How to make a 10kW bidirectional AC/DC converter

If you check the literature, prototype designs and evaluation boards for bidirectional power supplies are appearing everywhere. Why the sudden interest in bi-directionality? One of the main reasons is electric vehicles, or more exactly, their battery packs, as a storage medium for renewable energy.

Renewable energy is now a hot topic in many countries: it is the fastest growing energy source in the United States, with a 100% growth rate from 2000 to 2018¹ , the UK produced more power from zero-carbon energy than from fuel-based power stations for the first time last year, despite having more than 75% of the electrical power derived from fossil fuels less than a decade ago² and Austria, where RECOM has its energy-neutral headquarters, is at the forefront of the European green energy project, with around 72% of its electricity needs coming from zero-carbon sources³ . However, not every country has the advantages of a long coastline where nuclear power stations can be tucked away out of sight or convenient local snow-capped mountains and lakes. Most countries have to rely on wind, solar or small-scale river hydroelectric power which are not always very reliable sources; low river water levels in summer limit run-of-river power production and peak electrical demand often comes during windless days or at night.

One of the solutions to ensuring supply continuity is, of course, to use the combined electrical power stored in the batteries of electric vehicles (EV) to help balance out supply and demand in a so-called Vehicle-to-Grid (V2G) system. Within the next ten years, there is likely to be around 7 million electric vehicles in Germany alone, each with 20-100kWh of on-board battery capacity. Even if only 20% of that capacity is available at any one time, that still represents 140GW, or more capacity than 100 nuclear power stations.

The key to a successful V2G system is the combination of bidirectional energy flow and artificial intelligence. Most vehicles spend more than 95% of their time parked⁴. If an EV is plugged into a charging station while the owner is at work, the EV can determine whether to continue recharging its battery or to release part of its stored charge back to the grid at peak times, thus adapting its state-of-charge depending on known or predicted usage patterns. As the majority of daytime journeys are less than 37km per day³, it is not always necessary that the vehicle is fully charged or even remains so in between the daily journey times. But to do this requires a bidirectional charger/mains inverter to transfer the electrical power in both directions. Note that the bidirectional charging station does not need to be itself intelligent; the necessary processing power is already incorporated into the A.I. system built in to the electric vehicles.

Having established the potential need for millions of bidirectional AC/DC power supplies to meet the estimated increase in EVs by 2030, the next stage is to ask if it is commercially viable to build them. There are two relatively recent developments that have made bidirectional designs substantially simpler and cheaper to realize - the first is the introduction of new topologies that lend themselves particularly well to bidirectional current flow and the second is the maturation of new technologies such as Silicon Carbide (SiC) high power switching transistors, which are now price-competitive with the long-established Insulated Gate Bipolar Transistor (IGBT) technology, but with significantly better efficiency savings.

¹ https://www.c2es.org/content/renewable-energy/
² https://www.carbonbrief.org/analysis-uk-renewables-generate-more-electricity-than-fossil-fuels-for-first-time
³ https://mission2030.info/wp-content/uploads/2018/10/Klima-Energiestrategie_en.pdf
⁴ https://www.bmvi.de/SharedDocs/DE/Artikel/G/mobilitaet-in-deutschland.html

Unidirectional AC/DC battery chargers have been on the market for many years. They generally use the following arrangement, with some proprietary variations:


Mains powered battery chargers are essentially AC-DC converters (the PFC stage), followed by a DC-AC converter (the transformer driver stage), followed by an AC-DC converter (the rectifier and output filter stage), followed by a battery charging interface. Depending on the battery voltage and power levels, the transformer driver stage can use single-ended, push-pull, phase shift full bridge or LLC topologies, but in almost every battery charger application, power factor correction on the input and some sort of battery interface to handle reverse polarity protection and match the charging voltage and current profile to the cell chemistry is required.

To make such a design bi-directional could be realized by adding an inverter stage in parallel with the existing schematic:


However, this way of adding bi-directionality is inefficient in the use of the existing components and adds significant cost because two transformers are needed. If the market is for millions of such bi-directional power supplies, then the cost of each unit becomes a very significant factor. A better solution would be to use a topology that is inherently bi-directional with only one isolating transformer.


To go through the design considerations for such a product, we can analyse each stage in turn, comparing the conventional unidirectional topology with alternate bidirectional alternatives. As the desired result is a power supply that works in both directions, we can start at the end, as it is also the beginning.

The heart of any power converter is the power transformer. If the battery voltage is similar to the PFC bus voltage (around 400VDC), then a good choice for the power stage topology is CLLC, the bidirectional version of the popular resonant LCC topology. As CLLC is perfectly symmetrical, current can flow equally well in either direction, but as it relies on resonance, it works best with similar input and output voltages. However, if the battery voltage is lower (48V) or much higher (800VDC), then a phase-shifted full bridge (PSFB) is in many ways a better alternative because of the lower cost. In this example, the application was for a 48V lithium-ion battery pack, so a PSFB topology was chosen.

In the prototype 10kW design, an analog controller IC from Texas Instruments was used, the UCC28950. This may seem to be a surprising choice considering that digital controllers are a more modern alternative, however using a standard analog building block has several advantages:

1. It is an established part with well-known performance and proven reliability.
2. It has ZVS capability and light load management built-in, so the efficiency remains high over the entire load range
3. The timing for the synchronous rectification is also generated by the controller IC, sparing external components and cost.
4. It has built-in protection, so the output does not need to be constantly monitored for fault conditions, freeing up the microprocessor for other tasks.
5. It has built-in protection, so the output does not need to be constantly monitored for fault conditions, freeing up the microprocessor for other tasks.
6. It is inexpensive

The block-diagram of the power supply in battery charging mode is shown in Figure 5. The feedback loop is analog, using the output voltage to regulate the controller IC phase shift. In addition, the microcontroller measures the battery charging voltage and current and modifies the analog feedback loop to adjust them to the required levels. As such it is a hybrid – a digitally-controlled analog feedback loop. This arrangement gives all of the stability and failsafe operational benefits of the analog loop with the versatility and adjustment accuracy of a microprocessor-controlled trim control.

