IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO.

8, AUGUST 2020 6365

Digital Battery Management Design for

Point-of-Load Applications With Cell Balancing
Yin-Di Yang , Kai-Yu Hu , Member, IEEE, and Chien-Hung Tsai, Member, IEEE

Abstract—In portable devices, such as notebook com-

puters and tablet personal computers, one or more bat-
tery strings with multiple lithium-ion (Li-ion) cells have been
used since long. Life cycle and safety of the Li-ion battery
strings are crucial aspects. A battery management system
(BMS) is required for monitoring the status of each cell and
balancing their state of charge (SoC). However, most con-
ventional cell equalizers provide only cell monitoring and Fig. 1. Battery charge–discharge capacity concept.
equalization functions. In this article, a parallel architec-
ture is presented that not only achieves battery cell equal-
ization but also is capable of providing voltage regulation
for point-of-loads (POLs). In the proposed equalizer, sev-
eral isolated dc–dc converters are constructed. The input
terminal of each converter is connected to a battery cell,
and the output terminals are connected to one point for the
poststage POLs. A battery voltage acquirement system is
built for SoC estimation. A central BMS controller is realized
using a field-programmable gate array to receive sensed
signals and control all the functions. The entire system is
verified using the experimental results for a battery string Fig. 2. Classification of cell equalization techniques.
involving three NCR 18650B Li-ion cells connected in series.
Index Terms—Battery equalizer, battery management life-cycle reduction. An appropriate battery management sys-
system (BMS), bidirectional flyback converter, lithium-ion tem (BMS) with cell equalization [2] is crucial for these battery
(Li-ion) battery, low-power application.
packs.
I. INTRODUCTION Many methods have been developed in view of the cell equal-
ization demand [3]. Fig. 2 presents a classification of the com-
ITHIUM-ION (LI-ion) battery cells have many advan-
L tages such as high power density, no memory effect, long
life cycle, and low self-discharge rate [1]. Therefore, these cells
mon cell equalization methods. Passive balancing is the most
simple and cost-effective method and is commonly used in the
commercial products [4]–[7]. However, the efficiency loss and
are very suitable for the portable devices such as mobile phones, heating problems are significant in the passive balancing scheme
tablet personal computers, and notebooks. Specifically, for note- [8], [9]. Active balancing can be classified into several types:
books, multiple battery cells are connected to form battery Cell-to-Cell [10]–[19], Cell-to-Pack [20], [21], [26], [27], and
strings. Due to the variance between battery cells in each string, Pack-to-Cell [22]–[27]. In the Cell-to-Cell method, cell equal-
different charge capacities are obtained during the operation. As ization can be achieved between the specific cells through one
a result, cells might be overcharged or undercharged during the or more dc–dc converters. This method has a satisfactory bal-
operation, as presented in Fig. 1. This leads to cell damage or ance speed and is more efficient in terms of cost and control
simplicity. The Cell-to-Pack method discharges the cell with
the highest state of charge (SoC) to the entire battery pack or
Manuscript received October 10, 2018; revised February 1, 2019, April
23, 2019, and July 7, 2019; accepted August 18, 2019. Date of publi- string. By contrast, in the Pack-to-Cell method, the entire bat-
cation September 12, 2019; date of current version March 31, 2020. tery string energy is used to charge the cell with the lowest
This work was supported by the Ministry of Science and Technology, SoC. However, in both methods, the circuit devices encounter
Taiwan, under Project MOST 107-2221-E-006-201 and Project MOST
108-2221-E-006-145-MY2. (Corresponding author: Yin-Di Yang.) voltage stress because they have to deal with the entire battery
Y.-D. Yang and C.-H. Tsai are with the Department of Electrical En- string voltage through balancing. The balance speed of the two
gineering, National Cheng Kung University, Tainan 701, Taiwan (e-mail:, methods is relatively lower than that of the Cell-to-Cell method
[email protected]; [email protected]).
K.-Y. Hu is with the Department of Electronic Engineering, National because the energy does not transfer directly between cells but
Cheng Kung University, Tainan 701, Taiwan (e-mail:, s972882@gmail. transfers through the entire battery string.
com). All the aforementioned cell equalization methods only pro-
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org. vide the charge balancing function. However, the commercial
Digital Object Identifier 10.1109/TIE.2019.2940001 notebook computers require one or more voltage regulators,
0278-0046 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 6366 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 8, AUGUST 2020

Fig. 3. Conventional power system for a notebook with a BMS.

Fig. 5. Proposed BMS architecture.

in detail. In Section III, all operation modes including the nor-

mal operation and cell balancing modes are presented. Hard-
ware implementation and experimental results are presented in
Section IV to verify the proposed system. Finally, Section V
concludes this article.

