This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

A Rotor Position and Speed Estimation Method Using

an Improved Linear Extended State Observer
for IPMSM Sensorless Drives
Yanping Zhang, Student Member, IEEE, Zhonggang Yin, Member, IEEE, Cong Bai,
Gaolin Wang, Senior Member, IEEE, and Jing Liu

to the different operation speed, IPMSM sensorless drive

Abstract—A traditional linear extended state observer (LESO) technology can be divided into: methods based on back
does not have enough capacity to handle the rapidly-varying back electromotive force (EMF) model or flux linkage model in the
electromotive force (EMF). This key weakness limits the range of medium and high speed operation, and methods based
application of LESO in internal permanent magnet synchronous on salient characteristics in the range of zero speed and low
motor (IPMSM) sensorless control system. In this paper, an
speed operation [5], [6].
IPMSM sensorless drive method using an improved LESO
(ILESO) is proposed. The ILESO consists of pure integral and
IPMSM sensorless control technology based on salient
rapidly-varying sinusoidal interference estimator. The characteristics mainly includes high frequency signal injection
rapidly-varying sinusoidal interference estimator is added to the (HFSI) method. HFSI method needs to inject additional high
unknown interference estimation loop to estimate rapidly-varying frequency voltage into IPMSM. Since the rotor position
back EMF. And the frequency of the rapidly-varying sinusoidal information can be extracted from the high frequency current
interference estimator changes adaptively with the operating and only related to the salient characteristics of the rotor, the
frequency of the motor to guarantee the desired estimation HFSI method can well estimate the rotor position at zero speed
accuracy of back EMF in wide-speed range. This design makes it and low speed [7], [8]. However, the frequency range of the
possible to observe the back EMF in a relatively low bandwidth,
injected high frequency signal is between hundreds of hertz and
thereby solving the trade-off between LESO bandwidth and high
frequency noise filtering. Then, an enhanced phase locked loop several thousand hertz [9], which belongs to the sensitive range
(EPLL) is presented to improve the ability of tracking the of the human ear [10]. In addition, injected high frequency
rapidly-changing speed. The effectiveness of the proposed method signal will produce electromagnetic interference to the
is verified at a 2.0-kW IPMSM sensorless drive. surrounding electronic equipment and the additional torque
ripple cannot be ignored sometimes [11].
Index Terms—Interior permanent magnet synchronous motors, Sensorless drive technology based on back EMF model
sensorless drive, extended state observer, enhanced phase locked estimates the rotor position through the fundamental frequency
loop. back EMF [12]. Because there is no need for rotor salient
information and additional test signal, the position information
I. INTRODUCTION can be estimated only by fundamental motor model. And this

T HE interior permanent magnet synchronous motor

(IPMSM) has attracted more and more attention because of
its small size, large low-speed output torque and small air gap of
method is relatively simple to implement, so IPMSM sensorless
drive method based on back EMF is widely used. The back
EMF observer method main includes model reference adaptive
magnetic circuit [1], [2]. It has been widely used in subway, (MRAS) [13], extended Kalman filter (EKF) [14], sliding mode
electric vehicle and other fields [3]. IPMSM sensorless control observer (SMO) [15], extended state observer (ESO), etc. The
technology has achieved remarkable results in reducing system mismatch of motor parameters will lead to the performance
cost and improving system robustness [4]. Generally, according degradation of sensorless control based on MRAS. The EKF
sensorless method has good estimation performance in noisy
Manuscript received November 22, 2020; revised March 22, 2021; accepted
environments. However, due to the complexity of the algorithm,
May 19, 2021. This work was supported in part by Shaanxi Outstanding Youth
Fund under Grant 2020JC-40, in part by Key Laboratory of Power Electronic it is not easy to implement on a low-cost processor. Chattering
Devices and High Efficiency Power Conversion in Xi’an under Grant inevitably exists in SMO, which will reduce the steady state
2019219814SYS013CG035, in part by China Postdoctoral Science Foundation accuracy of drive system.
under Grant 2020M683524, in part by Nature Science Basic Research Plan in The active disturbance rejection control (ADRC) was
Shaanxi Province under Grant 2020JM-452, in part by Nature Science Basic proposed by Jingqing Han in the 1990s [16]. The ADRC is less
Research Plan in Shaanxi Province under Grant 2021JQ477, in part by Project
supported by the Technology Plan of Xi’an, China under Grant GXYD14.20.
dependent on the accurate mathematical model and has strong
(Corresponding author: Zhonggang Yin.) anti-interference ability [17]. Therefore, it has broad prospects
Y. Zhang, Z. Yin, C. Bai, and J. Liu are with the Department of Electrical to apply ADRC to IPMSM sensorless control system [18].
Engineering, Xi'an University of Technology, Xi'an, 710048, China (e-mail: ADRC consists of a tracking differentiator, an extended state
zhangyanpingmao@126.com;zhgyin@xaut.edu.cn; baicong123@xaut.edu.cn; observer (ESO), and a nonlinear state error feedback controller.
jingliu@xaut.edu.cn).
As the core of ADRC, the ESO has a decisive influence on the
G. Wang is with the School of Electrical Engineering and Automation,
Harbin Institute of Technology, Harbin, 150001, China (e-mail: disturbance rejection performance of the system [19]. The ESO
WGL818@hit.edu.cn). can estimate not only the known state variables, but also the

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

unknown interference of the system. Therefore, the ESO is can be rewritten as

widely used in uncertain systems [20].The ESO can be divided  px1 = x2 + b0 u + f ( x,t )
into linear ESO (LESO) and nonlinear ESO (NESO) [21].  (2)
NESO has high estimation accuracy and fast response speed,  px2 = pd (t )
but its application in engineering field is hindered due to the where x1=x, x2=d(t). Define h(t)=pd(t), and |h(t)|≤h0, h0 is the
large number of setting parameters and the complexity of boundary of h(t). The LESO of system (2) can be expressed as
stability analysis [22]. In contrast, the parameter setting of the ε = xˆ1 − x1

