Sourcing/making toroidal flyback transformers

[rs20]

I'd like to be in possession of a 1:1 isolation (flyback) toroidal transformer that has winding inductances of 30 uH, and can efficiently support the classic DCM flyback waveform which averages 1 A and peaks at 4A, 100kHz switching frequency.

- I've looked on element14, and can't really find any 1:1 toroidal transformers -- seems most of the range there is dedicated to AC-DC.

- I've tried using the EPCOS ferrite magnetic design tool; but it refuses to offer power calculations for toroidal cores for some reason? I did arrive at a non-toroidal solution* that I could source on element14, but it's hugely overspecced** and nasty looking.

- I sort of understand that gaps are important for flyback transformers, the energy is stored in the air gap something something***. But I remember reading somewhere that you could buy powder toroids that have little gaps between the grains (roughly speaking) that results in a "gap" being distributed evenly all the way around the toroid. This sounds wonderful, but how do you buy these? Do different "porosities" of powder get classified as different materials (like, does N87 have gaps in it?)

[ Also, if I see that a toroid won't saturate on me at a given DC bias, am I safe from excessive hysteresis losses on an AC waveform peaking at that DC bias? Or is saturation and hysteresis a completely different thing? ]

Any insights would be greatly appreciated, I always get in such a rut when it comes to magnetics. Why can't transformer core datasheets just have an hypothetical Isat and inductance for a single turn, that we can then just divide by N and multiply by N^2 (respectively)? The formulas they provide make you go through a bunch of work to arrive at the Isat for a 100 turn inductor, WTF is the point of the run-around? Gah! 

The end goal is a 24V DC to 24V DC @ 1A isolation converter. I know it's probably silly to use DCM for this, but I'd like to do it this way just for the laughs and ease of control.

* Specifically, ETD34 N87 cores with 1mm gap, 16 turns on both sides.
** Inductance has dropped by 1% at 8.9A bias.
*** OK, yes, adding a gap allows you to add more turns for a fixed final inductance, and as inductance goes up with N^2 but flux goes up with N, the gap reduces flux for a given inductance. Right?

[Circlotron]

Maybe some kind of common mode choke? Powdered iron of course.
http://au.element14.com/murata-power-solutions/52106c/choke-10mh-1-7a/dp/1767577

[T3sl4co1l]

Go with a cut core instead.  Ferrite toroids are wholly unsuitable because the permeability is high and the energy storage capacity is tiny.

Powder cores are unsuitable as well, due to losses.  They can be made to work, but aren't usually all that great.  And you really want some of the more unusual grades (mu from, oh, 10 to 35), not the mu=75 crap you see in consumer gear.

A 20W converter will fit in, oh, an EE20 sized core very easily.  Or anything like that (ETD, ER, P, RM..).  The core pieces need to be gapped.  By how much can be calculated.

DCM is just fine for this; you wouldn't worry about CCM or forward converters unless the current is inconveniently high (e.g., push-pull forward converters are popular for automotive applications, since it distributes the current better).  The usual point you want to change your mind is around 50-100W, depending.

An ETD34 is good for about a kilowatt.  You might've misinterpreted what the input numbers were meant to be.  Or the calculator's just that bad, I don't know.

As for physics/olosophy, they don't give straight circuit parameters because it's not in a circuit, it's just a core.  All the circuit parameters vary by number of turns (and their configuration, if it's a multi-legged core) and gap length, so it would be kind of silly to pick something arbitrarily.

For something with variable gap, like ferrite, the saturation flux is trivial to calculate.  To find saturation current, you need to know turns and gap length; the proportions are straight linear ratios in all cases.  (You can do the same procedure with powdered iron, but it's not very useful because the gap and permeability and everything will only ever be what they are for a given core.  So it doesn't matter if you calculate from flux or from amp-turns.)

Tim

[rs20]

Thanks Tim, always really appreciate your input!  So the toroidal cores with gap distributed throughout them are just a figment of my imagination?

