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/ACCESS.2018.2858199, IEEE Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier

Using High-control-bandwidth FPGA and SiC

Inverters to Enhance High-frequency Injection
Sensorless Control in Interior Permanent
Magnet Synchronous Machine
1,2 1 2 3 4 4
Wei Qian , Xi Zhang , Fanning Jin , Hua Bai , Member, IEEE, Dingguo Lu & Bing Cheng
1
Shanghai Jiaotong University, Shanghai, China
2
University of Michigan-Dearborn, Dearborn, 48128, USA
3
University of Tennessee, Knoxville, Tennessee, 37996, USA
4
Mercedes Benz North America, Redford Charter Twp, MI 48239

Corresponding author: Hua Bai (e-mail: [email protected]) and Xi Zhang (e-mail: [email protected] ).
This work was supported By Mercedes-Benz North America. It is also in part by National Science Fund of China (51677118), National Key R&D Plan Key
Special Project (2017YFE0102000) and Shanghai Municipal Inter-Governmental International Collaboration Project (16510711500).

ABSTRACT A high-frequency injection (HFI) sensorless control for Interior Permanent Magnet
Synchronous Motors (IPMSMs) with enhanced precision and widened speed range is proposed in this
paper. The injection frequency reaches up to 2 kHz under a 50~100 kHz Silicon Carbide (SiC)-based three-
phase inverter. In addition to the high switching frequency, the Field Programmable Gate Array (FPGA) is
utilized to achieve high control bandwidth (>200 kHz) when implementing the Field-oriented Control
(FOC) algorithm. The benefits of high switching frequency and high control bandwidth in senseless
controls are explained theoretically, i.e., leaving enough room for the injection frequency by using SiC
while tuning down the noise-to-signal ratio by using FPGA. Experimental results verified that such manners
improved the position estimation and lifted the injection frequency effectively, which further allows us to
widen the motor speed range under the HFI sensorless control from 0~500rpm with the conventional
Si+DSP design to 0~1200 rpm with the proposed SiC+FPGA.

INDEX TERMS Field-oriented control, FPGA, Interior PMSM, Sensorless, SiC.

I. INTRODUCTION for interior permanent magnet synchronous motors

Due to the prevalence of Electric Vehicles (EVs), the high- (IPMSMs), e.g., high-frequency signal injection, back EMF
power level multiple pole-pair permanent magnet estimation, model-based adaptive method, Kalman observer
synchronous motor (PMSM) is receiving more attention and sliding mode observer. The back EMF method is not
than ever [1], where field-oriented control (FOC) is competent for the angle estimation at the low speed, where
commonly used [2]. Traditional design flow uses the back EMF is weak [3]-[8]. The model reference
microcontrollers such as Digital Signal Processor (DSP) adaptive rule has high dependence on parameters of the
and PowerPC. However, in some cases that require a lot of motor. The Kalman filter observation is resource occupying
computation resources such as the sensorless motor control, and complex, and it also has high dependence on
one single microcontroller is not sufficient. With most of parameters of the motor [9]-[10]. The sliding mode control
microcontrollers implementing control algorithms in serial is a non-continuous system because of its observer.
sequence, the control bandwidth, i.e., the updating rate of "Chattering" of the system is hard to avoid, which seriously
the Pulse-Width Modulation (PWM) duty cycles, is highly affects the accuracy and reliability of the system. It is rare
limited. This explains why the PWM updating rate is to apply in EVs yet.
usually lower than the switching frequency, especially The high-frequency injection (HFI) method, at the
when the switching frequency is high. present stage, is one of very few methods with angle
There are several candidate sensorless control methods estimations applicable to zero or low speed. For such
1

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access
Author Name: Wei Qian, Xi Zhang, Fanning Jin, Hua Bai, Dingguo Lu and Bing Cheng

method, the precision of the angle estimation and the improves the position estimation, and 3) the increment of
effective range of the motor speed are always top concerns the injection frequency also allows the Low-Pass Filter
[11]-[16]. [11], [12] and [13] described how a HFI (LPF) to design a cut-off frequency distant from the
technique is used on IPMSMs to extract the rotor position fundamental signal, which decouples the fundamental back
information from the inherent saliency of the machine. [14] EMF from the HFI signal thereby widening the motor speed
has proposed a scheme by using a saturated flux linkage range applying HFI sensorless control. All these
model to distinguish the polarity of the rotor magnet at the assumptions need further verification in this paper.
standstill. However, most of works focused on the low The rest of this paper is organized as follows. Section II
injection frequency and low speed estimation. analyzed impact factors of the HFI sensorless estimation
Another high-frequency injection method is the square- method, particularly the control bandwidth and switching
wave signal injection [17]-[21], which has received frequency. Section III provided the system modeling based
attention in recent years. It is suitable for applications with on Xilinx System Generator (XSG), a Fixed-Point
the high dynamic response. In applications where low noise Arithmetic toolbox to automatically generate the HDL code
and low machine saturation effect are demanded, the for Xilinx 7K325 FPGA chip embedded in dSPACE
sinusoidal injection is more suitable [21]. MircoAutobox, as the foundation to realize the sensorless
Some researchers proposed the hybrid sensorless position control. Section IV presented hardware implementation and
estimation [22]-[24] for a wider range of the motor speed. experimental results. Section V concluded the contribution
However the accuracy of the position estimation is limited and innovation of this paper.
by either the switching frequency or the microcontroller.
[25] presented a Sliding Mode Observer (SMO) to extend II. HFI SENSORLESS CONTROL VS CONTROL
the range of speed, while the precision is still very low due BANDWIDTH AND SWITCHING FREQUENCY
to the low switching frequency (5 kHz). In [26], current Fig. 1 is a schematic diagram of a three-phase IPM
control algorithm is carried out every 100 μs. The switching drive, with its mathematical model in the d-q reference
frequency of the inverter and control bandwidth of the frame described by (1).
algorithm are low. The error reached 0.4 rad. In [27], the  dids
Vds  Rs ids  Ld  r Lq iqs
inverter uses the switching frequency of 5 kHz, and the  dt (1)
 diqs
estimated algorithm is running at 450 rpm electrical angle. 
In [28], the controller of PMSM uses Field Programmable Vqs  Rs iqs  Lq dt  r Ld ids  r m

