Auxiliary Power Supplies for High-Power Converter Submodules: State of the Art and Future Prospects

Recent developments in medium-voltage (MV) silicon and silicon carbide (SiC) power semiconductor devices are challenging state-of-the-art converter and auxiliary power supply (APS) designs. The APS is an important converter component, which energizes the gate-drive units and, therefore, has an influence on the overall reliability and efficiency of the converter system. There has, however, been comparably little research on how the APS of high-power converter submodules can be realized, in particular, for high-voltage applications. New, or improved, solutions may build on state-of-the-art topologies in the near future, but utilize MV SiC technology in the APS circuit itself to enable improved efficiency, reliability, simplicity, and compactness. Externally-fed APS concepts could provide several further advantages. Their various benefits on converter and system level may enable them to be a competitive solution for future APS concepts. Especially, light-based power supply systems are considered most useful since they offer extreme voltage isolation capability and immunity to electromagnetic interference. This article presents a review of the wide range of solutions for APSs, possible implementation options, and the most important design considerations. The different solutions are evaluated in a qualitative fashion, providing an overview of available APS concepts with regard to the requirements for high-power converter applications.


I. INTRODUCTION
I N ALMOST all high-voltage (HV) applications auxiliary power is required to energize gate-drive units (GDUs), control electronics, sensors, protection circuits, and further electronic equipment. The reference point of this voltage is often not on ground potential but on a high or even switched voltage. Depending on the application, i.e., the power and voltages involved, supplying this auxiliary power requires special attention and careful design. This may be even more important in future applications of HV converters. For example, it is anticipated that massive amounts of renewable energy will be transferred across continents in interconnected HV direct current (HVdc) systems [1]- [3], or even ultra-HVdc systems. However, in a meshed system it is important to ensure robustness and failure management. In case of a fault somewhere in the grid, the remaining system must be kept operational. Protection of converter equipment and the availability of a reliable auxiliary power supply (APS) system is crucial. Consequently, this requires robust and reliable solutions for the APSs in such systems.
In the future, it is likely that some of the HV converters will be equipped with newly developed power devices based on silicon carbide (SiC) [4]. Particularly, medium-voltage (MV) SiC devices with blocking voltages of 10-15 kV have gained significant interest in the recent years, and are increasingly available [5]- [7]. SiC bipolar devices are foreseen to be suitable for blocking voltages up to 50 kV [8]. From an overall converter design perspective, such a development could be very advantageous because of a reduced converter volume, weight, and complexity enabled by reducing the number of switches and associated auxiliary electronics [9].
Furthermore, the high blocking voltage could allow the use of two-level converter topologies in MV applications with higher voltage levels than what is possible with Si devices [7], [10]. However, there are still many challenges for the application of MV SiC devices, e.g., device packaging, gate drivers, and APS design. The main requirements for successful implementation of MV SiC devices are ultralow coupling capacitance and electromagnetic interference (EMI) immunity [7]. In addition, increasing the device voltage also increases the insulation requirements for the APS. For example, with a switch voltage rating of 50 kV, the nominal operating voltage of the device might be up to 30 kV. In many cases this also implies that the input voltage to the APS would be 30 kVDC. Powering the GDUs of such a converter would be far more challenging than the case with the Si IGBT converters of today. For these applications, APS solutions existing in the field are not feasible as they will inevitably become too complicated [11]- [13].
Therefore, in this article, APS requirements for high-power converter applications are derived and summarized in Section II. Section III presents a classification of APS concepts and important design considerations. A review and discussion of APS concepts is provided in Section IV; followed by Section V, which serves to evaluate relevant concepts. Based on the existing This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ solutions in the literature, the rest of this article focuses on a discussion of designs and approaches that can lead to improved future in APS systems. Finally, Section VI concludes this article.

