A Dynamically Adaptable Impedance-matching System for Midrange Wireless Power Transfer with Misalignment

To enable the geometrical freedom envisioned for wireless power transfer (WPT), fast dynamic adaptation to unpredictable changes in receiver position is needed. In this paper, we propose an adaptive impedance-searching system that achieves good impedance matching quickly. For fast and robust operation, the proposed method consists of three steps: system calibration, coarse search, and fine search. The proposed WPT system is characterized using distance variation and lateral and angular misalignment between coils. The measured results indicate that the proposed method significantly reduces searching time from a few minutes to approximately one second. Furthermore, the proposed system achieves impedance matching with good accuracy. The robust impedance-searching capability of the proposed system significantly improves power transfer efficiency. At 6.78 MHz, we achieve a maximum efficiency of 89.7% and a high efficiency of >80% up to a distance of 50 cm. When the center-to-center misalignment is 35 cm, the efficiency is improved from 48.4% to 74.1% with the proposed method. At a distance of 40 cm, the efficiency is higher than 74% for up to 60° of angular rotation. These results agree well with the simulated results obtained using a lumped-element circuit model.


Introduction
In recent years, the potential of power transmission through wireless means has attracted a great deal of research and commercial interest [1].This interest is closely related to the increasing number of mobile devices in our daily lives, which require frequent battery charging because of their high power consumption.To unify the charging protocol of various portable electronic devices, the Wireless Power Consortium (WPC) recently developed a new standard, Qi 1.1 [2].This standard development reflects the substantial interest in the wide range of wireless power charging applications.
Inductive coupling has been used for wireless power transfer (WPT) in many commercial applications.The drawback of wireless power charging using inductive coupling is that the operation is limited to a very short range.Efforts are being made to extend the operating range, for example, by using repeaters [3] and metamaterials [4,5].To extend the power transfer to a more convenient midrange distance, WPT based on resonant coupling has recently been demonstrated experimentally [6][7][8][9][10].This new approach increases the power transmission range by focusing the energy transfer to the narrow frequency region allowed by the high Q-factor.Thus, power transmission is extended to a midrange distance, enabling greater geometrical freedom for mobile device charging.
Nevertheless, the received power level is rather sensitive to both alignment and distance changes between the coils [11].Any change in coil position from the initial optimal location results in degraded transmission efficiency.The mechanical tuning method, in which the distance and/or angle between the coils are adjusted by physically moving coils, has been used to optimize power efficiency [6].To restore the original efficiency, manual impedance tuning has been used [12,13]; however, it is time-consuming and requires expertise to obtain the proper matching between coils.In a more automated approach, the frequency tracking technique has been proposed [14]; however, it uses a rather wide frequency band (from 6.17 MHz to 6.78 MHz), which can potentially cause frequency regulation issues.In [15], the authors reported a WPT system requiring an even larger bandwidth of 14 MHz (from 8 MHz to 22 MHz).Main consideration in using frequency tracking method is the availability of bandwidth, which complies with frequency regulation not to cause interference with other communicating devices.For power transmission, narrow bandwidth is usually enough.Therefore, impedance matching at a fixed frequency is the preferred approach to avoid interference issue.
In these previous works, however, only distance variation was considered; consequently, the results of these experiments cannot be used to evaluate efficiency performance for misaligned coils.
In this paper, we discuss a midrange WPT system that facilitates geometric freedom through fast adaptation capability.To quickly adapt to a dynamically moving coil, impedance matching is performed in three steps: system calibration, coarse search, and fine search.Using the coil parameters obtained in the system calibration step, the coarse search step estimates the coil position and determines the initial values for the impedance search.Then, the fine-search step refines the values in the matching network to improve the accuracy.The performance of the proposed WPT system in terms of searching time, matching accuracy, and efficiency is characterized under various practical ranges and positions of the coil.Measurements show that the proposed approach significantly reduces searching time.In addition, high efficiencies are achieved over a wide range of distance and alignment cases.These results show that the proposed approach has good potential to enable the geometrical freedom envisioned for the midrange WPT system.

