A Parameter Recognition Based Impedance Tuning Method For SS-Compensated Wireless Power Transfer Systems

This article has been accepted for publication in IEEE Transactions on Power Electronics.
This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2023.3302256
JOURNAL OF LATEX CLASS FILES, VOL. XX, NO. XX, DECEMBER 2022 1
A Parameter Recognition Based Impedance Tuning

Method for SS-Compensated Wireless Power
Transfer Systems
Gangwei Zhu, Student Member, IEEE, Jianning Dong, Senior Member, IEEE, Francesca Grazian, Student
Member, IEEE, and Pavol Bauer, Senior Member, IEEE
Abstract—This paper presents a parameter recognition based deviate from the nominal values [8]. Additionally, in practical
impedance tuning method for the impedance mismatch caused WPT applications, the metallic shielding plates are usually
by capacitance drift and coil misalignment in series-series (SS)- used in the transmitter (Tx) and receiver (Rx) coils to protect
compensated wireless power transfer (WPT) systems. First,
a parameter recognition method is proposed to identify the human and electronic devices from the stray magnetic field,
unknown parameters of the resonant circuits by only measuring while magnetic ferrites are often employed to improve the
the RMS values of the coil currents. No phase detection circuits coil coupling. As a result, the self-inductances of the coils
and auxiliary measurement coils are required. Furthermore, become sensitive to the spatial displacement between the
according to the recognized parameters, the reactance on both Tx and Rx pads as these metallic and magnetic materials
sides are minimized simultaneously by regulating the system
frequency and the phase shift angles of the active rectifier. interfere with the transmission of the magnetic flux [9]. The
Compared with the existing methods, the proposed parameter parameter deviations caused by the capacitance drift and coil
recognition method adopts a dynamic frequency approaching misalignment lead to detuning of the system, which reduces
strategy to avoid severe system detuning due to the bifurcation the transmission efficiency and results in a lower power
phenomenon. Moreover, based on the recognized parameters, factor [10]. Therefore, in order to obtain better transmission
the proposed impedance tuning method can simultaneously cope
with the parameter deviations caused by capacitance drift and performance, it is desirable for the system to consistently
coil misalignment on both sides without using extra circuits operate at the resonance state regardless of the capacitance
and switches. Experimental results show that the unknown drift and coil misalignment.
parameters of the resonant circuits are recognized accurately, Many works have been done to deal with the impedance
with the average relative errors all less than 3%. Additionally, mismatch caused by parameter deviations, which can be
by implementing the impedance tuning method, the dc to dc
efficiency of the WPT prototype is improved by 4.3%-15% in roughly divided into three categories: 1) using the variable
the experiments. capacitors or inductors [9], [11]–[15]; 2) tracking the resonant
frequency [16], [17]; 3) adopting the active rectifier [18], [19].
Index Terms—Impedance tuning, parameter recognition, coil
misalignment, capacitance drift, wireless power transfer (WPT). A commonly used approach for impedance tuning is to ad-
just the compensation impedance. In [11], a variable inductor
was implemented in the secondary side to dynamically tune
the receiver circuit. However, two extra conversion stages,
I. I NTRODUCTION
i.e., a diode rectifier and a buck converter, are required in
As an emerging technology in recent decades, wireless this method, which greatly increases the hardware costs. In
power transfer (WPT) enables safe, convenient, and automated [12]–[14], by introducing the selectable capacitor array, the
charging in many industrial applications, such as portable elec- compensation capacitance is discretely regulated. Neverthe-
tronics [1], underwater loads [2], implanted medical devices less, the use of numerous switches and capacitors increases
[3] and electric vehicles (EVs) [4]–[6]. To obtain a higher the system volume and costs. Additionally, this method fails
transmission efficiency, various compensation topologies are to continuously adjust the impedance. To reduce the number of
introduced in WPT systems to cancel the leakage inductance components and realize continuous tuning, controllable capac-
of the loosely coupled coils, where the most widely-used itors were used in [9] and [15]. In [15], a voltage-controlled
one is the series-series (SS) compensation [7]. In the SS- capacitor was proposed to achieve dynamic impedance tuning.
compensated WPT system, the series capacitors are designed Nevertheless, the analog amplifier used in this method limits
to be resonant with the coils at the nominal resonant frequency. the power transfer capability, and thus, it is only suitable for
However, due to the component tolerances, temperature vari- applications of a few watts. In [9], a pulse-width modulation
ations and aging effects, the compensation capacitance may (PWM) controlled capacitor was proposed to withstand the
detuning caused by the Tx inductance variation. This method
The work of Gangwei Zhu was supported by China Scholarship Council
under Grant 202106230112. (Corresponding author: Jianning Dong.) enables continuous capacitance tuning and is able to cope
Gangwei Zhu, Jianning Dong, Francesca Grazian, Pavol Bauer are with with high-power applications. However, the introduction of
the DC System, Energy Conversion and Storage (DCE&S) Group, Faculty of additional power switches in the PWM-controlled capacitors
Electrical Engineering, Mathematics and Computer Science (EEMCS), Delft
University of Technology, 2624 CP Delft, The Netherlands (e-mail: g.zhu- increases the power losses and hardware costs. Additionally,
2@tudelft.nl, j.dong-4@tudelft.nl, f.grazian@tudelft.nl, p.bauer@tudelft.nl). the capacitance drift is not considered in this method.
Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on August 19,2023 at 21:14:39 UTC from IEEE Xplore. Restrictions apply.
© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Power Electronics. This is the author's version which has not been fully edited and
Another typical method to deal with the impedance mis- deviations.

match is to track the resonant frequency. In [16], a power- To fill up the above research gaps, this paper proposes a new
frequency controller was proposed, which implements the parameter recognition based impedance tuning method. Firstly,
resonance tracking by synchronizing the switching signals a parameter recognition method is proposed to identify the
with the resonant current. In [17], a self-oscillating switching unknown parameters of the resonant circuits, and the original
technique was adopted, where the switching frequency of the features of this method are summarized as follows:
inverter automatically tracks the resonant frequency. However, 1) Compared to the conventional front-end parameter mon-
these methods can only realize zero-phase-angle (ZPA) input, itoring methods, the secondary-side information is intro-
whereas the reactance in the Rx side cannot be tuned. duced, and therefore, all of the unknown parameters of
To tune the Rx-side reactance, the active control of the the resonant circuits, including the self-inductances, the
rectifier was adopted in some methods. In [18] and [19], by mutual inductance and the compensation capacitances,
regulating the duty cycle and the phase shift angle of the are recognized accurately, without using any phase de-
active rectifier, the equivalent load impedance was regulated to tection circuits and auxiliary measurement coils;
minimize the Rx-side reactance and to realize the maximum 2) A dynamic frequency approaching strategy is proposed
efficiency tracking. However, the active rectifier control cannot to acquire the required data for parameter recognition,
simultaneously handle the impedance mismatch on both sides. which avoids severe system detuning during the tradi-
Recently, some researchers combine the resonant frequency tional frequency-sweep process;
tracking and the active rectifier control to tune the reactance 3) The parameter recognition is implemented at pre-start-
on both the Tx and Rx sides. In [20], an impedance tuning up, with the rectifier output short-circuited during this
control was proposed to deal with the self-inductance varia- process, which avoids significant load power fluctuations
tions caused by coil misalignment, where the switching fre- caused by frequency variations;
quency was controlled to realize ZPA input and a synchronous 4) A simple yet powerful heuristic algorithm called JAYA
switching technique was proposed for the active rectifier is introduced to recognize the unknown parameters.
to minimize the Rx-side reactance. However, the parameter Compared with the conventional genetic algorithm (GA)
deviations caused by capacitance drift were not discussed in and differential evolution (DE) algorithm, which need
this method. In [21], an impedance decoupling based method to tune the parameters of the differential weight and
was proposed to minimize the dual-side reactance caused by the crossover rate, the JAYA algorithm is free from
capacitance drift. Nevertheless, the phase detection circuits algorithm-specific parameters and thereby avoids the
and an auxiliary measurement coil are needed to measure the difficulty of tuning parameters [28].
reactance angles on both sides, which increases the hardware Furthermore, based on the recognized parameters, the reac-
costs and complexity. tance on both sides are minimized simultaneously by regu-
The phase detection circuits for reactance angle measure- lating the system frequency and the phase-shift angles of the
ments can be avoided by parameter recognition. In [22], active rectifier.
Dai et al. presented a pulse density modulation based pa- The remaining part of this paper is organized as follows.
rameter identification method to simultaneously estimate the Section II introduces the configuration of the SS-compensated
self-inductance and compensation capacitance on the Rx- system and analyses the impact of the impedance mismatch
side. However, only the Rx-side reactance is tuned in this caused by capacitance drift and coil misalignment. In Section
method. In [23], a parameter recognition based self-tuning III, the proposed parameter recognition method is described,
method was proposed to address a wide range of coupling and the corresponding impedance tuning method is presented.
coefficient variations, where both the mutual inductance and In Section IV, experimental results are given to validate this
the self-inductances can be identified without using the phase proposal. Finally, Section V concludes this paper.
information and auxiliary measurement coils. However, this
method cannot identify the compensation capacitances.
II. I MPEDANCE M ISMATCH C AUSED BY C APACITANCE
In practice, the impedance mismatch usually involves de-
D RIFT AND C OIL M ISALIGNMENT
viations of multiple parameters, and it is challenging to
tune the impedance when considering all of them. Therefore, A. System Configuration
previous studies mainly focus on parameter deviations in one As shown in Fig. 1, a typical series-series (SS) compensated
specific aspect, for instance, only considering the coil mis- WPT system is studied in this paper, where an active rectifier
alignment [9], [20], [23]–[25], the capacitance drift [21], or the is adopted in the secondary-side. The dc input voltage is
impedance mismatch on just one side [22]. To simultaneously represented by Uin , while the dc output voltage and current
tune the dual-side impedance mismatch under capacitance drift are denoted by Uout and Iout , respectively. Moreover, vab
and coil misalignment, the parameters of the whole resonant and vcd are the ac voltages of the inverter and rectifier; iP
circuit are required to be identified. Although the identification and iS are the transmitter and receiver coil currents; LP and
of multiple parameters can be realized by the frequency- LS represent the self-inductances of the coupled coils; CP
sweep based front-end monitoring methods [26], [27], these and CS are their corresponding compensation capacitors; RP
methods generally sweep the frequency around the nominal and RS are the equivalent loss resistances of the primary
resonant frequency, which leads to severe system detuning and secondary resonant circuits; RL is the resistive load.
√ The
and significant load power ripple especially under parameter mutual inductance of the coils is denoted by M = k LP LS ,
350 230 350 230

S1 S3 M S5 S7 Iout
iP CP CS iS 345
225
345
225
a + c 340 340
Uin + vcd
+ 220 220
vab LP LS Cfo Uout RL 335 335
-
b - - d 330
215
330
215
RP RS
S2 S4 S6 S8 325 210 325 210
Zin ZE 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10
Misalignment in x axis [cm] Misalignment in y axis [cm]
Fig. 1. Circuit diagram of the SS-compensated WPT system using the active (a) (b)
rectifier.
340 225 120 0.4
0.35
IP CP M CS IS 335 220 100
0.3
+ 80 0.25
VP ZE=RE+jXE 330 215
- 60
0.2
325 210 0.15

