Battery Management Systems

Battery Management Systems
Philips Research
VOLUME 1
Editor-in-Chief
Dr. Frank Toolenaar
Philips Research Laboratories, Eindhoven, The Netherlands
SCOPE TO THE 'PHILIPS RESEARCH BOOK SERIES'
As one of the largest private sector research establishments in the world, Philips
Research is shaping the future with technology inventions that meet peoples' needs and
desires in the digital age. While the ultimate user benefits of these inventions end up on
the high-street shelves, the often pioneering scientific and technological basis usually
remains less visible.
This 'Philips Research Book Series' has been set up as a way for Philips researchers to
contribute to the scientific community by publishing their comprehensive results and
theories in book form.
Ad Huijser
Design by Modelling
by
Henk. Jan Bergveld

Wanda S. Kruijt
Peter H.L. Notten
Philips Research Laboratories,

Eindhoven, The Netherlands
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-6108-9 ISBN 978-94-017-0843-2 (eBook)

DOI 10.1007/978-94-017-0843-2
Printed on acid-free paper
Explanation of cover:
The cover shows a transparent battery as an illustration of the use
of battery models for the design of Battery Management Systems
designed by Hennie Alblas
All Rights Reserved

© 2002 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2002
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming,
recording or otherwise, without written permission from the Publisher,
with the exception of any material supplied specifically for the purpose of
being entered and executed on a computer system, for exclusive use
by the purchaser of the work.
To Peggy, Bert and Pascalle
vii
Table of contents
List of abbreviations xiii
List of symbols XV
Series preface xxi
Preface xxiii
1. Introduction 1
1.1 The energy chain 1

1.2 Definition of a Battery Management System 3
1.3 Motivation of the research described in this book 4
1.4 Scope of this book 5
1.5 References 6
2. Battery Management Systems 9

2.1 A general Battery Management System 9
2.2 Battery Management System parts 10
2.2.1 The Power Module (PM) 10
2.2.2 The battery 14
2.2.3 The DC/DC converter 18
2.2.4 The load 19
2.2.5 The communication channel 19
2.3 Examples of Battery Management Systems 22
2.3.1 Introduction 22
2.3.2 Comparison of BMS in a low-end and
high-end shaver 22
2.3.3 Comparison of BMS in two types of cellular
phones 25
2.4 References 29
3. Basic information on batteries 31

3.1 Historical overview 31
3.2 Battery systems 33
3.2.1 Definitions 33
3.2.2 Battery design 35
3.2.3 Battery characteristics 36
Vlll
3.3 General operational mechanism of batteries 43

3.3.2 Basic thermodynamics 44
3.3.3 Kinetic and diffusion overpotentials 45
3.3.4 Double-layer capacitance 50
3.3.5 Battery voltage 52
3.4 References 52
4. Battery modelling 55
4.1 General approach to modelling batteries 55
4.1.1 Chemical and electrochemical potential 58
4.1.2 Modelling chemical and electrochemical reactions 59
4.1.3 Modelling mass transport 67
4.1.4 Modelling thermal behaviour 82
4.2 A simulation model of a rechargeable NiCd battery 86
4.2.2 The nickel reaction 89
4.2.3 The cadmium reactions 92
4.2.4 The oxygen reactions 97
4.2.5 Temperature dependence of the reactions 102
4.2.6 The model 103
4.3 A simulation model of a rechargeable Li-ion battery 107
4.3.2 The LiCo02 electrode reaction 108
4.3.3 The LiC6 electrode reaction 113
4.3.4 The electrolyte solution 117
4.3.5 Temperature dependence of the reactions 118
4.3.6 The model 118
4.4 Parameterization of the NiCd battery model 124
4.4.2 Mathematical parameter optimization 126
4.4.3 Results and discussion 131
4.4.4 Quality of the parameter set presented in section
4.4.3 under different charging conditions 138
4.4.5 Results obtained with a modified NiCd battery
model and discussion 144
4.5 Simulation examples 149
4.5.1 Simulations using the NiCd model presented in
section 4.2 149
4.5.2 Simulations using the Li-ion model presented in
section 4.3 155
4.6 Conclusions 162
4.7 References 165
ix
5. Battery charging algorithms 169

5.1 Charging algorithms for NiCd and NiMH batteries 169
5.1.1 Charging modes, end-of-charge triggers and
charger features 169
5.1.2 Differences between charging algorithms
for NiCd and NiMH batteries 175
5.1.3 Simulation example: an alternative charging
algorithm for NiCd batteries 177
5.2 Charging algorithm for Li-ion batteries 184
5.2.1 The basic principle 184
5.2.2 The influence of charge voltage on the
charging process 186
5.2.3 The influence of charge current on the
charging process 187
5.2.4 Simulation example: fast charging of a
Li-ion battery 188
5.3 Conclusions 191
5.4 References 192
6. Battery State-of-Charge indication 193

6.1 Possible State-of-Charge indication methods 193
6.1.1 Definitions 193
6.1.2 Direct measurements 195
6.1.3 Book-keeping systems 199
6.1.4 Adaptive systems 202
6.1.5 Some remarks on accuracy and reliability 203
6.2 Experimental tests using the bq2050 204
6.2.1 Operation of the bq2050 204
6.2.2 Set-up of the experiments 206
6.2.4 Conclusions of the experiments 211
6.3 Direct measurements for Li-ion batteries: the EMF method 212
6.3.2 EMF measurement methods 212
6.3.3 Measured and simulated EMF curves for the
CGR17500 Li-ion battery 214
6.3.4 Conclusions 219
6.4 A simple mathematical model for overpotential description 219
6.5 Proposed set-up for State-of-Charge system 225
6.5.1 The algorithm 225
6.5.2 Comparison with the bq2050 system 229
6.5.3 Comparison with systems found in the literature 230
X
6.6 Experimental tests with the system proposed in section 6.5 231
6.6.2 Set-up of the experiments 231
6.6.3 Experimental results 232
6.6.4 Discussion of the results 235
6.6.5 Conclusions of the experiments 237
6.7 Conclusions 238
6.8 References 239
7. Optimum supply strategies for Power Amplifiers

in cellular phones 241
7.1 Trends in cellular systems 241
7.2 The efficiency control concept 245
7 .2.1 Basic information on Power Amplifiers 246
7.2.2 Optimum supply voltage for optimum efficiency 250
7.3 DC/DC conversion principles 251
7.3 .1 Linear voltage regulators 25 2
7.3.2 Capacitive voltage converters 253
7.3.3 Inductive voltage converters 255
7.3.4 EMI problems involved in capacitive and
inductive voltage converters 258
7.3.5 Inductive voltage conversion for efficiency control 258
7.4 Simulation model derivation 258
7.4.1 DC/DC down-converter 258
7.4.2 Power Amplifier 260
7.5 Theoretical benefits of efficiency control 261
7.5.1 Simulation set-up 262
7.5.3 Conclusions 265
7.6 Experimental results obtained with a CDMA PA 266
7.6.1 Measurement set-up 266
7.6.2 Measurement results and discussion of part 1:
no DC/DC converter 267
7 .6.3 Measurement results and discussion of part 2:
with DC/DC converter 269
7 .6.4 Estimation of talk time increase in a
complete CDMA cellular phone 271
7.7 Application of efficiency control in a GSM cellular phone 274
7. 7.1 GSM power control protocol 274
7.7.2 Modifications in the Spark GSM phone 276
7.7.3 Measurement results and discussion 279
7. 7.4 Conclusions of the experiments 281
7.8 Conclusions 281
7.9 References 282
xi
8. General conclusions 285
About the authors 289
Index 291
xiii
List of abbreviations
ACPI Advanced Configuration and Power Interface
ACPR Adjacent Channel Power Ratio
ADC Analogue-to-Digital Converter
AM Amplitude Modulation
BER Bit Error Rate
BIOS Basic Input Output System
BMS Battery Management System
CAC Compensated Available Charge
cc Constant Current
CDMA Code-Division Multiple Access
Cd(OH)z Cadmium hydroxide
CHC Charging Control
cv Constant Voltage
DAC Digital-to-Analogue Converter
DCR Discharge Count Register
DCS Digital Cellular System
DMC Dimethyl carbonate
DQPSK Differential Quadrature Phase Shift Keying
DSP Digital Signal Processor
DTC DeskTop Charger
e Electron
EC Ethylene carbonate
ECC Energy Conversion Control
EDGE Enhanced Data rates for GSM Evolution
EM Electro-Magnetic
EMC Ethyl methyl carbonate
EMF Electro-Motive Force
EMI Electro-Magnetic Interference
ESR Equivalent Series Resistance
FDD Frequency Division Duplex
FDMA Frequency-Division Multiple Access
FM Frequency Modulation
FSK Frequency Shift Keying
OMSK Gaussian Minimum Shift Keying
GSM Global System for Mobile communication
H+ Proton
HzO Water
H 2 S04 Sulphuric acid
HVIC High-Voltage IC
ID Identification
IEC International Electrotechnical Commission
IIC Interface IC
KOH Potassium hydroxide
LED Light-Emitting Diode
LCD Liquid-Crystal Display
LR Linear Regulator
Li-ion Lithium-ion
LiCoOz Lithium cobalt oxide
xiv
LiMn20 4 Lithium manganese oxide

LiNi02 Lithium nickel oxide
LiPF6 Lithium hexafluorophosphate
LiC6 Lithium graphite
LMD Last Measured Discharge
LSE Least-Square Error
MSK Minimum Shift Keying
NAC Nominal Available Charge
NADC North American Digital Cellular
NiCd Nickel-cadmium
NiMH Nickel-metalhydride
Ni(OH)z/NiOOH Nickel hydroxide/nickel oxyhydroxide
NTC Negative-Temperature Coefficient resistor
Oz Oxygen
OQPSK Offset Quadrature Phase Shift Keying
Ox Oxidized species
PA Power Amplifier
Pb Lead
Pb0 2 Lead dioxide
PCB Printed-Circuit Board
PEO Polyethylene oxide
PFC Programmed Full Count
PM Power Module
PTC Positive-Temperature Coefficient resistor
PWM Pulse-Width Modulation
QAM Quadrature Amplitude Modulation
QPSK Quadrature Phase Shift Keying
Red Reduced species
RF Radio-Frequency
Rx Receive
SBC Smart Battery Charger
SBD Smart Battery Data
SBS Smart Battery System
SEI Solid Electrolyte Interface
SHE Standard Hydrogen reference Electrode
SLA Sealed lead-acid
SMBus System Management Bus
SMPS Switched-Mode Power Supply
SoC State-of-Charge
SoH State-of-Health
TCH Traffic Channel
TCM Timer-Control Module
TDD Time Division Duplex
TDMA Time-Division Multiple Access
Tx Transmit
UMTS Universal Mobile Telecommunication System
UPS Uninterruptible Power Supply
VRLA Valve-regulated lead-acid
Zn-MnOz Zinc-manganese dioxide
3G Third Generation
XV
List of symbols
Symbol Meaning Value Unit

A Electrode surface area m2
A Area of spatial element m2
Abat Battery surface area m2
AcJ'= Surface area of cadmium electrode m2
a; Activity of species i mollm3
a['f Activity of species i in the reference state mol/m3
a;b Bulk activity of species i mol!m3
a;s Surface activity of species i mol/m3
C; Concentration of species i mol/m3
C;b Bulk concentration of species i mol/m3
s
C; Surface concentration of species i mol/m3
Cch Chemical capacitance mot2/J
c,[ Electrical capacitance F
Crh Thermal capacitance J/K
cl Double-layer capacitance F
cdl
0 Double-layer capacitance per unit area F/m2
CH Helmholtz capacitance F
Co-c Gouy-Chapman capacitance F
Cpara Parasitic capacitance in DC/DC converter F
Cap rem Remaining battery capacity Ah
Cap max Maximum possible capacity that can be Ah
obtained from a battery
CF; Cost function for output variable i
D; Diffusion coefficient of species i m2/s
d; Diffusion layer thickness of species i m
DJ Anti-parallel diodes that model Butler-
Volmer relation for reaction j
dV!dtum Change of battery voltage in time, used V/s
as parameter in proposed SoC indication
system
E Error in 'battery empty' prediction of an %
SoC indication system
E; Potential of electrode i v
Eeq
I Equilibrium potential of electrode i v
gqbat Equilibrium potential of battery, or EMF v
gq* Apparent equilibrium potential v
EoI Standard redox potential of electrode i v
E"bat Standard redox potential of battery v
E"par Activation energy of parameter par J/mol
Ech Chemical energy J
Eel Electrical energy J
Erh ''Thermal energy" JK
Emax Maximum energy stored in capacitor or J
coil
xvi

Eq Energy term, normalized to current, used J/A
in simple overpotential description of
(Eq. 6.4)
F Faraday's constant 96485 C/mol
fc Channel frequency Hz
jRF RF frequency Hz
!switch Switching frequency in DC/DC converter Hz
Jt Relation between measured battery %
parameter and SoC in direct-measurement
SoC system
Function that translates coulomb counter %
contents into SoC based on the basis of V,
T and I measurements in a book-keeping
system
Exchange current for reaction j A
Anodic current A
Cathodic current A
Double-layer current A
Current level that determines state in A
proposed SoC indiction system
Supply current A
Diffusion flux of species i mol/(m2 .s)
Chemical flow molls
Heat flow w
Reaction rate constant for oxidation Unit
(anodic) reactionj depends
on
reaction
Reaction rate constant for reduction Unit
(cathodic) reaction j depends
on
reaction
Backward reaction rate constant Unit
depends
on
reaction
Forward reaction rate constant Unit
depends
on
reaction
Oxygen solubility constant mol/
(m3 .Pa)
Thickness of nickel electrode, grain size m
Thickness of positive electrode, grain size m
Thickness of negative electrode, grain m
size
Thickness of electrolyte in Li-ion model m
Number of electrons in reaction
xvii

m Number of spatial elements in the
electrolyte in Li-ion model
M Number of parameter sets in optimization
process
m; Molar amount of species i mol
mref Molar amount in the reference state mol
m 0cd Molar amount of cadmium nuclei at t=t0 mol
Mcd Molecular weight of cadmium kg/mol
n Number of electrons in reaction
n Number of capacitors in capacitive
voltage converter
N Number of nuclei
N Number of points taken into account in
optimization process
p Number of spatial elements in a system
p Pressure Pa
Pout Output power of PA in cellular phone dBm
Psup Supply power for PA w
Pch Chemical power w
Pet Electrical power w
Prh ''Thermal power" WK
par(T) Temperature-dependent parameter Unit
depends
on
parameter
pal' Pre-exponential factor for parameter par Unit
depends
on
parameter
Q Charge c
Qrh Heat J
Qed, Max Maximum capacity of cadmium electrode c
QNi,Max Maximum capacity of nickel electrode c
Qcd(OH)2,Ni Overdischarge reserve of Cd(OH)z at the c
nickel electrode
Overdischarge reserve of cadmium at the c
cadmium electrode
gnaxUCo02 Maximum capacity of LiCo02 electrode c
gnaxLiC6 Maximum capacity of LiC6 electrode c
r(t) Radius of hemispherical particles m
R Gas constant 8.314 J/(rnol.K)
Rch Chemical resistance Js/mol2
R, Electrolyte resistance Q
Rteak Resistance that models self-discharge in Q
Li-ion model
Rbypass On-resistance of bypass switch in DC/DC
converter
ESR of coil in DC/DC converter
xviii

Rloss Ohmic-loss resistance in DC/DC Q
converter
Rswitch On-resistance of switch in DC/DC

converter
Electrical resistance
Transformed antenna impedance seen at
collector/drain of final PA stage
Rth Thermal resistance KIW
Rs Series resistance of switch S Q
R.fl.k Ohmic and kinetic resistance Q
Rd Diffusion resistance Q
RdCd Time constant diffusion overpotential s
RkCk Time constant kinetic overpotential s
S; Switch i in DC/DC converter
SoCE First SoC in equilibrium state, just after %
re-entry from transitional state
SaCs First SoC when discharge state is entered %
SoC1 Last SoC in transitional state just before %
equilibrium state is entered
Time s
trem Remaining time of use s
tl,actual Experimental remaining time of use from s
the moment a discharge current is applied
until the battery is empty
T Switching period in DC/DC converter s
T Temperature K
Tomb Ambient temperature K
Toper Operating time s
Tperiod Period time of burst drawn by P A model s
Upos,i Interaction energy coefficient for LiCo02
electrode in phase i
Uneg,i Interaction energy coefficient for LiC 6
electrode in phase i
Inverse step function 1 for I~O
and 0 for
1>0
v Voltage v
vbat Battery voltage v
vcon Control voltage v
VEoD End-of-Discharge voltage v
Verror Error voltage in behavourial P A model Verror=O or v
Verror=I
Vnom Nominal supply voltage for P A v
Vsup Supply voltage v
Vsup,opt Optimum supply voltage, as applied in v
efficiency control
Vsup,min Minimum PA supply voltage at which v
linearity specification can still be met
xix
Symbol Meaning Value

Vswitch Switch drive voltage in DC/DC converter
V(t) Volume of deposited material at time t
V' Initial volume of hemispherical particles
at time to
Free gas volume inside battery
Var; Output variable in optimization process Depends

on Vari
Normalizing and weighing factor for Depends
output variable V ari on Vari
X Distance m
wj, xj, yj, Zj Reaction order of species in reaction j 11
X; Mol fraction of species i
X' 0 'phasetransition Mol fraction of Lt ions at which phase
transition occurs in positive electrode
~· 8phasetransition Mol fraction of Li+ ions at which phase
transition occurs in negative electrode
Z; Valence of ionic species i
a Transfer coefficient O<a<l
£l1h Heat transfer coefficient
y Activity coefficient 11
Ys Fugacity coefficient
8 Diffusion layer thickness

8 Duty cycle in DC/DC converter 0<8<1
8H Helmholtz layer thickness m
8PA Duty cycle of PA =ON/OFF ratio 0<8pA<1
.::1G" Gibbs free energy change under standard J/mol
conditions
.::1G"' Gibbs free energy change under standard J/mol
conditions, taking constant activity terms
for e.g. OH. or H 20 species into account
.::1G0 Change in (Gibbs) free energy in an J/mol
oxidation reaction
L1GR Change in (Gibbs) free energy in a J/mol
reduction reaction
!::.G0 Change in free electrochemical energy in J/mol
an oxidation reaction
t::.GR Change in free electrochemical energy in J/rnol
a reduction reaction
&! Change in a reaction's enthalpy J/rnol
LIS Change in a reaction's entropy J/(mol.K)
L1x Thickness of a spatial element m
Eo Permittivity in free space 8.85.10- 12 C2/(N.m2 )
e, Dielectric constant
rjJ Electrostatic or Galvani potential v
r/J, Electrostatic electrode potential v
¢1 Electrostatic electrode potential v
XX

¢J Electrostatic electrolyte potential v
f/Jox Electrostatic potential of Ox ions in v
electrolyte
Conduction period for switch in s

capacitive voltage converter
Conduction angle for P A rad
Chemical potential of species i J/mol
Electrochemical potential of species i J/mol
Standard chemical potential of species i J/mol
Gravimetric density of cadmium kg/m3
Stoichiometric factors for species i in a
reaction equation
Time constant s
Diffusion time constant s
Time constant associated with increase in s
overpotential in an almost empty battery
RF angular frequency radls
Overpotential of charge transfer reaction v
Diffusion overpotential v
Kinetic overpotential v
Ohmic overpotential v
Combined ohmic and kinetic v
overpotential
Increase in overpotential when the battery v
becomes empty
T}DCflJC DC/DC converter efficiency %
1Jmax Maximum attainable P A efficiency %
1Jmax, theory Maximum theoretical PA efficiency %
1JPA PA efficiency %
Oex Extended surface area ni
(pos,i Constant for LiCo~ electrode in phase i
(neg,i Constant for LiC6 electrode in phase i
Note I: Value of I has been assumed in this book

xxi
Series preface
As one of the largest private sector research establishments in the world, Philips
Research is shaping the future with technology inventions that meet peoples' needs
and desires in the digital age. While the ultimate user benefits of these inventions
end up on the high-street shelves, the often pioneering scientific and technological
basis usually remains less visible.
This 'Philips Research Book Series' has been set up as a way for our researchers to
contribute to the scientific community by publishing their comprehensive results and
theories in a convenient book form.
Dr Ad Huijser
Chief Executive Officer of Philips Research
xxiii
Preface
This book covers the interesting and multidisciplinary subject of Battery

Management Systems (BMS), in particular the design of BMS with the aid of
simulation models. The basic tasks of BMS are to ensure optimum use of the energy
stored in the battery (pack) that powers a portable device and to prevent damage
inflicted on the battery (pack). This becomes increasingly important due to the larger
power consumption associated with added features to portable devices on the one
hand and the demand for longer run times on the other hand. In addition to
explaining the general principles of BMS tasks such as charging algorithms and
State-of-Charge (SoC) indication methods, the book also covers real-life examples
of BMS functionality of practical portable devices such as shavers and cellular
phones.
Simulations offer the advantage over measurements that less time is needed to
gain knowledge of a battery's behaviour in interaction with other parts in a portable
device under a wide variety of conditions. This knowledge can be used to improve
the design of a BMS, even before a prototype of the portable device has been built.
Although the advantages are obvious, the application of simulation to the design of
BMS is relatively new, mainly due to the absence of useable battery models so far.
Therefore, a large part of this book is devoted to the construction of such simulation
models for rechargeable batteries. With the aid of several illustrations it is shown
that design improvements can indeed be realized with the battery models
constructed in this book. The battery models can be seen as 'transparent batteries' in
which the course of all reactions and the origin of certain externally visible battery
behaviour can be traced. This is also illustrated in the cover illustration of this book.
Apart from the battery models, models of other parts of a portable device are also
used. A battery's electrical and thermal environments can be modelled in this way.
After the formal definition of a BMS in chapter 1, chapter 2 provides more
detailed information on BMS. It is shown that the complexity and sophistication of a
BMS depend not only on the portable device in which it is included, but certainly
also on the battery type. The characteristics of the various parts that constitute a
BMS are explained and examples of BMS in different types of portable products are
given. The emphasis in these examples is on illustrating the influence of several
factors on the BMS complexity.
Chapter 3 starts with a brief historical overview that illustrates the development
of batteries from 250 BC until the present day. Commonly encountered battery-
related terms are defined and the main characteristics of the most common battery
types are listed. The operational mechanism of batteries is described in simple terms,
offering people without an (electro)chemical background some essential information
on the main processes that are responsible for battery behaviour. The several
contributions to the overall battery voltage are also defined.
Chapter 4 forms the core of this book and describes the subject of battery
modelling. This chapter should be read by those that would like to obtain detailed
knowledge on the occurrence of chemical and physical processes inside batteries and
how these processes can be modelled. A new method of battery modelling is
introduced that is based on the concept of physical system dynamics. The essence of
this concept is the analogy between energy storage and energy dissipation
definitions in various physical domains, such as the electrical domain, the chemical
domain and the thermal domain. As an advantage of this method, these domains are
xxiv
coupled enabling the transfer of energy from one domain to another. As a result, the
battery voltage, internal gas pressure and temperature can be simulated coherently
under a wide variety of conditions. The general principles of modelling are
described first in a step by step approach. The models are built up in the form of
network models, making them more suitable for use in an electronic environment
with electronic-circuit simulation tools such as SPICE. The general principles of
modelling are then successively applied to a NiCd battery and aLi-ion battery. For
both battery types the mathematical description of all the relevant processes and
reactions that take place inside that particular battery type are given. Chapter 4 also
covers the subject of finding the correct parameter values for the models. This is
done by means of a new mathematical method for the NiCd model. The simulation
examples at the end of the chapter illustrate the use of the constructed battery
models as 'transparent batteries'.
Chapter 5, 6 and 7 give more detailed information on the most important BMS
tasks. Simulation results obtained with the models described in chapter 4 serve as an
example of the use of these models to improve the design of BMS. Chapter 5
discusses the charging algorithms for NiCd, NiMH and Li-ion batteries encountered
in practice. Moreover, charging simulations are performed with both the NiCd and
Li-ion model discussed in chapter 4. A new charging algorithm based on a
maximum allowed battery temperature under the name thermostatic charging is
introduced based on simulation results. Measurement results are presented to
confirm the advantages revealed by the simulations. Possible causes for a decrease
in cycle life of Li-ion batteries when they are charged faster with increased current
and voltage values are investigated by means of simulations.
The subject of State-of-Charge (SoC) indication is discussed in chapter 6.
Three possible methods are identified and described. Experiments performed with a
commercially available SoC indication system are described and the shortcomings
of this system are identified. A new set-up for an SoC system for Li-ion batteries is
proposed in an attempt to overcome these shortcomings. The method is based on a
combination of the methods mentioned at the beginning of the chapter. The method
was developed using the battery models described in chapter 4. The system gives an
estimation of the remaining time of use during the valid conditions during discharge
of the battery. The chapter ends with a discussion of experimental results obtained
with the proposed system A comparison with the results obtained with the
commercially available system shows that many more experiments will have to be
carried out to further investigate the new system. Several suggestions for
improvements are given.
Chapter 7 compares several ways of powering a Power Amplifies (PA) in a
cellular phone with respect to efficiency. On the basis of trends in the cellular
telephony market it is shown that reducing the power consumption of the P A in talk
mode is becoming increasingly important. This can be achieved by increasing the
PA efficiency, which is defined as the ratio of the transmitted power and the power
drawn at the supply pin. The concept of efficiency control is introduced, which
implies that the PAis always powered with the minimum needed supply voltage. It
is shown that inductive DC/DC converters are most suitable in terms of efficiency to
generate this minimum supply voltage from the battery voltage. The benefits of
efficiency control are illustrated by means of simulations using the battery models of
chapter 4 and by means of measurements of practical PAs.
The various application examples of battery models discussed in this book
illustrate that these models are indeed beneficial for the design of BMS. The authors
would like to encourage the co-operation of people with different backgrounds such
XXV
as electrochemistry and electrical engineering to improve the design of BMS. As

illustrated in this book, a lot can be gained from this co-operation. This book looks
at the exiting field of battery management from both viewpoints. Therefore, the
authors hope to have laid a foundation for many more successful research and
development projects dedicated to getting the most out of our batteries. The future is
wireless!
Acknowledgements
The authors would like to thank the management of Philips Research Laboratories
for the opportunity to publish their research results in this book. Paul Regtien, Frans
Schoofs, Peter Blanken and Hans Feil are acknowledged for their substantial
contributions to the content of this book. The authors are also indebted to Rene van
Beek for performing many battery-related measurements described in this book. The
following persons are acknowledged for various contributions: Bram van den Akker,
Andre van Bezooijen, Jan Reinder de Boer, Ralf Burdenski, Vincent Cachard, Kok-
how Chong, Rene du Cloux, Valere Delong, Robert Einerhand, Wouter Groeneveld,
Saskia Hanneman, Anton Hoogstraate, Henk Jan Pranger, Siep Onneweer, William
Rey, Jurgen Rusch and Wouter de Windt. The authors would like to thank their
colleagues at Matsushita, Japan, for the co-operation during a joint battery project.
Domine Leenaerts, Gerard van der Weide and the Philips Translation Service are
acknowledged for proofreading of the original manuscript. Hennie Alblas is
acknowledged for drawing all figures in this book. Finally, the authors would like to
thank their families for the continuous support and understanding.
HENK JAN BERGVELD

WANDAKRUUT
PETER NOTTEN
Eindhoven, April 2002

Chapterl
Introduction
1.1 The energy chain

The demand for portable electronic consumer products is rapidly increasing.
Examples of fast-growing markets of portable products are notebook computers,
cellular and cordless phones and camcorders. Figure 1.1 shows the expected world-
wide shipment of cellular handsets until the year 2004 [1], with the total number of
handsets sold increasing dramatically. At present, almost half of the shipments of
cellular handsets are replacement sales, which means that people who already have a
cellular phone buy them. By 2004, almost all cellular handsets sold will be
replacement sales.
900
Figure 1.1: Expected world-wide shipments of cellular handsets in millions per type of sales
[I ]
In view of battery costs and environmental considerations, the batteries used to

power these portable devices should preferably be rechargeable. The demand for
rechargeable batteries is increasing in line with that for portable products. Table 1.1
shows the sales of the most important rechargeable battery types in Japan in 1998
[2]. The majority of the rechargeable batteries are manufactured in Japan. Table 1.1
shows that the number of batteries sold increased dramatically in 1998 relative to
1997. More information on rechargeable battery types will be given in chapter 3.
The energy yielded by a portable device in the form of, for example, sound or
motion, ultimately derives from the electrical energy supplied by the mains. The
conversion of energy from the mains to the eventual load inside a portable product
can be described as an energy chain [3], which is shown in Figure 1.2. The links of
the energy chain are a charger, a battery, a DC/DC converter and a load.
H. J. Bergveld et al., Battery Management Systems

© Springer Science+Business Media Dordrecht 2002
2 Chapter I
Table 1.1: Sales of the most important rechargeable battery types in Japan in millions in 1998
[2)
Battery type Sales in 1998 (1M pieces) Growth with respect to 1997
NiCd 588 85%
NiMH 640 112%
Li-ion 275 141%
t tmonitor and control

t t
Figure 1.2: The energy chain symbolizing the energy transfers from mains to load in a
portable product
Electrical energy from the mains is fed to the battery through the charger during
charging. The charger uses electromagnetic components like a transformer or an
inductor. Here, electrical energy from the mains is first transformed into magnetic
energy and then back into electrical energy. The battery stores the electrical energy
in the form of chemical energy. During discharge of the battery, chemical energy is
converted back into electrical energy.
The DC/DC converter is an optional link. It is not to be found in every portable
product. There are two reasons for its presence. First of all, the battery might deliver
a voltage which is not suitable for operation of the load. Secondly, each circuit part
of the load should be operated from the lowest possible supply voltage for efficiency
reasons, because a surplus in supply voltage is often dissipated in the form of heat.
In both cases, the DC/DC converter powers the load with the lowest possible supply
voltage, irrespective of the battery voltage. The DC/DC converter uses an inductor,
which translates the electrical energy from the battery into magnetic energy and
back into electrical energy again. In the load, the electrical energy from the DC/DC
converter is converted into sound, light, Electro-Magnetic (EM) radiation or
mechanical energy.
In order to make optimum use of the energy inside the battery, all conversions
of energy in the energy chain should be well understood and made as efficient as
possible. This is especially true when miniaturization of the portable product is
desired. Miniaturization is an important trend for many portable devices. Portable
phones, for example, are becoming smaller and smaller. When the volume of a
portable device decreases, the amount of dissipated power must also decrease. The
reason is that a smaller volume will yield a higher temperature when the amount of
dissipated power remains the same. There is also a trend towards increasing
complexity and functionality of portable products. An example is e-mail and internet
facilities added to the functionality of cellular phones. Adding complexity to the
total system adds to the total power consumption of the load. So assuring efficient
energy conversions becomes even more important. This can be achieved by
monitoring and controlling all the links in the energy chain. This is schematically
shown in Figure 1.2.
Introduction 3
There are algorithms that check and control the links in the energy chain. A first
example is a charging algorithm, which keeps track of the battery status and controls
the charger by interrupting the charging current when the battery is full. Charging
should not continue once the battery is full, because otherwise the battery
temperature will rise substantially and/or the battery might be damaged. This
decreases its capacity and usable number of cycles. Therefore, a proper charging
algorithm leads to a more efficient use of the battery and its energy.
A second example is an algorithm that determines the battery's State-of-
Charge (SoC). This information can be used to make more efficient use of the
battery energy. For example, it can be used as input for charge control, indicating
that the battery is full. Also, it is more likely that the user will wait longer before
recharging the battery when an accurate and reliable SoC indication is available on a
portable device. Less frequent recharging is beneficial for the cycle life of the
battery.
A third example is an algorithm that controls a DC/DC converter to power the
load with the minimum required supply voltage, dependent on the activity of the
load. An example of such a load is a Power Amplifier (PA) in a cellular phone. In
the case of a PA, the supply voltage may be lower for lower output power. This
leads to better efficiency. This example will be elaborated in chapter 7.
1.2 Definition of a Battery Management System

Three terms apply to the implementation of monitor and control functions in the
energy chain. These terms are battery management, power management and energy
management. As a rough indication, battery management involves implementing
functions that ensure optimum use of the battery in a portable device. Examples of
such functions are proper charging handling and protecting the battery from misuse.
Power management involves the implementation of functions that ensure a proper
distribution of power through the system and minimum power consumption by each
system part. Examples are active hardware and software design changes for
minimizing power consumption, such as reducing clock rates in digital system parts
and powering down system parts that are not in use. Energy management involves
implementing functions that ensure that energy conversions in a system are made as
efficient as possible. It also involves handling the storage of energy in a system. An
example is applying zero-voltage and zero-current switching to reduce switching
losses in a Switched-Mode Power Supply (SMPS). This increases the efficiency of
energy transfer from the mains to the battery.
It should be noted that the implementation of a certain function may involve
more than one of the three management terms simultaneously. This book will focus
on battery management and its inclusion in a system. A definition of the basic task
of a Battery Management System can be given as follows:
The basic task of a Battery Management System (BMS) is to ensure that

optimum use is made of the energy inside the battery powering the portable
product and that the risk of damage inflicted upon the battery is minimized.
This is achieved by monitoring and controlling the battery's charging and
discharging process.
Keeping in mind the examples of algorithms given in section 1.1, this basic task of a
BMS can be achieved by performing the following functions:
4 ~~1
• Control charging of the battery, with practically no overcharging, to ensure a

long cycle life of the battery.
• Monitor the discharge of the battery to prevent damage inflicted on the battery
by interrupting the discharge current when the battery is empty.
• Keep track of the battery's SoC and use the determined value to control
charging and discharging of the battery and signal the value to the user of the
portable device.
• Power the load with a minimum supply voltage, irrespective of the battery
voltage, using DC/DC conversion to achieve a longer run time of the portable
device.
1.3 Motivation of the research described in this book

As described in section 1.1, more efficient use of the energy inside a battery is
becoming increasingly important in the rapidly growing market for portable
products. Manufacturers of portable devices are consequently paying even more
attention to battery management. This is reflected in many commercial electronics
magazines, such as Electronic Design and EDN, containing examples of
implementations of battery management functions in a system Many examples are
to be found of ICs that implement certain charging algorithms [4]-[7]. Adding
intelligence to batteries in portable products to enable e.g. SoC monitoring is also
receiving a great deal of attention. The term 'smart battery' is a general buzzword
that pops up in many articles [5],[8]-[10]. However, no explanation of battery
behaviour is given in this kind of magazines. Therefore, the reason why one battery
management IC performs better than another is often not understood. Moreover, it is
hard to determine how the functionality of a BMS can be improved.
Besides in the magazines mentioned above, a lot of information on battery
management can also be found in the literature. For example, much attention is paid
to finding ways of accurately determining a battery's SoC [11]-[13]. The battery
management functions described in these articles are derived from extensive battery
measurements, for example measurements of battery discharge curves under various
conditions. Most of these measurements are very time-consuming. The conclusions
are moreover often empirical.
In practice, battery management functions are implemented in portable devices
by electrical engineers. These engineers usually treat the battery as a black box. It is
usually assumed that the battery is a voltage source with some series resistance.
However, in order to improve the functionality of a BMS, at least some
understanding of battery behaviour in the system is needed. A prototype of (part of)
the portable device is needed for measurements of actual battery charge and
discharge behaviour. On the other hand, simulation is a helpful tool in obtaining a
better understanding of the behaviour of complex systems under a wide variety of
conditions. Simulations take less time than measurements and no prototype is
needed. Therefore, the availability of simulation models for batteries would be very
helpful for the development of BMS.
Simulation models of batteries are given a lot of attention in the literature.
However, some models have been constructed by researchers with an
electrochemical background and are very complex and based on many mathematical
equations [14],[15]. Consequently, such models are usually not very suitable for
electrical engineers who have to simulate a battery in a system. Other models have
Introduction 5
been constructed by electrical engineers [16]. These models can be readily used in
system simulations, but do not lead to a better understanding of battery behaviour.
Therefore, simulations with these models do usually not lead to better views on
battery management.
This book presents the results of research into battery modelling obtained by
combining the expertise of electrical engineering with that of electrochemistry. The
result is a method for modelling batteries that can be readily applied to all kinds of
batteries [17]-[24]. The models result from translating (electro)chemical theory into
equivalent-network models using the principle of physical system dynamics [25].
This enables the use of conventional electronic-circuit simulators that electrical
engineers are accustomed to. In addition to the battery voltage and current, the
internal gas pressure and battery temperature can also be simulated integrally and
coherently under a wide variety of conditions. Apart from the electrical interaction
with the battery's environment, it is also possible to simulate the thermal interaction.
The modelling method even allows the simulation of effects such as venting under
high-pressure build-up inside the battery.
Obviously, close quantitative agreement between the results of simulations
using a battery model and measurements of battery behaviour is important. Part of
the research described in this book is devoted to optimizing this quantitative
agreement. The results of this research will be described in chapter 4.
As is revealed by the title of this book, the derived battery models will be
applied in the design of Battery Management Systems. The battery models can be
viewed as 'transparent', in which the course of the various reactions can be
investigated. For example, the charging efficiency can be easily plotted. Charging
efficiencies of different charging algorithms can be compared for optimization. An
example of an optimized charging algorithm that was found in simulations with a
battery model will be described in chapter 5 [24].
Internal battery behaviour that is normally hard to measure can be visualized
with the models. For example, overpotentials of all reactions inside a battery and
electrode equilibrium potentials can be easily plotted during the battery's operation.
This makes the model a very useful tool in the quest for more accurate SoC
indication algorithms. The model can be used to gain insight into the combined
action of the various processes taking place inside a battery. Based on this insight,
compact descriptions of battery behaviour can be derived. The results of research
efforts in this field will be described in chapter 6.
Sometimes a designer is not interested in what goes on inside a battery, for
example when simulating the run time of a portable device. In such cases the
designer is merely interested in the battery's discharge behaviour under various load
conditions. Part of the research described in this book is aimed at finding an
optimum method for powering a PA inside a cellular phone. As will be shown in
chapter 7, the battery models described in this book offer a simple way of comparing
the run times of various PA supply strategies in a cellular phone. In addition to
battery models, use will be made of simulation models of a DC/DC converter and a
PA to design the BMS.
1.4 Scope of this book

Chapters 2 and 3 provide general information required as a background in the
remaining chapters of this book. Chapter 2 describes the various parts of a BMS and
their functionality in more detail. Some examples of BMS in different portable
products are given to clarify the influence of several factors on the complexity.
6 Chapter 1
Chapter 3 deals with the central part of a BMS, which is the battery itself. Some
general information is given on the construction, types, operational mechanism and
behaviour of batteries.
The research results are described in chapters 4 to 7. Battery modelling forms
the core of this book. The construction of simulation models for rechargeable
batteries is described in chapter 4. Those who want to have a thorough
understanding of the background and construction of the models should read this
chapter. The adopted modelling approach is explained in detail and the model
equations for both a NiCd and a Li-ion battery model are derived. Further, the
efforts to improve the quantitative agreement between the results of simulations and
measurements are thoroughly discussed for the NiCd model. Chapters 5, 6 and 7
deal with the design of BMS. These chapters describe the use of the battery models
of chapter 4 and other models to find improved BMS schemes. One does not have to
read chapter 4 first in order to understand the content of these chapters.
Battery charging algorithms are discussed in chapter 5. It is shown that battery
models can readily be used in the development of new, more efficient charging
algorithms. Chapter 6 deals with the determination of a battery's SoC. Several
possible methods are compared. In addition, a new SoC indication system is
proposed and tested, based on simulations using a battery model and knowledge
obtained in battery measurements. An optimum method for efficiently powering
PAs in cellular phones is described in chapter 7. This strategy is named efficiency
control [26]. As a DC/DC converter is necessary to implement efficiency control,
some basic information on voltage conversion techniques is given. Measurement
results are discussed to define the benefits of efficiency control in practice and to
make a comparison with simulation results. Conclusions are drawn in chapter 8 and
recommendations are made for further research in the exciting field of BMS in
general and battery modelling in particular.
1.5 References
[1] From slide of Market Research Team Cellular, Philips Components, June 2000
[2] P.H.L. Notten, "Batterij Modellering & Batterij Management", EET project proposal,
March 2000
[3] The energy chain concept originates from Frans Schoofs
[4] R. Cates, R. Richey, "Charge NiCd and NiMH Batteries Properly", Electronic Design,
pp. 118-122, June 10, 1996
[5] L.J. Curran, "Charger ICs Reflect Shift to Smart Batteries", EDN, pp. S.lO-S.ll, July 3,
1997
[6] R. Schweber, "Supervisory ICs Empower Batteries to Take Charge", EDN, pp. 61-72,
September 1, 1997
[7] G. Cummings, D. Brotto, J. Goodhart, "Charge Batteries Safely in 15 Minutes by
Detecting Voltage Inflection Points", EDN, pp. 89-94, September I, 1994
[8] A. Watson Swager, "Smart-Battery Technology: Power Management's Missing Link",
EDN, pp. 47-64, March 2, 1995
[9] D. Freeman, "Integrated Pack Management Addresses Smart-Battery Architecture",
Begleittexte zum Design & Elektronik Entwicklerforum, Batterien, Ladekonzepte &
Stromversorgungsdesign, pp. 82-86, Miinchen, March 31, 1998
[10] C.H. Small, "ICs Put the Smarts in Smart Batteries", Computer Design, pp. 35-40,
August 1997
[11] J. Alzieu, H. Smimite, C. Glaize, "Improvement of Intelligent Battery Controller: State-
of-Charge Indicator and Associated Functions", J. Power Sources, vol. 67, no. 1&2, pp.
157-161, July/August 1997
Introduction 7
[12] J.H. Aylor, A. Thieme, B.W. Johnson, "A Battery State-of-Charge Indicator for Electric
Wheelchairs", IEEE Trans. on industrial electronics, vol. 39, no. 5, pp. 398-409,
October 1992
[13] 0. Gerard, J.N. Patillon, F. d' Alche-Buc, "Neural Network Adaptive Modelling of
Battery Discharge", Lecture Notes in Computer Science, vol. 1327, pp. 1095-1100,
1997
[14] D. Fan, R.E. White, "A Mathematical Model of a Sealed Nickel-Cadmium Battery", J.
Electrochem. Soc., vol. 138, no. 1, pp. 17-25, January 1991
[15] T.F. Fuller, M. Doyle, J. Newman, "Simulation and Optimization of the Dual Lithium
Insertion Cell", J. Electrochem. Soc., vol. 141, no. 1, pp. 1-10, January 1994
[16] S.C. Hageman, "Simple Pspice Models Let You Simulate Common Battery Types",
EDN, pp. 117-132, October 28, 1993
[17] W.S. Kruijt, H.J. Bergveld, P.H.L. Notten, Electronic-Network Modelling of
Rechargeable NiCd batteries, Nat.Lab. Technical Note 211/96, Philips Internal Report,
1996
[18] P.H.L. Notten, W.S. Kruijt, H.J. Bergveld, "Electronic-Network Modelling of
Rechargeable NiCd Batteries", Ext. Abstract 96, I9Ist ECS Meeting, Montreal, Canada,
1997
[19] W.S. Kruijt, E. de Beer, P.H.L. Notten, H.J. Bergveld, "Electronic-Network Model of a
Rechargeable Li-ion Battery", The Electrochemical Society Meeting Abstracts, vol. 97-
1, Abstract 104, p. 114, Paris, France, August31-September 5, 1997
[20] W.S. Kruijt, H.J. Bergveld, P.H.L. Notten, Principles of Electronic-Network Modelling
of Chemical Systems, Nat.Lab. Technical Note 261198, Philips Internal Report, October
1998
[21] W.S. Kruijt, H.J. Bergveld, P.H.L. Notten, Electronic-Network Model of a
Rechargeable Li-ion Battery, Nat.Lab. Technical Note 265/98, Philips Internal Report,
October 1998
[22] W.S. Kruijt, H.J. Bergveld, P.H.L. Notten, "Electronic-Network Modelling of
Rechargeable Batteries: Part I: The Nickel and Cadmium Electrodes", J. Electrochem.
Soc., vol. 145, no. 11, pp. 3764-3773, November 1998
Rechargeable Batteries: Part II: The NiCd System", J. Electrochem. Soc., vol. 145, no.
11, pp. 3774-3783, November 1998
[24] H.J. Bergveld, W.S. Kruijt, P.H.L. Notten, "Electronic-Network Modelling of
Rechargeable NiCd Cells and -its Application to the Design of Battery Management
Systems", J. Power Sources, vol. 77, no. 2, pp. 143-158, February 1999
[25] R.C. Rosenberg, D.C. Karnopp, Introduction to Physical System Dynamics, McGraw-
Hill Series in Mechanical Engineering, New York, 1983
[26] The name efficiency control originates from Ralf Burdenski
9
Chapter 2
This chapter gives general information on Battery Management Systems (BMS)

required as a background in later chapters. Section 2.1 stans with the factors that
determine the complexity of a BMS and shows a general block diagram. The
function of each pan in a BMS is discussed in more detail in section 2.2 and
examples of adding BMS intelligence are given. The BMS aspects of two types of
ponable devices are discussed in section 2.3. This serves to illustrate the theory
presented in sections 2.1 and 2.2.
2.1 A general Battery Management System

The concept of the energy chain was explained in chapter 1. Essentially, the links in
this energy chain already reflect the basic parts of a BMS. In more general terms, the
charger can be called a Power Module (PM). This PM is capable of charging the
battery, but can also power the load directly. A general BMS consists of a PM, a
battery, a DC/DC converter and a load.
The intelligence in the BMS is included in monitor and control functions. As
described in chapter 1, the monitor functions involve the measurement of, for
example, battery voltage, charger status or load activity. The control functions act on
the charging and discharging of the battery on the basis of these measured variables.
Implementation of these monitor and control functions should ensure optimum use
of the battery and should prevent the risk of any damage being inflicted on the
battery.
The degree of sophistication of the BMS will depend on the functionality of the
monitor and control functions. In general, the higher this functionality, the better
care will be taken of the battery and the longer its life will be. The functionality
depends on several aspects:
• The cost of the ponable product: In general, the additional cost of a BMS
should be kept low relative to the cost of the portable product. Hence, the
functionality of the monitor and control functions of a relatively cheap product
will generally be relatively low. As a consequence, the BMS will be relatively
simple. An example of the difference in BMS between a cheap and an
expensive shaver will be given in section 2.3.
• The features of the portable product: This is closely related to the product's
cost. A high-end product will have more features than a low-end product. For
example, a high-end shaver with a 'Minutes Left' indication needs more BMS
intelligence than a low-end shaver with no signalling at all.
• The type of battery: Some types of batteries need more care than others. An
example of the influence on the complexity of the BMS when moving from one
battery technology to a combination of two battery technologies will be given in
section 2.3.
• The type of ponable product: In some products, a battery will be charged and
discharged more often than in others. For example, a cellular phone might be
charged every day, whereas a shaver is charged only once every two or three

10 Chapter 2
weeks. The number of times a battery can be charged and discharged before it
wears out, together with the average time between subsequent charge cycles,
determines the lifetime of a battery in a device. So it is more important for this
number to be high in a cellular phone than in a shaver. This can be achieved by
making the BMS more intelligent. Therefore, amore sophisticated BMS is more
important in a cellular phone than in a shaver.
The intelligence needed for the BMS can be divided between the various parts. This
partitioning of intelligence is an important aspect in designing a BMS. The main
determining parameter in this respect is cost. Dedicated battery management ICs can
implement intelligence. Useful background information with many examples can be
found in [1]-[15]. Measured variable and parameter values and control commands
are communicated between the parts of the BMS via a communication channel. This
channel can be anything from a single wire that controls a Pulse-Width Modulation
(PWM) switch to a bus that is controlled by a dedicated protocol [16]-[18].
The structure of a general BMS is shown in Figure 2.1 . The partitioning of
intelligence is symbolized by placing a 'Monitor and Control' block in every system
part. The BMS shown in Figure 2.1 also controls a Battery Status Display. An
example is a single Light-Emitting Diode (LED) that indicates the 'battery low '
status. It can also be a string of LEDs indicating the battery's State-of-Charge (SoC)
or a Liquid-Crystal Display (LCD) that indicates the battery status, including the
SoC and the battery condition.
Communication channel
,_
Monitor
and
1'"""'190 Control
~ ~
r - - - - - - + - - - - - - - - - , r - - - i DC/DC
converter
Figure 2.1: A g eneral Battery Management System (BMS)
2.2 Battery Management System parts
2.2.1 The Power Module (PM)

The basic task of the PM is to charge the battery by converting electrical energy
from the mains into electrical energy suitable for use in the battery. An alternative
for the mains might be other energy sources, such as a car battery or solar cells. In
many cases, the PM can also be used to power the portable device directly, for
example when the battery is low. The PM can either be a separate device, such as a
Battery Management Systems 11
travel charger, or be integrated within the portable device, as in for example shavers.
Especially in the latter case, the efficiency of the energy conversion process has to
be sufficiently high, because the poorer the efficiency, the higher the internal
temperature of the portable device and hence that of the battery will be. Long
periods at elevated temperatures will decrease the battery capacity [19].
The monitor and control functions for the PM can be divided into two types.
First of all, the energy conversion process itself has to be controlled and, secondly,
the battery's charging process has to be controlled. The Energy Conversion Control
(ECC) involves measuring the output voltage and/or current of the PM and
controlling them to a desired value. This desired value is determined by the
CHarging Control (CHC) on the basis of measurement of battery variables such as
voltage and temperature. Moreover, the current flowing into the battery can be input
for the CHC function.
The ECC function is situated within the PM itself, when the PM is a separate
device. The CHC function can now be divided between the PM, the portable device
and the battery, assuming that this battery can be removed from the portable device.
The partitioning of the CHC function will depend on cost, but also on the employed
charging algorithm.
A removable battery can be charged separately on a so-called DeskTop
Charger (DTC). In addition to the standard battery that comes with a product, some
users also buy a spare battery. Cellular phone users who make many phone calls
during the day sometimes charge the standard battery and the spare battery. It is
often possible to simultaneously connect a standard battery in a portable product and
a separate spare battery to a DTC. The priority of charging must then be fixed in the
system. When the CHC function is incorporated inside the portable device, it has to
be present inside the DTC as well. The reason is that the user of the portable product
should be able to charge only the spare battery in the DTC. The CHC function inside
the portable product can then not be used.
Three simple examples of different ways of partitioning the CHC function are
given below. A separate PM and a detachable battery are assumed. In the first
example, the CHC function is incorporated in the PM. It is incorporated in the
portable product in the second example and it is incorporated with the battery in the
third example. A dashed box in the examples denotes the DTC indicating that it is
optional. For simplicity, the intelligence needed in the DTC to determine charging
priority is ignored.
Partitioning of the CHCfunction: example I

The first example is illustrated in Figure 2.2. A Negative-Temperature Coefficient
resistor (NTC) has been attached to the battery. This enables the measurement of the
battery temperature by contacting it through extra terminals on the battery pack. The
term battery pack will be explained in more detail in section 2.2.2. The battery
voltage can be easily measured, because the battery remains connected to the PM
during charging. Depending on the charging algorithm, the current that flows into
the battery will sometimes have to be measured. This can be done by measuring the
voltage across a low-ohmic resistor in series with the battery.
Based on the measured battery variables, the charging process is controlled
inside the PM using a suitable charging algorithm. The addition of a simple
identification means (ID in Figure 2.2) to the battery pack is desirable in case
batteries of different types and/or chemistries can be used with the portable product.
An example is the addition of a resistor to every battery pack, with one connection
to the battery minus terminal and another to an extra terminal on the battery pack.
12 Chapter 2
Each pack can then be identified by a certain resistor value, determined by the
manufacturer and programmed in, for example, the PM. Changing some parameter
values in the charging algorithm will often suffice in case batteries of the same
chemistry, but different capacities, can be used with the portable device. Different
charging algorithms often have to be used in case batteries of different chemistries
can be used. As all charge monitor and control functions are present inside the PM,
the DTC will only act as a mechanical and electrical connection between the PM and
the battery when it is used. Hence, no intelligence has to be added to it.
----I
Portable
MAINS Product
---_I
Figure 2.2: Incorporation of the charge monitor and control functions (CHC) inside the
power module (PM)
An advantage of this set-up is that no electronics have to be integrated with the

battery pack and no extra intelligence has to be added to the portable product.
However, the additional complexity of dealing with different types of batteries now
has to be added to the PM. A dedicated charge controller IC or a microprocessor can
accomplish the CHC function. A microprocessor offers the designer more flexibility
with respect to changes in the charging algorithm, for example when newer battery
chemistries will be used in future versions of the portable device.
Partitioning of the CHC function: example 2

The charge monitor and control functions have been incorporated inside the portable
product in the second example. This situation is depicted in Figure 2.3 . As in the
first example, only an NTC and a simple identification means (ID) have been added
to the battery pack. The latter is only needed in case different types of batteries
and/or chemistries may be used with the same portable product. The measurements
of the battery voltage, temperature and possibly current are processed in the portable
product and a control signal is sent to the PM.
The set-up of Figure 2.3 is especially advantageous when a microcontroller or
microprocessor is present inside the portable device. This is for example the case in
cellular phones and notebook computers. The charging algorithm can be easily and
flexibly programmed in this controller or processor. Moreover, the presence of a
micro-controller implies the possibility of programming an SoC algorithm based on
the measured battery parameters.
MAINS
portable product
A disadvantage of the set-up of Figure 2.3 is the presence of the CHC inside the
DTC. As discussed above, this DTC has to be able to charge a spare battery on its
own, without the presence of the portable device. Therefore, the same CHC as
programmed inside the portable product has to be added to the DTC, because no
charging intelligence is present inside the battery pack nor the PM.
Partitioning of the CHC function: example 3

The charge monitor and control functions have been added to the battery pack in the
third example. This situation is depicted in Figure 2.4.
· ·-·-·-·,
j DTC j
PM
Portable
MAINS Product
----I
battery pack
The measured battery voltage, temperature and/or current are processed inside the
battery pack in Figure 2.4 . The battery pack controls the PM autonomously through
a control wire. This means that it can even be charged directly by the PM, without
the need to connect it to the portable device or DTC. An advantage is that each
14 Chapter2
battery pack can now include its own charging algorithm. This obviates the need to
identify the battery to determine which charging algorithm should be used.
A disadvantage of the configuration shown in Figure 2.4 is that a considerable
amount of hardware has to be added to the battery pack, some of which is already
present in the portable product itself. A good example is the micro-controller
mentioned above. Also, the user has to pay for the same hardware each time he/she
buys such a battery when the CHC function is added to each battery pack.
In practice, the examples shown in Figure 2.2 and Figure 2.3 are encountered
most. Combinations of the configurations shown in these figures are also
encountered. Some practical examples will be given in section 2.3.
2.2.2 The battery

A battery's basic task is to store energy obtained from the mains or some other
external power source and to release it to the load when needed. This enables a
portable device to operate without a connection to any power source other than a
battery. Different battery systems with different chemistries and different
characteristics exist. Examples of some commonly encountered battery systems are
nickel-cadmium (NiCd), nickel-metalhydride (NiMH) and lithium-ion (Li-ion)
batteries. The characteristics of the various battery systems vary considerably, even
for batteries with the same chemistry, but, for example, a different design or
different additives. More information will be given in chapter 3.
The term battery pack is often used for detachable batteries. Depending on the
desired voltage and capacity of the battery, several batteries can be connected in
series and/or in parallel inside a battery pack. To avoid confusion, the basic battery
building blocks inside the battery pack are often referred to as cells. The series
connection of cells yields a higher total battery voltage at the same capacity in [Ah]
and the parallel connection yields a higher total battery capacity in [Ah] at the same
battery voltage. Apart from one or more cells, a battery pack may also contain other
components. Examples are the NTC and ID resistors mentioned in the previous
section. Moreover, some electronics might be present inside the pack. All is
included inside a plastic container with at least the plus and minus terminals of the
battery and possibly some other terminals for temperature measurement and/or
identification, as discussed in the previous section. In this book, the term battery will
be used for both bare batteries and battery packs.
It is important from a battery-management point of view to realize at this
moment that the various battery systems should be approached differently, for
example via different charging algorithms. Approaching different battery systems
differently is important in determining the electronics that has to be integrated with
the battery.
Electronic safety switch for Li-ion batteries

In the case of Li-ion batteries, an electronic safety switch has to be integrated with
the battery. The battery voltage, current and temperature have to be monitored and
the safety switch has to be controlled to ensure that the battery is never operated in
an unsafe region. The reason for this is that battery suppliers are particularly
concerned with safety issues due to liability risks. A voltage range, a maximum cur-
rent and a maximum temperature determine the region within which it is considered
safe to use a battery. The battery manufacturer determines these limits. Outside the
safe region, destructive processes may start to take place. Generally speaking, in the
higher voltage range these processes may eventually lead to a fire or an explosion,
whereas in the lower voltage range, they lead to irreversible capacity loss of the
battery.
The maximum voltage is dictated by two factors, the maximum battery
capacity and its cycle life. The cycle life denotes the number of cycles the battery
can be charged and discharged before it is considered to be at the end of its life. A
battery's life ends when the capacity drops below a certain level, usually 80% of its
nominal capacity. This is illustrated in Figure 2.5, which shows the maximum
battery capacity and the cycle life as a function of the voltage applied to the battery
during charging. The figure illustrates that the higher this voltage, the higher the
maximum battery capacity and the lower the cycle life will be. Around 4.1 to 4.2 V,
a 100 mV increase in battery voltage yields a 12% capacity increase, but a sharp
decrease in cycle life of 200 cycles.
• •
I
0 0
Cycle life I Capacity
150 600
i
Q)
::J 140 500 ~
i
0
iii CIS
> a.
CIS
iii 0
.E1:: 130 400 iii
0
1:: .E
1::
0
0 1::
300 0
Q)
120
.s
C)
;:.!!
0
1:: 0
\___
Q) 00
~ ~
200 .E
Q)
a. 110 Ul
CIS Q)
Ul 0
CIS >.
0
~ 100
0
CIS
a.
100
·=~
CIS
u 0
Q)
90 0 >.
()
4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7
Charging voltage - - - .
Figure 2.5: Maximum battery capacity and cycle life as a function of battery voltage for aLi-
ion battery (source: Panasonic)
The choice of the maximum voltage level for the electronic safety switch is a
compromise. The demands on the accuracy of this level are in the order of one
percent, because a certain maximum battery capacity and a certain minimum cycle
life have to be guaranteed. Obviously, the battery voltage should not reach the
maximum voltage level of the electronic safety switch during charging to prevent
the risk of current interruption during normal use. The demands on the accuracy of
the minimum voltage are less strict, as the voltage drops rather sharply when the
battery is almost empty. This means that a deviation in the voltage at which the
16 Chapter 2
battery is disconnected from the load by the safety switch does not influence the
usable battery capacity significantly. The minimum voltage needed to operate the
portable device should be chosen higher than the minimum voltage level of the
safety switch. This is to prevent the risk of the safety switch opening during normal
operation.
The maximum current specification resembles the specification of a fuse.
Below the maximum current level, which value is linked to the maximum capacity
of the battery, the safety switch should always conduct. Above this current level,
currents are allowed to occur for only a certain amount of time. How long that time
will be will depend on the thermal resistance of the battery to its environment and
the thermal capacity of the battery. The higher the current, the shorter the time it will
be allowed to occur to prevent the risk of the battery temperature becoming too high.
This means that the delay time between detection of the current and the opening of
the switch is smaller at higher current levels. A general set-up of an electronic safety
switch is shown in Figure 2.6.
Li-ion
+
battery
pack
v
SafetY
Switch T
Control
Control
Figure 2.6: General set-up of an electronic safety switch inside aLi-ion battery pack
The battery voltage and the voltage difference across the switch are measured. The
latter voltage difference can be used to measure the battery current. This voltage
difference also allows start-up of the IC when a charger is connected and the battery
voltage is lower than the lowest limit of the safety voltage range, which implies that
the switch is open. In that case, there will be a large voltage difference across the
opened switch depending on the maximum output voltage of the charger. This
voltage difference will allow the operation of some sort of start-up circuit inside the
safety-switch control IC. The temperature is usually measured inside the control IC,
but could also be measured using an NTC.
The maximum resistance of the electronic switch has to be of the same order of
magnitude as other series resistances in the current path. These other series
resistances include the internal battery impedance, the resistance of other passive
safety devices and the contact resistance between the battery and the load. An
example of a passive safety device is a Positive-Temperature Coefficient resistor

(PfC). This device gradually inhibits the current flow when the temperature
increases. In practice, values in the order of 100 mQ are found for the electronic
switch, for example an MOSFET. The cost aspect determines whether the switch is
realized in an IC process other than the control circuitry, or together with the control
circuitry on one IC. As the electronic switch is integrated with the battery, it is
important that the associated circuitry draws a negligible current from the battery. In
practice, values down to several f.!A are found.
Charge-balancing and pack supervisory electronics inside a battery pack

A different need for electronics arises when several cells are connected in series
inside a battery pack. This is for example often the case in notebook computers.
Because of differences between the individual cells inside the battery pack, a certain
imbalance in SoC may exist. For example, the cells might differ in the maximum
amount of charge that can be stored, or in their internal impedance. These
differences already occur due to manufacturing variance and will even increase
because of aging. Such aging effects may be worse for some cells than for others in
a battery pack. A reason could be that some cells become substantially hotter than
others. The occurrence of large temperature differences between the cells in a
battery pack is largely influenced by the position of the cells relative to the
temperature sensor. The weakest cell always determines the characteristics of the
battery pack.
Several examples of battery-pack supervisory ICs and charge-balancing ICs
can be found in the literature [20]-[22]. The essence of these ICs is to extend a
battery pack's useful lifetime by preventing major imbalances between the cells
inside the pack. A battery-pack supervisory IC measures each cell voltage
individually and interrupts the charge or discharge current when one of the voltages
exceeds or drops below a certain level. A charge-balancing circuit distributes the
charge applied to or drawn from the battery pack equally between the cells by
transferring charge from a cell with a high SoC to a cell with a lower SoC.
Alternatively, the cell with the highest SoC can be bypassed by a 'bleeder' resistor,
until the other cell(s) reach(es) the same voltage. Although both battery-pack
supervisory and charge-balancing circuits prevent major differences between cells,
the differences that arise from manufacturing spread can never be eliminated.
Other functions that can be integrated within a battery pack

Especially when electronic circuitry, such as a safety device or a charge-balancing
circuit, is already present in a battery pack, it is relatively cheap to add extra
functionality. The addition of some intelligence to a battery itself is advantageous,
because each type of battery has to be approached in a specific way. This is
especially true for portable products that can be used with different battery systems.
This eventually leads to a so-called smart battery, which can be defined as a
rechargeable battery with a microchip that collects and communicates present,
calculated and predicted battery information to the host system under software
control [6]. Features that are included in a smart battery are self-monitoring
functions, charge-regime control, communication with the system's host,
implementation of fault identification and protection and storage of information in a
memory. This includes information on the battery characteristics, present status and
the history of use. Hence, some of the information stored in the battery pack will be
fixed, whereas other information will be continuously updated.
18 Chapter 2
The fact that information is integrated with the battery makes it necessary to
standardize the format of the information to allow the use of batteries of
manufacturer A in a portable device of manufacturer B. Moreover, standardization
facilitates system design as it makes it possible to treat a battery as a standardized
system part, regardless of its voltage, chemistry and packaging. An example of this
will be presented in section 2.2.5, which deals with the communication channel.
2.2.3 The DC/DC converter

The basic task of a DC/DC converter in a portable product is to connect a battery to
the various system parts when the battery voltage does not match the voltage
needed. The battery voltage may be either too low or too high. In the first case,
DC/DC up-conversion is of course needed. In the latter case, DC/DC down-
conversion is needed when the battery voltage is higher than the maximum allowed
supply voltage of the load. Apart from this however, from the viewpoint of
efficiency it is always a good idea to convert the battery voltage into the minimum
supply voltage V min needed by the load for the following reason.
First of all, when a DC/DC converter is used, the design of the load can be
optimized for the minimum supply voltage instead of the whole voltage range of the
battery discharge curve. This will lead to a higher efficiency of the load in most
cases. Secondly, operating a system part with a higher supply voltage than necessary
implies a waste of energy. This can be inferred from Figure 2.7, which shows a
schematic representation of a typical discharge curve of a battery. Note that the area
under this curve represents V·/-t, which indeed expresses the energy obtained from
the battery in [J] . The minimum voltage Vmin needed to operate the load is also
a--.
Figure 2.7: Schematic representation of a typical battery discharge curve denoting energy
loss (shaded area) when the battery voltage is higher than the minimum supply voltage V min of
the load
The shaded area above Vmin expresses the energy that is lost in heat dissipation when
the load is directly connected to the battery due to the fact that the battery voltage
Vbar is higher than Vmin · Assuming that the battery voltage can be efficiently
converted into Vmim the energy loss can be considerably reduced. More information
on principles for converting battery voltages into other voltages will be given in
chapter?.
2.2.4 The load

The basic task of the load is to convert the electrical energy supplied by a battery
into an energy form that will fulfil the load's function. Many different types of
portable products exist, and hence many different types of loads. Some general
remarks on the connection of the battery to the load and the implementation of
battery management functions will be given below.
Different supply voltages

Different supply voltages are needed in most portable products. These voltages are
constantly changing as the technology progresses. The maximum possible supply
voltage before an IC breaks down is decreasing, because the feature size of ICs is
becoming smaller and smaller. However, the supply voltage needed for some
analogue functions will remain higher than the supply voltage needed for digital
system parts, for example on account of a need for a specific signal-to-noise ratio of
the analogue function. Most IC processes offer the possibility of designing analogue
circuits with a higher supply voltage than digital circuits. This can be achieved by
using special transistors with a thick gate oxide.
Physical aspects also influence the IC process in which the analogue system
part is realized and the supply voltage needed. Examples of such aspects are the
need for higher voltages for backlighting LCD screens in notebook computers or the
need for negative bias voltages for sorr..~ types of Power Amplifiers (PAs) in cellular
phones. In general, the supply voltages for the analogue system parts encountered in
portable devices show a larger variation in values than those of the digital system
parts. This means that for a given battery voltage a system designer has to provide
this variety of voltages from a single battery voltage using (a combination of) up-
conversion, down-conversion or up/down-conversion [23].
Power consumption of system parts

The power consumption of the various system parts of a portable device may vary
considerably. Take for example the PA in a cellular Global System for Mobile
communication (GSM) phone, which amplifies the Radio-Frequency (RF) signal to
the antenna. The P Atransmits in bursts in such phones. Currents of up to 2 A can be
drawn from the supply pin, depending on the efficiency and supply voltage of the
PA. This of course leads to different demands on the design of the power supply
circuitry around the PA than in the case of a system part that draws a current in the
rnA-range.
Use of existing components for BMS functions

The presence of a microcontroller or microprocessor in a portable product will ena-
ble the flexible implementation of many of the monitor and control functions needed
in a BMS. An example was shown in Figure 2.3.
2.2.5 The communication channel

The basic task of a communication channel in a BMS is to transfer monitor and
control signals from one BMS part to another. The complexity of a communication
channel in a BMS depends on the complexity of the system and the partitioning of
the intelligence needed for battery management. For example, the only
20 Chapter2
communication that takes place between the battery and the charger shown in Figure
2.2 involves the voltage across an NTC and the value of the identification resistor.
However, in the case of a smart battery, the batteries can communicate their status to
the system host and other system parts, such as the charger. The appearance of smart
batteries on the market resulted in a major trend towards standardization of the
information stored in batteries and the way it is communicated to other system parts.
Battery manufacturers teamed up with semiconductor and computer hardware
manufacturers to find a standard architecture for a communication bus between the
battery and the system host. The ultimate goal of standardization was to facilitate the
development of universal battery management solutions. In these solutions, each
battery, regardless of its type or chemistry, communicates information on its status
to other system parts in a defined way.
One of several ways of arriving at a standardized interface between a smart
battery and the rest of the system was suggested by Duracell in co-operation with
Intel in 1994 [5]-[7],[16]-[18]. This co-operation has resulted in the Smart Battery
System (SBS) specification. The SBS ownership is in the hands of several
companies, including the battery manufacturers Duracell, Energizer, Toshiba and
Varta and the semiconductor manufacturers Intel, Benchmarq, Linear Technology,
Maxim, Mitsubishi and National Semiconductor. The SBS specification is generic,
which means that it is independent of battery chemistry, voltage and packaging. The
specification consists of four parts:
• The System Management Bus (SMBus) specification.

• The Smart Battery Data (SBD) specification.
• The Smart Battery Charger (SBC) specification.
• The specification of the interface between the SMBus and the Basic Input
Output System (BIOS) in a portable computer.
Although the specification focuses specifically on the portable computer industry, it

was designed to be applicable to other portable products too, according to Duracell
and Intel. A block diagram illustrating the implementation of the specification is
given in Figure 2.8.
SMBus interface
I I I
Smart Charger
System Host and/or Smart Battery
other SMB Devices
Figure 2.8: Basic block diagram illustrating the implementation of the SBS specification
SMBus specifzcation
The SMBus uses a two-wire communication channel between the components and
an electrical layer according to the Philips I2C protocol with additional software.
Each SMBus component can be either a master or a slave, but there can only be one
master at a time. For example, the smart battery can signal a warning via the SMBus
when the battery is approaching low capacity, or a disk drive can ask the smart
battery whether there is enough energy left to save data.
SBD specification
The SBD specification defines the data that flows across the SMBus between the
smart battery, the SMBus host, the Smart Battery Charger and other devices. In
addition to software definitions, error detection, signalling and smart battery data
protocols, it includes a data set of 34 values of battery information. This includes
voltage, current, temperature and device type and identification data as well as
calculated and stored values, such as:
Run Time To Empty: the predicted remaining run time at the

present rate of discharge
Remaining Capacity: in units of either charge or energy
Relative State-Of-Charge: predicted remaining capacity as a
percentage of full charge capacity
Full-Charge Capacity: predicted pack capacity when fully
charged
Cycle Count: number of charge/discharge cycles of the
battery
SBC specification
The SBC specification defines three levels of chargers. A Ievel-l charger can only
interpret critical warning messages and cannot adjust its output. For example, it will
only interrupt its charge current when a short-circuit occurs. However, it cannot
adapt its charge current to the employed battery. This is done by a level-2 charger,
which is an SMBus slave device. It is able to dynamically adjust its output according
to the charging algorithm stored in the smart battery pack. This means that the
battery 'tells' the charger how it has to be charged. An example of such a set-up was
shown in Figure 2.4. A level-3 charger is an SMBus master device, which chooses a
stored charging algorithm by interrogating the battery. Hence, in the case of a level-
3 charger, the charger 'decides' by itself how the battery will be charged.
SMBus!BIOS interface specification

This part of the specification defines how the operating system and applications that
run on a portable computer can communicate with SMBus components, using the
BIOS layer as an intermediate. Software vendors Phoenix and SysternSoft have
developed software that supplies intelligent battery data to a computer user. This
enables the user to choose his/her own power management scheme interactively. For
example, the SysternSoft Smart Battery Software System and the Phoenix Smart
Battery Manager offer the computer users the possibility of choosing between better
economy and better performance by using a scroll bar. The software shows the
battery performance as a function of the power management scheme chosen by the
user. Moreover, the user may install his/her own critical warning messages.
Another example of a specification that defines the connection between
software and SMBus devices is the Advanced Configuration and Power Interface
(ACPI). It was developed by Intel, Microsoft and Toshiba and enables system
manufacturers to supply power management that is directed by the operating system.
22 Chapter2
Examples are the automatic turning on and off of peripherals, such as hard-disk
drives and printers, and the activation of the computer itself by peripherals.
Acceptance of the SBS specification by portable product manufacturers

The SBS specification offers flexibility with respect to the amount of intelligence
that is added to a battery. For example, a system designer is free to use only a
portion of the battery information specified in the SBD specification. In fact, the
system designer can design his/her own tailor-made smart battery system. One can
choose from a variety of ICs usable with the SMBus. The original vision of Duracell
was to manufacture standard smart batteries with fixed form factors and with all
intelligence, fulfilling the SBS specification, inside. This turned out to be too
limiting with respect to design freedom for system designers. A system designer
prefers to use custom-designed batteries and add intelligence in the form of specific
ICs. As a result, Duracell stepped out of the smart battery business for notebook
computers in 1997. However, the idea of using the SMBus and a specified interface
between the different building blocks has survived.
Although the SBS specification is claimed to be applicable to any portable
product, it is only used in notebook computers in practice. A microprocessor is
already at hand in this case and the inclusion of a communication protocol in the
BMS does not add too much to the cost of the portable product. The added cost of a
complicated bus and bus protocol might be too high in the case of other portable
devices. Dallas Semiconductor offers a cheaper alternative solution to smart
batteries that fulfil the SBS specification [17]. They have developed a range of
temperature measurement and identification ICs intended for integration in a battery
pack. Information is in this case communicated from the battery to the host
processor via a proprietary one-wire interface. This enables the design of a relatively
cheap Battery Management System
2.3 Examples of Battery Management Systems
2.3.1 Introduction
Examples of the BMS implemented in two types of portable products, a shaver and a
cellular phone, will be given in this section. The examples have been derived from
existing Philips products. In both cases, a distinction will be made between the BMS
in two types of the product. A distinction will be made between a low-end and a
high-end shaver. This will illustrate the influence of cost and features of the product
on the complexity and functionality of the BMS. The impact on the BMS of moving
from the NiMH battery technology to a combined use of NiMH and Li-ion batteries
will be described for the cellular phone. This will illustrate the impact of the type of
battery on the complexity of the BMS.
2.3.2 Comparison ofBMS in a low-end and high-end shaver

Figure 2.9 shows the BMS in a low-end shaver with strong emphasis on the cost of
the product. The shaver is powered by only one NiCd or NiMH battery, which has
been integrated inside the shaver. The PM has also been integrated in the shaver.
The battery has been connected to the motor via an ON/OFF switch.
Both the ECC and the CHC control functions have been included in a High-
Voltage IC (HVIC). This HVIC regulates the charge current by controlling a high-
voltage switch in a Switched-Mode Power Supply (SMPS) in a flyback
configuration [24] . The control signals used by the HVIC are the output voltage of
the PM, the voltage across resistor R1 and the voltage across the rectifying diode D.
The value of the output current depends on the mode of the HVIC. During charging,
the output current is high until 'battery full ' detection. The output current is
switched to a low value after this detection. A voltage regulation mode exists in
addition to these two charge modes. This mode occurs when the motor is switched
on during charging. In this case, the PM maintains a constant battery voltage and
hence a constant supply voltage for the shaver motor.
Low-end shaver
I------------------------------~
I PM
I
I ON / OFF
Mains I Mains
Filter
Motor
Figure 2.9: Battery Management System inside a low-end shaver
A certain amount of power needs to be transferred to the motor to enable shaving.

When the user is shaving with the shaver connected to the mains, the output current
of the PM will be relatively high. The reason for this is that the battery voltage is
relatively low, as only one NiCd or NiMH battery is used. Current measurement by
means of a series resistor is not attractive, because the high current value would lead
to a too high temperature in the shaver given the shaver dimensions. Therefore, the
value of the output current has to be determined in other measurements. This is
achieved by measuring the time for which current flows to the battery through diode
D, when the high-voltage switch on the primary side of the transformer is open. This
time is obtained by measuring the voltage across diode D [24]. Moreover, the peak
current at the start of the current flow through diode D is obtained by measuring the
peak voltage across R 1 at the mains side of the transformer at the moment when the
high-voltage switch is opened. As the current through diode D will ramp down
linearly to zero, this peak current combined with the period of time for which
current flows and the total switching period time T can be used to determine the
average current.
One way of determining the 'battery full' status of a NiCd or NiMH battery is
to measure the temperature rise of the battery at the end of charging. The difference
between the battery temperature and the ambient temperature is measured by two
NTCs in Figure 2.9. When the temperature difference between the battery and the
environment reaches a certain value that is included in the HVIC, the PM will
switch to a low current. This current will maintain the charge level of the battery by
24 Chapter 2
compensating self-discharge. Such a low current is commonly referred to as a

trickle-charge current.
The BMS for the low-end shaver has proven to be cost-effective. The simple
charging algorithm allows for enough charging cycles until the battery capacity
drops below an unacceptable level. However, the signalling of battery information to
the user is not implemented at all. This is an unacceptable situation for high-end
shavers. Therefore, a microcontroller is included in high-end shavers to provide all
sorts of signalling of information to the user through an LCD screen. As discussed in
section 2.1, the increased functionality will influence the complexity of the BMS.
The BMS of a high-end shaver is shown in Figure 2.10. The shaver is powered by
two NiCd or NiMH batteries in series. A Timer-Control Module (TCM) has been
included. Again, the batteries are connected to the motor via an ON/OFF switch.
:TcM
_____________ _shaver
High-end
ON / OFF
Motor
Figure 2.10: Battery Management System inside a high-end shaver
The ECC function has been included in the HVIC in the PM portion as in the
previous example. The HVIC controls the output current of the PM to the
appropriate value during charging, as communicated by the TCM module. The
HVIC switches to output voltage regulation when the motor is switched on during
charging. The current supplied by the PM in this mode is lower than in the similar
situation in the low-end shaver, which makes current measurement with resistor R 2
attractive.
The CHC function has been included in the TCM. The microcontroller
communicates to the rest of the system through the Interface IC (IIC). The IIC
measures the battery temperature and current. The temperature is measured through
an NTC that is in contact with both batteries. The battery current is measured via the
voltage across resistor R 3 . The IIC calculates the SoC of the battery by increasing the
value of a counter during charging and decreasing it during discharging at a speed
proportional to the value of the current. Hence, the charge flowing into and out of
the battery is measured, which is commonly referred to as coulomb counting. When
no external battery current is flowing, the counter is decreased at a programmed
speed to compensate for self-discharge of the battery. This speed is proportional to
the measured battery temperature, which means that the speed is higher at higher
temperatures. This is in agreement with real battery behaviour, with the self-
discharge rate being higher at higher temperatures. The accuracy of the SoC
indication method has proven to be sufficient for use in shavers.
The IIC will signal to the HVIC which charge current should be used
depending on the value of the counter. The IIC will signal the HVIC to switch from
the normal charge current to a lower top-off charge current when the counter reaches
its 80% full reference. Because the efficiency of the charging process drops at a high
current at the end of charging, lowering the charge current at that moment keeps the
charging efficiency high and will prevent a large increase in the battery temperature.
The IIC will signal the HVIC to switch to a low trickle-charge current when the
counter reaches its 100% full reference. The microcontroller is used to infer the
Minutes Left or Shaves Left value from the SoC value determined by the IIC. The
IIC will signal the battery condition to the user through the LCD.
2.3.3 Comparison of BMS in two types of cellular phones

The cost and features are the main drivers for the complexity of the BMS in the
shaver example. In the next example, moving from one battery chemistry to a
possible combination of two chemistries leads to a complete redesign of the BMS in
a cellular phone. The first type of cellular phone, which will be named type A
below, can only be powered by a NiMH battery pack. The second type, or type B,
can be powered by a NiMH battery pack or a Li-ion battery pack. Moreover, type B
is a newer model than type A. Therefore it is smaller and weighs less than type A,
which is in line with the trend towards smaller and lighter phones.
Figure 2.11 shows the BMS of a cellular phone of type A, which is powered by
a battery pack that includes four NiMH cells in series. This yields a total voltage of
the battery pack of 4.8 V. Other NiMH battery packs with higher capacities can be
used as accessories. The battery pack is detachable and can be charged separately
from the phone on a DTC. The arrangement for controlling the charging process is
depicted in Figure 2.lla. The power supply scheme for the various system parts
inside the phone is shown in Figure 2.11b. Unlike with the shaver, not all system
parts of the cellular phone are powered directly by the battery.
The PM of cellular phone A only includes the ECC. The V-I characteristic of
the power module is shown in Figure 2.11a. The output current during charging is
800 rnA. The battery pack voltage will not reach the maximum charger voltage of
8.2 V during charging. The charger output will only reach this maximum voltage
when an open-circuit occurs. The entire control of the charging process is located
inside the cellular phone itself. When the battery pack is connected, its type is read
from the value of the identification resistor (ID) inside the battery pack. This value
is looked up in a table stored inside the phone, which indicates the corresponding
type of battery pack. Depending on the capacity of the pack that has been connected,
the charge current of 800 rnA that flows into the phone from the PM will be
controlled by the PWM switch inside the phone in such a way that the charge time is
one hour. So the resulting charge current will be higher in the case of a pack with a
26 Chapter 2
higher capacity, the maximum charge current being 800 rnA. The NTC inside the
battery pack enables temperature measurement. The measured battery pack voltage
and temperature are fed to the microcontroller, which calculates the derivatives of
the measured values after some internal filtering. As is usually the case with Ni-
based rechargeable batteries, use is made of the specific shapes in the battery voltage
and temperature curves at the end of charging to determine the full state of the
battery. The calculation of the derivatives yields information on the shape of the
measured voltage and temperature curves. After a 'battery full' detection, the duty
cycle of the PWM switch will be changed to obtain a trickle-charge current that
maintains a full battery. As the trickle-charge current has to be small, the duty cycle
of the PWM switch will be very low, which means that the switch will be closed for
only a small percentage of the time. The battery pack may also be charged separately
on the DTC. This is indicated in Figure 2.11a by means of the dashed lines. The
PWM switch and the CHC function for controlling this switch are also included in
the DTC.
Mains
- PM
- Microcontroller
/
EJ ~~ CHC (NiMH)
I
Cellular phone A
I ~
v T Type
Battery INTCIIIDII
I
800mA -+I 1-------- I I
: V: T: :Type
: I : : ~~----------------~
DTC
(a) ·---
PA
3.3 V: Digital and analogue

signal processing circuits
(b) 4 V: RF circuits
Figure 2.11: BMS of a type A cellular phone with a NiMH battery pack: (a) Control of the
charging process (b) Power supply scheme for the various system parts
A 4.8 V PA is used to amplify the RF signals to the antenna in cellular phone A.

