PCB transmission line transformer

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

Should be pretty straightforward. Two identical windings, on layer 1 and layer 2, have a given width, spacing and dielectric the whole length. Interleaved windings of different construction, or layers connected in series, will have different results, but that's true of normal transformer windings (dispersive helical waveguide modes and such). You'd only want "wide" traces if you want a low impedance. You can also parallel up layers for that.

I've used planar transformers before. Rather lossy and space-inefficient with few layers. No PCB manufacturer can produce a board with better than about 0.2 winding factor, so it can't compare to the density typical of wound parts. On the other hand, it's quite low profile, and with enough layers, the losses are good enough. Performance is great, since the coupling is good, the winding compact with few turns, and high frequency core materials are available. Frequency response will of course matter more to your purposes, unless you want to do superfast kW pulses with selectable matching or something.

Ready made transformers, either made of PCB stacks, or just ridiculously thick PCBs of many layers, are available into the 5kW range, intended to operate at a sizable fraction of a MHz.

Cores are available on tape and reel from various magnetics manufacturers and distributors, e.g. Ferroxcube parts from Adams or Elna. You can pick E-E or E-I, among other subtle variations. The smaller parts are cute, and easy to misplace or drop.

The "I" parts are useful for other purposes because, really, they are more plate than "I" (e.g. Ferroxcube part numbers PLTxxx..). Handy when you need some heavy shielding and aluminum or copper plate just won't do, because they start melting.

I can only imagine what a loathesome hack it would be to construct one in PADS though...

Tim

The principal virtue of coax-wound transmission-line transformers is indeed that: Low leakage inductance. Using PCB transmission lines instead would yield higher leakage inductance, maybe similar to what you'd expect of twisted-pair windings, maybe a little worse.

Jeroen Belleman

wound them from micro-coax on ferrite toroids, with the shield being the pr imary and the inner conductor the secondary. Sub-ns speed and low leakage i nductance. But it's labor intensive.

ayers 1/3/5 could be one to three layers of spiral trace, primary, and 2/4/

eed that: Low leakage inductance. Using PCB transmission lines instead wou ld yield higher leakage inductance, maybe similar to what you'd expect of twisted-pair windings, maybe a little worse.

Microstrip is nasty. It's dispersive to boot. Stripline - with ground plane above and below a buried trace - isn't dispersive. The field spreads sidew ays to some extent, but it dies away very rapidly with distance, and should be quite a lot better than twisted pair. If the odd-numbered layers in Joh n's multiplayer PCB were all ground planes, strip-line traces on all the ev en numbered inner layers (not the last) would be pretty innocuous.

This will be a mosfet gate drive transformer, up to a few hundred ns, and I need very fast edges, namely low leakage L. 4KV isolation.

Good stuff, thanks.

PADS can draw any traces or copper patterns that any other layout software can do.

Inside a transformer??

Jeroen Belleman

I was concerned about that, but I don't know how to quantify it. I'd make a prototype, probably.

Shouldn't make any difference. The thing about transmission line transformers is the most of the action is confined to the space inside the transmission-line structure.

The high frequency performance is barely affected by the presence or absence of the surrounding ferrite, but the ferrite sustains the transformer action to much lower frequencies than the bare transmission line structure offers.

Den fredag den 13. juni 2014 17.25.49 UTC+2 skrev John Larkin:

is a transformer wound with micro coax really rated for 4KV ?

you can't find some small smd transformer that would work?

-Lasse

Pity about the shorted turns though. Using fairly wide aspect ratio traces on top of each other, all the way through the board, should reduce the dispersion a fair amount.

Cheers

Phil Hobbs

Probably. We'd test them, of course.

I've run RG58 at 20 KV, no problems. Teflon is rated somewhere between 1000 and

I've never seen a commercial coax-type transmission-line transformer. The MiniCircuits balun types wouldn't work.

Of course, we'd have no planes anywhere near the transformer region.

Right, which is why you wouldn't be using stripline. ;)

Cheers

Phil Hobbs

've wound them from micro-coax on ferrite toroids, with the shield being th e primary and the inner conductor the secondary. Sub-ns speed and low leaka ge inductance. But it's labor intensive.

r. Layers 1/3/5 could be one to three layers of spiral trace, primary, and

s indeed that: Low leakage inductance. Using PCB transmission lines instea d would yield higher leakage inductance, maybe similar to what you'd expec t of twisted-pair windings, maybe a little worse.

plane above and below a buried trace - isn't dispersive. The field spreads sideways to some extent, but it dies away very rapidly with distance, and should be quite a lot better than twisted pair. If the odd-numbered layers in John's multiplayer PCB were all ground planes, strip-line traces on all the even numbered inner layers (not the last) would be pretty innocuous.

Well caught Phil. Thanks. I wasn't thinking hard enough.

The "ground plane" layers would have to be ground strips, wider than the im pedance-defining central trace, but still forming a spiral rather than a cl osed loop. You wouldn't need a lot width difference to make the width of th e central trace impedance-defining.

