A Load-Independent Fission-Type Inductive Power Transfer System for 3D Reconfigurable IoT Array

In recent years, the number of electronic devices and appliances that support wireless charging has rapidly increased, owing to the research and development breakthroughs in inductive power transfer (IPT) technology. The existing wireless charging protocols, such as Qi and AirFuel, support only a few devices to be charged simultaneously. They cannot be directly utilized to construct a complex IPT network. This article introduces a reconfigurable three-dimensional (3D) IPT system based on series-parallel (SP) compensation for wirelessly powering multiple adjacent cubic Internet of Things (IoT) devices. These cubes can be configured into different 3D array combinations according to the application demands. After the 3D cube array is formed, a self-detection mechanism runs to turn off the uncoupled coils, such that to maximize the power transmission efficiency of the entire system. These cubes serve as not only IoT loads but also wireless power relays. In this way, the number of loads is not limited by the size of a single transmission coil. In addition, it maintains a constant drive voltage to each load device, regardless of the load variations in local or adjacent cubes. The constant voltage design eases local power management. A prototyped IPT cube array is manufactured and tested to validate the proposed idea. This fission-type IPT system makes a valuable exploration of the 3D IPT technology toward future ubiquitous and reconfigurable IoT arrays.


I. INTRODUCTION
Indctive power transfer (IPT) technology cuts the last cord between the power supply and electrical appliances. It offers many advantages such as safety, flexibility, and cleanliness. Thanks to these advanced features, IPT has become a promising power solution in some cutting-edge applications, such as implantable medical devices [1], intelligent home applications [2], [3], and electric vehicle [4]. Owing to the The associate editor coordinating the review of this manuscript and approving it for publication was Alon Kuperman . breakthroughs in IPT technology over the last decade, the number of devices that support wireless charging grows rapidly. However, most existing IPT systems only support one-to-one (single transmitter and single receiver) structure [5], as shown in Fig. 1(a). The specific connections limit the number of devices that can be charged simultaneously. The AirFuel Alliance wireless power transfer (WPT) system [6] can provide up to eight power receiver units (PRUs), as shown in Fig. 1(b). Nevertheless, it is still not comparable to the growing number of PRUs, especially for those in spatial distributed Internet of Things (IoT) applications. On the other hand, although there are some ambient energy harvesting technologies that can be utilized to power spaciously distributed IoT devices [7], [8]. The available power level is much lower than that the active IPT can provide. Therefore, it is necessary to develop three-dimensional (3D) IPT technology to meet the demand for flexibly configurable and more power-hungry IoT applications.
To satisfy the wireless charging demand of more load devices, the IPT systems supporting multiple PRUs have been studied in some literature [9], [10], [11], [12], [13]. By using the multi-load IPT system, the number of power inverters can be reduced; the power density increases [14]. In [9], an IPT system with two layers of primary coils is implemented for desktop peripherals, such as a mouse, keyboard, and loudspeaker at every position on the table. Other multiple PRUs IPT technologies were also observed from time to time, for example, the omnidirectional WPT charging bowl proposed in [10]. Multiple PRUs can be placed at any position in the bowl to be charged simultaneously without strict positional confinement. In [11], the concept of three roles in a multinode IPT system is proposed to balance the transferred power, transfer distance, and dynamic power variation requirements by changing the roles of different nodes.
As pointed out in [15], the cross-coupling among the multiple PRUs cannot be disappeared, leading to interference among different loads. On the other hand, from the load point of view, a constant voltage (CV) is preferred. Nevertheless, without a specific design, either the local or the neighbors' load variations might affect the local output voltage. To maintain the CV characteristic for the load, a dc/dc converter is used as a voltage regulator in [16] and [17]. In [18], multiple double-T resonant circuits are connected for multiple PRUs, such that to maintain CV outputs. Sun et al. [19] studied the frequency conditions of the multiple PRUs and realized the CV characteristics by making a series compensation to each coil.
Besides the number of receivers, the transfer distance and charging area are also of concern. These performances can be further increased and extended by introducing wireless power relays. In [16], a reconfigurable two-dimensional (2D) wireless charging system is implemented to extend the charging area using multi-hop and dynamic power routing. In [20], [21], [22], [23], [24], and [25], some domino-type inductive power relay systems are reported. In these designs, each relay consists of two coils, one for receiving the wireless energy from the upstream device and the other for transmitting it to the downstream device, respectively. They used appropriate circuit topologies and well-designed magnetic coil couplers to realize the CV characteristic in the seriesrelaying multi-load IPT systems.
In this paper, we proposed a novel fission-type IPT system based on series-parallel (SP) compensation as Fig. 1(c) shows. Every battery-free cubic cell in this system serves as a power relay with local CV output. The smart IPT cubes are reconfigurable to form any 3D array according to the application demand. They can detect the availability of the adjacent cells and disconnect the uncoupled coils, such that to enhance the energy transmission efficiency. The feasibility and performance of the proposed system have been evaluated in several experiments. The proposed fission-type IPT system has extended the wireless charging range of the neighboring IoT devices. The purpose is to explore the future wide-range IPT technology for ubiquitous IoT systems.
The rest of this paper is organized as follows. The system structure is introduced in Section II. The transfer capabilities analysis is presented in Section III. The feasibility and performance of the proposed system are evaluated in several experiments in In Section IV. Section V concludes this article.