10 11 12 13 14 15 10 11 12 13 14 15
Zin RP RS
(c) (d)
Fig. 2. Equivalent circuit model of the WPT system.
Fig. 4. Measured coil inductances under different coil misalignment directions
Shielding layer Receiver coil and air gaps: (a) self-inductance variations under the x-direction misalignment
(D=10 cm), (b) self-inductance variations under the y-direction misalignment
z
(D=10 cm), (c) self-inductance variations under different air gaps, (d) varia-
x
tions in mutual inductance and coupling coefficient under different air gaps.
y
TABLE I
S YSTEM PARAMETERS OF THE SS- COMPENSATED WPT SYSTEM
Transmitter coil
Ferrite layer Symbol Parameters Value Unit
LP Primary coil inductance 327.5 ∼ 335.5 µH
Fig. 3. Charging pads with the shielding and ferrite layers.
LS Secondary coil inductance 216.5 ∼ 222.7 µH
k Coupling coefficient 0.22 ∼ 0.35 /
CP 0 Nominal primary capacitance 10.45 nF
where k is the coupling coefficient. The equivalent circuit CS0 Nominal secondary capacitance 15.74 nF
model derived by fundamental frequency approximation is dP Degree of primary capacitance drift -0.2∼0.2 /
further presented in Fig. 2, where ZE is the equivalent load dS Degree of secondary capacitance drift -0.2∼0.2 /
RP Primary loss resistance 0.72 Ω
impedance regulated by the active rectifier; V̇P , I˙P , and I˙S RS Secondary loss resistance 0.48 Ω
are the phasor forms of the fundamental components of vab , RL Load resistance 100 Ω
iP , iS , respectively. According to the Kirchhoff’s voltage law, fN Nominal switching frequency 85 kHz
Uin DC input voltage 200 V
the steady state equation of the system is given by
V̇P = ZP I˙P − jωM I˙S

(1) coil inductances under different misalignment directions and
0 = jωM I˙P − (ZS + ZE )I˙S . air gaps are illustrated in Fig. 4 for the charging pads shown
The equivalent impedances of primary and secondary resonant in Fig. 3. As it can be observed from Fig. 4, the coil self-
circuits ZP and ZS are expressed as inductances slightly increase as the misalignment in the x- and
y- directions increases. More significant inductance variations
Zi = Ri + jXi , Xi = ωLi − 1/(ωCi ), (2) can be observed under different air gaps. As the air gap D
where the subscript i indicates the primary (P) or the secondary increases from 10 cm to 15 cm, the primary self-inductance
(S) side. Moreover, the equivalent load impedance ZE is LP drops from 335.5 µH to 327.5 µH, while the secondary
represented by ZE = RE + jXE , where RE and XE are self-inductance LS decreases from 222.7 µH to 216.5 µH.
the equivalent load resistance and reactance, respectively. Meanwhile, the increase in the air gap reduces the coupling
coefficient k from 0.35 to 0.22. In this paper, two cases, with
the air gap D at 10 cm and 15 cm, are selected as case-A and
B. Parameter Deviations Caused by Coil Misalignment and case-B for theoretical analysis, respectively.
Capacitance Drift On the other hand, due to the ambient temperature varia-
Ideally, for the SS-compensated WPT system, the compen- tions, the aging effect, and the manufacturing errors, signifi-
sation capacitors are designed to resonate with the coupled cant capacitance drift may occur in the practical applications.
coils at the nominal resonant frequency. Nevertheless, in prac- Therefore, it is essential to investigate the influence of the
tical applications, the parameters of the resonant circuits may capacitance drift. In this paper, both the primary and secondary
deviate from the nominal values due to the coil misalignment capacitance deviations are considered and expressed as
and capacitance drift.
CP = (1 + dP )CP 0 , CS = (1 + dS )CS0 , (3)
On one hand, as the aluminum shielding sheets and mag-
netic ferrites are generally used in the Tx and Rx pads, the where dP and dS indicate the degree of capacitance drift; CP 0
coil self-inductances vary with the spatial displacement. To and CS0 are the nominal compensation capacitances. Consid-
demonstrate the impact of coil misalignment, the measured ering component tolerances of the commercialized capacitors,
1000 dP=-0.2, dS=-0.2 dP=0, dS=-0.2 dP=0.2, dS=-0.2 1.0

0.98
dP=-0.2, dS=0 dP=0, dS=0 dP=0.2, dS=0
800 dP=-0.2, dS=0.2 dP=0, dS=0.2 dP=0.2, dS=0.2 0.95
0.9
0.97
Pout [W] 600 Power is limited
0.85
η[%]
η[%]
400 0.8
0.96
0.75 dS=0
dS=0
200 dS=0.2 0.7 dS=0.2
dS=-0.2 dS=-0.2
0.95 0.65
0 200 300 400 500 600 700 800 200 300 400 500 600 700 800
0 20 40 60 80
RE [Ω] Pout [W] Pout [W]
(a) (a) (b)

dP=-0.2, dS=-0.2 dP=0, dS=-0.2 dP=0.2, dS=-0.2
Fig. 6. Transfer efficiency η versus the output power Pout under parameter
2500 dP=-0.2, dS=0 dP=0, dS=0 dP=0.2, dS=0 deviations: (a) Case-A, (b) Case-B. Herein, the equivalent load resistance RE
dP=-0.2, dS=0.2 dP=0, dS=0.2 dP=0.2, dS=0.2
2000
is adjusted to obtain the same output power under different values of dS .
Pout [W]
1500
1000 Power is limited 2) Transfer Efficiency: Another effect of the parameter

deviations is on the transfer efficiency. Based on (1), the
500
transmission efficiency of the resonant tank is obtained as
0
0 20 40 60 80
RE [Ω] |I˙S |2 RE ω 2 M 2 RE
η = = 2 2
2 . 2
(b) Re[V̇P (I˙P )∗ ]
ω M RST + RP (RST + XST )
Fig. 5. Output power Pout versus the equivalent load resistance RE under
(5)
parameter deviations: (a) Case-A, (b) Case-B. As it can be observed in (5), the secondary lumped reactance
XST that exists in the denominator degrades the transfer
efficiency, whereas the primary reactance XP does not directly
the aging effects and the ambient temperature variations, the affect the efficiency. Assume XE = 0 and dP = 0, the
maximum degree of capacitance drift is set to ±20% to ensure efficiency η versus the output power Pout under parameter
that the proposed WPT system is able to deal with the majority deviations is shown in Fig. 6. As it can be observed in Fig.
of the capacitance drift cases in practice. The investigated 6(a), the transfer efficiency of the resonant tank is degraded
capacitance drift range follows the analysis in [21]. due to the capacitance drift, and the efficiency drop is more
With the parameter deviations caused by capacitance drift noticeable under light load conditions. Moreover, as shown in
and coil misalignment taken into account, the parameters of Fig. 6(b), since the air gap D is increased to 15 cm in case-B,
the SS-compensated system are listed in Table I. the secondary self-inductance LS is reduced from 222.7 µH to
216.5 µH. In this case, the secondary reactance XS becomes
larger when the secondary capacitance is also decreased, and
C. Impact of Parameter Deviations
thus, remarkable efficiency reduction can be observed when
1) Power Transfer Capability: According to (1), the output dS = −0.2.
power of the resonant tank is derived by According to the above analysis, the parameter deviations
caused by coil misalignment and capacitance drift reduce the
Pout = |I˙S |2 RE = ω 2 M 2 |V̇P |2 RE /DE , (4) system transfer efficiency and may result in a failure to obtain
the required output power. Therefore, it is desirable to cancel
where DE = (XP2 + RP 2 2
)(RST + XST2
) + 2ω 2 M 2 (RP RST − the dual-side reactance caused by parameter deviations, so that
4 4
XP XST ) + ω M , RST = RS + RE , and XST = XS + XE . the system can always operate at the resonance state.
As illustrated in (4), the parameter deviations of the coil induc-
tances and compensation capacitances influence the primary
III. P ROPOSED PARAMETER R ECOGNITION BASED
and secondary reactance XP and XS , which in turn affects the
I MPEDANCE T UNING M ETHOD
output power. Assume the equivalent load reactance XE = 0,
the output power Pout versus the equivalent load resistance A. System Detuning Caused by Frequency Variations
RE under parameter deviations is shown in Fig. 5. As it can To identify the unknown parameters, conventional parameter
be observed in Fig. 5(a), the output power curve has a peak recognition methods generally sweep the frequency around the
point when the degree of capacitance drift is dP = dS = −0.2. nominal resonant frequency [26], [27]. However, for the SS-
This means that the power transfer capability of the system is compensated WPT system, the nominal resonant frequency
limited no matter how the load resistance RE is adjusted, and may split into multiple resonant frequencies due to the param-
therefore, the system may not be able to reach the required eter deviations, which is the so-called bifurcation phenomenon
output power. Similar power limitations can also be observed [29]. Since these resonant frequencies are close to the nomi-
in Fig. 5(b). Compared with case-A, since the air gap D nal frequency, the traditional frequency-sweep process causes
of case-B is increased from 10 cm to 15 cm, the mutual severe system detuning under parameter deviations, which
inductance is correspondingly reduced from 95 µH to 58 µH, results in enormous coil currents and significant power ripple.
and thus, the system can reach more power. However, in some To demonstrate the influence of the system detuning on the
cases, the output power of the system is still limited due to the coil currents, the RMS value of the fundamental component
capacitance drift, for example, the case of dP = 0.2, dS = 0. of the primary coil current IP 1 under parameter deviations
40
Converntional dP=-0.2, dS=-0.2
35 frequency-sweep band dP=-0.2, dS=0
Peak current dP=-0.2, dS=0.2
30 point (75.9, 37.2)
A(5, 0.22, 101.8)
dP=0, dS=-0.2
dP=0, dS=0
25 dP=0, dS=0.2
MAX_IP1_Peak [A]
Nominal frequency
IP1 [A]
point (85, 4.6) dP=0.2, dS=-0.2

