EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: engrguy42 on May 28, 2020, 07:01:45 pm
-
Sorting thru some old boxes and found an old PC power supply. Was about to trash it and decided to hook up the scope and see if it was still breathing.
Attached is the AC coupled trace on the 12V output with about an amp of load on it. Wow, that's some switching noise...maybe 5v peak-to-peak.
Anyone know what the culprit might be? Not really interested in a science project right now, but if it's one of those "oh yeah, that's obviously xxxxx" I might tear it open and replace it.
Thanks.
-
One thing to know about old PC power supplies: they need a minimum load on the main output (+5 V) to work correctly. Try with 2 A.
After that, your measurements may look different.
-
One thing to know about old PC power supplies: they need a minimum load on the main output (+5 V) to work correctly. Try with 2 A.
After that, your measurements may look different.
Thanks. Yeah, tried almost 2A on the 5V and no difference.
-
How did you probe it?
-
How did you probe it?
Just a differential probe across the load resistor
-
I think this is some spikes are typical, from what I remember measuring. There is a huge amount of capacitance on the motherboard and various other filtering.
Try adding a big poly cap across measurement point and see if it changes much.
You can find some reviewers that measure the ripple under load, for example:
https://www.tomshardware.com/reviews/bitfenix-bf450g-power-supply,5614-9.html (https://www.tomshardware.com/reviews/bitfenix-bf450g-power-supply,5614-9.html)
-
I think this is some spikes are typical, from what I remember measuring. There is a huge amount of capacitance on the motherboard and various other filtering.
Try adding a big poly cap across measurement point and see if it changes much.
You can find some reviewers that measure the ripple under load, for example:
https://www.tomshardware.com/reviews/bitfenix-bf450g-power-supply,5614-9.html (https://www.tomshardware.com/reviews/bitfenix-bf450g-power-supply,5614-9.html)
Thanks, but 5 volts peak to peak on a 12v output is WAY out of line :D
Anyway, I called my little buddy "Bucky the Buck Converter" (only like $3, sitting in the drawer doing nothing) and he did a very nice job of cleaning things up.
-
That's in the MHz. As likely common mode as anything. What do you measure probing ground to ground?
Tim
-
Thanks, but 5 volts peak to peak on a 12v output is WAY out of line :D
Anyway, I called my little buddy "Bucky the Buck Converter" (only like $3, sitting in the drawer doing nothing) and he did a very nice job of cleaning things up.
Yeah you are right, grabbed a random PSU and its more like 500mV peak to peak, not 5V.
Maybe try with a normal probe instead of a differential, just make sure ground wire goes to black.
-
Hey, wait a minute...
That same trace occurs with a normal (not differential) probe plugged into the scope, and power supply turned on, but the probe itself is connected to NOTHING !!!
It's induced by ZOMBIES!!!! :scared:
-
Hey, wait a minute...
That same trace occurs with a normal (not differential) probe plugged into the scope, and power supply turned on, but the probe itself is connected to NOTHING !!!
Welcome to the wonderful world of working with power switching circuits. Huge amounts or radiated magnetic and electromagnetic fields, pulses on the power mains, it gets into everything. And, a lot of scopes don't have enough filtering on the mains input to keep it out of their sensitive internal circuits. I bet a good old Tek scope would at least show less of this noise.
Jon
-
Okay, well I didn't want to do it, but I guess I have to. Rolling out the big guns on this one.
Just ordered some ferrite cores to snap over the wires. BAM !! Now we'll see who's boss. :D
-
Hey, wait a minute...
That same trace occurs with a normal (not differential) probe plugged into the scope, and power supply turned on, but the probe itself is connected to NOTHING !!!
Welcome to the wonderful world of working with power switching circuits. Huge amounts or radiated magnetic and electromagnetic fields, pulses on the power mains, it gets into everything. And, a lot of scopes don't have enough filtering on the mains input to keep it out of their sensitive internal circuits. I bet a good old Tek scope would at least show less of this noise.
Jon
With the probe lead totally disconnected from the scope (unscrewed from the BNC) the trace is flat. But with the lead connected at the BNC but not connected to any devices, the same noise appears on the trace. Which implies to me that it's common mode noise generated by the power supply and being induced on the lead (even with the ground lead disconnected, BTW). Unless the BNC connector on the scope has some internal mechanism to zero the trace when nothing connected?
So it seems the scope goodness (ie, the "old scopes are awesome" fanboyism :D ) is kinda irrelevant isn't it, and it's more about cleaning up the emitted noise from the power supply? Which is why I'm going the ferrite core route.
EDIT: BTW, if I move the offending power supply from the bench down onto the floor (maybe 5 feet away), with the scope lead plugged into the scope but the probe connected to nothing, the noise on the trace drops down to almost nothing.
