SMPS for tube pedals

[R.G.]

Something which did seem odd.  Suppose we fry the diode due to overheating at 150degC.  The thermal resistance of the diode was something like 45 degC/W.  So we need 3.3W of diode loss.    I'm not seeing that type of number in the calculations.

That's one of the biggies, all right. The whole circuit doesn't seem to be moving enough power to make the diodes overheat, especially tie to wide, flat PCB areas. I've toasted my fair share of fast - and slow! - diodes, but this situation seems to not carry enough sheer power in bad conditions to overheat the diodes. Maybe. But it sure seems not to.

When that happens, I get out my textbooks, rules of thumb, pyramid amulets and slide rules and start looking at numbers.

[Rob Strand]

That's one of the biggies, all right. The whole circuit doesn't seem to be moving enough power to make the diodes overheat, especially tie to wide, flat PCB areas. I've toasted my fair share of fast - and slow! - diodes, but this situation seems to not carry enough sheer power in bad conditions to overheat the diodes. Maybe. But it sure seems not to.

I can imagine some extra reverse voltage due to the transformer output going negative during the on cycle and some ring from the transformer leakage over-voltaging the diode.   That would just zap the diode.   But for the diode to get progressively hotter as the output voltage and load is dialed up that has to be something else.   Very low output currents so forward loss unlikely too.

[printer2]

Back again. Going to try putting lipstick on a pig. I spent hours looking for a suitable replacement heatsink and finally gave up and searched through my junk. Cut apart a cpu heatsink and managed to mount it in a fraction of the time. Ordering replacement diodes as I do not trust the ones blowing. For good measure thinking of stuffing the distance between the transformer and the heatsink with compound to wick off heat from the transformer.

[Rob Strand]

For good measure thinking of stuffing the distance between the transformer and the heatsink with compound to wick off heat from the transformer.

It's not worth the mess.   Ferrite is a poor conductor of heat, and so is heatsink compound.  So in the end it's not going to achieve much cooling the entirity of the transformer.  Besides transformers can take a bit more than you might expect.  A better idea would be to put the edges with the heatsink and transformer in a more open air in the enclosure, away from any corners.

[printer2]

For good measure thinking of stuffing the distance between the transformer and the heatsink with compound to wick off heat from the transformer.

It's not worth the mess.   Ferrite is a poor conductor of heat, and so is heatsink compound.  So in the end it's not going to achieve much cooling the entirity of the transformer.  Besides transformers can take a bit more than you might expect.  A better idea would be to put the edges with the heatsink and transformer in a more open air in the enclosure, away from any corners.

It does not take much arm-winging to get me not to smear that stuff, hate it. I was not thinking of the heat bothering the transformer but rather to reduce any heat that gets to the board and therefore the diode. Did not think it would be a magic bullet but every quarter ounce eventually adds up to a pound (used to build and crash RC gliders). I am hoping to have the module on the bottom of a Tweed styled chassis with a heat shield between it and the tubes. Also being on the outside I want to use some perforated metal to shield against radiated EMI. 

[R.G.]