Gate Arrays (FPGAs), the switching frequency is 6 kHz, where Vds, Vqs, ids, iqs are the stator d-axis and q-axis voltage
HF signal is 1.2 kHz, the test speed is 100rpm, and the error and current, respectively. Rs is the stator resistance, Ld, Lq
also reaches 0.4 rad. are the d- and q- axis inductance, respectively, ωr is the
Authors of this paper believe the switching frequency angular velocity, and Ψm is the magnet flux linkage.
and control bandwidth are the two critical impact factors of
the HFI control. To further enhance the HFI performance, a Voltage Current
Sensor Sensor
high control-bandwidth microcontroller such as FPGA, and
S1 S3 S5
a high-switching-frequency inverter such as with SiC Current
Sensor
devices are musts. Essentially various functional modules Current
Sensor

inside the FPGA can be implemented in parallel instead of Vin

in series, which results in the updating rate of the FOC loop S2 S4 S6
up to hundreds of kHz, in contrast to 10 kHz of the
Gate Driver
conventional EV inverter. This potentially results in
updating the duty cycle of each switch every switching
period, yielding a faster control response and close to ideal Micro
Controller
sinusoidal motor current waveform [29]-[32]. When
FIGURE 1. The equivalent circuit schematic of the PMSM drive system.
combined with high-switching-frequency SiC devices [33]-
[34], the motor drive inverter system in this study expects a The d-q currents are obtained using Park transformation.
superior-performing HFI sensorless control, i.e., 1) the 2 2 

ultra-high switching frequency resulted from SiC devices cos  r cos( r  3 ) cos( r  3 ) 
allows the increment of the injection frequency, distant the iq    ia 
   2 2    (2)
injected signal away from the fundamental signal thereby id    sin  r sin( r  3 ) sin( r  3 )  ib 
facilitating the signal extraction, which meanwhile reduces i0    ic 
 
the audible noise, 2) by increasing the control bandwidth  1 1 1 
through the adoption of FPGAs, the switch duty cycle will  2 2 2 
be updated every switching period, which reduces the
distortion of fundamental and injection signals and further A. HFI MODEL
2

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access

The frequency of the injected signal needs be higher than ^

i qc
the fundamental frequency. The relationship between the Im[] sin ct LPF [] Errorpos
injected voltage and current can be approximated as (3).
V  L didc
^
i dc
 dc d Re[] cos2ct LPF [] Sign pol
dt
 (3)
V  L diqc
FIGURE 2. Heterodyning process to extract angle and polarity.

 qc q
dt The demodulated error signal is an input, which goes into
The subscript c denotes injected variables. The IPMSM a PI regulator and is ultimately fed back to form a closed
is well suited for the HFI technique to track a spatial loop. Fig. 3 shows the tracking loop used for the rotor
saliency, because of the inductance difference between d position estimation, where ωHF stands for the observed
and q axis in the rotor reference frame. The HFI method speed obtained from the HFI technique.
superimposes a d-axis sinusoidal carrier signal in the
Demodulation
estimated rotor reference frame onto the fundamental
component voltage. Specifically, the injected high ^
Error ω^ HF θ HF
frequency voltage complies with θ Δθ 1/S
k LPF PI
V^ dc  V cos  t

+

^
c c
(4) -


V qc  0
The ^ denotes the variable in the estimated d-q frame.
FIGURE 3. Tracking loop used for the rotor position estimation.
Furthermore, the current in the estimated rotor reference
frame can be expressed as follows. B. BEMF BASED STATE OBSERVER IN HIGH-SPEED
 ^i  Vc [( 1  1 )  ( 1  1 ) cos 2(  ^ )]sin  t  REGION
 dc 2 L L Ld Lq
r r c
The BEMF in the estimated rotor frame is
 c d q

 Vc2 d 2id ^ ^ ~

 2 2 ( d ) cos ( r   r ) sin c t
3 2 ed  r m sin  r
(8)

 c d
2 d (5) ^ ~
^ ^ eq  r m cos  r
 i qc  c (  ) sin 2(   r ) sin  t 
V 1 1
̂𝒓
̃ 𝒓 = 𝜽𝒓 − 𝜽
 r c where 𝜽
2c Ld Lq
 2 2 ^

 Vc d id ^ ^ q q
     r ) sin c t
2 2
 2 2 d 2 ( d ) cos ( r  r ) sin( r 
 c d ^
eq r m

where θr and θ̂r denote the actual and estimated rotor

position, respectively. From (5), the first item of q-axis
current is suitable for the position detection, and the second
item of the d-axis current is suitable for the polarity ^
^
ed
identification. Fig. 2 shows the heterodyning process to ~ d

extract the position and polarity information. (6) shows the d
position error term, which has an approximate linear
FIGURE 4. Actual vs observed back EMF in the sensorless control
relationship with the error between the actual and estimated
angle. This error term will be used as the input of a PLL
based tracking controller to drive the estimated angle to the The estimated BEMF along the d-axis has the same form
true value. (7) shows the polarity signal term. as the position error term of HFI scheme. So ed could be
used as the input of a PLL-based tracking controller to
^
V Lq  Ld ^

Errpos  LPF{i qsc sin c t}  c ( ) sin 2( r   r ) (6) drive the estimated angle to the true value. When angle
4c Ld Lq ̃ 𝒓 = 𝜽𝒓 − 𝜽
error 𝜽 ̂ 𝒓 is very small, assuming the rotor speed
2 2 and position remain constant during one control cycle, we
^
Vc d id ^

Sign pol  LPF{i dsc cos 2c t}   ( m ) cos ( r   r ) (7) can formulate the system state variables as:
3

