Help with PSU design - 2Kw unisolated Buck

[Vectrix150V]

Let me start off with saying this may be a possible entry to the hall of shame...

OK - I am trying to design a non-synchonous unisolated buck converter around a TL494, for the purposes of an open-source battery charger.
The goal is 160V DC at 12A, with constant current limiting (12A) and then constant voltage (160V). There is a battery management system input that is basically an opto across the soft start section to shut the supply down when any of the (44) cells go over their target voltage. The load is predictable and does not change fast (at all). The battery pack is isolated from the user and the chassis, so no nagging about safety please.

I'm using parts from the original charger (it was from a Vectrix VX1 that exploded and was beyond economical repair) - I'm using the output filter inductor of 270uH, wound on a very large E core with copper foil. It doesn't appear to be gapped, and is glued. I used a few other bits and pieces like the 105 degree low esr caps and the whole (sand cast) charger case. There is a 280uH PFC inductor as well I could use (same sort of construction - copper foil)

My design basically uses rectified 240V AC into 5 X 450V 330uf resovour caps (which are pre-charged using a large NTC which is bypassed with a relay after 3 seconds) and a TL494 as a single ended 100KHz PWM, driving an IRS2125 high-side driver, running a 25A 1200V IGBT with the voltage being sensed via a resistive divider, and current through a 0.01 Ohm shunt on the lowside. This is filtered through a 1K resistor and 0.1u cap. You would think this was a simple robust design. Errr... no.

The original design sort of worked. And I mean sort of - it was noisy (ie. squealed like a stuck pig) and was not very stable at all - adjusting the voltage down from anywhere would result in noise (hissing or squeal from the inductor) and the trace on the scope going everywhere. I put this down to poor PCB design. (Bad earth planning) It ate 1200V 25A IGBT's with abandon. They just shorted across all three pins.

The IRS2125 went straight into protection mode and refused to output anything until pin 3/4 were bridged - it wasn't surrent limiting but seems to be from spikes on the supply rail (was happy being run from a 12V battery) Even with a 15V zener, 0.1u MLCC ceramic and 100U directly across its rails it still refuses to work properly, so I left 3+4 bridged. Ugh. I don't like this hack. This is not related to the problem at hand though.

I since redesigned the PCB to include a ground fill plane, and removed a few superfluous bits (I originally buffered the voltage divider and amplified the shunt output X10) - no good. The real improvement came when I stuck a 0.1uf cap between pin 2 and 3 - the voltage control worked nicely, with no jitter on the oscilloscope and squeal from the inductor and adjusted between 0 and about 80% duty cycle (my duty cycle is a bit lower due to the use of the soft start circuitry on the DTC pin). The current control required a much larger (10uf) cap between pin 15 and pin 3. The IRS2125 is still being a PITA (if 3 + 4 are bridged temporarily it starts)

I know that this is an ugly hack but I am not sure how to calculate the compensation components - or if I have them even in the right place. There is an application note from TI where they use resistors to set the gain to 101 - not sure what this achieves.

So now I have a supply that is stable on an oscilloscope when fed with 120V DC. I had a 12 ohm load across it and it was perfectly happy. I could disconnect and reconnect it, and only hear a slight 'ping' from the inductor. But on mains voltage, there is a very noticible 'clack' from the inductor when the load is applied - this wasn't a problem with a 40W bulb as the load, but when I stepped it up to 150W, the IGBT died shorted (again!).

I'm not entirely sure what is killing them. I have a 15V Transzorb from G to S, the output cap is largeish (330uf) - I'm wondering if the inrush current is high enough to pop the IGBT (and the current limit loop is really, really slow to respond).

Can someone suggest what to do here? I have attached some schematics and also oscilloscope shots of the gate drive scoped referenced to ground - the 340V has a ground loop (since I don't have an isolated scope, the ground is referenced to ground, rather than 1/2 rectified mains potential).

Should I consider snubbers or Transzorbs across CE? What value snubbers?

I am considering moving to a more robust IGBT - An FGL60N100BNTD - 60A 1000V, or even an IRG80N60RUFD - 80A 600V (although 600V is closer to the bus voltage of 340V DC).

Let me know what to scope next, I have an old analog 10Mhz CRO, so catching anything fast at startup isn't likely.

Any help is most appreciated! I'm happy to provide the schematics in express PCB .sch format if this helps anyone to tear them apart as well.

