Hi all,
while designing a large isolated DC-DC converter (40 to 60V in, 360V out, 4kW), I ran into worries about flux walking.
The design is a full bridge of MOSFETs, one per leg, no paralleling, using MOSFETs rated at 100V and 1.4mΩ. One driver IC per MOSFET, located in its immediate vicinity, with signal coupling from the control IC to the drivers done by means of a small pulse transformer. Then one single big transformer, a bridge of SiC diodes, choke, filter capacitor. The doubts right now now are how to close the control loop.

My goal is to achieve high efficiency in this converter. Weight and size are no big concern. So I chose a large ferrite core (two E70 pairs stacked), with 3 primary and 30 secondary turns of Litz wire having a large number of strands. Secondary-primary-secondary layering. Running at just 20kHz. The result of this is well under 1% total transformer losses at all power levels, and also rather reasonable switching losses. Assuming that my calculations are right, of course... But there is a side effect: The primary inductance is pretty large, resulting in the magnetizing current being rather small in comparison to the load current. So I have doubts about whether peak current control is sufficient to rule out all possibility of flux walking trouble.

There can always be assymetries. Differences between MOSFETs, drivers, rectifier diodes, but there could also be some small undesired coupling between the power and control circuits, causing the conduction cycles of one side to be systematically longer than those of the other. When the magnetizing current is a good part of the total current, then peak current control certainly can handle that well! But when magnetizing current is small relative to load current, I fear that asymmetries and interference from the power circuit MIGHT end up causing enough flux walking, so that in one polarity the MOSFETs turn off under much higher peak load than on the other side. HOW, you ask, when peak current control is supposed to hold those peaks equal? Well, it's not instantaneous. And ferrite can saturate rather sharply. So, when the core goes into saturation, and the current starts shooting skyward, the current sensor has to detect that, the control IC has to react, its output has to switch, the signal has to propagate through the pulse transformer, the driver ICs, then those big MOSFETs have to turn off - and that whole process might take 300ns or perhaps more, and in that time the current will keep rising sharply!

Maybe peak current control is good enough to keep the MOSFETs from burning out, but two of the four might end up working with much higher switching loss than I had calculated. And that's bad for reliability, for EMI, and also for the overall efficiency of the converter.

At this relatively low input voltage and high power, I don't want to use a current sense resistor. It would dissipate something like 100W, and it might be hard to get its ESL low enough. But if I use current sense transformers, I need to use two of them, in the two half-bridges, so each core has time to reset, even when during operation at minimum input voltage the deadtime gets very short. Any differences between these two current sense transformers will directly cause flux imbalance in the big transformer, by making the control circuit extend the conduction time on one side.

I was toying with the idea of simply placing a DC-blocking capacitor in series with the primary, and then use average current control. But I would need roughly 200µF, with an ESR of well under 1mΩ, and low enough ESL, and that looks like a bank of perhaps 20 polypropylene capacitors of 10µF each in parallel. It's a bit inelegant. And I would still need peak current sensing for fast shutdown in case of trouble!

Another idea I had is deliberately increasing the magnetizing current, so that it's a more noticeable part of the total current, and the peak current controller can "see" it better. Putting a small airgap in the core would do that, and of course it would also increase the saturation current by the same proportion. But it would also cause a noticeable increase of switching loss and EMI, so I don't like that solution too much.

If you build such circuits, how do you handle this issue? Do you simply trust that peak current control will keep the flux centered well enough to prevent all trouble? Or do you put a capacitor in series to be safe? Or do you use a DC current sensor (a Hall sensor?) to sense the DC component in the primary current, and have the control circuit null it? Does that even work, in the presence of the large AC?

And a small collateral question: Is it safe and sound to use the IPTC014N10NM5 in this application? I'm a bit old-school, and have some trouble accepting the idea that such a small thing can switch 120A at 60V, but the datasheet and also my maths seem to say that it can...

Manfred

 

I would say that it's a risky plan to try to run that much
Current through a SMD Printed-Circuit-Board.
Especially when You say that you have plenty of space to work with.
You need some big-nasty SIC-FETs in a TO-247 Package.

I also would not recommend a "Full-Bridge" setup,
especially if You want maximum efficiency.