II. PROPOSED BMS

In this section, the entire cell equalizer system architecture
and power stage are presented, and the SoC estimation method
is introduced.
Fig. 4. Proposed BMS with power conversion function.
A. System Architecture
such as MAX17085B [28], for downstream point-of-loads A three series-connected Li-ion batteries BMS was developed
(POLs) and a light-emitting diode (LED) backlight driver such in this article. As presented in Fig. 5, three battery cells—cell1
as LP8545 [29], as presented in Fig. 3. The conventional BMS to cell3 —are connected in series to form a battery string (yellow
cannot achieve both cell equalization and power conversion for block). Multiple isolated dc–dc converters were adopted in this
POLs; thus, manufacturers encounter high-cost-related prob- article to solve the multiground problem and enable high-power
lems. To solve this problem, the cell equalization architectures application. Each input terminal of the multiple isolated dc–dc
have been proposed [30]–[32] by involving battery balancing converters is connected to a battery. All the output terminals
and voltage source provision for the poststage POLs. In the are connected together to form a voltage regulated point for the
Cell-to-Load method, the balancing function is achieved using poststage POLs; the point had a voltage of 3.3 V in this article.
multiple power converters and by sharing the energy of the A central digital BMS controller receives all sensed signals and
output load current on the basis of the SoCs of the battery cells. sends all control signals including the pulsewidth modulation
However, if there is no load current, there is no energy to be (PWM) duty cycle signals.
shared as well. The Cell-to-Load method does not perform cell The central digital controller is divided into three main blocks.
equalization appropriately in the absence of a load. 1) A battery monitoring center is presented as an orange
In this article, an architecture and a control strategy that block in Fig. 5. This block receives the terminal volt-
combine both cell equalization methods and provide a voltage ages of the three battery cells and three output currents.
source for downstream POLs were proposed, as presented in The battery terminal voltage drop due to the battery
Fig. 4. The main advantage of the proposed system is that impedance during discharging is compensated by using
it can accomplish battery balancing and can reduce voltage an IR compensator. An SoC estimator looks-up the SoCs
regulators for POLs simultaneously. The proposed converter of the three batteries on the basis of the compensated
mainly focuses on power conversion for low-voltage POLs terminal voltage.
(3.3 V here). Moreover, cell equalization can be conducted in 2) A mode selector selects different balancing modes on
the presence or absence of a load current. the basis of the situation, and then, the current calculator
This article is organized into several sections as follows. The calculates the balancing current in each converter on the
entire system architecture and circuit description is presented in basis of the SoC and the output load current. The afore-
Section II. Moreover, the block diagram of the central digital mentioned blocks are presented as the battery equalizer
BMS controller and the SoC estimation method is discussed in blue blocks in Fig. 5. The mode selector has three
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: DIGITAL BATTERY MANAGEMENT DESIGN FOR POINT-OF-LOAD APPLICATIONS WITH CELL BALANCING 6367

Fig. 6. Multiple flyback converters with series-connected input and parallel-connected output.

modes: Normal Operation, Cell-to-Load Balancing, and

Cell-to-Cell Balancing. The detailed operation principle
of these modes is presented in Section III. A current con-
troller increases or decreases the duty cycle on the basis of
the difference between the calculated and actual current
value.
Finally, the voltage regulation loop is presented as green
blocks in Fig. 5. A digital proportional–integral–derivative
(PID) voltage compensator conducts voltage regulation by using
a look-up table-based method; the digital input voltage signal
is encoded to a reasonable range to avoid limit cycle oscilla-
tion in the digital power system. Moreover, a 9-bit resolution Fig. 7. Discharge characteristic of an NCR-18650B battery cell.
dithering digital PWM (DPWM) is used to obtain low-cost high-
resolution duty signals.
This article was conducted using low-power specifications for digital pulsewidth modulation (PWM) modulator, as presented
verifying the functionality of the proposed architecture. How- in Fig. 5. Similar to that in a single flyback converter, there
ever, the architecture can be very easily modularized when in- are two operation stages in the continuous conduction mode
creasing the number of battery cells; thus, there is a possibility (CCM). As there is no phase-shift modularization mechanism,
of using the architecture for high-power applications. the three primary-side switches are simultaneously switched ON
in stage1 . The three battery cells start sending energy to Lm 1 –
B. Bidirectional Digital Flyback Converters Lm 3 . Then, in stage2 , Lm 1 –Lm 3 start sending energy to the sec-
ondary side for the output load. The duty signals here are only
A bidirectional flyback converter was used as an isolated calculated for voltage regulation; thus, it is an average value
dc–dc converter in this system. The converter has several ad- for three converters. However, when current control is required,
vantages, such as simple control, low cost for low-power ap- the duty signals differ from each other through appropriate duty
plication, and buck–boost operations for our design, when used adjustment [33]–[35].
in the isolated converters. Therefore, we replaced a diode that
is commonly used in the secondary side with a MOSFET for
a bidirectional operation. For the proposed architecture, multi- C. Battery Monitor
ple bidirectional flyback converters should be specifically con- Many methods have been proposed to estimate the
nected to each other. As presented in Fig. 6, each input of the accurate SoC of batteries [36]–[43]. In this article, we did not use
converters is connected in series with another input and each complicated methods but used the open-circuit voltage (OCV)
battery cell, whereas the output of the converters is connected in method, which was adopted in [20], because SoC estimation is
parallel to one terminal for performing voltage regulation. In this not the primary topic of discussion of this article. There is a
system, voltage regulation is achieved by the digital regulation high positive correlation between the terminal voltages of bat-
loop including voltage encoder, digital PID compensator, and tery cells and their actual SoCs. Fig. 7 presents an OCV–SoC
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 6368 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 8, AUGUST 2020

Fig. 9. Mode selection flow diagram.

