IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO.

6, NOVEMBER 2004 1635

Sensorless Control of the BLDC Motors

From Near-Zero to High Speeds
Tae-Hyung Kim, Member, IEEE, and Mehrdad Ehsani, Fellow, IEEE

Abstract—This paper presents the theory and implementation of with the combination of two differential equations governing
a novel sensorless control technique for the brushless dc (BLDC) the stator phase windings, has been used for this purpose. With
motor. The proposed new sensorless drive method solves the the speed-independent position function, the commutation
problem of the sensorless BLDC motor drives at very low speeds.
It provides a highly accurate and robust sensorless operation from instants can be estimated from near zero (1.5% of the rated
near zero to high speeds. For this purpose, an approach, a new flux speed) to high speeds. Since the shape of the position function is
linkage function is defined, that is speed-independent. The validity identical at all speeds, it provides a precise commutation pulse
of the proposed method is verified through both simulation and at steady state as well as transient state. The proposed method
experimental results and discussion. does not rely on the measured back-EMF; hence the need for
Index Terms—Brushless dc (BLDC) motor, control, sensorless. external hardware circuitry for sensing terminal voltages has
been removed. Moreover, the proposed sensorless algorithm
I. INTRODUCTION is very suitable for implementation on low cost, fixed point
digital signal processors (DSPs). A simulation model for the

T HE permanent magnet brushless dc (BLDC) motor is in-

creasingly being used in computer, aerospace, military, au-
tomotive, industrial and household products because of its high
sensorless BLDC motor drive has been developed to validate
the proposed methods, and experimental results are shown to
prove the methods.
torque, compactness, and high efficiency. The BLDC motor is
inherently electronically controlled and requires rotor position
information for proper commutations of current. However, the II. REVIEW OF SENSORLESS CONTROL METHODS FOR
problems of the cost and reliability of rotor position sensors BLDC MOTORS
have motivated research in the area of position sensorless BLDC
Sensorless BLDC Motor control has been a research topic for
motor drives. Solving this problem effectively will open the way
the last two decades. Many papers and patents have been pub-
for full penetration of this motor drive into all low cost, high re-
lished on this topic [9]. The main methods published in litera-
liability, and large volume applications.
ture on this topic especially for the trapezoidal back-EMF type
In the last two decades, many sensorless drive solutions have
of BLDC motors can be classified as follows.
been offered to eliminate the costly and fragile position sensor
for BLDC motors with trapezoidal back-EMFs [1]–[10]. The
back-EMF voltage sensing [1], [2], back-EMF integration [3], A. Back-EMF Sensing Techniques [1], [2]
[4], flux estimation [5], [6], detection of the freewheeling diodes The BLDC motor has a trapezoidal shape of the induced
conduction [7], and motor modification technique [11] are the back-EMF in the stator winding. Monitoring the phase
main categories of past sensorless solutions for BLDC motors. back-EMF measured from terminal voltages in the silent phase,
However, none work well at all speeds without accuracy, relia- the zero crossing of the back-EMF can be detected. Since
bility, and complexity problems, especially at low speed range. back-EMF is zero at standstill and proportional to speed, the
Typically, practical minimum speed of the conventional sensor- measured terminal voltage that has large signal-to-noise ratio
less drive is around 10% of the rated speed. Also, the position cannot detect zero crossing at low speeds. Also, the estimated
error from a phase shifter in transient state deteriorates the per- commutation points that are shifted by 30 from zero crossing
formance of sensorless drive [1], [2], [7]. These drawbacks of of back-EMFs have position error in transient state. With
sensorless BLDC motor drives have been an obstacle to the use terminal voltage-sensing method [1], an operating speed range
of this motor in various industrial applications. is typically around 1000–6000 RPM. The third harmonic
This paper presents a novel sensorless position detection back-EMF sensing method [2] provides wider speed range than
technique with a new physical concept based on a speed-inde- the terminal voltage-sensing method.
pendent position function for the BLDC motors. A physically
insightful speed-independent function of flux linkage, along B. Back-EMF Integration Technique [3], [4]
In this method, position information is extracted by inte-
Manuscript received March 14, 2003; revised November 21, 2003. This work
was presented at the Applied Power Electronics Conference and Exposition, grating the back-EMF of the silent phase. Integration starts
2003. Recommended by Associate Editor J. Ojo. when the open phase’s back-EMF crosses zero. A threshold is
The authors are with the Department of Electrical Engineering, set to stop the integration that corresponds to a commutation
Texas A&M University, College Station, TX 77843-3128 USA (e-mail:
[email protected]). instant. This method also has a problem at low speeds because
Digital Object Identifier 10.1109/TPEL.2004.836625 of the error accumulation problem.
0885-8993/04$20.00 © 2004 IEEE
1636 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 6, NOVEMBER 2004

