Power supply topology - will it work? (Control theory, stability)

[salbayeng]

Ah found it,  the OP says 150mA,  (for some reason I was thinking it was in 5-15mA range???)  OK an upper bound might be 1W or 3v at 150mA = 20ohms.
Probably don't need the "surge" feature as 150mA will charge the cap fast enough.??

One could probably achieve rough local current limiting by using a small NMOS (2n7002) to drag the big MOS gate down.  It's threshold is ~ 1.8v , so with a 10ohm source resistor, current limiting would occur around 180mA, (and you could put in a high impedance RC network if desired for "surge" protection),
For more accuracy, substitute a TL431, it will pull down the big MOS gate at 2.499v across the source resistor.  Although if one were using an opamp for CC mode, the local current limit would be superflous,  design effort might be better spent on thermal protection.

I can see how the boost cap for the PMOS would be useful now, with 150mA load current, a capacitance of 10uF might be required for adequate filtering so the PMOS may need to handle 10-20mA during reference changes to constrain the slew rate.

[T3sl4co1l]

Note that the source resistor contributes to loop stability, because it sets the maximum transconductance of the pass transistor.

Tim

[Yansi]

Thank you guys for more detailed analysis of the circuit. I have been few days quite overcome by my other work, so sorry for a small drop in communication. Hopefully I will draw the full schematic soon, so we can discuss a bit more in detail about the whole circuit.

Yes, the preregulator will have a filter attached on the output, you will see from the full schematic. (however there are still some unknown, what capacitances will be used there).

You think those 50V reserve accross D-S is too much? Well, cannot say much about that. Just wanted to be sure there is enough headroom for the voltage reg and also that the negative undershoot of the preregulator can not cause problems (currently 6V undershoot - but I think no easy solution for it, as the negative slope of voltage is given by the discharge rate of prereg output cap and windup occurs in the integrator in the control loop - please see my last post)

The overall goall is to make a decent (but not overcomplicated!) regulated PSU of 0-400V 0-300mA. As the current circuit seem very much viable, should not be much problem to scale it up for that goal. Bigger mosfets are on the way, so I might have them on friday.
I have chosen easily available and quite cheap IPW90R1K0C3 from Infineon, will be two of those devices in paralell (to increase SOA withstand capability of the linear stage against direct shorts or any wild overloads, where the mosfet will see full voltage accross D-S and full current for a moment). The source resistors should be left the same value, 22ohm per mosfet (11ohms together, so same drop at full 300mA output as we had so far with 150mA). But I might drop the value down if necessary.

By the way, what is better for the control loop? To have bigger or smaller source resistance? With higher resistance the mosfet should behave more like a current source device, more linear than with very small source resistance. Also should be better for a "SNR" side of thing (offset voltages will hide better if the signal value is higher, do not want to sense milivolts there). There is only half a watt being lost (0.15 squared times 22), 1W together both shunts at full output current. I don't thing it is a big deal, for a 120W output supply, do you?

I will use some constant current load on the output, to (maybe? over-)compensate the current leaking through the D-S resistor divider (feedback for the prereg).

Note that the 800K feedback resistor is actualy 330+470 K in series. Shall withstand 400 volts together easily.

Yansi

[T3sl4co1l]

More source resistance is better.  The maximum value is [V(opamp-supply) - Vgs(th)] / Iout(max), so in principle you could use a high voltage amp (Vsupply as high as Vds(dropout), say?) and make it stupidly linear.

But really, more than 10V gate to common isn't important.

Source resistor also affords you short-circuit, reverse and surge protection.  Put a zener from common to gate, and the transistor will limit current on its own, no matter how suddenly the output gets pulled around (say from connecting up a fully-charged electrolytic backwards, or several accidentally stacked somehow).

Putting active current and voltage limits on top of that makes things all the sweeter.

(You actually want the protection network a bit more complicated than a single zener, since reverse surge would do nasty things to the simple connection above.  A diode or zener at G-S, and a series resistor, in the common/ground return path I'm thinking, would be best.)

Tim

[Kleinstein]

A relatively large source resistor is needed to make the current sharing work well. The 22 Ohms are well large enough. The gate (especially gate to drain) capacitance might be a considerable limit to common mode rejection, e.g. ripple of the switched mode stage. So one might consider the earlier mentioned cascode circuit might be worth a thought with higher power (two quite large MOSFETs) - to a certain degree this an alternative to a more power full gate drive.