Fig. 8. Battery equivalent model and SoC estimation flow diagram.

measurement curve of a 18650B battery cell that can be used for

SoC estimation. However, there is a voltage mismatch between
the terminal voltages of the cells and their actual OCV during the Fig. 10. Normal operation of the proposed system.
charge–discharge operation [43]. When the discharge or charge
current appears on the battery cells, there is a voltage drop or
rise due to cell impedance. A battery impedance correction was
proposed in this article for relatively accurate SoC estimation. III. OPERATION OF BATTERY BALANCING
The battery equivalent circuit and the correction flow diagram This section is divided into several parts for explaining the en-
are presented in Fig. 8. In the SoC estimation flow, the battery tire operation of the proposed system. Initially, we specify how
impedance Zcell is calculated after the system startup, and the to select the balancing modes. Then, we discuss the operation of
impedance correction term updates with different load current multiple converters for conducting voltage regulation, for which
for OCV–SoC estimation. Since Zcell may not variate with load the balancing function is switched OFF. Subsequently, two cell
current, the impedance estimation only needs to be done once balancing modes are explained in detail. Finally, a complete
after the system startup. operation flow diagram of the entire BMS is presented.
The OCV with different discharge currents can be derived as As presented in Fig. 9, when the balance enable signal Bal_en
follows: is switched ON, on the basis of the different load situations, there
VOCV 1 = (Idischarge1 × Zcell ) + Vterm inal1 (1) are several modes in this article. When there is a load current
at the output, cell-to-load balancing is selected. However, when
VOCV 2 = (Idischarge2 × Zcell ) + Vterm inal2 . (2) there is no load, cell-to-load balancing cannot perform the equal-
ization appropriately. Thus, cell-to-cell balancing is selected
VOCV1 and VOCV2 are nearly equal to each other within the
when there is no load. The detailed operations are introduced in
short transition period of different discharging current. As a
the latter sections.
result, the battery impedance can be estimated using the fol-
lowing equation by measuring terminal voltage and discharging
current: A. Normal Operation (No Balancing)
Vterm inal1 − Vterm inal2 When the cell balancing function does not start, the entire
Zcell = . (3)
Idischarge2 − Idischarge1 system should still perform as a voltage regulator for the output
The first terminal voltage and the discharging current of each system. The output voltage was regulated to 3.3 V in this arti-
battery cells are recorded at the system reset state. And the sec- cle. As presented in Fig. 5, the output voltage signal is sent to
ond terminal voltage and the discharging current are recorded at the analog-to-digital converter to convert the signal before in-
the very beginning when the system is operated. As the discharg- putting it into the central digital BMS controller. After the BMS
ing current at the system reset state and the operated state are controller receives the digital voltage signal, it encodes the sig-
zero and nonzero, respectively, two different discharging current nal into an error signal by comparing the signal to a reference
can be obtained for sure. On the other hand, the SoC estima- voltage. Then, the error signal is passed through a digital PID
tion inaccuracy is mainly caused by the noise on the measured compensator for conducting the digital PWM duty cycle con-
terminal voltage. In order to guarantee an estimation accuracy, trol. Finally, the dead time control is also conducted for a safe
the measurement instant should exclude the power switch tran- operation. The energy flow through each converter is displayed
sition time of a switching converter. And the bypass capacitors in Fig. 10. The currents of the multiple converters are randomly
should add to minimize voltage ripple induced by the switching distributed because there is no current control in this operation
converter. mode.

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: DIGITAL BATTERY MANAGEMENT DESIGN FOR POINT-OF-LOAD APPLICATIONS WITH CELL BALANCING 6369

Fig. 11. Cell-to-load balancing function (SoC3 > SoC2 > SoC1 ). Fig. 12. Cell-to-cell balancing function (mode 1) (SoC3 > SoC2 >
SoC1 ).

B. Cell-to-Load Balancing Iref (k ) = Iout(k )

When a load current is drained to the load and the battery Vbat(k ) × (SoCk − SoCbal )
cells need to be balanced, the system employs the cell-to-load = N   × ILoad . (9)
balancing mechanism. The proposed cell-to-load balancing i=1 Vbat(i) × (SoCi − SoCbal )

mechanism targets to control discharge current of every battery

For the proposed cell-to-load balancing mechanism, the bat-
according to the SoC state. After certain discharge period
tery cell individually connects to the balancing circuit. And the
Tbalance , cell balance can be achieved. At the beginning, the
discharging current of each cell is sourced from the load, there
minimum SoC in the battery string with N battery cells will be
is no current flows between each cell. The convergence proof of
found as the target SoC (SoCbal ), as shown in (4). Since SoC is
the proposed cell balancing algorithm can be done considering a
the proportion of the charge in the battery cell to the maximum
single-cell condition. The relation between target SoC (SoCbal )
charge capacity, as shown in (5), cell balance can be achieved
and output current of the kth converter Iout(k ) can be modeled
by addressing the discharge current Iin(k ) on the battery cell k
as
after the discharge period Tbalance , as shown in (6). However,
the discharge currents are sourced from the load current ILoad Vout
f(Iout(k ) ) = SoCk − Iout(k ) · Tbalance − SoCbal .
which is equal to the summation of every output current Iout(k ) Vbat(k ) · η
of the converters, as shown in (7). And the relation between (10)
the desired discharge current Iin(k ) and the output current According to the convergence analysis of Newton’s method
Iout(k ) is shown in (8), where Vbat(k ) is the terminal voltage [44], the numerical results of Iout(k ) will convergence to Iref (k ) .
of the kth battery and η is the efficiency of the converter. Because f(Iout ), f  (Iout ), and f  (Iout ) are continuous in the
Assuming that the converters have the same efficiency, the interval 0 ≤ Iout ≤ Iload , f(Iref (k ) ) = 0, and f  (Iref (k ) ) = 0,
desired discharge current Iin(k ) can be implemented by the proposed cell balancing algorithm meets the convergence
controlling the output current as shown in (9) which can be constrain.
derived by substituting (7) and (8) into (6). As the result, the
SoC of every battery cell becomes SoCbal after the discharge C. Cell-to-Cell Balancing
period Tbalance .
As mentioned in Section III-B, cell-to-load balancing can be
In particle implementation, the output current of the converter
used to perform the cell equalization when there is a load but
connected to a battery with minimum SoC will be zero according
cannot be used when there is no load. This section explains
to (9), the converter will be shut down. The other converters
how cell balancing can be executed without considering the
control the output current by tuning the duty cycle after the
output load level. cell-to-cell balancing has three modes, and
output voltage regulates to the target voltage.
each mode is explained as follows. The operation of each mode
Take battery string with three cells as Fig. 11 (SoC3 >
can be thought as single cell to single cell balancing, so that the
SoC2 > SoC1 ), for example, converter no. 1 will be idle, con-
convergence can be guarantee.
verter no. 2 and no. 3 share load current according to (9).
1) Mode 1 (One to One): In mode 1, two converters connect
to the cell with maximum SoC and minimum SoC are active
SoCbal = min{SoC(k ) |k ∈ [1, 2, . . . , N ]} (4) and the other converters are idle. The cell with maximum SoC
charges the cell with minimum SoC through the converters. The
Qcell(k ) balancing current can be configured by the user. As a result, the
SoCk ≡ × 100% (5)
Q100% SoC of the active cells will be equal and mode 2 or mode 3 can
SoC1 − SoCbal SoC2 − SoCbal adapt afterward.
= = ··· Take a battery string with three cells, for example, one of
Iin(1) Iin(2)
the converters is idle, the other converters are discharged and
SoCN − SoCbal charged, respectively, as illustrated in Fig. 12. The converter that
= = Tbalance (6)
Iin(N ) is connected to the charged cell performs as a current loader for
the converter connected to the discharged cell.
Iout(1) + Iout(2) + · · · + Iout(N ) = ILoad (7)
2) Mode 2 (N−1 to One): Mode 2 is designed for balancing
Vout × Iout(k ) = Vbat(k ) × Iin(k ) × η (8) the battery string with N cells in which N − 1 of N cells have