Hence, this PA can be directly connected to the battery, as can be seen in Figure
2.1lb. Because phone A is a GSM phone, this PA will draw current in pulses, as
previously mentioned in section 2.2.4. When the battery pack is viewed as a voltage
source with series resistance for simplicity, the maximum output power obtainable
from this voltage source is proportional to the square of the voltage and inversely
proportional to the series resistance. The maximum power that can be supplied by
the battery pack will be relatively high, because the voltage is relatively high and the
series resistance in the current path from pack to PAis sufficiently low. As a result,
a relatively low value of the buffer capacitor Cb14 of 100 JJ.F is found to be sufficient
in practice to guarantee the correct supply voltage at the supply pin of the PA.
Hence, most of the peak current drawn by the PA during transmit bursts will be
drawn directly from the battery pack. The buffer capacitor serves mainly as a filter
that prevents the risk of high-frequency signals interfering with the PA operation.
Two Linear Regulators (LR) serve to power the remaining hardware inside the
cellular phone. More information on linear regulators will be given in chapter 7.
The BMS of a cellular phone of type B is shown in Figure 2.12. The phone is
supplied with a standard Li-ion battery pack that contains a single Li-ion cell of 3.6
V with an electronic safety switch, represented by the dashed box in Figure 2.12.
Moreover, a battery pack with three NiMH cells in series with a total voltage of 3.6
V can be bought as an accessory. No safety switch will then be included. The fact
that two different battery systems may be used with the cellular phone means that
the BMS will have to cover extra issues. The main issue is the difference between
charging algorithms for NiMH and Li-ion batteries. Again, the pack can be detached
from the phone and can be charged separately on the DTC, indicated by the dashed
lines. The arrangement for controlling the charging process is depicted in Figure
2.12a. The power supply scheme for the various system parts inside the phone is
illustrated in Figure 2.12b.
In addition to the ECC, the PM also includes part of the charging process
control. First of all, the V-1 characteristic is a function of the applied battery
chemistry. This can be seen in Figure 2.12a, where the maximum voltage limit is
lower for aLi-ion battery pack than for a NiMH battery pack. Secondly, the PWM
switch is now incorporated inside the PM. This is to avoid heat generation inside the
phone, as phone B is smaller than phone A. The partitioning of charge control will
depend on whether the battery pack is charged while connected to the phone or
separately on the DTC. Note that in addition to power line pair 1, through which the
charge current flows, two other wires are present. Wire 2 serves to control the PWM
switch inside the PM. Wire 3 is a battery voltage sense line. The use of wires 2 and 3
will be explained below.
When a battery pack is connected to the phone, its type will first be determined
by reading the value of the identification resistor (ID). The CHC function is
performed in the PM when the battery pack is of type Li-ion. Switch S1 will be
closed first. This will enable the PM to measure the battery voltage currentless via
wire 3. The PM will charge the Li-ion battery pack with a constant current of 650
rnA until the battery pack voltage reaches the value of 4.1 V. Then, the PM will
continue to charge at a constant voltage of 4.1 V. The battery pack temperature,
measured by the NTC, will only be used as a safeguard. This means that the charge
current will be interrupted by the opening of the PWM switch when the temperature
exceeds a certain value. Wire 2 will be used for this.
Switch S1 will be left open in the case of a NiMH battery pack. A maximum
voltage of 6.5 V will be used when the PM detects this on wire 3. The NiMH battery
pack voltage will not reach this value during normal charging. The charge current
will remain at 650 rnA until the microcontroller inside the cellular phone detects a
full battery in the same way as in phone A. After this detection, the PWM switch
inside the PM will be controlled by the phone to charge the battery with a trickle-
28 Chapter 2
charge current. This will take place via wire 2. More information on the charging
algorithms for both Li-ion and NiMH batteries will be given in chapter 5.
Mains - ECC~M 1 '1

-
bl:.' 2
'
2
\~
Microcontroller
{
3 3
PM CHC (NiMH)
I
Cellular phone B
t~ v T Type
I ~
6.5 v
(NiMH}
4,1 v
(Li-ion}
----- Battery lNTCIOQ]~~~~~fi~ I
650mA ~I
I
:---------: : '
: V: Ti Type
I o '
'' I : I
I
''
'' J-
I I --1 'f' '¥ 'f'
: :2
'I ' - - - ---------~ CHC (NiMH) I DTC
(a) 3
---------i CHC (Li-ion)
4 V: RF circuits
3.3 V: Digital and analogue
(b) signal processing circuits
Figure 2.12: BMS of a type B cellular phone with a standard Li-ion battery pack and optional
NiMH battery pack (a) Control of the charging process (b) Power supply scheme for the
various system parts
When the battery pack is charged on the DTC, its type will first be read. The voltage
sense line on wire 3 will be left open-circuit in the case of both types of battery
packs, because the DTC is equipped with a dedicated IC capable of controlling the
charging process of both NiMH and Li-ion batteries. As a result, the PM will always
use the 6.5 V maximum voltage limit, irrespective of the type of battery connected
to the DTC. All CHC functions will now be performed inside the DTC, unlike when
charging the pack on the phone. The PWM switch inside the PM will only be used
to derive the trickle-charge current in the case of NiMH battery packs, to stop
charging aLi-ion battery pack or to interrupt the charge current in the case of a fault
condition. Again, wire 2 will be used to this end.
Phone B uses a 4.8 V PA. A DC/DC up-converter is used to boost the battery
voltage from 3.6 V to 4.8 V. Figure 2.12b shows that the RF circuits have now been
connected to the output of the DC/DC converter through a linear regulator. This
linear regulator not only controls the battery voltage of 4.8 V to a 4 V supply voltage
for RF circuits, but it also serves as a filter to prevent voltage ripple from the output
of the DC/DC converter from interfering with the sensitive RF circuitry. More
information will be given in chapter 7. The other digital and analogue signal
processing circuits are connected to the battery pack through a linear regulator, as
was also the case in phone A.
Apart from the inclusion of a DC/DC up-converter between the battery pack
and the PA, an important difference with respect to phone A is the value of the
buffer capacitor Cbuf· The battery pack voltage is lower in phone B and the series re-
sistance in the current path is a lot higher. There are several reasons for this. First of
all, the safety measures used in Li-ion batteries in the form of electronic safety
switches and passive safety devices such as PTCs yield extra series resistance.
Secondly, the internal series resistance of a Li-ion battery is higher than the
resistance of three NiMH batteries in series. Thirdly, the DC/DC converter
introduces some extra series resistance. As a result, the peak current drawn by the
PA during the transmit burst now has to be drawn from the buffer capacitor. This
results in a value of roughly 2 mF.
2.4 References
[1] G. Jones, "Battery Management in Modem Portable Systems", Electronic Engineering,
pp. 43-52, November 1991
[2] B. Kerridge, "Battery Management ICs", EDN, pp. 100-108, May 13, 1993
[3] F. Goodenough, "Battery-based Systems Demand Unique ICs", Electronic Design, pp.
47-61, July 8, 1993
[4] F. Caruthers, "Battery-management Circuitry Gets Smarter", Computer Design's OEM
Integration, pp. 15-18, May 1994
[5] D. Maliniak, "Intelligence Invades the Battery Pack", Electronic Design, pp. 153-159,
January 9, 1995
[6] A. Watson Swager, "Smart-Battery Technology: Power Management's Missing Link",
EDN, pp. 47-64, March 2, 1995
[7] D. Freeman, "Freeing Portables from Battery Tyranny", Electronic Design, pp. 115-
121, July 10, 1995
[8] D. Freeman, "Integrated Pack Management Addresses Smart-Battery Architecture",
Stromversorgungsdesign, pp. 82-86, Mi!nchen, March 31, 1998
[9] F. Goodenough, "Microcontroller Grabs Data and Processes Analogue Data",
Electronic Design, pp. 139-144, April15, 1996
pp. 118-122, June 10, 1996
[11] F. Goodenough, "Battery Management ICs Meet Diverse Needs", Electronic Design,
pp. 79-96, August 19, 1996
[12] B. Kerridge, "Mobile Phones Put the Squeeze on Battery Power'', EDN, pp. 141-150,
April10, 1997
[13] L.J. Curran, "Charger ICs Reflect Shift to Smart Batteries", EDN, pp. S.10-S.11, July 3,
1997
[14] B. Schweber, "Supervisory ICs Empower Batteries to Take Charge", EDN, pp. 61-72,
September 1, 1997
[15] R. Nass, "Gas-Gauge IC Performs Precise Battery Measurements", Electronic Design,
pp. 39-42, October 13, 1997
[16] D. Stolitzka, "Smart Battery Standards Simplify Portable System Design", Electronic
Design, pp. 115-116, May 1, 1997
[17] C.H. Small, "ICs Put the Smarts in Smart Batteries", Computer Design, pp. 35-40,
August 1997
[18] J. Milios, "Harness the Power of the ACPI/Smart-Battery Standard", Electronic Design,
pp. 113-114, December 1, 1997
[19] D. Linden, Handbook of Batteries, Second Edition, McGraw-Hill, New York, 1995
[20] C. Pascual, P.T. Krein, "Switched-Capacitor System for Automatic Series Battery
Equalization", Proceedings of the Portable Power for Communications '97 Conference,
pp. 1-7, London, 1997
[21] L. Anton, H. Schmidt, "Charge Equalizer-Konzepte zum Optimalen Betrieb Seriell Ver-
schalteter Batterien", Begleittexte zum Design & Elektronik Entwicklerforum, Batterien,
Ladekonzepte & Stromversorgungsdesign, pp. 103-116, Miinchen, March 31, 1998
[22] W.F. Bentley, "Cell-Balancing Consideration for Lithium-Ion Battery Systems",
[23] L. Sherman, "Know the Peculiarities of Portable Power Supply Designs", Electronic
Design, pp. 94-104, July 7, 1997
[24] K.K. Sum, M.O. Thurston, Switch Mode Power Conversion; Basic Theory and Design,
Electrical Engineering and Electronics, vol. 22, Marcel Dekker, New York, 1984
31
Chapter 3
Basic information on batteries
Some basic information on batteries is given in this chapter. A historical overview in

section 3.1 describes the developments in battery technology. The characteristics of
the most important types of batteries that can be obtained on the market today are
summarized in section 3.2. Finally, the basic operational mechanism of batteries is
described in section 3.3. The information in this section will be used in chapter 4.
3.1 Historical overview

Batteries have been around for a long time. Earthen containers that served as
galvanic cells dating from 250 BC have been found in Baghdad [1]. These
containers were filled with iron and copper electrodes, together with an organic
acidic solution. This yielded a cell capable of supplying 250 rnA at a voltage of 0.25
V for approximately 200 hours. These cells were used to gild silver.
Two names are closely associated with the development of batteries and the
related science of electrochemistry. These names are Luigi Galvani and Alessandro
Volta [2]. Galvani performed an experiment in 1790, in which he suspended a frog
from an iron hook. With a copper probe he measured electric pulses, which he
believed originated in contractions of the muscles of the frog's legs. Volta attributed
these contractions to the current flowing between the iron and copper metals. To
prove that current could flow between two metals with an electrolyte in between, he
built a 'pile' consisting of alternating silver and zinc plates interleaved with paper or
cloth, which was soaked with an electrolyte. Hence, Volta was the first person in
modern times to have built an actual battery. He patented this structure in 1800. In
1834, Michael Faraday derived the laws of electrochemistry based on Volta's work.
This established a connection between chemical and electrical energy.
On the basis of Volta's work, other scientists also developed batteries of
various designs. A recurring problem with these batteries was gas formation at the
electrodes. This limited the available capacity, because the energy used to form the
gas could not be recovered. Leclanche found the most successful solution to this
problem in 1866 [2],[3]. He used manganese dioxide for the positive electrode,
mixed with carbon to improve conductivity. This mixture was fixed around a
graphite plate, which served as a current collector. He used zinc for the negative
electrode, while ammonium chloride served as the electrolyte. He achieved an open-
circuit voltage of 1.5 V with this battery. It could supply a medium current and had a
good shelf life. Others later improved the so-called Leclanche cell. The present-day
non-rechargeable batteries based on this principle, zinc-carbon or zinc-manganese
dioxide (ZnMn02) batteries, still dominate the household battery market on a world-
wide basis [4]. Their performance improved by 700% between 1920 and 1990.
The investigation of the behaviour of various metals in various electrolytes, in
particular diluted sulphuric acid, by Gaston Plante was another important event in
the development of batteries in 1859 [2],[3]. He constructed batteries as a sandwich
of thin layers of lead, separated by sheets of coarse cloth in a cylindrical container
filled with diluted sulphuric acid. The thin lead layers were connected as two
separate electrodes. He introduced a voltage difference between the two lead
electrodes, which charged the battery. By alternating this charge process with a

32 Chapter 3
subsequent discharge step, he eventually managed to form a positive electrode that

consisted of lead dioxide and a negative electrode consisting of finely distributed
lead. Repeated cycling could increase the thickness of these layers. After the
formation of the active layers, which eventually took some 24 hours, the charge
voltage of the lead/lead-dioxide couple rose to 2. 7 V.
In order to prevent breakdown of the lead plates, Plante later used thicker
plates, which led to quite heavy and bulky batteries that were used mainly for
stationary applications. One of the first applications of a battery of this type was in a
house-lighting installation with a dynamo, which served as charger, built by Henri
Tudor in 1882 [2]. The time-consuming formation process was later greatly
simplified by Camille Faure in France and Charles Brush in the USA in 1881 [2].
An important difference with respect to the Leclanche cell was that the cells
constructed by Plante were rechargeable. The so-called lead acid batteries are still
widely used nowadays, for example as car batteries for engine starting and vehicle
lighting, but they are used very little in consumer electronics.
Waldemar Jungner in Sweden and Thomas Edison in the USA laid the
foundation of the nickel-cadmium (NiCd) and nickel-iron alkaline storage battery
industry between 1895 and 1905 [2]. Like lead acid batteries, NiCd and nickel-iron
batteries are rechargeable. An important advantage of using alkaline solutions for
the electrolyte, instead of acidic solutions, is the possibility of using a wide range of
materials for the electrodes and containers; metals such as nickel would be affected
in an acidic electrolyte solution.
In 1839, William Grove carried out experiments to investigate the
decomposition of water into hydrogen and oxygen gas using platinum electrodes [2].
He observed that when the charging was stopped, a current started to flow in the
opposite direction due to the recombination of oxygen and hydrogen on the platinum
electrodes. These experiments laid the foundation for the development of fuel cells,
in which energy stored in fossil fuels is directly converted into electrical energy. The
active materials in a fuel cell are continuously supplied from a source external to the
cell and the reaction products are continuously removed from the cell. The work on
fuel cells also led to a better understanding of the behaviour of gas in porous
electrodes. This knowledge was later successfully applied to the development of
metal-air batteries, particularly with a zinc negative electrode and a porous air
electrode, based on carbon as positive electrode. These batteries are known as zinc-
air batteries.
In addition to the types of batteries mentioned so far, a wide variety of batteries
have been developed since Volta's 'pile', both rechargeable and non-rechargeable.
The advances in the development of portable electronic systems have had a
significant effect on the development of new types of batteries. However, only fairly
little progress have been made in improving battery characteristics such as energy
density, shelf life and reliability in comparison to the advances made in electronic
circuits [4]. While existing types of batteries are continuous! y being improved, new
types keep appearing. The introduction of the nickel-metalhydride (NiMH) battery
in 1990 and the lithium-ion (Li-ion) battery in 1991 were of a great importance for
portable consumer products. Research at Royal Philips Electronics played an
important role in the development of NiMH batteries [6],[7]. Apart from the
continuous need for higher energy densities, environmental concerns have also
boosted the development of these new types of batteries.
Basic information on batteries 33
3.2 Battery systems

An important distinction that can be made regarding batteries is that between
primary and secondary batteries. Primary batteries are non-rechargeable, whereas
secondary batteries are rechargeable. Different battery systems are available for both
primary and secondary batteries. Each battery system is characterized by its
chemistry. Examples of primary batteries are zinc-carbon (Leclanche, also known as
zinc-manganese dioxide (ZnMn02)), zinc-alkaline-Mn02 (also simply known as
alkaline batteries), zinc-air, mercuric-oxide, and lithium batteries [2]-[4]. The focus
in this book will be on secondary batteries [1]-[17]. Examples of secondary batteries
are sealed lead-acid (SLA) [11], NiCd [8], NiMH [5]-[8],[16], Li-ion [9],[14],[17],
Li-ion-polymer, Li-metal, zinc-alkaline-Mn02 [15], and zinc-air batteries. Note that
some battery systems are available in both a non-rechargeable and a rechargeable
form, such as zinc-alkaline-Mn02 batteries. Some characteristics of the most
important secondary batteries available on the market today or still under
development will be given in the remainder of this section.
3.2.1 Definitions
Various specific terms are often used in the literature and data sheets to specify the
characteristics of different battery systems. These terms are defined below
[3],[4],[10].
A. General definitions:
Cell: The basic electrochemical unit used to generate electrical energy

from stored chemical energy or to store electrical energy in the
form of chemical energy. A cell consists of two electrodes in a
container filled with an electrolyte. Definitions of these terms are
given below.
Battery: Two or more cells connected in an appropriate series/parallel

arrangement to obtain the required operating voltage and capacity
for a certain load. The term battery is also frequently used for
single cells. This terminology will also be adopted in this book,
except where a distinction between cells and batteries is needed. A
good example is a battery pack, which consists of several cells
connected in series and/or parallel. The term battery pack was
explained in chapter 2.
B. Capacity-related definitions:
Energy Density: The volumetric energy storage density of a battery,

expressed in Watt-hours per litre (Wh/1).
Power Density: The volumetric power density of a battery, expressed in

Watts per litre (W/1).
Rated Capacity: The capacity of a battery, expressed in Ampere-hours

(Ah), which is the total charge expressed in [Ah] that can
be obtained from a fully charged battery under specified
34 Chapter3
discharge conditions. These conditions are specified by

the manufacturer.
Specific Energy: The gravimetric energy storage density of a battery,

expressed in Watt-hours per kilogram (Whlkg).
Specific Power: The gravimetric power density of a battery, expressed in

Watts per kilogram (W/kg).
C. Design-related definitions:
Electrode: Basic building block of an electrochemical cell. Each cell

consists of a positive and a negative electrode. The cell
voltage is determined by the voltage difference between the
positive and the negative electrode.
Anode: The electrode at which an oxidation reaction (see section 3.3)

occurs. This means that the electrode supplies electrons to an
external circuit. The electron flow reverses between charging
and discharging. Hence, the positive electrode is the anode
during charging and the negative electrode is the anode during
discharging. Usually, a cell's anode is specified during
discharging and hence the name anode is commonly used for
the negative electrode.
Cathode: The electrode at which a reduction reaction (see section 3.3)

occurs. This means that the electrode takes up electrons from
an external circuit. Hence, the negative electrode is the
cathode during charging and the positive electrode is the
cathode during discharging. Usually, a cell's cathode is
specified during discharging and hence the name cathode is
commonly used for the positive electrode. To avoid
confusion, the electrodes will be named positive and negative
in this book.
Electrolyte: The medium that provides the essential ionic conductivity

between the positive and negative electrodes of a cell.
Separator: An ion-permeable, electronically non-conductive material or

spacer that prevents short-circuiting of the positive and
negative electrodes of a cell.
D. Application-related definitions:
C-rate: A charge or discharge current equal in Amperes to the

rated capacity in Ah. Multiples larger or smaller than the
C-rate are used to express larger or smaller currents. For
example, the C-rate is 600 rnA in the case of a 600 mAh
battery, whereas the C/2 and 2C-rates are 300 rnA and 1.2
A, respectively.
Cycle Life: The number of cycles that a cell or battery can be charged
and discharged under specific conditions, before the
available capacity in [Ah] fails to meet specific
performance criteria. This will usually be 80% of the rated
capacity.
Cut-offvoltage: The cell or battery voltage at which the discharge is

terminated. Also often referred to as End-of-Discharge
voltage. The symbol V&n will be used for this voltage
throughout this book.
Self-Discharge: Recoverable loss of capacity of a cell or battery. This is

usually expressed in a percentage of the rated capacity lost
per month at a certain temperature, because self-discharge
rates of batteries are strongly temperature-dependent.
Conditions have to be specified for certain definitions in order to be able to quantify

them For example, the rated capacity is strongly dependent on the discharge
current, battery temperature and cut-off voltage. In general, the capacity obtainable
from a battery will be lower at higher discharge currents, lower battery temperatures
and higher cut-off voltages. Usually, the rated capacity is specified at a current of
C/10 and at room temperature for secondary batteries. The specified cut-off voltage
is strongly dependent on the battery chemistry.
3.2.2 Battery design

Besides on operational conditions, such as discharge rate and battery temperature, a
battery's actual rated capacity will strongly depend on its design. In general, the
theoretical energy density, which depends on the battery chemistry, will be much
larger than the energy density specified in a battery's data sheet. This is due to the
presence of non-reactive components inside a battery, such as the current collectors
to which the electrodes are attached, the separator, the electrolyte and the container.
These components add to the battery weight and volume, but they do not contribute
to its energy density. For example, the energy density of a battery decreases when its
volume decreases, because the percentage of 'dead' volume of containers, seals and
other structural parts is larger in the case of smaller batteries. The energy density
and specific energy of every battery can be found by integrating the voltage-
dependent charge Q(V) over the applicable voltage range, similar to what was shown
in Figure 2.7, and dividing the result by the battery volume or weight, respectively.
A distinction can be made between cylindrical and prismatic batteries [3],[4].
The electrodes will be thin strips in most secondary cylindrical batteries. These
strips are rolled with the separator in between and placed in a cylindrical can.
Rolling the electrodes like this yields a relatively large surface area, enabling the
battery to supply larger discharge currents. However, this is achieved at the expense
of battery capacity, because the percentage of active material is lower than in a
design in which the electrodes are shaped like two concentric cylinders. This is due
to the fact that a current collector required to support the thin electrodes takes up
volume in a wound structure. A design with concentric cylinders maximizes the
amount of active material and is used mainly for primary batteries, e.g. zinc-carbon
and Zn-alkaline-Mn02.
36 Chapter3
In a prismatic battery, rectangular electrodes are stacked in flat rectangular boxes

with the separator in between. Prismatic batteries have a better form factor, because
they allow better utilization of the volume available inside a portable device. The
shapes and sizes of batteries are subject to standardization described in e.g. the IEC
(International Electrotechnical Commission) standard [3].
Most battery systems available on the market, whether cylindrical or prismatic
and irrespective of their size, are available in different types [3],[8]. These types
have been optimized for use in specific applications. Examples are high-rate
batteries, which allow relatively large discharge currents of up to several times the
C-rate, high-temperature batteries, which allow operation at higher temperatures,
high-capacity batteries and fast-charge batteries, which enable a very quick
recharge. These characteristics have been obtained by changing the design of the
electrodes or by adding compounds to the electrodes. For example, the use of so-
called sintered electrodes in batteries greatly decreases the internal impedance of the
cells. This allows the withdrawal of larger discharge currents.
An example of adding compounds to the electrodes of a NiCd battery will be
given in chapter 4. This is done to protect the battery during overcharging and
overdischarging. Overcharging means the continuation of charging when the battery
is full, whereas overdischarging means the continuation of discharging when the
battery is empty. The latter situation may occur when cells are connected in series in
a battery pack. In that case, the cell with the lowest capacity will be the first to reach
its empty state. The empty cell will be the first to be overdischarged when the total
battery pack voltage is still high enough to power the load. This is attributable to
manufacturing spread in characteristics, e.g. in capacity, of cells of the same type.
3.2.3 Battery characteristics

The main characteristics of the most important secondary battery systems are
summarized in Table 3.1 [1],[3],[4],[10],[15]. Ranges are given for the energy
density, specific energy, self-discharge rate and cycle life of almost all of the
systems in the table, because so many different types of battery systems are
available on the market.
NiCd batteries
The NiCd battery is commonly known as relatively cheap and robust. It is universal
and can still be found in many portable devices today. Most NiCd batteries can
supply large currents. It is possible to charge NiCd batteries in a relatively short
period of time because of their robustness. Charge times of only 10 minutes have
been reported [1]. The average cell voltage is 1.2 V. The characteristics of high
power delivery and short recharge times make NiCd batteries very popular for
power tools. Other applications include cordless phones, shavers, camcorders and
portable audio products. The positive nickel electrode is a nickel hydroxide/nickel
oxyhydroxide (Ni(OHh/NiOOH) compound, while the negative cadmium electrode
consists of metallic cadmium (Cd) and cadmium hydroxide (Cd(OH)2). The
electrolyte is an aqueous solution of potassium hydroxide (KOH). Major
improvements in energy density and specific energy have been obtained with respect
to earlier designs by using high porosity nickel foam instead of the original sintered
nickel as substrate for the active materials [4].
Table 3.1: Overview of the main characteristics of the most important secondary battery
systems
Battery system: NiCd NiMH Li-ion Li-ion-polymer SLA Rechargeable
alkaline
Average 1.2 1.2 3.6 3.6 2.0 1.5
operating
voltage [V]
Energy dens it) 90 .. 150 160.. 310 200.. 280 200 .. 250 70..90 250
Whll]
Specific energy 30 .. 60 50 ..90 90..115 100.. 110 20..40 20..85

Whlkg]
Self-discharge 10 .. 20 20..30 5.. 10 1 4 .. 8 0.2

rate [%/month
at 20"C
Cycle life 300 ..700 300.. 600 500..1000 200 200 .. 500 15 ..25
Temperature -20 .. 50 -20.. 50 -20 .. 50 ? -30.. 60 -30 .. 50

ange {"C)
Although very suitable for, for example, power tools, NiCd batteries have some
drawbacks. First of all, their energy density and specific energy are relatively low.
Secondly, NiCd batteries suffer from the so-called memory effect [1],[3],[4],[10].
This effect can be defined as a decline in effective capacity with repeated partial
charge/discharge cycles. As partial cycling continues, the battery will eventually
only be able to supply the capacity retrieved from the partial cycling. The battery
voltage drops significantly after this capacity has been removed and most portable
devices will stop functioning at that moment. However, putting the battery through
one or more complete charge/discharge cycles can restore full capacity. Most battery
experts attribute the memory effect to the Cd electrode. Although the effect is hard
to reproduce in laboratory experiments, it seems to result from the growth of large
crystals on the Cd electrode, which reduces the effective area [10]. As a final
drawback, the use of cadmium in NiCd batteries involves serious environmental
problems.
NiMH batteries
In response to the relatively low energy density and specific energy as well as the
environmental concerns associated with NiCd batteries, Sanyo Electric in Japan
introduced the first NiMH battery in 1990 [16]. A lot of useful research was
performed at Philips Research in the early seventies [6]. Table 3.1 illustrates that
NiMH batteries offer the same average operating voltage as NiCd batteries, with the
great advantage of a higher energy density. Applications include notebook
computers, cellular phones and shavers. In NiMH batteries a metalhydride (MH)
alloy has replaced the cadmium electrode. The positive electrode and the electrolyte
are more or less the same as in NiCd batteries.
The MH alloy is capable of storing hydrogen in a solid state. Two classes of
metal alloys are generally used in NiMH batteries, AB 2 and AB 5 alloys [3],[5],[6].
The AB 2 class of alloys consists of titanium and zirconium, while the alloys of class
AB 5 include rare-earth alloys based on lanthanum nickel. Almost 100% of the
38 Chapter3
commercially available NiMH batteries are based on the AB5 class of alloys [5],[6].
These alloys possess better rate capability and better stability characteristics, at the
cost of a lower specific energy than AB 2 alloys [3]. Although the chemistry of a
NiMH battery is similar to that of a NiCd battery, there are differences between the
two:
• NiMH batteries have a better energy density than NiCd batteries. This is due to
the fact that the MH electrode has a higher energy density than the Cd electrode
in NiCd batteries [3].
• The self-discharge rate of NiMH batteries is somewhat higher than that of NiCd
batteries. One of the factors that influence the self-discharge rate of a NiMH
battery is the ability of the MH electrode to retain the stored hydrogen under
storage conditions. The more hydrogen will be released under storage
conditions, the higher the self-discharge rate will be.
• Most manufacturers claim that NiMH batteries do not suffer from the memory
effect known from NiCd batteries, although examples can be found in literature
of the occurrence of the memory effect in NiMH batteries [3],[10],[12].
• In general, NiMH batteries are less robust with respect to overcharging than
NiCd batteries. This means that the charging algorithm of NiMH batteries
should be more accurate to prevent overcharging, especially when the charge
current is higher.
• A difference between the charging process of NiMH batteries and that of NiCd
batteries is that the net charging reaction in a NiMH battery is exothermic. This
means that heat is generated continuously during the charging. On the other
hand, the net charging reaction of a NiCd battery is endothermic. This means
that heat is consumed during the first phase of charging [3]. In the case of a
NiMH battery, this means that the battery temperature rises continuously during
charging, with a steep rise at the end of the charging. However, in a NiCd
battery the temperature remains relatively constant for most of the charging
process, with a steep rise at the end of the charging.
• Another difference between the charging process of NiMH and NiCd batteries
concerns the battery voltage profile, which is less pronounced in the case of
NiMH batteries than in the case of NiCd batteries at the start of overcharging.
This makes it more difficult to detect the 'battery full' condition. More
information on differences between the charging algorithms of NiMH and NiCd
batteries will be given in chapter 5.
Li-ion batteries
The first Li-ion battery was introduced by Sony in Japan in 1991. The chemistry of
Li-ion batteries differs significantly from that of Ni-based batteries [3],[9],[14]. Li-
ion cells offer the advantage of a high average operating cell voltage of 3.6 V,
because of the very negative standard potential of lithium with respect to the
standard hydrogen reference electrode (SHE). Moreover, Li-ion batteries have a
relatively high specific energy, which results in batteries that are lighter than Ni-
based batteries at the same capacity. Applications include palmtop and notebook
computers, cellular phones and camcorders. The electrodes in a Li-ion battery are
intercalation electrodes. Intercalation electrodes have a lattice structure in which
guest species may be inserted and extracted without any great structural
modifications in the host material. The operation of Li-ion batteries is based on the
transfer of lithium ions from the positive electrode to the negative electrode during
charging and vice versa during discharging. This is generally referred to in the
literature as the 'rocking chair' principle [3],[10],[14].
The positive electrode of a Li-ion battery consists of one of a number of
lithium metal oxides, which can store lithium ions. The oxides encountered most
frequently in commercially available batteries are lithium cobalt oxide (LiCo02 ),
lithium nickel oxide (LiNi02) and lithium manganese oxide (LiMn20 4 ). LiCo02 and
LiNi02 offer the advantage of a slightly higher capacity. LiMn20 4 is less toxic and
less expensive than the other materials [3]. The negative electrode of aLi-ion battery
is a carbon electrode, with the maximum ratio of the number of lithium ions and the
number of carbon atoms in the lattice being 1:6. The carbon electrode can be made
from graphite or petroleum coke. The use of graphite results in a higher capacity and
a flatter discharge curve than the use of petroleum coke [3].
The electrolytes used in Li-ion batteries are non-aqueous, as opposed to the
aqueous electrolytes used in Ni-based batteries. A salt dissolved in an organic
solvent serves as the electrolyte in Li-ion batteries. The choice of the organic solvent
is limited to those based on ethylene carbonate when graphite is used for the
negative electrode. Other solvents such as diethylene carbonate and propylene
carbonate can also be used when petroleum coke is used [10]. A popular choice for
the salt is lithium hexafluorophosphate (LiPF6). Important concerns with respect to
the electrolyte are compatibility with the chosen electrode materials, good ionic
conductivity and thermal and electrochemical stability. In general, the conductivity
of the mixtures used as electrolytes in Li-ion batteries is some two orders of
magnitude lower than the conductivity of the aqueous electrolytes used in Ni-based
batteries [3],[10].
Table 3.1 shows that, apart from a higher specific energy, Li-ion batteries also
have considerably lower self-discharge rates than Ni-based batteries. Moreover, Li-
ion batteries do not suffer from the memory effect. In applications, Li-ion batteries
have to be approached differently than Ni-based batteries, as far as both charging
and discharging are concerned. First of all, Li-ion batteries need a different charging
algorithm than Ni-based batteries. More information will be given in chapter 5.
Furthermore, Li-ion batteries are less capable of delivering large currents, expressed
in C-rate, than Ni-based batteries. Overdischarging Li-ion batteries leads to a
decrease in cycle life. Without further precautions, overcharging Li-ion batteries
leads to dangerous situations and may even cause a fire or an explosion of the
battery. Hence, it can be generally stated that overcharging and overdischarging of
Li-ion batteries is not allowed. This is an important difference from NiCd and NiMH
batteries. Therefore, strict control of the charging and discharging of the battery is
essential for safety reasons and for retaining a long cycle life. As described in
chapter 2, this leads to an increase in electronics for implementing the necessary
monitor and control functions in the form of an electronic safety switch. Combined
with a Positive-Temperature Coefficient (PTC) resistor used as a passive safety
device and a lower conductivity of the electrolyte, the total battery series resistance
of a Li-ion battery, including safety measures, can add up to around 300 mQ in
practical consumer batteries. When this series resistance is compared with the
relatively low series resistance of around 20 mQ in practical consumer Ni-based
batteries, the fact that the discharge capability of Li-ion batteries is lower than those
of Ni-based batteries is partly explained.
40 Chapter3
Li-ion-polymer batteries
As a successor of the Li-ion battery with a liquid organic electrolyte, the 'solid-state'
Li-ion battery started to appear on the market at the end of 1997. The basic
difference with respect to Li-ion batteries is that the electrolyte consists of a solid
ion-conducting polymer material [3],[4],[10]. The polymer electrolyte also serves as
a separator. Batteries of this type were referred to as Li-ion-polymer batteries in
Table 3.1.
The conductivity of polymer electrolytes is even lower than that of liquid
organic electrolytes [3]. Most liquid organic electrolytes have a conductivity in the
order of 10·3 (Qcm)" 1 at 20°C. A polymer electrolyte, such as the commonly used
polyethylene oxide (PEO) electrolyte, has a conductivity in the order of 10"8 (Qcm)" 1
at 20°C. As a result, the polymer electrolytes have to be kept very thin to achieve a
reasonable conductivity. Operation of the battery at higher temperatures also leads to
an improved conductivity of the polymer electrolyte. This is impractical for use in
many portable products. Improvements in conductivity at room temperature have
been obtained by adding liquid plasticizers, such as polypropylene carbonate. This
increases the conductivity to 10·4 (Qcm)" 1 at 20°C. Another approach involves the
use of so-called 'gelled' electrolytes. These electrolytes are obtained by trapping a
liquid solution of a lithium salt in an organic solvent in a polymer matrix. Using
such 'gelled' electrolytes, conductivities of up to 10·3 (Qcm)" 1 at 20°C have been
obtained.
Polymer electrolytes are less reactive with respect to lithium than liquid
electrolytes. This is beneficial for the safety of the battery. However, adding liquid
plasticizers or using 'gelled' electrolytes to increase the conductivity of the polymer
electrolyte leads to an electrolyte that is more reactive with respect to lithium and
part of the safety benefit is consequently lost. The fact that the polymers have to be
made very thin to have a reasonable conductivity moreover makes them vulnerable
to mechanical stress and pin-hole defects.
The use of a polymer electrolyte offers the possibility of fast production of the
cells using web equipment and the fabrication of thin cells [4],[10],[17]. The
electrodes, electrolyte and current collector are simply stacked in a sandwich
structure. Because the polymer keeps the structure together, no outside pressure of a
can is needed to form a battery. Instead, the housing can be made from a thin
laminated foil material. The Li-ion-polymer cells can be configured in many
possible dimensions, because they can be made with a flexible width, length and
thickness. This increases the energy density for a given battery cavity in a portable
product. Some battery manufacturers, such as Ultralife, have recently introduced Li-
ion-polymer batteries on the consumer market. The Ultralife batteries are based on
the use of a 'gelled' type of electrolyte. The technology of such batteries must
however still be substantially improved for them to acquire a significant market
share. The numbers cited in Table 3.1 are consequently subject to change.
Li-metal batteries
A further development of lithium-based batteries that is still in the research phase is
the replacement of the carbon negative electrode by a metallic lithium electrode.
Such batteries are commonly referred to as Li-metal batteries. The literature gives an
example of Li-metal batteries under development by the battery company Tadiran
[12],[13]. The biggest advantage of storing lithium in its metallic fonn, instead of its
ionic form surrounded by carbon atoms in the maximum ratio of 1:6, is a gain in
energy density and specific energy. However, the use of metallic lithium introduces
the severe problem of its very high reactivity. For example, metallic lithium will
react with any liquid electrolyte and forms a passivation film on the electrode after
every charge/discharge cycle. This process consumes a lot of lithium Therefore,
lithium has to be present in excess amounts, which lowers the energy density and
specific energy.
Research was initiated into the possibility of using a polymer electrolyte in Li-
metal batteries, because lithium reacts less with polymer electrolytes than with
liquid electrolytes. However, a problem that still arises in this case is the build-up of
surface irregularities on the lithium electrode during charging of the battery. These
irregularities are known as dendrites. There is a risk of short-circuiting by dendrites
that puncture the solid polymer film, because the solid-polymer electrolyte has to be
rather thin to ensure sufficient conductivity. At best, such short-circuiting shortens
the battery's cycle life. The substantial increase in temperature as a result of the heat
generated by the short-circuit current however also implies safety problems such as
flaming. Extensive research will have to be carried out to solve such safety problems
and increase the cycle life of Li-metal batteries.
SIA baneries
A relatively old battery technology, which can still be found on the secondary
battery market today, is that of the SLA battery [3],[4]. The chemistry of an SLA
battery is basically the same as that of an ordinary car battery. SLA batteries are
however maintenance-free. This means that the electrolyte does not have to be
replaced. SLA batteries contain only a limited amount of electrolyte, which is
absorbed in the separator or a gel. The positive electrode of an SLA battery is
formed by lead dioxide (Pb02), while metallic lead (Pb) in a high-surface-area
porous structure is used for the negative electrode. A sulphuric acid (H2S04 )
solution is used for the electrolyte. The average operating voltage of an SLA cell is 2
v.
Advantages of the SLA battery are its good rate capability and relatively low
self-discharge rate; see Table 3.1. Moreover, SLA batteries do not suffer from the
memory effect. They can be used under a continuous float charge. The reason for
this is that a starved electrolyte and an excess amount of material in the negative
electrode are used in the design. This facilitates the recombination of oxygen that is
formed during overcharging. As a result, large pressure build-up inside the battery is
prevented [3]. Moreover, the batteries are manufactured with a pressure valve. This
valve is resealable and opens only under extreme pressure build-up. The valve
prevents the introduction of ambient oxygen. The possibility of withstanding long
periods of float charge is advantageous for use in a back-up system. A final
advantage is that SLA batteries are low in costs.
A major disadvantage of SLA batteries is their low energy density and specific
energy. Especially when compared with Li-ion batteries, SLA batteries are heavy
and take up a large amount of space. An additional disadvantage is the problem of
irreversible capacity loss under deep discharge conditions. The deep discharge
condition may even occur due to self-discharge of the battery. This means that the
cycle life of an SLA battery drastically decreases when it is not frequently recharged
[1].
The term SLA is generally reserved for cylindrical batteries. The term Valve-
regulated lead-acid (VRLA) is used for prismatic types. VRLA batteries generally
vent at lower pressures. An example of an ultra-thin VRLA battery was introduced
by the company Bolder [11]. This battery is manufactured using a thin-metal film
technology. The rate capability is extremely high and these batteries can be charged
42 Chapter3
in short times of only several minutes, because the technology results in a very low
battery impedance. However, as the technology is based on lead acid, the energy
density and specific energy of such batteries are relatively low. This limits wide-
spread use in portable devices.
Rechargeable alkaline batteries

Secondary zinc-alkaline-Mn02 batteries based on the same chemistry as their
primary counterpart have become available on the market [3],[15]. Batteries of this
type were introduced by the Renewal company in the USA in 1993 [4]. As with the
primary type, the positive electrode consists of manganese dioxide (MnOz), the
negative electrode of zinc (Zn), and KOH is used for the electrolyte. An increase in
the capacity of the Zn electrode, a change in the separator design and additives to
both electrodes allow recharging [3]. Hence, although some charger manufacturers
claim to be able to recharge ordinary alkaline batteries, only the specially adapted
secondary types are really rechargeable.
The average operating voltage of a rechargeable alkaline cell during practical
use is 1.3 V. The rechargeable alkaline battery offers the advantage of a low self-
discharge rate and low cost. Disadvantages are the poor cycle life and the fact that
the initial capacity is lower than that of primary alkaline batteries. For example, the
initial capacity at 20°C is about 70% of the capacity of the primary battery [3].
Twenty discharge cycles can be realized before the capacity drops to 50% of the
initial capacity when the battery is discharged and subsequently charged from a
completely discharged state. This means that the available capacity decreases rapidly
with cycling. The achievable recharged capacity moreover decreases when the
battery has previously been deeply discharged.
Zinc-air batteries
In addition to the secondary battery systems described so far, various other systems
can be found in the literature. An interesting new development is the secondary zinc-
air battery [3],[10]. This battery system is still in the pioneering phase. Its chemical
structure is the same as that of its primary counterpart. The positive electrode is
made of carbon, the negative electrode consists of Zn, while KOH is used as the
electrolyte. The carbon electrode is exposed to the air and is only used as a reactive
surface, which is why it is known as an 'air' electrode. The voltage of a zinc-air cell
ranges from 1 V to 1.2 V.
The simplest way of 'recharging' a zinc-air battery is by replacing the Zn
negative electrode when all the Zn has been consumed. In a practical consumer
situation the battery should however be recharged electrically. The carbon 'air'
electrode then produces oxygen during charging and consumes oxygen during
discharging. An air-management system is needed for proper operation of a
secondary zinc-air battery. This system ensures the flow of air into the battery
required to supply current. On the other hand, the flow of air from outside is blocked
when no current is drawn. This prevents crystallization of carbonate in the porous air
electrode caused by carbon dioxide absorbed from the air. Such crystallization could
impede the introduction of air. Also, blocking the airflow prevents dehydration of
the cell or the opposite situation of flooding of the air electrode pores. It also reduces
the self-discharge rate of the battery.
Obviously, the air-management system adds to the volume and weight of the
battery. This decreases the resulting energy density and specific energy. The present
zinc-air batteries are moreover very sensitive to electrical abuse. To overcome this
problem, zinc-air batteries have built-in monitoring and charging circuitry. Some
manufacturers aim at the portable computer market, but no commercial battery has
so far been introduced.
3.3 General operational mechanism of batteries
3.3.1 Introduction
The general operational mechanism of a battery will be discussed in this section. In
its simplest definition, a battery is a device capable of converting chemical energy
into electrical energy and vice versa. The chemical energy is stored in the electro-
active species of the two electrodes inside the battery. The conversions occur
through electrochemical reduction-oxidation (redox) or charge-transfer reactions
[2],[3],[18],[19]. These reactions involve the exchange of electrons between electro-
active species in the two electrodes through an electrical circuit external to the
battery. The reactions take place at the electrode/electrolyte interfaces. When current
flows through the battery, an oxidation reaction will take place at the anode and a
reduction reaction at the cathode. The oxidation reaction yields electrons to the
external circuit, while a reduction reaction takes up these electrons from the external
circuit. The electrolyte serves as an intermediate between the electrodes. It offers a
medium for the transfer of ions. Hence, current flow is supported by electrons inside
the electrodes and by ions inside the electrolyte. Externally, the current flows
through the charger or load.
The charging and discharging of a battery are schematically illustrated in
Figure 3.la and b, respectively. In both cases, the negative electrode (-)is shown on
the left and the positive electrode (+) on the right. Figure 3.1 shows that oxidation
occurs at the positive electrode during charging, whereas reduction occurs at the
negative electrode. The reverse takes place during discharging.
Electrons Electrons
--<·-···· -)>-
(a) (b)
Figure 3.1: Electrochemical operation of a battery during charging (a) and discharging (b)
The charge-transfer reactions taking place at the positive and negative electrode can
be represented in a simplified form by:
44 Chapter3
Positive electrode (+):
BOx++ne (3.1)
Negative electrode (-):

k,,.
COx.+me DRed_ (3.2)
k.,.
where Ox and Red denote the oxidized and reduced species, respectively, and ka,+l-
and kc,+l- denote the reaction rate constants for the anodic and cathodic reactions,
respectively, whoses dimensions depend on the reaction equation. In each electrode,
oxidized and reduced species are present, which form a redox couple. Note that (3.1)
and (3.2) have been simplified for convenience. Other species will often be
involved, resulting in more complex reactions. During the charging of the battery, A
molecules of the reduced species Red+ are converted into B molecules of the
oxidized species Ox+ at the positive electrode, yielding n electrons (e). At the same
time, C molecules of the oxidized species Ox_ are converted into D molecules of the
reduced species Red_ at the negative electrode, taking up m electrons. The reactions
at both electrodes are reversed during discharging. The net reaction can be found by
adding (3.1) and (3.2). When n =m is assumed, this yields:
ch
A Red+ + C Ox_ ..: • B Ox+ + D Red_ (3.3)
d
3.3.2 Basic thermodynamics

The battery is in a state of equilibrium when no external current flows and the
reaction rates of the electrode reactions given by (3.1) and (3.2) in the right and left
directions are the same. The equilibrium potential (E'q) of each electrode can be
obtained with Nemst's equation [2],[3],[18],[19]:
(3.4)
for the positive electrode and
(3.5)
for the negative electrode, where Et is the standard redox potential of electrode i in
[V], R is the gas constant (8.314 J/(mol.K)), Tis the temperature in [K], n and m
denote the number of electrons involved in the charge-transfer reactions, F is
Faraday's constant (96485 C/mol), and a; is the activity of species i in [mollm3].
The activity a; is linearly proportional to its concentration (c;) and molar
amount (m;) according to:
}.m.
a.= y.c. = 1
(3.6)
1 1
Volume
where yis the dimensionless activity coefficient, which will be assumed to be unity
throughout this book, c; denotes the concentration of species i in [mollm3], m; is the
molar amount of species i in [mol] and the Volume in which species i resides is
given in [m3 ]. From (3.4) and (3.5) the following expression is obtained for the
equilibrium potential of the complete battery:
(3.7)
where again n = m has been assumed for simplicity. (3. 7) shows that the value of the
equilibrium potential of a battery depends on the ratio of the activities of the
oxidized and reduced species in the two electrodes. Therefore, the value of the
equilibrium potential depends on the State-of-Charge (SoC) of the two electrodes
and hence on the battery's SoC.
The Gibbs free energy change (L1G") of the battery under standard conditions is
the driving force enabling a battery to supply electrical energy to an external circuit.
L1G" in [J/mol] is given by [3],[19]:
(3.8)
The expression L1G"=+nFE'bar is also encountered in the literature, depending on the

convention that is used. This sign confusion is due to the fact that the sign of L1G"
depends on whether the reaction is an oxidation or a reduction reaction and the sign
of Ifbat is fixed [19]. More information will be given in chapter 4.
3.3.3 Kinetic and diffusion overpotentials

Electrical energy is supplied to or taken up from an external circuit in a non-
equilibrium situation. The rates in the left and right directions of the electrode
reactions of (3.1) and (3.2) are then not the same. For example, the reactions from
left to right prevail at both electrodes during the charging. The actual battery voltage
differs from the equilibrium value dictated by (3.7) when current flows through the
battery. This is due to polarization at the electrode/electrolyte interfaces, resulting
from the charge-transfer reactions. The difference between the actual electrode
potential E during current flow and the equilibrium electrode potential E'q of each
electrode is denoted as the overpotential rf' of the charge-transfer reaction
[2],[3],[6],[18],[19]. Hence,
(3.9)
where the sign of the overpotential rf' depends on the direction of the current that
flows into or out of the electrode. An oxidation reaction results in a current la, which
means that electrons flow out of the electrode. A reduction reaction results in a
46 Chapter 3
current 10 which means that electrons flow into the electrode. These currents are
equal and opposite in direction in a state of equilibrium, resulting in a zero net
current; see Figure 3.2. In a state of non-equilibrium however, one current is larger
than the other. For example, a net oxidation reaction will occur when Ia is larger than
lc. A net current la-Ic will then flow into the electrode and the overpotential will be
positive. A net reduction reaction will occur when la is smaller than lc and a current
la-Ic will then flow out of the electrode. The overpotential will then be negative. For
example, the overpotential will be positive at the positive electrode and negative at
the negative electrode during charging.
The overpotential can be seen as the driving force of the electrode reaction.
Both kinetic aspects and mass transport phenomena contribute to its value. Kinetic
aspects result in a kinetic overpotential ( r/') and mass transport phenomena result in
a diffusion overpotential (rf). This means that
(3.10)
The Butler-Volmer equation is valid for the relation between r/' and the net reaction
current I for each charge-transfer reaction [2],[3],[18],[19], i.e.
(3.11)
r
where expresses the exchange current for the charge-transfer reaction in [A] and a
denotes the dimensionless transfer coefficient, 0 < a < 1. In a state of equilibrium, I
r.
and rf are both zero and Ia = -lc = The exchange current I} of a charge-transfer
reactionj is given by [6],[19],[20]:
(3.12)
where A expresses the surface area of the electrode at which the reaction occurs in
at x
[m2], stands for the bulk activity of species i in [mollm3] and andy express the
dimensionless reaction orders of the Ox and Red species. For example, assuming
that the reaction given in (3.1) is the rate-determining reaction occurring at the
positive electrode, the values of x=B and y=A have to be used for obtaining r+ from
(3.12). The bulk activity is equal to the surface activity a' at the electrode/electrolyte
interface in the case of all species when the reaction rate is not limited by mass
transport processes. Hence, the concentration of each species that takes part in the
reaction is uniformly distributed throughout the electrode or the electrolyte. (3.12)
shows that the value of the exchange current depends on the values of the bulk
activities of the oxidized and reduced species, and hence on the electrode's SoC.
The current-voltage relations for both electrodes of the battery of Figure 3.1 are
shown in Figure 3.2. It is assumed that no mass transport limitation occurs, so rf is
zero. The arrows for I, Vbat and r/'+l· denote the situation during charging. Figure 3.2
shows that the net reaction current, represented by a solid line, results from currents
la and lc, represented by dashed curves, at each electrode. This is in agreement with
(3.11). The overpotential is indeed positive at the positive electrode and negative at
the negative electrode when I flows in the directions indicated for the charging. The
battery voltage Vbat equals the difference between the positive electrode potential E+
and the negative electrode potential E_. This means that, during the charging, Vbat is
larger than the battery equilibrium potential eeqbat· The direction of I at both
electrodes is reversed during discharging. In that case, Vbat will be smaller than
eeqbat· The net reaction current and the overpotentials are zero at both electrodes in a
state of equilibrium. Vbat then equals eeqbat·
The exchange current /+0 was chosen to be smaller than /_0 in the example
shown in Figure 3.2. As a result, the overpotential 7J/ is larger than 17-k· This also
r
follows from (3.11 ), where if has to be larger at smaller values to allow a certain
value of I at a constant temperature T. This means that in this example, the kinetics
of the reaction at the positive electrode are worse than at the negative electrode. The
battery voltage Vbat depends on the battery's SoC, because eeqbat depends on the
battery's SoC through (3.7), and 17} and 17-k depend on the SoC of the respective
electrodes through (3.11) and (3.12).
t
--v
Figure 3.2: Current-voltage relation for the positive (+) and negative(-) electrode without
mass transport limitations, rf=O [6]. The battery voltage Vbar• current I and kinetic
overpotentials rf +I- during charging are shown
If the value of the overpotential if in (3.11) is large enough, either Ia or Ic can be

neglected. In that case, the equation can be rewritten by taking the ln-form. This way
the Tafel relation is found, which is often encountered in the literature [3],[19].
When Ia >> Ic the Tafel relation reads:
(3.13)
The value of a can be inferred from the slope and the value of ? is found by
extrapolating the plot to if=O when if and I are plotted according to (3.13) and n is
a known quantity. In principle, this makes it easy to derive these values from
measured overpotentials and electrode currents.
So far it has been assumed that the bulk and surface concentrations are the
same. In practice, mass transport limitations may lead to a difference between the
surface and average concentration of reacting species. The reaction rate is influenced
when the electrode surface becomes depleted of or accumulated with species
48 Chapter3
consumea m or generated by the reaction, respectively. Mass transport of species to

and from electrode surfaces can occur in essentially three processes [3],[18],[19]:
(1) diffusion as a result of a concentration gradient;

(2) migration as a result of an electrical field;
(3) convection.
Both charged and uncharged species can diffuse. Typically, species will diffuse
from a region with a high concentration to a region with a low concentration. Only
charged particles can migrate. The potential change across a medium and so the
electrical field will be low when the medium through which the charged species is
transported is highly conductive. So migration then contributes little to mass
transport of the species. Convection occurs in, for example, zinc-air batteries, where
the air-management system ensures the circulation of air through the system.
One-dimensional diffusion in the x direction can be described using Pick's first
and second law [3],[18],[19]:
J .(x,t ) = -D .-"-'--.:....
lic;(x,t)
(3.14)
I I dx
and
&;(x,t) =D. t5 2 c;(x,t)

(3.15)
& 1 c:5x2
where Jlx,t) denotes the flux of the diffusing species i in [moV(m2.s)] at location x
[m] and time t [s], D; is the diffusion coefficient of the diffusing species i in [m2/s]
and clx,t) is the concentration of the diffusing species i in [moVm3] at location x and
timet.
Initial and boundary conditions have to be defined for each diffusion problem.
The initial condition specifies the concentration of the diffusing species at t =0 at all
locations x. For example, a species can be uniformly distributed throughout the
medium through which diffusion occurs at t = 0. The boundary conditions specify
the concentrations or concentration gradients at certain positions x for all times t. For
example, boundary conditions can be valid at an electrode/electrolyte interface or for
x approaching infinity. The solution of the diffusion equations will depend on the
initial and boundary conditions. Consider the situation depicted in Figure 3.3, where
the diffusing species i is extracted at x =0, for example at the electrode surface, in a
charge-transfer reaction. After some time t, c' may become close to zero, as
indicated in the figure, and the concentration profile becomes linear.
The species is supplied from the bulk through diffusion. The concentration of
the diffusing species is constant and equal to the bulk concentration cb at a distance x
larger than the diffusion layer thickness 0. An expression for the diffusion flux J can
be easily derived from Pick's first law (3.14) when the concentration profile is
stationary. This occurs when the concentrations at x =0 and x = 8, as well as 8 itself
remain constant. The maximum diffusion flux J; occurs when c• approaches zero:
]. = - D.(cb- c•) = - D.cb -

-D.ab
I
(3.16)
or
I I
I 0 0
c =ri'
~X
Figure 3.3: Concentration c;(x) of a diffusing species i during stationary diffusion in the x
direction
Because the flux is negative, it will flow from right to left in Figure 3.3. So, the
species indeed diffuses from a high to a low concentration.
The concentration profile of a species that takes part in a charge-transfer
reaction will give rise to an overpotential rf in addition to the overpotential r/. This
additional overpotential is given by [19]:
7J d =RT
-In-
nF
(c• J
cb
(3.17)
When the concentration of a reacting species approaches zero at the electrode

surface, the reaction rate becomes limited by diffusion and hence by the speed at
which the species is supplied at the electrode surface. Therefore, the current levels
off to a maximum value in the current-voltage characteristic of a reaction for which
diffusion is important, for example proportional to the maximum flux in (3.16) in
the case of stationary diffusion. The current-voltage relation described by (3.11) will
then no longer be valid. This is schematically shown in Figure 3.4, where the anodic
reaction is assumed to be limited by diffusion. The dashed curves denote currents 10
and lc according to (3.11). As can be clearly seen in the figure, the anodic current
levels off to a diffusion-limited current Id,max·
Taking a closer look at Figure 3.4, we see that the overpotential TJ may
approach infinity when the anodic reaction becomes diffusion-limited. In practice,
this will not be the case, because different charge-transfer reactions may take place
at each electrode. The competition between these charge-transfer reactions is
determined by the thermodynamics and kinetics of these reactions. For example,
when the rate of an oxidation reaction decreases because of kinetic or diffusion
50 Chapter3
limitations, a different oxidation reaction that involves different species will start to
compete with this oxidation reaction. Such competition between charge-transfer
reactions is very important in determining the characteristics of aqueous battery
systems. This will be described in chapter 4 for NiCd batteries.
I
t I d,max
-------
-v
t
Eeq
Figure 3.4: Current-voltage relationship for an electrode reaction in which the anodic reaction
is diffusion-limited at higher anodic reaction currents
The diffusion problem shown in Figure 3.3 can be easily solved because of its
stationary nature. However, it will often be impossible to solve diffusion equations
(3.14) and (3.15) analytically, for example because the diffusion layer thickness
cannot be considered constant. In such cases, the diffusion equations can be solved
numerically by taking the appropriate initial and boundary conditions into account.
Examples of ways of solving the diffusion problem for different diffusing species
can be found in the next chapter.
3.3.4 Double-layer capacitance

Charge separation will occur at the electrode/electrolyte interface when an electrode
is immersed in an electrolyte [3],[19]. The electrical charge present on the electrode
surface will attract ions of opposite charge in the electrolyte and will orient the
solvent dipoles. This will result in an electrical double-layer, described by the
Helmholtz theory in its simplest form. In this theory, the potential changes linearly
from the electrode potential qf to the electrolyte potential qi in a thin layer with
thickness OH· In its simplest form, oH equals the radius of the adsorbed hydrated ions
in the electrolyte. This layer is commonly referred to as the Helmholtz layer and can
be described by a constant capacitance CH.
In a more detailed approach, a layer consisting of ions of both positive and
negative charges exists in addition to the Helmholtz layer. This layer is denoted as
the Gouy-Chapman diffuse layer and forms a space-charge layer with a charge
density that decreases gradually into the electrolyte. This layer is wider for more
dilute solutions. The Helmholtz and Gouy-Chapman theories are schematically
illustrated in Figure 3.5.
The electrostatic potential in the Gouy-Chapman layer changes exponentially
with the distance from the electrode/electrolyte interface. As a result, the
corresponding capacitance is voltage-dependent, and hence non-linear. This non-
linear capacitance Ca.c is connected in series with CH. In addition to the Helmholtz
and Gouy-Chapman theories, other more complicated theories concerning the
electrical double-layer can be found in the literature [18]. These theories are
however beyond the scope of this book.
Electrode +
«P'
Gouy-Chapman layer
Helmholtz layer Helmholtz layer
-II-
(a) (b)
Figure 3.5: Helmholtz model (a) and Gouy-Chapman model (b) of the electrical double-layer
at an electrode/electrolyte interface. The electrode is assumed to be positively charged and
hence to attract negative ions from the electrolyte in this example. The electrode potential is
¢1 (s=solid) and the bulk electrolyte potential is ¢} (!=liquid)
The charging of the electrical double-layer with a current Idl can be described by:
d dA d
ld1 =-(AQ)=Q(E)-+A-Q(E) (3.18)
dt dt dt
where Q denotes the electrical charge stored in the double-layer per unit area in
[C/m2], A represents the surface area of the electrode/electrolyte interface in [m2]
and E is the electrode potential in [V], which equals f/1 - ¢1. When, as in the simple
Helmholtz theory, the double-layer capacitance is considered to be constant, and
hence independent of the electrode potential, and the surface area of the electrode in
contact with the electrolyte does not change, (3.18) reduces to the simple equation of
charging a capacitor, i.e.:
I =ACdl dE (3.19)
dl dt
where ct1, the double-layer capacitance per unit area, is equal to dQ!dE. The double-
layer capacitance is almost entirely determined by CHin the case of concentrated
electrolyte solutions commonly encountered in battery systems [2]. The value of the
double-layer capacitance is found to be in the order of 0.2 F/m2 in the case of a
smooth electrode/electrolyte surface [2],[18].
3.3.5 Battery voltage
To sununarize the theory presented in this section, the total battery voltage Vba, is
given by the difference between the electrode potentials E+ and E_, i.e.:
V:bat =E+ -E- =(Eeq±nk±nd)_IEeq+nk+nd)±l~R

~ + 'I+ 'I+ ~ - '1- '1- L.J Q (3.20)
The overpotentials rf and rf are positive at the positive electrode and negative at the
negative electrode when the battery is charged. This situation is reversed during
discharging. As a result of ohmic resistances in the electrodes and the electrolyte
(IR.a). the battery voltage increases with IIR.a during charging. This ohmic voltage
drop is subtracted during discharging. The potential development from the positive
to the negative electrode during charging is shown in Figure 3.6.
- Charger
-e
E_
Figure 3.6: Potential development inside a battery duringcharging; see (3.20)
Figure 3.6 shows an ohmic voltage drop E.a.+ in the positive electrode, E.a,elyt across
the electrolyte and E.a,. in the negative electrode. The equilibrium and overpotentials
exist across the electrode/electrolyte interfaces.
3.4 References
[1) M Prochaska, C. Fabich, "Stromkonserve, Akkutechnologien: Von Blei bis Lithium-
Ion", ELRAD, Heft 12, pp. 54-75, 1994
[2) M. Barak (Ed.), T. Dickinson, U. Falk, J.L. Sudworth, H.R. Thirsk, F.L. Tye,
Electrochemical Power Sources: Primary & Secondary Batteries, lEE Energy Series 1,
A. Wheaton &Co, Exeter, 1980
[4] R.A. Powers, "Batteries for Low Power Electronics", Proceedings of the IEEE, vol. 83,
no.4,pp.687-693,April1995
[5] J.R. van Beek, H.C. Donkersloot, J.J.G. Willems, "Rechargeable Hydride Electrodes for
Ni-H2 Batteries Based Upon Stable Hydrogen-Storing Materials", Proc. of the 141h Int.
Power Sources Symposium (Brighton, United Kingdom, September 1984), chapter 22 in

Power Sources, vol. 10, L. Pearce, Ed., pp. 317-338, London, 1985
[6] P.H.L. Notten, "Rechargeable Nickel-Meta!Hydride Batteries: a Successful New
Concept", lnterstitiallntermetallic Alloys, chapter 7 in NATO ASI Series E 281, vol.
151, F. Grand-jean, G.J. Long and K.H.J. Buschow, Ed., 1995
[7] P.H.L. Notten, J.R.G. van Beek, "Nickel-Meta!Hydride Batteries: from Concept to
Characteristics", Chemical Industry, vol. 54, pp. 102-115, 2000
[8] N. Furukawa, "Developers Spur Efforts to Improve NiCd, NiMH Batteries", lEE, pp.
46-50, October 1993
[9] I. Kuribayashi, "Needs for Small Batteries Spur Progress in Lithium-Ion Models", lEE,
pp. 51-54, October 1993
[10] M.J. Riezenman, "The Search for Better Batteries", IEEE Spectrum, pp. 51-56, May
1995
[11] B. Nelson, "High-Power Rechargeable Cells will be Available for Wireless
Applications", Wireless System Design, pp. 58-60, January 1997
[12] P. Dan, "Recent Advances in Rechargeable Batteries", Electronic Design, PIPS
Rechargeable Battery Technologies, pp. 112-116, February 3, 1997
[13] V. Biancomano, "Various Chemistries Protect PC Charges", Electronic Engineering
Times, pp. 107-119, July 14, 1997
[14] G.H. Wrodnigg, B. Evers, I. Schneider, J.O. Besenhard, M. Winter,
"Entwicklungstendenzen von Lithium-Batterien", Begleittexte zum Design & Elektronik
Entwickleiforum, Batterien, Ladekonzepte & Stromversorgungsdesign, pp. 17-26,
Miinchen, March 31, 1998
[15] R. Muller, "Grundlagen der Neuen Wiederaufladbaren AccuCell-Batterien",
Begleittexte zum Design & Elektronik Entwickleiforum, Batterien, Ladekonzepte &
[16] N. Furukawa, T. Ueda, "Sanyo NiMH batteries Outperform Li-ion", Nikkei Electronics
Asia, pp. 80-83, May 1996
[17] G. Smith, "Li-ion Units Lead Portable Charge", Electronic Engineering Times, pp. 114-
120, July 14, 1997
[18] J. O'M. Bockris, D.M. Drazic, Electro-Chemical Science, Taylor&Francis Ltd.,
London, 1972
[19] A.J. Bard, L.R. Faulkner, Electrochemical Methods: Fundamentals and Applications,
John Wiley&Sons, New York, 1980
[20] P.H.L. Notten, P. Hokkeling, "Double-Phase Hydride Forming Compounds: A New
Class of Highly Electrocatalytic Materials", 1. Electrochem. Soc., vol. 138, no. 7, pp.
1877-1885, July 1991
55
Chapter 4
Battery modelling
This chapter describes the development of simulation models of rechargeable

batteries. Battery behaviour is a result of a complex interaction between various
electrochemical and physical processes. Therefore, simulation models based on the
mathematical description of these processes are a useful tool in optimizing the
Battery Management System (BMS) used in portable devices. By applying all sorts
of external electronic and thermal stimuli to the model during simulations, the
designer of a BMS can not only investigate the development of the battery voltage
and temperature, but also the course of each of the various reactions that take place
inside the battery. This chapter discusses the development of battery models of this
type, which have not yet been reported in the literature in this form.
The general approach to modelling rechargeable batteries adopted in this
book is based on the theory of physical system dynamics. This theory is based on the
analogy between the definition of energy and power in several physical domains,
such as the electrical domain, the chemical domain and the thermal domain [1}. The
general approach is described in section 4.1. On the basis of this approach, the
development of a simulation model of a rechargeable NiCd battery and a
rechargeable Li-ion battery is described in sections 4.2 and 4.3, respectively. It is
shown that these models are based on a wide variety of parameters. Finding the
correct values for all the parameters that occur in the battery models is an
important task, because a close quantitative agreement between simulated and
measured battery characteristics is essential. The approach adopted for this task
and its application to the parameters of the NiCd model is described in section 4.4.
Some simulation examples based on the NiCd and Li-ion models are described in
section 4.5. Finally, conclusions are drawn in section 4.6.
4.1 General approach to modelling batteries

Modelling rechargeable batteries based on a mathematical description of the
processes that occur inside the battery has often been reported in the literature. Apart
from electrode models, such as Ni-electrode models [2],[3], complete battery models
that are based on the mathematical description of electrochemical and physical
processes can be found for SLA [4], NiCd [5],[6], Li-ion [7] and Li-polymer
batteries [8]. No use is made of an equivalent network and these models only
describe the batteries' voltage characteristics. None of these models considers the
battery temperature and internal gas pressure development during operation of the
battery and the interaction between these characteristics. In fact, the battery voltage,
pressure and temperature are closely linked in a sealed rechargeable battery and their
mutual influence is considerable. For example, both the internal partial oxygen
pressure and the battery temperature influence the rate of oxygen production and
recombination in NiCd batteries. These reactions do not contribute to effective
energy storage inside the battery. Therefore, the internal partial oxygen pressure and
the battery temperature influence the charging and discharging efficiency.
Besides purely mathematical models, battery models in the form of equivalent
electronic networks can also be found in the literature [9]-[12]. Three approaches
can be distinguished:

56 Chapter4
(i) The equivalent electronic network may consist of linear passive elements,
such as resistances, capacitances and inductances, to account for the
impedance of a battery [9]. However, these models generally fail to
describe the most important battery characteristics due to the complex non-
linear behaviour of batteries.
(ii) The equivalent electronic network may consist of linear passive elements in
combination with voltage sources, which describe the battery behaviour by
means of look-up tables [10],[11]. In this approach, the simulated battery
characteristics can only be realized through interpolation of the tabulated
data. Moreover, the accuracy of this type of model depends strongly on the
amount and reliability of the tabulated data.
(iii) An equivalent electronic-circuit model for a NiCd battery based on non-
linear passive and active elements configured around a PIN-diode model
has been reported [12]. Again, this model does not take into account the
influence of the internal partial oxygen pressure and battery temperature on
the battery's behaviour. Also, many parameters in this model do not have
any electrochemical meaning.
An advantage of these models is that they can be simulated together with

surrounding electronic circuits, such as a charger, using a conventional electronic-
circuit simulator. This enables a system designer to simulate a battery, together with
its surrounding BMS. However, the electronic-network models above have some
drawbacks. Therefore, a new approach will be presented in this chapter.
An electrochemical/physical model for rechargeable batteries will be derived
first in the approach of battery modelling adopted in this book. This model describes
the thermodynamics, charge-transfer kinetics and mass transport limitations of the
various reactions. It also describes the physical processes that occur during
(over)charging, resting and (over)discharging. All reactions and processes will be
described in the form of mathematical equations. Then, the mathematical equations
will be clustered for each process and each cluster will be represented by linear
and/or non-linear equivalent network elements. These network elements are
combined in a single network.
On the basis of the principles of physical system dynamics [1], three domains
are distinguished, notably the electrical, chemical and thermal domains. These
domains are coupled, which enables the transfer of energy from one domain to
another. The electronic-network elements, electrical resistance R,1 in [Q] and
electrical capacitance C,1 in [F], form the electrical domain, where voltage in [V]
and current in [A] determine the battery behaviour. The (electro)chemical-network
elements, chemical resistance Rch in [Jslmol2] and chemical capacitance Cch in
[moe/J], form the chemical domain, where battery behaviour is based on chemical
potential in [J/mol] and chemical flow in [moVs]. Key factors in this domain are, for
example, diffusion profiles inside an electrode and the amount of oxygen present
inside an aqueous battery. This amount can evidently be translated into a partial
oxygen gas pressure. Finally, the thermal-network elements, thermal resistance R1h
in [KIW] and thermal capacitance C1h in [J/K], form the thermal domain, where
battery behaviour is dictated by temperature in [K] and heat flow in [W]. The battery
temperature development is determined in this domain in conjunction with the
electrical and chemical domains. A similar approach has also been successfully
applied to model transformers, taking into account the electrical and magnetic
domains [13].
Battery modelling 57
The advantage of the modelling approach adopted in this book is the possibility of
simulating a battery in its electronic and thermal environment. The battery voltage,
temperature, internal gas pressure, the state of all the charge-transfer reactions and
the mutual interaction between all these variables are taken into account. This as
opposed to the electronic-network models found in the literature mentioned above,
which only describe electrical battery behaviour. The origin of specific battery
behaviour can be examined for example by studying the rates of the various battery
reactions, because the models developed in this book rely on electrochemical and
physical battery theory. In simple terms, the models act as 'transparent batteries',
enabling us to view the course of all relevant processes that take place under various
user conditions. An advantage is that all the parameters in the models have an
electrochemical meaning and can therefore, in principle, be determined in
experiments, obtained from the literature or in some cases estimated within
reasonable limits. This section deals with the general principles applied in modelling
batteries on the basis of the principles of physical system dynamics. The basic
building blocks that can be used to build the equivalent network of a complete
battery are described [14]. Examples in which these building blocks will be used to
model specific battery systems will be given in later sections for NiCd and Li-ion
batteries.
The basic principle of physical system dynamics is that several domains can be
distinguished in physics, in which an analogy exists between the definitions of
energy and power. An effort and a flow variable can be defined in each domain, the
product of these variables expressing power in [W]. Energy can be stored in each
domain or exchanged with other domains. The analogy between energy and power
definitions in the three relevant domains in batteries, the electrical, chemical and
thermal domains, is clarified in Table 4.1. The effort variables of these domains are
given in the first row, the flow variables in the second row. The power variables are
shown in the third row. The displacement variables in the fourth row of Table 4.1
are defined as the integral of the flow variables over time. The energy, capacitance
and resistance variables are listed in the fifth, sixth and seventh row, respectively.
Table 4.1: Analogy between the electrical, chemical and thermal domains
Electrical domain Chemical domain
Quantity Symbol Unit Quantity Symbol Unit
Voltage v v Chemical potential IL J/mol
Current I A Chemical flow Jcb moVs
Electrical power Pe1=Vl w Chemical power p cb=jJ.Jcb w
Charge Q=It c Molar amount m=Jcht mol
Electrical energy Ee1=Pe1t J Chemical energy Ecb=Pcbt J
Electrical capacitance Cel=QN F Chemical capacitance Ccb=m/IJ. mol2/J
Electrical resistance Rei=VII Q Chemical resistance Rch=jJ./Jch Jstmoe
Thermal domain
Quantity Symbol Unit
Temperature T K
Heat flow Jth w
"Thermal power" Plh=TJ!h WK
Heat Qth=Jtht J
"Thermal energy" ~=Ptht JK
Thermal capacitance 4=Q~ J/K
Heat resistance Rth=T/Jth KIW
58 Chapter4
Note that with the choice of heat as the displacement variable in the thermal domain,
the terms "thermal power" and "thermal energy" do not have any physical
meanings. The product of effort and flow indeed has the unit [W], as in the other
domains, when the entropy S is used as the displacement variable and hence the flow
variable becomes the entropy flow rate dS!dt. However, it is common practice to use
heat as the displacement variable, as is shown in Table 4.1, because it is generally
not practical to use entropy as the displacement variable. The analogy with other
domains is in this case more mathematical than physical [1].
4.1.1 Chemical and electrochemical potential

The chemical potential J.l; of an uncharged species in [J/mol] is defmed as
[15],[16]:
.u, = .u; + RTln( a~.:r ) (4.1)
where 11-f denotes the standard chemical potential of species i in [J/mol], R is the gas
constant (8.314 J/(mol.K)), T stands for temperature in [K], a; denotes the activity
of species i in [mol/m3 ] and a{~' is the activity of species i in the reference state in
[rnollm3].
A similar definition holds for the chemical potential of a charged ionic species
i. Then the term electrochemical potential is used, with symbol Ji; . An extra term
that involves the electrostatic or Galvani potential ,P [16], valid in the place where
the ionic species i is present, is added to the normal chemical potential definition in
this case [15],[16].
(4.2)
where Z; denotes the valence of ionic species i. For example, z = +n for a positive
ion An+ and z = -m for a negative ion B"'·. F is Faraday's constant (96485 C/mol)
and ,Pis the electrostatic potential in [V].
The activity of a species is linearly proportional to its concentration, and hence
to its molar amount, in a certain volume; see chapter 3. On the basis of the analogy
of Table 4.1, both the chemical potential of an uncharged species i and the
electrochemical potential of a charged ionic species i can be modelled by means of a
chemical capacitance, where the 'charge' on the capacitance corresponds to the
molar amount m; of the species and the 'voltage' across the capacitance amounts to
the species' chemical potential Jl.i. The potential of the bottom plate of the
capacitance with respect to grouod equals the term z;F,P in (4.2) in the case of a
charged species. The chemical capacitance models the storage of energy in the
chemical domain for both uncharged and charged species. The chemical
capacitances that store uncharged and charged species are shown in Figures 4.1(a)
and (b), respectively.
+
+ +
1-li m; j:i;
Z;;~iY
(a) (b)
Figure 4.1: Chemical capacitance representing (a) chemical potential and storage of
uncharged species i and (b) electrochemical potential and storage for charged ionic species i
An expression for the inverse of both capacitances in Figure 4.1 can be inferred
from the derivative of the chemical potential with respect to the molar amount, i.e.
1 _ dp _ d [
cch(m)- dm- dm fl
o
+
RT l ( a
n aref
)J
(4.3)
=d- ( p o +RTln-
dm
( m
mnf
)J =RT
m
-
As can be inferred from (4.3), the capacitances in Figure 4.1 are non-linear and have
the unit [mol 2/J], which is in accordance with Table 4.1.
4.1.2 Modelling chemical and electrochemical reactions

In addition to the electrochemical reactions involving the exchange of electrons
between electroactive species described in chapter 3, chemical reactions may also
take place inside a battery. These reactions involve uncharged species only.
Consider, for example, the following chemical reaction:
kr
A+B• )C+D (4.4)
kb
where k1 denotes the forward and kb the backward reaction rate constant,
respectively. The unit of both constants depends on the details of the reaction
equation.
The reactions in both directions take place at the same rate in a state of
equilibrium and the net molar amounts of species A, B, C and D remain unchanged.
The sum of the chemical potentials of species A and B equals the sum of the
chemical potentials of species C and D in this case. However, one reaction takes
place at a higher rate than the other in a non-equilibrium situation and the net result
is a chemical flow from the two species with the highest total chemical potential to
the species with the lowest total chemical potential. The chemical flow for the
chemical reaction in (4.4) in non-equilibrium situations is given by [15]:
60 Chapter4
(4.5)
The chemical reaction of (4.4) can be represented by a network model by combining

four chemical capacitances, described by (4.3), with a chemical resistance. The
'charges' on the chemical capacitances represent the molar amounts m; of the four
species taking part in the reaction and the chemical resistance represents the
chemical flow lch according to (4.5). This is shown in Figure 4.2.
IJ.c
J.le J.lo
Figure 4.2: Simulation model in the chemical domain of a chemical reaction according to
(4.4)
The chemical resistance is defined by pJ!lch• where f.iR is the chemical potential
across the resistance and lch is the chemical flow through the resistance, see Table
4.1. Therefore, the chemical resistance in Figure 4.2 depends on k1, kb and the molar
amounts of all the reacting species, which makes it non-linear. The chemical flow
J,h is directed from left to right in Figure 4.2 when the sum of the chemical
potentials of species A and B is higher than the sum of the chemical potentials of
species C and D. The molar amounts of species A and B will decrease in this case,
whereas the molar amounts of species C and D will increase. The chemical flow lch
is directed from right to left in the opposite situation. The relation between k1 and kb
can be inferred from the state of equilibrium, when the activities a; of all the species
are equal to their equilibrium or bulk activities ab and the chemical flow J,h is zero.
Hence,
(4.6)
Moreover, the following equation holds for the chemical potentials in equilibrium:
(4.7)
bb) -RTin(refref)
=RTin( acaD ac aD
ab ab a ref a ref
A B A B
and hence:
(4.8)
The ratio kfkb is constant at a constant temperature. This means that the ratio of the
activities of C and D and A and B, as shown in (4.6), remains constant when the
reaction is in equilibrium and the temperature remains constant.
A chemical reaction will take place in the chemical domain only. As described
in chapter 3, electrochemical reactions enable the conversion of electrical energy
into chemical energy and vice versa. Hence, interaction between the electrical and
chemical domains is realized through an electrochemical reaction. Therefore, the
electrical and chemical domains must be coupled in a network model of an
electrochemical reaction.
Consider the following simple electrochemical reaction that takes place at the
surface of a well-conducting electrode, immersed in an electrolyte solution:
k,
OX'+ + ne· .: ~ Red (4.9)
k.
Consider, for example, that the Ox species is present in the electrolyte, while the Red
species is present in the electrode. An example is an iron electrode (Fe) that is
immersed in an electrolyte containing Fe2+ ions. Consider a net reduction reaction
(R), with the reaction directed to the right prevailing. Electrons are consumed by the
reaction in this case, which results in a cathodic electrode current. Ions OX'+ from
the electrolyte transform into Red atoms in the electrode by taking up n electrons
from the electrode. Adopting the sign convention proposed by Vetter [16], the
change in free energy LlGR for the reduction reaction R equals -nFE'q in this case,
with E'q being the equilibrium potential of the electrode. By definition, the value of
LlG can be inferred from EV;f-li for both the reduction reaction R and the oxidation
reaction 0, with V; being the stoichiometric factors for species i in the reaction
equation [16]. By definition, these factors V; are positive for reaction products and
negative for reactants. The Red species is the reaction product and OX'+ and
electrons are the reactants in the case of a reduction reaction. LlG depends only on
chemical potentials and not on electrochemical potentials.
It also holds that "f. V JI; =0 for any electrochemical reaction in a state of
equilibrium [16]. In the literature, the summation "f. V ;71; is defined as f).G , which
is the change in free electrochemical energy [15],[16]. The electrochemical
potentials and hence also the electrostatic potentials are taken into account in this
term. In a state of equilibrium, ka=kc and an expression for the equilibrium potential
can be found by considering either the reduction reaction or the oxidation reaction.
In the case of the change in free electrochemical energy !l.GR (in [J/mol]) for the
reduction reaction of (4.9) this yields:
62 Chapter 4
!:J.G R = JiRed - (Jlox + nJiJ

= f-lRed -{pox + ZoxFC/Jox + nf.le + nzeFC/Je) (4.10a)
=f.lRed -{pox + nFf/Jt + nJ.le - nFf/Js )= 0

where ¢f=tPe is the electrostatic potential of the electrode in [V], i.e. the electrostatic
potential valid for the electrons, ¢1=1Pox is the electrostatic potential of the electrolyte
in [V], i.e. the electrostatic potential valid for the OX'+ ions, Zox= +n is the valence of
the Oxn+ ions and Ze= -1 is the valence of the electrons. The change in free energy
LlGR (in [J/mol]) can now be inferred from:
!:J.GR = f-lRed - f.lox - nf.le
= r'Red + RTln(aRed)
llo
ref
)
QRed
1{ =~
(4.10b)
+~· + RT + np; + nRTin( :~ )J
= 1/lRed
f .. o _ r'Ox
o -nr'eo)+ RTln(aReda;;t)
11 11
ref
QRedaOx
= nF(f/Jt - f/Js)
where (4.10a) was used in the last step. The activity of electrons has been assumed
equal to a/'1 in (4.10b ), because the activity of electrons does not change noticeably
in an electrode with a very good electronic conductivity [15]. An expression gq can
now be found from (4.10b):
(4.10c)
(4.1 Oc) shows that the electrode is indeed in equilibrium, because the electrode
potential Eelectrode equals the equilibrium potential E'q. The expression for gq is in
agreement with Nemst's equation given in chapter 3. In addition, it can be inferred
from (4.1 Oc) that:
(4.11)
The signs of LlG and D.G reverse for the oxidation reaction of (4.9), in which case
the reaction directed to the left prevails, which means that the changes in free energy
and free electrochemical energy become /'iG 0 =flax+ nf1,- fl&d = -t::.GR and
/'iG0 = liox + nJie - JiRed = --t::.GR. In that case, LIG0 equals nFE'q, because the sign
of E'q is fixed. Hence, the same expression for E'q in (4.10c) is found when the
oxidation reaction is considered. The sign convention of LIG may lead to confusion
when reading the literature. The sign convention proposed by Vetter and already
used above will be adopted throughout this book [16].
The chemical flow associated with the electrochemical reaction in (4.9) is
given by [15]:
(4.12)
where A denotes the electrode surface area in [m2 ] and ka and kc denote the reaction
rate constants for the anodic (0) and cathodic (R) reactions, respectively. Their unit
will depend on the reaction equation. a denotes the transfer coefficient, which is
dimensionless, and 0<a<1; x andy reflect the reaction orders. It will be shown later
in this section that (4.12) is in fact the Butler-Volrner equation, expressed in the
chemical domain. Note that when kc>ka. the absolute value of t::.GR is larger than
that of /'iG0 . When ka>ko the reverse situation occurs and the direction of lch
reverses. Neither the change in free electrochemical energy t::.G nor the chemical
flow lch can be measured directly. However, they can be derived from the electrode
potential and current, respectively.
The coupling between the electrical and chemical domains can be modelled by
means of an ideal transformer that couples the electrical effort variable voltage V to
the chemical effort variable chemical potential J.l and the electrical flow variable
current I to the chemical flow variable chemical flow lch; see Table 4.1. The
transformer coefficient is 1/nF, where n is the number of electrons involved in the
electrochemical reaction that couples the two domains.
V=L
nF (4.13)
I= nFJch
The transformer is assumed to be ideal, which means that no energy is stored in the
transformer. This means that Vl=f.llch· The ideal transformer is a purely
mathematical component used to link different physical domains [1]. The network
model for an electrochemical reaction can be found using the ideal transformer. This
is shown in Figure 4.3. The direction of the current /electrode and chemical flow fch• as
well as the sign of r/' and the relation between LIG0 and E'q are shown for ka>kc in
(4.9). This means that the oxidation reaction is assumed to prevail.
The electrical and chemical domains separated by an ideal transformer can be
clearly recognized in Figure 4.3. The coupling between the equilibrium potential E'q
64 Chapter4
in the electrical domain and the change in free energy .1G0 in the chemical domain
through the transformer coefficient 1/nF is indicated in the figure. Moreover, the
coupling between the electrode current I and the chemical flow lch is indicated. The
direction of letectrotk and chemical flow lch and the signs of <1G and r/' are reversed
for the reduction reaction.
The electrode and equilibrium potentials are denoted by the symbol E in the
electrical domain, the kinetic overpotential by the symbol r/' and the electrostatic
potentials of the electrode and electrolyte by ¢1 and ¢1, respectively. The
electrode/electrolyte interface is represented by a linear capacitance CO, which
denotes the double-layer capacitance. The electrode potential Eetectrotk=¢1-¢1 occurs
across this double-layer capacitance, as described in chapter 3. The anti-parallel
diodes 0 and R model the exponential relation between the electrode current I and
the kinetic overpotential r/' according to the Buder-Volmer equation, while the
electrical port of the ideal transformer models the equilibrium electrode potential E"q
as defined in (4.10c).
Eelectrode
-
Electrical domain
Ielectrode
+ TJk -
0
+
• -
1--------'--.
Eeq =~Gd(nF)
=
-
l=nFJch
N=1/(nF)
-
A ,....-------!
.+~Go-
Chemical domain Jch = 1/(nF)

nz8 Fcj>9 ZoxF«i>ox
+I_ + +II __
+ I+
nile JlRed
Figure 4.3: Simulation model in the electrical and chemical domains for an electrochemical
reaction according to (4.9). The definition of E'q, the signs of LIGand r/' and the direction of
the current and chemical flow hold for an oxidation reaction (0)
Each capacitance in the series connection in the chemical domain corresponds to a

species that takes part in the charge-transfer reaction; see (4.9). The polarity of the
capacitances depends on whether the species appear on the left side or the right side
of (4.9) and the chemical potential of the electrons is multiplied by n, in accordance
with (4.10a). The chemical flow lch is given by (4.12). In accordance with Figure
4.1, the potential at the bottom plate of the capacitance that stores the molar amount
of electrons is nz.FlPe and the potential at the bottom plate of the capacitance that
stores the molar amount of oxidized species is ZoxFf/Jox· From Figure 4.3, the change
in free energy LlG0 at the chemical port of the ideal transformer is inferred to be:
fl.Go = -fl.GR = nf-l. + 1-lox - 1-lRed

(4.14)
=> E"q = fl.Go = - fl.GR
nF nF
This means that the network shown in Figure 4.3 is in accordance with the sign
methodology discussed earlier for both the oxidation and the reduction reaction. An
expression for r/' in terms of I:!.G0 can be inferred from Figure 4.3:
k _ E Eeq _ r~t• r~tl fl.G 0

'17 - electrode - - 'I' - 'I' - nF
= ¢• - f/Jl - n~ {tl.Go - nF(f/Jt - f/Js ))

