Test methodology

[quest for silence]

Originally Posted by ITelektro

My English is bad but I hope I was able to explain what is SMPS ripple and how it occurs, if there is any questions....

Well, at first glance I don't think the problem lies there: broadly speaking I know what ripple is, at least for linear technologies, while I don't know measuring techniques, particularly for SMPS.

I'm just wondering how different time window on the oscilloscope may affect a given output voltage and relevant artifacts, which should not change (or not that much) whatever you will do to measure them (on the oscilloscope settings, with reference to probes and ground loop I guess it would be another whole story: but as I said just above, I don't know measuring techniques/methodologies).

[ITelektro]

Originally Posted by quest for silence

Well, at first glance I don't think the problem lies there: broadly speaking I know what ripple is, at least for linear technologies, while I don't know measuring techniques, particularly for SMPS.

OK, it can be useful for other members. Measuring the ripple is one of the simpler measurements of power analysis. But probing technique is important.

]
Same PSU (Rigol DP811A @40v 5A load) with ground lead and without.

I'm just wondering how different time window on the oscilloscope may affect a given output voltage and relevant artifacts...

It will not. Ideally, the measurements with oscilloscope will not affect the signal that is measured.
If the question should be: why do we have different measurement results?
Then the answer is quite simple, because if we measure in the us range we will observe only that part of the sample. Much like when we look at something with a microscope or with the eye. The important thing is; what we want to observe. In our case we observe SMPS ripple, that particular part. It is on same freq as PWM, usually around 100kHz.
If you want to see the full range of frequency vs. gain, then an oscilloscope is not an instrument that should be used. In that case it should be considered a spectrum analyzer or oscilloscope FFT function.

FFT shows the signal peak at low frequencies (about 100hz) because it is linear power supply.

[ITelektro]

It's been a while.....

Originally Posted by Travis

The voltage on the bulk capacitor has ripple at ~100Hz (2x the line freq). Since the DC/DC stage has a limited ripple suppression ratio at this frequency, part of that ripple may go through the entire DC/DC stage and remain in the DC output.

Measuring ripple in ms range may miss high freqvency wave forms like switching transient noise or random noise, even switching ripple. It will depend of equipment that we use. But, the ripple component is relatively robust and will only be slightly reduced by the addition of additional capacitance on the output. The noise component has a much lower energy content and can usually be successfully attenuated by using capacitance at the product’s output and at the input to critical load components. Thus, in most cases measuring ripple in ms range will give most relevant results. Anyway, best way is to show both waveforms especially with low sample rate osciloscopes and in case when periodic random noise occur.

Meanwhile lot of questions rise up.....

Originally Posted by Behemot

Do you have differential probe? Do you have decouplers and what are they? Where do you measure? Do you have all connectors plugged into something? These all things come into play. Not even the norm specifies all of them and they pretty much heavily affect the results. We all mostly don't talk about the fact none of us (correct me if I am wrong) has differential probe but as long as our results are +- the same,

Intel recommendation is common method of measuring ripple, even Intel does not describe where exactly hook the probe, other documents are.
The capacitors should be located at the oscilloscope probe tip and close to the output pins of the product.Therefore, the most accurate reading is obtained
when the oscilloscope connection point is located directly at the capacitors(0.1μF).
Differential measurment will rejact most common mode signal, free connectors will not affect the results.
Differences betwen 1X passive, 10X pasive, 1X differential, how insufficient bendwith affect results and lot more try to answer in my article obout ripple measurment.
Can be found on my web. Hope for translation soon.
Constructive critiques are welcome, thanks.

[crmaris]

Lol I also own the same scope (keysight), soldering station (Welder), similar desoldering station, the same bench PSU (Rigol). Wow!

I like your bench table

[crmaris]

when you measure ripple the most important is to provide a clean source! If not then what ever you use, differential or normal probe, will be rendered useless.

Also all professional grade breakout boards have the decoupling caps near to the scope's connection board, as itelectro mentions.

For me differential probes aren't needed. I have a good pair of those which I only use when i want to measure really high voltages or when I want to reassure that my scope won't break. A good and fast scope is what you need (not so fast though unless you want to measure transient response). Even a plain stingray scope with its joke bandwidth catches the same amount of ripple that an ultra expensive scope reports.

However there is a trick here. Some PSUs although show normal ripple, once pushed periodically they have spikes. In order to catch those spikes you need to monitor for quite a long time and use a trigger action or just log in real time the scopes output and have software report those spikes (I use the second method since I am a software/control freak).

In addition, if you put extra sensors inside the PSU you should be extra careful because those sensors can increase ripple in some cases! Learned this the hard way.

Finally, everything that is connected to the PSU, including the scope, has also to be fed by a totally different line since in some cases even a Chroma source isn't enough to provide a clean input to the DUT.

[Behemot]

Getting different line may be somewhat tricky. I at least put an isolation transformer between the line and the measured PSU…

As for the insufficient bandwidth of the ×1/×10 probes, I've seen that, it's nicely covered by EEVblog in one of the videos.

[ITelektro]

I watched video yesterday, a great work by Dave's, as usual.
Although there Passive probes that are suitable for these measurements, but they are quite rare and expensive. Usually 2X probe.
Anyway, it's just part of the story with conclusion; "there really is no “right” probe selection for any given application. There are only “right” oscilloscope/probe combination selections"

Originally Posted by crmaris

Lol I also own the same scope (keysight).....