C. Flux Linkage-Based Technique [5], [6] Where, the first term, , represents the self flux
In this method, the flux linkage is calculated using measured linkage of phase A; the second and third term stand for the
voltages and currents. The fundamental idea is to take the mutual flux linkage with phases B and C; and the fourth term,
voltage equation of the machine and by integrating the applied , stands for the flux linkage due to the permanent magnet
voltage and current, flux can be estimated. From the initial that is attached on the rotor. The first three terms of (2) are
position, machine parameters, and the flux linkages’ relation- function of current and position, and the is the function
ship to rotor position, the rotor position can be estimated. This of position. Therefore, the flux profile has a close relationship
method also has significant estimation error in low speed. with the dynamic performances. Since the BLDC motor uses
Improper error of parameters and sampled current is reason for permanent magnets, the permeability of the attached permanent
accumulation error at low speeds in which the voltage equation magnet is considered as that of air, and hence, typically, the
is integrated in a relatively large period of time. motor has small inductance variation. Therefore, if the motor
is operated within the rated current, the saturation effect of in-
D. Freewheeling Diode Conduction [7] ductance due to current level is usually neglected. For the sur-
face-mounted permanent magnet (SMPM) type of BLDC mo-
This method uses current flowing through a freewheeling tors, the permanent magnets are roundly attached on the round
diode in silent phase. For a short period after reaching zero surface of the rotor, and hence, the inductance variation by rotor
crossing of the back-EMF in silent phase, a tiny current is position is negligibly small. Based on the characteristics of the
flowing through freewheeling diode; during the active phase SMPM type of BLDC motors, the flux profile can be simplified
switches are turned off under alternate chopper control. This as (3) along with the following assumptions:
silent phase current starts to flow in the middle of the commuta-
1) the motor is operated within the rated condition and hence
tion interval, which corresponds to the point where back-EMF
the saturation effect due to current level is neglected;
of the open phase crosses zero. This method also has position
2) the leakage inductance is negligibly small and hence
error of commutation points in transient state. The most serious
neglected;
drawback of this method is the requirement of six additional
3) iron losses are negligible;
isolated power supplies for the comparator circuitry to detect
current flowing in each freewheeling diode.
(3)
III. PROPOSED SENSORLESS CONTROL METHOD
Most popular and practical sensorless drive methods for Substituting (3) into (1) gives
BLDC motors rely on speed-dependent back-EMF. Since the
back-EMF is zero or undetectably small at standstill and low (4)
speeds, it is not possible to use the back-EMF sensing method
In balanced three-phase BLDC motors
in the low speed range. Also, the estimated commutation points
that are shifted by 30 from zero crossing of back-EMFs have
position error in transient state. The flux estimation method (5)
[5], [6] also has significant estimation error at low speed, in
which the voltage equation is integrated in a relatively large Where, , and represent the self-inductance and mu-
period of time. In this paper, to overcome the above drawbacks, tual-inductance, respectively. Substituting (5) into (4) gives
a novel method, based on a new speed-independent function, is
proposed, one that is based on a new physical insight. (6)

