HV power supply build - circuit safety components.

[Doug Coulter]

There are no 3kv zeners -- and if there were, they'd be a bad choice for HV and any stored energy. They stink at impulse loads. ZNR type devices are like zeners but so squishy they need to be at about twice the volts you normally run -- so you need 2x overrating on what they are protecting minimum. There's no easy way to get around that problem, other than prevention in the first place, and people like Spellman do both to get to decent reliability -- look at those 6 amp diodes in a low current supply -- and twos in parallel at that! I see similar efforts inside my big Glassman too, it's kind of the universal solution to this class of problem. You can't just go down on the ZNR voltage, as they start drawing current, heat up, and they die young from that even if you can spare the power -- they are "weak allies" -- not rated for any continuous power handling, just rare events.

I kind of doubt your circuit as shown can be protected from the AC coupling caps discharging in an output arc at all, with all the supplies running at ground potential. There's just too much stored energy implied in those. A huge reverse polarity diode across the output might help there, but in that circuit there's no place to put it that will work. You're kind of on uncharged ground here

I think you'd be better off either putting them all in parallel and driving a normal stack (you might find they sync up easily, worth a shot), or just put a doubler on each one, and hook a bunch in series with separate floating supplies some way -- but that's only my opinion. Both of those things were done in the past with at least enough success to get into the books as working solutions.

You might have to sacrifice a couple to testing...

Remember even a tiny capacitor can deliver huge peak amperes in a quick discharge - it's only limited by its parasitic series impedances, which can be quite low. Add high voltage, and that's not much of a current limit. And if you use series resistors, well, that kinda stinks too for other reasons (and the resistors may arc over themselves instead).

[chrismb]

I kind of doubt your circuit as shown can be protected from the AC coupling caps discharging in an output arc at all, with all the supplies running at ground potential.

Well, I'm not bothered about the supplies, but the external DC power source that feeds them. That incoming DC supply, one is connected to the common ground - well, it doesn't have to be but in making it so, then any HV trying to escape by that route will go via the +ve DC supply. So connecting a TVS between the + and - should then catch the transients. In regards dissipation - that's why I was asking whether a spark gap plus silicon would likely be required to prevent it 'getting out'. The caps I will be using will only be about one J in each of the upper ones (I am currently planning to use 2200pF). The typical TVS I'm looking at switch at around the 1ps rate (according to spec sheet) and the inductance in 1 foot of cable should mean it is unlikely all to come through at a current in excess of what a good TVS (or few) can handle, no?

[Doug Coulter]

You didn't say you were trying to protect the LV supply. Heck, that's fairly easy -- just make sure it's got a big cap across it and can withstand some overvoltage -- trivial for a 30v supply. That's just protecting the cheapest and easiest to replace part anyway (but also the easiest to protect). A joule is FAR TOO MUCH for a multiplier cap. But it won't matter because it's going to be in series with a 10-27 pf cap inside the potted supply -- you're just hosed there, that's a huge showstopper. You can't safely hook any other cap in series to take more volts without quite a lot of diddling around -- it the external cap is bigger, then the little one inside the potting will see most of the voltage when it's applied -- boom. If it's even smaller, it's too small a net capacity, and you have no current capability and the voltage output is "even more squishy'". Capacitors in series act just like a resistive voltage divider for AC -- as in the rising edge of the applied voltage. Most of the volts show up across the smaller (higher impedance) cap in direct ratio to the cap sizes. This effect is used in scope probes to keep the divide ratio constant in the presence of strays in the DC divider resistors, for example.

Remember than in an arc to ground, unlike the DC filter caps, the AC side of the multiplier caps have to go through the diodes to the output. This is a serious danger to at least the last diode (and the rest if all the caps aren't perfectly matched so they all discharge the same rate). A typical cap as you suggest might have well under one ohm effective series resistance. OK, at 3 kv, that's three thousand peak amps available -- diode lets out the magic smoke every single time.

Put an ohmmeter across the HV output of your little supplies and see -- it's almost certainly going to read open circuit due to the built in ballast (current limit to the output) cap. Unless you can get inside of that -- the entire scheme cannot work. The ones I pay about 10 times that much for make this easy as they're not potted. But they are about $15 each for 12 ma at 1kv output.

As I've mentioned, the TVS characteristics are a very rounded knee (not sharp like a zener) so if you use most at the spec "pass voltage" they won't limit hard till about twice that. Lowest spark gaps I've seen are a few hundred volts to trigger, and by that time I'd assume a 30v supply has been toasted. Spark gaps have a different characteristic once triggered -- they then pull all the way to ground more or less.

