Impact of scope noise when measuring signals

Can't say I'm an expert, but here goes at my understanding:

How does the scope noise measured when the channel is disconnected from a probe - noise floor? - impact on the measured signal?

1. How do I take that into account when interpreting the signal?  For example, could I take the mean RMS value shown in image 1, from the mean RMS value in image 2?  Does it mean the scope just isn't capable of measuring signals around the min-max RMS values?
2. I don't touch the probe once I've started taking measurements - it's connected on test points - but even a fractional movement can change the measured signal to the extent I can't trust that I've made a proper probe connection, if that makes sense.  Specifically, the amplitude of the 'spikes' can grow or shrink although the 'core' signal remains pretty much as-is; by 'core' I mean the bright green portion of what is shown.  Does that imply anything about the spikes not related to my circuit - environment noise?  Probing issue?
   I've turned off other devices and lights in the room with no impact.  I get the same signal measuring directly across an output cap (albeit a 10uF electrolytic.)  I've zoomed in to the signal in the image below - although regular, it doesn't look like switching noise:

Scope noise with absolutely nothing attached to the input connectors mostly comes from the front end amplifier and whatever little can be coupled into the input connectors. Terminating the inputs reduces the coupled components. This is the noise "floor" - there is no reliable way the oscilloscope can usually resolve below this, although with reduction in bandwidth through averaging, this may be possible. The noise will have both relatively "white" AC components and possibly a DC "offset" component depending on how well it is calibrated, and some of these may also drift over time with changes in operating temperature and ageing of components.

Connecting a probe, even with nothing on the other end of it, is expected to increase the noise, sometimes significantly as sources of RF, EMI, ESD are coupled into the oscilloscope inputs via the probe lead which can act a bit like an antenna. The values will change as you move the probe around because of the way the lead may pick up noise will change depending on the geometry in which the lead/probe makes to the source of the noise. Some DC offset may even be measurable due to thermoelectric voltages being built up at junctions between dissimilar metals. If you use the flying ground clip lead, that would also be a good source of noise. Other potential sources include using the front USB connection and rear LAN/USB connections on big-box units, or poor grounding.

If you use a 10x probe, you're bound to have even more noise as the signal is divided down in return for the full bandwidth. Using 1:1 coupling is usually best if you're trying to resolve very small signals, but bandwidth will often be limited ~<10MHz.

Spiky noise could be a result of many possibilities - it could exist because the probe contact has shifted very slightly during measurement (e.g. contact crackle), it could be "radiated" into the setup (e.g. Wi-Fi router, mobile phone), it could be coupled ESD static, it could even be buggy oscilloscope ADC firmware especially when interleaving multiple ADCs where they are not perfectly matched. It can even be from within the oscilloscope - some cheaper units aren't always internally well shielded or have the cleanest power supplies. Other times, you could be measuring actual artifacts in amplifier circuits - e.g. popcorn noise which is a real signal.

As for whether you can take the RMS values and subtract them - yes and no. The RMS average calculated across the whole buffer acts as an averaging mechanism. Noise (if purely white) averages out leaving the signal component alone. But if there is any "bias" in your noise, then averaging that would cause a bias in the output. Averaging the signal over a longer period to eliminate white noise contribution may seem like a good idea, but then you become vulnerable to instability and drift of the actual intended reading. Finally, if the signal is below the floor, you just don't know whether you're reading the oscilloscope's own internal biases and non-linearities as you're really only using a few LSB of the relatively "low" precision converter (as compared to using an appropriate DMM/SMU). In the end, your resulting signal-to-noise ratio may not be large enough to have confidence you are measuring an actual signal, rather than the oscilloscope front-end drift!

Generally speaking, what you are trying to do doesn't seem sensible to do with an oscilloscope input "bare" - I wouldn't trust the measurement results even from a higher-end scope for signals that don't form at least 10mV peak-to-peak purely because the noise will affect the computed measurements. What some people do is to use an active probe with a sensitive preamplifier front-end to measure such small signals using an oscilloscope - something like https://www.alphalabinc.com/product/lna10/ although as expected, because of the characteristic of white thermal noise, what you gain in measurement stability comes at a cost of bandwidth. This is also where some rather expensive power-rail probes come into play as well if measuring the power supply ripple-and-noise is your thing - e.g. R&S RT-ZPR40 but everything comes at a price ...