The drop on the linear output stage does not need to be much larger than the undershoot / drop of the pre-regulator. So something like 10 V should be enough. Overshoot of the pre-regulator might be relevant for worst case power loss for the FETs.

I am not that convinced the choice of MOSFET is good - this are modern fast switching types, that tend to have a not so well suited SAO. Though the high voltage helps. The more natural choice would be old types.

[Yansi]

Look carefuly: The mosfet that I have chosen is DC rated. I don't know a single one "older" type of high voltage mosfet that has a DC SOA. Ant the SOA limit is only power limit in reality (no second breakdown as BJT ones have) - which is what one would expect from a mosfet that is rated for DC linear operation.

[Kleinstein]

The problem with modern MOSFETs is that they do have an effect similar to second brake-down, it just got a different name. So if the curve only shows the a simple power limit, there is a good reason not to fully trust that curve.  The old first generation MOSFETs usually did not have this extra limit, but modern types usually have. However I also do not know a suitable type.

It is well possible that FET could work. Under normal conditions it will not see more than 50 V for more than about a few ms (if at all). Also many of the MOSFETs can survive quite some power, but one never knows. Especially many modern FETs are not made and tested for linear operation, but for fast switching and low R_on from a small die.

[Yansi]

If the SOA states only the power limitation, there should not be any other limitation. If there is, the manufacturer is lying and it would be a big trouble for them then.

Also not much mosfets are with DC SOA today, so if you find one that is, I think one could rely on that.

[T3sl4co1l]

SuperJunction FETs apparently are "free of" 2nd breakdown.  Which is remarkable, as they have the highest current density among MOSFETs, too.  I haven't seen any articles discussing this aspect of the technology.

As far as I can tell, there was only ever one reason why MOSFETs claimed to avoid 2nd breakdown: they sucked.  Until recent advances (Trench, VDMOS), the current density was always significantly lower than that of BJTs.  The gain is lower, too.  The power levels required to cause 2nd breakdown would've been perhaps 10 times that for a contemporary BJT, or even more -- nothing that could ever possibly be achieved, given the limitations of the package.

There's no such thing as "free from 2nd breakdown", only that it be inaccessible within package dissipation limits.

Which, by the way: I don't think I've ever seen a SOT-23 MOSFET with 2nd breakdown limits.  There's just not enough heat flow through the package to cause a gradient in the die!

IGBTs have even higher power densities, with the sensitivity (gain and tempco) of BJTs.  Though, remarkably, I have seen a few datasheets (not many) which show power-limited DC SOA.  I find this suspicious, and wouldn't recommend using them that way without extensive qualification testing.

I have no problem with using MOSFETs where the SOA says they're okay.  Qualification is always good, of course.

Related: I tested an IRF740 to have RthJC something like 60% of the datasheet (maximum) rating!  Those old chips have one thing in common: huge dies.  In comparison, a brand new Fairchild FDPF6N60 (I think) expired damned close to its ratings, like 5% over.  Tiny die!

Tim

[Yansi]

Oh yea, that's why I fuckin hate those new "150A" mosfets in DFN (and similar) SMD packages, where the claimed cannot be ever reached in real applications.

IRF3205 and IRFP4110 is still the top of quality products I have ever used. (And those old IRF540 and similar are still the good ones). Huge dies!

BTW, the courier was almost unimaginably fast, they delivered me the parts today (although I expected tomorrow at best). So now I have the 90R1K0 Infineon mosfets prepared for some more weekend toying. Also bought some STF3NK80Z for the prereg, as those STP6N80K5 I have used so far in the prototype (both in the prereg and linreg part :-) ) to be too good for the prereg, will try the cheaper ones. Should do good enough too. (Also I like the fully plastic body more for this application)


As far as the prereg voltage setpoint over the drain-source of the linreg: Yes, I might lower the voltage, I admit those 50V might be more than what is really needed. (it is an easy mod, also I plan to leave a trimpot for this in the final design).

Y.

[salbayeng]

For the pre-regulator, I would suggest a slightly different approach

(a) make a fast, low gain inner loop that controls the voltage output of the pre-regulator so it tracks a reference voltage, the pre-reg output would then have smallish capacitors, and probably a fairly high ripple.  The loop gain would be fairly low , e.g. the pre-reg output might droop 10v under full load, (so GH~40) , and there would be no integrators to wind up.