System Model
Figure 1a shows the WPT system using resonant coupling.The source voltage is VS, and the source and load resistances are RS and RL, respectively.The source and load loops are inductively coupled to the multi-turn spiral resonators (Resonators 2 and 3).With a large turn ratio, RS and RL are converted to a large effective resistance in parallel with the LC resonator [20], creating the high Q-factor needed for resonant coupling.The distance between the internal resonators is d.The center-to-center distance (lateral misalignment) between the resonators is ρ.For the receiver, consisting of resonator 3 and the load loop, there may be angular misalignment θ.The power transmission can be affected by: Because the size of the designed coils is small compared with the wavelength, we build the system model using lumped elements, as shown in Figure 1b.The parameters of each coil are described by lumped RLC elements, where Ri (i = 1-4) represents ohmic and radiation losses of the coil.The direct-coupling coefficients between each of the two adjacent coils are k12, k23, and k34.The cross-coupling coefficients are k13, k14, and k24.Applying basic circuit theory, we obtain the current Ii in each coil by using: where: At great distance, the strength of the cross-coupling coefficients is substantially smaller than that for direct coupling; we can therefore use the approximation: To simplify the analysis, we assume that all coils are resonating at the same frequency, ω0 = 2πf0.When RS and RL are connected to the source and load loops, respectively, the Q-factors of the loops are affected by these resistors.Then, the impedance values (passive and reciprocal) can be expressed as: where Qi is the quality factor.In the WPT system, the coupling coefficient and Q-factor are two key parameters.Solving Equation (1) using the Zij given in Equation ( 2), currents I1 in the source loop and I4 in the load loop are obtained, as shown in Equations ( 3) and (4): By applying Equations (2) to Equations ( 3) and (4), we obtain more compact expressions for these currents: where: To achieve impedance matching at the input port, we examine the input impedance ZIN by looking into the source loop.Using: we obtain: In the case of R1 << RS, Equation ( 7) is simplified to: When one of the geometric parameters (d, ρ, θ) changes, it affects the coupling coefficients, thereby resulting in change of ZIN.A smart matching network senses this change in ZIN so that the converted impedance ZIN,match is matched again to RS. Implementation of the matching network is described in Section 3.
At close distance, such as in an over-coupled region [8], we must consider the effect of cross-coupling on ZIN.The cross-coupling causes shifts in the frequencies of the transmission peaks.The frequency shift can be explained using the phase of the two modes [8], where the strength of the cross-coupling term determines the amount of the frequency shift.With the cross-coupling coefficients included, we solve Equation (1) for I1 and obtain: where: Using Equation ( 9), it can be shown that the system model with cross-coupling creates imaginary parts in ZIN.