II. APPLICATIONS AND REQUIREMENTS
Various parts of high-power converters should have isolated APSs. Some of the more common examples of such converter parts include GDUs in HVDC converters, static compensators (STATCOMs), MV motor drives, and converters in electric trains [15]. Possible HV applications other than high-power converters include sensors on high potential, e.g., monitoring of HV power transformers, transmission lines, and switchgear [16]. Real-time monitoring can facilitate advanced and prescriptive maintenance of power system components. It is to be expected that smart metering and monitoring on high potential will become more important in HVDC grid applications in order to to ensure system reliability and stability [17]. Low-cost, low-power, and high-efficient electronic sensor nodes for harsh industrial applications are presented in [18].
Several APS requirements for high-power converter applications are summarized and described in Table I. Sufficiently high isolation voltage and output power are core requirements. It should be noted that some of the listed requirements are mutually dependent. Therefore, it is necessary to find an optimum trade-off between them depending on the specific application, which defines the functional requirements of the respective APS solution. Fig. 1 shows how the first seven requirements can be derived. The necessary isolation voltage level, V iso , is either defined by the system voltage or by the submodule voltage, V sm . Depending on the source of the auxiliary supply voltage (external / internal) it is V iso,ext or V iso,in . This is discussed in detail in Section III. The EMI immunity is affected by both the dv/dt and the switching frequency f sw . The dv/dt is in turn affected by the type of semiconductor, its drain-source voltage V ds , switching current I d , and gate-drive strategy. Other requirements like reliability, efficiency, compactness, and a fast start-up time are highly application-specific.
Several factors affect the total required power of an APS system P APS , as shown in Fig. 1 and given as follows: An estimation of the power requirements of various applications is shown in Fig. 2. As regards sensors, very little power is needed for communication electronics P com and control P ctrl . Concerning the APS for submodules, however, the most important contribution is the power required to supply the GDUs P GDU . A powered GDU is fundamentally needed to turn a  [14], [26], and [27]). (a) Internal APS concepts. (b) External APS concepts. power device ON and OFF, respectively. The power demand of each gate driver depends on the used power semiconductors and their switching frequency [10]. Depending on the application and employed converter topology the number of GDUs N GDU , which have to be supplied by one APS, may vary. In the case of the modular multilevel converter (MMC), which is shown in Fig. 3(a), N GDU depends on the submodule topology, e.g., two for half-bridge, four for full-bridge, or seven for semifull-bridge submodules [19]. Today, the total required APS output power P APS for MMC submodules ranges from a few watts to some tens of watts depending on the semiconductor device [15].
Thyristors and integrated gate commutated thyristors (IGCTs) are current controlled devices with a high GDU power consumption. Novel SiC MOSFETs, on the other hand, are voltagecontrolled devices and require very little gate charge. The small gate capacitance of SiC devices promises reduced power consumption in the gate-drive circuit when compared to established technology based on silicon transistors [20]. In [21], it is founded that a comparably low gate-drive power consumption is necessary for MMCs with silicon IGBTs or SiC power devices, mainly because of the low switching frequency of an MMC. For switching frequencies below 1 kHz, the power consumption is well below 1 W. SiC JFET-based supercascodes [22]- [25] feature especially low gate-drive power consumption because a single low-voltage (LV) MOSFET at the bottom of the series chain controls the entire series switch. The input capacitance of this LV MOSFET is low, and thus, the gate charge requirements of the whole supercascode is low compared to paralleled MV MOSFETs or even series-connected MV MOSFET modules.

A. Sources of Auxiliary Supply Voltage
There are two ways of supplying auxiliary power. One approach is to supply this power from the input voltage of the power electronic system. In case of a converter, this might be the submodule capacitor on a floating potential, hereafter called internal APS. The second option is to take the power from a central power source with the same ground reference as that of the entire system, hereafter called external APS. Fig. 3 shows the two APS sources based on the example of an MMC half-bridge submodule. The challenges and benefits of both methods are summarized in Table II.
It can be concluded that several advantages are achievable by having control over the GDUs even when the submodule capacitor is completely discharged, e.g., improved robustness, because most power semiconductors (IGBTs and IGCTs) are not guaranteed to block their rated voltage with unpowered GDUs [14]. External power to the GDUs could also enable the use of normally ON semiconductor devices in the high-power sector. SiC JFETs, for example, even though offering superior robustness [28], switching performance, and low specific onstate resistance [29], have been less favourable choices for power switches in the past, since a loss of GDU power results in a short circuit of the submodule capacitor.

