Electronics > Projects, Designs, and Technical Stuff

High voltage adjustable power supply 3kV/1A

(1/2) > >>


     I would like to share one of my ongoing projects. The plan is to have an 3kV/1A adjustable isolated power supply. I want to use it for HF tube amplifier experiments. The main requirements are:

- adjustable voltage 300-3000V
- voltage regulation is nice to gave. But not mandatory if voltage drop is <15% between 10% to 90% load
- isolated from mains
- simplest transformer possible (no push-pull)
- fully protected to overtemperature and short circuit
- adjustable current limit (no constant current mode)
- immunity from strong RF fields

This is the block diagram I came out with:

So far I designed board B, C and D.
Board B contains the main rectifier, soft start circuit and a 15V auxiliary supply for IGBT drivers. There is also an output to notify the PWM controller that the soft start is over.

Board C contains the H bridge, gate drivers and protection circuits. The gate drivers (2EDL23I06PJ) have a current sense input. I set the threshold to 45A. I expect this protection to be activated only in case of a catastrophic event. There will be another current sense circuit at DC output with a  <2A threshold. For over-temperature protection I plan to  use a washer type NTC mounted on the screw of every IGBT. The IGBTs are IGW40N65F5.

Board D is the high voltage rectifier. I used SiC diodes to avoid the need to use capacitors to balance the voltage during reverse recovery.

     I didn't decided yet what transformer to use. I'm thinking on a really large U core type. It seems to be more easy to provide adequate insulation because of the extra available space. I want to use a larger that needed core section to avoid winding many hundreds of turns. The switching frequency will be around 80kHz.
Please feel free to add any critics or suggestions

The power factor will be poor, with a simple bridge rectifier and capacitor. You'll probably need a 32A breaker. How about adding an active power factor correction stage?

At this moment I don't want to increase the complexity even more by adding a PFC circuit. I'll just add a slow fuse. Maybe later if everything works fine I would consider adding one. This supply could be a great candidate to demonstrate the usefulness of such a circuit.

Hmm, how exactly do you expect variable voltage, without regulation?  (Just that you don't NEED it, I guess?)

Is this a resonant design?  The transformer is unspecified so I guess it could be planned as an LLC with integral leakage inductance.  The secondary capacitance will be significant.  Still, no output filtering at all is shown, if I'm not mistaken?

Do recommend full wave doubler rectifier, you only need a 1500V secondary and 3kV worth of diodes (so, probably 4 in series would still be a good idea).

Might as well get co-pack IGBTs, you're planning on putting diodes in with them anyway it seems (another indicator of resonant? -- forward doesn't technically need them, though diodes at all are likely still a good idea due to leakage, even if smaller ones).

Not sure why the gate clamp diodes (D2, etc.).  If the gate blows up, the driver's toast either way; typically a transistor fails as a three-way short, thus delivering some fraction of its fault current to the gate pin.  You need big TVSs and extra series resistance to actually have a driver survive.  Add in a TVS to the 15V rail, and what would look like ESD protection between the logic source and the gate drivers (well, the optos in this case), if you want to treat the IGBTs and drivers as disposable and replace them after a fault occurs.

The opto is... rated for enormous isolation, and pricey!  I don't see a need for this; typical digital isolators these days are rated for more than enough dV/dt and voltage to do the job.  You probably have worse things to worry about if >5kV somehow appears across your isolation barrier...

Oh what the heck, it's slowed down a hell of a lot too, C5, C6, C11, C12 are way too big.  The optos are 100/200ns and the drivers another 300/400ns (depending on rise/fall, I don't remember which is which), there's at least half a microsecond already between what the controller is doing, and what's actually happening.  The RCs will add another almost a microsecond!

The discrete flip-flop seems... incongruous with the rest of the circuit?  There's also no opto conveying its state to the controller, that's not great.  RESET seems to be hanging.

Hm, all the temp indicator LEDs are wired hard in parallel.  I suppose there's supposed to be diodes to FAULT, just a typo.  FYI, you may want hysteresis to eliminate chatter, and/or latching to maintain which source caused the fault.  For extra bits of flip-flops, consider a CD4043 or the like.

Speaking of CMOS, CD4011 or the like will be just fine for flip-flops.  Or 4043s, or 4013s, etc.

Or don't run any logic on 15V, use 74HC on 5V, you have it handy already.  Which is, I think just because the opto receivers need 5V?

BTW, beware of self-heating of the thermistors.  470R is kind of a strong pullup, even from +5V.  I prefer >=10k thermistors.  (Although I often run them from 12V, which is about the same situation, hah.)  You can always set the threshold voltage lower (using higher pullups), too.

Oh and the input, no way in hell you need that much capacitance, about 2mF should be enough.  One computer grade cap should do, or a handful of snap-in types.  Preferably the latter, as they're quite low inductance -- probably fine to place them beside the inverter and skip out on the local (film?) bypasses.  Or just a few uF local, may still be handy (DC Link MKP type are cheap and dense, well suited as the description suggests).