LESO is convenient, and the stability analysis is easy, so it is  pxˆ1 = xˆ2 +b0 u + f ( xˆ1 ,t ) − β1 ε (3)
popular in engineering practice [23].  pxˆ = − β ε
 2 2
In recent years, more and more researchers pay attention to
where x̂1 the estimated x1, x̂2 is the estimated unknown
ESO as a sensorless driving technology. In [24], a LESO was
proposed to estimate the equivalent back EMF of IPMSM, and interference d(t), and initially the approximation x̂2 =constant. ε
the rotor position and speed were obtained by the equivalent is the estimation error. β1 and β2 are gains of the LESO.
back EMF. A LESO was presented to estimate the speed of The model of the IPMSM in the αβ axes (further detailed in
induction motors in [25]. An improved ADRC consists of two Appendix) [12], [29], [30] can be expressed as
LESO was proposed in [26]. Two LESOs are used to estimate uα   Rs 0  iα   pLq 0  iα  eα 
EMF and internal interference, respectively. However, due to  =   i  +  0    + e  (4)
the coupling between the two LESOs, it is impossible to  β 
u 0 R s β  pLq   iβ   β
estimate back EMF and extended disturbance completely with
independently. In [27], the second-order LESO is constructed to  eα   −sinθr 
realize the observation of the equivalent back EMF, and then the   = ( Ld − Lq ) id ωr + ωr ψ f    (5)
e
 β  cosθr 
rotor position is calculated by the inverse tangent function. In
order to combine the advantages of the LESO and NESO, a where [uα, uβ]T and [iα, iβ]T are the α-axis and β-axis stator
linear-nonlinear switching ESO was proposed to estimate voltages and stator currents, respectively. Rs, ψf, Ld, Lq, id, θr and
back-EMF in [28]. The frequency of the back EMF is the same ωr are stator resistance, permanent magnet flux linkage, d-axis
as the operating frequency of the motor. So the back EMF is a inductance, q-axis inductance, d-axis stator current, actual rotor
very fast-varying sinusoidal disturbance when the IPMSM is position, and actual rotor electrical angular speed, respectively.
running at high speed. However, ESO has a low-pass filtering [eα, eβ]T is the α-axis and β-axis equivalent back EMF.
property for the estimated back EMF, which will cause the Equation (4) can be rewritten as
estimated back EMF phase to lag behind the actual back EMF iα  1 uα  1  Rs 0  iα  1 eα 
phase at high speed, thus making the estimated rotor position
p =  −   −   (6)
iβ  Lq u β  Lq  0 Rs  iβ  Lq eβ 
inaccurate. Although high ESO bandwidth can reduce the phase According to the basic idea of the LESO, the system input
lag of estimated back EMF, it also increases the sensitivity of and the unknown interference of (6) can be expressed as
ESO to noise.
In order to accurately estimate the back EMF of IPMSM and b0 u = uαβ / Lq

thus accurately estimate the rotor position and speed, an  f ( x, t ) = − Rs iαβ / Lq (7)
IPMSM sensorless drive method using an improved LESO 
(ILESO) is proposed in this paper. The ILESO is developed d αβ = −eαβ / Lq
from conventional LESO. The pure integrator is maintained to Then, the corresponding LESO of (6) is established as
track DC or extremely low frequency disturbances. Besides, the follows:
rapidly-varying sinusoidal interference estimator (RSIE) is ε1 = xˆ1 − iαβ

inserted into the disturbance estimation loop to estimate  pxˆ1 = xˆ2 + b0 u + f ( xˆ1 , t ) − β1 ε1 (8)
rapidly-varying back EMF. This design allows ILESO to  pxˆ = − β ε
observe the back EMF with a relatively low bandwidth, which  2 2 1

resolves the tradeoff between the bandwidth of LESO and the where x̂1 =[ xˆ1α , xˆ1 β ]T represents the estimated stator currents
high-frequency noises filtering. Moreover, in order to make the iαβ. x̂2 =[ xˆ2 α , xˆ2 β ]T represents the estimated unknown
rotor position error converge to zero during frequency ramp, an
disturbances dαβ. ε1=[ε1α, ε1β]T represents the estimation error,
enhanced phase locked loop (EPLL) is presented. Finally, the
experimental results show the validity of the proposed method. and ε1α =xˆ1α − iα and ε1 β =xˆ1 β − i β .
Because (4) has symmetrical structures, the performance of
II. IPMSM SENSORLESS DRIVE BASED ON LESO the LESO for equivalent back EMF estimation is analyzed by
taking α-axis as an example. According to (6) and (8), the LESO
First order single input single output system can be expressed
estimate error dynamics can be obtained as
as
px = f ( x, t ) + d (t ) + b0 u (1)  pε1α = ε2α − με1α − β1ε1α
 (9)
where p is differential operator, x is the state variable, u is the  pε2α = − β2 ε1α − hα (t )
system input, b0 is the control gain, f(x, t) is known interference, where ε2 α = xˆ2 α − d α , μ=Rs/Lq, hα(t)=pdα. Equation (9) is
and d(t) is unknown interference. transformed to s-domain and expressed as
By extending unknown interference d(t) as a new state, (1)

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

 sε1α = ε2 α − με1α − β1ε1α inaccurate estimation rotor position. This will worsen sensorless
 (10) driving performance of IPMSM, especially at high speed.
 sε2 α = − β2 ε1α − sd α
êα
According to (9), the transfer function can be obtained
ε1α −s
= (11)
d α s 2 + ( μ + β1 ) s + β2
ε2 α − s( s + μ + β1 )
= 2 (12)
d α s + ( μ + β1 ) s + β2
Now that the observation error has been derived, the
observed variable xˆ2 α is derived as
xˆ2 α = ε2 α + d α
− s( s + μ + β1 ) β2 Fig. 2. The simulation results of the actual equivalent back EMF and the
= dα + dα = 2 dα (13)
s + ( μ + β1 ) s + β2
2
s + ( μ + β1 ) s + β2 estimated equivalent back EMF by the LESO.

The characteristic polynomial (13) is the Hurwitz stable III. IPMSM SENSORLESS DRIVE BASED ON AN ILESO
λ ( s) = s 2 + ( μ + β1 ) s + β2 = ( s + ω0 ) 2 (14)
Because the unknown disturbance dαβ is a very fast-varying
where ω0 is the bandwidth of LESO. According to (14), [β1, sinusoidal disturbance at high speed. Therefore, the LESO is not
β2]T=[2ω0-μ, ω02 ]. sufficiently capable to deal the unknown disturbance dαβ. In
According to (7), (13) and (14), the transfer function of the order to improve the ability of LESO to estimate rapidly varying
estimated back EMF and actual back EMF can be obtained sinusoidal interference dαβ, an ILESO is proposed.
eˆ ω02
fα ( s) = α = (15) A. IPMSM Sensorless Control Structure Based on ILESO
eα ( s + ω0 ) 2
Fig. 3 presents the system block diagram of IPMSM
where eˆα is estimated equivalent back EMF in the α-axis. sensorless control based on ILESO. To reduce the cost of
Fig. 1 presents the Bode plots of fα(s), with ω0 ∈ IPMSM sensorless control system, the voltage sensors are not
{500,1000,1500} (rad/s). As seen in Fig. 1, the LESO has a installed. The reference voltage uα* and u *β are used to estimate
low-pass filtering property for the estimated back EMF. The
the equivalent back EMF. The reference voltage uα* , stator
LESO can accurately estimate the equivalent back EMF at low
speed. However, the frequency of the equivalent back EMF and current iα and estimated angular velocity ωe are used as the input
unknown disturbance dαβ is the same as the operating frequency of ILESO. The output of the ILESO is the estimated unknown
of the motor. So the unknown disturbance dαβ is a very disturbance dα. The estimation process of the estimated
fast-varying sinusoidal disturbance at high speed. Therefore, the unknown disturbance dβ is similar to dα. Then, the estimated
phase of the estimated back-EMF based on traditional LESO unknown disturbance dαβ are used as the input of EPLL. The
seriously lags the phase of the actual back-EMF at high speed. output of EPLL is the estimated rotor position and speed.
Vdc
Although the high LESO bandwidth ω0 can reduce the phase lag +
id∗ u*d u*α
of the estimated back EMF, it also increases the sensitivity of PI dq
ωr* +
PI MTPA
id*
−
id
SV
VSI
the LESO to noise. − iq∗ + u*q u*β PWM
ωe
PI αβ
− θe
iq θe
id iα ia
dq αβ
ib
iβ
ωe
αβ abc
ic

dα
θe IESO
uα*
EPLL IPMSM
dβ
IESO
u *β
ωe
Fig. 3. The system block diagram of IPMSM sensorless control based on
ILESO.