B. Implementation
In this section, possible options for internal and external APS implementations are discussed on the basis of Fig. 4, where isolation barriers are indicated in red and potential levels are depicted with different shades of gray (the darker the higher the potential). In addition, the challenges in designing low-power HV step-down dc-dc converters are pointed out and possible realizations are presented.
Today, the preferred APS source for HV converters is usually the submodule capacitor V sm . The reason for this choice is that the APS follows the electric potential of the submodule capacitor, such that a galvanic isolation of several tens or hundreds of kilovolt is not necessary, and the isolation voltage is merely a few kilovolt V iso,int .
The power conversion can be achieved in two stages [15], as shown in Fig. 4(a). In the first stage, a nonisolated low-power  conversion is realized to perform step-down and regulation functions [30]. In high-power systems, such as HVdc converters, this stage is challenging because the input voltage is in the range of several kilovolts, and thus, a voltage conversion with a high stepdown ratio is needed for supplying the auxiliary electronics [31]. Since their power consumption is usually very small compared to the rated power of such a system, low-current MV power transistors are required. Such switch devices, however, are only used in special applications, and therefore, have the disadvantage of high cost and low availability. Possible realizations of high conversion ratios by means of series-connection are given in Table III. For the second stage, galvanically isolated power can be supplied by a conventional dc-dc converter on basis of flyback, forward or resonant topologies [10], or an induction-based APS concepts. The design of the first stage defines the required input voltage range of the second stage [32]. The isolation stage is a key feature of the APS, and can be considered as the major difficulty associated with designing an APS for MV SiC devices, in particular, in MV converter systems. The challenges of implementing such systems are addressed in Section V-A and in [33]- [35]. Internal APSs of HV MMCs may, however, have lower requirements for interwinding capacitance and isolation strength, depending on the submodule capacitor voltage as described in [13] and [34].
A challenge for the second stage is the high peak voltage stress and switching speed of novel MV SiC semiconductors [7], [33], [36]. A dv/dt of more than 100 kV/µs might occur between primary and secondary side of the transformer, and therefore, across the interwinding coupling capacitance C iso,int . This may lead to common mode (CM) currents through the isolation of the transformer, which can possibly cause EMI problems or circuit malfunction. This is a major concern, and in the worst case not only malfunction may result, but even destruction by excessive capacitive currents. Examples of destruction include burnt circuit board traces, molten transformer windings, destroyed input stages of logic or analog circuits, etc. The secondary effects of the described destruction may be even worse as the complete converter operation may be compromised.
The implementation of an external APS is shown in Fig. 4(b). Here, the auxiliary power is supplied from a ground-referenced source V ext . The isolation required to bridge the potential difference between ground and system voltage might be much higher than the submodule voltage, as in the case of an internal APS V iso,ext V iso,int . This is not easily realizable depending on the application.
For both implementations option it should be noted that any device, which is not connected to the ground of the APS, must have its own isolated supply channel V iso,ch . Thus, multiple galvanically separated outputs are required to supply the individual GDUs in a submodule. Two such outputs are indicated in illustration Fig. 4, exemplifying an APS implementation for a half-bridge submodule.

IV. REVIEW AND DISCUSSION OF APS CONCEPTS
An overview of different APS concepts is shown in Fig. 5. These concepts are categorized in two groups according to their source of auxiliary power, i.e., internal and external APS concepts. The internal concepts deal primarily with the conversion of a HV input into a LV output, whereas the external concepts deal with the distribution of voltages assuming that a power source is available. The APS concepts can be further classified, based on their underlying energy transfer mechanism, into "conduction-based," "induction-based," and "radiation-based" concepts.