Parameter Extraction
In our WPT system, the source and load loop coils each have a radius of r1 = 20 cm.All coils are fabricated with copper of wire radius a1 = 0.15 cm.Inductance Lloop for the source and load loops is obtained using [21]: where μ0 is the permeability of free space.Figure 2 shows the coil geometry of resonators 2 and 3. Inductance Lspiral (in μH) of this multi-turn flat spiral coil is calculated using the Wheeler formula [22]:  When the axes of the circular filaments are parallel, the mutual inductance M is obtained using Maxwell's method [23].For two parallel single-turn coils having radii rm and rn, M can be expressed as: where J0 and J1 are Bessel functions of the zero-th and first order, respectively.This expression does not contain wire radius a; it is assumed that ratios a/rm and a/rn are sufficiently small [24].Using Equation ( 12), we calculate the coupling coefficient kij,calc between the two multi-turn coils using: where N2 and N3 are numbers of turns, and Li and Lj are the inductances obtained using Equations ( 10) and (11).
For lateral and angular misalignment between circular filaments, Grover [25,26]developed a method to calculate M based on Neumann's formula.Using the method of vector potential, Babic et al. [27] developed a general mutual inductance formula requiring simple numerical integration.To reduce computational effort, we use this method to obtain the coupling coefficients.
To extract coupling coefficient kij,meas by experiment, we use the frequency splitting occurring in the near distance expressed as: where fH and fL are the higher (odd mode) and lower (even mode) split frequencies of the magnetically-coupled resonators, respectively [28].At great distance, the coupling is weak and the frequency splitting disappears.Because the WPT system of Figure 1 is a two-port network, we can capture the system characteristic using transmission coefficient |S21|.Then, kij,meas is extracted by fitting the system model of Figure 1b to the measured |S21|.Figure 3a shows measured frequency splitting versus distance.The splitting frequency ranges from fL = 5.92 MHz to fH = 8.13 MHz. Figure 3b shows |S21| versus distance d measured using fL, fH, and f0 = 6.78MHz, respectively.When d is smaller than 40 cm (the boundary between the over-coupled and under-coupled regions [8]), frequency splitting occurs, and |S21| is larger when measured at fL than at either f0 or fH. Figure 4a presents a comparison between the measured and calculated coupling coefficients (k23, k14, and k13 = k24) as functions of d (without misalignment, ρ = 0 cm and θ = 0°).The coupling coefficients decrease rapidly with d, and the calculated results show good agreement with the measured data.Figure 4b shows the measured and calculated coupling coefficients as functions of ρ for d = 30 cm. Figure 4c shows the coupling coefficients as a function of θ for d = 40 cm.The coupling coefficients decrease slowly for θ < 60°, whereas they converge rapidly to 0 when θ approaches 90° (for example, k23, k13, and k14 are 2 × 10 −5 , 9 × 10 −6 , and 7 × 10 −6 , respectively, at θ = 89.99°).Table 1 shows the measured electrical parameters of the coils using an Agilent 8714ES network analyzer (Santa Rosa, CA, USA).The data is measured at f0 = 6.78MHz, which is the resonant frequency of the coils.For resonant frequency tuning, the values of the external variable capacitors in the source and load loops are carefully tuned using the network analyzer.For resonators 2 and 3, we use the self-capacitance existing in the resonator geometry [8].In the experiment, we reduce resonator length by cutting it little by little until we obtain the resonant frequency.The calculated inductances using Equations (10) and (11) are Lloop = 1.31 μH and Lspiral = 38.98 μH, which agree with the measured data within a 3.1% difference.To realize variable inductor Ltune, we use the lumped element approach for compact realization [29].Ltune consists of fixed inductor Lfix and variable capacitor Cvar1 in a parallel configuration, as shown in the inset of Figure 5. From the complex impedance created by Lfix and Cvar1, Ltune can be expressed as: where: is the resonance frequency of the variable inductor, Lpar = 60 nH is the parasitic inductance from the board trace, and Cpar = 15 pF is the parasitic capacitance (average value) of the relay switch.When ZIN changes due to coil movement, the tunable matching network searches for values of Ltune and Cvar2 for impedance matching and adapts ZIN,match to the new coil position.To facilitate a dynamically moving receiver, fast adaptation capability is desirable for the matching network.
The switches are realized using an electrically controlled relay [30].When the relay is actuated by 5 V, the switching time is approximately 10 ms.The relay can switch to as high as 250 V and 2 A. We employ a total of 20 switches, with 16 switches (SWS,1-SWS,8 and SWP,1-SWP,8) used to control the values of Cvar1 and Cvar2.The four switches (SW1-SW4) provide flexibility in realizing the matching network.With SW2 = SW3 = ON while other switches are OFF, the matching network is bypassed.The bypass configuration is used in the system calibration step (Figure 6).The port impedance of the components connected to the matching network corresponds to RS = 50 Ω.Although the current setup uses 50 Ω due to constraints from the components and measurement equipment, the value of RS does not need to be 50 Ω.To allow for impedance matching in a general case of source impedance ZS = RS + j•XS, the matching network provides two different configurations.
With SW1 = SW3 = ON and SW2 = SW4 = OFF, the series-L/shunt-C configuration is realized.Then, the desired values for tune L and var2 C for impedance matching can be expressed as: where: With SW2 = SW4 = ON and SW1 = SW3 = OFF, the shunt-C/series-L configuration is realized, and the desired values for tune L and var2 C for impedance matching can be expressed as: where: By substituting Equations ( 16) and (18) into Equation ( 15), we calculate the initial values for Cvar1: The fine-search step (Figure 6) refines the values of Cvar1 and Cvar2 for impedance matching; i.e., tune L → tune L and var2 C → var2 C  .
To realize finely tunable impedance, a small Lfix value is desirable in the Ltune.However, a small Lfix leads to narrow inductance and impedance tuning ranges.To determine a suitable value for Lfix, we examine ZIN and the magnitude of the reflection coefficient ГIN = (ZIN − 50)/(ZIN + 50) of the WPT system.Figure 7 shows the measured ZIN when d is varied from 15 cm to 100 cm in 5 cm steps.ZIN is measured by connecting the source loop to port 1 of a network analyzer.ZIN is capacitive in the near distance, whereas it changes to inductive when d > 40 cm in our system.When d changes from 15 cm to 100 cm, |ΓIN| varies from 0.242 to 0.908.Additionally, shown in Figure 7, are the measured ZIN values when misalignment ρ changes from 5 cm to 45 cm in 5 cm steps at d = 15 cm.In this case, |ΓIN| varies from 0.066 to 0.924.Considering the trade-off between the achievable impedance range and impedance step, we choose Lfix = 440 nH.Using the Lfix value, the tunable matching network covers the |ΓIN| range of our WPT system.For the ZIN simulation, we input both the coil parameters (Table 1) and coupling coefficients (Figure 4) into the system model of Figure 1b.We observe good agreement between the measured and simulated ZIN, indicating the accuracy of the extracted parameters.