(4.15)
- fl.Go
nF
In a similar way, it can be found for the reduction reaction that 'f/k = I:!.GR .
nF
(4.12) can be easily transferred to the electrical domain by realizing that
l=nFlch and using (4.15). The resulting equation links I and r/' and is in accordance
with the Butler-Volmer equation defined in chapter 3. The analogy between the
various parameters in the electrical and chemical domains is summarized in Table
4.2.
Instead of modelling an electrochemical reaction in the chemical and electrical
domains, it is also possible to represent all the elements of the model in the electrical
domain only. This can be accomplished by transferring the model equations from
the chemical domain to the electrical domain. An example was given above for
(4.12). Chemical capacitances can be represented by electrical capacitances with a
voltage f.l/nF across them and an electrical charge q=nFm on them. The following
equation is then obtained for the voltage across each capacitance:
V =_.!:!:.._ =_.!:!:.._o + RT In(_!!_)

nF nF nF a'ef
= yo + RT In(_!!!_) (4.16)
nF mre1
=V o +RT-I n
nF
( q-
q'ef
J
where r has again been assumed to be unity.
66 Chapter4
Table 4.2: Analogy between variables in the electrical and chemical domains shown in Figure
4.3
Electrical domain Chemical domain Relation
Description Symbol Unit Description Symbol' Unit
Current I A Chemical lch moVs l=nFJch
flow
Equili- gq v Change in J/mol

flGi=
brium free energy E"q = 11Go
potential Iv,,u,
i nF
- -11GR
nF
Standard E' v Change in AG;= J/mol
redox standard Eo= 11G~
Potential free energy Iv,,u;
i nF
-11GR
0
--
nF
Kinetic r/' v Change in AGj = J/mol
k -11G0
over- free
potential electro- Iv,,U,
i
rJ =
nF
chemical
energy 11GR
=--
..
nF
Note 1: J =0 orR, 1 ts spectes m reacuon equation
The same result as that obtained in transferring equations from the chemical to the
electrical domain is obtained when using the simulation model presented in Figure
4.3 when all circuit elements in the chemical domain are "shifted" through the ideal
transformer to the electrical domain. Series and parallel connections of elements in
the chemical domain remain series and parallel connections in the electrical domain,
respectively. The impedances of circuit elements in the chemical domain have to be
multiplied by !1, with N=llnF, as can be understood by dividing the top by the
bottom equation in (4.13). Hence, it can be found for the electrical capacitances,
using (4.3) and q=nFm:
(4.17)
where Cch(m) is inferred from (4.3). (4.17) can also be found by calculating dV!dq
from (4.16). The representation of the simulation model of Figure 4.3 in the
electrical domain only is shown in Figure 4.4.
Eelectrode
-
Ielectrode -- 0
A
I
IJ.o/nF
+
1------illl-----f
llReJnF
+
Figure 4.4: Simulation model of Figure 4.3 in the electrical domain only
Figure 4.4 illustrates that the transformer has now vanished from the model.
4.1.3 Modelling mass transport

Mass transport to and from the electrode surfaces inside a battery can influence the
rate at which the electrochemical reactions take place, as was described in chapter 3.
In electrochemistry, it is quite common to model mass transport of species by means
of electronic-network models. For example, the low-frequency behaviour of the
battery impedance is described by the Warburg impedance in most textbooks that
deal with basic electrochemistry [17],[18]. As early as 1900, Warburg derived an
equation for the impedance of diffusion-controlled processes at a planar electrode.
The Warburg impedance is basically an RC-ladder network or transmission line,
which describes the diffusion of species to and from the electrode surface. An
expression for the Warburg impedance can be found under certain limiting
conditions, using expressions for the non-linear capacitances and resistances in the
RC-ladder network [19].
An example of an approach to modelling ionic mass transport under the
influence of joint diffusion and migration using equivalent network models can be
found in the literature [20]. However, this approach is not based on the principles of
physical system dynamics. Therefore, a different approach is presented in this
section, which is based on the theory of physical system dynamics. It will be shown
that the expressions derived for the resistances and capacitances in the RC-ladder
network, which describes the one-dimensional diffusion of species, yield the same
RC time as the RC time calculated in [20]. The approach adopted in the derivation is
the same as that employed in arriving at a numerical solution to the problem
Diffusion in one dimension is considered and the system is divided into spatial
elements of the thickness Lix. Unlike in efforts aimed at arriving at a mathematical
numerical solution to the problem, it is not necessary to use discrete time steps
because time discretization has already been performed by a circuit simulator during
the simulation of the network model. The division into spatial elements of a
chemical system in which mass transport takes place in the x-direction only is
clarified in Figure 4.5.
68 Chapter 4
x::O X: I
Figure 4.5: Division of a chemical system into p spatial elements with a thickness L1x and a
volume AL1x between x=O and x=l
The activity a4 of the species in element 4 is shown as an example in Figure 4.5,

where the activity coefficient y has again been assumed to be unity for simplicity.
Mass transport law

The driving force behind the movement of a species is a gradient in the
electrochemical potential. Mass transport by convection will not be considered in
this book. The direction of the resulting chemical flow is from a region in which the
electrochemical potential is high to a region in which the electrochemical potential is
low. For an uncharged species, an expression for the chemical flow Jch(x) at position
x can be inferred from [15]:
1 (x) =_ ADc(x) dji = ADc(x) dfl

ch RT dx RT dx
=_ ADc(x)
RT
() (flo + RT
dX
ln( c(x) ))
cref
(4.18)
= _ ADc(x) ( RT dc(x)) =-AD dc(x)
RT c(x) dx dx
- !:lfl
=
where Jch(x) denotes the chemical flow of the transported species at position x in
[molls], as defined in Table 4.1, Rclx) is the associated chemical resistance at
position x in [Js/rnol 2 ], see Table 4.1, c(x) is the concentration of the uncharged
species at location x in [rnoVm3], c''1 is the reference concentration in [moVm3] and
D denotes the diffusion coefficient of the transported species in [m2/s]. (4.18) shows
that a chemical potential difference iJ.fJ, across a chemical resistance Rch yields a
chemical flow lch· This is in fact Fick's first law, which was introduced in chapter 3.
For a charged species, (4.18) becomes:
jch(x)=
ADc(x) (Jji ADc(x)( RT dc(x) + zF df/J(x))
RT dx RT c(x) dx dx (4.19)
=-AD dc(x) _ zF ADc(x) d¢J(x) = -!:J.ji
dx RT dx Rch (x)
(4.19) is known as the Nemst-Planck equation [15]. It expresses the mass transport
of a charged species by combined diffusion and migration. Using (4.2), the term
!:l.Jl can be rewritten as:
!:l.Jl = dJl. !:J.x = (RT-l- da(x) + zF d¢J(x) )!:J.x (4.20)

dx a(x) dx dx
A similar expression can be found for L1f-l using (4.1 ). In that case, the term that
contains rp is zero. Using (4.20) for !:l.Jl or a similar equation for L1f-l, both (4.18) and
(4.19) yield the same expression for Reix) at position x. It should be realized that
chemical resistance is defined as a positive variable, which means that a positive
chemical flow is directed from the terminal with the highest electrochemical
potential to the terminal with the lowest electrochemical potential. Suppose a species
flows from x1 to x2o where x2 >x1 and Jl1 > J12 • The gradient in electrochemical
potential from x 1 to x2 is negative, whereas the associated chemical flow is positive.
By definition, L1f-l and !:l.Jl are negative in this case.
- !:l.Ji = Rch(x)Jch(x)
= -R (x)(AD dc(x) + zF ADc(x) d¢J(x))
ch RTox ox (4.21a)
= -(Rr _1_ da(x) + zF d¢J(x) )!:J.x

a(x) dx dx
in which (4.19) was used in the second line with consideration of the sign and
(4.20) was used in the third line. The activity a will now be introduced in the second
line, with again r=1, while the third line will be re-arranged such that a similar term
appears as in the second line. Realizing that the second line equals the third line in
(4.21a), the following expression is then obtained:
70 Chapter4
R (x)(AD aa(x) + zF ADa(x) a¢(x))

ch ax RT ax
(4.21b)
=(AD aa(x) + zF ADa(x) a¢(x)) RTI:lx
ax RT ax ADa(x)
Now, an expression for Rch(x) can be inferred from (4.2lb):
R h (x) = RTI:lx = RTI:lx = RTI:lx 2

(4.21c)
c ADa(x) AD(:~) m(x)D
Mass transport at position x through a chemical resistance Rchfx) described by

(4.21c) is indicated in Figure 4.6.
Figure 4.6: Representation of mass transport at position x by a chemical resistance Rch(x),

using the analogy of Table 4.1
Conservation of mass
The conservation of mass is expressed by the following equation:
dm =/::J (4.22)
dt ch
In Figure 4.5, conservation of mass means that the change in time of the molar
amount mi in a spatial element i, between locations ( i-1 )Ax and iL1x, is determined by
the net chemical flow &c/ This flow is the difference between the chemical flow at
location x=(i-l)L1x and that at location x=iL1x. Using the analogy between the
electrical and chemical domain of Table 4.1, the conservation of mass can be
described by a chemical capacitance Cchi for each spatial element i, as shown in
Figure 4.1, where the amount mi denotes the molar amount in spatial element i. This
is visualized in Figure 4.7. An expression for the chemical capacitance Cchi was
derived in (4.3).
Initial and boundary conditions

Initial and boundary conditions have to be defined for each mass transport problem
to enable its solution. For example, an initial condition could be that the activity, and
hence the molar amount under the assumption that r= 1, of the transported species is
uniformly distributed throughout the system and is the same for all x at t=O.
m;
Figure 4.7: Representation of mass conservation law in spatial element i (see Figure 4.5) by
means of a chemical capacitance
Two surface boundaries occur for the chemical flow in a chemical system in which
one-dimensional mass transport is considered, for example at x=O and x=l in Figure
4.5. The transported species can either be able to cross the boundary to a
neighbouring system or be restricted to the chemical system through which it is
transported. In the first case, the chemical flow across the boundary usually will
depend on the concentration gradient across the boundary, whereas in the second
case the flow across the boundary will be zero.
Mass transport in the absence of an electrical field

Transport by migration will not occur when no electrical field is present. Pick's law
can then be used for mass transport; see (4.18). Combining Pick's law, the condition
of mass conservation of (4.22) and initial and boundary conditions, one-dimensional
mass transport of a species in the x-direction in the absence of an electrical field can
now be described with the aid of an RC-ladder network. This is shown in Figure 4.8
for the considered chemical system of Figure 4.5.
RTruc2
Rch ((p- 1)~x) =m((p--1)ruc)D
x=O X = (p--1) rue x=p~x=l
\
Jch ~ 1)rue)
spatial spatial spatial spatial

element 1 element2 element p--1 element p
Figure 4.8: Simulation model in the chemical domain describing one-dimensional mass
transport of a species in the x-direction in the absence of an electrical field
72 Chapter4
The molar amounts m; in spatial elements i are present in chemical capacitances Cch;
in Figure 4.8. A chemical potential, described by (4.1), is present across each
chemical capacitance. A difference in chemical potentials between two neighbouring
spatial elements will give rise to the transport of the species by pure diffusion from
one chemical capacitance to its neighbour through a chemical resistance Rch· With
Rch• defined in (4.21c), and Cch, defined in (4.3), an RC time of L1x2/D is found for
each spatial element. This is indeed in agreement with the derivations of Homo et
al., where R=Lh!D and C=k. [20].
As an example, the boundary conditions have been taken so that the species is
injected from a neighbouring system in the form of a chemical flow lch(O) at the
surface at x=O. The species is not allowed to cross the boundary at x=l. The
system's initial condition can be enforced by the initial values of m; at t=O.
Consider again the electrochemical reaction given by (4.9). When both the
oxidized and the reduced species can diffuse through the system and no electrical
field is present, the network model of Figure 4.3 changes into the network model of
Figure 4.9. As in Figure 4.3, the directions of currents and flows as well as the
polarities of L1G and r/ are shown for an oxidation reaction. The directions and
polarities reverse for a reduction reaction. An RC-ladder network, as shown in
Figure 4.8, has been added in the chemical domain in Figure 4.9. The boundary at
x=O is the electrode/electrolyte interface at which the electrochemical reaction takes
place, while the boundary at x=l is the electrode/current-collector interface. The
molar amounts on the capacitances in the RC-ladder network represent the molar
amounts of the oxidized and reduced species throughout the electrode, respectively.
Spatial element 1 starts at x=O, spatial element p ends at x=l. A difference between
the molar amounts in spatial element I and those inside the electrode will give rise
to mass transport by diffusion through the electrode. The chemical resistances Rch(x)
are given by (4.21c), in which Dox is used for the oxidized species and DRed for the
reduced species. Neither the oxidized nor reduced species can cross the boundary at
x=l. At the boundary at x=O, the species are allowed to cross the boundary by means
of the electrochemical reaction. The bulk activity of electrons is assumed to be equal
to the surface activity. Therefore, no ladder network is used for electrons.
Figure 4.9 shows that the chemical potentials across the capacitances that
represent spatial element 1 at x=O are denoted as surface chemical potentials J.i. As
a result, the potential across the transformer winding in the electrical domain is now
obtained from the following equation. Again, the activity of electrons is assumed to
be equal to the reference activity a/•1 as in (4.10b). In addition, the chemical
potential of electrons is assumed uniform throughout the electrode, which means
that f.l:=f.le·
(4.23)
The equilibrium potential now depends on the surface activities a• of the oxidized
and reduced species, according to (4.23). This potential can be denoted as the
apparent equilibrium potential Eq*. The difference between Eq* and the true
equilibrium potential Eq, which was defined in (4.1 Oc ), is the fact that Eq is related
to the bulk activities ab of the reacting species. Bulk activities are the same as
surface activities in a state of true equilibrium. The difference between Eq* and Eq
is the diffusion overpotential rf, which can be found by rewriting (4.23).
(4.24)
The expression for the diffusion potential is in agreement with the general
expression for a diffusion overpotential given in chapter 3; see (3.17). Eq equals Eq*
and rf is zero, when ab equals a• for the Ox and Red species.
It is not possible to discern between Eq and rf when the simulation model is
shown in the form of Figure 4.9, because both variables are present in Eq*. It is
sometimes convenient to make this distinction. It becomes possible to distinguish
between Eq and rf when the definition of the chemical capacitances is altered. This
is shown in Figure 4.10, in which the model has also been represented in the
electrical domain as additional illustration. As was discussed in section 4.1.2, the
representation of the complete network in the electrical domain can be achieved by
"shifting" the elements in the chemical domain through the ideal transformer to the
electrical domain.
The electrical capacitance Cbutk represents the series connection of three
capacitances which store the bulk molar amounts of all species that take part in the
reaction. Only one capacitance has been drawn for simplicity. The voltage across
this capacitance equals Eq, as defined in (4.10c). By definition, the voltage across
each capacitance in the RC-ladder network is proportional to the difference between
the chemical potential in the applicable spatial element and the bulk chemical
potential ;1. This bulk chemical potential is the same for all spatial elements. The
chemical potential will be ;1 in all the spatial elements at true equilibrium. As a
result of this definition, the capacitances describe the amount of material in spatial
element i relative to the bulk amount of material. The following equation illustrates
this for the voltage across the surface capacitances Cj, where j can be Ox or Red.
1 • b RT RT
V. =-(p,. -p,.)=-ln-=-ln-
a; m; (4.25)
1 nF 1 1 nF ab nF mb
J J
74 Chapter4
Eelectrode
Electrical domain
-
Ielectrode
+ TJk
0
• --
+ Eeq• = dGd(nF) -
I= nFJch
·-
1---------,
N = 1/(nF)
Chemical domain
Jch = 1/(nF)
X=O
nile J.l1. Ox= ll8x J.l1. Red = J.l~ed
Spatial element 1
- + ~+ 1----1
- - - 1 _11-----r--'+
X = dx - - + f\h, Ox (dx) f\h, Red (dx)

J.l2,ox J.l2,Red
Spatial element 2 : + 1-1_----_-il + :
l J.lp, Ox J.lp, Red l
Spatial element p - - - - -
X=l
~~- -'~
C _ mp,Red
ch- AT
Figure 4.9: Simulation model for an electrochemical reaction according to (4.9) in the
electrical and chemical domains, where mass transport of the Ox and Red species takes place
in the electrode in the absence of an electrical field. The definition of £"'1, the signs of L1G and
r/' and the directions of the current and chemical flow hold for an oxidation reaction
It can be easily verified that the diffusion overpotential defined by (4.24) is present
across these two surface capacitances:
s b
1'/d = !-lox - llox (4.26)
nF
For the Ox and Red species, f.l=Ji' in all the spatial elements in a state of
equilibrium. Therefore, the voltage across all capacitances in the RC-ladder network
is zero and rf is zero in a state of equilibrium. The value of each resistance Rch has
been multiplied by fi=( 1/nFl because of the transition from the chemical to the
electrical domain. Likewise, the value of each capacitance Cch has been divided by
N2 in the electrical domain.
Eelectrode
_ _ _ TJd _ __
+
X=O
-
Ielectrode
Spatial element 1 -
/
~x-~x
nF
X= ill<
Spatial element 2 - - - - •
: b b :
! l!p, Ox - J.I.Ox llp, Red - !!Red !
: nF nF :
Spatial element p - - - - • ~r-1----. . . ,p
Cal=
X= I
Figure 4.10: Adapted simulation model of Figure 4.9 with altered definitions of the
capacitances to allow distinction between the true equilibrium potential gq and the diffusion
overpotential if. The model is represented entirely in the electrical domain
Mass transport in the presence of an electrical field

Mass transport of a charged species will take place by combined diffusion and
migration when an electrical field is present. An expression for the electrical field is
needed to describe this mass transport, which can be found by solving the Poisson
equation. Assume that the transported species are positive and negative ions with
equal and opposite charges, for example ions of a fully dissociated binary salt
A"+B"-. In a spatial element with volume ALlx, the Poisson equation reads [15]:
(4.27)
where &0 is the electrical permittivity in free space (8.85.10- 12 C2/(N.m2)), e, is the
dielectric constant of the medium through which the ions are transported, Z; denotes
the valence of the transported species, z=+n for the A"+ ions and z=-n for the 13"-
ions. m+ and m. denote the molar amounts of An+ and B"- ions in volume ALlx,
respectively.
The basic building block for modelling the Poisson equation for a fully
dissociated binary salt An+B"- is shown in Figure 4.11 for both the chemical and the
electrical domain. The model describes a spatial element with a volume ALlx, as
introduced in Figure 4.5, and its connection to the neighbouring spatial elements. It
76 Chapter4
consists of four capacitances connected together in a central node point. The

capacitances ce to the left and right of the central-node point denote geometric
parallel-plate capacitances with a surface area A and a thickness Llx. The electrostatic
potential f/Jc is valid at the centre of the spatial element, whereas f/Jz is valid at the
centre of the neighbouring spatial element to the left and f/J, is valid at the centre of
the neighbouring spatial element to the right. The geometric capacitances denote the
capacitances between the centre points of the spatial elements, because the distance
between the centres of the spatial elements is Llx.
C' _ E0 E,A
ch- ax(nF} 2
(a) (b)
Figure 4.11: Simulation model of the Poisson equation in (a) the chemical domain and (b) the
electrical domain
The ratio of the capacitances in the chemical and electrical domains is N2 , as

discussed before. The 'charge' on the capacitance above the central-node point in
Figure 4.11a denotes the molar amount of negative ions B"-, while the 'charge' on
the capacitance below the central-node point represents the molar amount of positive
ions An+_ The unit of these capacitances is [mof/J], whereas it is [F] in Figure 4.1lb.
It can be seen that the networks illustrated in Figure 4.11 indeed describe the
Poisson equation of (4.27). The mass conservation law of (4.22) holds for the
central-node point for the network in the chemical domain. Hence, it can be found
from Figure 4.11 a that:
(4.28)
This is in fact the equivalent of Kirchhoff's current law in the electrical domain.
(4.28) can be re-arranged, where the equation for Cch E in Figure 4.11 a is used in the
second line of the following equation:
(- nF¢J1+ 2nF¢Jc- nFf/J, 'f;;h = m A"• - m 8.-

=} (-f/Jt +2¢Jc -¢J,}tF Eo~E,AY =rnA•• -mn·- (4.29)
&nF
nF&
::::} (-f/Jt + 2¢Jc - f/J,) = - - (m A"• - mn•-)
e0 e,A
The Poisson equation according to (4.27) can be found from (4.29), with the aid of a
basic mathematical equation for the second derivative of the electrostatic potential f/J
in the centre of the spatial element:
(4.30)
The Poisson equation in the electrical domain can be found in an analogous way.
Now that the basic building block for modelling the Poisson equation has been
defined, the mass transport of charged species by combined diffusion and migration
under the influence of an electrical field can be modelled. As an example, consider
an electrochemical cell with two metal electrodes and an ideal solution of a binary
salt M'+X'· that acts as electrolyte. Assume that only the metal ions M'+ can react
with metal M at the electrodes and that the X'. ions do not react with any species.
M'+ ions are injected into the ideal solution by an oxidation reaction at the positive
electrode, whereas M'+ ions are withdrawn from the ideal solution by a reduction
reaction at the negative electrode when an external voltage source is applied to the
cell. This is illustrated in Figure 4.12. Both M'+ and X'- ions are transported by
combined diffusion and migration in the electrolyte.
The electrical field is directed from the positive electrode to the negative
electrode. The positive ions M'+ will migrate in the direction of the electrical field,
whereas the negative ions X'- will migrate in the opposite direction. The
concentration gradients of positive and negative ions are identical throughout the
electrolyte, because charge neutrality is sustained. The concentration gradient for
both ions is directed from the positive electrode to the negative electrode, because
positive ions are injected at the positive electrode and removed at the negative
electrode. Therefore, mass transport by diffusion is directed from the positive
electrode to the negative electrode for both ions. The model for mass transport in the
absence of an electrical field illustrated in Figure 4.8 can be combined with the
model shown in Figure 4.11 for modelling the Poisson equation. For this purpose,
the electrolyte is divided into p spatial elements, as shown in Figure 4.5. This yields
the model for one-dimensional mass transport in the x-direction in the electrolyte by
78 Chapter 4
combined diffusion and migration for a system such as that depicted in Figure 4.12.
This is illustrated in Figure 4.13. The network model of Figure 4.13 can also be
modelled in the electrical domain, as was discussed in section 4.1.2.
Vapp~;ed
+
{\
\_)
+ -
Diffusion
MM ., ..1,
Migration
1·:
~~ ~~>
Diffusion
x,_
Migration
X=O X=l
M~M".+ne M,...+ne~M
Figure 4.12: Schematic representation of an electrochemical cell with two metal (M)
electrodes placed in an ideal solution using a M"+x•- binary salt as the electrolyte, with the
M"+ and X'- ions being transported by combined diffusion and migration
The upper part of the network in Figure 4.13 describes the mass transport of the
positive ions M'+, while the lower part describes the mass transport of the negative
ions X"-. The molar amounts of the positive ions M"+ in spatial elements i are present
on the capacitances Cchi.M"+, while the capacitances Cchi,xn- store the molar amounts
of the negative ions )(''. A chemical potential J.-l is present across each capacitance.
The interface between the positive electrode ( +) and the electrolyte is located at
x=O. Here, positive ions M'+ are injected into the electrolyte by an electrochemical
Mn+
reaction. This electrochemical reaction takes place at a rate of lch (0). The
electrostatic positive electrode potential is f/Je.+· The interface between the negative
electrode and the electrolyte is located at x=l. Here, ions M'+ are removed from the
electrolyte by an electrochemical reaction. The rate at which this electrochemical
reaction takes place is l chM"+(l). The electrostatic negative electrode potential is f/Je.-·
The negative ions)('- are not allowed to react at the electrode/ electrolyte interfaces.
The electrostatic potentials, multiplied by nF, at the centres of all the spatial
elements are present at the central node points, as indicated in Figure 4.13. The
electrostatic electrode potentials ¢',,+ and f/Je.- are shown on the left and right,
respectively. The sum of the electrostatic potential at a certain location multiplied by
zF and the chemical potential J.-l yields the electrochemical potential Jl at that
location, as described by (4.2). Note that for the positive ions, z=+n, whereas for the
negative ions, z=-n. Therefore, a positive Jl+ is present at the top node points,
whereas a negative Jl_ is present at the bottom node points. Differences between the
electrochemical potentials 71+ of neighbouring spatial elements yield a chemical

flow through the chemical resistances in the upper part of Figure 4.13. In an
analogous manner, differences in ji_ lead to a chemical flow through the chemical
resistances in the lower part of Figure 4.13. The chemical resistances were defined
in (4.21c), in which the appropriate diffusion coefficients have to be used for
positive and negative ions, respectively. The expression for the geometric
capacitances Cche is given in Figure 4.1la. The distance from the centre to the
electrode surface in spatial elements 1 and p is only Lh/2. Therefore, L1xl2 instead of
L1x should be used in the expressions of Cch E,edge_
....._Electrostatic
c:;.ltdge potential at the
boundaries and
and certres
of spatial
elements 1.. p
muhiplied by nF
---+- Eledrochemical
Boundary:
X=O
-t - x
-t
=(p--1)ax
=~~:nsx"-
in spatial elements
Boundary:
X=l
1..p
Figure 4.13: Simulation model in the chemical domain for one-dimensional mass transport of
a charged species in the x-direction in the presence of an electrical field in, for example, the
system depicted in Figure 4.12
Figure 4.13 shows that the potential across the capacitances CchE.edge equals nF(¢1-¢/)
for both electrodes. The potential of the positive electrode ¢1 equals ~e.+ and the
electrolyte potential ¢1 in spatial element 1, which is directly adjacent to the
electrode surface, equals ~~- For the negative electrode, ¢/=~•. - and ¢/=fA,. This
means that the capacitances Cch E,edge are actually the double-layer capacitances in the
chemical domain! As a result, the distance L1xl2 is actually the thickness of the
Helmholtz layer; see chapter 3. It will be shown in section 4.3 how the network of
Figure 4.13 can be used to model an electrolyte and how it has to be combined with
the positive and negative electrode models.
The direction of migration and diffusion flows in the network of Figure 4.13
will now be considered. A complicating factor is that species with positive and
negative signs have to be considered in one network. Chemical flows are commonly
defined as a flow of positively charged particles. Therefore, all the species will be
considered to be positively charged for simplicity in the explanations below, which
means that chemical flows in the lower part of the network of Figure 4.13 express
80 Chapter4
the flow of the ions X as if they were positive. In reality, the flow of the negative
ions r· will be in the opposite direction. The electrostatic potentials fP; will decrease
from x=O to x=l, because the positive electrode potential is higher than the negative
electrode potential. As a result, a chemical migration flow from the left to the right
occurs in the upper part of the network, which means that positive ions migrate from
left to right under the influence of the electrical field. The chemical migration flow
in the lower part of the network is also directed from the left to the right, which
means that negative ions will migrate from right to left in reality. At the same time,
the injection of positive ions at x=O and their removal at x=l will lead to a
concentration gradient of positive ions from the left to the right. This will lead to a
diffusion flow of positive ions from the left to the right.
The electrostatic potentials in the spatial elements will be constant when the
electrical field throughout the electrolyte is stationary, which means that it does not
change in time. Hence, the flow through the geometric capacitances Cch e will be
zero. As a result, all chemical flow through the chemical capacitances in the upper
part of Figure 4.13 will flow through the capacitances in the same spatial element in
the lower part. Assume for the moment that this has been the case since t=O, which
means that the molar amount of the M'+ ions is the same as that of the r· ions in a
spatial element. As an example, assume that nFrp1 equals 10 J/mol in spatial element
1 and that the chemical potential across capacitances Cc/ for both the M'+ and the
r· ions is 1 J/mol. The latter statement implies that fl' has been assumed equal for
both ions for simplicity. Furthermore, assume that nFfP2 equals 9 J/mol in spatial
element 2, because the electrostatic potential decreases from the left to the right, and
that the chemical potential across both the capacitances Cch 2 in spatial element 2 is
0.5 J/mol. This means that the concentration gradient will indeed be directed from
the left to the right. As a result, JI1+ equals 11 J/mol and JI2 .+ equals 9.5 J/mol,
which can be inferred from applying Kirchhoffs voltage law in the chemical
domain. This leads to a chemical flow from the left to the right of 1.5 molls when
Rch is assumed to be I Jslmof for simplicity. This flow can be separated in a
migration flow of 1 molls and a diffusion flow of 0.5 molls, which both flow from
the left to the right, which is in agreement with Figure 4.12. In the bottom part of the
network, an electrochemical potential of 9 J/mol is present on the left-hand side of
Rch in spatial element 1, while an electrochemical potential of 8.5 J/mol is present at
the right-hand side in spatial element 2. Assuming the same value of 1 Jslmotl for
Rch• this leads to a chemical flow of 0.5 molls from the left to the right. This flow
can be separated in a migration flow of 1 molls from the left to the right and a
diffusion flow of 0.5 molls from the right to the left. This leads to a net chemical
flow by combined diffusion and migration of 0.5 molls from the left to the right. The
actual migration and diffusion flows of the negative ions will be in the opposite
direction, as stated above. Hence, the migration is directed from the right to the left
and the diffusion is directed from the left to the right. This is in accordance with
Figure 4.12. In summary, the migration part of the chemical flow will be caused by a
difference in electrostatic potentials between two spatial elements for each chemical
resistance in Figure 4.13, while the diffusion part will be caused by a difference in
chemical potential.
In the case described above, in which the concentration profiles of positive and
negative ions are the same, charge neutrality holds. This means that the molar
amounts of positive ions equal the molar amounts of negative ions. The Poisson
equation of (4.27) shows that the second derivative of the electrostatic potential is
zero in this case. This means that the electrical field inside the electrolyte is not a
function of the location x. Hence, the difference between f/J; and f/Jj across each
geometrical capacitance Cc{ is constant, so a linear electrostatic potential gradient is
present across the electrolyte.
The second derivative of the electrostatic potential is not zero in the Gouy-
Chapman diffuse layer, as described in chapter 3. This means that, according to the
Poisson equation of (4.27), there will be no charge neutrality in the spatial elements
present in this Gouy-Chapman layer. In order to avoid charge neutrality inside a
spatial element, the current through the geometrical capacitance Ccheon the left-hand
side must have been unequal to the current through the capacitance on the right-hand
side for at least some time. This leads to a difference in molar amounts of positive
and negative ions and hence to the absence of charge neutrality. It has been assumed
above that a stationary situation occurred since t=O. In reality, the electrical field has
to be built up since t=O. Let us therefore consider the build-up of an electrical field
inside an electrolyte in some more detail. Starting at t=O, the current that flows
through the capacitance Cch • on the left-hand side of a spatial element in one of the
two Gouy-Chapman layers at the electrode/electrolyte interfaces will differ from the
current that flows through the capacitance Cch• on the right-hand side. The current
flowing through the capacitances Cch • on the left- and right-hand sides will be equal
for spatial elements outside the Gouy-Chapman layers. As soon as a stationary
situation occurs at t=tb and the electrical field no longer changes in time, the current
through all capacitances Cch• will be zero. Because the current through all
capacitances Cc/ outside the Gouy-Chapman layers has been equal from t=O
onwards, the electrostatic potential across each capacitance will have the same
value. Therefore, the electrostatic potential gradient will indeed be linear. However,
this will not be the case for capacitances Cch • inside the Gouy-Chapman layers.
Therefore, the second derivative of the electrostatic potential in these two regions
will not be zero. Moreover, although the current flowing through the upper chemical
capacitance will equal the current flowing through the lower chemical capacitance
from t=t1 onwards, charge neutrality will never be attained in those spatial elements
inside the Gouy-Chapman layer. The reason for this is that the current through the
upper capacitance has been unequal to the current through the lower capacitance
from t=O to t=t1 .
The theory of mass transport presented in this section can be summarized as follows:
• The chemical resistance, defined in (4.2lc), models the general mass transport
laws of (4.18) and (4.19).
• The combination of geometric capacitances and chemical capacitances placed
around a central node point models the Poisson equation; see (4.27) for the
chemical domain.
• The principle of mass conservation is modelled by means of an RC-ladder
network, in which the difference between the chemical flows entering and
leaving a spatial element is integrated in a chemical capacitance.
• Boundary conditions at the interface of the considered system through which
species are transported, such as an electrolyte solution, and neighbouring
systems, such as electrodes, can be modelled by means of chemical flow
sources at these interfaces. The chemical flow sources have a zero value and can
be omitted when the species cannot be transported across the interface.
82 Chapter4
• Initial conditions of the mass transport problem can be introduced by placing

initial molar amounts on all the chemical capacitances at all locations in the
system
4.1.4 Modelling thermal behaviour

Heat production or consumption of an electrochemical cell has not been considered
in the previous sections. A well-known thermodynamic law links the change in free
energy L1G of a cell to the change in enthalpy AH, the change in entropy L1S and the
temperature Tat which the reactions takes place [15),[16],[21],[22]:
(4.31)
A simple example will be given to explain the meaning of this law. Consider a
simple experiment in which a battery is charged from an equilibrium situation. The
charge current is chosen infinitesimally small. As a result, the overpotentials can be
considered zero, and hence no losses occur in the system when the battery is
successively charged and discharged. This means that the amount of energy that has
been supplied during charging can be fully retrieved during discharging. The
equilibrium potentials of the electrodes can be considered constant, because the
current is so small that the equilibrium hardly changes. The battery can be regarded
as a system on which electrical work is performed by a current source that charges
the battery. The battery is to be considered to be in contact with a thermal reservoir
[21]. Heat transfer can occur between the battery and the thermal reservoir. As an
arbitrary choice, the charging reaction is considered to be exothermic, which means
that heat is transferred from the battery to the thermal reservoir. A similar derivation
as given below can be made when the charging reaction is assumed to be
endothermic. The situation is sketched in Figure 4.14a in which subscripts c mean
charging. The battery is shown in contact with a thermal reservoir and a current
source that performs work on the battery. Besides signs indicating energy and
entropy changes, potentials and currents, the figure also shows the effect of the
thermodynamic law of (4.31) ort the values of L1G, AH and (-TL1S) of the battery.
Figure 4.14a shows that an oxidation reaction (0) occurs at the positive
electrode and a reduction reaction (R) at the negative electrode. Note that £"'·+ >0
and F:q,· <0. This has been chosen for simplicity, but of course the explanation
below will also hold for other polarities. The polarities simply depend on which
potential is chosen as a reference in the system Positive electrical energy LIG0 is
supplied to the positive electrode, because both the electrode potential and the
current directed into the electrode are positive. Positive electrical energy LIGR is also
supplied to the negative electrode, because both the electrode potential and the
current directed out of the electrode are negative. The changes in free energy LIG0
and LIGR are indeed positive when the sign convention described in section 4.1.2 is
adopted. This is shown directly below the battery in Figure 4.14a, with a plus sign
being used for an oxidation reaction and a minus sign for a reduction reaction. The
total electrical energy supplied to the battery is positive and equals LIGc=Ebarlct,
which is simply the product of the supplied electrical power Ebarlc and time. It can
also be inferred from the summation of the energy LIG0 supplied to the positive
electrode and the energy L1GR supplied to the negative electrode, because zero losses
have been assumed. This leads to the expression L1Gc=nFEbat illustrated below the
battery in Figure 4.14a. This is in accordance with the sign convention adopted in
this book, because L1G=+nFE holds in the case of a non-spontaneous reaction in all
the electrochemical literature. This will indeed be the case in a charging reaction, as
work has to be performed on the battery to 'force' the charging reactions to occur.
+
Heat
Thermal Ll.Hc > 0 Thermal - TLl.S Ll.Hd < 0
reservoir Ll.S0 < 0 reservoir Ll.Sd > o
t t t t
Oxidation Reduction Reduction Oxidation
<I.G0 = nFE' > 0 <I.GA=- nFE-> 0 <I.GA =- nFE' < 0 <I.G0 = nFE- < 0
Figure 4.14: (a) Charging a battery (subscripts c mean charging) with an infinitesimally small
current, so that overpotentials are zero, so no losses occur and the equilibrium potential does
not change. The energy L1Gc supplied during charging can be fully retrieved (L1Gd) (subscripts
d mean discharging) during discharging. Changes in free energy (L1G), enthalpy (L1H) and
entropy (LIS) of the battery are shown. The charge reaction has arbitrarily been chosen to be
exothermic (LlSc<O) (b) Same as (a), but for discharging the battery with an infinitesimally
small current, the reaction is now endothermic (LlSd>O) by definition
The change in entropy in the battery during charging will be negative (L1Sc<0),
which means that the entropy decreases, because of the arbitrary choice of an
exothermic charging reaction. This decrease in entropy in the battery must be
balanced by transfer of heat to the thermal reservoir [21]. This is shown in Figure
4.14a. The consequence of (4.31) is depicted at the bottom of Figure 4.14a. The
change in free energy of the battery L1G0 supplied by the current source, is divided
between a change in enthalpy &lc and transfer of heat -TL1S to the thermal reservoir.
Note that -TL1S will be positive, because L1Sc<O and a positive temperature T have
been assumed. The numbers given in the figure are arbitrary and merely serve to
illustrate the principle of (4.31). The battery's free energy changes from 0 to 2 J/mol
(L1Gc=2 J/mol>O), which leads to a change in enthalpy from 0 to 1 J/mol (L1Hc=1
J/mol>O) and a change in heat from 0 to 1 J/mol (- TL1S =1 J/mol>O), which is
delivered to the thermal reservoir.
The battery will be discharged when the current source is replaced by a
resistor. Again, an infinitesimally small current is assumed. Now, a reduction
reaction will occur at the positive electrode and an oxidation reaction at the negative
electrode. Both L1G0 and L1GR will now be negative, as indicated below the battery in
Figure 4.14b in which subscripts d mean discharging. The change in the battery's
free energy during discharging L1Gd, which equals the work performed on the
84 Chapter4
environment, is again inferred from the summation of L1G0 and L1GR, leading to
L1Gd=-nFEbat· This means that the minus sign convention (LlG=-nFE) that is often
encountered in the literature actually holds for the spontaneous reaction direction.
Note that L1Gd=-L1G0 which means that the energy supplied during charging will be
retrieved completely. This is in agreement with the assumption of zero losses. As
(4.31) is of course still valid, the discharge reaction has to be endothermic, or
L1Sd>0. This means that the battery's entropy will increase. This can be seen at the
bottom of Figure 4.14b, in which the battery free energy changes from 2 J/mol to 0
(L1Gd =-2 J/mol<O) resulting from a change in enthalpy from 1 J/mol to 0 (L1Hd =-1
J/mol<O) and a change in heat from 1 J/mol to 0 (- TL1Sd = -1 J/mol<O). This heat (-
TL1S) is obtained from the thermal reservoir.
The above experiment merely served to give an idea of what happens with
respect to heat in an ideal case. In reality, losses will occur in a battery when it is
charged or discharged. The heat flow generated inside a battery by electrochemical
reactions i is given by [22],[27]:
Jin,elch
th
=~
LJ
-T!::.Si III+~ [:n
F 1 LJ t'ft
+/2Rohmic
total (4.32)
; n; ;
where lrhin,etch is the heat flow in [W], generated by the battery, I the battery current
in [A], I; is the partial current of electrochemical reaction i in [A], T the battery
temperature in [K], L1S; denotes the entropy change associated with electrochemical
reaction i in [J/(mol.K)], n; the number of electrons exchanged in electrochemical
reaction i, 1}; the total overpotential due to kinetic aspects and mass transport
associated with electrochemical reaction i in [V] and R101athmic denotes the total
ohmic series resistance inside the battery in [Q].
The heat flows defined in the second and third terms of (4.32) will always be
positive, directed from the battery to the thermal reservoir in Figure 4.14. Note that
in the second term, I; and 1]; will always have the same sign by definition. This
means that, in practice, the battery will always heat up because of these two terms,
irrespective of the direction of the current. This makes sense, as these two terms
represent the losses that occur inside the battery as a result of overpotentials and
ohmic series resistance, respectively. These heat flows can be seen as 'one-way'
flows of heat from the battery to its thermal environment. These 'one-way' flows
occur during both charging and discharging, and hence the amount of energy
supplied during charging will never be fully retrieved during discharging.
The first term can be either positive or negative. It is positive when LlS;<O,
which is valid for an exothermic reaction. This means that the battery will heat up as
a result of all the exothermic reactions that take place. The battery will cool down as
a result of the endothermic reactions, for which LlS;>O. This was explained in Figure
4.14. A reaction that is exothermic during charging is endothermic during
discharging and vice versa. The heat flow in the first term is defined as the amount
of heat, -TL1S in [J/mol], multiplied by a chemical flow 1/nF in [molls]. This is in
agreement with the theory presented earlier in this section. In a purely chemical
reaction, the second and third terms in (4.32) will be zero, because no electrical
current I is involved. The first term will be a heat flow defined by -TLlSJch·
The temperature of a battery is determined by the net amount of heat inside the
battery and the thermal or heat capacitance of the battery, according to the analogy
between the electrical and thermal domains in Table 4.1. The net amount of heat is
built up by a net heat flow lth net inside the battery, which is determined by the total
heat flow Dthin,i, generated internally by all the chemical and electrochemical
reactions, and the heat flow Jthout from the battery to the environment by means of
conduction, convection and radiation. Hence:
}net=~ Jin,i _}out

th £.J th th (4.33)
The heat flow from the battery to the environment is given by [1]:
(4.34)
where Jthout denotes the total heat flow from the battery to the environment in [W],
ll1h the battery heat transfer coefficient in [W/(K.rn2)], Abat the battery surface area in
[m2], Tamb stands for the ambient temperature in [K], and Rthbat denotes the thermal
resistance from the battery to the environment in [KIW]. This thermal resistance is
the sum of the conductive, convective and radiative thermal resistances that describe
the heat transfer from the battery to the environment.
The net heat flow from inside and outside the battery yields the accumulated
amount of heat inside the battery. This net amount of heat can be calculated using:
Qthnet =JJ thnet (t)dt =cbat

th
•T (4.35)
where Qthnet is the net amount of heat inside the battery in [J] and Cthbat the thermal
or heat capacitance of the battery in [J/K].
A simple first-order network model can be constructed that models the thermal
behaviour of a battery using the analogy between the electrical and thermal domains
in Table 4.1 and the equations described in this section. This is illustrated in Figure
4.15.
As can be seen in Figure 4.15, the net heat flow Jthnet inside the battery is
integrated in the thermal capacitance Ct/at. The battery temperature T bat is present
across the battery thermal capacitance. In a similar way, the ambient temperature is
present across the ambient thermal capacitance Cthamb· A temperature difference
between the battery and the environment will give rise to a heat flow lthour between
the battery and the environment, with the sign depending on the sign of the
temperature difference. Changes in the battery temperature will have virtually no
effect on the ambient temperature when the ambient thermal capacitance is much
larger than the battery thermal capacitance. Then the battery will not be able to heat
up the environment. Only a single battery temperature is calculated in this simple
thermal-network model. This means that the battery temperature is considered to be
uniform throughout the battery. More sophisticated thermal-network models could
take into account the changes in temperature occurring from one location inside the
battery to the other.
86 Chapter4
J,~et ~
-
J out
th
Rbat
th
+ +
athnet Tbat
0 thamb Tamb
cbat
th
- camb
th
Figure 4.15: Simulation model in the thermal domain that describes the temperature
development in a battery
4.2 A simulation model of a rechargeable NiCd battery
4.2.1 Introduction
The general principles described in the previous section will be used in this section
to construct a network model for a rechargeable NiCd battery [23]-[28]. As many
different types of NiCd batteries exist, there is no single model that is valid for all
types.
A schematic representation of a mature design of a sealed rechargeable NiCd
battery is given in Figure 4.16. This design was chosen as the basis for deriving the
battery model in this section. An example of a NiCd battery based on this design is
the P60AA manufactured by Panasonic. The three basic elements in Figure 4.16 are:
the positive nickel electrode, the negative cadmium electrode and a concentrated
KOH electrolyte solution that provides ionic conductivity between the electrodes.
To prevent electrical contact between the positive and negative electrodes, the two
electrodes are separated by a thin porous layer of insulating material, the separator.
Main storage reactions that take place during charging and discharging
The electrochemical storage reactions are given in the region denoted as storage
capacity in Figure 4.16. The arrows pointing upwards denote the charging reactions,
while the arrows pointing downwards denote the discharging reactions. All reaction
equations have been arranged with the same amount of electrons for comparison. In
their simplest form, the main storage reactions are given by:
k..M
Ni(OH) 2 + OH. +===~ NiOOH + H 20 +e· (4.36)
k.,,NI
for the nickel reaction and
Cd+20H. (4.37)
k•. CJ
for the cadmium reaction. During charging, Ni(OHh is oxidized to NiOOH at the
nickel electrode. At the same time, Cd(OHh is reduced to metallic Cd at the cad-
mium electrode. The reverse reactions take place during discharging. Current flow is
supported by electrons inside the electrodes and by OH' ions inside the electrolyte.
Externally, the current flows through the charger or load. K+ ions do not react at the
electrode interfaces. Their behaviour can be compared to that of the xn· ions in
Figure 4.12, i.e. the net current supported by K+ ions can become zero.
nickel cadmium
electrode electrode
2 + 2H20 +4e
-
Over-
r-------f .. 02/--
-------l
Cd(OH)2 charge
--
40H'
4H20
2Cd charge
u
4NiOOH
storage QNI, Max chp~e
Ocd, Max
capr______ _ ~Cdl!-"1< t______ j_

4 f-r
dis hjuge discharge
4Ni(OH)2
40H
li:
-- --
Ocd, Cd Over-
Cd(OH)2 Cd
discharge
~ 02
40H-::::: 0
Separator
+
KOH electrolyte
Figure 4.16: Schematic representation of a sealed rechargeable NiCd battery
Side-reactions taking place during overcharging and overdischarging

Besides the main storage reactions, electrochemical side-reactions also take place,
especially during overcharging and overdischarging. These side-reactions include
oxygen production and reduction reactions, schematically represented by the curved
arrows in Figure 4.16, and a cadmium side-reaction at the nickel electrode,
represented by the grey Cd(OH)z area in the nickel electrode. All grey areas in
Figure 4.16 symbolize the excess material that is added to the NiCd battery to
protect it from excessive pressure build-up during overcharging and overdis-
charging. How this is achieved will be discussed below.
Overcharging
An overcharge reserve of excess Cd(OH)z is added to the cadmium electrode.
Hence, the nickel electrode is the capacity-determining electrode, which is reflected
by the fact that QNi,Max is smaller than Qcd,Max in Figure 4.16. As a result, the amount
of Ni(OH)z will approach zero before all the Cd(OH)z has been consumed at the
cadmium electrode during the charging process. An alternative oxidation reaction
has to occur at the nickel electrode. This reaction is the production of oxygen,
according to:
(4.38)
Oxygen is produced at the nickel electrode through the oxidation of OH. ions into
oxygen and water molecules. Metallic Cd is still formed at the cadmium electrode,
88 Chapter 4
because of the excess Cd(OH)z. Oxygen molecules can diffuse to the cadmium
electrode when the amount of oxygen inside the battery increases. There, the
oxygen-reduction reaction starts to compete with the Cd-forming reaction. The
oxygen molecules can be reduced to OR ions, according to:
kc,02
0 2 + 2H2 0 + 4e· ---.40H. (4.39)
The production of oxygen at the nickel electrode and removal of oxygen at the
cadmium electrode are schematically represented by the top curved arrow in Figure
4.16.
Overdischarging
An overdischarge reserve of excess Cd(OHh, generally denoted as a depolarizer, is
added to the nickel electrode in the amount Qcd(OH!2.Ni• as indicated in Figure 4.16.
Moreover, some metallic cadmium is added to the cadmium electrode in an amount
Qcd.Cd smaller than Qcd(OH! 2,N;- This is done to enforce similar oxygen production and
reduction processes as those occurring during overcharging. In this way, the
production of hydrogen gas at the nickel electrode is prevented and a hydrogen
recombination cycle is avoided [29]. This is important because of the poor electro-
catalytic activity of the cadmium electrode with respect to the Hz oxidation reaction,
which would lead to an undesirable very high internal Hz pressure. Overdischarging
takes place in two steps.
Because of the excess of metallic Cd in the cadmium electrode, the amount of
NiOOH in the nickel electrode will diminish before all the metallic Cd has been
consumed in the cadmium electrode. An alternative reduction reaction now has to
occur at the nickel electrode. The excess Cd(OH)z in the nickel electrode will be
reduced to metallic Cd during the first overdischarging step and the reverse
oxidation reaction will still occur at the cadmium electrode. Hence, the following
reactions occur during the first overdischarging step.
k,·,Cd
Cd(OHh + 2e· _____. Cd + 20H" (4.40)
at the nickel electrode, and

ka,CJ
Cd + 20H"---. Cd(OHh + 2e· (4.41)
at the cadmium electrode.

The second overdischarging step starts when discharging proceeds to the stage
at which all the metallic Cd has been consumed at the cadmium electrode. Oxygen
gas will be produced at the cadmium electrode in an alternative oxidation reaction.
This leads to an increase in the amount of oxygen inside the battery. As a result, the
oxygen reduction reaction starts to compete with the Cd(OHh reduction reaction at
the nickel electrode and a similar situation as during overcharging occurs. The
reactions of (4.39) and (4.40) compete at the nickel electrode, while the reaction of
(4.38) takes place at the cadmium electrode. The production of oxygen at the
cadmium electrode and the removal of oxygen at the nickel electrode during the
second overdischarging step are schematically represented by the bottom curved
arrow in Figure 4.16.
Figure 4.16 shows that three types of reactions can take place inside a NiCd
battery. More details on these reactions will be given in the next three sections, in
which use is made of the theory described in chapter 3 and section 4.1. The activity
effects of electrons will be neglected, or a, and a;•f will be assumed to be the same,
as discussed in section 4.1.2.
4.2.2 The nickel reaction

The main storage reaction for the nickel electrode was given above in (4.36). It is
assumed that the volume of the active material does not change during charging and
discharging and that NiOOH, which is the Ox species, and Ni(OH) 2 , which is the
Red species, are the only species present within the region of the nominal storage
capacity. In other words, it is assumed that the sum of the mol fractions x; of the
species i is unity, or:
(4.42)
XNiOOH + XNi(OH), = 1
where the mol fractions X; are defined by (y= 1 assumed):
a;
X; =------'---- (4.43)
aNiOOH + aNi(OH) 2
where i is either NiOOH or Ni(OH)z. This means that the electrode is considered to
be a solid-solution electrode, with the activities of the species being variable and
their sum constant. Mol fraction xN;ooH corresponds to the State-of-Charge (SoC) of
the NiCd battery, because NiOOH is the species that is formed during charging.
Thermodynamics of the nickel reaction

The change in free energy L1GN;,o of the oxidation reaction (0) that occurs at the
nickel electrode during charging can be found in the same way as shown above in
(4.10a); see also (4.36):
. a aref
+ RTln( a NtOOH aref
HzO N•COH), ow
)
(4.44)
ref ref
a Ni(OH), a OW a NiOOHa H 2 0
a .
=!1Go M~
+ RTln( NtOOH
a
H 20
aref
Nt(OH)z
~
aref )
ow
~
aNi(OH) 2 aowaNiOOHaH2 0
The a 0 w activity can be considered to be constant when, as a simplification, the

electrolyte concentration is considered to be constant and uniformly distributed
across the battery. Activity aH2o is also constant, because of this simplification. The
simplification is reasonable when the currents applied to the battery are not too high.
The ln term containing these constant activities can then be added to ,1(1'Ni,O· The
contribution of the ar•f terms is constant and will also be added to ,1(1'Ni.O· The
90 Chapter4
following expression is now found for the equilibrium potential of the nickel
reaction:
eq !1GNiO
E M"-- '
1 nNiF
(4.45)
where nNi=l, according to (4.36). The same expression can be found for the
equilibrium potential when the reduction or discharge reaction of (4.36) is
considered. This was shown in section 4.1.2. All activities are bulk activities, which
are equal to the surface activities due to true equilibrium, because (4.45) expresses
the true equilibrium potential. As will be described below, H+ ions are removed from
the nickel electrode during charging and inserted during discharging. It has been
found in experiments that electrical and elastic interaction between these H" ions and
between the H+ ions and the nickel and oxygen atoms in the electrode does not lead
to a deviation from the Nemst-like behaviour of the equilibrium potential in (4.45)
[30].
The value of Jt>Ni in a strong alkaline solution has to be used, which is 0.52 V
vs the Standard Hydrogen Electrode (SHE) at 298 K, because the electrolyte in the
NiCd battery is a concentrated 8 M KOH solution and the constant concentration of
the electrolyte is present in Jt>Ni through a 0 w [31].
Kinetics of the nickel reaction

The kinetics of an electrochemical reaction are determined by the value of the
exchange current through the Butler-Volmer relationship, as was discussed in
chapter 3. In agreement with this theory and (4.12), the exchange current value for
the nickel reaction is determined by:
(4.46)
where ANi refers to the surface area of the nickel electrode, ka.Ni and kc,Ni are the
anodic and cathodic rate constants of the nickel reaction, respectively, and wNi• xNi•
YNi and ZNi denote the reaction orders of the corresponding species in the nickel
reaction. Again, the bulk activities of the reacting species are used in (4.46), because
the exchange current is derived from the equilibrium situation.
Diffusion limitation of the nickel reaction

The bulk activities of the reacting species in (4.36) may differ from the activities at
the electrode surface. The nickel reaction is generally assumed to proceed through
diffusion of H+ ions through the active electrode material. H+ ions are formed during
charging, which diffuse to the electrode/electrolyte interface, where they recombine
with OR ions to form H20. The opposite occurs during discharging. This is
described by the following reaction equations, which yield (4.36) by addition.
Ni(OHh +===~ NiOOH + W + e· (4.47)
kf
H+ + OH • • H20 (4.48)
kb
A schematic one-dimensional representation of the nickel electrode during charging
is shown in Figure 4.17. It is assumed that the positions of the nickel atoms are fixed
and that charge transport occurs by means of electrons and H+ ions only. The
movement of electrons is considered to be much faster than that of H+ ions, so mass
transport limitation is attributed to H+ ions only. It is further assumed that the nickel
electrode is a perfect conductor. As a result, the electrical field inside the electrode
will be negligible and mass transport through migration can be neglected. Mass
transport through convection can also be neglected and hence only diffusion of H+
ions is considered.
Current collector Electrolyte
•-e-
• •
H+-+-
N~i3+
X=O
Figure 4.17: Schematic one-dimensional representation of the nickel electrode during

charging
Electrons are removed from the lattice at the nickel atom sites during charging and
H+ ions are formed in the oxidation of Ni(OHh to NiOOH. The H+ ions move
through the lattice from one nickel atom site to another, which leads to an apparent
flux of Ni(OHh and NiOOH species. Concentration profiles of Ni(OHh and NiOOH
arise in the electrode when the charge-transfer kinetics is fast in comparison with the
92 Chapter4
mass transport rate of H+ ions. The W ions eventually arrive at the

electrode/electrolyte interface, where they form H20 with OH ·ions. The diffusion
of H+ ions through the electrode in the absence of an electrical field, as shown in
Figure 4.17, can be described as explained in section 4.1.3. An apparent equilibrium
potential ENieq• can be defined, which is the summation of the true equilibrium
potential ENi•\ defined in (4.45), and a diffusion overpotential TJN/ according to
(4.26), when the surface activities of the NiOOH and Ni(OHh species differ from
the bulk activities. The apparent equilibrium potential can be inferred from (4.45) by
using the surface activities for the NiOOH and Ni(OH)2 species. Hence,
(4.49)
4.2.3 The cadmium reactions

The cadmium reaction equation was given in (4.37). The reaction can occur at both
electrodes, as described in section 4.2.1. Only the cadmium electrode will be
considered in the remainder of this section, for simplicity. The same equations also
hold for the cadmium reaction that may occur in the nickel electrode.
The cadmium electrode is considered to be a heterogeneous system composed
of two separate phases, which are metallic Cd, the Red species, and an oxide layer
consisting of Cd(OH)2, the Ox species. This as opposed to the nickel electrode,
where two species NiOOH and Ni(OHn co-exist in a single phase. It is assumed that
electron transfer takes place at the interface of the Cd species and the electrolyte
only and not at the Cd(OH) 2/electrolyte interface. This assumption is based on the
fact that Cd(OH)2 is a poor electronic conductor [32]. The reaction of (4.37) takes
place via a dissolution-precipitation mechanism [33]. This mechanism is a two-step
reaction according to:
kj
Cd(OH)z + 20H. oE ) Cd(OH)/
k. (4.50)
kc.OJ
Cd(OH)/ + 2e" +===~ Cd + 40H.
k,QJ
Soluble intermediate Cd(OH)/" ions are formed by a chemical reaction during
charging. Subsequently, these ions can be reduced at the solid-solution interface,
resulting in the formation of a Cd layer. The actual transfer of electrons takes place
at the Cd surface only, because electrons are only exchanged here. During
discharging, the reverse reaction sequence occurs.
Thermodynamics of the cadmium reactions

The change in free energy LlGcd,R in the reduction reaction (R) that occurs at the
cadmium electrode during charging can be found in the same way as for (4.1 Ob ); see
also (4.37).
+ RT ln( a cd Ia \2 a Cd(OHlz
~ on- J ref J (4.51)
ref I ref \2
acd \a ow J aCd(OHlz
= AGo + RT ln[ a
Cd
Ia \2aref
~ on- J Cd(OH) 2
J
Cd ,R ref I ref \2
acd \a on- J acd(OHlz
The a0 H' activity will be considered to be constant, as described in section 4.2.2. In

addition, the activities of the solids Cd(OH)z and Cd are equal to the corresponding
reference activities by definition. The ln term containing a0 u- can now be added to
LlG°Cd.R• while the ln terms that contain the activities of the Cd(OH)2 and Cd species
are zero. The following expression is then found for the equilibrium potential of the
cadmium reaction:
2
a on-
- !1G~d.R - RT ln
are1 (4.52)
on-
where ncd=2, according to (4.37). Again, the same expression can be derived from
the oxidation or discharge reaction in (4.37). Note that (4.52) expresses the true
equilibrium potential, which means that the bulk activity a0 H' should be used. The
equilibrium potential for the cadmium reaction is constant, irrespective of the state
of the reaction, because all activities are constant. The value in a strong alkaline
solution should be used for E'cd• which is -0.81 V vs SHE at 298 K [31], because
the concentration of the electrolyte is present in E' Cd through a 0 w
Kinetics of the cadmium reactions

The kinetics of the cadmium reaction are described by the Butler-Volmer equation.
The exchange current value for the cadmium reaction is determined by:
94 Chapter4
(4.53)
where ka,ed and kc.ed are the anodic and cathodic reaction rate constants of the
cadmium reaction, respectively, Xed• Yed and Zed are the reaction orders of the
corresponding species and Aed refers to the free metallic cadmium surface at which
the charge transfer takes place. All bulk activities in (4.53) are constant. It is
assumed that during charging, the cadmium is first formed in the form of single
nuclei, which grow further into metallic particles, which eventually form a layer.
The reverse process takes place during discharging. As a result, the free metallic
cadmium surface area Aed is a function of the amount of cadmium in the cadmium
electrode and hence of the SoC. It is further assumed that the time interval during
which the nuclei are formed is sufficiently small to allow us to consider the
nucleation process to be instantaneous. This means that all nuclei are formed when
the charging starts at time t 0 • During normal battery operation, Cd will already be
present in the cadmium electrode at the beginning of the charging due to the
overdischarge reserve Qed,ed; see Figure 4.16. This means that nuclei will indeed
already be present at time t0 •
Hemispherical nuclei are assumed, which grow or shrink at the rate at which
the reaction of (4.37) takes place. Suppose that the formation of metallic cadmium
during the charging starts with N nuclei at t=t0 • The 'extended' electrode surface
area Oex of N hemispherical particles at any time t>to can be inferred from 0ex=N1t
?(t), where the radius r(t) of each hemispherical particle can be inferred from the
particle's volume. This is the area of the particle that is in direct contact with the
surface. It is assumed that the number of particles remains constant. Only the size of
the particles increases during charging or decreases during discharging. In addition,
it is assumed that all the particles have the same volume at each time t. The volume
of a hemispherical particle is half that of a sphere, or (213)~. The total volume V(t)
of the deposited material can be obtained by taking N times the volume of a
hemisphere, i.e.
vo +_g_ JJ cd(t)dt
2 M I
V(t) = N -m 3 (t) = (4.54)
3 Pcd ,.
where V(t) is the total volume of theN hemispherical particles in [m3], N the number
of hemispherical particles, r(t) the radius of hemispherical particles in [m], V' the
initial volume of the N hemispherical particles at t=t0 in [m3], Med the molecular
weight of cadmium in [kg/mol], Ped the gravimetric density of cadmium in [kg/m3]
and led is the chemical flow associated with the cadmium reaction of (4.37) in
[molls]. led is positive during charging and the volume of the N hemispherical
particles increases. During discharging, led is negative and the volume of the N
particles decreases.
An expression for the extended surface area (}ex can be inferred from (4.54).
First, an expression for the radius r(t) will be derived:
3 -
=( - M JJ cd (t)dt
3
1 [ vo + _g__
1
r(t) ]]
2 Nn Pcd I"
I
(4.55a)
Cd [ mcd + JJ cd(t)dt
3
3 -
=( - M-
1 -
1
0
]]
2 Nn Pcd 10
where m°Cd denotes the molar amount of the cadmium nuclei at t=t0 • This amount
equals the molar amount of the overdischarge reserve Qcd,Cd (see Figure 4.16) during
normal battery operation, which means without the occurrence of deep discharges,
where part of the overdischarge reserve has been consumed. Surface area e.x is now
inferred from:
e.x = Nm 2 (t) =-Nn

-2 (
3M cd
- - -[
o
J
mcd + I J Cd (t)dt ]]3
(Nn )3 2 Pcd 10
2
(4.55b)
= (Nn'fi '(3M
_ _g__
2 Pcd
[m~d + JJ cd(t)dt
1
t0
]]
3
The actual surface area covered by metallic cadmium Acd becomes smaller than e,x
when the growing cadmium particles start to merge, because a correction has to be
made for the overlapping areas. The occurrence of overlapping areas is clarified in
Figure 4.18.
The relation between the actual surface area covered by metallic cadmium and
e,x can be inferred from Avrami' s theorem [34]:
(4.56)
where Acdmax denotes the maximum surface area that can be covered by cadmium
particles, i.e. the surface area of the cadmium electrode in [rrr].
Diffusion limitation of the cadmium reactions

The only species that can be depleted at the electrode/electrolyte interface in the case
of the cadmium reaction are OH' ions. It was assumed for the nickel reaction that the
surface activity of OH' ions equals the bulk activity. This means that diffusion
limitation of the nickel reaction due to depletion of OH- ions was assumed to be
negligible. This is justified, because the diffusion of H+ ions is dominant in
determining the diffusion overpotential for the nickel reaction. This is explained by
the diffusion constant of W ions, which is four orders of magnitude smaller than the
96 Chapter4
diffusion constant of OH- ions [35],[36]. Diffusion of OH- ions has been considered
for the cadmium reaction for completeness, because OH- ions are the only species
that can be depleted from the electrode/electrolyte surface in this case. This means
that the surface activity of OH- ions may differ from the bulk activity. As was shown
in (4.24), the apparent equilibrium potential Ec/q* can be written as the sum of the
true equilibrium potential of (4.52) and a diffusion overpotential t'Jcl. which
depends on the difference between the bulk and surface activities of on ions.
Figure 4.18: Overlapping metallic cadmium nuclei growing on the electrode surface
s 2
a on-
-l!l.G~a.R- RT In
a'•f
Eeq*- on-
Cd -
ncaF
2
b (4.57)
a on-
-l!l.G~a.R - RT In
+E-i~~.YJ
a'•!
on-
ncaF n ca F ~s J a 08 _
Eeq d Eo d
= ca +TJca = ca +TJca
The mass transport through diffusion of On ions can be similarly described by

means of an RC-ladder network, as described above for the mass transport of Ir'
ions in the nickel electrode. However, a simplification can be made for OH- ions by
considering stationary diffusion with a linear concentmtion gradient across a thin
diffusion layer with a fixed thickness d 0 u-. This assumption is allowed when the
time necessary to build the diffusion layer is much smaller than a typical battery
charge/discharge time. The diffusion layer thickness can be assumed to be in the
order of l )..lm for OK ions, while the diffusion constant DoH· is in the order of w- 10
m2/s [36]. A general law relates the thickness of the diffusion layer to the diffusion
constant and the time needed to build the diffusion layer according to ~ou-1(7!D0 u-)
[37]. The time needed to build the diffusion layer in the case of oH· ions is 0.3 ms,
using the numbers mentioned, which is indeed a lot smaller than normal
charge/discharge times. On the other hand, the diffusion constant for H+ ions DH+ is
in the order of 10"14 m2/s [35]. The time needed to build a diffusion layer of 1 J..lm is
indeed 3 s. Therefore, stationary diffusion cannot be assumed for W ions and an
RC-ladder network should be used.
In the case of stationary diffusion, the diffusion current lowdif can be inferred
from Fick's first law, as described in chapter 3, with I= nFJ and a= Jt:. Hence,
(4.58)
where n=2 and the activity coefficient You- will again be considered to be unity.
Area Acd defined by (4.56) has to be used, because the charge transfer reaction takes
place at the cadmium surface only. The kinetic current Icln is described by the
Butler-Volmer equation for the cadmium reaction of (4.37), with the exchange
current f'OJ being given by (4.53). A simple way of including the effect of OH"
diffusion on the overpotential of the cadmium reaction is to find an expression for
the overall reaction current. This current is determined by the kinetic current for
small overpotentials and by the diffusion-limited current lou- dif for large
overpotentials; see Figure 3.4. The overall reaction current can be found in [15],
assuming a mixed kinetic-diffusion controlled reaction:
(4.59)
Of course, the effect of OK ion diffusion on the cadmium electrode potential can
also be modelled by means of an RC-ladder network. This has not been considered
in the present NiCd model for simplicity.
4.2.4 The oxygen reactions

The reaction equation for oxygen production was given above, by (4.38), whereas
that for oxygen recombination was given by (4.39). When oxygen production occurs
at one electrode, oxygen recombination will occur at the other electrode and vice
versa, as described in section 4.2.1. Where applicable, the electrode at which the
reaction takes place will be mentioned in the text.
Thermodynamics of the oxygen reactions

The change in free energy AG0 2> 0 in the oxidation reaction given by (4.38), which
occurs, for example, at the nickel electrode during overcharging, is given by:
98 Chapter4
!:1G02 ,0 =f.lo 2 + 2p,H20 + 4p,e_ - 4f.low

= 110
1""'02
+ 2~""'HzO
110 + 4110 4110
~""'e-- ~""'oH-
(4.60)
=!:1G~.0 + RT ln(a-7
2• a'eJ J
0
Oz
The a0 w and a820 activities have again been considered to be constant for simplicity
and their reference activities have been taken the same. Hence, the ln term that
contains these constant activities is added to L1G0 02 0 to yield the term Aao'a2 0 , while
' '
the ln term that contains the reference activities is zero. The following expression for
the equilibrium potential of the oxygen reaction is inferred from (4.60):
(4.61)
=
where n02 4, according to (4.38). Again, the same expression can be derived by
considering the reduction reaction of (4.39). The bulk activity of 0 2 should be used,
which equals the surface activity in a state of true equilibrium, because (4.61)
expresses the true equilibrium potential. The value of E:~in a strong alkaline is 0.40
V vs SHE at 298 K [31].
The oxygen inside the battery is present in either the gas phase (g) in the free
volume inside the battery or dissolved in the electrolyte (1). The oxygen dissolution
reaction can be represented by:
kJ
Oil) oE ~ Olg) (4.62)
kz
where k1 and k2 are the corresponding reaction rate constants. Henri's law applies in
the case of an ideal gas [21]. This law relates the activity of oxygen dissolved in the
electrolyte to the oxygen pressure in the gas phase through:
(4.63)
where (4.63) also holds for the activity in the reference state a0 / '1 and the reference
oxygen pressure Po/'1, which is taken to be 1 atmosphere (=105 Pa). Moreover, yg
denotes the fugacity coefficient in [mol/(m3.Pa)] and K02 the oxygen solubility
constant in [mol/(m3 .Pa)]. Using (4.63), the equilibrium potential E 0 /q of (4.61) can
be rewritten as:
(4.64)
Kinetics of the oxygen reactions

The exchange current (f' 02l Ni for the oxygen production reaction that occurs at the
nickel electrode surface ANi during overcharging is given by:
(4.65)
where the bulk activities of OR ions and water are again considered to be constant
for simplicity, ka,o/i and kc,o/i are the anodic and cathodic reaction rate constants of
the oxygen production reaction at the nickel electrode, respectively, and x 0 2' y 02 and
z02 again denote the reaction orders. In order to reduce the complexity of the
mathematical treatment, in the simple expression for the anodic branch of the
Butler-Volmer equation, which describes the oxygen evolution, it is assumed that
the surface coverage of oxygen molecules remains negligibly low and that the
concentrated KOH electrolyte behaves like an ideal solution. An expression similar
to (4.65), but in which ANi has been replaced by Acd and with different values for the
reaction rate constants, holds for the oxygen production reaction that occurs at the
cadmium electrode during the second overdischarging step. This was described in
section 4.2.1.
The oxygen that has been formed at the nickel electrode during overcharging
will eventually be reduced at the cadmium electrode, as was explained in section
4.2.1. The oxygen can be transported to the cadmium electrode through the
electrolyte or through the gas phase. It is most likely that the oxygen will be
transported mainly through the free gas volume inside the battery [38], since the
solubility of oxygen in the highly concentrated KOH electrolyte solution is
relatively low, i.e. 4.10-5 mol/! in 8 M KOH [39]. The presence of water is essential
for the recombination reaction, as can be inferred from the oxygen recombination
reaction of (4.39). Therefore, the dissolution of oxygen in the electrolyte and its
transport to the electrode surface through the electrolyte is an important step in the
oxygen reduction reaction. The kinetics of the oxygen reduction reaction are
100 Chapter4
described by the cathodic branch of the Butler-Volmer equation. The exchange

current of the oxygen reduction reaction that takes place at the free metallic
cadmium surface Acdduring overcharging is inferred from:
(4.66)
where the charge transfer coefficient ao2R may have a different value than the charge
transfer coefficient ~ for the oxygen production reaction and the same remark
holds for the anodic and cathodic reaction rate constants ka,o2 Cd and kc,o 2Cd in relation
to those in (4.65). An expression similar to (4.66) can be used for the oxygen
reduction at the nickel electrode during the second overdischarging step. This
reaction then competes with the cadmium reduction reaction of (4.40) that takes
place at the surface of the exposed cadmium at the nickel electrode Ac/i. Hence, the
area of this cadmium should be used in the equation for the exchange current of the
oxygen reduction reaction that takes place during the second overdischarging step.
Diffusion limitation of the oxygen reactions

For the oxygen reactions, water is assumed to be present in abundance. Therefore,
mass transport limitation can only occur as a result of depletion of either 0 2
molecules or oH· ions. The diffusion layer thickness is assumed to be equal for both
species and in the order of 1 j..Lm The diffusion constant for both species is in the
order of 10·10 m2/s [36],[39]. Therefore, stationary diffusion can be assumed for both
species, as discussed in the previous section. This leads to a diffusion limitation
current for OH- ions, defined by (4.58), for which the appropriate area A; should be
used, which will depend on the electrode at which the 0 2 reaction occurs. A similar
expression can be found for stationary diffusion of 0 2 molecules through a diffusion
layer with a thickness d02 . This diffusion occurs for the reduction reaction of (4.39):
(4.67)
where n=4, D 02 is the diffusion coefficient of oxygen in [m2/s] and the activity
coefficient y02 will again be considered to be unity. The appropriate area should be
used for A;.
(4.58) and (4.67) show that the diffusion limitation currents of Olf ions and Oz
molecules depend on the bulk concentrations of OH- ions and 0 2 , respectively. As
the bulk activity of OH" ions will be a lot higher than that of 0 2 molecules in the
electrolyte due to the strong alkaline electrolyte, only diffusion limitation of 0 2
molecules is considered for the oxygen reactions in the NiCd model. A difference
between the oxygen bulk activity and the oxygen activity at the electrode surface
gives rise to a diffusion overpotential77o/ The apparent equilibrium potential Eo/q*
can be written as the sum of the true equilibrium potential Eo/q of (4.61) and 1'/o/.
in accordance with (4.24):
(4.68)
Whether or not diffusion limitation plays a role in the reaction will depend on the
value of the overpotential. This was shown in Figure 3.4, in which it was illustrated
that the electrode current levels off to the diffusion-limited current for large
overpotentials. The electrode current is however determined mainly by the reaction
kinetics at relatively low overpotentials. A mixed kinetic-diffusion-controlled
reaction takes place in the intermediate region, as considered before in section 4.2.3.
The values of the overpotentials for the oxygen reactions which take place at the
nickel and cadmium electrodes during overcharging and overdischarging differ
substantially.
Oxygen is produced at the nickel electrode during overcharging, as explained
in section 4.2.1. The reaction takes place at an electrode potential that is at least 120
mV more positive than the equilibrium potential for the oxygen reaction. This can be
understood by considering the difference between E'Ni of 0.52 V vs SHE and E'02 of
0.40 V vs SHE. In this case it is assumed that the kinetics of the reaction can be
simply described by the anodic branch of the Butler-Volmer relationship. It is hence
assumed that the overpotential is generally not large enough for diffusion limitation
of oxygen to be a limiting factor. This means that the reaction current does not level
off to the diffusion-limited current at the usual values of the overpotential.
The oxygen reduction reaction during overcharging takes place at a very high
overpotential because the cadmium electrode potential is very negative with respect
to E'02 . Here, it is reasonable to assume that mass transport of oxygen through the
electrolyte to the free metallic cadmium surface becomes a limiting factor. Whether
the electrode current will equal the diffusion-limited current or be less will depend
on the parameter values of the oxygen reaction and the oxygen gas pressure through
the bulk activity of oxygen; see (4.67). Therefore, as for the cadmium reaction, a
mixed kinetic-diffusion-controlled reaction is assumed for the oxygen reduction
reaction. Hence, the reaction current for the oxygen reduction reaction is given by:
(4.69)
Again, the diffusion of oxygen can also be modelled by an RC-ladder network. This
has not been considered in the present NiCd model for simplicity.
Oxygen is produced at the cadmium electrode and reduced at the nickel
electrode during the second overdischarging step, as explained in section 4.2.1. The
situation described for overcharging is now reversed. The oxygen evolution at the
cadmium electrode takes place at a relatively low overpotential. Therefore, only the
102 Chapter 4
anodic branch of the Butler-Volmer equation is considered in the NiCd model. The
reduction reaction takes place at the cadmium surface at the nickel electrode at
relatively high overpotentials, like the oxygen reduction during overcharging. This
again leads to the use of a mixed kinetic-diffusion-controlled reaction in the model,
as described by (4.69).
Oxygen gas pressure calculation

The relation between the oxygen gas pressure inside the battery and the molar
amount of oxygen is described by the general gas law [21]:
RTm0
- 2
(4.70)
where m 01 denotes the molar amount of oxygen in the gas phase in [mol] and Vg is
the free gas volume inside the battery in [m3]. The solubility of oxygen in the highly
concentrated KOH electrolyte solution is poor, as described above. Therefore, the
contribution to m01 of oxygen dissolved in the electrolyte is neglected [38]. The
amount of oxygen in the gas phase will depend on the difference between the
amount of oxygen produced at one electrode and the amount reduced at the other.
Therefore, the net molar amount of oxygen inside the battery can be found by
integrating the net chemical flow of oxygen inside the battery:
dmo, = }Ni + }Cd _ JR,Ni _ JR,Cd

dt o, o, o, o,
/Ni + /Cd _ 1 R,Ni _ 1 R,Cd (4.71)
- o, o, o, o,
n0 ,F
where the latter term is inferred from the relation between the chemical and
electrical domains, as explained in section 4.1.
4.2.5 Temperature dependence of the reactions

The temperature dependence of the reaction rate constants k; and diffusion
coefficients D; in the equations is described by the Arrhenius relation for all
reactions that take place inside the NiCd battery, which is given in a general form in
[15]:
-E;ar
par(T) = paro exp(--) (4.72)
RT
where par(T) denotes a temperature-dependent parameter, which can be any reaction

rate constant k; or diffusion coefficient D;, par0 denotes the pre-exponential factor
and E'"par is the activation energy of par in [J/mol]. The temperature dependence of
the standard redox potentials E'"; of the reactions i is given by [15],[31]:
E; (T) = E; (298[K]) + (T- 298[K]) dE:

dT (4.73)
= E: (298[K]) + (T- 298[K]) f:..S;
n;F
where L1S'; denotes the entropy change associated with reaction i in [J/(mol.K)]. It is
assumed that LIS; is constant within the limited temperature range in which batteries
usually operate. The sign of LIS; will depend on whether reaction i is endothermic or
exothermic, as explained in section 4.1.4.
4.2.6 The model

The complete simulation model for a sealed rechargeable NiCd battery based on the
schematic representation given in Figure 4.16 is shown in Figure 4.19. This model is
based on the general principles described in section 4.1. Figure 4.19a represents the
equivalence of the electrochemical behaviour and Figure 4.19b that of the thermal
behaviour. In Figure 4.19a, the nickel electrode model is shown on the left side and
the cadmium electrode model on the right side. The electrodes are modelled by a
series resistance R; that represents the electronic conductivity of the electrode
material, in series with a parallel connection of several reaction paths and a
capacitance C;dl that represents the double-layer capacitance.
The electrodes are separated by a resistance R., which models the ionic
conductivity of the electrolyte and the separator. This is a simplification of the
model describing mass transport as a result of diffusion and migration in section
4.1.3 (see Figure 4.13) in which OH" ions have replaced the Mn+ ions, because both
ion types react at the electrode surfaces. K+ ions have replaced the xn· ions, because
neither ions react at the electrode surfaces. The bulk activity of the oH· ions in the
electrolyte is assumed to be constant and uniformly distributed, as described above.
This means that no diffusion profile is present for the OH" ions and, because of
charge neutrality, the same remark holds for the K+ ions. The bulk and surface
activities of the oH· ions are assumed to be the same for the nickel and oxygen
reactions, as described above. Only for the cadmium reactions is a difference
between the surface and bulk activities of the oH· ions considered. This will be
modelled in the anti-parallel 0 and R diodes, as will be described below, and not by
two chemical capacitances describing the surface and bulk activities of the OH" ions,
respectively. This means that the molar amounts of the OH- and K+ ions are constant
in each of the spatial elements 1,2, ..p in the network shown in Figure 4.13.
Consequently all these chemical capacitances can just as well be omitted from the
network. The resistance string for K+ ions will then become floating and can also be
omitted. No current will flow through the geometric capacitances when the electrical
field is assumed to be stationary. Hence, these capacitances can also be omitted.
This means that transient effects in the electrolyte will not be considered. What
remains is a string of resistances that describes the migration of OR" ions from one
electrode to another. This string of resistances can be transferred to the electrical
domain and can be simply modelled by a resistance R, with a constant relatively low
value in the mQ-range. This gives rise to a relatively small voltage drop across the
electrolyte. This low voltage drop is due to the high concentration of the electrolyte.
104 Chapter4
In section 4.3 it will be shown for aLi-ion battery model how the electrolyte can be
modelled taking into account mass transport as a result of diffusion and migration.
Nickel IElectrolyte I Cadmium
dl
eq•
RNI + EN; Ro - E~ + Red
c~ C~(OH)2
R (a)
CCd(OH)2
ch
(b)
Figure 4.19: Simulation model in the electrical, chemical and thermal domains of a
rechargeable NiCd battery based on the schematic representation of Figure 4.16. (a)
Equivalence of electrochemical behaviour, (b) Equivalence of thermal behaviour
Three parallel reaction paths can be distinguished in the part of the model
representing the nickel electrode, which from top to bottom denote: (i) the nickel
storage reaction, (ii) the oxygen side-reaction and (iii) the cadmium side-reaction.
Two parallel reaction paths can be distinguished in the part of the model
representing the cadmium electrode, which from top to bottom denote: (i) the
cadmium storage reaction and (ii) the oxygen side-reaction.
Two domains can be distinguished in Figure 4.19a, notably the electrical and
the chemical domain. Energy storage in the electrical domain is modelled by the
double-layer capacitances CU, that in the chemical domain by chemical capacitances
Cch;, where i denotes the reacting species. The coupling between the electrical and
chemical domains is represented by ideal transformers, as described in section 4.1.
Each reaction path j is modelled as an ideal transformer in series with two
anti parallel diodes Dj, with one diode representing the oxidation (0) reaction and the
other the reversed reduction (R) reaction, as indicated in the figure.
Only changes in the activities of NiOOH and Ni(OHh are considered for the
nickel reaction, as discussed in section 4.2.2. Therefore, only the capacitances
CctwoH and ccti(OHJ2 are present in the chemical domain. The source LJG'~; has
been defined in (4.45) for an oxidation reaction. This means that this source includes
all jl' values and hence the jl' term is omitted in the electrochemical potentials J.1.
across capacitances Cct;ooH and ccti(OHJ2. LJG'~; also includes the constant
activities of water and OK ions; see (4.45).
The diffusion process of H+ ions in the x-direction inside the electrode is
modelled by an RC-ladder network, as can be seen in Figure 4.19a. The electrode
has been divided into spatial elements as shown in Figure 4.5. A network similar to
that shown in Figure 4.8 is used, because no electrical field is present; see section
4.2.2. The capacitances on the left-hand side contain the molar amounts of the
NiOOH species in spatial elements i and the capacitances on the right-hand side
contain the molar amounts of the Ni(OH) 2 species in spatial elements i. The top of
the RC-ladder network denotes the electrode/electrolyte interface at x=O, where the
actual electrochemical reaction takes place, whereas the bottom denotes the
electrode/current collector interface at x=lNi· The diffusing W ions can only cross
the electrode/electrolyte boundary due to the electrochemical reaction, whereas the
electrode/current collector boundary cannot be crossed. The chemical resistances
Rc/i(OHJ2 and RchNiOOH are given by (4.21c), in which the diffusion coefficient of H+
ions in the nickel electrode (DH+) is used for both resistances. The resistances have a
value which depends on the position x (see Figure 4.8) because the value of the
resistances depends on the molar amount m of the corresponding species at position
x, according to (4.21c).
A concentration profile of Ni(OH)z and NiOOH species throughout the
electrode material leads to a change in chemical potential of both species throughout
the electrode. However, this change in chemical potential can only be measured in
electrical terms as an extra voltage drop r/ across the electrode/electrolyte interface,
as described by (4.24). Hence, the apparent equilibrium potential ENieq• for the nickel
reaction is present across the electrical port of the transformer. The current-
overpotential relation for the diodes DN; is described by the Butler-Volmer equation,
with the exchange current !'Ni being given by (4.46).
Only the change in the activity of oxygen is considered for the oxygen
reaction, as described in section 4.2.4. This is expressed by the capacitance Cch02 ,
which contains the total molar amount of oxygen present inside the battery. The
contribution of oxygen dissolved in the electrolyte to the molar amount of oxygen is
neglected; see section 4.2.4. Therefore, capacitance Cch 0 2 contains the molar amount
of oxygen present in the gas phase, which is directly linked to the internal oxygen
106 Chapter4
pressure through (4.70). The sources L1G"'0 2 are defined by (4.60). The net chemical
flow of oxygen production (see (4.71)) is into Cch 02 , whereas that of oxygen
reduction is out of Cch02 . This has been realized in the model shown in Figure 4.19
by connecting Cch0 2 to the oxygen reaction path at both electrodes with the signs
indicated.
Only one capacitance describing the bulk molar amount of oxygen is present in
the model. Hence, no distinction is made between the surface and bulk activities of
oxygen in the chemical domain. This means that the voltage across the electrical
port of the ideal transformer is the true equilibrium potential; see (4.61). The
diffusion overpotential rf is taken into account in the current-voltage relation for the
anti-parallel diodes D 02 . Hence, the overpotential rf' across the anti-parallel diodes
equals r/'+rf. No diffusion limitation of oxygen is to be expected for the oxygen
evolution reaction at either electrode; see section 4.2.4. Therefore, the anodic branch
of the Butler-Volmer equation is used for the oxidation diodes. Here, (4.65) will be
used for the nickel electrode and a similar expression with different parameters for
the cadmium electrode. Diffusion limitation is to be expected in the oxygen
reduction reactions. Therefore, (4.69) will be used to describe the current-voltage
relationship for the reduction diodes in the nickel and cadmium electrode models,
using the appropriate parameter values.
The activities of all the reacting species except OH- can be considered to be
constant in the cadmium reaction, as mentioned above. The molar amounts of both
Cd and Cd(OH)2 have been modelled by a chemical capacitance in Figure 4.19 in
consistence with the rest of the model. However, it should be noted that in this case
the chemical potential across these capacitances is actually constant. Therefore,
these capacitances are actually sources with a constant chemical potential. The
molar amount of Cd and Cd(OH)2 can be found by integrating the chemical flow
through these sources.
No capacitance that describes the molar amount of OH- ions is to be found in
the cadmium electrode model. The bulk activity of OH- ions has been taken into
account in source L1G"~d; see (4.52). The voltage across the electrical port of the
transformer is the true equilibrium potential for the cadmium reaction. Diffusion
limitation is taken into account in the current-voltage relationship for the anti-
parallel diodes Dcd• in a similar way as described for the oxygen reaction. The
current-voltage relationship is described by (4.59) for both the oxidation and the
reduction diodes in the nickel and cadmium electrodes. The same value is used for
ked in the kinetic current Iclin for both electrodes. However, the maximum surface
area Acdmax in (4.56) is taken to be lower for the nickel electrode than for the
cadmium electrode for the dependency of Acd on the amount of cadmium, because
Qcd(OHJz.Ni is much smaller than Qedmax+Qcd,Cd in Figure 4.16.
The thermal behaviour outlined in Figure 4.19b is modelled in the same way as
discussed in section 4.1.4. The contributions to the internal! y generated heat flow are
described by (4.32). The calculated battery temperature is used in the model
equations to model the temperature dependence.
During current flow, the current divides itself over the available reaction paths
in each electrode. Which reactions will occur will depend on the value of the
equilibrium potential of each reaction and the rate at which each reaction can take
place. In principle, the redox reaction with the lowest value of the equilibrium
potential will thermodynamically be more favourable than a redox reaction with a
higher value of the equilibrium potential. Applying Kirchhoffs voltage Jaw to two
parallel reaction paths reveals that the thermodynamically favourable reaction has a
higher overpotential 1] across the antiparallel diodes. However, this does not
necessarily mean that this reaction will actually take place, because the reaction rate
might be too low, for example due to a low exchange current or diffusion limitation.
More information on the parameter values used in the NiCd model will be given in
section 4.4.
4.3 A simulation model of a rechargeable Li-ion battery