A. Internal APS Concepts
The following internal APS concepts are presented and discussed one by one: 1) voltages dividers, 2) snubber circuits, 3) tapped-inductor buck (TI-buck) converters, 4) monolithic isolated converters, 5) input-series output-parallel (ISOP) converters, and 6) multicell dc-dc converters. Multilevel converters, such as diode-clamped and flying capacitor converters, are not considered to be feasible alternatives due to their nonmodular structures making it increasingly difficult to scale for higher conversion ratios, the need for balancing resistors, and more complex control [37]- [39].
1) Voltage Dividers: For applications with comparably low input voltage and low auxiliary power requirements, the power can be supplied by voltage dividers. If a simple resistive divider is used, a subsequent network is required to stabilize the output voltage depending on input voltage and output power variations. This subsequent step could be a zener diode circuit for dissipating surplus power. Another possible realization would be to use a MOSFET operating in the active region in series with the LV resistor R lv . Several other options are also possible.
Voltage dividers are simple and reliable solutions, but suffer from a relatively low efficiency [14], [15]. When several tens to hundreds of watts of auxiliary power are required (e.g., IGCT-based submodules), they would be unacceptably lossy. Furthermore, the efficiency varies inversely with the input voltage. Thus, the low efficiency and associated power loss might not be acceptable for submodules, employing MV semiconductor switches. Another potential drawback is that any voltage divider circuit would need a certain input voltage before the circuit becomes operational. This may cause a considerable start-up time.
To increase the input voltage range, while still having an acceptable efficiency, the concept of active voltage dividers, presented in [12] and shown in Fig. 6, can be employed. In the first step, the voltage is reduced to an intermediate level, and then used as an input to an isolated two-output dc-dc converter, i.e., the second stage in Fig. 4(a). The voltage ratio is decided by the ratio of the upper resistor R hv and the equivalent resistance of the lower branch. A prototype with 7-kV input voltage has been demonstrated in [12] and [13]. It employs a 4.5-kV Si MOSFET in the lower arm voltage regulator and is designed as APS for a 10-kV SiC MOSFET as the main switch.
The simple resistive voltage divider comprises only passive components, and is therefore likely to have high EMI resilience. However, for the active version, the final level of resiliency also depends on the converter circuits and has to be proven experimentally for the particular application. Regarding compactness, the major driver of volume is the power dissipation, which is potentially unacceptably large for a passive voltage divider. For the active voltage divider, the transformer of the second stage might be voluminous depending on the isolation voltage range.
2) Snubber-Based Power Tapping: Depending on the choice of main power semiconductor switches of the converter and the design of the main commutation circuit, there may be a need for a snubber circuit to protect the switches from voltage spikes at turn-OFF or excessive current rates at turn-ON. Such snubbers will charge and discharge energy during the switching transitions of the main power switches. Typically, a large portion of the energy that is charged during one switching transition will be dissipated in a resistor during the next transition. An example of such a circuit is a simple RC snubber, which can be used to reduce the peak voltage at turn-OFF and damp the following oscillations. In the RC snubber, energy is dissipated in the resistor at both turn-ON and turn-OFF. One alternative to tap energy from the snubber circuit would be to use some of the energy that is dissipated at turn-OFF for the power supply of the gate driver [40]. Such a circuit is shown in Fig. 7.
The main drawback of snubber-based power tapping is that no power is obtained if no switchings are performed, for instance because of tripping due to protection. Problems may also occur due to changes in switching frequency or switching speed, which affects the power transfer in the snubber. This may call for large storage capacitors in the APS. In many cases, an additional alternative powering method must be used if the main power switches should stay reliably in the OFF state for extended durations. In such cases, a resistive voltage divider, or zener diode regulator, R & D z in Fig. 7, is one option. The power rating of this unit can be very small because no energy is used for charging or discharging of the gate of the main switch. This also means that the start-up time depends on this voltage divider.
Snubber-based power tapping can be used up to extremely HV levels, because the elements that are subjected to the HV are included in the snubber circuit itself. Due to the simplicity of the circuit, the solution can be considered as very reliable with excellent EMI resilience. In case, a snubber is anyway needed for the operation of the main circuit, the power tapping of the snubber will not add any significant volume to the circuit. Therefore, such an APS design can be considered very compact. Moreover, no additional losses are added by the power tapping in this case, which means that an excellent efficiency can be claimed.
3) TI-Buck Converter: The TI-buck converter topology, as shown in Fig. 8, has several advantages over the regular buck converter. Notably, the efficiency and switch utilization at HV conversion ratios can be significantly improved, even at high power levels. Modeer et al. [14] presented an implementation using a novel autonomous HV valve, in which the voltage sharing and gate-drive control is realized. The advantages of the proposed concept are several. First, no isolated signal is needed between HV and LV sides, eliminating the need for pulse transformers or optoisolators [30]. The control functionality is fully Fig. 7. Snubber-based power tapping. located on the LV side. Also, soft switching with zero-voltage switching (ZVS) turn-ON and large capacitive snubbers at turn-OFF is enabled, which reduces both switching losses and EMI in the HV valve. It also greatly simplifies the series-connection of semiconductors needed to implement the HV valve. Simulations and measurements indicate high efficiency (80%-90%) [14], allowing passive cooling.
The TI-buck converter has the potential to ensure high levels of reliability. Since there are no signal isolation components, these points of failure are eliminated. The reliability can be further improved by using redundant devices in the HV valve. Moreover, as the circuit utilizes resonant transitions, the resonant modes are well defined, and therefore, it is likely that the circuit has a good noise resiliency. This is yet to be proven experimentally.