-
Another data point...
Covering the power supply in a tin foil hat (aluminum foil) cuts the peak switching noise viewed on the scope (with a lead connected to no devices) from around 1 volt peak to about 0.25 volts peak.
Yes, I actually got a roll of Reynolds Wrap and covered the power supply. :-+
So yeah, a tin foil hat is sometimes necessary. :D
Anyone know where I can buy a Faraday cage? :D
-
The loop area of your probe is acting as a nice antenna for the switching noise radiating off of the power supply and the common-mode noise currents flowing on the output cables. The common-mode noise currents cause a changing magnetic field in the vicinity of your power supply, which in turn get converted to a differential mode voltage that shows up on your oscilloscope input.
When power supply manufacturers measure ripple, usually they use a "ripple probe", that is a probe with a very small enclosed loop area loop area. The best way to do this is to use a probe tip BNC socket which allows you to plug the oscilloscope probe directly into a coaxial connector which is then wired to the power supply output. Manufacturers typically also specify a small capacitance at the end of the cable (10uF-100uF + 0.1uF ceramic). Most power supply manufacturers also use a 20MHz bandwidth limit on the oscilloscope. To further reduce EMI pickup, make sure to use a 1x probe.
With this setup, you have a more realistic idea about the power supply's actual ripple. Of course, even with the "real" ripple of the power supply, near-field radiated emissions can still cause system problems. As you saw with your accidental antenna probe, it can be quite easy to pick up noise, and that noise is real. If you have a system with lots of high sensitivity (high impedance) nodes and/or large PCB loop areas, then radiated switching noise can still be a problem for you.
-
The loop area of your probe is acting as a nice antenna for the switching noise radiating off of the power supply and the common-mode noise currents flowing on the output cables. The common-mode noise currents cause a changing magnetic field in the vicinity of your power supply, which in turn get converted to a differential mode voltage that shows up on your oscilloscope input.
When power supply manufacturers measure ripple, usually they use a "ripple probe", that is a probe with a very small enclosed loop area loop area. The best way to do this is to use a probe tip BNC socket which allows you to plug the oscilloscope probe directly into a coaxial connector which is then wired to the power supply output. Manufacturers typically also specify a small capacitance at the end of the cable (10uF-100uF + 0.1uF ceramic). Most power supply manufacturers also use a 20MHz bandwidth limit on the oscilloscope. To further reduce EMI pickup, make sure to use a 1x probe.
With this setup, you have a more realistic idea about the power supply's actual ripple. Of course, even with the "real" ripple of the power supply, near-field radiated emissions can still cause system problems. As you saw with your accidental antenna probe, it can be quite easy to pick up noise, and that noise is real. If you have a system with lots of high sensitivity (high impedance) nodes and/or large PCB loop areas, then radiated switching noise can still be a problem for you.
Thanks, but is there a fix? I'm far less concerned with accurate measurement than I am with minimizing the switching noise in the first place. I assumed some well-placed ferrite cores on the power supply wires might do the trick, since this is MHz-range switching noise. Anyone know of a fix? I assume this is fairly common in workbench situations where guys have 126 different pieces of equipment stacked on top of each other.... :D
-
I guess the point I was trying to make is: I think it's mostly your measurement. The switching noise you see on your scope is not necessarily what the powered system actually "sees". Yes, the power supply is emitting a bunch of HF junk, but it doesn't necessarily mean that it causes issues. (Rather, it doesn't mean that it actually gets picked up by the system and is converted into the large peak-to-peak value that you saw.) Also, if the power supply is from a real, reputable manufacturer then it should have underwent 3rd party EMI testing that proves compliance with international EMI requirements. These requirements set levels for acceptable levels of EMI. If you want to cut down on the radiated emissions, then you are definitely on the right path with adding ferrites and shields.
-
Even with a differential probe, it requires good technique to get meaningful results. I usually end up using direct differential coaxial connections which modern differential probes do not support, but old ones do.
Also beware that modern differential probes tend to fall out of calibration due to poor construction and of course lack service documentation. So expect poor common mode rejection unless you verify it yourself.
-
Even with a differential probe, it requires good technique to get meaningful results. I usually end up using direct differential coaxial connections which modern differential probes do not support, but old ones do.
Also beware that modern differential probes tend to fall out of calibration due to poor construction and of course lack service documentation. So expect poor common mode rejection unless you verify it yourself.
Apparently there are a lot of measurement experts out there... :D
But are there any SMPS experts? The real issue is why this power supply is emitting this switching noise. Anyone know what may have failed in order for all of this switching noise to occur?
-
Hi,
If you share a few pictures of the power supply it may help.
Jay_Diddy_B
-
Like I say it's basically an ATX computer power supply. I checked inside and it's got a lot of PFC components. I think it came off an HP desktop, so it should be reasonably well designed.