II. SYSTEM CONFIGURATION
The proposed fission-type IPT system is formed by many cubic IoT units. Each cube is an independent IoT node. Various sensors, digital microprocessors, screens, and network interfaces are embedded inside each cube to make it intelligent. When a cube is connected to a neighboring unit, it receives power from the neighboring unit for local usage and also relays the electricity to its neighbors. Fig. 2 shows the conceptual IoT array with reconfigurable displays, which are powered by the proposed IPT technology. Before configuration, all cubes are off work and randomly placed as shown in Fig. 2(a). In order to satisfy different display demands, the IoT cubes can be configured in either VOLUME 11, 2023  one-dimensional (1D), 2D, or 3D arrays, as shown in Fig. 2 On each face of a cube, there is a printed circuit board (PCB) coil for power exchange from/to the neighboring cube. Such a 3D cubic design enables the wireless power relay in any spatial direction, which significantly extends the range and convenience of electric power delivery in the 3D space.

A. CUBE UNIT
In the fission-type reconfigurable IPT cube system, all cubic cells have an identical circuit topology, as shown in Fig. 3(a). To analyze the system, some basic definitions are given as follows. Each cube consists of six circuit branches, which represent the coils on the six faces of a cube and their corresponding compensating components. The six branches are connected in parallel with an equivalent ac load R ac . Owing to the PCB manufacturing consistency, the coil inductance and equivalent series resistance (ESR) of each coil are almost identical; therefore, they are denoted as the same notions for simplicity. Each branch consists of a coil inductor L, a series compensating capacitor C 1 , and a parallel compensating capacitor C 2 . Each coil has an ESR r. The operation frequency of the power source is f . The corresponding angular frequency is ω = 2πf . The mutual inductance between two adjacent cubes is denoted as M . The parameter M can also be regarded as a constant value because the relative position between two cubes is fixed by using some additional permanent magnetic locks. Therefore, the coupling coefficient k = M /L between adjacent cubes is also a constant number. The quality factor of a coil is defined as follows Six electronic switches S 1 -S 6 control whether or not the corresponding branches are connected to the system. If there is an adjacent cube detected, the system turns on the corresponding branch for bidirectional wireless power transfer; if not, it disconnects the very branch.
In the proposed circuit topology, the values of the compensating capacitors C 1 and C 2 are designed as follows such that the equivalent impedance of each cube exhibits a resistive value and the load voltage magnitude can maintain constant.