Some context on the oscilloscope pictures - the really, really bad one is prior to the compensation components being fitted - the one with some jitter is at 340V but with an earth loop. These are taken at the gate of the IGBT - 5V/Div with a 10x probe (so 50v/div). There is a 15V transzorb between G and E.

[japlytic]

Re: Help with PSU design - 2Kw unisolated Buck

Welcome, Vectrix150V!
An appropriately rated high voltage differential probe should be used on your oscilloscope for testing this design.

[ben7]

Re: Help with PSU design - 2Kw unisolated Buck

That extra diode after the inductor is not needed

[Vectrix150V]

Re: Help with PSU design - 2Kw unisolated Buck

Yep, I know, this is actually for a battery charger, and the battery is permanently connected to the charger - I don't want the battery to discharge through the voltage feedback network, or the load resistor.

I solved my issues with the current limiting - put a load resistor across the output of the supply (prior to the second diode) - 3.3K worked fine at 150V (and I had a 50W metal case resistor that was that size). The supply regulates fine, and immediatly reacts to a change in load - you can switch a 1000W load on and off and not blow something up.

The largest problem I was having was that the IGBT's I purchased from Ebay appeared to be counterfeit. They were dying at 300W or so, unacceptable!

I ran out of IGBT's and ended up using an IRFP460 in its place (with a 400V transorb from Drain to source in leau of a snubber).

Managed to drag 1.5Kw out of it before the Mosfet shorted (it was unstable, and beating at a subharmonic, and I would assume that this exceeded the SOAR of the Mosfet).

Now, I need to stabilise the feedback loop for the current loop (voltage loop is fine) and try either two IRF460's in parallel or a larger IGBT...

[ben7]

Re: Help with PSU design - 2Kw unisolated Buck

Yep, I know, this is actually for a battery charger, and the battery is permanently connected to the charger - I don't want the battery to discharge through the voltage feedback network, or the load resistor.

I solved my issues with the current limiting - put a load resistor across the output of the supply (prior to the second diode) - 3.3K worked fine at 150V (and I had a 50W metal case resistor that was that size). The supply regulates fine, and immediatly reacts to a change in load - you can switch a 1000W load on and off and not blow something up.

The largest problem I was having was that the IGBT's I purchased from Ebay appeared to be counterfeit. They were dying at 300W or so, unacceptable!

I ran out of IGBT's and ended up using an IRFP460 in its place (with a 400V transorb from Drain to source in leau of a snubber).

Managed to drag 1.5Kw out of it before the Mosfet shorted (it was unstable, and beating at a subharmonic, and I would assume that this exceeded the SOAR of the Mosfet).

Now, I need to stabilise the feedback loop for the current loop (voltage loop is fine) and try either two IRF460's in parallel or a larger IGBT...

You could try a higher power MOSFET, like the SIHG47N60S-E3 from vishay.

[kaboom]

Re: Help with PSU design - 2Kw unisolated Buck

I solved my issues with the current limiting - put a load resistor across the output of the supply (prior to the second diode) - 3.3K worked fine at 150V (and I had a 50W metal case resistor that was that size). The supply regulates fine, and immediatly reacts to a change in load - you can switch a 1000W load on and off and not blow something up.

Nice! Now the current from the inductor always has someplace to go, load or no load.

What about your 12A limit? Compensation? IE- 12A at the voltage of a discharged battery bank vs 12A at "nominal" voltage?

Get a differential probe! You're lucky you've got an unbonded ground rod or HRG, otherwise, you would've burned up your scope probe ground. Possibly the scope, too. You'd have shorted the lower half of the line rectifier.

[Vectrix150V]

Re: Help with PSU design - 2Kw unisolated Buck

Compensation - like I said, I have the voltage loop nicely compensated (don't get into positive feedback or strange subharmonic oscillations) but the current loop isn't so good - you can hear some noise from the inductor. Last time I cheated and kept adding larger caps between pin 15 and pin 3 - before I put the 10K rsistors in to keep the input impedence stable, I think I had 10uf across it (!!)

I think I really need to do some more reading up on how the compensation works - I'm only using a basic (type 1) compensation network.