8c d d
2 2

x  Ax  Bu (9)
y  Cx
3

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access
Author Name: Wei Qian, Xi Zhang, Fanning Jin, Hua Bai, Dingguo Lu and Bing Cheng

where V2 Lq  Ld
T T T where K LS  ( ) , K HS  r m .
^ ^ ^ ^   ^ ^  ^ ^  4c
x  ids iqs ed eq  u  Vds Vqs  y  ids iqs  Ld Lq
     
^
 R ^ Lq 1   1  f tr ( r )
  s r 0  L 0 
 Ld Ld Ld   d 
 ^   1  1 0 0 0 
L
A    r d
R
 s 
1 
B 0  C 1

0 
Lq Lq Lq   Lq  0 1 0 0 
   0 
 0 0 0 0   0 
   0 0 
 0 0 0 0 
^
An asymptotic state observer is employed. The observer r
0 LS HS
can be described as:
 FIGURE 6. Transition function.
x r  Ax r  Bu  L( y  Cx r ) (10)

The block diagram of hybrid observer is shown in Fig. 7.

 Rs Ld ^ 
  L  2o Lq
r  A PLL-based tracking controller is used. Since this paper is
 d  focused on the HFI method, we will not detail the BEMF
 Lq ^ Rs  and the hybrid observer.
L    r   2o 
 Ld Lq 
 L 2 0  ^

 d o  r
iabc
 0  Lqo2  Low Speed
Observer
errpos
1/ 2 K LS
where , ^
ω0 is the bandwidth of the observer, ϛ is the damping wr ^
ftr ( r ) ^
r
^
coefficient. With a designed gain matrix L, the observer ^
wr
1
PI
generates the observed vector xr, containing the estimated wr ^
1  ftr ( r ) s
BEMF, as shown in Fig. 5. Vabc ed
High Speed 1/ K HS
iabc Observer
u 1
y
( sI  A)
^ ^
B C r wr


L
 FIGURE 7. Hybrid observer block diagram

C
^ ^ D. THE SELECTION OF THE INJECTION FREQUENCY
 x 1 x Due to the limitation of conventional micro-controllers and
B
 s
power devices, the switching frequency of the motor drive
A system is only 5-10 kHz, which explains why majority of
State observer
injection frequency falls into the range of 500~1000 Hz.
The room between the switching frequency and injection
FIGURE 5. State observer to estimate back EMF.
frequency is limited, which restrains the design of the filter.
Our study will apply a simple first-order LPF design. Here
C. HYBRID OBSERVER IN TRANSITION REGION
ωcut is the cut-off frequency.
To realize the full-speed-range sensorless control, HFI
 1
and BEMF are combined by using a hybrid observer. The  H ( j )  2
error information obtained from HFI scheme and BEMF    
̃𝒓 .  1   (12)
based method has the identical form: 𝒆𝒓𝒓𝒐𝒓 = 𝑲𝒔𝒊𝒏𝜽   cut 
These two error signals could be used for the PLL-based 
tracking controller with the necessary normalization. A new H ( j )   tan 1 (  )
error signal which combined these two together is defined  cut
̂ 𝒓 ) is a transition function that determines
by (9). 𝒇𝒕𝒓 (𝝎 If ωcut is close to the injection frequency, the position
which error signal should be relied on, as shown in Fig. 6. signal will be distorted in terms of the phase and amplitude,
^ errpos ^ e which jeopardizes the precision of estimation. If ωcut is near
error  ftr ( r )  [1  ftr ( r )] d (11)
the switching frequency, where the harmonics are severe, it
2 K LS K HS
also worsens the estimated position. Therefore, it is
preferred to leave enough gap between the motor

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access

fundamental frequency, injected frequency and the going through a tooth pitch, yielding the magnetic guiding
switching frequency. wave as
Note both the fundamental waveform (voltage and Z Z
  , t   ( A0   Ai cos(i  f t ))   Aj cos( j  ) (13)
current) and the injected harmonics will be modulated by p p
the PWM strategy, e.g., comparing the reference with where Z is the number of stator slots, p is the pole pair of
carrier triangles and generating the switch duty cycle. the motor, A0   Ai cos(iZ p f t ) is the average magnetic
Therefore both the fundamental and the injected signal conductance, Aj is the amplitude of subharmonic magnetic
carry harmonics. When the motor is running at high speed conductance, θ is position angle and ωf is the fundamental
with relatively low-frequency injected signal, as shown in angular velocity.
Fig. 8 where the injection frequency is 500Hz, potentially It can be seen from (13) that slotting harmonics can
some of the low-order current harmonics will be overlapped rapidly reach the HFI frequency with the increase of speed,
with the injected-frequency fundamental component. Even which seriously confuses the extraction of the injected
an excellent filter cannot 100% separate all these signals. signal. Detailed simulation is carried out in Fig. 10.
When the speed of motor goes up, such overlap becomes
worse.

PM speed
frequency Injection
frequency

Signal overlap
Difficulty design for LPF

(a) 20 rpm & 500 Hz HFI

FIGURE 8. The motor current spectrum at 500Hz injection frequency.

In Fig. 9 where the injection frequency is 2 kHz, the

overlap between injection frequency and fundamental
frequency is little, which results in the fidelity of position
information even for the higher speed. Of course this
requires a higher switching frequency. To set the injection
frequency of 2 kHz, 10 kHz switching frequency is not (b) 400 rpm & 500 Hz HFI
enough anymore. SiC devices will be the candidate for >10
kHz switching in high-power applications.

(c) 150 rpm & 2 kHz HFI

FIGURE 9. The motor current spectrum at 2 kHz injection frequency.

While conventional analysis regards the BEMF of the

motor as the pure sinusoidal waveform, the motor has
slotting effect, resulting in uneven magnetic conductance.
The interaction between magnet motive force and non-
uniform magnetic conductance induces the slotting
harmonic EMF. When the motor rotates, the average (d) 800 rpm & 2 kHz HFI

magnetic conductance is periodic pulsation with the time of

5

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access
Author Name: Wei Qian, Xi Zhang, Fanning Jin, Hua Bai, Dingguo Lu and Bing Cheng

FIGURE 10. Relation between the effective injection signals with

different motor speed at the same injected amplitude in three phase
current.

