0-70V, 0-5A Lab Power Supply Design

[H713]

I do not like to trust datasheets for this kind of thing- the intense competition in the power MOSFET market means that the numbers can be a somewhat idealistic.

Testing is easy enough to do. A big transformer, bridge rectifier and some decently sized caps, plus a power supply with a 10 turn pot to control the gate. I tested the 2SK3675 and the FQA8N90C and found them to be quite robust in this application, pushing them past 150 watts of dissipation. They don't handle HV linear operation too well, but they have seemed fairly tough in this application.

As an interesting aside, the little IRF640 actually held up pretty well- probably why BSS used them in the EPC780.

The HV PS design thread I started a while back had much discussion about the issue of MOSFET DC SOA. Discussion remained quite civilized.

[schmitt trigger]

Just a thought:
because of the risks already mentioned with failing Mosfets, would a crowbar circuit be a good idea?

Of course the crowbar's threshold could not be a fixed value, but would have to track the actual voltage output setting.

[Neomys Sapiens]

I would refrain from digital control other than providing the set values via D/A converters.

I think that those PS that use multiple MosFets in a linear mode must have had the benefit of a builder who could ask everything from a supplier, like very very tight selections. The only appliance which I have personal experience with was a custom unit which was used to weld hermetic cases for hybrids. If one of the Mosfets did break ranks even a bit, mass extinction ensued. While it is certainly feasible, you would have to do the selection yourself, which could be costly. If you want to try anyway, APT, IH (now Renesas) and IR had appnotes on the topic.

Where I see problems is the 2HE requirement. Even assuming that you use fans with tunnel-type heatsinks, the heatsink has to have some capacity for heat which means mass, and therefore size.

I'm hoping to mitigate the issues with the FETs in a few ways. For one, I'll be using 900V rated parts with a fairly high R_dson. Additionally, I will be using quite a few of them in parallel to have as much overhead as possible. The idea is that there will be enough beef in the series pass transistors that they will survive even if the tap-switching fails. They will also have some rather generous source resistors in order to help with the current sharing.

The voltage/current/power headroom was not what I had in mind. In the mentioned application, the MosFets died during parameter changes due to differences in transconductance. So you'll have to match for this and have ample beef in the driver stage to keep them in line. The worst case is when one is sticking out during a change in drive which coincides with a change in the load circuit.
Result: BANG zoing tingle crash Moo : literally!, as I saw TO3 caps leaving the base violently, when we tried to rebuild the thing with new MosFets.
Should not discourage you. The appliance I'm referring to was handling far higher currents and a load circuit which, depending of the tool condition and the cycle time, was varying apruptly.

[David Hess]

A pair of more common transformers could be used with a toggle switch to wire the secondaries for series and parallel operation.  Unfortunately though for audio applications, higher current will be needed at higher voltage so this is of limited use.  But the same limitation works well with fold-back current limiting; in an audio application, full current is not needed at low voltages so build that into the power supply to considerably relax the power dissipation requirements.

Or use multiple small transformers with their secondaries in series to provide the taps.  Two transformers with center tapped or dual secondaries yield 4 output levels.

[H713]

I expect that the 3 levels that would be available with the Antek transformer (two primaries, two secondaries) would be sufficient. Maybe not the best, but sufficient. As was mentioned, it's pretty easy to add windings as well, but we'll see if that's necessary.

Doing some back-of-the-envelope, worst-case-scenario napkin math, if the voltage rails were 90 volts (and never sagged under load), with 3 steps I the maximum power that would ever have to be dissipated would be 150 watts. That's a lot, but not unmanageable, and in practice it would never be that high because 1) the rails will be less than 90 volts, and 2) they will sag under load. Even if I only used 3 FETs per side, each FET would stay below 50 watts of dissipation. For the sake of having a safety factor, I'll probably use five or six.

[graybeard]

I see a couple of fundamental flaws in your schematic.

The emitter of Q6 is directly driving of M1 through M4.  There is no current path to keep that transistor turned on.  You need to add some sort of current path to keep Q6 in it's active region.  Perhaps a resistor from gate to source across the MOSFETs.