If you can come up with a bunch of floating DC input supplies, most of the problems go away, but of course, it's no longer cheap. No surprise, someone else would be doing it if it was cheap or easy this way.

[chrismb]

it's going to be in series with a 10-27 pf cap inside the potted supply -- you're just hosed there, that's a huge showstopper. You can't safely hook any other cap in series to take more volts without quite a lot of diddling around -- it the external cap is bigger

Blast! You are right! Hadn't dealt with that little hiccup.

The CCFL supplies I have bought before were also unpotted and, actually, I was expecting these to be, also. In fact the previous had two outputs so that two of those 4 lines could be connected to, capacitance-free. It didn't occur to me that I would end up with supplies that were both single outlet AND potted!

Oh well, but this may still be OK because if there is a cap on only one line (which is usual?) then I will pick that one to use to ground the supply, then the other AC output will be sans capacitor. [Edit: or is it? hmmm.... not sure what voltage the inductor coil will float to, now you got me thinking about it.] Unless they've gone stuck in two caps, then life will get complicated. I'm not yet throwing hats in just yet, but I admit you have me worried now....

If you can come up with a bunch of floating DC input supplies, most of the problems go away, but of course, it's no longer cheap. No surprise, someone else would be doing it if it was cheap or easy this way.

But I think the way I have proposed is cheap and easy! I put two of these together today and got 4kV across a 470kOhm stack. I reckon that's pretty darned good, for $2.50!

These things can each push ~12W into the stack, and if you add enough of those together then you are ending up with some pretty significant power ratings. It seems a good way of accomplishing 'stiff' stages at the top of a stack. I'm also surprised by how low the decoupling capacitor can be in value. I've used 10nF down to 470pF and there is little to choose between the results. (Actually, this might be figuring into the internal capacitor issue - if it is bigger than that, then it is big enough.)

What seems to count more (and I am not sure I understand why) is the stage capacitor. I would have thought that the decoupling capacitor also has to carry a certain energy to recharge the stage capacitor (i.e. the link cap should be bigger than the stage cap) but it doens't seem to work out that way. Don't know why - it beats my understanding. Probably something to do with inductance again!!!

That being said, I have a few 'measurement' issues which you, Doug, will no doubt be familiar. I'll start a new thread for that question....

[Doug Coulter]

Nothing is cheap if you have to fix it all the time....and that super cheap deal you got implies they're surplus and won't be available later when you need more.

Every CCFL supply I've had grounds one end of the transformer (it's not just an inductor) secondary, and couples out through a tiny cap which acts like the current limiting ballast for the fluorescent tube, limiting current. If there are two outputs, they both get separate caps from the same transformer secondary "hot" side wire. I've never seen a 12 w one with a single output as they don't make the bulbs for that (at least my suppliers don't list any and they carry a bunch of parts from many brands for lots of apps -- not to say it's impossible, but I've not seen one) so something you're saying doesn't add up. Might be wise to unpot one and find out what's really inside.

It won't be stiff with that ~27 pf cap in the output (and many stages in series on top of that). I use .01's when I want stiff (like to run a phototube or ion source). Then I have to limit the current input to the thing to keep from frying it if the load does something stupid.

I think you're going to find the first real arcs once you have a bunch of stages toasts most of this, but I'd be happy to be wrong (and CliffS and his counterpart at Glassman who've spent many decades working this problem all day every day would be quite redfaced if it truly works and they missed it, not to mention a bunch of other career HV engineers). Of course, all those caps you're having to add make it more expensive to make in real production -- the HV ratings really cost, which is why the standard stack design keeps the volts/cap constant and low.

Since the transformer secondary is not perfectly coupled to the primaries (usually one push pull for the transistor outputs and one for the royer oscillator feedback) the load will affect the oscillators, and making something resonant with the leakage inductance will have "funny" effects on it all. In the ones I use, you can unground the other side of the secondary winding and float it a little -- but not much, the insulation there isn't that great. The only things I've seen that just use an inductor (tapped at that) and kickback are the camera flash supplies -- and they are super inefficient when tested here.

[chrismb]

Every CCFL supply I've had grounds one end of the transformer

When you say 'ground', Doug, what is that grounded to? Should I see continuity through to the input wires?

I've never seen a 12 w one with a single output as they don't make the bulbs for that

It's 4W rated. I'm just running it at a higher voltage, and it seems to run fine like that.

It won't be stiff with that ~27 pf cap in the output (and many stages in series on top of that). I use .01's when I want stiff (like to run a phototube or ion source).