I did some quick (and very dirty) estimation on the switching diode losses for the circuit in question. The US3M looks like it ought to work, mostly, which might account for the stories of it working until the voltage gets high. The UF4007 surprised me in that it has a shorter trr than the US3M, 75nS vs 85nS; I had expected the purpose-packaged US3M to be better, but both are pretty good.
Both have similar thermal characteristics. The US3M has a thermal resistance of 26C/W to its terminal when mounted on 0.53" X 0.73" pads of 2-ounce copper. This reflects the fact that most of the heat is conducted out the terminals into the PCB copper, not through the body. The UF4007 has a similar situation, in that it specifies thermal resistance of 30C/W to its leads. It too relies on the PCB copper for heat sinking. The UF4007 doesn't specify a PCB pad or area size, only a max power dissipation of 2W.
Some of the very dirty stuff is in estimating (not calculating or simulating) the reverse recovery losses. If the diode is conducting current into a cap full of 300Vdc, then tries to stop conducting, it takes trr seconds to stop conducting. During that time, it has to conduct the reverse current and (to a first approximation) the DC voltage from the output capacitor. This is a big assumption based on thinking that the transformer capacitances and reflected voltage conditions can pull the cathode nearly to ground(ish) to get the capacitor voltage across the diode.
Another dirty assumption was that the reverse recovery current was a square pulse of 1A. The datasheets show a curve for reverse recovery current peaking at 1A, and trr being at recovery to 1/4 of that. Rather than do the integration graphically or with math in my head, I reasoned that it has to be better than 1A for trr. So each turn off event could be about 300V * 1A = 300W. This power only lasts for trr, so the energy is 300W*85nS, or 25.5uJ. Doing that at a switching frequency of 100kHz gives 2.55W. This goes down with switching frequency, of course. If it's 50kHz, the dissipation is only 1.28W, etc.
Two-ish watts in a diode with 30C/W thermal resistance would make the junction about 60C over ambient, and so yes, it gets hot. The same reasoning applies to both diodes. That makes me feel better about my intuition that there wasn't enough power being converted to heat the diodes. There is, it's just that there's quite a bit going into the diodes.
I then wondered - why does the UF4007 survive? Probably because there are two of them and they are close-enough matched to share dissipation. Or that the UF4007 semiconductor process and packaging is somehow better inside. The PCB area isn't particularly great compared to the datasheet notes on the US3M, so maybe that's an issue.
The high voltage output is an issue if my thinking isn't way off. The diode reverse losses would go up with output voltage. They would also go up linearly with switching frequency.  Yeah - if my thumbnailing isn't too far off, the diode could get into heat stress and not live long because of thermal effects as the output voltage goes up. Turning the output voltage down lets it cool off. There might be an option to lower the switching frequency if the inductor can take the longer on-time, or if the inductor can be subbed for one with enough energy storage at the lower frequency.

[printer2]

I did some quick (and very dirty) estimation on the switching diode losses for the circuit in question. The US3M looks like it ought to work, mostly, which might account for the stories of it working until the voltage gets high. The UF4007 surprised me in that it has a shorter trr than the US3M, 75nS vs 85nS; I had expected the purpose-packaged US3M to be better, but both are pretty good.
Both have similar thermal characteristics. The US3M has a thermal resistance of 26C/W to its terminal when mounted on 0.53" X 0.73" pads of 2-ounce copper. This reflects the fact that most of the heat is conducted out the terminals into the PCB copper, not through the body. The UF4007 has a similar situation, in that it specifies thermal resistance of 30C/W to its leads. It too relies on the PCB copper for heat sinking. The UF4007 doesn't specify a PCB pad or area size, only a max power dissipation of 2W.
Some of the very dirty stuff is in estimating (not calculating or simulating) the reverse recovery losses. If the diode is conducting current into a cap full of 300Vdc, then tries to stop conducting, it takes trr seconds to stop conducting. During that time, it has to conduct the reverse current and (to a first approximation) the DC voltage from the output capacitor. This is a big assumption based on thinking that the transformer capacitances and reflected voltage conditions can pull the cathode nearly to ground(ish) to get the capacitor voltage across the diode.
Another dirty assumption was that the reverse recovery current was a square pulse of 1A. The datasheets show a curve for reverse recovery current peaking at 1A, and trr being at recovery to 1/4 of that. Rather than do the integration graphically or with math in my head, I reasoned that it has to be better than 1A for trr. So each turn off event could be about 300V * 1A = 300W. This power only lasts for trr, so the energy is 300W*85nS, or 25.5uJ. Doing that at a switching frequency of 100kHz gives 2.55W. This goes down with switching frequency, of course. If it's 50kHz, the dissipation is only 1.28W, etc.
Two-ish watts in a diode with 30C/W thermal resistance would make the junction about 60C over ambient, and so yes, it gets hot. The same reasoning applies to both diodes. That makes me feel better about my intuition that there wasn't enough power being converted to heat the diodes. There is, it's just that there's quite a bit going into the diodes.
I then wondered - why does the UF4007 survive? Probably because there are two of them and they are close-enough matched to share dissipation. Or that the UF4007 semiconductor process and packaging is somehow better inside. The PCB area isn't particularly great compared to the datasheet notes on the US3M, so maybe that's an issue.
The high voltage output is an issue if my thinking isn't way off. The diode reverse losses would go up with output voltage. They would also go up linearly with switching frequency.  Yeah - if my thumbnailing isn't too far off, the diode could get into heat stress and not live long because of thermal effects as the output voltage goes up. Turning the output voltage down lets it cool off. There might be an option to lower the switching frequency if the inductor can take the longer on-time, or if the inductor can be subbed for one with enough energy storage at the lower frequency.