A. Principle of the Proposed Method For a balanced wye-connected BLDC motors, the three-phase
currents always meet the following equation:
Each active phase in an ac motor can be described by a first
order differential equation. The general voltage equation of one (7)
of the active phases is given by
Using (7), the (6) is simplified as
(1)

where is the active phase voltage, is the phase resistance, (8)

is the phase current, is the rotor position, is the
total flux linkage of the active phase, and “ ” is the number of where represents the phase inductance under balanced condi-
phases. The flux linkage in the active phase includes both self tions. The last term in the voltage equations is so called back-
and mutual flux linkages. For three-phase BLDC motors, the EMF, and the term is redefined as
total flux linkage of the phase A is

(2) (9)
KIM AND EHSANI: SENSORLESS CONTROL OF THE BLDC MOTORS 1637

where stands for the back-EMF constant. It is seen that

in (8) is expressed as a constant value times a flux linkage func-
tion that is changed only by the rotor position as shown in (9).
The is a flux linkage form function that is a function
of rotor position. Since some manufacturers do not provide a
motor neutral point, the line-to-line voltage equations are used
as follows:

(10)

(11)

(12)
Fig. 1. G function waveforms.
where stands for the instantaneous speed. The is a
line-to-line flux linkage form function that is a function of the
functions ( , and ) wave-
rotor position. Now we define a new function, , as
form based on (15). The standard commutation instant is when
(13) the functions are changed from positive infinity to nega-
tive infinity as in Fig. 1.
Then, can be derived as The dc link bus voltage, and currents are sampled with
40-kHz sampling frequency. To avoid the deterioration of the
G function signal due to the noise from measured values, the
sparse calculation method is applied. The method is especially
(14) useful at very low speeds such as 0.25 Hz (15 RPM). The
number of sampled data in the 60 commutation period reaches
Since the function itself has a one to one relationship to more than 10 000 at around 15 RPM. If the calculation
with rotor position, it is possible to use this function for position frequency for G function is set to have 0.5 resolution, one
estimation. But as shown in (14), to know the function, calculation cycle can include more than 100 sampled data. The
the instantaneous speed term, that is unknown for dynamic op- collected data in one cycle are arrived at one sample which is
erations, is required to calculate the function. time average of the data to decrease noise in the calculation
of , phase currents, and di/dt terms. To polish the critical
B. Derivation of the Speed-Independent Position Function
values in the G function especially at low speeds, the calcu-
To eliminate the instantaneous speed term, , that causes lation frequency is modified proportionate to speed to collect
trouble in using the function for position estimation, more data at low speeds. The calculation frequency is changed
one line-to-line function is divided by another line-to-line to maintain 0.5 resolution while the speed is changing. This
function, and the divided new speed independent function delay due to the sparse calculation can cause the deterioration
is named . For example of the acceleration or deceleration performance. The speed-in-
dependent function method along with the sparse calculation
provides 15 Hz/s acceleration or deceleration performance.

C. Commutation Strategy
After careful investigation of the shape of function, to
(15) provide the best indirect position-function that is speed-inde-
The digitized expression of the function is (16), shown at the pendent with high sensitivity at each commutation point, the
bottom of the page. ratio of two line-to-line functions is sequentially utilized.
Where, , and are switching status of phase From the sequential combination of two line-to-line func-
A, B, and C and known values by the controller. Fig. 1 shows the tions at each commutation interval, the ideal function that

(16)
1638 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 6, NOVEMBER 2004

Fig. 3. Estimation of the position using the characterized G function.

that the commutation signal can be generated at the peak point

that is the most sensitive part of the function. From the
available stator current and dc link voltage, the function is
computed based on the equations of Table I in real time.
Since the waveform of the function is identical at the
entire speed range, it can be characterized at steady state in a
look-up table, and used as a position reference for sensorless
operation at all speeds. If the control purpose is just to follow a
square wave reference current for constant torque, the commuta-
tion points that are detected at the peak point of the func-
tion are sufficient as position information, but for special pur-
Fig. 2. Sensorless operation using the proposed method (from top to bottom:
line-to-line H functions; G function; phase current; generated commutation
pose control strategies, for example, an advanced angle control
signal). for the field weakening operation, it is required to know more
position information between commutation points.
Fig. 3 shows the procedure to estimate position between com-
TABLE I
G FUNCTION AT EACH MODE mutation points. From the measured current and line-to-line
voltages that are calculated by multiplication of the dc bus
voltage and switching status, function is computed at
each mode. Then, position is estimated by the look-up table of
the characterized function.

