Current limiting for supercapacitor charging

[PeterF]

Hi,

I'm working on a project where we have a supercapacitor bank with 10 series-connected capacitors.
These are to be used together with an external PSU, to keep the system going if the external PSU for some reason is temporarily shut off.
(They are also used for some peak current handling that the PSU might not handle).


I would like to use somewhere around 1-4 amps from the input (24V DC) PSU to charge the capacitors, when power is applied.

I want to avoid having the PSU going into its' current limiting, so I need to some how limit the current with which the capacitor bank is charged.


Do you have any suggestions for how to simplest implement such an input current limiting?
I've played around with a SEPIC design using the LT3956 with the application note mentioned in the datasheet.
(http://cds.linear.com/docs/en/datasheet/3956f.pdf)

I am however not too pleased with the low power output and I am not sure how much I could safely scale up this design.
I am also worried that I've missed a really simple way of solving this problem.


Thanks in advance for any tips! 

[Marco]

What's wrong with a buck setup with a pmos, but with only current limiting for the control loop with some hysteresis?

[sanwal209]

You can use series resistor or simple approach will be to use a lamp in series. The lightbulb will have a higher resistance when it's hot (higher current, when the battery is dead) and a lower resistance when it's cool (low current, battery approaching full charge).

[PeterF]

Thanks for your replies!

I don't love the complexity and bulkyness of a switched DC/DC solution (as the one I have now) for running up to 4-5 Amps @ 24V.

I just wanted to know if someone could come up with a simpler way of solving it.

[Marco]

You might get away with an ina301-q1 to turn on a p-mosfet for a current limited buck (it has a comparator with hysteresis built in). Low number of components.

[PeterF]

I'm actually persuing the resistor suggestion a bit now.

Will try to see how much power I can practically burn in surface mounted resistors on a quite limited area on the PCB.

[Poe]

I've used sanwal209's lamp solution in the past for very large banks. 

For an industrial product I've used NTC resistors:
https://www.digikey.com/short/qcvd5d
B57234S0150M000

When the power supply turns on, current is limited to about an amp.  As it heats, resistance drops and so might current, but at least it charges faster than just a fixed resistance.

[PeterF]

Thanks!

Perhaps I should look further into the NTC path, but I want something preferably SMD and it should handle at least 2-3 Amps.

Maybe paralelling a bunch of NTC:s could work?

[ocset]

Well if I was you I would just put an accurate current clamp on the output of the charger PSU.
Eg get an error amplifier in there, and shovel a reference into it, then get the error amp to brake your smps (by eg pulling the comp pin down etc) when the current goes above the value XXX.
You would have two error amps…one for normal regulation, and the other for current clamping…and just make the clamper kick in when the overcurrent happens.

[Marco]

If you're just going to burn the power you might as well do it with silicon, it can't survive the high temperatures resistors can ... but they tend to have better thermal coupling to the PCB and you don't really want something to get very hot any way.

As I said though, buck with a ina301-q1 might work with a very low component count. You don't need any voltage regulation loop at all, since it can simply be allowed to go upto supply. Just need to limit current through an inductor with an on/off P-MOSFET without trying to switch it too often.

[Cerebus]

Thanks!

Perhaps I should look further into the NTC path, but I want something preferably SMD and it should handle at least 2-3 Amps.

Maybe paralelling a bunch of NTC:s could work?

Sounds like something as simple as a constant current source made out of a depletion mode MOSFET with a resistor in the source and the gate tied to the bottom of that resistor would do the trick. It will obviously drop out of constant current mode compliance once the capacitors charge. Or, just pick a depletion mode MOSFET with an I_dss of 2-3 A and strap the gate to the source - only do this if you can tolerate the wide I_dss tolerance of a MOSFET.  An IXTA3N100D2 would fit the bill for either scenario. Watch out for the substrate diode if a potential reverse current flow would be a problem.

[NiHaoMike]

If the inrush limiting is used infrequently enough that the efficiency of a buck converter is not worth the space, maybe use some small incandescent bulbs (e.g. car light bulbs) as a current limiter? It would be better thermal wise than resistors or linear current limiting, since a large percentage of the heat is radiated. You can add a MOSFET to bypass the bulbs once the inrush limiting is complete.

[PeterF]

Thanks!

Perhaps I should look further into the NTC path, but I want something preferably SMD and it should handle at least 2-3 Amps.

Maybe paralelling a bunch of NTC:s could work?

