PWM controlled Variable High Voltage Power Supply design.

[intabits]

I have a project in mind that could use a processor controlled variable high voltage power supply.
Playing around in LTSPice, I've come up with the circuit below that works very well in simulation.

I'd be interested in any comments or suggestions you might have on the design, particularly relating to making it work as a real circuit, and any issues that need to be considered.

It's the classic simple emitter follower voltage regulator, but instead of a zener voltage reference, it's governed by a PWM signal after smoothing through a 2nd order passive low pass filter. Because it operates at voltages higher than logic levels, the PWM signal is level shifted by Q1 and Q2 before going into the filter.

My project won't need high currents, but I've used a Darlington pass element to reduce the currents in the filter, and level translator.
Also, making Q2 double as a constant current source greatly helped with response linearity, settling time, and reduced power dissipation.

The output is very linear with respect to the PWM duty cycle, bending only a little at highest output power.
It appears to work reasonably well with loads from 50 Ohm to 10K.

The 2nd order filter increases the settling time, (not an issue with my project), but it does eliminate a small amount of ripple that gets past the first stage. (less of an issue, with Q2 being a CCS)

The circuit works for a large range of VSource voltages, from 10 to 120V. Most components are not critical over this supply range.
Only R3 and R5 need to change, as indicated in the .param directives. Some adjustment of R5a is needed to trim the output offset voltage.
This adjustment also allows compensation for the VBE of the pass element and other component tolerances.

The controller could monitor the output voltage via an ADC, and then tweak the duty cycle if higher accuracy is required.

At higher VSource voltages, power dissipation in R3 & R5 might become an issue. For a 100v source voltage they dissipate a few hundred mW. The simple answer is bigger resistors.

The transistors types used are just what comes with LTSpice, and there's still plenty of room to experiment with component values, and/or adjust them for actual circuit conditions. 





LTSpice schematic of simulated supply

LTSpice file (I'm in a hurry to post this, as I have to go out. But haven't worked out how to include the .ASC file. Any/Best methods?)



Output Voltage Response to Input PWM Duty cycle.
0-100% in 10% steps (100v source, 50 Ohm load)



Response linearity (100v source, 50 Ohm load)



Load regulation for 20, 50, 100, 200, and 500 Ohm loads,
over the 100% duty cycle range in 25% steps. (100v source)
Showing regulation loss at high load (20 Ohm).



Output Voltage Response to Input PWM Duty cycle
0-100% in 10% steps (10v source, 50 Ohm load)



Response linearity (10v source, 50 Ohm load)
Showing degraded linearity at low source voltages.

[T3sl4co1l]

I think you will find much better results if you simply filter the PWMDAC, then use a conventional op-amp, level shifter and follower configuration.  That way, the PWM is derived from a precision reference voltage*, the op-amp is simply tracking a reference value, and feedback regulates the output without any attention required from the PWM source.  You have absolutely no problems with high voltage switching, switching speed adding errors, and supply voltage varying the whole thing**.

*This can be done by buffering the PWM signal through a buffer supplied from a precision (say 2.5, 4.096, 5.0, etc.) supply, or by using a SPDT (or pair of SPST) analog switches wired between VREF and GND.  This addresses the error when PWM is supplied from an MCU with dirty 3.3V/5V/whatever supply.

**Unless of course you want the whole thing to be proportional to input, I don't know.  Usually "power supply" requirements are the opposite of that; but there are still some applications where that's expected.  "DC transformers" for example.

Also give some attention to current limiting, and probably power limiting as well.  A typical method is simply a series sense resistor, with a transistor wired across it such that, if the sense voltage exceeds the transistor's turn-on voltage (Vbe), the transistor pulls down on the output transistor's base/gate drive, pulling it somewhat off.

Power is simply the voltage drop multiplied by the current flow.  If you limit current to 1A say, then at the maximum 120V drop (into a short circuit load), you need to dissipate 120W.  That's a big transistor.  That's also beyond the SOA rating of many transistors -- beware, the full power rating may not be available at high voltages, always read the datasheet and make sure it has the specs needed for a linear application.

You can save on power dissipation, under heavy load or fault conditions, using a foldback current limiter, or an electronic fuse (output turned on and off altogether) function.  Foldback is easy, it's a modest tweak to the current limit circuit.  Fuse behavior is more complicated, and at least needs a few transistors or ICs to implement so I won't go into detail at this time, just for information as a typical option.

Tim

[mag_therm]

DC Safe Operating Area SOA and junction temperature Tj  are going to be a problem for a 50 ohm load with 50 to 100 volt input.

It might be feasible with say a pair of To 263 Silicon Carbide BJT which can go toward Tj = 175 C.

But I've never tried those in a Darlington with a Silicon driver .

