ATX PSU 3.3V rail too low - dead dual diode?

[HwAoRrDk]

I'm attempting to troubleshoot a PC ATX power supply that has a faulty 3.3V rail. The output is too low, about 2.7V. All other voltage rails are perfectly fine, by the way.

I found a third-party schematic that appears to match what I have (same controller chip, etc, albeit for a 250W model, not the 300W I have), to trace the circuitry of the 3.3V rail.

The dual-diode package I've circled on the schematic (D13, an SBL1040CT in my case) appears to be completely shorted between centre pin and the other two in either polarity. It measures about 5 ohms on resistance. I'm assuming the fact I'm measuring it in-circuit should not make any difference and give a false positive, right?

How would this diode being shorted still allow the PSU to give some kind of output on the 3.3V rail?

I suppose I should de-solder it from the board and double-check. Also, I guess I should put the 3.3V output on the scope to see if it's oscillating and meter is showing 2.7V as just the average.

P.S. L3 and C22 are not populated on my PSU, with L3 being linked out.

[wraper]

How would this diode being shorted still allow the PSU to give some kind of output on the 3.3V rail?

Look on the schematic. There is 4.7 OHM resistor on the output to GND and transformer winding is connected to GND too. Nothing wrong with low resistance across the diode.

[wraper]

Rather look what's with D28, D29 and R73.

[HwAoRrDk]

I 'scoped the 3.3V output. It is oscillating, between approx. 4V and 1.8V at exactly 100 Hz.

Oh, and I forgot to mention the PG signal never goes high.

[wraper]

Did you connect any load to all outputs?

[wraper]

I 'scoped the 3.3V output. It is oscillating, between approx. 4V and 1.8V at exactly 100 Hz.

Oh, and I forgot to mention the PG signal never goes high.

If 5V is ok, the D28, D29 and R73. should not allow it to drop below 3V.

[HwAoRrDk]

Look on the schematic. There is 4.7 OHM resistor on the output to GND and transformer winding is connected to GND too. Nothing wrong with low resistance across the diode.

Oh, I see. I did not notice the centre tap on the transformer. Yes, I suppose measuring either half of the dual diode it would instead be going through the transformer to ground and then back through R49.

If 5V is ok, the D28, D29 and R73. should not allow it to drop below 3V.

D28, D29 and R73 are not fitted on this PSU.

Did you connect any load to all outputs?

Not when I was 'scoping it, no. But when I was originally diagnosing it (and discovered the problem), it was behaving the same with a load - giving 2.7V output on the 3.3V rail.

[fzabkar]

I would replace C23.

[Pfriemler]

4R7 as a permanent load on the 3.3 V rail? Never.

And why 100 Hz? Are the primary capacitors ok?

[RoGeorge]

L1 and the square magnetic hysteresis symbol drawn near L1 tells that L1 is the mag-amp coil that does the post-regulation for the 3.3V.  If R49 (4.7\$\Omega\$) is OK, check the mag-amp components that drive L1 (not shown in the schematic crop attached in the OP).

In PC supplies, the 3.3V requires precise voltage regulation (has the tightest voltage error specs from all the other output voltages), therefore the 3.3V is not just a tap of the 5V or 12V.  The 3.3V must have post-regulation independent of the other output voltages, and that's why the need of a mag-amp post-regulator (by L1 in your schematic).

The mag-amp works with low DC current.  Typically the DC controll current for L1 is driven by some voltage reference and some low power transistor, so it is somewhere on the PCB in the low voltage side, not on the radiator with the power diodes.

I've added a few links about mag-amps in this other post: https://www.eevblog.com/forum/repair/is-it-worth-trying-to-repair-a-800w-pc-power-supply/msg4740566/#msg4740566  (no need to read my long reply there, click only the mag-amp links from there)

[wraper]

4R7 as a permanent load on the 3.3 V rail? Never.

Why not? Resistors provide a minimal load, so power rails with group stabilization do not go totally haywire without sufficient load present. Modern PSUs designs may avoid using them for better efficiency ratings but a while back nobody cared that much.

[wraper]

Not when I was 'scoping it, no. But when I was originally diagnosing it (and discovered the problem), it was behaving the same with a load - giving 2.7V output on the 3.3V rail.