From the simulation in Fig. 10, it can be seen that for the
same signal injection amplitude, when the motor speed is
low, even the low frequency signal injection still has a good
signal-to-noise ratio. When the motor speed is increased,
the harmonics of back EMF greatly reduces the signal-to-
(a) Estimation with 2 kHz injection frequency (yellow curve: estimation
noise ratio of the effective signal, resulting in HFI failure. angle)
When the injected signal is at higher frequency, the back
EMF harmonics will not affect the signal-to-noise ratio of
the injected signal. This is in aligned with the previous
theoretical analysis.
The challenge then becomes the selection of the injection
frequency. When the injection frequency increases, the
inductive impedance of the three-phase motor will increase
accordingly (XL=ωLm). To keep the same-amplitude
injected current, we need increase the amplitude of the
(b) Estimation with 10 kHz injection frequency (yellow curve: estimation
injected voltage, which may cause modulation index > 1, angle)
i.e., over modulation. Since the injection frequency is much
higher than the fundamental frequency, it is assumed that
the fundamental frequency and BEMF are almost constant
regardless of the injection signal. In addition, for simplicity
of the analysis, the injection signal is replaced with the
representative sinusoidal voltage (A∙sinωct). Therefore in
the steady state the injected HF current is:
A c Lm
ic  sin(ct  a tan ) (14)
Ra  (c Lm )
2 2 Ra

where Ra is phase internal resistance of PMSM, Lm is the (c) Angle error with 2 kHz injection frequency
motor leakage inductance, ic is the injected current, A is the
amplitude of the injected voltage, and ωc is injection
frequency. Assume the maximum modulation index is 1,
the fundamental angular velocity is ω0 and k is the phase
electromotive coefficient (Vrms/rad/s). The amplitude of
injected and fundamental voltage should not go beyond the
DC-bus limit, i.e.,
V
A  2k0  DC (15)
3
(d) Angle error with 10 kHz injection frequency
Combining (14) and (15) yields (16), setting the limit
between the fundamental and the injected frequency. FIGURE 11. Estimation and errors with different injection frequency
under the switching frequency 100 kHz and the speed 300rpm.
VDC
Ra 2  (c Lm )2 ic  2k0  (16)
3
E. IMPACT OF THE CONTROL BANDWIDTH
Following figures show simulation results of the
While it has been widely accepted that the higher the
estimation error under 2 kHz and 10 kHz injection
frequency scenarios. The switching frequency is 100 kHz switching frequency the better the modulation performance,
and the motor speed is 300 rpm. It can be seen that higher which can be simply realized by SiC devices over
injection frequency can avoid signal noise caused by high conventional Si devices, in this section we will focus on the
speed and reduce the estimation error effectively. When the control bandwidth, i.e., the updating speed of the algorithm.
injection frequency is 2 kHz, the error is obviously greater This has not been fully covered in previous literatures yet.
than that of injection frequency 10 kHz, though such error As shown in Fig. 12, a schematic diagram of different
might be acceptable in the applications. trigonometric waves (determining switching frequency)
generated by the up/down counter and different control
6

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access

bandwidth generation of SVPWM, the faster the updating injection contains following components:
rate (green signal) the finer the tuned signal, yielding  q V fq
smaller harmonics. Sometimes the switching frequency can i f  sin( f t )
be high, however, the control performance is still not  Zq

 q Vc
q
satisfactory if the updating rate (control bandwidth) is low. (17)
Fig. 13 is the spectrum contrasts of the motor phase ci  sin(c t )
 Zq
current. Both have 100 kHz switching frequency and 2 kHz
 Vq
injected signals. Their control bandwidth is different, i.e., iPWM
q
 PWM sin(PWM t )
10 kHz and 100 kHz, respectively. Obviously, the higher  Zq
the control bandwidth, the cleaner the signal thereby the
Here Zq is the three-phase equivalent impedance in the d-q
easier to extract the useful signal. q q q
system. Vf , Vc , VPWM are effective voltage after the
Frequency 2 Same Signal modulation in d-q system. ωf, ωc, ωPWM are the fundamental
with different
Frequency 1
update frequency, the injection frequency and the switching
q
frequency of the inverter, respectively.if is the fundamental
B q
A current of q-axis, ic is the current component of the
injection signal in the q-axis, iqPWM is the high frequency
harmonic component introduced by the switching
frequency. (13) multiplied by sin(ωct) yields:
V fq
i qf sin(c t )  sin( f t ) sin(ct )
Zq (18)
Frequency1+ A
V fq

Zq
 cos( c   f )t  cos(c   f )t 
Frequency2+A
Vcq
icq sin(ct )   cos(2ct )  1 (19)
2Z q
Frequency2+B
q
VPWM
FIGURE 12. Impact of the control bandwidth.
q
iPWM sin(ct )  sin(PWM t )sin(c t ) 
Zq (20)
q
VPWM
 cos(PWM  c )t  cos(PWM  c )t 
Zq
With the influence of the control bandwidth, sinωct is
delayed at least for τ. Here τ is the updating period, the
inverse of the updating frequency. This applies to the
injected signal, PWM and Analog-to-Digital Converter
(ADC). The lower the control bandwidth, the bigger value
of τ.
Id Injection

- +
(a) With the control bandwidth of 10 kHz. + PI
Ud
SVPWMPMSMADC
Clarke
BPF
Id_ref and Park LPF

τ
Sin(wt)

FIGURE 14. Delay caused by control bandwidth.

