SPI ATX 350w refurbish

[pentium]

Normally I'd look for another power supply but I'm extremely short sources on where to find another PSU right now so I have to make a compromise.

AT form factor high wattage power supplies are hard to come by these days so I'm grabbing a cheap low wattage ATX power supply and making that work.
I pulled an SPI 350W rated (probably 250w) ATX supply out of an untested Pentium D machine. Now that I have it on he bench I can tell that the lone 16v 3300uf cap is bulged and the remaining 10v 3300uf and 10v 2200uf are TEAPO, so it's safe to say they'll need to go as well. First issue is all I got are Panasonic 16v 2200uf caps. I know normally it's not a good idea to use lower value caps but this thing is going into a less demanding system (300 peak max) so will it be fine?

Second is that with PWR_ON grounded and a hard drive plugged in as a load the thing sqeals like a madman. The +5 and +12 are nice and stable and +5 has 4mv AC ripple but +12 has 128mv. That doesn't seem right. Should I assume the squeal is attributed to the extremely high ripple, not enough of a load or simply it being a crappy Sparkle?

[kaboom]

Re: SPI ATX 350w refurbish

get the teapos outta there!


the panasonics will be fine for testing- don't run it to full load
don't forget those teapos, being rather questionable, may have only been 2700u, especially if they're rather compact

load it with lamps or resistors to the approx demand of the system it will eventually power, then scope it and examine the ripple

if you're attempting to measure ripple with a DMM, those won't read high freq AC too well- if that's the case, ripple will usually be higher than indicated

the squealing isn't ripple per se; it's the feedback loop oscillating

replace all caps including small ones (you already know this) and recap the aux supplies

don't forget the two 2.2-4.7u/50v caps in the base ckts of the two switching transistors, if it's a half bridge.

[pentium]

Re: SPI ATX 350w refurbish

Yep. After I replaced the caps I DID have in stock I loaded down the rails and it powered up with no squeal and already a lot less ripple. This is indeed a candidate to complete the rebuild.

[kaboom]

Re: SPI ATX 350w refurbish

1. Get the caps off the load resistors.

2. Return the low side common from the load to the common (black) wire(s) from the output of the supply, not to the chassis of the unit under test.

3. Connect scope probe gnd to case of PSU and probe tip to an unused wire from the same rail (under test) you want to measure:

You want to measure ripple on +5. Connect load between one of your red wires and any black wire; connect probe tip to one of those red wires which does not have the load attached. Again, connect scope gnd to case. This puts scope gnd/common very close to the secondary common of the supply under test, and also puts the probe tip very close to the rail whose ripple you want to measure; you use one of the unloaded red wires (in the case of +5) as a "jumper" to go directly to the output of the supply.

You also never connect the scope common to the return wire of the load at the load; here you have common potential rise, leading to common mode currents up the scope probe shield. That current goes thru the scope case, up the third wire of the scope, down the third wire of the PSU's line cord, and back to the secondary low side. This is an unwanted parallel path for testing; remember, this is secondary current (with ripple) going up your scope cable. Any AC component will couple into the center conductor. Depending on phase, it can give false indications of greater or lesser than actual ripple.

The scope common must go to the lowest point on the secondary- either the case or an unused/unloaded "black wire" with a computer SMPS.

In order to truly measure ripple, you cannot have any resistive elements (like loaded wires to resistor banks) before the point where you pick the sig off with scope.

See example:



M2= ripple at supply
M1= ripple after being shunted by filter consisting of test lead/clip resistance (R1/R2), load resistance (R3), and the "lying capacitor" (C1) across the load
M3= DC return potential rise- results in current up the scope probe cable shield
M4= AC return potential rise- also results in current up the probe cable shield, but being AC, couples into the signal wire! A recipe for misleading/false "safe" ripple indications.

All wire has resistance, and so has voltage drop while carrying current! This is true both at the load end of a "supply" wire, as well as at the load end of a "return" wire! See M3 & M4 again- the low side of the load is not at GND potential, but actually higher!

[pentium]

Re: SPI ATX 350w refurbish

1. Get the caps off the load resistors.

Those are the Panasonic 2200's. You literally just told me doing this for testing was fine.

2. Return the low side common from the load to the common (black) wire(s) from the output of the supply, not to the chassis of the unit under test.

You can't see it but there's two yellow wires. One shorts PWR_ON to the chassis and the other is +12.

3. Connect scope probe gnd to case of PSU and probe tip to an unused wire from the same rail (under test) you want to measure:

I already did the ripple check again and things have improved even with the wrong value caps.

I just wanted to follow-up.

[kaboom]

Re: SPI ATX 350w refurbish

Those are the Panasonic 2200's. You literally just told me doing this for testing was fine.

No, that's not what I said at all.

One- I never specifically said to put them across the load resistors- the diagram shows what will happen.

Two- "try these caps for testing" means to solder them into the unit in question, even if temporary to evaulate how it responds, as has been done here and elsewhere.


I don't know how you made that leap; it doesn't even make sense to do that, especially while leaving the old ones in place. Should've pulled the board, pulled the 3300/16v teapo, and whichever teapo on +5 before the pi choke, and replaced/subbed both with one of those panasonics.

[PeteS in CA]

Re: SPI ATX 350w refurbish

Sparkle isn't crap. MAYBE not as good as Delta or Lite On, but ...

Anyway, being rated for 350W it might be 10 years old or older. Even the best electrolytics have a finite life. Probably some of the O/P caps were high impedance, resulting in oscillation and high O/P ripple. That's why installing good caps fixed the squealing. I'd still get some good, proper value, caps for the rebuild (maybe Nichicon PW or PA series, or similar?). If the P/S is >/= 10 years old, the fan may be worn or worn out. Depending on the currents you need, uprating (as needed) to rectifiers rated for at least 30% higher than the needed current would be a good idea.