D. Current Control
To control currents, a simple PI or hysteresis controller can
be used for the proposed sensorless method. Also, to calculate
line-to-line voltages in equations of Table I at each mode, the
three phase activated current control is utilized. It means that a
silent phase current is controlled as zero.

E. Starting Procedure
In typical sensorless operation of the BLDC motor, the forced
is independent of speed and contains continuous position infor-
alignment of the rotor is a way of setting an initial position
mation is derived. It means that only the positive part of the
[1]–[4]. Two phases are excited to cause the rotor to shift and
function waveform in Fig. 1 is utilized. It is noted that
lock into position. If the rotor is not in the desired position, the
the standard commutation point is the peak point of ideal
forcing torque from the excited phases causes it to rotate and
function that is the most sensitive part of the function as shown
stop at the desired position. After energizing two of the three
in Fig. 2. Therefore, the position estimation based on the
motor phases for enough time to ensure the rotor will lock into
function can provide the best accuracy at commutation points
position, the next commutation signal advancing the switching
of the current which are the most important instants for the sen-
pattern by 60 is given, then immediately the proposed sensor-
sorless operation.
less algorithm using the function is applied to detect the
Table I shows the position estimation equations at each mode
next commutation instant. The 60 rotor movement is enough
in Fig. 2. Two line-to-line functions are used in each mode
to detect the commutation instant using the position estimation
as Table I. As shown in Fig. 2, at mode 1, is used as
equations in Table I. After the first detection of the commuta-
a position estimation equation; after a 60 electrical angle, at
tion point, both torque and speed control is possible using the
mode 2, is utilized. The time duration of each mode in
estimated speed from time duration of each commutation point.
Fig. 2 corresponds to 60 by electrical angles.
When the function reaches a predefined threshold value,
the motor is commutated. The threshold value is defined based IV. SIMULATION RESULTS
on the current rising time and desired advanced angle. From the The proposed sensorless control algorithm has been verified
sequential combination of the functions based on Table I, through numerical simulation. Figs. 4 and 5 illustrate the perfor-
the function can be made as shown in Fig. 2. It is noted mance of the proposed sensorless method at 50 and 1000 RPM,
KIM AND EHSANI: SENSORLESS CONTROL OF THE BLDC MOTORS 1639

where , and are the air-gap flux density,

rotor radius, length, speed, number of turns for phase A,
harmonic coefficient, and rotor position, respectively.
The current commutation is performed when the func-
tion reaches a predefined threshold value. The threshold value
can be modified for proper advanced angle control. Based on
the function, the virtual commutation signal is generated
as in Figs. 4 and 5.
Fig. 6 shows the starting performance and speed response of
the proposed sensorless control method. Fig. 6(a) and (b) show
the excited current and speed waveform for the forced alignment
procedure and transient state. The currents are initially excited
from phase A to B and controlled with the hysteresis control
strategy. From zero to 3.5 s, the motor is locked into a position
and the next commutation signal is given to run the motor and to
detect the rotor position. During the first 60 periods, the
function is computed to estimate the rotor position and detect
the first commutation point. Fig. 6(c) shows the function
waveform for the starting of the motor. Even in transient state,
Fig. 4. Simulation results of the proposed sensorless method at 50 RPM (from
top to bottom: phase current; G function; virtual commutation signal; speed). precise commutation points can be obtained by the func-
tion. The comparison between the virtual Hall sensor that is gen-
erated by the function and the real Hall sensor in transient
state is presented in Fig. 6(d).