4) Monolithic Isolated Converters:
Flyback converters are a viable option for MMC submodule APSs [15], [41]- [43], even for high power levels. One of their main drawbacks is the need for a large number of series-connected switching components for HV applications due to their monolithic (nonmodular) configuration [44]. Novel SiC semiconductors with high blocking voltage could solve this issue, although the availability of MV low-power devices still remains an issue.
In the TI-buck converter, discussed in the previous section, the HV switch can be realized by autonomous series-connected switching cells, which are powered by the reverse current through the HV switch prior to the turn-ON. A similar solution would, however, not work for the flyback converter due to the differences in their operating principles. This makes the realization of the HV switch in the flyback converter more challenging. In [11], a flyback converter with the main switch realized as a series-connection of several switches, as shown in Fig. 9, has been presented. This concept, also discussed in Table III, allows an extension of the input voltage range along with the possibility to achieve redundancy for the main switch. However, if the main switch is realized by a single MV switching device, no redundancy is provided and a failure of this would cause a total shutdown. Thus, good overall reliability of a flyback converter can be achieved by design.
In the case of series-connected switches, those need separate, isolated GDUs, which have been implemented using pulse transformers in [11]. This will have a significant adverse impact on the compactness of the circuit. Accordingly, since there is no easy way to achieve a HV switch, the input voltage range is limited. However, if MV low-current SiC devices were available, very high input voltages would be possible.
Since the circuit relies on switch-mode transitions, it is important that the gate control of the switching device is very robust, such that clean switching trajectories are obtained. If this cannot be achieved, it may cause excessive switching losses and EMI problems. In contrast to the TI-buck converter, which allows ZVS turn-ON and larger snubber capacitors, a normal flyback converter may therefore have less EMI immunity. This has still to be proven experimentally. Flyback circuits are, however, known to be efficient, and there is no reason to believe that it is different in this case. 5) ISOP Converters: ISOP flyback converters are an attempt to solve the difficulty of high input voltage and low output power by connecting the inputs of several isolated dc-dc converters in series and parallel-connecting the outputs [42], [45]. This concept is shown in Fig. 10 and also described in Table III. Efficiency, simplicity, and cost-effectiveness of each individual stage are clear advantages. Furthermore, the voltage balancing among the converters is considerably simpler than the balancing of individual switches of a monolithic implementation as the timing requirements are lower [46].
However, this solution brings its own challenges: input capacitor voltage balancing, isolation among the flyback stages, start-up synchronization, as well as overvoltage and short-circuit protection. Control methods have been published to accomplish an equal sharing of the input voltage [31]. Start-up of an implementation with master and slave converters requires all stages to be powered beforehand; otherwise input voltage balance is lost.
The ISOP concept enables redundancy, both in voltage and power, without adding excessive cost, which enables a very high reliability. However, the main drawback is the need for a number of transformers with high isolation voltage, which may contribute to a comparably bulky system. The ISOP is based on multiple monolithic isolated converters, and is therefore likely to have similar EMI resilience.

6) MultiCell DC-DC Converters:
Multicell dc-dc converters, Fig. 11, remove the need for HV isolation transformers by cascading nonisolated converter stages, which is given in Table III. A high step-up dc-dc converter based on this approach is presented in [47] by stacking basic boost converter elements. A buck-boost bidirectional topology is used for each module in [31] and [48], while a cascade of step-down stages is proposed in [49] and [50]. A result of the cascaded structure is that the current magnitude is not the same for all stages, but rather increases linearly along the cascade, i.e., the conditions for a stage near the top of the cascade are quite different from those of a stage further downstream. A 200-W converter has been presented in [50], which confirms that a high power capability is, nevertheless, possible.
The converter features LV stresses of the employed components and a comparably simple control scheme [31]. Each module can operate without an external power supply or external gate signals (on-board power supply). The basic concept can be modified with coupled inductors in order to isolate the converter output, if required. However, a disadvantage resulting from the coupling is that the control of all switches has to be synchronized [31].
The input voltage range is similar to the TI-buck, monolithic isolated, and ISOP converter concepts. For HVs, MV SiC switches would be necessary. It is likely that this circuit has a good EMI resilience, because the internal control circuits are autonomous, and thus, do not rely on external control signals. However, due to the fact that many components are required and these components have different stresses, it is harder to state an equally high reliability compared to the previous concepts.