20 dP=0.2, dS=0
dP=0.2, dS=0.2
15
10
5
0
60 70 75 80 90 95 100 110 120 130
Frequency [kHz]
B(81, 0.35, 11.6) k
RE [Ω]
(a)
100
Converntional
Fig. 9. Value of M AX IP 1 P eak under different values of RE and k.
dP=-0.2, dS=-0.2
90 frequency-sweep band dP=-0.2, dS=0
Peak current
80 point (75.7, 95.8) dP=-0.2, dS=0.2
dP=0, dS=-0.2
70 dP=0, dS=0 under case-B, the output power at the nominal frequency point
Nominal frequency dP=0, dS=0.2
60
point (85, 1.3) is 800 W, with the primary coil current at 4.6 A. However, IP 1
IP1 [A]
dP=0.2, dS=-0.2
50 dP=0.2, dS=0
40 dP=0.2, dS=0.2 increases to 37.2 A at 75.9 kHz under dP = 0.2, dS = −0.2.
30 More severe system detuning can be observed under smaller
20
load resistances, as shown in Fig. 7(b). When the equivalent
10
0 load resistance RE is 5.8 Ω under case-B, the output power
60 70 75 80 90 95 100 110 120 130
Frequency [kHz] at the nominal frequency point is 200 W, with the primary
(b) coil current IP 1 at 1.3 A. Due to the frequency variations
Fig. 7. Value of IP 1 under frequency variations and parameter deviations: and capacitance drift, IP 1 dramatically increases to 95.8 A
(a) RE = 24.1 Ω (case-B), where the output power at the nominal frequency at 75.7 kHz under dP = 0.2, dS = −0.2. It should be noted
point is 800 W, (b) RE = 5.8 Ω (case-B), where the output power at the that, in Fig. 7, the inverter output is set to the full duty cycle.
nominal frequency point is 200 W. Herein, the dc input voltage of the system
is 200 V, the nominal switching frequency is 85 kHz, and the inverter output Although decreasing the inverter duty cycle is able to reduce
is set to the full duty cycle. the coil currents, it results in a remarkable disparity in the
current amplitudes under different cases. This poses a great
Maximum value of IP1_Peak challenge to the measurement accuracy of the coil currents.
Maximum value of IP1_Peak
Furthermore, observing the current curve in Fig. 7 reveals
IP1_Peak [A]
IP1_Peak [A]
that the peak current point can be derived by sweeping the

frequency. Take case-B as an example, by sweeping the fre-
quency from 60 kHz to 130 kHz, the peak value of the primary
dP dP coil current IP 1 P eak under different values of RE and degree
dS dS
of capacitance drift is illustrated in Fig. 8. As shown in Fig.
(a) (b) 8, the maximum value of IP 1 P eak , i.e., M AX IP 1 P eak,
can always be found in the case of dP = 0.2, dS = −0.2,
where the degree of the primary and secondary capacitance
drift reaches the upper boundary and the lower boundary,
IP1_Peak [A]
IP1_Peak [A]
respectively. Moreover, by sweeping the parameters of the

equivalent load resistance RE and the coil coupling coefficient
k, the value of M AX IP 1 P eak under load variations and
dP dP
dS dS coupling changes is shown in Fig. 9. As it can be observed, the
(c) (d) value of M AX IP 1 P eak is inversely related to both RE
Fig. 8. Value of the peak primary coil current IP 1 P eak under different and k. In other words, when RE or k is small, more severe
values of RE and degree of capacitance drift in case-B: (a) RE = 5 Ω, (b) system detuning may occur due to the frequency variations
RE = 25 Ω, (c) RE = 45 Ω, (d) RE = 65 Ω. and parameter deviations. Specifically, the worst case shown
in Fig. 9 is the point A, where the value of M AX IP 1 P eak
and frequency variations is depicted in Fig. 7, where the reaches 101.8 A when RE = 5 Ω and k = 0.22, while the best
frequency-sweep band is set to 75∼95 kHz to ensure that case is the point B, where the value of M AX IP 1 P eak is
sufficient frequency points can be collected [26]. In Fig. 7, 11.6 A under RE = 81 Ω and k = 0.35.
the equivalent load resistance RE is set to different values The increased coil currents not only significantly increase
to investigate the impact of load variations, while case-B is the current stress of the resonant tank, but also potentially
selected as an example to demonstrate the influence of coil damage the semiconductor devices due to the rise of their
misalignment. Moreover, 9 different capacitance drift cases junction temperature.
are illustrated with dotted solid lines in different colors to Another impact of the system detuning is on the system
show the influence of capacitance drift, where the maximum output power Pout . As shown in Fig. 10, when the equivalent
degree of capacitance drift is set to ±20%. As illustrated in load resistance RE is 24.1 Ω under case-B, significant power
Fig. 7(a), when the equivalent load resistance RE is 24.1 Ω ripple is observed under frequency variations and parameter
6000 ON ON
Converntional dP=-0.2, dS=-0.2
5000
frequency-sweep band dP=-0.2, dS=0 S1 S3 M S5 S7 Iout
Peak power point dP=-0.2, dS=0.2
iP CP CS iS
(75.9, 5575) dP=0, dS=-0.2
4000 a + c
dP=0, dS=0
Nominal frequency dP=0, dS=0.2 Uin + +
vcd
Pout [W]
point (85, 800)

dP=0.2, dS=-0.2 vab LP LS Cfo Uout RL
3000 -
dP=0.2, dS=0
dP=0.2, dS=0.2 b - - d
RP RS
2000 S2 S4 S6 S8
iP iS
1000 LPF
OFF LPF
OFF
PWM Generator
0 iP1 iS1
60 70 75 80 90 95 100 110 120 130 RMS Value RMS Value
Frequency [kHz] Extraction Extraction
Variable
Frequency f IP1 IS1
Dynamic Frequency
IS1 Wireless
Fig. 10. System output power Pout under frequency variations and parameter Approaching Communication
deviations when the equivalent load resistance RE is 24.1 Ω in case-B. Herein, {IP1_i, Is1_i, ωi} Direction of data flow
the input conditions of the system are the same as those in Fig.7.
The JAYA Algorithm Recognized
{LP, LS, CP, CS, M}
deviations. In some cases, the output power of the system may Fig. 11. Block diagram of the proposed parameter recognition method
considerably exceed the load tolerance, eventually causing implemented at pre-start-up, where the rectifier output is short-circuited during
this process. Herein, the measured IS1 is transmitted from the secondary
irreparable damage to the load. controller to the primary controller for parameter recognition.
According to the above analysis, it can be concluded that
the traditional frequency-sweep based parameter recognition where ZR is the reflected impedance of the secondary resonant
methods may lead to severe system detuning under parameter circuit, which can be expressed as
deviations, resulting in considerable coil currents and signifi-
cant power ripple. Therefore, to ensure the safe operation of (ωM )2
ZR = . (7)
the system, it is essential to avoid wide frequency variations RS + j[ωLS − 1/(ωCS )]
near the nominal resonance point. The amplitude of the equivalent input impedance is denoted
by |Zin |. If the switching angular frequency ω is switched
to another value, a new value of |Zin | can be obtained.
B. Proposed Parameter Recognition Method Furthermore, by sweeping the switching frequency ω, multiple
This paper proposes a new parameter recognition method sets of {ωi , |Zin i |} can be acquired as
to recognize the unknown parameters of the capacitors and VP 1 i
|Zin i | = = f (RP , RS , LP , LS , CP , CS , M, ωi ), (8)
coils, i.e., LP , LS , CP , CS and M , as shown in Fig. 11. IP 1 i
First, the parameter recognition process is carried out at pre- where ωi (i = 1, 2, · · · , m) is the i-th selected frequency point;
start-up, and the rectifier output is short-circuited during this m is the number of the measured frequency points; VP 1 i and
process. As a result, significant load power ripple during the IP 1 i are the RMS values of vP 1 and iP 1 at the i-th frequency
conventional frequency-sweep process is avoided. It should be point; vP 1 is the fundamental component of the inverter output
noted that in [30] and [31], the rectifier output is also short- voltage vab . According to the analysis in [26], assuming RP
circuited to implement fast mutual inductance identification. and RS are given, {LP , LS , CP , CS , M } can be estimated by
However, in this paper, all the unknown paramters of the multiple sets of {ωi , |Zin i |}.
resonant circuits are considered, and the number of unknown However, it is difficult to identify {LS , CS , M } accurately
paramters are much larger than that in [30] and [31]. Then, the with only the front-end information, as there are countless
primary and secondary coil currents iP and iS are measured sets of {LS , CS , M } that lead to an almost consistent re-
and fed to two separate low pass filters (LPFs). With the flected impedance ZR . Suppose there is a new set of solution
LPFs, the fundamental components of iP and iS , i.e., iP 1 {L′S , CS′ , M ′ } satisfying L′S = λLS , CS′ = (1/λ)CS , M ′ =
and iS1 , are extracted. Furthermore, the RMS values of iP 1 √
λM , where {LS , CS , M } represents the correct solution and
and iS1 , i.e., IP 1 and IS1 , are obtained by the RMS value λ is any non-zero positive real number. Generally, when the
extraction modules. According to the values of IP 1 and IS1 , switching frequency ω is deviated from the secondary resonant
a dynamic frequency approaching strategy is proposed to √
frequency ωS (ωS = 1/ LS CS ), XS is much larger than
determine the selected frequency points. By recording the RS . Therefore, the reflected impedance ZR ′
of the new set of
values of IP 1 and IS1 at different frequency points, multiple solution satisfies
sets of {ωi , IP 1 i , IS1 i } are acquired. Finally, based on the
′ (ωM ′ )2 λ(ωM )2
acquired data, the unknown parameters are derived by the ZR ≈ = ≈ ZR .
JAYA algorithm. In the following section, the implementation j[ωL′S − 1/(ωCS′ )] jλ[ωLS − 1/(ωCS )]
(9)
of the proposed method will be described in detail.
Since λ can be any value, there are countless sets of solution
1) Recognition Model: When the rectifier output is short-
{L′S , CS′ , M ′ } such that the reflected impedance ZR ′
is almost
circuited, the equivalent input impedance Zin of the SS-
identical with the correct reflected impedance ZR . This means
compensated system is derived by
that it is difficult to accurately recognize {LS , CS , M } with
1 only the primary-side information. Therefore, in this paper, the
Zin = RP + j(ωLP − ) + ZR , (6) secondary coil current iS is also measured. By introducing the
ωCP
secondary-side information, the relationship between IS1 and Start
IP 1 is established as Increase the frequency from fL