Well, yeah, I have been using 10nF's, but today I was trying out different ratings and it didn't seem to make much difference. So what I have in mind is that I'll drill down to the cap, tap it there and replace the existing cap with a higher voltage rating. The thing is, you just made me realise that whichever side of the transformer is held to ground, there is still that small ballast cap propping the supply up. Now, if I am pulling good juice out of it like this (which I am!) then I need not replace that cap with anything bigger. And as that is going to be, as you say, 30pF or so, all I need are 100pF 30kV caps, which are much less that a $/cap.

This is going to work out cheaper than I expected!!

I think you're going to find the first real arcs once you have a bunch of stages toasts most of this, but I'd be happy to be wrong (and CliffS and his counterpart at Glassman who've spent many decades working this problem all day every day would be quite redfaced if it truly works and they missed it, not to mention a bunch of other career HV engineers).

Well, sorry, yes I think they have missed this option. I have looked very hard and have not found any previous accounts or attempts of using multiple independent power supplies feeding a single stack before.

A conventional stack just seems wrong, after contemplating this. Why have stiff lower stages and floppy upper stages, when you can feed the same power into each with a decoupling cap? If all the power goes in the bottom, you've got too much loss at the excessively-stiff bottom and too soft and squidgy stages at the top!

Of course, all those caps you're having to add make it more expensive to make in real production -- the HV ratings really cost, which is why the standard stack design keeps the volts/cap constant and low.

As above. I'm now aiming for 100pF ratings.

Since the transformer secondary is not perfectly coupled to the primaries (usually one push pull for the transistor outputs and one for the royer oscillator feedback) the load will affect the oscillators, and making something resonant with the leakage inductance will have "funny" effects on it all. In the ones I use, you can unground the other side of the secondary winding and float it a little -- but not much, the insulation there isn't that great. The only things I've seen that just use an inductor (tapped at that) and kickback are the camera flash supplies -- and they are super inefficient when tested here.

I'll have to munch through detail like this as it hits me in practice. I make absolutely no claim to be an electronic's whizz - quite the opposite! I'm an 'ideas' man who tinkers with hardware based on a few [possibly misconceived] simple ideas, but I keep fiddling until it works. It usually does, and the things that don't work... well, I've just not fiddled with them long enough, then!!

[Doug Coulter]

This is displaying so many misconceptions I'm not sure where to begin. Hooking another 30 pf cap in series with the original one nets 15 pf and half the DC will still appear across the internal one.

Yes, you should probably see continuity between the ground of the HV output and the low voltage input ground. All mine do, anyway. Was it too much effort to just get out an ohmmeter Why would they add expensive isolation to something that needs none in the intended application?

Yes conventional stacks have issues. If you try to make the lower stages stiffer by using larger coupling caps there, you then have the coupling caps discharging through all the diodes in an arc, whereas if they're all equal, at least only the top diode gets turned back on hard (and fried). Sometimes tricks are used with conventional stacks, like feeding them in the middle (analyze the circuit) which only needs two large HV coupling caps, instead of large ones per stage. The manufacturers are after all trying for robustness and much more power than this makes, at a price - and have regulation as well. You can get that in your design to a point by regulating the input DC based on a signal from a HV divider off the output.

Yes, the output power of these will probably be better at higher input feed if things can handle that - it's the current that cooks the transistors and limits power at the nominal input voltage. Of course pushing that will make it easier to fry the transistors with an arc kickback overvolt back through the transformer.

You are still hooking a bunch of power supplies in series, and it still won't be "stiff" it will be 1/N as stiff as one. If you don't limit output arc current with a big R (which reduces stiffness further) the kickback to the transformer primaries, since these are single ended, will still fry the transistors.

That power supply book JohnF recommended would be really good for you and answer nearly all your questions -- you might even wind up understanding transformers and inductors!
Switching Power Supply Design, second edition
Abraham I. Pressman
ISBN 0-07-052236-7

[chrismb]

This is displaying so many misconceptions I'm not sure where to begin. Hooking another 30 pf cap in series with the original one nets 15 pf and half the DC will still appear across the internal one.

I wasn't planning to (now). As I said, I'll drill down to a track between the transformer and the existing cap, and cut it out of the loop.

Yes, you should probably see continuity between the ground of the HV output and the low voltage input ground. All mine do, anyway. Was it too much effort to just get out an ohmmeter

Just wanted to get an impression of whether this is 'typical' (irrespective of what mine are like). The thing I didn't mention (I kinda thought I'd put down enough 'pros' above already!) is that where you question whether I will get more of these inverters, it doesn't matter, because any-old AC supply will feed a stage. Need not be the same frequency nor even voltage! Each stage is a capacitor-isolated ac doubler stacked on top of the last. In fact, it would make sense to keep lower frequency supplies towards the bottom and higher frequencies at the top so as to make the requirements on the coupling caps less at the top end.

