2MHz toroidal core?

Thats a lot of Al at 2MHz. Are you sure you need that much?

Is this for a lossy element or a low loss one? If you want lossy look for type 77 material.

You can stack toroids to make them into thicker cores. This raises Al.

--
--
kensmith@rahul.net   forging knowledge

"Ken Smith" a écrit dans le message de news:dr84dh$q01$ snipped-for-privacy@blue.rahul.net...

wrote:

Yup. Almost that much.

That's for a low noise resonant converter some of us seem to have at the moment, and will power my low noise preamplifier head.

The PS have to pass 10-12W with ultra low common mode current, +/-15V output. I now have a proto running on my bench, made from... unknown salvaged EMI management ferrites. They are 20x10x10mm (Od,Id,h) and I measured them at

800 Mu_i. I also now have drawn the Mu'/Mu" curves, so that'll help identifying the manufacturer.

At 2MHz, 2x6 turns it's 36uH and 100mA magnetizing current. I don't want much more and I can't have more turns (leakage inductance will be unmanageably high). And to render the thing a bit more involved, the low common mode current target excludes de facto the MnZn ferrites: the core needs to be high resistivity to avoid primary-core-secondary capacitive coupling. The ones I have Right now I have 0.35pf primary/secondary parasitics which I manage to lower down to about 4fF (yes, not pF) common mode coupling with some careful shielding and a bit of tricks.

Too bad I only have 2 cores: I just needed 10 to build the 5 PSU I need.

Losses are not that critical due to the relatively low power, but I don't want 5W either.

Yup but that'll rise the primary/secondary parasitics too and I can't afford that. And clean winding is already difficult enough with one core. With two, arghh...

--
Thanks,
Fred.

Have a look at earlier threads on how to minimize leakage inductance. You can go a long way with that. Paul Mathews

"Paul Mathews" a écrit dans le message de news: snipped-for-privacy@o13g2000cwo.googlegroups.com...

I'm very well aware of the way to reduce leakage inductance. Unfortunatly this invariably leads to higher interwinding parasitics which is a definitive no-no here.

Think: Vout=15V, F=2MHz -> dv/dt>188V/us

I want common mode currents under 1uA, preferably 1/3ua

This translates to under 5fF interwinding capacitance, or capacitance dissymetry if you want to play the balancing game.

Try to achieve that with interwound windings.

--
Thanks,
Fred.

In article , Fred Bartoli wrote: [...]

If it wasn't for your interwinding capacitance issue, I'd suggest twisting the windings together.

First a couple of things I'm sure you've already thought of:

The leakage inductance makes this:

L1 T1 L3 ------))))----+----- --------------)))))---- ! ! ! ) ) ( L2 ) ) ( ) ) ( ! ! ! --------------+----- --------------------

T1 is a ideal transformer.

I assume that the primary of this is effectively across the capacitor of a tuned circuit. (At least at higher power) This would be a normal way to do things.

If the leakage inductance L3 can be predicted, you could place a capacitor in series to cancel it. This makes it a double tuned system but the Q of the secondary side will be quite low so it shouldn't be a major problem.

You could put a capacitor across the secondary and make it a highish Q tuned circuit. Unfortunately for your application, I don't think this is in the cards.

Now for my real suggestion:

I've made lower interwinding capacitance at the cost of leakage inductance by linking cores with single grounded turn. I'm thinking that you may be able to do the same in a more extreme form like this:

Rod soldered into hole / ************* * * *

"Ken Smith" a écrit dans le message de news:drapkh$6vq$ snipped-for-privacy@blue.rahul.net...

wrote:

Ken, you've almost described what I'm going to do (well have already done on the proto), but with some minor modifications.

Effectively 2 cores, with a "one" turn in between. The first core is pushpull hard driven at 100% duty ratio. Then the series of both cores leakage inductances (about 500nH each xformer, seen from the 'single turn' side) is resonated out with a cap. The secondary side of the second xformer has 2 small parallel caps to limit dV/dt and thus parasitic currents closed on the shield.

Then the differences: primary side of the second core has a center tap to ground in order to balance parasitics currents to the primary side shield. Just a single grounded turn doesn't do. The secondary side is already balanced. The inter-cores loop has more than one turn to reduce loop current (10W have to pass there) and ease these parasitic currents balancing. Plus we need an even turn number to make a center tapped winding :-) Then we need shielding between the "1 turn" primary and the isolated side secondary. 0.1pF capacitance is already way too much and if you carefully think about the voltage distribution along the turns, CM voltage can't be perfectly balanced, even in a perfect world. So we have complete shielding of both sides, with both shields crossing the second core between primary and secondary.