Three-Step Searching Method
To select a search method for the values of Cvar1 and Cvar2, we examine the two-port network of the WPT system (Figure 1) using |S21|.Figure 8 shows the simulated |S21| values for d = 15 cm, 40 cm, and 80 cm.The results show a single peak with a smooth slope, which agrees with a previous report [18].Examining all impedance points, however, is too time consuming to use in an environment in which coils move dynamically; in our system, the time required to actuate a relay switch is approximately 10 ms.Because Cvar1 and Cvar2 change from 10 pF to 1320 pF in 10 pF steps, the dimensions of the search matrix for Cvar1 and Cvar2 will be 132 × 132 = 17,424.Then, the required time is approximately 3 min.To reduce search time and quickly adapt to dynamically moving coils, we propose an efficient searching method for the tunable matching network.Figure 6 presents a flowchart of the proposed searching method.The method consists of three steps: system calibration, coarse search, and fine search.The operations performed in each step are as follows: Step 1: The system calibration step is performed one time to calculate and measure the system parameters.Using the coil parameters in (13), kij,calc as functions of d are obtained.With the matching network in the bypass configuration (SW2 = SW3 = ON while other switches are OFF), the reflected power Preflect at the input of the matching network is measured as d is changed from 15 cm to 100 cm in 5 cm steps.The data are used to calculate the initial value of ZIN for Step 2.
Step 2: In the coarse-search step, the distance d is estimated to obtain an initial value for the impedance search.The receiver (resonator 3 and load loop) can be positioned in either an over-or under-coupled region.When we use a single frequency f0 = 6.78MHz, frequency splitting occurs and two distances can exist for which |S21| (thus, the reflected power Preflect) is the same, as shown in Figure 3b.Therefore, the WPT system cannot distinguish between the two regions.To detect the region where the receiver is positioned, we measure Preflect while scanning the frequency in the range from 5.5 MHz to 8.5 MHz (a low power level may be used during frequency scanning).At a near distance d, belonging to the over-coupled region, frequency splitting creates multiple transmission peaks [8].In the case in which there are more than two local minimum peaks in Preflect, therefore, the system determines that the receiver is positioned in an over-coupled region.With increasing d, coupling decreases until the transmission peaks converge.Therefore, at a distance d, belonging to the under-coupled region, Preflect has one minimum peak.By detecting the number of transmission peaks, the region (over-or under-coupled) is identified.Then, the distance to the receiver is estimated by comparing the measured Preflect with the data obtained in Step 1.Using the estimated distance, the corresponding coupling coefficients are obtained using Figure 4a.By combining them with the coil parameters shown in Table 1, we calculate Zij (i,j = 1-4) and ZIN using Equations ( 2) and ( 9), respectively.Using Equations ( 16)-( 19) and ZIN, we obtain the values of tune L and var2 C  for the matching network.
Step 3: In the fine-search step, the accuracy of the impedance matching is improved by refining the values in the matching network.The method used in this step is based on the hill-climb searching technique [31], which has been modified to enhance searching speed.In this step, the routine continuously evaluates the neighboring points in the direction of minimizing Preflect until a valley point is found.To speed up the searching process, previously visited points are marked and eliminated during the evaluation step.A detailed description of the proposed method is as follows.Let the two-dimensional values of Cvar1 and Cvar2 (with 10-pF steps) be mapped to row indices n and column indices m of square-matrix Z (n, m = 1-132).Let z(n,m) be the current location at row n and column m in Z (z ϵ Z).Let L be the list of points that have been visited during the searching process in Z.Then, the system performs the process as follows: (1) Using the matching component values calculated in Step 2, set the initial starting point z(n,m)ϵ Z.
(2) Add point z to the list of visited points L, or L = {z}.
(3) Determine the set of eight neighboring points A(z) based on current location z. (4) Find the set of unvisited neighboring points R(z) using A(z) and list L.
(5) Using the detected Preflect at the coupler, find the best matching point zopt among the eight neighboring points in R(z).(6) Check whether zopt remains unchanged.If yes, the system process ends.If no, add the already examined points R(z) to the list L; i.e., L = L U {z'| z' ϵ R(z)}.(7) Assign new starting point z using local optimal point zopt for the next loop, and proceed to Step 3 (3).