So you are saying, "It depends?"

Just kidding. Your back of a napkin analysis is at the level I can absorb easily. Now I have a a few more data-points to use with Digikey's selection.

Thank you.

[printer2]

"Oh, you were looking for Elegance? Keep walking and it will be three doors to the left. This is Just Get The Damn Thing Done"

Only got it up to 19W with the power supply I have at the moment and the 350V caps I got in the circuit. The brick does not like when the module is set for 350V on startup and shuts down and restarts. Turning down the set voltage and then turning it up it does not get upset. Set it to 350V and 50mA through a 6V6 in SE.

The heatsinks are pretty cool. With ambient being 24 C I have 37 C on the foil (0.2 mm, 0.1 mm that I folded on itself and run some solder between them) and 33 C on the big old heatsink after an hour and a half. OK that was fun, how about if I put the original heatsink back? This time 19.7W out I got 35.2 C on the foil, 38 C on the heatsink. Seems the foil really does help. I will have to run more power through it, maybe even with a higher supply voltage.

 

[printer2]

Swapped the 6V6 for a 6N3C Russian tube, on 310V I had 80 mA, 25W out of the module and the original heatsink heated up to 55 C. I did not get to measure the foil as the diode popped. By the time I got the thermistor on the foil it was also measuring 55 C so I am sure it got hotter.

The module with a 6V6 in SE (359V 55 mA) seems like it would be fine with my little makeshift heatsink. I need to look at replacement parts for the diode with sucking the heat away in mind.

[PRR]

6V6 in SE (359V 55 mA)

That's a lot of heat for a 6V6. 19.7W, maybe 17.7W if you allow for bias and screen. Rated 12-14W. Also 315V. Yes, there is a report of Fender pushing modern Champs to that power zone.

Your Russian 6L6-alikes may be fine.

[printer2]

6V6 in SE (359V 55 mA)

That's a lot of heat for a 6V6. 19.7W, maybe 17.7W if you allow for bias and screen. Rated 12-14W. Also 315V. Yes, there is a report of Fender pushing modern Champs to that power zone.

Your Russian 6L6-alikes may be fine.

Yeah I know, but I did not have a better way of dissipating the power. Nothing glowed red so I was not too worried. I have more tubes than I will ever need so not a big concern. About 25 NOS 12V6's in the upper left, a hand full of used also beside them.

[Rob Strand]

Some of the very dirty stuff is in estimating (not calculating or simulating) the reverse recovery losses. If the diode is conducting current into a cap full of 300Vdc, then tries to stop conducting, it takes trr seconds to stop conducting. During that time, it has to conduct the reverse current and (to a first approximation) the DC voltage from the output capacitor. This is a big assumption based on thinking that the transformer capacitances and reflected voltage conditions can pull the cathode nearly to ground(ish) to get the capacitor voltage across the diode.
Another dirty assumption was that the reverse recovery current was a square pulse of 1A. The datasheets show a curve for reverse recovery current peaking at 1A, and trr being at recovery to 1/4 of that. Rather than do the integration graphically or with math in my head, I reasoned that it has to be better than 1A for trr. So each turn off event could be about 300V * 1A = 300W. This power only lasts for trr, so the energy is 300W*85nS, or 25.5uJ. Doing that at a switching frequency of 100kHz gives 2.55W. This goes down with switching frequency, of course. If it's 50kHz, the dissipation is only 1.28W, etc.

I'm not convinced about the 1A estimate agreeing with what is happening in the circuit.  The forward currents are quite low in this circuit.  I was planning to get back to the cause myself but coming up some realistic estimates is actually a fair amount of effort.   (I'm sure that's why you used the 1A value - I get it, we something to work with!!)

There's also a technicality in that the power loss is some factor less than the V*I product.  These factors creep in when V and/or I changes with time; factors like 1/2, 1/4, 1/6.   All fair enough but it does push the power dissipation estimate down.

The Trr values in the datasheets are only valid for the test conditions in the datasheet.  So when the current and voltages in the circuit are different to the test circuit you get different recovery times.