s
IP 1 RS2 + [ωLS − 1/(ωCS )]2 ωLS − 1/(ωCS ) Measure the RMS value of the
= ≈ . coil currents IP1(i) and IS1(i)
IS1 (ωM )2 ωM
(10) IP1(i)>IM ? or No Update the frequency
Define the ratio of IP 1 to IS1 as Tii , i.e., Tii = IP 1 /IS1 . IS1(i)>IM ? f(i) using (15)
Then, the ratio Tii′ of the new set of solution {L′S , CS′ , M ′ } Yes
Decrease the frequency from fH
satisfies
ωL′S − 1/(ωCS′ ) λ[ωLS − 1/(ωCS )] √ Measure the RMS value of the
Tii′ ≈ ′
= √ ≈ λTii . coil currents IP1(i) and IS1(i)
ωM λωM
(11) IP1(i)>IM ? or No Update the frequency
The only solution of λ such that (9) and (11) hold simultane- IS1(i)>IM ? f(i) using (16)
Yes
ously is λ = 1. This means that by introducing the information
End
of iS , the correct solution of {LS , CS , M } is unique and all
the unknown parameters can be estimated accurately. Fig. 12. Flow chart of the proposed dynamic frequency approaching strategy.
To introduce the information of iS into the recognition
model, the equivalent gain from IS1 to VP 1 is given by
This means that the measurement process for the inverter
VP 1 (RP + jXP )(RS + jXS )
|ZP S | = = |jωM − |. (12) voltage can be avoided. Based on the recognition model,
IS1 jωM the unknown parameters {LP , LS , CP , CS , M } are searched
Similarly, by sweeping the switching frequency ω, multiple within the empirically selected lower and upper bounds.
sets of {ωi , |ZP S i |} can be obtained as 2) Dynamic Frequency Approaching: Based on the above
VP 1 i analysis, it is essential to measure iP and iS at multiple
|ZP S i| = = g(RP , RS , LP , LS , CP , CS , M, ωi ). frequency points for parameter recognition. However, accord-
IS1 i
(13) ing to the analysis in Section III-A, when the parameters of
Furthermore, with the information on both sides, the recogni- the capacitors and coils deviate from the nominal values, the
tion model for the unknown parameters is derived as frequency variations near the nominal resonance point may
lead to severe system detuning, resulting in enormous coil
min J = ||VP 1 − VP 1est || + ||VP 1 − V̂P 1est || (14) currents. To constrain the coil currents in a safe range, a
s.t. VP 1est = |Zin |IP 1 , V̂P 1est = |ZP S |IS1 , LP L ≤ LP dynamic frequency approaching strategy is proposed, as shown
≤ LP H , LSL ≤ LS ≤ LSH , CP L ≤ CP ≤ CP H , CSL ≤ CS in Fig. 12. In Fig. 12, fL , fH , IM , as well as the update
≤ CSH , and ML ≤ M ≤ MH , where equations (15) and (16), are pre-designed based on the system
parameters and stored in the microcontrollers in advance.
|Zin | = diag{|Zin i |}(i = 1, · · · , m), However, the values of IP 1 and IS1 are updated based on
the measured results in practical operation. The detailed steps
|ZP S | = diag{|ZP S i |}(i = 1, · · · , m),
of this strategy are as follows. First, the frequency is swept
IP 1 = [IP 1 1 , IP 1 2 , · · · , IP 1 m ], from the lower bound fL at various frequency points while
IP 1 and IS1 are measured. The frequency-sweeping process
IS1 = [IS1 1 , IS1 2 , · · · , IS1 m ], is stopped when IP 1 or IS1 exceeds the threshold value IM .
VP 1est = [VP 1est 1 , VP 1est 2 , · · · , VP 1est Then, the frequency is swept from the upper bound fH and is
m ],
terminated when IP 1 or IS1 is larger than IM again.
V̂P 1est = [V̂P 1est 1 , V̂P 1est 2 , · · · , V̂P 1est m ], In this paper, the current threshold IM is set as 10 A based
VP 1 = [VP 1 1 , VP 1 2 , · · · , VP 1 on the current capability of the system, while fL and fH are
m ].
set to 65 kHz and 125 kHz. The basic principle of selecting
Herein, IP 1 is the measured RMS values of iP 1 at multiple fL and fH is to ensure that IP 1 and IS1 at both fL and fH are
frequency points; VP 1est is the estimated RMS values of vP 1 lower than 10 A under any case. Take case-B as an example,
derived by |Zin |IP 1 ; IS1 is the measured RMS values of the values of IP 1 and IS1 when the rectifier output is short-
iS1 ; V̂P 1est is the estimated RMS values of vP 1 derived circuited is shown in Fig. 13. As illustrated in Fig. 13(a), fL
by |ZP S |IS1 ; VP 1 is the measured RMS values of vP 1 ; is required to be lower than fA1 to ensure that the value of
||VP 1 − VP 1est || represents the norm of the voltage differ- IP 1 is smaller than 10 A at the frequency point fL under any
ences between VP 1 and VP 1est ; ||VP 1 − V̂P 1est || is the case, while fH should be higher than fB1 to guarantee IP 1
norm of the voltage differences between VP 1 and V̂P 1est . is always smaller than 10 A at fH . Similarly, as shown in
It should be noted that when the dc input voltage Uin and the Fig. 13(b), fL is also required to be less than fA2 , and fH is
duty cycle of the inverter ac voltage vab remain unchanged, needed to be greater than fB2 to ensure that the value of IS1
the RMS value of vP 1 does not vary with the switching does not exceed the threshold. Furthermore, define fA min as
frequency. In our paper, the dc input voltage is configured at the minimum value of fA1 and fA2 , fB max as the maximum
200 V, and the inverter output is fixed at the full duty cycle. value of fB1 and fB2 , the values of fA min and fB max under
15 dP=-0.2, dS=-0.2 dP=0, dS=-0.2 dP=0.2, dS=-0.2 Start

D1(70,15.7)
dP=-0.2, dS=0 dP=0, dS=0 dP=0.2, dS=0
dP=-0.2, dS=0.2 dP=0, dS=0.2 dP=0.2, dS=0.2 Initialize the population
A1(fA1,10) B1(fB1,10) Calculate the fitness value of

10
each individual using (14)
IP1 [A]
C1(69,9.2) IM=10A
Identify the best and worst
individuals xbest and xworst
5
Update each individual using (17)
Yes Is the new individual No

0
60 fL 70 80 90 100 110 120 fH 130 better than the old one?
Frequency [kHz]
New individual
Keep the old individual
replaces the old one
(a)
15 dP=-0.2, dS=-0.2 dP=0, dS=-0.2 dP=0.2, dS=-0.2
dP=-0.2, dS=0 dP=0, dS=0 dP=0.2, dS=0 No Is the number of iterations
dP=-0.2, dS=0.2 dP=0, dS=0.2 dP=0.2, dS=0.2
reaches Genmax?
B2(fB2,10) Yes
10
A2(fA2,10) Output the optimal individual
IS1 [A]
IM=10A and its fitness value
End
5
Fig. 15. Flowchart of the JAYA algorithm.
0
60 fL 70 80 90 100
Frequency [kHz]
110 120 fH 130 some cases, and the measurement errors should be considered
in these cases. Therefore, if the frequency interval is small
(b)
under the low-current frequency points, the measurement
Fig. 13. Values of IP 1 and IS1 when the rectifier output is short-circuited
under case-B: (a) IP 1 , (b)IS1 .
errors might be significant in some cases. Besides, a small
frequency interval will also make the frequency-sweep process
130 time-consuming. Considering the abovementioned reasons, a
fB_max B(0.35, 121.7) dynamic frequency interval is adopted in this paper. When the
120 fA_min
frequency is increased from fL , the update equation is
110
Frequency [kHz]
100
f (i + 1) = f (i) + τ × ceil[IM − max(IP 1 (i), IS1 (i))], (15)
case-A
90 where ceil is the round-up function; τ is a fixed factor,
80 which is set to 0.2 to ensure that the coil current does not
70
increase dramatically as it approaches to the threshold IM .
A(0.35, 65.1)
Similarly, when the frequency is decreased from fH , the
60
0.2 0.25 0.3 0.35 0.4 updated equation is
Coupling coefficient k
f (i + 1) = f (i) − τ × ceil[IM − max(IP 1 (i), IS1 (i))]. (16)
Fig. 14. Values of fA min and fB max under different values of k when
the rectifier output is short-circuited. As shown in (15) and (16), the frequency interval is dynam-
ically adjusted between 0.2 kHz and 2 kHz according to the
RMS values of the coil currents. Based on the above analysis,
different values of k when the rectifier output is short-circuited by using the proposed dynamic frequency approaching strat-
is illustrated in Fig. 14. To guarantee the coil currents at fL egy, the coil currents are constrained safely, and the required
and fH are lower than 10 A under any case, as shown in Fig. data for parameter recognition is acquired efficiently.
14, fL is required to be lower than 65.1 kHz, and fH should 3) JAYA Algorithm: After the required data is obtained,
be larger than 121.7 kHz. Therefore, fL and fH are set to the optimization problem described in (14) is needed to be
65 kHz and 125 kHz, respectively. solved to recognize the unknown parameters. In this paper, a
Moreover, the interval of the frequency points is not fixed heuristic algorithm called JAYA is introduced to derive the
but dynamically adjusted according to the measured current unknown parameters. Generally, there are two mainstream
values. The motivations of the dynamic frequency interval are approaches for solving unknown parameters in the WPT sys-
as follows. On one hand, when the frequency approaches to tems: the least square approximation (LSA) and the heuristic
the peak current point, the coil currents rise dramatically. Take algorithms. The LSA was implemented in [32] to recognize
dP = dS = 0.2 under case-B as an example, as shown in Fig. the mutual inductance and load resistance, while it was used
13(a), IP 1 is 9.2 A when the frequency is 69 kHz, whereas for identifying multiple loads in [27] and the secondary-side
it rapidly increases to 15.7 A at 70 kHz. As a result, if the reactance in [22]. The heuristic algorithms, which can find
frequency interval is large under the high-current frequency optimal solutions more efficiently than the traditional LSA
points, the coil currents may greatly exceed the threshold, for the multi-variable systems [26], are also usually used,
endangering the safe operation of the system. On the other such as the genetic algorithm (GA) in [33], and the adaptive
hand, when the frequency is close to fL and fH , as illustrated differential evolution (ADE) algorithm in [26]. However, for
in Fig. 13(a), the coil currents might be quite small under conventional GA algorithm and differential evolution (DE)
algorithm, various algorithm-specific parameters, including the S1 S3 M S5 S7 Iout

iP CP CS iS
crossover and mutation rates, need to be tuned. Although the a + c
ADE algorithm avoids the difficulty of tuning parameters, the Uin + +
vab LP LS vcd Cfo Uout RL
-
adaptive cross-over and mutation rates need to be calculated b - - d
RP RS
in each iteration, which increases the complexity of the al- S2 S4 S6 S8
gorithm. Compared with the GA and ADE algorithms, the
PWM Generator Phase Syn &
JAYA algorithm is free from algorithm-specific parameters ω
Pulese Gen
Direction of data flow β,φ
(except for two common parameters, i.e., the population size Equation (18)
Wireless Equation (21)
Load
LP,CP RL Detection
Psize and the maximum number of generations Genmax ), and Recognized
Communication
RE,XE
Equation (19) Pref
only one update equation is required in each generation [28]. {LP,CP,LS,CS,M} LS,CS,M,
ω,VP
LS,CS,M,
ω,VP
The flowchart of the JAYA algorithm is depicted in Fig. 15,
and the process is detailed as follows: [Initialization] First, Fig. 16. Block diagram of the impedance tuning method. Herein, the
a random population x with Psize individuals is generated parameters {LS , CS , M , ω, VP } are transmitted from the primary controller
to the secondary controller for calculating the required phase shift angles of
in the search space (within the lower and upper bounds), the active rectifier.
where xp,q represents the q-th parameter of the p-th individual
(p = 1, 2, · · · , Psize ; q = 1, 2, · · · , N ), and N is the number Herein, it should be noted that the secondary reactance can
of the unknown parameters. Specifically, in our paper, x rep- be slightly overcompensated to realize zero-voltage-switching
resents the population with Psize (Psize =50) individuals, and (ZVS) for the inverter [22]. According to [18], the equivalent
each individual consists of N (N =7) unknown parameters. The load impedance provided by the active rectifier is
unknown parameters are RP , RS , LP , LS , CP , CS , and M ,
 RE = 42 RL cos2 (φ)(1 − cos(β))

respectively. [Fitness] Then, the fitness value of each individ- π (20)
ual is calculated based on the cost function described in (14).  XE = 4 RL sin(φ)cos(φ)(1 − cos(β))
2
[Update] According to the fitness value of each individual, π
the best individual xbest = [xbest,1 , xbest,2 , · · · , xbest,N ] with where β is the duty-cycle angle of the rectifier input voltage
minimum fitness value and the worst individual xworst = vcd , and φ is the phase shift angle between V̇S and I˙S .
[xworst,1 , xworst,2 , · · · , xworst,N ] with maximum fitness value Substituting (20) into (21), β and φ are derived by
are selected, and the population is then updated by
 φ = arctan( XE )