Vcq
sin(c t ) sin(c t   )
Zq (21)
V cos
q
V s in  q
 c
 cos 2ct  1  sin 2ct
c
2Z q 2Z q
(21) shows in the non-ideal case, the delay τ not only
(b) With the control bandwidth of 100 kHz. reduces the effective signal intensity Vcqcos 2Z q , but also
FIGURE 13. Impact of the different control bandwidth. creates more harmonic components. This is consistent with
the simulation results in Fig. 13.
Assume the current of the q-axis after the high frequency
7

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access
Author Name: Wei Qian, Xi Zhang, Fanning Jin, Hua Bai, Dingguo Lu and Bing Cheng

As a summary, the ideal setting of related frequency is

shown in Fig. 15, which describes the frequency relation of
the fundamental wave, the LPF, the injection signal, the
inverter switch and the control logic. It indicates:
1) A high control bandwidth reduces the distortion of the
fundamental and injection signals, which in return reduces
the effort of the filter design. Because FPGA is good at
parallel processing, it greatly shortens the operation time of
the FOC algorithm thereby allowing the switch duty cycle
to be updated every switching period. This has been
simulated in Fig.13 able to improve the signal to noise ratio
of the injected signal, thereby ease of the position signal
extraction. (a) Three phase current with the dead band of 2 µs
Mag

Switching
Cut-off frequency
frequency of and
non ideal control
LPF bandwidth

Fundamental Signal
frequency injection
(speed range) frequency
range

(b) Three phase current with the dead band of 250 ns

Frequency

FIGURE 15. The relationship among fundamental frequency, cut-off

frequency of LPF, switching frequency and control bandwidth.

2) A high switching frequency allows the wide range of

the injection frequency and motor fundamental frequency.
By using SiC devices, the inverter can endure high
switching frequency. Not only the weight and volume of
passive components but also the motor size can be greatly
shrunk [37]. In addition to allowing higher injection
frequency and widening the control bandwidth, SiC devices (c) Three phase current harmonics with the dead band of 2 µs
also adopts smaller dead band, which reduces the phase-
current distortion. During the dead band, the top and bottom
semiconductor devices are both off, with the current
commutating through the body diode. The voltage error
caused by dead band in each PWM cycle is

Td  Ton  Toff
Vp  Vd (22)
T
Here Td is dead band, Ton is turn-on interval for devices, Toff
is turn-off interval for devices, T is the switching period, Vd (d) Three phase current harmonics with the dead band of 250 ns
is DC bus voltage. Since SiC devices have much shorter FIGURE 16. Phase current Impact from the dead band
turn-on and turn-off delays, the dead band can be set as
~200 ns or even shorter, in contrast to >1µs in Si inverters.
The following simulation results (Fig. 16) show that a short III. SYSTEM MODELING BASED ON XSG TOOLBOX
dead band reduces the distortion of the phase current. In this paper, the model-based design (MBD) method is
utilized for the FOC algorithm, as shown in Fig. 17.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access

Implemented in Xilinx FPGA DC Bus

- +
Id Injection

- +
Id_ref + PI
Ud Ualfa
SVPWM
Inverse
Iq Park PWM-VSI
- Transform PWM
Iq_ref + PI
Uq Ubeta

^
^
 r
Vdqs ia
^
High ^ Clarke and ib

1/s r
PI frequency idqs Park ic
injection Transform Vdc

^
 r PMSM

FIGURE 17. Control diagram of whole system.

Since the FPGA supports a modular design, we divided FIGURE 19. Experimental platform.

the FOC algorithm into four modules (stages), with each TABLE I
module being built in the Simulink using XSG toolbox. PARAMETERS OF TESTED PMSM
Then the VHDL code for the FPGA can be automatically
Item Value Unit
generated. The model is built from the Xilinx conventional
module embedded in Matlab, aiming at the ease
maintenance and fast development. Vivado will then burn Rated Voltage 230 V
Current Continuous 3.0 A
functions into the FPGA and form a logical hardware array. Current Peak 9.6 A
At the same time, we can also see FPGA resources Rated Power 6.9 kW
consumed by the entire algorithm, as shown in Fig. 18. The Max Speed 3800 rpm
Rated Torque 2.44 Nm
HFI sensorless control only occupies small portion of the Peak Torque 7.38 Nm
FPGA. d-axis inductance 12 mH
q-axis inductance 34 mH
Stator Resistance 6.98 Ω

In Fig. 20, when the injection frequency is 500 Hz and

motor speed is 32.5 rpm, we can use HFI to estimate the
angle as shown in (a), indicating the angle error of 8 o.
When the speed increases to 139 rpm, the angle error
becomes larger and the control needs transition from HFI to
BEMF, using the hybrid observer. When the speed goes to
295 rpm, the angle is solely estimated by the BEMF
observer with the HFI disabled.

FIGURE 18. The occupancy and FPGA utilization.

IV. HARDWARE EXPERIMENTAL RESULTS

In order to verify the effectiveness of the proposed
design, the hardware setup is shown in Fig. 19, which
contains a SiC-based three-phased inverter, one IPMSM
and one MicroAutoBox with Xilinx XC7K325 FPGA chip
embedded. Parameters of the tested IPMSM are presented
in Table I. Test results, primarily the actual and estimated
angle, can be obtained from the control desk provided by
dSPACE. In the experiment, we set the control bandwidth
equal to the switching frequency.
(a) Estimated at 32.5 RPM, using HFI method.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access
Author Name: Wei Qian, Xi Zhang, Fanning Jin, Hua Bai, Dingguo Lu and Bing Cheng

(b) The position estimation when reversing the direction.

(b) Estimated at 139 RPM, using HFI + Back EMF.

(c) The position estimation under 1200 rpm.

(c) Estimated at 295 RPM, using pure back EMF.

FIGURE 20. Using hybrid controller for angle estimation.

With experiments carried out under 2 kHz injection (d) Estimation error with 500 Hz injection at 300 rpm electrical angle.
frequency and 100 kHz switching frequency, Fig. 21 shows
the comparison between actual and estimated position
under speed of 300 rpm, forward and reverse rotation and
1200 rpm, respectively. The purple signal is the real
position whereas the blue signal is the estimated position.
The error between the real and estimated position when
motor speed is 1200 rpm is <2.3%. No hybrid or BEMF
observer is needed. Experimental results verified that the
estimated angle is well aligned with the experimental data.
(e) Estimation error with 2000 Hz injection at 600 rpm electrical angle.