Experimental Results
Figure 9 presents a system block diagram of the proposed WPT.The system includes a signal generator, a radio-frequency (RF) power amplifier [32], a 20 dB directional coupler [33], a power detector [34], a tunable matching network, switch drivers, and a control unit.The RF power amplifier has 50-Ω output impedance (RS = 50 Ω), providing an available power Pavail = 30 dBm (20 Vp-p in 50 Ω load).An Agilent N1914A power meter and 8481B power sensor (RL = 50 Ω) are used to measure the output power Pout.The coupler and power detector are used to measure reflected power Preflect.Through the coupled reflected port, Preflect is scaled down so that it is within the dynamic input range (−50 dBm to 20 dBm) of the detector.The detector generates a linear voltage from 2.3 V to 0.5 V, which is inversely proportional to power.The control unit MSP430 includes a 10-bit analog-to-digital converter (ADC), which coverts this voltage to digitized data.Based on the data, the control unit executes the impedance searching routine (Figure 6) to minimize Preflect.Therefore, this work focuses on optimizing the power transfer (transmission) efficiency without a rectifier at the load.The transmission efficiency is defined by the ratio of Pout to Pavail.For high-power applications, system energy efficiency, which includes the loss in the power source, is usually considered.Under the impedance matching condition, half of the power is dissipated in the power source, limiting the system efficiency to 50%; reducing the loss in the power source is desirable [1].In this work, however, design of a low-loss power source, i.e., an RF power amplifier, is out of scope.Furthermore, this work emphasizes enabling the geometric freedom envisioned for mobile devices.Therefore, we consider transmission efficiency for low-power devices operating in midrange applications, in which the loss of the power source is not of primary concern [1].
Figure 10 shows the experimental setup for the proposed WPT system.The coil structure is similar to one used in a previous work [35]; however, the resonant frequency is tuned to 6.78 MHz in this work.The dynamic impedance-matching capability of the fabricated WPT system is characterized under the change of three geometric parameters: distance d, lateral misalignment ρ, and angular misalignment θ.

Distance Change
Figure 11 shows measured impedance ZIN,match at the input port of the matching network (Figure 9).When we compare the results of the two methods, we note that, compared to the one-step method, the initial starting point is closer to the center of the Smith chart when the three-step method is used.Thus, the trajectory of ZIN,match is significantly shorter, indicating a faster searching capability.
The number of examined points (including the eight neighboring points around zopt) during the fine search is equal to the number of relay switching.Using the one-step method, the number of points are 250, 221, and 307 at d = 15 cm, 40 cm, and 80 cm, respectively.Using the three-step method, it is 94, 47, 18 at d = 15 cm, 40 cm, and 80 cm, respectively.For practical application of the WPT system, operating lifetime should be considered.Based on the datasheet [30], the relay allows the maximum 10 8 number of switching operations before end-of-lifetime.Based on this data, we can estimate the allowed number of impedance matching operations.Compared to the one-step method, the number of relay switching operation is significantly reduced using the three-step method, extending the operating lifetime of the WPT system.
To examine the searching speed, we measure the output of power detector Vdetect as a function of time.Figure 12 shows the measured results at three distances: d = 15 cm, 40 cm, and 80 cm.While the impedance matching is performed, Vdetect gradually increases with time (|ГIN,match| decreases with time).Using the one-step method, the searching time is 2-3 s.Using the three-step method, the times to reach the final point are 0.94 s, 0.47 s, and 0.18 s for d = 15 cm, 40 cm, and 80 cm, respectively.In the case in which all impedance points are sequentially examined, the required time is approximately 180 s.It is evident that the three-step method significantly reduces the searching time: using the proposed three-step method, the searching time is reduced more than 180 times.
The detector output Vdetect provides an alternative means for assessing the accuracy of impedance matching.Using the datasheet in [34], the measured Vdetect is converted to Preflect.From the relationship between ГIN,match and Preflect, we obtain: reflect IN,match avail coupler where ILcoupler = 0.8 dB is the insertion loss of the coupler.When the three-step method finishes impedance matching, |ГIN,match| obtained using Equation ( 21) is −26.1 dB, −42.7 dB, and −25.4 dB for d = 15 cm, 40 cm, and 80 cm, respectively.These results indicate good impedance matching, and correlate well with the results obtained using ZIN,match.Efficiency improvement with the proposed method is measured as a function of d.To examine the impedance matching effect, we measure the transmission efficiency.Considering that small mobile receivers typically do not have sufficient power and space for the system of a tunable matching network, the matching network is realized in the transmitter.Therefore, the impedance mismatch effect at the load loop of the receiver is removed by using the method in [35].Figure 13 presents a comparison of the efficiency versus d.Without a matching network, the efficiency rapidly degrades when the distance deviates from the optimum coupling distance of approximately 35 cm.However, using the proposed tuning method, efficiency is significantly improved; it increases by 57.5% and 47.1% at d = 15 cm and d = 70 cm, respectively.In the near distance (d < 35 cm), efficiency is higher than 88% with a maximum value of 89.7%.These results additionally demonstrate the measured efficiency using the mechanical tuning method [35].Using Equation ( 8), we obtain the condition for the coupling coefficient k23,match when ZIN is matched to RS, which is expressed as: For given coil parameters, the condition Equation ( 22) specifies the k12 and k34 required to achieve k23,match.When one of the geometric parameters (d, ρ, θ) changes, we can adjust k12 and k34 to satisfy Equation (22).The insertion loss of the matching network is 0.45 dB (average value).Because of the insertion loss, the peak efficiency achieved using the proposed tuning is 2.3% lower than the one obtained using mechanical tuning.However, changing k12 and k34 by moving coils is not convenient; moreover, it is rather slow in practice.The measured results show that the efficiency achieved using the proposed tuning method closely follows that of the mechanical tuning method.The results demonstrate that the slow mechanical tuning can be replaced with the fast electric tuning capability of the proposed WPT.