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 6370 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 8, AUGUST 2020

Fig. 13. Cell-to-cell balancing function (mode 2) (SoC3 = SoC2 >

SoC1 ).

Fig. 14. Cell-to-cell balancing function (mode 3) (SoC3 > SoC2 =

SoC1 ).

similar SoC and one of N cells has lower SoC than the others.
N − 1 of N battery cells discharge equally through appropriate
current control, which can be configured by a user as well.
Take the battery string with three cells, for example, two
battery cells are discharged, and the remaining cell is charged. Fig. 15. Proposed system operation flow.
Fig. 13 presents the detailed operation of mode 2.
3) Mode 3 (One to N−1): Mode 3 is designed for balancing
the battery string with N cells in which N − 1 of N cells have
similar SoC and one of N cells has greater SoC than the others.
The battery cell that has the highest SoC charges than the other
cells until all of the cells has the equal SoC.
Take the battery string with three cells, for example, two cells
have the same SoCs but their SoC values are less than the SoC of
the third cell. The battery cell that has the highest SoC charges
the other two cells. The detailed functional operation of this
mode is presented in Fig. 14.
4) Operating Flow: After illustrating all the operation
modes of the system, the entire system operation flow is pre-
sented in this section. The proposed algorithm is mainly segre-
gated into three parts: system initialization, voltage regulation,
and cell balancing. After the system starts, it receives the ter-
minal voltages from the three battery cells and estimates their
Fig. 16. Entire battery management system, including the battery
SoCs through the OCV method. Then, the voltage regulation is string with the voltage acquirement system, a multiple flyback converter,
performed as aforementioned. Finally, if the three battery cells and a central BMS controller.
are balanced, no operation is performed, but the voltage regula-
tion is performed. However, if they are not balanced, the system
switches to either cell-to-load balancing or cell-to-cell balanc- DE1-SoC field-programmable gate array (FPGA)-based board.
ing on the basis of the load situation. The entire system flow The electrical specifications of the entire system are listed in
diagram is presented in Fig. 15. Table I, and the related devices used in this system are listed in
Table II.
IV. REALIZATION AND EXPERIMENTAL RESULTS
A. System Implementation B. Normal Operation (No Balancing)
To verify the performance of the proposed equalizer, a BMS In this section, the experimental results of the voltage regula-
was incorporated into the hardware for the three Li-ion cells. tion function are presented when the Li-ion batteries operate at
Fig. 16 presents the entire proposed system including a battery a cell terminal voltage range of 2.7–4.2 V. The currents of the
string involving three NCR 18650B Li-ion cells that are con- multiple converters were slightly different due to the unbalance
nected in series, the multiple flyback converters as a balancing of the hardware implementation. Fig. 17 presents the case in
circuit, and a central BMS controller developed using the Altera which the three cells were set to 4.2 V, and the load current was
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: DIGITAL BATTERY MANAGEMENT DESIGN FOR POINT-OF-LOAD APPLICATIONS WITH CELL BALANCING 6371

TABLE I
ELECTRIC SPECIFICATIONS OF THE PROPOSED SYSTEM

TABLE II
BATTERY TYPE AND MAJOR DEVICE PARAMETERS
Fig. 18. Waveform of the output voltage (yellow) and output currents
of the multiple converters (red, green, and purple). The three cells were
set to 2.7 V, and output load current was 0.9–1.8 A.

Fig. 19. Converter efficiency at input cell voltage set to 4.2 V, load
current from 0.2 to 2.4 A.