(a) The position estimation under 300 rpm.

(f) Estimation error with 500 Hz injection at 1200 rpm electrical angle.

10

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access

(g) Estimation error with 2000 Hz injection at 1200 rpm electrical angle.
(d) Three-phase current under 2 kHz injection frequency.
FIGURE 21. Position estimation and estimation error.
FIGURE 22. Three-phase duty cycle signals and current.

Fig. 22 shows reference waveforms under the HFI Experimental results were summarized as Fig. 23.
method. They are compared with the triangle wave to
produce the PWM, which is used to control three phase
inverters. Due to the injection of high frequency signals, the
reference waveform is superimposed with the injection
signal. The signal component will also be involved in the
modulation. The yellow, blue and purple lines are reference
waveforms of three phases. Furthermore, the injected
voltage harmonics is imposed to the fundamental, yielding
the fuzzy three-phase duty cycle signals.
Error(rad)

PM Speed
Injection （rpm）
Frequency
（Hz）

(a) Three-phase duty cycle signals with 500 Hz injection.

FIGURE 23. The relationship among error, injection frequency and
PMSM speed.

It shows the relationship among error of the angle

estimation, injection frequency and speed. It can be
concluded that with the increasing of the injection
frequency, the precision of the position estimation increases
too. In addition, within the tolerable angle estimation error,
the higher the injection frequency the wider the applicable
range of speed using the HFI sensorless control.
Furthermore, Fig. 24 is the curved surface based on the
(b) Three-phase duty cycle signals with 2000 Hz injection. experimental data, showing the relationship among error of
angle estimation, switching frequency, control bandwidth
and injection frequency. Increasing either the switching
frequency (control bandwidth) or injection frequency will
improve the precision of estimation. Note the experiment
sets the switching frequency equal to the control bandwidth.

(c) Three-phase current ripple under 500 Hz injection frequency.

11

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access
Author Name: Wei Qian, Xi Zhang, Fanning Jin, Hua Bai, Dingguo Lu and Bing Cheng

Dynamic Performance PMSM,” IEEE Transactions on Industrial

Electronics, vol. 57, pp. 2092-2100, June 2010.
[4] Xinda Song, Bangcheng Han, Shiqiang Zheng, Jiancheng Fang,
“High-Precision Sensorless Drive for High-Speed BLDC Motors
Based on the Virtual Third Harmonic Back-EMF,” IEEE
Transactions on Power Electronics, vol. 33, pp. 1528-1540, Feb.
2018.
[5] R. G. Krishnan, T. B. Isha, P. Balakrishnan, “A back-EMF based
sensorless speed control of permanent magnet synchronous
machine,” Power and Computing Technologies (ICCPCT), Kollam,
Error(rad)

India, 2017, pp.1-5.

[6] Sukanta Halder, Pramod Agarwal, S P Srivastava, “MTPA based
Sensorless Control of PMSM using position and speed estimation by
Back-EMF method,” Power Systems (ICPS), New Delhi, India, 2016,
pp. 1-4.
[7] Gang Liu, Chenjun Cui, Kun Wang, Bangcheng Han, Shiqiang
Zheng, “Sensorless Control for High-Speed Brushless DC Motor
Injection Switching Frequency & Based on the Line-to-Line Back EMF,” IEEE Transactions on Power
Frequency control bandwidth（kHz） Electronics, vol. 31, pp. 4669-4683, July 2016.
（Hz）
[8] Cheng Zhou, Changying Xie, Maosong Zhang, “Speed-sensorless
FIGURE 24. Angle error, switching frequency (control bandwidth) and control for induction motor based on back-EMF integration,”
injection frequency. Industrial Electronics and Applications (ICIEA), Auckland, New
Zealand, June 2015,pp. 1376-1379.
[9] Jin-Woo Lee, “Adaptive sensorless control of high speed PMSM
with back EMF constant variation,” Power Electronics and ECCE
V. CONCLUSIONS Asia (ICPE-ECCE Asia), Seoul, South Korea, June 2015. pp. 1400-
This paper presented a high-frequency injection 1404.
sensorless position estimation by using FPGA and SiC [10] Yingqiang Li, Ming Yang, Jiang Long, Zirui Liu, Dianguo Xu,
“Current sensorless predictive control based on extended kalman
inverter. Due to the fastness and high compactness of filter for pmsm drives,” Transportation Electrification Asia-Pacific
FPGA, the control bandwidth can be much higher than (ITEC Asia-Pacific), Harbin, China, Aug. 2017,pp. 1-6.
most of conventional microcontroller. The high switching [11] D. Raca, M. C. Harke, and R. D. Lorenz, “Robust magnet polarity
frequency and control bandwidth together make the high estimation for initialization of PM synchronous machines with near
injection frequency (~2 kHz) possible and in result secure zero saliency,” IEEE Trans. Ind. Appl., vol. 44, no. 4, pp. 1199-1209,
Jul./Aug. 2008.
both the high precision of position estimation and the wide
[12] H. Kim, K.-K. Huh and R. D. Lorenz, “A Novel Method for Initial
applicable range of the motor speed. Experimental results Rotor Position Estimation for IPM Synchronous Machine Drives,” in
validated the effectiveness of proposed sensorless position Conf. Rec. IEEE IAS Annu. Meeting, Oct. 2003, vol. 2, pp.1173-
estimation in FPGA. The resources occupied inside the 1180.
FPGA was calculated as well. [13] M. Corley, R. D. Lorenz, “Rotor Position and Velocity Estimation
for a Permanent Magnet Synchronous Machine at Standstill and High
Given the advantage of SiC devices is well known in Speeds,” in Conf. Rec. IEEE IAS Annu. Meeting, Oct. 1996, vol. 1,
pp. 36–41.
power electronics domain, we skipped the discussion of the
[14] Y.Jeong, R. D. Lorenz, “Initial Rotor Position Estimation of an
SiC inverter and only focused on the wide control Interior Permanent-Magnet Synchronous Machine Using Carrier-
bandwidth of FPGAs. From this aspect, this paper proves Frequency Injection Methods,” in Conf. Rec. IEEE IEMDC’03,
that the high control bandwidth of FPGAs effectively Madison, WI, USA, 2003, pp. 1218-1223.
maximizes the high-switching-frequency capability of [15] H. Kim, M. C. Harke, and R. D. Lorenz, “Sensorless control of
interior permanent magnet machine drives with zero-phase lag
wide-bandgap semiconductors. Though this paper has lifted position estimation,” in Conf. Rec. IEEE-IAS Annu. Meeting,
the injection frequency to 2 kHz, the future work is to seek Pittsburgh, PA, USA, 2002, pp. 1661-1667.
the optimal injection frequency of the sensorless control. [16] D. Raca, P. Garcia, “Carrier-Signal Selection for Sensorless Control
of PM Synchronous Machines at Zero and Very Low Speeds,” in
Conf. Rec. IEEE-IAS Annu. Meeting, 2008, pp. 1-8.
[17] Liu, J. M., and Z. Q. Zhu, “Sensorless control strategy by square-
waveform high-frequency pulsating signal injection into stationary
REFERENCES reference frame,” IEEE Journal of Emerging and Selected Topics in
[1] C.R.Harahap, R. Saito, H.Yamada, T.Hanamoto, “Speed control of Power Electronics, vol. 2, no. 2, pp.171-180, 2014.
permanent magnet synchronous motor using FPGA for high
frequency SiC MOSFET Inverter,” Journal Engineering Science and [18] Kim, Dongouk, et al, “Suppression of injection voltage disturbance
Technology, pp. 11-20, 2014. for high-frequency square-wave injection sensorless drive with
regulation of induced high-frequency current ripple,” IEEE
[2] P. K. Sharma and A. S. Sindekar, "Performance analysis and Transactions on Industry Applications, vol. 52, no. 1,pp. 302-312,
comparison of BLDC motor drive using PI and FOC," 2016 2016.
International Conference on Global Trends in Signal Processing,
Information Computing and Communication (ICGTSPICC), Jalgaon, [19] Xu, P. L., and Z. Q. Zhu, “Novel square-wave signal injection
2016, pp. 485-492. method using zero-sequence voltage for sensorless control of PMSM
drives,” IEEE Transactions on Industrial Electronics, vol. 63, no. 12
[3] Fabio Genduso, Rosario Miceli, Cosimo Rando, Giuseppe Ricco pp. 7444-7454, 2016.
Galluzzo, “Back EMF Sensorless-Control Algorithm for High-