With over-sized Cores, I would estimate that "Flux-Walking" is very unlikely to be an issue,
on the other hand, using a "Pulse-Transformer" can potentially be problematic.
There are plenty of totally isolated FET-Drivers that would do a fine job, without a Pulse-Transformer.
Then Drive them with a small, stand-alone, Isolated, Switching-Regulator-Module.
You need the Low-Voltage for the controller in any case, so You may want to get 2 of them.

I like "Current-Based" Controllers for their ability to
"cover-up" the problems and quirks, and just work, despite a possibly non-ideal design.
And, they don't necessarily have to have massive power losses as You described,
~100-Watts seems really excessive, are You sure You didn't make a mistake on the Math ?
You will have that much Voltage-Drop just on the Circuit-Board if You go with SMD Construction.
The Voltage-Drop in a "purposefully-under-sized" Ground-Wire
is all you'll probably need for a Current-Sense-Resistor.

An "Old-School" Though-Hole Controller-Chip will do a fine job,
and can be built on Perf-Board, with all Through-Hole-Components,
instead of taking-on the tedious job of designing an SMD Circuit-Board.
.
.
.
Just an example to give You some ideas ..........
.

.

.
.
.

 

while designing a large isolated DC-DC converter (40 to 60V in, 360V out, 4kW),

40V 4kv = 100A Use thick copper. Do not use "thermals". Run parallel copper in all layers. Most board houses will run thick copper on the internal layers if you ask.
I do not know your plan but...... 0 to 200A or 80 to 120A ramp?? Think about 100A at 1.4m ohm = watts. I think you need lower RDSon. In through hole 100V parts I think 1.05m ohm is the best.
Using bridge there should not be ring so you can use low voltage parts.
Using bridge the two MOSFETs are in series, so the RDSon is doubled.
The idea from LowQCab of using a center tapped transformer will have 1/2 the RDSon loss because the current is passing through only one MOSFET. But the voltage will be 2x+ring. You will need to sit on the ring hard.
IRF100P218AKMA1 100V 1.28m ohm
Going to a 200V MOSFET and the RDSon increases.

Can you do surface mount?
SVC6H890N Cannot find in stock. 0.5 m ohm 80V

Have you looked at modules?

 

Dear "LowQCab",

> I would say that it's a risky plan to try to run that much
> Current through a SMD Printed-Circuit-Board.

In fact I don't plan to do that! It would need to be a thermally enhanced board with super extra thick copper, and I cannot make those, and ordering it from overseas is too much hassle.

My plan is to solder each of those SMD MOSFETs with its thermal pad to a copper base plate, cut from a piece of bus bar, and use those assemblies just like leaded parts! Simply bolt them by these base plates to a heatsink, with a silpad in between, in close proximity, and build the whole high current / fast signal circuit (MOSFETs, bypass caps, transformer primary) with wide straps of 0.5mm thick copper sheet. Maybe 20-30mm wide or so.

> You need some big-nasty SIC-FETs in a TO-247 Package.

SiC-FETs, for low voltage and high current? Can you point me to any? All I know are good for high voltages, but have much too high RdsON to use at 40V, 120A.

And TO-247? It has at least 10nH source lead inductance. That's completely unusable without a Kelvin lead. With Kelvin lead, perhaps... but not as good as those tiny SMD MOSFETs with their 7 parallel and very short source leads!

I would absolutely LOVE to use MOSFETs in packages that are easier to handle, and more robust, than those SMDs. It's just that I can't find any! And I don't want to build parallel arrays of MOSFETs, because they just won't properly share the switching loss, no matter what one does. I use paralleled TO-220 MOSFETs in applications where the point is lowering the RdsON, and switching loss is irrelevant. But in a 20kHz circuit, switching loss is very significant, so paralleling doesn't work too well. I know that millions of Chinese inverters use paralleled TO-220 MOSFETs, but I also know that they die like flies...

That's why I want to use single MOSFETs, and the only suitable ones I found are SMD ones.

> I also would not recommend a "Full-Bridge" setup, especially if You want maximum efficiency.

What's the rationale behind that? I decided in favor of the bridge, precisely because it is what produces best efficiency! And that's for the following reasons:

- A push-pull needs MOSFETs of more than twice the voltage rating, compared to a bridge. These have 3-4 times as much RdsON. So despite having only a single MOSFET loss per side in the push pull, instead of two in series with the bridge, the bridge's conduction loss is lower.

