An Accurate Lithium-Ion Battery Gas Gauge Using

Two-Phase STC Modeling

Hsueh-Chih Yang and Lan-Rong Dung
Department of Electrical and Control Engineering
National Chiao Tung University
Hsinchu, Taiwan, R.O.C.
Email: [email protected]

Abstract The lithium-ion (Li-ion) battery is a time-varying,

nonlinear component. Its discharge characteristic is dependent on
discharge current, loading change scheme, ambient temperature,
and initial state-of-charge (SoC), and hence its remaining time
can vary with the discharge operating conditions in a manner
of nonlinear relationship. The non-linearity behavior makes the
accurate gas-gauge of Li-ion battery very difficult. This paper
presents an efficient scheme to simplify the estimation of battery
service time with high-degree of accuracy. According to the
typical discharge characteristic of Li-ion batteries, we applied a
two-phase single-time-constant (STC) model for the gas-gauging
strategy and parameterize the discharge operating conditions
in terms of the first-phase gradient, the knee voltage, and the
second-phase gradient, instead of using complex curve-fitting
equations. As shown in the experimental results, the accuracy
of predicted remaining time is less than 1% for constant current
cases, and 10% for loading change cases.

I. I NTRODUCTION
With increasing demands of portable electronics like notebook PCs, cell phones, PDAs, and digital cameras, lithium-ion
(Li-ion) and polymer Li-ion batteries are now very popular
in the daily lives. The state-of-charge (SoC) of a battery is
essential to users and power management policy such like the
Advanced Configuration and Power Interface (ACPI) [1]. A
batterys SoC is its available capacity expressed as a percentage of its rated capacity. Knowing the amount of energy left
in a battery compared with the energy it had when it was new
gives the user an indication of how much longer a battery will
continue to perform before it is exhausted or needs recharging.
However, the discharge behavior of a battery differs from that
of a capacitor, and simple equations alone can not determine
the SoC. The complicated electrochemical reactions make
predicting the battery present remaining capacity and residual
service lifetime difficult.
A number of researchers have reported models for predicting the battery remaining capacity or service lifetime. An
electrochemical model, DUALFOIL, based on concentratedsolution theory was reported in [2]. It is accurate and general
enough to handle a wide range of Li-ion cells, which also
provides the extensive use of its companion simulator software [3]. Since electrochemical models are accurate but time
consuming in practice. An efficient macromodel for Li-ion
batteries was presented in [4], where the battery is modeled
by a PSPICE circuit consisting of voltage sources and linear
passive elements. [5] approximates a discrete-time circuit

1-4244-0755-9/07/$20.00 '2007 IEEE

model by using VHDL language. Models based on PSPICE

and discrete-time VHDL are faster though less accurate. [6]
proposed a high-level diffusion-based analytical model. They
consider the concentration evolution of the active materials in
the battery during a discharge process in a finite region. By
identifying the model parameters, the diffusion-based model
can predict the battery lifetime given the discharge profile
in advance. Researchers also proposed a battery emulator for
experiments with battery-aware designs in [7]. The emulator
is an intelligent power supply that mimics the behavior of a
battery by running a battery simulation program in real-time. It
senses the current load and responds by controlling the output
voltage as an actual battery would, but it is computationally
intensive. A widespread review of battery models and batteryaware issues can be found in [8]. The accurate and fast online
estimation of the SoC and the service lifetime of a battery still
remains a challenge today.
We aim to develop an accurate and low-complexity gas
gauge of a Li-ion battery in practical cases. The gas gauge
takes the rate-recovery and rate-capacity effects into account
and is capable of predicting the overall discharge curve. We
simplified the battery model with accuracy and consider it as
a two-phase component as shown in Fig. 2 where an internal
resistance Rint has two components R1 and R2 . The discharge
curve of a Li-ion battery which was shown in Fig. 3 can be
divided into two parts: an exponential voltage decay of the
first phase characterized by discharging a fully charged single
time constant resistor-capacitor circuit (STC RC circuit) with a
long time constant, = RC, and a linear change of the second
phase characterized by discharging a small capacitor. The
t
exponential decay, 1 e , can be expanded as a Taylor series,
n
t

