A good homebrew 500V 1A power supply

[Wolfgang]

Hi,

I am looking for some ideas to build a GOOD industry-grade linear power supply for 500V/1A

"Industry grade" means:

- powered by 230VAC
- Ripple and noise below 1mV RMS / 5mVpp in a bandwidth of 1MHz
- 1 A output, currrent limit settable from 10mA to 1A, with 1mA precision
- regulation from input better than 100ppm, from output better than 100ppm over full range
- Voltage settable from 0 to 500V, with 0.1V precision
- Misuse tolerant, short circuit proof.
- Thermally stable and protected
- Below 1J stored energy in the output caps

I made some smaller supplies, look here:

https://electronicprojectsforfun.wordpress.com/power-supplies/high-voltage-lab-power-supplies/

IMHO, some difficult points are:

- pass transistors (I would suggest IXYS linear MOSFETs)
- transformer range switching / preregulator design
- current and voltage regulator loop design
- a really stable reference (LM399 ?)
- overcurrent detection and protection - maybe several lines of defense are necessary
- decoupling of meters and settings from the HV side.

Any ideas ? Old service manuals from top of the line equipment ?

[Dave]

The ripple and noise requirements you provided are quite stiff. What purpose will this supply serve?

I'd use a floating regulator for something like this, control it all with a DAC and have the ADC for the readback there as well. Then just use some optocouplers with reasonable isolation rating to pass the data to the ground referenced UI. Perhaps SPI, I2C or some other data bus that is easy to isolate.

For the pass element I'd use a BJT cascode, because it would be cheaper than specialty IXYS parts and easier to find replacements, should something to poof.

Make sure that the transformers that you use have sufficient dielectric strength between primary and secondary. Also consider how you're going to attach the toasty bits to the heatsinks - those thermally conductive condoms that cover the whole package come to mind.

Last but not least - resistor thermals and noise. Because you want to keep noise to a minimum you want relatively low resistances in the feedback network, but at the same time you don't want the resistors heating up and drifting around, because it would hurt stability/precision.

I just tossed a bunch of ideas here, feel free to use/discard them as you see fit.

[Wolfgang]

Hi,

the purpose is to power some low noise, but power hungry tube equipment (special noise diodes running high current).

A floating design makes sense. I have used this design in other PSUs and it worked well with mechanical pots to adjust voltage and current, like this:

https://electronicprojectsforfun.wordpress.com/power-supplies/high-voltage-lab-power-supplies/a-250v-150ma-variable-linear-power-supply/

Of course, this one could be blown up to 1A and more voltage, but I'm afraid that ripple and noise specs could not be met without a rework.

Do you have a sample circuit for a BJT cascode for high voltage ? I have only seen them for low power.

The transformer problem will be solved by using standard off-the-shelf isolation transformers with 2 chambers (speced for 4kV isolation).

Resistor noise in the feedback network could be mitigated by making separate AC and DC paths.

In order to meet ripple requirements, a preregulator is an option in my view, something like here, only more powerful:

https://electronicprojectsforfun.wordpress.com/power-supplies/high-voltage-lab-power-supplies/a-1kv-50ma-linear-power-supply/

[David Hess]

- Ripple and noise below 1mV RMS / 5mVpp in a bandwidth of 1MHz

1mVrms over 1MHz is 1000nV/SqrtHz.  At a gain of 100, that is 10nV/SqrtHz maximum from the reference and error amplifier amplifier which is not feasible without filtering but filtering is not difficult.  The output can be filtered with an LCR network and the reference can be filtered to remove high frequency noise.

- Below 1J stored energy in the output caps

So let's see; 1 joule and 500 volts is 8 microfarads.  That is small enough to require some care with frequency compensation.  Low output capacitance will make low noise more difficult but I agree that output capacitance should be minimized

- pass transistors (I would suggest IXYS linear MOSFETs)

A bipolar design would probably need three $5 output transistors in series or even six in three series two parallel.  On the other hand, a MOSFET design would need more than one part to handle the power dissipation anyway.

- a really stable reference (LM399 ?)

There are lots of suitable references but they will all need to be filtered.  A buried zener like the LM399 will have the best low frequency noise.

- overcurrent detection and protection - maybe several lines of defense are necessary

Considerable leeway in the power and safe operating area requirements can be gained by limiting current at low output voltages.  Do you really need 1 amp over the entire 500 volt output range?

Any ideas ? Old service manuals from top of the line equipment ?

The HP 6448B supported up to 600 volts at 1.5 amps but it used SCR phase control for regulation.  Their lower current high voltage power supplies use SCR preregulation followed by a series regulator.