Lateral Misalignment
Figure 14 shows the measured trajectory of ZIN,match as misalignment ranges from ρ = 0 cm to 40 cm in 10 cm steps.In the proposed WPT system, this misalignment is not explicitly detected (only d is estimated using Preflect in Step 2).Therefore, there will be error in the estimated ZIN.At ρ = 10 cm, the measured ZIN is 12.6 − j24.4 Ω, whereas the estimated value is 8.6 − j16.8 Ω.Although the estimated ZIN differs somewhat from the actual value, the results, presented in Figure 14, show that ZIN,match gradually converges to the center of the Smith chart.This result indicates robust searching capability that is successful in dealing with mismatch caused by ρ.In addition, we note that the ZIN,match trajectory of the three-step method is shorter than that of the one-step method.At ρ = 40 cm, |ГIN,match| improves from −12.2 dB to −31.9 dB (and from −1.1 dB to −32.0 dB) using the three-step method (and using the one-step method).In the range where ρ changes from 0 cm to 40 cm, the worst case of measured |ГIN,match| at the final point is −28.1 dB for ρ = 0 cm, showing a good impedance match.
Figure 15 shows the measured Vdetect and |ГIN,match| as functions of time.As the impedance matching process is performed, |ГIN,match| gradually decreases with time.In the range where ρ changes from 0 cm to 40 cm, the worst case of measured |ГIN,match| using the three-step method is −26.2 dB, indicating good impedance matching.When we compare the search time for the case of ρ = 0 cm, the search time is not significantly affected by ρ.The time to reach the final value is <1.2 s using the three-step method.Using the one-step method, the time is approximately 3 s (worst case).Although the misalignment causes a shift in the estimated ZIN, we see that the proposed method can nevertheless find a good impedance match within a time period suitable for tracking dynamically moving coils.without matching.Using the proposed tuning method, relatively constant efficiency is obtained for up to ρ = 30 cm.At d = 15 cm, the measured efficiency improves by 39.1% and 35.1% at ρ = 10 cm and ρ = 40 cm, respectively.At d = 30 cm, the efficiencies with and without a tuning network both decrease with increasing ρ.Using the proposed method, the efficiency is significantly improved.At ρ = 35 cm, it increases from 48.4% to 74.1%.The simulated efficiencies are also shown, which agree well with the measured data.