- A push pull needs its MOSFETs to turn off against more than twice the voltage, compared to the bridge. Also the higher voltage MOSFETs are slower. The result is higher total switching loss in the push pull, despite having just 2 devices.

- A bridge dumps the energy of the leakage inductance right back into the supply, via the MOSFET's body diodes. A push-pull circuit instead dumps it either into a snubber (loss), or into the MOSFETs in avalanche conduction (loss and risk).

- The split primary of a push pull transformer has a lower copper utilization than the single primary of a bridge transformer. So, at a given transformer size, transformer efficiency is also better for the bridge.

In favor of the push pull is just the simpler circuit, particularly the driving scheme. At low power level that's very important. But at 4kW, the simplicity advantage isn't that important.

So, a bridge it will be, unless you can convince me otherwise! ;-)

> With over-sized Cores, I would estimate that "Flux-Walking" is very unlikely to be an issue,

Given the low operating frequency I chose, even at low core loss the operating flux density isn't that low. Even a modest amount of flux de-centering will start to cause single-sided saturation.

> on the other hand, using a "Pulse-Transformer" can potentially be problematic.
> There are plenty of totally isolated FET-Drivers that would do a fine job, without a Pulse-Transformer.

Yes, that's right. I might change my position on how to drive my MOSFETs. I preliminarily chose those individual drivers (TC4422A, TC4452 or similar), with pulse transformer signal coupling, because they are quite powerful (9A peak current), and the whole scheme is pretty fast. Optosiolated drivers are a bit weaker and slower. High driving current is very useful to speed up the switching and reduce switching loss, even at 20kHz, and faster signal propagation is particularly useful for overcurrent protection in case of faults. But I have to admit that it's tempting to use optodrivers, for the simplicity.

I don't want to use high/low side, non-isolated drivers, because I want galvanic insulation between the controller and the primary power circuitry.

> Then Drive them with a small, stand-alone, Isolated, Switching-Regulator-Module.
> You need the Low-Voltage for the controller in any case, so You may want to get 2 of them.

My intention in that regard is to make a small flyback regulator, running off the 40-60V input, that provides multiple insulated outputs. An UC3844 driving a small, moderate voltage MOSFET. Each additional floating output takes just an extra winding of a few turns, a 1N4148 diode and a capacitor. Voltage tracking between them is pretty good.

Another option is to buy several ready-made small DC-DC converters. They are cheap these days. But they need to have low enough input-output capacitance, and be resistant to high dv/dt. So I prefer making my own auxiliary supply! Better safe than sorry.

> I like "Current-Based" Controllers for their ability to
> "cover-up" the problems and quirks, and just work, despite a possibly non-ideal design.

Yes, and it's easier to stabilize them. With my voltage-mode switchers I usually have to cut-and-try to get the control loop stable, despite all calculation. With current mode it usually works right away.

> And, they don't necessarily have to have massive power losses as You described,
> ~100-Watts seems really excessive, are You sure You didn't make a mistake on the Math ?

I'm pretty sure I got it right. I was assuming the use of a current-mode control IC like the UC3856. It typically uses 1V peak for current sensing. I have some of those at hand, so I'm partial to using them. In my application the peak current will be 120A or so. When the input voltage is at the low end of the range, current is almost continuous, ramping between roughly 85 and 120A. Roughly 105A RMS. The sense resistor would need to be roughly 8mΩ, and ower dissipation would be 88W. Right?

I could of course use a more modern IC, that uses just 0.5V for current sensing. Or program the UC3856 for 0.5V, which is also possible. But noise immunity gets worse. And the power dissipation in the sense resistor would still be too high.

So I really need to use current transformers. One or two, depending on how I do it. For peak current control, I don't see how to do it with just one.

> An "Old-School" Though-Hole Controller-Chip will do a fine job,
> and can be built on Perf-Board, with all Through-Hole-Components,
> instead of taking-on the tedious job of designing an SMD Circuit-Board.

Yes, and that's how I plan to build the control circuit. And the high power part, well, "Stonehenge Style", using a chunky heatsink as base plate!


> Just an example to give You some ideas ..........