B. Proposed ILESO
Fig. 4 shows the proposed ILESO. The pure integrator in
Fig. 1. The Bode diagrams of the fα(s). interference loop is kept to track DC or extremely low
Fig. 2 shows the simulation results of actual equivalent back frequency disturbances. Besides, the RSIE is added to the
EMF and estimated equivalent back EMF when speed unknown interference estimation loop to estimate the
command is set at 50 Hz, and ω0=1000 rad/s. As seen in Fig. 2, rapidly-varying back EMF. The block diagram of the RSIE is
the phase of the estimated equivalent back EMF lags behind the shown in Fig. 5. From Fig. 5, the inserted RSIE can be presented
phase of the actual equivalent back EMF, which results in as

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

K1 ( s + g ) d αβ The RSIE is inserted into the LESO to estimate the

G ( s) = =
(16) rapidly-varying back EMF. Since the frequency of the back
ε1 ( s + g ) 2 + ωe2
EMF and the operating frequency of the motor are the same for
where K1 and g are adjustable coefficients. The infinite gain of
the IPMSM, to estimate the rapidly-varying back EMF, ωe is the
the integrator is shifted to ωe.
estimated speed in (16) and (23), and ωe is obtained by EPLL.
Rs b0u According to (23), the Bode plots of eˆα /eα are shown in Fig. 6,
Lq
−+ + with ω0=1000 rad/s, K1=3.14β2, g=0.1, and ωe=314 rad/s.
iαβ − + x̂1 1 − Compared with Fig. 1, the proposed ILESO has higher gains at
ε1
s
− around the ωe=314, and the phase lag is reduced around the
β1 d αβ ωe=314 compared with the conventional LESO, which means
1 + that ILESO can accurately estimate the equivalent back EMF in
β2
s
+
a relatively low bandwidth compared to traditional LESO.

Fig. 4. The structure diagram of the proposed ILESO.

ωe
+ 1
ε1 K1 × ÷ d αβ
− s +g

1
×
s +g

RSIE
Fig. 5. The structure diagram of the RSIE.
From equation (4), it can be seen that the α-axis voltage
equation and the β-axis voltage equation are symmetrical
structures, so the following takes the analysis of the α-axis
voltage equation as an example. An ILESO can be expressed as
 ε1α = xˆ1α − iα Fig. 6. the Bode plots of eˆα /eα.

 pxˆ1α = xˆ2 α + b0 u + f ( xˆ1α , t ) − β1 ε1α (17) To further verify the effectiveness of the ILESO for estimated
 pxˆ = − β ε − G (t )ε fast-varying sinusoidal interference, the simulation results of
 2α 2 1α 1α
actual equivalent back EMF and estimated equivalent back
where G(t) is time domain expression of G(s). According to (6) EMF is shown in Fig. 7 when speed command is set at 50 Hz,
and (17), ILESO estimate error dynamics can be obtained as and ω0=1000 rad/s, K1=3.14β2, g=0.1. As seen in Fig. 7, the
 pε1α = ε2α − με1α − β1ε1α estimated back EMF by ILESO can track the actual equivalent
 (18)
 pε2α = − β2 ε1α − G (t )ε1α − hα (t ) back EMF well without amplitude attenuation and phase error.
Equation (18) is transformed to s-domain and expressed as Compared with Fig. 2, the ILESO can accurately track the phase
of the equivalent back EMF. Because the phase information of
 sε1α = ε2 α − με1α − β1 ε1α
 (19) equivalent back EMF is directly related to the rotor position, so
 sε2 α = − β2 ε1α − G ( s )ε1α − sd α tracking the phase information of equivalent back EMF is very
According to (19), the transfer function can be obtained important for accurate estimation of rotor position.
ε1α −s
= (20) eˆα
d α s 2 + ( μ + β1 ) s + G ( s ) + β2
ε2 α − s( s + μ + β1 )
= (21)
d α s 2 + ( μ + β1 ) s + G ( s) + β2
According to (21), the observed variables xˆ2 α is derived as
G ( s) + β2
xˆ2 α = ε2 α + d α = (22)
s 2 + ( μ + β1 ) s + G ( s ) + β2
According to (7), (16), and (22), the transfer function of
estimated back EMF and actual back EMF can be expressed as
eˆ Fig. 7. The simulation results of actual equivalent back EMF and estimated
f α (s) = α equivalent back EMF by ILESO.
eα
C. Parameters of the ILESO Adjustment Guidance
β2 s 2 + ( K1 + 2 β2 g ) s + K1 g + β2 g 2 + β2 ωe2
= From Fig. 4 and Fig. 5, the ILESO has four adjustable
s + ( μ + β1 + 2 g ) s3 + ( g 2 + ωe2 + 2 β1 g + 2 μg + β2 ) s 2
4

parameters, namely the β1, β2, g and K1. The parameter tuning
+( β1 g 2 + μg 2 + β1ωe2 + μωe2 + K1 + 2 β2 g ) s + K1 g + β2 g 2 + β2 ωe2 guidelines are summarized as follows.
(23)

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

1) Parameters β1 and β2 adjustment guidance cos(θe )

In order to maintain the excellent performance of traditional
LESO in tracking DC or extremely low frequency disturbances, + Ex ∆θ + ωe 1 θe
dα × Kp
parameters β1and β2 in the proposed ILESO are adjusted similar + + s

to traditional LESO. The parameters β1and β2 of the LESO and dβ × Ki

1
s
can be simply designed according to the desired bandwidth of PD LF VCO
the LESO [31]. As shown in (14), when the value of ω0>0, all sin(θe )

the poles of the characteristic equation (14) lie on the negative Fig. 8. Block diagram of the second-order PLL.
half of the plane, which means the system is stable. And once ω0 The transfer function θe/θr of second-order PLL can be
is chosen, the parameters β1 and β2 can be determined according expressed as
to (14) to be β1=2ω0-μ, β2= ω02 . The bandwidth ω0 should be θ Ex K p s + E x K i
GPLL 2 = e = 2 (25)
large enough to ensure that the dynamics of the LESO. On the θr s + E x K p s + E x K i
other hand, a larger bandwidth ω0 will make the LESO and
where Ex=Eαβ/Lq, and Eαβ is the amplitude of equivalent back
ILESO more sensitive to measurement noise, and the bandwidth
EMF, Kp and Ki are adjustable gain. From (25), the error transfer
ω0 is constrained by many factors, such as discretization period,
functions Δθ/θr are described as
desired dynamic tracking performance, etc. Thus, in practice,
the bandwidth ω0 can be selected as a tradeoff between ∆θ s2
GPLL 2e = = 2 (26)
disturbance tracking performance and immunity to sampling θr s + Ex K p s + E x K i
noise. Considering these factors, the bandwidth is designed to where Δθ=θr-θe.
be 1500 rad/s in this paper. The phase jump, the frequency jump and the frequency ramp
2) Parameters g and K1 adjustment guidance in s-domain can be expressed
According to (16), the characteristic equation of the RSIE m
can be expressed as G pj = (27)
s
s 2 + 2 gs + g 2 + ωe2 = 0 (24)
n
From (24), when the value of g>0, all the poles of the G fj = 2 (28)
s
characteristic equation (24) lie on the negative half of the plane,
c
which means the system is stable. When the value of g<0, all the G fr = 3 (29)
poles of the characteristic equation (24) lie on the positive half, s
which means the system is unstable. Therefore, in order to where m, n and c are the gains of the phase jump, frequency
ensure the stability of the RSIE, g>0. The greater the value of g, jump and frequency ramp, correspondingly. When the Gpj, Gfj
the greater the stability margin of the RSIE. However, the and Gfr are input to PLL respectively, and applying the final
greater the value of g, the smaller the gain at frequency ωe, and value theorem to the phase-error transfer functions of the
the worse the frequency selection ability. Considering these second-order PLL, which can be described as
factors, the g is chosen to be 0.1 in this paper. ms 2
∆θ pjf = lim sGPLL 2 e G pj = lim 2 =0 (30)
In order to maintain the same gain as the interference loop s →0 s →0 s + E K s + E K
x p x i
based on pure integral and the unknown interference estimation
ns
loop based on the RSIE, K1 can be expressed as K1=β2K2, where ∆θ fjf = lim sGPLL 2 e G fj = lim =0 (31)
K2 is adjustable coefficient. Larger K2 enables higher gains of s →0 s →0 s 2 + Ex K p s + E x K i
RSIE at the ωe, which helps improving the estimation accuracy c c
of the RSIE. However, the high gains tend to bring destabilizing ∆θ frf = lim sGPLL 2e G fr = lim = (32)
s →0 s →0 s + E x K p s + Ex K i E x K i
2