B. External APS Concepts
The following external APS concepts are presented and discussed one by one: 1) bootstrap circuits, 2) isolation transformers, 3) inductive power transfer (IPT), 4) radio frequency (RF), and 5) optical links.
1) Bootstrap Circuits: So-called bootstrap circuits have been used in many applications to facilitate power supply in circuits where two or more ground levels are used. A typical application is power supply of high-side GDUs [51], [52], where an isolated power supply (for instance by means of a transformer) would have to be used otherwise. The bootstrap circuit has also been proposed to be used for power transfer to GDUs of diode-clamped multilevel converters [53].
The basic operation of powering high-side drivers with bootstrap circuits is simple. When the low-side switch is in the ON state, the filter (or bootstrap) capacitor of the high-side driver can be charged via a HV bootstrap diode in a charge-pump fashion. If the maximum voltage of the bootstrap diodes is not sufficient for the specific capacitor voltage used in the submodule, each bootstrap diode has to be realized as a series-connection of two or several diodes. This redundancy may, however, significantly enhance the reliability, and if combined with several parallelconnected channels, an excellent reliability may be achieved. A drawback of the concept is that it generates current spikes when charging and discharging the capacitors. These current spikes may cause problems with EMI, which remains to be proven experimentally.
2) Isolation Transformers: The volume of an APS is mainly defined by the volume of the isolation transformer. HV isolation transformers with a closed magnetic core are usually heavy, bulky, and expensive [54]. Dimensions of a 100-W dry-type cast coil transformer, which is capable to withstand the stress of 20 kVrms for 10 s without partial discharge, are 20 × 20 × 20 cm at a weight of about 5.5 kg [55]. However, in a converter with a series-connected switch (see Table III), a comparably large number of such transformers are required. Due to their cost and size, this is a significant drawback. Air or tape-insulated transformers with discrete windings or PCB windings are not suitable for HV applications, because of large required creepage and safety distances to other circuit components [5]. A modular APS for MV MMCs based on an LLC resonant converter with individual floated voltage sources containing high-frequency and HV-isolation transformers is presented in [56].
Using MV SiC devices, calls for the development of compact APSs with very low coupling capacitance. Special transformer designs complying with these requirements have been presented in [5], [7], and [10]. Rothmund et al. [5] demonstrated a highly compact isolation transformer with a size of only 16 × 16 × 14 mm and a maximum output power of 2 W is demonstrated, which can be integrated into an SiC power module. Another compact design intended for the supply of several GDUs at the same time is proposed in [10]. The mentioned studies report ultralow coupling capacitances in the range of 1-3 pF. Ground shields are suggested to be used in order to further reduce the effective capacitance [7]. It should be noted that commercially available isolated power supplies, e.g., for 3.3 and 6.5 kV IGBTs, feature coupling capacitance in the range of 20-25 pF [57]- [59].
It can be concluded that the voltage range of isolation transformers is limited. Even if higher voltages are possible, using for instance oil filled containers, such concepts necessarily become voluminous and costly. Isolation transformers, therefore, cannot be used to power individual submodules externally from the ground-level of HVDC converters. However, they may be used as a second stage of the APS to feed power to the high-side switch of a submodule, as shown in Fig. 4(a).
3) Inductive Power Transfer: The application of IPT systems for APSs has been investigated in [32], [54], and [60]- [64]. Reported insulation voltages reach up to 55 kVrms. The power is delivered by two (or more) coils [64], which can be considered as a coreless transformer. Multiple isolated outputs are possible [32], [60]. Experimental results demonstrate that comparably high output power levels of 250 W at high dc-dc efficiencies between 85-95% can be achieved [32].
Since the electric breakdown field strength of air is comparably small, a large air gap and large creepage paths are necessary to ensure a high isolation voltage. It should be noted that the efficiency drops rapidly with increased distance between the coils. Depending on the overall design, the air gap of presented prototypes is in the range of 20-80 mm. The IPT concept is, therefore, not suited for an integration in a highly compact module [5]. Another drawback of the concept is its weak tolerance to EMI. The transmitter coil generates strong near field radiation and, is potentially a considerable source of EMI [62]. To protect the GDU from inductively coupled EMI, it has to be placed or shielded properly, because the radiated EMI may otherwise cause serious malfunction of the extremely EMI sensitive GDU, especially in the case when MV SiC devices are used.
However, the air gap offers the advantage of very low parasitic capacitance, which significantly increases CM impedance of the GDU. A design with a coupling capacitance of only 1.72 pF is presented in [62]. The IPT system can be optimized with respect to voltage stability, efficiency, component stress, and complexity. An operating concept without control feedback is shown in [61]. 4) Radio Frequency: One potential way to transmit power and data via electromagnetic radiation is via RF waves, so-called WPT, see Fig. 13. Until recently, the amount of energy that could be captured at the device level was very limited, and RF WPT technology could only be used for extreme low-power applications [65]. However, there are recent articles showcasing RF WPT for relatively high-power loads, e.g., 20-W output at 40 MHz [66], and a circuit outputting 500 W at 4.6 MHz [67].
It should be noted that if the power level is increased, there is a risk of exceeding the safety standards for radio. Additionally, RF might interfere with the communication of other devices and cannot be transmitted beyond 1 m. It therefore remains questionable whether RF technology can meet the key APS requirements of high-power converters given in Table I, in particular sufficient EMI immunity. 5) Optical Link: Another way to transmit power and data via electromagnetic radiation is via light, also shown in Fig. 13. The optical link can be set up via fiber optics or air. Both technologies, POF and OWPT, are described and discussed in the following.
The POF concept has been in use since the 1990s, for instance as power source for HV current sensors [68], [69]. The reasons for choosing POF systems over conventional power supplies could be the need for very high galvanic isolation, HV protection, EMI sensitivity, weight reduction, spark protection, corrosion resistance, or high magnetic fields. POF systems include a laser, an optical fiber, an optical power converter (OPC), and an LV dc-dc converter. The isolation capacitance is extremely small compared to a solution with isolation transformer. This enables virtually infinite dv/dt immunity and negligible small CM currents through the optical fiber [7]. Until recently, POF solutions have been comparably costly [5], limiting the applicability to highly specialized equipment with large cost margins [18], [70]. However, significant advances in laser diode and multijunction solar cell technology have been made [16], [71], where power and voltage levels could be increased and cost predictions for the future have become very optimistic. A key role in this development is a new OPC technology [72]- [74], facilitating conversion efficiencies exceeding 60%. OPCs with more than 20-W output power and 30-V output voltage are available [26]. The POF technology is particularly interesting for future MV SiC MOSFETs, as presented in [75], offering a high isolation capacitance and potentially a small footprint. Different implementations of a POF-based APS for MMC submodules are suggested in [21]. Charging of multiple devices is possible by additionally introducing optical elements to manipulate and control the light beam, e.g., splitters [76].
A drawback of the POF concept is that the efficiency of the total link might be relatively low [77]. As regards HV applications, there is also a general concern in relation to reliability, especially for high-power POF systems. Recently, such reliability issues have been addressed in [21] together with power and data transmission requirements. As mentioned before, POF technology offers superior noise immunity [6]. However, when placed close to a GDU of a HV converter, the receiver might need to be shielded.
In the OWPT concept, also referred to as power-beaming, a laser light beam is directed towards a receiver in a straight line over air. This approach is similar to POF, but without the need for an optical fiber, eliminating the risk of disconnection or breaking the same and, thereby, increasing reliability. Experimental demonstrations of OWPT have been published in [78] and [79]. A long-range, free-space power-beaming system with 400 W of power across 325 m has been achieved in the context of military research. More and more companies are, however, recently focusing on powering wirelessly for civil uses, e.g., home and industrial appliances, promising a significant reduction in installation and maintenance expenses for the latter. Current OWPT technology delivers power up to 3 W at a distance of 5-10 m [80]. Challenges of the OWPT concepts are the correct alignment of transmitter and receiver, and safety concerns due to objects or persons in the beam path. As regards the first, a laser with narrow beam divergence is preferable. This way the transmitter and receiver can also be designed smaller. Concerning safety, rapid-response interlock systems can be implemented to ensure that the laser emitter terminates if the beam path is blocked and check that the path is clear before turning it back on.