This video has a good demo of the effect of the circuit on the actual currents and recovery times.   
You can pretty much dial up any recovery time you want.
#201: Basics of Reverse Recovery Time in a Diode

6:00  Demo Start
7:40  Increase forward bias/Forward current ==> recovery time to increase.
8:00  Increase reverse bias ==> decrease recovery time

The flat-bottomed recovery waveforms are a result current limiting due to the signal source resistance.

The schematic seems to have a bug.  I think the 100nF timing cap is incorrect.  IIRC the predicted switch time for 100nF is very low, less than 10kHz.   Your 50kHz to 100kHz values make more sense, that's the type of numbers I was using last week.   It's not possible to resolve the switch time unless someone measures the switching waveforms or the cap.

[Rob Strand]

I tried to put some basic numbers to the design.  However, we can only go far on specifics since we don't know the switch frequency and we don't know the winding ratio or inductance of the transformer.

Nonetheless, I've assigned some values which allow 20W output at 200V.  I chose an on-time of 10us roughly ball-parking a 50kHz to 100kHz switch frequency range.   With the values shown, the duty cycle and discontinuous conduction off time are feasible numbers.

Bottom line is:
- it's possible the forward diode current is the cause of the problem.
- that would mean the transformer winding ratio isn't quite right.

For the simulation I manually change the secondary inductance and the switching period to maintain 200V output.

The results are in the text.




As for the video posted earlier showing the diode frying.   Perhaps the simultaneous increase in output voltage and load current (due to fixed resistive load)  just pushes the *forward* diode current too far.

[R.G.]

I'm not convinced about the 1A estimate agreeing with what is happening in the circuit.  The forward currents are quite low in this circuit.  I was planning to get back to the cause myself but coming up some realistic estimates is actually a fair amount of effort.   (I'm sure that's why you used the 1A value - I get it, we something to work with!!)

I'm not convinced either. It's a really dirty estimate.  :-) 

I could rationalize it by the filter cap pouring charge back into the junction for a few nanoseconds while something mysterious was happening in the transformer or primary. Anyway, the recovery current is probably not dependent on the forward current, only that the junction won't actually support any reverse voltage until the charges in the junction are swept out by reverse current flow. The filter cap could do that for 1A, maybe; depends on what's sucking on the anode side of the diode keeping the anode more negative than the couple of hundred volts on the filter cap.

About all I can say for myself is that it does seem to come out with a result that matches the symptoms, kind of.

Edit; hit send too soon.
Yes, a forward current of an amp or two in a diode with a 1.8v max forward drop for roughly the same 85nS gets to about 3W as well. Could be a transformer issue. If I had transformer specs I could do a much less rough estimate for that. Maybe it's both forward and reverse at the same time.

About now, if I were trying to get this to work, I'd whip in a massively bigger horse. Mouser has the Vishay VS-E5TX1512S2L-M3 for $2.12 each, and it might be coaxable into the pads on the PCB, although it's not a real fit. It's rated for 15A and only 29ns trr. It would make for interesting theater.

[Rob Strand]

I'm not convinced either. It's a really dirty estimate.  :-) 

I could rationalize it by the filter cap pouring charge back into the junction for a few nanoseconds while something mysterious was happening in the transformer or primary. Anyway, the recovery current is probably not dependent on the forward current, only that the junction won't actually support any reverse voltage until the charges in the junction are swept out by reverse current flow. The filter cap could do that for 1A, maybe; depends on what's sucking on the anode side of the diode keeping the anode more negative than the couple of hundred volts on the filter cap.

About all I can say for myself is that it does seem to come out with a result that matches the symptoms, kind of.

Edit; hit send too soon.
Yes, a forward current of an amp or two in a diode with a 1.8v max forward drop for roughly the same 85nS gets to about 3W as well. Could be a transformer issue. If I had transformer specs I could do a much less rough estimate for that. Maybe it's both forward and reverse at the same time.

The symptoms looked more like reverse issues to me too.

Well I had a closer look at the details and 1A reverse current isn't unreasonable.  For 300V out and 40W output load: With the 500uH secondary I could get a peak reverse current of 1.3A with 1.1W dissipation due to the reverse current.   With a 200uH secondary both those figure are approximately doubled.  The dissipation due to forward currents is quite low, like 250mW or so.  Given we know nothing of the transformer ratio that's about as far as I'm willing to go.  To get to the bottom of it someone needs to pin down the details of the transformer and the switch frequency.
FWIW, I  set-up the transformer with a lower inductance so the Toff time was longer.  Using half the previous values, ie. Lp=8.5uH and Ls=250uH, the reverse power in the diode dropped considerably.   It highlights the need for details of the actual unit.