V. DSP-BASED IMPLEMENTATION AND

EXPERIMENTAL RESULTS
Fig. 7 shows an overall system block diagram of the proposed
new sensorless drive. In this experiment, the PI controller is used
for each phase current control. The phase terminal voltages are
calculated based on the dc link voltage, switching status of the
inverter, and voltage drops of the inverter switches. When the
dc link voltage has a large ripple due to the size of a dc link
capacitance, a voltage sensor is required to sense the dc link bus
voltage.
The experimental setup consists of a 1-HP, 5.5-A, 1000 RPM,
four-pole BLDC motor drive loaded with a permanent magnet
dc generator as shown in Fig. 8. The entire drive system is con-
trolled by a low-cost, fixed-point digital signal processor (DSP),
Fig. 5. Simulation results of the proposed sensorless method at 1000 RPM TMS320F243.
(from top to bottom: phase current; G function; virtual commutation signal;
speed). Table II shows the four-pole BLDC motor specification to
examine the performance of the developed sensorless drive
technique.
respectively. The phase current can be controlled by any con-
trol strategy such as PI or hysteresis controller. In this simula- Figs. 9 and 10 show the experimental waveforms of the
tion, each phase current is controlled with the hysteresis control function, virtual Hall sensor signal made by the function,
strategy. phase current, and speed, respectively, at 50 and 20 RPM. Also,
The back-EMF, , is modeled for simulation Fig. 11 shows the phase current, function, and speed wave-
using Fourier series based on the measurement of BLDC motor form at 900 RPM. The function is computed in the DSP
that is used in experiment as and sent to a D/A (digital/analog) converter to observe it through
an oscilloscope.
To observe the accuracy of estimated commutation points, the
(17)
virtual Hall sensor signal that is generated by the function
is compared with the real Hall sensor signal as in Fig. 12.
For fast simulation, only the first three terms of the Fourier Each edge indicates the commutation instant. It is noted that
series are utilized to construct the back-EMF profile as the position estimation error of the virtual Hall sensor is less
than 5 by electrical angle even at this low speed.
Fig. 13 shows the starting performance of the proposed sen-
(18) sorless method. The machine is initially aligned by exciting two
1640 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 6, NOVEMBER 2004

Fig. 6. Speed response and performance for the starting and transient state (current control mode): (a) phase current, (b) speed response, (c) G function, and
(d) virtual and real Hall sensor signals.

TABLE II
MOTOR SPECIFICATION (LL: LINE-TO-LINE)

Fig. 7. Block diagram of the proposed sensorless drive.

phases from zero to 2 s as in Fig. 13. After the forced align-

ment, during the first 60 commutation interval, the func-
tion is computed and estimate commutation points as Fig. 13.
The starting procedure from standstill to 50 RPM as in Fig. 13 Fig. 8. Experimental test-bed: (a) overall system and (b) BLDC motor and dc
is smooth and has no serious distortion of speed. generator load.
Fig. 14 shows the performance of the sensorless operation
under a sudden load change. As shown in Fig. 7, a dc generator torque is controlled by current control, and from a step change
is used as a load and connected by a resistance bank. The motor of the resistance value at a certain time, the load is changed.
KIM AND EHSANI: SENSORLESS CONTROL OF THE BLDC MOTORS 1641

Fig. 9. Sensorless operation at 50 RPM [200 ms/div.] (from top to bottom: G

Fig. 11. Sensorless operation at 900 RPM [20 ms/div.] (from top to bottom:
function; virtual Hall sensor signal [5 V/div.]; phase current [2 A/div.]; speed
phase current [10 A/div.]; G function; speed [600 RPM/div.]).
[125 RPM/div.]).

Fig. 12. Comparison of the virtual Hall sensor signal and real Hall sensor
Fig. 10. Sensorless operation at 20 RPM [200 ms/div.] (from top to bottom: signals at 50 RPM [100 ms/div.] (from top to bottom: phase current [2 A/div.];
G function; virtual Hall sensor signal [5 V/div.]; phase current [2 A/div.]; speed G function; virtual Hall sensor signal [5 V/div.]; real Hall sensor signals
[125 RPM/div.]). [5 V/div.]).