I'm actually cheating with the scope - I connect the probe tip to the gate or drain, and the scope ground to mains ground (NOT the negative rail) - I would pop the ELCB/RCD if I didn't, since it would be connecting half rectified mains potential to ground. (In Australia we run a MEN system - multiple earth neutral - where the neutral is bonded to earth - single phase, not split phase as per USA).

And yes, I should get a differential probe

The current limit is set up as a foldback - so as the load increases, the voltage is dropped back so the load always sees 12A across it - a discharged lithium battery bank is capable of drawing 100A+ out of a charger if it doesn't have this type of limiting.

This is a CC/CV charger - it limits the current to 12A until the voltage reaches 160V (Constant current stage) and keeps the voltage at 160V (constant voltage) as the current tapers off when the cells are fully charged (minus self discharge and the 7mA draw of the onboard vehicle DCC converter).

I might try a heavier MOSFET as suggested too.

[kaboom]

Re: Help with PSU design - 2Kw unisolated Buck

Compensation - like I said, I have the voltage loop nicely compensated (don't get into positive feedback or strange subharmonic oscillations) but the current loop isn't so good - you can hear some noise from the inductor. Last time I cheated and kept adding larger caps between pin 15 and pin 3 - before I put the 10K rsistors in to keep the input impedence stable, I think I had 10uf across it (!!)

Yes, that would certainly count as cheating!

I think I really need to do some more reading up on how the compensation works - I'm only using a basic (type 1) compensation network.

I'm actually cheating with the scope - I connect the probe tip to the gate or drain, and the scope ground to mains ground (NOT the negative rail) - I would pop the ELCB/RCD if I didn't, since it would be connecting half rectified mains potential to ground. (In Australia we run a MEN system - multiple earth neutral - where the neutral is bonded to earth - single phase, not split phase as per USA).

Well, it's the high impedance of the probe that's "protecting" you then.

Is there a single transformer, probably 11kV 1p->240V 1p feeding your house? Or are you on a larger "secondary level" distribution system? The latter consists of a large (150-333kVA, or more) 416Y/240V transformer feeding several blocks. Each house gets one of the phases and the neutral. That's similar to the way hotels and large apartment complexes are fed in the US, but on a larger scale. And 416Y/240, instead of 208Y/120.
I have attached a diagram of a typical 120/240V 1p 3w bank.

BTW, "split phase" is a holdover from many years ago. It actually describes the typical induction motor, having a centrifugal start switch (also sometimes a start cap) capable of running on single-phase power.

http://forums.mikeholt.com/showthread.php?t=79044
http://forums.mikeholt.com/showthread.php?t=124915

And yes, I should get a differential probe

The current limit is set up as a foldback - so as the load increases, the voltage is dropped back so the load always sees 12A across it - a discharged lithium battery bank is capable of drawing 100A+ out of a charger if it doesn't have this type of limiting.

Which can ruin your whole week.

This is a CC/CV charger - it limits the current to 12A until the voltage reaches 160V (Constant current stage) and keeps the voltage at 160V (constant voltage) as the current tapers off when the cells are fully charged (minus self discharge and the 7mA draw of the onboard vehicle DCC converter).

I might try a heavier MOSFET as suggested too.

Which means it has to be stable at 12A/160V and 12A/~115V. [(3/4.2V)*160V]

I like the idea of parallel MOSFETs. Two of the same type will only dissipate 1/4 the heat as one, with the same current.

[Vectrix150V]

Re: Help with PSU design - 2Kw unisolated Buck

The power is actually three phase 415V to the house, and a neutral - it somes in from underground (on 20mm2 cables!) Its a newish estate, so this is not unusual - there is a large 11Kv transformer down the street. Its rare for new houses not to have 3 phase in from the street.

All three phases are used for the ducted airconditioner, and then it is split for different circuits. The neutral is bonded to earth at the switchboard (everyone's switchboard).

I'm having problems modelling the compensation - I'm basing my testing on a resistive load, but this is not the same as a battery - at all (its a very large capacitor with a low ESR).

Should I consider removing the 1K resistor and 0.1u capacitor from the current feedback? I'm wondering if the filter here is causing a phase shift that pushes it into positive feedback.

I also considered another idea of using an MC33341 and IGBT driver IC as per an interesting application note where it is used as a switchmode current controller (its in its datasheet) - might work to drive the buck converter rather than the TL494?

I'm reluctant to do this though, I seem to be pretty close with the design I already have (and don't feel like troubleshooting another design).