[printer2]

Finally got back to it. Frequency 72.5 kHz, 300V 79 mA. Flattened out a couple of pennies (make sure they are copper kiddies) and did some fancy soldering as well as painted them and the original heatsink flat black, still bare copper in the picture.





I did not get a lot of time on it but the heatsinks felt a little warm to the touch.

[PRR]

pennies (make sure they are copper kiddies)

Where do you get those?? US Cent has been 2.5% Copper since 1982, our VietNam era. Bad drives good out of circulation so you won't find two of the 95% Copper cents today (unless you pay as a collector). I think Canada has quit cents. UK Penny is steel with electroplate copper.

Copper roof flashing is a thing and I have seen small squares sold individually for chimney step-flashing, and have used that material as fins on LM377 chips.

[printer2]

pennies (make sure they are copper kiddies)

Where do you get those?? US Cent has been 2.5% Copper since 1982, our VietNam era. Bad drives good out of circulation so you won't find two of the 95% Copper cents today (unless you pay as a collector). I think Canada has quit cents. UK Penny is steel with electroplate copper.

Copper roof flashing is a thing and I have seen small squares sold individually for chimney step-flashing, and have used that material as fins on LM377 chips.

After my copper foil experiment (scraps from shielded a MRI suite) I thought I would just buy a small piece of sheet, the smallest they had was a one foot square piece, clicked on the tab, OMG $97. Forget that. Yes we got rid of the penny up here a number of years ago, had a coffee can of pennies from back in the day. We went over from copper to plated steel in 1998 (think some might have been zinc alloy or something, a magnet would not pick them up). Pre-1982 were thicker, I separated about 50 of them when I thought that would be more than enough. I flattened them to get a little more area but afterward though just putting a flat spot on the penny and soldering it in place would do. I was concerned about the transformer pins but I would just trim the area of the penny close to to them and call it a day.

I need to do a little more testing to see how much I can pull out of the modules without too much effort. Also if they will upset the 2.7A laptop supply as they do the 2A one I was using that gets overloaded on startup. Maybe they might be worth using if you have a couple of cents to throw at them (sorry, I could not help that).

[PRR]

Changing/debasing the coinage is no fun, and I see Canada went through both Zinc and Iron (and a dash of Nickel) on the way. I see that throwing pennies in a jar is a long tradition, so they may not be that hard to find. (But check your copper spot-price first!)

You won't want to hear that decades ago I finished much of a kitchen in copper sheet. Even that may have been a lucky buy: I remember trimming with "coppertone" which was dyed Aluminum. (They just retired the MRI trailer here with a much bigger unit; should I have been down there the night before "helping"?)

EDIT: bad timing.....  https://www.youtube.com/shorts/Lm6_xOXndfk

[Rob Strand]

Finally got back to it. Frequency 72.5 kHz, 300V 79 mA. Flattened out a couple of pennies (make sure they are copper kiddies) and did some fancy soldering as well as painted them and the original heatsink flat black, still bare copper in the picture.

I did not get a lot of time on it but the heatsinks felt a little warm to the touch.

Very cool.

I'll try to match something with your numbers.   I'm assuming oscilloscope waveform is the output ripple at the first cap?   (As it sure looks like the diode current passing through the output cap ESR.)

It would be good to get the PWM output from the chip.  It's not possible to extract the MOSFET on-time from the diode current waveform.  The diode current waveform is when the switch is off but it is shorter than the total off time.   So we don't know the on-time or the total off-time.

Almost forgot I found this,
https://dalmura.com.au/static/YH11068A.pdf

It quotes 75kHz operating frequency, which is in agreement with you measurement.    I need to look at the datasheet for that chip.  I was assuming it's constant off time and variable off time, which is how the MC34063 works, but it could be different.


Nonetheless I might be able to narrow the times based on the 8V to 32V input and 40V to 390V output.   Your  72.5 kHz, 300V 79 mA (and 12V in?) would then pin things down in the middle of range.