You have the four MOSFETs in parallel.  In SPICE where they match ideally this will work fine, but in real life the MOSFETs will not match.  You will need to include a small source degeneration resistor series with the source of each MOSFET, and you will need to do your best match the devices on a curve tracer.  You might have better luck replacing the MOS series pass devices with NPNs, each with a emitter degeneration resistor to help mismatches and prevent thermal runaway.

There may be other flaws, but those two stuck out at me.

[H713]

The simulation file is a rough topology. No doubt there will be numerous compensation capacitors that need to be added as well, and I suspect that I will need grid stoppers to stop this thing from going ballistic into capacitive loads. The need for a gate-source resistor is a good point- I had initially drawn the schematic using NPN transistors and neglected to add this when I decided to switch to FETs.

I couldn't be bothered to add the source resistors to the LTSpice simulation since its main purpose was to verify the voltage and current controls. Source resistors will be employed in order to minimize issues with current sharing. I was initially planning on using rather generous source resistors of about 1 ohm, which would then be mounted to the heatsink.

I also considered using BJTs for the series pass transistors, but the performance wasn't as good. They are still on the table, the MJL21194 being the most likely candidate. Suitable BJTs will be more costly than FETs, require significant current to drive the base (most BJTs of this power level have a relatively low hfe), and have a smaller SOA than the FETs that I have been looking at. As mentioned previously, in testing the FQA8N90C, I was unsuccessful in killing it, even with 6 amps at 45 volts (or something like that). Numerous times I dumped the entire 65 volt, 8800 uF capacitor bank into the FET under test, with a peak current of something like seven amps, and it survived. More testing still needs to be done on the FETs to ensure they can take what I want to throw at them, however, they performed better than the BJTs in the testing I have done thus far.

I'll probably test a 2SC5200 and an MJ21194 just to see what they can realistically take, but as of right now the FETs look far more interesting. Many linear audio amplifiers using paralleled HEXFETs have been built, so I expect I should be able to work out the current sharing, possibly using fairly large source resistors mounted to the main heatsink.

I'm quite busy at the moment, with numerous other projects that take priority, so testing for this power supply has progressed rather slowly.

[H713]

Welp, I should have listened. Even with huge 1 ohm source resistors, the FQA8N90C-F109 FETs won't share current well at all, in some cases one FET will be handling 3A while another will take less than an amp. I'm hesitant to go above 1 ohm for the source resistors, since anything above that will have to be at least of the 10W variety, ideally 25W. This starts to get pretty big and inefficient. 

With some reasonable (if not particularly tight) matching I believe one could get a pair of these FETs to share current reasonably well, and it could make for an excellent rugged and low-cost bench supply in the 2-3A range.

I'm planning to use NJW21194G BJTs ($3.55 in 10 qty at DigiKey) and I'll just have to live with using BJTs in a darlington configuration. I'm going to try really, really hard to avoid having to use a triple. Triple EFs are a pain and they can be a b!tch to stabilize, so hopefully I can get away without the second driver transistor.

I did some preliminary testing using a pair of 2SC3281s with an MJE15032 driver. My stock of 2SC3281s are pulls, and using a DMM on diode check to "kind of sort of" match the vbe, I picked the two worst. They still current share just fine with 0.22R emitter resistors. This isn't  really surprising, since BJT power amps rarely bother to match the output devices too closely, and they usually us 0.22R emitter resistors.

I still have to test my current limiting circuit, but the voltage regulation part works fine and seems to be reasonably easy to stabilize, mostly by borrowing techniques from power amp design including base stoppers, output inductor (which only helped a tiny bit) and an output zobel. This is in addition to the compensation cap, of course.

I'm going to design this power supply such that the transistor SOA can take the full 80V across them at 6A. I don't want them to ever fail. The tap switching will only be to minimize power dissipation. My plan is to control tap switching, fan speed and crowbar range through a microcontroller. I am aware that this can raise issues if the microcontroller crashes, so extra care will be needed to make sure that it goes into a safe fault mode rather than blowing up. Since I'm "digitally challenged", I'll be using training wheels on the microcontroller- an Atmega328 programmed using the Arduino IDE.