Angular Misalignment
Figure 17 shows the measured trajectory of ZIN,match when there are angular misalignments ranging from θ = 0° to 75° in 15° steps.The results show that ZIN,match gradually converges to the center of the Smith chart using either method.At θ = 30°, |ГIN,match| improves from −14.1 dB to −53.7 dB (and from −2.8 dB to −53.5 dB) using the three-step method (and using the one-step method).Although the final |ГIN,match| values are similar, we achieve a significantly shorter trajectory of ZIN,match using the three-step method.When θ changes from 0° to 75°, the worst case of measured |ГIN,match| at the final point is −31.8 dB, indicating that the proposed system can handle significant angular misalignment.Figure 18 shows the measured Vdetect and |ГIN,match| as functions of time for different θ.Using the three-step method, the maximum search time is approximately 1.7 s, which occurs for θ = 30°.In the range where θ changes from 0° to 75°, the worst case |ГIN,match| is −35 dB.These results show that the proposed method can dynamically track and find a good impedance match under a wide range of angular misalignments.Figure 19 shows the measured efficiency when θ changes from 0° to 75° at d = 40 cm.For up to θ = 60°, the efficiency is higher than 74%.For θ = 75°, the efficiency improves from 26.6% to 58.2% using the proposed method.We note that the proposed system achieves an efficiency >58% at θ = 75°.Considering the high angle of rotation, this efficiency performance is notable.To verify the measured data, we use the coupling coefficients of Figure 4c.Using the method in [27], we obtain k23 = 0.029, k12 = k34 = 0.29, k13 = k24 = 0.013, and k14 = 0.009 at θ = 75°.These numbers and the electrical parameters (Table 1) are used in the equivalent circuit model shown in Figure 1b.The simulated data show good agreement with the measured data.These results demonstrate excellent efficiency performance of the proposed system, which can deal with significant angular rotation.To evaluate dynamic adaptation to unpredictable receiver position changes, we select a sequence of test cases, as shown in Figure 20.Each test case along the trajectory is indexed from 1 to 12, and it is built with a random combination of d, ρ, and θ. Figure 21a shows measured ZIN,match and |ГIN,match| along the trajectory index.The maximum |ГIN,match| is 0.053 (−25.5 dB), which occurs at index-8 (d = 80 cm, ρ = 20 cm, and θ = 15°).Figure 21b shows the measured efficiencies along the path.Among the selected test cases, the maximum efficiency is 90.5% at index 3, while the minimum efficiency is 36.6% at index 6.The highest efficiency improvement of 42.8% is achieved at index 7.Overall, the proposed three-step method provides fast adaptation (<1 s) to changes of various receiver positions.For practical application of a WPT system, characterization of leakage/fringe fields from WPT is important [36].Therefore, we measure the leakage/fringe fields of the four-coil WPT without a ferrite core or aluminum shielding.A safety guideline for electromagnetic exposure is available from IEEE C95.1 standard [37].The leakage magnetic B-field is measured at 5 cm from the edge of the resonator (35 cm from the center).We use a Narda 8715 survey meter equipped with an 8732D magnetic field probe.To find a worst-case high-power condition under which human safety is still secured, we increase Pavail to 7.8 W. At d = 100 cm, the load coil receives about 2 W.Under these conditions, the measured B-field is 1.34 A/m at the side of resonator 2, gradually decreasing to 0.15 A/m at the side of resonator 3. The IEEE C95.1 standard allows about 2 A/m at 6.78 MHz.Under these operating conditions, therefore, the proposed WPT system can be used within the safety guidelines.

Conclusions
We propose an efficient three-step impedance searching method for a dynamically adaptable midrange WPT system.In the system calibration step, the coil parameters are obtained.In the coarse-search step, the coil position is estimated and the initial values for the matching network are determined.In the fine-search step, the impedance matching accuracy is improved by refining the values in the matching network.Measurements show that the proposed three-step method achieves good impedance matching within a short time and is therefore suitable for seamless wireless power operation.In addition, characterizations under several distance and alignment changes demonstrate the robustness of the proposed method.Using this method, the power transfer efficiency is significantly improved.In cases in which distance changed, we achieve a high efficiency of more than 50% for up to d = 80 cm, with a maximum efficiency of 89.7%.In the case of lateral misalignment, efficiency improved by 39.1% and 35.1% at ρ = 10 cm and ρ = 40 cm, respectively, at d = 15 cm.In the case of angular misalignment of θ = 75°, the efficiency improved from 26.6% to 58.2% at d = 40 cm.These results indicate that the proposed system can handle significant lateral and angular misalignment as well as distance changes.These results will be useful for convenient provision of power in dynamic environments in which mobile devices frequently change position.Moreover, further improvement in the dynamic adaptation is expected by adding communication capability between the receiver and the transmitter [14,15,38].