(b) form a reference voltage that is  equal to output voltage + desired_MOSFET_drop , so you are making an outer loop that controls the MOSFET_drop

While it might be argued that (a) and (b) together are algebraically the same as what you already have, but in practice you can employ different filtering to the signal path in (b)
(This approach also avoids the voltage divider leaking current across the MOSFET)
Because your buck stage is back to front, in practice the feedback for (a) is a difference of Bus_plus - Vout_minus, and the reference voltage in (b) is generated from another difference amp Vout_plus - Vout_minus. You won't need an opto-coupler either, and you can feed in your desired_MOSFET_drop using the 5vref on the UC3842. The downside is you may now be leaking current across the pre-reg MOSFET.
An alternative is to use a current mirror as a level shifter to replace the Diff amp

---------
An alternative control for the pre-reg might be hysteretic control , this is always stable (or always unstable, depending on your viewpoint!)

-----
Something else to possibly throw in series with one of the supply lines is a largish bobbin type inductor with a diode across it, but depending on where you hook your power supply caps, you probably wont need this (and your supply is current source essentially). This helps avoid extreme levels of short circuit current when discharging capacitors. Basically the inductor sucks up the capacitor voltage for the first millisecond or so, and discharges the capacitor at hundreds of amps instead of thousands, the internal resistance of the inductor also limits the short circuit current, and the large slug of copper sucks up a lot of joules. The diode is there to recirculate the stored current when the short circuit is removed. If there is a semiconductor involved , having the inductor there pushes the locus down in the SOA curves to a safe region, and gives control loops time to respond.

[Yansi]

Hello guys!
After some long time, I have finaly found some time for more hobby electronics toying. This is the result:
A fully working CV/CC programmable power supplies 0-300V 0-100mA.

A dual channel 12bit  DAC with a 12bit ADC (for output current measurement) have been added together with an isolated DC/DC and digital SPI isolation. The module requires just the unregulated input of 300+ V and 5V for the controls. The floating section is powered from the 5V through the isolated DC/DC.

These are based exactly on what you have helped me a lot with in the above comments. These two will be used in a later project.  The larger power supply with the switchmode pre-regulator we have discussed (400V 300mA) is still under development (waiting till I have more time to design some custom PCBs for it).

Many thanks! Will post more when I get more time to finish this supply.

Y.

[marcopolo]

Any news?

[Yansi]

Hello, unfortunately not many news - what exactly do you have on mind?

The programmable modules are still sitting on the shelf, waiting for the control board to be assembled. (I have some troubles on the control board to be solved there too - I need a DAC output within the range of  +5 to -55V).

The original idea of 400V 300mA power supply with switchmode pre-regulator is currently at rest, due to other high priority work and access to a proper 3kW 500V PSU from ElektroAutomatik. I could make some pretty interesting series of episodes of the design failures in those two EA power supplies I have repaired.

Also, I have helped my friend to design a linear power supply using this topology, I will post an image when it'll be finished.

[marcopolo]

Hello,

Something like the final schematic 

[Neomys Sapiens]

Hello, unfortunately not many news - what exactly do you have on mind?

I could make some pretty interesting series of episodes of the design failures in those two EA power supplies I have repaired.

OOOh, yes please! Black cloud of shame over Viersen!

[Yansi]

Well... just as a sneak-peek:  Our team bought a 10kW up to 1000V 30A power supply from Electro Automatik Gmbh (EA-PS 81000), to charge our prototype EV traction accumulators with it.  The damn thing exploded like at least! four times (and at least three times was sent for warranty repair), and under the most bizarre circumstances.

It has even exploded once, sitting turned on on a table, with no load. BANG! Something blew inside.  Most of the time, the power supply had like 10-20 hours of run time before the big bang happened.

You know, these blow ups were one hell scary, if it did that, while the traction accumulator was connected to it.  Fortunately, we have had it always connected through fuses and safety disconnect relays and the damage never occurred on the secondary side of the PSU.

One time we desperately needed to power supply (but it was again kaputt gemacht), so we have requested a temporary replacement. We were given a different type (1U rack-mount 3kW 500V one, don't know the type off-hand). And guess what! Once upon a time at the place wee needed it the most, it blew like hell too, just after I have plugged it in the wall and flipped the switch. It has just shown the obligatory "Device is starting up" and then all hell broke loose inside. 

So, ended up with two exploded very expensive power supplies which one of them was not even ours, the company finally stopped communicating with us (as far as I was told so - I was not the one in charge of "communication").