Oh, and I just noticed your main diagram suggests 20kV isolation, which, I don't have a clue where you'd even find an aux transformer rated for that.  Another custom windup?  You'll need to keep shopping for optos, that's only rated 16kV.  Again, I don't get it, you'll have to put in real work to get anywhere close to that kind of offset.  Like, intentionally charging the poor thing on top of a CRT's 2nd anode, or, stacking it on a live Tesla coil.  Or an electrically-small antenna transmitting some ~kW.  None of which makes any lick of sense for a transmitter power supply...

Likewise, RF immunity is easily provided by metal enclosure and heavily filtering all connections passing through it.  Use a mains inlet filter.  Create HV feedthrus for your output.  Bypass grounds to chassis (typically with small Y1 rated caps), or ground to ground as the case may be.  Actual RF inside should be ~volts, and most of that being electric field from the transformer and rectifier.

So, needless to say, this thing will be two dozen kinds of deadly in operation, but especially so if it's actually, intentionally, operationally, getting up to voltages like that.  Like, I personally would not operate it without being on the far side of a lockout-gated metal fence, dangerous.  (Power companies have test rooms like this, the safety procedures are quite ornate -- and with good reason, even if they're only doing a few mA in there for a given test.)

Also, again, if you're going to have that much voltage to isolate, then uh... how the hell are you gonna control the poor thing?  All the controls are suggested as being at output level.  It'd honestly be safer to wire everything on the primary side, where it's expecting only a measly 2.5kV (or 3-4 depending on CAT) transient above the paltry 240VAC mains...

So, show stoppers:
- I don't see ANY indication for AC current sense.  Maybe you're putting a current transformer in the wiring diagram?  Dunno.  The control MUST be current mode, either for a square-wave forward converter (peak or average current mode), or for resonant (current limiting).  (Voltage mode is a bad idea, best forgotten as a relic of primitive times gone by...)
- No feedback from the inverter.  So the fault can latch and the controller just assumes everything is fine.  And also has no way to reset it.
- You haven't shown the last sheet yet, but beware of short-circuit (transient) fault currents, especially with regards to the current sense, and anything else touching it (and also, for S&Gs, calculate the peak output current under such conditions ;) ).

And the real worry point:
- You seem to have made enough mistakes that it suggests you're a novice when it comes to switching power supplies, or high power projects.

Please, PLEASE, get some experience first.

You can get all the experience you need, just making piddly little, like, 12V 1A converters.

It all scales just fine.

Well, inverter parasitics don't scale, but that's part of the fun, make a few different ranges of voltages and currents, and make shitty layouts to exaggerate the strays.

Discover why I'm so insistent about current-mode operation.

Then, and only then, scale up to the low kW, and maybe to the low kV, and build this project.

You can also discover why MCU controls are a bad idea -- unless you have much, much more experience.  Mind -- I have no idea what you were planning, but I want to head this off too, just in case.

Would it surprise you to know, as many years experience as I have, I still have yet to design a software SMPS control?  It's true.  There are several reasons, some practical (who cares, who would pay for that when I can drop in a more-or-less proven IC that does the job already--), but some also to do with experience.  Software IS bugs.  Only very, very good software is relatively free from bugs.  And I don't know that I'd say any software project is truly free from bugs, aside from the most trivial cases.  (There are "provably correct" software projects out there -- some even of quite surprising scale -- but this shifts the burden of proof onto the specifications themselves, which may in turn contain "bugs".  Writing software compliant to a spec, is at least as hard as writing a truly comprehensive spec.)  Let alone my software, which honestly is at or above average from what I've seen of other embedded projects, but is hardly, like, medical or aerospace grade.

No, an SMPS isn't likely to cause immediate bodily harm like a robotic surgeon or flight computer could, but... well, we're also talking about voltages that can reach out and touch you from whole centimeters away.  So, do you really want to take the chance that you might have bugs in there?..

And so, that's the major reason why I don't, and prefer hardware which I'm both more confident in to begin with (but that's just me), and more confident in being able to produce a "correct" design, based on good design principles (like minimizing internal state).

Your design so far, by the way, does have this going for it, or the beginnings of it anyway -- if the driver fault signals work as planned, they should be able to turn off the IGBTs well within their maximum short-circuit time, and the circuit simply clicks off (the click will likely be audible!), rather than causing a cascade failure (typically, one transistor fails, then the other transistor switches into it, drawing short-circuit fault current; the driver, optos, maybe mains rectifier, maybe some output stuff, fail from there).  Note that, as shown, this WILL NOT protect against transformer shorts to ground, as the high side transistors are not monitored -- for that, you may wish to consider additional circuitry, perhaps drivers with desat detection as well (or instead), etc.  I've had excellent results with this, even just using a discrete desat circuit; thanks to this, plus a hardware fault latch, I'd just about be able to fit all the transistors I've ever blown, in the palm of one hand.  (And the main reason I can't, is because I've blown a couple of industrial modules, that aren't quite palm-sized; if you allow stacking, then sure.)  Other people seemingly almost brag about having blown "buckets" of transistors in their careers (and this, well before retirement age, I might add); I cannot say I agree with such an attitude.