B. CONSTANT VOLTAGE OUTPUTS
The equivalent circuit model of a cubic unit is shown in Fig. 3(b). According to (2), the capacitive reactance C 1 balances some of the inductive reactance of the coil inductance L, making the remaining inductance equal to M . For a general case, suppose that there exist six identical cubes adjacent to the six faces of the studied cube, as illustrated in Fig. 3(b). The phasors of branch currents of the six adjacent cubes are denoted as I 1 to I 6 , respectively. The phasors of load ac voltages of the adjacent cubes are denoted as V 1 to V 6 , respectively. In the proposed fission-type IPT system, a face of a cube must be aligned and abut against one of the six faces of another cube, in order to exchange power between the two cubes. The cross-coupling between the non-abutting coils is weak; therefore, it is neglected in the circuit model analysis. Based on the nodal analysis, the governing equation of a studied cube is formulated as follows where N is the number of effective adjacent cubes. The quality factor Q is a constant much larger than one. Its typical value is often greater than 100 when the operation frequency is 6.78 MHz. The coupling coefficient k is less than one. With these prerequisites, (4) can be simplified into Further, when the load resistance R ac is much larger than the ESR r, the ac voltage across the equivalent ac resistor can be further simplified into the following equation From (6), V ac the voltage phasor across R ac is the average of those ac voltages in its adjacent cubes. V ac is independent of the equivalent ac load under the ideal conditions. Given that V ac describes the voltage of an arbitrary cube in the 3D-IPT network, a stable solution is attained when all the cell voltages are equal. Therefore, the CV feature is realized against the load variations and fluctuations in the distributed cubes.

C. ZERO-PHASE-ANGLE (ZPA) ANALYSIS
When the reactive parts in a unit cube is balanced, it exhibits a pure resistive impedance to its adjacent cube. A ZPA is achieved between the branch voltage and current. When all cubes exhibit a resistive outlook, the drive inverter gives the minimum switching loss. Therefore, the ZPA analysis is of important to optimize the power transmission efficiency. In this proposed design, ZPA can be realized by turning off all the uncoupled branches of a cube. Fig. 4(a) shows how the switches work to realize the ZPA characteristics. When two cubes are connected, their coupled branches should be switched on, While other uncoupled branches are turned off. Given that the uncoupled branches are disconnected or open-circuited, the simplified circuit is shown in Fig. 4(b). In the corresponding T-model, which is shown in Fig. 4(c), if the ESR r is small enough to be neglected and the resonant conditions (2) and (3) are satisfied, the branch input impedance is just  The ZPA characteristic is realized under such an ideal zero ESR condition. For a nonzero ESR lossy case, the equivalent input impedance of a single branch is derived as follows The features of equivalent impedance in a non-ideal case that cannot be simply observed using the closed-form expression; therefore, it requires numerical simulation. Fig. 5 compares the branch input impedance at different coupling coefficients, when r, f , and L are set to 0.407 , 6.78 MHz, and 2.05 µH, respectively. As it can be observed from the figure, with a relatively large coupling coefficient, i.e., when k ≥ 0.3, the real part of the branch input impedance is almost  linearly proportional to R ac , while the imaginary part is almost negligible. As long as the coupling coefficient between two adjacent coils in this design is about 0.77, larger than 0.3, the branch input impedance is very close to a pure resistive one, i.e., the ZPA characteristic is achieved.

III. CONTROL METHOD
As mentioned in Sub-section II-C, the ZPA characteristic is satisfied only when the uncoupled branches are disconnected.
The switches should be turned on or off automatically according to the existence of the corresponding adjacent cubes. An embedded microcontroller (MCU) is utilized to carry out this neighborhood detection. A conventional mobile embedded system is usually powered by batteries [16].