I should clarify, the Epcos calculator did tell me that the ETD34 is overkill; the problem is that element14 has a woeful selection of (Epcos) gapped cores (i.e., ETD34 is the smallest AFAICT), so I was constraining myself to that.

Now, instead, I've come up with the design below, but I'll have to make do with ungapped cores and using kapton tape to hack in a gap* while I wait for a colleague to make a digikey order. I know that the kapton tape will make 3 separate gaps, but it'll do for experimentation.

E20/10/6, N87, 0.5mm gap, 20 turns --> 30uH inductor with 4% permeability rolloff at 4A.



EDIT: Fixed broken image link.

[The Electrician]

So the toroidal cores with gap distributed throughout them are just a figment of my imagination?

Not at all.  Low loss powder cores exist:

http://www.mag-inc.com/products/powder-cores/mpp-cores

I haven't checked pricing.

Not quite as low loss are these: http://www.mag-inc.com/products/powder-cores/kool-mu

[T3sl4co1l]

Thanks Tim, always really appreciate your input!  So the toroidal cores with gap distributed throughout them are just a figment of my imagination?

Nah.  Well sort of.  On a microscopic scale, yes, a distributed gap is a reasonable description.  But you don't have any way to measure or adjust that gap, so it's better to use a bulk approximation and say it's <rated mu_r> material instead.

Indeed, once you've gapped a core (or purchased pre-ground cores), you can just as well call it mu_eff, where the path length is as stated (l_e), the gap length being absorbed into the equivalent.  This works by adding the core and gap reluctances, then dividing by the original length.

Or you can go with the air gap equivalent, which means the core length l_e gets divided by mu_r.  That's how much l_g it's equivalent to.  (And if you do some typical examples, you'll see it's quite small indeed -- which is why ungapped ferrites don't store much energy.  Energy density, by the way, goes as B^2 / (2 * mu_r * mu_0).  Which is also a pressure -- N/m^2 == J/m^3.  This is called the Maxwell stress.  If you know the difference in B-field at either end of the armature in a solenoid, you know how much force is acting on it.)

Tim

[calexanian]

I buy my ferrite materials from Amidon Associates. They cater to the ham radio community and will sell small quantities of whatever materials you may need.

[url]http://www.amidoncorp.com//url]

[rs20]

Awesome, thanks everyone. I was doing some further reading, and came across a mention that inductors for boost converters should have a "gapped"* core (which makes sense, the theory of operation is almost identical between a boost and flyback converter). This got me thinking, all the power inductors I buy for my boost and buck converters are not obviously physically gapped, and they are "recommended for DC-DC converters" and have "low core losses" etc etc. Here's an example datasheet, so that we're on the same page [heh]. I'm now very conscious that this "datasheet" is rather lacking in actual numbers about core losses. Nevertheless, assuming the 2307-RC is a fine 33uH boost converter inductor, could I not simply buy a 2307-RC and wind on a second coil myself with the same number of turns to get a 33uH:33uH flyback transformer? Does that look like an exotic low-mu material to thy trained eyes? Granted, it's significantly bigger and heavier than a E20 or E25 core, so maybe not so much.

Note that my question above is purely out of curiosity, the hint about using P240 sandpaper to make my own gaps is a really great idea. I can just stock a bunch of ungapped cores and a piece of sandpaper, rather than stocking every imaginable permutation of core size and gap. So that's what I'm going to do. Thanks so much for the suggestion, AcHmed99! I'll try out an E20 first (I probably don't actually need the full 20W.)

* Or (according to Tim's points), low mu_r?

[T3sl4co1l]

Hmm, I've got a 2200- series choke, which looks like (and I confirmed with Bourns that) it's a Kool-Mu type.  Good for, say, 40% current ripple at 100kHz or thereabouts.