( )
1
n=0

n! , where t is time. Since t
, the first phase,

then, was approximated as a linear change. Consequently,
applying the approximation combining two straight lines (L1 :
at + b) and (L2 : ct + d) agrees well with the measured
data. We named the intersection "Knee" where the voltage was
specified as VK , thereby being aware of the curve entering the
second phase. Briefly, we focus attention on statistics of the
slopes of two straight lines and VK to characterize the curve of
terminal voltage during constant-current discharge at constant
temperature. In addition, the rate-recovery effect or relaxation
effect means the battery can recover from its lost efficiency

866

if given a chance to rest during periods of reduced load. The

discharge curve with dynamic current loads can be estimated
by carrying out a set of current load tests and cooperating with
surface fitting and interpolation techniques.

3.9
3.8

II. M ODEL DESCRIPTION

3.7

The objective of this section is to explain the proposed

battery model for the gas-gauging strategy and parameterize
the discharge operating conditions.

3.6
Battery Voltage (V)

Region 1

A. Description of the discharge characteristics

Region 2

3.5

Region 3

3.4
3.3
3.2

A typical discharge voltage versus time characteristic for

a Li-ion battery is given in Fig. 1. Clearly there are four
distinct regions; two nonlinear regions (Region 1 and Region
3) and two linear regions (Region 2 and Region 4). From an
user point of view the linear regions are the most important.
The Region 1, occurring at the start of discharge, typically
occupies a small proportion of the total discharge duration.
By observing Fig. 1, the discharge curve can be divided into
two parts shown in Fig. 3: an exponential voltage decay of the
first phase characterized by discharging a fully charged single
time constant resistor-capacitor circuit (STC RC circuit) with
a long time constant, , as described by
= RC

3.1
3

Region 4
0

200

400

(1)

1000 1200
Time (sec)

1400

1600

1800

2000

R2

R1

(2)

Fig. 2.

Since t
, the first phase, then, was approximated as a linear

change. Consequently, applying the approximation combining
two straight lines (L1 : at + b) and (L2 : ct + d) agrees well
with the measured data. We named the intersection "Knee"
where the voltage was specified as VK , thereby being aware of
the curve entering the second phase. Briefly, we focus attention
on statistics of the slopes of two straight lines and VK to
characterize the curve of terminal voltage during constantcurrent discharge at constant temperature.
Observing Fig. 4, for the first 800 sec the discharge current
was 0.5A. Then, the current load was reduced to 0.4A. Under
this condition, the battery recovers its voltage and afterward,
discharges in a manner of exhibiting the same behavioral trend
of the discharge curve under 0.4A till the end of discharge.

Equivalent circuit representation of lithium-ion battery.

3.9

Knee

3.8
3.7

Battery Voltage (V)

3.6

Discharge with long

RC time constant

3.5
3.4
3.3

Discharge with
small capacitor

3.2
3.1
3

B. Determining model parameters

To validate the proposed equivalent model, the model parameters of a specific battery must be identified experimentally
first. We use the battery emulator which was presented in
[7] to obtain the discharge curves under a set of constant
discharge rates (0.2A, 0.3A, 0.4A, and 0.5A) and a set of
two-stage constant discharge rates. The major objective of

800

Fig. 1. Typical discharge voltage versus time characteristic illustrating the

four regions.

and a linear change of the second phase characterized by

discharging a small capacitor. Therefore, we conclude that
a Li-ion battery can be modeled as a two-phase component
as shown in Fig. 2 where an internal resistance Rint has
two components R1 and R2 . The exponential decay can be
expanded as a Taylor series where t is time.
n

1 t
1  ( t
)

e =

n=0 n!

600

200

400

600

800

1000 1200
Time (sec)

1400

1600

1800

2000

Fig. 3. Typical discharge voltage versus time characteristic divided into two
phase with an intersection Knee.

867

3.9
3.8
3.7

Battery Voltage (V)

3.6
3.5

0.5A to 0.4A
0.4A
0.5A

3.4
3.3
3.2
3.1
3

200

400

600

800

1000

1200

1400

1600

1800

2000

Time (sec)

Fig. 4.

Geometric representation of sensitivity analysis of VK .

Fig. 7.