A. TRANSIENT STAGE
In the proposed IPT network, an ac/dc bridge rectifier and the following voltage regulator are utilized in each cube for supplying the digital voltage to the embedded MCU circuit; therefore, no additional battery is needed. Starting from the default state, in which all the switches of the receiver cube's branches are closed, i.e., all branches are connected at the beginning, as shown in Fig. 6(a). If there is no adjacent cube detected, the input impedance of a branch is expressed as follows Therefore, the impedance of an uncoupled branch is inductive. Its inductance is 2M , as shown in Fig. 6(b). No matter a branch is connected or disconnected, it will not short the load; therefore, such a topology guarantees the safe operation of the circuit. The cube voltage V ac rises when one of the branches is coupled to the existing network. The ac voltage goes across the bridge rectifier and charges the filter capacitor C rec , such that the MCU can be started up when V rec rises above a specific voltage threshold.
In real applications, the equivalent ac load R ac is embodied with a rectifier, a filter capacitor, a dc/dc converter, and a dc load, as illustrated in Fig. 6(b). The ac voltage is rectified to dc voltage through the full bridge rectifier and capacitors. At the relatively light load condition, the rectified dc voltage V rec = π |V ac | /4. Fig. 7 shows the transient waveform showing the accumulating charging process of C rec . Within 0.2 milliseconds, the rectified voltage V rec reaches the startup threshold of the dc/dc chip. This embedded MCU is then activated after a constant digital voltage is established and stabilized.

B. DETECTION METHOD
After activation, all cubes keep detecting their branches from time to time to perceive the approaching and detaching events in their neighborhood. If a new inactive cube approaches an activated cube, its secondary-side impedance, as illustrated in Fig. 6(a), can be derived as follows The reflected impedance looking from the primary-side activated cube is expressed as follows The input impedance of the activated cube is obtained as follows Eq. (12) demonstrates that when an inactivated cube joins the system, it first introduces an inductive impedance to  the coupled branch of its relaying neighbor cube. After the embedded MCU is started up, it detects the occupancy of its neighboring slots and disconnects the uncoupled branches, such that its reflected impedance can attain a pure resistive value. Fig. 8 shows the voltage waveform and spectrum of a branch coil within V coil1 to V coil6 , whose corresponding probes are shown in Fig. 6(a). When a cube is coupled to one of the branches, the corresponding coil voltage is shown in Fig. 8(a). if no cube approaches, the coil voltage magnitude is much smaller, as shown in Fig. 8(b). From the spectrum shown in Fig. 8(c) and (d), we can see that the dc components with or without a neighboring cube are almost the same. The major difference lies in the magnitudes of their fundamental harmonic. Given this observation, the different information associated with the coil voltage can be used to identify the occupancy of the adjacent face slots of a cube. An ac peak detection circuit is designed for such an identification task. Fig. 9 shows the auxiliary circuit and output waveform. In the circuit, the dc component is firstly filtered by the capacitor C b . Following that, the high-order harmonics are filtered by a low-pass filter and voltage divider, which are formed by R 1 , R 2 , and C p1 . Finally, the pure ac voltage is converted into dc by a peak detector, which is formed by D 1 , C p2 , and R 3 . The output dc voltage is fed to an analog-to-digital converter (ADC) for quantification and sent to the MCU for generating the corresponding switch control commands.
The cubes first run an initialization program and finalize the configuration according to a specific spatial arrangement. The detection and control processes are summarized as follows. When the MCU is activated, it carries out a neighborhood detection function to identify the coupled or uncoupled branches. The MCU turns off the uncoupled branches. After an initial delay of three seconds, the reconfigurable process is done.

