Building an adjustable high voltage bench power supply

Been contemplating switching over to the new-fangled Linear Technology 3080 three terminal regulator. This one has some nice features, specifically a very low internal noise floor and programmable output with just a single resistor. Several Spice simulations show that is has both plusses and minuses with respect to the LM317. In addition to lower noise floor, the other plus is about 3 dB better power supply ripple rejection than the LM317. Minuses are that it has about 8 dB worse load regulation than the LM317. So, for now, I do not see that the benefits outweigh the drawbacks in this application, so I’m sticking with the LM317 as I continue to plod through this thing.

Not all LM317s are created equal! The On Semiconductor LM317T oscillates madly at about 55 KHz when used in this application. It has done this consistently with three different samples, so I don’t believe it’s just a “bad” sample I have. So far, the ST Micro LM317T performs admirably—extremely quiet DC coming out and is able to hold its liquor…until you exact max current (350 mA) at somewhere around 30V output, which means the fets are blocking somewhere around 520V DC. Under those conditions, the STMicro can’t control the gates, and poof… Oddly there is no oscillation that I can see on the scope at the moment of failure.

There are two additional brands of LM317 I’m going to try: Texas Instruments and Fairchild. I also ordered a few STMicro LM217s (a bit better performance than the 317). Perhaps one of these will do the trick. I have high hopes for the TI brand. TI seems to make high quality stuff.

By the way, it did occur to me (several times) that maybe this design simply can’t do what I’m asking it to do, namely the hard task of supporting max current at min voltage output. I realize this is a hard requirement, but a full-proof design must be able to support it. It’s not realistic to ask the operator to “please make sure you don’t ever short the leads”—it’s going to happen once or twice I just know it. If I were making a “simple” relatively small adjust range that did not need the requirement of surviving a full-on short, I’d be done weeks ago. Indeed, this has been a most challenging project.

But I’m not to the point of throwing in the towel yet, however I admit I've been tempted a time or two (or three) over the last few weeks.

If these other brands of LM317 do not work out, I think I will have exhausted all of my available options, although I could try a smaller form factor fet (TO-220 outline instead of the TO-247 outline I’m currently using). That might give the LM317 an easier load to work with….dunno.
 
If you want fool proof design, you need much more complex circuit. Try to find HP publshed application notes regarding power supplies. It was issued somewhere in early 1970s. It describes how they designed bench power supplies that can regulate both current and voltage. You can also find service manual of their high voltage power suply of that era and replicate circuit, of cause using modern components. That would be a better path then continue frying mosfets.
 
I'll see if I can find some HP schematics, they might provide some insight I obviously lack. However, if I move away from the Maida approach, my inclination would probably be to try some sort of hybrid--for example pass elements made from 6L6 power tubes with a transistor error amplifier of some sort.

Before I completely change directions, I really would like to understand why this circuit is not kind to the fets. Sure doesn't seem like I am running them out of their safe area of operation. I mean, the fets are spec'd for 950V drain to source voltage, and at something like 10A each. I'm running them at a fraction of that voltage and current (anywhere from 2 to 4 fets in parallel). The failure seems to happen consistently at about 350V to 400V on the drains, 35V on the sources, and at 350 mA output, where the current load is (supposedly) split across the 4 fets. With 4 fets in parallel, the gate capacitance would be higher than just a single fet, so I wonder if that has something to do with it.

The only thing that makes sense is I'm somehow exceeding the gate-source voltage (about 30V I think on these fets), but these fets have built in projection zeners, so again, doesn't quite make sense to me. Logically I'm deducing there is really some sort of oscillation going on but I just can't see it.
 
Have you tried a small cap across your current limit adjustment pots? it would change something... I think it reacts too fast and over/under shoots
 
I have not tried that, Scott, because I took out the current limit transistor. All that's there now is a single 30 ohm resistor for current limiting, which limits to about 350 mA over an 11V drop of the 317. The limiting is extremely smooth, no oscillation.
 
I'm done with mosfets

So I ran one more very carefully controlled test. With all the other tests I ran there was always the possibility of other factors that I felt might have caused the fets to fail--oscillation coming and going under seemingly unpredictable conditions, problems with active current limiters, fets not properly secured to the heat sink, wiring errors, stuff like that.