4.3.1 Introduction
The general principles described in section 4.1 will be used in this section to
construct a network model for a rechargeable Li-ion battery [40],[41]. Again, there
are many different types of Li-ion batteries. The CGR17500 Li-ion battery
developed by Panasonic will be modelled in this section. This battery consists of a
LiCo02 positive electrode and a graphite LiC 6 negative electrode. The electrolyte of
aLi-ion battery consists of a salt dissolved in an organic solvent; see chapter 3. The
dissolved salt is LiPF6 in the case of the CGR17500 Li-ion battery considered in this
section, which yields Lt and PF6. ions in the organic solvent. This solvent is a 1:3
mixture of ethylene carbonate (EC) and ethyl methyl carbonate (EM C). The use of
EC as a co-solvent is of major importance for the performance of the Li-ion battery,
because it forms a protective layer on the graphite electrode [40]. This protective
layer is denoted as a Solid Electrolyte Interface (SEI) and it allows only Li+ ions to
be transported to and from the graphite electrode. It prevents the risk of solvent
molecules entering the graphite electrode together with these ions [43]. The SEI is
electrically insulating and, when sufficiently thick, it suppresses the decomposition
of the electrolyte at the electrode surface. Besides the two electrodes and the
electrolyte solution, a separator is present inside the battery. A Positive-Temperature
Coefficient resistor (PTC) is also present inside the battery for safety reasons; see
chapter 3. The SEI will not be considered in the Li-ion model described in this
section. A PTC can be modelled by a series resistance. The CGR17500 is a
cylindrical battery, with a diameter of 17 mrn and a height of 50 mrn. An external
series safety switch has to be added in applications, as described in chapters 2 and 3.
The overall reaction during charging and discharging of the considered Li-ion
battery, in which X Li+ ions are involved, is given by:
ch
xLiCo02 +xC6 ..: • xCo02 +xLiC6 (4.74)
d
Li+ ions are moved back and forth between the electrodes, as expressed in (4.74),
which explains the frequently encountered term 'rocking-chair battery'. The Li+ ions
move from the LiCo02 electrode to the graphite electrode during charging. They
move in the other direction during discharging. Both electrodes are so-called
intercalation electrodes, with the u+ ions being guests that can be inserted into or
extracted from the host lattice. In an intercalation electrode, the host lattice does not
change significant! y upon insertion of the guest atoms [44].
Unlike in the NiCd battery model presented in the previous section, no side-
reactions have been modelled. Side-reactions that could take place inside a Li-ion
battery include Li metal deposition and electrolyte decomposition. These side-
reactions could be responsible for aging effects leading to gradual capacity decrease
during the battery's lifetime. So these side-reactions are important. However, they
will be left out of the model presented in this book, because of insufficient
108 Chapter4
knowledge of their characteristics. Hence, in each of the electrodes only one reaction
is included in the model. The two electrode reactions will be described separately in
the next sections. Moreover, the behaviour of the electrolyte will be described.
4.3.2 The LiCo02 electrode reaction

Li+ ions are extracted from the LiCo02 electrode during charging and inserted into it
during discharging. The extraction and insertion of Li+ ions occurs through
oxidation and reduction, respectively, as described by the following basic reaction
equation for one Li+ ion:
k.,po.
LiCo02 ~===~ Co02+Lt +e· (4.75)
kc.rxn
where ka,pos and kc,pos are the anodic and cathodic rate constants of the LiCo02
(pos=positive) electrode reaction, respectively. The electrode is modelled in a way
comparable with that used to model of the nickel electrode in section 4.2. The
electrode is considered to be a solid-solution electrode, with LiCo0 2 being the Red
species and CoOz the Ox species. The Ox species actually denotes the portion of the
electrode material from which Lt ions have been extracted. It is assumed that
LiCo02 and Co02 are the only species present in the electrode. As a result, the sum
of the mol fractions of LiCo02 and Co02 is unity. As in the NiCd model, volume
changes in the active material during charging and discharging are neglected. The
positions of the Co02 centres in the electrode lattice are fixed and the Li+ ions are
transported through the electrode. Hence, the LiCoOz species can be compared with
the Ni(OHh species in the NiCd battery and the Co02 species with the NiOOH
species. The Li+ ions can be compared with the H+ ions in the NiCd battery. The mol
fraction of Co02 corresponds to the SoC of the Li-ion battery, because the Co02
species is formed during charging. The mol fraction of LiCo02 equals the mol
fraction of Li+ ions inside the positive electrode, i.e. (y=1 assumed):
X upos -- xpos
UCo0 2
= -----=---
aUCo02
(4.76)
aucoo2 + aco0
2
In theory, the mol fraction xtf0 ' can vary between 0, when all the Li+ ions have been
extracted, and 1, when all the lattice sites have been filled. In the latter case there are
no longer any vacancies for Li+ ions. In practice, only part of this full capacity can
be used reversibly under normal operating conditions. How large this part will be
will depend on the material used for the positive electrode. For example, the
reversible range of xupo" is less than or equal to 0.5 in the case of the material used in
the CGR17500 Li-ion battery [18]. For the CGR17500 battery this means that the
LiCoOz electrode will only cycle between xv' = 0.5 and Xupos = 0.95 under normal
operating conditions. This means that the rated nominal capacity, derived from the
range 0.5< xtlos <0.95, will be only a little less than half of the full capacity of the
LiCo02 electrode in this case. The mol fraction of the Co02 species can be derived
in a similar way as shown in (4. 76). This mol fraction equals 1- xtf0 '.
Thermodynamics of the LiCo02 electrode reaction

The change in free energy L1Gucoo2.o in the oxidation reaction that occurs at the
LiCo02 electrode during charging can be found in a similar way as shown in (4.10).
As discussed above, the LiCo02 electrode is modelled in a similar way as the nickel
electrode. The equilibrium potential of the nickel electrode can be described by the
Nernst equation; see (4.45). However, in the case of the extraction and insertion of
Lt ions from and into an intercalation electrode, such as the LiCo02 electrode, the
electrical and elastic interaction between the intercalated u+ ions and between u+
ions and the host atoms has to be accounted for [44],[45]. This interaction can be
described with the aid of an extra interaction energy term. The change in free energy
can be inferred from (4.75) as follows:
!:!G LiCo02 ,o =f.J,Co0 2 + f.j u• + /1 e- - f.j LiCo02
- RTU posxfrs + RT(pas
(4.77)
where Upos denotes the dimensionless interaction energy coefficient in the LiCo0 2
electrode, RTUpos is the interaction energy in [J/mol], xdos is defined in (4.76), and
(,os denotes a dimensionless constant in the LiCo02 electrode, where RT(,os is an
interaction energy term which does not depend on xd08 • As before, bulk activities
have been used in (4.77) because the equilibrium condition is valid. All reference
activities a'•! have been assumed the same in the last line, as in earlier derivations,
and the activity au+ of u+ ions has been assumed to be the same as its reference
activity for simplicity. This has been done because the electrolyte concentration is
approximately 1 M and no concentration profile is present in the electrolyte in a
state of equilibrium. As before, the activity of electrons has been taken the same as
the reference activity a/'1. The following expression is then found for the
equilibrium potential of the LiCo02 electrode:
-
!:lGLiCoO2• o
nLiF
= !:!G'/_;coo,,o + -RT-
nuF
[1n(-acoo,
- - J- U posxLipos + r pos ] ~ (4.78)
nLiF aLiCo02
_ o
- Eucoo + -RT- [ 1n( 1 - xLos J- uposXupas + r pos ]
~
' nuF xf;s
where nu = 1 (see (4.75)) and the extra terms due to the interaction between the
intercalated u+ ions and between these ions and the host material are clearly
identifiable. The same expression can be derived by considering the reduction
110 Chapter4
reaction of (4.75). This was shown in section 4.1.2. Both activities in the ln term
have been replaced by the corresponding mol fractions, as defined in (4.76).
The expression for F:qucoa 2 in (4.78) is in agreement with an expression for the
equilibrium potential for the LiCo02 electrode given in [45]. The expression for the
equilibrium potential is similar to the Nemst-like equilibrium potentials derived in
previous sections in this chapter when the interaction energy between the
intercalated Lt ions is zero, or Upos= 0.
The equilibrium potential drops more steeply with increasing xd0 ' in the case
of repulsive interaction, which occurs when Upos >0. Repulsive interaction means
that if we place a Li+ ion and an electron at one site in the lattice, the nearest
neighbour sites will be forbidden for the next ion that is inserted [45]. The repulsive
interaction can lead to the formation of a 'superlattice' inside the electrode, in which
the intercalated Li+ ions are ordered at fixed distances, so that interactions are
avoided as much as possible, which means that maximum distance between nearest
neighbours is achieved [44]. One can easily imagine a repulsive electrical interaction
between ions of the same charge. If Li+ ions are intercalated in layers, one can
imagine that repulsive elastic interaction will result when aLi+ ion is present in layer
n, thereby leaving less space for intercalation directly above and below the ion in
layersn-1 andn+l.
The equilibrium potential will drop less steeply than with pure Nemst-like
behaviour with an increasing xupos for attractive interaction, where Upos<O. Attractive
interaction means that if we place a Li+ ion and an electron at one site in the lattice,
the nearest neighbour sites will be preferred for the next insertion [45]. In an
extreme case, the equilibrium potential can become constant for a certain range of
xd0 '. Phase separation will then occur and the intercalated Li+ ions will condense in
two separate phases [44],[45). A phase is characterized by a certain lattice structure.
The energy of interaction between the phases is negligible in comparison with the
interaction energy within each phase [44]. In simple words, the Li+ ions stick
together in two separate phases. Screening of the Li+ ions by electrons present in the
lattice will occur, which prevents that the Li+ ions show repulsive behaviour.
Therefore, attractive electrical interaction can take place. One can imagine attractive
elastic interaction in the above example in which the Lt ion was intercalated in
layer n in the lattice. As a result, more space will be created for the sites directly
next to this ion in the same layer.
When Li+ ions are intercalated in a host lattice, phase transitions between
possible lattice structures may occur depending on the number of Li+ ions present in
the host lattice [44]. Each possible lattice structure is characterized by a certain
amount of free energy, which will vary during intercalation. Depending on the free
energy values of each of the possible lattice structures, a first-order phase transition
will take place to the lattice structure with the smallest amount of free energy. The
two phases may exist together over a certain range of xu' when attractive
interaction occurs. This was discussed above.
Two first-order phase transitions occur within the LiCo02 electrode at
approximately xdos =0.25 and xd0 ' =0.75 [46]. The phase transition at Xupos = 0.25
is not noticed during normal battery operation, because the battery is not cycled in
this region, as previously discussed. The other phase transition is noticed as a
change in the slope of the true equilibrium potential as a function of xd0 ' . This
change in the slope of the equilibrium potential is realized in the present Li-ion
battery model by a change in the interaction energy between the intercalated Li+ ions
from a value Upos.I in phase 1 to a value Upos,z in phase 2. The values of the
dimensionless constants (pos,I and 'os, 2 in phases 1 and 2, respectively, are chosen so
that a smooth transition is achieved from the equilibrium potential in phase 1 to that
in phase 2. This modelling approach is similar to the one described in [45]. The two
equilibrium potentials in the different phases for the considered CGR17500 Li-ion
battery are described by:
f~E'q )
UCo0 phase! -
2
_ EoUCoO, + -RT [l (1-
xff" ) - U pos,IXUpos + ~r pos,l
F n ---p;;;-
nu Xu
J (4.79)
for 0.75 ~ xLrs ~ 0.95, and
f~E'q )
UCoO, phase2
_ Eo
- LiCoO, + -RT
F n
nu
[l (1-xf;' ) _
pos
Xu
U pos
pos,2XLi + ~r pos,2 J (4.80)
for xdos < 0.75. The equilibrium potential is found to be independent of xLrs in
phase 1, which means that this is a two-phase region and the interaction is attractive
[46]. Phase l and phase 2 actually co-exist in this region of x~•. The slope of the
equilibrium potential is steeper than predicted by pure Nemst behaviour in phase 2.
Hence, the interaction becomes repulsive in phase 2. In order to obtain a smooth
transition of the equilibrium potential at the phase transition, the relation between
'os,I and (pos,2 is defined as:
(pos2 =(Upos,2 -Upos,l)x:;:,..,,wuition +(pos,l

(4.81)
= (U pos,2- U pos,l) • 0.75 + ( pos,l
It should be noted that a smooth transition can also be obtained by assigning a
different value to the standard redox potential e>ucoa2 in each of the two phases. The
terms (pos,I and (pos, 2 can then of course be omitted in (4.79) and (4.80), respectively
[45].
Kinetics of the LiCo02 electrode reaction

The reaction kinetics of the charge-transfer reaction that takes place at the surface of
the LiCo02 electrode are described by the Butler-Volmer relationship. The exchange
current value for this reaction is determined by:
(4.82)
where ALiCoo2 denotes the surface area of the LiCo02 electrode and Xucoo2 and Yucoo 2
the reaction orders. Again, bulk activities have been used, because an expression for
P is always derived under equilibrium circumstances. As a simplification, it is in the
present model assumed that the kinetics of the reaction can be described by the same
parameter values in both phases.
112 Chapter4
Diffusion limitation of the LiCo02 electrode reaction

Diffusion of Lt ions is important in determining the rate at which the reaction of
(4.75) occurs. Diffusion of u+ ions inside the electrode is important in determining
the reaction rate. The movement of electrons is again assumed to be much faster
than the movement of u+ ions and mass transport limitations inside the electrode are
therefore attributed to Lt ions only. Moreover, a negligible electrical field is
assumed to be present inside the electrode. Mass transport through convection is
neglected. Therefore, only mass transport due to diffusion of Li+ ions inside the
electrode material is considered. Concentration profiles of the LiCo02 and the Co02
species arise because the Li+ ions diffuse through the electrode material. This can be
modelled in the same way as described above for the nickel electrode.
u+ ions enter or are retrieved from the electrolyte in the same form, as opposed
to the situation at the nickel electrode described in section 4.2.2, where It' ions form
water molecules or are obtained from water molecules at the electrode surface. The
Li+ ions are transported further in the electrolyte. Accumulation or depletion of Li+
ions can occur at the electrode surface, because Li+ ions are not present in
abundance in the electrolyte, and a concentration profile can consequently be
formed. So, mass transport limitation of Lt ions in the electrolyte must be
considered. This influences the surface activity a 5u+ of u+ ions and hence also the
potential across the electrode/electrolyte interface. Rearrangement of the terms in
(4.78) yields an expression for the true equilibrium potential E'qucoo2 and the
diffusion overpotential rfucoo2• which in summation yield the apparent equilibrium
potential E'q*ucooz· The latter potential can be inferred from (4.78) by using surface
activities instead of bulk activities. The activity au+ of the u+ ions must now be
included, which was neglected in the equilibrium situation in (4.78); see (4.77).
The second term in the last line of (4.83) describes the overpotential across the
electrode/electrolyte interface due to the mass transport of u+ ions through the
electrode material. The third term in the last line of (4.83) expresses the effect of a
concentration profile of Li+ ions in the electrolyte solution.
E UCo0
eq•
2
_
-
EoUCo0
2 +-- [1 [
RT n -a~oo,
, - - ) - U posXupos +\>pos
r ]
nuF aueoo,
- - l[a~,)
+RT n-
nuF a';j.
=Eo.
UCoo,
+ RT
nU
F
[l (a~oo,a~oo,a~coo, )-u
n s b b
a LiCo02 a coo, a UCo02
P?'
posXu
+ \>r pos ]
+RT
--nl [a~,)
--
nuF a';j.
-_ EoLiCo02 + -RT [1 [ a~oo,

- n - b - -) -
U po.,X U
pos + \>r pos ]
nuF aucoo, (4.83)
RT acoo aucoo
+--ln[ s '
RT
b)
au,
' +--ln - - [')
nuF a~coo,a~oo, nuF a';j.
RT
-_EoLiCo02 + - - n --;;;;-
nuF Xu
[1 (1-
xf;'' ) - U posX U
pos + \>r pos ]
4.3.3 The LiC6 electrode reaction

Lt ions are inserted into the LiC6 electrode during charging and extracted from it
during discharging. The insertion and extraction of u+ ions occurs through
reduction and oxidation, respectively, as described by the following basic reaction
equation for one Lt ion:
(4.84)
where ka,neg and kc,neg are the anodic and cathodic reaction rate constants of the LiC6
(neg=negative) electrode reaction, respectively. The electrode is modelled in the
same way as in the model of the LiCo~ electrode outlined in the previous section.
The electrode is considered to be a solid-solution electrode, with UC6 being the Red
and C6 the Ox species. It is assumed that UC6 and C6 are the only species present in
the electrode and that the sum of their mol fractions is unity. The positions of the C 6
centres in the electrode host lattice are fixed and the u+ ions are transported through
114 Chapter4
the electrode and can fill the sites intended for guest atoms. The mol fraction of LiC6
in this case corresponds to the SoC of the Li-ion battery, which equals the mol
fraction of u+ ions inside the electrode, or (y=lassumed):
(4.85)
The mol fraction of the C6 species can be derived in a similar way and equals 1-
xun•g. Unlike with the LiCo02 electrode, Xu neg can be cycled from nearly zero to 1 in
the LiC6 electrode, which means that the electrode can be completely filled by u+
ions. The value of xuneg will remain somewhat larger than zero after the first
charge/discharge cycle. This is because the SEI layer is formed during the first
charge/discharge cycle, as described in section 4.3.1. This means that the LiC6
electrode in the CGR17500 battery is cycled between Xu neg =0.06 and Xu neg =1 under
normal operating conditions.
Thermodynamics of the LiC6 electrode reaction

As the LiC6 electrode is an intercalation electrode, just like the LiCo02 electrode,
the interaction between the intercalated Lt ions and between these ions and the host
atoms has to be accounted for. Therefore, a similar expression holds for the change
in free energy LlGuc6.R of the reduction reaction that occurs at the LiC6 electrode
during charging:
(4.86)
where Uneg denotes the dimensionless interaction energy coefficient in the LiC 6
electrode, RTUneg the interaction energy in [J/mol], xuneg was defined by (4.85), and
(,.g denotes a dimensionless constant in the LiC6 electrode. Bulk activities have
been used, reference activities have been taken the same, the activity au+ of u+ ions
in the electrolyte has been assumed to be the same as the reference activity in a state
of equilibrium and the electron activity has been taken to be the same as its reference
activity, as before. This yields the following expression for the equilibrium potential
of the LiC 6 electrode reaction:
E eq
uc.
+ RTF ac. J-u

[1n( auc neg xneg
L1 + \:>neg
r ] (4.87)
nu 6
_e
- LiC6 + -RT- [1 (1- xZ~g J U
ll neg -
neg
neg X Li
r ]
+ \::>neg
nLiF Xu
The same expression can be derived starting from the oxidation reaction in (4.84).
Mol fractions have been used instead of activities in the last line of (4.87); see
(4.85).
The so-called staging phenomenon occurs in the LiC 6 electrode [47],[48]. This
phenomenon implies that the Li+ ions do not occupy all the available layers at the
beginning of the intercalation process. A number of k-1 empty layers exists between
layers in which u+ ions are present when the electrode is in the k'h stage. The phase
of the electrode is characterized by the stage or stages in which the electrode resides.
Only high stages are present when charging starts. A phase transitions will occur
each time a new stage arises. This is caused by empty layers that also start to be
filled. Four phase transitions that take place between five phases in the LiC 6
electrode have been reported in the literature [47]. Two stages co-exist in most of
these five phases. The equilibrium potential is constant in these regions, as discussed
above. One stage might be present or two other stages might co-exist after a phase
transition and the equilibrium potential changes to a new value. For example, a
phase transition may occur from the co-existence of stages 2 and 3 to the co-
existence of stages I and 2. The magnitude of such a potential change depends
strongly on the inter- and intra-layer interaction energies [48].
A good example of a lattice-gas model for the staging phenomenon in the
graphite electrode can be found in [48]. However, only two phases and one phase
transition at xuneg =0.25 will be considered in the present Li-ion model as a
simplification. The reason is that measurements have shown that phase transitions at
other values of xuneg have only a negligible effect on the equilibrium potential
[40],[41]. The two phases have been modelled in the same way, as described for the
LiCo02 electrode. The two equilibrium potentials in the two different phases in the
considered CGR17500 Li-ion battery are described by:
fEeq )
\ LiC• piulsel -
_Eo
LiC6 + -RT [ [1-xZ~g
F 1n neg
JU
-
neg
neg,IXLi + \::>r neg,l ] (4.88)
nu xu
for Xu neg < 0.25 and

116 Chapter 4
for xuneg ;:: 0.25. The dimensionless constants (,., 8. 1 and Sneg. 2 are defined in the same
way as for (4.81).
Kinetics of the IiC6 electrode reaction

The reaction kinetics of the reaction that takes place at the surface of the LiC 6
electrode are described by the Butler-Volmer relationship. The exchange current
value for the reaction is determined by:
(4.90)
where Auc6 denotes the electrode surface area and Xuc6 and Yuc6 the reaction orders.
As a simplification, it is in the present model assumed that the kinetics of the
reaction can be described by the same parameter values in both phases.
Diffusion limitation of the LiC6 electrode reaction

As with the LiCo02 electrode, diffusion of Lt ions both inside the electrode and in
the electrolyte plays a role in determining the rate of the reaction. An expression for
the apparent equilibrium potential E'q*uc6 can be found in the same way as shown in
the derivation of (4.83), using (4.86) and (4.87). This potential is the summation of
the true equilibrium potential Equc6 and the diffusion overpotential due to diffusion
limitation of the electrode reaction of u+ ions inside the electrode and inside the
electrolyte. For the latter diffusion, the activity of Lt ions in (4.86) has to be taken
into account. This is shown in the following equation.
.
Euc. [1n( 1- x~:g )
o
+ -RT U neg X Lineg + ':>neg
r ]
eq
-_ E LiC6 - neg -
nuF Xu
(4.91)
As in (4.83), the second term in the last line expresses the overpotential due to mass
transport of u+ ions inside the electrode. The last term expresses the overpotential
due to mass transport limitation of u+ ions in the electrolyte.
4.3.4 The electrolyte solution

Lt ions are formed at one electrode in the considered Li-ion battery and injected
into the electrolyte at the electrode/electrolyte interface. These Li+ ions are removed
from the electrolyte at the other electrode and allowed to enter the electrode through
the electrolyte/electrode interface. The electrolyte of the considered CGR17500 Li-
ion battery consists of a LiPF6 salt dissolved in an organic solvent, as discussed in
section 4.3.1. As a result, u+ ions and PF6 ions co-exist in the electrolyte solution.
Both ions are transported inside the electrolyte under the influence of concentration
gradients through diffusion and under the influence of an electrical field through
migration. The influence of convection is neglected. The ionic conductivity of the
non-aqueous electrolyte used in Li-ion batteries is relatively low, as opposed to the
electrolyte in the NiCd battery considered in section 4.2. The conductivities of
typical non-aqueous electrolytes are in the range of 10"3 to 10"2 (Qcmf 1' whereas
typical conductivity values of aqueous electrolytes are in the range of 10"1 to 1
(Qcmf 1 [43]. Therefore, the electrical field inside the electrolyte cannot be
neglected in the Li-ion model, as was done in the NiCd battery model.
The total current that flows through the electrolyte is the sum of the diffusion
and migration contributions of both ion types, as defined by (4.19) for the chemical
domain. The diffusion coefficient Du·'1Y' of the Li+ ions in the electrolyte should be
used for the Li+ ions and zu•= 1. Likewise, the diffusion coefficient of PF6 ions in
the electro! yte D PF"6elyt should be used for the PF6 ions and ZPF"6 = -1. The movement
of the ions is such that the solution is electrically neutral anywhere inside the
electrolyte, except in the very thin Gouy-Chapman layers at the electrode surfaces.
Figure 4.20 illustrates the movement of ions in the electrolyte during the charging of
the Li-ion battery considered in this section.
Figure 4.20 illustrates the injection of Li+ ions into the electrolyte at the
positive electrode and their extraction from the electrolyte at the negative electrode.
The injection occurs at the rate of the LiCoOz electrode reaction, as discussed in
section 4.3.2, whereas the extraction occurs at the rate of the LiC6 electrode reaction;
see section 4.3.3. The PF6 ions are not allowed to cross either of the
electrode/electrolyte interfaces. The net current supported by these ions can become
zero because the contributions to the total current of the PF6 ions by diffusion and
migration have opposite signs. In that case, a steady-state situation is reached and all
current in the electrolyte is supported by Lt ions.
The self-discharge rate of Li-ion batteries is considerably lower than that of
NiCd and NiMH batteries; see chapter 3. However, it cannot be neglected. The
origin of self-discharge in Li-ion batteries is not exactly known. It consists of a
reversible part, which implies a temporary capacity loss that can be recovered, and
an irreversible part, which is a permanent capacity loss that cannot be recovered.
The reversible part could be attributable to electronic conductivity of the electrolyte
solution, whereas the irreversible part could be attributable to electrolyte
decomposition.
In the ideal case, the electrolyte solution would be a perfect electronic insulator
and current flow inside the electrolyte would be supported only by ions. However,
the electrical resistivity of the electrolyte is not infinitively large in practice. On the
basis of an average self-discharge rate of 5% per month at room temperature (see
chapter 3), a rated nominal capacity of 720 mAh and a battery voltage of 4 V, an
equivalent resistor Rteak of 80 kQ is found to be present between the positive and
negative electrodes of the CGR17500 battery.
118 Chapter 4
u• _ Diffusion u•
UCoO:z-+C:OO.+e-+u• --+
Migration _
Diffusion
Migration
Figure 4.20: Schematic representation of the charging of aLi-ion battery, in which u+ and
PF6 ions are transported through the electrolyte by means of combined diffusion and
migration
4.3.5 Temperature dependence of the reactions

The temperature dependence according to the Arrhenius relation of (4.72) should be
taken into account for the reaction rate constants k; and the diffusion coefficients D;.
The temperature dependence given by (4.73) should be considered for the standard
redox potentials £'1. Moreover, the temperature dependence of self-discharge should
be considered. The temperature dependencies mentioned above have not yet been
taken into account in the present Li-ion model version.
4.3.6 The model

The simulation model for a CGR 17 500 Li-ion battery is shown in Figure 4.21. This
model is based on the general principles given in section 4.1. No thermal network is
shown. A network similar to the one in Figure 4.19b can be easily included, as soon
as the temperature dependencies have been added to the model equations. The
LiCo02 electrode model is shown on the left-hand side in Figure 4.21 and the LiC 6
electrode model on the right-hand side. The electrolyte model is shown in the middle
connecting the two electrodes. The self-discharge of the battery is modelled by R1eah
as described above. This resistor is connected between the positive electrode and the
negative electrode in the electrical domain.
The model of the CGR17500 battery shown in Figure 4.21 is built up almost
entirely in the chemical domain. The only components present in the electrical
domain are the ohmic series resistances Rucoo2 and Ruc6 of the electrodes and the
self-discharge resistor R1eak· This has been done to clarify how the mass transport
model of Figure 4.13 has been included in a battery model. No side-reactions have
been modelled, as stated in section 4.3.1, and hence only one reaction path is present
in each electrode.
Energy storage in the chemical domain is modelled by chemical capacitances
Cch• with i denoting the reacting species. The coupling between the electrical and
chemical domains is represented by ideal transformers, as described in section 4.1.
Both electrode reactions have been modelled in the same way. The kinetics of both
reactions have been modelled in the chemical domain. This means that the change in
free electrochemical energy !:J.G is now present across the anti-parallel diodes Dj;
see Table 4.2. One of the anti-parallel diodes represents the oxidation reaction and
the other represents the reversed reduction reaction. This has been indicated in the
figure.
UCo02electrode electrolyte LiC 6 electrode
A leak
Electrical domain
...,
.+,middle
Jchllrge
+---
Chemical domain
Ja.··l
x=O
Spalilllelemanl1
Spalilllelemanl2
Spatial element p
Figure 4.21: Simulation model in the electrical and chemical domains of a CGR17500 Li-ion
battery
The species Co02, LiCo02 and u+ ions take part in the reaction at the LiCo02
electrode, as expressed by (4.75), and therefore the chemical capacitances that
contain the molar amounts of these species are present in the LiCo02 electrode
model. All j.l' values have been omitted in the chemical potentials IJ across the
capacitances of the Co02 and LiCoOz species, because these values are already
120 Chapter4
represented in source LJCflucoo2; see (4. 77). Moreover, the term ( -Upos xdos +'os) has
been added to the description of the capacitances of the Co02 and LiCo02 species to
obtain (4.83).
The diffusion of Li+ ions inside the electrode is described by an RC-ladder
network. The electrode has again been divided into p spatial elements each with a
thickness /Jx. The thickness of the electrode is lpos· The capacitances on the left-hand
side contain the molar amounts of the Co02 species in the spatial elements and the
capacitances on the right-hand side contain the molar amounts of the LiCo02 species
in these elements. The top of the RC-ladder network denotes the
electrode/electrolyte interface at x=O, where the actual electrochemical reaction
takes place. The bottom denotes the electrode/current collector interface at x=lpos·
The diffusing Li+ ions can pass the electrode/electrolyte boundary only in the
electrochemical reaction; they cannot pass the electrode/current collector boundary.
The chemical resistances Rchcoo2 and Rchucoo2 are given by (4.21c), with the
diffusion coefficient Du•pos of the Lt ions in the positive electrode being used for
both resistances. Again, the value of the resistances depends on the molar amount m
of the Li+ ions in the spatial elements according to (4.21c), which equals the molar
amount of the LiCo02 species. This means that resistances Rchcoo2 and RchLiCo02 have
the same values in each spatial element. Different spatial elements will have
different values for Rc/002=Rchucoo2 when a diffusion profile of Li+ ions occurs.
Such a diffusionprofile will give rise to a diffusion overpotential rfucoo2 across the
electrode/electrolyte interface, as defined by (4.83). The relation between the change
in free electrochemical potential AG across the diodes Ducoo2 and chemical flow
lch is described by the Butler-Volmer equation in the chemical domain; see (4.12).
The exchange current f' ucao2 is given by (4.82).
The LiC6 electrode has been modelled in the same way as the LiCo02
electrode. The electrode has again been divided into p spatial elements for
simplicity, with the electrode thickness having been defined as lneg· The number of
spatial elements does not have to be the same as in the positive electrode, because
the electrode thicknesses or grain sizes do not have to be the same. The diffusion
coefficient Du•"'B of the Lt ions in the negative electrode is used for the chemical
resistances. The exchange current for the anti-parallel diodes is given by (4.90).
The electrolyte has been modelled in the chemical domain, as described in
section 4.1.3 and shown in Figure 4.13. The electrolyte has been divided into m
spatial elements, as shown in Figure 4.21, where the electrolyte thickness is l,1yr· The
electrolyte model comprises capacitances Cchu• and Cc/1"6, described by (4.3), as
well as the geometric capacitances Cch e defined in Figure 4.11. The resistances Rchu•
and Rc/1"6 are described by (4.21c), with the resistance values again depending on
the spatial element. The diffusion coefficients Dpf' /yrand D,.;• etyr of the PF6 ions and
Li+ ions in the electrolyte are used in the resistance expressions, respectively. The
resistances Rchu• and Rc/1"6 have different values in the same spatial element when
the diffusion coefficients of the ions differ. The top part of the network describes the
mass transport of the PF6 ions as a result of diffusion and migration, while the
bottom part describes the mass transport of the Li+ ions as a result of diffusion and
migration. The Lt ions are allowed to cross the electrode/electrolyte interfaces at
both the LiCo02 electrode at x=O and the LiC6 electrode at x=letyr· This is
represented in Figure 4.21 by the connections between the capacitances Cchu• and
the electrode reactions at x=O and x=lelyr· The PF6 ions are not allowed to cross these
boundaries because they do not participate in the electrochemical reactions.
The geometrical capacitances Cch e.edge denote the double-layer capacitances C 1
in the chemical domain; see section 4.1.3. The chemical capacitance equals the
electrical capacitance C 1 divided by (nuF/, as expressed by (4.17). This is indicated
in Figure 4.21 for both electrodes. The electrostatic potentials at the
electrode/electrolyte interfaces at x=O and x=le1y, expressed in the chemical domain,
are also indicated in the figure. A chemical electrostatic potential difference of
nuF( ¢1ucoo2 -¢fucoo2 ) is present across the chemical double-layer capacitance of the
positive electrode. This chemical potential difference is nuF( ¢1uc6 -¢fuc6 ) for the
negative electrode. The electrostatic potentials ¢1ucoo 2 and ¢1uc6 at the electrical
ports of the ideal transformers are present at the indicated nodes. The electrostatic
potential at the node at which the two ideal transformers are connected equals the
electrostatic potential tPeLytmiddle in the middle of the electrolyte, because the
electrolyte model is completely symmetrical. This can be easily envisioned by
splitting the battery in half. The voltage across the left-hand ideal transformer covers
the positive electrode potential and half of the potential across the electrolyte. The
right-hand ideal transformer covers the potential across the other half of the
electrolyte and the negative electrode potential.
The direction of a charge current !charge is indicated in the electrical domain in
Figure 4.21. The currents through the electrical ports of both ideal transformers will
always be the same because of the network's structure. The direction of the chemical
flow lcharge in the chemical domain is shown for a steady-state situation. This means
that the electrostatic potentials and the diffusion profiles remain unchanged. ]charge
then flows through the chemical capacitances of the Co02 and LiCo02 species at the
electrode/electrolyte interface at the positive electrode and the LiC 6 and C6 species
at the negative electrode. The molar amounts of Co02 and LiC6 increase, as is
reflected by the polarity of these capacitances, whereas the molar amounts of
LiCo02 and C6 decrease. The current flows only through the chemical resistors Rchu•
in the electrolyte because of the steady-state situation. No current flows through the
chemical capacitances containing the molar amount of Li+ ions in the electrolyte at
the interfaces at x=O and x=lelyt because the diffusion profile of Li+ ions in the
electrolyte remains unchanged in a steady-state situation.
As a further illustration of the theory presented in section 4.1, the model of
Figure 4.21 will be represented in the electrical domain only. This can be done as
described in section 4.1.2. In Figure 4.22, the network for the chemical domain has
been transferred to the electrical domain by 'shifting' it through the ideal
transformers. Therefore, all impedances have been multiplied by N2 =llr, as nu=l.
Moreover, a distinction between the true equilibrium potential and the
diffusion overpotential due to diffusion of the Li+ ions in the electrodes, as described
by (4.83) and (4.91), will be made for both electrodes. This was illustrated above in
Figure 4.10. The results of this exercise are shown in Figure 4.23. A capacitance
describing the bulk activities of the Ox and Red species has been added to the
network in both electrodes (labelled 3 for LiCo02 electrode and 4 for LiC 6
electrode). In addition, the RC-ladder networks describing the diffusion of Li+ ions
inside the electrodes have been changed slightly by moving the chemical resistances
to one side of the RC-ladder network. This has been done to show that the electrical
potential inside the electrode hardly changes thanks to the good conductivity of the
electrodes. This can be inferred from the direct connection between the
122 Chapter4
electrode/current collector interfaces at x=lpos and x=lneg• respectively, and the

electrode/electrolyte interfaces at x=O.
LiCoO, electrode electrolyte LiC, electrode
X= 1,., _.-J H 1---

Figure 4.22: Simulation model of the CGR17500 Li-ion battery in the electrical domain only.
The model is based on that shown in Figure 4.21, as indicated in the text
The networks shown in Figures 4.22 and 4.23 are identical, as can be understood by
applying Kirchhoff's voltage law to the interface between two spatial elements in
the RC-ladder network. The circular arrows indicate this. Figure 4.23 shows that the
electrostatic potential inside the electrode is constant, which means that the same
potential is valid at x=O and x=lpos or x=lneg• and that it hence equals ¢1. The
electrochemical potential may vary throughout the electrode due to concentration
gradients. The resulting diffusion overpotential can only be measured at the
electrode/electrolyte interface.
The application of Kirchhoff's voltage law to both electrodes in Figure 4.23
illustrates that the voltage across the electrode/electrolyte interface, ¢1-¢1, equals the
sum of various contributions. They are the true equilibrium potential E:q, the
diffusion overpotential due to diffusion of Lt ions inside the electrodes 1Jd, the
kinetic overpotential r/ across the diodes Dj and the voltage across the capacitance
describing the activity of the Li+ ions at the interface. This latter voltage is in
agreement with the last term in (4.83) and (4.91). The total battery voltage can hence
be obtained from:
vbat = IR LiCo02 + IR uc. + E UCo02 + E elyt - E uc.

= 1Rucoo2 + 1Ruc + (¢Lco0 2 - ¢'~iCo02 )
6
+ (¢'~iCo0 2 - ¢'Lc (¢Lc ¢'Lc.)

6 )- 6 -
(4.92)
f.l u•-)
+'f].k +-m_,
uc. F
nu
where Eetyt is the electrostatic potential across the electrolyte.
LiCo02 electrode electrolyte LiCselec:trode
x=O x=O
4 +--
IJ.~o0 2- J..l~oo 2 - + - +
nuF
J.l~ieoo 2- Jl~eoo 2
nuF
jl~- Jl~iCoO?
nuF
J.l~1c 6- J.l~1c 6
nuF
YH
Figure 4.23: Alternative simulation model based on that shown in Figure 4.22 with modified
definitions of capacitances to allow a distinction to be made between the true equilibrium
potential and the diffusion overpotential due to diffusion of u+ ions in the electrodes.
Moreover, the RC-ladder networks in both electrodes have been rearranged
124 Chapter4
An expression for the overpotential due to diffusion of u+ ions in the electrolyte can
be obtained from (4.92):
(4.93)
where aJ,L;• and am,u• are the activities of the u+ ions in the spatial elements 1 and m
of the electrolyte, respectively.
Until now the focus has been on the construction of electronic-network models
for rechargeable batteries in general (section 4.1) and NiCd (section 4.2) and Li-ion
batteries (this section) in particular. The parameters in these models have to be given
proper values before simulations can be performed. This will first be discussed in
the case of the NiCd model in the next section, in which simulation results will be
compared to measurement results. More simulation results with the NiCd model will
be described in section 4.5.1. The parameters of the Li-ion model described in this
section will be given in section 4.5.2, followed by a comparison between the results
of simulations and measurements.
4.4 Parameterization of the NiCd battery model
4.4.1 Introduction
The simulation results obtained with the battery models described in the previous
sections must of course agree as well as possible with results obtained in practice.
So the results of the simulations must be quantitatively compared with the results of
measurements performed under the same conditions. This section describes the first
results of research performed to obtain optimum quantitative agreement between the
results of simulations using the NiCd model and measurements [49].
The NiCd model described in section 4.2 is characterized by a chemical and
mathematical description of the processes involved, the structure of the simulation
model and a set of parameter values. In order to obtain close agreement between
simulated and measured battery behaviour, both the design of the model itself and
the values of the parameters should be considered. The iterative process for
optimizing the battery model to obtain close agreement between the results of
simulations and measurements is schematically outlined in Figure 4.24.
The first three parts on the left-hand side of Figure 4.24 represent the steps in
the model development outlined in previous sections. In the last two parts, the
results of simulations are obtained and compared with the results of measurements.
The quantitative agreement between the results of simulations and measurements
can be improved in several ways on the basis of the results of the comparison. This
is represented by the feedback loops shown on the right-hand side of Figure 4.24.
First of all, the chemical description can be modified, for example by adding to
the model the description of a process that was not considered before. This is
represented by the feedback loop that points at the first step in Figure 4.24.
Secondly, the mathematical description of the considered processes can be modified,
as represented by the second feedback loop. Both modifications can be said to
modify the model itself. Thirdly, the values of the parameters in the model can be
optimized, leaving the original structure of the model intact. This is represented by
the third feedback loop.
Three approaches can be used to find parameter values for the battery model, as
indicated in Figure 4.24. In the first approach, the parameter values are obtained
from the literature, such as the number of electrons that take part in the reactions,
standard redox potentials, etc. In addition, some design-related parameters may be
obtained from battery manufacturers or data books. However, the values of some
parameters cannot be found in the literature or have been obtained under conditions
that do not hold for the batteries considered here. In that case, the second approach
may be useful.
Define chemical
description of
battery processes Modify chemical description
+
Find mathematical
description of
considered processes Modify mathematical description
+
Translate processes
into equivalent
network model
+
Simulate
Modify parameter values using:
1. Literature
+ 2. Parameter measurements
3. Mathematical strategy; see section 4.4.2
Compare the results
of simulations with
measurements
Figure 4.24: Set-up of the modelling strategy and iterative process of parameter optimization
and model modification aimed at improving the quantitative agreement between the results of
simulations and measurements
The parameter values are found by means of measurements in the second approach.
It should be noted that most measurements of battery parameters, such as exchange
currents and diffusion coefficients, take a considerable amount of time, because the
parameters have to be obtained from a series of voltage and current measurements,
which have to be performed on the separate electrodes. Some parameter values will
be difficult to obtain even by means of measurements.
A third approach will often have to be used, because not all battery model
parameters can be found via the first two approaches. This third approach will be the
main focus of this section. It involves the application of a mathematical strategy,
based on the comparison of battery curves obtained from simulations and
measurements. In principle, the entire set of battery model parameters can be
obtained using this strategy. However, the complexity of the approach increases
126 Chapter4
with the number of parameters to be obtained, so this number should not be too
high. The strategy will be explained in more detail in section 4.4.2.
Most parameter values in a battery will show a certain spread in practice. This
means that whichever of the three above-mentioned approaches is applied, a set of
parameter values obtained for a certain battery will not necessarily hold for all
batteries of that type. An optimized parameter set will be derived for a single battery
in this section.
The results obtained from applying the strategy of section 4.4.2 will be
described in section 4.4.3. The results of the simulations obtained with the optimized
battery model will be compared with results obtained in measurements in a charging
experiment using a single charge current, after which the optimized parameter
values will be listed and discussed. The applied mathematical optimization strategy
will also be discussed. The quantitative agreement between the results of simulations
using the optimized battery model and the results of measurements at various charge
currents will be discussed in section 4.4.4. A change in the model based on these
results will be suggested in the form of a modified mathematical description, as
indicated in Figure 4.24. Finally, in section 4.4.5, the strategy of section 4.4.2 will
again be applied at various charge currents using a modified model.
4.4.2 Mathematical parameter optimization

The battery model can be regarded as a black box in the mathematical strategy
described in this section. This interpretation of the battery model is shown in Figure
4.25.
Parameters
Input variables Output variables

Battery model
black box
Figure 4.25: Black box interpretation of the battery model
The black box processes a number of input variables and yields a number of output
variables. Parameters determine the relationship between the input and output
variables. The strategy comprises three main parts, notably (i) defining an
admissible parameter space, (ii) identifying a suitable sub-region within this
admissible parameter space and (iii) finding the optimum parameter set within the
resulting reduced parameter space. Although the battery model is primarily regarded
as a black box, knowledge of battery behaviour will be used in the strategy.
Examples can be found in the explanation of the three main parts of the strategy
below.
The strategy was used in simulated and measured charging experiments
involving the NiCd model and a Panasonic P60AA NiCd rechargeable battery,
respectively. The design of the NiCd model was based on that of the type of NiCd
battery used in the measurements, as described in section 4.2. In the charging
experiment, the battery model and the real battery were charged at a 1 C charge rate
for two hours at 25°C; see chapter 3 for the definition of C-rate. The simulations
were performed using the electronic-circuit simulator PSTAR©, developed at Royal
Philips Electronics N.Y. The pressure inside the battery was measured during the
experiments with a pressure transducer (Transamerica Instruments, type no. 4 702-
10) mounted in a hole drilled in the bottom of the battery casing. The battery
temperature was measured with a Pt-100 thermocouple placed on the bare metal on
the long side of the AA-size casing. More information on the three parts of the
strategy is given below.
(i). Defining an admissible parameter space

The input and output variables and the parameters to be taken into account are
identified in the first part of the strategy. The input variables include the current
Ucharge) that flows into the battery and the ambient temperature (Tamb)· The output
variables include the battery voltage (V), temperature (7) and internal oxygen gas
pressure (P), and the derivatives of these variables. The derivatives are included
because the V, P and T profiles show clear curvatures during overcharging. The
inclusion of derivatives in the output variables makes it easier to compare the results
of simulations with those of measurements. Another reason for including the
derivatives in the output variables is to minimize the number of extreme results
obtained in the simulations in the next two steps of the strategy.
As mentioned above, the number of parameters considered should not be too
large. Therefore, the parameters with fixed values, such as the number of electrons
that take part in a reaction, or with only a limited influence on the output variables in
the considered experiment are to be kept constant. They do not participate in the
optimization process. The influence of the parameters on the output variables was
determined in a sensitivity analysis. In addition, knowledge of battery behaviour was
used and, for example, parameters with no or only a negligible influence on the
charging behaviour were omitted. A certain range is defined for each parameter
considered on the basis of the assumed possible spread in each parameter's value. 34
parameters were selected as variables in the present optimization process, yielding a
34-dimensional parameter space. Within this parameter space, 340 parameter sets
were generated by randomly but uniformly spreading the values for each of the 34
parameters throughout the specified range. This was done using the Latin Hypercube
method [50], which was improved by further spreading the points in the parameter
space with the aid of a simple dedicated software tool to minimize differences in the
distances between the neighbouring points.
The admissible parameter space was defined in a charging simulation for each
parameter set, as described above. The simulated V, P and T curves were visually
inspected and those parameter sets that yielded highly unrealistic results were
discarded. Roughly 10% of the original number of parameter sets were discarded
because of this reason. This is another example of using battery knowledge in the
mathematical process. The model behaviour thus obtained for parameter sets within
the admissible parameter space was already in fair agreement with the actual
battery's behaviour.
(ii). Identifying a suitable sub-region I reducing the parameter space

In the second step, the simulated output variables of the battery model were
compared with the corresponding measured variables of the real battery. This was
done for each parameter set within the admissible parameter space defined in the
first step. The basic principle of the second step is illustrated in Figure 4.26.
Figure 4.26 illustrates that each simulated and corresponding measured output
variable, Var;, was compared using a cost function, CF;, given by:
128 Chapter 4
=L ~.J (Vm;,sim,J - Va'i,meas,J J

N
CF; (4.94)
J=l
where i ranges from 1 to 6, because six output variables are being considered. The
factor W;J denotes the normalizing and weighing factor for the output variable Var;.
Simulations Measurements
1. Perform simulation 2. Compare output

for each parameter variables for each
set in admissible simulation with
parameter space measured output
variables
Par1 P~':. rk Par34

Input Output variables Output variables Input
variables Var1,sim• i = 1...6 variables
Icharge lcharge
Var1 =V
Var2 = dV/dt 3. Find sub-region in admissible
Var3 =T parameter space where all CFi
Var4 = dT/dt have minimum values by limiting
Var5 =P the original ranges of the parameter
Var6 =dP/dt Par1 ... Par34 (see Figure 4.27)
Figure 4.26: Schematic representation of the second step in the optimization process
As can be seen in (4.94), the cost function is a simple Least-Square Error (LSE)
function that is weighed and normalized for each of the six output variables. The
index j increases from 1 at the beginning of the charging curves to N at the end of
the charging curve. N was chosen to be 7200 and the points at which the
comparisons were made were each spaced 1 second apart. The use of a quadratic
cost function is important, because positive and negative differences are weighed
equally.
The output variables need to be normalized because of the large differences in
the absolute values of the six output variables. The cost functions need to be of the
same order of magnitude when the cost functions CF; are combined into a single
cost function in step (iii) below. The variables were normalized by dividing the
squared difference between the simulated and the measured values, shown in (4.94),
by the squared difference between the simulated value and the average measured
value. The average measured value is a single value that represents the average of all
the values measured along the entire curve.
The values were weighed so as to be able to concentrate on certain parts of the
curves. In the case of battery charging curves, the most pronounced behaviour in
terms of V, P and T occurs after one hour of charging. Therefore, we focussed our
attention on the points in the overcharging region in the case of each output variable
Var; by increasing the weight factor towards the end of the curve for each output
variable. This is yet another example of the use of battery knowledge in the
mathematical strategy. A consequence of this, however, is that the battery behaviour
at the beginning of the charging will not be optimized to the same extent as the
behaviour during overcharging.
The ranges of the 34 parameters Par1•. Par34 were further reduced on the basis
of the comparison of the results of simulations and measurements for all parameter
sets in the admissible parameter space. This was done to reduce the size of the
original parameter space. The new ranges were chosen so as to minimize the cost
functions CF; in the new sub-region. The results of the simulations thus obtained for
the parameter sets in the new sub-region were consequently already quite close to
the measurements. This close agreement was achieved for all six defined output
variables, as all cost functions CF; already have minimum values in this reduced
parameter space.
The basic principle of the reduction of the range of parameters Park> k=1..34, is
schematically represented in Figure 4.27. Six cost functions, CF;, were calculated
for each parameter set. Supposing that there are M parameter sets, then in the end M
values of CF1 are available, M values of CF2 , etc. The values of CF1 can then be
ranked, as can the values of CF2o etc. For each cost function, the parameter sets that
yield the lowest values for CF; can then be identified. This information can be used
for the actual reduction of the ranges of the 34 parameters. In the situation
illustrated, the five best parameter sets were selected for each cost function. The
values of parameter Park in these parameter sets are shown along the horizontal lines
at the bottom of Figure 4.27. Note that the five best parameter sets of different cost
functions are not necessarily the same. This can be inferred from the fact that there
is no vertical line with six dots on it. The new range is defined as the region in
which the greatest number of parameter values for Park are to be found, as shown in
the figure. Parameter values within the new range will now yield minimum values
for all six cost functions, CF;. The new range was obtained by considering the
uniform distribution of Park within its original range.
(iii). Finding the optimum parameter set within the reduced parameter space
In the third step of the strategy, the optimum parameter set is found within the
reduced parameter space defined in the previous step. This is done via a standard
minimization of a cost function by means of the commonly used simplex method.
The cost function comprises the cost functions for the individual output variables as
follows:
6
Costfunction = LCF: (4.95)
j;l
In our experiments, we performed the third step with the Optimize© optimization
software developed by Royal Philips Electronics N.Y. Starting from a parameter set
130 Chapter4
in the reduced parameter space obtained in the previous step, the software minimizes
the cost function in (4.95) by performing many simulations involving different
parameter sets in an iterative process. The software stops when the cost function
changes remain within a certain preset window. Note that all six output variables are
considered simultaneously in this step.
Results of comparisons of the output variables of

simulations and measurements for each parameter set
in admissable parameter space (see Figure 4.26):
Par. set Cost functions for each parameter set:

1 CF11 CF21 CF31 CF41 CFs1 CFe1
2 CF12 CF22 CF32 CF42 CFs2 Cfe2
... ... ... ... ... ... .. .
M CF1M CF2M CF3M CF4M CFsM CFeM
t
Ranking cost functions for each output variable separately:
Ranked cost functions

(Lowest values at the top of the table)
CF1 ... CFe
1-
Low: CF13a ... CFass
CF1245 ... Cfe1oo
CF113e ... CFe13
Mrows
CF11s ... CFa210
CF17a ... CFe1o1
... ... .. .
High: CF1314 ... CFe2a
-
t
Values of Park in the five parameter sets yielding the five lowest
values of the cost functions CF;
Minold
... .•••_. • • •
Cost
functions:
CF1
CF2
CFa ••
• • ··-· _.
•••
,.__ Example: Values of Park
in the parameter sets
38,245,136,15 and 78
CF4
__.
•-··
CF5
CF8 •• ,...._ Example: Values of Park
in the parameter sets
New range 55, 100,13,210 and 101
Minnow Max08w
Original range
Figure 4.27: Simplified schematic representation of the definition of a new range for
parameter Part. In the last step shown at the bottom, the values of Park in five parameter sets
that yield the five lowest values of cost function CF; are shown on each horizontal line
4.4.3 Results and discussion
Comparison ofthe results of simulations with the results of measurements

The final results obtained with the mathematical strategy described in the previous
section are shown for the simulated and measured V, P and T curves in Figure 4.28,
Figure 4.29, and Figure 4.30, respectively. These figures all show the measured
curves, labelled meas, plus the curves obtained in the simulation in which the
optimum parameter set obtained with the strategy described in the previous section
was used. The latter curves have been labelled opt. The curves that were obtained in
the simulation using the initial parameter set are also shown. These curves have been
labelled start. These parameter values can also be found in [26], [27] and [28].
1.6
t
vopt
1.5 vmeas
~
>
1.4
1.3
1.2
0.0 0.5 1.0 1.5 2.0
t [hours] ------+-
Figure 4.28: Simulated (V,.,.rr [V] and Vopt [V]) and measured <Vmeas [V]) battery voltage
versus time (t [hours]) for charging a NiCd P60AA Panasonic battery at a 1C-rate for 2 hours
at 25°C ambient temperature. The battery model developed in section 4.2 was used in the
simulations. In addition to the quantitatively optimized simulated voltage <Vop1), the voltage
curve at the time when the mathematical strategy was first applied using the parameter set
found in [26]-[28] is also shown (V,1ar1)
Figure 4.28 shows that the results obtained in mathematical optimization represent a
significant improvement over the V,1art curve. For charging times larger than roughly
one minute, the maximum difference between the optimized simulated voltage curve
Vopt and the measured voltage curve Vmeas is only 18 mY, which amounts to 1.4% in
the case of a nominal battery voltage of 1.25 Y. The difference between the results
of simulations and those of measurements seems reasonable, because the measured
curve, Vmeaso will show a certain spread in practice. There are several reasons for
this. First of all, the measurement accuracy is finite. In this case, allowance must be
made for a voltage measurement resolution of 1 mY. This means that the difference
observable in Figure 4.28 is greater than can be explained by measurement accuracy
only. Secondly, the battery voltage measured during the charging of several NiCd
batteries of the same type will display a certain spread. Unfortunately, no reliable
statistical data is available on this spread. Therefore, it is at present not known
132 Chapter4
whether the difference between V meas and Vopt observable in Figure 4.28 is within the
spread of battery behaviour in practice.
4.0
t 3.0
~
"'0.... 2.0
a.
1.0
0.0
0.0 0.5 1.0 1.5 2.0
t [hours] ~
Figure 4.29: Simulated <Psran [105 Pa] and Pop~ [lOS Pa]) and measured (Pmeas [105 Pa])
internal oxygen gas pressure versus time (t [hours]). See Figure 4.28 for conditions
The simulated and measured battery voltage curves Vopt and V meas are more or less
identical in shape, except for the parts representing the first 20 minutes. First of all,
it should be noted that weighing has been applied, which may have led to greater
discrepancies between the results of simulations and measurements at the beginning
of charging. This can be seen especially in the first few minutes. This difference
may also be attributable to processes occurring in the measured battery for which no
allowance has been made in the model. A possible example of such a process is
hysteresis of the battery's equilibrium potential. Measurements have shown that the
equilibrium potential of a NiCd battery depends on whether it is determined after the
battery has been charged or discharged [30]. In the latter case, the equilibrium
potential is lower than in the former case. As the battery was charged after a
discharging step in the measurements, its equilibrium potential will have changed
from a low value during the discharging to a higher value during the charging. This
could explain the differences between the simulated and measured voltage curves at
the beginning of the charging, because the battery model does not allow for any
occurrence of hysteresis yet. When we take a closer look at this difference we see
that the optimized curve obtained with the model is indeed higher than the measured
curve. No appropriate mathematical description of the hysteresis of the equilibrium
potential is as yet available. Further research will have to be carried out to check
whether the inclusion of hysteresis in the model will improve the quantitative
agreement between the results of the simulations and the measurements. Note that
this is an example of model modification, as shown in Figure 4.24.
Figure 4.29 shows that the P opt curve represents a considerable improvement
over the Psrarr curve. The difference between the end values of the simulated
optimized curve Popt and the measured curve Pmeas is 18 kPa or 0.18 bar. This
difference amounts to 9% of the measured end value of 200 kPa. As with the voltage
curves, the two potential causes of spread in the measured curve have to be taken
into account. A resolution of I kPa holds for the measurement accuracy, so the
difference observable in Figure 4.29 cannot be explained by measurement accuracy
only.
t 40.0
0
~
I-
20.0
0 . 0 + - - - -.......----.------"""T'"-----!
0.0 0.5 1.0 1.5 2.0
t[hours] -
Figure 4.30: Simulated (Tstart [ 0 C] and Topt [°C]) and measured (Tmeas [°C]) battery
temperature versus time (t [hours]). See Figure 4.28 for conditions
The difference between the optimized and measured pressure curves Popt and Pmeas
during overcharging suggests that the equilibrium between oxygen production and
recombination was reached at different oxygen pressures in the simulations and the
measurements. Many parameters have to be considered in calculating a battery's
behaviour in terms of oxygen production and recombination. It therefore seems
reasonable to assume that the observed difference is partly attributable to incorrect
parameter values. The P meas and P opt curves are identical in shape and the point at
which the oxygen pressure starts to rise is the same in both curves. Moreover, the
shape of the simulated Popt curve is much better than that of the Pstart curve during
overcharging. Nevertheless, the discrepancy observable between the results of the
simulations and the measurements prompted further research into the kinetics of the
oxygen reactions taking place inside the battery [30].
Another improvement is observable between the simulated Tstart and Topt curves
shown in Figure 4.30. However, the difference between the Topt and Tmeas curves still
amounted to 7°C at the end of the charging. This difference cannot be explained by
the accuracy of the measurements which is within 0.1 °C. The present one-
dimensional model for the thermal battery behaviour is probably far too simple. The
battery temperature is assumed to be homogeneously distributed throughout the
entire battery in the present model. The thermal model involves only one heat
capacitance and one heat resistance to the outside world; see Figures 4.15 and 4.19.
Taking the parameters for heat capacitance and heat resistance into account in the
optimization process did not lead to further improvements. Further research is
needed to investigate the influence of using distributed networks on the quantitative
agreement between simulated and measured temperatures. For such a network, the
battery must be divided into several parts and each part must be able to have its own
temperature and will have to be modelled separately. It will then be possible to
134 Chapter 4
consider temperature gradients inside the battery, too. The difference between the
simulated and measured battery temperatures will also be partly responsible for the
observed differences between Vopt and Vmeas and between Popt and Pmeaso because V
and P are temperature-dependent both in the model and in practice.
Discussion of the parameter values

The parameter values of the NiCd model are summarized in Tables 4.3 and 4.4.
Some of the battery parameters shown in these tables were inferred from orthogonal
model parameters. Parameters are orthogonal when they can be varied
independently, which means that a variation in one parameter does not lead to a
change in another parameter. The battery parameters have been listed so that they
can be understood and compared with measured values more easily. An example is
the double-layer capacitance of the two electrodes given in Table 4.4. The values of
these double-layer capacitances have actually been calculated using the following
simple equation from the model:
(4.96)
where C0 dl denotes the capacitance per area in [F/m2] and A the area of the electrode
in [m2].
Table 4.3 lists the parameters that were not varied in the optimization process.
The values were obtained from the literature and from the supplier of the P60AA
NiCd battery. They include design-related and electrochemical parameters. As a
reference to earlier sections, the fourth column lists the equations or figures in which
the parameters were used. There where parameters have been used in more than one
equation, the first equation in which they were used is referred to in the tables. The
appropriate section number is given there where a parameter was only mentioned in
the text. The parameter values given in Table 4.3 remained unchanged throughout
the simulations and therefore hold for both the start and the opt curves shown in
Figures 4.28, 4.29 and 4.30.
Table 4.4 shows the parameters considered in the application of the strategy
described in the previous section. For example, the double-layer capacitances em
have been inferred from (4.96). In the optimization process, the value of C/1
remained fixed at 0.2 F/m2 for both electrodes [27], whereas the areas A were varied
separately for each electrode. Consequently, the ratios of the starting values and the
optimized values are the same for A and ~1 for each of the two electrodes. All the
parameters were orthogonal in the actual parameter sets used in the application of
the strategy.
The parameters given in Table 4.4 have again been split up into design-related
and electrochemical parameters. The second column shows the starting values.
These values are in accordance with those presented in previous publications on the
NiCd model [26]-[28] and were used for all the curves labelled start in Figures 4.28,
4.29 and 4.30. The third column shows the optimized parameter values used for all
the curves labelled opt in Figures 4.28, 4.29 and 4.30. For validation purposes, the
available measured values have been included in the fourth column. These
measurements were obtained with an 80AAS NiCd battery, which differs from the
type of battery considered in this book, but is based on the same battery system [30].
However, the indicated values serve merely to place the optimized parameter values
in a realistic perspective. Finally, the sixth column lists the equations or figures in
which the parameters have been used, using the same procedure as for Table 4.3. It
should be noted that all temperature-dependent parameters, which are all Dt, E;0 , and
It parameters, are specified at 25°C. The temperature dependence of these
parameters is determined by the corresponding value of the activation energy E'. All
!;0 values are specified at 50% SoC, while all/~0 values are specified at the reference
pressure of 105 Pa=l bar.
Table 4.3: Fixed parameter values for the NiCd model described in section 4.2 which were
not considered in the parameter optimization process
Parameter Value Unit Reference
Design-related parameters
m2 (4.34)
A bar 2.7·10" 3
c,hbal 14 J/K Figure 4.15
Qed. Max 5000 c Figure 4.16
Qed(OH)z.Ni 1000 c Figure 4.16
Qed. ed 800 c Figure 4.16
R bar 14 KIW (4.34)
th
RNi 2 mQ Figure 4.19
Red 2 mQ Figure4.19
Electrochemical parameters
(f'o)el 1.0·10"3 A Section 4.2.4
(f'o)e/ 1.0·10"3 A (4.66)
Med 112.4 g/mol (4.54)
nNi 1 - (4.45)
ned 2 - (4.52)
noz 4 - (4.61)
Ped 8.6·10 6 g/m3 (4.54)
Different exchange currents have been used in Tables 4.3 and 4.4 for oxygen
evolution and recombination occurring at the two electrodes, (f'o)el and (f'o)e/ in
Table 4.3 and (f'o)NiR and (f'o)N; 0 in Table 4.4; see Figure 4.16. This was explained in
section 4.2.4. In Table 4.3, the two exchange currents have been given the same
value, which remained constant in the simulations [27), whereas they have different
values in the starting and optimized parameter set given in Table 4.4 [27].
Not wishing to draw final conclusions from the optimized parameter values,
we would nevertheless like to make some general remarks. First of all, good
agreement with the measured parameter values is observable in the case of
parameters £" Ni• £' Do 2 , E' ed• E' Ni• p Ni• .1Sed and .1SNi• in other WOrds, seven Of the
eleven compared parameter values. The difference of a factor of two observed for an
r
exchange current such as Ni is not that great, especially when the range of two
orders of magnitude used for this parameter in the optimization strategy is
considered. The optimized values of parameters E'Ni• E'D 02 and PN; are particularly
close to the measured values given in Table 4.4, whereas the starting values differ
substantially from the optimized and measured values. This inspires great
confidence in the quality of the model and the approach that was used to find the
optimized parameter values.
136 Chapter4
Table 4.4: The parameters of the NiCd model described in section 4.2 which were considered
in the parameter optimization process. Values presented in previous publications [26]-[28]
and the values obtained using a 80AAS NiCd battery [30] have been included to enable
comparison
Parameter Value in [26], Value in Measured Unit Reference
[27] and [28] optimized value
parameter set
ANi 11.8 10.9 m2 (4.46)
Acd 23.3 32 m2 (4.53)
cNi
dl 2.4 2.2 F Figure4.19
cCd
dl 4.7 6.4 F Figure4.19
QNi,Max 3060 3724 c Figure 4.16
R, 10 9.1 mQ Figure 4.19
[Ni 5.5·10'6 58.3·10'6 m Figure 4.17
Vg 1.7•10'6 m3 (4.70)
1.5·10'6
Electrochemical parameters
ab01r 8.0·103 6.8·10 3 mollm3 (4.46)
dOz 3.0·10'7 5.4·10'7 m (4.67)
do~r 1.0·10'6 1.3·10·6 m (4.58)
D 0 .lu+ 5.0·10' 14 169.9·10- 14 m2/s Section 4.2.6
Do.2H+, - 80.8·10' 14 m21s Section 4.2.6
see note l
D"Oz 5.0·10' 10 9.6·10' 10 m2/s (4.67)
D 0oJr 5.0·10' 10 4.0·10' 10 m2/s (4.58)
E"cd 40.0·103 50.9·103 36.2·103 J/mol (4.72)
£"Ni 40.0·103 24.8·10 3 24.3·103 J/mol (4.72)
E"o2 40.0·103 76.8·10 3 J/mol (4.72)
E"ou+ 9.6·10 3 17.8·103 J/mol (4.72)
E"oo2 14.0·103 9.2·10 3 9.4·103 J/mol (4.72)
E"oau· 14.0·103 8.7-103 J/mol (4.72)

£"Cd -0.81 -0.83 -0.81 V vs SHE (4.52)
E"Ni 0.52 0.53 0.55 V vs SHE (4.45)
E"Oz 0.40 0.40 v VS SHE (4.61)
{' Cd 2.3·103 27.2·10 3 0.52 A (4.53)
f'Ni 36.0 0.82 0.33 A (4.46)
(f'o)NF 117.6•10'9 1.8•10'6 2.2·10·2, see A (4.65)
note 2
(f'ojN/ 0.5·10' 3 18.8·10'3 A Section 4.2.4
fXcd 0.5 0.55 - (4.53)
aNi 0.5 0.39 - (4.46)
1:02 0.5 0.23 0.06, - (4.65)
see note 2
LISCd, see note 3 -204 -161.9 -191 J/(mol.K) (4.73)
LISNi· see note 4 130.3 115.9 127.4 J/(mol.K) (4.73)
LISo2. see note 649.0 702.7 J/(mol.K) (4.73)
4
Note I: During the curve-fitting process, the diffusion coefficient D 0 u+ was made dependent on the local
mol fraction of H' ions (xu+) in the nickel electrode. The local value of xu+ inferred from capacitance C;
was used for each spatial element i in the RC-ladder network. The following dependence of D"u+ on Xu+
was used [35]:
Note 2: These values hold for sufficiently large current and overpotential values
Note 3: The sign holds for a reduction reaction, so during charging for the cadmium reaction at the
cadmium electrode [31]
Note 4: The sign holds for an oxidation reaction, so during charging for the nickel reaction and the
oxygen-production reaction at the nickel electrode [31]
Secondly, the parameters f' cd and E' cd differ significantly from the measured values.
A possible reason for this is the fact that in the first step the range of f' cd was chosen
such as to ensure that the overpotential of the cadmium reaction would be relatively
small, or in the order of 1 mV. This means that it was assumed that the kinetics of
the cadmium reaction are much faster than those of the nickel reaction. This small
cadmium reaction overpotential, and hence the parameters affecting its value, had of
course little influence on the overall battery voltage. This made it more difficult to
find correct values for these parameters, as they do not influence the battery voltage
curve to any great extent. It seems reasonable to assume that this is the reason why
the optimization program failed to find an exact value for E'Cd· The measured value
of f' cd shows that the assumption that the kinetics at the cadmium electrode are
much faster than those at the nickel electrode is not correct in the case of the 80AAS
NiCd battery. The measured value of the parameter f' cd is not even within the range
chosen for f' Cd in the simulations. Therefore, further research will have to be done
into the parameter values of the cadmium reaction of the P60AA type of battery
considered here.
Thirdly, a major difference between the optimized and measured values was
observed in the case of parameters (f'o)N; 0 and ao2 . As can be seen in Table 4.4, the
optimized value obtained for (f'o}Nt is lower than the measured value, whereas the
optimized value of ao2 is higher than the measured value. These two parameter
values determine the amount of oxygen that is produced at the nickel electrode
during overcharging. We used the anodic branch of the Butler-Volmer equation to
express the production of oxygen at the nickel electrode during charging, which is
repeated here for reference:
(4.97)
As can be inferred from (4.97), a high value off'02 in combination with a low value
of ao2 will mathematically yield the same value for l0z as a low value of f' 02 in
combination with a high value of Clo2 at the same overpotential 110z and at a constant
temperature T. Using the values given in Table 4.4, the combination of the
optimized values of f' 02 and ao2 and the combination of measured values yield
values for 110z of 378 mV and 377 mV, respectively, at the considered charge rate of
1 C and at a temperature of 318 K, which is the average temperature in the
138 Chapter4
overcharge region. Although the discrepancy between the measured and optimized
parameters for the oxygen evolution at the nickel electrode can be explained by
basic mathematics, this discrepancy prompted a more detailed study of the kinetics
of the evolution of oxygen at the nickel electrode. More information will be given in
the next section.
Discussion of the mathematical optimization strategy

Some remarks can be made on the applied mathematical strategy. First of all, the
low weight of the cost function at the start of the charging may have led to the
observed discrepancy between the simulation and measurement results. Further
research into the possibility of improving the quantitative agreement at the start of
the charging should shed more light on this matter. Including the occurrence of
hysteresis in the description of the equilibrium potential, as described above, could
also help if it should prove impossible to improve the agreement by means of
mathematical optimization only.
The mathematical parameter improvement method proved to be a good means
for focusing on certain parts of the model in which improvement is still needed. This
has sparked research in certain specific directions, such as the investigation of the
kinetics of the oxygen evolution reaction at the nickel electrode described above. So
all in all, the method has not only led to a significant improvement in agreement
between the results of simulations and measurements, but has also prompted further
research aimed at improving the model.
Although the model can primarily be regarded as a black box for mathematical
parameter improvement, as shown in Figure 4.25, (electro)chemical and battery
knowledge was used in several steps in the process. Using this knowledge is crucial
to prevent unrealistic results as much as possible. Moreover, several iterations were
performed to obtain the final results presented in this section. This means that some
guidance of the process is still needed. However, the mathematical parameter
optimization method can be readily applied to battery systems other than NiCd
systems.
The results presented in this section were obtained in a charging experiment
using one charge current only. In the next section, the quantitative agreement
between the results of simulations and measurements will first be investigated in the
case of the optimized parameter set given in Table 4.4 under various charging
conditions.
4.4.4 Quality of the parameter set presented in section 4.4.3 under different
charging conditions
Charging simulations have been performed at different currents in order to
investigate the quality of the model simulations using the optimized parameter set
presented in the previous section. The simulation results were compared with
measurements obtained at the same charge current. Figures 4.31 and 4.32 show the
simulated and measured V, P and T curves obtained after charging at a 0.1 C charge
rate at a 25°C ambient temperature. Figures 4.33 and 4.34 show similar curves
obtained at a 2 C charge rate at 25°C ambient temperature. In both cases, a 100%
overcharge was applied, as in the case of the 1 C curves presented in the previous
section. As can be inferred from the figures showing the curves obtained at these
charge currents, the quantitative fit between the simulated and measured curves is
worse than that obtained for the charge rate of 1 C applied in the optimization
process. Some remarks will be made on both cases below, concentrating on the most
prominent differences between the measurements and the simulation results.
1.6 0.6
vopt 0.5
t
1.5
Vmeas 0.4 t;f
~ 1.4 Pmeas 0.3 'b
> ;s
Popt a.
0.2
1.3
0.1
1.2 0.0
0.0 5.0 10.0 15.0 20.0
t [hours] ~
Figure 4.31: Simulated <Vopt [V]) and measured (Vmea.r [V]) battery voltage plotted along the
left y-axis and simulated (Popt [105 Pa]) and measured (Pmeas [105 Pa]) internal oxygen
pressure plotted along the right y-axis versus time (t [hours]) obtained in charging a NiCd
battery at 0.1 Cat 25°C ambient temperature using the NiCd model described in section 4.2 in
the simulations using the optimized parameter values given in the third column of Table 4.4
The simulated battery voltage during overcharging is higher than the measured
voltage in Figure 4.31. The measured oxygen pressure started to rise early in the
charging process already. In the simulation however, the oxygen pressure increased
quite abruptly after around 13.5 hours. The pressure levelled off to a steady-state
value of 21 kPa after approximately 15.5 hours. Figure 4.32 shows that the shapes of
the simulated and measured temperature curves are similar. However, the simulated
temperature at the end of the charging is somewhat higher than the measured
temperature, which was also the case in Figure 4.30.
The shape of the simulated battery voltage curve differs significantly from that
of the measured battery voltage curve obtained in the 2 C charging experiment in
Figures 4.33 and 4.34. The main difference concerns the position of the peak, which
occurred earlier in the charging process in the simulation. As will be explained in
detail in section 4.5, the peak in the battery voltage curve marks the beginning of the
production of oxygen at the nickel electrode and the reduction of oxygen at the
cadmium electrode. This can also be seen in Figures 4.33 and 4.34, in which the
internal oxygen gas pressure and the battery temperature started to rise earlier in the
charging process in the simulation. The value of the simulated battery temperature at
the end of the charging was again higher in the simulation.
An explanation for the differences between the results of the simulations and
the measurements described above can be found by studying the measurement
results given in Table 4.4 more closely. For this purpose, a Tafel plot of the current-
versus-overpotential measurements of the oxygen-evolution reaction occurring at the
nickel electrode is shown in Figure 4.35 [30]. A Tafel plot is a convenient way of
plotting current-voltage relationships of electrochemical systems according to
140 Chapter4
(4.97), because the value of a can be inferred from the slope of the plot, while the
value off' can be found by extrapolating the curve until the overpotential r/' is zero;
see chapter 3.
1.6 40
vop,
t t
1.5 35
vmeas
~ 1.4 30
IT
2.....
> I-
Topt
Tmeas
1.3 25
1.2 20
0.0 5.0 10.0 15.0 20.0
t [hours] ~
Figure 4.32: Simulated <Vopt [V]) and measured <Vmeas [V]) battery voltage plotted along the
left y-axis and simulated (Topt ["C)) and measured (Tmeas ["C)) battery temperature plotted
along the right y-axis versus time (t [hours]). See Figure 4.31 for conditions
1.6 4.0
t
1.5 3.0 t
~
~ 1.4
>
2.0 'b ...
Q.
1.3 1.0
1.2 0.0
0.0 0.25 0.5 0.75 1.0
t [hours] ~
Figure 4.33: See Figure 4.31; the charge rate was 2 C in this case
Figure 4.35 shows that the kinetics of the oxygen production reaction can be
described as a two-step process, with the values of a and f' differing for curves (a)
and (b). In fact, as mentioned in footnote 2 of Table 4.4, the measured values of
(I'o)Nio and ao2 in the table were derived at high current and overpotential values,
i.e. from curve (b) in Figure 4.35 [30]. A value of 0.26 for a.;2 and a value of 3.1·10-6
[A] for (f'o)N;0 can be obtained from curve (a). Curve (c) was obtained from curves
(a) and (b) in a calculation similar to that used for a mixed kinetic-diffusion
controlled reaction current; see (4.59) and (4.69). This means that:
(4.98)
1.6 80
t t
1.5 60
> 5'
1.4 40 L
> 1-
1.3 20
1.2 0
0.0 0.25 0.5 0.75 1.0
t[hours] ~
As can be seen in Figure 4.35, the calculated curve (c) agrees very well with the
curve based on measurements at moderate currents. Note that the battery voltage has
been plotted along the horizontal axis. The overpotential for the oxygen evolution
reaction at the nickel electrode has to be used to infer f' and a from the Tafel plots.
This overpotential can be found by subtracting the oxygen equilibrium potential
from the nickel electrode potential, i.e. TJo2=EN;-E'q02 • The oxygen equilibrium
potential equals the standard redox potential E'02 at the reference pressure of 1 bar,
hence E'q 02=E'02=0.40V; see section 4.2.4 and (4.64). The nickel electrode potential
can be inferred from the battery voltage by the addition of the cadmium electrode
potential, i.e. EN;=EN;cd+Ecd· Neglecting the relatively small overpotential at the
cadmium electrode, the cadmium electrode potential is the same as the equilibrium
potential EccFE'qccFE'ccF-0.81 V; see section 4.2.3 and (4.52). In summary, this
means that the oxygen overpotential can be inferred from the battery voltage plotted
along the horizontal axis by TJo 2 =EN;cd-0.81-0.40= EN;cd -1.21 V.
Figure 4.36 shows the same measured Tafel plots as shown in Figure 4.35
along with the Tafel plot calculated for the anodic branch of the evolution of oxygen
in the NiCd model; see (4.97). The optimized parameter values given for (f'o)N; 0 and
a.;2 in Table 4.4 were used in the calculation of this curve.
142 Chapter4
1.3 1.4 1.5 1.6 1.7 1.8 1.9
E[V]
Figure 4.35: Measured Tafel plots representing the oxygen production reaction occurring at
the nickel electrode during overcharging at 20°C [30]. The voltage of the NiCd battery used
has been plotted along the horizontal axis, while the battery current I has been plotted on a
logarithmic scale along the vertical axis. The marks were obtained in measurements. Curve
(a) is the linear approximation at low currents, curve (b) at high currents and curve (c) is the
calculated transition from curve (a) to (b) for moderate currents, see (4.98)
It can be inferred from Figure 4.36 that the measured (b) and calculated (d) Tafel
plots intersect at a current of approximately 600 rnA. This is in fact the visualization
of the mathematical explanation based on (4.97) that was given for the difference
between optimized and measured parameter values for (f'o)Ni0 and ao2 in section
4.4.3. Although the measurements in Figures 4.35 and 4.36 were performed using a
80 AAS NiCd battery, it is reasonable to assume that a similar two-step approach
holds for the P60AA NiCd battery, too. Figure 4.36 will now be used to explain the
main differences between the simulation and measurement results described earlier
in this section.
For the low current of 0.06 A (0.1 C-rate), curve (d) predicts a larger
overpotential for the oxygen evolution reaction in the simulations than the value
predicted by curve (a) for the measurements. This means that in the simulations, the
rate of the oxygen evolution reaction was lower than in the measurements, resulting
in a later start of the oxygen evolution. This is indeed observable in Figure 4.31. As
the overpotential of the oxygen production was higher in the simulations, the battery
voltage was also higher during overcharging. This is also observable in Figure 4.31.
1.3 1.4 1.5 1.6 1.7 1.8 1.9
E(V]
Figure 4.36: Same measured Tafel plots (curves (a)-(c)) as shown in Figure 4.35. Also
included is the Tafel plot calculated for the evolution of oxygen at the nickel electrode
according to (4.97) (d) as was included in the NiCd model presented in section 4.2, using
optimized parameter values for (f'o}Ni0 and «o2; see Table 4.4
At the high charge current of 1.2 A (2 C-rate), the measured battery behaved as
predicted by curve (b) in Figure 4.36, in which the region of interest lies to the right
of the intersection of curves (b) and (d), because of the high current value. The
overpotential for the evolution of oxygen was larger in the measurements than in the
simulations. The reverse situation as for 0.1 C occurred. The rate of oxygen
evolution was higher in the simulations than in the measurements. This led to an
earlier rise in the internal oxygen gas pressure and a lower battery voltage during
overcharging in the simulations than in the measurements. This is exactly what is
It can be concluded that the main differences between the simulation and
measurement results in the presented figures can be explained by the fact that the
evolution of oxygen in a NiCd battery seems to proceed in two steps, whereas a one-
step process was included in the NiCd model described so far. A model modification
to include the two-step process and subsequent parameter value optimization will be
considered in the next section.
144 Chapter 4
4.4.5 Results obtained with a modified NiCd battery model and discussion
The modification of the mathematical description of the oxygen evolution reaction
that occurs at the nickel electrode suggested in the previous section is in fact an
example of the second feedback loop in Figure 4.24. It was shown in the previous
section that the anticipated improvement should occur at charge currents ranging
from small to large values. The mathematical strategy presented in section 4.4.2 was
applied in section 4.4.3 for one charge current only. In this section, step (iii) of the
strategy will be repeated using a modified NiCd model, taking into account the two-
step behaviour postulated for the production of oxygen at the nickel electrode, for
various charge currents simultaneously. In addition to using two different values for
(f'o)N? and Q02, We alSO made both Q02 parameters temperature-dependent in
accordance with the experimental results [30].
The considered charge rates in the simulations were 0.1 C, 1 C and 2 C. (4.95)
was used as cost function and the contributions of the different charge currents were
simply added. The different contributions of the different charge currents were not
weighed in any way. Step (iii) of the strategy was simply repeated for different
charge currents to obtain a first impression of the improvement that can be realized
with the modified NiCd model relative to the results presented in the previous
section. The ambient temperature was again 25°C. The same measured curves were
used as presented in section 4.4.4. The optimized parameter set of Table 4.4 was
used as a starting point for all the parameters, except for those describing the oxygen
evolution at the nickel electrode. The same ranges as determined in section 4.4.3
were used. The measured values inferred from Figure 4.36 were used for the newly
added parameters describing the two-step oxygen-production process [30]. The
ranges of these parameters were based educated guesses. The results obtained for a
charge rate of 0.1 C are shown in Figures 4.37 and 4.38.
Comparison of the simulated pressure curve shown in Figure 4.37 and the
simulated curve shown in Figure 4.31 shows that the repetition of step (iii) using the
modified model caused the simulated pressure curve to shift to the left. This is the
desired result, because the oxygen production started too late in the simulations in
Figure 4.31. So, the kinetics of the oxygen-production reaction at the nickel
electrode indeed improved in the simulations. The shape of the simulated curve did
not change significantly and the gradual increase in oxygen pressure indicated by the
measured curve was not observed. However, the moment at which the pressure
increased significantly in the simulation now coincided with the moment at which
the pressure increased more in the measurements. The improved kinetics of oxygen
production at the nickel electrode did not lead to a significant change in the
simulated voltage curve. Comparison of the simulated temperature curve in Figure
4.38 with the simulated temperature curve in Figure 4.32 shows that the repetition of
step (iii) using the modified model led to better agreement of the simulated and
measured end values.
1.6 0.6
~
t 1.5
Vopt
Vmeas
0.5
0.4
> ~
1.4 Pmeas 0.3 "'0
T-
Pot Q.
0.2
1.3
0.1
1.2 0.0
0.0 5.0 10.0 15.0 20.0
t [hours] ~
Figure 4.37: Simulated (Vopt [V]) and measured (Vmeas [V]) battery voltage plotted along the
left y-axis and simulated (P0 pt [1frl Pa]) and measured (Pmeas [1frl Pa]) internal oxygen
pressure plotted along the right y-axis versus time (t [hours]) for charging a NiCd P60AA
battery at 0.1 C at 25"C ambient temperature using the NiCd model described in section 4.2
with the modified mathematical description for the oxygen-production reaction at the nickel
electrode in the simulations. The parameter values for the simulated curves were obtained by
repeating step (iii) of the strategy presented in section 4.4.2 for charge rates of 0.1 C, 1 C and
2 C simultaneously
1.6 40
t 1.5
Vopt
vmeas
35
t
~
> 30 L
(3'
1.4
1-
1.3 25
1.2 20
0.0 5.0 10.0 15.0 20.0
t[hours] ~
Figure 4.38: Simulated (V0 p1 [V]) and measured (Vmeas [V]) battery voltage plotted along the
left y-axis and simulated (Topt ["C]) and measured (Tmeas ["C]) battery temperature plotted
along the right y-axis versus time (t [hours]). See Figure 4.37 for conditions
The results obtained for a charge rate of 1 C are shown in Figures 4.39 and 4.40.
146 Chapter4
1.6 4.0
t 1.5 3.0
t
a;'
a..
~ 1.4 2.0 "'0
> ~
a..
1.3 1.0
1.2 0.0
0.0 0.5 1.0 1.5 2.0
t [hours] ___..
Figure 4.39: See Figure 4.37; the charge rate was 1 C in this case; compare to Figures 4.28
and 4.29
1.6
t
1.5
~ 1.4
>
1.3
-
1.2
0.0 0.5 1.0 1.5
t [hours]
Figure 4.40: See Figure 4.38; the charge rate was 1 C in this case; compare to Figure 4.30
Comparison of the results presented in Figures 4.39 and 4.40 with those given in
section 4.4.3 shows that the agreement between the simulation and measurement
results changed. This is quite understandable, because the optimization considered in
this section was performed for three currents simultaneously. The difference
between the simulated and measured battery voltages has increased to around 30 mV
in the mid-region in Figure 4.39, which is still acceptable for a nominal battery
voltage of 1.25 V.
The difference between the final simulated and measured pressures has become
2.6 kPa instead of 22 kPa in Figure 4.29. This is a significant improvement.
However, the pressure-versus-time curve is steeper in Figure 4.39 than in Figure

4.29, where the curve shown in Figure 4.29 is in better agreement with the
measurement. Apparently the simulated oxygen-production process started more
abruptly as a result of the better kinetics of the oxygen-production reaction at the
nickel electrode.
Comparison of the quantitative agreement between the simulated and measured
battery temperatures in Figures 4.40 and 4.30 shows that the simulated end value has
improved significantly. The difference is 1.3°C in Figure 4.40, which is smaller than
the difference of 7°C in Figure 4.30. The results obtained for a charge rate of 2 C are
shown in Figures 4.41 and 4.42.
1.6
I
>
1.5
~
1.4 2.0 ~
>
0..
1.3 1.0
-
1.2 -1--,---=;==;;;;;;;;;;;;;:~--,----r--.---+ 0.0
0.0 0.25 0.5 0.75 1.0
t [hours]
Comparison of Figures 4.41 and 4.42 on the one hand with Figures 4.33 and 4.34 on
the other hand shows that in the case of V, P and T the quantitative agreement
between the simulation and measurement results has improved significantly. The
agreement is now good for the voltage curve during practically the entire charging
period. The maximum difference in value between the simulation and measurement
results amounts to 15 mV, or 1.2%, and occurred at the end of the charging. This is
comparable with the results presented in section 4.4.2 for a charge rate of 1 C.
The difference between the simulation and measurement results in the final
value of the internal oxygen gas pressure is 7 kPa in Figure 4.41, which is a great
improvement over the value shown in Figure 4.33. The simulation and measurement
results obtained at the moment at which the oxygen production starts still differ
slightly. In the simulation, the oxygen production started a little earlier than in the
measurement. Nevertheless, the shape of the pressure-versus-time curve is in close
agreement with that representing the results of the measurement.
The simulated battery temperature shown in Figure 4.42 is a lot closer in shape
to the measured curve than the curve illustrated in Figure 4.34. The difference in the
final values between the simulation and measurement results is 4.6°C, which is
better than the agreement between the simulation and measurement results obtained
for the charge rate of 1 C; see Figure 4.30.
148 Chapter4
1.6 80
t 1.5 60
t
()
2.....
~ 1.4 40 1-
>
1.3 20
1.2 0
0.0 0.25 0.5 0.75 1.0
t [hours] _____..
In summary it can be stated that for the low charge rate of 0.1 C, the model
modification has only partly resulted in the desired effect. The kinetics of the
oxygen-production reaction at the nickel electrode improved as expected and the
oxygen reaction began earlier in the charge process. However, the simulated
pressure curve is still too steep and the battery voltage during overcharging remains
too high. For a 1 C-rate, the differences between the simulated and measured battery
voltages are a little higher than those presented in section 4.4.3, but they are still
acceptably small. The end values of the internal oxygen pressure and especially the
temperature have improved. The improvement with respect to the results presented
in the previous section is evident for the 2 C charge rate. The quantitative agreement
between the simulation and measurement results has become even better for this
current than for the 1 C charge rate considered in section 4.4.2.
A possible explanation for the marginal improvement in the results at lower
charge currents could be the fact that the kinetics of the evolution of oxygen at the
nickel electrode are different in a P60AA battery than as shown in Figure 4.35 for a
80AAS battery. Measurement of the Tafel plots shown in Figure 4.35 for the P60AA
battery should shed some light on this matter. Step (iii) was moreover performed
without considering possibilities of weighing the contributions of the various charge
currents in the cost function of (4.95). Another reason could be the fact that the
optimum parameter set given in Table 4.4 was taken as a starting point. The strategy
presented in section 4.4.2 could be extended for various charge currents and the
process could be repeated from scratch. More investigations will have to be carried
out to find out whether applying weighing to the cost function or starting the process
described in section 4.4.2 again are useful ways of improving the results for lower
charge currents.
The optimized parameter set of Table 4.4 will be used in the simulations
discussed in the next section, because the experiments described in this section
merely served to give a first impression of possible further improvement of the
simulations. Therefore, the proposed two-step oxygen-production reaction at the
nickel electrode will not be considered in the next section.
4.5 Simulation examples
4.5.1 Simulations using the NiCd model presented in section 4.2

In this section some simulation results will be presented that were obtained using the
NiCd battery model presented in section 4.2. The parameter values of the optimized
parameter set given in Table 4.4 were used. First, the results of the charging
simulation using a 1 C charge rate at 25°C ambient temperature, presented in Figures
4.28, 4.29 and 4.30 will be described in greater detail. For reference, the simulated
V, P and T curves have been plotted in a single figure below.
~ 1.5
> 40 I-
2:
a.
1.4
20
1.3
1.2 L---......:===:::t==::::::::.:.____...JO.O 0
0 1.0 2.0
t[hours]-
Figure 4.43: Simulated battery voltage (V [V]), internal oxygen gas pressure (P [bar]) and
temperature (T rc]) as a function of time (t [hours]) during charging at 25°C ambient
temperature at a 1 C-rate. The optimized parameter values given in Table 4.4 were used
The phenomena observable in the V, P and T curves shown in Figure 4.43 can be
understood by considering the simulation model of the NiCd battery shown in
Figure 4.19. The current will flow from the left (positive terminal) to the right
(negative terminal) during charging. The charge current flows through the nickel
reaction path at the nickel electrode (DN; and electrical port of the transformer), until
the moment when the oxygen pressure starts to rise. The molar amounts increase in
the chemical capacitances Cch NiOOH and decrease in the chemical capacitances
Cc/iiOHi 2 in the RC-ladder network. The current flows through the cadmium reaction
path (Dcd and the electrical port of the transformer) in the cadmium electrode. The
molar amount increases in capacitance Cchcd and it decreases in capacitance
CchCd(OH!2 •
The overpotential across diodes D~ for the oxygen reaction at the nickel
electrode is higher than that for the nickel reaction across diodes DN;. because the
value of E'Ni is higher than the value of E'o; see Table 4.4. However, although the
higher overpotential for the oxygen reaction may predict a higher reaction current
than for the nickel reaction, according to the Butler-Volmer equation, a negligible
amount of oxygen is formed at the nickel electrode at the beginning of the charging.
150 Chapter4
This is due to the relatively low reaction rate of the oxygen evolution reaction,
expressed in a low value for (f'o)Nio in (4.65). No current flows through the
cadmium reaction path at the nickel electrode, because under normal charging
conditions only Cd(OH)2 is present in the nickel electrode and no metallic Cd, which
could be oxidized into Cd(OH)2 • No current flows through the oxygen path at the
cadmium electrode either, because only a negligible amount of oxygen gas is present
during the first stages of charging.
When the charging is continued, the decrease in the molar amount of the
Ni(OH)2 species in the nickel electrode will lead to a lower value of f' Ni through
(4.46). As a result, a larger overpotential r/'Ni across diodes DNi occurs at the same
value of the current, as can be inferred from the Butler-Volmer equation given in
chapter 3. At the same time, the apparent equilibrium potential E'q*Ni rises according
to (4.49). The increase in both r/'Ni and E'q*Ni results in a substantial increase in the
battery voltage. Consequently, the overpotential of the oxygen reaction across
diodes D~ in the nickel electrode model also rises, as can be understood by applying
Kirchhoffs voltage law to the parallel combination of the nickel and oxygen
reaction paths. The equilibrium potential of the oxygen reaction remains unchanged
until the oxygen pressure starts to rise; see (4.64). An increase in the overpotential of
the oxygen reaction eventually allows some oxygen to be formed. This in turn leads
to an increase in (f'o)Nio according to (4.65), because the activity of oxygen ab~
increases, and the oxygen reaction will gradually take over from the nickel reaction.
Overcharging now commences. The increase in ab~ leads to an increase in oxygen
pressure according to (4.63).
Owing to the available oxygen, current may now also start to flow through the
oxygen reaction path at the cadmium electrode, which represents the reduction of
oxygen. As a result, the oxygen reduction flow in the chemical domain will have the
opposite direction of the oxygen evolution flow induced at the nickel electrode,
according to the signs associated with the transformer ports in Figure 4.19. This
prevents an unlimited increase in the internal oxygen pressure.
Table 4.4 shows that there is a substantial difference between the standard
electrode potentials It' cd and It'0z As a result, there is a substantial difference
between the equilibrium potentials of the cadmium and oxygen reactions.
Application of Kirchhoffs voltage law to the parallel connection of the cadmium
and oxygen reaction paths at the cadmium electrode reveals that there will be a large
overpotential for the oxygen reaction across diodes D0z at the cadmium electrode. A
reaction with a large overpotential will generate a lot of heat according to (4.32).
Hence, the heat flow source l 1hin in Figure 4.19b representing the heat generated by
the oxygen reduction reaction at the cadmium electrode will attain a large value. As
a result, the battery temperature will start to rise more in this overcharge region.
Both the overpotentials 17 and the standard electrode potentials It' are temperature-
dependent through (4.72) and (4.73), respectively. Therefore, the battery voltage
decreases, because of the rising temperature. This is due mainly to decreasing
overpotentials, as observed in simulations. This can be understood from the fact that
the exchange currents f increase with increasing temperatures, because the reaction
rate constants ki increase according to (4.72).
The distribution of current between the nickel reaction path and the oxygen
reaction path is very important in determining a battery's charging efficiency. Only
current that flows through the nickel reaction path leads to effective accumulation of
charge inside a battery. Current that flows through the oxygen reaction path leads to
pressure build-up and oxygen recombination. Eventually, it leads to energy loss in

the form of heat dissipation. The charging efficiency is strongly influenced by both
the charge current and the battery temperature. More information on charging
efficiency will be given in chapters 5 and 6.
The diffusion of H+ ions becomes a limiting factor of increasing importance for
the rate of the nickel reaction when higher charge currents are applied. This means
that the apparent equilibrium potential gq•Ni starts to rise earlier in the charging
process, because of a higher value of the diffusion overpotential rfNi; see (4.49).
This in tum leads to an earlier start of the production of oxygen in the charging
process, and hence to a lower charging efficiency.
The ratio of the rates of the nickel and oxygen reaction decreases as the battery
temperature increases. This is due to the difference in the temperature dependence of
the two reactions, which is revealed by the difference in activation energies; see
Table 4.4. Figure 4.44 shows the simulated partial currents that flow through the
nickel and oxygen reaction paths in the nickel electrode during charging at a
constant rate of 0.1 C at ambient temperatures of 0 °C and 60°C. The sum of the two
partial currents equals the charge current. The oxygen reaction takes over a lot
earlier in the charging process at the higher ambient temperature of 60°C, which
leads to a poorer charging efficiency at higher temperatures. This has indeed been
confirmed in experiments.
t 0.08
.... ____ __
~ 0.06 k-------~---
.§ ...... ............ I
......,...,./ Ni
0.04
.........
60°C)',-x
0.02
\ "' '
,. ..........
.....
.... ...... ......