Your first example: That circuit doesn't implement current mode control. It's a voltage mode control circuit, with average current limiting. There is no active flux centering whatsoever. And I strongly disagree with the MOSFET selection! Those 650V-rated SiC-FETs, with their 80mΩ (when cold!) RdsON, are totally out of place in a 14.5V, 40A circuit. At full power, running hot and with increased RdsON, they will cause close to 200W of total conduction loss! That's about 35% conduction loss!!! They will need very good heatsinking, just to survive.

Also, 5W gate resistors? For devices rated at just 43nC gate charge, and operating at 10kHz? Even the smallest SMD resistor you can find would be loafing along!

SiC cascodes like those are great in a certain range of applications: Fast switching of relatively high currents, at several hundred volt and higher. Along with easy driving. But in low voltage circuits, much cheaper common silicon MOSFETs are much better.

Well, my questions remain open! I still haven't decided whether to use a series capacitor and average current control with a single current sense transformer, or save the series capacitor and use peak current control with two separate current sense transformers.

Manfred

 

Controlling peak current might help prevent flux walking under certain circumstances (such as un-matched switching on alternate duty cycles) but the real solution is to have the same volt-seconds across the coil (core) on both half cycles. Of course a little gap in the magnetic path makes the circuit a little less sensitive to this effect.

 

my questions remain open!

Current mode stops flux walking. I use Darin resistors to monitor current. I love CTs but they do not pass DC and thus will not work well if flux walking is the problem.
I use JLPCB in China and they make 2oz copper on all layers.
I have not done an Aluminum PCB. I think they should make a great heatsink.

At near 100A, you might think about two power supplies running in parallel. In the lab next door, they have four supplies running in parallel. (phase shifted 25%) The input and output ripple is very low, using small capacitors.

 

Dear Ron,

> 40V 4kv = 100A

Yes, 100A worst-case average, plus a little extra to cover losses. I want 4kW at the very output, not at the input.

> Use thick copper. Do not use "thermals".

Definitely thick copper. Just not so thick as to run into excessive skin effect.

> Run parallel copper in all layers.

I was writing my previous reply while you were writing this. You surely have just read that I do not plan on mounting those SMD MOSFETs on a PCB!

> Most board houses will run thick copper on the internal layers if you ask.

But at what price? And it's too much hassle. I only want to build a single unit.

> I do not know your plan but...... 0 to 200A or 80 to 120A ramp??

The latter.

> Think about 100A at 1.4m ohm = watts. I think you need lower RDSon.

Let's make that about 1.8mΩ, considering that the silicon will be warm. And 105A RMS, to cover losses, and the higher RMS per average for the ramp. So, 105A for half the time, per each of the 4 MOSFETs. That makes 10W per MOSFET, 40W total conduction loss. 1% of the total power. That seems acceptable to me. 1% was my design goal.

> In through hole 100V parts I think 1.05m ohm is the best.

But through-hole parts suffer too much from source lead inductance. The switching loss goes way up!

> Using bridge there should not be ring so you can use low voltage parts.

I have my doubts as to how well I can control the overshoot. Even with short, wide copper strap interconnections, and a lot of bypass caps soldered right over each half bridge, I expect several nH of stray inductance in the clamping circuit. So there will be some overshoot. In one detailed calculation I did, I arrived at roughly 16V of overshoot. That's on top of the max input voltage, 60V, so 76V total. 80V-rated MOSFETs should just barely do, but I feel safer using 100V rated ones.

> Using bridge the two MOSFETs are in series, so the RDSon is doubled.

Yes. Considered above.

The idea from LowQCab of using a center tapped transformer will have 1/2 the RDSon loss because the current is passing through only one MOSFET.

But that MOSFET will need at least a 200V rating, and maybe 250V. So, much higher RDSon. More than twice as much.

> But the voltage will be 2x+ring. You will need to sit on the ring hard.

Yes. Heavy snubbers, throwing away some energy. And a 120A snubber isn't that easy to make. It needs a total inductance of no more than a very few nH, or it won't be effective enough.

IRF100P218AKMA1 100V 1.28m ohm

I had looked at that one during project definition. Did you look at the switching speed rating? With 50V and 100A, which is quite representative of my project, and very low gate resistance, 110ns rise time, 120ns fall time, in resistive switching. But mine is inductive switching... Anyway we can use these figures to compare to the SMD transistor I pre-selected (I might still change this decision, if I find a better one): For the same voltage and current, and an even slightly lower gate resistor, 30ns rise time, 30ns fall time! 4 times as fast! That means A WHOLE LOT less switching loss!