RE
x′p,q = xp,q +r1 (xbest,q −|xp,q |)−r2 (xworst,q −|xp,q |), (17) π 2 XE ) (21)
 β = arccos(1 −
where xbest,q and xworst,q are the q-th parameter of the best 2RL sin(2φ)
and worst individuals, respectively; r1 and r2 are two random Considering the range of β ([0,π]) and φ ([−π/2,π/2]), the
numbers within [0,1]. In (17), the term r1 (xbest,q − |xp,q |) re- maximum RE and XE that can be provided by the active
veals the tendency of approaching the best individual, whereas rectifier are 8RL /π 2 and ±4RL /π 2 , respectively.
the term −r2 (xworst,q − |xp,q |) indicates the tendency of
escaping from the worst individual. [Selection] The updated IV. E XPERIMENTAL V ERIFICATION
individual x′p,q is accepted if it gives a better fitness value. All
A. Experimental Setup
the accepted individuals at the end of iteration are retained and
become the input of the next generation. [Termination] The To verify the feasibility of the proposed parameter recogni-
optimization process is stopped when the termination criteria tion based impedance tuning method, experiments are carried
is satisfied. Here, the termination criteria is the number of out on a SS-compensated WPT prototype using the charging
iterations reaches Genmax . pads in Fig. 3, as shown in Fig. 17. A dc power supply
is utilized to provide the required power, and two rheostats
C. Impedance Tuning are connected in series as the load resistor. The impedance
Based on the recognized parameters, an impedance tuning tuning algorithm and PWM generation are implemented in the
method is adopted, as shown in Fig. 16. To minimize the LaunchPad F28379D. Two separate H-bridge converters are
primary reactance XP , the switching frequency ω of the adopted as the inverter and the active rectifier, respectively.
inverter is adjusted to The experimental data and waveforms are recorded by an
oscilloscope (YOKOGAWA DLM2054) and the dc to dc
1
ω=√ . (18) efficiency of the system is measured by a power analyzer
LP C P (YOKOGAWA WT500). More details of the prototype are
Additionally, the duty cycle of the inverter ac voltage vab is listed in Table II.
maintained at the full duty cycle. Then, the equivalent load
impedance ZE is regulated by the active rectifier to minimize B. Parameter Recognition
the secondary reactance XS and to obtain the required output
power Pref , i.e., The accuracy of the proposed parameter recognition method
 is verified in 10 cases of the parameter deviations, as listed
2 2
 RE = ω M Pref
 in Table II. The parameters of the coil inductances and
VP2 (19) compensation capacitances in Table III are measured by an
 XE = −XS = −(ωLS − 1 )

impedance analyzer (Agilent 4294A).
ωC S
Oscilloscope Launchpad F28379D 0

-0.063dB -0.208dB
-5
Magnitude (dB)
-7.1dB
-10 Fundamental frequency band:
65 kHz~86 kHz
Power analyzer
-15
Inverter 3-rd harmonic frequency band:
195 kHz~258 kHz
DC power supply Rectifier -20
CS
-23.3dB
CP -25
Charging pads
-30
Load resistor 101 10 2
Frequency (kHz)
Fig. 19. Magnitude-frequency response of the designed digital LPF. Herein,

the maximum frequency point that can be reached when the frequency is
increased from fL is calculated as 86 kHz based on the parameter deviation
Fig. 17. Experimental setup of the SS-compensated WPT system. range listed in Table II.
TABLE II TABLE III

C ASE S TUDY FOR PARAMETER R ECOGNITION PARAMETERS OF THE JAYA A LGORITHM
No. LP [µH] LS [µH] M [µH] CP [nF] CS [nF] Symbol Value Symbol Value
A1 335.5 222.7 95 9.9 17.32 LP L 300 µH LP H 350 µH
A2 335.5 222.7 95 11.53 16.5 LSL 200 µH LSH 250 µH
A3 335.5 222.7 95 11.53 14.88 ML 50 µH MH 120 µH
A4 335.5 222.7 95 9.9 13.21 CP L 5 nF CP H 15 nF
A5 335.5 222.7 95 9.07 14.88 CSL 10 nF CSH 20 nF
B1 327.5 216.5 58 9.9 17.32 RP L 0.5 Ω RP H 0.9 Ω
B2 327.5 216.5 58 11.53 16.5 RSL 0.3 Ω RSH 0.7 Ω
B3 327.5 216.5 58 11.53 14.88 Psize 50 Genmax 5000
B4 327.5 216.5 58 9.9 13.21
B5 327.5 216.5 58 9.07 14.88
inevitable due to the capacitive input impedance Zin . However,
since the data acquisition stage only lasts for hundreds of
milliseconds (see section IV-D), the extra switching loss
v v ab[200V/div] v ab v ab [200V/div]
ab
caused by hard switching during this process can be ignored,
and the voltage spikes at the switching transient can also be
iP i [5A/div]
P
iP i [10A/div]
P
suppressed by optimizing the resistor-capacitor (RC) snubber
circuits in practice [34], [35]. Another noteworthy point is that
i
S i [5A/div]
S
iS iS[10A/div] the extracted coil currents, as shown in Fig. 18(a), contain
Time [10μs/div] Time [10μs/div] a small amount of 3-rd harmonics when the frequency is
(a) (b) increased from fL (65 kHz). Therefore, a digital second-order
LPF is implemented to filter out the harmonic components of
the coil currents when the frequency is increased from fL ,
v ab iP vab [200V/div] v iP v [200V/div]
ab ab
and the magnitude-frequency response of the designed LPF
is illustrated in Fig. 19. It should be noted that when the
i [10A/div]
P i [5A/div]
P
frequency is decreased from fH (125 kHz), the digital LPF is
not required due to the 3-rd harmonics are well-attenuated by
iS[10A/div] iS i [5A/div]
S
the high-impedance characteristics of the resonant circuits in
iS
Time [10μs/div] Time [10μs/div] the high frequency band, as it can be observed in Fig.18(c) and
(c) (d) 18(d). After the harmonics are filtered by the LPF, the RMS
Fig. 18. Measured operating waveforms of case-A1 under different frequency
values of the fundamental components of the coil currents,
points: (a) 65 kHz, (b) 70.4 kHz, (c) 107.7 kHz, (d) 125 kHz. i.e., IP 1 and IS1 , are then extracted by the rms function of
MATLAB. The extracted RMS values of IP 1 and IS1 under
To acquire the required data for parameter recognition, the case-A1 are illustrated in Fig. 20. As it can be observed, by
rectifier output is short-circuited and the system frequency implementing the proposed dynamic frequency approaching
f is regulated based on the dynamic frequency approaching method, IP 1 and IS1 are accurately obtained under sufficient
strategy shown in Fig. 12. By dynamically adjusting the system frequency points.
frequency f , the primary and secondary coil currents iP and Furthermore, based on the values of IP 1 and IS1 , the JAYA
iS are measured at multiple frequency points. The measured algorithm is implemented to recognize the unknown parame-
data are recorded by the oscilloscope and then extracted ters of the resonant circuits. The searching constraints of the
to MATLAB for processing. Fig. 18 shows the recorded unknown parameters {LP , LS , M , CP , CS } for the JAYA
waveforms of case-A1 under different frequency points. It algorithm are listed in Table IV. Since the loss resistances
should be noted that when the frequency is increased from RP and RS may vary with the system operating conditions,
fL (65 kHz), hard switching of the inverter power switches is RP and RS are considered as the unknown parameters as
15 15 15
Calculated Calculated LP
Experimental Experimental CP
Average Relative Error[%]

LS
10 10
CS
10
IS1 [A]
IP1 [A]
IM=10A IM=10A
M
5 5
5
0 0
60 70 80 90 100 110 120 130 60 70 80 90 100 110 120 130
Frequency [kHz] Frequency [kHz]
(a) (b) 0
8 10 12 14 16 18 20 22
Fig. 20. Extracted RMS values of IP 1 and IS1 in case-A1: (a) IP 1 , (b)IS1 . Number of extracted frequency points
350 5 Fig. 22. Average relative errors (AREs) of case-A1 under different numbers
10 5
of frequency points.
Relative Error[%]
345 4
9.95
LP[µH]
340 3 4
Relative Error[%]
9.9 45 40
335 2
3
Convergent Fitness value J

CP[nF]
Convergent Fitness value J

330 1 9.85 40
35
325 0 2 35
9.8
230 5 30
30
1
Relative Error[%]
4 9.75
225 25 25
LS[µH]
3 9.7 0
220 20
2 18.5 5 20
215
15
1
18 4 10 15
210 0 Relative Error[%]
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
96 5 No. of Recgnition No. of Recgnition
17.5 3
CS[nF]
Relative Error[%]
95.5 4
17 2 (a) (b)
M[µH]
3
95
2 16.5 1 Fig. 23. Convergent fitness values for the studied 10 cases: (a) case-A, (b)
94.5 1 case-B.
94 0 16 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
No. of Recgnition No. of Recgnition 3
(a) (b) 2.5
Fig. 21. Recognized results for case-A1: (a) coil inductances LP , LS and 2
ARE [%]
M , (b) compensation capacitances CP and CS .