Rate-recovery effect.

the simulation is to construct a function which closely fits

a discrete set of known data points. Theoretically, all the
parameters are multivariable functions of SoC, discharge rate,
loading change scheme, and temperature. However, within
certain error tolerance, we simplified parameters to be independent or linear functions of discharge rates in a constanttemperature application. We use the fitted data of a set of
simulation data of VK and slopes of L1 and L2 under a set
of constant discharge rates to estimate discharge curves well
for additional constant rates. This is accomplished in Fig. 5 by
linearly interpolating between values for VK and slopes a and
c.
Fig. 8.

C. Sensitivity analysis of model parameters

The sensitivity analysis of model parameters can be performed by using geometric representation as shown in Fig. 7
and 8. The sensitivity analysis presented the effects of the
amount of deviation battery service lifetime caused by modeling error. Considering the scenario as shown in Fig. 7, we
estimated Knee voltage with error, as described by
VK =

VK

VK

L = VK (

1
1
)
|a| |c|

changes.
L = (VK Vcut )(

(4)

Approximately, in Eq.(4), the magnitude of VK is 104 , and

|a| and |c| range between 105 104 and 103 102
respectively. Consequently, we deduced that the magnitude of
L is less than 101 . In addition, the slope of L2 , c, contributes
error to battery service lifetime as well. The magnitude of
fitting error of slope c, c, as shown in Fig. 6(d), ranges
between 105 104 . For this reason, the fitting error of c
will produce service lifetime error less than 10sec by Eq.(5).
Moreover, to obtain precise estimation results, the slope of L1
must be carefully determined, especially when discharge load

1
1
c
 ) = (VK Vcut )(  )
|c| |c |
cc

(5)

III. E STIMATION ALGORITHM OF A BATTERY S SERVICE

LIFETIME

(3)

Therefore, the estimation error of battery service lifetime, L,

was derived by

Geometric representation of sensitivity analysis of c.

A. Notation
The following notation is used in the remainder of this
paper.
Ts
Cr
Lr
Vcut
V (n)
VK (n)
I(n)
T (n)
t(n)
mL1 (n)
mL2 (n)
L(n)

868

sampling period;
current used to determine rated capacity;
rated battery service lifetime;
cut-off voltage;
present terminal voltage;
present Knee voltage;
present load current;
present temperature;
present time;
present slope of L1 ;
present slope of L2 ;
present estimated battery service lifetime.

3.85

-0.001

0.0

3.75
-1.0x10

3.70

a
VK

3.65

-1.5x10

3.60

-2.0x10

-4

-4

-4

Slope of L 2 (V/sec)

-5

Slope of L 1 (V/sec)

-5.0x10

Battery Voltage (V)

-0.002

3.80

-0.003
-0.004
-0.005
-0.006
-0.007
-0.008

3.55
-2.5x10
0.20

0.25

0.30

0.35

0.40

0.45

-4

0.20

0.50

0.25

0.30

0.35

(a)

0.45

0.50

(b)

(a) Parameters VK and a for a set of constant discharge rates. (b) Parameter c for a set of constant discharge rates.

Fig. 5.

x 10

0.40

Battery Discharge Rate (A)

Battery Discharge Rate (A)

Fitted surface

-3

x 10

-4

3
4
2
3
1

1
0

-1

-1
-2

-2

-3
0.5

0.5
0.5

0.4

0.5

0.4

0.4
0.3

0.4
0.3

0.3
0.2

The second discharge rate (A)

0.2

The first discharge rate (A)

0.3
0.2

The second discharge rate (A)

(a)
x 10

0.2

The first discharge rate (A)

(b)

-4

x 10

Fitted surface

-5

3
8
2

2
0

-1

-2
-2
-4
-3
-4

0.5

0.5
0.4

0.4

0.5

0.5

0.4

0.3

0.3

0.4

0.3

0.3
0.2

0.2
The second discharge rate (A)

The second discharge rate (A)

The first discharge rate (A)

(c)

The first discharge rate (A)

(d)

Fig. 6. (a) S_VK (I1 , I2 ) = 0.0089I1 + 0.0072I2 + 0.0008, fitted surface of VK . (b) Fitting error of VK . (c) S_L2 (I1 , I2 ) = 103 (0.8541I1 +
0.8369I2 + 0.3229), fitted surface of L2 . (d) Fitting error of L2 .