That's why I preferred the SMD MOSFET. The difference is simply due to the much lower source lead inductance of the SMD package. Of course, to take advantage of it, I need to connect my driver's ground return directly to the first source pin of each MOSFET.

So, while that TO-247 MOSFET has marginally better RDSon, saving maybe 4W total in conduction loss, the SMD MOSFET allows dramatically faster switching, saving AT LEAST 50W in total switching losses! At least that's the potential loss saving I see. I don't know whether I will get my wiring and driving good enough. If anybody on this forum has practical experience with such high current SMD multipin MOSFETs, it would be nice if he could pipe in and tell me if this really works, or if there are pitfalls I haven't seen! I have not yet used any such SMD MOSFET in such a high current application. I simply don't know if it's realistic to switch 120A with such a device, and really get much lower switching loss than with a TO-247. Just my maths say so. But I have quite a bit of experience building RF power amplifiers, so I know about the importance of source lead inductance also from the practical side...

> Going to a 200V MOSFET and the RDSon increases.

Yes, a lot.

> Can you do surface mount?

I can, but I don't enjoy it. I'm nearing age 60, and my eyes aren't what they used to be. Let's put it this way: Using some 1206 passives is no problem, while 0805 starts to get critical, and I won't use 0603 or smaller. An IC with 1.27mm pin pitch is OK, while one with 0.5mm pin pitch is a curse. In many of my more recent projects I have combined through-hole ICs with 1206 passives, on single-layer boards. That's quite easy to do, and very effective. I often use that same combination on perfboards.

SVC6H890N Cannot find in stock. 0.5 m ohm 80V

That one is a naked die! No way for me to use it. And available only by special order. I guess one has to order a million of them, for the factory to make them. After installing and bonding such a die, probably the total RDSon will be up to more than 1mΩ.

> Have you looked at modules?

Yes, but haven't found any usable ones. A lot of modules are available for line voltage, but not for low voltages. Also they tend to be expensive.

I have also looked at MOSFETs in SOT-227 ISOTOP packages, with screw terminals and Kelvin source connection, but there are few and they are less good than the modern SMD ones.

Manfred

 

> Controlling peak current might help prevent flux walking under certain circumstances (such as un-matched switching on alternate duty cycles) but the real solution is to have the same volt-seconds across the coil (core) on both half cycles.

Easier said than done! The problem is imperfections of all sorts.

> Of course a little gap in the magnetic path makes the circuit a little less sensitive to this effect.

I might just do that. Just a small gap! Increasing the magnetizing current to 10% or so of the max load current.

 

The Multi-Pin TO-262 Packages are thinking outside the box, 2-Thumbs-Up !!!
But I think that You will have to put so much HEAT into your Copper in order to melt Solder
that it will also melt the Plastic-Package.
Copper can store a massive amount of HEAT Energy,
and it will take a long time for that Energy to dissipate,
so the Package will have to withstand that HEAT for a too long period of Time.

 

> Current mode stops flux walking.

Well... it has to! The point is whether it stops the flux walking BEFORE reaching saturation, or only a little into it! If it's the latter, then there will be increased switching loss.

And exactly at what point or level it stops the flux walking, will depend on how strong any existing asymmetry is, and how large the magnetizing current is compared to the load current...

> I use Darin resistors to monitor current.

Darin resistors? You caught me there. I have never heard of them. Neither has Google!

> I love CTs but they do not pass DC and thus will not work well if flux walking is the problem.

Yep, tricks are needed to sense DC with CTs, or in any case a more complex circuit. If I go the peak current control way, without series capacitor, I will probably use two CTs, like this:


R1 along with the turns ratio defines output voltage per current. R2 is for core reset, during the time each high side MOSFET is off. R2 needs to be high enough so that just the small magnetizing current creates enough voltage across the CT, to be equal or greater than the voltage appearing during conduction time. If the CTs have enough core loss, R2 with its two diodes might not be required. But if the CTs have too much core loss, they might not reset properly, and possibly start running into saturation at the end of each conduction cycle.

Any difference between the two CTs, or the two forward diodes, will directly make the current controller cause a peak current difference between the sides. That's a worry I have. At least the sensing resistor is just one for both sides.

> I use JLPCB in China and they make 2oz copper on all layers.