You cannot expect ATX PSU to work properly without any load. Many outright shut down. Loading resistors by themselves usually do not provide enough load for proper operation. And even if PSU works, voltages are often off because group stabilization cannot deal with it.

[HwAoRrDk]

L1 and the square magnetic hysteresis symbol drawn near L1 tells that L1 is the mag-amp coil that does the post-regulation for the 3.3V.  If R49 (4.7\$\Omega\$) is OK, check the mag-amp components that drive L1 (not shown in the schematic crop attached in the OP).

Attached is the full schematic. I can't vouch for it being 100% correct, obviously, but every part I've compared has matched, barring minor details like exact part numbers (e.g. D13 is an SBL1040CT), some component values (e.g. C23 is 1000 uF) and unpopulated components.

I had been looking at that part of the circuit around Q7. Was attempting to check it, but got quite confused at the results until I realised a jellybean 8550 PNP transistor has a different pin-out to an 8550S. However, even after resolving pin-out confusion, I'm still getting approx. 0.18V on diode test both ways across one pair of pins (forget which, I neglected to write it down). I suspect I need to de-solder it to test properly.

R49 seems okay, and measures 4.3 ohms in-circuit. Looks like a 3W size. The colour bands on it don't make sense, though: orange/brown/black/gold. If we assume the first two are poorly-tinted and are in fact yellow/orange, that makes more sense - but black, making 43 ohms? It's definitely not grey or silver either.

By the way, all the electrolytic caps in this PSU - with the exception of the large mains-voltage caps C1 & C2 - are fairly new because I re-capped the whole thing about 1.5 years ago.

I'm thinking maybe if Q7 is bad I can bodge in a BC327 as replacement, even though the pin-out is different order. The specs seems fairly similar. BC327 are the only TO-92 PNP transistors I have at hand.

You cannot expect ATX PSU to work properly without any load. Many outright shut down. Loading resistors by themselves usually do not provide enough load for proper operation. And even if PSU works, voltages are often off because group stabilization cannot deal with it.

Ah, c'mon, this isn't the 1980s anymore. I don't recall coming across any ATX PSU in the last 20 years that didn't produce in-spec voltages with no external load.

And I already said the behaviour on the 3.3V rail is the same low 2.7V avg. output regardless of whether there is a load or not. And by load, I mean hooked up to a PC motherboard, HDD, FDD, GPU, optical drives.

[RoGeorge]

I would check Q7, that seem to be the transistor that controls the current through the magnetic amplifier L1, and thus the voltage of the 3.3V rail.  Desolder Q7 for a proper check.

Same, verify all the R and D and C around Q7.  If necessary, desolder one terminal only to properly measure their value.  If all parts are good, might be the fault of the C2002 IC.  Inside it is the voltage reference to be compared with the 3.3V (by the Q7).

[HwAoRrDk]

Okay, it appears I lied when I said all the other voltage rails were fine. They read okay on a multimeter, but on the oscilloscope some are definitely not fine.

I put a 750 mA load (some old stupidly-overpowered CPU heatsink fan from my junk bin) on the 12V rail and 'scoped all the voltage rails from the main 20-pin connector.

The 12V rail looks pretty horrible. 5V and -12V have some significant saw-tooth ripple too. I should mention that I 'scoped this PSU back when I had re-capped it and the noise level on all rails was brilliant - it was the quietest PSU I owned, beating a much newer one! So everything here (with possibly the exception of the 5V SB) is definitely outside the norm.

[HwAoRrDk]

I would check Q7, that seem to be the transistor that controls the current through the magnetic amplifier L1, and thus the voltage of the 3.3V rail.  Desolder Q7 for a proper check.

Same, verify all the R and D and C around Q7.  If necessary, desolder one terminal only to properly measure their value.  If all parts are good, might be the fault of the C2002 IC.  Inside it is the voltage reference to be compared with the 3.3V (by the Q7).

Thanks, I will do that.

If the controller IC is bad, then I guess it's the scrap heap for this PSU, as it's unlikely I'd be able to get a replacement chip. Shame, I had hoped this PSU would serve for longer, as it's the only one I have that features a -5V rail, so handy for vintage PC equipment.