In this case, the motor speed is changed from 50 to 180 RPM. in turn, causes a decrease of the motor efficiency. The instanta-
Under the load change, in transient period, reliable sensorless neous torque and power equations of BLDC motors are
operation is guaranteed based on the speed-independent
function.
Theoretically, the speed independent function is independent
with the load variation if the operational speed is bigger than the (19)
stated minimum speed. But, in practical operation, the sensor-
As an example, Fig. 15 shows the ideal shape of back-emf,
less drive has the acceleration/deceleration performance such
current, and total torque under commutation point error in
as 15 Hz/s (mechanical frequency) due to the limited resolu-
indirect position sensing of BLDC motors. Based on Fig. 15, the
tion caused by calculation frequency. The sensorless drive can
maximum average torque and power per electrical cycle without
handle load changes that make less speed variation than 15 Hz/s.
any position errors are calculated as

VI. ESTIMATION ERROR OF THE PROPOSED METHOD (20)

The results of the position estimation error are the ripple of (21)
total torque and decrease of the available torque per commu-
tation period, hence, the decrease of torque/ampere ratio. This, where represents the back-EMF constant.
1642 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 6, NOVEMBER 2004

Fig. 13. Sensorless starting [0.5 s/div.] (from top to bottom: G function; virtual
Hall sensor signal [5 V/div.]; phase current [2 A/div.]; speed [125 RPM/div.]).

Fig. 15. Effect of rotor position estimation error on the electrical torque:
(a) available torque and a load line and (b) three-dimensional (3-D) waveform.

TABLE III
AVAILABLE TORQUE PER COMMUTATION PERIOD

dc link voltage measurement

error ;
Parameter error (phase
Fig. 14. Performance under sudden load change [200 ms/div.] (from top
resistance);
to bottom: phase current [2 A/div.]; G function; virtual Hall sensor signal Parameter error (phase
[5 V/div.]; real Hall sensor signal [5 V/div.]). inductance);
and Current measurement error
If there is a commutation angle error, , the available torque .
per commutation period and motor efficiency under the commu- Based on the above error sources, the worst-case numerator
tation angle error can be calculated as Table III. and denominator are simulated and compared with the numer-
To calculate the estimated position error, the numerator and ator and denominator that do not contain the error sources. In
denominator of (16) can be represented with possible error this case, the sign of each error sources is defined to make the
sources as follows. worst scenario. With this approach, the estimated position error
1) Numerator with possible error sources as (22), shown at can be calculated based on the amount of each error sources.
the bottom of the page. If the worst-case position error is calculated, the avail-
2) Denominator with possible error sources as (23), shown able torque per commutation period can be calculated
at the bottom of the next page, where the error sources are based on the Table III. As an example, Fig. 16(a) shows

(22)
KIM AND EHSANI: SENSORLESS CONTROL OF THE BLDC MOTORS 1643

(a) (b)
Fig. 16. Available torque per commutation period based on the current reference with a dc generator load.

the available torque and speed characteristic with as-

sumed error sources (
A, dc link voltage measurement error V,
resistance variation % inductance variation %)
when different current references are given. To define the
minimum speed of the sensorless drive method under pos-
sible error sources, the characteristics of the electromagnetic
torque, load torque, and cogging torque of the BLDC motor
are carefully investigated. The cogging torque is the function
of rotor position. Since the cogging torque can oppose the
electromagnetic torque at certain position, the maximum value
of cogging torque is added to the load torque and compared
with the electromagnetic available torque to define the speed
limitation as shown in Fig. 16(a). A dc generator is used as load
in the experimental test-bed. In this case, the load torque has
Fig. 17. Impact of inductance error on the position estimation error.
a linear relationship with the motor speed. The dotted straight
line presents the dc generator load plus magnitude (maximum
value) of the cogging torque. In the proposed method, the constant inductance value is as-
The intersected points in Fig. 16(a) indicate the operating sumed and used for the G function calculation. In actual case,
points of the sensorless drive. In this case, the minimum speed the inductance value is varying by rotor position having max-
is at around 15 RPM where the dotted line becomes a tangen- imum and minimum value because of the local saturation. How-
tial line of an available torque curve. Fig. 16(b) shows the 3-D ever, for the SMPM type of BLDC motors, the variation is usu-
waveform of Fig. 16(a). The available electromagnetic torque ally small (less than 20% of the dc offset value). Authors used
and load torque surfaces are shown in 3-D space. The inter- the dc value (average value) of the inductance that is initially
sected line of two surfaces shows the operating speed when the measured for the G function calculation. The resultant position
different current command is given. With these assumed error estimation error due to the constant inductance value is shown
sources, the minimum speed is at around 15 RPM. in 3-D space as Fig. 17 based on (19) and (20).