effects on the resulting system. Consequently, the gain K2 is

chosen to be 3.14. Thus K1=β2K2=3.14β2 in this paper. As seen in (30)-(32), for the Gpj and Gfj, the rotor position
error can converge to zero through the second-order PLL.
D. EPLL However, for the frequency ramp input Gfr, the rotor position
After obtaining accurate phase information of the equivalent error cannot converge to zero through the second-order PLL.
back EMF, the rotor position is usually estimated by phase The rotor position error is related to the gain of the frequency
locked loop (PLL). Therefore, designing a reasonable PLL is jump c, back EMF amplitude Eαβ and integral gain Ki. The Eαβ is
the key step to accurately estimate the rotor position. proportional to the speed of the motor, so it cannot be adjusted
1) Traditional second-order PLL for a specific drive system. The gain c is relatively large for
Usually, the second-order PLL is used to estimate the rotor industrial applications requiring fast acceleration and
position. A typical second-order PLL consists of three parts: the deceleration. To suppress the rotor position error by increasing
phase detector (PD), loop filter (LF), and voltage-controlled the integral gain Ki for the frequency ramp input in the condition
oscillator (VCO). Fig. 8 shows block diagram of the of rapid acceleration and deceleration. However, this will
second-order PLL. reduce the anti-interference ability of PLL.
2) EPLL
To make the rotor position error converge to zero during
frequency ramp, an EPLL is presented in this paper. And a rotor
position error normalization technique is adopted to make the

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

dynamic performance of the EPLL independent of the IV. EXPERIMENTAL VERIFICATION

equivalent back EMF amplitude. The block diagram of the Fig. 10 shows the 2.0 kW IPMSM experimental platform, and
EPLL is shown in Fig. 9. the permanent magnet synchronous motor connected with the
cos(θe ) testing IPMSM provides the required load torque for the testing
IPMSM. The testing IPMSM parameters are shown in Table I.
dα ÷ ×
+ ∆θ ωp ( s 2 + as + b) ωe 1 θe
s2 s The main control chip of the inverter is TMS320F28335
+
d α2 + d β2
produced by TI, and the switching frequency of IGBT is 4 kHz.
To verify the accuracy of the estimated rotor position, a resolver
dβ ÷ × was installed on the rotor shaft of the testing IPMSM. The actual
PD LF VCO
position obtained by the resolver is only used to compare with
sin(θe ) estimated rotor position and does not participate in the
Fig. 9. Block diagram of the EPLL. closed-loop control.
The transfer function θe/θr based on the EPLL can be TABLE I
TESTING IPMSM PARAMETERS
expressed as
Parameter Value Parameter Value
θ ω p s 2 + aω p s + bω p
GEPLL = e = 3 (33) Rated power 2.0 kW Rated current 5.8 A
θr s + ω p s 2 + aωp s + bω p Rated voltage 380 V Rated torque 19 N·m
where a, b and ωp are adjustable gains. Rated frequency 66.67 Hz Rated speed 1000 rpm
From (33), the transfer functions Δθ/θr can be expressed as
∆θ s3
GEPLLe = = 3 (34)
θr s + ω p s + aωp s + bω p
2

When the Gpj, Gfj and Gfr are input to EPLL respectively,
applying the final value theorem to the phase-error transfer
functions of the EPLL, and it is described as
ms 3
∆θEpjf = lim sGEPLLe G pj = lim 3 = 0 (35)
s →0 s → 0 s + ω s + aω s + bω
2
p p p

ns 2
∆θEfjf = lim sGEPLLe G fj = lim =0 (36)
s →0 s →0 s + ω p s + aω p s + bω p
3 2 Fig. 10. 2.0-kW IPMSM vector control platform.

cs A. Verification of Estimated Equivalent Back EMF

∆θEfrf = lim sGEPLLe G fr = lim =0 (37)
s→0 s →0 s + ωp s + aω p s + bω p
3 2 Fig. 11~Fig. 13 present the experimental comparisons of the
estimated equivalent back EMF and the position estimation
where the ΔθEpjf, ΔθEfjf and ΔθEfrf are rotor position steady-state
error at 150 rpm, 1000 rpm and 1800 rpm with no load,
errors when the Gpj, Gfj and Gfr are input to EPLL, respectively.
respectively. Fig. 11(a)~Fig. 13(a) show the experimental
As seen in (35)-(37), when the Gpj, Gfj and Gfr are input to EPLL
results of the LESO. Fig. 11(b)~Fig. 13(b) show the
respectively, the rotor position error can converge to zero
experimental results of the ILESO. In order to compare the
through the EPLL. Therefore, EPLL has better position and
fairness of the experiment, the bandwidth ω0 of the LESO and
speed estimation performance than second-order PLL.
the ILESO is set to ω0=1500 rad/s, and the same PLL (the EPLL)
Equation (33) can be written as
is used in the experimental results of Fig. 11~Fig. 13. The
(2 + l )ζωn s 2 + (1 + 2lζ 2 )ωn2 s + lζωn3
GEPLL = (38) amplitude of estimated equivalent back EMF is normalized to 1
( s + lζωn )( s 2 + 2ζωn s + ωn2 ) in Fig. 11~Fig. 13. As seen in Fig. 11, the phase of the estimated
where ω p = (2 + l ) ζωn , aω p = (1 + 2lζ 2 )ωn2 , and bω p = lζωn3 , equivalent back EMF matches the actual rotor position based on
LESO and ILESO, and the estimated rotor position error is
ωn is the natural frequency, ζ is the damping coefficient, and l is
within 10 degree. Therefore, the LESO and ILESO can estimate
the coefficient of the system. Form (38), the characteristic
the equivalent back EMF very well when the equivalent back
equation of the EPLL can be obtained as follows
EMF is composed of low-frequency components. As seen in Fig.
s 3 + (2 + l ) ζωn s 2 + (1 + 2lζ 2 )ωn2 s + lζωn3 = 0 (39)
12(a), the phase of estimated equivalent back EMF does not
If three poles of (39) s1,2=−ζωn±jωn(1−ζ2)1/2, and s3=−lζωn match the actual rotor position seriously based on the LESO,
have negative real parts, the EPLL is stable. The dominant pole and the position estimation error reaches 24 degree. However,
approximation method is adopted to design EPLL parameters in as seen in Fig. 12(b), the phase of the estimated equivalent back
this paper, lζωn≥5ζωn, so l=5 is chosen. The ζ=0.707 to obtain a EMF matches the actual rotor position based on the ILESO, and
good damping dynamic response. The choice of ωn is a trade-off the estimated rotor position error is within 10 degree. Compare
between acquisition time, overshoot, stability and filtering the experimental results in Fig. 13, the phase of the equivalent
ability. Taking these factors into account, ωn=260 rad/s is back EMF estimated by the LESO lags behind the actual rotor
chosen in this paper. position seriously, and the rotor position estimation error
reaches 41 degree. However, the phase of the estimated
equivalent back EMF matches the actual rotor position based on

