A supercapacitor bank on a DC bus is defined by two numbers that are decided together: how far its voltage is allowed to swing, and how much current the converter must pass at the bottom of that swing. Widen the window and the cell count falls while the current rating climbs in proportion. Narrow it and the bank grows for the same usable energy. Most of the trouble in this part of a design comes from fixing one number without the other.
Earlier pieces in this series split a duty cycle between a supercapacitor and a battery by time scale, then worked out a bank size from one hoist cycle. That second article ended with 56 F across a 600 V bus, sized to recover 4.4 MJ on each descent. It stopped there on purpose. Getting those 4.4 MJ in and out of the bank is a converter problem, and the converter is what this article is about.
Choosing the Voltage Window
On a battery, the voltage window is mostly a property of the chemistry. A lithium cell runs from about 4.2 V to 3.0 V and you take what you get. On a supercapacitor the window is a design choice, bounded only by the cell’s rated voltage at the top and by whatever the converter can still work with at the bottom.
The top end is fixed arithmetic. A string of N cells rated at 2.7 V can sit at 2.7N volts and no higher, so a 600 V bus sets a minimum of 223 cells in series. Designers normally leave a little headroom and use 222 cells at 2.7 V, which comes to 599.4 V, or switch to the 3 V series and land on 200 cells.
The bottom end is free. It can be 90% of the top or 33%, and that single choice moves both the cell count and the converter rating.
What Each Window Costs in Cells
The energy a bank can deliver is the difference between the energy it holds at the top of the window and what is left at the bottom:
Eusable = ½ C (Vmax² − Vmin²)
Divide that by the full energy ½CVmax² and the capacitance drops out. What remains is a pure ratio: 1 − (Vmin/Vmax)². That ratio is the whole story, because it says how much of the bank you paid for and how much you can actually use.
| Vmin as a share of Vmax | Usable share of bank energy | Capacitance needed for the same usable energy |
|---|---|---|
| 90% | 19% | 3.9 times the 2:1 case |
| 80% | 36% | 2.1 times |
| 75% | 44% | 1.7 times |
| 60% | 64% | 1.2 times |
| 50% | 75% | 1.0 |
| 33% | 89% | 0.84 times |
Two things fall out of that table. The first is that a narrow window is brutally expensive in cells: letting the bank sag to only 90% of its top voltage throws away 81% of what you bought. The second is that the gains flatten fast. Going from a 2:1 window to 3:1 adds 14 points of usable energy and costs the converter twice the current. The last stretch of the swing is where the money is worst spent.
Converter Rating Follows the Bottom of the Window
The number that gets missed is the current at the low end. A converter moving a fixed power P sees a current of P divided by the bank voltage, so its worst case is at the bottom of the window, not the middle.
Take 220 kW, which is roughly what a container crane hoist recovers on a descent of about twenty seconds. Across a 600 to 450 V window the current runs from 367 A at the top to 489 A at the bottom. Push the same window down to 300 V and the bottom current becomes 733 A. The power has not changed. The current the converter must carry has gone up by half.
The general rule is simple enough to keep in your head: for a fixed power, the current at the bottom of the window is the current at the top multiplied by Vmax/Vmin. A 2:1 window doubles it. A 3:1 window triples it.
This is why a converter quoted at “220 kW” can still trip on a crane. If the rating was written around nominal bank voltage, the low end of the swing is outside its current limit, and the drive discovers that on the first long descent with a full load.
Compare Against Direct Parallel Before You Commit
Not every system needs a converter, and the alternative deserves a look before dismissing it. Connect the bank straight to the DC bus and the bus voltage becomes the window.
On a 600 V bus held within plus or minus 10%, the bank sees 540 to 660 V. That is a 0.82 ratio, which works out to 33% usable energy. Recovering the same 4.4 MJ as the hoist example takes 61 F instead of 56 F. Put that against the 32.6 F a 2:1 converter window would need for identical usable energy and the comparison is 61 against 33.
So direct parallel costs about 1.9 times the cells, and it also removes the conversion loss and the converter itself. On a system where the bus is genuinely stiff and the recoverable energy is modest, skipping the converter is the better engineering. On a system that needs a wide swing to keep the cell count sane, the converter earns its place by widening the window rather than by any virtue of its own.
Put plainly, a converter does not create energy. It buys a wider voltage window, and it bills you in current.
Precharge at This Scale Is Not a Resistor Job
A discharged bank has to be brought up to bus voltage before it can be connected, and the energy involved is the full ½CV². For 56 F at 600 V that is 10.1 MJ, or 2.8 kWh.
A precharge resistor sized for a few seconds of inrush has to dissipate that energy. Ten megajoules through a resistor in a minute is 168 kW of dissipation, which is not a component, it is a heater. Direct-parallel designs at this capacitance generally cannot precharge through a resistor at all. They need a current-limited source, which in practice means a converter or a controlled charger doing the job anyway.
Even with a converter the arithmetic sets the pace. Charging follows I = C dv/dt, so a 100 A limit on 56 F takes 336 seconds to reach 600 V from empty. That is five and a half minutes. Precharge at this scale is a scheduled operating state with its own sequence and interlocks, not a relay closing at switch-on.
Smaller systems behave the same way at smaller scale. A 5 F bank on a 48 V rail holds 5.8 kJ, which a resistor can absorb in under a minute at a couple of hundred watts. The physics does not change; the numbers just stop being awkward.