Yes, the output power of these will probably be better at higher input feed if things can handle that - it's the current that cooks the transistors and limits power at the nominal input voltage. Of course pushing that will make it easier to fry the transistors with an arc kickback overvolt back through the transformer.

For sure. But are you saying it isn't worth building and trying??.. You're not losing your sense of trying something out, Doug, are you, to get real data out?

You are still hooking a bunch of power supplies in series, and it still won't be "stiff" it will be 1/N as stiff as one.

It'll be as stiff as a pile of stiff power supplies in series. For sure! But I bet it'll be stiffer than a single-supply-fed stack.

If you don't limit output arc current with a big R (which reduces stiffness further) the kickback to the transformer primaries, since these are single ended, will still fry the transistors.

If they don't want to play, let 'em blow! I'll figure the problem later!! So far, everything is working 'in spec' (nothwithstanding the package rating for the inverters is 12V, but the internals shouldn't be exceeding their ratings).

[Doug Coulter]

Nah, it's worth building and trying if you can accept it might all pretty much all fail in one shot. It won't be as stiff as a well designed single fed stack (assuming what that's fed with is stiff enough). Hard to compare without specifics of each. You might not see the failure modes I'm predicting till you get a good number of these stacked up - you'll definitely see failures on the internal coupling cap above maybe a handful of KV, certainly by 10+, since it's not rated for that and will have half the volts across it (maybe more if your added one has any leakage or is bigger -- and air itself has some at these volt levels). So you're going to have to dig potting compound no matter what. It actually does matter what you feed things with, mixed frequency is "interesting" but in the standard stack, particularly the full wave version with 180 deg drive phase, the situation lets you use far smaller caps because of what's drawing current through what part of the cap stacks when -- things work out really nicely for that. One reason that people use all same sized caps is to handle what happens in an arc discharge or any sudden load. The small ones can be driven reverse polarity in such a case, which turns on a diode to "protect" it, more often than not, losing the diode in the process. The other huge advantage to a full wave stack is any arc current through the coupling caps cancels in the transformer driving it, instead of blowing back into the primary and the drive semiconductors.

Remember your ballast resistor on the output has to not arc over internally itself with full supply volts across it. A small carbon R isn't going to cut it for that, nor will a 10w wirewound -- tried them too. The pros use up to about 50 2" long carbons in series/parallel. My main fusor uses a cartridge resistor 14" long -- and I've cooked a couple of those arcing between the turns.

The "ballast cap" in these is designed to drop the output to about 200v for a 1kv one at rated current in a fraction of a half cycle for the bulb characteristics. Not very stiff at all.

So, anywho, why are you making this? Building a Fusor?

[chrismb]

I have a very nice 'non-destructive' 30kV insulation tester which I have put to work testing these CCFLs. This is very clever and does not cause damage [that I have ever noticed], only looking for the first onset of 'extra electrons' finding their way through. (Actually, its ability to protect the test item is a bit unfortunate in this scenario, because I was trying to use it to make the cap fail - to hopefully fail it closed circuit!)

Between the output lines and the input, there is no connection. It begins to breakdown at around 8Kv. There is around 8GOhm between -ve and one of the output lines, and 5Gohm to the other. Similarly, between output lines and the case (which is not connected to the input lines) the resistances are around 5 and 3GOhm. Presumably, the one with the higher resistance must be the one with the cap in series? I do not need to guess, though, next week I will X-ray it and see what goes on inside, and where I need to drill into it to bypass that cap.

So, anywho, why are you making this? Building a Fusor?

Good question. My device (more details will flood in at some stage - just in the thick of things at the moment and don't even have much of any time to be in the workshop making progress!!!) should operate in a fusor-style 'DC mode', and seeing as I am geared up for neutrons right now, I though I'd wind up the volts and see if I can get some experience with measuring the pesky things first. I know I'm getting the right potentials in it, because I've measured them in the chamber with probes, but I had never originally set up for high volts and I have, only, some 6kV DC-DC converters (designed for floating loads) which are only rated to 3.5kV isolation between in-out. I've put a couple 'back to back' (ground in the middle) such that there has been a differential in the chamber of >12kV, but even that [running the full 6kV where there is also the ground reference of the input lines present] is exceeding the level and I don't want to stack them on top of each other for high volts, I suspect they would die like that, and for no good reason because they can't put out more than a mA, which hasn't been a problem so far but in DC mode at 12kV they are already drawing that much current, so I will clearly need more at higher volts.