Help designing DC/DC stepdown 400v -> 24v 125A 3000W

[boatman]

Hello everyone,

I am working on a tethered drone power supply. a long wire with 400v will be attached to the drone and I need to turn that into 24v @60A 125max. Yes I know that is a crazy about of voltage and current. I have found some ~3000$ options out there but they are for charging electric cards and are very heavy. I am sure I can make a lighter for less money and more redundant. The Drone is 10 gram/watt, so if I can save 10 grams, I can get away with one more watt of waste heat. also, because it will be mounted to a drone, I can vent heat easily. 

1st idea: Using the amazing power supply designer at https://webench.ti.com/power-designer/ is spat out a very nice UCC28951QPWRQ1 Phase-Shifted Full-Bridge circuit but it looks like I would need to make/order custom inductors and transformers for it. it also doesn't show me the current needs for those transformers, so I would need to work that out as well. There are some other issues with the design it shows me, such as, it cant find FETs that can handle the current and Voltages... I can fond some on Digikey. I am also worried that if it is using ideaFETs in its calculations the other component values might be off. I can also not simulate this design in webench. So with all these issues I was thinking of just designing my own.
Has anyone here make a similar design using the UCC28951 based circuit from webench? or can offer any ideas for sourcing/making those transformers?

2nd idea, I would like to use a couple Current-Mode PWM Controllers working at a high frequency to allow for smaller transformers. if I can get maybe 6 of them to work together each would only need to output around 20A. This would give it good redundancy. and allow for easier testing. I was thinking of a MAX5071A?
Does anyone have a suggestion on PWM controller chip that would be good for this application? or even a better idea for circuit topology?

Side note: The tether will be around 150m long. It is basically a transmission line at that point and I am hoping that a get away with just using a 2mF. yes I know a ~500v 2mF cap is a lot. I dont know what the transient current needs of the drone will be until we start testing.

More Side Notes: this will be directly attached to the backup batteries on the drone. This will help smooth out any voltage drops and act as a backup. I was thinking of adding a UPS, but that would add too much weight and from my testing the batteries handle everything just fine without.

Transformer info:
Ns1toNp = 1, Ns2toNp = 1, Lp = 1.3 mH
Ns1toNp = 105.49 m, Ns2toNp = 105.49 m, Lp = 288.38 µH
Ns1toNp = 100, Lp = 20 mH, Leakage_L = 20 µH

Looking forward to hearing everyone's thoughts.

[mikeselectricstuff]

Take a look at Vicor for high-performance DC/DC module solutions.
For minmum weight you need highest frequency, and you'll likely only find that in a module solution

For example this 1600w converter : https://www.digikey.co.uk/en/products/detail/vicor-corporation/BCM6123TD1E2663T00/7389608
Claimed 97.4% peak efficiency, 41g weight ( plus heatsink!)

You're not going to get anywhere near that performance in a discrete soluiton.
and you can parallel them, so that  gets you your 3kw very easily, and assuming 95% real-world efficiency, only 150w of heat to deal with

Note some Vicor devices are a bit weird in that they are fixed-ratio rather than constant-voltage, not sure why.

 At least you'll have the downdraught for cooling...

[ejeffrey]

Me trying to parse Vicors explanation is that by having fixed interleaved phases they can cancel switching ripple extremely efficiently and maintain very high converter bandwidth.  This lets you minimize the lov voltage capacitance you need on favor of capacitance on the input rail where you only need 1/N^2 as much capacitance to get the same energy storage.  The higher voltage capacitors are larger than low voltage ones but you generally win on area.

So their recommended operation  for these converters is to have an approximately 1:1  regulated converter (say 48-40V) followed by a fixed current multiplier.  They claim this can still be overall lower volume and higher efficiency than a single regulated step down converter.  They certainly are high performance designs and we have a project out for prototype build right now using them.  I don't know how much of their performance is actually due to this architecture, but that is their claim.

[boatman]

Vicor for high-performance DC/DC modules you say?  I did see those on digikey but didn't know they could do parallel operation =0 Didn't read the data sheet for them closely enough.
I will check those out. Thanks. 

[mikeselectricstuff]

There is an appnote on parallel operation :
https://www.vicorpower.com/documents/application_notes/vichip_appnote16.pdf

[PCB.Wiz]

400v ... I need to turn that into 24v @60A 125max.
Does anyone have a suggestion on PWM controller chip that would be good for this application? or even a better idea for circuit topology?

More Side Notes: this will be directly attached to the backup batteries on the drone. This will help smooth out any voltage drops and act as a backup. I was thinking of adding a UPS, but that would add too much weight and from my testing the batteries handle everything just fine without.

Interesting application.
It is highly focused, so you can use that to your advantage.

You can move the battery control to the ground side, where weight/losses matter much less.
ie you send a variable appx 400V that is scaled by a fixed ratio, in the drone, to what is needed.
Maybe fault current sense can be managed on the ground too ?

If you scale by an exact amount, you can create a multiple phase sync'd design that can greatly reduce the ripple currents and so reduce the capacitors.

eg a 16 phase design, (assuming no loss) needs nominally 384V in for 24V out.
All FETS are always conducting at 6.25%, one after the other, so the cable current is about 8A, plus inductor ripple.
Each inductor and FET stage needs to manage up to appx 8A, so they are simpler to design and optimize.

Addit You could also check a dual 4 phase design. First stages uses 4 x 25% conduction time to do 384 to 96V and second stages use 4 x 25% for 96V to 24V
This allows lower RDS fets as the voltage drops as the current increases.