But the image of the noise I posted looks to be a very typical SMPS output waveform, with switching noise plus ripple. The only issue is why the switching noise peaks are so freakin' high and not filtered out. I'm assuming there's a standard circuit in these things whose job is to filter out those switching peaks. Once I have that figured out I can focus on it and see if there's a bad component.
-
Anyway, I was looking at an Analog Devices paper on switching noise for these SMPS's, and it says:
"Switching transition noise is typically in the range of 10 MHz to 300 MHz. It is much higher in frequency than the switching frequency of a switching regulator. For attenuation of this noise on the output of the power supply, an LC filter, commonly used to reduce the output ripple voltage, may not be the right choice. Ferrite beads are much better suited to attenuate such high frequencies."
It looks like the noise I'm seeing is in the 40MHz range, which matches what the paper says. And as I mentioned before, and apparently this paper confirms, I ordered some ferrite beads the other day to see if that helps. Or maybe there's a dead C in an internal LC filter circuit...
EDIT: Cool...I just checked and my ferrite cores are arriving today. :-+ It's an assortment of like 20 different sizes, so I guess I'll be experimenting today. :D
-
Well, looks like a main switching edge coupling through... Whether it's a two-switch forward or flyback, or traditional half bridge I'm not sure.
Check primary and secondary side grounding, and by that I mean RF grounding, capacitors and all that. Missing or failed Y caps would be a possibility. Can always put it through a line filter if it doesn't seem to be anything onboard.
Tim
-
Hi,
A photograph of the pcb would really help the discussion.
You describe the this an 'old' pc power supply. The techniques have changed over the years.
I have seen cases where the interference suppression components have been removed, to save cost, probably after the power supply past FCC testing.
You could have dried up output capacitors,
You could have damaged X or Y caps.
Like I say a few good photos will help a lot.
I have several decades of SMPS experience.
Regards,
Jay_Diddy_B
-
Thanks. Here's the power supply guts. Didn't see any empty "important component belongs here" type spaces. And lots of PFC yumminess on the top there...
-
Well, FWIW, I tried wrapping some of my new ferrite cores around the +/- wires coming out of the power supply and it's pretty disappointing. Yeah, I know that it's a science project where you're supposed to match the core material to the frequency, blah, blah, blah, but I just grabbed some no-name (and no spec) cores off Amazon to see if they'd make a difference.
The only difference I noticed was when wrapping a larger core around both wires...it knocked the switching peak down by maybe 30%, but no affect on the ringing frequency or anything else. So I'm guessing this power supply is just emitting some nasty switching noise EMF and it's radiating into the room and getting picked up by the DUT circuit wires and my probes.
Hell, it's in a metal case and connected to ground, etc., so all I can figure is there's something amiss inside the case, like a bad cap or something.
Anyway, looks like this will go in the scrap bin, or maybe I'll get the energy to desolder some components and put them in my parts drawers.
But I know this will be nagging at me to figure out... :D
-
Well no, not on the top in the photo, that's line filter, more or less a 5 pole line filter, plus some ferrite bead and extra ceramic caps on the mains entry, by the looks of it. Definitely a PFC choke on the left though, the black core probably being Kool-Mu. Not sure why there are so many devices on the heatsink, I'd expect four (1 MOSFET and 1 diode for PFC; 2 MOSFET for output, most likely 2-switch forward).
Most likely failure I think would be something missing from ground. Measure the caps. Check the secondary side grounding, it should be shorted to the case, possibly in multiple locations.
Could still be some kind of electrical failure, other than failed filter components; a transistor accidentally shorted to a heatsink, or insulation failure in the transformer maybe, could do that I suppose (if unlikely without also shitting the bed completely).
Tim
-
Interesting...
I checked continuity between the power supply "ground" (black wires) and the power supply case. Nothing. I did the same with another ATX power supply. Same thing. I'm wondering if the secondaries are designed to be isolated, and maybe rely on the main PC ground when they plug in to the motherboard?
Also, I did a bit more testing. Scope turned on with a single test lead/probe connected to the BNC, but the probe left unconnected, and it's just resting on top of the power supply. The power supply was plugged in, but NOT turned on. Which means it was running in standby mode presumably.
And the attached is the noise I got. Of course it varied as I moved the probe around and more or less got induced in the leads. But wow, even in standby mode this thing is emitting that much RFI.
BTW, I also connected a wire between the black/ground wire and the case ground thru a ferrite core, and that made no difference.
So again we're back to the question: how do you filter something like this, and what could be going wrong here? Or is it normal?
EDIT: ...and if I add a 50 ohm resistor across the probe the noise goes away. And when I turn the PS on, the switching noise comes back. Is there something obvious and dumb I'm missing?