1.5
well. Additionally, the population size Psize and the maximum 1
generation numbers Genmax of the JAYA algorithm are also 0.5

shown in Table IV. The parameters are independently recog-
0
nized by the JAYA algorithm for 10 times. The JAYA algorithm A1 A2 A3 A4 A5 B1 B2 B3 B4 B5
Case No.
is implemented in MATLAB on a computer with a Intel(R)
Core(TM) i7-1185G7 CPU, and the average computation time Fig. 24. Average relative errors (AREs) for the studied 10 cases.
for each recognition is 1-3 s. Fig. 21 presents the recognized
parameters for case-A1 in each recognition. As shown in
fitness values for all of the studied 10 cases are demonstrated
Fig. 21, the unknown parameters of case-A1 are recognized
in Fig. 23. The convergent fitness values are steady for all
accurately, with the relative errors in each recognition all less
cases in each recognition, which validates that the JAYA
than 2%. Besides, as it can be observed, the recognized results
algorithm can always find the global optimal solution. The
of each independent recognition are basically consistent, which
AREs for the studied 10 cases are then presented in Fig.
validates the stability of the JAYA algorithm. Fig. 22 shows
24. As shown in Fig. 24, the unknown parameters of the
the average relative errors (AREs) of case-A1 under different
resonant circuits are recognized accurately, with the AREs of
numbers of frequency points. As shown in Fig. 22, the AREs
all cases less than 3%. Moreover, the standard deviations of the
are significant when the number of extracted frequency points
recognized results under each case are shown in Fig. 25. The
is small. This is because when the amount of extracted data
standard deviations of the unknown parameters LP , LS . CP ,
is insufficient, it becomes difficult to converge to the optimal
CS and M are less than 0.5 µH, 0.6 µH, 0.012 nF, 0.035 nF
solution in the iterative process due to inevitable measurement
and 0.2 µH, respectively, with the relative standard deviations
noise, and the results of each recognition is not stable. This
(RSD) all less than 0.3%.
phenomenon can also be observed in other similar works [22],
[27]. However, the AREs converge to a stable level as the num-
ber of extracted frequency points increases. It should be noted C. Impedance Tuning
that the number of frequency points can be tuned by tuning After the unknown parameters are accurately recognized,
the factor τ in (15) and (16) to ensure sufficient frequency the impedance tuning method is implemented to allow the
points can be obtained for accurate parameter recognition. system operate close to the resonance state. To demonstrate
Moreover, all of the 10 cases listed in Table III are studied. the effectiveness of the impedance tuning method, experiments
To verify the stability of the JAYA algorithm, the convergent are carried out in two cases of parameter deviations, i.e.,
0.012 0.3 0.5 0.3
0.01 fs=85kHz
RSD[%]
0.2
LP[µH]
0.008 0.2
0.25 v
ab
v
ab
fs=92kHz
RSD[%]
0.1
CP[nF]
0.006
0.004 0.1
0 0 i P
0.002
0.6 0.3
i P
RSD[%]
0.4 0.2
LS[µH]
0 0
0.035 0.3
0.2 0.1 φ=-33.2°
0.03 i S
i S
0.025
0.2
0 0
v cd v
RSD[%]
0.2 0.3 cd
CS[nF]
0.02
Time [5μs/div] v ab [200V/div] i [5A/div]
P Time [5μs/div] v ab [200V/div] i [10A/div]
P
RSD[%]
0.015 0.2
M[µH]
0.01
0.1 0.1 v cd [200V/div] i [20A/div]
S v cd [200V/div] i [10A/div]
S
0.1
0.005
0 0 0 (a) (b)
A1 A2 A3 A4 A5 B1 B2 B3 B4 B5 A1 A2 A3 A4 A5 B1 B2 B3 B4 B5
Case No. Case No.
(a) (b)
Fig. 25. Standard deviations for the studied 10 cases: (a) coil inductances
LP , LS and M , (b) compensation capacitances CP and CS .
vab fs=85kHz v ab fs=91kHz
i
P
(c) (d)
iP
Fig. 27. Measured operating waveforms and dc to dc efficiency of case-B5

i S
i S
φ=-22.9° when delivering 700 W power: (a) operating waveforms before tuning, (b)
v cd v cd operating waveforms after tuning, (c) measured dc to dc efficiency before
Time [5μs/div] v ab[200V/div] i [10A/div]
P Time [5μs/div] v
ab [200V/div] i [10A/div]
P
tuning, (d) measured dc to dc efficiency after tuning.
vcd [200V/div] i [10A/div]
S
v
cd [200V/div] i [10A/div]
S
100 100
(a) (b)
95
95
Efficiency [%]
Efficiency [%]
90
90 85
80
85
Before tuning 75 Before tuning
After tuning After tuning
80 70
200 300 400 500 600 700 800 200 300 400 500 600 700 800
Output Power [W] Output Power [W]
(a) (b)
Fig. 28. Measured dc to dc efficiency of case-A5 and case-B5 under different
(c) (d)
power points: (a) case-A5, (b) case-B5.
Fig. 26. Measured operating waveforms and dc to dc efficiency of case-A5
when delivering 600 W power: (a) operating waveforms before tuning, (b)
operating waveforms after tuning, (c) measured dc to dc efficiency before iP are almost in phase (iP slightly lags behind vab to realize
tuning, (d) measured dc to dc efficiency after tuning. ZVS for the inverter). To illustrate the effectiveness of the
impedance tuning, the measured dc to dc efficiency of case-A5
case-A5 and case-B5. For case-A5 and case-B5, the primary before and after tuning is further presented in Fig. 26(c) and
and secondary compensation capacitances are 9.07 nF and 26(d). As it can be observed, when delivering 600 W in case-
14.88 nF, with the degree of capacitance drift at -13.21% and A5, the dc to dc efficiency of the WPT system is increased
-5.46%, respectively. Besides, compared with case-A5, the air from 88.3% to 95.5%. The measured operating waveforms and
gap of case-B5 is increased from 10 cm to 15 cm, and thus, efficiency of case-B5 when delivering 700 W power are then
the coil coupling coefficient k is decreased from 0.35 to 0.22. demonstrated in Fig. 27. Based on the identified parameters
Due to the abovementioned parameter deviations caused by and carrying out the impedance tuning, as shown in Fig. 27(a)
coil misalignment and capacitance drift, both the primary and and 27(b), the system frequency is increased from 85 kHz to
secondary resonant circuits are detuned. 92 kHz, and the phase shift angle φ is adjusted to −33.2◦ .
Fig. 26 presents the measured operating waveforms of case- The reactance on both sides are minimized simultaneously, and
A5 when delivering 600 W power. Before the impedance therefore, the dc to dc efficiency of the system is improved
tuning, as shown in Fig. 26(a), the system frequency is 85 kHz, from 86% to 91.9%.
and the secondary ac voltage vcd and current iS are in phase. Furthermore, the measured dc to dc efficiency of case-A5
Moreover, due to the system detuning, the primary current and case-B5 under different output power points is shown in
iP leads the primary ac voltage vab . By implementing the Fig. 28. By implementing the impedance tuning method, the
impedance tuning method based on the recognized parameters, dc to dc efficiency is significantly improved. As shown in Fig.
the operating frequency is increased to 91 kHz to minimize 28(a), the efficiency is improved by 4.3%-7.7% after achieving
the primary-side reactance XP , while the phase shift angle impedance tuning in case-A5. More significant efficiency
between vcd and iS , i.e., φ, is regulated to −22.9◦ to minimize optimization can be observed under case-B5, as illustrated
the secondary-side reactance XS . After the impedance tuning, in Fig. 28(b), the efficiency improvement after tuning ranges
as illustrated in Fig. 26(b), the primary voltage vab and current from 5.9% to 15%.
Primary Measurement Board Secondary Measurement Board Wireless Communication

800 Modules
(20,730) Power is limited
600
Output Power [W]

Power is limited (80,618)
400
200
Case-A5
Case-B5
0 Primary Signal Processing Board Secondary Signal Processing Board
0 20 40 60 80
RE [Ω]
Fig. 31. Hardware modules for the close-loop experiments.

