trigger transformer

There are also magnetic isolators that may have enough isolation, although 4 KV sounds pretty high. They need regular pulses to keep the "AGC" centered. (Really, more of an automatic zero control.) NVE is one maker of these.

Jon

You could do the same with capacitance I guess? Saves two inductors!

Should work fine but I would not do that with regular FR4. Isolation voltage is not the same as working voltage because the stuff degrades. There is usually about an order of magnitude difference.

Also, absolutamente no sharp corners or edges. Even on other layers and I'd make sure of that for at least a 1" radius.

We used some NVE couplers once, but they could get confused and output the wrong state statically. Not good.

When exposed to 4kV and not rated for that as a working voltage that would likely become worse over time.

At sub-ns frequencies I would have thought you could simply pass a pulse through the board capacitively. If necessary through 2+ layers of board. I've not tried this though.

NT

Den tirsdag den 24. juni 2014 18.07.40 UTC+2 skrev John Larkin:

isn't that pretty much how the icoupler work except the coil are on a die

page 22.

-Lasse

I like the cap idea, it keeps the fields contained, while the inductor broadcasts everywhere. (should that be whilst?)

George H.

Why look for ferrite-core solutions at all? Big voltages are handled with large spaces, and an air-core inductor seems appropriate. I wouldn't clock, but would use set/reset inputs, That gives you more flexibility in determining initial conditions.

So, this is an easy job for a hobbyist in his basement, but it's gonna be a mess for someone who would prefer a pick-and-place component, and a crowded circuit board.

Avago ACML-7410 claims 3ns pulse width distortion on 36ns prop delay using magnetic coupling between chips. Rated over 5kVAC for 1 minute only. But sounds like you want to do that on pcb scale?

piglet

Way too slow and jittery, anyhow.

I was thinking of using surfmount, maybe 0805, wound ceramic-core (no ferrite) inductors for speed.

All I want to do is fire a one-shot.

I could use a, say, 390 nH 0805 inductor. If I apply 100 volts in 1 ns (which is feasible) the current rises at 0.25 amp per ns. So I can push in a, say, 4 ns pulse before things get too extreme.

With 100 volts on the "primary", a coupling ratio of 1/20 gives me a 5 volt pulse on the "secondary", plenty to fire my one-shot. The secondary inductor might have more turns, if needed. Or add an NPN transistor to make it more sensitive.

Rough numbers like that maybe. Might work. I guess I need to measure some couplings to see what that ratio really is. Too hard to calculate. Gotta select a PCB material that lets me put 4KV across it without getting too thick, too.

he opposite side, as a transformer? Whack the primary with a big fast pulse , and it might make enough on the other side to fire a 1G123. Gotta try it, I guess.

ld be too slow.

Really? The Analog Devices magnetically coupled parts look like they might come close - depending on exactly what you actually need to do. Channel mat ching can be 2nsec or better for channels going the same direction. Here's the data sheet for a two channel part.

The "C" version has a 24nsec maximum propagation delay, but the temperature dependence is typically 1.5psec/K if you pay for the premium grade (C) par t.

Minimum pulse width is 10nsec. It's sort of cute. Your printed circuit tran smission line transformer would be nicer, but bulkier.

The original (Analog ADUM and etc. series) are statically stable, achieved using a gated astable on the TX side. They can be flipped, but reset within a few microseconds I think it is. The RX side also has a missing pulse detector so it fails safe if the input side loses power.

The dV/dt is quite good. Maybe not 4kV in nanoseconds, but gee, that's what fiber optics is for.

Tim

The pix I see of 0805 packages, show metal endcaps; that is going to complicate the flux on-axis (it's like having shielding in the wrong place). And, 0805 packages only put the dipole poles 2mm apart; your coupling will suffer if you can't get within, say, 6mm center-center distance (4mm insulation spacing?). Could you go with an 1812 package ? Bourns CM453232 series seems not to be end-capped.

If the coupling is good enough, your energy goes into the secondary (and the primary impedance isn't as high as the 390 nH value in the datasheet). The receive side still benefits from lowest available input impedance, probably the common-base preamp is better than feeding straight to a MOS logic gate.

off

the

pulse, and it

guess.

would be

Way expensive but, Ghz class FO link should do the trick.

?-)

or lose the fibre, just a pair of LED/optotr

NT

opposite side, as a transformer? Whack the primary with a big fast pulse, and it might make enough on the other side to fire a 1G123. Gotta try it, I guess.

AS JL said, opto-transistors are too slow. GHz fibre-otic links use faster receivers, which are a lot more expensive.

You are stuck with using a heavily reverse-biased photo-diode of some kind - which give you rotten transfer ratios - so you need very wide-band high gain amplification to get a decent amplitude signal out. Perfectly practical, but not cheap.

Maybe dual (differential) capacitors like the ISO150, which could be made with PCB material.

That could function in the presence of significant low frequency noise on the 4kV bus.

Best regards, Spehro Pefhany