[Kleinstein]

Current sharing with fets is tricky. There is a way to use 1 extra bjt per fet for feedback to ensure current sharing. The bjts form a kind of differential amp from the current sharing resistors. This is used in some supplies. See here:
https://www.eevblog.com/forum/projects/why-mosfet-is-used-as-a-power-transistor-on-the-aim-tti-ql355tp/msg3137964/#msg3137964

The bjts may still be the easier choice.

[magic]

I have heard of some crazy people using FETs to drive power BJTs to increase "beta" and "fT" of the driver...
No clue how it turns out in practice. A readily apparent problem is the need for higher supplies.

[H713]

I have heard of some crazy people using FETs to drive power BJTs to increase "beta" and "fT" of the driver...
No clue how it turns out in practice. A readily apparent problem is the need for higher supplies.

Kyocera did this in one of their receivers back in the day... sounded good, but it had an annoying tendency to blow up. I considered it, but I my experience with doing this in power amps has been that stability is even more tricky than with the EF triple.

At this point, I think the BJTs will be an easier option. I suspect that I can get away without using a triple, if am willing to pass more current through Q8. With the MJE15032 I am using here, I anticipate this will be just fine.

[David Hess]

I have heard of some crazy people using FETs to drive power BJTs to increase "beta" and "fT" of the driver...
No clue how it turns out in practice. A readily apparent problem is the need for higher supplies.

Bipolar transistor beta and fT drop at high currents so that sort of makes sense although I cannot say that I have seen that configuration.  A composite transistor with power MOSFET driven by a bipolar transistor is common though to boost the transconductance of the MOSFET.

[Pawelr98]

A readily apparent problem is the need for higher supplies.

Not really a problem.



The pump charge can produce the required extra voltage without any extra winding, using existing transformer.
No special elements required, just plain 50Hz charge pump.
Zener may be added  for limiting the max voltage though.

With this extra voltage the Vdrop is at maximum the Vcesat of BJT which is ~0.6-1V or almost nothing for mosfet transistors.
During normal operation the unregulated voltage can be barely higher than the output voltage, providing good efficiency.

Tap switching is one thing to consider.
With simple comparators and maybe some logic gates it shouldn't be a problem.
SCR preregulator can be used but has to be filtered well.
There's also main capacitor switching.
Lower capacitance means lower mean voltage which in turn means lower average power dissapation.
Ripple will go up and this has to be accounted for when testing and when setting rules for the switching.

Series resistors somewhere within the design is not efficient but does a decent job at protecting the pass element.
I like to use those when testing "rough" loads which can be unpredicible.
But this I only used for HV 150-400V DC supply. 0-300ohm 25W wirewound potentiometer that goes in series with the transformer secondary.
If the load is too great then the main capacitor voltage will sag, along with OCP limiting the current.
By adjusting the resistance I can allow higher voltage at certain currents.

If one searches through some military surplus parts then finding 1ohm adjustable high power resistors is possible.
But the same can be implemented by a combination of relays/transistors and normal high power resistors.

[H713]

Here's the test setup so far. Seems to work okay, there are a few changes that I need to make in order to optimize stability and ripple rejection. I should be able to get it quite a bit better than it is now, and I suspect I can lower the value of C4 considerably, which is always nice since the current limiting is a bit more effective.

Stabilizing the current limiting circuit was not easy. I had initially thought that I could get it to behave with a small cap from the output of the op-amp to the inverting input, but all this did was lower the frequency of the saw-tooth oscillation. Ultimately the only thing that did anything useful was the series RC circuit consisting of C1 and R18. This of course made the voltage regulator part of the circuit a little squirrelly. A 470 pF from the output to the inverting input of the voltage regulator op-amp seemed to make things worse, so this may come down to careful tweaking of values. I've never had much success with simulating or calculating compensation capacitors, as it is difficult to account for parasitic properties of components. I'm sure some of the real analog geniuses (who are a whole lot smarter than me) are able to calculate / simulate values and have them work perfectly.

A need for higher supplies would not have been an issue. An easier solution to the charge pump is to simply spec a slightly higher voltage power transformer. One of the goals here is to try and minimize how much noise and ripple the regulator circuit has to reject. I don't care how efficient this thing is as long as it can still happen in a 2U enclosure (probably from a Crest CA9). Switching converters are efficient, but the noise they introduce can be difficult to shut up. I thought really hard about using a buck converter as a preregualtor, but I decided that I'd rather deal with slightly reduced efficiency by using tap switching than have to worry about that noise making its way to the output.