With this last test I am very sure everything else was in order. But since one of the fets failed in the same way as before under roughly the same conditions as before, I can say now with very high confidence that this design cannot work with the voltage adjust range and current range I want.

So, what do do? I don't have a strong desire to build an extremely complicated thing, but I do need the basics to work well, and I really am still desirous for all the great electrical characteristics that the LM317 offers--very good power supply ripple rejection (something like 60 dB), EXCELLENT load regulation (~75 dB), and very low output resistance (< .25 ohm). So I've decided I'm going to take a break for a while, but when I come back to it, I'm going to look at a hybrid Maida using power tubes (the 6L6) as the pass devices, and the LM317 as the regulator.

I've already attempted the use of depletion mode mosfets in the actual build, and the power supply purrs along nicely when you keep the currents and voltages at sane levels, so I am confident power tubes, which have similar electrical characteristics as depletion mode type mosfets, but arguably are way more rugged at the extreme voltage and current boundary conditions, will work well too.

Also, a shout out to the Texas Instruments LM317--it is built like a tank. Its tab (back plane) in the TO-220 outline is at least twice as thick as the other brands, so I will continue to use this brand moving forward in my testing. I will of course need to use the High Voltage version now, since the LM317 will need to drop somewhere in the neighborhood of 30V to 40V when working with the 6L6 tube as the pass element (the grid is about 30V below the cathode at the average operating point I'll be running them at). Certainly there will be some details to work out such as floating filament and screen grid supplies, but that should be straight forward.

Was this a waste of time? NO! I have learned a ton about how transistors work! Was it smart use of money? Hmm, just don't tell my wife! :).
 
Hiya,

Sorry about the Mosfet adventure.

Kinda discouraging even from here in the peanut gallery.

I give you credit and praise for being thorough and not quitting at the first sign of magic smoke.

I know you will get it sorted out and taking a break is a awesome idea.

Frannie
 
This circuit works very well with FIXED output voltage. When variable output is required in that high range (0-600V), much more complex circuit is required, I am afraid.
 
I give you credit and praise for being thorough and not quitting at the first sign of magic smoke.

I know you will get it sorted out and taking a break is a awesome idea.

Thanks! What a ride this one has been.

This circuit works very well with FIXED output voltage. When variable output is required in that high range (0-600V), much more complex circuit is required, I am afraid.

And having lived through this, I'll add that this design works well when the input voltage is no more than about 30V or so above the desired output voltage. But even if you had fixed output (believe me, I tried it), when you attempt to block more than about 400V across the fet at high currents, the gates crap out.

Also, I found that a moderate amount of variability is okay, maybe up to 30% down from the max output voltage. Under those conditions, this design appears to be stellar, BUT ONLY if you do not need to design for an output short. If there is reasonable chance of a short occurring, the mosfet approach falls short when the drain voltage exceeds about 400V.

This all might work if you doubled the number of pass devices (to 8 or 10 or so), but I'm not going to go to that level of experimentation.
 
+1... Your perseverance with this build is great!! This is the kind of dedication that makes the impossible happen... I've been plowing along with the 811 amp, and experimenting and pondering the results for a while, is the only way to succeed...
Best of luck, I'm sure you'll get it...
 
Thanks p1. So the HP approach back then was basically a series pass design using a BJT with a cascaded BJT error amplifier. Probably standard procedure there. How they they did current limiting with that "mixing" amplifier is quite nifty though.

My observation is that they are using a BJT (a current controlled device) as the pass element, not a fet (a voltage controlled device). I expect that was deliberate. My hunch is BJTs when used as pass devices, are better able to control voltage at the extremes, and especially at max current and min voltage out, exactly the conditions under which MOSFETs at least seem to be vulnerable.

Well, it seems power tubes can do this too. Seems to me power tubes receive this kind of punishment on a regular basis in power amps when driven hard--at times blocking nearly all of the B+ voltage when driven deep into AB1 or AB2 territory. So I don't know...I'm not excited about building big heat sinks and paralleled darlington pairs. Power tubes it seems have built in heat sinks (the glass), so I'm leaning toward using power tubes as the pass devices.