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

Position estimation Estimated back

the ILESO at 1800 rpm, and the estimated rotor position error is eˆα eˆ β

EMF (pu)
1
within 10 degree. As seen in Fig. 11(a)~Fig. 13(a), with the 0
increase of the frequency of the equivalent back EMF, the phase -1

Actual position
of the equivalent back EMF estimated by the LESO lags the 360

(deg)
actual phase more serious, which results in the estimated rotor 180

error (deg)
position lagging behind the actual rotor position. As seen in Fig. 4 1 deg ree
0
50
11(b)~Fig. 13(b), with the increase of the frequency of the 0
equivalent back EMF, the ILESO can still accurately estimate -50
the phase of equivalent back EMF. Therefore, compared with Time (10 ms/div)
(a)
the LESO, the ILESO can accurately estimate the equivalent

Position estimation Estimated back

eˆα eˆ β

EMF (pu)
back EMF at high speed. 1
0
Position estimation Estimated back

eˆ α eˆ β
EMF (pu)

1 -1
0

Actual position
-1 360

(deg)
180
Actual position

error (deg)
360 0
(deg)
180 50 1 0 deg ree
error (deg)

0 0
50 9 deg ree -50
0 Time (10 ms/div)
-50 (b)
Time (100 ms/div) Fig. 13. Experiment results of estimated equivalent back EMF and position
(a) estimation error at 1800 rpm. (a) LESO. (b) ILESO.
Estimated back

eˆα eˆ β
EMF (pu)

1
B. Verification of Sensorless Operating at Different Speeds
0
-1 Fig. 14 shows the experimental results with no load when
Actual position

speed command changes as 150∼300∼600∼1000∼800∼

360
500∼400∼150 rpm. Fig. 14(a) and Fig. 14(b) present the
Position estimation

(deg)

180
error (deg)

0 experimental result of the LESO and ILESO, respectively. As

50 8 d egree seen in Fig. 14 (a), the higher the operating frequency of the
0
-50 motor, the greater the error of rotor position estimated by the
Time (100 ms/div) LESO. As seen in Fig. 14 (b), compared with the LESO, the
(b)
estimated rotor position error is within 10 degree during the
Fig. 11. Experiment results of estimated equivalent back EMF and position
estimation error at 150 rpm. (a) LESO. (b) ILESO. entire speed operation.
Estimated back

êα ê β
Position estimation Actual speed

1000
EMF (pu)

1
800
(rpm)

0
600 1000

Estimated speed
-1
400 800
Actual position

(rpm)
360 200 600
Position estimation

(deg)

180 0 400
error (deg)

200
error (deg)

0
50 2 4 deg ree 50 2 6 deg ree 0
0 0
-50 -50
Time (20 ms/div) Time (1 s/div)
(a) (a)
Position estimation Estimated back

eˆα eˆ β
Position estimation Actual speed

1000
EMF (pu)

1
800
(rpm)

0
600 1000
Estimated speed

-1
400 800
Actual position

(rpm)

360 200 600

(deg)

180 0 400
error (deg)

200
error (deg)

0
50 9 deg ree 50 10 degree 0
0 0
-50 -50
Time (20 ms/div) Time (1 s/div)
(b) (b)
Fig. 12. Experiment results of estimated equivalent back EMF and position Fig. 14. Experimental results with operating at different speeds. (a) LESO. (b)
estimation error at 1000 rpm. (a) LESO. (b) ILESO. ILESO.

C. Verification of Sensorless Control Under Load Contiontions

Fig. 15 presents the experimental waveforms under rated load
and given speeds of 200 rpm, 400 rpm, 600 rpm, 800 rpm and
1000 rpm respectively. Fig. 15(a) and Fig. 15(b) present the
experimental results of the LESO and the ILESO, respectively.
When the operating frequency of the motor increases to 800 rpm,

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

there is a large position estimation error based on the LESO, and 20% rated load disturbance is applied to the test motor, there is a
the sensorless control cannot be completed. However, when the large position estimation error based on the LESO, and the
proposed ILESO is adopted, the position estimation error is sensorless control cannot be completed. From Fig. 17 (b), the
within 10 degree when speed command changes as rotor position estimation error within 10 degree during the
200∼400∼600∼800∼1000 rpm, and the system can still different load disturbance based on the ILESO.
operate stably at 1000 rpm with rated load. 1000

Actual speed
800
Actual speed

1000

(rpm)
800 600
(rpm)

400

Position estimation
600
200

Position estimation
400

error (deg)
0 50
200

error (deg)

a-phase current
0 0
50
10 -50
a-phase current

(A)
10 -50 0
(A)

-10
0
Time (2 s/div)
-10
Time (1 s/div) (a)
(a) 1000

Actual speed
800
Actual speed

1000

(rpm)
800 600
(rpm)

400

Position estimation
600
200
Position estimation

400

error (deg)
0 50
200
error (deg)

a-phase current
0 50 0
10 -50
a-phase current

(A)
10 -50 0
(A)

0 -10
Time (2 s/div)
-10
Time (1 s/div) (b)
(b) Fig. 17. Experimental results at 1000 rpm with load change as
Fig. 15. Sensorless operating at different speeds with rated load. (a) LESO. (b) 0%∼20%∼40%∼60%∼80%∼100% of rated load. (a) LESO. (b) ILESO.
ILESO. Fig. 18 shows experimental results of sensorless operation
Fig. 16 shows experimental results at 500 rpm with a step based on the ILESO at 1000 rpm with a step rated load
rated load disturbance. Fig. 16(a) and Fig. 16(b) present the disturbance. The rotor position estimation error within 15
experimental results of the LESO and the ILESO, respectively. degree at 1000 rpm during the step load disturbance based on
From Fig. 16 (a), the rotor position estimation error based on the ILESO. Thus, the ILESO can operate at 1000 rpm with the
LESO exceeds 25 degree during the step load disturbance. From step rated load.
Fig. 16 (b), the rotor position estimation error based on ILESO 1000
Actual speed

within 15 degree during the step load disturbance. 800

(rpm)

600
Actual speed

400

Position estimation
500
(rpm)

200

error (deg)
250
Position estimation

0 50
0
a-phase current

0
error (deg)

50 10 -50
(A)

0 0
a-phase current

-50
-10
10 Time (2 s/div)
(A)

0
Fig. 18. Experimental results at 1000rpm with a step rated load disturbance.
-10
Time (1 s/div) Fig. 19 presents the sensorless drive results with rated load
(a) disturbance at 40 rpm (4% rated speed). Fig. 19(a) and Fig. 19(b)
present the experimental results of the LESO and the ILESO,
Actual speed

