I use my USB scope (Picoscope 2205) to measure the voltage ripple of a switched power converter (wall wart that converts from AC to 5VDC). I use the AC coupling setting on the scope to isolate the ripple and noise.
I have a few challenging questions about the background noise on my test setup. My USB scope is connected to a laptop which runs on battery and is not connected to any other peripherals that could be a source of noise. The peak-to-peak ripple that appears before I power-on the switch is 8mV, which is typical of what I've read in other forums on this topic for USB scopes. When I power-on the switch, the peak-to-peak ripple displays at 50mV.
My questions are:
1) Does the 50mV peak-to-peak ripple consist of the 8mV of background ripple plus 42mV of ripple from the switch? Or is the background ripple hidden within the switch ripple rather than additive to it, in which case the switch alone is the source of all 50mV of ripple?
Signals add or subtract, they do not "hide" in one another. What you will see, though, is if one signal is huge and the other tiny, the tiny one will become insignificant. That doesn't mean it's not there, but the large signal will push it into insignificance. Of course, in your case with not so huge a signal, you have noticed that the "tiny" signal has become significant.
It's likely that the background noise of the scope is more or less wide band white noise. This will be added to the noise from the SMPS which will have a distinct frequency depending on its switching frequency (say 100 kHz or so).
2) If my scope can detect all higher frequencies involved, does it matter to the total ripple measurement if the frequency of my background noise is different than the frequency of the switch ripple? In other words, when measuring peak-to-peak ripple, is it customary to include multi-frequency components of the waveform, or do you try to isolate discrete frequency ranges and measure them separately?
Typically when measuring power supplies, a bandwidth limit of 20 MHz is used. And yes, switching off this limit drastically increases the noise on the measured signal.
Noise is a multi frequency phenomenon, when distributed evenly across the (audible) spectrum, it's often called white noise. Ripple is limited to rectification of the 50/60 Hz mains (full wave rectified: 100/120 Hz) and in an smps also the frequencies used in the switching process and their harmonics.
Square waves will have much more high frequency content (harmonics) than sinusoidal signals.
Rather than just look with a scope at the ripple, also use a spectrum analyser (or the FFT function, if your scope has it). That will show you the noise floor all across the spectrum and peaks rising above this noise floor where the switching noise (with harmonics) is.
3) Assuming that the background noise is additive to the switch ripple, what is the common method to account for the background noise so I'm not overstating the ripple from the switch? If they were at completely different frequencies, I could apply a filter to remove the background noise. But let's say they have some overlap in frequency. In this case, do I simply perform a manual subtraction of the background noise when reporting results?
Assuming that the background noise is spread across the spectrum, it's no use trying to filter out specific frequencies. You must see how much that random freqency content rides on the frequency content of the smps and account for it. Depending on your scope's options, you could play around with settings like persistence and averaging to give you a cleaner signal.
My scope supports math channels for transformations of the signal, but I can't figure out a way to subtract a constant noise value from the oscillating waveform that centers on zero.
You can add or subtract noise to or from a single frequency signal just fine. But there is a catch.
I have added a screendump (Screen1) of 1 Vpp noise on CH3 (pink) added to a 5 Vpp 1 kHz square wave on CH1 (yellow) to simulate noise injected into a clean signal. The math trace represents this polluted signal. It is the purple trace and it is magnified to make the result clearer. Unmaginified it would sit right where the yellow trace is but with the added noise.
The vertical scale (horizontally placed) cursors are placed 5 V apart at + and - 2.5 V. This is where the clean 1 kHz square wave would be but as you can see, the noise adds to and subtracts from this making the peaks vary randomly with the noise's 1 Vpp amplitude.
Thanks for replying to my questions. Yes, I watched the video (before I posted my questions) and I just watched the segment again at your suggestion. In this video, Dave addresses a lot of topics but not all the ones I have questions about. A couple examples:
- He refers to ALL background noise as Common Mode noise. This doesn't apply to my setup as I have multiple sources of noise from my test setup (USB scope connected to a laptop running on battery) which are probably not Common Mode related but originate from my PC's internal voltage regulator, hard drive, fan, etc.
- As he investigates the source of his Common Mode noise and finally finds it, the solution he takes is to eliminate the noise. I can fully understand this is the preferred approach. However, I am resigned to keep the noise in my test setup, and I would just like to subtract it out from the gross measurement leaving me with the net signal. I will try the two-channel subtraction method that Dave demonstrates ("poor man's differential probe") and I'll post back with results.
Any additional comments are welcome.
In your setup, you would have to take into account the amount of noise to be able to determine the smps's ripple. But please note that the noise you see riding on this signal may not be just the background noise of the scope, escpecially if the noise's amplitude is significantly higher than that of a terminated scope channel without any input signal. If it is higher, you will have picked up noise from other sources, and measuring low level signals this easily happens, as Dave's video showed. So the techniques shown to prevent this "pollution" of your measurement are really neccessary to prevent false readings and drawing the wrong conclusions.
Edit: so you can add and subtract any frequency, but the catch is that noise is random. Subtracting one noise signal from the other will never result in a noisefree signal, unless they're identical, like common mode noise.
To show this, the second attachment (Screen2) has CH1 and CH3 connected to two different outputs of the signal generator set to generate the same kind of noise signal. They may look identical, but coming from two different outputs, they're not really (or rather: really not). As a result, subtracting one from the other hardly leads to a flat line. That's what you get with random noise and the background noise from CH1 of your scope is likely not identical to that of CH2.
For the last attachment, I tied both CH1 and CH3 to the same generator output. Look at how much flatter the math trace is now that both signals are identical (it's not completely flat due to minute differences in cables and scope inputs, but you get the idea).