[schmitt trigger]

From your posts and your blog, I can see that you are a very experienced engineer.

So there is not much advice I can give you, other than to consider a soft-start circuit and redundant protections in your design.
I.e, an output relay, in addition to disabling the main series pass transistor.

[Wolfgang]

Hi Dave,

thanks for the comments; I am fine with most expect for your suggestion of cascaded bipolars as pass transistors.

Do you have a schematic that uses this in a commercial supply ?

All the "beanstalks" and other series solutions I tried are so slow that both ripple rejection and load change recovery became a problem.

Thanks
  Wolfgang

[MaBerY]

I'm involved in the same process. I looked at several designs and I concluded that for that kind of currents and voltages some preregulation is needed to keep SOA and part counts low. The HP890A  used thyrisors to alter the rectified cycles and that seems a nice solution to keep the pass circuit simple.

I also found Tom Biskupic's blog article which looks promising https://wahz.blogspot.com/2015/10/lab-power-supply-project-pre-regulator.html

Another approach is the Frank Kneifel design http://www.jogis-roehrenbude.de/Roehren-Geschichtliches/Roehren-Netzteile/400V-Netzteil/Kapitel1.htm  using a TCA785 dimmer circuit feeding the transformer. I think that design is more noisy due to the phase angle control.

Since I have a few classic transformers in my junk drawer also carrying the 6.3 heater windings I did not want to do the regulation on the primary side but on the secondary side like the HP approach.

[lordvader88]

What are u going to use it for ?


I would like a fairly similar HV source for vacuum tubes. Sooner or later I will have to buy or make 1.

Something seems pretty wrong about using SMPS for HV for tubes, so I'll have to buy a nice new transformer.

[Wolfgang]

Hi,

I'm gonna use it for tube circuitry, too. I'll update my webpages as soon I have a design ready.
My current idea is a combination of a range switching scheme for an isolation transformer (230 -> 2x 115V) plus a doubler scheme.

Some info is already here (not complete) :

https://electronicprojectsforfun.wordpress.com/power-supplies/preregulator-circuits-for-linear-power-supplies/

[Dr.Krieger]

Just a noob's question:
Why go with that (low-Vce?) "BJT cascode" thing instead of using a non-IXYS, cheap 1000V Transistor?
I see a BUF420AW 1000V 30A in Mouser for about $10, that's not so bad if you could use one for the job, right?
I know that you would have thought of something like that (or an old BUX89A), i just don't know why you don't want to go with it...
Good luck with your PSU designing sir!

Mouser link:
https://www2.mouser.com/ProductDetail/STMicroelectronics/BUF420AW?qs=ljbEvF4DwOMtKKFpP3X7UQ==
Datasheet:
https://gr.mouser.com/datasheet/2/389/buf420aw-954520.pdf

[Wolfgang]

Unfortunately there is a pitfall using even powerful MOSFETs in linear mode.
They become thermally unstable and burn out even when the power level is way below their dissipation limit, similar to the second breakdown of bipolars.

Googling "MOSFET linear mode" or "Spirito Effect" gets you the info you need.

IXYS can sell their stuff only because its needed in linear power supplies, electronic loads, current regulators or other stuff where MOSFETs are not running in
saturated mode. Nobody would buy them if the cheap switching stuff would survive this, too.

They cost a bit more than switching FETs, but their half-life is so much longer it pays off.   
In a HV Supply, a punched pass transistor could immediately kill your load. That makes the cost of a safe pass device well spent.

[David Hess]

Why go with that (low-Vce?) "BJT cascode" thing instead of using a non-IXYS, cheap 1000V Transistor?
I see a BUF420AW 1000V 30A in Mouser for about $10, that's not so bad if you could use one for the job, right?

Wolfgang identified the problem.

The temperature coefficient of threshold voltage reverses at high drain-to-source voltage so the MOSFET cells suffer from current hogging like a bipolar transistor would suffer from current crowding under the same conditions leading to secondary breakdown.

IXYS and some others make special linear rated power MOSFETs which have better safe operating area curves at high voltages.  MOSFETs intended for switching applications often do not even show this in their specifications.

[floobydust]

A HV power supply like this needs to incorporate at least dual-slope SOA protection, instead of just hoping the pass transistors' are tough enough with simple current limiting.

Output current and voltage across the pass transistors can be looked at, this is common in audio power amplifiers for example.
Michael Kiwanuka: SOA Protection for Audio Power Amplifiers is a good read.