500 respectively. From Fig. 19, when the rated load disturbance is
(rpm)

250 suddenly added or subtracted, the maximum rotor position

Position estimation

0
errors estimated by the LESO and the ILESO are 39 degree and
error (deg)

50
0 35 degree, respectively.
Position estimation Actual speed
a-phase current

-50 eˆα eˆ β
40
(rpm)

10
0
(A)

a-phase Actual position

0
-10
(deg)

Time (1 s/div) 360

(b) 180
error (deg)

Fig. 16. Experimental results at 500 rpm with rated load disturbance. (a) LESO. 50 0
(b) ILESO. 0
current (A)

-50 9 deg ree 10

Fig. 17 shows the experimental results at 1000 rpm with load 0
change as 0%∼20%∼40%∼60%∼80%∼100% of rated load. -10
Time (1 s/div)
Fig. 17(a) and Fig. 17(b) present the experimental result of the (a)
LESO and ILESO, respectively. As seen in Fig. 17 (a), when

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

Position estimation Actual speed

eˆα eˆ β

Actual position Estimated back

Flage
40 1
(rpm)

EMF (pu)
0 0

a-phase Actual position

eˆ α
1

(deg)
360 0
180 -1

Position estimation
error (deg)

50 0 360

error (deg)
0 180

(deg)
current (A)
-50 9 deg ree 10 50 0
1 4 deg ree 1 7 degree
0 0
-10 -50
Time (1 s/div) Time (1 s/div)
(b) (a)
Fig. 19. Experimental results at 40 rpm with rated load disturbance. (a) LESO.

Actual position Estimated back

Flage
1

EMF (pu)
(b) ILESO.
0
eˆα
1
D. Verification of Motor Parameter Mismatch
0
Fig. 20 shows the experimental results of the sensorless drive -1

Position estimation
360
with q-axis inductance Lq mismatch at 1000 rpm. Flag=0

error (deg)
180

(deg)
indicates that the q-axis inductance Lq is the nominal value, 50 0
1 2 deg ree 1 4 deg ree
Flag=1 indicates that the q-axis inductance Lq changes to 2 0
-50
times the nominal value. Fig. 20(a) and Fig. 20(b) present the Time (1 s/div)
experimental results of the LESO and the ILESO, respectively. (b)
From Fig. 20(a), when the q-axis inductance Lq changes to 2 Fig. 21. Experimental results of the sensorless drive with stator resistance Rs
times the nominal value, the estimated rotor position error mismatch at 40 rpm. (a) LESO. (b) ILESO.
increases significantly. Compared with the LESO, the rotor E. Verification of EPLL
position error of the ILESO increases less when the q-axis
Fig. 22 shows experimental results at the fast acceleration and
inductance Lq changes. Therefore, the proposed method is less
deceleration with no load, and the speed slew rate is 3300 rpm/s
sensitive to the variations of the q-axis inductance Lq.
during acceleration, and the speed slew rate is 1100 rpm/s
Actual position Estimated back
Flage

1 during deceleration. Fig. 22(a) and Fig. 22(b) present the

EMF (pu)

0
eˆα
1
experimental results of the second-order PLL and the EPLL,
0 respectively. In order to compare the fairness of the experiment,
-1
the same ILESO is used to estimate the equivalent back EMF in
Position estimation

360
the experimental results of Fig. 22. From Fig. 22 (a), the rotor
error (deg)

180
(deg)

50 2 4 deg ree 30 d eg ree 0 position estimation error is about 12 degree during acceleration
0 based on second-order PLL, and the rotor position estimation
-50
Time (40 ms/div) error is about 5 degree during deceleration. Thus, the
(a) second-order PLL cannot achieve zero steady-state position
error during acceleration and deceleration. From Fig. 22 (b), the
Actual position Estimated back
Flage

1
EMF (pu)

0 rotor position estimation error is within 5 degree based on the

eˆα
1 EPLL during acceleration and deceleration.
0
-1
Position estimation Actual speed

1800
Position estimation

360 1350
(rpm)
error (deg)

180 900 1800

Estimated speed
(deg)

50 0 450 1350
9 deg ree 11 degree
(rpm)

0 0 900
-50 450
error (deg)

Time (40 ms/div) 0

(b) 20
Fig. 20. Experimental results of the sensorless drive with q-axis inductance Lq 0
mismatch at 1000 rpm. (a) LESO. (b) ILESO. -20
Time (1 s/div)
Fig. 21 shows the experimental results of the sensorless drive
(a)
with stator resistance Rs mismatch at 40 rpm. Flag=0 indicates
Position estimation Actual speed

that the stator resistance Rs is the nominal value, Flag=1 1800

1350
(rpm)

indicates that the stator resistance Rs changes to 2 times the 900 1800
Estimated speed

nominal value. Fig. 21(a) and Fig. 21(b) present the 450 1350
(rpm)

experimental results of the LESO and ILESO, respectively. As 0 900

450
seen in Fig. 21, when the stator resistance Rs changes to 2 times
error (deg)

0
the nominal value, the position estimation error estimated by the 20
ILESO is less increased than that estimated by the LESO. 0
-20
Therefore, compared with the LESO, the ILESO is less sensitive Time (1 s/div)
to the variations of the stator resistance mismatch Rs. (b)
Fig. 22. Experimental results at the fast acceleration and deceleration with no
load. (a) second-order PLL. (b) EPLL.

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

V. CONCLUSION iα = id cos θr − iq sin θr

 (A-6)
In this paper, an IPMSM sensorless drive method using an iβ = id sin θr + iq cos θr
ILESO is proposed to accurately estimate the IPMSM
equivalent back EMF and thus accurately estimate the rotor According to (A-6), (A-5) can be expressed as
position and speed. The proposed ILESO consists of pure  ∆Lp ( 2id cos θr − iα ) + p (ψ f cos θr ) 
V (θr ) =  
 ∆Lp ( 2id sin θr − iβ ) + p (ψ f sin θr ) 
integral and the RSIE to estimate the total disturbance of the
sensorless drive system. The RSIE is added to the unknown
interference estimation loop to estimate rapidly-varying back  −∆Lp (iα ) + p ( ψ f + 2∆Lid ) cos θr  
 
EMF. The experimental results demonstrate that the proposed = (A-7)
 −∆Lp (i ) + p (ψ + 2∆Li ) sin θ  
ILESO can also correctly track very fast-varying equivalent  β  f d r 
back EMF at high speed. Compared with the traditional LESO
 −∆Lp (iα ) + p ( ψ f + ( Ld − Lq )id ) cos θr  
sensorless method, the maximum errors of the estimation  
=
position with load disturbance and different speed operations  −∆Lp (i ) + p (ψ + ( L − L )i ) sin θ  
are obviously reduced, and the rotor position estimation error  β  f d q d r 

does not increase with the increase of motor operating According to (A-7), (A-4) can be expressed as
frequency.  uα  iα   L 0  iα 
  = Rs   + p   
APPENDIX u β  i β   0 L  i β 
 −∆Lp (iα ) + p ( ψ f + ( Ld − Lq )id ) cos θr  
The voltage equation of an IPMSM in the dq axes can be  
+
described as  −∆Lp (i ) + p (ψ + ( L − L )i ) sin θ  
ud  id   pLd 0  id   0 − Lq  id   0   β  f d q d r 