My subconscious is reminding me of a more balanced structure - two side-by- side traces - but I'd have to get into my microwave reference books (two ra ther mathematical paper-backs) to see what they'd look like in context, and I think that they are back in Sydney.

Bill Sloman, Sydney - but in Corgoloin, Burgundy, France, right now.

And, with less than a ns of total prop delay, picoseconds of dispersion wouldn't be an issue in driving mosfet gates.

Of course, Sloman is the magnetics expert here, and we know nothing about this stuff.

I'll have a glass of Nuit St. Georges in honour of your holiday. ;) My family and I are going to tour the WW1 battlefields in August. (*)

In the limit of many layers, a stack of alternating P-S-P-S traces on top of each other gives you the same field configuration as stripline.

(The argument is the same as the way you derive the method of images, only backwards.)

Doing that with some finite number of layers like 6 or 8 should reduce the dispersion by quite a bit, I would think.

Cheers

Phil Hobbs

(*) We're staying in Talbot House at Poperinghe, which is sort of a living museum. (Hopefully some of my receivables come in before then. I have a bit of a concentration of slow-paying customers at the moment.)

You'd have to make sure the core wire, with insulation intact, comes out a fair distance.

Not likely to find much in SMD, though these are pretty cool:

Cp ~ 20pF, LL ~ 0.8uH (equivalent, total, end-to-end, open/short circuit test).

Or in other words, ~150 ohms Zo, 40MHz Fo. Good down to ~10ns rise time, some squigglies beyond there.

A bit high impedance for gate driving. You'll want to follow it up with a buffer, for most cases.

They're more-or-less scramble/multifilar wound with "triple insulated" wire, hence the high isolation and bandwidth. Likewise, LL is evenly distributed over all four sections (so, the two halves of each CT winding are about equally coupled as to either part of the secondary).

The CT winding gives darned good CMRR in balanced mode operation. Think I tested common mode with an avalanche pulse generator and measured at least

So, could be handy for isolating data over a difficult isolation barrier (like, nanosecond gate drives). Say, zero-DC-coded LVDS, at a slightly higher than normal line impedance (150-200 ohms, rather than 100). Kind of a bulky, old-fahsioned and expensive way to do it though, considering monolithic transformer logic couplers are available these days.

Tim

Incorrect -- the LL is quite large, given the large core-to-shield distance. All that space fills up with magnetic field fairly easily.

Ditto a twisted pair: the space between wires, and the space around wires themselves (especially if thick insulation is used) provides quite a lot of leakage, relatively speaking.

The only structures that provide exquisitely low leakage are massively interleaved structures. But they have so much capacitance, they become hard to drive.

Hmm...

Wouldn't it be magical if the problem weren't just lumped equivalent leakage, or lumped equivalent capacitance, but an impedance defined by the ratio thereof?

Indeed, that is exactly the case. Which is why -- when operated at the system impedance -- a transmission line transformer can work at much higher frequencies than Fo = 1 / (2*pi*sqrt[LL*Cp]). Likewise, the impedance is, say, 50 ohms = sqrt(LL/Cp), or whatever the characteristic impedance of the cable happens to be.

If you try to operate it at much lower impedance (say for power transfer purposes), the LL appears massive, and the operating frequency must be limited. If you try much higher impedances (er, not that anyone uses tubes anymore, but a transformer for interstage coupling of wideband tube amps would apply here), you'll have the exact same problem from the capacitance.

Coax has one advantage in that, if you can arrange the system, or geometry, so as to be able to ignore the voltage and current on the outside of the shield conductor, the impedance will remain constant despite being wound into a transformer. In contrast, a twisted pair is not easy to wind in such a way that adjacent turns -- and layers -- do not couple to each other.

Most times, you cannot ignore said current, and one conductor (or both, in the twisted-pair case) necessarily takes on a solenoidal or toroidal resonator, or parasitic-capacitance-between-layers aspect, which limits bandwidth to not much higher than Fo due to the various dispersive, anharmonic peaks and valleys that result from such construction, which invariably load down the circuit and are extremely difficult to anticipate and compensate for. In these cases, your only choice is to make the transformer smaller (shorter electrical length = higher Fo), which in turn limits low frequency performance (not as much core available inside the thing).

Tim

how about this tiny one, meant for 100Mb ethernet I believe

-Lasse

Isolation?

...Like...can that thing actually handle *any* voltage at all!?

I serious can't see any isolation rating on that thing. If it meets LAN specs, it MUST be 1500V hi-pot. But they don't say. Scary!!

Definitely not enough for silly things (wink) like 4kV gate drives, of course. At least for more than a few cycles.

Ditto any transformer or DC-DC isolator or whatever that doesn't specifically claim "reinforced". For example, Murata has a line of "3kV barrier" DC-DC isolators that they hardly guarantee for SELV. Internal construction is just enameled wire over enameled wire on a toroid core, scary for any kind of voltage.

Nice thing about planar transformers is, if you're using internal layers for the isolation barrier, you automatically get reinforced spec. I mean, I guess technically it's a "single layer" insulator... but then, is it? It's multiple layers of prepreg. Not sure. Far as I know it counts though.

Tim