---- ......
0 o~---5~--~1~0-~~-1~5------~20
t[hours) _ . .
Figure 4.44: Simulated partial nickel (INi [A]) and oxygen (/02 [A]) currents as a function of
time (t [hours]) in the nickel electrode during charging at a rate of 0.1 C at 0 °C and 60°C
ambient temperature. The optimized parameter values given in Table 4.4 were used
When batteries are not used for some time, they will lose stored energy as a result of
self-discharging, especially at higher temperatures. This phenomenon can be easily
understood by considering the simulation model of a NiCd battery in Figure 4.19
when no load is applied. The value of the apparent equilibrium potential gq•Ni is
higher than that of the equilibrium potential gq ~ because It'Ni is higher than It'0z
Therefore, a loop current will flow counterclockwise through the parallel
152 Chapter4
combination of the nickel and oxygen reaction paths in the nickel electrode. This
leads to a decrease in stored energy at the nickel electrode and an increase in the
molar amount of oxygen and the internal oxygen gas pressure. The oxygen produced
is reduced at the cadmium electrode, which leads to a decrease in stored energy in
the cadmium electrode through a similar loop current. Figure 4.45 shows the
simulated SoC at different ambient temperatures. At 25°C, about 70% of the charged
capacity is lost after 100 days. This corresponds well to the reported self-discharge
of 60% of the charged capacity of a typical sealed NiCd battery after three months of
storage at 25°C [18]. At higher temperatures, the rates of all the electrochemical
reactions become higher through (4.72) and consequently the loop currents in the
nickel and cadmium electrode models increase. As a result, the self-discharge rate
increases at higher temperatures, as can be inferred from Figure 4.45. A higher self-
discharge rate at higher ambient temperatures is commonly known from practice and
the literature. For example, a 100% loss of charged capacity through self-discharge
has been reported at 40°C for a typical sealed NiCd battery in a period of a little over
2 months [18]. This corresponds quite well to the curve simulated at a 45°C ambient
temperature in Figure 4.45. The opposite situation occurs at lower temperatures. A
20% loss of charged capacity after 3 months is reported in the literature for a typical
sealed NiCd battery at an ambient temperature of 0°C [18]. Again, this corresponds
perfectly to the simulated curve shown in Figure 4.45.
t
(.)
0
C/)
--- ---
...............
25 50 75 100
t[days] ~
Figure 4.45: Simulated SoC as a function of time (t [days]) at different ambient temperatures.
The optimized parameter values given in Table 4.4 were used
Figure 4.46 shows the simulated V, P and T curves during discharging and
overdischarging at an ambient temperature of 25°C at a 1 C discharge rate. A 0 V
region is clearly recognizable in the voltage curve after approximately 1.5 hours,
followed by a region with a reversed battery voltage. In the latter region, both the
temperature and the pressure increase and level off to a steady-state value. This is
commonly encountered in practice when overdischarging NiCd batteries [18]. What
happens during overdischarging can be understood by considering the simulation

model of a NiCd battery shown in Figure 4.19.
t 1.5 r----v::-:-----....
I 1.0 discharge region overdischarge region
.I
~ 1·----------i~l·+--------+16 a.. 'iii' 60 IT
1:.....
>M ~
~-·············· =.
0 / 4 a.. 40
...,................ :'
-0.5 p··································.:
/ _____e_____ 2 20
-1.0
-1.5 ..___ __.___ __.__ ___.__ __.____._ ____.o 0
0 0.5 1.5 1 2.5
t[hours]-
Figure 4.46: Simulated battery voltage (V [V]), internal oxygen gas pressure (P [lOS Pa]) and
battery temperature (T [0 C]) curves as a function of time (t [hours]) during discharging and
overdischarging at an ambient temperature of 25°C at a 1 C discharge rate. The optimized
parameter values given in Table 4.4 were used
The current direction in Figure 4.19a is from the right (negative terminal) to the left
(positive terminal) during discharging. During normal discharging, when the battery
voltage remains at a positive, relatively constant level, current flows through the
cadmium reaction path at the cadmium electrode and the nickel reaction path at the
nickel electrode. The battery is effectively discharged in this case. The molar
amount of Cd decreases in the cadmium electrode and the molar amount of Cd(OHh
increases. The molar amount of NiOOH decreases in the nickel electrode, while that
of Ni(OH)z increases. The nickel electrode will run out of NiOOH before all the Cd
has been consumed at the cadmium electrode because of the excess amount of Cd at
the cadmium electrode, as explained in section 4.2.1. Consequently, the exchange
current f' Ni of the nickel reaction will decrease at the end of the discharging. This
will lead to a larger overpotential r/'Ni· Moreover, the value of the apparent
equilibrium potential Kq*Ni will decrease, mainly because diffusion limitation of H+
ions leads to a high value of the diffusion overpotential rfNi· As a result, less current
flows through the nickel reaction path and a competing reaction will gradually take
over. No current can flow through the oxygen reaction path at the nickel electrode
because no oxygen is yet present to be reduced. Therefore, the current will start to
flow through the cadmium reaction path at the nickel electrode. The resulting battery
voltage will approach 0 V because identical reactions now occur at both electrodes.
This is the first stage of overdischarging. A lot of heat is generated in this reaction,
because the current through the nickel reaction path does not become zero
instantaneously and the overpotential of the nickel reaction will have become very
large because the electrode potential has reversed. This is illustrated by the
temperature rise in Figure 4.46 during the period when the battery voltage is 0 V. As
154 Chapter4
soon as all current has started to flow through the cadmium reaction path at the
nickel electrode the temperature will decrease again, because the overpotential of
this reaction will be substantially smaller than the nickel overpotential during
overdischarging.
When the amount of excess Cd in the cadmium electrode decreases to zero, the
current will start to flow through the oxygen reaction path in the cadmium electrode.
As a result, the molar amount of oxygen, and hence the oxygen gas pressure, will
increase. The potential of the cadmium electrode will now reverse because of the
difference in the signs of E' cd and E'o; see Table 4.4. The oxygen-reduction reaction
will start to compete with the cadmium-reduction reaction at the nickel electrode.
This is the second stage of overdischarging. The reverse situation as described for
the charging process will occur. The cadmium electrode will act as a positive,
oxygen-forming electrode, whereas the nickel electrode will act as a negative,
oxygen-reducing electrode. As a result, the oxygen pressure will level off, as can
indeed be seen in Figure 4.46. Again, this reaction in particular generates a lot of
heat because of the high overpotential for the oxygen reduction reaction at the nickel
electrode, which leads to a significant increase in temperature.
Voltage reversal may occur when several batteries are connected in series. It is
more correct to use the term cells in this case, as explained in chapter 3. Series or
parallel connections of cells in battery packs are often used in practice in order to
obtain a higher voltage at the same battery capacity in [Ah] or a higher battery
capacity in [Ah] at the same battery voltage, respectively. Figure 4.47 shows the
simulated battery pack voltage and the internal oxygen gas pressure of one of its
cells when six NiCd cell models are connected in series. The battery pack was
discharged in a simulation using a 1 C discharge rate at 25°C ambient temperature.
Two cases were simulated. In case (a), all six cells inside the pack were given the
same capacity as indicated in Table 4.4. In case (b), only one cell was given the
same capacity as in Table 4.4 and the others were given capacities decreasing in
steps of 10% of the capacity indicated in Table 4.4 to reflect the spread in cell
capacity. This spread depends on the type, manufacturer, usage history and age of
the pack. The heat resistances between the six cells and between the cells and the
environment were taken to be zero for simplicity. In more advanced simulations,
thermal models of the heat flows between the cells in the battery pack or from
individual cells to the environment could also be considered.
Curves (a) in Figure 4.47 have the same shape as the V and P curves in Figure
4.46. All the cells will become empty at exactly the same moment because all the
cells have exactly the same capacity, and the battery pack voltage V(a) is simply the
sum of six identical voltage curves. The pressure curve P(a) is shown for only one
cell, but it is the same for all six cells. The oxygen pressure rises when all the cells
have reversed their voltages, which can likewise be observed for one cell in Figure
4.46. No pressure build-up will take place in practice in the case of identical cell
capacities because when the battery pack voltage decreases and approaches 0 V, the
electronic equipment will no longer operate and the discharge of the battery pack
will stop.
Curves (b) in Figure 4.47 clearly show the result of the different capacities
inside the battery pack. Curve V(b) shows the total pack voltage and curve P(b)
shows the pressure build-up inside the cell with the lowest maximum capacity. This
cell will be the first fully discharged cell and its voltage will approach 0 V. This can
be seen in curve V(b) as the first drop in the battery pack voltage. The battery pack
voltage is still high enough to power the load, and discharging will continue because
the other cells still have a voltage of around 1.3 V. Subsequently, the cell with the
second lowest capacity reaches a voltage of approximately 0 V, which is represented
by the second drop in the battery pack voltage. Then the cell with the third lowest
capacity reaches a voltage of 0 V and a third drop in the battery pack voltage is
observed. The fourth drop in the battery pack voltage is due to voltage reversal of
the cell with the lowest maximum capacity. As a consequence, the oxygen pressure
inside this cell will increase, as can be inferred from curve P(b). A vertical line has
been added to the figure at that moment for convenience. The battery pack voltage at
this moment is still around 2.5 V, which may allow discharging to continue even
further. High pressure-build-up inside the cells may eventually lead to venting and
loss of electrolyte, which degrades the cell's performance. The simulation clearly
shows that the battery pack is as strong as its weakest cell. Therefore, the capacities
of cells inside a battery pack should be as similar as possible and the load should be
shut off at an appropriate voltage level. The choice of this voltage level will be a
compromise between the run time of the load and the lifetime of the battery pack.
The simulation clearly shows the need for a Battery Management System as already
described in chapter 2.
10 4
t
~ 5
t
'iii'
Q.
> ltl
0
I
I
=.
0 I 2 Q.
I ••
I : •••• •••••••••• .,!:~..----·--..-.---..---..
p (b)if f P (a)
1:
-5 1:
1:
1:
-10 0
0 2.5
t [hours] ---+-
Figure 4.47: Simulated voltage (V [V]) and internal oxygen gas pressure (P [10' Pa]) as a
function of time (t [hours]) of a battery pack with six NiCd cells connected in series during
discharging at a 1 C-rate and at an ambient temperature of 25°C. Curves (a) represent the
situation when all six cells have the same capacity; see Table 4.4. Curves (b) represent the
situation when the six cells decrease in steps of 10% of the capacity indicated in Table 4.4
4.5.2 Simulations using the Li-ion model presented in section 4.3

Some simulation results obtained using the Li-ion battery model presented in section
4.3 will be presented in this section. The values of the parameters used in this model
were found in a simpler way than for the NiCd battery model. The number of
parameters used in the Li-ion model is considerably smaller than that used in the
NiCd model, mainly because of the absence of modelled side-reactions. The curves
simulated with the Li-ion model were fitted to measured curves obtained with the
156 Chapter4
CGR17500 Li-ion battery from Panasonic [40]. As a starting point, many parameter
values relating to the battery's design were obtained from the battery manufacturer
and other parameter values were, where possible, obtained from the literature. In
addition, the parameters used in the equations for the equilibrium potentials for the
positive electrode in (4.79) and (4.80) and for the negative electrode in (4.88) and
(4.89) were found by curve fitting, using curves measured for the equilibrium
potentials of the separate electrodes.
The parameter values for the exchange currents f' LiCoo2 and f'uc6 were found in
impedance measurements performed with the CGR17500 battery [40]. Polar plots of
the real and imaginary parts of the impedance of a battery can be used to derive
parameters relating to the kinetic and diffusion processes occurring inside the
battery, as will be described in more detail in chapter 6. However, it is not possible
to distinguish between the contributions of the two individual electrodes to these
processes. Therefore, it has in this book been assumed that the kinetic parameters
inferred from the measured battery impedance can be assigned exclusively to the
LiCoOz electrode. This means that it is assumed that the kinetics of the LiC 6
electrode are relatively fast in comparison with those of the LiCo02 electrode. The
value of the double-layer capacitance CJucoa2 was also inferred from the impedance
plot under the assumption mentioned previously. The double-layer capacitance value
of C 1uc6 was chosen identical to CJucoo2.
The parameter values used in the simulations using the Li-ion model described
in this section are summarized in Table 4.5. As in Table 4.4, a distinction has been
made between design-related parameters and electrochemical parameters. The fourth
column of this table lists the equations or figures in which the parameters have been
used. Temperature dependence of the model equations has not been implemented
yet, as mentioned in section 4.3. Therefore, temperature-related parameter values
have not been included in Table 4.5 and all the parameter values are valid at 25°C.
The f'; values are specified at 50% SoC. A PTC has been included in series with the
battery for safety reasons (see chapter 3), which is also considered in the model by
means of a series resistance. Its value RPTc can be found in Table 4.5.
Some examples of simulations performed using the Li-ion model with the
parameter values of Table 4.5 will be described in the remainder of this section.
First, some charging simulations will be discussed and the simulation results will be
compared with measurements obtained with a CGR17500 Li-ion battery. A Li-ion
battery is charged in two modes, a constant current (CC) mode followed by a
constant voltage (CV) mode. More information will be given in chapter 5. Figure
4.48 shows the simulated and measured voltage and current curves for charging a
CGR17500 Li-ion battery at a constant rate of 1 C in CC mode and a constant
voltage of 4.1 V in CV mode. The charge time is two hours.
The CC and CV modes can be easily distinguished in Figure 4.48. The current
is constant in CC mode, while the battery voltage rises gradually to the value it
attains in CV mode. In the latter region, the battery voltage remains constant while
the battery current drops. As can be seen in the figure, the quantitative agreement
between the simulation results and the measurements is quite good. The largest
discrepancy between the simulation results and measurements is observable in the
voltage curves. As in the NiCd model, the simulated charging voltage is higher than
that measured at the beginning of the charging. A possible reason for this, apart from
incorrect parameter values, could again be the occurrence of processes in the
measured battery which have not been taken into account in the Li-ion model. The
change from CC mode to CV mode occurs somewhat later in the charging process in
the simulations because the simulated voltage is lower than the measured voltage at
the moment when the battery approaches the maximum voltage level in CV mode.
Table 4.5: Most important parameter values of the Li-ion battery model described in section
4.3
Parameter Value Unit Reference
cnuc6 0.4 F Figure 4.21
c;dlUCo02 0.4 F Figure 4.21
ct=ucoo2 5184 c See note 1
(l"axuc6 3060 c -
lpos 10·10-6 m Figure 4.21
lneg 10·10"6 m Figure 4.21
letyt 2.8·10"4 m Figure 4.21
Rteak 80·10 3 Q Section 4.3.4
Rrrc 50·10-3 Q -
e, 10 - Figure 4.11
Electrochemical ~arameters
Aelyt 20·10-3 m2 See note 2
Du,P"s 1.5·10-13 m2/s Section 4.3.6
Du•neg 1.5·10-13 m2/s Section 4.3.6
Du•elyt 2·10- 11 m2/s Section 4.3.6
D elyt 3·10- 11 m2/s Section 4.3.6
Pr 6
E:'ucoo2 3.89 V vsLi (4.78)
E"uc6 0.04 V vsLi (4.87)
f'ucoo 2 251·10-3 A (4.82)
f'uc6 15.3 A (4.90)
nu 1 - (4.78)
Upos,l -3 - (4.79)
Upos,2 20 - (4.80)
Uneg,l -10 - (4.88)
Uneg,2 -2 - (4.89)
xmsphasetrunsition 0.75 - (4.81)
;i"'Bphasetransition 0.25 - Section 4.3.3
zu• +1 - Section 4.3.4
ZPr6 -1 - Section 4.3.4
aucoo2 0.4 - (4.82)
auc6 0.15 - (4.90)
""s,l 0 - (4.79)
"-s.Z 17.25 - (4.80)
(n.g,l 0 - (4.88)
"'-K.2 2 - (4.89)
Note 1: The rated capacity of the LiCoO:z electrode, i.e. 0.72Ahx3600(seclhour)=2592 C, is only roughly
half of the full capacity, because XLi is only cycled between 0.5 and 0.95, as described in section 4.3.2.
Hence, the value of the full capacity is a factor of two higher than the rated capacity and this value has
been used in the equations describing the behaviour of the LiCo02 electrode
Note 2: Surface area electrolyte, which is the area of spatial element in the electrolyte; see Figure 4.5
158 Chapter4
t 4.2
v .... --- 0.8 t
4.0
~ 0.6~
> 3.8
3.6 0.4
3.4
0.2
3.2
3.0 0
0 30 60 90 120
t [min] -----..
Figure 4.48: Simulated (marked lines) and measured (solid lines) voltage (V [V]) and current
(I [A]) curves as functions of time (t [min]) for charging aLi-ion battery at a 1 C charge rate
in CC mode and a 4.1 V charge voltage in CV mode for two hours in total at a battery
temperature of 25°C
Figures 4.49 and 4.50 show the simulated and measured battery voltage and current
curves, respectively, for charging a Li-ion battery at different charge rates, ranging
from 0.1 C to 1 C. With each charge current, CC/CV charging is stopped when the
battery current drops below 10 rnA in CV mode. This ensures that almost equal
capacities in [Ah] are reached for each charge current. This can be inferred from the
figures. The quantitative agreement between the measurements and the simulation
results is better for lower charge currents for both the voltage and current curves.
The voltage simulated during the initial stages of charging is higher than the
measured voltage, as observable in Figure 4.48, whereas this situation is reversed
later in CC mode. Therefore, the CV mode later takes over from the CC mode in the
simulations.
Figure 4.51 shows the charge time necessary to charge the battery to a capacity
of 770 mAh as a function of the current in CC mode when the charge voltage in the
CV region is 4.1 V. Charging is stopped when the battery current in CV mode
reaches the value of 10 rnA.
The simulated and measured charge times in Figure 4.51 show an almost
perfect fit, as opposed to the simulated curves in Figures 4.49 and 4.50, which show
some deviations from the measured curves. Figure 4.51 indicates that the charge
time initially decreases strongly as a function of the charge current. However, the
charge time hardly decreases when the charge rate becomes larger than 1 C. This is
because the charge time is almost completely determined by the time at which the
battery is charged in CV mode at higher charge currents. In this mode, the charge
time is independent of the charge current. The overpotentials in the battery are much
higher than at smaller charge currents when higher charge currents are applied in CC
mode. As a result, the CV mode is reached much quicker in the charging process.
More information on the influence of the parameters Imax and Vmax on the charge time
and other factors will be given in chapter 5.
4.2~----------------------------------~
i
>
>
0 0.2 0.4 0.6 0.8

Charged capacity [Ah] )o
Figure 4.49: Simulated (marked lines) and experimental (solid lines) battery voltage (V [V])
curves as functions of the charged capacity [Ah] for charging aLi-ion battery with various
charge currents in the CC mode and a 4.1 V charging voltage in the CV mode at a battery
temperature of 25"C. The charge rate increases from bottom to top: (o) 0.1 C, (•) 0.5 C, ( +) 1
c
i
0.8
1c
0.6
~
0.4
o.sc
0.2
0.2 0.4 0.6 0.8

Charged capacity [Ah] )o
Figure 4.50: Simulated (marked lines) and experimental (solid lines) battery current(/ [A])
curves as a function of the charged capacity [Ah]. See Figure 4.49 for conditions
Figure 4.52 shows simulated and measured battery voltage curves as functions of the
discharged capacity during discharging at various discharge currents. The figure
shows that the quantitative agreement between the results of the simulations and the
measurements is quite good for discharge rates lower than 1 C. However, a
discrepancy is observable in the region from 0 to 0.3 Ah of discharged capacity for
the discharge rates of 1 C and 1.5 C. In this region, the simulated battery voltage is
higher than the measured battery voltage. This difference at higher discharge
currents could not be reduced in further optimization of the model parameters. This
suggests the absence in the model of a process which does occur in real life.
Alternatively, the mathematical description of a process included in the model might
be incorrect.
160 Chapter4
I...
40
Qin =0.77Ah
'iii'
Vmax =4.1 V
::I 30
0
E.
CD
2'
...."fi"' 20
0 0.3 0.6 0.9 1.2 1.5

lcharge [A]--+-
Figure 4.51: Simulated (o) and experimental (+) charge time necessary to charge a Li-ion
battery to a capacity of 770 mAh (tcharge [hours]) as a function of the charge current (/charge
[A]) during CC/CV charging with Vmax = 4.1 Yin CV mode at a battery temperature of25°C
I
4.2
4.0
~
>
3.6 0.1 c
0.25C
0.5C
3.4
0.75C
3.2
3.0 &~ 1.5C

0 0.2 0.4 0.6 0.8
Discharged capacity [Ah)--+-
Figure 4.52: Simulated (marked lines) and experimental (solid lines) battery voltage curves
(V [V]) as functions of the discharged capacity [Ah] during discharging aLi-ion battery with
various discharge currents at a battery temperature of 25°C. The discharge rate increases from
top to bottom: (o) 0.1 C (X) 0.25 C (•) 0.5 C (L\) 0.75 C (•) 1 C (.A.) 1.5 C
The kinetics of the LiCo02 electrode were assumed to be much slower than the
kinetics of the LiC6 electrode, as discussed previously. Therefore, the kinetics of the
LiCo02 electrode have far more influence on the battery voltage than the kinetics of
the LiC 6 electrode in the simulations. A possible explanation for the difference
between the simulation results and the measurements at higher discharge currents in
the region from 0 to 0.3 Ah discharged capacity is that in the model, the kinetics of
the LiCo02 electrode have been assumed to be independent of the phase formation.
As discussed in section 4.3.2, a phase transition occurs in the LiCo02 electrode at
xu=0.15. This means that, starting from a fully charged battery, or xu=0.5 in the
LiCo~ electrode, the phase transition at xu=0.15 occurs at approximately 0.36 Ah
discharged capacity. This value of 0.36 Ah agrees more or less with the value at
which the discrepancy between the simulation results and the measurements occurs
in Figure 4.52. It seems reasonable to assume that the model predicts the battery
voltage during discharging quite well for xu>0.15 for all discharge currents, because
the kinetics modelled for the LiCo02 electrode apparently apply to this region.
However, a deviation occurs at higher discharge currents for xu<0.15, which could
be explained by a deviation in kinetics occurring in reality which has not been taken
into account in the model. It is also possible that the kinetics of the LiC 6 electrode
play a more important role in determining the voltage in reality.
Another remarkable feature that can be seen in Figure 4.52 is that the total
discharged capacity is more or less the same for all discharge currents. This
behaviour differs from the behaviour of, for example, NiCd batteries, where the
discharged capacity varies more with the discharge current [18]. An explanation for
this difference is that the diffusion of H+ ions in the nickel electrode of a NiCd
battery is limited due to the relatively large diffusion length in comparison with that
of a LiCo02 or LiC6 electrode in aLi-ion battery. For example, SEM photographs of
a nickel electrode of a SOAAS NiCd battery have shown that the size of the grains is
50 IJ.m [40]. This value corresponds well to the value of 58.3 IJ.m given for the
thickness of the nickel electrode lNi in the optimized parameter set of Table 4.4. This
value is about five times higher than the grain size in a LiCo0 2 or LiC 6 electrode;
see lpos and lneg in Table 4.5. Therefore, the influence of the diffusion of H+ ions in
the nickel electrode of a NiCd battery is greater than the influence of the diffusion of
Li+ ions in the electrodes of a Li-ion battery. Diffusion limitation within the
electrode causes part of the electrode material to become inaccessible during
discharging. This effect becomes stronger with an increasing discharge current,
which may explain the dependency of the discharged capacity on the discharge
current in NiCd batteries.
Figure 4.53 shows the simulated overpotentials that occur in the Li-ion model
during discharging at a rate of 1.5 C. It can be seen in Figure 4.53 that, except at the
end of the discharging process, the diffusion of u+ ions in the electrodes hardly
contributes to the total overpotential. This means that the phenomenon that part of
the electrode material becomes inaccessible during discharging is not dominant in a
CGR17500 Li-ion battery. This explains the observed independence of the
discharged capacity with respect to the discharge rate between 0.1 C and 1.5 C. The
simulations show that the total overpotential is determined mainly by the reaction
overpotential of both electrodes (r/1o1aJ) in the considered Li-ion battery and the
overpotential across the electrolyte due to diffusion of u+ ions <rfu+). The
overpotential due to the diffusion of Lt ions in the electrolyte is obtained from
(4.93).
162 Chapter4
>
-0.10
d
'llu+ in LiCo02
-0.15
ld = -1.08A (1.5C}
-0.20 L------~--------1----........I.---L----1
0 0.2 0.4 0.6 0.8
discharge capacity [Ah] --.....
Figure 4.53: Simulated overpotentials for the Li-ion model during discharging at a 1.5 C-rate
at a battery temperature of 2s•c
4.6 Conclusions
In this chapter it has been shown that the general approach to battery modelling
based on physical system dynamics can be used to model basic building blocks that
can be used to model various battery system The general approach has been applied
to two battery systems in this chapter, NiCd batteries and Li-ion batteries. The
application of the general approach to NiCd batteries leads to a battery model with
which the battery voltage (V), current (1), temperature (1) and internal oxygen gas
pressure (P) can be simultaneously and coherently simulated under a wide variety of
conditions. The battery voltage and the behaviour of the electrolyte of Li-ion
batteries have been simulated. The chosen approach allows other processes to be
added to the models, as long as a mathematical description is available. Moreover, it
can also be applied to other types of batteries and battery systems. As long as the
basic structure of the battery in terms of, for example, electrode capacities and
volumetric characteristics, the types of reactions taking place, etc., as well as the
mathematical description of all the processes are known, the basic building blocks
described in this chapter can be used to develop a network model of the battery. An
advantage of the applied approach is that the parameters used in the model all have a
physical or chemical meaning.
To enhance the predictive nature of the models, a good quantitative agreement
between the results of simulations and measurements is important. Such agreement
depends on the design of the model, including such aspects as the considered
processes, the correctness of the mathematical descriptions of the considered
processes and the values of the parameters in the model. The emphasis in section 4.4
has been on parameter estimation. A method has been used that should increase the
chances of finding the optimum parameter set as much as possible, because the
estimation of many parameter values of a complex non-linear model is a
complicated process. The first results presented in section 4.4.3 show that it is
possible to fmd a good quantitative agreement between simulation and measurement
results when the model is optimized for one operating condition. This good
quantitative fit holds for V, P and T simultaneously. Comparison of optimized and
measured parameter values, where available, shows that the chosen approach yields
parameter values close to the measured values.
The quantitative agreement between the simulation results and the
measurements decreases when other operating conditions are considered. A
modification in the mathematical description of one of the modelled processes has
been considered in a first attempt to improve the quantitative agreement for more
than one charging condition. Part of the iterative process has been repeated with this
modified model. The results are promising and the improvements realized so far
conform to our expectations. However, especially the oxygen pressure curves
observed in the simulations and the measurements still differ considerably in shape
for low charge currents. Hence, more iterations in the iterative optimization process
are needed to further improve the quantitative agreement. Even with simple manual
tuning of parameters quite good quantitative agreement can already be obtained for
the Li-ion model, mainly because it involves fewer parameters.
The simulation examples given in section 4.5 show that the models can be used
to gain insight into the processes that take place inside the batteries. The models act
as a 'transparent battery', in which, for example, the development of the
overpotentials of the individual reactions can be studied. The user can determine
which externally observable characteristics correspond to which process, as has been
illustrated in various examples. Besides for gaining insight into what happens inside
a battery, the models can also be used to generate V, I, P and T quickly and
repeatably. This offers the designer of a BMS a quick means for predicting battery
behaviour early in the design process. An example given in this chapter has shown
the effect of connecting cells with unequal capacities in series in a battery pack. A
possible mechanism for breakdown of the pack and ways of preventing it have been
described. More examples of using the model in various forms will be given in the
following chapters.
The quantitative agreement is not optimum in all situations, as has been
illustrated in some examples. Besides incorrect parameter values another reason
could be the omission of certain processes from the models. The lack of proper
physical or chemical and associated mathematical descriptions of some of these
processes is the reason for the omission. These processes can be readily included in
the model once the corresponding mathematical descriptions have been found.
As a first example of these omissions, aging effects, which lead to the gradual
loss of battery capacity when a battery ages, have not been included, because a good
physical or chemical description of this process is still lacking. One of the possible
reasons for capacity loss could be irreversible reactions taking place during the life
of a battery. These reactions consume active material and form non-active species
that do not take part in energy storage. Such an irreversible reaction could be
modelled as a parallel path to the main reactions.
A second example of an omission from the current battery models is the effect
of electrode porosity on the battery behaviour. A real electrode consists of a large
number of grains of active material that are connected to the current collector. There
are pores between the grains, which should be filled by the electrolyte solution to
enable the transfer of ions to and from the interface between the electrode grains and
the electrolyte solution. However, depletion of the electrolyte solution from the
pores has been proven to have a significant influence on the behaviour of a battery
[51]. A possible way of modelling the effect of porosity is to extend the models from
the one-dimensional approach that has been described in this chapter to a two-
164 Chapter4
dimensional approach. The electrodes can be modelled as many small identical

electrodes connected in parallel, with each small electrode denoting a particle of
active material [40]. Each grain is modelled in the same way as in the electrode
models described for the NiCd and Li-ion battery models. The grains are on one side
all connected to one node that denotes the current collector. On the other side, each
grain contacts the model of the electrolyte at a different spot. In the case of aLi-ion
battery model, the electrolyte model should be a two-dimensional extension of the
model described in this chapter. For the NiCd model, this means that the approach of
modelling the electrolyte by a simple resistance must be abandoned. The electrolyte
should be modelled by a two-dimensional RC-ladder network that describes the
diffusion and migration ofOH" and K+ ions in the electrolyte.
A third example of an omitted process is the hysteresis that occurs in the
equilibrium potential. It has been found in measurements with both NiCd and Li-ion
batteries that the equilibrium potential of a battery during charging is higher than the
equilibrium potential during discharging. No theoretical explanation has been
reported in the literature so far. Therefore, this process cannot be included in the
models at this moment. However, it is believed that the occurrence of this process
might provide an explanation for the differences between the simulation results and
the measurements in both battery models at the initial stages of charging.
As a fourth example, the memory effect has not been included in the NiCd
model. This effect causes a reversible capacity loss when the battery is repeatedly
partially charged and discharged. Although, as mentioned in chapter 3, it is
commonly believed that the memory effect is associated with crystallization
processes that occur at the cadmium electrode, no mathematical description of this
process is known at this moment.
A fifth and final example is the omission of side-reactions from the Li-ion
model. One of the side-reactions that may occur is the formation of metallic lithium
from Li+ ions. This reaction will occur at potentials that are negative with respect to
a lithium reference electrode. Such potentials can occur at the negative electrode
under certain conditions. Metallic lithium is extremely reactive, and any deposited
Li atoms will readily react with the electrolyte to form a passivation layer or SEI;
see section 4.3.1. This will lead to a loss of Li+ ions in the system, and hence to a
capacity loss because the formation of a passivation layer is an irreversible process.
Hence, modelling this side-reaction is also a way of modelling capacity loss. The
lithium deposition reaction can be modelled as a path parallel to the main reaction at
the negative electrode [40]. Both the main and the side-reactions can induce currents
through the capacitance that contains the molar amount of Li+ ions at the negative
electrode surface. The SEI layer can be modelled by a resistance connected in series
with the electrode model. The value of this resistance can be made to depend on the
amount of material that is converted in the side-reaction in order to model an
increasing resistance as a function of cycle life.
Apart from the addition of processes to the models, the temperature-
dependence of the various modelled processes also deserves some attention. It has
been shown in section 4.4 that, although reasonable quantitative agreement can be
obtained between simulated and measured temperature curves, the present
temperature model is probably too simple. An extension has been proposed, in
which the battery temperature is no longer homogeneously distributed through the
battery (pack). This will require further attention in future research. No temperature
dependence of model parameters has yet been considered in the present Li-ion
model. This can readily be included in the future.
4. 7 References
[2] J. Bouet, F. Richard, Ph. Blanchard, "A Discharge Model for the Ni Hydroxide Positive
Electrode", in Nickel Hydroxide Electrodes, D.A. Corrigan, A.H. Zimmerman (Eds.),
PV 90-4, The Electrochemical Society Proceedings Series, Pennington N.J., pp. 260-
280, 1990
[3] J.W. Weidner, P. Timmerman, "Effect of Proton Diffusion, Electron Conductivity and
Charge-Transfer Resistance on Nickel Hydroxide Discharge Curves", J. Electrochem.
Soc., vol. 141, no. 2, pp. 346-351, February 1994
[4] E. Karden, P. Mauracher, F. SchOpe, "Electrochemical Modelling of Lead/Acid
Batteries under Operating Conditions of Electric Vehicles", J. Power Sources, vol. 64,
no. 1, pp. 175-180, January 1997
[5] D. Fan, R.E. White, "A Mathematical Model of a Sealed Nickel-Cadmium Battery", J.
Electrochem. Soc., vol. 138, no. 1, pp. 17-25, January 1991
[6] D. Fan, R.E. White, "Mathematical Modelling of a Nickel-Cadmium Battery- Effects of
Intercalation and Oxygen Reactions", J. Electrochem. Soc., vol. 138, no. 10, pp. 2952-
2960, October 1991
[7] T.F. Fuller, C.M. Doyle, J. Newman, "Simulation and Optimization of the Dual lithium
Ion Insertion Cell", J. Electrochem. Soc., vol. 141, no. l, pp. 1-10, January 1994
[8] C.M. Doyle, T.F. Fuller, J. Newman, "Modelling of Galvanostatic Charge and
Discharge of the Lithium/Polymer/Insertion Cell", J. Electrochem. Soc., vol. 140, no. 6,
pp. 1526-1533, June 1993
[9] S.A. llangovan, S. Sathyanarayana, "Impedance Parameters of Individual Electrodes
and Internal Resistance of Sealed Batteries by a New Non-destructive Technique", J.
Applied Electrochem., vol. 22, no. 5, pp. 456-463, May 1992
[10] S.C. Hageman, "Simple PSpice Models Let You Simulate Common Battery Types",
EDN, pp. 117-132, October28, 1993
[11] S.C. Hageman, "PSpice Models Nickel-Metal-Hydride Cells", EDN, p. 99, February 2,
1995
[12] S. Waaben, I. Moskowitz, I. Frederico, C.K. Dyer, "Computer Modelling of Batteries
from Non-linear Circuit Elements", J. Electrochem. Soc., vol. 132, no. 6, pp. 1356-
1362, June 1985
[13] J.J.L.M. van Vlerken, P.G. Blanken, "Lumped Modelling of Rotary Transformers,
Heads and Electronics for Helical-Scan Recording", IEEE Trans. Magn., vol. 31, no. 2,
pp. 1050-1055, March 1995
[14] W.S. Kruijt, H.J. Bergveld, P.H.L. Notten, Principles of Electronic-Network Modelling
of Chemical Systems, Nat.Lab. Technical Note 261/98, Philips Internal Report, October
1998
[16] K.J. Vetter, Electrochemical Kinetics: Theoretical and Experimental Aspects, Academic
Press, New York, 1967
[17] M. Barak (Ed.), T. Dickinson, U. Falk, J.L. Sudworth, H.R. Thirsk, F.L. Tye,
Electrochemical Power Sources: Primary & Secondary Batteries, lEE Energy Series 1,
A. Weaton&Co., Exeter, 1980
[18] D. linden, Handbook of Batteries, Second Edition, McGraw-Hill, New York, 1995
[19] J.R. MacDonald (Ed.), Impedance Spectroscopy, John Wiley&Sons, New York, p. 59,
1987
[20] J. Homo, M.T. Garcia-Hernandez, "Digital Simulation of Electrochemical Processes by
the Network Approach", J. Electroanal. Chern., vol. 352, no. 1&2, pp. 83-97, June 1993
[21] G.M. Barrow, Physical Chemistry, Fourth Edition, McGraw-Hill, Tokyo, Japan, 1979
[22] C.H. Hamann, W. Vielstich, Electrochemie II: Elektrodenprozesse, Angewandte
Elektrochemie, Taschentext 42, Verlag Chemie, Weinheim, 1981
166 Chapter4

Rechargeable NiCd Batteries", Ext. Abstract I-5-05, 461h ISE Meeting, Xiamen, China,
1995
[24] W.S. Kruijt, H.J. Bergveld, P.H.L. Notten, Electronic-Network Modelling of
Rechargeable NiCd batteries, Nat.Lab. Technical Note 211196, Philips Internal Report,
1996
Rechargeable NiCd Batteries", Ext. Abstract 96, 191" ECS Meeting, Montreal, Canada,
1997
[26] W.S. Kruijt, H.J. Bergveld, P.H.L. Notten, "Electronic-Network Modelling of
Rechargeable Batteries: Part I: The Nickel and Cadmium Electrodes", J. Electrochem.
Soc., vol. 145, no. 11, pp. 3764-3773, November 1998
Rechargeable Batteries: Part II: The NiCd System", J. Electrochem. Soc., vol. 145, no.
11, pp. 3774-3783, November 1998
Rechargeable NiCd Cells and its Application to the Design of Battery Management
[29] W. Randszus, H.A. Kiehne (Eds.), Gasdichte Nickel-Cadmium Akkumulatoren, Varta
Batterie AG, Hannover, Germany, 1978
[30] P.H.L. Notten, A.A. Jansen van Rosendaal, W.S. Kruijt, H.J. Bergveld, "The oxygen
evolution and recombination kinetics in sealed rechargeable NiCd batteries", in
preparation, 2002
[31] S.G. Bratsch, "Standard Electrode Potentials and Temperature Coefficients in Water at
298.15K", J. Phys. Ref Data, vol. 18, no. 1, pp. 1-21, 1989
[32] Gmelins Handbuch der Anorganischen Chemie, Verlag Chemie GmbH, Leipzig-Berlin,
8th Ed., vol. 33, p. 74, 1925
[33] P.H.L. Notten, Chronopotentiometric Measurements at Ni/Cd Batteries and Separate
Electrodes, Nat.Lab. Technical Note 268/92, Philips Internal Report, 1992
[34] M. Avrami, "Kinetics of Phase Change: II. Transformation-Time Relations for Random
Distribution of Nuclei", J. Chern. Phys., vol. 8, no. 2, pp. 212-224, February 1940
[35] S. Motupally, C.C. Streinz, J.W. Weidner, "Proton Diffusion in Nickel Hydroxide
Films: Measurement of the Diffusion Coefficient as a Function of State-of-Charge", J.
Electrochem. Soc., vol. 142, no. 5, pp. 1401-1408, May 1995
[36] Y. Okinaka, "On the Oxidation-Reduction Mechanism of the Cadmium Metal-
Cadmium Hydroxide Electrode", J. Electrochem. Soc., vol. 117, no. 3, pp. 289-295,
March 1970
[37] J. O'M. Bockris, D.M. Drazic, Electro-Chemical Science, Taylor&Francis Ltd.,
London, 1972
[38] S.U. Falk, A.J. Salkind, Alkaline Storage Batteries, John Wiley&Sons, New York, 1969
[39] R.E. Davis, G.L. Horvarth, C.W. Tobias, "The Solubility and Diffusion Coefficient of
Oxygen in Potassium Hydroxide Solutions", Electrochim. Acta, vol. 12, no. 3, pp. 287-
297, March 1967
October 1998
[41] W.S. Kruijt, E. de Beer, P.H.L. Notten, H.J. Bergveld, "Electronic-Network Model of a
Rechargeable Li-ion Battery", The Electrochemical Society Meeting Abstracts, vol. 97-
1, Abstract 104, p. 114, Paris, France, August 31- September 5, 1997
[42] R. Pong, U. von Sacken, J.R. Dahn, "Studies of Intercalation into Carbons Using
Nonaqueous Electrochemical Cells", J. Electrochem. Soc., vol. 137, no. 7, pp. 2009-
2013, July 1990
[43] K. Brandt, "Historical Development of Secondary Lithium Batteries", Solid State
lonics, vol. 69, no. 3&4, pp. 173-183, August 1994
[44] W.R. McKinnon, R.R. Haering, "Physical Mechanisms of Intercalation", in Modem

Aspects of Electrochemistry, vol. 15, B. Conway, R.E. White, I. O'M. Bockris (Eds.),
Plenum Press, New York, pp. 235-304, 1983
[45] T. Ohzuku, A. Ueda, "Phenomenological Expression of Solid-State Redox Potentials of
LiCo02 , LiCo"2Ni 1r202, and LiNi0 2 Insertion Electrodes", J. Electrochem. Soc., vol.
144, no. 8, pp. 2780-2785, August 1997
[46] J.N. Reimers, J.R. Dahn, "Electrochemical and In Situ X-Ray Diffraction Studies of
Lithium Intercalation in Li,Co0 2", J. Electrochem. Soc., vol. 39, no. 8, pp. 2091-2097,
August 1992
[47] T. Ohzuku, Y. Iwakoshi, K. Sawai, "Formation of Lithium-Graphite Intercalation
Compounds in Nonaqueous Electrolytes and Their Application as a Negative Electrode
for a Lithium Ion (Shuttlecock) Cell", J. Electrochem. Soc., vol. 140, no. 9, pp. 2490-
2498, September 1993
[48] P.A. Derosa, P.B. Balbuena, "A Lattice-Gas Model Study of Lithium Intercalation in
Graphite', J. Electrochem. Soc., vol. 146, no. 10, pp. 3630-3638, October 1999
[49] H.J. Bergveld, W.J.J. Rey, R. du Cloux, W.S. Kruijt, P.H.L. Notten, "Quantitative
Optimization of Electronic-Network Models for Rechargeable Batteries", to be
submitted,2002
[50] M.D. McKay, R.I. Beckman, W.J. Conover, "A Comparison of Three Methods for
Selecting Values of Input Variables in the Analysis of Output from a Computer Code",
Technometrics, vol. 21, no. 2, pp. 239-245, February 1979
[51] C.M. Doyle, J. Newman, A.S. Gozdz, C.N. Schmutz, J.M. Tarascon, "Comparison of
Modelling Predictions with Experimental Data from Plastic Lithium Ion Cells", J.
Electrochem. Soc., vol. 143, no. 6, pp. 1890-1903, June 1996
169
Chapter 5
Battery charging algorithms
One of the links in the energy chain is the charger, as explained in chapter 1. In the
charger, electrical energy from the mains is transferred via magnetic energy into
electrical energy suitable for charging the battery. Two types of control can be
distinguished in the charger or power module, notably control of the energy
conversion process and control of the charging process; see chapter 2. The latter
control process is implemented in the form of a charging algorithm. A charging
algorithm is an example of a battery management function with which the battery
itself is monitored and the energy conversion process in the charger is controlled in
order to charge the battery in an efficient way.
Batteries with different chemistries require different charging algorithms.
Examples of charging algorithms for NiMH and Li-ion batteries have been
described in section 2.4. The charging algorithms for these battery chemistries will
be discussed in greater detail in this chapter. The charging algorithms for NiCd and
NiMH batteries will be dealt with in section 5.1, the charging algorithm for Li-ion
batteries in section 5.2, in which the influence of the current and voltage parameters
on the charging process will also be discussed.
An example of the use of the NiCd battery model for comparing two different
charging methods will be given in section 5.1.3. Simulation results will be compared
with measurement results. The results of simulations using the Li-ion model will be
used to demonstrate the difference between normal and fast charging in section
5.2.4. A possible mechanism for the loss of cycle life encountered in fast charging in
practice will be derived from the results of simulations. Finally, conclusions will be
drawn in section 5.3.
5.1 Charging algorithms for NiCd and NiMH batteries
5.1.1 Charging modes, end-of-charge triggers and charger features

NiCd and NiMH batteries are usually both charged with a constant current. When
fast charging is applied, this current is switched to a lower value when a certain end-
of-charge trigger occurs. In practice, this trigger is based on time, voltage or
temperature. More than one trigger is used in most chargers. The first trigger
occurring during the charge process leads to a change in the charge current. By
switching to a lower current, the battery is protected from excessive internal
pressure build-up and temperature rise.
As a general rule, the higher the charge current, the more care should be taken
to avoid overcharging by switching off the high current in time. Therefore, the
triggers that are used depend first of all on the applied charge current. Moreover,
some triggers are more suitable for NiCd and some for NiMH batteries. This will be
discussed in section 5.1.2. Finally, the choice of a certain end-of-charge trigger
depends on the battery's acceptable cycle life. The greater the amount of
overcharging is allowed by applying a certain end-of-charge trigger, the lower the
battery's cycle life will be, because overcharging has a negative influence on the
cycle life. In general, the charging algorithm has to be more sophisticated when the
amount of overcharging has to be limited, because deriving triggers from

170 Chapter 5
measurements is more complicated in that case. Charge control, and hence the
battery management system, consequently become more complicated when a battery
with a longer cycle life is required. This was already described in section 2.1.
Charging modes
The charging modes are categorized by the value of the charge current as indicated
below [1]. The battery will be charged with more than 100% of its rated capacity in
most cases to obtain a fully charged battery. This is needed because some of the
applied energy is lost in side-reactions. Moreover, several charging modes will
sometimes be combined. An example is the application of the sequence fast
charging, top-off charging and trickle charging. See chapter 3 for the definition of
C-rate for charge currents.
1. Low-rate charging
In the case of low-rate charging the charge rate is around 0.1 C and the charging is
stopped by a timer. The charge time is 14-16 hours in the case of a fully discharged
battery and a 150% applied charge. Low-rate charging is used mainly in cheap
overnight chargers, in which the charging 'algorithm' is very simple, because only a
timer is used. In the simplest case, the battery is charged by a DC voltage source
with a series resistance. The charge current decreases, because the battery voltage
gradually rises during the charging. The term 'semi-constant current charging' is
then used. In practice, examples can be found of continuous charging with a low
rate.
2. Quick charging
A 0.3 C charge rate is applied in quick charging. The charge time for a fully
discharged battery when the total charge applied to the battery ranges from 120% to
150% is 4 to 5 hours. Quick charging is usually ended with a timer.
3. Fast charging
In the case of fast charging, the charge rate is higher than or equal to 1 C in practice.
This means that the maximum charge time is just over one hour. The actual charge
time will depend on the battery's SoC before charging and on the amount of
overcharging resulting from the chosen end-of-charge trigger. For example, a 120%
charge with a 1 C charge rate implies a charge time of 1 hour and 12 minutes when
the charging is started with a completely empty battery. Ultra-fast chargers with
charge rates over 4 C, hence with charge times less than 15 minutes, have been
reported for NiCd batteries [2],[3). It should be noted that the charged capacity will
be lower than 100% in this case. Proper end-of-charge triggers based on voltage and
temperature measurements have to be implemented for fast charging of both NiCd
and NiMH batteries to prevent the risk of serious damage to the batteries.
4. Top-off or equalization charging

Top-off charging at a rate of around 0.1 Cis applied directly after fast charging. It is
more often applied to NiMH batteries than to NiCd batteries. Battery manufacturers
sometimes recommend it. As the term indeed implies, top-off charging is used to
charge a battery to a full 100% charge after the fast-charge current has been
switched off. Another reason to apply top-off charging occurs when several cells are
charged in series, as in a battery pack. Top-off charging is then applied to ensure
that all the cells are fully charged after an end-of-charge trigger has been generated
by the combination of cells. Therefore, the term equalization charging is sometimes
Battery charging algorithms 171
encountered in the literature [2], although the process does not lead to equal cell
capacities. In most cases, the top-off charging period is terminated by a timer, for
example after half an hour or after one hour.
5. Trickle or maintenance charging

A relatively small trickle or maintenance charge rate of 0.03 C to 0.05 C is applied
to the battery to compensate for self-discharge. Trickle charging serves to maintain a
100% charged battery when it is left in the charger for some time. Trickle charging
begins directly after fast charging has ended when no top-off charging is applied.
Alternatively, it can also start directly after quick or top-off charging has ended,
when applicable. There is generally no time limit for trickle charging, so it can
continue for an indefinite amount of time.
6. Reflex or 'burp' charging

This method of charging is not often encountered in practice. It involves charging a
battery with current pulses followed by short rest periods with zero current, a short
period of discharge and another short rest period. The inventors of this charging
scheme claim that a battery can be charged more efficiently with Reflex charging.
They argue that the discharge pulses or 'burps' help to maintain a low internal gas
pressure and battery temperature. However, no hard proof of this is given. An
example can be found in [3].
End-of-charge triggers
Depending on the charging mode, several end-of-charge triggers or combinations of
these triggers can be used [1]-[4]. The end-of-charge triggers most frequently
encountered are listed below. After a short description, these triggers will be
compared in terms of the amount of overcharging they cause. This is performed in
Figure 5.1, which compares most of the end-of-charge triggers mentioned below.
1. Timed end-of-charge trigger

A timer is used in this method to switch off the charge current. It can be used as a
stand-alone end-of-charge trigger to terminate low-rate or quick charging. Moreover,
a timed end-of-charge trigger is often used to terminate a period of top-off charging.
However, detrimental overcharging may occur in the case of higher charge currents
when only a timer is used to end the charging of a battery which was not fully
discharged when the charging began. The timed end-of-charge trigger is therefore
used only as a fail-safe limit in such cases. This means that a timer will eventually
force a switch-over of the charge current to a trickle-charge current when all other
end-of-charge triggers fail.
2. Maximum temperature end-of-charge trigger

As described in chapter 4, the temperature of NiCd batteries will rise quite steeply at
the end of charging. The same holds for NiMH batteries. This rise in temperature
can be used to derive end-of-charge triggers. For example, a battery can be charged
until a maximum battery temperature has been reached. Obviously, a disadvantage
of this method is the fact that a battery will virtually not be charged at high ambient
temperatures, whereas it may be substantially overcharged at low ambient
temperatures. Therefore, a maximum battery temperature is usually used as a fail-
safe limit for fast charging.
I72 Chapter 5
3. LJT end-of-charge trigger

With a LJT end-of-charge trigger, the ambient temperature is also taken into account.
Two temperature sensors have to be used, one of which measures the ambient
temperature and the other the battery temperature. Charging is stopped when the
difference between battery and ambient temperatures reaches a certain
predetermined value. This value depends on several factors, including the battery
size, configuration, heat capacity and, where applicable, the number of cells. It must
therefore be determined separately for each type of battery before implementation.
Care should be taken that the battery temperature does not influence the ambient
temperature sensor. The ambient temperature sensor is usually attached to a dummy
thermal mass which is thermally isolated from the battery. This end-of-charge
trigger is usually used to terminate fast charging.
4. LJT/LJt end-of-charge trigger

In order to set off a ..1T/..1t end-of-charge trigger, the change in time of the battery
temperature has to be greater than a predetermined value. In this case, the influence
of the ambient temperature is virtually eliminated. A reasonable predetermined value
is I 0 C/min. This end-of-charge trigger is usually used to terminate fast charging.
5. Maximum voltage end-of-charge trigger

The charge current is in this case switched to a trickle-charge current when the
battery voltage reaches a certain predetermined value. This is a highly unreliable
method, because the battery voltage depends on many factors, such as the battery's
series resistance, temperature and age and the type of battery concerned. Currentless
measurement of the battery voltage can eliminate the influence of the series
resistance. This is applied in most fast chargers that use end-of-charge triggers based
on battery voltage.
The problem that arises when a maximum voltage is used to switch the charge
current to a trickle-charge current is twofold. When the maximum voltage level is
chosen too low, the battery will not be fully charged. When it is taken too high, the
danger exists that the battery voltage will never reach this level. This is especially
true because the battery heats up at the end of charging which leads to a drop in
battery voltage. A maximum battery voltage is used only as a fail-safe limit for fast
and quick charging on account of the problems mentioned above.
6. -..1V end-of-charge trigger

The voltage of a NiCd battery is a function of temperature and therefore it drops
slightly during overcharging; see chapter 4. The charging of NiCd batteries is often
stopped when the battery voltage or the average cell voltage in a battery pack has
dropped IO to 20 mV. -LJV end-of-charge triggers are less suitable for NiMH
batteries because the drop in voltage at the end of the charging is significantly lower.
A voltage drop of I 0 m Vper cell is recommended to set off the trigger in the case of
NiMH batteries. The allowed drop in voltage should not be too small, as this would
make the trigger sensitive to noise. On the other hand, a too large value will lead to
detrimental overcharging. This is especially true at lower charge currents and
elevated ambient temperatures, because then the battery voltage drop will be lower
and it will take longer for the trigger to be set off. This end-of-charge trigger is not
recommended for charge rates below 0.5 C, because the -LJV value is then hardly
detectable. This end-of-charge trigger is usually used to terminate fast charging.
7. Voltage plateau or dV/dt=O end-of-charge trigger

This trigger is based on the shape of the battery voltage curves during the charging
of NiCd and NiMH batteries. As described above, before the voltage drop occurs
both types of battery show a voltage plateau. Here, the first derivative of the battery
voltage is zero, in other words dV/dt=O. Care should be taken to ensure that the
circuitry with which this end-of-charge trigger is implemented is sufficiently
sensitive, because the battery voltage may also be fairly flat in the mid-region of
charging. Of course, this should not trigger the circuitry. The battery voltage
samples are usually filtered by calculating the running average value over several
samples before calculating the derivative. This is done to reduce the influence of
noise disturbances on the calculation of the derivative. The trigger is set off when
the derivative is zero. The voltage plateau end-of-charge trigger is usually used to
terminate fast charging.
8. Inflection-point or tfV!dr=O end-of-charge trigger [2]

Like the previous two methods, this method is based on the general shape of the
battery voltage curve during charging. Before the battery voltage reaches a plateau,
an inflection point occurs. At this point, the first derivative of the battery voltage is
at its maximum and hence the second derivative of the battery voltage is zero, or
tfvldr=O. Similar remarks as given above for the dV/dt end-of-charge trigger hold
with respect to sensitivity and filtering. The trigger is set off when the calculated
second derivative is zero. This trigger is not very often encountered in commercial
fast chargers because the number of calculations that have to be performed is rather
large.
9. Pressure-based end-of-charge trigger

In NiCd batteries, the internal gas pressure rises at the end of charging as illustrated
in chapter 4. This also holds for NiMH batteries. This phenomenon could
theoretically be used to derive a trigger. The pressure could be measured by a
pressure sensor that is in contact with the battery's interior through a hole drilled in
the battery casing. This is however not very practical. Alternatively, the pressure
could be measured by strain gauges set around the battery casing. However, no
commercial application of strain gauges has so far been introduced.
Comparison of the end-of-charge triggers

The end-of-charge triggers based on voltage and temperature measurements are
summarized in Figure 5.1. The use of a maximum time (tmax), temperature (Tmax) and
voltage (Vmax) as a fail-safe limit is also shown. The depicted voltage and
temperature curves hold for a NiCd battery charged at a rate of 1 C. Although
similar, these curves will look somewhat different for NiMH batteries, as will be
explained in the next section.
Figure 5.1 shows that use of a -LlV end-of-charge trigger leads to significant
overcharging. In this figure, the -LtV threshold Par1 has been chosen to be 20 mV as
an example. More sophistication should be included in the system to ensure less
overcharging with a voltage-based end-of-charge trigger. Figure 5.1 clearly
illustrates that overcharging is virtually prevented with the inflection-point or
tfV!dr=O end-of-charge trigger. As a consequence, the temperature and internal gas
pressure hardly rise, as explained in chapter 4. However, calculating the second
derivative of the battery voltage involves considerably more signal-processing effort
than determining a -LtV voltage drop.
174 Chapter 5
~
>1.4
40
20
0.5 1.0 1.5

t(hours]-
Figure 5.1: Overview of end-of-charge triggers based on voltage and temperature

measurements. Typical shape of battery voltage (V [V]) and temperature (T reD for charging
a NiCd battery at a 1 C charge rate at 28°C ambient temperature. These curves have been
drawn by hand as an illustration of their typical shape
Both temperature-related triggers result in less overcharging than the -LlV end-of-
charge trigger, as can be seen in Figure 5.1. Obviously, the values of parameters
Par2, the threshold for the LiT trigger, and Par3, the threshold for the LlT/Llt trigger,
should be chosen low enough, because otherwise the battery temperature rises too
high. Less overcharging and overheating means less loss of cycle life. Figure 5.2
shows the advantage in terms of cycle life of using a LlT/Llt trigger instead of a -LlV
trigger in a NiMH battery [1]. The cycling conditions were charging at a 1 C-rate,
resting for 0.5 hour, discharging to a 1 V battery voltage and resting again for 2
hours.
A battery's cycle life is usually defined as the number of charge/discharge
cycles that a battery can experience until 80% of the initial capacity remains. Figure
5.2 shows that, according to this definition, the cycle life is only 500 cycles when a-
LlV trigger is used, whereas the cycle life is more than 700 cycles when a LlT/Llt
trigger is used. Hence, although the implementation of the latter trigger requires
more effort as it implies including a temperature sensor and calculating the
derivative of the measured temperature versus time, it yields a better cycle life.
Charger features
Apart from one or more end-of-charge triggers and fail-safe methods, such as
maximum time or temperature, various battery parameters are also checked before
charging is started. First of all, most fast chargers check for any faults, such as a
short-circuit or an open-circuit. Such faults could indicate a damaged battery. In
addition, most fast chargers check the initial battery voltage and temperature. The
battery voltage and temperature both have to be within a certain range to enable fast
charging. The battery is in an overdischarged state when the initial battery voltage is
below a minimum level, for example below 0.9 V. In this case, it has to be charged
with a relatively small current until the battery voltage has risen above this
minimum level. This is done to prevent battery damage. Afterwards, fast charging
can proceed with a higher current. The battery can only be charged with a relatively
small current when the initial battery temperature is out of range, for example
outside the range from 0°C to 40°C. Fast charging with a higher current may proceed
when the battery temperature re-enters the required range.
t 80!--------------=~~------~~_J
*-.?:-
l4o
(.)
200 400 600

Number of cycles -
Figure 5.2: Battery capacity as a function of the number of charge/discharge cycles of a

NiMH battery using either a -LtV or a LiT/At end-of-charge trigger to terminate fast charging at
a 1 C charge rate (measured curves redrawn from [1] with permission)
Premature set-off of any of the applied end-of-charge triggers may occur at the
beginning of fast charging, for example on account of battery voltage peaks.
Therefore, the use of some of the triggers is inhibited in the initial stages of fast
charging in most commercial fast chargers.
Especially NiCd batteries may suffer from the memory effect; see chapter 3. A
method for restoring full battery capacity involves forcing a complete discharge
until the battery voltage has become 0.9 V. In order to accomplish this, some
commercial fast chargers offer an optional discharge feature. The user can start the
discharging by pushing a button. The user is recommended to fully discharge the
battery after a predetermined number of charge/discharge cycles.
5.1.2 Differences between charging algorithms for NiCd and NiMH batteries
Although the charging characteristics of NiCd and NiMH batteries are quite similar,
each type of battery requires its own charging algorithm, mostly because NiMH
batteries are more sensitive to overcharging than NiCd batteries. In the case of these
batteries, the end-of-charge triggers should therefore be chosen more carefully,
because of their impact on the battery's cycle life. In general, the triggers that are
recommended for NiMH batteries are more sophisticated and lead to less
overcharging than the triggers used for NiCd batteries. It is not recommended to
charge NiMH batteries in less than one hour, because of their sensitivity to
overcharging.
Figure 5.3 gives a schematic representation of the main differences in charging
characteristics between NiCd and NiMH batteries [1]. The voltage curves in this
figure look similar. However, the shape of the voltage curve of the NiMH battery is
less pronounced than that of the NiCd battery. This difference becomes increasingly
176 Chapter 5
evident at lower charge rates and higher temperatures, although the voltage
characteristic of a NiCd battery also becomes less pronounced under these
conditions. Because of this, -LlV triggers are less suitable for NiMH batteries, as
already discussed in the previous section. Another reason for this is that such
triggers lead to considerable overcharging, to which NiMH batteries are more
sensitive than NiCd batteries.
1.6
~
t 1.5
Cl)
Cl
ctl
'S> 1.4
G)
(.)
1.3
0 40 80 120 160
(a) Charge input, % of typical capacity -
t 60
, ,,
, ,,
5'
e..... 40
....
, .. , ;
!!! ............
.... ........ ........ NiCd

:I
(ij
....
Cl) _
0.
E 20
.2!
G)
(.)
0
0 40 80 120 160
(b) Charge input, % of typical capacity -
Figure 5.3: Comparison of typical characteristics of NiCd (dashed curves) and NiMH
batteries (solid curves) during charging at a moderate constant-current charge rate: (a) voltage
characteristic (b) corresponding temperature development (measured curves redrawn from [1]
with permission)
As can be seen in Figure 5.3, the temperature of NiCd batteries rises only at the end
of charging. The moment at which the temperature starts to rise more or less
coincides with the steeper rise in battery voltage. This is due to the side-reactions
that take over from the main storage reactions; see chapter 4. The net charge
reaction, which takes place roughly before 80% charge input has been reached in
Figure 5.3, is endothermic in a NiCd battery. As a result, the temperature remains
relatively constant in this region. In a NiMH battery however, the net charge
reaction is exothermic. This means that the temperature rises even far before 80% of
the end of the charge process is reached. The exothermic nature of the side-reaction
leads to an even greater increase in battery temperature when the battery approaches
full charge. This is similar to the increase in temperature in NiCd batteries at the end
of charging.
Another difference in temperature behaviour is observable during
overcharging. At moderate charge rates, the battery temperature will level off to a
steady-state value in NiCd batteries. This was illustrated in several charge curves in
chapter 4. In NiMH batteries, however, the battery temperature continues to rise
during overcharging [1]. This and the higher battery temperature near the end of
charging in NiMH batteries makes it clear why overcharging should be prevented as
much as possible in NiMH batteries. A high battery temperature is a disadvantage
with respect to cycle life, and high temperatures occur already relatively early during
overcharging.
The type of end-of-charge trigger most frequently used in NiCd batteries is the
-L1V trigger, because of its simplicity and the fact that NiCd batteries are relatively
robust with respect to overcharging. An alternative has to be used when the charge
rate is lower than, for example, 0.5 C, because the voltage drop then becomes
virtually undetectable. In that case, a simple alternative could be timed charging
with thermal cut-off protection.
The end-of-charge trigger recommended for NiMH batteries is either the
dV!dt=O or one of the temperature-difference triggers L1T or L1T/L1t. As was shown in
Figure 5.1, these end-of-charge triggers lead to less overcharging than -L1V triggers.
Figure 5.2 showed that L1T/L1t triggers yield a significant improvement in cycle life
over -L1V triggers. dV/dt=O triggers may also be used when no clear peak is
observable in the voltage characteristic, for example at low charge currents or high
temperatures. A combination of a dV!dt =0 and a L1TIL1t trigger was used for a NiMH
battery in the example of the cellular phone in section 2.4. A L1T trigger was used in
the example of the low-end shaver.
5.1.3 Simulation example: an alternative charging algorithm for NiCd

batteries
The battery and charging electronics are packed close together in a small volume in
some portable products, such as shavers. The main problem encountered in practice
is a high internal temperature, which has a negative influence on the battery's cycle
life. A simulation of charging a NiCd battery inside a shaver will demonstrate this
problem in the first part of this section. An alternative charging algorithm has been
simulated as a possible solution [5]. Such simulations have to be performed using a
thermal model of a shaver, which will be described next.
Thermal shaver model

The charger is a Switched-Mode Power Supply (SMPS) in flyback configuration [6].
It consists of a rectifying mains buffer, a transformer, a High-Voltage IC (HVIC)
that regulates the charge current, a rectifying secondary diode and an electrolytic
smoothing output capacitor. An example of such a charger inside a shaver was
shown in Figures 2.9 and 2.10 in chapter 2.
178 Chapter 5
A simplified thermal model of the electronics in the shaver that are relevant during
charging can be made on the basis of the results of temperature-difference and
relaxation measurements. Three heat-producing components can be distinguished in
the shaver:
• The NiCd battery.