869

B. Algorithm description
As explained previously, the discharge characteristic has
four distinct regions. The actions carried out during estimating
are defined by the region selected. By monitoring the present
voltage gradient, V (n), three kinds of geometric estimation
methods can be applied to predict the residual service lifetime
by L(n) t(n) and the remaining capacity by Coulomb
counting method accurately. Fig. 9 summarizes the resulting
estimation algorithm, which can be divided into several parts.
In any of them, the timing functions are generated by interruptions, and every one of them has its own subroutine. Additional
characteristics of these parts are briefly reported below.
Measurement:
From the beginning of the algorithm, we monitor the amount
of load current, I(n), going out of the battery, the batterys
terminal voltage, V (n), and the temperature, T (n), as well as
the time, t(n), that passes.
Cut-off Voltage Detected:
Once the algorithm has been started, a very important question
is when to finish it. We set a cut-off voltage as the termination
condition.
Voltage Slope Calculation:
To obtain the voltage slope of the first phase, the calculation
has been made, as described by
V (n)
mL1 (n)

= V (n) V (n 1)
V (n)
=
Ts

(6)
(7)

Slope Gradient:
In Region 1 and Region 3, the slope gradient is nonlinear
and gradually smooth, afterwards, entering into linear Region
2 and Region 4 respectively. The present discharge region
was determined by two thresholds, T H1 and T H2 . The slope
gradient was defined by
gradient = |V (n) V (n 1)|

(8)

Knee Estimation and Voltage Slope Estimation:

As mentioned previously, each specified load current can be
used to find the corresponding Knee and the slope of L2 with
the data in Fig. 5. VK and mL2 (n) will not be changed till the
load current changes in the proposed algorithm.
VK (n)
mL2 (n)

= VK (n 1)
= mL2 (n 1)

(9)
(10)

Current Load Changed:

When encountering the current measurement noise, we can
add one more threshold, T H3 , which is responsible to sensing
current loading changes.
Knee Compensation and Voltage Slope Compensation:
VK and mL2 (n) need to be compensated once the load
changing has been detected. We compensate VK and the slope
of L2 for a variable-current discharge load by two fitted
surfaces, S_VK (I1 , I2 ) and S_L2 (I1 , I2 ), shown in Fig. 6 (a)
and Fig. 6 (c) respectively. We can use the fitted equations to

find the compensation values while load current changes from

I1 to I2 . The compensation methods were formulated by
VK (n)
mL2 (n)

= VK (n 1) + S_VK (I(n 1), I(n)) (11)

= mL2 (n 1) + S_L2 (I(n 1), I(n)) (12)

Service Lifetime Estimation (A):

This part of algorithm is used to deal with the scenario that the
battery discharges during Region 2. VK and mL2 (n) should be
determined by either estimating or compensating. The present
information point, (t(n), V (n)), is going to move through the
slope direction, mL1 (n), and then we can predict the time
when the Knee voltage, VK , will be reached. Subsequently,
by using the direction of mL2 (n), L(n) can be determined
at the time when the Vcut is reached. The formulation was
derived as below.
VK (n) (V (n) mL1 (n)t(n)) VK (n) Vcut
+
L(n) =
mL1 (n)
mL2 (n)
(13)
Service Lifetime Estimation (B):
The scenario that the load current changes during the period
when entering into Region 3 and Region 4 has been excluded
in the discussion. Consequently, the estimating methods performed in Region 3 and Region 4 are identical. The estimation
results do not come with much error.
V (n) Vcut
(14)
L(n) = t(n) +
mL2 (n)
Service Lifetime Estimation (C):
The slope in Region 1 degrades rapidly. If we apply Service
Lifetime Estimation (A) in Region 1, the estimation results
will be far away from the truth. Observing that the period of
Region 1 is relatively short, we provide a simple equation for
reference.
I(n)
(15)
L(n) =
Cr
C. Results
The battery emulator, B#, [7] is used to simulate the battery
voltage characteristics for comparison. B# is a programmable
power supply that emulates the behavior of a battery. It
measures the current load, calls a battery simulation program
to compute the voltage in real time, and controls a linear
regulator to mimic the voltage output of a battery. We perform
a series of simulations with dynamic current load by modifying
the parameters as shown in Fig. 10. The column labelled
"err%" in Table I refers to the percentage of the difference
between the battery service lifetime obtained by using the
proposed gas gauge hardware and the B# emulator. As shown
in the comparison results, the accuracy of predicted lifetime
is less than 10% for loading change cases.
IV. C ONCLUSION
We have developed an accurate and low-complexity gas
gauge for Li-Ion battery. The gas gauge takes the rate-recovery
and rate-capacity effects into account and is able to predict the
discharge time. The battery is modeled as a two-phase STC
network. As shown in the results, the proposed gas gauge can

870

START

Discharge in the first phase (Region 2)

Service lifetime estimation (A)

Measurement

Yes
STOP

Service lifetime
estimation (C)

Voltage slope
compensation

Knee estimation

Knee
compensation

Cut-off voltage
detected ?