I sincerely do appreciate the tip on that schematic though, as I actually might use some of those ideas in who-knows-what I might do in the future. :thmbsp:
 
When HP designed that supply, there were no power mosfets. yet, so they used BJT. They had to use stacked pass regulator too, when high current was involved due to limitations of BJT (in another unit). You can use that circuit but substitute BJT with mosfet. Key is how voltage and current limit circuits work together with a separate reference source. Another feature is no reference to ground, so it can work as either positive or negative source.
 
Thanks p1. So the HP approach back then was basically a series pass design using a BJT with a cascaded BJT error amplifier. Probably standard procedure there. How they they did current limiting with that "mixing" amplifier is quite nifty though.

My observation is that they are using a BJT (a current controlled device) as the pass element, not a fet (a voltage controlled device). I expect that was deliberate. My hunch is BJTs when used as pass devices, are better able to control voltage at the extremes, and especially at max current and min voltage out, exactly the conditions under which MOSFETs at least seem to be vulnerable.

Well, it seems power tubes can do this too. Seems to me power tubes receive this kind of punishment on a regular basis in power amps when driven hard--at times blocking nearly all of the B+ voltage when driven deep into AB1 or AB2 territory. So I don't know...I'm not excited about building big heat sinks and paralleled darlington pairs. Power tubes it seems have built in heat sinks (the glass), so I'm leaning toward using power tubes as the pass devices.

I sincerely do appreciate the tip on that schematic though, as I actually might use some of those ideas in who-knows-what I might do in the future. :thmbsp:


At that point in time a fet with that voltage and current capability was just a dream .

i would suggest something like this , the best of both technologies

http://www.mouser.com/ProductDetail/Infineon/IHW30N100T/?qs=gLoIpP5ymzqoNvrUoQKftA==

http://www.mouser.com/ds/2/196/IHW30N100T_Rev2_7-78629.pdf

http://www.mouser.com/Search/ProductDetail.aspx?qs=OSPYYm892W6CqUqwZ5fHDg==
http://www.mouser.com/ds/2/149/FGA50N100BNTD2-244159.pdf


A two stage regulator might be a good idea also so the drop across the pass device never see's the the entire voltage drop .
 
Last edited:
At that point in time a fet with that voltage and current capability was just a dream .

i would suggest something like this , the best of both technologies

http://www.mouser.com/ProductDetail/Infineon/IHW30N100T/?qs=gLoIpP5ymzqoNvrUoQKftA==

http://www.mouser.com/ds/2/196/IHW30N100T_Rev2_7-78629.pdf

http://www.mouser.com/Search/ProductDetail.aspx?qs=OSPYYm892W6CqUqwZ5fHDg==
http://www.mouser.com/ds/2/149/FGA50N100BNTD2-244159.pdf


A two stage regulator might be a good idea also so the drop across the pass device never see's the the entire voltage drop .
These are for switching. I didn't see anyone using IGBT in linear mode.
 
I built a variable HV supply based on Mototola's MC1466L floating regulator back in the early '70s. Worked pretty well until I managed to destroy the chip somehow. Something like this, without the variable xformer.

2009716225133902.gif
 
Let's see, my last post was in May. In about June of this year I built up a whole rig of matched and paralleled BJTs in a Darlington configuration and subbed it in for the MOSFETs for the pass device. As usual things worked really well until at the extremes, the extremes being about 300 mA current draw dropping ~500V across the pass devices. Given my level of sophistication (which isn't much) and my knowledge, I'm not able to get this design to work in it's current form.

If I were to limit the output voltage to ~300V max or something, or if I were to somehow be able to detect an over current situation and reduce the unfiltered input voltage in those situations so that in the worst case (short circuit conditions) the pass device never needed to drop more than ~300V, the thing would work beautifully. Unfortunately, that second approach is beyond my understanding and the first approach doesn't give me the voltage adjustment range I want.

I've put this project aside for now while I built my son an amp for a Christmas present. Then I'll get back to this, probably early next year.
 
Back
Top Bottom