Sounds like something as simple as a constant current source made out of a depletion mode MOSFET with a resistor in the source and the gate tied to the bottom of that resistor would do the trick. It will obviously drop out of constant current mode compliance once the capacitors charge. Or, just pick a depletion mode MOSFET with an I_dss of 2-3 A and strap the gate to the source - only do this if you can tolerate the wide I_dss tolerance of a MOSFET.  An IXTA3N100D2 would fit the bill for either scenario. Watch out for the substrate diode if a potential reverse current flow would be a problem.

I actually did a quick experiment with that exact part a while ago, but thought that it wouldn't work here since 24V*3A = 72W that has to be burnt in the MOSFET (in the beginning of charging, when the caps are at 0 V).

Or am I missing something with dimesioning?
(figure 3 here:
https://www.infineon.com/dgdl/Infineon-Application_Note_Applications_for_Depletion_MOSFETs-AN-v01_00-EN.pdf?fileId=5546d4624cb7f111014cd63d1a197d94
)

[PeterF]

If you're just going to burn the power you might as well do it with silicon, it can't survive the high temperatures resistors can ... but they tend to have better thermal coupling to the PCB and you don't really want something to get very hot any way.

As I said though, buck with a ina301-q1 might work with a very low component count. You don't need any voltage regulation loop at all, since it can simply be allowed to go upto supply. Just need to limit current through an inductor with an on/off P-MOSFET without trying to switch it too often.

Thanks!

This is starting to sound appealing to me. I've never designed a buck with a current error amp in this way, do you have any quick example schematic or document link I could take a look at?

[The Soulman]

I'm working on a project where we have a supercapacitor bank with 10 series-connected capacitors.

Below are some questions, I'm not being rude just want to point you in the right direction. 

Did you calculate the total capacitance for and ESR for 10 series-connected capacitors?
Did you implement a way to balance these capacitors?
Most 24V ups's have a 20V battery cut-off, did you calculate the run time wit your capacitors?

Back to your question, if you don't need a high re-charge rate a resistor for charging and a schottky diode connected in parallel
for discharging should work well enough.

[Cerebus]

I actually did a quick experiment with that exact part a while ago, but thought that it wouldn't work here since 24V*3A = 72W that has to be burnt in the MOSFET (in the beginning of charging, when the caps are at 0 V).

Well, for a start you're missing (although you hint at it) that your average power dissipation is going to be 1/2 that simply because the bank is charging. Also, it's not going to be for long, a few seconds, tens of seconds. You can always reduce the current delivered (larger R_d thus larger V_gs and thus lower I_ds) to lower the dissipation and accept a few seconds longer charging time. Engineering trade-offs, as usual. In return you get a cheaper, simpler, more robust (if properly dimensioned) solution than switching converters or whatever.

Edited to add: You haven't said what kind of capacity you're aiming at. I'm assuming here that the hold-up time of you solution is measured in seconds, not minutes. I'm assuming that because you're talking about using super capacitors. If I'm mistaken and you're looking at minutes of hold-up time with super capacitors then you've obviously get very deep pockets and will indeed need the gold-plated option of another power converter.

[PeterF]

I'm working on a project where we have a supercapacitor bank with 10 series-connected capacitors.

Below are some questions, I'm not being rude just want to point you in the right direction. 

Did you calculate the total capacitance for and ESR for 10 series-connected capacitors?
Did you implement a way to balance these capacitors?
Most 24V ups's have a 20V battery cut-off, did you calculate the run time wit your capacitors?

Back to your question, if you don't need a high re-charge rate a resistor for charging and a schottky diode connected in parallel
for discharging should work well enough.

Hi,
you're not rude - thanks!

Yes, the total capacitance is around 31F.

And I have a way of balancing the capacitors implemented as well, using these fancy parts at the moment:
http://www.aldinc.com/pdf/ALD810022.pdf

Anyone have any experience using them?


I've both calculated the run-time and tested it multiple times, so I'm quite happy with it at the moment.
(The capacitors I'm using are basically filling up most space available in the casing, so I can't really go up anymore in capacity)

[PeterF]

I actually did a quick experiment with that exact part a while ago, but thought that it wouldn't work here since 24V*3A = 72W that has to be burnt in the MOSFET (in the beginning of charging, when the caps are at 0 V).