Hi Tim,

Thank you for the huge reply :)
I'll try to clarify most topics.

First some extra introduction.
My background is embedded software engineering. Electronics is just a hobby. I'm mostly interested in RF circuits.
          I have no experience is SMPS besides low power/voltage converters you normally have around microcontrollers. At home I worked a lot with high voltages both DC and RF because of my interest in high power tube amplifier. I know what parts I can touch/probe and what things can blow. I’m aware of the dangers and I plan anyway to supply first the IGBT bridge from a low voltage current limited DC supply. Maybe as low as 24V if the IGBTs switch correctly. I can increase the shunts to check the protection timing at low currents and so on. I also have a 1.4kW 230V 1:1 separation transformer for later tests.

My plan is to make a hard switching open loop supply for a class AB2 amplifier. This amplifier is a very dynamic load. Its current varies with voice modulation between tens of mA to full current ratings. A current mode control would not be suitable.

Later I will try to use a voltage feedback from the output for the PWM controller (SG3525)
I think that a resonant design is too complicated and as far as I understood it is best suited for a more constant load.

Now back to your questions:

I expect to change the output voltage by changing the duty cycle. Most likely I'll use a SG3525 PWM controller. I'll use a feedback voltage from the output if there will be no stability issues.

About a voltage doubler rectifier, this was one few other options I considered. Besides connecting in series a few smaller supplies. Since it will not work at 50Hz, the voltage multiplication at this power level might be a viable solution. But I didn't find a good enough reason to use it.

Actually in my first design I was planning to use a IGBT pack. There are models that contain also the gate drivers and temperature sensors. All major power semiconductors companies have a family of them. This option was very appealing. The reason I drop the idea was because they were all limited to 20kHz. With a few exceptions that have an outrageous price. Such a low switching frequency would result in more transformer turns that I have patience to wind.

The purpose of D2 is to limit the severity of a short circuit. During a short circuit there would be a current flowing through the gate-drain Miller capacity. This will rise even more the gate voltage slowing the IGBT turn off. There is an app note from Infineon where they recommend this approach.

About opto. Most common types are rated to 5kV. Now this value is specified for a limited time. In my worst case I would have 3kV unlimited time. I don't know how to de-rate those 5kV for unlimited time. That is why I pick those fancy optocouplers. The value of C5,5,11 and 12 are the one recommended in the datasheet. They should add about 2-300nS delay. I will adjust them with the real thing running. Depending on what would be the level of interference.

      The discrete flip-flop, together with many other things is copy pasted from Infineon's reference design for this gate driver.
I didn't have any good reason to replace it.
      I don't forward the fault condition from flip flop to PWM controller to avoid an extra optocoupler. If I use the supply in open loop mode only then this is not a problem.

Indeed, there is a mistake with temperature indicator LEDs being in parallel and without a latch!

For the 50Hz filter capacitors I made a simulation to check the ripple voltage. I can't find it now, but I think with those values the ripple is 5% or something at full load ... I will review it. I also happen to have about 200 new Epcos low ESR capacitors of that type :)

Those optocouplers are rated for 20kV not 16kV. That is why that 20kV isolation on the first picture.
It is overkill, but it will not add much extra cost relative to the total.

Yes, I will make a custom auxiliary transformer. I will order a toroidal core with just the primary winding, and I will add myself the rest. There is a nice toroidal company very close to my home.

3kV is quite a usual voltage for high power tube amplifiers. With the right protection measures it is safe to use it.

Now the showstoppers:

"I don't see ANY indication for AC current sense"
No current mode was planned, see above

"No feedback from the inverter."
There are two fault latches. The first one is the discrete transistors one. This turns off the gate drivers. PWM signal are keep coming from PWM controller. In open loop mode this is not a problem but otherwise you are right. The controller will increase the duty cycle because low output voltage and then it will start with highest voltage. This must be fixed.

The last sheet is not ready yet. The current sense will be based on a INA200. I like it because it has also a latch and comparator.

I don't plan to use an MCU to control it, but an analog method. Something like SG3525.
I write enough sw at work. This is a hobby project and there is no fun to do the same thing at home.

I consider that a transformer short to ground is unlikely to happen. The transformer will be oversized, and it should run cool. I plan to use some high-quality tape to isolate it and it should not be any problem unless I don't install correctly.


[0] Message Index

[#] Next page

There was an error while thanking
Go to full version