• The HVIC that controls the charge current.
• The remaining supply parts (transformer, rectifying mains buffer, diode).
The thermal model of the electronics in the shaver can be coupled to the thermal
model of the battery shown in Figure 4.19(b) using the principles of physical system
dynamics [7]. The thermal shaver model takes into account the thermal resistance
R,h from the battery to its surroundings and the thermal properties of the battery's
surroundings. This thermal model therefore replaces R ,h and C,hamb in Figure 4. 19(b)
and is connected at node Tbar· This is illustrated in Figure 5.4, in which the
remaining supply parts are denoted as supply [5].
Tamb
Rsi-c Tamb
Tamb= 25°C
Figure 5.4: Thermal and electrical simulation model of a shaver with an integrated NiCd
battery and charger (see also Table 5.1)
The complete network connected to node Tbar is represented in the thermal domain.
Each heat-producing component is represented by a heat source P;, which models
the dissipation, and a heat capacitance C;. This combination is coupled to the
constant ambient temperature, modelled by a constant temperature source Tamh• and
to the two other heat-producing components by means of heat resistances R;. The
ambient temperature remains constant because the shaver itself will not heat up its
environment. However, the heat-producing components inside the shaver will
significantly influence each other's temperature. An additional heat resistance Rs;.c
from the silicon to the outside of the package of the HVIC is taken into account. The
actual charging process is modelled in the electrical domain by means of a ucrrent
source ! charge·
A 50% efficiency has been found for the power supply in practice. This means
that the dissipated power equals the electrical power ! chargeV bar supplied to the
battery. Again, it has been found in practice that approximately 1/3 is dissipated in
the IC and 2/3 in the remaining supply parts. The ambient temperature and the
temperatures of the remaining supply parts, HVIC package and battery are
represented at the nodes T amb• Tsupply , T 1c and That• respectively. The values of the
components in the thermal model are summarized in Table 5.1. These values were
derived from measurements.
Table 5.1: Values of the components of the thermal network shown in Figure 5.4
Component Description Value Unit
R1 Heat resistance from remaining supply parts to ambient 47 KIW
Rz Heat resistance from battery to remaining supply parts 93 KIW
RJ Heat resistance from battery to ambient 41 KIW
R4 Heat resistance from remaining supply parts to HVIC 50 KIW
Rs Heat resistance from battery to HVIC 166 KIW
R6 Heat resistance from HVIC to ambient 232 KIW
Rsi-c Heat resistance from internal silicon to outside package 15 KIW
ofHVIC
C1c Heat capacitance ofHVIC 2 J/K
Csupply Heat capacitance of remaining supply parts 8 J/K
chnl Heat capacitance of battery (see Figure 4.19(b )) 36 J/K
Simulation of a galvanostatic charging algorithm inside a shaver

The result of the simulation of the use of a galvanostatic charging algorithm is
shown in Figure 5.5. Galvanostatic charging means that the current remains constant
until it is switched to a trickle-charge current. This is applied in shavers nowadays.
The simulation model shown in Figure 5.4 was used. The parameter set of Tables
4.3 and 4.4 was used in the NiCd battery model and the values specified in Table 5.1
were used for the heat resistance and capacitance of the battery. The charging
efficiency is defined as the ratio of the effectively stored electrical charge and the
total applied electrical charge. A constant charge current !charge of 1.2 A was applied
for 1 hour.
Figure 5.5 shows a rise in battery temperature to 62°C at the end of charging.
The charging efficiency clearly drops at the end of the charging, which is partly due
to the elevated battery temperature. A lower charging efficiency at a higher battery
temperature was demonstrated earlier in Figure 4.44. Moreover, the self-discharge
rate is higher at elevated temperatures. This was shown earlier in Figure 4.45. The
battery's State-of-Charge (SoC) drops as a result of self-discharge after the charging
has stopped, because the high battery temperature slowly drops to the ambient
temperature; see chapter 6 for a definition of SoC. After 2 hours, the battery's SoC
is 94%, while the charging efficiency at that point is 80%.
Simulation of a thermostatic charging algorithm inside a shaver

The result of simulating an alternative charging algorithm is shown in Figure 5.6.
The same simulation conditions as in Figure 5.5 were chosen, except that the charge
current !charge was made dependent on the battery temperature in such a way that the
battery temperature never exceeded 35°C. This may be called thermostatic charging,
i.e. charging with a maximum temperature. A total charge time of 1 hour was used,
as in the galvanostatic simulation illustrated in Figure 5.5.
Figure 5.6 illustrates the maximum battery temperature of 35°C and the
decrease in charge current during charging. The drop in charging efficiency is
virtually zero due to the lower battery temperature. Even after two hours, the
charging efficiency was still 99.5%, as opposed to the 80% observable in Figure 5.5.
180 Chapter 5
Moreover, the drop in SoC was virtually zero after the charge current was
interrupted. The SoC was 80% after one hour of charging, which is 14% lower than
the value in Figure 5.5. However, the applied electrical energy was considerably less
than in the case of galvanostatic charging and this led to less dissipation. An
advantage of thermostatic charging is that a considerably longer cycle life may be
anticipated, due to a reduction in maximum battery temperature. The results of the
simulated galvanostatic and thermostatic charging experiments inside a shaver are
summarized in Table 5.2.
l lll
1.6 0
100
charging efficiency
lcharge
:i 1.2 75 ~ u
e...
~
e...
>. (.)
C) 1- 0
c::::
Q) en
0.8 50 :§
=
Q)
Cl
c::::
-~
0.4 «<
25 .r:.
C)
0 0
0 0.5 1.0 1.5 2.0
t[hours] ~
Figure 5.5: Simulated charge current (lclwrge [A]), State-of-Charge (SoC [%]), charging
efficiency [%] and battery temperature (Tbat [°C]) as functions of time (t [hours]) for charging
for one hour with a current of 1.2 A at 25°C ambient temperature. A NiCd battery model with
the parameter set of Tables 4.3 and 4.4 was used in the simulation model shown in Figure 5.4
l lll
1.6 100
charging efficiency
SoC
:i 1.2 75 ?fl. u
e...
~
e...
>. ()
C) 1- 0
c::::
Q) en
0.8 50 ~Q)
Cl
c::::
Tbat -~
0.4 «<
25 .r:.
C)
0
0.5 1.0 1.5 2.0
t[hours] ~
Figure 5.6: Same curves as in Figure 5.5 for the simulation of an alternative charging
algorithm. In this case, the battery temperature was not allowed to rise above 35°C
Table 5.2: Comparison of the results of the simulated galvanostatic (Figure 5.5) and
thermostatic (Figure 5.6) charging experiments inside a shaver
Method: t.ha..... [min] SoC[%] Charging eff. [%] TmaxrCJ
Galvanostatic 60 94 80 62
Thermostatic 60 80 99.5 35
In the above example the same charge times were chosen for galvanostatic and
thermostatic charging. Galvanostatic and thermostatic charging can also be
compared on the basis of charge times. This will be considered next, in both
simulations and measurements.
Simulations and measurements offast thermostatic charging

Thermostatic charging can also be applied for faster charging at low battery
temperatures. This will be demonstrated below by means of simulations and
measurements. For simplicity, the simulations and experiments were performed
using a NiCd battery in free space. Therefore, the thermal environment was taken to
be the same as in Figure 4.19(b) in the simulations and the thermal shaver model
was omitted. This means that a thermal resistance Rrh of 25 KIW was assumed from
the battery to the environment and a battery heat capacitance C1hbar of 14 J/K was
used. These values are in accordance with the values in Table 4.3. The environment
itself had a fixed temperature of 25°C, which was forced in the simulation by using a
constant temperature source connected to node Tamb in the thermal model part of
Figure 4.19(b ). The battery was charged in either a galvanostatic or a thermostatic
mode. A current of 1.2 A was used for the galvanostatic charging. The
measurements were performed using a Gates 600 mAh NiCd battery. A charge time
of 40 minutes was chosen for the measurements because the capacity of this battery
is lower than that of the NiCd model used in the simulations. A charge time of 60
minutes was applied in the galvanostatic simulation.
The charge current started at a value of 4 Ain the case of thermostatic charging
and was lowered to zero in eleven steps. Ten closely spaced temperature thresholds
in a range from 29°C to 31.5°C were used in both the simulations and the
measurements. The current was stepped down each time the battery temperature
reached the value of a temperature threshold. The current was set to zero and the
charge time ended when the battery temperature reached the final threshold of
31SC. This implementation enabled us to use the available equipment in a
relatively simple set-up. The two charging methods are summarized in Table 5.3.
The simulated results of the galvanostatic charging are shown in Figure 5.7.
The figure shows that the battery was overcharged, which led to a battery
temperature of 54°C. The attained SoC value was 95% and the charging efficiency
was 81% at the moment when the charge current was switched off after 60 minutes.
The simulated result of thermostatic charging is shown in Figure 5.8. The
current steps listed in Table 5.3 can be clearly recognized in this figure. A charge
time of 35 minutes resulted in this case. The SoC was 89% and the charging
efficiency was 94% at the end of the charging. The highest battery temperature, of
31.5°C, occurred at the end of the charging. Hence, although the charge time used in
the thermostatic simulation was around a factor of two lower than that used in the
galvanostatic simulation, the SoC was about the same. In addition to the benefit of a
charge time reduced by a factor of two, the battery temperature remained more than
20°C lower than in the galvanostatic charging simulation. The results obtained with
the two charging methods in simulations are summarized in Table 5.4.
182 Chapter 5
Table 5.3: Characteristics of the galvanostatic and thermostatic charging methods used in the
simulations and measurements
Temperature T rCJ Charge current I [A]
Galvanostatic Thermostatic
T<29.25 I = 1.2 A for all temperatures for a 4
29.25 :S T < 29.5 fixed time of 60 minutes in the 3.6
29.5 :S T < 29.75 simulations and 40 minutes in the 3.2
29.75 :S T < 30 measurements 2.8
30 :S T < 30.25 2.4
30.25 :S T < 30.5 2
30.5 :S T < 30.75 1.6
30.75 :S T < 31 1.2
31 :S T < 31.25 0.8
31.25 :S T < 31.5 0.6
T~ 31.5 0
Table 5.4: Comparison of the results of simulated galvanostatic and thermostatic charging
according to Table 5.3
Method: trn...... [min] SoC[%] Charging eff. [%] TmaxrCJ
Thermostatic 35 89 94 31.5
The measured results of the galvanostatic and thermostatic charging experiments are
compared in Table 5.5.
t 2.0 .....--------oo:::::------------. 100

I SoC
t t t
1.6
iS
s
1- (.)
1::
Q)
50~
0.8 Cl
1::
-~
25 Ill
0.4 £
0
20 40 60 80
t[min]-
Figure 5.7: Simulated charge current Ucharge [A]), State-of-Charge (SoC [%]), charging
efficiency [%] and battery temperature (Tbat ["C]) as functions of time (t [min]) for
galvanostatic charging of a NiCd battery model for one hour using 1.2 A at 25"C ambient
temperature. The parameter set of Tables 4.3 and 4.4 was used
i 6.0
charging efficiency
SoC
100
iii
<( 75 ~
~
u
~
(/!..
- 4.0 >. u0
u I-
I::
QJ en
"(3
50 :E
QJ
Cl
2.0 I::
"6>
Tbat
25 ~
.s:::
u
0
20 40 60 80
t [min) ------+-
Figure 5.8: Same simulated curves as in Figure 5.7 for thermostatic charging according to
Table 5.3 at 25°C ambient temperature
Table 5.5: Comparison of the results of measured galvanostatic and thermostatic charging
according to Table 5.3
Method: 1chanre (min1 SoC [%1 Charging eff. [% 1 Tmax (°C1
Thermostatic 20 98 94 31.5
Comparison Tables 5.4 and 5.5 illustrates the reduction by a factor of two of the
charge time realized with thermostatic charging in both the simulations and the
measurements. The SoC for galvanostatic charging is somewhat higher than for
thermostatic charging in the simulations. The SoC was practically equal in the case
of galvanostatic and thermostatic charging in the measurements. The charging
efficiency in thermostatic charging was higher than that in galvanostatic charging in
both the simulations and the measurements. The simulated and measured
efficiencies agreed very well with each other.
The temperature at the end of the galvanostatic charging was only slightly
higher than that at the end of the thermostatic charging in the measurements; see
Table 5.5. In the case of the simulations, this difference was significantly higher.
This means that the battery was overcharged more severely in the galvanostatic
charging simulation. The SoC and the temperature at the end of the charging will
decrease when the charge time of 60 minutes is lowered in the simulations. In that
case, the simulation result of galvanostatic charging will approach the measurement
result shown in Table 5.5. However, the difference in charge time between
galvanostatic and thermostatic charging will then be less than a factor of two.
Further measurements indicated that even a charge time of 12 minutes to an
SoC of 94% was possible. The charge current was not stepped down further than 2
A in that case and the maximum charge temperature was chosen 33.5°C. This
illustrates that the thermostatic charging algorithm can be optimized towards
184 Chapter 5
minimum charge time (12 minutes versus 20 minutes) at the cost of a slight increase
in battery temperature (33.5°C versus 31.5°C).
Simulation offast thermostatic charging inside a shaver

Similar simulations of fast thermostatic charging were performed for a NiCd battery
inside a shaver using the simulation model illustrated in Figure 5.4. The temperature
thresholds were chosen to be a little higher in order to obtain a reasonable SoC
because of the higher overall battery temperature. The conditions are listed in Table
5.6 and the simulation results are summarized in Table 5.7.
Table 5.6: Thermostatic charging regime used to simulate the charging of a NiCd battery
inside a shaver; see also Figure 5.4 and Table 5.1
Temperature T rc1 Ch1U'2e current I [A]
T<37 4
37 :5T<39 3.6
39 :5T <40 3.2
40:5T<41 2.8
41 :5T<41.5 2.4
41.5 :5 T < 42.0 2
42.0 :5 T < 42.2 1.6
42.2 :5 T < 42.4 1.2
42.4 :5 T < 42.6 0.8
42.6 :5 T < 42.8 0.6
T~42.8 0
Comparison of Tables 5.7 and 5.2 shows that the same SoC can be attained with the
thermostatic charging regime of Table 5.6 as in galvanostatic charging in 4 minutes
less charging time. This difference in charging time is a lot less impressive than the
result obtained in charging a NiCd battery in free space. However, the previous
disadvantage of 14% less SoC has now been eliminated. The temperature at the end
of the charging was still 19.2°C lower than in the case of galvanostatic charging.
Hence, the advantage of an anticipated longer cycle life still holds.
5.2 Charging algorithm for Li-ion batteries
5.2.1 The basic principle

ALi-ion battery has to be charged with a so-called CC/CV regime. This was already
described and illustrated in chapter 4; see Figure 4.48. The basic voltage and current
characteristics for charging aLi-ion battery are repeated in Figure 5.9.
-
Figure 5.9: Battery voltage (V) and current (l) characteristics during the application of a
CC/CV charging regime to aLi-ion battery
A battery manufacturer specifies both the values of the current Icc and the maximum
voltage Vmax· The charge rate in CC mode ranges from 0. 7 C to 1 C, which is related
to the allowed current density. The value of Vmox depends on the type of Li-ion
battery. It can be either 4.1 V ± 1% or 4.2 V ± 1%. The reason for the tight
requirements with respect to the accuracy of this voltage was explained in chapter 2;
see Figure 2.5. The value of Imin is usually chosen to be between a 0.05 C and a 0.1
C-rate. Alternatively, the total charging time is usually set to 2 hours when tmax is
used to terminate the charging. As an illustration, the charge time tc~wrge when lmin is
used to end charging is smaller than tmax in Figure 5.9.
There is a difference between using lmin or tmax to terminate the charge. The
advantage of using Imin is the fact that the battery is then charged to the same
capacity, expressed in [%] of the full battery capacity, at the same battery
temperature and internal impedance. In addition, the time for which the battery is in
CV mode is minimized, which is advantageous for the battery's cycle life. The time
for which the charge regime is in CV mode may be relatively long when tmax is used
to end the charging, especially when an almost full battery is being charged. The
difference between using I min or tmax is explained in more detail below.
The current that flows into the battery in CV mode is determined by the
difference between the externally applied voltage Vmax and the battery's equilibrium
potential E'qbat· This difference equals the total overpotential 7Jba/otai, including the
ohmic voltage drop. As charging proceeds in CV mode, E'qbar rises in dependence on
the battery's SoC and because Vmax is fixed, the current will drop accordingly. This
happens irrespective of whether Imin or tmox is used to end charging.
The total overpotential depends on the battery's SoC and impedance and on
current and temperature; see chapter 6. When the battery's temperature and internal
impedance remain constant, the total overpotential depends on current I and SoC
only. When charging is always stopped at the same current lmm, this will always
occur at the same SoC under the condition of constant temperature and internal
impedance. This SoC can be found from the equation E'qba,(SoC)=Vmax-7Jba,to1al(lm;,.,
SoC), where SoC is the only unknown quantity. This equation expresses that the
186 Chapter 5
battery voltage at the terminals during charging, Vmax• equals the sum of the battery
equilibrium potential E'qbat and the total overpotential 1'/ba,rotal; see (Eq. 3.20) and
chapter 6. However, when charging is stopped at tmax charging will not necessarily
always stop at the same current value. For example, when a half-full battery is
charged with a charge time tmax the current at the end of charging will be different
from the situation where an empty battery has been put on charge for the same
charge time of tmax· Therefore, the SoC at the end of charging will not be the same,
because the equation E'qbalSoC)= Vmax-1'Jba/01ai (I, SoC) will yield different SoC
values for different values of I. Note that a difference in battery temperature or
impedance between charge cycles will always lead to differences in SoC at the end
of charging, irrespective of the use of either Imin or tmax·
A Li-ion battery may only be charged within a certain battery voltage and
temperature range. Charging is inhibited when either the battery voltage or
temperature is outside this range. A charger will check for fault conditions, such as
short-circuits or open-circuits, before charging a battery. An electronic safety switch
is always connected in series with a Li-ion battery (see chapter 2), which will
interrupt the current flow when the battery voltage or temperature drifts beyond the
safety ranges. These safety voltage and temperature ranges are of course wider than
the ranges within which charging is allowed. Hence, the safety switch will still be
able to interrupt the charge current when the charger safety fails. The Positive-
Temperature Coefficient resistor (PTC) also acts as an extra safety device (see
chapter 2) interrupting the charge current when the safety IC fails.
5.2.2 The influence of charge voltage on the charging process

A higher value of Vmax leads to a higher capacity, as illustrated in chapter 2, but at
the same time the cycle life decreases drastically. The reason for the higher capacity
is the wider range of the mol fraction of Lt ions (x~.r') over which the positive
electrode can be cycled. It was described in chapter 4 that Xupos cycles between 0.5
and 0.95 in the positive LiCo02 electrode under normal conditions, with a mol
fraction of 0.5 corresponding to a fully charged battery. When the value of Vmax is
increased, xdos is allowed to become lower than 0.5, which means that more Li+
ions are removed from the positive electrode. This leads to a larger capacity.
A possible reason for the loss of cycle life is the decomposition of the
electrolyte at a relatively high voltage. A high voltage is present at the positive
electrode when the battery is charged in CV mode. The decomposition reactions also
involve u+ ions, which irreversibly react with compounds of the electrolyte to form
decomposition products. The longer the battery remains at the relatively high CV
voltage, the more electrolyte will be decomposed, using up valuable active material,
thereby lowering the capacity. So besides high values of Vmax• the application of Vmax
for long periods of time should also be avoided. This was already mentioned in the
previous section.
Another possible reason for the loss of cycle life is the fact that Li metal
deposition is allowed to occur at the electrode surface when the potential of the
negative LiC6 electrode becomes lower than 0 V. The Li metal atoms formed at the
electrode surface can react with electrolyte compounds and form a passivation layer
on the remaining metallic Li. This layer insulates the remaining lithium underneath,
which results in a loss of capacity. There will moreover be a risk of dendrite
formation, which could lead to an internal short-circuit.
The value of Vmax is specified with a high degree of accuracy, because battery
manufacturers want to guarantee a certain capacity and a certain cycle life. A
specified accuracy of ±1% means an accuracy of ±41 mV for a value of Vmax of 4.1
V. Figure 2.5 showed that such an accuracy already implies a fair amount of spread
in a battery's capacity and cycle life. Therefore, some battery manufacturers
demand an accuracy even smaller than ±1% in CV mode.
The value of Vmax at the battery terminals is extremely important in determining
both the charged capacity and cycle life. Additional series resistance in the current
path between the battery and the charger should be avoided, because Vmax has to be
accurately controlled at the battery terminals. This can be achieved in two ways.
The first possibility is to measure the battery terminal voltage when no current
flows and to control this voltage to Vmax. This can be referred to as 'resistance-free'
measurement. In this case, the charge current is interrupted for a short period during
which the voltage is measured at the battery terminals. The current value or duration
of the next current pulse can be adapted, depending on the difference between this
measurement and the desired value of Vmax. The second possibility is to use two
additional currentless wires to sense the battery voltage while the charge current is
still flowing through the power lines. Again, the sensed voltage is to be controlled to
Vmax· Both methods are compared in Figure 5.10. A resistance Rseries models the total
series resistance that is always present outside the battery. This resistance includes
the resistance of the wires, the wire contacts, the PTC and the safety switch. It is
divided equally between both wires in the figure. This figure is based on a very
simple model of the battery that shows only the ohmic internal series resistance
Rohmic• the charge-transfer overpotential rf'bat• which includes all kinetic and
diffusion overpotentials, and the total equilibrium potential E'qbat· It is assumed that
rf'bat hardly changes when the current is interrupted for a short time because of the
relatively long RC time.
As can be seen in Figure 5.10a, interruption of the current not only eliminates
the influence of Rserieso but also the influence of Rohmic inside the battery. This
explains the term 'resistance-free' measurement. The sum E'qbat+rf'bat is controlled
to Vmax· The use of voltage sense wires only eliminates the influence of Rserieso as can
be seen in Figure 5.10b. The reason is that current flows continuously throughRohmic
and the voltage IRohmic cannot be eliminated. A voltage of E'qbat+rf'bat+IRohmic is
sensed and controlled to a value of Vmax· Hence, a later transition from CC to CV
mode occurs in the case of Figure 5.10a than when voltage sense wires are used as
in Figure 5.10b. This later transition would be beneficial for reducing a battery's
charge time, because the current in CC mode is higher than that in CV mode.
However, how 'resistance-free' measurement influences a battery's cycle life is not
known.
5.2.3 The intl.uence of charge current on the charging process

The use of a higher value of Icc would instinctively lead to a lower charge time.
However, the CV mode will be entered earlier during the charging process, because
the total battery overpotential and the ohmic voltage drop inside the battery are
higher at a higher charge current. The charge time does not depend on the charge
current in CV mode, but on the course of the battery equilibrium potential. This
effect was illustrated by the results of a simulation and a measurement example in
Figure 4.51. The total charge time hardly decreases when Icc is made to be higher
than a 1 C-rate.
An increase in Icc does not lead to a decrease in charge time, but it has a
negative influence on the cycle life. This is because a battery has a relatively high
voltage Vmax for a relatively long time. As described above, this enables the
188 Chapter 5
generation of decomposition products. The increase in both Icc and Vmax and an
additional compensation for the ohmic battery resistance Rohmic• as shown in Figure
5.10a, decrease the charge time. However, this will have an even more dramatic
effect on a battery's cycle life.
V =Vmax + I(Rseries + Rohmic) Battery
Charger
(a)
V = V max + I Rseries
Battery
Charger
(b)
Figure 5.10: Two methods for controlling Vmax in CV mode: (a) 'resistance-free'
measurement by means of current interruption and (b) using additional currentless voltage
sense lines. The battery is represented by a simple model
5.2.4 Simulation example: fast charging of a Li-ion battery

The results of simulations using two different charging regimes will be described in
this section [8]. The usual charging parameters Icc (0.7 C-rate or 0.5 A) and Vmax
(4.1 V) were used in charging regime A, as recommended by Panasonic. Increased
values of Icc (2.5 C-rate or 1.8 A) and Vmax (4.25V) were used in charging regime B.
The parameters of the two charging regimes are summarized in Table 5.8. The
simulated charge time needed to reach the nominal capacity of 720 mAh has been
included in the fourth column. The build-up of capacity during the simulations is
illustrated in Figure 5.11.
Table 5.8: Results of simulations using two different CC/CV charging regimes, the Li-ion
model described in chapter 4 and the parameter set of Table 4.5
Charging regime Icc [A] Vmax [V] tcJuu.ae to nominal capacity of 720 mAh
rmin]
A 0.5 4.1 95
B 1.8 4.25 30
Table 5.8 and Figure 5.11 show that the charge time to the nominal capacity is only
30 minutes in the case of charging regime B, whereas it is about 95 minutes in the
case of charging regime A. Hence, increasing Icc and Vmax indeed leads to a decrease
in charge time. Moreover, Figure 5.11 shows that charging a battery to a higher
voltage leads to a significant increase in capacity. The reason for this was explained
in section 5.2.2. The rest of this section will focus on the possible causes of capacity
loss observed when charging regime B is applied.
t 1.0
8
~ 0.8
~
~
0.6
~
(.)
0.4
0.2
120
t[min]-
Figure 5.11: Simulated build-up of capacity ([Ah]) versus time (t [min]) for charging regimes
A and B; see Table 5.8
There are two possible mechanisms for the reduction in cycle life, as explained in
section 5.2.2. In the first place, electrolyte decomposition may occur at the LiCo0 2
electrode due to the high electrode potential. In the second place, Li metal
deposition may take place at the LiC6 electrode, which is induced when the electrode
potential drops below 0 V. The breakdown mechanisms themselves have not yet
been implemented in the Li-ion model. However, the electrode potentials can be
easily visualized in the simulations. That way some initial insight into the
probability of either of the breakdown mechanisms occurring can be obtained. The
development of the LiCo~ electrode potential is illustrated in Figure 5.12 for
charging regimes A and B.
Figure 5.12 shows that the positive electrode potential never exceeds the value
of 4.1 V in the case of charging regime A. The maximum electrode potential is
about 4.15 V in the case of regime B. The results of electrolyte decomposition
experiments using a LiMn20 4 positive electrode can be found in the literature [9].
The electrolyte concerned consisted of a mixture of ethyl carbonate (EC), dimethyl
carbonate (DMC) and lithium hexafluorophosphate (LiP_F6), which is slightly
different from the electrolyte used in the CGR17500 Li-ion battery. In these
experiments, decomposition of the considered electrolyte at a positive LiMn20 4
electrode started at a voltage of 5 V. Although the positive electrode material in the
CGR17500 battery is LiCo02 instead of LiMn20 4 and the electrolyte composition is
slightly different, it seems reasonable to assume that electrolyte decomposition
cannot be the main cause of the capacity reduction observed at the simulated
electrode potential values. However, it should be noted that experiments will have to
be performed to investigate electrolyte decomposition at a LiCo02 electrode to
confirm this.
190 Chapter 5
4.2
t
>
::J 4.1
~
>
4.0
3.9 .___ _ __,__ _ _ _L..-_ _ __,__ _ ____.

0 30 60 90 120
t[min]-
Figure 5.12: Simulated LiCo02 electrode potential versus aLi metal reference electrode ([V])
as a function of time (t [min]) for charging regimes A and B; see Table 5.8
The negative electrode potential simulated for the two charging regimes is shown in
Figure 5.13. The negative electrode potential does not drop below 0 V during the 95
minutes needed to charge the battery to 720 mAh in the case of regime A. However,
the electrode potential drops below 0 V in the 30 minutes needed to charge the
battery to 720 mAh in the case of charging regime B. The electrode potential drops
to -120 mV when charging is continued with regime B.
A possible cause of capacity loss in the case of regime B could be Li metal
deposition according to the simulation result shown in Figure 5.13, because the
negative electrode potential drops below 0 V. Care should be taken in interpreting of
absolute values in these simulations, because the quantitative agreement between the
results of the simulations and measurements has yet to be investigated under various
conditions. For example, the simulation shows that 30 minutes of charging with
regime B results in a negative electrode potential of -12 mV. It still has to be
investigated by means of further measurements whether this value is in agreement
with measurements, and whether this negative potential is large enough to allow a
significant amount of Li deposition.
Simulations to determine modified negative electrode capacity

Simulations using the Li-ion battery model can be used to experiment with different
electrode capacities. Such simulations can easily be used to explore, for example,
the impact of the capacity of the negative LiC 6 electrode on the charging behaviour,
because the simulations only take a limited amount of time and involve no safety
risks. Figure 5.14 shows the simulated negative electrode potential versus aLi metal
reference electrode during charging with regime B at different negative electrode
capacities. The negative electrode capacities used in the simulations range from 792
mAh to 1275 mAh. The nominal value is 850 mAh or 3060 C; see also Table 4.5,
parameter (!"axLic6 •
0.3
t
c.
0.2
::J
g? 0.1
>
0
-0.1
-0.2
0 30 60 90 120
t[minJ-
Figure 5.13: Simulated LiC6 electrode potential versus aLi metal reference electrode ([V]) as
a function of time (t [min]) for charging regimes A and B. The negative electrode capacity
was 850mAh
Figure 5.14 shows that an increase in the negative electrode capacity leads to a shift
in the moment at which the negative electrode potential drops below 0 V. For
example, an increase in the negative electrode capacity from the nominal value of
850 mAh to 935 mAh, leads to an increase in time from 27 minutes to 34 minutes.
This means that a 10% increase in the nominal negative electrode capacity would
already prevent the risk of the negative electrode potential dropping below 0 V
during fast charging, because the fast charging time is 30 minutes. This would
prevent the assumed possible Li metal deposition during fast charging with regime
B. It is assumed that a change in the negative electrode capacity hardly influences
the battery voltage and that the charge time consequently remains 30 minutes.
Dedicated experiments should be performed to check this.
5.3 Conclusions
1n this chapter charging algorithms for NiCd, NiMH and Li-ion batteries have been
described. Charging algorithms are an important aspect of the implementation of a
Battery Management System, as batteries with different chemistries have to be
charged using different dedicated charging algorithms.
It has been shown that simulations using the battery models developed in this
book are useful tools in designing such charging algorithms. Although not all
processes have been included in the models, such as side-reactions that lead to
irreversible capacity loss, simulations serve as quick means for finding potentially
better charging algorithms or possible reasons for poor cycle life performance.
Additional experiments will have to be performed to verify any significant,
interesting phenomena found in the simulations. The results of the simulations help
focus the attention in experiments. This leads to a decrease in measurement time. In
the case of the NiCd example, measurements with the thermostatic charging
algorithm confirmed the phenomena observed in the simulations. In the case of the
192 Chapter 5
Li-ion example, measurements of, for example, the electrode potentials and
morphology are still required. Moreover, the processes of electrolyte decomposition
at the positive electrode and Li metal deposition at the negative electrode should be
studied in greater detail. The results of such studies can then possibly be
implemented in the model when a mathematical description of these processes
becomes available. This will enable simulation of degradation processes in the
future and could lead to redesign of the battery itself.
0.3
~
t 0.2
:.J
~ 0.1
>
0
-0.1
-0.2
0 20 40 60
t[min]-
Figure 5.14: Simulated LiC6 electrode potential versus aLi metal reference electrode ([V]) as
a function of time for charging regime B and different negative electrode capacities (/"""uc6
5.4 References
[2] G. Cummings, D. Brotto, J. Goodhart, "Charge Batteries Safely in 15 Minutes by
Detecting Voltage Inflection Points", EDN, pp. 89-94, September 1, 1994
[3] F. Goodenough, "Battery-Based Systems Demand Unique ICs", Electronic Design, pp.
47-61, July 8, 1993
pp. 118-122, June 10, 1996
Rechargeable NiCd Cells and its Application to the Design of Battery Management
[6] K.K. Sum, M.O. Thurston, Switch Mode Power Conversion; Basic Theory and Design,
Electrical Engineering and Electronics Series, vol. 22, Marcel Dekker, New York, 1984
October 1998
[9] J.M. Tarascon, D. Guyomard, "New Electrolyte Compositions Stable over the 0 to 5
Voltage Range and Compatible with the Li 1+xMn20Jcarbon Li-ion Cells", Solid State
lonics, vol. 69, no. 3&4, pp. 293-305, August 1994
193
Chapter 6
Battery State-of-Charge indication
One of the tasks of a Battery Management System is to keep track of a battery's

State-of-Charge (SoC), as discussed in chapter I of this book. Information on the
SoC can be used to control charging and discharging and it can be signalled to the
user. An accurate SoC determination method and an understandable and reliable
SoC display to the user are important features for a portable product. However,
many examples of poor accuracy and reliability can be found in practice. This can
be pretty annoying, especially when a portable device suddenly stops functioning
whereas sufficient battery capacity is indicated. A poor reliability of the SoC
indication system may induce the use of only part of the available battery capacity.
For example, the user may be inclined to recharge the battery every day, even when
enough battery capacity is indicated on the portable device. This will lead to more
frequent recharging than strictly necessary, which in turn leads to an earlier wear-
out of the battery. The effect of inaccuracy of SoC indication can be even worse
when the SoC value is also used to control charging. The battery is either not fully
charged or it is overcharged. In the former case, the battery will be recharged more
often than needed, which will lead to an earlier wear-out. In the latter case, frequent
overcharging will lead to a lower cycle life. Some methods with which an SoC
indication system can be realized will be described in this chapter.
The definition of SoC and related terms will be given first in section 6.I.
Existing SoC indication methods will be categorized and some practical examples
will be given. Measurement results obtained with a commercially available SoC
indication IC from Benchmarq (bq2050) will be described in section 6.2 as an
illustration of a system encountered in practice. The main shortcomings of this
system will be identified. A simple mathematical model for a Li-ion battery will be
derived in sections 6.3 and 6.4 on the basis of simulations using the battery model of
chapter 4. On the basis of the results presented in sections 6.3 and 6.4, a set-up for a
new SoC system will be proposed in section 6.5. Some measurement results obtained
with the proposed set-up will be described in section 6.6 and the results will be
compared with those obtained with the bq2050. Conclusions will be drawn in
section 6. 7.
6.1 Possible State-of-Charge indication methods
6.1.1 Definitions
The term SoC may be confusing. The main reason for this is that a distinction must
be made between the charge or energy inside the battery and the portion of this
charge or energy that will actually be available under the current discharge
conditions. A significant difference may occur between these two, for example when
a battery is being discharged at low temperatures. The charge or energy that can then
be supplied by the battery is significantly smaller than the charge or energy that is
actually present. The question now arises whether SoC expresses the charge or
energy that is present or the portion of the charge or energy that will be available
under the current conditions. In either case, the term SoC may express: (1) the actual
SoC of the battery, which is the quantity we are trying to estimate, (2) the estimated

194 Chapter6
SoC, which is a calculated variable and (3) the SoC displayed to the user. Type (2)
will usually equal type (3). For psychological reasons however, type (3) may
temporarily differ from type (2) for the sake of consistency.
Consistency is a complex issue, because batteries show complex behaviour
sometimes. This may lead to incomprehension by the user, even when the SoC is
correctly estimated and displayed. For example, when a user is in a cold
environment, the battery will appear empty earlier than at room temperature. This
will lead to incomprehension when this is displayed to a user and he/she moves back
to room temperature, because then the displayed SoC increases without recharging.
Although correct, this 'inconsistency' could be hidden from the user. The issue of
consistency will not be dealt with further in this book, so type 3 will not be further
considered.
In an attempt to avoid further confusion, the following definitions will be used
in this book:
State-of-Charge (SoC):
The charge (in [Ah]) that is present inside the battery. The SoC reflects an
estimated value in most cases in this chapter. SoC can also be expressed in [%]
of the maximum possible charge, for example on a bar graph in which 100%
reflects a full battery and 0% reflects an empty battery.
Remaining capacity (Cap,,,.):

The charge (in [Ah]) that is available to the user under the valid discharge
conditions. Again, Cap,,m reflects an estimated value in most cases. This means
that Cap,,m is equal to or smaller than SoC, depending on the conditions.
Cap,,m can also be expressed in [%] of the maximum possible charge, for
example on a bar graph in which 100% reflects a full battery and 0% reflects
an empty battery.
Remaining time of use (t,,m):

The estimated time that the battery can supply charge to a portable device
under the valid discharge conditions before it will stop functioning when the
battery voltage will drop below the End-of-Discharge voltage V&o·
Charge obtained from the battery (Q0 u1):

The charge in [Ah] obtained from a battery during discharging. The value of
Qour is zero at the start of each subsequent discharge.
The term SoC will also be used as a collective noun, such as in the title of this
chapter. This means that an SoC indication system may estimate the battery's SoC
and/or Cap,,m and/or t,,m; see the definitions given above. In the explanation of the
principles of SoC indication methods in the remainder of this section, the term SoC
will be used as a collective noun for simplicity. The distinction between SoC and
Cap,em will be made in later sections.
The remaining time of use will be most interesting for a user of a portable
device. This time trem in [s] can be inferred from Cap,.m in two ways, depending on
the type ofload:
Battery State-of-Charge indication 195
Caprem . 3600 = [Ah]. [s]

trem = J [A] [h] (6.1)
for a current-type load([) and
v•.,
f Cap rem (V )dV (6.2)
t = V EoD • 3600 = [Wh] [s]
rem p [W ]' [h]
for a power-type load (P) with Vba, expressing the battery voltage at the moment trem
is estimated. The integral of Caprem over the voltage range applicable during the
subsequent discharge expresses the energy obtained from the battery. It is
impossible to exactly anticipate future load conditions. Therefore, trem can only be
inferred from the past current or power consumption, assuming that it remains
constant until VEoo is reached. As an alternative, a worst- and best-case value of trem
can be displayed. The best case applies to the minimum expected load, while the
worst case applies to the maximum expected load. For example, a cellular phone
may indicate talk time-left and standby time-left. The current will increase for a
power-type load when the battery voltage decreases during discharge. This has to be
taken into account in the estimation in (6.2). In the remainder of this chapter, only
current-type loads will be considered for simplicity. Therefore, Caprem will be
expressed in [Ah] in accordance with the definitions given above.
6.1.2 Direct measurements

The direct measurement method is based on a reproducible and pronounced relation
between a measured battery variable and the SoC. This battery variable should be
electrically measurable in a practical set-up. Examples of such battery variables are
battery voltage ( V), battery impedance (Z) and voltage relaxation time ( i) after
application of a current step. Most relations between battery variables depend on the
temperature (D; see chapters 3 and 4. Therefore, besides the voltage or impedance,
the battery temperature should also be measured. The relation J! between the
measured battery variable and the SoC, in which d means direct and T the
temperature as a parameter, can be stored in the system. The basic principle of an
SoC indication system based on direct measurement is shown in Figure 6.1.
Figure 6.1: Basic principle of an SoC indication system based on directmeasurement
The main advantage of a system based on direct measurement is that it does not
have to be continuously connected to the battery. The measurements can be
performed as soon as the battery has been connected, after which the SoC can be
directly inferred from the function//.
196 Chapter 6
The main problem is determining the functionj/, which should describe the relation
between the measured battery variable and the SoC under all applicable conditions,
including spread in battery behaviour. Conditions include the discharge current,
which may vary quite a lot depending on the application, temperatures (for example
outdoor use involves a substantially wider temperature range than indoor use),
storage times, etc. A battery's behaviour depends strongly on these conditions. In
general, the greater the amount of variation in conditions in practical use, the less
accurate a system based on direct measurement will be. The reason for this is that it
is difficult to derive j/ for all conceivable conditions. To make things worse, all
batteries will wear out during use. This implies a change in battery behaviour during
lifetime. Adaptive systems may be able to cope with battery spread and aging. This
will be described in more detail in section 6.1.4. Possible battery variables will be
discussed below.
Battery voltage (V)

Many existing SoC systems are based on voltage measurement, for example in
cellular phones. The battery voltage is then measured and translated into a number
of bars shown in a bar graph. The accuracy and reliability of these systems is
generally poor, especially when temperature dependence has not been taken into
account and in the case of batteries with a limited voltage variation for most of the
discharged capacity. For example, the voltage variation from a full to an empty
NiCd or NiMH battery is limited to approximately 200 mV. The discharge curve
depicted in Figure 6.2 illustrates why SoC systems based on voltage measurement
will usually have a poor accuracy. The voltage variation LiV around voltage V can
result from variations in current, resistance or temperature. Due to the relatively flat
discharge curve, this voltage variation will lead to a considerable variation .:lSoC
around the estimated SoC value.
t
v
SoC SoC --+

~SoC
Figure 6.2: Schematic discharge curve illustrating why SoC indication systems based on
voltage measurement will usually have a poor accuracy
In order to get more insight in the various components that constitute the battery
voltage Vbat• consider the simplified battery model shown in Figure 6.3. The EMF
(Electro-Motive Force [1],[2]) voltage denotes the sum of the equilibrium potentials
E'q of the two battery electrodes. The r/' voltage is the sum of the kinetic
overpotentials of the two electrodes, while rf expresses the sum of the diffusion
overpotentials. Finally, the rf voltage is the ohmic overpotential, which depends on
the total series resistance in the battery. The sign and value of all the overpotentials
depend on the direction and value of the battery current (I).
--
n
~I
~--+
Vbat
T-
Figure 6.3: Simplified battery model in the electrical domain
Kinetic and diffusion overpotentials will gradually increase or decrease when the
battery current is applied or interrupted, respectively. This occurs with a certain time
'constant', which is expressed by two RC combinations, RkCk and RdCd, in Figure
6.3. The non-linear resistances Rd and Rk depend on the battery's SoC and general
condition, such as storage time and age and on current and temperature. This
dependence is expressed by arrows in Figure 6.3. The value of RIJ is variable in a
portable device with a detachable battery pack, for example, because of
contaminated contacts. In other words, all the overpotentials depend on many
factors. All these contributing factors have to be taken into account in the function//
when the battery voltage is measured during current flow, which is virtually
impossible. The voltage relaxations have to be taken into account when the current
is interrupted. Possible methods for measuring only the EMF of a battery will be
described in more detail in section 6.3.
An example of changing battery characteristics in order to improve the
accuracy of an SoC system based on voltage measurement is the Infolithium system
applied in Sony camcorders [3]. In the batteries used in this system, the applied
electrode materials realize a steeper discharge curve, which is beneficial for the
accuracy; see Figure 6.2. However, the accuracy of such a system is still limited
when the portable device is used under a wide variety of conditions.
Battery impedance (Z)

The battery impedance is a frequency-dependent complex quantity, which describes
the relation between battery voltage and current. It can be measured in the frequency
domain and derived from measurements in the time domain [4]. In the former case,
the AC impedance valid in a certain bias point is measured. This is referred to as
impedance spectroscopy [1],[5]. In the latter case, the voltage difference before and
198 Chapter6
after the application of a current step can be divided by the magnitude of the current
step. However, the time at which the voltage samples are taken then strongly
influences the value inferred from the measurements. Therefore, it is not correct to
use the term impedance for this value.
A simplified polar impedance plot is shown in Figure 6.4a, where / 1<[2. The
negative imaginary axis has been indicated at the top, as is common practice in
electrochemistry [1],[5]. This yields capacitive behaviour in the first quadrant and
inductive behaviour in the fourth quadrant. An example of measuring voltage
differences L1V1 and L1V2 after the application of a current step L1l is shown in Figure
6.4b.
t
-lm (Zbatl
f~O./
Capacitive
t 45°
Re(Zbatl--
Inductive R0
f~oo
(a) (b)
Figure 6.4: (a) Measurement of battery impedance in the frequency domain: simplified polar
plot showing battery impedance as a function of frequency, (b) Measurement of voltage
differences LIV1 and LIV2 after the application of a current step & in the time domain
At low frequencies, the battery impedance will be represented by a straight line in

the polar plot under a 45° angle; see Figure 6.4a. This part of the impedance plot
corresponds to the process in which diffusion limitations are dominant. A semicircle
will appear at higher frequencies. This semicircle, with diameter Ret• is due to kinetic
aspects of the charge transfer reactions. Frequency /J equals l/(21rRetC1), in which
Ret denotes the charge transfer resistance inferred from RTI(nFF) (see chapters 3 and
4), while C0 expresses the double-layer capacitance. In practice, this semicircle
corresponds to the electrode reaction with the poorest kinetics. A second semicircle
corresponding to the electrode reaction with the fastest kinetics will then be virtually
undetectable. Ret will be in the order of 0.2 Q at 50% SoC in practical Li-ion
batteries [6]. The impedance is real and equal to the total ohmic series resistance Rn.
which may have a value up to 300 mQ in practical Li-ion batteries (see chapter 2) at
the frequency h. which is in the order of 10kHz in practical Li-ion batteries [6]. The
impedance of a practical battery will be inductive at frequencies larger than f 2 ; see
Figure 6.4a. The electrode geometry and the internal connections determine the
inductance.
Figure 6.4b illustrates that the value of L1V/L11 inferred from measurements at
the time t1 is smaller than the value of L1V;/'L11 inferred at the time t2 • The value of
L1V/L11 corresponds to R12 in Figure 6.4a. However, the value of L1V;/'L11 will also
depend on kinetic and possibly even diffusion-related aspects of the battery voltage.
It is therefore not correct to use the term impedance for these values, although this is
common practice in the electrochemical literature [4].
Measurement of battery impedance as a function of frequency as shown in
Figure 6.4a is not practical for SoC indication in a portable device, because a signal
with a frequency sweep has to be applied. Some dependence of the impedance on
the SoC can be found in a laboratory set-up, but this dependence is usually smaller
than the dependence on temperature [4]. For this reason, and the fact that this form
of measurement is not very practical, it is not applied in practice. However,
impedance spectroscopy is very useful for studying battery behaviour in detail in a
laboratory set-up when it is used to characterize separate electrodes.
The measurement of Rn at the time t 1 shown in Figure 6.4b is used mainly in
portable products as a means of indicating the battery's condition (State-of-Health or
SoH) [4],[7],[8]. Battery wear-out can be detected by an increased series resistance.
Therefore, assessing the value of Rn by simply applying a current step can serve to
test whether the battery is of poor quality and should be replaced. This is used
mainly in industrial applications like Uninterruptible Power Supplies (UPS), in
which a large number of batteries are placed in series and parallel.
Voltage relaxation ( T)
Apart from just taking two voltage samples before and after the application of a
current step as shown in Figure 6.4b, the complete battery voltage relaxation curve
can also be measured and used as a criterion for the SoC. This technique is referred
to as chronopotentiometry, which is the equivalent of impedance spectroscopy in the
time domain. However, because no reliable information on the SoC can be obtained
in impedance spectroscopy, the same remark holds for chronopotentiometry.
Therefore, this technique is not applied in practice.
6.1.3 Book-keeping systems

Book-keeping systems are based on current measurement and integration. This can
be denoted as coulomb counting, which literally means 'counting the charge flowing
into or out of the battery'. This yields an accurate system when all the charge
applied to the battery can be retrieved under any condition, at any time, which
means that Cap,.m is always the same as the SoC. However, chapters 3 and 4 have
shown that things are a lot more complicated in batteries. Therefore, other battery
variables such as voltage and temperature are also measured in practice. The basic
principle of an SoC indication system based on book-keeping is shown in Figure
6.5.
The function fb\r.~> where bk means book-keeping and V, I and T are
parameters in the function, is based on the content of the coulomb counter, which is
the integral of current I over time. Other battery variables (V and D are measured to
compensate the content of the coulomb counter for the battery behaviour. The value
of I itself is also used to compensate the counter.
The following battery behaviour can be compensated for in a book-keeping
system:
• Charging efficiency: Side-reactions occur at the end of charging in, for

example, NiCd and NiMH batteries. As a result, not all charge applied to the
battery is effectively stored. The charging efficiency depends on the SoC, I and
T. This phenomenon was illustrated by means of a simulation in section 4.5.
200 Chapter 6
Coulomb counting
I ldt
Figure 6.5: Basic principle of an SoC indication system based on book-keeping
• Discharging 'efficiency': Depending mainly on the SoC, T and/, only part of

the available charge inside a battery can be retrieved. The term 'efficiency' is
somewhat misguiding here, because it is possible to retrieve the charge under
different conditions. For example, a battery that may seem empty after it has
been discharged with a relatively high current can still be discharged further
after a rest period and/or with a lower current. In general, less charge can be
obtained from a battery at low temperatures and/or large discharge currents. A
battery's age also influences the discharging efficiency, for example due to an
increased internal resistance of the battery. The compensated SoC value equals
Caprem under the valid discharge conditions.
• Self-discharge: Any battery will gradually Jose charge, which becomes
apparent when the battery is left unused for some time. This cannot be measured
by a coulomb counter, as no net current flows through the battery terminals. The
self-discharge rate of a battery depends strongly on temperature, as well as on
the SoC. A simulated example of self-discharge of a NiCd battery was given in
section 4.5.
• Capacity loss: The maximum possible battery capacity in [Ah] decreases when
a battery ages. Capacity loss depends on many factors. In general, the more the
battery is misused, for example overcharged and overdischarged on a regular
basis, the larger the loss will be. Some types of batteries have been optimized
towards maximum cycle life, which means that their capacity decreases more
slowly as a function of the number of charge/discharge cycles [1]. Voltage
measurement is used to deal with capacity loss in most commercial book-
keeping systems. An example will be given in section 6.2, in which the capacity
parameter is updated based on the measurement of a battery discharge curve.
• Storage effects/memory effect: Some types of batteries will temporarily show
deviating behaviour after long times of storage. The behaviour will depend not
only on the storage time, but also on temperature. Some types of batteries, such
as NiCd batteries, show the memory effect as described in chapter 3. This will
lead to a temporal and reversible capacity loss, as opposed to the irreversible
capacity loss due to aging.
A practical example of a book-keeping system is shown in Figure 6.6. Apart from an

analogue measuring function, involving AID conversion and preprocessing, a
microcontroller or microprocessor is used to perform the actual book-keeping
calculations in the form of functionfh\u Two types of memory are needed. Basic
battery data, such as the amount of self-discharge as a function of T and the
discharging efficiency as a function of I and T, is read from the ROM. The RAM is
used to store the history of use, such as the number of charge/discharge cycles,
which can be used to update the maximum battery capacity.
r---.V
Analogue
preprocessing :
T • Amplification
• Filtering
• AID conversion
Charger
information _ ___.
Figure 6.6: Practical set-up of a book-keeping system
An important difference with respect to a direct measurement system is the presence

of the variable 'time' in the functionfh\r, 1. The time reference tis obtained from a
crystal oscillator (Xtal). As a result, a memory at least has to remain connected to
the battery at all times. A microcontroller inside the device can be used to calculate
the SoC and store regular SoC updates in the memory inside the battery pack with a
time stamp when a battery pack is connected to a portable device. When the pack is
detached, the microcontroller can read the latest time stamp upon renewed
attachment and estimate the time for which the battery has been unused. In extreme
cases, the entire system can be integrated within the battery pack. An example will
be given in section 6.2.
Errors will accumulate over time in a book-keeping system. Possible error
sources include measurement inaccuracy and limited accuracy of the compensations
infh\r. 1. Therefore, the system must be calibrated from time to time. The estimated
SoC is reset to an assumed value in a calibration point. For example, the charger can
signal to the SoC system when an end-of-charge trigger occurs; see chapter 5. This
serves as a 'battery full' calibration point, as indicated in Figure 6.6 by the arrow
labelled Charger information. The effectiveness of this calibration of course
depends on the accuracy of the 'battery full' trigger. Moreover, the system can be
calibrated to 'empty' when the battery voltage drops below a certain level, defined
as 'battery empty'. However, this is rather tricky, because when a battery is
discharged with a large current at a low temperature, the voltage will drop below the
'battery low' level while the battery still contains a considerable amount of charge.
The reason for this is the discharging 'efficiency' phenomenon described above.
The compensations inf h\r. 1are rather empirical in most practical systems [9],
[10]; see also section 6.2. However, the behaviour cannot be accurately extrapolated
when the battery is exposed to conditions for which the empirical equations have not
been derived. In an ideal case, battery behaviour is described in the form of an
electrochemicallphysical model such as the models described in chapter 4 [11], in
which the SoC is calculated on the basis of the external inputs V, I and T. Such a
comprehensive description could be too complicated to handle in practical systems.
202 Chapter 6
A combination of an empirical and an electrochemical/physical model will be

suggested in section 6.5.
6.1.4 Adaptive systems

The main problem in designing an accurate SoC indication system is the
unpredictability of both battery and user behaviour. Battery behaviour depends
strongly on conditions, including age, some of which may be unanticipated.
Moreover, spread in behaviour of batteries of the same type and batch makes life
more difficult. A possible solution is to add adaptivity to a system based on direct
measurement, book-keeping or a combination of the two. The basic principle of
adding adaptivity to an SoC indication system is depicted in Figure 6. 7.
Estimated battery
Update of information behaviour
(Specific battery behaviour)
e Adaptive
Basic set of information f-----'~Hf+----r Control
(Standard battery behaviour) Unit
Battery model description
Observed battery
Charge/d ischarge behaviour
conditions: I, Tamb•V yb
Figure 6.7: Basic principle of an adaptive SoC indication system
The basis in Figure 6.7 is the battery model description. The measured battery
variables hat• T bat and Vbat are the inputs of this model, which estimates battery
behaviour in the form of output vector f.n
on the basis of these inputs. Vector f.n
contains at least the SoC, but could also contain additional battery variables, such as
an estimated value of the battery series resistance. Another possibility would be to
estimate the battery voltage on the basis of the hat and Tbat measurements, and to
compare this estimated value with the measured value V bat· The model may contain
the function f! of section 6.1.2 or the function fb\r.1 of section 6.1.3, or a
combination of the two. The system starts with a basic set of information, which
describes standard battery behaviour for the type of battery concerned.
Adaptivity of the model is based on a comparison of f., with observed battery
behaviour in the form of vector Y;; . This comparison is made whenever possible. It
results in an error signal £, which is input to an Adaptive Control Unit. The unit
updates the information in the model by updating parameter values or even by
changing the model description. As a result, the model is adapted on the basis of
behaviour specific to the battery to which the system is connected and the error
between estimation and observation is minimized.
A first simple example of adaptivity was already given in section 6.1.3. The
parameter maximum battery capacity is updated from time to time in a system,
taking into account capacity loss on the basis of observed differences between
estimated and observed SoC values. A second example of an adaptive system can be
found in [12]. The model for a lead-acid battery is based on electrochemical and
physical theory. Adaptivity is based on the comparison of estimated and measured
values of the battery charge/discharge current and battery voltage. The errors
between estimations and measurements are used to update the parameters in the
battery model by means of a Kalman filter. A similar approach can be found in [13].
A third example can be found in [14], where the relation between the charge
removed during a sampling period and the discharge current is updated using
estimations of the battery's EMF, which has a direct relation with the SoC. More
detailed information on this system will be given in section 6.5.3. A fourth example
of adding adaptivity to a system on the basis of direct measurement can be found in
[15]. Here, the discharge time remaining until a predefined end-of-discharge voltage
has been reached is estimated using a predefined function. This function describes
the course of the discharge voltage of a standard battery for various discharge
currents. The parameters in this function are updated on the basis of measurement of
the battery's voltage. More information will be given in section 6.5.3. A fifth
example of adding adaptivity to an SoC system is found in [16]. Here, a neural
network is used to model battery behaviour in the form of a discharge curve.
In conclusion, it is fair to say that some kind of adaptivity is needed in an SoC
system to take changes in battery behaviour over time into account. Many examples
in which adaptivity has been successfully applied can be found in the literature.
However, it is premature to assume that by simply adding a learning ability, the
battery model will eventually take all relevant battery behaviour into account to
enable accurate SoC estimation. The extent to which this is possible depends
strongly on the battery itself, and even more on the variety of conditions under
which the battery is used.
6.1.5 Some remarks on accuracy and reliability

The accuracy of an SoC indication system is limited by systematic and random
errors. Systematic errors result from an incorrect or incomplete inclusion of battery
behaviour in a system. For example, a systematic error in the estimation will occur
when no dependence on discharge current of the remaining capacity Cap rem has been
taken into account in a book-keeping system. Systematic errors may lead to a
situation in which a portable device stops operating while the estimated remaining
capacity Caprem is still larger than 0%. Therefore, most SoC systems deliberately
give a pessimistic estimation. This is achieved by defining a safety margin or reserve
capacity to take into account all uncertainties in the estimation. The reserve capacity
is still available for use when the indication shows 0% of remaining capacity. This
leads to an even larger value of the systematic error. In general, the less accurate the
battery behaviour is described in the system, the greater the systematic error will be.
The magnitude of the systematic error depends on the condition of use. Some
systems make only a small systematic error under standard conditions. However,
under non-standard conditions, such as discharges at low temperatures, the
systematic error increases.
Random errors lead to a probability distribution around the estimated SoC and
Cap rem values. Random errors are caused by spread in behaviour of batteries of the
same type and by measurement inaccuracy. Random errors are taken as the basis for
204 Chapter 6
system accuracy definitions in most commercial presentations of SoC indication

systems, mainly in the form of measurement accuracy. This can be achieved by
defining the system's error for a condition under which the systematic error is small
or even negligible. However, the systematic error becomes significantly larger than
the random error especially under non-standard conditions. This leads to a much
lower accuracy.
A large estimation error under conditions that are frequently encountered in
practice leads to a low perceived reliability. Consequently, the user decides to
recharge the battery more often than needed, which leads to more charge/discharge
cycles in the same time. This may lead to earlier wear-out of the battery. Therefore it
is important to define precisely under which conditions the battery is likely to be
used. The estimation error under these conditions at least should be acceptable.
6.2 Experimental tests using the bq2050
6.2.1 Operation of the bq2050

The bq2050 IC from Benchmarq is a book-keeping system for Li-ion batteries. It has
been designed for integration in a battery pack. The battery voltage, current and
temperature are measured. The system is based on coulomb counting, as explained
in section 6.1.3. The counter is compensated for charge and discharge current as
well as temperature. Self-discharge compensation corrected for temperature is
applied. Finally, a means for updating the maximum available battery capacity is
included to deal with capacity loss. The estimated SoC and remaining capacity
Cap,em can be displayed on a Light-Emitting Diode (LED) display or communicated
to a host processor in the system through a one-wire serial interface.
The system uses several registers for storing estimated and measured battery
variables, such as Cap,.m and temperature, and registers for storing an identification
code and programming pin settings. The possible settings include the maximum
capacity, the initial SoC when the system is first connected to the battery, and the
type of Li-ion battery, which is programmed by selecting the type of negative
electrode present in the Li-ion battery: graphite or coke. Moreover, the self-
discharge correction can be enabled or disabled and two different values can be
selected for VEov· The first value, VEoD,Jo is used as an early warning, while the
second value, VEov, 2• serves as a 'battery empty' reference.
Four registers are important in the calculation of remaining capacity in the
bq2050:
• NAC: Nominal Available Charge

• CAC: Compensated Available Charge
• LMD: Last Measured Discharge
• DCR: Discharge Count Register
The operation of the bq2050 can be understood from the diagram in Figure 6.8 [10].
All four registers count in mAh units. The NAC and CAC registers are the actual
results of the capacity estimations. Basically, NAC is a coulomb counter, which is
only compensated for self-discharge. The CAC register reflects the capacity that is
accessible under the current discharge conditions or the charge that is stored under
the current charge conditions. It is obtained from the NAC register by multiplying its
contents by a factor smaller than or equal to one. Therefore, the value stored in the
CAC register is always equal to or smaller than the contents of the NAC register.
The multiplication factors change and CAC acquires a different value as soon as the
conditions (I and/or T) change. In terms of the definitions given in section 6.1.1, the
NAC register contains the estimated SoC value, while the CAC register contains the
estimated Cap,.m value.
The maximum battery capacity is stored in the LMD register. The NAC
register, and consequently the CAC register, never count higher than the value of the
LMD register. The content of the LMD register equals the content of the PFC
(Programmed Full Count) register when the system is first connected to the battery.
This PFC register can be programmed by the user. The DCR register plays an
important role in updating the LMD register to take capacity loss into account. The
update mechanism will be described in more detail later in this section.
The NAC register is increased proportional to the integral of charge current
over time when the battery is charged. The NAC register is also decreased with the
self-discharge rate, because self-discharge is a continuous process. This rate is
dependent on the content of the NAC register divided by a temperature-dependent
factor J.J.T), which is smaller at higher temperatures, resulting in a higher self-
discharge rate at higher SoC and temperature. This was described in chapter 4 and is
in agreement with practical observations [1]. For temperatures lower than 10°C,
multiplication of the content of the NAC register by 0.9 yields the content of the
CAC register. This is presumably done to take a lower charging efficiency at lower
temperatures into account.
Charge: Discharge:
Icharge Idischarge
T ~ 1o•C: 108 (T) = 1 I < 0.5 C: Ide, 1= 1

T T < 1o•C: 108 (T) = 0.9 I 2: 0.5 C: Ide, 1 < 1 '---.,-'--'
•
l
• NACmax = CACmax = LMD Only for 'valid'

discharge from
• NAC = =
CAC 0 if EDV flags NAC =LMDto
set at the start of charging EDV1 flag set
Figure 6.8: Diagram of calculation of remaining capacity in bq2050 (modified from [10])
The NAC and CAC registers are reset to zero before charging starts when both
'battery empty' (EDV) flags have been set during the previous discharge cycle. This
is an example of calibration, as discussed earlier. The battery empty flags are two
bits that are set when the battery voltage drops below VEoD,I and V Eov, 2• respectively.
This does not necessarily mean that the NAC and CAC registers are zero at that
206 Chapter 6
moment, because the remaining charge inside the battery could still be considerable
when the battery voltage drops below VEoD• as explained above. The EDV flags may
not be set at discharge rates larger than 2 C to prevent the risk of the error due to
resetting the NAC and CAC registers to zero becoming too large. In agreement with
the general definition in chapter 3, the C-rate is defined on the basis of the battery
capacity C, which equals the content of the LMD register in the bq2050.
When the battery is discharged, the content of the NAC register decreases
proportional to the integral of the discharge current over time. The content of the
CAC register may be smaller than that of the NAC register during current flow to
take the discharging efficiency into account. This is achieved by multiplying NAC
with the factors !de. I and fde(T). The higher the discharge current and the lower the
temperature, the smaller these factors will be. For discharge rates smaller than 0.5 C
and temperatures above 10°C, the contents of the NAC and CAC registers are equal,
because the correction factors are unity in that case.
The content of the DCR register is increased during discharging. The content
of the DCR register is reset to zero when the NAC register reaches its maximum
value during charging, which is NAC=LMD. The DCR register is increased on the
basis of coulomb counting and self-discharge, as long as NAC>O during
discharging. The situation may occur that NAC is decreased to zero while the
battery voltage is still larger than VEoD.J· The DCR register is then only increased by
coulomb counting and self-discharge is ignored during the subsequent discharge
time, because NAC=O leads to a zero self-discharge rate; see Figure 6.8.
The content of the DCR register is used only to update the LMD register after a
'valid' discharge. This is a discharge that starts at NAC=LMD, which means that
discharging starts from a 'full' battery, according to the system The discharge
process may moreover not be interrupted with recharges of more than 256 NAC
counts, and self-discharging during the whole discharge process may not exceed
more than 18% of the PFC value. This means that leaving the battery on the shelf
and discharging it completely by means of self-discharge will not lead to a battery
capacity update. Finally, a temperature below 0°C when the V&v. 2 level is reached
inhibits a 'valid' discharge. Summarizing the demands for 'valid' discharge, one
could say that a 'valid' discharge is a discharge from a full counter to a battery
voltage below VEoD.l with hardly any interruptions and at a reasonable temperature.
6.2.2 Set-up of the experiments

A CGR17500 Li-ion battery from Panasonic was connected to the Printed-Circuit
Board (PCB) of the evaluation kit of the bq2050. The serial port of the IC was read
by a computer, which made a log file of the status of the internal registers. The
battery was also connected to automatic cycling equipment, which also retained a
log file in which voltage, current and the integral of current over time were stored.
The measurement set-up is shown in Figure 6.9. The bq2050 was programmed with
the following settings:
• One Li-ion cell with a graphite negative electrode

• Maximum capacity: PFC=704 mAh (this value is the closest to the nominal
capacity of the CGR17500 battery of 720 mAh)
• VEoD values: VEoD,J =3.04 V (early warning), VEoD,l =2.94 V (battery empty)
• Self-discharge compensation enabled
• NAC=LMD=PFC on initiation
Computer
Cycling equipment
Log file:
SoC, t,
register
status
Figure 6.9: Measurement set-up for experiments using the bq2050 IC
Two types of experiments were performed to check the accuracy of the bq2050:
Experiment 1
The battery was successively charged and discharged in the first experiment.
Charging was performed with a current of 500 rnA (0.7 C-rate) in CC mode and a
voltage of 4.1 V in CV mode and a charge-termination current of 10 rnA, as
explained in chapter 5. Discharging was performed with DC currents with different
values until the battery voltage dropped below 2.8 V, which is below VEov. 1 and
VEoD.l· Each discharge cycle was directly followed by a charge cycle. The battery
was discharged in successive cycles with 150 rnA (0.2 C-rate), 700 rnA (::::1 C-rate),
2 A (2.7 C-rate), 700 rnA and again with 150 rnA. The experiment was performed at
25°C and at -5°C.
Experiment 2
The charging algorithm in the second experiment was identical to that used in
experiment 1. Both DC currents with different values and pulse-shaped currents
were applied during discharging. Again, discharging proceeded until the battery
voltage reached 2.8 V and each discharge cycle was followed by a charge cycle. The
battery was first discharged with 150 rnA, then with 2 A, then twice with 300 rnA.
The battery was discharged twice with a pulse-shaped current after this, with the
profile illustrated in Figure 6.10. Such pulse-shaped currents can be found in cellular
phones based on the Time-Division Multiple Access (TDMA) principle, such as
GSM (Global System for Mobile communication) phones. More information on
cellular telephony will be given in chapter 7. The value of 300 rnA is somewhat
lower than the average value of the pulse-shaped current. Experiment 2 was
performed only at 25°C.
Note:
Current and time values
apply for GSM specification
150 mA
-t
577 J.lS 7 X 577!-<S
Figure 6.10: Pulse-shaped current used in experiment 2

208 Chapter6
Initiation in both experiments

The system and battery were initialized before both experiments were started to
bring them in a well-defined state. The settings of the bq2050 were chosen such that
NAC equalled LMD and hence PFC upon connection of the battery to the system.
The battery was first discharged to 2.8 V with a current of 150 rnA. This first
discharge was a 'valid' discharge at a temperature of 25°C, because it proceeded
from NAC=LMD all the way down to the setting of both EDV flags. The new value
for LMD then equalled the capacity that was removed from the moment the battery
was connected until the moment the EDV flags were set. This is not necessarily the
maximum battery capacity. The battery was subsequently charged using the same
procedure as described above. The bq2050 updated its LMD value again after a
subsequent 'valid' discharge with 150 rnA, which corresponded to the full battery
capacity because the battery was completely charged. To be on the safe side, the
battery was charged again using the same procedure, discharged with 150 mA and
charged. The result of this initiation procedure is a full battery and an initialized
bq2050.

Definition of error in empty estimation (E)
The battery user will perceive an empty battery when the battery voltage drops
below 2.8 V. Assume that this happens at t=tempry· In an ideal case, the CAC register
will be zero at exactly t=fempty· An error E of the 'empty' estimation will occur in
practice when the CAC register is either zero before t=tempty or still has a content
larger than zero at t=tempry· In the first case, the error E will be defined as the charge
Qout that was obtained from the battery from the moment when the CAC register
became zero until t=tempty divided by the actual LMD value. The error will then be
defined as negative, which means that the estimation is pessimistic. In the latter
case, the error E will be defined as positive. The content of CAC at t=tempry will then
be divided by the LMD value. This situation will be most annoying to a user.
Accuracy considerations
The error E may either be inferred from E=(QouiLMD )-100%, with E<O, or from
E=(CAC1=temp/ LMD)·JOO%, with E>O as explained above. This means that the
accuracy .t1E with which this error E will be determined is given by the following
general definition:
llx llLMDJ
!ill =lEI· ( ~ + ILMDI
(6.3)
where x can be either Qout or CAC,=,empry depending on the sign of E as discussed

above and the error itself can be represented by E:ME to take inaccuracy into
account. The value of the LMD register is changed in units of 1 mAh, which means
that the accuracy of the register content is ±0.5 mAh. .tJLMD will be taken 0.5 mAh,
because it is a parameter that is not updated on a regular basis. However, the CAC
register is a variable and will be updated all the time when current flows. In order to
quantify LiCA C, we should realize that the bq2050 stored data once every minute in
the experiments. On the other hand, the cycling equipment stored battery voltage
data every time the voltage changed with 2 mV. The measurement error in the
battery voltage made by the cycling equipment is only ±0.15 mV. The discharge
curve will be very steep when Vbat:;2.8 V at t:;tempry· For example, a slope of 0.04
mAhlmV can be inferred from the discharge curve at 381 rnA in Figure 6.11. This
means that the moment t:;tempty is represented with high precision in the log file of
the cycling equipment. However, the uncertainty in the time when the CAC register
attains a certain value in the bq2050 log file will be 1 minute at maximum. This
means that the maximum error L1CAC in [mAh] that was made when E was derived
from comparing the log files of the bq2050 and the cycling equipment was (1/60)·/,
in which current I is expressed in [rnA].
The part of the error L1Qout that was made at the beginning of the determination
of Qout is equal to L1CAC, because Qout started on the basis of the content of the CAC
register, i.e. CAC:;O. The part of the error L1Qout that was made at t:;tempty is
negligible. This can be understood as follows. The discharge curve will be very
steep at t:;tempry and the voltage measurement accuracy is ±0.15 mV. This will lead to
an error of 0.3 mV·0.04 mAhlmV:;O.Ol mAh in the case of a variation of 2·0.15
mV:;0.3 mV in measured battery voltage. This is negligible compared to the part of
the error L1Q0 u1 that was made at the start of Qout· In summary, the error L1Qout will be
assumed the same as L1CAC in the tables below.
Results
The results of experiments 1 and 2 are summarized in Tables 6.1 and 6.2,
respectively. The experiments started at the first row and ended at the last row in the
tables. For example, a charge Qout of 14 mAh was removed from the battery from the
moment when the CAC register content had become zero until t:;fempry at a discharge
rate of 0.2 Cat 25°C, see the first row, third column in Table 6.1. The LMD register
had a value of 792 mAh at that moment. Hence, the error was negative then and its
value was found from E=(-141792)·100% = -1.8%. L1E was calculated using (6.3),
which yielded L1E=l.8·((2.5/14)+(0.51792))=0.3%, in which L1CAC=2.5 mAh was
found from (1/60)·150=2.5 mAh. This means that E=-1.8±0.3%. During the next
discharge at a rate of 1 C, the content of the CAC register was 6 mAh at t=tempry· The
error was (+61792)-100%= +0.8% in this case, because the LMD register content
remained unchanged. L1E was found using (6.3): L1E=l.6%, in which
LlCAC=(l/60)·700=11.7 mAh was used. Therefore, £:;0.8 ±1.6%.
Table 6.1: Errors (E) in estimation of empty battery by bq2050 in experiment 1

ldlscluuv I.u..:t....... E±lili at 25"C [%]1 E:ME at -S"C [%] 1
[A] [C-rate]
0.15 0.2 -1.8±0.3 -6.7±0.4
0.7 1 +0.8±1.6 +2.3±1.7
2 2.7 +16.7±4.2 +57.8±4.8
0.7 1 0±1.52 -86.2±1.7
0.15 0.2 -3.8±0.3 -15.6±0.4
210 Chapter6
Table 6.2: Errors (E) in estimation of empty battery by bq2050 in experiment 2

[A]
ldlschiU'IIe ldlscha.... [C-rate] E::ME at 25°C [% ]1
0.15 0.2 -574.0±7.7
2 2.7 +20.4±4.7
0.3 0.4 -2.6±0.7
0.3 0.4 -2.0±0.7
PD 2A/0.15A (1/8)3 0.5 -0.8±0.9
PD 2A/0.15A (118)3 0.5 -1.7±0.9
..
Note 1: E= (Q~, afterCAC=O)ILMD (-) orCAC..,""P'YILMD (+),see (6.3) fordeflmhon of L1E
Note 2 <Table 6.1. fourth row): E=O in this case, which leads to L1E=O according to (6.3). This is not
realistic. Therefore, L1E was defmed as .dCACJLMD in this case, because the uncertainty in the real
moment when the CAC register became zero is 1 minute.
~See Figure 6.10
Table 6.1 shows that the largest errors occur at large discharge currents and low
temperatures. The estimation is indeed pessimistic for a discharge rate of 0.2 C,
because of the negative error values, and below 5% at room temperature. However,
the errors can become positive at discharge rates of 1 C and larger, especially at low
temperatures. Positive errors should be avoided as much as possible, as stated above.
The errors are also mostly negative in Table 6.2 for discharge rates smaller than 1 C
with the exception of the first PD discharge, where the error can also be positive due
to accuracy limitation of the derived E value. The error is positive for a discharge
rate larger than 1 C. However, the error for the 150 rnA current in the first row is a
lot larger than in Table 6.1 for 150 rnA at 2SOC. This will be explained below.
Another interesting phenomenon can be observed in Table 6.2. The pulse-
current yields reliable estimations, because the errors are very small. A low-pass
filter with a bandwidth of only 0-8 Hz is used across the current sense resistor on the
test PCB of the bq2050. Therefore, only the average current is measured. The result
of experiment 2 shows that this is indeed allowed. This can also be understood from
the battery models in chapter 4 and the simple battery model shown in Figure 6.3.
All AC current will flow through the capacitors in Figure 6.3 when the frequency of
the first harmonic of the current is high enough, and only the average current will
flow through Rk and Rd. This is definitively the case for the current shown in Figure
6.10 for realistic time constants of 10 ms due to the charge-transfer processes and 1
minute or more due to the diffusion-controlled processes. This means that the
charge-transfer and diffusion-controlled processes are determined by the average
current. The current profile will only cause an additional voltage ripple around the
discharge curve one obtains with the equivalent DC current of 381 rnA. This voltage
ripple equals (2 A-150 mA)*RQ- This is illustrated in Figure 6.11.
The large error for the first 150 rnA discharge current in Table 6.2 can be
understood as follows. The battery was almost empty at the beginning of experiment
2. The first discharge is a 'valid' discharge and this led to an update of the LMD
register to the capacity removed during the first discharge, which was only 70 mAh
in experiment 2. The C-rate is expressed in terms of the capacity stored in the LMD
register; see also the C-rate definition in chapter 3. This means that even 150 rnA
was larger than a 2 C-rate in experiment 2 according to the bq2050 definition. This
means that the capacity was not updated further, because the EDV flags were not set
for discharge rates higher than 2 C. This was solved by resetting the IC at the
moment the battery was fully charged again after the discharging with 0.15 A.
4.0
1 3.8
~3.6
Cll
~
0
> 3.4
3.2
3.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Oout [Ah]
Figure 6.11: Measured discharge curves of the CGR17500 Li-ion battery. The top curve is
for a DC discharge current of 150 rnA. The third curve from the top is for a DC discharge
current of 381 rnA. The bottom curve is for a DC discharge curve of 2 A. The two remaining
curves labelled PD form the discharge curve including voltage ripple for the current profile
shown in Figure 6.10, which has an average value of 381 rnA
Another discrepancy in the LMD update mechanism was discovered during the
experiment at 25°C; see Table 6.1. The discharge current measured by the bq2050
during one of the discharge cycles in the initiation phase was somewhat higher than
the real current. Therefore, the LMD register was updated to 792 mAh from the
DCR register, which is higher than the actual battery capacity. As a result, the NAC
register never reached the value of the LMD register again during the charging and
none of the subsequent discharges were therefore 'valid'. Aging effects can no
longer be taken into account in this case, because the LMD register will not be
updated again.
6.2.4 Conclusions of the experiments

• At discharge rates lower than 0.5 C and at room temperature, the estimation
error is smaller than 5%, which is acceptable in an application.
• The accuracy of the estimations is low during discharging at low temperature (-
5°C) and at rates of 1 Cand larger at all temperatures.
• Measuring pulse-shaped currents with a low bandwidth of 0-8 Hz yields reliable
estimations of Caprem·
• The capacity updating mechanism does not work properly under certain
conditions.
Hence, the main shortcomings of the system are estimations at large discharge
currents and/or low temperatures and the capacity updating mechanism.
212 Chapter 6
6.3 Direct measurements for Li-ion batteries: the EMF method
6.3.1 Introduction
The main problem in measuring a battery's voltage for SoC indication is the fact that
the overpotentials depend on so many factors; see chapter 4 and section 6.1.2. An
alternative would be to estimate the EMF only. The EMF is found to be an accurate
indicator of the SoC for lead-acid batteries with little dependency on the battery's
temperature and age [14]. This makes the EMF a very attractive candidate for SoC
indication. This raises the question whether this is also the case for other types of
batteries. EMF curves for Li-ion batteries will be described in this section [17].
Indications can be found in the literature that these curves' dependence on SoC is
very reproducible [18]. This is confirmed by the measurements and the simulations
discussed in this section. Various methods for assessing a battery's EMF will be
discussed first in section 6.3.2. Note that the EMF is valid at equilibrium, which
means that no external current flows. Therefore, SoC indeed relates to the charge
inside the battery; see the definitions in section 6.1.1.
6.3.2 EMF measurement methods

Three methods are available for assessing the EMF:
• Voltage relaxation
• Linear interpolation
• Linear extrapolation
Voltage reloxation
Figure 6.3 shows that the battery voltage will eventually relax to the EMF value
after current interruption. This may take a long time, especially when a battery is
almost empty. Moreover, it is difficult to determine when the battery voltage has
completely relaxed. One can try to estimate the EMF value on the basis of the
voltage relaxation curve to speed up the estimation time without waiting until the
battery voltage has fully relaxed [14]. Voltage relaxation measurements are
performed under currentless conditions, as opposed to the two other methods
described below.
As discussed in chapter 4, hysteresis is observable in the EMF curve. This
means that the EMF value for a certain SoC depends on whether that SoC was
reached by means of charging or by means of discharging. The effect of hysteresis is
taken into account when the EMF curve is obtained through voltage relaxation after
either charging or discharging steps.
Linearinte~okulon
The average battery voltage is determined from the battery voltages during charging
and discharging with the same currents for linear interpolation. The basic concept of
linear interpolation is illustrated in Figure 6.12, which shows that the battery voltage
Vc is higher than the EMF during charging, whereas it is lower (Vd) during
discharging. The average voltage obtained from voltages Vc and Vd is the EMF when
ld=-lc and the battery impedance is symmetrical. The battery temperature must
moreover be the same when Vc and Vd are measured. Linear interpolation is applied
in Figure 6.12 for two different SoC values, yielding two different EMF values. In
order to determine the complete EMF-SoC curve, one can start with an empty
battery, charge the battery completely with current lc and then discharge the battery
completely again with current ld=-lc· The EMF curve is the average of the charging
and discharging curves. The current values should not be taken to be too high,
because at high lc values the battery voltage will be charged in CV mode earlier, in
which the battery voltage remains fixed and the current decreases, as described in
chapter 5. Hence, the prerequisite for linear interpolation is not fulfilled. A different
impedance value has been assumed in states 1 and 2 in Figure 6.12, as can be
inferred from the different slopes of the interpolated lines. The battery impedance
may be linearized and will then be symmetrical by default when sufficiently small
values are used for lc and h
..···········..··• Vc,2
v,_, r""/
v,' j--
. . . . . . . r vc.1
EMF1:SoC 1
- !discharge lcharge-
Figure 6.12: Basic concept of linear interpolation to assess the EMF voltage of a battery
It should be possible to apply a charge and a discharge current to access the EMF
through linear interpolation. This becomes rather impractical when a battery is not
attached to a charger, as will usually be the case in practical use. Moreover,
hysteresis is ignored with linear interpolation, because a single EMF value is found
for each SoC value. The occurrence of hysteresis will result in errors when the EMF
curve, stored in the system's ROM, was determined by means oflinear interpolation
by the manufacturer and the EMF is determined using a different method during
operation of the portable device, for example voltage relaxation. This should be
taken into account in the system design.
linearexuapokulon
The battery voltages obtained with different currents with the same sign and at the
same SoC are linearly extrapolated to a current of zero value in this case. The basic
concept of linear extrapolation is illustrated in Figure 6.13. The linear extrapolation
method is illustrated for discharge currents ld.a and ld.b• where a larger discharge
current leads to a lower voltage. Again, two different EMF values are determined at
two different SoC values and the battery temperature must be the same during the
determination of Va and Vb. The use of more measurement data for extrapolation
increases the accuracy. However, the larger the discharge currents, the larger the
deviation from a straight line will be due to the non-linear overpotential-versus-
214 Chapter6
current characteristic. A polynomial fit through all the measured voltages should in
such cases be used.
1=0
- !discharge lcharge-
Figure 6.13: Basic concept of linear extrapolation to assess the EMF voltage of a battery
A manufacturer can use extrapolation to assess the complete EMF curve by

successively charging and completely discharging a battery with various currents.
The x-axes of the various discharge curves will be aligned when the battery is
charged to the same SoC every time. This can be achieved by ending the CV mode
with a minimum current, as discussed in chapter 5. Assessing a complete EMF curve
through extrapolation takes a considerable amount of time, because the battery has
to be fully charged and discharged several times.
Linear extrapolation can also be used to assess the EMF in a portable product
in practical use, as opposed to linear interpolation, because linear extrapolation can
be realized simply by applying two discharge currents with different values and
linearly extrapolating the accompanying battery voltages. However, it should be
kept in mind that a steady-state situation has to occur after the application of each
current. When this is not the case, this will influence the EMF value estimated in
linear extrapolation. Hysteresis is taken into account with extrapolation. The EMF
value valid for discharging will be found when discharge currents are used. This is
another difference with respect to linear interpolation.
6.3.3 Measured and simulated EMF curves for the CGR17500 Li-ion battery
Measured EMF curves obtained with the aid of voltage relaxation and linear
interpolation methods will be discussed first in this section [19]. The measurements
were performed at three different temperatures, of 0°C, 25°C and 45°C, using three
identical CGR17500 Li-ion batteries simultaneously, with the results obtained for
the three batteries being averaged for each condition. The batteries were tested in
over 600 charge/discharge cycles to investigate the influence of battery age on the
EMF curves. Extrapolation was not considered in the measurements, because of the
great amount of time involved in such cycle tests. All the batteries were activated to
make sure that they were in the same condition before starting the experiments. The
activation cycles included charging the batteries in CC/CV mode, at a 0.7 C charge
rate in CC mode and a 4.1 V voltage in CV mode, with a minimum current of 35
rnA. The batteries were discharged at a rate of 0.5 C.
The voltage relaxation measurements were performed by charging and discharging

the batteries in small 15 mAh increments at a rate of 0.1 C. Each charge or discharge
step was followed by a rest period of 30 minutes, after which the voltage was
sampled. This voltage was assumed to be equal to the EMF. Experiments with rest
periods of 720 minutes showed that the difference in voltage after 30 and 720
minutes of resting was only in the order of 2 mV, while the complete voltage
relaxation was in the order of 25 mV. Therefore, the rest period of 30 minutes was
assumed adequate to allow the battery voltage to relax almost completely. The
batteries were charged and discharged in 48 steps. Charging continued at a rate of
0.1 C after the last charging step until the voltage of 4.1 V was reached. This
resulted in a defined full state. Discharging continued at a rate of 0.1 C after the last
discharging step until VEoD (3 V) was reached, which led to a defined empty state.
The batteries were charged at a rate of 0.1 C until 4.1 V in the interpolation
measurements. After a rest period of 30 minutes, the batteries were discharged at a
rate of 0.1 C to 3 V. In order to average the results of the charging and discharging
cycles, each n'h discharge curve was averaged with each (n+1)'' charge curve. A
fully discharged battery was taken as a capacity reference to obtain an unambiguous
capacity-axis definition. This means that the point at which the battery voltage
dropped below 3 V was defined as 0 mAh. For example, the interpolated EMF value
at 10 mAh was found by averaging the discharge voltage measured 10 mAh before
the battery voltage dropped below 3 V with the charge voltage measured 10 mAh
after the subsequent charging process started.
The maximum capacity was defined as the capacity obtained from the battery
during the discharging to 3 V at a rate of 0.1 C in both the voltage relaxation and the
interpolation measurements. The battery had to be charged at a rate of 0.1 C to 4.1 V
before being discharged. This maximum capacity value was used to plot the EMF
curves on a relative axis. A voltage measurement resolution of I mV has to be taken
into account in the case of all voltage measurements obtained with the cycling
equipment employed here. A comparison of the EMF curves obtained in voltage
relaxation and interpolation is shown in Figure 6.14.
Figure 6.14 shows that the two curves obtained in voltage relaxation after
charge and discharge steps are not identical. This is due to the hysteresis effect
discussed above. The greatest hysteresis occurs at a capacity of roughly 30% and
amounts to approximately 40 mY. The voltage difference is significantly larger than
the measurement accuracy, which was the same as discussed in section 6.2.3, and
cannot be explained by a too short relaxation time, as a relaxation time of 720
minutes only marginally moved the lines closer together. The interpolated EMF and
the mathematical average of the voltage relaxation curves are practically identical.
This inspires greater confidence in the consistency of the two measurement
techniques. Only a small deviation occurs when the battery is almost empty, which
is partly due to longer relaxation times during discharging. In this case 30 minutes
rest time is not enough.
Figure 6.15 shows the EMF curves obtained in voltage relaxation after
discharge steps at different temperatures. It illustrates that the temperature
dependence of the EMF is zero at 50% SoC. At SoC values higher than 50%, the
EMF decreases at lower temperatures. At values lower than 50%, the EMF increases
at lower temperatures. The maximum error in the SoC in the temperature range from
0°C to 45°C occurs around 30% SoC and amounts to 8% when only the EMF curve
at 25°C is taken into account in an SoC indication system. The dependence of the
EMF on the temperature will not be taken into account in the remainder of this
chapter.
216 Chapter6
4.2.---------------------------------------------·
" Average OCV I

- Interpolated EMF
a ocv discharge
• OCV charge
3.0~--~--~----~--~----~--~----~---L----L---~
100 80 60 40 20 0
SoC[%]
Figure 6.14: Comparison of the EMF curves obtained in voltage relaxation and interpolation
measurements. The EMF values obtained in the voltage relaxation measurements are
represented as squares, where OCV means Open-Circuit Voltage; measurement points were
obtained after charging (+) and after discharging (0). The curve (x) represents the
mathematical average of the (+) and (0) curves. The solid line represents the EMF curve
obtained in interpolation. The x-axis shows SoC [%],normalized to maximum capacity (see
text) and all measurements were performed at 25"C
4.1
I
4.0
3.9
>
';' 3.8
..........
f3.7
>
3.6
3.5
3.4 • · • o•c
--2s•c
3.3
-4s•c
3"2 100 80 60 40 20 0
SoC[%] ______.
Figure 6.15: Measured EMF curves obtained in voltage relaxation after discharge steps at
different temperatures: o•c, 25"C and 45"C. The x-axis shows SoC [%], normalized to
maximum capacity (see text)
Figure 6.16 shows the EMF curves obtained in interpolation at different

temperatures. It is in agreement with the phenomena in Figure 6.15. This confirms
the consistency between the two measurement methods. However, the SoC at which
the temperature coefficient is zero differs. It occurs at 50% in Figure 6.15, but at
60% in Figure 6.16. It was found that it occurs at 70% in the EMF curve derived
from voltage relaxation after charge steps; this has not been included in the figures.
Hence, ignoring the hysteresis effect due to interpolation averages out the SoC value
at which a zero temperature coefficient occurs. The EMF curves will differ at
different temperatures when plotted on an absolute x-axis, because the absolute
capacity obtained from a battery is lower at lower temperatures. On an absolute
scale, a capacity of 718 mAh was found for o•c, 739 mAh was found for 25"C and
755 mAh for 45"C.
4.2~-------------------------------------------------,
4.0
13.8
?.:
~ 3.6
~
g
3.4
- - - interpolated EMF; o•c
-interpolated EMF; 25•c
3.2 -interpolated EMF; 45°C
3.0~--~--~----~--~--~----~--~--~----~--~
100 80 60 40 20 0
SoC[%] - +
Figure 6.16: Measured EMF curves obtained in interpolation at different temperatures: o•c,
25"C and 45"C. The x-axis shows SoC [%], normalized to maximum capacity (see text)
Figure 6.17 shows EMF curves interpolated for different charge/discharge cycles.
The EMF curves are practically the same for all cycle numbers when plotted on the
normalized capacity axis. The maximum difference is only 10 mV. The EMF curves
plotted with an absolute capacity axis will differ, because the absolute battery
capacity decreases when the battery becomes older. The absolute capacity values
that were found for cycles 14, 289 and 615 were 733 mAh, 707 mAh and 660 mAh,
respectively.
An EMF curve can also be simulated with the aid of the Li-ion battery model
described in chapter 4. The EMF value can be simply obtained in simulations by
adding the true equilibrium potentials of the two electrodes. The model is simply
charged or discharged and the EMF is plotted versus SoC [%].Figure 6.18 shows
the simulation results for three different temperatures, of o•c, 25"C and 45"C. The
capacity obtained from the battery until the battery voltage dropped below 3 V was
taken as the capacity reference when the model was discharged.
218 Chapter 6
4.2 . . . . . - - - - - - - - - - - - - - - - - - - - - - - - - .
& 3.6 f-.

s
0
> 3.4 ,__ 4 interpolated EMF (17/18}
~interpolated EMF (181/182}
"interpolated EMF (607/608}
3.2 f-.
• interpolated EMF (616/617}
3.0L--~~----~~--~~---~~--~~----~1--~I----~1--~I--~
100 80 60 40 20 0
SoC[%] _____..
Figure 6.17: Measured interpolated EMF curves for different charge/discharge cycles at
25°C. Cycle 17/18 (a), cycle 181/182 (-¢>),cycle 607/608 (0), cycle 616/617 <•).
The x-axis
shows SoC[%], normalized to maximum capacity (see text)
4.2
4.0
i
~3.8
G
u.
:2
w
3.6
- - - simulated EMF; ooc

3.4 -simulated EMF; 25°C
-simulated EMF; 45°C
3.2 L--------~---------1-------~------l.----...1
100 80 60 40 20 0
SoC [%] ----+-
Figure 6.18: Simulated EMF curves for three different temperatures: 0°C, 25°C and 45°C
The simulated EMF curves shown in Figure 6.18 correspond very well with the
measured interpolated curves in Figure 6.16. The general trend of a negative
temperature coefficient at a low SoC and a positive temperature coefficient at a high
SoC agrees with both Figures 6.15 and 6.16. The curves obtained in the interpolation
of measurements and the simulated curves do not take into account the effect of
hysteresis. Hysteresis was taken into account in the measured curves obtained with
voltage relaxation in Figure 6.15. Therefore, the agreement between the results of
the simulations and the interpolation measurements seems logical. The simulated
EMF curve is inferred from E"'ucoo2- E'quc6 , see (Eq 4.78) and (4.87). The only
temperature dependence that is taken into account in the present Li-ion battery
model is in the terms RTinuF· Apparently, this yields the correct temperature
behaviour.
The influence of aging can be studied by simulating of the EMF curve with a
720 mAh and a 600 mAh maximum capacity. After normalization it was found that
the two curves were exactly identical. (Eq 4. 78) and (Eq 4.87) show that this is not
surprising. Only the relative capacity in the form of xu occurs in both equations, and
not the absolute battery capacity.
6.3.4 Conclusions
The measurements discussed in this section reveal that the EMF curve versus
normalized capacity in [%] shows little dependence on temperature and number of
cycles in the case of the investigated CGR17500 Li-ion battery. The results of the
simulations with the Li-ion battery model developed in chapter 4 show the same
behaviour. These results will be used in section 6.5 to define an alternative SoC
indication system.
6.4 A simple mathematical model for overpotential description