Fig. 17. Waveform of the output voltage (yellow) and output currents
of the multiple converters (red, green, and purple). The three cells were
set to 4.2 V, and the output load current was 1.2–2.4 A.

changed from 1.2 to 2.4 A. The output voltage was controlled

at approximately 3.3 V even though the transient state. By con-
trast, Fig. 18 presents the case in which the three cells were set
to 2.7 V, and the load current was changed from 0.9 to 1.8 A.
The output voltage was well controlled throughout the entire
operation range of the Li-ion battery cells.
Figs. 19 and 20 are the measurement curves of the converter
efficiency at input cell voltage set to 4.2 and 2.7 V. The maxi- Fig. 20. Converter efficiency at input cell voltage set to 2.7 V, load
mum efficiency is about 85% and average efficiency is greater current from 0.2 to 2.0 A.
than 70%.
because they were not controlled in this case. Fig. 21(b) presents
C. Cell-to-Load Balancing the same waveform but when the balancing function was ON.
In this section, we present the experimental results of the We observed that the output voltage was still 3.3 V as well con-
cell-to-load balancing mechanism. The steady-state waveforms trolled. However, the currents of the multiple converters were
of the multiple converters are illustrated in Fig. 21. Fig. 21(a) adjusted to different levels on the basis of the SoCs of the three
presents the 3.3 V output voltage and the output current of the cells (in this experiment, SoC1 was 85%, SoC2 was 90%, and
multiple converters when the balancing function was OFF. The SoC3 was 95%). As presented in Fig. 22, the current can be
currents of the multiple converters were randomly distributed appropriately controlled even when the load current varies in
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 6372 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 8, AUGUST 2020

Fig. 21. (a) Steady-state waveform when the balancing function is OFF.
(b) Steady-state waveform when the balancing function is ON (output
voltage: green; output currents of the multiple converters: blue, red, and
orange; SoC1 : 85%, SoC2 : 90%, and SoC3 : 95%; output load current:
0.6 A).
Fig. 24. Battery terminal voltage when the balancing function is ON
(initial SoCs: SoC1 : 85%, SoC2 : 90%, and SoC3 : 95%).

Fig. 22. (a) Transient-state waveform when the balancing function is

OFF. (b) Transient state waveform when the balancing function is ON
(output voltage: green; output currents of the multiple converters: blue,
red, and orange; SoC1 : 85%, SoC2 : 90%, and SoC3 : 95%; output load Fig. 25. (a) Output voltage (yellow) and output currents of the three
current: 0.6 A). converters (green, red, and purple). (b) Output voltage (yellow) and cur-
rents of the three cells (green, red, and purple); initial SoCs: SoC1 :
87.5%, SoC2 : 90%, and SoC3 : 95%.

Fig. 23. Battery terminal voltage when the balancing function is OFF
(initial SoCs: SoC1 : 85%, SoC2 : 90%, and SoC3 : 95%).
Fig. 26. Battery terminal voltage when the balancing function is ON
(initial SoCs: SoC1 : 87.5%, SoC2 : 90%, and SoC3 : 95%).

the transient state. As a result, the terminal voltages of the three

battery cells are presented in Figs. 23 and 24 when the balanc-
ing function is OFF and ON, respectively. When the balancing 1) Mode1 (One to One): In mode 1, the SoCs of the three
function was OFF, the terminal voltages of the three cells did battery cells were initialized as follows: SoC1 : 87.5%, SoC2 :
not converge for a period of time, that is, their SoCs still have 90%, and SoC3 : 95%. On the basis of the discussions presented
gaps with each other. However, as the operation of the balancing in Section III, the three cells were charged, idle, and discharged,
function was initiated, the terminal voltages of the three cells respectively. Fig. 25 presents the output voltage, output cur-
converged after an equalization time period. rents, and actual battery cell currents. Due to the power loss of
the converter, the current levels of the two sides were different.
However, this does not affect the equalization function. The ter-
D. Cell-to-Cell Balancing
minal voltage recorded is presented in Fig. 26, and we observed
Compared with the cell-to-load balancing mechanism, when that the balancing function is working through a period of time.
the load current does not present, the system enters the cell- 2) Mode 2 (N−1 to One): In mode 2, the SoCs of the three
to-cell balancing mechanism, as aforementioned in the section. battery cells were initialized as follows: SoC1 : 85%, SoC2 :
In this section, we present the experimental results of the three 95%, and SoC3 : 95%. As aforementioned, two cells (cell2 and
modes of cell-to-cell balancing. cell3 ) discharged to the other cell (cell1 ). Fig. 27 presents the
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: DIGITAL BATTERY MANAGEMENT DESIGN FOR POINT-OF-LOAD APPLICATIONS WITH CELL BALANCING 6373

Fig. 27. (a) Output voltage (yellow) and output currents of the three
converters (green, red, and purple). (b) Output voltage (yellow) and cur-
rents of the three cells (green, red, and purple); initial SoCs: SoC1 : 85%,
SoC2 : 95%, and SoC3 : 95%. Fig. 30. Battery terminal voltage when the balancing function is ON
(initial SoCs: SoC1 : 85%, SoC2 : 85%, and SoC3 : 95%).

TABLE III
COMPARISON OF ACTIVE CHARGE EQUALIZATION ARCHITECTURES

Fig. 28. Battery terminal voltage when the balancing function is ON

(initial SoCs: SoC1 : 85%, SoC2 : 95%, and SoC3 : 95%).

1
Overall comparison, it could be improved through design.

cell3 . The terminal voltages of the three battery cells converged

at an equalization time, as presented in Fig. 30.