Funnily enough, the shields, which have to extend right in the middle of the core (but be careful, don't close the loop :-), see high induced voltage in the center leg (a few volts), and some adjustment is necessary to compensate for the minute dissymetry that will inevitably appear (flux leakages are pretty high) and create some unwanted common mode current. The rather crude proto (dead bug on a plain GND plane) show that the induced CM voltage into both shields has to be carefully equaled. Once that's done, the CM current is well below 1uA, down to the 200nA level on each harmonics. I probably can expect better with a clean PCB and tightly coupled tracks (this isn't well controlled now).

Now, I've also found some available cores (43 material) so I should have a more definitive view of all this next week, before making a clean PCB.

--
Thanks,
Fred.

In article , Fred Bartoli wrote: [...]

A comment about "great minds" could have been put here but wasn't.

Ok got it.

Huh? Where's the cap? Is it in series with the single turn? At least that's what I think you said. That cap ends up wit a lot of current in it doesn't it?

It needs to be mounted very near the cores and needs to have a very low ESR. I guess you could take my pumbing parts idea and put the cap on one of the pipe-caps. A ring of smaller values would work better than one big one.

I assume you ground one point on this near the output side core. Did you also ground the core material too? Most cores are conductive. If you put a small blob of conductive epoxy onto them you can hold a fine wire in place to ground them.

Do you really need those extra caps? It seems like it would be better to try to not put the sharp edges into the core.

... I've got to go make a salad for dinner so more later

--
--
kensmith@rahul.net   forging knowledge

"Ken Smith" a écrit dans le message de news:drbsg0$g05$ snipped-for-privacy@blue.rahul.net...

wrote:

on

:-)

xformer,

Yup, it's there. About 2.2A RMS and 20V RMS "only", but still within reach for good reasonably sized capacitors. That's also one reason why I only have a 2:1 and 1:2 ratio.

I just hate plumbing :-) Seriously, I'm not sure I got all the picture of what you drafted.

... Getting back to your drawing ...

Hmmm, yes, I didn't understand but now I am. That's clever and will indeed help keep the leakage inductance low, but alas this won't solve the capacitive coupling at all. It will obviously provide low coupling from one core side to the other, but to the detriment of the capacitance to the *common* shield which is the copper tube, washer,... And now, where do you reference all the plumbing? You've just replaced the interwinding capacitive coupling by a worse (parasitics will be higher) winding to other side ground coupling.

See below.

I think you're speaking of the single turn there. You can ground it as near to the core as you like, even in the core's symetry plane and have it perfectly balanced, there are still the 15V secondaries voltage swings that'll induce CM currents. You *have* to have 2 shields between the windings (keep in mind the fF order of magnitude), one referenced to each side, to keep the windings capacitive current from flowing across the barrier.

Yup, *most* cores are conductive. That's why I banned MnZn and use NiZn ones. Of course, one can "ground" the core, but which side? And what to do with the other side winding parasitics to core? I've ordered some MnZn cores too to play with, but this is just to confirm they won't suit and I'm ready to bet they will be really bad.

But you're giving me some idea to maybe further reduce the residual capacitance that crosses the barrier through the core cross section.

limit

Yup. You can do whatever you want to 'not put the sharp edges into the core', the leakage inductance does its work and gives you nice ones on the output.

have

an

the

the

in

compensate

induced

level

tightly

--
Thanks,
Fred.

In article , Fred Bartoli wrote: [...]

It depends on what coupling you are worried about. The copper plumbing is at (or very near) some voltage ground. You get a large capacitance to "ground" but remove all the capacitance to the AC voltage on the primary.

The capacitor in at the input side's pipe cap. The ground connection is at the output sides pipe cap.

Perhaps it is worse for your actual situation. I had assumed that a capacitance to ground was OK to add. If not, oh well, the idea cost you nothing and was worth every penny.

Yes, I was thinking in terms of the single turn system I had used in the past. I didn't care so much about currents. I needed a total capacitance of the system under 1pF. I used a thinish copper wire and thickish insulation to keep the copper far from the windings. There was also a gap between the two cores.

Hummm... Yup your right.

Yes, for your application you need more copper parts.

I'm not sure I understand this comment. In the two core case, I assumed that each core would be grounded to the local ground.

I just had another though.

Would multiple thin link windings be better than one thicker one? Unfortunately, the capacitance function has a log of the diameter in it so this may not work.

How about 3 cores?

--
--
kensmith@rahul.net   forging knowledge