To make the circuit shown in Figure 5 bidirectional, the microprocessor needs to take over control of the synchronous rectification FETs on the “output” side, and use the high current filter inductor to make a current-fed push-pull boost converter.


In a current-fed topology, the current through the transformer windings must never fall to zero, so both Q_A and Q_B are switched on simultaneously and alternately switched off. The supply current IL is limited by the inductor inductance. On the primary (now secondary) winding, a stepped output will be generated at a peak voltage equal to the turns ratio.


On the previously primary side of the transformer, the PSFB controller is disabled by the microcontroller and the primary side transistors are not driven. However, the SiC transistors all contain both a body diode and a free-wheeling diode in parallel, so that the voltage generated by the transformer winding is passively full-wave rectified and the output stored on the PFC capacitor. An additional voltage feedback loop allows the microcontroller to regulate the reverse energy flow.


The biggest advantage of this topology is its simplicity and the double-use of all of the main components for both forward and reverse energy flow. This significantly reduces the costs compared to the entirely symmetrical arrangement of many other bidirectional designs that use Dual Active Bridge (DAB) arrangements with four switching transistors on each side and two power inductors in addition to the transformer.

It must be noted that the primary side switching transistors need to have robust body diodes to function as passive rectifiers, even with external diodes in parallel. Only SiC or IGBT transistors can be used in this way as the body-diode structure inherent in Si-MOSFETs can suffer from latch-up and fail.

Power factor correction is required for single phase mains-powered power supplies with more than 75W. The simplest active topology uses a boost converter to raise the rectified mains input to higher than the peak input voltage so that the current control into the PFC capacitor can be made continuous and synchronized to the sinusoidal input voltage to give a power factor close to unity:


This circuit suffers from two major problems: D1-D5 mean that the circuit is unidirectional only and the losses in the diode bridge and boost converter stage can be significant. However, a relatively stable DC bus voltage is a prerequisite for FBPS and many resonant power stages that cannot adapt to a wide input voltage range.

A more widely adopted topology for higher power designs is the totem-pole bridgeless PFC topology. The name totem-pole comes from the American Indian’s use of a carved tree trunk to make images of various gods and powerful chiefs as a totem to ward off evil spirits and bring good fortune to the tribe. As these carved figures were stacked above each other on the vertical pole, the name was appropriated to describe vertically stacked transistors in layout schematics. In a totem-pole bridgeless design one set of transistors is used to switch the input synchronized with the 50/60Hz mains frequency and another set switches at a much higher frequency (100kHz) to form the boost converter. The combination requires no rectification diodes and so eliminates their associated losses.


SiC transistors offer the advantage of having a robust body diode with a low Qrr, so make ideal components for the high frequency switching transistors Q1 and Q2. The low Qrr and RDS,on permit the PFC stage to operate in constant conduction mode (CCM) which lowers the THD and increases the overall efficiency. CCM also simplifies the EMC filtering, another cost benefit. The low frequency mains synchronizing transistors Q3 and Q4 can be Si-FETs. A further advantage of the totem-pole topology is that is can operate bi-directionally.

A further development of the totem-pole PFC topology is the three-phase version with neutral point clamping (NPC). Many three phase supplies do not include a neutral wire, consisting of L1, L2, L3 and ground only. The addition of some steering diodes in the so-called Vienna rectifier topology creates a NPC point that is at the mid-point of the DC bus voltage. The use of NPC means that the currents in the three phases are more evenly balanced, which reduces the site of the EMC filter components.


The solution shown in Figure 11 is sufficient for many high power PFC applications, but it is not perfect. The NPC diodes can be replaced by back-to-back transistor pairs to increase the efficiency at a higher component cost, but that would still not eliminate the triangle-wave common mode voltage at the junction of the PFC output capacitors.


This common mode voltage is unavoidable and causes additional currents to circulate in the EMC filter, reducing its effectiveness and adding to the overall power dissipation. A solution to this issue is to use a technique called an ‘unfolder’ to inject an anti-current into the neutral point junction to cancel out the common mode voltage. The source of this current is a separate buck converter powered from the high voltage bus.


A full bi-directional bridgeless PFC design requires a complex controller that can accurately co-ordinate all of the various timing and synchronization requirements. A digital vector space controller is a good solution for the programming:


The block diagram of the final bidirectional solution looks something like this:


A 10kW prototype battery conditioner was built to confirm the working specifications. The application was not an EV charging station, but used as part of a battery conditioning system (new manufactured battery packs need to be charged, discharged and then recharged to reach their full capacity – by recycling the stored energy back into the mains supply, this process was made more economic), however, the basic principles of operation were similar to what is described in this article. The prototype has an efficiency of more than 96% and a power factor of more than 0.99 in both directions.

At the beginning of this article, the question was asked “Who needs a bidirectional AC/DC converter?” The answer seems to be “most likely, millions of us.” With such a high demand, there will be many solutions offered to solve this problem, so this article covers only one such potential arrangement. However, for all high volume applications, the main considerations besides reliability and performance is low unit cost. Many of the bidirectional topologies offered are fully symmetrical: the output is a mirror-image of the input. This seems logical. However, by modifying a non-symmetrical design so that it can work in both directions can offer a lower cost by avoiding duplication of components and reducing the overall BoM.