12

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access

[20] Ni, Ronggang, et al, “Square-wave voltage injection algorithm for Wei Qian received the B.S. degree
PMSM position sensorless control with high robustness to voltage
in Mechanical Engineering and
errors,” IEEE Transactions on Power Electronics, vol. 32, no. 7, pp.
5425-5437, 2017. Automation from Nanjing University
[21] Alaei, Ahmadreza, et al, “Dynamic performance analysis of high of Aeronautics & Astronautics
frequency signal injection based sensorless methods for IPM (NUAA), Nanjing, China, in 2005
synchronous motors,” Power Electronics, Drives Systems and and the M.S. degree in Aircraft
Technologies Conference (PEDSTC), 2018, pp. 151-156. Design from Nanjing University of
[22] E. Trancho，E. Ibarra，A. Arias, “A novel PMSM hybrid sensorless Aeronautics & Astronautics
control strategy for EV applications based on PLL and HFI,”
Industrial Electronics Society, IECON 2016-42nd Annual
(NUAA), Nanjing, China, in 2013.
Conference of the IEEE，Florence, Italy, Oct. 2016, pp. 6669-6674. He is currently a Ph.D. candidate at
[23] D. Hind et al., “A hybrid sensorless control solution for an the School of Mechanical
automotive drive application,” 2017 IEEE International Electric Engineering, Shanghai Jiao Tong University (SJTU),
Machines and Drives Conference (IEMDC), Miami, FL, 2017, pp. 1- Shanghai, China. His research interests include soft
6. switching power converters, power electronics devices and
[24] Z. Chen, Z. Zhang, R. Kennel and G. Luo, “Hybrid sensorless electric motor control systems.
control for SPMSM With multiple saliencies,” IECON 2015 - 41st
Annual Conference of the IEEE Industrial Electronics Society,
Yokohama, 2015, pp. 001188-001193. Xi Zhang received the B.Sc. degree in
[25] H. M. Dawood Habbi and A. A. A. Al-Khazraji, “FPGA based vector Applied Mathematics and the B.E.
control of PM motor using sliding mode observer,” 2015 Modern degree in Information and Control
Electric Power Systems (MEPS), Wroclaw, 2015, pp. 1-5. Engineering from Shanghai Jiao Tong
[26] Slimane Medjmadj, Demba Diallo, Mohammed Mostefai, “PMSM University (SJTU), Shanghai, China in
Drive Position Estimation: Contribution to the High-Frequency
Injection Voltage Selection Issue,” IEEE Transactions on Energy 2002. He received the M.E. and Ph.D.
Conversion, pp. 349-358, March 2015. degrees in Power Electronics and
[27] Ying Fan, Rui Wang, and Li Zhang, “Sensorless Control of Five- Electric Power Drive from SJTU, in
phase IPM Motor Based on High-Frequency Sinusoidal Voltage 2004 and 2007, respectively. From
Injection,” Electrical Machines and Systems (ICEMS), Aug. 2017, September 2007 to July 2009, he held
pp.1-5.
a postdoctoral position with the Department of Electrical
[28] Zhixun Ma, Jianbo Gao, and Ralph Kennel, “FPGA Implementation
of a Hybrid Sensorless Control of SMPMSM in the Whole Speed and Computer Engineering, the University of Michigan-
Range,” IEEE Transactions on Industrial Informatics, pp. 1253- Dearborn, Dearborn, MI, USA. He is currently an associate
1261, Aug. 2013. professor with the Institute of Automotive Engineering and
[29] K.Shirabe. M. Swamy, J.K. Kang, M.Hisatsune, Y.Wu, D.Kebort, National Engineering Lab for Automotive Electronics and
and J.Honea, “Advantages of high frequency PWM in AC motor
drive applications,” IEEE Energy Conversion Congress and
Control Technology, SJTU. His research interests include
Exposition, Releigh, NC, 2012, pp. 2977-2984. power management strategies, power electronics devices
[30] Zhaoyong Zhou, Tiecai Li, T, Takahashi, E. Ho, “FPGA realisation and electric motor control systems for alternative-fuel
of a high performance servo controller for PMSM,” APEC '04. vehicles.
Nineteenth Annual IEEE, 2004, pp. 1604-1609. Fanning Jin (S’16) was born in
[31] Ying-Shieh Kung, Ming-Hung Tsai, “FPGA-based speed control IC China in 1992. He received the B.E.
for PMSM drive with adaptive fuzzy control,” IEEE Transaction on
degree in Automation from
Power Electronics,vol. 22, No. 6, pp. 2476-2486, 2007.
Huazhong University of Science &
[32] Hanamoto, Tsuyoshi, et al, “Digital hardware circuit using FPGA for
speed control system of permanent magnet synchronous Technology, Hubei, China, in 2015,
motor,” ICEM 2008. 18th International Conference, 2008, pp. 1-5. and M.S.E degrees in electrical