Widening the Window Makes Balancing Mandatory
Series strings need cell voltages held equal, and the reason is unforgiving. One cell running above its rated voltage degrades faster than its neighbours, its capacitance falls, it then swings further on the next cycle, and the gap widens. There is no self-correcting mechanism.
The passive answer is a resistor across every cell, sized so it carries far more current than the cell’s leakage spread. Supercapacitor leakage sits in the tens to hundreds of microamps at rated voltage. A resistor drawing 1 mA at 2.7 V is 2.7 kΩ and burns 2.7 mW. Across 222 cells that is 0.6 W of continuous loss, which is acceptable in exchange for holding every cell within a few tens of millivolts of its neighbours. The selection guide covers how the cell families differ in leakage and self-discharge, which is what sets that spread in the first place.
Active balancing shuttles charge between cells instead of burning it, and draws less standing power. It also adds switching components to every cell or every few cells, and each one is a failure mode inside an enclosure nobody wants to open.
Whichever scheme you choose, a wide window makes it non-negotiable. A bank that spends most of its life near Vmax gives leakage spread days to accumulate, and the cell that drifts high is the one that fails. A narrow window that keeps the bank well below rated voltage is far more forgiving of a weak balancer, which is another hidden cost of choosing the wide window.
How Much of the Bank the Control Is Allowed to Use
The last piece is a control decision rather than a hardware one, and it usually gets made by accident.
A bank sized to recover 4.4 MJ per descent will hold a full 4.4 MJ only if the control runs it from Vmax all the way to Vmin on every cycle. Anything less is a partial recovery, and the shortfall shows up as heat in the braking resistor. Deciding the reserve floor explicitly is what separates a system that behaves predictably from one that hits its limits under load.
Two constraints normally set that floor. The drive needs enough headroom to absorb a second descent before the first has been consumed, so the floor is set by the worst realistic sequence of events rather than by the average cycle. And the converter’s current limit caps how fast the bank can be emptied, so a deep discharge followed immediately by another descent will hit the limit before it hits Vmin.
Both cases need a defined fallback. Keeping the braking resistor installed is the usual answer, and it is the right one. A bank that reaches full charge mid-descent has nowhere to put the energy, and the resistor gives the drive somewhere safe to send it while keeping the crane operable.
Where This Goes Wrong on the Bench
Three failures account for most of the trouble in this part of a design.
The first is a converter rated at nominal bank voltage, which trips on overcurrent at the bottom of the window. It usually appears as a fault on long descents and clears on short ones, which sends people looking at the load rather than the rating.
The second is a cell that drifts high over weeks of standing at Vmax. It rarely announces itself. What shows up is a bank that delivers less energy than the model predicts, because the weakest cell in the string limits the whole string’s usable voltage while the balancer quietly cooks.
The third is a control that trusts its own capacitance figure. Supercapacitor cells lose capacitance over years, and a bank 20% down from its installed value delivers 20% less energy for the same voltage swing. A control that stops at Vmin on the assumption that Vmin means a known amount of energy will under-deliver, and the gap widens as the bank ages. Re-deriving the usable energy figure from a periodic capacitance check is what keeps the assumption honest.
What Comes Next
That closes the three-part run on hybrid energy storage: the time-scale split, the energy account for one cycle, and the converter that connects them. The next application series looks at a different duty entirely, where the bank is asked for very high power for a few hundred milliseconds at a time and the thermal design inside the module matters more than the string configuration.
Questions About Supercapacitor DC-DC Converter Sizing
How do I choose the minimum bank voltage?
Start from the converter’s current rating, because that is what the choice actually costs. Work out the current at your candidate Vmin from P/Vmin, check it against the converter’s continuous and peak ratings, and compare the resulting capacitance with the 1 − (Vmin/Vmax)² table. If the current increase outweighs the cell reduction, the window is already wide enough.
Do I need a converter at all?
Only if you intend to use a wider window than the bus allows. On a bus held within plus or minus 10%, direct parallel gives about 33% usable energy and needs roughly 1.9 times the cells of a 2:1 converter window. If the recoverable energy is small or the cell count is not a constraint, connecting straight to the bus is simpler and loses nothing.
Can I skip the balancing resistors?
Not on a series string that sits near rated voltage. Leakage varies between cells, and without a bypass path the variation accumulates until one cell runs above its rating. Passive resistors drawing about 1 mA per cell cost a few milliwatts each and keep the string within tens of millivolts, which is enough for most duty.
Why does my bank deliver less energy than the calculation says?
Three usual reasons: the control is not using the full window, one cell is drifting high and limiting the string, or the cells have aged and lost capacitance. Check the actual voltage swing at the bank terminals first, then cell voltages under load, then capacitance against the installed value.
Working out a bank and its converter for a specific duty? Send Xuansn the bus voltage, the power and duration of the event, the number of cycles per shift and the window you are considering. Our engineers manufacture the 2.7 V and 3 V cell series, the 3 V 600 F cells used in strings of this size, the 3.8 V lithium-ion series and the 5.5 V modules that smaller banks are built from, and will come back with a cell count, a string configuration and the current the converter has to carry at the bottom of the window. Reach us at coco@xuanxcapacitors.com, call +86-769-8166 8821, or send your specification through the contact page.