Well, for a start you're missing (although you hint at it) that your average power dissipation is going to be 1/2 that simply because the bank is charging. Also, it's not going to be for long, a few seconds, tens of seconds. You can always reduce the current delivered (larger R_d thus larger V_gs and thus lower I_ds) to lower the dissipation and accept a few seconds longer charging time. Engineering trade-offs, as usual. In return you get a cheaper, simpler, more robust (if properly dimensioned) solution than switching converters or whatever.

Edited to add: You haven't said what kind of capacity you're aiming at. I'm assuming here that the hold-up time of you solution is measured in seconds, not minutes. I'm assuming that because you're talking about using super capacitors. If I'm mistaken and you're looking at minutes of hold-up time with super capacitors then you've obviously get very deep pockets and will indeed need the gold-plated option of another power converter.

Yes I was just very bluntly using the maximum power, because the capacitors take minutes to charge up and not seconds.

And yes, price is not primarly the issue here - since, as you hinted, the capacitors are not cheap in themselves.

I am however looking for as simple and quickly designed solution as possible since time is a factor, and I'd like to have a simple design that is easier to verify and understand for future support of the project.

[tszaboo]

It doesn't get simpler than this:
http://www.onsemi.com/pub/Collateral/NSI50350AS-D.PDF
Connect 3 of them in parallel for ~1A. It will pop, if the capacitors are not charged fast enough from 0, but I think those capacitors will never be really 0. Place a diode in series if you want reverse current protection. Also comes in different packages, and currents.

[Cerebus]

I am however looking for as simple and quickly designed solution as possible since time is a factor, and I'd like to have a simple design that is easier to verify and understand for future support of the project.

Always desirable. My reckoning is that if you can't immediately understand an old design when you dig it a few years later (and hence have forgotten what you did) then either the design is too complicated or the schematic needs a big, clear notes block next to the tricky bit.

Anyway, 31F - so 32.25 mV/s/A  or 248s to reach 24V @ 3A, say 4 minutes. Yeah, that's gonna be a pain. About 9kJ to lose if you go the MOSFET constant current source route.

You could always go for the "Lunar Rover" phase change cooling with a big lump of paraffin wax.    Actually that's not crazy, just for giggles I checked and that's about the amount of heat required to melt 45g of wax at 45-65C. If you've got a defined minimum cycle time you might get away with it.

[PeterF]

Perhaps I'll stick to the SEPIC design I ripped off the app note from above, I'll just need to read up a bit more on how to verify it properly.
It does work, but I'd like to boost the power output a tad, and ensure that nothing overheats in the long run.

And then comes all the EMC verification on top of that 

[PeterF]

It doesn't get simpler than this:
http://www.onsemi.com/pub/Collateral/NSI50350AS-D.PDF
Connect 3 of them in parallel for ~1A. It will pop, if the capacitors are not charged fast enough from 0, but I think those capacitors will never be really 0. Place a diode in series if you want reverse current protection. Also comes in different packages, and currents.

Would that work?

If I'd like to handle >2 (preferably 3-4) Amps with that solution, would it be possible to handle the heat on a PCB?

I gave up on the idea of burning away most energy with a non-switched solution when I realized that I would need to burn so many watts (peaking at ~72W for 3A)

[tszaboo]

Well, it has terrible, <50% efficiency, and it would be hot. But these have NTC characteristics, so they would self regulate the current if it gets hot. In theory. I actually designed this, as a solution into something, in a very similar application: Tickle charging lead acid battery. Never got used, so I did not test it though. The thing is, if you are making a one off, or very small production batch, I take the simplest solution. Designing a proper, trusted SEPIC could take days, especially if you have EMI requirements, if you get it right first.

[Buriedcode]

I realise you don't want a switching solution, but given the amount of heat you'll have to dissipate , I would avoid the linear approach and go with a basic flyback switcher.  Because you are not trying to create a steady smooth supply, you won't have to worry too much about regulation.  Also, current sources are better for charging caps (not wasting 50% of the input in a resistor).

So a coupled inductor (1:1) rated for several amps, a generic current-mode boost converter chip (probably with an external switch at these currents).  Even a constant on-time control circuit would do the job. A simple comparator could be used to stop the charging circuit once the caps have reached the target voltage. 

You could control the input current by setting the peak current on the primary (the on-time), or the off-time.

[Marco]

ina301-q1

The reason I suggested this IC because it does high side current sensing with a built in comparator, whereas I thought all buck LED drivers were low side current sensing (which you presumably don't want).

Looking a little further though, I saw the TPS92515-Q1. A 2A internal FET constant current LED driver with high side current sensing, that should be ideal for you. Drop out is 0.2V max, which I doubt is a problem.