No

(Region 1)

Voltage slope
estimation

No

Voltage slope
calculation

Yes
Current load
changed ?

No

Slope gradient
>0?

No

Slope gradient
< TH 1 ?

Yes

Voltage slope
> TH2 ?

No

Yes (Region 3)

Yes
Service lifetime
estimation (B)

Voltage slope
estimation

Knee estimation

Discharge in the second phase (Region 4)

Fig. 9.

Flowchart of estimating present remaining capacity and residual service lifetime.

TABLE I

Load Current (A)

IR

C OMPARISON R ESULTS .
TP
200
300
100
400
300
100
500
300
200
600
700
200
600
200
400
200
600
100
400
400
500
500

IP
IQ
TP

TQ
Time (sec)

Fig. 10.

Dynamic load current profile example.

estimate the remaining discharge time with high degree of

accuracy.

TQ
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
400
400
500

IP
0.8
0.8
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.5
0.5
0.5
0.5
0.2
0.2
0.2
0.2
0.5
0.5
0.5

IQ
0.4
0.4
0.5
0.5
0.6
0.4
0.4
0.8
0.5
0.5
0.6
0.4
0.4
0.8
0.8
0.4
0.4
0.8
0.8
0.4
0.4
0.4

IR
0.4
0.4
0.5
0.5
0.6
0.4
0.4
0.8
0.5
0.5
0.6
0.4
0.4
0.8
0.8
0.4
0.4
0.8
0.8
0.3
0.3
0.3

[7](sec)
1741
1624
1151
912
724
1881
1649
562
1250
1347
1113
1879
1760
557
648
2001
2187
553
790
3265
3196
3147

proposed(sec)
1764
1532
1128
838
703
1927
1490
557
1172
1344
1159
1855
1670
524
659
2059
2259
524
782
3197
3275
3264

err%
1.31
-6.71
-1.98
-8.16
-2.90
2.44
-9.67
-0.92
-6.21
-0.23
4.17
-1.30
-5.13
-5.87
1.76
2.92
3.27
-5.22
-1.06
-2.08
2.48
3.72

R EFERENCES
[1] (Oct. 2006). Advanced Configuration and Power Interface Specification, Revision 3.0b. [Online]. Available: http://www.acpi.info/spec.htm
[2] T.F. Fuller, M. Doyle, and J. Newman, Modeling of galvanostatic charge
and discharge of lithium-ion insertion cells, J. Electrochem. Soc., vol. 141,
no. 1, pp. 19, Jan. 1994.
[3] (May 2005). FORTRAN Programs for Simulation of Electrochemical Systems. [Online]. Available: http://www.cchem.berkeley.edu/ jsngrp/fortran.html
[4] S. Gold, A PSPICE macromodel for lithium-ion batteries, in Proceedings of the 12th Battery Conference, pp. 915, 1997.
[5] L. Benini, Discrete-time battery models for system-level low-power
design, IEEE Trans. Very Large Scale Integr. (VLSI) Syst., vol. 9, no.
5, pp. 630640, Oct. 2001.

[6] D. Rakhmatov and S.B.K. Vrudhula, An analytical high-level battery

model for use in energy management of portable electronic systems, in
Proc. IEEE Int. Conf. Comput.-Aided Des., pp. 488493, Nov. 2001.
[7] C. Park, and J. Liu, and P.H. Chou, B#: a battery emulator and
power profiling instrument, The International Symposium on Low Power
Electronics and Design (ISLPED), pp. 288293, Aug. 2003.
[8] R. Casas and O. Casas, Battery sensing for energy-aware system design,
Computer, vol. 38, no. 11, pp. 4854, Nov. 2005.

871