Good to know. I just had a look, and they even offer shipping by mail! That's important for me, because the flashy courier companies don't ship to my neck of the woods, but the postal service does.

> I have not done an Aluminum PCB. I think they should make a great heatsink.

Neither have I, I mean, never done an own design on them. But in my professional work I have worked with them, both aluminium and copper base. The base is good, but the Achilles heel is the thermal conductivity of the insulation. It's typically worse than thermal grease, mica, and silpad! In fact, depending on exactly which kind of aluminium base PCB gets used, it would be too bad to handle the dissipation of the SMD MOSFETs I'm considering. That's why I prefer soldering them to a copper base plate, and bolting that one to a heatsink, via a Silpad. The Silpad area is then much larger, achieving low enough thermal resistance (and of course a few pF higher unwanted capacitance...).

> At near 100A, you might think about two power supplies running in parallel.

Yes, or four of them, or even a dozen very simple flyback converters, each one 30° away from the next one. I have considered all those options, and the jury is still out. But right now I'm leaning towards building a single big unit. The reason is efficiency. Bridges end up more efficient than push pulls, or (beware!) flybacks. But building several complete bridge circuits, each with its own controller, drivers, FETs, transformer, current sensing, rectifier, choke, etc, tends to need a lot of parts.

> In the lab next door, they have four supplies running in parallel. (phase shifted 25%) The input and output ripple is very low, using small capacitors.

Yes - as long as you can get all those circuits close enough together! Otherwise the inductance between them will again require more capacitors. But the idea of phase interleaving is what makes flybacks a possible contender, despite its fierce ripple both on the input and output. Still, flybacks with passive snubber are too lossy, and flybacks with active snubbers are again too complex, to use a dozen of them. And with fewer than a dozen, peak current becomes a problem, at least when using TO-220 MOSFETs.

Other than that, I love discontinuous current mode flybacks! Very simple, and very well behaved.

 

> The Multi-Pin TO-262 Packages are thinking outside the box, 2-Thumbs-Up !!!
But I think that You will have to put so much HEAT into your Copper in order to melt Solder
that it will also melt the Plastic-Package.

I have done that with RF power transistors in plastic cases. It's not so different. 63-37 solder melts at 183°C, if my memory isn't kidding me. It's a simple matter to apply some solder paste, put the transistor on top, with some contraption that holds it in place and presses it slight down, then heat the copper block from below. As soon as the solder melts and the transistor seats down in place, remove the heat source. The plastic won't get over 200°C, and that only briefly. It's rated for more than that.

If worried about the time spent hot, it would be possible to cool down the copper quickly, with a small amount of water, but I have never found that necessary, and I would worry about thermal stress.

And that big SiC FET you linked has RDSon of 15mΩ when cold. That's more than 10 times worse then the little and very much cheaper SMD silicon FET! A quad of them would incur in roughly 400W of total conduction loss. That's not usable. That SiC FET is great for its intended application range, but that range doesn't include my application!

 

Everybody-place-your-Bets !!!!!

I say You will never get over about ~93% Efficiency no matter what You do,
so it comes down to just how much Time and Money you're willing
to spend to get that last 0.5%.

I prefer over-kill, robustness, and extra-stability, wherever possible,
at the expense of the last ~7 to ~8% in efficiency.
.
.
.

 

I say You will never get over about ~93% Efficiency no matter what You do,

It looks like I can achieve 95% in this converter. About 1% MOSFET conduction loss, 1% switching loss, less than 1% total transformer loss, slightly over 1% rectifier loss, and roughly 1% in various assorted other losses, including the choke, capacitor ESR, and the power needed for control and driving.

When operating at partial power, the percentual loss for control and driving gets higher, but several of the other losses get percentually smaller, so I hope to achieve 95% over most of the power range!

I'm not asking for any more than that... ;-)

 

Go for it !!!
I'd love to see the results !
However, I've yet to see Theory, and careful Math calculations,
exactly match what happens in reality when it comes to designing Switching-Supplies.
It can become very time consuming and frustrating.
.
.
.

 

However, I've yet to see Theory, and careful Math calculations,
exactly match what happens in reality when it comes to designing Switching-Supplies.
It can become very time consuming and frustrating.

I know... Been there, done that, quite a few times!
The older I get, the closer the behaviour of my circuits matches the calculations - but never fully!
And the bad side of getting old: The longer I take to actually build anything!