Mine is touchy

[ITelektro]

Originally Posted by crmaris

when you measure ripple the most important is to provide a clean source! If not then what ever you use, differential or normal probe, will be rendered useless.

Maybe, but first of all we must define what is clean source or, in other words what is not clean source.
Every PSU have some PSRR (power supply rejection ratio) so noise will be reject depend of that value. I think, the difference between measured ripple of PSU on Croma source and PSU on isolating transformer will be negligible.

For me differential probes aren't needed.

In some cases, I agree. But in some other cases measured ripple results will be at least twice high depend of technique and probing solution.
In ripple measurement oscilloscope and probing solution must be defined or test results are questionable since they cannot be reproduced.

Worst case scenario is;
Attachment 2497
10X probe VS Diff. Probe. (scope is calibrated, differential probe is verified on calibrated instrument (rigol dg1022)).
If ripple is lower amplitude, difference in results will be even higher.
This is important for newer PSU, because we can expect lot of PSU whit ripple lower than 10mV. Signal to noise ratio on 10x probe will be even higher.
This is example;
Attachment 2499
Solution like 1x or 2x probe will give lot better results but only in case if probe have enough bandwidth.
Attachment 2498
Here we can se 100mV 20Mhz signal measured with 1X probe 7Mhz bandwidth.
Even signal is 100mV measuring results is only 31.5mV.
There are some cheap and good solution but I will leave that for some other time.
So, yes 10x probe is in some case good enough but all this above is a reason why Intel specs recommends differential probe.

[crmaris]

I use both an isolation transformer and an AC source. Clean voltage is the key in order to properly measure ripple. Also you have to use a dedicated line (which i also do). If you have excess noise in this line then nothing of the above will help 100%. The AC source has the proper EMI filtering in order to filter the line's noise but its main purpose is to help you apply every waveform you want, and basically it keeps the voltage steady in order to have the same exactly parameters in every test regardless of the PSU's capacity.


Ripple below 10mV is of less importance according to my opinion. What if you measure 5mV and it actually is 7 or 8mV. On top of that you even will lose that if you don't use a logging software to monitor the entire load test (which can last from 5-15 minutes at a time) and have in hand the entire ripple session of the PSU (with a good scope you might be able to bypass this though).

I have a good set of differential probes, only use it to measure dangerous (very high voltage) stuff. I am aware of the affordable diff probe solutions, since I searched about the rebranded models that are on the market before I acquired mine

ATX spec recommends lots of stuff, however I seriously doubt if an electronics engineer has written it or checked it (at least the latter versions). See the silly high ripple limits that it has for example, which remain the same for many years now.

[ITelektro]

Originally Posted by crmaris

I use both an isolation transformer and an AC source. Clean voltage is the key

Define term "clean voltage", please.

Ripple below 10mV is of less importance according to my opinion. What if you measure 5mV and it actually is 7 or 8mV.

4mV and 8mV, double difference is not irelevant. It is only if its easy for PSU manufacturer to archive that. Is it? I dont think so.
I suggest in order to constructive discussion leave personal opinion beside.
Hope for link or screenshoot of measurement.

On top of that you even will lose that if you don't use a logging software to monitor the entire load test (which can last from 5-15 minutes at a time) and have in hand the entire ripple session of the PSU (with a good scope you might be able to bypass this though).

I agree...Logg. software is one way. Another way is statistics option that lot modern scopes have.

I have a good set of differential probes, only use it to measure dangerous (very high voltage) stuff. I am aware of the affordable diff probe solutions, since I searched about the rebranded models that are on the market before I acquired mine

I think that you understand this well, but it can confuse some young players.

High voltage differential probes are not suitable tool for ATX PSU ripple measurement. Attenuation ratio probe is typically more than 20x(10x,20x,100x,1000x)..... Signal will be attenuated by factor of 20 and noise will be multiplied by same factor.
So, for low amplitude signals like ripple (for modern ATX PSU) high voltage differential probe is not recommended. It is same case like 10x passive probe.
Second thing is, high voltage differential probe has significantly lower CMRR(Common-mode rejection ratio). Typically 1:100 while low voltage diff. probe has ratio 1:3000.
All this is like tolls in workshop. We have screwdrivers, hammers, wrenches... We use tool that is suitable for job.
There is no way to get absolute precise results, but its important to know limitation of tools that has been used.
Regarding that, "clean source or no clean source" will affect results, more important is how much expressed in mathematical form rather then like opinion.
Like we can see, in some cases probing techniques and probe selection will significantly affects results. In some cases 30% and more. Is this relevant? We can argue, but it is fact not opinion.
For cheap and good option I thought of 50ohm path. I have pick up all components from local electronic store, will post here results soon.

ATX spec recommends lots of stuff, however I seriously doubt if an electronics engineer has written it or checked it (at least the latter versions). See the silly high ripple limits that it has for example, which remain the same for many years now.

I dont know, Crmaris. Even I repaired aprox 10k motherboard I am not convinced to put credibility engineer that has written that document. I need good reason for that.
I have good example. Santa rosa platform and Merom CPU on Toshiba motherboards. Most common problem is capacitor on Vcore voltage. System will be stable untill ripple exceed 100mV, its about 10%. First conclusion is, motherboard are not so sensitive to ripple voltage.
Second thing is design standard. If is motherboard designed to handle that amount of ripple and work stable there is no problem, it should be.
I agree, lower is better but its not about that, its about what is enough regarding motherboard design.
There is one thing to consider too, influence on capacitor life....
However, reason for differential probe recommendations still stays as best option for making low voltage measurements.