E. Comparison With Prior Studies

Fig. 29. (a) Output voltage (yellow) and output currents of the three
converters (green, red, and purple). (b) Output voltage (yellow) and cur- Finally, we compared some balancing architectures that
rents of three cells (green, red, and purple); initial SoCs: SoC1 : 85%,
SoC2 : 85%, and SoC3 : 95%.
use the flyback converter with our proposed cell equalization
method, as presented in Table III. Compared with Cell-to-Pack
and Pack-to-Cell, our proposed system, first, has a lower volt-
age stress of the balancing converter (Li-ion battery cell volt-
output voltage because this is the most crucial factor. Moreover, age and 3.3 V) because the other two methods have to deal
the figure presents the three output currents and the battery with the voltage of the entire pack. Second, the balancing
cell currents. As presented in the figure, the second and third speed of the proposed system is qualitatively faster because
converter’s currents were positive, which implies that cell2 and the energy is transferred directly through high or low SoC
cell3 were discharged. By contrast, the first converter’s current cells without passing through the entire battery pack. Third,
was negative, and the cell1 was charged by the other two cells. easy modularity is another advantage of the proposed sys-
The balancing record is illustrated in Fig. 28. tem. When the cell number in the battery string increases, the
3) Mode 3 (One to N−1): In the last mode, the SoCs of voltage stress on each converter remains the same, the dupli-
the three battery cells were initialized as follows: SoC1 : 85%, cate converter modules can be used. The proposed cell bal-
SoC2 : 85%, and SoC3 : 95%. Similarly, in mode 2, the cell ance algorithm can be implemented by extending desired N in
that had the highest SoC (cell1 ) discharged to the other two (9) and adding current/voltage sampling analog to digital con-
cells (cell2 and cell3 ). Fig. 29 presents the output voltage and verters (ADCs) on the controller. Also, the balancing speed
currents of the three converters. The first and second converters’ will not decrease with the increase of cell number. Compared
currents were negative because cell1 and cell2 were charged by with the Cell-to-Load architecture, we can achieve a balancing
Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 6374 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 67, NO. 8, AUGUST 2020