[Th3_uN1Qu3]

Re: Help with PSU design - 2Kw unisolated Buck

You asked me for help via PM - i get little spare time nowadays, i find myself always doing jobs here and there for some extra $$. A high power battery charger isn't trivial to design, and i feel like you went wrong by choosing the TL494 here. Current mode would be the way to go.

I feel like most of your issues would be vastly simplified if you went for a 3842. Voltage mode control and high current = lots of blown parts guaranteed. With current mode you can use a high value source resistor and limit the possibility of damage while ironing out the quirks.

[Vectrix150V]

Re: Help with PSU design - 2Kw unisolated Buck

Thanks for your reply.

Current mode control with a buck converter is an interesting idea.

A 3842 could be used with a low-side MOSFET/IGBT with a current sensor in the current path, and the voltage loop could be done with a TL431/Optocoupler combination, that drops pin 1 to ground when the target voltage is hit. Needs some more thought, but not a bad solution. Saves messing with instrumentation amps etc for differential voltage measurement.

The more I think about this - the better it seems in terms of parts cost and niceties like pulse by pulse current limiting (might stop my Mosfets from self-destructing quite so often). TBH I have never used current-mode control before, it will be interesting.

More recent update would be to use the UCC3802 vs 3842 due to better current sensing and a few other niceties.

[kaboom]

Re: Help with PSU design - 2Kw unisolated Buck

A 3842 could be used with a low-side MOSFET/IGBT with a current sensor in the current path, and the voltage loop could be done with a TL431/Optocoupler combination, that drops pin 1 to ground when the target voltage is hit.

I also would've suggested this. It would seem Andy (Unique) beat be to it!

Needs some more thought, but not a bad solution. Saves messing with instrumentation amps etc for differential voltage measurement.

Which is a nice little bonus.

The more I think about this - the better it seems in terms of parts cost and niceties like pulse by pulse current limiting (might stop my Mosfets from self-destructing quite so often). TBH I have never used current-mode control before, it will be interesting.

More recent update would be to use the UCC3802 vs 3842 due to better current sensing and a few other niceties.

That's the best argument so far in favor of current-mode control, especially when the battery characteristics are considered.

I think it's interesting to note that all the cheap PC power supplies, as bad as they can be, appear less prone to blowing up, as current-mode control has largely replaced voltage-mode. Those Leadmans may put out horrible ripple, but even as far back as the LP-6100s, they were current mode with a 384x. Shorts from wire trimmings, and other faults, notwithstanding, of course.

Good luck!

[Vectrix150V]

Re: Help with PSU design - 2Kw unisolated Buck

I have had a bit of a think and a look at some datasheets, and have come up with the attached design.

Looking at using a UCC3804 (rather than a 02) - This is because it has a 50% maximum duty cycle (don't need any higher, 340/2 = 190V, only need 150-160V) and the current loop won't need slope compensation as a result.

It also has blanking on the current sense input (though I left a 1K resistor and 0.1u cap for filtering, they can be left off if not needed), and a pull down on the gate drive output, and a 15V internal zener, so another two less parts needed.

I think this should work OK, at least I should stop exploding mosfets...

Can I get people to have a look and scrutinize this for stupid ideas?

I think the UCC3804 has half the frequency output with the same timing components selected (I have 95Khz according to my calculations) due to the internal flip/flop.

[ben7]

Re: Help with PSU design - 2Kw unisolated Buck

No glaring problems that I can see...

Oh, why the 270K resistor, when you have an external 15v PSU?

[kaboom]

Re: Help with PSU design - 2Kw unisolated Buck

No glaring problems that I can see...

Oh, why the 270K resistor, when you have an external 15v PSU?

I'd remove that, and just connect the small bias supply directly to pin 7 of the 3804.

Looking at using a UCC3804 (rather than a 02) - This is because it [b]has a 50% maximum duty cycle[b] (don't need any higher, 340/2 = 190V, only need 150-160V) and the current loop won't need slope compensation as a result.

Sounds good. Less work, and less "accident prone." You won't have to worry about the MOSFET(s) while adjusting the slope comp.

It also has blanking on the current sense input (though I left a 1K resistor and 0.1u cap for filtering, they can be left off if not needed), and a pull down on the gate drive output, and a 15V internal zener, so another two less parts needed.