[wraper]

Check large electrolytic caps on the primary side (do not forget to discharge them first). 100Hz ripple on the output can mean they are dead. Also measure recified mains voltage too.

[DavidAlfa]

Ah, c'mon, this isn't the 1980s anymore. I don't recall coming across any ATX PSU in the last 20 years that didn't produce in-spec voltages with no external load.

Neither had I seen any atx psu requiring external load for a very long time, but a week ago I repaired a dead Fujitsu PSU (Just a dry capacitor), without any load the 5V rail would rise to 6V and trigger the OVP.
Then added a 100 Ohm load, all fine!

So better provide some load when testing.

[HwAoRrDk]

I de-soldered Q7 and measured it. All checks out okay: E-B and C-B both measure exactly 0.7V; B-E, B-C, C-E, E-C all open.

Measured and inspected all the passives around Q7:

R48 = 5.6 (4.7)
R52 = 5.0 (4.7)
R54 = 100 (100)
R55 = 106 (100)
R56 = 506 (510)
R59 = 1.495k (1.5k)
D25 = 0.493V
D26 = 0.495V
C31 = not shorted

All seem okay. R48 & R52 are a little high, but that's probably my crappy meter leads. I don't have an LCR meter, so I can't measure C31, only do a basic check it's not shorted. But as I said before, it's quite new.

Which means I don't have any means of measuring the large primary caps C1 & C2 either. So I might just take a chance and replace them (should have done it before with the others, but I couldn't get any at the time ), and if that doesn't fix it, then I guess it's the controller.

[HwAoRrDk]

De-soldered C1 and C2 as well. Even if I can't measure them, it's pretty obvious now that they're toast - one had such a pool of electrolyte under it, it was sizzling as I was de-soldering!

[Pfriemler]

Check large electrolytic caps on the primary side (do not forget to discharge them first). 100Hz ripple on the output can mean they are dead. Also measure recified mains voltage too.

That's what I mentioned 6 hours earlier - and I was right, as we see now... 

4R7 as a permanent load on the 3.3 V rail? Never.

Why not? ...

700 mA rsp. 2,3 watts dumb load under all conditions? It's a waste of energy, if not a poor design. A whole PC's do not need more than 8 watts with an idle desktop. But you are right - nobody cared about efficency for a long time.

[wraper]

700 mA rsp. 2,3 watts dumb load under all conditions? It's a waste of energy, if not a poor design. A whole PC's do not need more than 8 watts with an idle desktop. But you are right - nobody cared about efficency for a long time.

If you'd opened some older ATX PSUs, it would be hard to find one without these resistors. Yes wasted power, but it's better than PSU outputting wrong voltage with no sufficient load present on all of the rails and destroying something.

[RoGeorge]

4R7 as a permanent load on the 3.3 V rail? Never.

Why not? ...

700 mA rsp. 2,3 watts dumb load under all conditions? It's a waste of energy, if not a poor design. A whole PC's do not need more than 8 watts with an idle desktop. But you are right - nobody cared about efficency for a long time.

That resistor must be there for the magnetic amplifier to work properly.  That is a design to regulate very high currents with a simple magnetic core, no switching transistors.  It's a very clever way to control big currents with minimal effort:

Magnetic Amplifier Control for Simple, Low Cost, Secondary Regulation by Bob Mammao
https://www.ti.com/lit/ml/slup129/slup129.pdf

The 3.3V rail can drive 10-20A.  If the post-stabilisation were to be made with switching semiconductors, switching 10-20A with semiconductors would probably dissipate more than those 2.3W needed by the mag-amp's 4.7 ohm resistor.

Now if the whole PC needs only 8W, that's more like a Raspberry Pi, maybe a laptop.  A wall adapter would be enough.  High end desktops or servers can easily draw 200-500W at an average load, and about 50-100W when idle.

[Pfriemler]

Lesson learned, thx to @RoGeorge and @wraper.

[HwAoRrDk]

It's fixed! It was indeed just the primary side capacitors. Fitted new ones and now the output is perfect and noise free all round - with or without external load.

Thanks all for the help.