We ended up voiding the warranty on both PSUs deciding to see where the issue was (the manufacturer never told us, they have just swapped the modules inside and sent it back with a paper attached to it how within specification the unit was after repair).

Later we have learned, that the temporary 3kW power supply exploded due to short circuit on the DCbus after the PFC circuit - due to them (Elektro Automatik) using a fucking damn stupid thin kapton foil as an insulation for the components on heatsinks, and that further due to cheaping out on a €3000 or €4000 power supply, to omit the obligatory Y2 caps to couple the heatsink to ground (to reduce RFI, it was rather directly sitting on the hard DCbus+ potential, increasing the potential (no pun intended) of the big bang. 

The kapton foil failed, due to probably a burr being present on a TO247 shunt resistor (stamped aluminium back side of it) in the PFC circuit that have over time pierced the kapton, or the resistor failed itself, resulting in the damage of the insulative kapton tape. Not really sure which one first, that would require further writing, which I currently do not have time to do.

The other PSU, the big 10kW guy is a completely different matter. Although still kapton tape insulation on the heatsinks,  the failure mode was completely different - probably arc-over on badly designed PCB.  Combined with the fact, we used the PSUs on mobile basis in all kinds of different environments (but never damp, mind you!) coupled with absolutely NO particle or dust filters on the fans and lack of conformal coating, was a disaster waiting to happen.

The 665V DCbus jumped across to the control circuitry and blew the shit out of it. I have once tried repairing it with my friend (after spending numerous tens of hours reverse engineering the schematics of that over-complicated thing), but it blew again in the same spot just as we were slowly bringing the PFC section back to life, which was just pita and a show stopper.

The smaller 3kW PSU was successfully repaired. Fortunately just the DCbus was shorted, no damage in any of the electronics occurred.  I have replaced the fucking kapton with kerafol. It is working flawlessly ever since (and hopefully will still be). We still posses the 3kW PSU, they never wanted it back. If they will, we will likely charge them the repair costs.

So the lesson learnt here:

1) Elektro Automatik PSUs are not really worth the money they are sold for. Although nice and shiny on the outside, the PCB designs inside are just pure hell. It is I'd say obvious, the designer haven't had much of a clue* how to do layout of switchmode power stuff, the power components are all over the place with in my opinion just extreme lengths of parasitics within the current loops and the control circuitry aint any better: Looks like someone dropped a bucket of SMDs on a PCB and let the auto-router software loose. (It was absolutely evil to trace the control circuitry). Also some shape features on the PCB were found very funny looking.

*or he wasn't let to do it better in the first place.

2) do not use kapton tape/foil as (the only) thermally conductive insulation material, between heatsink and components. Can be easily nicked and punched through. (And also I think it even doesn't have that good thermal properties, does it?)

As I have said, this would make a few very interesting episodes by itself, now I will just end with a couple of photos of the aftermath.

Inside of the 10kW PSU
Where the arcover occured (There is straight DCbus right on the pinheader)
Mess after the PSU was raided by us reverse engineering almost every single PCB of it.
Inside of the 3kW PSU
You could probably tell where the  sharp burr was

[Yansi]

Hello,

Something like the final schematic 

The original idea of building the 400V 300mA PSU is not completed yet, see above. The circuit is a huge spaghetti ball across a couple of veroboards, including the tracking pre-regulator (nothing special there either, just an inverted buck topology with 500V at the input side with an UC3843 in control of it).

The schematic partials I think are still available in this thread.  I do not see much value of posting a schematic of an unfinished circuit/product, I see more value in the thread being as a guide how to design such PSU and verify its performance.

Also, currently I am toying with a completely switchmode based 200W regulated PSU of 0-350V - so I have something to employ my brain cell with, as a medicament for my otherwise very bad mood (due to other things in my life) :-/

[001]

Hello,

Something like the final schematic 

The original idea of building the 400V 300mA PSU is not completed yet, see above. The circuit is a huge spaghetti ball across a couple of veroboards, including the tracking pre-regulator (nothing special there either, just an inverted buck topology with 500V at the input side with an UC3843 in control of it).

Thank You for Your work! It is amasing project
But can You help me with schematic  of analog variant since I`m no need any microcontrollers?

[Yansi]

Switchmode power supplies do not typically require any microcontrollers in them. (By analog I meant linear, sorry for the possible confusion).

Also attaching photo of the switchmode supply I am working on to have some stress relieved.

PS: Please fix the quotes in your answer