How to measure PSU noise and ripple correctly

[slurry]

This seems to be an easy task, just connect your multimeter and see what you get.
...or is it a little bit more tricky than that? 

What would you consider being the "standard" method to measure the dreaded noise and ripple from our beloved power supplies?
I cant seem to find very much useful info on the www (lmgtfy please  )
Of course i watched #594 (thanks Dave) but i want more!


but i want more!

My setup today:
I use a scope, isolated from mains.
Measure with 20MHz filter, averaging 1000 times
and AC-coupled (duh..) 1Mohm input.
Using 1m of RG58 coax (~82pF), shortest possible leads to terminal and a BNC connector to the scope.

- Should i use a capacitive load at the scope terminal, around 10µF? (saw that in a datasheet somewhere)
S Should a scrap my coax and use a scope probe, but then i will have a long ground wire, i can use a BNC adapter on the probe and then a BNC-banana adapter... 
- Should i do my readings at the minimum, middle and maximum voltage if a lab psu or is it always at maximum voltage? (the ripple and noise can vary greatly over the range)
- What should i consider when i measuring the PSU under load, there is no dutycycle to take in consideration on a lab PSU so should a max it out?

Thoughts and facts are welcome, and by the way, a happy new year! 

[David Hess]

I usually attach a short coaxial pigtail with a BNC connector on the end to the power supply output or low impedance signal that I want to monitor.  Then a x1 or x10 oscilloscope probe can be connected using a probe tip to BNC adapter and there is no problem with the loop area of the ground lead picking up noise.

The alternative in a noisy environment is a differential probe.

[BravoV]

For common desktop computer's ATX PSU, there is already a documented standard on how to measure the noise & ripple, and as David pointed out, using a differential probe is the recommended tool.

Its clearly documented the measurement details here -> ATX12V Power Supply Design Guide

[tautech]

Bravo's ATX example best mimics a "real world" scenario with some load, bulk and bypass capacitance like a "real" circuit. Some SMPS units I've imported (not Siglent) needed 1 further parameter added in the test setup in order to meet their manufacturers spec:
Scope waveform Averaging.

With a DSO to clean up the displayed waveform, you use Averaging more than you'd think.

Which brings out the old favourites:
https://www.eevblog.com/2014/04/10/eevblog-601-why-digital-oscilloscopes-appear-noisy/
https://www.eevblog.com/2014/04/27/eevblog-610-why-digital-scopes-appear-noisy-part-2/

[Kleinstein]

Using a scope to get a defined 20 MHz Bandwidth is a good idea. However averaging mode will do to much and reduce the noise bandwidth - this how it works to get a clean waveform even with lots of noise. So this is OK to look at the ripple part, but not for the noise.

So it is more like using a sufficiently large window (low deflection speed) to get averaging for the build in RMS calculation. One might have to check the DSO - which are the correct setting and maybe take into account the DSOs own noise.

The short BNC - Coax cable should be OK, if longer termination at one end is a good idea. But careful with 50 Ohms termination at the scope and a higher DC voltage. With a supply with a floating output, there should be no need for a differential probe - with a grounded supply one might need one, or an insulation transformer to run the supply.

Noise and ripple can depend on the load. Using it close to maximum power is often the highest ripple. So it could be worth measuring with a high load (e.g. 90% of max) and a low one.

[bingo600]

I usually attach a short coaxial pigtail with a BNC connector on the end to the power supply output or low impedance signal that I want to monitor.  Then a x1 or x10 oscilloscope probe can be connected using a probe tip to BNC adapter and there is no problem with the loop area of the ground lead picking up noise.

The alternative in a noisy environment is a differential probe.

How short is "short" ??

Would a 10cm length be ok ?

/Bingo

[David Hess]

Using a scope to get a defined 20 MHz Bandwidth is a good idea. However averaging mode will do to much and reduce the noise bandwidth - this how it works to get a clean waveform even with lots of noise. So this is OK to look at the ripple part, but not for the noise.

Averaging is great for measuring line regulation, load regulation, and switching ripple.  An external trigger is often necessary for this to work.

So it is more like using a sufficiently large window (low deflection speed) to get averaging for the build in RMS calculation. One might have to check the DSO - which are the correct setting and maybe take into account the DSOs own noise.

Be real careful about using automatic RMS measurements for noise.  DSOs which use the display record for measurements are unlikely to produce the correct result; for instance the Rigol DS1000Z series does not.

The short BNC - Coax cable should be OK, if longer termination at one end is a good idea. But careful with 50 Ohms termination at the scope and a higher DC voltage. With a supply with a floating output, there should be no need for a differential probe - with a grounded supply one might need one, or an insulation transformer to run the supply.

One reason I like this is that it does away with the problematical 50 ohm termination in a DC environment although a 50 ohm AC termination could be used.  The short length of the pigtail and low additional capacitance will have no effect on low impedance signals.  I usually end up using the coaxial pigtail around switching power supplies.

[Conrad Hoffman]

I'm a dinosaur. For linear supplies, I slap a 1A7A high gain plug-in in my old scope and use a 24" piece of coax with grabbers on the end to connect. I set the plug-in filters for the bandwidth of interest. If doing audio work, noise at 1 MHz isn't really what I'm interested in, so I might limit to 30 kHz. I've never found meters very revealing of the things I want to know when it comes to ripple and noise, though they certainly work if you interpret them correctly.

Now, with switchers it's a whole 'nuther matter. The late Jim Williams of Linear Technology does a far better job of describing high frequency measurements in their app notes than anything I can say. Highly recommended, probably start with AN-47 as it covers probes. Basically, any loop area in the ground connection throws everything off, so special probe tips or even dedicated connectors are the rule.

[macboy]

Bravo's ATX example best mimics a "real world" scenario with some load, bulk and bypass capacitance like a "real" circuit. Some SMPS units I've imported (not Siglent) needed 1 further parameter added in the test setup in order to meet their manufacturers spec:
Scope waveform Averaging.

With a DSO to clean up the displayed waveform, you use Averaging more than you'd think.

Which brings out the old favourites:
https://www.eevblog.com/2014/04/10/eevblog-601-why-digital-oscilloscopes-appear-noisy/
https://www.eevblog.com/2014/04/27/eevblog-610-why-digital-scopes-appear-noisy-part-2/

If you use averaging, you eliminate random noise and the result is only the ripple component. That can be a valuable measurement, if ripple is what you want, rather than ripple+noise. Most scope inputs are not sensitive enough to accurately directly measure noise, so you need to use a separate low noise amplifier (LNA) to amplify the noise (while adding as little extra noise as possible) prior to measuring on the scope. You want the measured noise signal to be at least 10x higher magnitude than the background noise (i.e. what you see with the same setup and the power supply turned off).

[David Hess]

For linear supplies, I slap a 1A7A high gain plug-in in my old scope and use a 24" piece of coax with grabbers on the end to connect. I set the plug-in filters for the bandwidth of interest. If doing audio work, noise at 1 MHz isn't really what I'm interested in, so I might limit to 30 kHz. I've never found meters very revealing of the things I want to know when it comes to ripple and noise, though they certainly work if you interpret them correctly.

I sometimes do the same thing with a 7A22 which is a descendant of the 1A7A.  It is gratifying when a spot noise measurement agrees with other noise measurements and the design.