IV. EXPERIMENT
An experimental fission-type reconfigurable 3D-IPT system with four battery-free cubes is fabricated and tested to evaluate the feasibility and performance of the proposed design. The experimental setup is shown in Fig. 10(a). The prototyped cubes are hollow. All digital components in a conceptual IoT application can be allocated inside the cube. The transmitting coils only occupy the frame surface. The four cubes are configured into one-dimensional [ Fig. 10(b)], twodimensional [ Fig. 10(c)], and three-dimensional [ Fig. 10(d)] layouts, respectively, for testing the IPT performance in different configurations. The system is powered by a dc power source (WPS3010B, Wanptek Inc.) with a 30 V supply voltage. Each face of a cube has a four-turn PCB coil to form a magnetic coupler with its adjacent cube. A DSP (digital signal processor) evaluation board (LAUNCHXL using TMS320F28377S, Texas Instrument Inc.) is used to generate a high-resolution pulse width modulation (HRPWM) for driving a full bridge inverter. The full bridge inverter is built with four GaN (gallium nitride) switches (GS61008T, GaN System Inc.) and operates at 6.78 MHz. An MCU (ESP32, Espressif Inc.) is programmed to detect and control the active switches in each cube. The MCU inherently includes Bluetooth and Wi-Fi modules, which enables a convenient switch between the reconfigurable state and the stable energy transmission state. The switches are implemented with mechanical relays (TX2SA-L2, Panasonic Inc.). The ratio of the output power to the input power of the selected mechanical relays at isolation and contact conditions at 6.78 MHz are 65 dB  and 0.05 dB, respectively, enabling each WPT branch to be completely turned on or off. The high-frequency ac power is rectified to dc power by a rectifier bridge consisting of four discrete diodes (DFLS240L, Diodes Inc.). The output of the diode bridge rectifier is connected to a close-loop regulated buck converter (TPS5430, Texas Instruments Inc.) for providing a 5 V regulated output voltage to the MCU and the extra load. In this study, the extra load in each cube is a 5 resistor. The rated output power of each cube is set to 5 W. The coupling coefficient of any two nonadjacent coils is measured to be less than or equal to 0.068, which is more than ten times smaller than that of two adjacent coils (0.77 in this design). Therefore, the cross-coupling among the nonadjacent coils can be neglected. The parameters of the other passive components are listed in Table 1. Moreover, in the transmitter drive, as the full-bridge inverter gives a square-wave voltage output, its compensating capacitor C 2 , which is originally placed in parallel with the inverter output and transmitting coil, should be moved to either a cube in the system, such that to avoid the transient short-circuit condition at the switching instants.

A. VALIDATION OF BASIC 3D-IPT FUNCTION
The detection algorithm is implemented by the MCU in each cube. The components' parameters in the ac-peak-detector circuit are listed in Table. 2. Fig. 11 illustrates the transient processes of the three configurations. The rectified voltage V rec values in the four cubes in the three cases reach their maximum value within 24 ms. It means that the MCUs can properly configure and activate the system within 24 ms. At steady state, the rectified voltages in all cubes and all cases are close to the 30 V input voltage. We can also find that the cubes that are closer to the transmitting coil can more quickly receive sufficient energy and attain a stable state. After a short delay, the coil voltages V coil in each cube reach the steady state, whose voltages are illustrated in Fig. 12(a). The ac component peak of a coil voltage is obtained after being fed into a peak-detector circuit. The output voltage is within the range of ADC input; therefore, it can be quantified by the ADC in the activated MCU. Fig. 12(b) shows the output voltage V det after the peak detection circuit. It is not a constant dc voltage. We use the average value of 60 samples to identify the difference with or without an adjacent cube. If there exists an adjacent cube, the average detection voltage V det of the corresponding branch is about 1.2 V. If there has no adjacent cube, V det is very low, around 150 mV. In this way, the MCUs successfully find out the uncoupled branches and turn off their corresponding switches, such that the transmission efficiency can be optimized. Fig. 13 shows the steady-state waveform of the ac voltage in each cube. The ac voltage of a cube is out of phase to that of its neighboring coupled cube. In case I, ac voltages in cubes 1 and 3 have the same phase, while those in cubes 2 and 4 are out of phase. In case II, ac voltage in cubes 1 and 4 have the same phase, while those in cubes 2 and 3 are out of phase. In case III, ac voltage in cubes 2, 3, and 4 have the same phase since they are all adjacent to cube 1. The ac voltage magnitudes in all cubes and cases are very close to a constant value.