For gapping, there's usually 4 ways about it:
1. As little as possible: ideally, ground or lapped faces for absolute minimum gap.  Purposes: pulse transformers, common mode chokes.
2. Not quite minimal (say, < 2 mils): factory ends (usually ground well enough), or just a little gap (paper or plastic film).  Purposes: power transformers, where the slightly higher magnetizing current is desirable for assisting startup transients, improving switching, or operates with other components as a resonant converter.  Very large value inductors for small signal applications.
3. Small (say, 1-10 mils): commonly seen where you have an oversized core handy, or need to use one because of other considerations (lower capacitance by sectioning the windings, high voltage clearance).
4. Normal (say, 10-100 mils): typical for power inductor applications, where the mu_r is reduced to a level suitable for copper windings*, usually 10-60 or thereabouts.  This gives a good compromise between core vs. winding losses (for most materials), efficiency and size.

Mind that, for a C or E shape core, the gap occurs twice.  The total gap length is total over the loop, so don't forget to double it.

*The permeability is only there at all, because copper isn't quite as conductive as we'd like for most switching purposes.  If it were, we could use air core coils of similar size, without making compromises on efficiency.  But since it isn't, we need permeable cores to help it along.  If we were stuck with a higher resistance material, like aluminum or zinc, for wiring, the ideal permeability for inductor cores would be higher (and their size would be larger and/or efficiency poorer, as a direct result).

So if you're stockpiling one or a few cores, for assorted purposes (transformers and inductors), it's probably not necessary to reserve "all possible values", just a few special ones covering purposes such as these.

Personally, I have a pile of EE33 (a somewhat odd size) cores, and bobbins, so I just put on as many turns as I need, for whatever gap I put in the thing.  They're good up to about 100W capacity with good efficiency, and more with cooling and/or other approaches (e.g., forward vs. flyback converters).  Usually I gap inductors by supergluing a small piece of bare FR-4 of handy thickness (~30 mil) on each leg of one core piece.

Example:
http://seventransistorlabs.com/Images/HVPower2.jpg
This one was made with paper in the gap, for reasons around about #2 and #3. It's only about 20W, so I have lots of extra space.  I also kind of want to use that space, because it's a high voltage output.

Tim

[T3sl4co1l]

You could try to do that with that Bourns Toroid, I wouldn't waste my time though. They don't give enough information, so that may be a clue that its basically meant for input or output filtering. The 1kHz test frequency and only DC current rating is also a clue.

Here is what an inductor data sheet should look like.

Bourns, in general, and many other manufacturers, are, I guess, just too lazy to bother handing out data.  Fancier or newer parts may have it, but you always pay extra for it.

The one manufacturer I know of that does publish excellent resources for most of their parts, is CoilCraft.  They're also cheaper than almost everyone, by the way (manufacturer-direct sales!).  They have SPICE models available ... although be warned, they are AC steady state only, useless for transient simulation in most SPICE engines.

Tim

[joeqsmith]

So the toroidal cores with gap distributed throughout them are just a figment of my imagination?

Not at all.  Low loss powder cores exist:

http://www.mag-inc.com/products/powder-cores/mpp-cores

I haven't checked pricing.

Not quite as low loss are these: http://www.mag-inc.com/products/powder-cores/kool-mu

Just an FYI, their spreadsheet has some problems.  They are working on correcting it and should release it soon.   The older program works fine.

[station240]

Tortech in Australia make custom transformers, including toroidal types. I'm sure there are other companies that do also.

[Circlotron]

the hint about using P240 sandpaper to make my own gaps is a really great idea.

For moment I thought that meant to use the thickness of the sandpaper as a gap spacer. 
In that respect I have used drafting film with great success.

[Circlotron]

I seem to recall that making the pole faces non-parallel across the air gap altered the rate of change of inductance as saturation approached. Not sure what it's purpose was now.

[T3sl4co1l]

That can be handy for swinging chokes, where you get the needed energy storage at high flux but better ripple filtering at low flux.