  = Rs   +  0  i  + ωr   + 
u
 q i
q  pL q q  Ld 0  iq   ωr ψ f  iα  0  iα 
 Lq
 (ψ f + ( Ld − Lq )id ) cos θr  
= Rs   + +  
Lq  iβ 
p p
(A-1)
i β  0  (ψ + ( L − L )i ) sin θ  
where [ud, uq]T and [id, iq]T are the d- and q-axis stator voltages  f d q d r 

and currents, respectively. (A-8)

The transformation matrix Tdq-αβ(θr) from dq axes to αβ axes For IPMSM, id is always commanded to be zero or negative,
can be expressed as and ωr ψ f >> ( Ld − Lq ) p (id ) in steady state [29]. The last term
cos(θr ) − sin(θr )  of (A-8) can be rewritten as
Tdq −αβ (θr ) =   (A-2)
 sin(θr ) cos(θr )   (ψ f + ( Ld − Lq )id ) cos θr  
 
According to (A-1) and (A-2), the IPMSM model in the αβ p
axes can be expressed as  (ψ + ( L − L )i ) sin θ  
 f d q d r 

u α   L + ∆L cos(2θr ) ∆L sin(2θr )  iα 

  = p    cos θr ⋅ p (ψ f + ( Ld − Lq )id ) + ( ψ f + ( Ld − Lq )id ) p (cos θr ) 
u β   ∆L sin(2θr ) L − ∆L cos(2θr )  i β  = 
(A-3)  sin θr ⋅ p ( ψ f + ( Ld − Lq )id ) + (ψ f + ( Ld − Lq )id ) p (sin θr ) 
iα   − sin θr   
+ Rs   + ωr ψ f   −
 θ r
= ωr (ψ f + ( Ld − Lq )id ) 
sin
i β   cos θr  
where L=(Ld+Lq)/2 and ΔL=(Ld-Lq)/2. Rearranging (A-3), the  cos θ r 

following equations can be obtained: (A-9)

uα  iα   L 0  iα  According to (A-8) and (A-9), the model of the IPMSM in the
u  = Rs i  + p    αβ axes [12], [29], [30] can be obtained, as shown in (4) and (5).
 β  β  0 L  i β 
(A-4)
 cos θr sin θr  id    − sin θr  REFERENCES
+ ∆Lp   + ωr ψ f 
  sin θr − cos θr  iq    cos θr 
 [1] G. Zhang, G. Wang, D. Xu, and N. Zhao, “ADALINE-network-based
   PLL for position sensorless interior permanent magnet synchronous
The last two terms of (A-4) can be rewritten as motor drives,” IEEE Trans. Power Electron., vol. 31, no. 2, pp.
 cos θr sin θr  id    − sin θr 
1450-1460, Feb. 2016.
V (θr ) = ∆Lp   + ωr ψ f  G. Luo, R. Zhang, Z. Chen, W. Tu, S. Zhang, and R. Kennel, “A novel
  sin θr − cos θr  iq  
[2]

    cos θr  nonlinear modeling method for permanent-magnet synchronous motors,”
IEEE Trans. Ind. Electron., vol.63, no. 10, pp. 6490-6498, Oct. 2016.
 cos θr sin θr  id   ψ f cos θr  [3] G. Wang, H. Hu, D. Ding, N. Zhao, Y. Zou, and D. Xu, “Overmodulation
= ∆Lp       + p  (A-5) strategy for electrolytic capacitorless PMSM drives: voltage distortion
 
  sin θr − cos θr  iq    ψ f sin θr  analysis and boundary optimization,” IEEE Trans. Power Electron., vol.
35, no. 9, pp. 9574-9985, Sept. 2020.
 ∆Lp ( id cos θr + iq sin θr ) + p(ψ f cos θr )  [4] G. Wang, D. Xiao, G. Zhang, C. Li, X. Zhang, and D. Xu, “Sensorless
=  control scheme of IPMSMs using HF orthogonal square-wave voltage
 ∆Lp(id sin θr − iq cos θr ) + p(ψ f sin θr )  injection into a stationary reference frame,” IEEE Trans. Power Electron.,
vol. 34, no. 3, pp. 2573-2584, Mar. 2019.
By applying the following inverse Park transformation to the
currents