See, this is what I hate about this freakin' electronics stuff. All these weird zombie induced effects that suddenly appear then go away for no reason. What a PITA. You waste all this time trying to squash stuff that doesn't exist.
-
Okay, so I re-did my measuring per the attached image. Added a ferrite core on the twisted leads coming from the ATX supply 12V, added a few capacitors across the 12V, and used one of those low noise ground attachments to the probe. And this seems to have significantly reduced the measured noise. Still the switching spikes are there, but they're down in the 100mV peak range.
So I'm guessing this is much closer to the actual value of peak switching noise...maybe 50-100mV peak.
Okay, I feel better now... :D
-
Capacitive coupling of high frequencies. Which because the open circuit probe is such a high impedance (a few pF), doesn't take much.
How about this, clip 1k across the probe and measure:
- Probe GND to enclosure: probably nothing
- Probe tip to enclosure: probably nothing, or incidental noise
- Probe ground to secondary ground: some noise
- Probe tip to secondary ground: full noise
Hopefully 60Hz hum doesn't interfere with this test...
You're measuring the noise in otherwise-ordinary circumstances (probing circuits) because of the RF voltage drop along the ground clip and other unbalanced ground paths; and the CMRR of the probe cable itself might not be fantastic.
It seems the enclosure is a good enough ground, and the secondary leads are carrying CM switching noise. The noise is a voltage generated between enclosure and secondary. It probably has a high impedance (100s ohms, kohms?), so that it can be easily shunted to ground with some bypass caps to the enclosure, or direct grounding. As long as you have (had?) it open, this is easy enough to add.
Ferrite beads will do little to nothing, because their impedance is modest (~100s ohms). You need either a very high impedance (10s kohm+?), which isn't a practical value CMC at this frequency or amperage, or a low impedance shunting it back to the source.
Which also suggests additional filtering if desired; run the secondary through a CLC (CM) filter mounted to the enclosure. The last C also gives a low impedance, so ferrite beads may be useful again, say if you need to dampen a bit of ringing elsewhere.
Tim
-
Even with a differential probe, it requires good technique to get meaningful results. I usually end up using direct differential coaxial connections which modern differential probes do not support, but old ones do.
Also beware that modern differential probes tend to fall out of calibration due to poor construction and of course lack service documentation. So expect poor common mode rejection unless you verify it yourself.
Apparently there are a lot of measurement experts out there... :D
But are there any SMPS experts? The real issue is why this power supply is emitting this switching noise. Anyone know what may have failed in order for all of this switching noise to occur?
If the measurement is not accurate, and this type of noise measurement is very difficult to make, then the switching noise may not exist. Your results are consistent with a poor measurement so I and others have suggested going back and verifying your measurement setup by doing things like shorting the differential inputs at the ground connection.
I would go further and measure the high frequency common mode rejection of the differential probe since that is a common (ahem) problem these days.
-
Even with a differential probe, it requires good technique to get meaningful results. I usually end up using direct differential coaxial connections which modern differential probes do not support, but old ones do.
Also beware that modern differential probes tend to fall out of calibration due to poor construction and of course lack service documentation. So expect poor common mode rejection unless you verify it yourself.
Apparently there are a lot of measurement experts out there... :D
But are there any SMPS experts? The real issue is why this power supply is emitting this switching noise. Anyone know what may have failed in order for all of this switching noise to occur?
If the measurement is not accurate, and this type of noise measurement is very difficult to make, then the switching noise may not exist. Your results are consistent with a poor measurement so I and others have suggested going back and verifying your measurement setup by doing things like shorting the differential inputs at the ground connection.
I would go further and measure the high frequency common mode rejection of the differential probe since that is a common (ahem) problem these days.
Thanks, but the goal here was to decide, fairly quickly, if the power supply should be tossed or if it's useful. If it was legitimately generating what I initially thought (around 1-2 volts switching peaks) it should be repaired or tossed. But at this point I've determined that, with some simple changes in how I'm measuring (as described in my last post), the actual number is closer to 0.05-0.1 volts peak. Which to me says it's fine. Whether it's off by 0.126 femtovolts is somewhat irrelevant to the overall goal here. As I said, this isn't worth a science project.
What I did learn was, to my surprise, the ability of these things to generate large amounts of induced switching noise, especially since they are in grounded metal cases with tons of internal filtering. And also that a well-placed ferrite core can have an impact (maybe 20% in this case), though relatively small. And those low noise ground attachments to the scope probes are a big help, as is using a 1x multiplier rather than x10. As is making sure all the wiring on the breadboard is tight. And also doing some trial and error with additional filter capacitors.
BTW, I also yanked another ATX supply and got similar results, so it further confirms the belief that this thing hasn't failed. Now I just need to figure out what the hell I'm gonna do with it... :D