function even when there is no load current. However, the cost [12] C. Karnjanapiboon, K. Jirasereeamornkul, and V. Monyakul, “High effi-
of the proposed system may be slightly higher if the number of ciency battery management system for serially connected battery string,”
in Proc. IEEE Int. Symp. Ind. Electron., Seoul, South Korea, Jul. 2009,
battery cells is increased. pp. 1504–1509.
[13] H.-S. Park, C.-E. Kim, C.-H. Kim, G.-W. Moon, and J.-H. Lee, “A mod-
V. CONCLUSION ularized charge equalizer for an HEV lithium-ion battery string,” IEEE
Trans. Ind. Electron., vol. 56, no. 5, pp. 1464–1476, May 2009.
An architecture and a control method that combined both bat- [14] Q. Ouyang, J. Chen, J. Zheng, and H. Fang, “Optimal cell-to-cell balancing
tery cell equalization and voltage regulation for the poststage topology design for serially connected lithium-ion battery packs,” IEEE
Trans. Sustain. Energy, vol. 9, no. 1, pp. 350–360, Jan. 2018.
POLs were proposed as a combo BMS in this article. In ad- [15] N. Cui, Y. Shang, Q. Zhang, and C. Zhang, “A direct multi-cells-to-multi-
dition to the normal voltage regulation operation, the system cells equalizer based on LC matrix converter for series-connected battery
provides two cell equalization mechanisms—Cell-to-Load and strings,” in Proc. IEEE Appl. Power Electron. Conf. Expo., San Antonio,
TX, USA, Mar. 2018, pp. 680–683.
Cell-to-Cell—on the basis of the load situation. Although the [16] X. Cao, Q.-C. Zhong, Y.-C. Qiao, and Z.-Q. Deng, “Multilayer modular
two balancing modes were demonstrated separately, there was a balancing strategy for individual cells in a battery pack,” IEEE Trans.
possibility of combining the two modes (i.e., supplying energy Energy Convers., vol. 33, no. 2, pp. 526–536, Jun. 2018.
[17] Y. Ma, P. Duan, Y. Sun, and H. Chen, “Equalization of lithium-ion battery
to POLs and charging the lowest SoC cell at the same time) in pack based on fuzzy logic control in electric vehicle,” IEEE Trans. Ind.
relative low load current situation. A central digital BMS con- Electron., vol. 65, no. 8, pp. 6762–6771, Aug. 2018.
troller was developed using an FPGA-based development board [18] Q. Ouyang, J. Chen, J. Zheng, and Y. Hong, “SOC estimation-based quasi-
sliding mode control for cell balancing in lithium-ion battery packs,” IEEE
to receive all sensed signals and send all control signals includ- Trans. Ind. Electron., vol. 65, no. 4, pp. 3427–3436, Apr. 2018.
ing the duty cycle PWM signals. Therefore, this architecture [19] S.-W. Lee, K.-M. Lee, Y.-G. Choi, and B. Kang, “Modularized design
presented the high modularization flexibility if the number of of active charge equalizer for Li-ion battery pack,” IEEE Trans. Ind.
Electron., vol. 65, no. 11, pp. 8697–8706, Nov. 2018.
battery cells is increased. As all isolated dc–dc converters could [20] A. M. Imtiaz and F. H. Khan, “Time shared flyback converter based
be implemented into this article, we selected a bidirectional fly- regenerative cell balancing technique for series connected Li-ion battery
back converter because it has a simple control and low cost for strings,” IEEE Trans. Power Electron., vol. 28, no. 12, pp. 5960–5975,
Dec. 2013.
realizing low-power functions. The entire system was verified [21] H.-S. Park, C.-E. Kim, G.-W. Moon, J.-H. Lee, and J. K. Oh, “Two-
by conducting experiments on a battery string involving three stage cell balancing scheme for hybrid electric vehicle lithium-ion battery
NCR 18650B Li-ion cells connected in series. Multiple flyback strings,” in Proc. IEEE Power Electron. Spec. Conf., Orlando, FL, USA,
Jun. 2007, pp. 273–279.
converters were connected with each other, which provided a [22] A. Xu, S. Xie, and X. Liu, “Dynamic voltage equalization for series-
voltage source of 3.3 V at 8 W. The peak efficiency of the entire connected ultracapacitors in EV/HEV applications,” IEEE Trans. Veh.
system was approximately 85% for the average measurement. Technol., vol. 58, no. 8, pp. 3981–3987, Oct. 2009.
[23] Y. Shang, C. Zhang, N. Cui, J. M. Guerrero, and K. Sun, “A crossed
The detailed specifications and comparison table were presented pack-to-cell equalizer based on quasi-resonant LC converter with adaptive
in this article as well. fuzzy logic equalization control for series-connected lithium-ion battery
strings,” in Proc. IEEE Appl. Power Electron. Conf. Expo., Charlotte, NC,
REFERENCES USA, Mar. 2015, pp. 1685–1692.
[24] C. Zhang, Y. Shang, Z. Li, and N. Cui, “An interleaved equalization
[1] G. Ning, R. E. White, and B. N. Popov, “A generalized cycle life model architecture with self-learning fuzzy logic control for series-connected
of rechargeable Li-ion batteries,” Electrochim. Acta, vol. 51, no. 10, battery strings,” IEEE Trans. Veh. Technol., vol. 66, no. 12, pp. 10923–
pp. 2012–2022, Feb. 2006. 10934, Dec. 2017.
[2] H. Rahimi-Eichi, U. Ojha, F. Baronti, and M.-Y. Chow, “Battery manage- [25] M.-Y. Kim, C.-H. Kim, S.-Y. Cho, and G.-W. Moon, “A cell selective
ment system: An overview of its application in the smart grid and electric charge equalizer using multi-output converter with auxiliary transformer,”
vehicles,” IEEE Ind. Electron. Mag., vol. 7, no. 2, pp. 4–16, Jun. 2013. in Proc. 8th Int. Conf. Power Electron., ECCE Asia, Jeju, South Korea,
[3] G.-L. Javier, R.-C. Enrique, M.-M. M. Isabel, and G.-M. A. Miguel, May 2011, pp. 310–317.
“Battery equalization active methods,” J. Power Sources, vol. 246, pp. [26] M. A. Hannan, M. M. Hoque, S. E. Peng, and M. N. Uddin, “Lithium-
934–949, Jan. 2014. ion battery charge equalization algorithm for electric vehicle applica-
[4] Texas Instrument, “16-Cell EV/HEV integrated battery monitor and pro- tions,” IEEE Trans. Ind. Appl., vol. 53, no. 3, pp. 2541–2549, May/Jun.
tector,” bq76PL455A-Q1 Datasheet, Apr. 2015 (Revised Nov. 2016). 2017.
[5] Analog Devices, “Lithium ion battery monitoring system,” AD7280 [27] C.-H. Kim, M.-Y. Kim, and G.-W. Moon, “A modularized charge equalizer
Datasheet, Apr. 2011. using a battery monitoring IC for series-connected Li-ion battery strings in
[6] Analog Devices, “Multicell battery monitors,” LTC6804 Datasheet, Oct. electric vehicles,” IEEE Trans. Power Electron., vol. 28, no. 8, pp. 3779–
2016. 3787, Aug. 2013.
[7] Maxim Integrated, “12 channel, high voltage sensor, smart data- [28] Maxim Integrated, “Integrated charger, dual main step-down con-
acquisition interface,” MAX11068 Datasheet, Jun. 2010. trollers, and dual LDO regulators,” MAX17085b Datasheet, Rev. 2, Mar.
[8] T. H. Phung, J. C. Crebier, and Y. Lembeye, “Voltage balancing converter 2012.
network for series-connected battery stack,” in Proc. IECON 38th Annu. [29] Texas Instrument, “High-efficiency LED backlight driver for notebooks,”
Conf. IEEE Ind. Electron. Soc., Montreal, QC, Canada, Oct. 2012, pp. LP8545 Datasheet, Apr. 2010 (Revised Dec. 2013).
3007–3013. [30] M. M. U. Rehman, F. Zhang, M. Evzelman, R. Zane, and D. Maksimovic,
[9] T. H. Phung, A. Collet, and J.-C. Crebier, “An optimized topology for “Control of a series-input, parallel-output cell balancing system for electric
next-to-next balancing of series-connected lithium-ion cells,” IEEE Trans. vehicle battery packs,” in Proc. IEEE 16th Workshop Control Model.
Power Electron., vol. 29, no. 9, pp. 4603–4613, Sep. 2014. Power Electron., Vancouver, BC, Canada, Jul. 2015, pp. 1–7.
[10] P. A. Cassani and S. S. Williamson, “Design, testing, and validation of [31] M. Shousha, T. McRae, A. Prodić, V. Marten, and J. Milios, “Design and
a simplified control scheme for a novel plug-in hybrid electric vehicle implementation of high power density assisting step-up converter with
battery cell equalizer,” IEEE Trans. Ind. Electron., vol. 57, no. 12, pp. integrated battery balancing feature,” IEEE J. Emerg. Sel. Topics Power
3956–3962, Dec. 2010. Electron., vol. 5, no. 3, pp. 1068–1077, Sep. 2017.
[11] F. Baronti, G. Fantechi, R. Roncella, and R. Saletti, “High-efficiency [32] M. Shousha, A. Prodić, V. Marten, and J. Milios, “Design and implemen-
digitally controlled charge equalizer for series-connected cells based on tation of assisting converter-based integrated battery management system
switching converter and super-capacitor,” IEEE Trans. Ind. Inform., vol. 9, for electromobility applications,” IEEE J. Emerg. Sel. Topics Power Elec-
no. 2, pp. 1139–1147, May 2013. tron., vol. 6, no. 2, pp. 825–842, Jun. 2018.

