Quick experiments on measuring noise. Excuse the crap photos - they are phone ones and the curtains are shut and the light off.
Output DG811 noise 1V RMS into 50R. I'm not sure what the noise power/bandwidth relation is on this.
Well firstly I did the old two signal thing and tried to overlap them until there is no peak as you do with an analogue scope. That did not work because the channel display of the active channel completely overlays the other one rather than adds to it.
If you turn off intensity grading it is more profound:
Now if you do one channel and measure the RMS voltage it comes out at 0.666 volts.
Let's try the 34401A ... 0.123 volts
Now the 3465A ... 0.145 volts.
Worth noting that the RMS figures might actually be fairly accurate as the "noise power" is proportional to bandwidth. The 34401A is true RMS responding with about 1MHz of bandwidth (measured). The 3465A is average responding with about 800KHz of bandwidth (measured) but peaks slightly above the first dip so that might explain that. The scope is, well 200MHz in theory so that's going to read a lot more.
So conclusions:
1. Digital scopes are shite for measuring RMS noise using traditional methods.
2. They might be ok using direct RMS measurements but that would need validating.
3. Buy a power or RMS voltmeter with thermal RMS sensor and an analogue scope
4. Noise is difficult to measure and you must consider instrument bandwidth.
Further experiment: Develop a passive 50 ohm low pass filter that is under the bandwidth of all instruments and has close to zero insertion loss thus equalising and measure again on all devices.
Edit: 5th order chebyshev LPF should do the job...