( 1 )Figure 1 .
Figure1ashows the WPT system using resonant coupling.The source voltage is VS, and the source and load resistances are RS and RL, respectively.The source and load loops are inductively coupled to the multi-turn spiral resonators (Resonators 2 and 3).With a large turn ratio, RS and RL are converted to a large effective resistance in parallel with the LC resonator[20], creating the high Q-factor needed for resonant coupling.The distance between the internal resonators is d.The center-to-center distance (lateral misalignment) between the resonators is ρ.For the receiver, consisting of resonator 3 and the load loop, there may be angular misalignment θ.The power transmission can be affected by: (1) distance d; (2) lateral misalignment ρ; (3) angular misalignment θ; and (4) RL change.In this work, we focus our investigation on achieving high transmission efficiency under the change of geometric parameters (Cases 1-3).
b = 5 cm is the depth of the winding, rm = 27.5 cm is the mean radius, and N = 6 is the number of turns, respectively.

Figure 2 .
Figure 2. Dimension of multi-turn flat spiral for inductance calculation.

Figure 5
Figure 5 shows the schematic of the tunable matching network.The matching network consists of a variable inductor Ltune, switches SWS,k and SWP,k, and capacitor banks CS,k and CP,k (k = 1-8).To realize a fine search step, CS,k and CP,k each consists of a set of eight capacitors (10 pF, 20 pF, 40 pF, 50 pF, 100 pF, 200 pF, 400 pF and 500 pF) with 100 V ratings.By selecting a combination of switches SWS,k and SWP,k, Cvar1 and Cvar2 change from 10 pF to 1320 pF in 10 pF steps.Previous works used the capacitor matrix only[16][17][18].To realize the flexible matching network, we include a variable inductor as well as the capacitor matrix.

Figure 5 .
Figure 5. Schematic of tunable matching network; and circuit schematic of tunable inductor Ltune is shown in the inset.

Figure 6 .
Figure 6.Flow chart showing the impedance searching method.

Figure 7 .
Figure 7. Measured and simulated ZIN for distance and alignment changes.

Figure 9 .
Figure 9. Block diagram of the proposed WPT system.

Figure 10 .
Figure 10.Experimental setup for the proposed WPT system.
Figure11shows measured impedance ZIN,match at the input port of the matching network (Figure9).When the system examines the eight neighboring points around the current location, the value of impedance zopt in the fine search (Step 3) is recorded.The impedance is measured at three distances (d = 15 cm, 40 cm, and 80 cm) with ρ = 0 cm and θ = 0°.If only the fine-search step is used, the impedance searching starts from the center of the search matrix (this is called a one-step method).A three-step method uses system calibration (Step 1), coarse search (Step 2), and fine search (Step 3).When the searching process is performed, we observe that ZIN,match gradually converges to the center of the Smith chart.The ZIN,match is related to the reflection coefficient because ГIN,match = (ZIN,match -50)/(ZIN,match + 50).At d = 80 cm, |ГIN,match| improves from −0.57 dB to −25.6 dB using the one-step method.Using the three-step method, it improves from −14.9 dB to −25.6 dB.The final |ГIN,match| is −28.1 dB and −44.8 dB at d = 15 cm and 40 cm, respectively.These results indicate that the proposed tunable matching network achieves good impedance matching.

Figure 11 .
Figure 11.Measured trajectories of the input impedance for three distances.

Figure 12 .
Figure 12.Measured detector output and reflection coefficient versus time at three distances: d = 15 cm, 40 cm, and 80 cm.

Figure 13 .
Figure 13.Efficiency comparison with and without a matching network.

Figure 14 .
Figure 14.Trajectory of the measured input impedance for different misalignments from ρ = 0 cm to 40 cm for d = 15 cm.

Figure 15 .
Figure 15.Measured detector output and reflection coefficient versus time for different center-to-center misalignments from ρ = 0 cm to 40 cm, d = 15 cm.

Figure 18 .
Figure 18.Measured detector output and reflection coefficient versus time for different angular misalignments, d = 40 cm.

Figure 19 .
Figure 19.Efficiency comparison as a function of angular misalignment.
misalignment (deg.)simulated measured w/o matching measured with matching

Figure 20 .
Figure 20.A sequence of test cases under changes of three geometric parameters.

Figure 21 .
Figure 21.Measured results along the trajectory index: (a) input impedance and (b) efficiency.

Table 1 .
Measured electrical parameters of coils.