In practice, current is drawn from a battery during discharging and overpotentials
(TJ) occur; see Figure 6.3. A battery may appear empty to a user even if the estimated
EMF value still estimates an SoC larger than 0%, because the battery voltage drops
below VEoo· This is illustrated in Figure 6.19, where SoC has been plotted on the
horizontal axis to explain this effect, which can be translated to a time axis using the
value of the discharge current. Discharging starts at A% SoC and the battery voltage
drops with an overpotential 1]. As a result, the battery voltage drops below VEoo at
B% SoC. This means that the remaining capacity Cap,.m is zero at B% SoC; see the
definitions in section 6.1.1. It should be possible to estimate this under the
discharging conditions concerned by including a mathematical description of 1] in
the system.
Simulations with the battery models presented in chapter 4 yield a first
impression of the course of 1] under various conditions. This is a lot less time-
consuming and a lot simpler than conducting measurements, because overpotentials
at various discharge rates and various SoC values can be easily plotted after
simulations that only take a few seconds to run. Figure 6.20 shows the
overpotentials simulated with the Li-ion model using the parameter values given in
chapter 4 for various discharge currents when the battery is discharged from a full
state (100% SoC), half full state (50% SoC), 25% SoC and 10% SoC. For easy
comparison of the overpotential curves at various discharge currents in one figure,
Qout has been plotted on the horizontal axis. Again, this axis can be translated to a
time axis using the value of the discharge current. This has to be considered when
relaxation processes are concerned. A Positive-Temperature Coefficient resistor
(PTC) was placed in series with the battery, see chapter 4, but no safety IC
resistance was considered for simplicity.
All the overpotential curves in Figure 6.20 represent similar behaviour. The
overpotential quickly reaches a certain level in the initial stages of discharging,
220 Chapter 6
which level depends on the discharge current. This is attributable to the ohmic
voltage drop and a relatively fast relaxation of the kinetic overpotentials. Then, a
slow relaxation process occurs, after which the battery voltage again reaches a more
or less stable level. Simulation results in Figure 4.53 illustrated that this relates to
the build-up of a transport overpotential of Li+ ions in the electrolyte. Finally, the
overpotential increases again when the battery is almost empty. This is attributable
to the increase in the kinetic and diffusion overpotentials due to diffusion limitation
of Li+ ions in the electrodes when the battery approaches the empty state. The
increase in diffusion overpotential is dominant according to simulations; see Figure
4.53.
t EMF
~
......... .... ________________ A% Bo/o
v
~
-------r-.. . ~l',,
TJ ......
:
[
'\
[ \
! I
[ \
VEoDL-------------------------------~~--~-
SoC[%]-
Figure 6.19: Schematic representation of EMF (dashed) and discharge (solid) curves leading
to zero remaining capacity (Caprenr=O) at B% SoC
For practical use in an SoC indication system, the behaviour shown in Figure 6.20
should be described in an equation that is as simple as possible and involves a
minimum number of parameters. The parameter values should be easily obtainable
in a limited number of measurements. In principle, such an equation can be inferred
from the overpotential description used in the more complicated battery models
described in chapter 4; see for example (4.92). In this chapter, a more or less
empirical equation will be used for obtaining the first results of overpotential
estimation in a proposed SoC indication set-up, which will be described in the next
sections. This equation is given below [20]:
1J(Qin ,T,I,t) =1Jruc (T,l,t) +1Jd (T,I,t) + 'YJq(Qin, T,l,t) =
l { R.,.(T)+R,(T{ -e'';~, )+ U,~~~)(T)(i-e'•;', JJ (6.4)
where 17 is the total overpotential (in [V]), which is composed of three contributions.
Overpotential T/ac results from the combined effect of the ohmic and kinetic
overpotentials, overpotential T/d is due to diffusion of Li+ ions in the electrolyte and
T/q describes the increase in overpotential when the battery becomes empty. I is the
applied discharge current (in [A]), Rac(T) is the temperature-dependent 'ohmic and
kinetic' resistance (in [Q]), RJ.T) is the temperature-dependent 'diffusion' resistance

(in [Q]) and t:J.T) is the temperature-dependent 'diffusion' time constant (in [s]). Rae
equals R.a+Rk in Figure 6.3, under the assumption that RkCk is very small, and Rd and
Td =RdCd can also be seen in Figure 6.3. The dimensionless function Uq(l) is an
inverse step function, with Uq( I)= 1 for discharging (1:50) and Uq{1)=0 for charging (I
>0). Variable Eq{T) is the temperature-dependent energy, normalized to the current I,
which describes the increase in overpotential when the battery becomes empty (in
[J/A]). Finally, Tq(T) is a temperature-dependent time constant associated with the
increase in overpotential in an almost empty battery. Parameter Eq(T) is a measure of
the energy that cannot be obtained from the battery when I increases. Hence, this
parameter expresses the difference between Cap,.,. and SoC. (6.4) must be
expressed in differential form to describe the overpotential development under
changing current conditions. The variable Q;n(t) is the charge present inside the
battery at timet, which means that Q;n expresses SoC in [C]. (6.4) illustrates that the
overpotential will instantaneously attain a value IRac(T), after which the additional
overpotential IRJ.T) builds up with a time constant t:J.T). The final term in (6.4)
leads to an increase in overpotential due to a decrease in Q;n when the battery
approaches the empty state. This occurs with time constant Tq(T). The contribution
of TJq to the total overpotential decreases drastically when an empty battery is
charged, because the diffusion limitation of Lt ions in the electrodes vanishes.
Therefore, Uq(l) is made zero when a charge current is applied (I >0).
r-Q.1
~
I="-Q.2
-o.3
-Q.4'-----'----'----'----'-----'-----'---..LL----'
0 200 400 600 800
Oout [mAh] ---+
Figure 6.20: Simulated overpotentials TJ [V] versus charge obtained from the battery (Q ••,
[mAh]) for various discharge rates (0.1 C, 0.3 C, 0.7 C and 0.9 C) for a complete discharge
(1), discharge from 50% SoC (2), discharge from 25% SoC (3) and discharge from 10% SoC
(4) at 25"C. The Li-ion model described in chapter 4 was used in the simulations
222 Chapter6
Only five parameters occur in (6.4), some of which have an electrochemical

meaning. All parameters can be found with the aid of curve fitting with battery
discharge curves. However, the value of Rg, can also be inferred from an impedance
diagram as the sum of RSJ and Rc1; see Figure 6.4. The parameter Rd could in
principle be inferred from an impedance diagram, but this is difficult in practice.
The reason for this is that the shape of the impedance plot, of which the simplified
version was shown in Figure 6.4, will depend on the nature of the diffusion process
in practice. The measurements moreover have to be conducted at low frequencies
leading to a change in the battery's condition, because the measurements take a long
time. For the same reason, it is also difficult to infer parameter Td from an impedance
plot. Parameter Eq can only be obtained by fitting with discharge curves obtained at
different currents. Parameter -zq can only be inferred from fitting (6.4) to experiments
in which the current is interrupted when 1'Jq has attained a significant value. This will
lead to the occurrence of relaxation curves, which can be used to fit -zq. All the
parameters have to be derived at different temperatures.
All parameter values in (6.4) will be obtained by fitting measured complete
battery discharge curves at various discharge currents in the remainder of this
section. The use of impedance plots for obtaining some of the parameters will not be
considered for simplicity. The measured overpotentials will be determined from the
difference between discharge curves obtained at different discharge currents as an
alternative to conducting tedious overpotential measurements by means of voltage
relaxation. Figure 6.21 shows four experimental overpotential curves (solid). A
CGR17500 Li-ion battery was discharged at various discharge rates including 0.1 C,
0.2 C, 0.4 C, 0.8 C and 1 C, at 25°C. Four differential curves were constructed, on
the basis of 0.1 C as the smallest discharge rate, with the differential rates being 0.1
C (=0.2 C-0.1 C), 0.3 C, 0.7 C and 0.9 C.
The dashed curves in Figure 6.21 represent the calculated overpotentials that
resulted from curve fitting of (6.4) to the measured differential overpotential curves.
First, each of the four differential curves was fitted individually. The average value
for the four fits was then calculated for each parameter. These average parameter
values are revealed in the figure caption. The dashed curves shown in Figure 6.21
were calculated with these average parameter values. This illustrates that the fits are
quite good, even with average parameter values. It shows that the overpotentials at
various discharge currents can be fitted well in the case of a practical Li-ion battery
with a single set of parameters. Parameter Tq was chosen to be zero for simplicity,
because its effect is negligible in the case of complete discharge curves. This can be
understood from (6.4), where the exponential term in 1'Jq has become zero already
long before the term inversely proportional to Q;n starts to increase. This parameter
becomes important when the overpotential behaviour of an almost empty battery has
to be estimated and the current is often changed or interrupted in this region.
The measured differential curves shown in Figure 6.21 agree quite well with
the simulated overpotential curves shown in Figure 6.20, which were obtained with
the battery model described in chapter 4, in terms of order of magnitude and general
shape. However, the relaxation process described by parameter Td in (6.4) was
slower in the simulations than in the measurements. Care should be taken with the
representation of the x-axis in [mAh], as explained above. The quantitative
agreement between the results of the simulations and the measurements still needs
improvement. It should be noted that the measurements shown in Figure 6.21 were
not obtained with a high degree of accuracy. Therefore, new dedicated
measurements were performed to derive the parameter values in (6.4). These
measurements were performed at different temperatures, because the overpotentials

depend strongly on temperature. A CGR17500 battery was discharged using two
discharge currents at various temperatures. The discharge rates were 0.1 C and 0.5
C, which led to a single differential rate of 0.4 C. The resulting differential discharge
curve was fitted at four temperatures of -l0°C, 0°C, 25°C and 45°C. The result is
shown in Figure 6.22. The parameter values obtained at different temperatures are
listed in Table 6.3. Figure 6.22 shows very good fits of (6.4) with the measurements
obtained at all the temperatures, especially at the moment when the overpotential
begins to increase when the battery is empty. This is important, because (6.4) will be
used to estimate the battery's remaining time of use.
0.00
-------
I
-0.10
~ -0.20
>
<I
-0.30
-0.40
200 400
Oout [mAh] _______.
Figure 6.21: Experimental differential battery discharge curves in the case of a complete
discharge obtained for differential rates 0.1 C, 0.3 C, 0.7 C and 0.9 C as a function of the
charge obtained from the battery Q••, [mAh] at 25°C (solid curves). The dashed curves show
the overpotentials calculated with the aid of (6.4) using parameter values valid at 25°C: Ra =
0.11 Q, Rd= 0.17 Q, 1'd= 74 sec, Eq= 28.8 J/A and 1'q=0
Table 6.3 shows that ROk decreases as the temperature increases. The kinetic part of
ROk will be larger at lower temperatures, because all electrochemical reactions
proceed more slowly at lower temperatures. The ohmic part of ROk decreases as the
temperature decreases. Hence, the largest part of ROk will be attributable to the
kinetics of the reactions at -10°C. The reverse situation takes place at 45°C, where
the ohmic part is dominant. The ohmic part of ROk will be around 100 :mQ in
practical batteries at 25°C, excluding the safety measures. The fits show that the
increase in ohmic resistance as the temperature increases is not as high as the
increase of the kinetic part as the temperature decreases, because ROk is only 0.11 Q
at 45°C.
224 Chapter6
The diffusion coefficient increases with the temperature according to the Arrhenius
relation; see (4.72). This leads to a lower diffusion overpotential, and hence to a
lower value of Rd, at higher temperatures. Time constant Td also decreases with an
increasing temperature, because it is derived from RdCd; see Figure 6.3. This can
indeed found in Table 6.3. More energy can be obtained from the battery at higher
temperatures because of the better kinetics and higher diffusion rates at higher
temperatures. Therefore, Eq decreases as the temperature increases, as can indeed be
seen in Table 6.3. Again, Tq was chosen to be zero at all the temperatures. Table 6.3
shows that the parameters in Figures 6.21 and 6.22 are in the same order of
magnitude at 25°C. The differences are partly due to the inaccuracy of the
measurement results shown in Figure 6.21. Moreover, spread in battery behaviour
has to be accounted for in parameter sets for different batteries of the same type.
Many more measurements should be performed to obtain a better understanding of
parameter spread between batteries. In addition, tests should be performed to
investigate the influence of aging on the parameter values. This will not be
considered here. The parameter values used in Figure 6.22 will be used in the
remainder of this chapter.
0.00
-0.10
~
i -0.20
~
-0.30
-0.40
-0.50
0 200 400 600 800
Oout [mAh] ____.
Figure 6.22: Experimental differential battery discharge curves obtained for a differential rate
of0.4 Cas a function of the charge obtained from the battery Q0 ., [Ah] at -10°C, 0°C, 25°C
and 45°C. The dashed curves show the overpotentials calculated with the aid of (6.4) using
the parameter values listed in Table 6.3
Table 6.3: Values for the parameters in (6.4) for different temperatures; see Figure 6.22. The
values obtained at 25"C in Figure 6.21 have been included for reference
Parameter: -to•c o•c 2s•c 45"C 25"C (Figure 6.21)
Rru: [Q] 0.45 0.29 0.15 0.11 0.11
R.J [Q] 0.57 0.37 0.16 0.09 0.17
'td [s] 165 169 132 42 74
Eq [J/A] 468 216 7.2 7.2 28.8
'to [s] 0 0 0 0 0
6.5 Proposed set-up for State-of-Charge system
6.5.1 The algorithm

The results of sections 6.3 and 6.4 will be used in this section to derive a new set-up
for an SoC indication system which is intended to overcome the shortcomings of the
investigated bq2050 system The proposed system is a combination of a direct-
measurement system and a book-keeping system [20]. The EMF method described
in section 6.3 is used for the direct measurement. The EMF curve, which is stored in
the system, must be determined for the type of battery concerned using any of the
methods described in section 6.3.2. The overpotential description of (6.4) is used for
the book-keeping. The proposed algorithm operates in five different states:
• Initial state
• Equilibrium state
• Charge state
• Discharge state
• Transitional state
The system's estimations can be shown to the user in each state in the form of a
value of SoC expressed in [%] and a remaining time of use trem available under the
valid discharge conditions; see the definitions in section 6.1.1. The state diagram of
the algorithm is shown in Figure 6.23.
The algorithm starts up in the initial state, when the battery is first connected to
the SoC system The algorithm then shifts to the appropriate state, depending on
whether the battery is charged, discharged or in equilibrium The initial state is re-
entered only when the battery has been disconnected from and reconnected to the
system The SoC in the initial state is determined by measuring the battery voltage
and translating this into an SoC value with the aid of the EMF curve. This is only an
estimation, because there is no guarantee that the battery voltage is close to the EMF
at the moment when the battery is connected to the system. However, at least an
estimation of the initial SoC can be made, which is not possible with the Benchmarq
system described in section 6.2.
A current smaller than or equal to lum is drawn from or flows into the battery
and the battery voltage has relaxed from previous currents in the equilibrium state.
The parameter lum is defined in the system Its value should not be taken too large,
because then the measured voltage will differ too much from the EMF. On the other
hand, the equilibrium state will hardly be entered when the value of lum is taken too
small. For example, the value of fum can be chosen a little higher than the current
drawn by a cellular phone in the standby mode. The battery voltage will be close or
equal to the EMF when the phone is in the standby mode and the battery voltage has
226 Chapter 6
relaxed from the current previously drawn in talk mode. For example, a current of 1
rnA will only yield a deviation from the EMF of 300 j..tV at the battery terminals in
the case of a realistic series resistance of 300 mQ. The kinetic and diffusion
overpotentials will be negligible under this condition. This means that in the
equilibrium state, the SoC can be determined by means of the EMF method by
simply measuring the battery voltage and translating it into an SoC value with the
aid of the EMF curve. The validity of the '1IIg1im and battery voltage stable'
condition is checked before the equilibrium state can be entered. This is illustrated
in Figure 6.23.
II I :S llim and voltage not stable
II I :S llim and voltage stable

Initial state
Figure 6.23: State diagram of the proposed SoC indication system
The algorithm resides in the charge state when the battery is charged with a positive
current larger than lum· It is in the discharge state when the battery is discharged
with a negative current larger than 11im in absolute terms. The SoC is determined by
coulomb counting in both states. In addition to the SoC, lrem is also estimated in the
discharge state. This will be described later in this section. A starting value of SoC
for coulomb counting is always known when either the charge or discharge state is
entered. This starting value is obtained with the aid of the EMF method when the
state is entered from the initial or equilibrium state. Alternatively, the charge state
can be entered from the discharge state and vice versa. In this case a starting value
is also available.
The transitional state is entered during a change from either the charge or
discharge state to the equilibrium state. The current is smaller than or equal to 11im in
this state. It is determined whether the equilibrium state may be entered or not, as in
the initial state. This means that the voltage has to be stable or relaxed to allow a
transition to the equilibrium state. The system shifts back into either the charge or
the discharge state when this is not the case and an absolute current larger than llim
starts to flow again. Coulomb counting continues in the transitional state as long as
the battery voltage has not fully relaxed.
Figure 6.24 shows the transitions between states in a practical situation. The
figure was obtained in simulations using the battery model of Figure 6.3, with the
complete overpotential being described by (6.4), and a measured EMF curve was
entered in the model.
CD
Cl
~
g 3.96
3.88 t..__.;.:;__ ..:.;;;......&-:;;.__ _...J..o~~=~o,;.:...;:::.....;,.;.;.~~~.....:.........:..;.._:...-..J

0 800 1600 2400 3200 4000
t [s] ----+
Figure 6.24: Simulated EMF (top curve [V]), discharge voltage (bottom curve [V]) and
transitions between states as a function of time [s] in the proposed SoC indication system. The
battery model shown in Figure 6.3 was used in the simulations, with the overpotential
description according to (6.4) and an EMF curve derived from measurements
The algorithm is in the equilibrium state at t=O, because a current smaller than or
equal to lum then flows out of the battery and the battery voltage is stable. The latter
condition can be checked by calculating the derivative of the battery voltage over
time and comparing it with the value of the parameter dV/dtlim· The SoC is
determined with the aid of the EMF method. A current larger than lum is drawn from
the battery at t=lOOO sand the battery voltage then drops. The algorithm shifts from
the equilibrium to the discharge state. In addition to the SoC, the system also
estimates t,,m on the basis of the SoC value in this state. The discharge current
changes to a value smaller than or equal to !lim at t=1600 sand the transitional state
is entered. The SoC is determined by coulomb counting in both the discharge state
and the transitional state. Coulomb counting continues in the transitional state,
because a current smaller than or equal to Tum still flows in this state. The battery
voltage has relaxed by t=2400 s and the algorithm then changes to the equilibrium
state again.
A first important aspect that must be borne in mind in the system is that the
unit of SoC is [C] in coulomb counting, because the integral in time of the battery
current is determined. The SoC is determined on a percentage scale in the EMF
228 Chapter6
method; see section 6.3. Therefore, the content of the coulomb counter in the charge
and discharge state has to be translated into a percentage scale, too. The maximum
battery capacity Capmax on an absolute scale is needed for this purpose. Some means
of updating the value of Capmax has to be taken into account to cope with capacity
loss due to aging. A simple method for achieving this is illustrated in Figure 6.25.
Transitional state
Equilibrium state: Discharge state :J..:,-E_q_u_ili_b_riu_m_s_ta_t_e
--------~
~
i~ --~!l ___________
t SoC8 [%] SoCE[%]
i; CaPmax [C] = soc.[%]-

100
SoCe [%] • Oout [C]
I I dletermined with
L---------'------~---'--__.--t- - EMF method
Figure 6.25: Simple method for updating Capmax to take capacity loss into account
A necessary condition in Figure 6.25 is for the system to run through a sequence of
states: equilibrium state, discharge state, transitional state and equilibrium state.
The new value of Capmax is simply calculated by relating the charge Qour drawn from
the battery mainly in the discharge state and a little in the transitional state to the
difference in SoC (SoC5-SoCE) before and after discharging as derived from the
EMF method. The method can be made more complex by enforcing a minimum
value of Q0 ur for the update to be valid. Moreover, only small changes in Capmax
should be allowed. Finally, the update mechanism should be allowed to occur only
under 'standard' conditions, for example a discharge rate that is not higher than 1 C
and a temperature that is within the range from 10°C to 40°C. A similar updating
mechanism can be implemented during charging.
A second important aspect to be borne in mind in the discharge state was
illustrated in Figure 6.19: Caprem may differ from the SoC during discharging. Both
Caprem and trem can be estimated under the prevailing discharge condition and from
the present SoC value. Note that the SoC value itself will be updated continuously
with the aid of coulomb counting in the discharge state and will remain available at
all times. In the present algorithm, the time at which the voltage will drop below
VEoD• or EMF-rJ=VEoD is estimated. The difference between that time and the present
time yields the remaining time of use trem· To estimate the time at which EMF-
'YJ= VEoD• the future overpotential development must be estimated under the
prevailing discharge conditions by increasing the time and decreasing the value of
Q;n in (6.4). This is achieved by expressing (6.4) in a differential form and by
calculating the anticipated changes d'f/ in time dt. At the same time, the future EMF
development must be estimated from Q;r/Capmax. The estimation of future values of
rJ and EMF can be seen as a 'fast-forward' anticipation of battery behaviour under
the prevailing discharge condition. The time at which the battery voltage will drop
below VEoD has to be re-estimated when the discharging condition changes. This
leads to an updated value of trem· The future overpotential may also be estimated for
a condition which does not currently apply. It may be a nice feature to show trem
under different conditions. For example, a cellular phone may display the remaining
standby time, the remaining talk time under the current conditions and the remaining
talk time at low temperatures [20].
6.5.2 Comparison with the bq2050 system

The proposed system can be compared with the bq2050 system; see section 6.2.
• The SoC system proposed in this section can be regarded as a book-keeping

system, which is calibrated each time the battery relaxes to equilibrium. This is
achieved by resetting the SoC value calculated withthe aid of the book-keeping
system to the value obtained with the EMF method. More calibration points will
be encountered during use than is the case with the bq2050 system, which is
only calibrated when the battery voltage drops below the EDV threshold during
a 'valid' discharge. The periods for which coulomb counting is applied will be
shorter in the case of the proposed system and hence the errors will accumulate
over shorter time periods.
• The two systems' methods for expressing the difference between the SoC and
Caprem during discharging are comparable. A correction factor for the coulomb
counter is used in the bq2050 during discharging to calculate the CAC from the
NAC register depending on the discharging conditions. Both NAC and CAC
remain available to the user. The coulomb counter (NAC) reflects the SoC,
while a corrected counter (CAC) shows Caprem for the current discharging
conditions. In the proposed system, trem is estimated from the SoC, with both
variables remaining available to the user. The way in which Cap,.m or trem is
derived from the SoC value changes with the discharging conditions in the case
of the two systems. However, a difference between the two systems concerns
the method itself that is used to estimate Cap,,m from the SoC. Only a limited
number of correction factors is used in the bq2050 system, which depend on the
temperature and current. A fast-forward estimation of battery behaviour is
performed for each condition in the proposed system This yields different
behaviour for a whole range of currents and temperatures.
• The updating mechanism of Capmax presented in this section differs from the
updating mechanism used in the bq2050. A 'valid' discharge should take place
with the bq2050 system before the value of CapmtlX (called LMD in the bq2050)
can be updated. Capmax can be updated more frequently in the proposed
mechanism, for example every time the condition illustrated in Figure 6.25 is
satisfied.
• The system proposed in this section estimates the initial SoC by itself in the
initial state. The bq2050 is a pure book-keeping system, so the initial SoC has to
be programmed by the user. Problems may occur, depending on the initial SoC
and also on the subsequent discharging and charging conditions. This was
shown in section 6.2, in which a wrong capacity update occurred on the basis of
the initial condition and the subsequent discharging.
• Self-discharge is taken into account in different ways in the two systems. The
coulomb counter counts down at a speed proportional to its contents and
temperature in the bq2050 system This is not necessary in the system proposed
230 Chapter 6
here. The self-discharge will eventually lead to a decrease in EMF in

equilibrium. As a result, the SoC will be automatically updated.
6.5.3 Comparison with systems found in the literature
System 1
The concept of using book-keeping when current flows and the EMF method when
the battery voltage has stabilized has also been described for lead-acid batteries [14].
The system was applied in field tests using electric wheelchairs and yielded
satisfying results. A pronounced relationship between the EMF and the SoC in [%]
exists for lead-acid batteries, which is even linear. Two interesting differences with
respect to the system proposed in this section can be found in the system proposed in
[14].
First of all, a different correction is applied to the coulomb counter during
discharging in the wheelchair system. A linear relation is assumed between the
discharge current I and the battery capacity removed per sampling period J.t. C, in
which Cis a correction factor. The difference between Cap,.m estimated with the aid
of book-keeping and the SoC estimated using the EMF method is determined when
the battery is in equilibrium. C is modified on the basis of this difference. This
means that the effects of discharging efficiency and capacity loss are combined.
These two effects are separated in the system proposed here. The 'correction factor'
in the form of an estimated future overpotential behaviour is updated every time the
condition changes, even when the battery does not return to equilibrium A separate
updating mechanism of Capmax leads to a better insight into the actual value of
Capmax, which can serve as battery health monitor.
A second difference concerns the transition to equilibrium state. The algorithm
in the system proposed here simply waits until the battery voltage has stabilized.
This can take a considerable time, depending on the battery's condition. An elegant
approach was adopted in the wheelchair system The EMF value is estimated from
an assumed relaxation profile of the battery voltage when the discharge current is
interrupted. This allows calibration without the necessity to wait until the battery
voltage has stabilized. It would be interesting to further investigate the possibilities
of including this in the proposed system, because an overpotential description is
available. This approach was however not adopted in the experiments described in
the next section.
System 2
An empirical function describing the entire battery discharge voltage is used for
lead-acid batteries in [15]. The remaining time of use until VEoD is reached is
calculated under the current discharge conditions on the basis of this empirical
function. No distinction is made between the EMF and the overpotential. It is
unclear how the system behaves when no or little current flows, because the
empirical function is obtained by fitting discharge curves for significant currents.
The concept of using a description of battery voltage during current flow to estimate
I rem is similar to the system described in this section. The parameters in the empirical
function in [15] are updated on the basis of a comparison of the battery voltage
estimated by the empirical function and the measured battery voltage. This
adaptivity is an interesting subject for further research for the system described here;
the parameters in (6.4) could then be updated on the basis of measurements. This
was not taken into account in the experiments described in the next section.
6.6 Experimental tests with the system proposed in section 6.5
6.6.1 Introduction
Some preliminary test results obtained with the system proposed in the previous
section will be described in this section. Again, a CGR17500 battery was used. The
main aim of this section is:
• to investigate the accuracy of the estimation of trem under different load

conditions and at different temperatures. Moreover, the behaviour of the system
when load changes occur during discharge will be investigated;
• to investigate the calibration of the book-keeping system in equilibrium state;
• to investigate the capacity updating mechanism proposed in Figure 6.25.
6.6.2 Set-up of the experiments

Two experiments have been performed. The battery was initialized with a standard
regime before each experiment and at each temperature. This regime comprised
three charge-discharge cycles, using a 500 rnA current in CC mode and a 4.1 V
voltage in CV mode, with a minimum current of 35 rnA, in the charge regime. The
battery was discharged at a 0.5 C-rate to a voltage of 2.8 V. A 30-minute rest period
was allowed between the charging and discharging. Both experiments were
performed at 25°C and at 0°C. This means that the initiation was performed four
times.
Experiment 1
Experiments similar to those carried out with the bq2050 were carried out in the first
experiment to facilitate comparison with the results presented in section 6.2. The
charge regime was the same as during initiation. A rest period of 30 minutes was
allowed between the charging and discharging. A rest period of 30 minutes after
discharges to 2.8 V at rates higher than 0.2 C was followed by another discharge at a
0.2 C-rate to 2.8 V and another 30-minute rest period. This was done to ensure an
empty battery. The battery was successively discharged at rates of 0.2 C, 1 C, 2.7 C,
and with the pulse current defined in Figure 6.10. Besides for comparison with the
results presented in section 6.2, the results of the first experiment also serve the first
main aim mentioned above.
Experiment 2
Successive charge and discharge cycles, separated by rest periods of 30 minutes,
were applied again in the second experiment. The charge cycles were the same as in
the first experiment. The same remark as above holds for discharging to 2.8 V at
rates larger than 0.2 C. Three different discharge regimes were successively applied:
1. Discharge with 11=0.2 C-rate for 60 minutes, followed by a discharge with

lz=2.7 C-rate, l 2 >l1o to 2.8 V.
2. Discharge with h for 10 minutes, followed by a discharge with h to 2.8 V.
3. Discharge at a 0.013 C-rate (10 rnA) for 15 minutes, followed by a discharge at
a 0.5 C-rate for 30 minutes, another discharge at a 0.013 C-rate for 30 minutes,
and finally a 0.5 C-rate discharge to 2.8 V.
232 Chapter 6
Discharge regimes 1 and 2 yield information on the first aim with respect to load
changes. A discharge with a small current is followed by a discharge with a high
current in discharge regime 1. The reverse occurs in discharge regime 2. Discharge
regime 3 serves the second and third aims in this section. It leads to the situation
illustrated in Figure 6.25. Therefore, Capmax can be derived as discussed above.
Moreover, the book-keeping part is calibrated at SoCE. Regime 3 was repeated
twice, with an intermittent full discharge at a rate of 0.1 C. The latter discharge was
incorporated between the two 3 regimes to derive the real value of Capmax at that
moment.
The experiments were organized in time as follows. First, the order standard
regime, experiment 1, standard regime, experiment 2, was performed at 25°C. The
algorithm continuously monitored the SoC throughout this period. Then the same
order was repeated at 0°C. Again, the algorithm continuously monitored the battery.
6.6.3 Experimental results

The parameters obtained in Figure 6.22 were used in (6.4); see Table 6.3. A value of
720 mAh was used for Capmax, which is the nominal battery capacity. To allow fair
comparison with the results obtained with the bq2050, the measured value of Capmax
was not used. The values of lum and dV!dtum in the algorithm were chosen to be 20
rnA and 15 J..tV/s, respectively.
Results of experiment 1
The results of experiment 1 are summarized in Table 6.4. The remaining time of use
under the prevailing discharge conditions was estimated by the algorithm at three
moments during each discharge. This estimated time was compared with the actual
time from that moment until the battery voltage dropped below 2.8 V for the first
time. The first moment was chosen to be the moment when the discharge current
was applied. The second moment was chosen somewhere halfway through the
discharging, and the last moment was chosen at which either the estimated or the
real remaining time of use became zero. Estimated times of less than 10 seconds
were regarded as zero.
In order to allow fair comparison with the errors calculated in the case of the
bq2050 in section 6.2, the errors E were calculated as follows:
E =( tl,estimated - ti,actU/11 J·1 00% (6.5)

t!,actUill
where i==1,2 or 3. This means that the difference between estimation and reality is
expressed relative to the full discharge time t 1,actUill with the current for which the
remaining time of use is estimated. This is indeed a fair comparison with the results
presented in section 6.2, because the difference between the CAC register
(estimation) and the real remaining capacity was expressed relative to the LMD
register in that section, with the LMD being a measure of the full capacity. The
accuracy ..1E with which E can be determined will be discussed further below.
Table 6.4: Results of experiment 1: Estimated and actual remaining times of use at three
moments during discharging at various discharge rates
(a)25"C
l.slscharlo tl.esllmaled tlrst L\t [s)
E±L\E [%) t2.est1mated 2.aetual [s) L\t [s] E±L\E[%)
rc-ratel [s] fsl
0.2 17694.1 18141.3 -447.2 -2.47±0.01 8160.5 8612.1 -451.6 -2.49±0.01
1 3564 3643.2 -79.3 -2.18±0.01 1702.7 1772.3 -69.6 -1.91±0.01
2.7 1244.2 1030.4 213.8 20.75±0.03 728.9 513.8 215.1 20.88±0.03
pot 14805 15338.2 -533.2 -3.48±0.01 7028.6 7426.3 -397.7 -2.59±0.01
ldlscharae[C-rate) t3.estlmated [sl t3.actual [S) L\t [s) E±L\E [%)

0.2 0 454.7 -454.7 -2.51±0.01
1 0 80.5 -80.5 -2.21±0.01
2.7 207.9 0 207.9 20.18±0.03
Pot 0 526.3 -526.3 -3.43±0.01
ldlscllano tl.estlmated tt.actual L\t [s) E±L\E [%) t2.estlmated 2.ac1ua1 [s) L\t [s) E±L\E [%)
rc-ratel fsl fsl fsl
0.1" 34092.1 33539 553.1 1.65±0.01 17106.4 16568 538.4 1.61±0.01
1 3060 3194.2 -134.2 -4.20±0.02 1439 1555.9 -116.9 -3.66±0.02
2.7 115.4 88.9 26.5 29.81±0.37 69.3 42.7 26.6 29.92±0.37
pot 13952.6 13877.2 75.4 0.54±0.01 6535.2 6471.3 63.9 0.46±0.01
I.u.m•..., [C-rate) t3.esumated [sJ t3.actual [s) L\t [s) E±L\E[%)

0.1 542.1 0 542.1 1.62±0.01
1 0 150.2 -150.2 -4.70±0.02
2.7 26.7 0 26.7 30.03±0.37
pot 0 169.3 -169.3 -1.22±0.01
N!lliU.; See Ftgure 6.1 0. However, due to an error m the programmmg of the cycling eqUipment, the duty
cycle of the 2A pulses was 1180 instead of 118. This led to an average current of 173 rnA
Note 2: The result at a 0.1 C-rate has been included in this table because of measurement problems
encountered in the cycle at a rate of 0.2 C at 0°C
Accuracy considerations
On the basis of (6.5), the following expression can be derived for the accuracy t1E
with which E can be determined. The error can again be expressed by E::f:LlE:
(6.6)
where t1ti,estimated and .1ti,actuat are the accuracies with which the estimated and actual
remaining times of use can be determined, respectively. We now have to estimate
these accuracies, where .1ti,actual =.1tI.actual=t1tacruat can be assumed. Let us ftrst
consider t1ti.esnmated· This time is inferred from t,mp1y-tp, with the battery being empty
at t=tempty CVbat= VEoD) and tp being the time at which the estimation is made. The time
tempty was determined internally with high precision from the stored EMF curve and
the overpotential function of (6.4). It has therefore been assumed that there is no
measurement error in the determination of this time. The time tp is inferred from a
time point in the log file of the cycling equipment. These times have an accuracy of
234 Chapter6
±0.06 s. This means that .1t;,estimated has been assumed 0.06 s in the tables in this
section.
The actual remaining time of use is also found from 1emp~y-tp. The same remarks
with respect to accuracy as discussed above hold for tp. However, now the time tempty
was inferred from measurements instead of internal calculations. As discussed
above, tempty is the moment when the battery voltage has dropped below 2.8 V. The
cycling equipment stored measured voltage values when they had changed by 2 mV.
This means that in worst case, tempty was determined when Vbat=2.8 V-2 mV.
Assuming a slope of the discharging curve of 0.04 mAhlmV as before, the error
expressed in [hr] can now be found from ((0.04 mAhlmV)·2mV)/l, in which I is
expressed in [rnA]. The total error in the actual remaining time of use .1tactual
expressed in [s] can now be found from (0.08 mAhl/)·3600·r<).Q6.
Results of experiment 2
The results of experiment 2 for discharge regimes 1 and 2 are summarized in Table
6.5. Again, the accuracy of the trem estimations was determined at three moments.
This time, the first moment was chosen to be the moment at which the discharge
current changed, which was from h to 12 in discharge regime 1 and from h to 11 in
discharge regime 2. The second moment was chosen halfway through the
discharging with the changed current (/2 in regime 1 and 11 in regime 2). The third
moment was chosen to be the moment at which either the estimated or actual time
became zero, with estimated times smaller than 10 seconds again being regarded as
zero.
Table 6.5: Results of experiment 2: Estimated and actual remaining times of use at three
moments during discharging with discharge regime l and discharge regime 2
(a) 25°C
lc~~sc~~a ..... tt.estlmated tl.actual At [s] E±AE[%] t2.esUmated t2.actual At [s] E±AE[%]
[C-rate] [s] [s] [s] [s]
!Reg. 1. 985.1 789.5 195.6 24.78±0.04 686.1 494.3 191.8 24.29±0.04
p.2~2.7
!Reg. 2: 9643.6 10106.5 -462.9 -4.58±0.02 4933.8 5425.9 -492.1 -4.87±0.02
~.7~0.2
l.tlsch....... [C-rate] t3...umaled [s] t3.ac:1ua1 [s] At [s] E±AE[%]

!Reg. 1: 0.2~2.7 195.2 0 195.2 24.72±0.04
IR_eg. 2: 2.1~0.2 0 449.6 -449.6 -4.45±0.02
Table 6.5 (continued): Results of experiment 2: Estimated and actual remaining times of use
at three moments during discharging with discharge regime I and discharge regime 2
(b) 0°C
lc~~scban~e tt.esllmaled tt.actual ~t [s] E~E[%] tl.estlmated tl.actual ~t [s] E~[%]
[C-rate] [s] [s] [s] [s]
Reg. 1: 57.1 55.8 1.3 2.33±0.48 29.2 27.6 1.6 2.87±0.48
0.2--72.7
Reg. 2: 10835.7 10551.1 284.6 2.70±0.02 5809.6 5536.1 273.5 2.59±0.02
2.0 1--70.2
Idtscha,... [C-rate] t3.eslimaled [s] t3.actual [s] ~t [s] E~E [%]

!Reg. 1:0.2--72.7 1.8 0 1.8 3.23±0.48
!Reg. 2: 289 0 289 2.74±0.02
~.0 1 --70.2
Note 1: Due to measurement problems encountered m regune 2 at 00 C, an alternallve cycle With a change
from a 2 C-rate instead of a 2.7 C-rate was used in this table
The results of discharge regime 3 of experiment 2 are summarized in Table 6.6. The
variables necessary to calculate Capmax are listed for both applications of the regime.
The value of Capmax derived from the intermediate discharging at a rate of 0.1 C has
been included for comparison. The value of SoC in transitional state, just before the
battery entered the equilibrium state (SoC,), and the first value in the equilibrium
state (SoCE) are listed for the sake of the second aim in this section. The difference
between SoC, and SoCE illustrates the effect of calibration.
Table 6.6: Results of discharge regime 3 of part 2: SoC8, SoCE, Qout and calculated Capmax
(see also Figure 6.25), Capmax derived from full discharge at a rate of 0.1 C, and SoC in the
transitional state just before the battery entered the equilibrium state (SoC,)
(a) 25°C
Application of SoCs SoCE Qout [C] CaPmax.est. Capmax. meas. SoC1
re2.3 [%] [%] [mAh] 1 [mAh]1 [%]
First 100 74.3 651.2 702.9 735.1 74.9
Second 100 74.2 651.1 701.0 735.1 74.9
Application of SoCs SoCE Qout [C] CapiiWI.est. Capmax. meas. SoC1

rel!.3 [%] [%] [mAh] 1 [InAh]1 [%]
First 97.1 68.0 654.1 625.7 670.9 71.9
Second 93.0 65.0 655.3 650.1 670.9 67.8
Note 1: Estunated from SoCs, SoCE and Q••,; see Figure 6.25
Note 2: Measured from full discharge at a rate of 0.1 C
6.6.4 Discussion of the results
Experiment 1
Table 6.4 shows that the absolute error Lit remains constant throughout the complete
discharging time. This was to be expected from the applied algorithm, because the
conditions do not change from t1o through t2 until t3 . As a result, the time when the
battery is estimated to be empty does not change, and hence the estimated times
t~estimared decrease with the actual remaining times of use t;,actual· The results shown in
Table 6.4 can be compared with the results obtained with the bq2050 presented in
Table 6.1. This has been done in Table 6.7.
236 Chapter 6
Table 6.7: Comparison of errors in empty estimation with the bq2050 and the SoC algorithm
proposed in section 6.5 under identical circumstances at 25°C and similar circumstances at
low temperatures (-5"C in the case of bq2050 and 0°C in the case of the algorithm presented
in section 6.5)
I [C· E±AE of bq2050 at E±AEofnew E±AE of bq2050 at E±~ofnew
rate] 25°C[%] algorithm at t 3 at -5°C [%] algorithm at t 3 at
25"C [%] 0°C [%]
0.2 -1.8±0.3 -2.51±0.01 -6.7±0.4 1.62±0.01
1 0.8±1.6 -2.21±0.01 2.3±1.7 -4.70±0.02
2.7 16.7±4.2 20.18±0.03 57.8±4.8 30.03±0.37
PD -1.3±0.92 -3.43±0.01 3 Not available -1.22±0.01 3
Note I: Thts measurement was perlormed at a rate of 0.1 C mstead of a rate of 0.2 C
Note 2: See Figure 6.10; this error is the average of the errors in both pulse discharges in Table 6.2
Note 3: See Figure 6.10; a duty cycle of 1180 instead of 118 was used
Inspection of Table 6.7 shows that the estimations obtained with the bq2050 yield
smaller errors than those obtained with the new algorithm at 25°C at rates of l C and
smaller. Moreover, the estimation of t,.m obtained with the new system has not
improved at the large discharge rate of 2. 7 C compared to the Cap,.m estimations
obtained with the bq2050. However, the error in the estimation obtained with the
new algorithm is still lower than 5% and also negative, i.e. pessimistic, at rates of 1
C and lower encountered in practical portable devices. This will usually be
acceptable. It should also be noted that better results will be obtained with the
proposed algorithm once the real value of Capmax of 735 mAh (see Table 6.6) has
been introduced instead of the nominal capacity of 720 mAh. The estimations at low
temperatures and large discharge currents obtained with the new algorithm are
considerably better, although the estimation error is still positive and large, hence
annoying for a user in practice. The error in the estimation is again smaller than 5%
at discharge rates of 1 C and lower.
Experiment 2
The results presented in Table 6.5 illustrate the effect of previous discharge currents
on the estimations when the current is changed during discharging. Some of the
results given in Tables 6.5 and 6.6 have been included in Table 6.8 to further clarify
this effect.
Table 6.8: Errors in estimated remaining time of use for full (Table 6.4) and partial (Table
6.5) discharges at rates of 0.2 C and 2.7 C at 25"C and o•c
Temperature E±~ of partial E±AE of full E±~ of partial E~offull
["C] discharge at 0.2 discharge at 0.2 discharge at 2.7 discharge at 2. 7
C-rate [%] 1 C-rate [%] 2 C-rate [%] 1 C-rate [%] 2
25 -4.45±0.02 -2.51 ±0.01 24.72±0.04 20.18±0.03
0 2.74±0.02 1.62±0.01 3 3.23±0.48 30.03±0.37
Note I: See Table 6.5
Note 2: See Table 6.4
Note 3: Discharge rate ofO.l C instead of0.2 C
Table 6.8 shows that the error in the estimation increases somewhat, but not
dramatically, for partial discharges using a certain current. For example, the error is
-2.51% when the battery is completely discharged at a rate of 0.2 C at 25°C. This
error increases to -4.45% for a partial discharge at a rate of 0.2 C when the battery
has first been discharged at a large rate of 2.7 C for 10 minutes. This error is hence
still pessimistic and smaller than 5%. This inspires greater confidence in the applied
approach of overpotential estimation, because apparently the overpotential
relaxation resulting from the application of the 2. 7 C discharge rate is taken into
account correctly. The estimation of t,,m for a partial discharge at a rate of 2.7 C at
0°C is much better than that at full discharge. More experiments will have to be
carried out before a definite conclusion can be drawn concerning this improved
accuracy for partial discharges with large currents at low temperatures.
The results shown in Table 6.6 show that the capacity updating mechanism
proposed in Figure 6.25 yields estimations of Capmm: that are lower than the actual
value. This can be understood from the fact that SoCE is lower than SoC, in all cases.
The battery was in a state of equilibrium when the first discharge at a rate of 0.5 C
was started. This means that SaCs was based on the EMF method. The charge Qour
represents the charge that was drawn from the battery when the battery was in the
discharge state and subsequently in the transitional state. The latter state was
entered when the discharge current at a 0.5 C-rate was interrupted after 30 minutes.
The SoC was updated by means of coulomb counting in these two states. A value of
720 mAh was used for Capmax to translate the absolute values of SoC, calculated by
coulomb counting, into a relative SoC in [%]. This value can be recalculated from
Table 6.6 by relating the difference between SoCs and SoC, to Qout• which means
that SoC1 replaces SoCE in the capacity updating equation given in Figure 6.25. The
battery voltage relaxed and the current of 10 rnA was lower than fum (20 rnA) in the
period of 30 minutes after the first application of the 0.5 C discharge rate. The
equilibrium state was re-entered during this period after the change in battery
voltage in time dropped below 15 J..LV/s. The value of SoCE was the first SoC value
that was again based on the EMF method. This value was somewhat lower than SoC,
in all cases, because the battery voltage had not fully relaxed. This effect was more
pronounced at low temperatures, because the overpotential that had to relax was
larger (R.ac and Rd were larger) and therefore also the associated time constants were
larger. Therefore, the difference between SoC, and SoCE is larger at 0°C in Table 6.6.
As a result, the difference between the estimated and measured Capmax value was
larger at 0°C; see the capacity updating equation given in Figure 6.25. However, the
fact that Capmax is lower at lower temperatures is indeed revealed by the algorithm.
An improvement in the conditions under which the equilibrium state may be re-
entered should lead to more accurate estimations of Capmax.
The effect of calibration is revealed by the difference between SoC, and SoCE
in Table 6.6. The calibration is not fully accurate, because the battery voltage has
not fully relaxed. The value of SoC increased a little further once the equilibrium
state was re-entered in the experiments, because the battery voltage relaxed further.
Due to the relatively long rest periods of 30 minutes applied in the experiments, the
effect of this 'premature' calibration on the accuracy of the estimations was limited.
The effect cannot be seen at all throughout the equilibrium state, when the battery
voltage has fully relaxed until the next charge or discharge state is entered.

The main aim of this section was to investigate the accuracy of the estimations of
remaining time of use under various conditions, including changing load conditions.
In addition, the proposed capacity updating mechanism was investigated. No
definite conclusions can be drawn, because only few experiments have been carried
out.
238 Chapter 6
The estimation errors obtained with the new algorithm are only slightly larger than
those obtained with the existing bq2050 system. The estimation for discharges with
large currents at low temperatures is better with the new system than with the
bq2050. Especially the estimation for a partial discharge at a 2. 7 C-rate at 0°C has an
acceptable accuracy. The accuracy of estimations at large currents at room
temperatures is not better than that obtained with the bq2050 system and is still too
low. This can be improved by investigating the fit of (6.4) with measurements with
large discharge currents. The parameters for (6.4) that were used in the experiments
were derived only at discharge rates smaller than 1 C. The new system works well
when load changes occur, because the differences in the accuracy of estimations for
full and partial discharges are limited. The capacity updating mechanism needs
improvement, which should aim mainly at improving the conditions under which the
equilibrium state may be re-entered. This will also have a beneficial effect on the
system's calibration. However, problems with the capacity updating mechanism of
the kind described for the bq2050 system in section 6.2 cannot occur with the new
system.
The expected advantages of the new system over the bq2050 were listed in
section 6.5.2 and included (i) improved accuracy due to more calibration points, (ii)
an improved capacity updating mechanism and (iii) a realistic estimation of the
initial SoC. Unfortunately, it is not possible to draw solid conclusions with respect
to these possible advantages on the basis of the few experiments. Experiments
should be performed with both systems under many more conditions than presented
in this book to quantify these advantages in practice. Both systems should eventually
be tested in field tests.
6.7 Conclusions
The subject of SoC indication has been discussed in this chapter. Three basic
methods for SoC indication were identified in sections 6.1 and 6.2 and their main
characteristics and problems have been discussed. The main problem involved in
direct-measurement systems is the inclusion of all possible battery and usage
conditions in the function J/, which links the measured battery variable to the SoC.
The main problem in book-keeping systems is to define reliable calibration
opportunities that occur often enough during the battery's use.
The focus in sections 6.3 and 6.4 was on Li-ion batteries. Two usable
phenomena for SoC indication were identified. First of all, the EMF method was
discussed as a possible candidate for direct measurement. Measurements and
simulations show that the EMF curve does not depend on many parameters. It is not
dependent on aging of the battery and its dependence on temperature is limited.
Secondly, a simple way of describing overpotentials has been derived from
simulations, which can be used to determine the remaining capacity under various
discharge conditions. The battery model described in chapter 4 greatly enhances
understanding of both the EMF method and the overpotential description. The
simulations show good agreement with practical conditions. Hence, the use of the
battery models described in chapter 4 in designing Battery Management Systems has
again been proven useful.
A new SoC indication system was proposed and discussed in sections 6.5 and
6.6, in which information presented in the previous sections was applied. The
system is a combination of direct measurement by means of the EMF method and
book-keeping using a simple description of overpotentials. The book-keeping part is
calibrated through direct measurement in the form of the EMF method. Hence, the
benefits of both methods are combined. The results of some preliminary tests
performed with the new system have been discussed. The accuracy of estimations
under various conditions, including load changes, is satisfactory. In comparison with
the results obtained with the existing bq2050 book-keeping system, mainly the
estimations at large discharge currents and low temperatures are better with the new
system However, the estimations obtained with the bq2050 system at moderate
currents and at room temperature are better than those obtained with the proposed
system
Many more tests will have to be carried out in the future to investigate the new
system in greater detail. For example, tests with partial charges and discharges with
an almost empty battery should be performed. The parameter 'fq in (6.4) should be
given a value for this purpose. The results of the capacity updating mechanism will
improve when a more accurate EMF estimation can be made on re-entry of the
equilibrium state. Apart from experimenting with different values of dV/dtum• one
should also consider the approach adopted in [14], where the EMF value is
estimated from the voltage relaxation curve. An improvement in this area will have a
positive effect not only on the capacity updating mechanism, but also on the
accuracy of the system's calibrations. Finally, adding adaptivity to the system to
update the parameters in (6.4) should be considered. This adaptivity should improve
the system's capability of coping with aging effects and parameter spread for
different batteries of the same type.
6.8 References
[1] D. Linden, Handbook of batteries, Second edition, McGraw-Hill, New York, 1995
[2] G.M. Barrow, Physical Chemistry, Fourth Edition, McGraw-Hill, Tokyo, Japan, 1979
[3] Sony Infolithium webpage: http://www.seJ.sony.com/SEUconsumer/ handycaml worry-
free/infolithium.html
[4] F. Huet, "A Review of Impedance Measurements for Determination of the State-of-
Charge or State-of-Health of Secondary Batteries", J. Power Sources, vol. 70, no. 1, pp.
59-69, January 1998
October 1998
[7] M. Kim, E. Hwang, "Monitoring the Battery Status for Photovoltaic Systems", J. Power
Sources, vol. 64, no. 1, pp. 193-196, January 1997
[8] W. Cantor, "Monitoring Lead-Acid Batteries in UPS Systems", Electronic Design, pp.
123-128, January 12, 1998
[9] J. Alzieu, H. Smimite, C. Glaize, "Improvement of Intelligent Battery Controller: State-
of-Charge Indicator and Associated Functions", J. Power Sources, vol. 67, no. 1&2, pp.
157-161, July/August 1997
[10] Benchmarq Microelectronics, bq2050 Lithium-Ion Power Gauge IC, Datasheet, June
1995
[11] P.H.L. Notten, H.J. Bergveld, W.S. Kruijt, "Battery Management System and Battery
Simulator", Patent US6016047, W09822830, EP0880710, filed 11 October 1997
[12] C.E. Barbier, H.L. Meyer, B. Nogarede, S. Bensaoud, "A Battery State-of-Charge
Indicator for Electric Vehicle", Proc. of the Institution of Mechanical Engineers,
Automotive Electronics, ImechE 142, pp. 29-34, May 1994
[13] P. Liirkens, W. Steffens, "Ladezustandsschiitzung von Bleibatterien mit Hilfe des
Kalman-Filters", etlArchiv, Bd. 8, H. 7, pp. 231-235, 1986
240 Chapter 6
[14] J.H. Aylor, A. Thieme, B.W. Johnson, "A Battery State-of-Charge Indicator for Electric
Wheelchairs", IEEE Trans. on industrial electronics, vol. 39, no.S, pp. 398-409,
October 1992
[15] A.M. Pesco, R.V. Biagetti, R.S. Chidamber, C.R. Venkatram, "An Adaptive Battery
Reserve Time Prediction Algorithm", Proc. of the Jl'h international
telecommunications energy conference, Session 6.1, pp. 1-7, Firenze, Italy, October 15-
18, 1989
[16] 0. Gerard, J.N. Patillon, F. d' Alche-Buc, "Neural Network Adaptive Modelling of
Battery Discharge', Lecture Notes in Computer Science, vol. 1327, pp. 1095-1100, 1997
[17] F.A.C.M. Schoofs, W.S. Kruijt, R.E.F. Einerhand, S.A.C. Hanneman, H.J. Bergveld,
"Method of and Device for Determining the Charge Condition of a Battery", Patent
W0062086, EP0002787, filed 29 March 2000
[18] J. Molenda, A. Stoklosa, T. Bak, "Modification in the Electronic Structure of Cobalt
Bronze LixCo02 and the Resulting Electrochemical Properties", Solid State Ionics, vol.
36, no. 1&2, pp. 53-58, October 1989
[19] R.E.F. Einerhand, S.A.C. Hanneman, State-of-Charge Indication Method: Li-ion
Batteries, Nat.Lab. Technical Note 190/98, Philips Internal Report, September 1998
[20] H.J. Bergveld, H. Feil, J.R.G. van Beek, "A Method of Predicting the State-of-Charge
as well as the Use Time Left of a Rechargeable Battery", Patent application
PHNUJ00675EPP, filed 30November 2000
241
Chapter 7
Optimum supply strategies for Power
Amplifiers in cellular phones
The market for cellular phones is growing rapidly and cellular phones rank among
the most prominent portable devices. All these phones are battery-powered and long
use times on a single battery charge are required. The Power Amplifier (PA) draws
most of the current from the battery in talk mode. Therefore, a reduction in this
current will lead to a significant increase in talk time. This chapter covers powering
the PA with the optimum supply voltage for each output power, which leads to a
decrease in supply current. As described in chapter 1, this is one of the basic tasks
of a Battery Management System.
Section 7.1 gives some basic information on cellular systems and describes the
trends in these systems. These trends lead to an increased desirability of reductions
in the PA supply current. The concept of efficiency control enabling such reductions
in supply current will be introduced in section 7.2. A DC/DC converter is needed for
efficiency control. Different ways of translating one DC voltage into another DC
voltage will be discussed in section 7.3. Simulation models will be developed for a
DC/DC converter and a PA in section 7.4. Simulations can be performed with these
models and the Li-ion battery model developed in chapter 4 to define the theoretical
benefits of efficiency control. These simulations will be described in section 7.5.
Section 7.6 discusses PA measurements to verify the theoretical benefits derived in
section 7.5. The battery model described in chapter 4 will be used to estimate talk
time improvements in a complete cellular phone. Results of measurements of
applying efficiency control in a complete cellular phone will be discussed in section
7. 7. Conclusions will be drawn in section 7.8.
7.1 Trends in cellular systems

Cellular phones are used in cellular systems [1],[2], which form part of the large
group of wireless systems. A cellular system is organized in the form of cells, which
complement each other and each cover a certain geographical area. One base station
is present in each cell, which can accommodate several handsets. Information to and
from these handsets is transmitted in the form of modulated signals. The base station
controls the links with the various handsets. Communication from handset to base
station is referred to as the up-link, while communication in the other direction is
referred to as the down-link.
Many different types of cellular systems exist nowadays. Each cellular system
is characterized by the employed frequency band, the applied modulation technique
and by the way in which the available channels are distributed over the handsets.
The frequency band is divided into several channels, whose number and width
depend on the cellular system. Each handset is allowed to use a certain channel.
Data is transmitted by means of signals, which are modulated around the centre
channel frequency. Most of the current cellular systems are based on digital
modulation techniques, as opposed to conventional analogue modulation techniques,
such as AM (Amplitude Modulation) and FM (Frequency Modulation). The main
reason for using digital modulation techniques is the greater channel capacity and
signal integrity in comparison with that available with analogue techniques. This

242 Chapter7
increases the quality of the established links between the handsets and the base
station. Moreover, digital modulation provides compatibility with digital data
services and higher data security [1]-[3].
Data in digital modulation techniques is represented by one- or multiple-bit
symbols. These symbols are transmitted by modulating a carrier, with, for example,
the amplitude and phase of the transmitted signal reflecting the symbol's value.
Examples of digital modulation techniques include QPSK (Quadrature Phase Shift
Keying), FSK (Frequency Shift Keying), MSK (Minimum Shift Keying) and QAM
(Quadrature Amplitude Modulation). A detailed description of digital modulation
methods is beyond the scope of this book. A comprehensive introduction to digital
modulation can be found in [3].
Channel allocation to separate the different users of the available spectrum is
organized by means of multiplexing methods in all communication systems. In
cellular systems, multiplexing is already achieved by geographical separation. Two
handsets are allowed to transmit in the same channel when they are present in
different cells, because of the inherent geographical distance between them Apart
from geographical multiplexing as applied in cellular systems multiplexing methods
based on frequency, time and code are also applied in communication systems.
These multiplexing methods are denoted as access methods. The three main access
methods are FDMA (Frequency-Division Multiple Access), TDMA (Time-Division
Multiple Access) and CDMA (Code-Division Multiple Access) [1]-[3].
Each transmitter and each receiver uses a different frequency in FDMA
systems. Therefore, each transmitter or receiver is identifiable by the frequency it
uses. In TDMA systems, data is communicated in bursts, which are separated in
time. The minimum time unit is a time slot. When handset A transmits data in a
certain channel in time slot N, handset B can transmit in the same channel in time
slot N+ 1. As a result, each channel is shared by several handsets and each handset is
identifiable by the time at which it sends or receives data. Sharing of the same
channel by various handsets is realized in a different way in COMA systems. Here,
several handsets transmit their data in the same channel simultaneously and a unique
code is added to each signal. The receiving party correlates the received signal with
the unique code of the other party, with whom communication is in progress. As a
result, only the signal of interest is derived from the received signal. In this case, the
added code enables identification of transmitters and receivers.
In addition to the access methods described above, two different duplex
methods are available. The transmitter and the receiver are multiplexed on the same
frequency in TDD (Time Division Duplex) systems, as opposed to TDMA systems,
in which several transmitters or receivers are multiplexed. A simple example is a
two-way radio, with which one has to press a button to talk and release it to listen.
The transmitter and receiver use separate frequencies or frequency bands in FDD
(Frequency Division Duplex) systems. Some examples of cellular systems available
on today's market are summarized in Table 7.1.
A simplified transmit path from base-band to antenna in a cellular phone is
shown in Figure 7.1. The information that has to be transmitted originates in the
base-band part in the form of bits. These bits are translated into a modulated data
stream, which is up-converted to the channel frequency .fc by a Radio-Frequency
(RF) mixer. The PA amplifies this signal to the antenna, where some means of
filtering and impedance transformation is present between the output of the PA and
the antenna. This leads to a loss of power between the PA and the antenna, which is
usually referred to as insertion loss. Important issues in Figure 7.1 within the scope
of this book are power control and the demands on signal quality at the antenna.
Optimum supply strategies for Power Amplifiers in cellular phones 243
Table 7.1: Overview of cellular systems on today's cellular telephony market [3]
GSM900 NADC CDMA DCS1800
Geography Europe North America North America, Korea, Europe
Japan
Introduction 1992 1992 1995-1997 1993
Frequency 935-960MHz 869-894MHz 824-849 MHz (down, 1710-1785

Range down down US) MHzdown
890-915 MHz 824-849MHz 869-894 MHz (up, US) 1805-1880
up up 832-834, 843-846, 860- MHzup
870 MHz (down, Japan)
887-889,898-901,915-
925 MHz (up, Japan)
Access TDMA/ TDMAI COMA TDMA/

Method FDMA FDMA FDMA
Duplex FDD FDD FDD FDD

Method
Modulation 0.3GMSK 7t14DQPSK Handset: QPSK 0.3GMSK

(1 bit/symbol) (2 bits/symbol) Base station: OQPSK (1 bit/symbol)
( 1 bit/symbol)
Mobile 3.7 mW-2 W 2.2mW-6W 10nW-1 W 250mW-2W

output
power
Channel 200kHz 30kHz 1.23 MHz 200kHz

spacing
Number of 124 freq. ch., 832 freq. ch., 3 20 (US), 12 (Japan) 3000-6000
channels 8 times1ots users per ch.
perch. (992) (2496)
The base station orders the handset to transmit at a certain output power level for
power control. The control range and the step size depend on the cellular system
Table 7.1 shows different output power ranges for different cellular systems.
Moreover, the power control is organized in the form of a protocol, which also
depends on the cellular system. This protocol specifies the organization in time of
the output power changes. For example, it specifies how often output power changes
may occur, and how long a transition from one output power to another takes. An
example for GSM (Global System for Mobile communication) will be given in
section 7. 7. The output power of the P A has to be ramped up and down during a
transmit burst in TDMA systems, such as GSM, according to a specified power-time
template. This is referred to as the ramping specification, which prevents unwanted
signals in other parts of the frequency spectrum
Power control is realized in a closed loop in most cellular phones, because of
the stringent demands made on it. This is shown in Figure 7 .1, where the measured
output power p meas is forced to be equal to the reference output power p reft dictated
by the base-band part, in a closed control loop. The base-band part obtains
244 Chapter7
information on the required value of P,e~ from the base station. The bandwidths of
these loops can be up to 1 MHz, because the ramping times in current cellular
systems are in the range of tens of microseconds. The bandwidth and temperature
stability of this loop have to be high enough to be able to control the output power to
the right level, irrespective of PA tolerances and temperature variations.
--
Frequency domain demands:
Channel width
Vsup
nme domain
demands:
Output Pmeas
power
Pre!
Figure 7.1: Simplified transmit path in a cellular phone from base-band to antenna
Stringent demands apply to the quality of the signal transmitted at the antenna. The
demands in the frequency domain limit the power transmitted in neighbouring
channels and further away from the carrier frequency fc· This prevents violation of
communication outside the own frequency channel. It is implemented by a spectral
mask for the neighbouring or adjacent channels, which defines the allowed Adjacent
Channel Power Ratio (ACPR). An example is shown in the top right part of Figure
7.1. A similar spectral mask exists over a much wider frequency range, which
defines how much power may be transmitted at distances further away fromfc, for
example in frequency bands of other cellular systems.
Similar quality demands apply in the time domain on the basis of a maximum
allowed Bit Error Rate (BER) for each cellular system. The receiver on the other
side may not be able to distil the correct symbols when a transmitter violates the
demands on the signal quality in the time domain, and the BER may be too high. An
example of a demand specified in the time domain is the vector error, which is the
allowed deviation in amplitude and phase of a vector that represents a certain
symbol in the 1/Q plane [3]. This is shown in the bottom right part of Figure 7.1.
Several trends can be distinguished in cellular systems, which are summarized
below:
Trend 1: Higher ON/OFF ratio of the PA

The first trend is towards a higher ON/OFF ratio of the PA. In the European GSM
system, the ON/OFF ratio increases from 1/8 in the case of conventional GSM
system to 3/8 in the case of newer generations of the system, such as the EDGE
(Enhanced Data rates for GSM Evolution) system This means that three out of eight
instead of one out of eight time slots are used for data transmission. The third
generation (30) UMTS (Universal Mobile Telecommunication System) will be
based on the CDMA principle, with the PA always being ON. The PA current
consumption will have a greater impact on the talk time when it is ON more often.
Trend 2: Non-constant envelope modulation techniques

The second trend is towards non-constant envelope digital modulation techniques.
The reason for this is the more efficient use of the available spectrum for these
techniques, which is needed to accommodate the increasing number of phones and
their increasing functionality. For example, the conventional GSM system uses
OMSK (Gaussian Minimum Shift Keying); see Table 7 .1. This is a constant-
envelope modulation method with which the amplitude of the modulated signal
remains constant. 8PSK modulation will be applied for the EDGE system, leading to
a non-constant envelope signal. The 30 UMTS system will use QPSK modulation,
which is also a non-constant envelope modulation method due to the raised cosine
filtering for band compression.
The transmit path has to be linear with non-constant envelope modulation,
because information is present in the amplitude of the transmitted signal. A simple
way of increasing the linearity of a PA is back off its output power or to drive the
PAin a less efficient but more linear class, such as class A. This leads to an increase
in the PA's current consumption, and hence to a decrease in talk time, as will be
shown in later sections.
Trend 3: Lower average output power of handsets

A third trend is towards a lower average output power, because of the increasing
number of cells and the associated decrease in cell size. Each cellular system is
characterized by a certain maximum output power, as was shown in Table 7 .1. The
PAs are designed to be able to transmit at this maximum output power. The P A is
only used optimally at this maximum output power, as will be shown in the next
section. Assuming a fixed supply voltage, the PA will draw significantly more
current than is necessary at lower output powers. Hence, although a decrease in the
average output power transmitted by the handsets leads to less overall power
consumption, the PA is no longer used optimally and draws more current than is
necessary.
As will be clear from the three trends described above, a reduction of a PA's supply
current is becoming increasingly important. One possible way of achieving this will
be described in the next section.
7.2 The efficiency control concept

The efficiency of a PA ( TJPA) is in this book defined as:
(7.1)
where Pout is the radiated output power and Psup is the power drawn at the supply pin
to enable the amplification. An increase in the PA efficiency for a certain output
power Pout leads to the desired reduction in the PA supply current lsup drawn from
the supply voltage Vsup as illustrated by (7.1).
Voltage V,up is either the battery voltage or the output voltage of a DC/DC
converter in present-day cellular phones. In the latter case, the supply voltage of the
246 Chapter?
P A remains constant during the discharging of the battery. In the former case, the
supply voltage of the PA varies during the discharging, according to the battery
discharge curve. An optimum supply voltage leading to an optimum efficiency
exists for each output power. This will be discussed later in this section. The
efficiency control concept [4] involves the use of this optimum PA supply voltage
for each output power. This ensures that the PA is always used at the maximum
possible efficiency. For a better understanding, the next section gives some more
detailed information on PAs and their efficiency.
7.2.1 Basic information on Power Amplifiers

A simple representation of a PA is shown in Figure 7.2 [5],[6]. This simple
representation holds for the PA classes (A, AB, B and C) considered in this book.
The PAis represented by its final stage only. Several stages will be used in series in
a real PA to ensure the proper power amplification. The transistor shown is bipolar,
but FET devices are also used in practice. When the collector voltage of the
transistor approaches 0 V saturation will introduce non-linear effects. The DC bias
network at the base of the illustrated transistor has been omitted for simplicity. The
RF signals are represented by simple non-modulated sine waves with frequency !RF·
The LC output filter formed by cblocko cfilter and Lfilter resonates at !RF and is assumed
to be ideal, which means that its quality factor Q is so high that the voltage across
Cfilter and Lfitter is always sinusoidal with frequency !RF• even if the current through it
is non-sinusoidal. Non-sinusoidal currents will arise from distortion due to the large-
signal behaviour of the transistor. The assumption of a high Q implies narrow-band
operation [5]. Finally, the resistor Rantenna represents the antenna load.
Coil Lbias• often referred to as 'Big Fat Inductor', is a short-circuit in the case of
DC current and hence the DC voltage at the collector of the transistor equals Vsup·
Lbias is an open-circuit in the case of AC (RF) signals, which means that the actual
voltage at the collector can be higher than Vsup· The maximum amplitude of the AC
signal at the collector is Vsup• because the minimum voltage at the collector is 0 V
and the DC voltage is Vsup· Therefore, the maximum voltage at the collector is 2Vsup·
The blocking capacitor Cbtock prevents DC power consumption in Rantenna and is a
short-circuit for RF signals. The ideal LC filter suppresses higher harmonics of the
input signal, which are caused by distortion as discussed earlier. As a result, only the
signal with the fundamental RF frequency !RF is present across Rantenna·
A well-known theorem states that maximum power transfer occurs when the
load impedance matches the source impedance. The efficiency will be only 50% in
this case. Such an efficiency is unacceptable for the output power levels encountered
in cellular phones. Therefore, PAs are generally designed to supply a specified
output power to a load with the highest efficiency possible [5]. Hence, the objective
is to obtain maximum output power, not maximum power gain. Assume that the
load resistance observed at the collector in Figure 7.2 is Rtoad· Maximum output
power occurs when a full signal swing is present across Rtoad• which is a sinusoidal
voltage with an amplitude Vsur Therefore, maximum power is found from:
pout,max =R
--
v.:p (7.2)
2 load
Vsup RFout
RF;n
Rantenna
Figure 7.2: Simple representation of aPA [5],[6]
The antenna impedance Rantenna is fixed at 50 Q in most cellular phones. The