(23)
1644 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 19, NO. 6, NOVEMBER 2004

VII. CONCLUSION
This paper presented a novel sensorless drive method for
BLDC motors. This technique makes it possible to detect the
rotor position over a wide speed range from near zero to high
speeds. The capability of position detection at around 1.5% of
the rated speed makes the starting procedure much simpler than
conventional methods. Also, the proposed approach provides
a precise commutation pulse even in transient state because
of the speed-independent characteristic of the function.
From the simulation and experimental results, the validity
of the developed sensorless drive technique using the new
speed-independent function is successfully verified. Based on
the successful experimental results, the proposed sensorless al-
gorithm with wider speed range can be implemented in various
industrial and other applications.
Fig. 18. Commutation angle error at 20 RPM (from top to bottom: phase
current [1 A/div.]; G function; virtual Hall sensor signal [5 V/div.]; absolute REFERENCES
value of the commutation angle error [2.5 (mechanical)/div.]).
[1] K. Iizuka et al., “Microcomputer control for sensorless brushless motor,”
IEEE Trans. Ind. Applicat., vol. 27, pp. 595–601, May/June 1985.
[2] J. Moreira, “Indirect sensing for rotor flux position of permanent magnet
ac motors operating in a wide speed range,” IEEE Trans. Ind. Applicat.,
vol. 32, pp. 401–407, Nov./Dec. 1996.
TABLE IV [3] R. C. Becerra, T. M. Jahns, and M. Ehsani, “Four-quadrant sensorless
WORST-CASE COMMUTATION ANGLE ERROR (EXPERIMENTAL RESULT) brushless ECM drive,” in Proc. IEEE APEC’91 Conf., 1991, pp.
202–209.
[4] T. M. Jahns, R. C. Becerra, and M. Ehsani, “Integrated current regulation
for a brushless ECM drive,” IEEE Trans. Power Electron., vol. 6, pp.
118–126, Jan. 1991.
[5] N. Ertugrul and P. Acarnley, “A new algorithm for sensorless operation
of permanent magnet motors,” IEEE Trans. Ind. Applicat., vol. 30, pp.
126–133, Jan./Feb. 1994.
[6] R. Wu and G. R. Slemon, “A permanent magnet motor drive without
a shaft sensor,” IEEE Trans. Ind. Applicat., vol. 27, pp. 1005–1011,
Sept./Oct. 1991.
The position error is less than 0.3 by mechanical angle in [7] S. Ogasawara and H. Akagi, “An approach to position sensorless drive
the speed range from 20 to 1000 RPM if the applied current for brushless dc motors,” IEEE Trans. Ind. Applicat., vol. 27, pp.
is less than rated value. The position error is calculated when 928–933, Sept./Oct. 1991.
[8] A. Consoli, S. Musumeci, A. Raciti, and A. Testa, “Sensorless vector and
the sparse calculation method is utilized. As shown in Fig. 17, speed control of brushless motor drives,” IEEE Trans. Ind. Electron., vol.
the impact of the inductance value error (below 20%) on the 41, pp. 91–96, Feb. 1994.
proposed position estimation method is negligible. [9] J. P. Johnson, M. Ehsani, and Y. Guzelgunler, “Review of sensorless
methods for brushless dc,” in Proc. IEEE IAS’99 Conf., vol. 1, 1999, pp.
In Fig. 18, the experimental results about the commuta- 143–150.
tion angle error using the experimental test-bed are shown. [10] M. Tomita, M. Satoh, H. Yamaguchi, S. Doki, and S. Okuma, “Sensor-
The bottom signal in Fig. 18 shows the absolute value of the less estimation of rotor position of cylindrical brushless dc motors using
eddy current,” in Proc. IEEE International Workshop on Advanced Mo-
commutation angle error at 20 RPM. The worst commutation tion Control, vol. 1, 1996, pp. 24–28.
angle errors at different speeds based on the experimental [11] T. Kim and M. Ehsani, “Advanced sensorless drive techniques for brush-
less dc motors,” U.S. Patent 60/438,949, 2004.
results are shown in Table IV. To find the value of worst-case
at each speed, the motor is operated for a long time with the
sensorless algorithm, and the worst value of the commutation
angle error is stored in a memory of the DSP. The stored value
is assumed as the worst-case commutation angle error at that
Tae-Hyung Kim (S’00–M’04) received the B.S. and
speed. The minimum speed of the sensorless operation depends M.S. degrees from Korea University, Seoul, in 1994
on measurement error and the influences of the motor parameter and 1998, respectively, and the Ph.D. degree from
Texas A&M University, College Station, in 2003, all
variations. With the test-bed conditions in Fig. 8, sensorless in electrical engineering.
operation down to 15 RPM is realized. If more accurate mea- Since 2003, he has been a Postdoctoral Research
surement devices and an A/D converter that has higher data bits Associate with the Advanced Vehicle and Motor
Drives Research Center, Texas A&M University.
are used, the minimum speed can be further decreased. His research interests include sensorless control of
With the proposed method, the minimum speed of the sen- BLDC motor drives, power conversion system of
BLDC generator, hybrid electric vehicles and sys-
sorless operation has been improved to be around 1.5% of the tems, inverter topology for ac motor drives, and application of DSP controllers
rated speed. for motor drives.
KIM AND EHSANI: SENSORLESS CONTROL OF THE BLDC MOTORS 1645