Try anything from a .01-.033u cap, if you can. Maybe as high as a .056. Don't block the Isense for too long...

Put the pull-down resistor directly across the MOSFET G and S, if you're doing point-to-point wiring. An 18V zener from gate to source is also a good idea. Cathode of said zener goes to the gate.

I think this should work OK, at least I should stop exploding mosfets...

Can I get people to have a look and scrutinize this for stupid ideas?

I think the UCC3804 has half the frequency output with the same timing components selected (I have 95Khz according to my calculations) due to the internal flip/flop.

If you're unsure, you can always put a resistor between the cap bank and the buck converter. Or even a light bulb.

[Vectrix150V]

Re: Help with PSU design - 2Kw unisolated Buck

Thanks for the replies, this inspires a bit more confidence.

The UCC3804 actually has a maximum of 12V when driven from a low impedence source voltage, so I think my 15V bias supply is probably a bit too high - The bias supply was a 2.5W integrated smps transformer (yeah cheating) but I think I might use a capacitor/bridge/zener combo to make 12V to run the soft-start relay (after a 3 second delay) and provide VCC to the chip. I'll remove that resistor, the bias supply should be adequate.

I'll be building this onto a PCB with a ground plane fill design, but will add those parts anyway.

Now trying to find a low inductance, high wattage resistor for the current sensing.

The UC3842 has an interesting section on slope compensation when exceeding 50% duty cycle - it uses a transistor on the RC pin to resistively sum the PWM ramp pulses into the Isense pin.

I don't expect to exceed 50% duty cycle (I think? Since the voltage is related to duty cycle in discontinuous mode anyway).

[Vectrix150V]

Re: Help with PSU design - 2Kw unisolated Buck

Well, I have the current loop nicely compensated, but the voltage loop is a mess for some reason.

That is, I can add a load and the circuit works perfectly with changing input voltage to maintain a current across it (including >50% duty cycle, which is a problem with curent mode) - but at light loads, I can't seem to get it stable. I need it to be stable from 100mA to 12A, but it only seems to be happy at higher loads. (CCM?)

Rather frustrating!

[Th3_uN1Qu3]

Re: Help with PSU design - 2Kw unisolated Buck

Low load regulation may be a mess for some reason, but maybe not for the reason you think it is.

I can't be bothered making the calculations - please mention the frequency you are switching at, and the values of all compensation components including the RC on Isense. I'll try and help once you provide this data.

[ben7]

Re: Help with PSU design - 2Kw unisolated Buck

I can't be bothered making the calculations - please mention the frequency you are switching at, and the values of all compensation components including the RC on Isense. I'll try and help once you provide this data.

I agree.

Like, at how much output power does it become stable? Etc...

[Vectrix150V]

Re: Help with PSU design - 2Kw unisolated Buck

Thanks for the replies - I will take some more measurements but I think that the problem is due to the optocoupler not being in the linear region - in the thread 'UC3842 basics' it was discovered (on page 5) that a low load instability problem was due to an insufficient load on the phototransistor, and insufficient current through the LED.

The current loop is stable now (I am feeding back some of the oscillator ramp through an emitter follower into the current sense input) - before this the frequency halved once it hit 50% duty cycle - now no issue (buck topology so I 0-100% duty if possible). I also added a shottky diode from pin 3 to pin 6 (this clamps negative going spikes into the current sense input which confuses the UC3842) - from an application note named 'solving the problem of pulse-skipping oscillations' which I was encountering as well (this trick is in some datasheets of the UC8342 and not others - sigh).

I am using a 4N25 (avg CTR 100), so a 1.2K resistor gives 4mA load, so will aim for 4mA through the optocoupler LED as well. The 150K resistor I was using would have resulted in a miniscule load on the optocoupler.

I am looking at an application note () from Micrel where this is done as well.

I am running the circuit from a bank of series 12V SLA's which means I can adjust the input voltage in steps of 12V - the current loop behaves itself, but the voltage loop instability is due to load, not input voltage, and is not affected by changing compensation components.

The compensation is very simple - it consists of a 0.1 cap across the terminals of the TL431, and a 150K resistor from pin 1 to pin 2 I'll redraw the schematic (it has changed a lot) and post it up. Looking at the micrel app note the resistor between 1 and 2 is way too big.

I'll get back with some more results soon.