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: DIGITAL BATTERY MANAGEMENT DESIGN FOR POINT-OF-LOAD APPLICATIONS WITH CELL BALANCING 6375

[33] C. Jamerson, T. Long, and C. Mullett, “Seven ways to parallel a magamp,” Kai-Yu Hu (M’17) received the B.S. degree
in Proc. 8th Annu. Appl. Power Electron. Conf. Expo., San Diego, CA, in electrical engineering in 2015 from National
USA, Mar. 1993, pp. 469–474. Cheng Kung University, Tainan, Taiwan, where
[34] S. Luo, Z. Ye, R.-L. Lin, and F. C. Lee, “A classification and evaluation he is currently working toward the Ph.D. degree
of paralleling methods for power supply modules,” in Proc. 30th Annu. in electrical engineering with the Department of
IEEE Power Electron. Spec. Conf. Rec., Charleston, SC, USA, Jul. 1999, Electronic Engineering.
vol. 2, pp. 901–908. His research interests include fast transient
[35] Y. Huang and C. K. Tse, “Circuit theoretic classification of parallel con- dc–dc converters and power integrated circuit
nected DC–DC converters,” IEEE Trans. Circuits Syst. I, Reg. Papers, design.
vol. 54, no. 5, pp. 1099–1108, May 2007.
[36] J. Xu, M. Gao, Z. He, J. Yao, and H. Xu, “Design and study on the state
of charge estimation for lithium-ion battery pack in electric vehicle,” in
Proc. Int. Conf. Artif. Intell. Comput. Intell., Shanghai, China, Nov. 2009,
pp. 316–320.
[37] L. Ran, W. Junfeng, W. Haiying, and L. Gechen, “Prediction of state
of charge of Lithium-ion rechargeable battery with electrochemical
impedance spectroscopy theory,” in Proc. 5th IEEE Conf. Ind. Electron.
Appl., Taichung, Taiwan, Jun. 2010, pp. 684–688.
[38] V. Pop, H. J. Bergveld, D. Danilov, P. P. L. Regtien, and P. H. L. Notten,
Battery Management Systems, Accurate State-of-Charge Indication for
Battery-Powered Applications. New York, NY, USA: Springer, 2008.
[39] N. Otero, H. Rahimi-Eichi, J. J. Rodriguez-Andina, and M.-Y. Chow,
“FPGA implementation of an observer for state of charge estimation
in lithium-polymer batteries,” in Proc. Int. Conf. Mechatron. Control,
Jinzhou, China, Jul. 2014, pp. 1646–1651.
[40] F. Baranti, R. Roncella, R. Saletti, and W. Zamboni, “FPGA implemen-
tation of the mix algorithm for state-of-charge estimation of lithium-ion
batteries,” in Proc. IECON 40th Annu. Conf. IEEE Ind. Electron. Soc.,
Dallas, TX, USA, Oct. 2014, pp. 5641–5646.
[41] H.-T. Lin, T.-J. Liang, and S.-M. Chen, “Estimation of battery state of
health using probabilistic neural network,” IEEE Trans. Ind. Inform.,
vol. 9, no. 2, pp. 679–685, May 2013.
[42] F. Sun, X. Hu, Y. Zou, and S. Li, “Adaptive unscented Kalman filtering for
state of charge estimation of a lithium-ion battery for electric vehicles,” Chien-Hung Tsai (M’96) received the B.S. de-
Energy, vol. 36, no. 5, pp. 3531–3540, May 2011. gree in electrical engineering from the Tatung
[43] C.-H. Lin, C.-Y. Hsieh, and K.-H. Chen, “A Li-ion battery charger with Institute of Technology, Taipei, Taiwan, in 1985,
smooth control circuit and built-in resistance compensator for achieving and the M.S. and Ph.D. degrees in electrical
stable and fast charging,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 57, engineering from National Taiwan University,
no. 2, pp. 506–517, Feb. 2010. Taipei, Taiwan, in 1989 and 1996, respectively.
[44] K. E. Atkinson, An Introduction to Numerical Analysis. Hoboken, NJ, In 1989, he joined the Design Automation
USA: Wiley, 1989. Department of Electronic Research and Ser-
vice Organization, Industrial Technology and
Yin-Di Yang received the B.S. degree from Na- Research Institute, Hsinchu, Taiwan, where he
tional Chiao Tung University, Hsinchu, Taiwan, in was involved in providing computer aided design
2013, and the M.S. degree from National Cheng services for mixed-signal design teams. During 1996 to 2001, he served
Kung University, Tainan, Taiwan, in 2019, both in as an Associated Professor and Chairperson with the Department of
electrical engineering. Electronic Engineering, Lung-Hua University of Science and Technol-
During 2015 to 2016, he served as an Asso- ogy, Taoyuan, Taiwan. From 2001 to 2004, he served as the Deputy
ciated Electrical Engineer in charge of designing Manager and Director with ALi Inc., Taipei, Taiwan, where he was in
portable navigation devices with Garmin Corpo- charge of designing mixed-signal integrated circuits for the optical-drive
ration, New Taipei City, Taiwan. He is currently systems (CD and DVD). Since 2005, he has been in the Department of
working as a Digital integrated circuit (IC) Design Electrical Engineering, National Cheng Kung University, Tainan, Taiwan,
Engineer with Novatek Microelectronics Corpo- where he is currently a Professor. His main research interests include
ration, Hsinchu, Taiwan. His research interests include battery manage- mixed-signal and digitally controlled system-on-chip design for the power
ment system, power IC design, and digital IC design. management applications.

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on April 16,2024 at 11:58:10 UTC from IEEE Xplore. Restrictions apply.