V. EVALUATION AND FUTURE PROSPECTS
A qualitative evaluation of the presented APS concepts regarding the technical requirements of Table I is given in Table IV. In addition, a visual representation of this table is shown in Fig. 14, aiming to facilitate the identification of relative benefits and shortcomings of the various APS concepts. Moreover, a discussion on the selection of a suitable APS concept for MV and HV applications is provided. Finally, future prospects of internal and external APS systems are discussed in the rest of this section. On the one hand, those are based on the technology development of APS concepts, and on the other hand, technological trends with respect to converter main circuits, e.g., MV SiC devices and UHV submodules.
In general, the suitability of a certain APS concept is highly dependent on the requirements of the application. Sometimes, it can be advantageous to combine several of these concepts. Examples of promising combinations are discussed later in this section. In the majority of cases, the most important factor deciding the choice of an APS technology is the isolation (or input) voltage. For example, the most extreme case regarding isolation voltage is to supply power to a submodule of an HVdc  1 First stage of a two-stage internal APS (see Fig. 4). 2 Requires a power infeed to the submodule at the lowest voltage level. 3 Considered as external APS for MV converters, and as second stage of an internal APS for HVDC converters (see Fig. 4). 4 Depends on the required compactness of the design. 5 Rapid development of more powerful and reliable laser/receiver systems. converter externally. The only realistic option in this case is to use one of the optical concepts. As the voltage level goes down, additional options are possible. At the lowest voltage levels, the technical challenges are no longer significant. State-of-theart APS concepts are sufficiently good for today's converters, but a reduction of complexity would anyway be beneficial, for instance, in terms of reliability and cost. In some cases, snubber-based power tapping could provide such a complexity reduction. However, their need for a resistive voltage divider (or equivalent) during long OFF states of the main power switches is a drawback.
APSs for MV and HVdc converter submodules shall be discussed separately in the following, in order to exemplify the strength and weaknesses of the presented APS concepts regarding the specific requirements of those applications.