Tricky to do for switchers, as that shifts the control loop response.

Tim

[rs20]

Well, it ain't toroidal, but I think it's beautiful! Now to wash my hands of N87 dust (it's good for you, right?).



(PS/ Australian 50 cent pieces are freakin' massive. That's an EF20.)

Thanks everyone!

EDIT: Fixed broken image link.

[rs20]

IIRC, it's 0.6mm wire, with the skin depth being 0.22mm, that's almost all skin. Furthermore, I think the "that's too heavy to carry 100kHz" mentality is a little bit too far; even if a heavier gauge of wire has a worse Rac/Rdc fraction, the Rac still improves in inverse proportion with diameter (while Rdc improves with the inverse square proportion, hence why the ratio appears to get "worse"). In short, yes, I've wasted a little bit of copper there, but I saved time on winding and my Rac will still be quite low.

Thanks for the tip on leakage inductance; I was indeed wondering whether the outer turns would interact more strongly with the outer parts of the core.

[rs20]

Oh, I didn't mean to be implying any smart-assness, sorry -- was just saying that lots of people (not you) think thicker conductors are worse at carrying higher frequencies; and double-checking that my reasoning was OK. Sorry!

[Yansi]

Just note: The iron powder cores are also not suitable for DCM flybacks because of the huge change of B through the DCM cycle. Iron powders don't like much swing in flux density, they usually have quite high hysteretic loss. There are graphs in the datasheets you can use for estimation of the loss.

[rs20]

Just note: The iron powder cores are also not suitable for DCM flybacks because of the huge change of B through the DCM cycle. Iron powders don't like much swing in flux density, they usually have quite high hysteretic loss. There are graphs in the datasheets you can use for estimation of the loss.

So is a gapped N87 MnZn ferrite core OK for this purpose?

[T3sl4co1l]

FYI, skin effect is worsened by proximity -- the windings down in the middle experience a 'pinching' effect from all the amp-turns around them, easily increasing Rac/Rdc by a factor of 5!  If you ground down the center leg for air gap, then the fringing flux around that will also contribute to losses (eddy currents induced by the highly divergent field, which exists within, oh, 3 gap lengths around the gap, let's say).

Also, kinda looks like a lot of turns for ~100kHz and 24V, so we'll see I guess.

Tim

[T3sl4co1l]

Yeah, there's actually a formulation for optimal distribution of windings to minimize or equalize losses; it's kind of goofy, but with 3D printers available these days, not terribly far-fetched.

Geez, I know it used to be public, but I can't find it!  Anyone know what I'm talking about?  Did they pull it down?

Tim

[T3sl4co1l]

You mean, a powder core in the center?  Yeah, of the right grade, that would be workable.  Composites are probably a pain to put together, though.  (I've seen combo toroids, intended for swinging chokes and that sort.  Much easier to put together!)

Commercial designs use center gaps, because there's less external fringing and EMI.  You can double-gap a core with spacers easily enough, just mind that there'll be more external field.

Tim

[rs20]

My silly, expensive design (optimized for personal learning more than any real engineering consideration) is coming along nicely, but I thought I'd do some searching for the typical chips used for this sort of thing. However, I was surprised to note that the example flyback schematics in every chip datasheet I've seen has been non-isolated; i.e., the analog feedback is just connected back to the feedback on the primary side. It's not obvious to me how to sneak optical isolation into this arrangement, accurate analog optical isolation is hard. I imagine the reference should be on the secondary side, rather than integrated into the chip.

Anyway, no need to go into great detail, I'd just like a few datasheets to look through -- which chip(s) would you typically choose to run a small, isolated (DCM?) flyback converter like this?

[T3sl4co1l]

UC3842 is a classic.  Lots of e.g. TPS series parts that would work as well, saving a few percentage points on efficiency by using a faster, more efficient controller (or integrated switch if you like).

Tim