Fig. 29. Illustration of the limited output power due to the parameter
deviations under case-A5 and case-B5.
S1 S2 S3 S4
v ab [500V/div]
v ab[500V/div]
v ab fs=91kHz v ab fs=92kHz
i [20A/div]
P
i i [20A/div]
i
P P
v cd [500V/div]
P
Increase f from fL Decrease f from fH
φ=-14.8° φ=-31.6° v cd [500V/div]
i [20A/div]
S
iS i S
i [20A/div]
S
v
cd v cd
Time [500ms/div] Time [50ms/div]
Time [5μs/div] v
ab [200V/div] i [10A/div]
P Time [5μs/div] vab [200V/div] i [10A/div]
P (a) (b)
v
cd [200V/div] i [10A/div]
S vcd [200V/div] i [10A/div]
S
Fig. 32. Experimental results of the close-loop experiments under case-A5:
(a) (b) (a) complete implementation of the proposed method, (b) enlarged view of
the data acquisition stage.
3
Using the oscilloscope
2.5
Close-loop experiments
ARE [%] 2
1.5
0.5
(c) (d)
0
Fig. 30. Measured operating waveforms and dc to dc efficiency of case- LP CP LS CS M
Recognized Parameters
A5 and case-B5 when delivering 800 W power: (a) operating waveforms
after tuning in case-A5, (b) operating waveforms after tuning in case-B5, (c)
measured dc to dc efficiency after tuning in case-A5, (d) measured dc to dc Fig. 33. Comparisons on the recognized results in the close-loop experiments
efficiency after tuning in case-B5. and those using the oscilloscope.
It should be noted that due to the parameter deviations processing boards are installed on top of the LaunchPads.
in case-A5 and case-B5, the output power of the system Additionally, the NRF24L01+ modules shown in Fig. 31 are
is limited, as illustrated in Fig. 29. Therefore, Fig. 28 only employed for the dual-side wireless communication. Accord-
presents the measured efficiency in the power range of 200- ing to [36], the adopted module takes 339 µs for a single-
600 W before tuning in case-A5, and the measured efficiency byte payload to complete a transmission and acknowledgment
in the power range of 200-700 W before tuning in case-B5. between the transmitter and the receiver. It is worth noting that
Nevertheless, by implementing the impedance tuning method, the proposed method only requires the measurement of steady-
the system can reach the rated power (800 W) under both case- state RMS values of the coil currents, and the communication
A5 and case-B5, as shown in Fig. 30. As it can be observed, latency does not affect the accurate measurement of these
by regulating the system frequency and the phase shift angles values. Based on the above hardware modules, the close-
of the active rectifier according to the recognized parameters, loop experiments are carried out under case-A5, where the
the system is able to reach 800 W output power, with the dc to experimental results are demonstrated in Fig. 32.
dc efficiency at 96% under case-A5 and 92% under case-B5, As shown in Fig.32(a), the complete implementation of the
respectively. proposed method includes 4 different stages, where S1, S2,
S3, S4 represent the pre-start-up stage, the data acquisition
D. Close-loop Verification stage, the JAYA algorithm execution stage and the normal
To further verify the feasibility of the proposed method charging stage, respectively. After the system starts, it firstly
in practice, the close-loop recognition and tuning process is enters the data acquisition stage, where the rectifier output is
implemented in the TMS320F28379D microcontrollers. Two short-circuited and the inverter frequency is adjusted based on
measurement boards and two signal processing boards, as the proposed dynamic frequency approaching strategy. During
shown in Fig. 31, are utilized to measure the primary and this stage, as shown in Fig.32(b), the inverter frequency f
secondary coil currents, respectively, where the measurement is firstly increased from fL , and then it is decreased from
boards are connected to the resonant circuits and the signal fH after the coil currents exceed the threshold IM . The data
TABLE IV
C OMPARISONS WITH OTHER WORKS ON IMPEDANCE TUNING
Wireless Rated
Reference Compensation Considered parameters Parameter recognition Impedance tuning
communication Power
[9] LCC-S LP Phase detection SCC No 66.8 W
Frequency tuning &
[20] LCC-S M, LP , LS Phase detection No 3.3 kW
semi-active rectifier
Phase detection &
[24] LCC-LCC M, LP , LS Two SCCs No 3 kW
the gradient descent algorithm
Phase detection & Frequency tuning &
[21] SS CP , CS Yes 50 W
auxiliary coils semi-active rectifier
Pulse density modulation & Frequency tuning &
[22] SS LS , CS No 145 W
the LSA algorithm active rectifier
Dynamic frequency approaching & Frequency tuning &
This article SS M, LP , LS , CP , CS Yes 800 W
the JAYA algorithm active rectifier
acquisition stage stops when the coil currents exceed the TABLE V
threshold again. In the experiments, 20 different frequency C OMPARISONS WITH OTHER WORKS ON PARAMETER RECOGNITION
points are extracted with the time for each frequency point
Recognized Data Acquisition Max.
configured at 20 ms, and therefore, the total time for the whole Ref.
parameters & Algorithm Error
Time
data acquisition stage is 400 ms. After the data acquisition, Multiple Traditional frequency
[27] / 1 /
the JAYA algorithm is implemented to solve the unknown loads sweeping & LSA
parameters, which takes around 3.2 s in the experiments. In the Traditional frequency
[32] M, RE / /
close-loop tests, the maximum average relative error (ARE) of sweeping & LSA
the recognized results for case-A5 is 2.6%, which is slightly M, RE , Traditional frequency
[26] <3% 17.95 s
CS , LS sweeping & ADE
larger than that of using the oscilloscope. The recognition
Pulse density
results in the closed-loop experiments are compared with [22] CS , LS
modulation & LSA
<5% 2∼5s
those using the oscilloscope in Fig. 33. The discrepancy in
Harmonic detec-
estimated results is attributed to the difference in measurement [37] M, RE 6.9 % 7 ms
tion & calculation
errors. In the close loop experiments, the current transducer This M, LP , LS , Dynamic frequency
<3% ≈ 3.6 s
(LAH 50-P) is adopted to measure the coil currents, with the article CP , C S approaching & JAYA
frequency bandwidth at 200 kHz, and the sampling rate of 1 The symbol “ / ” indicates the relevant details are not provided in
the DSP is configured at 1 MHz. Compared with the 50 MHz the reference.
bandwidth of the current probe (KEYSIGHT N2782B) and the
12.5 GHz sampling rate of the oscilloscope, the measurement caused by capacitance drift and coil misalignment on both
errors in the close-loop experiments are increased due to sides. In this paper, the primary-side reactance is tuned by
the limited bandwidth of the adopted measurement system. regulating the inverter frequency, while the secondary-side
Consequently, the recognition errors in the close-loop exper- reactance is tuned by the active rectifier. Compared with those
iments are slightly larger than that of using the oscilloscope. methods using SCCs, the proposed method avoids the extra
In practice, the current sensor with high bandwidth and the power losses caused by SCCs. Since all parameters of the
analog-to-digital converter (ADC) with high sampling rate resonant circuits are considered, the proposed method requires
can be employed to further minimize the measurement errors, the secondary-side information to accurately identify LS , CS ,
thereby achieving higher recognition accuracy. Finally, based and M . Therefore, the wireless communication is needed in
on the recognized parameters, the system frequency as well as this paper.
the phase shift angles of the rectifier are regulated to realize the Moreover, the proposed method is compared with other
impedance tuning, and the system enters the normal charging existing parameter monitoring methods in Table V. For multi-
stage. parameter estimation, it is of great importance to measure
the coil currents at various frequency points. The existing
E. Comparison with Other Works methods mainly acquire this information from two distinct per-
To illustrate the difference between the proposed method spectives: frequency sweeping [26], [27], [32] and harmonic
and other existing impedance tuning methods, detailed com- detection [22], [35]. However, for the harmonic-detection-
parisons are summarized in Table IV. based methods, the number of parameters that these methods
Compared with other reported methods, the main con- can accurately estimate is restricted by the limited harmonic
tribution of the proposed method is all parameters of the frequency points. On the other hand, the traditional frequency-
resonant circuits are recognized simultaneously, without using sweeping-based methods are susceptible to severe system
extra phase detection circuits and auxiliary coils. Based on detuning and load power ripple under wide-range parameter
these identified parameters, the proposed method is able to deviations. To this end, a dynamic frequency approaching
simultaneously deal with a wide range of parameter deviations strategy is proposed in this paper, with the rectifier output
short-circuited during the parameter recognition process. This proposed parameter recognition strategy also shows promise
approach avoids significant load power ripple, and is able in optimal power flow control and foreign object detection.
to efficiently and safely measure the coil currents under a These potential applications will also be considered as part of
sufficient number of frequency points. As shown in Table V, our future work.
the maximum estimation error of the proposed method is com-
parable to other existing works. Additionally, the estimation
time of the proposed method is around 3.6 s, which is also R EFERENCES
fast enough for the stationary charging applications.
[1] S. R. Hui and W. W. Ho, “A new generation of universal contactless
battery charging platform for portable consumer electronic equipment,”
V. C ONCLUSIONS AND F UTURE W ORK IEEE Transactions on Power Electronics, vol. 20, no. 3, pp. 620–627,
2005.
In this paper, a parameter recognition based impedance tun- [2] C. R. Teeneti, T. T. Truscott, D. N. Beal, and Z. Pantic, “Review of
ing method is proposed for the SS-compensated WPT systems. wireless charging systems for autonomous underwater vehicles,” IEEE
Journal of Oceanic Engineering, vol. 46, no. 1, pp. 68–87, 2019.
The proposed parameter recognition method is implemented [3] D. Ahn and S. Hong, “Wireless power transmission with self-regulated
at pre-start-up, with the rectifier output short-circuited during output voltage for biomedical implant,” IEEE Transactions on Industrial
this process. Therefore, the proposed method is more suitable Electronics, vol. 61, no. 5, pp. 2225–2235, 2013.
for stationary applications, whereas it is not desirable to be [4] A. Ahmad, M. S. Alam, and R. Chabaan, “A comprehensive review of
wireless charging technologies for electric vehicles,” IEEE transactions
used in the dynamic wireless charging cases. Notably, as the on transportation electrification, vol. 4, no. 1, pp. 38–63, 2017.
proposed method is load-independent and the total time for the [5] P. Machura and Q. Li, “A critical review on wireless charging for electric
entire process is only several seconds, it can also be repeated vehicles,” Renewable and Sustainable Energy Reviews, vol. 104, pp.
209–234, 2019.
during the charging process to deal with some abnormal [6] G. Zhu, J. Dong, W. Shi, T. B. Soeiro, J. Xu, and P. Bauer, “A mode-
behaviors, for instance, a sudden movement of the receiver coil switching-based phase shift control for optimized efficiency and wide
or significant variation in operating temperature. Compared zvs operations in wireless power transfer systems,” IEEE Transactions
on Power Electronics, vol. 38, no. 4, pp. 5561–5575, 2022.
with the existing methods, the proposed parameter recognition [7] W. Zhang and C. C. Mi, “Compensation topologies of high-power
method avoids severe system detuning and significant load wireless power transfer systems,” IEEE Transactions on Vehicular
power ripple, and is able to accurately identify all of the Technology, vol. 65, no. 6, pp. 4768–4778, 2015.
unknown parameters of the resonant circuits. The accuracy [8] Y. Wang, Z. Yang, and F. Lin, “Design and implementation of wireless
power transfer systems with improved capacitor error tolerance,” IEEE
of the parameter recognition is experimentally validated in Transactions on Industrial Electronics, vol. 69, no. 5, pp. 4707–4717,
10 cases of parameter deviations, with the relative average 2021.
errors of the recognized results all less than 3%. Thanks to the [9] D.-H. Kim and D. Ahn, “Self-tuning lcc inverter using pwm-controlled
switched capacitor for inductive wireless power transfer,” IEEE Trans-
pre-start-up parameter recognition, the parameter deviations actions on Industrial Electronics, vol. 66, no. 5, pp. 3983–3992, 2018.
caused by coil misalignment and capacitance drift on both the [10] L. Shi, P. Alou, J. Á. Oliver, J. C. Rodriguez, A. Delgado, and J. A.
primary and secondary sides can be addressed without adding Cobos, “A self-adaptive wireless power transfer system to cancel the
reactance,” IEEE Transactions on Industrial Electronics, vol. 68, no. 12,
any extra switches and circuits, which is the main contribution pp. 12 141–12 151, 2020.
of this proposal. Based on the recognized parameters, an [11] R. Mai, P. Yue, Y. Liu, Y. Zhang, and Z. He, “A dynamic tuning method
impedance tuning method is carried out to cope with the utilizing inductor paralleled with load for inductive power transfer,”
IEEE Transactions on Power Electronics, vol. 33, no. 12, pp. 10 924–
impedance mismatch caused by parameter deviations. By 10 934, 2018.
regulating the system frequency and the phase shift angles of [12] Y. Lim, H. Tang, S. Lim, and J. Park, “An adaptive impedance-matching
the active rectifier, the reactance on both sides are minimized, network based on a novel capacitor matrix for wireless power transfer,”
and therefore, the dc to dc efficiency of the WPT prototype IEEE Transactions on Power Electronics, vol. 29, no. 8, pp. 4403–4413,
2013.
is enhanced by 4.3% to 15% in the experiments. Close-loop [13] T. C. Beh, M. Kato, T. Imura, S. Oh, and Y. Hori, “Automated impedance
verification is also implemented to further verify the feasibility matching system for robust wireless power transfer via magnetic reso-
of the proposed method in practice, with the time consumption nance coupling,” IEEE Transactions on Industrial Electronics, vol. 60,
no. 9, pp. 3689–3698, 2012.
for the whole parameter recognition process at around 3.6 s [14] J. Kim, D.-H. Kim, and Y.-J. Park, “Free-positioning wireless power
in the experiments. transfer to multiple devices using a planar transmitting coil and switch-
It is worth mentioning that this paper only presents the able impedance matching networks,” IEEE Transactions on Microwave
Theory and Techniques, vol. 64, no. 11, pp. 3714–3722, 2016.
application in the typical and widely-used SS compensation [15] R. W. Porto, V. J. Brusamarello, L. A. Pereira, and F. R. de Sousa, “Fine
network. However, in recent years, hybrid compensation struc- tuning of an inductive link through a voltage-controlled capacitance,”
tures, such as LCC-S, LCC-LCC, etc., are gaining more and IEEE Transactions on Power Electronics, vol. 32, no. 5, pp. 4115–4124,
2016.
more attention. For these hybrid networks, since the number [16] M. Moghaddami, A. Sundararajan, and A. I. Sarwat, “A power-frequency
of passive elements is much greater than that of the SS controller with resonance frequency tracking capability for inductive
compensation, the implementation of parameter identification power transfer systems,” IEEE Transactions on Industry Applications,
vol. 54, no. 2, pp. 1773–1783, 2017.
is more difficult, and additional information is required to
[17] A. Namadmalan, “Self-oscillating tuning loops for series resonant induc-
be measured. Considering the increasing popularity of these tive power transfer systems,” IEEE Transactions on Power Electronics,
hybrid structures, the research on how to further extend the vol. 31, no. 10, pp. 7320–7327, 2015.
proposed method to the hybrid compensation will be regarded [18] R. Mai, Y. Liu, Y. Li, P. Yue, G. Cao, and Z. He, “An active-
rectifier-based maximum efficiency tracking method using an additional
as our future work. Another noteworthy point is that, in measurement coil for wireless power transfer,” IEEE Transactions on
addition to its application for active impedance tuning, the Power Electronics, vol. 33, no. 1, pp. 716–728, 2017.
[19] A. Berger, M. Agostinelli, S. Vesti, J. A. Oliver, J. A. Cobos, and Gangwei Zhu (Student Member, IEEE) was born
M. Huemer, “A wireless charging system applying phase-shift and in Hunan, China, in 1997. He received the B. S.
amplitude control to maximize efficiency and extractable power,” IEEE degree in electrical engineering from Central South
Transactions on Power Electronics, vol. 30, no. 11, pp. 6338–6348, University, Changsha, China, in 2018, and the M.S.
2015. degree in electrical engineering from Shanghai Jiao
[20] S. Ann and B. K. Lee, “Analysis of impedance tuning control and Tong University, Shanghai, China, in 2021. He is
synchronous switching technique for a semibridgeless active rectifier currently working toward the Ph.D. degree in Delft
in inductive power transfer systems for electric vehicles,” IEEE Trans- University of Technology, Delft, The Netherlands.
actions on Power Electronics, vol. 36, no. 8, pp. 8786–8798, 2021. His research interests include advanced control and
[21] K. Song, G. Yang, H. Zhang, X. Huang, J. Jiang, Y. Lan, X. Huang, modulation for wireless power transfer.
J. Li, and C. Zhu, “An impedance decoupling-based tuning scheme for
wireless power transfer system under dual-side capacitance drift,” IEEE
Transactions on Power Electronics, vol. 36, no. 7, pp. 7526–7536, 2020.
[22] R. Dai, R. Mai, and W. Zhou, “A pulse density modulation based receiver Jianning Dong (Senior Member, IEEE) received
reactance identification method for wireless power transfer system,” the B.S. and Ph.D. degrees in electrical engi-
IEEE Transactions on Power Electronics, vol. 37, no. 9, pp. 11 394– neering from Southeast University, Nanjing, China,
11 405, 2022. in 2010 and 2015, respectively. He was a Post-
[23] W. Li, G. Wei, C. Cui, X. Zhang, and Q. Zhang, “A double-side Doctoral Researcher with the McMaster Automotive
self-tuning lcc/s system using a variable switched capacitor based on Resource Centre, McMaster University, Hamilton,
parameter recognition,” IEEE Transactions on Industrial Electronics, ON, Canada. Since 2016, he has been an Assistant
vol. 68, no. 4, pp. 3069–3078, 2020. Professor with the DC System, Energy Conversion
[24] Z. Luo, Y. Zhao, M. Xiong, X. Wei, and H. Dai, “A self-tuning and Storage (DCE&S) Group, Delft University of
lcc/lcc system based on switch-controlled capacitors for constant-power Technology (TU Delft), Delft, The Netherlands. His
wireless electric vehicle charging,” IEEE Transactions on Industrial research interests include electromechanical energy
Electronics, vol. 70, no. 1, pp. 709–720, 2022. conversion and contactless power transfer.
[25] W. Li, Q. Zhang, C. Cui, and G. Wei, “A self-tuning s/s compensation
wpt system without parameter recognition,” IEEE Transactions on
Industrial Electronics, vol. 69, no. 7, pp. 6741–6750, 2021. Francesca Grazian received her BSc degree
[26] Y. Yang, S.-C. Tan, and S. Y. R. Hui, “Front-end parameter monitor- in Electrical Engineering from the University of
ing method based on two-layer adaptive differential evolution for ss- Bologna in 2016, and her MSc and PhD degrees in
compensated wireless power transfer systems,” IEEE Transactions on Electrical Engineering from TU Delft in 2018 and
Industrial Informatics, vol. 15, no. 11, pp. 6101–6113, 2019. 2023, respectively. Her main research interests are
[27] J. Yin, D. Lin, C. K. Lee, T. Parisini, and S. Hui, “Front-end monitoring Power Electronics and Wireless Power Transfer for
of multiple loads in wireless power transfer systems without wireless battery charging. Since January 2023, she works as
communication systems,” IEEE Transactions on Power Electronics, Electrical R&D Engineer at Laser Precision Solu-
vol. 31, no. 3, pp. 2510–2517, 2015. tions in Amsterdam, the Netherlands. Dr. Grazian
[28] R. Rao, “Jaya: A simple and new optimization algorithm for solving was a recipient of the Best Poster Award at the ECPE
constrained and unconstrained optimization problems,” International European PhD School in 2022, Gaeta, Italy, and the
Journal of Industrial Engineering Computations, vol. 7, no. 1, pp. 19– Best Paper Award at the International Conference on Power Electronics and
34, 2016. Motion Control (PEMC) in 2022.
[29] C.-S. Wang, G. A. Covic, and O. H. Stielau, “Power transfer capability
and bifurcation phenomena of loosely coupled inductive power transfer
systems,” IEEE transactions on industrial electronics, vol. 51, no. 1, pp.
148–157, 2004.
[30] J. Zeng, S. Chen, Y. Yang, and S. Y. R. Hui, “A primary-side method
for ultrafast determination of mutual coupling coefficient in milliseconds Pavol Bauer (SM) received his Master’s degree in
for wireless power transfer systems,” IEEE Transactions on Power electrical engineering from the Technical University
Electronics, vol. 37, no. 12, pp. 15 706–15 716, 2022. of Kosice, Kosice, Slovakia, in 1985, and his Ph.D.
[31] Y. Yang, S. C. Tan, and S. Y. R. Hui, “Fast hardware approach degree from the Delft University of Technology,
to determining mutual coupling of series–series-compensated wireless Delft, The Netherlands, in 1995.
power transfer systems with active rectifiers,” IEEE Transactions on From 2002 to 2003, he was with KEMA (DNV
Power Electronics, vol. 35, no. 10, pp. 11 026–11 038, 2020. GL), Arnhem, The Netherlands, on different projects
[32] J. Yin, D. Lin, T. Parisini, and S. Hui, “Front-end monitoring of the related to power electronics applications in power
mutual inductance and load resistance in a series–series compensated systems. He is currently a Full Professor with the
wireless power transfer system,” IEEE Transactions on Power Electron- Department of Electrical Sustainable Energy, Delft
ics, vol. 31, no. 10, pp. 7339–7352, 2015. University of Technology, and the Head of DC
[33] D. Lin, J. Yin, and S. R. Hui, “Parameter identification of wireless power Systems, Energy Conversion, and Storage Group. He is also a Professor with
transfer systems using input voltage and current,” in 2014 IEEE energy the Brno University of Technology, Brno, Czech Republic, and an Honorary
conversion congress and exposition (ECCE). IEEE, 2014, pp. 832–836. Professor with the Politehnica University Timisoara, Timisoara, Romania.
[34] Toshiba Electronic Devices and Storage Corporation, “Power He has published over 120 journal articles and 500 conference papers in
mosfet selecting mosfets and consideration for circuit design”, his field. He is an author or co-author of 8 books, holds 7 international
2018. [Online]. Available: https://toshiba.semicon-storage.com/info/ patents, and organized several tutorials at international conferences. He has
application note en 20180726 AKX00064.pdf?did=13416 worked on many projects for the industry concerning wind and wave energy,
[35] Fairchild Semiconductor, “Resistor-capacitor (rc) snubber design for power electronic applications for power systems such as Smarttrafo; HVDC
power switches”, 2014. [Online]. Available: https://www.digikey.com/ systems, projects for smart cities such as photovoltaic (PV) charging of electric
en/articles/resistor-capacitor-rc-snubber-design-for-power-switches vehicles, PV and storage integration, contactless charging; and he participated
[36] Nordic Semiconductor, “nRF24L01 single chip 2.4 GHz radio in several Leonardo da Vinci and H2020, and Electric Mobility Europe EU
transceiver”, 2006. [Online]. Available: https://www.sparkfun.com/ projects as a Project Partner (ELINA, INETELE, E-Pragmatic, Micact, Trolley
datasheets/Components/nRF24L01 prelim prod spec 1 2.pdf 2.0, OSCD, P2P, and Progressus) and a Coordinator (PEMCWebLab.com-
[37] J. Liu, G. Wang, G. Xu, J. Peng, and H. Jiang, “A parameter iden- Edipe, SustEner, Eranet DCMICRO).
tification approach with primary-side measurement for dc–dc wireless- Prof. Bauer is the Former Chairman of Benelux IEEE Joint Industry Appli-
power-transfer converters with different resonant tank topologies,” IEEE cations Society, Power Electronics and Power Engineering Society Chapter,
Transactions on Transportation Electrification, vol. 7, no. 3, pp. 1219– the Chairman of the Power Electronics and Motion Control Council, a Member
1235, 2020. of the Executive Committee of European Power Electronics Association,
and also a Member of the International Steering Committee at numerous
conferences.

