Multi active bridge converters for multiport power conversion

Wichtigste Erkenntnisse

• A multi active bridge saves transformers and switches, and it moves the engineering risk from hardware into control.
• Every phase shift affects every port, so a decoupling strategy has to be proven across the full load range, voltage range and each direction of power.
• Switched models run in real time let you cover every port combination and fault case before a prototype is built.

A multi active bridge converter replaces several separate isolated converters with one transformer, and the price of that saving is a control problem you have to prove across every port combination.

Sites that pair generation with storage are already common. Berkeley Lab counted 543 hybrid projects installed across the United States through 2024, and a typical site still joins solar, batteries and the grid through separate conversion stages. A multiport converter collapses those stages into one magnetic link, which cuts parts, volume and conversion steps.

That shared link is why the hard part is control. Power sent from one port lands on all the others unless the controller compensates, and the number of operating cases grows with each winding you add. Your power electronics will behave only as well as the controller has been tested. Real-time simulation is the practical way to test all of it before copper and silicon are committed.

What a multi active bridge converter does with one transformer

A multi active bridge converter connects three or more DC ports through a single high frequency transformer with one winding per port. Each port has its own full bridge that applies a square wave voltage to its winding. Power moves between ports according to the phase shift between those square waves, in either direction.

The simplest relative is the dual active bridge, which has two ports and one phase shift. Add a third winding and a third bridge and you have a triple active bridge. Picture a 400 V battery, an 800 V DC bus and a 48 V auxiliary supply sharing one core. Turns ratios handle the voltage differences, and every port stays isolated.

Linking those three ports pairwise with dual active bridges takes three transformers and 24 switches. The triple active bridge does it with one transformer and 12. The cost is a shared magnetic path, so no port acts alone.

How power flows between ports in a triple active bridge

Power in a triple active bridge flows from the port whose voltage leads in phase to the port that lags. Between any two ports, the transferred power rises with both port voltages and with the phase shift, and falls as switching frequency or the leakage inductance between them increases. Each port’s total is the sum of its exchanges with the others.

A generic two-port calculation shows the scale. With both voltages referred to 400 V, a 50 kHz switching frequency, 20 microhenries of link inductance and a 30 degree phase shift, the standard phase shift equation gives roughly 11.1 kW. Add a third port with its own phase shift and port 1 sends power to port 2 and port 3, and port 2 also trades with port 3, so three exchanges run at once. Two phase angles, each measured against a reference bridge, set all three.

Why port coupling makes multiport control harder than it looks

Port coupling means one control action moves power at every port. Each phase shift is measured against the same reference, so adjusting the angle of one bridge alters its exchange with all the other bridges at once. 

“A controller built from independent loops will see each loop disturb its neighbours.”

Consider a triple active bridge feeding a regulated load while a battery charges. A load step makes the load loop increase its phase lag to draw more power. That same shift pulls extra power out of the battery port, so the battery current dips until its own loop reacts. The two loops then correct against each other and oscillate.

Leakage inductances set how much each pair of ports interacts, and those values move with winding layout and manufacturing tolerance. A decoupling scheme tuned for the design values will be slightly wrong on the built unit.

Operating point matters too. The relationship between phase shift and power is nonlinear, so a controller that is well damped at light load can ring at full power. You can’t sign off on one operating point and assume the rest.

Control strategies that decouple power flow across the ports

Decoupling control makes each port respond to its own setpoint while the others hold steady. The three common routes are a decoupling matrix in software, separated loop bandwidths, and a transformer designed so one port acts as a stiff voltage reference. Each one trades design effort against performance.

A decoupling matrix inverts the small-signal gains between phase shifts and port powers, then feeds the result forward so a change at one port is cancelled at the others. Loop separation gives the load port a fast loop and the battery port a slow one. The hardware route gives one port very low leakage, so the remaining ports see it as a voltage source.

Whichever route you pick, the proof is the same. Each strategy has to hold across the load range, the voltage range and every direction of power. That’s a test matrix, and a single bench run won’t cover it.

How to model a multiport converter for controller testing

A multiport converter model for controller testing has to reproduce the switching events, the multiwinding transformer and the timing of every gate signal. Averaged models are useful for loop design. Controller validation needs a switched model, because phase shift errors, dead time and lost soft switching only appear at the switching level.

Start with the transformer. A three-winding unit reduces to an equivalent network of leakage inductances, one between each pair of ports, plus a magnetizing branch. Take those values from the winding design or from a short circuit test on each pair, because a guessed value gives you a controller tuned for a converter that doesn’t exist.

Timing is the next constraint. At 50 kHz a half period lasts 10 microseconds, so a phase shift of 1% of that half period is only 100 nanoseconds. A simulator that samples gate signals more coarsely will misreport transferred power. OPAL-RT runs the switched converter model on FPGA-based simulators with the physical controller connected in a hardware-in-the-loop setup, so the gate timing your controller produces is what the simulated bridges receive.

Where hardware prototypes fall short for multiport energy routing

A hardware prototype can only exercise the port combinations you can afford to build sources and loads for. Energy routing through a multi active bridge means each port sources or sinks on command, so a full test needs a bidirectional supply on every port. Few labs have that at rated power.

The count rises fast. A three-port converter has 6 distinct source and sink patterns, a four-port unit has 14 and a five-port unit has 30. Multiply each by the voltage corners and load levels, then add faults. A fault test on a prototype also risks the only transformer you have.

Scale makes the gap expensive. Roughly 890 GW of storage capacity was seeking grid interconnection across the United States at the end of 2024. Each of those projects routes power among cells, generation and the grid, and each routing case is a controller state somebody has to verify.

A simulated plant lets you run the cases a prototype makes costly or risky.

• Every source and sink pattern at each voltage corner
• Load steps on one port while another port reverses direction
• Loss of a port during power transfer
• Transformer leakage values at both ends of the tolerance band
• Short circuit and open circuit faults on each bridge

What full port coverage means for a multiport program

Full port coverage means your controller has been run through every direction of power, at every voltage corner, before the first prototype is energized. It’s the difference between knowing the control works and hoping the untested cases resemble the tested ones. With a shared transformer, they rarely do.

“It’s the difference between knowing the control works and hoping the untested cases resemble the tested ones.”

The converter itself is the easy win. The control risk stays hidden until a port combination nobody tested shows up in the field. Discipline here is plain work. You list the cases, run them all and fix what rings.

That’s the role we see for simulation at OPAL-RT. Engineers who run the complete matrix in real time arrive at the prototype with a controller they already trust, and the hardware build becomes a confirmation. You’ll spend bench hours on efficiency and thermal results.