electronics Society, 2003, pp. 897-901.
[25] J. J. Awerbuch and C. R. Sullivan, “Control of ultracapacitor-battery hybrid power source for vehicular applications,” in IEEE Conference on Global Sustainable Energy Infrastructure: Energy2030, 2008, pp. 1-7.
[26] C. W. Deisch, “Simple switching control method changes power converter into a current source,” in Proc. IEEE Power Electronics Specialists Conference, 1978, pp. 300-306.
[27] “Voltage Regulator Module (VRM) and Enterprise Voltage Regulator-Down (EVRD) 11.1 Design Guidelines,” Intel Corp., Santa Clara, CA, Tech. Rep. 321736-002, Sept. 2009.
[28] X. Zhou, P.-L. Wong, P. Xu, F. C. Lee, and A. Q. Huang, “Investigation of candidate VRM topologies for future microprocessors,” IEEE Trans. on Power Electronics, vol. 15, no. 6, pp. 1172-1182, Nov. 2000.
[29] “VRM 8.5 DC-DC Converter Design Guidelines,” Intel Corp., Santa Clara, CA, Tech. Rep. 249659-002, March 2002.
[30] P. T. Krein, “Power semiconductors in converters,” in Elements of Power Electronics. New York, NY: Oxford University Press, 1998, pp. 451-473.
[31] A. Khaligh and P. Chapman, “Reduction of output capacitance in dc-dc converters using anticipated load transients,” in Proc. IEEE Applied Power Electronics Conference, 2008, pp. 818-823.
[32] S. Lim, J. Fan, and A. Huang, “Transient Voltage Clamp (TVC) circuit design based on constant load line impedance for Voltage Regulator Module (VRM),” IEEE Trans. on Industrial Electronics, to be published.
[33] J. T. Mossoba and P. T. Krein, “Modeling of unbalanced multiphase buck converters with applications to voltage regulator module control,” in Proc. IEEE Applied Power Electronics Conference, 2005, vol. 3, pp. 1424-1429.
[34] P. Midya, P. T. Krein, and M. F. Greuel, “Sensorless current mode control — an observer-based technique for DC-DC converters,” IEEE Trans. on Power Electronics, vol. 16, no. 4, pp. 522-526, July 2001.
[35] J. W. Kimball, J. T. Mossoba, and P. T. Krein, “A stabilizing, high-performance controller for input series-output parallel converters,” IEEE Trans. on Power Electronics, vol. 23, no. 3, pp. 1416-1427, May 2008.
153
[36] J. R. Huang, S. C.-H. Wang, C. J. Lee, E. K.-L. Tseng, and D. Chen, “Native AVP control method for constant output impedance of DC power converters,” in Proc. IEEE Power Electronics Specialists Conference, 2007, pp. 2023-2028.
[37] P. T. Krein, “Approximate methods for control design,” in Elements of Power Electronics. New York, NY: Oxford University Press, 1998, pp. 621-658.
[38] A. Emadi, S. S. Williamson, and A. Khaligh, “Power electronics intensive solutions for advanced electric, hybrid electric, and fuel cell vehicular power systems,” IEEE Trans. on Power Electronics, vol. 21, no. 3, pp. 567-577, May 2006.
[39] Powerstream H083448, Product datasheet, June 3, 2010. [Online]. Available: http://www.powerstream.com/p/H083448.pdf
[40] R. C. Kroeze, “Electrical battery model for use in dynamic electric vehicle simulations,” M.S. thesis, University of Illinois at Urbana-Champaign, 2008.
[41] P. T. Krein, “Overview of feedback control for converters,” in Elements of Power Electronics. New York, NY: Oxford University Press, 1998, pp. 576-616.
[42] Cornell Dubilier type DCMC Capacitor, Product datasheet, Jan. 25, 2010. [Online]. Available: http://www.cde.com/catalogs/DCMC.pdf
[43] K. J. Kelly, M. Mihalec, and M. Zolot, “Battery usage and thermal performance of the Toyota Prius and Honda Insight during chassis dynamometer testing,” in Proc. IEEE Battery Conference on Applications and Advances, 2002, pp. 247-252.
[44] J. Wei and F. C. Lee, “Two-stage voltage regulator for laptop computer CPUs and the corresponding advanced control schemes to improve light-load performance,” in Proc. IEEE Applied Power Electronics Conference, 2004, vol. 2, pp. 1294-1300.
[45] International Rectifier IRF6795MPbF, Product datasheet, July 19, 2010. [Online]. Available: http://www.irf.com/product-info/datasheets/data/irf6795mpbf.pdf
[46] International Rectifier IRF6711SPbF, Product datasheet, July 29, 2010. [Online]. Available: http://www.irf.com/product-info/datasheets/data/irf6711spbf.pdf
[47] P. T. Krein, “Dc-dc converters,” in Elements of Power Electronics. New York, NY: Oxford University Press, 1998, pp. 123-132.
[48] R. W. Erickson and D. Maksimović, “AC equivalent circuit modeling,” in Fundamentals of Power Electronics, 2nd ed. New York, NY: Kluwer Academic Publishers, 2001, pp. 188-255.
154
[49] R. W. Erickson and D. Maksimović, “Current programmed control,” in Fundamentals of Power Electronics, 2nd ed. New York, NY: Kluwer Academic Publishers, 2001, pp. 439-481.
[50] G. C. Verghese, C. A. Bruzos, and K. N. Mahabir, “Averaged and sampled-data models for current mode control: A re-examination,” in Proc. IEEE Power Electronics Specialists Conference, 1989, pp. 484-491.
[51] R. B. Ridley, “A new small-signal model for current-mode control,” Ph.D. dissertation, Virginia Polytechnic Institute and State University, Blacksburg, VA, 1990.
[52] J. J. Awerbuch and C. R. Sullivan, “Filter-based power splitting in ultracapacitor-battery hybrids for vehicular applications,” presented at the IEEE Workshop on Computers in Power Electronics, Boulder, CO, 2010.
[53] R. W. Erickson and D. Maksimović, “Converter transfer functions,” in Fundamentals of Power Electronics, 2nd ed. New York, NY: Kluwer Academic Publishers, 2001, pp. 293-302.
[54] D. L. Logue and P. T. Krein, “Simulation of electric machinery and power electronics interfacing using MATLAB/SIMULINK,” in Proc. Workshop on Computers in Power Electronics at COMPEL, 2000, pp. 34-39.
[55] “Matlab/Simulink electric machine and power electronics toolbox,” Simulink toolbox, Nov. 13, 2009. [Online]. Available: http://energy.ece.illinois.edu/ PEToolbox.zip
[56] P. T. Krein, “Geometric control,” in Elements of Power Electronics. New York, NY: Oxford University Press, 1998, pp. 663-676.
[57] H. K. Khalil, “Second-order systems,” in Nonlinear Systems, 3rd ed. Upper Saddle River, NJ: Prentice Hall, 2002, pp. 54-59.
[58] “Q8 Data Acquisition System user’s guide version 1.0,” Product manual, 2003. [Online]. Available: http://www.ece.clemson.edu/crb/ece495/Equipment/ Quanser_Q8_Manual.pdf
[59] M. J. Tavernini, B. A. Niemoeller, and P. T. Krein, “Real-time low-level simulation of hybrid vehicle systems for hardware-in-the-loop applications,” in Proc. IEEE Vehicle Power and Propulsion Conference, 2009, pp. 890-895.
[60] I. Buchmann, “Internal battery resistance” in Batteries in a Portable World, Electronic book, 2001. [Online]. Available: http://www.buchmann.ca/Chap9-page3.asp