B. RELUCTANCE AGAINST LOAD VARIATION
The proposed 3D-IPT circuit topology has an inherent strong rejection to load variation. The output dc voltage is relatively constant under different or fluctuating load resistance. In this experiment, we change R 4 the load resistor of cube 4 from 15 (light load) to 5 (heavy load) in a sudden, while VOLUME 11, 2023  keeping others load unchanged, i.e., R 1 , R 2 , and R 3 = 5 . Fig. 14 shows the rectified dc voltages in the three cases when experiencing sudden load variations. In case I, this load variation causes a slight voltage fluctuation in cube 4, about 4 V out of 30 V, for 30 ms. Such a load variation also influences the rectified voltage in cubes 1, 2, and 3 through the coupling network. The voltage of the cube, which is closer to the transmitting coil, is less affected. After 30 ms, this voltage fluctuation diminishes. In cases II and III, the voltage fluctuation caused by the load change in cube 4 has very little influence on cubes 2 and 3. In general, this system shows a strong load variation rejection ability. Such a unique feature eases the design of the following dc/dc regulator stage.

C. EFFICIENCY AND LOSS ANALYSIS
In this work, the output power is consumed by the external load in each cube. i.e. 5 resistor. Given that the dc/dc regulator supplies a constant 5 V output voltage, the output power is fixed at 5 W in each cube. Therefore, the total output power of the four cubes is 20 W. The power utilization efficiency η is calculated as the ratio of P out , the total dc power of the four extra loads in four cubes after the regulator, to P in , the dc input power before the transmitter inverter η = P out P in = P out P loss + P control + P out (13) where P loss and P control are the power losses of the power electronics and the microcontroller. The loss power contains conduction and switching losses in the inverter, rectifier, IPT  coil links, and dc/dc regulators. The efficiency of IPT coil links is up to 90%-95%. The end-to-end power efficiency of the system under different configurations is listed in Table 3.
The efficiency can gain a 2%-3% improvement after detecting the neighborhood availability and optimizing the network by opening the uncoupled branches. The voltage and current waveform of the inverter output are illustrated in Fig. 15. The current I inv lags behind the voltage V inv a little bit. Therefore, the zero-voltage switching (ZVS) characteristic can be achieved; the MOSFET switching loss is minimized.
In general, the experimental results prove the feasibility of the proposed system. From the comparison with other stateof-the-art 3D multi-load WPT systems, which is listed in Table 4, the performance of the proposed system is commensurate with other cutting-edge designs. Moreover, this fissiontype wireless 3D-IPT design processes a configurable feature, which is promising to support the easy configuration of IoT applications in the future smart home or smart city scenarios.

V. CONCLUSION
In this paper, a fission-type 3D-IPT (three-dimensional inductive power transfer) system was proposed to meet the multiple-device power supply requirements in future ubiquitous and re-configurable IoT applications. A series-parallel (SP) compensation network was utilized to realize the zero phase angle (ZPA) and constant voltage (CV) characteristics in each wireless coupled cube unit. The cubes not only receive power from their upstream cubes but also serve as power relays to wirelessly transfer power to their downstream cubes. The CV property implies a strong load-variation rejection in all cubes. It eases the design of the following voltage regulation stage and the digital load. A neighborhood detection circuit detects the coupling condition of all branches in a cube and turns off those uncoupled branches. Such actions realize the zero-phase-angle (ZPA) property and further improve power utilization efficiency. To validate the proposed idea, a 20 W 3D-IPT experimental system was prototyped and carefully evaluated. The prototype consists of one transmitter and four independent cubes as relays or end-users in any spatial arrangement. The experimental results showed that the coil-link efficiency is above 90%, while the end-to-end transmission efficiency of the whole system ranges from 60% to 70%. The proposed 3D-IPT system provided a new idea and engineering practice for effectively building the 3D reconfigurable IoT array, which is useful for future ubiquitous IoT applications. A very close neighboring position is required between two cubes in this design, such that the compensation condition can be ensured. This restriction of neighboring distance might be loosened in future new designs.