QuoteHow exactly did you arrive at 190uV RMS of noise? I assume you used the built in RMS measurement capability. That *should* work however 190uV RMS of noise seems high to me and I would want to verify that RMS measurements work correctly with wide bandwidth noise. Maybe something weird is going on there which seems to be a ubiquitous problem with Rigol DSOs.
Yes I used the built-in function AVG. The noise is measure in normal acquisition mode. In High Res mode were you average over a number of over captured values (not sure about what it exactly does), the noise goes down to 85uV. I'm not talking about trace averaging. If I do trace averaging, I still measure around 70uV. That could indicate there is some offset, or a systematic error (crosstalk, quantisation error?)
Using averaging or high resolution acquisition mode is note going to return a proper or comparable measurement of noise. Either will attenuate high frequency noise (and are useful for this reason); my measurements were over the full bandwidth of roughly 100 MHz. (1) The fact that you measured 85uV RMS and 70uV RMS with high resolution and averaging indicates to me that something is very wrong; the noise should have been much much lower because of limited bandwidth.
What should work is an automated RMS measurement of the trace without averaging or high resolution mode enabled.
The measurement should *not* change significantly at different sample rates or record lengths (although it may look visually different); if it does, then something is wrong and this is easy to demonstrate. The RMS value is equal to the standard deviation which is what the tangential measurement I used returned. (2)(3) If half of the sample points are removed by using a lower sampling rate, the standard deviation remains the same. If half of the sample points are removed by halving the record length, then the standard deviation also remains the same. This applies *even if* the noise bandwidth is undersampled so aliasing is not important (!) which is how RF sampling voltmeters can make RMS measurements into the microwave RF bands. (4)
The measurement should *also not* change significantly when operating in real time or with a single shot acquisition. I only bring this up because Rigol may be doing some DSP voodoo on its real time display to produce index grading which could interfere with an accurate RMS measurement so this should be checked. If the results differ, then I would trust the single shot measurement more than the real time measurement.
(1) I am ignoring shape factor of the bandpass. All of my measurements were on instruments with a single pole Gaussian rolloff which increases the measured noise by 1.6 times because noise above the 3dB bandwidth is also included. Do not sweat the small stuff until the big stuff is taken care of.
(2) A couple years ago I verified that tangential RMS noise measurement on an analog oscilloscope agreed with RMS measurement using sampling RMS voltmeter and automated RMS measurement on a good DSO. I mean it should, right? So does it? Actually, tangential measurement agreed so well while some DSO RMS measurements did not that I now trust analog tangential RMS measurement more than random DSO measurements.
(3) I do not think the Rigol can do it but some higher end DSOs can produce histograms and make standard deviation measurements. The standard deviation measurement should return the same value as the RMS measurement.
(4) This suggests another test which can be done besides changing the time/div or record length. The Rigol has to reduce sample rate when more channels (1 GS/s for 1 channel, 500 MS/s for 2 channels, 250 MS/s for 3 or 4 channels) are used but this should *not* affect the RMS noise measurement made on one channel. If the measurement changes, then something is broken.