A Parameter Recognition Based Impedance Tuning Method For SS-Compensated Wireless Power Transfer Systems

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Parameter Recognition Based Impedance Tuning Method For SS-Compensated Wireless Power Transfer Systems

Uploaded by

Copyright:

Available Formats

This article has been accepted for publication in IEEE Transactions on Power Electronics.

A Parameter Recognition Based Impedance Tuning

Another typical method to deal with the impedance mis- deviations.

350 230 350 230

325 210 0.15

1000 dP=-0.2, dS=-0.2 dP=0, dS=-0.2 dP=0.2, dS=-0.2 1.0

(a) (a) (b)

1000 Power is limited 2) Transfer Efficiency: Another effect of the parameter

point (85, 4.6) dP=0.2, dS=-0.2

that the peak current point can be derived by sweeping the

respectively. Moreover, by sweeping the parameters of the

point (85, 800)

secondary-side information, the relationship between IS1 and Start

IP 1 is established as Increase the frequency from fL

15 dP=-0.2, dS=-0.2 dP=0, dS=-0.2 dP=0.2, dS=-0.2 Start

A1(fA1,10) B1(fB1,10) Calculate the fitness value of

Yes Is the new individual No

IM=10A and its fitness value

Fig. 15. Flowchart of the JAYA algorithm.

algorithm, various algorithm-specific parameters, including the S1 S3 M S5 S7 Iout

Oscilloscope Launchpad F28379D 0

Fig. 19. Magnitude-frequency response of the designed digital LPF. Herein,

TABLE II TABLE III

Average Relative Error[%]

Convergent Fitness value J

Convergent Fitness value J

(a) (b) 2.5

M , (b) compensation capacitances CP and CS .

well. Additionally, the population size Psize and the maximum 1

generation numbers Genmax of the JAYA algorithm are also 0.5

0.012 0.3 0.5 0.3

vab fs=85kHz v ab fs=91kHz

Fig. 27. Measured operating waveforms and dc to dc efficiency of case-B5

Primary Measurement Board Secondary Measurement Board Wireless Communication

Output Power [W]

Fig. 31. Hardware modules for the close-loop experiments.

φ=-14.8° φ=-31.6° v cd [500V/div]

You might also like