A. MV Converters
MV converters operate in most cases with elevated switching frequencies, and thus, may have a higher GDU power consumption compared with HVdc converters depending on the employed semiconductor device technology. Certain internal APS topologies are more feasible if a high power transfer is needed, for instance, TI-buck, ISOP, or multicell converters.
Since system voltages of MV systems are in the range of 1-52 kV [55], also induction-based concepts are viable. Several external GDU and APS solutions based on special isolation transformer designs [5], [7], [10] and IPT systems [32], [61], [81] have been reported for MV applications, such as large motor drives and transformer-coupled STATCOMs [55]. As regards FACTS applications, a fast start-up speed is beneficial [30], which can be provided by all external APS concepts, as shown in Fig. 14(b).

B. HVDC Converters
For HV applications, such as HVdc and FACTS, the system voltage can be hundreds of kilovolts (up to 1 MV). Such extreme requirements on isolation systems and distances make supplying power from ground-referenced transformer-based systems infeasible. The corresponding design challenges have only permitted internal APSs to be a viable option for high-power HV converters.
External powering was considered to be not practical for these applications [82], because the possible transferable power of optical APS concepts has been too low until recently. The utilization of MV SiC devices is foreseen to reduce the power demand, potentially allowing the use of low-power APS concepts, like voltage dividers. However, MV SiC devices may at the same time motivate to increase the switching frequency of MMCs to facilitate smaller, lighter systems that are more efficient, and likely less expensive. Reliability is of paramount importance in industrial HV applications and, therefore, a major design criterion. Thus, costly and less efficient APS solutions should not necessarily be excluded. Also, additional redundancy and large margins might be acceptable if such measures promise to enhance the reliability of the converter.

C. Future Internal APS Concepts
It is likely that MV SiC power devices will be employed in HV submodules in the future, enabling capacitor voltages of up to 30 kV. Thus, extreme isolation requirements may apply not only for external, but also for internal APS concepts. However, stateof-the-art APS concepts are not suitable for such submodule ratings. Robustness against electromagnetic noise will become even more important, because GDUs and control electronics may be easily interfered due to the fast switching and high dv/dt of MV SiC devices [83].
Future solutions may build on state-of-the-art topologies, but utilize MV SiC technology in the APS itself to enable improved efficiency, reliability, simplicity, and compactness of those topologies. Flyback and TI-buck converters may be an interesting alternative if an MV supercascode SiC switch [23], [25] replaces the series-connected MOSFETs. The active voltagedivider-based APS in combination with a second, isolating converter is also a promising concept for HV submodules if the required power is relatively low. Higher input voltages may be possible with SiC power devices with higher voltage ratings, such as the SiC supercascode.

D. Future External APS Concepts
The design process and construction of an MMC could be significantly simplified if the gate signals and auxiliary power are directly supplied to the submodules without the need of tapping from the capacitor. Black start capability is a big advantage, since auxiliaries could be powered before the main circuit. Hence, access to control equipment, sensors, and monitoring data is possible without the need for complicated start-up procedures.
Since light-based systems fulfill the main requirements of extreme voltage isolation and immunity to EMI, they are considered most useful. HVdc converter stations seem to be well suited for the implementation of optical APS systems, since those work most reliably in controlled environments (no beam interruptions and effects of weather phenomena), and in environments where objects that require power are static. Furthermore, multiple receivers can possibly be easily charged from the same transmitter, with the ability to scale to high power levels.
Due to the high rate of development in the field of optical powering, it is not unlikely that such APS concepts may be more cost effective than the state-of-the-art APS concepts within a few years.

VI. CONCLUSION
This article presented a review of state-of-the-art APS concepts for high-power converter submodules, investigated alternative concepts including active voltage dividers and snubberbased power tapping, and also discussed recently presented solutions, such as optical power concepts.
State-of-the-art topologies are described and evaluated, and it is founded that the choice of solution is highly dependent on the requirements of the application regarding input voltage and output power range, EMI immunity, compactness, start-up time, etc. The challenges in designing low-power HV step-down dc-dc converters are pointed out and possible realizations are presented.
In the future, it is also likely that MMCs and other HV equipment will employ MV SiC power devices. This may call for other solutions than supplying the submodule APSs internally from the submodule capacitor. Some of the state-of-the-art concepts can be adapted to higher voltage levels by using MV low-current devices based on SiC technology. One way to devise new APS concepts could be by combining existing solutions. POF and OWPT technologies are under rapid development, and it is likely that the performance increases while cost reduces in the next few years, making optical powering a very competitive technology. From 1994 to 2011, he was a Development Engineer with ABB, Västerås, Sweden, in various powerelectronics-related areas, such as railway traction systems and converters for HVdc power transmission systems. He is currently an Associate Professor with KTH. He is the Inventor or Co-Inventor of 12 granted patents and 13 patents pending, and has authored or coauthored more than 75 scientific papers published at international conferences or in journals. His research interests include new converter topologies for power transmission applications and grid integration of renewable energy sources.