[magic]

Kyocera did this in one of their receivers back in the day... sounded good, but it had an annoying tendency to blow up. I considered it, but I my experience with doing this in power amps has been that stability is even more tricky than with the EF triple.

At this point, I think the BJTs will be an easier option. I suspect that I can get away without using a triple, if am willing to pass more current through Q8. With the MJE15032 I am using here, I anticipate this will be just fine.

Bipolar transistor beta and fT drop at high currents so that sort of makes sense although I cannot say that I have seen that configuration.  A composite transistor with power MOSFET driven by a bipolar transistor is common though to boost the transconductance of the MOSFET.

As usual, lots of creative analog technology is found on DIYAudio.
For example, John Curl from Parasound expressed preference for that kind of topology in the first few pages of this thread; it looks like the motivation in his case is just a general preference for FETs combined with the difficulty of protecting FET outputs from shorts. So that's another commercial product which uses such output stage and they make claims of good reliability. What kind of sorcery goes into making it work, we don't know.

[Kleinstein]

For the high power level one can hardly get around tap switching or some preregulator (e.g. SCR or switched mode).
Tap switching can be electronic with the pass-elements in series. So the transistors would only see a reduced voltage. This can be easier on the soa.

It depends on the topology if the extra drive voltage is  problem. For the usual floating regulator it is not.

[H713]

Agreed, DiyAudio is the playground for some very talented analog designers. For an audio amplifier, I wouldn't hesitate to play with such a topology. The big issue is probably less about oscillation and more an issue of thermal stability, but it's probably no worse than the EF triple in that regard.

For any sort of lab power supply below 37 volts, it's really hard to justify doing this kind of thing these days when you could use a uA723, LM317 or other IC to control it. Sure there are some challenges to work around (like the high reference voltage in the 723), but these are easier problems to work around than the challenges that come from doing it with op-amps and transistors.

I still don't love my current limiting circuit, I'm sure there's a more elegant way to do it. The traditional way (which is how the 723 does it) is great, except for that the vbe of the current limiting transistor changes with temperature. As a result, I find that it's both touchy and drifts like crazy. If anyone knows of a more elegant way to do what I'm trying to achieve here (and maybe one that's easier to stabilize), I'd love to hear about it.

[magic]

TLV431 is sort of transistor with more thermally stable (but also higher) Vbe.

[Kleinstein]

The usual topology for a lab supply (especially with more than some 25 V) is the floating regulator. Here current regulation is naturally with an OP.
The main positive aspect of the simple Vbe type current limit is the simplicity and speed - so it could still be used as an additional fast emergency limit.

The lm317 is not really helping for lab supply and the lm723 is not much more than a possibly (some versions) good reference and a poor OP with access to the compensation. Using OPs and a separate reference is not more complicated. One has to think about the compensation anyway.

[David Hess]

Here's the test setup so far. Seems to work okay, there are a few changes that I need to make in order to optimize stability and ripple rejection.

A base-emitter shunt resistor should be added across the 2N3055 output transistors.  Otherwise there is no way to remove charge turning the transistor off which implicates the closed loop stability.  Darlington transistors have this shunt resistor built in for exactly this reason.  If emitter ballast resistors are used, which should be done also for current sharing, then the shunt resistor can encompass the emitter ballast resistors also.

The alternative is to drive the base of the output transistor with a class-ab driver instead of class-a like you have.

[Neomys Sapiens]

@H731: would you possibly post the LTSpice-file for the schematic above?

[H713]

Sorry, I meant to post this earlier.

[b_force]

For just audio purposes I would make a much simpler design.
Which is a VARIAC followed by a isolation transformer with symmetrical outputs.
Just make a simple bridge rectifier followed by a MOSFET regulator (capacitance multiplier) where the voltage is set by resistors.

For repair purposes this is also a nice approach, because quite often you want to turn up the voltage very slowly.
Having a VARIAC in general is nice to have anyway, so you kill two birds with one stone (poor birds )

[Pawelr98]

The usual topology for a lab supply (especially with more than some 25 V) is the floating regulator. Here current regulation is naturally with an OP.
The main positive aspect of the simple Vbe type current limit is the simplicity and speed - so it could still be used as an additional fast emergency limit.