- Gough

Can't say I'm an expert, but here goes at my understanding:

How does the scope noise measured when the channel is disconnected from a probe - noise floor? - impact on the measured signal?

1. How do I take that into account when interpreting the signal?  For example, could I take the mean RMS value shown in image 1, from the mean RMS value in image 2?  Does it mean the scope just isn't capable of measuring signals around the min-max RMS values?
2. I don't touch the probe once I've started taking measurements - it's connected on test points - but even a fractional movement can change the measured signal to the extent I can't trust that I've made a proper probe connection, if that makes sense.  Specifically, the amplitude of the 'spikes' can grow or shrink although the 'core' signal remains pretty much as-is; by 'core' I mean the bright green portion of what is shown.  Does that imply anything about the spikes not related to my circuit - environment noise?  Probing issue?
   I've turned off other devices and lights in the room with no impact.  I get the same signal measuring directly across an output cap (albeit a 10uF electrolytic.)  I've zoomed in to the signal in the image below - although regular, it doesn't look like switching noise:

Scope noise with absolutely nothing attached to the input connectors mostly comes from the front end amplifier and whatever little can be coupled into the input connectors. Terminating the inputs reduces the coupled components. This is the noise "floor" - there is no reliable way the oscilloscope can usually resolve below this, although with reduction in bandwidth through averaging, this may be possible. The noise will have both relatively "white" AC components and possibly a DC "offset" component depending on how well it is calibrated, and some of these may also drift over time with changes in operating temperature and ageing of components.

Connecting a probe, even with nothing on the other end of it, is expected to increase the noise, sometimes significantly as sources of RF, EMI, ESD are coupled into the oscilloscope inputs via the probe lead which can act a bit like an antenna. The values will change as you move the probe around because of the way the lead may pick up noise will change depending on the geometry in which the lead/probe makes to the source of the noise. Some DC offset may even be measurable due to thermoelectric voltages being built up at junctions between dissimilar metals. If you use the flying ground clip lead, that would also be a good source of noise. Other potential sources include using the front USB connection and rear LAN/USB connections on big-box units, or poor grounding.

If you use a 10x probe, you're bound to have even more noise as the signal is divided down in return for the full bandwidth. Using 1:1 coupling is usually best if you're trying to resolve very small signals, but bandwidth will often be limited ~<10MHz.

Spiky noise could be a result of many possibilities - it could exist because the probe contact has shifted very slightly during measurement (e.g. contact crackle), it could be "radiated" into the setup (e.g. Wi-Fi router, mobile phone), it could be coupled ESD static, it could even be buggy oscilloscope ADC firmware especially when interleaving multiple ADCs where they are not perfectly matched. It can even be from within the oscilloscope - some cheaper units aren't always internally well shielded or have the cleanest power supplies. Other times, you could be measuring actual artifacts in amplifier circuits - e.g. popcorn noise which is a real signal.

As for whether you can take the RMS values and subtract them - yes and no. The RMS average calculated across the whole buffer acts as an averaging mechanism. Noise (if purely white) averages out leaving the signal component alone. But if there is any "bias" in your noise, then averaging that would cause a bias in the output. Averaging the signal over a longer period to eliminate white noise contribution may seem like a good idea, but then you become vulnerable to instability and drift of the actual intended reading. Finally, if the signal is below the floor, you just don't know whether you're reading the oscilloscope's own internal biases and non-linearities as you're really only using a few LSB of the relatively "low" precision converter (as compared to using an appropriate DMM/SMU). In the end, your resulting signal-to-noise ratio may not be large enough to have confidence you are measuring an actual signal, rather than the oscilloscope front-end drift!

Generally speaking, what you are trying to do doesn't seem sensible to do with an oscilloscope input "bare" - I wouldn't trust the measurement results even from a higher-end scope for signals that don't form at least 10mV peak-to-peak purely because the noise will affect the computed measurements. What some people do is to use an active probe with a sensitive preamplifier front-end to measure such small signals using an oscilloscope - something like https://www.alphalabinc.com/product/lna10/ although as expected, because of the characteristic of white thermal noise, what you gain in measurement stability comes at a cost of bandwidth. This is also where some rather expensive power-rail probes come into play as well if measuring the power supply ripple-and-noise is your thing - e.g. R&S RT-ZPR40 but everything comes at a price ...

- Gough

Thanks Gough, some very useful information in this.  I've responded to everyone in one post below.