Handling 120A with single SMD devices is a new area for me, and so is a bridge circuit that works with such a high load current to magnetizing current ratio as this one. That's why I poured my worries onto this forum!

I'm making up my mind to go the standard peak current control path, using an UC3856, and two separate current sensing transformers to sense the current in the two high-side MOSFETs. And no series capacitors. And if problems show up at testing, see how to fix them.

 

As an alternative , how about adding one or more hall-effect sensors to verify that the flux dances back and forth equally, rather than tending to saturate more in one direction than the other. Then a separate, controlled, scheme to re-center the magnetization when that becomes needed?
Not by any means an alternative to the very best design, but with real world parts and not needing to be the ultimate in cheapest possible product, it is an option to consider.

 

As an alternative , how about adding one or more hall-effect sensors to verify that the flux dances back and forth equally, rather than tending to saturate more in one direction than the other. Then a separate, controlled, scheme to re-center the magnetization when that becomes needed?
Not by any means an alternative to the very best design, but with real world parts and not needing to be the ultimate in cheapest possible product, it is an option to consider.

There are two questions/problems to this approach:

- Where do I place a Hall effect sensor, or two? To actually measure the flux density, I would need to airgap the core, with a gap wide enough to insert the sensor. That would make the inductance go way down, and magnetizing current way up, increasing losses. If instead I place a Hall effect sensor attached to the ungapped core, or in a small recess machined into the core, it will see almost no flux until the core saturates, and at that point it will suddenly get a lot of flux. So it would act as a saturation sensor, instead of a linear flux sensor. It would not allow keeping the flux centered without touching saturation.

- Hall effect sensors (at least the ones I know) aren't fast enough for a square wave or triangle wave at 20kHz. The question is then: Will this lack of speed act as a linear low pass filter, so that the sensor still correctly measures the average (DC component) of the flux density? If so, great! But I don't know if that would be so.

 

GOOD POINTS about hall sensors. They would not be the best choice, I see that now. I admit to making an error, so I offer an alternative.: A separate winding over the segment of core most likely to lean toward saturation, and the waveform would indicate less demagnetization in one direction if that started to happen. And because the monitor winding would not be part of the power system it could just serve as a trigger to alter the drive in one direction just a bit to back off on the saturation trend. It would be a "residual flux" control servo, running independent of the rest of the system. Very uncommon but possibly starting a trend if perfectly matched components did not become common soon enough.

 

Hello there,

Hall effect sensors work good enough for sensing current in the current paths which can act fast to detect overloads. However, a good way to reduce flux walking is to integrate the voltage across the primary of the transformer and use that as feedback to gently control the pulse widths of the bridge transistors. This can also reduce audible noise coming from the output transformer which can really be very annoying to workers that have to stay close to the converter on a daily basis.
If the voltage integrates positive, then the imbalance is in one direction, and if it integrates negative, it's the other direction. When it is zero, the volt seconds are the same each way. The change should not have to be too large or it could indicate that something more serious is wrong.
You'll have to work out the details of the implementation for your setup.

 

I'm looking at Hall sensors again. I have a few CSA-1V here, which are rated for ±10mT range, 100kHz bandwidth and 8µs response time. They sense flux parallel to the IC. Maybe they can pick up enough stray field from the core to give a usable signal, if mounted flat against the core. I would have to play with one of those sensors, and try...

Sensing via a separate winding seems pointless to me, unless some novel and complex circuitry with AC pre-biasing or such is used. That windingby itself won't detect DC imbalance, because windings don't induce DC. And even when the core starts to saturate, the induced voltage will not collapse right away. It will do so only when the saturation is so deep that the current has shot up enough to cause a large ohmic drop. And that's far too late!

Integrating the voltage across the primary looks like a rather touchy business to me. The problem is that all circuits have offset, and all circuits also have a drift in that offset. Any offset in the integrator will drive the power circuit such that a permanent, constant net DC component appears in the voltage over the primary. That will cause a DC current in the primary, depending on that voltage and the winding resistance. To get this DC current low enough, the integrator's worst-case offset needs to be low enough, and that might be hard to achieve. In my transformer the ratio of winding resistance to saturation current is such that only about 1mV of total DC voltage component is acceptable across the primary. The integrator and all related circuitry would need to be better than that, in terms of offset.

I will think it all over.