The lm317 is not really helping for lab supply and the lm723 is not much more than a possibly (some versions) good reference and a poor OP with access to the compensation. Using OPs and a separate reference is not more complicated. One has to think about the compensation anyway.

Traditional Vbe current limiter was, in my experiments when designing a fairly fast digital I-V curve tracer, a much more stable approach compared to differential opamp circuit.

If one can get a hand on a NPN Germanium transistor then this approach is very efficient and simple.
Not very thermally stable (as Ge parts have bigger thermal drift) but still an interesting addon.
I have used a GT404 russian NPN Ge transistor for testing my design.
Required just 50-55mV Vbe drop to activate the current limiter.
Collector was connected to COMP pin on the 723 chip.
Compared to usual 650mV drop this is pretty much nothing.
Had to limit the current to around 2A. Voltage wise (leakage currents, gain) it was boucing around in the +- 50mA region which is OK.
When enclosed in a small space with hot elements it can and most likely will have a very noticeable thermal drift but I have not tested it in my design as it was not meant for long term operation, just few seconds at most.

A need for higher supplies would not have been an issue. An easier solution to the charge pump is to simply spec a slightly higher voltage power transformer. One of the goals here is to try and minimize how much noise and ripple the regulator circuit has to reject.

Also a thing to note.
With bipolar transistors it makes more sense to sense the current at the emitter.
Base currents with a few big power transistors is not really negligible. It could be as much as 100mA (beta of 50).

Charge pump is a simple design.
Two diodes, two capacitors.
There is no switching intererence/ripple as it operates strictly on 50/60Hz voltage on the main rectifier bridge.
Components cost pennies and make the control circuitry supply "independent" of the main supply.
That is you can drop 1V at the pass transistors instead of say 3V because of control circuitry requiring higher voltage just to go through pass transistors Vbe drops.
Of course one may simply use an extra winding (with toroid transformers just add turns of any wire around the core) or something like that to provide that extra voltage for the control circuitry.

To give an example of why such design improves efficiency.
Main supply provides 72V under load.
If you use the main supply for both control circuitry and pass elements then the maximum output voltage will be 69V as there's 1V missing to overcome the pass element base-emitter junction voltage drop.
To get around this, you bump the supply voltage to 73V. At 5A it means extra 5W which has to be dissapated.

With a charge pump (or extra winding, doesn't matter) the main supply voltage is still 72V but the control circuitry has 80V available.
Now the maximum output voltage can be 71V. At 70V you only drop 2V*5A=10W instead of 15W like in previous example.

And when choosing a off-the-shelf transformer it's often not possible to say "oh, I would like an 0.5Vac extra on the main winding to get over the regulator Vdrop". So the 5W difference can evolve into 10W or more.

Also if done right then the control circuitry will get a ripple-free voltage for controlling the pass elements.
Another diode+capacitor connected to the main filter capacitor means almost no ripple as the current draw of the control circuitry is pretty much nothing compared to the main pass elements.

[H713]

@b_force, the problem is a lack of adjustable current limiting. When doing initial tests on a circuit that I don't necessarily trust, either because it is new or intermittently failing. Once I am confident that it is behaving well,  I may turn that maximum current up to something like 3A- just so I don't vaporize PCB traces if something stupid happens.

Again, the charge pump is an interesting idea but I don't think it's necessary. The heatsinks being used here are out of a Crest CA9, and they will be used with forced air cooling. I'm aware that I can improve efficiency by running the control circuitry off a higher voltage, but in this case it is not necessarily a priority.

I may play some more with the Vbe current limiter. The speed of this type of current limiting circuit meant that it worked very well in my HV power supply, which has suffered countless output shorts without issue. I may also be able to get around the thermal drift issues through physical design to make sure that the transistor used stays at a relatively constant temperature. If I can come up with a way to allow variation between about 100 mA and the full current (6A), without a stupidly twitchy control, then that will work. I may be able to get around the issue with a switch for high and low ranges.


Also, what is not shown in the schematic yet is a diode-capacitor network to feed the control circuitry, which drastically improves the ripple rejection.