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

[5] X. Luo, Q. Tang, A. Shen, and Q. Zhang, “PMSM sensorless control by permanent-magnet synchronous motor,” IEEE Trans. Ind. Electron., vol.
injecting HF pulsating carrier signal into estimated fixed-frequency 63, no. 5, pp. 3019-3027, May 2016.
rotating reference frame,” IEEE Trans. Ind. Electron., vol. 63, no. 4, pp. [25] C. Du, Z. Yin, J. Liu, Y. Zhang, and X. Sun, “A speed estimation method
2294-2303, Apr. 2016. for induction motors based on active disturbance rejection observer,”
[6] G. Wang, M. Valla, and J. Solsona, “Position sensorless permanent IEEE Trans. Power Electron., vol. 35, no. 8, pp. 8429-8442, Aug. 2020.
magnet synchronous machine drives-a review,” IEEE Trans. Ind. [26] L. Qu, W. Qiao, and L. Qu, “An enhanced linear active disturbance
Electron., vol. 67, no. 7, pp. 5830-5842, Jul. 2020. rejection rotor position sensorless control for permanent magnet
[7] J. Lee, Y. Kwon, and S. Sul, “Signal-injection sensorless control with synchronous motors,” IEEE Trans. Power Electron., vol. 35, no. 6, pp.
tilted current reference for heavily saturated IPMSMs,” IEEE Trans. 6175-6184, Jun. 2020.
Power Electron., vol. 35, no. 11, pp. 12100-12109, Nov. 2020. [27] B. Wang, C. Liu, and J. Shang, “Sensorless control of permanent magnet
[8] Q. Tang, A. Shen, X. Luo, and J. Xu, “IPMSM sensorless control by synchronous motor based on extended state observer,” in Proc. IEEE Int.
injecting bidirectional rotating HF carrier signals,” IEEE Trans. Power Conf. Elect. Mach. Syst., Aug., 2019.
Electron., vol. 33, no. 12, pp. 10698-10707, Dec. 2018. [28] F. Jiang, K. Yang, S. Sun, H. Zhang, L. Tang, and A. Liu, “An improved
[9] G. Zhang, G. Wang, H. Zhang, H. Wang, G. Bi, and X. Zhang, extended state observer based on linear-nonlinear switching strategy for
“Pseudorandom-frequency sinusoidal injection for position sensorless PMSM sensorless control,” in Proc. IEEE Int. Conf. Elect. Mach. Syst.,
ipmsm drives considering sample and hold effect,” IEEE Trans. Power Oct. 2018.
Electron., vol. 34, no. 10, pp. 9929-9941, Oct. 2019. [29] J. Liu, T. A.Nondahl, P. B. Schmidt, S. Royak, and M. Harbaugh, “Rotor
[10] Y. Zhang, Z. Yin, J. Liu, R. Zhang, and X. Sun, “IPMSM sensorless position estimation for synchronous machines based on equivalent
control using high-frequency voltage injection method with random EMF,” IEEE Trans. on Ind. Appl., vol.47, no. 3, pp. 1310-1318, May/Jun.
switching frequency for audible noise improvement,” IEEE Trans. Ind. 2011.
Electron., vol. 67, no. 7, pp. 6019-6030, Jul. 2020. [30] G. Zhang, G. Wang, D. Xu, R. Ni, and C. Jia, “Multiple-AVF
[11] Z. Lin, X. Li, Z. Wang, T. Shi, and C. Xia, “Minimization of additional cross-feedback-network-based position error harmonic fluctuation
high-frequency torque ripple for square-wave voltage injection IPMSM elimination for sensorless IPMSM drives,” IEEE Trans. Ind. Electron.,
sensorless drives,” IEEE Trans. Power Electron., vol. 35, no. 12, pp. vol.63, no. 2, pp. 821-831, Feb. 2016.
13345-13355, Dec. 2020. [31] G. Wang, R. Liu, N. Zhao, D. Ding, and D. Xu, “Enhanced linear ADRC
[12] G. Wang, T. Li, G. Zhang, X. Gui, and D. Xu, “Position estimation error strategy for HF pulse voltage signal injection-based sensorless IPMSM
reduction using recursive-least-square adaptive filter for model-based drives,” IEEE Trans. Power Electron., vol. 34, no. 1, pp. 514-525, Jan.
sensorless interior permanent-magnet synchronous motor drives,” IEEE 2019.
Trans. Ind. Electron., vol. 61, no. 9, pp. 5115-5125, Sept. 2014.
[13] A. Pal, S. Das, and A. Chattopadhyay, “An improved rotor flux space
vector based MRAS for field-oriented control of induction motor drives,”
IEEE Trans. Power Electron., vol. 33, no. 6, pp. 5131-5141, Jun. 2018.
Yanping Zhang was born in Shannxi, China, in 1989.
[14] Z. Yin, G. Li, Y. Zhang, J. Liu, X. Sun, and Y. Zhong, “A speed and flux
He received the B.S. degree in electrical engineering
observer of induction motor based on extended Kalman filter and markov
from Xi’an Polytechnic University, Xi’an, China, in
chain,” IEEE Trans. Power Electron., vol. 32, no. 9, pp. 7096-7117, Sept.
2013, and the M.S. degree in power electronics and
2017.
electrical drives in electrical engineering in 2017
[15] G. Wang, Z. Li, G. Zhang, Y. Yu, and D. Xu, “Quadrature PLL-based
from the Xi’an University of Technology, Xi’an,
high-order sliding-mode observer for IPMSM sensorless control with
China, where he is working toward the Ph.D. degree
online MTPA control strategy,” IEEE Trans. Energy Convers., vol. 28,
in power electronics and electrical drive since 2017.
no. 1, pp.214–224, Mar. 2013.
His research interest includes high performance
[16] J. Han, “From PID to active disturbance rejection control,” IEEE Trans.
sensorless control of synchronous motor.
Ind. Electron., vol. 56, no. 3, pp. 900-906, Mar. 2009.
[17] C. Du, Z. Yin, Y. Zhang, J. Liu, X. Sun, and Y. Zhong, “Research on
active disturbance rejection control with parameter autotune mechanism Zhonggang Yin (M’13) was born in Shandong,
for induction motors based on adaptive particle swarm optimization China, in 1982. He received the B.S., M.S. and Ph.D.
algorithm with dynamic inertia weight,” IEEE Trans. Power Electron., degrees in electrical engineering from Xi'an
vol. 34, no. 3, pp. 2841-2855, Mar. 2019. University of Technology, Shaanxi, China, in 2003,
[18] G. Wang, R. Liu, N. Zhao, D. Ding, and D. Xu, “Enhanced linear ADRC 2006 and 2009, respectively.
strategy for HF pulse voltage signal injection-based sensorless IPMSM In 2009, he joined electrical engineering department
drives,” IEEE Trans. Power Electron., vol. 34, no. 1, pp. 514-525, Jan. of Xi'an University of Technology, where he is
2019. currently a professor. His research interests include
[19] G. Zhang, G. Wang, B. Yuan, R. Liu, and D. Xu, “Active disturbance high-performance control of ac motor, and digital
rejection control strategy for signal injection-based sensorless IPMSM control of power converters.
drives,” IEEE Trans. Transport. Electrific., vol. 4, no. 1, pp. 330-339,
Mar. 2018. Cong Bai was born in Shaanxi, China, in 1993. She
[20] L. Zhao, Q. Li, B. Liu, and H. Cheng, “Trajectory tracking control of a received the B.S. degree in electrical engineering
one degree of freedom manipulator based on a switched sliding mode from Xi’an University of Technology, Xi’an, China,
controller with a novel extended state observer framework,” IEEE Trans. in 2015. She is currently working toward the Ph.D.
Syst., Man, Cybern., Syst., vol. 49, no. 6, pp. 1110-1118, Jun. 2019. degree in electric machines and electric apparatus
[21] C. Liu, G. Luo, X. Duan, Z. Chen, Z. Zhnag, and C. Qiu, “Adaptive from Xi’an University of Technology, Xi’an, China.
LADRC-based disturbance rejection method for electromechanical servo Her main field of interest is high-performance control
system,” IEEE Trans. on Ind. Appl., vol.56, no. 1, pp. 876-889, Jan./Feb. of linear motor.
2020.
[22] J. Li, Y. Xia, X. Qi, and Z. Gao, “On the necessity, scheme, and basis of
the linear–nonlinear switching in active disturbance rejection control,”
IEEE Trans. Ind. Electron., vol. 64, no. 2, pp. 1425-1435, Feb. 2017.
[23] W. Tan, and C. Fu, “Linear active disturbance-rejection control: analysis
and tuning via IMC,” IEEE Trans. Ind. Electron., vol. 63, no. 4, pp.
2350-2359, Apr. 2016.
[24] B. Du, S. Wu, S. Han, and S. Cui, “Application of linear active
disturbance rejection controller for sensorless control of internal

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2021.3085126, IEEE
Transactions on Power Electronics

IEEE TRANSACTIONS ON POWER ELECTRONICS

Gaolin Wang (M’13) received the B.S., M.S. and

Ph.D. degrees in Electrical Engineering from Harbin
Institute of Technology, Harbin, China, in 2002,
2004 and 2008 respectively.
In 2009, he joined the Department of Electrical
Engineering, Harbin Institute of Technology as a
Lecturer, where he has been a Full Professor of
Electrical Engineering since 2014. From 2009 to
2012, he was a Postdoctoral Fellow in Shanghai Step
Electric Corporation, where he was involved in the
traction machine control for direct-drive elevators. He has authored more than
100 technical papers published in journals and conference proceedings. He is
the holder of 10 Chinese patents. His current major research interests include
permanent magnet synchronous motor drives, high performance direct-drive
for traction system, position sensorless control of AC motors, efficiency
optimization control of PMSM, and digital control of power converters.
Dr. Wang serves as a Guest Associate Editor of IEEE Transactions on Industrial
Electronics, an Associate Editor of IEEE Access, IET Electric Power
Applications, and Journal of Power Electronics.

Jing Liu was born in Anhui, China, in 1982. She

received the B.S., M.S. and Ph.D. degrees in
electronic engineering from Xi'an University of
Technology, Shaanxi, China, in 2003, 2006 and 2009,
respectively.
In 2009, she joined electronic engineering
department of Xi'an University of Technology, where
she is currently an associate professor. Her research
interests include the power semiconductor devices
and their application to power electronic devices.

0885-8993 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Newcastle University. Downloaded on July 11,2021 at 19:38:50 UTC from IEEE Xplore. Restrictions apply.