[intabits]

I think you will find much better results if you simply filter the PWMDAC, then use a conventional op-amp, level shifter and follower configuration. 

Thanks for your reply. You're right, of course.
I was just enamoured with the simplicity of my little circuit that used no op-amps, and seemed to work so well.

I'll start experimenting along the lines you suggest.

As to SOA, that was just exploring the limits. My actual application will probably be less than 50V, and less than 500mA.
Though more pass elements could be added to the Darlington if required.
And short circuit protection is a real world issue that must be added.

Also, accuracy is not very important, within 250mV is fine, as long as the controller can measure the actual value. But if accuracy can be easily improved, then it should be...

[David Hess]

I completely agree with T3sl4co1l about using local feedback to control the output voltage.

Instead of a high voltage level shifter with gain, I might consider a bootstrapped operational amplifier:

https://www.edn.com/bootstrapping-your-op-amp-yields-wide-voltage-swings/

But either are very feasible.  If high bandwidth is required, then a discrete transconductance based design is more suitable; these were used for analog oscilloscope z-axis amplifiers to get 50+ volts swing in 20 nanoseconds or less.

[intabits]

Thanks for the intersting link. (lots of KW instead of KOhm typos!?)

...to get 50+ volts swing in 20 nanoseconds or less.

Speed is no issue. In my application, a 50V swing in 20 seconds would almost be acceptable.

[T3sl4co1l]

Thanks for your reply. You're right, of course.
I was just enamoured with the simplicity of my little circuit that used no op-amps, and seemed to work so well.

Of course if you want to use discrete transistors, that's fine too -- for the stated accuracy, and with the voltage levels in this circuit being convenient, it doesn't take many to pull off.  Bonus, add another transistor for better filtering the PWM, get even lower ripple out, or something.  (Which is also a good idea for using a dual op-amp, actually.)

(The filtering is practically free: you already need at least one RC filter, preferably several RCs; just wrap them around the op-amp instead, and there you are, a sharper active filter.  Doing this with just transistors, takes a few extra resistors for bias (so, there is that bother), and performs almost as well as the op-amp.)

Thanks for the intersting link. (lots of KW instead of KOhm typos!?)

Heh, check the article date -- 1999, the original probably used font tags to make W.  Unicode support was poor back then.  On the upside, omega is Greek's 'W' so it still makes [trans]literal sense.

Speed is no issue. In my application, a 50V swing in 20 seconds would almost be acceptable.

That's actually a bit interesting in itself... presumably you need it on for much longer than that, and that puts some limits on certain aspects of the circuit/system.  For example, a minute at 500mA is 30 coulombs, it would take quite a capacitor to deliver that much charge -- if continuous power weren't available.  (Which I'm guessing is irrelevant, as you'll have a mains powered transformer or something like that handy.)  Heatsinks... might not have to be fully sized (for continuous duty), but you definitely can't get away with naked devices over that time scale (which can handle rated power for, oh, maybe fractional seconds).  Either way, nothing of consequence, it's basically going to be your average power supply...

Tim

[intabits]

I didn't mention any details of my application for this circuit because I thought it might be of more general interest. However, as you've pointed out, it has limitations that make it unsuitable for general use.

My application for this is an automated electrolytic capacitor reformer and tester. In this scenario, it would be more of a "variable voltage source", rather than a general purpose power supply. It would be called upon to deliver only milliamps during reforming, and maybe a few hundred milliamps for quite short periods to do a fast charge when required. These currents would be appropriately limited by resistors, and the whole application would be mindful of keeping things with the SOAs and other limits. The controller providing the PWM would also be monitoring various voltages via a 4 channel ADC.

So thinking about it some more, rather than investigate other options, I'm inclined to proceed with my little circuit in this application.

Addition of Q5 and R8 provides basic current limiting.     






Why an "automated electrolytic capacitor reformer and tester"?
Well, I have a *lot* of electrolytics to process.
This is most, but not all of them.




 

[poorchava]

I don't really see the point of this. Get a microcontroller, build a half bridge or full bridge, wind a trafo, add some diodes and lc filter. Works from 1V to about 500V output, from single watts to kW range. Where SMPS gets hard is if you need one of the parameters to be especially good, such as step response,ripple or accuracy. Otherwise even a simple feedback with a P regulator works ok.

This is even simpler if input voltage is low. Up to like 50W the entire primary side can literally be 1 chip and 2 capacitors.

[intabits]

(Sorry, I thought this thread had run its course, so I haven't checked in for a while.)

I don't really see the point of this. Get a microcontroller, build a half bridge or full bridge, wind a trafo, add some diodes and lc filter.

Your method is in an area where my experience is limited, which is probably why it didn't occur to me.
But it suits my requirements very well, and I have wanted to tinker with that sort of thing in a general way for some time.
My current project gives me an incentive to experiment with doing things that way. Thanks.