Mehrdad Ehsani (S’70–M’81–SM’83–F’96) re-

ceived the Ph.D. degree in electrical engineering
from the University of Wisconsin-Madison in 1981.
Since 1981, he has been with Texas A&M Univer-
sity, College Station, where he is now a Professor of
electrical engineering and Director of the Advanced
Vehicle Systems Research Program. He is the author
of over 300 publications in pulsed-power supplies,
high-voltage engineering, power electronics, and
motor drives. He is coauthor of three books on power
electronics and motor drives and a contributor to an
IEEE Guide for Self-Commutated Converters and other monographs and books.
He is the author of over 20 U.S. and EC patents. His current research work is
in power electronics, motor drives, hybrid electric vehicles, and systems.
Dr. Ehsani received the Prize Paper Award in Static Power Converters and
motor drives at the IEEE-Industry Applications Society at the 1985, 1987, and
1992 Annual Meetings, the James R. Evans Avant Garde Award from the IEEE
Vehicular Technology Society, in 2001, and the IEEE Undergraduate Teaching
Award in 2003. In 1992, he was named the Halliburton Professor in the College
of Engineering at A&M. In 1994, he was also named the Dresser Industries Pro-
fessor in the same college. In 2001 he was named the Dow Chemical Faculty
Fellow of the College of Engineering at Texas A&M University. He is an As-
sociate Editor of the IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS and
the IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY. He has been a member
of IEEE Power Electronics Society AdCom, past Chairman of PELS Educa-
tional Affairs Committee, past Chairman of IEEE-IAS Industrial Power Con-
verter Committee and past chairman of the IEEE Myron Zucker Student-Faculty
Grant Program. He was the General Chair of IEEE Power Electronics Specialist
Conference for 1990. He has been an IEEE Industrial Electronics Society and
Vehicular Technology Society Distinguished Speaker and IEEE Industry Appli-
cations Society past Distinguished Lecturer. He is Chairman of Vehicular Tech-
nology Society Vehicle Power and Propulsion Committee. He was elected to the
Board of Governors of IEEE-VTS in 2003. He is also a Registered Professional
Engineer in the State of Texas.