[33] Lai, Rixin, et al, “High-voltage high-frequency inverter using 3.3 kV engineering from the University of
SiC MOSFETs,” Power Electronics and Motion Control Conference Michigan-Dearborn, Michigan,
(EPE/PEMC), Novi Sad, Serbia, 2012, pp. DS2b.6-1 - DS2b.6-5. USA, in 2017, respectively. His
[34] Shen, Miaosen, Shashank Krishnamurthy, and Mihir Mudholkar, research interests include Field-
“Design and performance of a high frequency silicon carbide
inverter,” Energy Conversion Congress and Exposition (ECCE), Programmable Gate Array (FPGA) utilization in various
Phoenix, AZ, USA, 2011, pp. 2044 - 2049. control algorithm (e.g., motor control) and hybrid energy
[35] Lai, Chunyan, et al, “Direct Calculation of Maximum-Torque-per- storage system for EV applications.
Ampere Angle for Interior PMSM Control Using Measured Speed
Harmonic,” IEEE Transactions on Power Electronics, pp.1-1, 2018. Hua Bai received B S and PHD
[36] Jihua Ma, Jin Zhao, and Jiajiang Sun, “A novel PMSM speed control degree in Department of Electrical
scheme based on sliding-mode and fuzzy disturbance
observer,” Industrial Electronics Society, IECON 2017, Beijing, Engineering of Tsinghua
China, 2017, pp. 1704-1710. University, Beijing, China in 2002
[37] Li, Yingjie, Di Han, and Bulent Sarlioglu, “Design of high-speed and 2007, respectively. He was a
machines using silicon-carbide based inverters,” Energy Conversion post-doc fellow and research
Congress and Exposition (ECCE), 2015, pp. 3895-3900.
scientist in Univ of Michigan-
Dearborn, USA, in 2007 and 2009,

13

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 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/ACCESS.2018.2858199, IEEE Access
Author Name: Wei Qian, Xi Zhang, Fanning Jin, Hua Bai, Dingguo Lu and Bing Cheng

respectively. He was an assistant professor in Department interests include control systems, electric machines, and
of Electrical and Compurter Engineering, Kettering power electronics in electric/hybrid vehicle applications.
University, MI, USA in 2010~2016. In 2017~2018 he
joined University of Michigan-Dearborn as associate
professor. He is currently the associate professor in EECS,
University of Tennessee, Knoxville. His research interest is
the power electronic modelling, control and integration
including variable frequency motor drive system, high
voltage and high power DC/DC converter, wide-bandgap
devices and hybrid electric vehicles.

Dingguo Lu (S’09-M’18)
received the B. Eng. degree in
mechanical engineering from
Zhejiang University, Hangzhou,
China, in 1997, the M.S. degree
in mechanical engineering and the
Ph.D. degree in electrical
engineering from the University
of Nebraska–Lincoln, Lincoln,
NE, USA, in 2009 and 2018, respectively. He is currently
an engineer with the Mercedes-Benz Research &
Development North America, Inc., Redford, MI, USA. His
research interests include electric machines and drives,
power electronics, renewable energy systems, machine
learning and its applications in hybrid-electric vehicles.

Bing Cheng (S’90–M’97) received

the B.S. and M.S. degrees from
Northeastern University, Shenyang,
China, in 1982 and 1984, respectively,
and the Ph.D. degree from the
University of Massachusetts, Amherst,
in 1992, all in electrical engineering.
From 1992 to 1994, he was with
Cleveland Machine Controls, where he
was responsible for ac induction motor
control development for industrial drives. In 1994, he
joined Ford Motor Company— Ecostar Electric Drives,
LLC, which was acquired by Ballard Power Systems,
Siemens VDO, and Continental Corporation. As a principal
engineer, he performed research and development work on
motor control software development, power electronic and
system simulation for fuel-cell and hybrid vehicles. From
2010 to 2015, he worked at Fiat Chrysler Automobiles as
the E-Motor Controls and Integration manager, where he
was responsible for motor control, motor calibration and
software development for all the battery/hybrid electric
vehicles programs. In 2015, he joined Mercedes-Benz
Research and Development North America as a motor
control and calibration manager. He is currently
responsible for E-drive motor control research and software
development for hybrid and electric vehicle applications.
He is also an Adjunct Professor in Electrical Engineering
Department at McMaster University since 2013. His

14

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.