[T3sl4co1l]

Yes, phase interleave is very attractive, even at fairly modest power levels I would say.  LLC resonant is probably the best architecture, saving some (but not all, actually*) switching loss, though I'm not aware of any phase synchronized controllers available and you might be cooking your own at that point (there are STM32, DSP something or other, etc. dev platforms to get started with).  PSPWM has the advantage that it's easier to synchronize, but does incur some losses.

*SiC is preferred over Si-SJ for this reason; SJ has one loss mechanism that can't be reduced by ZVS switching.  SiC do not suffer this mechanism.  Or use GaN, if you're feeling up to the challenge of such speed..!

Losses at least shouldn't be a problem, with such airflow available.  There is still some priority, as you don't want to be putting in huge heat sinks, pipes and spreaders to get that heat around and out.

Tim

[Wolfram]

*SiC is preferred over Si-SJ for this reason; SJ has one loss mechanism that can't be reduced by ZVS switching.  SiC do not suffer this mechanism.  Or use GaN, if you're feeling up to the challenge of such speed..!

If you're talking about Coss hysteresis losses, these are not exclusive to SJ structures, but present in most MOSFETs, also WBG ones. They are particularly bad for MEMI SJ structures, but can also be significant at higher frequencies for other transistors, particularly GaN MOSFETs and SiC JFET cascodes. https://infoscience.epfl.ch/record/273264/files/Insights_Coss_Losses.pdf https://ieeexplore.ieee.org/document/9265823

[T3sl4co1l]

The one I found particularly motivating was:
COSS Measurements for Superjunction MOSFETs: Limitations and Opportunities; G. D. Zulauf, J. Roig-Guitart, J. D. Plummer and J. M. Rivas-Davila
which it looks like used to be online and has since been pulled. :sus: But in any case, suffice it to say there's a couple watts attributable to the effect, even at fairly modest frequencies (low 100s kHz), for average size devices say at 400V 5A switching, and much less occurs in planar types, including now-obsolete* types, SiC, and GaN to some extent.

*I think they just re-marketed them, actually; "linear" type MOSFETs have all the hallmarks of traditional planar designs.  But "they" is far fewer than it used to be; I think only IXYS is making them now.  So, doubly marked up in price...

Sure, it's not "none", there are other effects, like substrate effects in GaN, but as seen from the above plots, it's nowhere near as bad as Si.  Si you can see the switching on the scope, at very ordinary time scales (us).

The exception being cascode devices, where the internal node incurs losses in a similar way to SJ, and for similar reasons I suppose.  A similar catch applies to planar Si of course: they're just so goddamned big to get comparable ratings, that you burn as much in gate drive power.  (Which is to say: SJ is absolutely an improvement! A remarkable one at that; just not a unilateral one.)

Tim

[boatman]

16 phase design, (assuming no loss) needs nominally 384V in for 24V out.
All FETS are always conducting at 6.25%, one after the other, so the cable current is about 8A, plus inductor ripple.
Each inductor and FET stage needs to manage up to appx 8A, so they are simpler to design and optimize.

I like this idea. Do you, or someone else, have a chip recommendation or data sheet?

[mariush]

Do you need a single 125A rail or could you maybe have 2 smaller power supplies, like one for each motor or something like that?

TME.eu has a 2160w psu from Meanwell, BIC-2200-24-CAN, up to 93% efficient : https://www.tme.eu/en/details/bic-2200-24can/built-in-power-supplies/mean-well/

Remove the ac rectification and you'd have up to 2.1kW in one psu

[mag_therm]

Is there alternative to sending DC up the link?
Now I am a staunch proponent of DC distribution, but there is an alternative.
(From my induction heating background and sometimes using Pyrotenax and coax for special applications)
Generate about 700V or more of trapezoidal AC in range 15 ~ 25kHz on the ground such that it is within ratings of the RF coax. Use that for the link. Will be lighter than 400V DC.
Up the top, use an amorphous core step down transformer with fast diode rectifier (I did some ferrite and amorphous hand held transformers for brazing ,one rated at about 20kVA that could be held in one hand although it was water cooled) 0.9 Tesla @ 25 kHz

You need 12 V @ 125A DC intermittent? That is 1500 Watt, a bit over 2 Amp in the coax. A termination network for the trapezoidal AC might allow elimination of the DC choke too.

[coromonadalix]

to me the Vicor fit the bill,  expensive a bit,  but as previously written, you'll get nowhere as theses performances  ....  with DIY  solutions

[PCB.Wiz]

16 phase design, (assuming no loss) needs nominally 384V in for 24V out.
All FETS are always conducting at 6.25%, one after the other, so the cable current is about 8A, plus inductor ripple.
Each inductor and FET stage needs to manage up to appx 8A, so they are simpler to design and optimize.

I like this idea. Do you, or someone else, have a chip recommendation or data sheet?

Because the regulation is done elsewhere, you just need 16 pulses, so a sync binary counter and decoders will do this, feeding decent gate drivers with shoot through protection inbuilt.
74HC4514 can do 16 decodes in one package, or you can use 2 x 74HC138/238 or 74HCS138/74HCS238  8 bit decoders in two packages + counter.

The spare Enable line in these, can be fed from a current limit detector, to give immediate fault drive removal.

[NiHaoMike]

Another idea: use common universal motors with simple PWM control from the DC supply, easy way to at least prove the concept. Could replace the universal motors with HV BLDC later on.