maximum output power at the antenna in, for example, a GSM phone is 2 W; see
Table 7.1. For a typical supply voltage of 3.6 V, which results from a direct
connection of the PA supply pin to aLi-ion battery, the load at the collector R1oad has
to be 3.24 Q to achieve the maximum required output power of 2 W; see (7.2).
Therefore, an impedance transformation from 3.24 Q to 50 Q is needed. This is
realized by the LC-filter in Figure 7.2. Cblock resonates with Lfilter at the signal
frequency !RF• while cfilter and Lfilter resonate at 3fRF• i.e. the third harmonic of the
input frequency. Series resonance of Cblock and ~11,, will lead to the necessarry low
impedance R1oad of 3.24 Q at !RF· The parallel resonance of Cfilter and Lfilter leads to
the high impedance of 50 Q seen at R1oad· This leads to a suppression of the third
harmonic in the output signal. An impedance transformation network was also
shown in Figure 7 .1. The impedance transformation network in Figure 7.2 has been
assumed lossless. However, due to series resistances of the inductors and capacitors,
a power loss of at least 1 dB will occur in practice. The PA has to transmit at least I
dB more than the desired output power at the antenna, because the output power at
the antenna is specified in the system.
Four classic PA classes are encountered in the literature [5]-[7], notably class
A, class AB, class B and class C. The difference between these classes arises from
the bias conditions. The transistor conducts 100% of the time in class A operation, it
only conducts 50% of the time in class B operation, between 50% and 100% of the
time in class AB operation and less than 50% of the time in class C operation. As a
result, the class C PA is the most efficient, followed by the class B and the class AB
PAs. The class A P A is the least efficient. This can be understood by considering
that power is dissipated in the transistor when current flows while the collector
voltage is larger than zero. The basic equation for the collector current for all classes
is given by:
(7.3)
248 Chapter?
where lvc is the current through Lbias and iRFsin( WRFt) is the current through Cblock·
The different classes depend on the value of lvc. while lc remains greater than or
equal to zero. The collector currents of the different P A classes are depicted in
Figure 7.3.
t
lc
loc,A ----
loc,Ae---
loc, e 0
!i~! t-
~~~
loc.c ~
2«l>AB 1------------------------
2Cl>A
Figure 7.3: Collector currents of different P A classes
The conduction angle 24> is shown in Figure 7.3 for all classes and for an arbitrary
output power. The maximum theoretical efficiency 1Jmax,theory occurs at maximum
output power Pout, max for all classes and can be calculated using (7 .1) [5], where I sup
is the DC content of the collector current lc. The output power follows from the
fundamental component of the current ic, because this fundamental component is the
only current allowed to flow in Rantenna· The maximum theoretical efficiency
1Jmax,theory is then found to be:
2<1> - sin(2<P)
(7.4)
'flmax.rheory = 4(sin(<P)- <P cos(<I>))
In practice, the maximum attainable efficiency 1Jmax at maximum output power

Pout,maxwill be somewhat lower than the theoretical maximum efficiency 1Jmax,theory·
This is amongst others caused by non-zero minimum collector voltage and
inevitable losses in interconnect and filter components. A summary of the PA
classes A, AB, Band Cis given in Table 7.2.
Table 7.2: Summary of PA classes A-C: Conduction angle 2C/J [rad], maximum theoretical
efficiency TJmax.theory [%] and linearity class
Class: 2<1> [rad]: llmax.theorv [% ]: Linearity class
A 27t 50 ++
AB 7t<2<1><27t 50< TIIIIIX < 78.5 +
B 7t 78.5 -
c 2<1><7t 78.5 < Tlmax < 1001 --2
Note 1: The maxunum effic1ency of 100% for class C occurs at zero output power
Note 2: Class C PAs are not often encountered in practice because they demand substantial filtering due
to their non-linearity. This leads to an expensive system solution
The efficiency will be lower than TJmax at output powers lower than Pow,max· For each
of the classes in Table 7 .2, the efficiency expressed in P outo Vsup and TJmax is given by:
(7.5)
Substitution of (7.2) into (7.5) shows that the efficiency is indeed TJmax at
Pow=Pout,max· The value of TJmax can be found from PA data sheets and has an upper
limit of TJmax,theory for each PA class. Moreover, the efficiency is proportional to the
root of the output power Pout· In practice, the value of Vsup will be chosen such that,
given the value of the transformed antenna load impedance observed at the collector
(Rtoad), maximum output power can be transmitted; see (7.2). The efficiency will be
lower than TJmax when Vsup has the same value at output powers lower than Pout,max·
This can be understood from the way in which power is controlled in practice. Either
the bias of the PAis changed or a linear regulator in the supply line ensures a lower
voltage at the collector/drain of the output transistor. The difference between these
two power control methods is illustrated in Figure 7 .4. The voltage Vcon denotes the
control voltage needed to establish a variation in output power. Only the final stages
of the PAs have been depicted for simplicity.
Vsup
Vsup
Vcon---1
V'sup
Lbias
Filter and
impedance
matching
network
Figure 7.4: Two different approaches to power control in PAs: (a) bias control for bipolar
PAs, (b) drain control by means of a linear regulator for FET PAs
250 Chapter 7
The bias control method shown in Figure 7 .4a is used for bipolar PAs. The result is
an output voltage with less than full swing for output powers lower than Pous,max·
This leads to an efficiency lower than 1'/max• because 1'/max is only attained at full
swing. The power control method with the linear regulator is used for FET PAs. In
that case, a FET switch has to be present in the supply line, as indicated in Figure
7.4b, to switch the device off completely, because most FET devices are 'normally-
on'. Therefore, power control is realized on the drain-side of the FET by using the
switch as a linear regulator. This results in a voltage V'sup lower than Vsup• as
indicated in the figure, and hence a lower output power. Although the voltage across
R1oad observed at the input of the impedance matching network will in this case have
an amplitude of V'sup• which is full swing, the efficiency is still lower than
maximum. The reason for this is that the efficiency of the linear regulator decreases
at lower values of V'sup and a fixed value of Vsup· More information on linear
regulators will be given in section 7.3.
Other P A classes in which the output transistor is used as a switch are also
encountered in the literature [5]-[7]. These PA classes are D, E and F. The
theoretical efficiency of the PAis 100% in these cases, because an ideal switch has
zero voltage across it when current flows, whereas zero current flows when the
switch does not conduct. The efficiency is lower than I 00% in practice. These PAs
are highly non-linear, resulting in more filtering at the output. This leads to a more
expensive system. Therefore, PAs of classes D, E and F have not yet been used in
cellular phones. The remainder of this chapter will concentrate on class A, AB and B
type PAs.
7.2.2 Optimum supply voltage for optimum efficiency

In order to have an efficiency of 1'/max at all output powers, Vsup should be made
dependent on Pour· This is the essence of efficiency control and applies to both
bipolar and FET PAs. The optimum value Vsup,opr as a function of Pout can be
obtained from the general equation (7.5):
(7.6)
where R1oad reflects the transformed antenna impedance again. When the supply
voltage for maximum output power Pout,max is named Vnom• which is the nominal
supply voltage described in the data sheet of the PA, (7.6) can be rewritten as:
vsup,opt = (7.7)
using (7 .2). Some means of DC/DC conversion is needed for efficiency control to
translate the battery voltage into the desired value of Vsup,opt· The advantages of
efficiency control are listed below:
• The maximum efficiency of the PA, as specified in the data sheet, is maintained
over a wider range of output powers, as opposed to only at maximum output
power.
• The application of a well-defined supply voltage for each output power may
make it easier to design the PA. The reason for this is that the PA designer now
only has to take a limited range of supply voltages into account during the
design, as opposed to a whole range of supply voltages when the PA is
connected directly to the battery. This advantage already holds when a DC/DC
converter with a fixed output voltage is used.
• The PA designer can choose the optimum nominal supply voltage when a
DC/DC converter is present in the system anyway. Impedance transformation is
needed to be able to transmit at the desired output power, as was explained in
section 7.2.1. (7.2) shows that the higher the nominal supply voltage for
maximum output power is chosen to be, the higher the load observed by the PA
will be allowed to be and the lower the ratio of the 50 Q antenna impedance and
the impedance R 1oad observed at the PA side will become. This lowers the losses
in the impedance transformation network, because the parasitic resistances in
this network are now smaller relative to R1oad·
The implementation of efficiency control in a cellular phone does not imply drastic
changes to the transmit architecture. An example for a GSM phone will be given in
section 7. 7.
The transmit path needs to be sufficiently linear when the cellular system
employs non-constant envelope modulation; see section 7 .l. Full swing RF signals
at the output of the P A are then not allowed, because saturation of the output
transistor will lead to an unacceptable level of non-linearity. In that case, a
somewhat higher value than that found from (7.6) and (7.7) has to be chosen for
Vsup.opr · This will lead to a smaller improvement in efficiency. A quantitative
example will be given in section 7.6, which shows that a considerable improvement
can still be realized in practice.
7.3 DC/DC conversion principles

In general, two approaches exist for converting one voltage into another. The first
approach involves a time-continuous circuit with a dissipative element, whereas the
second approach involves a time-discrete circuit with an energy-storage element.
The first approach is only suitable for converting a higher voltage into a lower
voltage, which is down-conversion, whereas the second approach enables up- and
down-conversion. The two approaches are illustrated in Figure 7.5, in which the
battery voltage V;n is converted into Your·
(a) (b)
Figure 7.5: Two approaches to convert a battery voltage into another voltage. (a): Time-
continuous with dissipative element (V;n > Vour only) (b) Time-discrete with energy-storage
element
252 Chapter 7
The dissipative element in Figure 7.5a remains connected between the battery and
the load. The efficiency of the voltage conversion will always be lower than 100%,
because of its dissipative nature. The energy-storage element in Figure 7.5b is first
connected to the battery to store energy, after which it is connected to the load to
supply this energy. The efficiency of the voltage conversion process is 100% in the
theoretical case in which no energy is lost in the energy-storage element and
switches. The type of time-discrete voltage converter employed and the char-
acteristics of the employed components will determine the efficiency in practice. An
energy buffer Cbuf is necessary, because of the time-discrete nature of converters of
this type.
7.3.1 Linear voltage regulators

The configuration of Figure 7.5a is commonly known as a linear voltage regulator
[8]. A more detailed schematic representation of a linear voltage regulator is shown
in Figure 7.6. It consists of a transistor, which may be of any type, controlled by a
regulator, which compares a fraction of Vout with a reference voltage Vref' Linear
regulators vary in the used type of transistor and its drive circuit. The transistor is
operated in the saturation region in the case of FETs and the linear region in the case
of bipolar transistors. This means that the output current 10 .,. and hence Vout• will
hardly change when V;n changes. However, a certain minimum voltage difference
V;n- Vout• which is the dropout voltage, has to be present for proper operation. A
simple resistor represents the load in Figure 7 .6.
Figure 7.6 illustrates that the maximum efficiency 1] is the ratio of V0 u1 and Vin·
A decrease from this maximum value is caused by the current consumption of the
linear voltage regulator itself, including the oparnp, voltage reference and current
through R 1 and R 2• Hence, 10 "' is smaller than lin· The efficiency of the linear voltage
regulator will be higher in the case of smaller differences between V;n and Vout• as
can be understood from the ratio. Therefore, the term Low DropOut regulator (LDO)
is often encountered in practice when a low voltage drop can be accommodated, the
term 'low' being a relative term.
V;n > Vout
Load
{V;n - Vout)m;n = dropout voltage
~ lout
Figure 7.6: Basic schematic representation of a linear voltage regulator
Linear voltage regulators are encountered in portable devices when the difference
between the battery and load supply voltages is not too high, because the efficiency
will then still be acceptable. An advantage of linear regulators is that they are cheap
and small because they do not need an energy-storage element, which usually takes
up quite some volume. Linear regulators are often used as filters, because variations
in Vin are suppressed in Vout· Although not included in Figure 7.6, small capacitors
with values specified in the data sheet are added to the input and output of a linear
voltage regulator in practice. The output capacitor improves the response to load
changes and usually implements regulator loop frequency compensation.
7.3.2 Capacitive voltage converters

A converter as shown in Figure 7 .5b has to be used when there is a large difference
between the battery voltage and the voltage needed by the load, or when the load
voltage should be higher than the battery voltage. The energy-storage elements can
either be capacitive or inductive. The basic principle of a capacitive voltage
converter is shown in Figure 7.7. Such a configuration is commonly referred to as a
charge pump. The switches are operated from a non-overlapping clock signal with
periods C/J1 and cf?.. The capacitors are connected in parallel to the battery for the up-
converter for period C/J1 and each capacitor is charged to the battery voltage. The
capacitors are connected in series for period C/7., added to the battery voltage, and
connected to the output buffer capacitor Cout· At no load, this leads to an output
voltage V0 u1=(n + 1)·Vi., where n is the number of capacitors.
Load
(a) 2 n
V;n > Vout

V;n r-- - - - - - " - -- - - -
Cout Load
~2 ~2
(b)
Figure 7.7: Basic schematic representation of a capacitive voltage converter: (a) Voltage up-
converter (b) Voltage down-converter
The capacitors are connected in series for the down-converter for period C/J1 and the
total series connection is charged to the battery voltage. The capacitors are
connected in parallel for period cf?. and connected to the output buffer capacitor Cout·
This leads to an output voltage V0 .,= Vi,/n at zero load, with n being the number of
254 Chapter7
capacitors. By combining parallel and series connections for periods lP1 and ~.
non-integer conversion factors can be realized.
Efficiency considerations
Consider the charging of an ideal capacitor by a voltage source V through a switch S
with an on-resistance Rs. This is illustrated in Figure 7.8, from which the following
equation can be derived for the energy Ec(t) stored inC as a function of timet:
(7.8)
And for the energy ER(t) dissipated in Rs as a function of timet:
(7.9)
+ VR(t)- v .:!.
S Rs l(t} = Rs e'
~+
-t
+ VR(t) = V • e'
v C Vc (t) Vc(t) = V • (1
-t
- e')
't=RsC
Figure 7.8: Charging a capacitor C by a voltage source V through a switch S with an on-
resistance Rs
Ec(t) and ER(t) both become Emax= ~cyl when t approaches infinity. This means
that, irrespective of the value of Rs. equal amounts of energy are stored and
dissipated. This also holds when Rs is zero or is not constant. The latter case occurs,
when the capacitor Cis charged through a transistor. Table 7.3 shows the course in
time of Ec(t) and ER(t). For clarity, these energies have been normalized to the
maximum energy Emax·
Table 7.3: Normalized energies Edt) and ER(t) and normalized voltage Vc(t)/V at normalized
times t/'t
tl't Vc(t)/V [%] Ec(t)!Emax [%] ER(t)/Emax [%]
0.25 22.4 5 39.7
0.38 31.6 10 53.3
0.69 50 25 75
1 63.2 40 86.5
1.23 70.7 50 91.4
2.97 94.9 90 99.7
3.68 97.5 95 99.9
Table 7.3 shows that storing the first 5% of Emax causes an energy loss of 39.7% of
Emax in Rs. However, when the same amount of energy is stored starting at 90%
(90% ~ 95% of Emax), the energy loss is only 0.2% of Emax· Hence, although the
energy storage occurs fast at the beginning of charging, because it only takes 0.25't
to store the first 5% of Ema:n it is rather inefficient. The efficiency is 5/(5+39.7) =
11.2%. The efficiency increases to 5/(5+0.2)=96.2% when Vc approaches its end
value of V, but it takes 0.71t to store 5% of Emax from 90% to 95%. This means that
the capacitors should be charged (almost) to the source voltage V in order to have a
high efficiency and the capacitance should be high, so that the voltage drop is
minimal in the discharge period. Hence, only a fraction of the stored energy is active
in the transfer process.
The frequency at which the energy is transferred or the amount of the energy
that is transferred each cycle need to increase when the required output power of the
charge pump is high. The combination of high efficiency and high power leads to
very large capacitance values, because storing energy takes more time at the end of
the charging process. These large capacitors cannot be integrated with the switches
and control circuitry on a single IC in practice. External capacitors should be used
instead. When efficiency is not a big issue and/or the load current is small, smaller
capacitors can be used, which creates the possibility of integrating them.
7.3.3 Inductive voltage converters

The basic principle of using an inductive voltage converter is depicted in Figure 7.9.
This type of converter is generally referred to as a DC/DC converter. As with ca-
pacitive converters, up-conversion and down-conversion of the battery voltage are
possible. The time-discrete nature of the two converters can be clearly recognized by
the switches.
V;n < Vout V;n> Vout
Vout
Load
(a) (b)
Figure 7.9: Basic schematic representation of an inductive voltage converter: (a) Voltage up-
converter (b) Voltage down-converter
Switch S 1 is closed first and energy is stored in the inductor for both the up- and the
down-converters. The inductor current ramps up linearly in the ideal case of zero
series resistance. Switch S 1 is opened at a certain moment, depending on the desired
ratio of V;n and V0 u1 and the value of the load. The current will flow through the
diode D, because the inductor current cannot change instantly, and the polarity of
the voltage across the inductor changes, which leads to a decrease in the energy
stored in the inductor and the current will ramp down accordingly. The energy that
was stored in the inductor for the period when S1 conducted will now be transferred
to the output buffer capacitor Cow· Switch S2 is not strictly necessary in either case.
256 Chapter?
When it conducts in the second phase, the efficiency will be higher than without the
switch, since the voltage across diode D will be lower than without the bypass
switch. Bypassing the diode by switch S2 is commonly referred to as synchronous
rectification.
The waveform of the current through the inductor is shown in Figure 7 .I 0 for
both the up- and the down-converter. For simplicity, it is assumed that the inductor
current is never zero. This is called continuous conduction mode, as opposed to
discontinuous conduction mode, in which the inductor current decreases to zero each
switching cycle. The depicted waveform is valid for a high-efficiency converter, in
which the voltage drop across the switches during current conduction can be
neglected.
dl= vt.M
t-
Figure 7.10: Inductor current h for the converters of Figure 7.9 in continuous conduction
mode with ideal switches
The conduction period of S1 is t5 T. The duty cycle t5 is defined as the ratio of the
time for which S1 conducts and the total switching time T and has a value O<o<l.
As a consequence, S2 conducts for a period (1-0)·T. In steady-state, in which case the
input voltage, load current and duty cycle are fixed, all the energy that was stored in
the inductor for the period t5 Twill be transferred to CoUJ during ( 1-o)· T. This means
that the Volt·second product for the coil is equal in both periods, or
Vd0T)·t5T=Vd(l-o)T)·(l-o)"T. For the up-converter, this leads to:
v.m8T =tv
\ ou
, - v.m XI- o...,.,... ==> vout
)1 V.
=_1_
l-0 (7.10)
m
whereas for the down-converter it leads to:
(7.11)
(7.10) and (7.11) illustrate that the ratio of Vin and Vout can be controlled by
controlling the duty cycle 0. The ratio of the output and input voltages is greater than
unity in the case of the up-converter and smaller than unity in the case of the down-
converter, because 0<8<1. An up/down-converter can be constructed by combining
the configurations of Figure 7.9a and b. Four switches are necessary in that case and
two conducting switches are always present in the current path. This leads to an
efficiency that is poorer than that of single up- or down-converters, because then
only one conducting switch is present in the current path.
The number of components needed to achieve a certain ratio of Vin and Your is
much smaller than in the case of capacitive converters. Only the inductor and diode
are realized with external components in practice and the switches and the controller
can be integrated. Another difference is that the ratio of Vm and Your in inductive
converters can be changed by controlling 8. In the case of a capacitive converter, the
circuit topology has to be changed to achieve this or additional losses must be
introduced in the form of an additional linear voltage regulator. Moreover, an
essential difference between capacitive and inductive voltage converters is the effi-
ciency of energy storage.
Efficiency considerations
Consider the storage of energy in an ideal inductor L from a voltage source V
through a switch S with an on-resistance R5. A voltage source is again to be
considered, because this book deals with batteries as power sources and a battery
can be regarded as a voltage source. This is shown in Figure 7 .11.
v -t
l(t)=R8 ·(1-e7 }
+ + VR(t) =V • ( 1 - e=!-)
-t
v VL (t) VL(t) =V ·e 7
L
't= Rs
Figure 7.11: Charging an inductor L by a voltage source V through a switch S with an on-
resistance Rs
The energy Edt) stored in Las a function of timet can be obtained from:
(7.12)
where/=:. VJR 5 • When t approaches infinity, the energy stored equals Emax=:. lf2L/2• The
energy ER(t) dissipated in R as a function of time t is obtained from:
(7.13)
An important difference with respect to a capacitive converter is that ER(t) can be

made zero when Rs is made zero. Hence, energy can be stored in an inductor from a
voltage source with 100% efficiency in the ideal case when Rs is zero, as opposed to
storing energy in a capacitor from a voltage source, in which case a certain amount
of energy of ~cy2 is always lost, irrespective of the value of Rs. when the capacitor
is charged from 0 V to a voltage V. This means that inductive voltage converters can
258 Chapter?
achieve a 100% efficiency with ideal components. This is not the case for capacitive
converters. This holds for a voltage source as input, which is the case in battery-
powered equipment. Inductive voltage converters are for this reason found in many
portable products.
Although an efficiency of 100% is theoretically possible with an inductive
voltage converter, losses will occur in practice, resistive and capacitive losses.
Resistive losses are caused by series resistance in the current path, for example an
Equivalent Series Resistance (ESR) of the inductor and on-resistances of the
switches. Moreover, each switch has a parasitic capacitance, which leads to
capacitive switching losses in practice.
7.3.4 EMI problems involved in capacitive and inductive voltage converters

The time-discrete voltage converters can cause Electro-Magnetic Interference
(EMI), for example, due to voltage ripple on the output buffer capacitor caused by
the ESR, or due to current loops with varying currents that span a significant area.
These problems can be avoided by carefully choosing the components, filtering and
Printed-Circuit Board (PCB) layout. An output voltage control scheme with a fixed
switching frequency outside the frequency bands of interest could also be
considered. A linear regulator often follows an inductive voltage converter when the
load is highly sensitive to EMI, as in RF circuitry in cellular phones. The voltage
difference across the linear regulator should be kept low for efficiency reasons, as
discussed earlier. However, the linear regulator has the advantage of offering
additional filtering [9]. Moreover, the addition of a linear voltage regulator to the
output of an inductive voltage up-converter implies the possibility of realizing a
voltage up/down-converter [10). Finally, one should realize that the output spectrum
of a down-converter will be more favourable than that of an up-converter. The
reason is the difference in shape and frequency content of the current that flows
through the output capacitor for both converter types.
7.3.5 Inductive voltage conversion for efficiency control

The battery voltage has to be converted into a variable PA supply voltage for
efficiency control; see section 7.2. An inductive voltage converter is most suitable
for this from an efficiency point of view. Moreover, the ratio of ~n and Vout can be
controlled relatively easily. This makes it easy to change from one converter output
voltage to another. However, the influence of voltage ripple at the converter output
on the PA's RF behaviour should be carefully investigated. This will be discussed in
sections 7.6 and 7.7.
7.4 Simulation model derivation

Simple simulation models for a DC/DC down-converter and a PA will be derived in
this section. Both models will be used in section 7.5 to compare different supply
strategies of a PA by means of simulation.
7.4.1 DC/DC down-converter

The efficiency of a DC/DC converter is obtained from:
n _ pout _ pout
'I DC/DC - - - - - (7.14)
~n,total pout + ~oss
The power loss P 1oss includes ohmic loss due to parasitic series resistance R 1oss and
switching losses. The total switching losses are obtained from fswitchCpara V 2switch•
where !switch is the switching frequency, Cpara is the total parasitic capacitance that
has to be charged and discharged each switching cycle, and Vswitch is the switch drive
voltage. Vswitch should be as high as possible to minimize the switch resistance. Vswitch
is made equal to V;11 , because this is the highest voltage available in a down-
converter. This leads to the following expression for TJoctDc:
(7.15)
The output current and power of a DC/DC down-converter are limited. Figure 7.9b
illustrates that for a duty cycle O=l, switch S1 is always conducting and the
converter is nothing more than a connection from V;n to Vout through an always-
conducting switch and a coil. This situation will eventually occur when V;n decreases
towards Vout· The maximum current lout that can be supplied at a certain difference
between V;n and Vout is now obtained from:
I =Vin- Vout (7.16)

out R
loss
Maximum output power Pow,max occurs when the load resistance R1oad equals R1oss·
The efficiency will be 50% in this case and Pout, max is obtained from:
p = v;~ (7.17)
out,max 4 Rloss
A conventional PA behaves as a current source and the value of this current I sup
drawn at the PA supply pin depends on the requested output power and the
efficiency of the PA; see section 7.4.2. The battery voltage is the input of the
converter and the supply voltage requested for the PAis the output when a DC/DC
down-converter powers the PA. The converter will no longer be able to supply
enough current at the requested supply voltage of the PA below a certain minimum
voltage difference between the input and the output of the converter; see (7 .16). As a
consequence, the PA cannot supply the requested output power. The series
resistance between the battery and the PA can be lowered in this case by connecting
a low-ohmic bypass switch in parallel to the converter. The battery voltage may now
drop a little further, since the bypass configuration has a lower resistance than R 1oss·
This will extend the talk time of a cellular phone. The PA supply voltage now varies
with the battery voltage through:
260 Chapter?
(7.18)
The simulation model for a DC/DC down-converter that will be used in section 7.5
is depicted in Figure 7.12.
I. _ Vout lout
'"- TlDCIOC V;n
Figure 7.12: Simulation model for the DC/DC down-converter used in section 7.5
Efficiency 1'/DCIDC in Figure 7.12 is given by (7.15). Switches SA and S 8 are assumed
to be ideal. Switches SA conduct when V 1n-Vout is larger than 10 u1R 1om where lout is the
current drawn at the output. Switch S8 does not conduct in this case and the output
voltage is fixed by voltage source Vout· When V 1n· Vout becomes smaller than loutRtoss•
only switch S8 conducts and the voltage at the output of the converter is given by
(7.18), with Vbat= Vtn and lsup=lout·
7.4.2 Power Amplifier

The behaviour of the PA was described in section 7.2. The PA will be simply
modelled as a current source in the simulations in section 7.5, having the value:
[ = ~utf)PA (7.19)
sup V
'f'JPA sup
where Pout is the requested output power and OpA is the duty cycle of the PA, which
is smaller than 1 in the case of TDMA systems. Vsup is the PA supply voltage, which
is the voltage across the current source that models the PA, and 'I'JPA is given by:
11 = vsup,opt ( 11 _ v -vnom sup,opt ) (7.20)

'IPA Vsup '/max Q.S
where the first term is found from (7.5) and (7.6). A modification is made to 'I'Jmax in
(7 .20) when Vsup,opt is lower than Vnom• hence when Pout is lower than P out,rnox; see
(7. 7). The value of 0.8 was derived from measurements obtained with a practical PA
applied in GSM phones [11]. This models the decrease in 1'/max with respect to the
theoretical maximum efficiency 17max,theory due to the increasing importance of

saturation effects at lower supply voltages. The course of 1JPA as a function of the
supply voltage for maximum output power P 1 =Pout,max and a lower output power P 2
is shown in Figure 7.13.
TJmax
Tl max < Tlmax

1
-1
Error ~
-----,-------
I
I
I
Error
i
TIPA
V sup, op12 Vsup, opt1 = V nom Vsup-
Figure 7.13: PA efficiency (1/pA) versus supply voltage (Vsup) for maximum output power P 1
and for P 2<P1
An optimum supply voltage Vsup,opt is shown for both output powers, where Vsup,opt is
lower at lower output powers, which follows from (7.7). The PA efficiency will
decrease inversely proportionally to Vsup when Vsup is higher than Vsup,opt; see (7.20).
The maximum attainable efficiency decreases with the output power in accordance
with (7.20). The PAis no longer able to transmit at the requested output power when
Vsup drops below Vsup,opt· This is indicated by the arrows labelled 'Error' in Figure
7.13. This can be taken into account in a PA simulation model by changing an error
signal Verror from 0 (Vsup?Vsup,opt) to 1 (Vsup<Vsup,op1). The PA simulation model is
+ +
t lsup according
to (Eq. 7.19)

Figure 7.14: Simulation model for the PAused in section 7.5
7.5 Theoretical benefits of efficiency control

Different supply strategies of a PA will be compared by means of simulations in this
section. In addition to the simulation models developed in the previous section, the
Li-ion model of chapter 4 will be used to translate the efficiencies of the various
supply strategies into operating times Toper The strategies differ in the application of
DC/DC conversion. The situation where no DC/DC conversion is applied and a PA
is directly powered by a battery will be taken as a reference. A DC/DC converter
262 Chapter?
will be used for converting the battery voltage into the PA supply voltage in all other
cases. The converters differ in efficiency, type, use of a bypass switch and control of
the output voltage. The relative change in Toper with respect to the reference situation
will illustrate the possible benefits or drawbacks of the use of each of the DC/DC
converters.
The efficiency of a converter, which is mainly determined by the on-resistance
of the switches, of course influences Toper to a large extent. Therefore, converters
with different on-resistances of the switches will be compared. The desired supply
voltage of the PA will be lower than the battery voltage in many practical cases.
Therefore, most converters considered in the simulations will be down-converters.
However, down-converters will not be able anymore to supply the desired PA
power when the battery voltage approaches the desired PA supply voltage; see
section 7 .4.1. This situation does not occur when an up/down-converter is used.
Therefore, this converter type will also be considered in one of the simulations.
Alternatively, the battery voltage at which down-converters fail to supply the
voltage requested by the PA can be decreased by applying a bypass switch; see
section 7 .4.1. The use of such a switch will be considered in some simulations to
investigate the influence on Toper· Of course, the on-resistance of the bypass switch
determines its effectiveness. Different values for this on-resistance will therefore be
considered. Finally, the output voltage of the converters will either be fixed or
variable according to (7.7). The latter situation will illustrate the possible benefits of
efficiency control. Seven supply strategies in total will be compared in simulations,
including the reference situation.
7.5.1 Simulation set-up

The PA model was connected directly to the battery model in simulation 1. This is
the reference situation. A DC/DC down-converter model was placed between the
battery model and the PA model in simulations 2-6. A DC/DC up/down-converter
was placed between the battery and the PA in simulation 7. All the simulations
started with a full battery and ended either when the battery was empty, i.e. when the
battery voltage had dropped below 2. 7 V, or when Verror in the PA model had
become 1, indicating that the PA was no longer able to transmit at the desired output
power level. Toper was determined for a range of output powers and for each of the
supply strategies. The parameter values used in the PA model are listed in Table 7.4.
Table 7.4: Parameter values for the PA model of section 7.4 used in simulations 1-7
Parameter: Value: Unit:
PIIIIX 35 dBm
Pout Variable dBm1
Ynom 3.2 v
Tlmax 45 %
cSPA 3/8 -
Note I: P [dBm]=JO·Jog(P [mW]), so 0 dBm =I mW
The specifications of 35 dBm for Pmax and 3/8 for OpA were derived from the GSM
standard, in which the future increase of OpA from 118 to 3/8 was anticipated. The
maximum output power of the PA is 35 dBm, because a power loss of 2 dB between
the PA and the antenna due to impedance matching and filtering was taken into
account; see Figure 7.1. The values of Vnom and 'flmax are in agreement with present-
day GSM PAs. The values of Pour increased from 21 dBm to 35 dBm in 2 dB
increments. This increment size is used in the power control of PAs in GSM cellular
phones, as will be described in more detail in section 7. 7.
The supply strategies followed in simulations 2-6 differ in the parameters used
in the DC/DC down-converter model. These differences concern:
• The value of the on-resistance of the switches (Rswitch) and the ESR of the coil
(Rcou), where Rswitch+Rcou=Rtoss· This influences the efficiency of the converter.
• The use of a bypass switch and, when used, its on-resistance (R!Jypass).
• The control of the output voltage of the converter: V out is fixed and equal to Vnom
of the P A or Vout is variable and equal to Vsup,opt for each Pout (efficiency
control).
The parameter values for the DC/DC down-converter model shown in Figure 7.12
used in simulations 2-6 are listed in Table 7.5.
Table 7.5: Parameter values for the DC/DC down-converter model of Figure 7.12 used in
simulations 2-6
Parameter: Sim. 2 Sim.3 Sim.4 Sim. S Sim. 6 Unit:
Yout 3.2 3.2 3.2 3.2 Var. v
Rswitch 0.15 0.15 0.055 0.055 0.15 Q
Rcoil 0.05 0.05 0.022 0.022 0.05 Q
Rbypass 109 0.1 109 0.05 109 Q
Cpara 20 20 20 20 20 pF
fswitch 400 400 400 400 400 kHz
Efficiency control was applied in simulation 6. Extra low-ohmic values for Rswitch
and Rcoil were used in simulations 4 and 5. All values of Rswitch and Rcoil were
obtained from several data sheets of DC/DC converters and coils [12],[13]. The use
of a bypass switch with different on-resistances was considered in simulations 3 and
5. A value of 1 GQ was used for Rbypass in the other simulations, which means that
no bypass switch is used. The value of Cpara was based on the values for practical
converters, whereasfswuch was chosen in accordance with data sheets [12].
A fixed efficiency was used for the up/down-converter in simulation 7. No
efficiency control was considered, and therefore an output voltage of 3.2 V was
used. Simulations with design software for a DC/DC up/down-converter [14] were
used to derive an efficiency of 85%. This is an average value for various input
voltages, a switch resistance of 100 mQ, a coil ESR of 20 mQ, a switching
frequency of 400kHz and a load corresponding to a PA output power of 35 dBm.

The results of simulations 1-7 are summarized in Table 7.6, which lists the relative
change in Toper with respect to the results of simulation 1. Therefore, the
improvement with respect to this simulation realized in simulation 1 itself is 0% at
all output powers. Several conclusions can be drawn from Table 7.6:
• A benefit of using a down-converter with a fixed output voltage of Vnom• instead

of connecting the PA directly to the battery, is the positive effect on the
operation time of the PA. Compare the results of simulations 2-5 with the result
of simulation 1.
264 Chapter 7
• A benefit of using a down-converter with lower-ohmic switches and a coil with

a lower ESR is an increase in the operation time of the PA. Compare the results
of simulations 4 and 5 with those of simulations 2 and 3.
• A benefit of using a bypass switch with the down-converter is a slight increase
in operation time at maximum output power. The effect on Toper is negligible at
lower output powers, in comparison with the situation without a bypass switch.
Compare the result of simulation 3 with that of simulation 2 and the result of
simulation 5 with that of simulation 4 at maximum output power.
• Efficiency control yields a significant increase in Topero especially at lower
output powers. Compare the result of simulation 6 with the results of all the
other simulations.
• A drawback of using an up/down-converter is a decrease in Toper at all output
powers. Moreover, the silicon area of this converter is larger than that of a
down-converter due to the two extra switches. Compare the result of simulation
7 with the results of all the other simulations.
Table 7.6: Simulation results obtained for different PA supply strategies. The table shows the
relative changes in operation time Toper of the PA in [%] with respect to the result of
simulation 1, in which the PA was connected directly to the battery. The relative changes are
shown for different values of the output power Pout of the PA
Simulation Chan2e in T ...,.. relative to simulation 1 [%]
Pout [dBm]
35 33 31 29 27 25 23 21
1: No DC/DC 0 0 0 0 0 0 0 0
2: DC/DC down, +6.4 +10.4 +12.0 +13.3 +14.3 +15.3 +16.0 +16.6
R,,.;lch=0.15Q,
Rcou=0.05Q, no bypass
switch
3: DC/DC down (see +8.3 +10.4 +12.0 +13.3 +14.4 +15.3 +16.0 +16.6
2)R,....,=0.1Q
4: DC/DC down, +11.3 +12.6 +14.0 +14.8 +15.5 +16.3 +16.8 +17.2
R,,.;"•=0.055Q,
Rcoa=0.022Q, no
bypass switch
5: DC/DC down (see +11.7 +12.7 +13.9 +14.8 +15.6 +16.2 +16.8 +17.2
4) R.,pau=0.05Q
6: DC/DC down (see +6.4 +41 +80.8 +133 +196 +274 +275 +275
2) v,.,= v,..,opt for all
Pow (efficiency
control)
7: DC/DC up/down, -3.1 -4.2 -3.1 -2.5 -1.9 -1.3 -0.9 -0.5
no bypass switch
7]=85%
Although not shown in Table 7.6, it was found in all the simulations that the
operating time of the PA ended at maximum output power, because Verror became 1.
This occurred at a battery voltage of 3.2 V in simulation 1. This is illustrated in
Figure 7.13 at maximum output power, where Vnom equals 3.2 V. This occurred at a
battery voltage higher than 3.2 V in simulations 2-6. It is tempting to say that Toper
will be lower when it ends at a higher battery voltage, but this is not the case.
Inspection of the column for 35 dBm obtained in simulations 2-6 shows that Toper
with a down-converter is still higher than without a down-converter. The reason for
this is the higher total efficiency of the solution with a down-converter in the
discharge region where the battery voltage is still considerably higher than 3.2 V.
When the combination of a PA and a down-converter is regarded as a black box

connected to a battery, the minimum battery voltage at which this black box can still
transmit at maximum output power is higher than in the case of a bare PA. Any
output power can be transmitted at any battery voltage when an up/down-converter
is used in the black box. However, the simulations with an efficiency of 85% show
that a drawback is the decrease in Toper at all output powers. Alternatively, one could
consider the use of a bypass switch that is switched on in the critical situation. This
decreases the battery voltage at which the maximum output power can still be
supplied by the black box, as described in section 7 .4.1.
The PA model of Figure 7.14 draws average power when OpA is smaller than 1.
For example, for OPA=3!8, the model draws power continuously with a value of 3/81h
of the power that would be drawn during three time slots. This occurs in practice
when a large buffer capacitor is used at the output of a DC/DC converter. The peak
power during the transmit burst is then drawn from this buffer capacitor while the
DC/DC converter supplies average output power. This is a sensible thing to do from
an efficiency point of view. Smaller switches can be used in the converter, leading to
smaller switching losses. At the same time, the large peak currents are drawn from
the capacitor via a low-ohmic path, assuming that the buffer capacitor has a low
ESR.
In order to obtain insight into what happens when no large buffer capacitor is
used, some additional simulations were performed. A simple modification can be
made to the PA model of Figure 7.14 to make it draw peak power instead of average
power. The PA supply current is shown in Figure 7.15. An extra parameter Tperiod
was added to the model. The PA draws peak power for the time OpATperiod·
Simulations 1 and 4 were performed once again using OPA=3/8 and Tpenoo=l s. The
change in Toper observed in simulation 4 relative to the value in simulation 1 is
shown in Table 7.7 for two output powers. Simulation 4 was used, because lower
losses are especially important when peak power is drawn.
Table 7. 7 shows that Toper decreases at maximum output power when a DC/DC
down-converter is used without a large buffer capacitor at the supply pin of the PA.
However, it still makes sense to use a DC/DC down-converter at output powers
lower than maximum, even without the application of efficiency control. Toper is
lower in absolute value when peak power is drawn from either the DC/DC down-
converter or directly from the battery. Therefore it makes sense to use a buffer
capacitor.
Table 7.7: Change in Toper observed in simulation 4 (with a DC/DC down-converter) relative
to the value in simulation 1 (no DC/DC converter) with a PA drawing peak instead of average
power; see Figure 7.15
P.,,, [dBm] Chan2e in T •.,.. relative to the value in simulation 1 [%]
35 -22
33 +6
7.5.3 Conclusions
The operation time Toper of a PA with the specifications listed in Table 7.4 powered
by a down-converter is higher than that of a PA connected directly to a battery.
Increasing the converter efficiency by using lower-ohmic switches has a positive
effect on Toper for all PA output powers. The use of a bypass switch only increases
Toper at maximum output power. The use of an up/down-converter with a realistic
efficiency of 85% has a negative effect on Toper- Toper increases substantially relative
266 Chapter?
to a fixed output voltage when the converter output is made variable with efficiency
control. This is especially true at output powers lower than maximum. Therefore, the
efficiency control supply strategy will be investigated in more detail by means of
measurements in the remainder of this chapter.
i
I _ Pout
sup- TIPAVsup
-------- r - -
........ t-
5pATperlod
Tperiod
Figure 7.15: PA supply current !,up of the modified PA model
7.6 Experimental results obtained with a CDMA PA

The benefit of efficiency control will be demonstrated in this section by means of
measurements obtained with a CDMA PA with a variable supply voltage from a
DC/DC down-converter [15]. The Li-ion battery model from chapter 4 will be used
to estimate the actual talk time improvement realized in a cellular CDMA phone on
the basis of the measured efficiencies. The PA was an EC 2068 CDMA PA from
EiC Corporation and the converter was a TEA1210 DC/DC down-converter
manufactured by Philips. A voltage source was used instead of a battery. For
simplicity, this voltage source will be named battery in the remainder of this section.
7.6.1 Measurement set-up

The measurements have been split into two parts:
• In part 1, no DC/DC converter was used and the PA was connected directly to
the battery and characterized; see section 7.6.2.
• In part 2, a DC/DC down-converter was used to power the PA. The results of
part 1 of the measurements were used to detennine the desired output voltage of
the converter; see section 7.6.3.
The efficiency of the P A was measured in part 1 of the measurements for a range of
battery voltages Vbat from 3.9 V to 2.7 V, which is the applicable voltage range for a
Li-ion battery, and for a range of output powers Pout,RF· Sense lines were used to
ensure the desired voltage at the supply pin of the PA. The set-up for part 1 of the
measurements is illustrated in Figure 7.16.
The RF power meter was calibrated before use. Pout,RF was controlled by means
of the RF input power Pm,RF• supplied by a CDMA signal generator. For each value
of Vbat• P;,RF was chosen so that P out,RF equalled the requested output power.
Whether or not the CDMA specification was met was checked via the values of
ACPR1 and ACPR2 with a spectrum analyzer. ACPR1, which should be at least -44
dBc, was measured at 885 kHz from the centre frequency fc, whereas ACPR2 was
measured at 1.98 MHz from.fc. ACPR2 should be at least -60 dBc. The unit dBc
expresses the suppression with respect to the carrier signal (c). This is an example of
a specification in the frequency domain, as described in section 7.1. The minimum
PA supply voltage Vsup,min at which the ACPR specification could still be met was
determined for each value of P out,RF· The PA efficiency was determined for each
value of V bat larger than v.UP.mim as indicated in the upper left comer of Figure 7 .16.
Note that Vsup,min can be higher than VsUP,opt· This will be explained in section 7 .6.2.
Pout RF
TJpA= - - ' -
Vbat lsup Battery
Voltage
Source
Measured supply
current lsup
COMA
signal generator
with P;n, RF such
that Pout, RF
is required
output power
fc = 836.5 MHz
Figure 7.16: Set-up for part 1 of the measurements: no DC/DC converter
Figure 7.17 shows the set-up for part 2 of the measurements. In these measurements,
the DC/DC down-converter powered the PA with Vsup,min for each value of Pout,RF
found in part 1 of the measurements. Hence, efficiency control was applied. The
battery voltage Vbat was applied at the input of the DC/DC converter using sense
lines. The ACPR specification was checked for each value of Vbat· In addition, the
efficiency 'fltotal of the total solution including the DC/DC down-converter was
determined, as indicated in the upper left comer of Figure 7.17.
7.6.2 Measurement results and discussion of part 1: no DC/DC converter

The results of part 1 of the measurements are summarized in Tables 7.8 and 7.9, in
which the results are compared with calculations with the P A model in section 7 .4.
268 Chapter?
Table 7.8: Results of part 1 of the measurements: PA connected directly to the battery
Vbat Pout.RF = 28 dBm = 0.63 W Pout.RF= 14 dBm= 25 mW
[V]
lsup [rnA] TJPA {%} TJPA {%}: lsup{mA} TJPA {%} TJPA {%}:
model1 model1
3.9 544 29.7 29.6 161 4 5.2
3.5 556 32.4 33 161 4.5 5.9
2.7 - - - 160 5.8 7.6
Note 1: See section 7.4: Parameter values from PA (EC 2068) data sheet: V..m=3.5 V, P..., = 0.63 W,
71m..=33%
Pout RF
Tltotal = --·- Battery
Vbat lsup
Voltage
Source
Measure supply
current lsup
=
Vout Vsup,min
for each requested
Pout, RF ~verter
r-------.,
COMA
signal generator
with P;n RF such
that Pout, RF
is required
output power
fc =836.5 MHz
Figure 7.17: Set-up for part 2 of the measurements: Efficiency control was implemented with
the aid of a DC/DC down-converter
Several conclusions can be drawn from Table 7.8:
• The P Aindeed behaves as a current source, because lsup is almost constant when
vbat changes.
• The PA can no longer supply the requested maximum output power of 28 dBm
at a battery voltage of 2.7 V. This is in accordance with the results of the
simulations discussed in section 7.5.
• The efficiency is exceptionally low at an output power of 14 dBm at all battery
voltages. This is the average output power of a CDMA phone [16].
• The PA model predicts the PA efficiency quite reasonably. The predicted
efficiency is higher than the measured efficiency at 14 dBm. This is because the
CDMA system uses QPSK modulation. Therefore, the PA has to be linear, as
was explained in section 7 .1. The model assumes that the PA can have a voltage
swing of Vsup in the output stage. However, a voltage headroom has to be taken
into account, because the PA has to be linear, which leads to a lower efficiency.
This was discussed in section 7.2.2.
Several conclusions can be drawn from Table 7.9:
• V,up,min is lower at lower output powers, as was to be expected on the basis of

the considerations outlined in section 7 .4.
• The voltage headroom, explained above, is clearly observable in this table, in
which Vsup,min is higher than Vsup,opt to meet the linearity specification of the
considered CDMA system.
• The PA efficiency at Vsup,mim which is the maximum attainable efficiency,
decreases as the output power decreases. This decrease is larger than that
allowed for in the model which was fitted to a GSM PA; see (7.20).
Table 7.9: Measured Vsup,min and calculated Vsup,opt values

PouddBm] V sup,mln [V]: measured V sup,opt [V]: model' llPA [%]at Vsup,mln:
measured
28 3.4 3.5 33.3
24 2.3 2.1 28.8
20 1.6 1.4 23.9
16 1.2 0.9 17.9
14 1.0 0.7 15.4
Note 1: See sectiOn 7.4: Parameter values from PA (EC 2068) data sheet: Ynom=3.5 V, Pmm = 0.63 W,
7]m..=33%
7.6.3 Measurement results and discussion of part 2: with DC/DC converter

The DC/DC converter had a minimum output voltage of 1.3 V due to the design of
the converter. Therefore, a value of 1.3 V was used instead of Vsup,min for the output
powers 16, 14 and 12 dBm, which unfortunately makes fair comparison difficult.
The efficiency of the DC/DC converter was inferred from the measurements as
follows:
:::> TJ _ Tlrotal
TJrotal = TJDCIDC TJ
PA,@V,.,.mm DC IDC - (7.21)
'fJ PA,@ V,u,u,min
where r'JPA• valid at Vsup,min• from the last column of Table 7.9 was used. The
efficiency r'JPA at 1.3 V, instead of Vsup,mim had to be used in the calculations for
output powers of 16 dBm and lower. This efficiency can be easily inferred from
Table 7. 9 by taking the current source behaviour of the PA into account.
Table 7.10 summarizes the results of part 2 of the measurements. The
efficiencies of part 1 have been included in this table to enable comparison. This
enables the calculation of the relative change in efficiency when efficiency control is
applied. The measured DC/DC efficiencies are also compared with those calculated
with the DC/DC converter model of section 7.4.
Several conclusions can be drawn from Table 7.10(a):
• Efficiency control yields a 5.4% improvement in efficiency when Vbar is 3.9 V.

270 Chapter 7
• The PA is no longer able to transmit the requested output power of 28 dBm

according to the CDMA specification at a battery voltage of 3.5 V. This was
still possible in part 1 of the measurements, in which no DC/DC converter was
used; see Table 7.8. This is in agreement with the conclusions presented in
section 7.5. The combination of a DC/DC converter and a PA ceases to meet the
specification at maximum output power at a higher battery voltage than without
a DC/DC converter. This situation can be improved somewhat by using a
bypass switch at maximum output power, as discussed in section 7.5.
• The measured 1JDC!vc is lower than the 1JvCtDc calculated with the model of
section 7.4. This is because the TEA1210 used in the experiments was not
optimized for use as a down-converter. The calculated value should be seen as
an upper limit of the efficiency.
Table 7.10: Comparison of the efficiencies obtained in parts 1 (no DC/DC converter) and 2
(with DC/DC converter) of the measurements demonstrating the effect of efficiency control.
The measured DC/DC converter efficiency TJvcmc is compared with the calculated value
obtained with the simulation model of section 7.4
(a) P••~RF = 28 dBm
vh•• [VJ Pnn.Rl'=28 dBm= 0.63 W
7JPA {%} TJrotal [%] Efficiency 7JDCIDC {%} TJDCIDC {%}
no DC/DC, see (with DC/DC, improvement[%] (measured) (calculated
Table 7.9) eff. control) when eff. control is with the model
applied of section 7.4/
3.9 29.7 31.3 5.4 94 97.8
3.5 32.4 . . - -
2.7 - - - - .
Note 1: See section 7.4: V;.=Vbat, V••.=3.4 V, 1•.,=556 rnA, R,w~,c•=10 m.Q, Rcoi/=40 m.Q, Rby,... = 1 GO (no
bypass switch), c..,=l nF,f...,c.=600kHz [12]
(b) PouJ,RF = 14 dBm

Vbat [V] P.,. t.RP= 14dBm=25mW
TJPA {%} TJtotad%] Efficiency TJDODc [%] TJDODc [%]
no DC/DC, see (with DC/DC, improvement[%] (measured/ (calculated
Table 7.9) eff. control) when eff. control is with the model
applied of section 7.4/
3.9 4 9.9 148 83.4 94.6
3.5 4.5 10.1 124 85.1 95.4
2.7 5.8 10.7 84 90.1 96.7
Note 1: Wilh 7]PA=11.9% at a supply voltage of 1.3 V, mferred from Table 7.9
Note 2: See section 7.4: V,.=Vbat. V•..=1.3 V, 1•.,=162 rnA, R.w~"•=10 m.Q, Rco"=40 m.Q, Rby..,, = 1 GO (no
bypass switch), C..,=1 nF,f...~c•=600 kHz [12]
Several conclusions can be drawn from Table 7.10(b):
• The higher the voltage difference between the input and the output of the
DC/DC converter, the greater the gain in efficiency when efficiency control is
applied. This is because the PA still operates at a reasonable efficiency when
efficiency control is applied, whereas 1JPA is quite low when the PA supply
voltage is larger than Vsup,min· The maximum attainable efficiency is not 15.4%,
as predicted by Table 7.9, but 11.9%, because the PA is operated at 1.3 V,
which is higher than Vsup.min·
• The efficiency increases at all battery voltages because the output voltage of the
DC/DC converter is 1.3 V, which is a lot lower than the battery voltage.
• The efficiency of the DC/DC converter calculated with the aid of the model of
section 7.4 is too optimistic. The same explanation holds as for Table 7.10(a).
Figure 7.18 shows the PA efficiency as a function of Paur,RF at Vba1=3.5 V with and
without efficiency control. The gain in efficiency obtained with efficiency control
increases when Paut,RF decreases. This is in accordance with the results presented in
section 7.5. It no longer holds at output powers of 16 dBm and lower, for which the
'IJPA and 'IJroral curves move closer together again, because the DC/DC converter
powered the PA with 1.3 V in all three cases, which is higher than Vsup,min at these
output powers. Although 'IJraral has a value at 28 dBm in Figure 7.18, the CDMA
specification was in fact not met at this output power with efficiency control at 3.5
v.
_,.
i
~ 20
L
30
./
Tltotal
./
-'
~--- v v
~--= /
- ----
./"""' ,.....~PA
10
v
0
10 14 18 22 26 30
Pout, RF [dBm] -
Figure 7.18: Efficiency without (TJPA• from part 1) and with (TJrotal• from part 2) a DC/DC
down-converter at various output powers Pout,RF• Vba,=3.5V
7.6.4 Estimation of talk time increase in a complete CDMA cellular phone

Only the PA itself has been discussed so far. To predict the increase in talk time in
an actual CDMA cellular phone with efficiency control, some additional information
is needed. This information includes:
1. the power consumption of the rest of the phone;

2. the percentage of time for which the phone transmits at each of the possible
output powers;
3. the battery voltage course during discharging.
The power consumption of the rest of the phone determines the output power up to
which it still makes sense to improve efficiency. For example, when the rest of the
phone consumes 1 W, a decrease in the power consumption of the PA from 2 W to 1
W will make a lot of difference in the phone's total power consumption. However, a
decrease in PA power consumption from 2 mW to 1 mW hardly will influence the
phone's total power consumption. The lower the power consumption of all the
phone parts except the P A, the lower the output power for which efficiency control
still makes sense.
272 Chapter7
The percentage of time for which the phone transmits a certain output power should
be obtained from the network provider. Efficiency control does not lead to much
improvement at maximum output power. Efficiency control is not very beneficial
when the phone transmits at this output power a lot of the time. However, the
general trend is towards transmitting at lower output powers most of the time, as
described in section 7 .1.
A battery discharge curve can be measured, which is time-consuming, or it can
be simulated with the aid of the Li-ion battery model described in chapter 4, which
takes only a limited amount of time. The efficiency improvement figures from Table
7.10 and Figure 7.18 will be translated into talk time improvements in simulations
below. We will first consider the reduction in power consumption of a complete
CDMA phone at a battery voltage of 3.5 V; see Figure 7.19. The power
consumption of the rest of the phone was assumed to be 924 mW, which value was
obtained from a practical CDMA phone. For the occurrence in time of the measured
output powers, a rough estimate was used based on information obtained from the
literature [16]. The average output power was taken to be 14 dBm.
Figure 7.19 illustrates that the total power consumption of the phone can still
be decreased by 23% with efficiency control, even at a very low output power of 12
dBm. This can be understood as follows. Although the output power is only 16 mW,
the PA efficiency both with and without a DC/DC down-converter is lower than 10
%. Therefore, the PA still consumes a significant amount of power and an
improvement in this power consumption still contributes to a decrease in the phone's
total power consumption.
50 50
i
Assumed
occurrence of
Pout,AF
intime [%]
40
30
- 40
30
i
Decrease in power
consumption
of a complete COMA
phone when a DC/DC
down-converter is
20 used to power PA
at optimum supply
voltage [%]
10 10
0 0
12 14 16 20 24 28
Pout,RF [dBm]-
Figure 7.19: Reduction in power consumption of a complete CDMA phone at various output
power levels P out,RF (in [dBm]) and the assumed occurrence in time of these output powers (in
[%]).Battery voltage Vba1=3.5 V
The talk time simulations were performed using the set-up shown in Figure 7.20.
Li-ion
battery model
Figure 7.20: Set-up used for talk time simulations. The battery model was developed in
chapter 4
Current hx is either drawn by the PA when no DC/DC down-converter is present or

is drawn by the combination of a PA and a DC/DC down-converter. Its value was
chosen to be equal to l sup in parts I and 2 of the measurements described earlier in
this section. Table 7.8 showed that Irx is more or less constant at constant output
power when no DC/DC down-converter is present. Table 7 . I0 illustrates that the
total efficiency is more or less constant when a DC/DC down-converter is used for
efficiency control. This means that the P A with DC/DC converter behaves as a
constant power source with a value dependent on the PA's output power. The values
of lrx at different output powers are summarized in Table 7.11. The PA output power
is divided by the total efficiency of Figure 7.18 in the situation with a DC/DC
converter. The value used for Prest was 924 mW, as in Figure 7.19.
Table 7.11: lrx for various P0 u1 RF values with and without a DC/DC down-converter
ITx Pout,RF = Pout,RF= Pout,RF= Pout,RF= Pout,RF= Pout,RF=
value 28dBm 24dBm 20dBm 16dBm 14dBm 12dBm
With p out ,_I_ p out , _1_ pout · - ' - pout · - ' -
P.ut ,_1_ P.ul ,_l_
DC/DC
0.31 v ba, 0.26 vbal 0.21 vbat 0.14 vbat 0.10 V bat 0.07 v bal
Without 550mA 365mA 251 rnA 183 rnA 161 rnA 144mA
DC/DC
The occurrence percentage in time of P out.RF was taken from Figure 7.19, which led
to the sequence for P out,RF in time illustrated in Figure 7.21. This was repeated every
100 s during the complete discharge of the battery.
According to the CDMA specification the combination of a DC/DC down-
converter and a PA can no longer transmit the requested 28 dBm output power at
battery voltages of 3.5 V and lower. The same holds for a separate PA at battery
voltages of 3.4 V and lower. This effect was ignored in the simulations for
simplicity. Hence, both with and without the DC/DC down-converter, the
simulations proceeded from a full battery until the battery voltage dropped below
2.7V.
The simulated increase in talk time with efficiency control was 24%. Besides
on the assumed Prest value, time distribution of output powers and a simulated
battery discharge curve, this figure is based on measurement results obtained with a
CDMA PA. Another assumption that has not yet been addressed involves the power
control protocol. It was assumed that the supply voltage of the PA immediately had
the right value as soon as an output power change occurred. In practice, the time that
is available for this supply voltage change will strongly depend on the power control
protocol. An example of including efficiency control in a complete GSM phone will
274 Chapter 7
be given in the next section. Information on the power control protocol is needed for
this.
i 28 1 - ' I ,_!
'E 24
Ill
~
E. .._
u.
20
a:
s0
16
a..
12
10 15 20 40 80 100 110
t[sec] -
Figure 7.21: Sequence for Pow,RF (in [dBm]) as a function of time (t [sec]) used in the talk
time simulations
7.7 Application of efficiency control in a GSM cellular phone

The main aim of this section is to show how efficiency control can be implemented
in an actual cellular phone, such as the GSM Spark phone manufactured by Philips
[11]. First, some background information will be given on the GSM power control
protocol. Then, the modifications of the phone needed to include efficiency control
will be described. Finally, measurement results will be discussed and compared with
results without efficiency control. The measurement results will also be compared
with the results of calculations using the simulation models of section 7 .4.
7.7.1 GSM power control protocol

Several classes exist in the GSM system. Each class is determined by, amongst other
factors, a maximum power and a power control range. This range is from 5 dBm to
33 dBm at the antenna in the case of the Spark phone. The output power of the P A
varies between 7dBm and 35 dBm, with an insertion loss of 2 dB between the PA
and the antenna.
The base station controls the output power of the handset in steps of 2 dB. It
communicates the required output power to the handset. Communication between a
GSM handset and a base station is organized in so-called reporting periods, each
reporting period consisting of 104 TDMA frames and lasting 480 ms [17]. Each
TDMA frame lasts 4.616 ms and consists of 8 slots of 577 llS each. The structure of
a reporting period is shown in Figure 7.22.
Each frame accommodates four Tx/Rx (Transmit/Receive) pulse combinations
(the Tx and the Rx pulse each fit into a slot) for communication with four handsets,
each Tx!Rx pair being separated by three slots. One pair is shown in Figure 7.22 as
an illustration. The frame number (between 0 and 103), at which a reporting period
starts, will depend on the TCH (Traffic CHannel) number, which may have eight
values (TCHO .. TCH7). For example, a reporting period lasts from frame 0 to 103 in
the case ofTCHO and from frame 13 to frame 12 in the next group of 104 frames in
the case ofTCHl, etc.
The desired output power level is communicated in the Rx slots of the so-
called SACCH frames, four of which are present in each reportingperiod. The frame
numbers for the SACCH frames depend on the TCH number. For example, in the
case of TCHO, the SACCH frames are located at frame numbers 12, 38, 64 and 90.
All four SACCH frames have to be received to enable successful decoding of the
information from the base station concerning the desired output power in the next
reporting period. Hence, decoding this information can commence when TDMA
frame 90 has been received in the case of TCHO. The change to the new output
power in the next reporting period takes place at a rate of 2 dB every 60 rns (13
frames). The change commences in the first frame of the next reporting period. An
example of the timing of a 4 dB increase in output power in a TCHO channel is
given in Figure 7.23.
Decoding the information in the four received SACCH frames takes four
frames. Therefore, it becomes available in frame 95 in TCHO, which is 37 rns before
the end of the reporting period. The actual change in output power starts in frame
number 0 of the next reporting period. Therefore, the time between the Tx pulse in
frame 103 and the Tx pulse in frame 0 in the next reporting period is available to
change the output power by the first 2 dB. This time amounts to 4 ms.
- - j - TCH1 --j-TCH2
0 12 25
I I I lsi I I I I I I I
4.616 ms - - j - TCH3 --j-TCH4
26 38 51
- - j - TCH5 --j-TCH6
52 64 77
__,.- TCH7 --r-TCHO
78 90 103
I I I !!I I lsi I I I I I I I
Txn············· ·-------------] S: SACCH frames for TCHO
.~xn
-
!
577 !.IS
Figure 7.22: Structure of a reporting period in the GSM system

276 Chapter?
Reporting period n --t- Reporting period n +1

90 95 103 0 1213
:::Jsl I I I
t
Information on
I I I I/ ) l..,\ 1\.\I I I 2IdBIchange
I I in I_/I__./I_)/ / ~J \.·. ~ · .:~·
output power level 2 dB change in \
output power level \
\
\
output pow~_•·••.ep--~
1eve1 ~/
in reporting period
0~~ ~\\\
0
n + 1 becomes
available 0
Tx Rx Tx Rx Tx Rx Tx Rx
Figure 7.23: Example of a 4 dB output power increase in a TCHO channel
Output power changes take place in small steps and relatively slowly in time,
according to the GSM protocol. The DC/DC converter for efficiency control has to
be able to change its output voltage within 4 ms. This needs to be done once every
60 ms at the most.
7.7.2 Modifications in the Spark GSM phone

The PA is usually connected directly to the battery in a Spark phone. The battery
consists of four NiMH cells connected in series and has a nominal supply voltage of
4.8 V. The PA in the Spark phone is a BGY204 from Philips. This PA has a
maximum efficiency 7Jmax of 55%, a nominal supply voltage Vnom of 4.8 V and a
maximum output power Pout,max of 35 dBm The PA connections in the Spark phone
and the organization of power control are depicted in Figure 7.24. This figure is a
more detailed extension of Figure 7.1. Impedance matching and filtering have been
omitted for simplicity. Instead, an antenna switch for changing from Rx to Tx and
vice versa is shown.
Determine
desired pout
andTx info
Decode
i
received
information Rx info
System RadioCoDec (analogue)
controller (AD/DA)
(digital)
Figure 7.24: PA connections and power control organization in a Spark phone

The PA is connected to an RF input (RF in), which derives from the system
controller through a Digital-to-Analogue Converter (DAC) and an RF Tx block,
including a mixer. The PA amplifies this signal to a signal RF out, which is
transmitted at the antenna. The PA has a power control pin, which is connected to a
power controller. Power control is implemented as shown in Figure 7.4(a), because
the BGY204 is a bipolar PA.
Data is received at the antenna during the Rx slots. The Rx info is digitized by
an Analogue-to-Digital Converter (ADC) after being down-mixed. The digital
information is interpreted in the system controller. The data that has to be
transmitted in the next Tx slot is determined. Information on a new output power is
received at certain designated times, as explained in section 7.7.1. When needed, the
system controller programs the power controller with a new value immediately after
the last transmit burst with the present output power.
The power controller provides the proper control voltage to the PA, starting
immediately before the actual burst starts. It ramps up the power according to the
predefined power-versus-time template; see section 7.1. The ramping time is 20 j.LS
in the case of GSM. The power controller maintains the output power dictated by the
base station after the ramping with a tolerance of ±1 dB. The signal RF in is
amplified to the antenna during the burst. At the end of the transmit burst, the power
controller ramps down the output power of the PA. Power control is performed in
closed-loop mode, as discussed in section 7 .1.
A DC/DC down-converter was included between the battery and the PA supply
pin as a modification to the existing Spark phone. Existing hardware was used in the
experiment. A TEA1204 DC/DC down-converter from Philips was used, whose
output voltage was made controllable by adding some external hardware which
consumed negligible power. Three possible output voltages could be generated,
depending on the control voltage Vcon applied to the external hardware:
1. the battery voltage could be passed through directly. Switch S 1 in Figure 7.9b
conducted continuously in this situation;
2. a fixed output voltage of 3. 7 V could be generated;
3. a fixed output voltage of 3.4 V could be generated.
The values of Vsup,opt can be calculated from the PA parameters Vnom and P max for
each PA output power using (7.7). Table 7.12 shows the values of Vsup.opt for the
output powers at which the measurements were performed. The voltage applied with
the DC/DC converter is shown in the fourth column. A PA supply voltage higher
than Vsup,opt was used at output powers of 29 dBm and lower, because only three
voltages are possible due to restrictions of the employed hardware. Hence, no full
benefit of efficiency control could be obtained in the experiments.
Table 7.12: Optimum supply voltages Vsup,opt according to (7.7) and actual supply voltages
obtained from the DC/DC down-converter for various output powers of the P A and at the
antenna
Pout at antenna PoutatPA Vsup,opt according Actual supply voltage using
[dBm] [dBm] to (7.7) [V] DC/DC down-converter [V]
33 35 4.8 4.8
31 33 3.8 3.7
29 31 3 3.4
27 29 2.4 3.4
25 27 1.9 3.4
23 25 1.5 3.4
278 Chapter 7
The PA connections and power control scheme in the modified Spark phone with
efficiency control are illustrated in Figure 7.25. The original power control loop was
left intact. The supply voltages of all the other system blocks, except the PA, were
not altered.
The control voltage for the DC/DC converter was supplied by a spare DAC in
the Radio Coder-Decoder (CoDec). The software in the Spark phone was expanded
slightly with a simple table, which translated the requested output power at the
antenna into a control voltage Vcon [18]. The system controller programmed the right
Vcon value in the spare DAC when the requested output power value was
programmed into a register of the power controller. Hence, when needed, the
DC/DC converter changed from one output voltage to another, immediately after the
Tx burst with the previous output power. The employed DC/DC converter was not
able to draw current from the output capacitor to make a step down in output
voltage. Therefore, the PA drew current from the buffer capacitor at the PA supply
pin in the first Tx bursts with a new output power. The voltage was kept at that value
by the DC/DC converter as soon as the voltage had decreased to the voltage at
which the DC/DC converter was installed.
Vcon
Determine
desired pout
Tx info and
Vcon from
table
t
Decode
received
information Rx info
System RadioCoDec (analogue)
controller (AD/DA)
(digital)
Figure 7.25: PA connections and power control scheme in modified Spark phone with
efficiency control
Besides adding a DC/DC converter and adding a table to the software, an additional
buffer capacitor was placed at the supply pin of the PA. A total capacitor value of 1
mF was obtained by adding capacitors to the 100 J.lF already present in the original
phone. The reason for this has already been explained in section 7.5. The bursts
(GSM is a TDMA system) should be drawn from a buffer capacitor with a low ESR
value. The value of 1 mF was found to be a good trade-off between efficiency and
capacitance value.
7.7.3 Measurement results and discussion

The original Spark phone shown in Figure 7.24 was compared with the modified
Spark phone of Figure 7.25 in the experiments. A 4.8 V voltage source was used in
both cases, instead of a battery. This voltage will be called Vbat in the remainder of
this section. The phone was operated at different output powers, which were
checked with a spectrum analyzer. The current consumption figures of the complete
phone can be compared. Table 7.13 shows the measurement results. The phone was
operated at channel 62, which is in the middle of the GSM Tx frequency band; see
section 7 .1.
Table 7.13 shows that the power consumption of the modified phone is 1.5%
higher than that of the standard phone at maximum output power. This is because
the DC/DC converter acts as a 'pass-through' for the battery voltage in the modified
phone at maximum output power. Hence, an extra series resistance of S1 and the
ESR of the coil are present in the supply path; see Figure 7.9(b). The PA current
consumption can be calculated at maximum output power with a supply voltage of
4.8 V and an efficiency of 55%, valid at 4.8 V. The difference when the values of
P101a1 for the standard and modified phone are compared at maximum output power is
attributable to an extra series resistance of 133 mQ in the modified phone. The same
currents flow into the PA in both cases. This value is in good agreement with the
real value of 70 mQ obtained for the switch resistance connected in series with 50
mQ for the ESR of the coil. The series resistance can be lowered by applying a low-
ohmic bypass switch, as described in section 7.5. The values of P101at given in Table
7.13 are average powers, which were obtained by dividing the power transmitted in
the burst by eight.
Table 7.13: Power consumption of complete phone measured in a standard Spark phone and
a modified Spark phone with efficiency control
Pout at Pout at Standard Spark Modified Spark (eff. control)
antenna PA Vba, VpA Ibat Ptota! 1 Vbat VpA Ibat Ptota!
[dBm] [dBm] [V] [V] [rnA] [mW] [V] [V] [rnA] [mW]
33 35 4.8 4.8 325 1560 4.8 4.8 330 1584
31 33 4.8 4.8 286 1373 4.8 3.7 272 1306
29 31 4.8 4.8 261 1253 4.8 3.4 240 1152
27 29 4.8 4.8 242 1162 4.8 3.4 224 1075
25 27 4.8 4.8 228 1094 4.8 3.4 214 1027
23 25 4.8 4.8 214 1027 4.8 3.4 207 994
Pout at antenna [dBm] Cllan2e in Ptntol for modified Spark [%]

33 +1.5
31 -5.1
29 -8.8
27 -8.1
25 -6.5
23 -3.3
The decrease in power consumption as a function of the output power of the

modified phone is in agreement with Figure 7.19 for a CDMA phone. Hence,
efficiency control has the same effect on the power consumption of the complete
phone, irrespective of the type of cellular system. No effect, or even a negative
effect, as can be inferred from the 1.5% increase in total power consumption in
280 Chapter 7
Table 7.13, was observed at maximum output power. The decrease in power
consumption of the complete phone first becomes greater when the output power
decreases. This is because the difference between the battery voltage and the
optimum PA voltage Vsup.opt increases. Hence, more can be gained from efficiency
control. However, the benefit of efficiency control becomes less when the output
power decreases further, because the power consumption of the rest of the phone
remains constant. This is related to the ratio of power savings in the PA and the total
power consumed by the phone. The decreases in power consumption were less in
this experiment than in the experiments with the CDMA P A described in the
previous section, because the Vsup of the P A in the GSM phone was in most cases
higher than V,up,opt• unlike in the CDMA PA experiment. Therefore, the full benefit
of efficiency control was not realized in the experiments with the GSM phone.
The results shown in Table 7.13 can be verified with the aid of the PA model
of section 7.4. The PA power consumption can be calculated with the PA model for
both the standard and the modified phone and for each output power. A power
consumption of 842 mW is found for the rest of the phone when this calculated PA
power consumption is combined with the measured P 1orat value for the standard
phone at maximum output power; see Table 7.13. This corresponds to a current of
175 rnA at 4.8 V, which is in agreement with the results of the measurements. The
changes in power consumption measured for the modified phone (see Table 7.13)
can now be compared with the results of calculations in which the power
consumption of the rest of the phone is assumed to be the same at all output powers.
Moreover, the change in power consumption in the case of ideal efficiency control
can be calculated. In that case, the DC/DC converter supplies the PA with Vsup,opt at
all output powers. The results of this exercise are shown in Table 7.14. A fixed
efficiency of 90% was assumed for the DC/DC converter in all calculations. The PA
duty cycle OpA was chosen to be 118, in accordance with the GSM system for which
the Spark phone was built.
Table 7.14: Change in power consumption for a complete Spark phone relative to that of a
standard Spark phone obtained in (a) measurements (see Table 7.13), (b) calculations based
on the measurement results of Table 7.13 and the PA model of section 7.4, and (c)
calculations of ideal application of efficiency control on the basis of the measurement results
of Table 7.13 and the P A model of section 7.4
Pout at Pout at (a) Change in Ptotal (b) Change in Ptotal (c) Change in Ptotal
antenna PA measured for calculated from (a) calculated for ideal
[dBm] [dBm] modified Spark with the P A model efficiency control
[%] [%] [%]
33 35 +1.5 +1.5 +1.5
31 33 -5.1 -5 -5
29 31 -8.8 -7.7 -11
27 29 -8.1 -6.7 -13.9
25 27 -6.5 -5.7 -15
23 25 -3.3 -4.8 -14.8
The calculated changes in column (b) of Table 7.14 are in good agreement with the
measured changes in column (a). Hence, the benefits of efficiency control are again
predictable with the models derived earlier in this chapter. The power consumption
can moreover be decreased substantially by optimizing the DC/DC converter design
and making it suitable for more output voltages. This is especially true at lower
output powers, as can be seen in column (c). A similar exercise should be performed
for real talk time calculations; see section 7.6. This exercise will not be repeated
here.
Specific demands in both the frequency and the time domain exist for the signal
transmitted at the antenna. For example, switching transient tests were performed
using the modified Spark phone. The spectrum of the complete transmit burst was
measured with a spectrum analyzer and had to be within a spectral mask in these
tests; see section 7 .1. The only violation occurred at an output power of 31 dBm and
in channel 124, which is in the far right of the Tx frequency band. This violation was
caused by PA saturation because the supply voltage of the PA was slightly lower
than Ysup,opt of 3.8 V. The supply voltage at which saturation occurs depends on the
frequency of the transmitted signal. Moreover, the insertion loss between PA and
antenna is frequency-dependent and the losses in these passive circuits are higher at
the edge of the frequency band. Therefore, Ysup,opt should also be made frequency-
dependent. The violation disappeared when the PA supply voltage was increased a
little.
Besides switching transient tests, many other tests have to be performed to
determine the quality of the modulated signal at the antenna. All the necessary tests
with the same P A and DC/DC converter as used in the experiments showed that the
voltage output ripple of the DC/DC converter does not lead to violation of any
specification [19]. The main reason for this is that the PA has a good power supply
rejection ratio due to the high-bandwidth power control loop; see Figures 7.24 and
7.25. Placing buffer capacitors at the PA supply pin with a low ESR still makes
sense, notably to reduce the ripple, but also to increase the system efficiency.

The behaviour of the PA and the passive circuits between the PA and the antenna
have to be characterized in more detail, for example as a function of frequency and
temperature, in order to design a cellular phone with efficiency control. Then, an
optimum PA supply voltage can be defined for a range of conditions, including
output power, frequency and temperature. All frequency and time domain
specifications should be met under these conditions. The output power and
frequency are known quantities, while the temperature can be measured. The table
included in the base-band part now becomes somewhat more complicated. However,
the basic method of including efficiency control is still the same as described in this
section.
7.8 Conclusions
Several trends in today's cellular telephony systems lead to a need for more efficient
PAs. Simple calculations show that an optimum PA supply voltage Vsup,opt exists for
each output power. With efficiency control, the PAis powered with Ysup,opt at each
output power, irrespective of the battery voltage.
The brief description of DC/DC conversion principles given above has shown
that inductive converters offer the best opportunity of delivering powers required by
a PAin an efficient way. Any input battery voltage can be controlled to any output
voltage value, irrespective of the load. Therefore, inductive DC/DC converters are
the best candidates for implementing efficiency control in a cellular phone. The
output voltage of the DC/DC converter is then controlled to Ysup,opt at each output
power.
282 Chapter?
Simulations and measurements have shown that efficiency control yields a

significant reduction in PA power consumption, mainly at output powers lower than
maximum. The PA power decreases only slightly at maximum output power, or
even increases slightly, depending on the difference between the battery voltage and
the PA supply voltage. The use of a bypass switch helps to decrease the battery
voltage at which the PA can still supply maximum output power. The bypass switch
does not yield any improvement at output powers lower than maximum In general,
the benefits of efficiency control are greater in the case of greater differences
between the battery voltage and V,up,opt· The results obtained with the PA simulation
model in particular were found to be in good agreement with measurements.
Besides the realized reduction in PA power consumption, the actual benefit of
efficiency control in a cellular phone depends on three other factors. First of all on
the power consumption of the rest of the phone, secondly on the PA output power
distribution in time and thirdly on the battery discharge curve. The power
distribution in time is determined by the cellular system The results presented in
this chapter show that the battery models of chapter 4 can be used together with
simulation models for a PA and a DC/DC converter to predict talk time
improvements realizable with efficiency control in an actual cellular phone. These
improvements can also be translated into battery size reductions when talk time
improvement is not really an issue for a phone manufacturer. This leads to a smaller
and lighter phone with the same talk time. Simulations with the battery model enable
a system designer to quickly obtain an idea of what a battery management function,
such as efficiency control, can offer.
Outlook
The implementation of efficiency control in a GSM phone shows that only minimal
effort is required. A DCIDC converter has to be added between the battery and the
PA. This DC/DC converter is already present in many cellular phones. The output
voltage of especially Philips' inductive DC/DC converters, which have a digital
controller, can easily be made controllable. The control signal can even be obtained
directly from base-band in a digital format in a dedicated design. A table should be
implemented in software to translate requested output powers into corresponding
control voltages. The inclusion of frequency and temperature information in this
table should be considered because the value of V,up,opt at which the output signal of
the PA still meets all the cellular system specifications depends on the frequency
and temperature. This was not considered in the simulations and measurements
discussed in this chapter, because most measurements were performed at room
temperature and in the middle of the cellular system frequency band. More PA
measurements should be performed to determine the influence of the temperature
and frequency on Vsup,opt· However, the basic principle of implementing efficiency
control in a cellular phone remains the same as described in this chapter.
7.9 References
[1] T.S. Rappaport, Wireless Communications-Principles & Practice, Prentice-Hall, Upper
Saddle River, New Jersey, USA, 1996
[2] J.D. Gibson (Ed.), The Mobile Communications Network, CRC press 8573, IEEE press
PC5653, 1996
[3] Digital Modulation in Communication Systems- An Introduction, Application Note
1298, Hewlett Packard, 1997
[4] The name efficiency control originates from Ralf Burdenski
[5] T.H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits, Cambridge
University Press, Cambridge, United Kingdom, 1998
[6] S.C. Cripps, RF Power Amplifiers for Wireless Communications, Artech House,
Norwood MA, USA, 1999
[7] J. Smith, Modem Communication Circuits, McGraw-Hill Series in Electrical
Engineering, Electronics and Electronic Circuits, New York, USA, 1986
[8] F. Goodenough, "Low-Dropout regulators Get Application Specific", Electronic
Design, pp. 65-77, May 13, 1996
[9] B. Kerridge, "Mobile Phones Put the Squeeze on Battery Power", EDN, pp. 141-150,
April10, 1997
[10] L. Sherman, "Know the Peculiarities of Portable Power Supply Designs", Electronic
Design, pp. 94-104, July 7, 1997
[11] H.J. Bergveld, Increase of Efficiency of GSM Power Amplifiers by Varying the Supply
Voltage: Measurement Results of a Demonstrator Built with Spark, Nat.Lab. Technical
Note 093/97, Philips Internal Report, April1997
[12] Data sheets TEA12xx, Philips Semiconductors
[13] Data sheets from various coils, Coilcraft
[14] F.J. Sluijs, C.M. Hart, D.W.G. Groeneveld, S. Haag, "Integrated DC/DC Converter with
Digital Controller", International Symposium on Low Power Electronics and Design,
Monterey, USA, pp. 88-90, August 10-12, 1998
[15] H.J. Bergveld, A. van Bezooijen, A.J. Hoogstraate, Efficiency Improvement of a CDMA
PA Using a DC/DC Down-converter: Measurement Results, Nat.Lab. Technical Note
185/99, Philips Internal Report, June 1999
[16] P. Asbeck, J. Mink, T. Itoh, G. Haddad, "Device and Circuit Approaches for Next-
Generation Wireless Communications", Technical Feature, Microwave Journal,
February 1999, Figure 8
[17] Final draft pr ETS 300 578:1994 (GSM 05.08 version 4.11.0)
[18] H.J. Bergveld, F.A.C.M. Schoofs, "A Communication System, Device and Method",
Patent US6298222, W09931798, EP0960474, filed 12 July 1998
[19] M. Simon, RF-Amplifier + DC/DC-Converter: Test Report, ABZ05-PA-0001, Philips
Internal Report, Philips Technology Centre for Mobile Communication, January 1996
285
Chapter 8
General conclusions
The subject of Battery Management Systems (BMS) is very broad and involves
many disciplines. The main focus in this book has been on simulations of BMS. As
stated in chapter 1, simulations are a helpful means for obtaining an understanding
of the behaviour of complex systems under a wide variety of conditions.
Measurements of battery behaviour generally take a lot of time. With battery
simulation models, simulations of battery behaviour in a portable device under
development can be performed swiftly. The increased understanding of this
behaviour can be applied to improve the BMS functionality. So far, no battery
models were available for this purpose. Therefore, battery models with which
improvements in BMS designs can indeed be realized have been described in this
book.
The simulated behaviour should always be checked by comparing it with
measurements. A great advantage in this respect is however that many situations can
already be simulated in a relatively short time and the best possible situation can be
selected before the start of the measurements. This can serve to give direction to the
measurements, which significantly shortens the amount of time involved in the
experiments. For example, an optimum charging algorithm can be selected in
simulations. This algorithm can then be investigated in more detail by means of
measurements. Tedious testing of different possibilities by means of measurements
is prevented in this way.
The application of physical system dynamics to battery modelling is new. It
turns out to be an elegant approach, in which simulation models of different battery
types can be constructed on the basis of a single general set of building blocks. Once
a mathematical description of the reactions and processes inside the battery has been
derived, a model can be built for each type of battery 'brick by brick'. As a result,
the models are based on physical and electrochemical theory and the various
processes can easily be recognized in the model. The model serves as a 'transparent
battery', in which the courses of the various reactions can be visualized. Because of
the applied approach, a battery is represented by an equivalent network model. This
enables simulation in the conventional circuit simulators commonly used by
designers of portable devices. The battery models can therefore be simulated in both
thermal and electrical interaction with their surroundings in the portable device. The
best example of this was given in chapter 5, in which the NiCd model interacted
both electrically and thermally with the surrounding system parts in a shaver.
The various examples given in this book illustrate that the desired complexity
of a battery model depends largely on the application. The entire model was needed
in the shaver example in chapter 5 because the focus not only was on battery
voltage, but also on the combined effect of voltage (V), current (/), internal gas
pressure (P) and battery temperature <n in interaction with the other shaver parts.
However, although possible, the model was not used as a 'transparent battery' in this
example, because a shaver designer is not interested in what goes on inside the
battery, but only in the SoC build-up and temperature development with the
different charging algorithms.
Another example, in which a complete network model is definitely needed, is
the charging simulation of Li-ion batteries in chapter 5. This example clearly shows

286 Chapter 8
that the developed models can readily be used by battery manufacturers to optimize
the design of their batteries. Without the need to develop a wide range of prototypes
and perform many time-consuming experiments, insight into the influence of certain
design parameters on battery behaviour can be gained very quickly. Moreover, it is
very easy to plot the simulated individual electrode potentials. In practice,
measurement of these electrode potentials requires the insertion of a reference
electrode inside the battery, which is not trivial. Therefore, the 'transparency' of the
battery model, in the form of the easy accessibility of all the equilibrium potentials
and overpotentials, is an advantage here.
In some cases the entire battery model will have to be used to gain insight in
battery behaviour, after which the behaviour can be described in a simpler form.
This was illustrated in chapter 6. The Electro-Motive Force (EMF) method was
developed on the basis of the results of simulations and measurements. The
overpotential description needed for 'time left' predictions was derived in
simulations using the entire model. The resulting simpler battery description was
used in a State-of-Charge (SoC) system. However, such a simple description can
also be used in simulations in which a designer is only interested in battery voltage
and does not care at all about what goes on inside the battery. A good example of
such simulations is the talk-time simulation performed in chapter 7. In this case, a
battery model was only needed to generate a discharge curve under various load
conditions. A model based on an EMF curve and the overpotential description of
chapter 6 will suffice in this case.
A good quantitative agreement between the results of simulations and
measurements is important for increasing the usefulness of the models. It has been
shown that good quantitative agreement can be obtained in a charging experiment at
a 1 C-rate in the NiCd model with the aid of a dedicated mathematical method. This
good agreement was obtained for the V, P and T curves simultaneously with a single
parameter set. This inspires confidence in the applied modelling approach.
Moreover, the outcome of the comparison of parameter values obtained with the
mathematical method and available measured parameter values is promising,
because most of the optimized parameter values were close to the measured values.
However, parameter optimization alone is not enough, because quantitative
optimization is an iterative process that also involves model modifications. With the
model modification described in chapter 4 the quantitative agreement was improved
for various charge currents simultaneously, although there was still room for
improvement. However, the observed improvement was in the right direction, which
again inspires confidence in the quality of the model. Various comparisons of
simulation and measurement results obtained with the NiCd model discussed in
other parts of this book have shown that usable predictions can already be made with
the present NiCd model and parameter set. Examples are the self-discharge
simulation discussed in chapter 4 and the charge simulations in chapter 5. Good
quantitative agreement of charge and discharge curves simulated and measured
under various conditions has also been demonstrated for the Li-ion model. However,
temperature dependence will still have to be included in this model. Moreover, side
reactions have not been implemented yet.
A large part of this book was devoted to the construction of the models
themselves. The most important BMS improvements that have been realized with
the constructed models are listed below:
General conclusions 287
• A cause of and solution for breakdown of cells in battery packs have been
identified. The model enables simulations more complex than those presented in
this book, in which the thermal behaviour of cells at different positions in the
pack can also be taken into account.
• A new charging algoritlun denoted as thermostatic charging has been found in
simulations. The simulation results have been verified with measurements. This
charging algoritlun is very suitable for use in small portable devices with built-
in chargers, because such devices often suffer from high internal temperatures
during charging. The battery performance will degrade faster when the battery
is exposed to high temperatures many times. High temperatures are prevented
with thermostatic charging, whereas the same charged capacity can still be
achieved in the same amount of time as in a standard charging regime with a
constant current.
• A possible solution for capacity loss in Li-ion batteries during fast charging has
been identified. Experiments will have to be performed to check whether the
proposed solution indeed has the desired effect. If those experiments confirm
the simulation results, fast charging with less capacity loss will become possible
for Li-ion batteries after the capacity of the negative electrode has been
increased by 10%.
• A new SoC indication method has been developed on the basis of the results of
simulations. The results are promising, although many more experiments and
improvements are still needed. The accuracy of the predicted remaining time of
use is better than 5% for discharge rates of 1 C and lower. The predictions for
large discharge currents at low temperatures are more reliable than those
obtained with the existing bq2050 book-keeping system. However, the
predictions obtained with the bq2050 at moderate currents and at room
temperature are better than those obtained with the proposed system. Some
suggestions for improvement of the proposed system were given in chapter 6.
• The battery models have been used as easy means for predicting talk time
improvements realizable with efficiency control, together with the simulation
models of a DC/DC converter and Power Amplifier (PA). A talk time
improvement of 24% has been predicted for a Code-Division Multiple Access
(CDMA) phone. This prediction is based partly on measurement results
obtained with a CDMA PA. Measurements with a complete Global System for
Mobile communication (GSM) phone have shown that reductions in phone
power consumptions of up to 10% can be realized with efficiency control. This
reduced phone power can also be translated into a longer talk time.
289
About the authors

Dr. Henk Jan Bergveld was born in Enschede, The Netherlands,
in 1970. He studied electrical engineering at the University of
Twente in Enschede from which he obtained his M. Sc. degree
(cum laude) in 1994. He joined the Philips Research
Laboratories in Eindhoven, The Netherlands, in 1994. His
research interest was the design of Battery Management Systems
and the modelling of rechargeable batteries to improve these
designs. He worked on battery modelling with the co-authors of
this book. He also worked with various product groups within
the Philips organization to help improve the battery management functions in e.g.
shavers and cellular phones. The results of these research activities resulted in his
PhD degree, which he obtained (cum laude) from the University of Twente in 2001.
He is currently a senior scientist in the Mixed-signal Circuits and Systems group
focussing on the integration of Radio-Frequency transceivers in CMOS technology.
Dr. Wanda S. Kruijt was born in Badanah, Saudi-Arabia, in

1963. She studied Chemical Engineering in Amersfoort, The
Netherlands, and subsequently Physical Chemistry at the
University of Utrecht, The Netherlands, from which she obtained
her M. Sc. Degree in 1989. She then started a research study on
the initial stages of electrochemical phase formation. This work
resulted in her PhD thesis entitled "Light Scattering from a
Nucleating Surface" and her PhD degree from the University of
Utrecht in 1993. She then joined the Philips Research
Laboratories, where she continued her electrochemical specialism in the study and
modelling of physical and (electro)chemical processes in rechargeable batteries.
During this time she worked closely with both other authors of this book. She joined
the Business Group Luminaires of Philips Lighting in 1998. She is currently
working as a project manager at LumiTec, the pre-development group of BG
Luminaires. She is responsible for the development and industrialization of large
area LCD backlights.
Prof. dr. Peter H.L. Notten was born in Eindhoven, The

Netherlands, in 1952. He was educated in analytical chemistry
and joined the Philips Research Laboratories in 1975. While
working at these laboratories on the electrochemistry of the
etching of Ill-Y semiconductors he received his PhD degree
from the Eindhoven University of Technology in 1989. Since
then his activities have been focussing on the research of
hydride-forming electrode materials for application in
rechargeable NiMH batteries and as switchable optical mirrors.
At present he is responsible for the research programme "Portable Energy" within
the Philips Research Laboratories. Since 2000 he has been appointed as part-time
professor at the Eindhoven University of Technology where he is heading the group
"Electrochemical Energy Storage". His main interests include modelling of various
rechargeable battery systems and battery materials research.
291
Index
A
AB 2, AB 5 alloys 37,38 Battery systems
Access method 242 NiCd
ACPR 244,267 Oxygen reaction 97-103
Activation energy 102 Thermodynamics 97
Activity of a species 44,45 Kinetics 99
Activity coefficient 45,68 Diffusion limitation
Adaptive control 202 100
Air-management system 42,48 NiMH 37,38
Anode 34 Li-ion 37-39
Antenna impedance 247,251 LiCo02 electrode reaction
Arrhenius relation 102 108-113
A vrami' s theorem 95 Thermodynamics
108
B Kinetics 111
Base-band 242-244 Diffusion limitation
Base station 241,274 112
Battery 33,43 LiC6 electrode reaction
Aging effects 163,200 113-116
Battery lifetime 10,17 Thermodynamics
Battery Management System 3 114
Battery pack 14,33,154,155,170 Kinetics 116
Battery voltage 52,122,196 Diffusion limitation
Battery impedance 116
156,197-198,222 Li -ion-polymer 37,40
Capacity loss 200 Li-metal 40
Cell 33,154,155,170 Rechargeable alkaline 37,42
Cylindrical battery 35 SLA 37,41
Primary battery 33 VLRA 41
Prismatic battery 36 Zinc-air 42
Secondary battery 33 ZnMn~ 31
Storage effects 200 BER 244
Battery systems Brush, Charles 32
NiCd 36,37,86-103 Butler-Volmer equation
Nickel reaction 89-92 46,63-65,137
Thermodynamics 89
Kinetics 90 c
Diffusion limitation C-rate 34
91 Capacitive voltage converter 253-255
Cadmium reaction 92-97 Cathode 34
Thermodynamics 93 CDMA 242,243,266-273
Kinetics 93 Cellular
Diffusion limitation Cell 241,245
95 Handset 1,241
Phone 5,9,25-29,229,241
System 241,243
Channel 241
Charge neutrality 77,80,81
292 Index
Charge pump 253 Dendrite 41,186

Charge-transfer reaction 43,44 Depolarizer 88
Charger 1,2 Diffusion 48,72,117
Charger features 174 Diffusion layer thickness 48,97
Travel charger 11 Diffusion length 161
DTC 11-14,26-28 Stationary diffusion 49,97,100
Charge-balancing circuit 17 Digital modulation 241,242
Charge controller IC 12 Non-constant envelope 245,251
Charging Discharging 'efficiency' 200
Charging algorithm 3,5,11 Displacement variable 57
Li-ion 156-159,184-191 Dissolution-precipitation mechanism
NiCd 169-184 92
NiMH 169-177 Double-layer capacitance
Charging control 11-14 51,64,79,134
Charging efficiency Down-link 241
25,55,150,151,179,199 Duplex method 242
Charging mode 170
Charging priority 11 E
Equalization charging 170 EDGE 244,245
Fast charging Edison, Thomas 32
170,181-184,188-191 Efficiency
Galvanostatic charging 2,11,18,178,245,248,249,252,
179-183 254,255,257-261,267,269
Low-rate charging 170 Efficiency control
Maintenance charging 171 6,246,250,261-274,277-282
Quick charging 170 Effort variable 57
Reflex charging 171 Electrical capacitance 56,57
Thermostatic charging Electrical domain 56,57
179-184 Electrical double-layer 50
Top-off charging 170 Electrical field 75,77,80,81
Trickle charging 171 Electrical resistance 56,57
Chemical capacitance 56-59 Electrochemical potential 58
Chemical domain 56,57 Electrochemical reaction 61
Chemical flow 57,59,63 Electrode 34,43
Chemical potential 57,58 Electrode porosity 163
Chemical power 57 Intercalation electrode 38,107
Chemical reaction 59 Sintered electrode 36
Chemical resistance 56,57,69,70 SHE 38,90
Communication channel 10,19-22 Solid-solution electrode 89,108
Conduction angle 248 Electrolyte 34,43,117
Convection 48 Aqueous electrolyte 39
Cost function 127-130 Conductivity 117
LSE cost function 128 Decomposition 186,189
Coulomb counting 25,199 'Gelled' electrolyte 40
Cut-off voltage 35 H2S04 41
Cycle life 3,4,15,35,169,174,186 KOH 36,42,86
Non-aqueous electrolyte 39
D PEO 40
DC/DC converter Polymer electrolyte 40
1,2,5,18,28,255,258-260 Electrostatic potential 58
Index 293
G
EMF 196,212-219 Galvani, Luigi 31
EMF measurement methods Galvani potential 58
Voltage relaxation 212 Gas law 102
Linear extrapolation 213 Gibbs free energy 45,61,66,82-84
Linear interpolation 212 Free electrochemical energy
End-of-charge trigger 61,66
169,171-174,201 GMSK modulation 245
Inflection point 173 Gouy-Chapman theory 50
Maximum temperature 171 Diffuse layer 50,81,117
Pressure-based 173 Grove, William 32
Timed 171 GSM 19,26,207,243-245,262,274
Voltage plateau 26,173 Reporting period 274,275
~T 23,172
~TI~t 26,172 H
-~V 172 Heat 2,57,82-86
End-of-Discharge voltage 35,194 Heat capacitance 84,85
Endothermic reaction 82,84,177 Heat flow 84,85
Energy Heat transfer coefficient 85
Energy chain 1,2 Helmholtz theory 50
Chemical energy 2,33,57 Helmholtz layer 50,79
Energy Conversion Control Henri's law 98
11-14 High-voltage IC 22,177
Energy density 33
Electrical energy 2,10,33,57 I
Energy Management 3 IEC standard 36
Magnetic Energy 2 Impedance spectroscopy 197,199
Mechanical energy 2 Impedance transformation
Specific energy 34 242,247,251
Thermal energy 57 Inductive voltage converter 255-258
Enthalpy 82,83 Continuous conduction mode
Entropy 58,82-84,103 256
Equilibrium potential 5,44,62,64,66 Discontinuous conduction mode
Apparent equilibrium potential 256
73 Duty cycle 256
Hysteresis 132,164,212-219 EMI problems 258
Exchange current 46 Infolithium system 197
Exothermic reaction 82-84,177 Insertion loss 242,281
Interaction energy 109
F Attractive interaction 110
Faraday, Michael 31 Repulsive interaction 110
Faure, Camille 32
FDD 242
FDMA 242
J
Jungner, Waldemar 32
Fick's laws 48,69,97
Flow variable 57
Frequency band 241,281 K
Fuel cell 32 Kirchhoff
Fugacity coefficient 99 current law 76
voltage law 80,106,122,150
294 Index
L
Latin hypercube method 127 Physical system dynamics 5,56,57
Lattice structure 110 Plante, Gaston 31
Lattice-gas model 115 Poisson equation 75-77
Leclanche cell 31 Power
Lithiumdeposition 164,186,189-191 Power Amplifier
Linear regulator 27,252,253,258 3,19,242-251,259-261
Low DropOut regulator 252 Bipolar PA 250
Load 2,19 Class A,AB,B,C 246-249
Current-type load 195 Class D,E,F 250
Power-type load 195 Duty cycle 260
FETPA 250
M Power control 243,250,273-278
Power density 33
Mass transport 48,67-82,103,112
Power management 3
Mass conservation law 70,71
37,164,175,200 Power module 9-14
Memory effect
Specific power 34
Microcontroller 12,19,24,26-28,200
Microprocessor 12,19,200 PTC 17,29,39,107,156,186
48,7 5, 117 PWM 10,25-28
Migration
Mol fraction 89,108,114
Q
QPSK modulation 242,245,268
N
Nemst's equation 44,62
Nemst-Planck equation 69 R
NTC 11,23,26 Ramping specification 243
Nucleation process 94 Rated capacity 33
Reaction rate constant 44,59,63
Reaction order 46
0
36,87, 150,169 Reduced species 44
Overcharging
Reduction reaction 34,61
Overdischarging 36,88,152-154
Remaining capacity 194
Overpotential 5,45,219-225
Remaining time of use 194
Kinetic overpotential
'Rocking chair' principle 39,107
46,64,66,197,220
Diffusion overpotential
46,49,73,197,220 s
Ohmic overpotential 197,220 Safety switch 14-17,27,39,107,186
Oxidation reaction 34,61 SEI 107,114,164
Oxidized species 44 Self-discharge
25,35,117,151,152,179,200,229
p Separator 34,86,107
Shaver 9, 22-25,177-181
Pack supervisory electronics 17
Thermal shaver model 177,178
Parameter optimization 126
Simulation 4,5,285
Parameter space 127
Smart battery 4,17,20
Parameter space reduction
SBD specification 21
127-129
SBC specification 21
Orthogonal parameters 134
SBS specification 20-22
Partitioning of intelligence 10-14
SMBus specification 20
Passivation layer 164,186
Spectral mask 244,281
Phase separation 110
Staging phenomenon 115
Phase transition 110
Index 295
w
State-of-Charge (SoC) VVarburgimpedance 67
3,4,89,108,114,193,194
SoC indication 195-239
Accuracy 203,208,233
Random error 203
Reserve capacity 203
Systematic error 203
Adaptive systems 202,203
Book-keeping systems
199-202,204-211
Calibration 201,229,237
Direct measurements
195-199
Reliability 204
State-of-Health (SoH) 199
Stochiometric factor 61
Supply voltage 18,19,246,250
SMPS 3,22
Synchronous rectification 256
T
Tafel relation 47
Tafel plot 139,142,143
TDD 242
TDMA 242
TDMA frame 274
Thermal capacitance 56,57,84,85
Thermal domain 56,57
Thermal resistance 56,57,85
Thermodynamics 44
Time slot 242,274-276
Transfer coefficient 46,63
Transformer 2
Ideal transformer 63
Transformer coefficient 63
Trickle-charge current 24,26,171
Tudor, Henri 32
u
UMTS 244,245
Up-link 241
UPS 199
v
Valence of ionic species 58,75
Vector error 244
Volta, Alessandro 31
Voltage ripple 28,258

Battery Management Systems

Uploaded by

Copyright:

Available Formats

Battery Management Systems

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Battery Management Systems

Uploaded by

Copyright:

Available Formats

What is the scope and purpose of the Philips Research Book Series?

What is the scope and purpose of the Philips Research Book Series?

What are some of the main components of a Battery Management System discussed in the book?

What are some of the main components of a Battery Management System discussed in the book?

Battery Management Systems

SCOPE TO THE 'PHILIPS RESEARCH BOOK SERIES'

Henk. Jan Bergveld

Philips Research Laboratories,

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

ISBN 978-90-481-6108-9 ISBN 978-94-017-0843-2 (eBook)

Printed on acid-free paper

All Rights Reserved

List of abbreviations xiii

Series preface xxi

1.1 The energy chain 1

2. Battery Management Systems 9

3. Basic information on batteries 31

3.3 General operational mechanism of batteries 43

5. Battery charging algorithms 169

6. Battery State-of-Charge indication 193

7. Optimum supply strategies for Power Amplifiers

8. General conclusions 285

About the authors 289

LiMn20 4 Lithium manganese oxide

Symbol Meaning Value Unit

Symbol Meaning Value Unit

Symbol Meaning Value Unit

Symbol Meaning Value Unit

Rswitch On-resistance of switch in DC/DC

Symbol Meaning Value

Var; Output variable in optimization process Depends

8 Diffusion layer thickness

Symbol Meaning Value Unit

Conduction period for switch in s

Note I: Value of I has been assumed in this book

This book covers the interesting and multidisciplinary subject of Battery

as electrochemistry and electrical engineering to improve the design of BMS. As

HENK JAN BERGVELD

Eindhoven, April 2002

1.1 The energy chain

In view of battery costs and environmental considerations, the batteries used to

H. J. Bergveld et al., Battery Management Systems

t tmonitor and control

1.2 Definition of a Battery Management System

The basic task of a Battery Management System (BMS) is to ensure that

• Control charging of the battery, with practically no overcharging, to ensure a

1.3 Motivation of the research described in this book

1.4 Scope of this book

This chapter gives general information on Battery Management Systems (BMS)

2.1 A general Battery Management System

H. J. Bergveld et al., Battery Management Systems

Figure 2.1: A g eneral Battery Management System (BMS)

2.2 Battery Management System parts

2.2.1 The Power Module (PM)

Partitioning of the CHCfunction: example I

An advantage of this set-up is that no electronics have to be integrated with the

Partitioning of the CHC function: example 2

Partitioning of the CHC function: example 3

2.2.2 The battery

Electronic safety switch for Li-ion batteries