My original hypothesis that the front end attenuates in line with Nyquist to avoid aliasing was incorrect. It doesn't attenuate as the sample rate decreases.
I'd originally derived this from measuring rise time - fail on my part!
Today I hooked up an Agilent E4433B RF signal generator, set it to 200MHz 10dBm with a 50 ohm through attenuator at the scope.
Here are the results, 1ch, 2ch, 3ch, 4ch and 4ch+LA. Vrms doesn't change by any significant amount.
Note the differences between Freq and Counter measurements when the LA is turned on. I thought the counter measurement might always give the correct answer, but no.
Howard - thanks a lot for pointing this out. A few days ago, I made
similar findings here regarding high frequency aliasing, albeit without plugging in and considering the effect the digital probe may have (actually, I assumed it wouldn't change analog sample rate since it's not been documented). The counter discrepancy that you observed between the so-called "Hardware Frequency Counter" and the "measurement window counter" when enabling the digital channels is quite interesting and may give us a hint where the "bottleneck" is. May I suggest to perform the same test with the digital channels while only one and two analog channels are active to see if this discrepancy between the two counters is present as well (obviously, with high enough input frequency to cause aliasing)?
My take on the shortcoming of the scope hardware is memory bandwidth! The "hardware" counter operates on the data stream from the ADC directly, before it gets sent to the sampling memory while the measurement window counter probably operates on the decimated shadow memory which resembles the sample memory (from a temporal point of view) if the time base is set fast enough.
What really surprises me about this is that there's two additional RAM footprints available that --
at least in the photos of the pre-release version of the DHO900 -- had been found to be populated. Assuming that these additional memory chips provide storage for digital input sample data and waveform data for the AWG, I wouldn't expect this to slow down the analog channels. An FPGA should be able to process all this independently in parallel, as long as it's chosen with a sufficient amount of logic cells... So we actually may be up to a surprise if we look into a recently produced DHO900 scope -- the additional RAM foorprints may still be unpopulated.
If it's due to the lack of FPGA resources, Rigol will have to accept the charge of having skimped on the choice of the Zynq version that they used on the DHO900. Whatever the reason for this observation is (I mean the further reduction of sample rate when enabling the digital channels), I consider this a major flaw that most probably cannot be corrected with a firmware update. This was the last "nail in the coffin" that finally triggered my decision to return the DHO914S after the Bode Plot disaster.
I may get a DHO804 at a later time since there are many quite nice solutions in this scope as long as it's used purely as an "analog" scope without the additional functions that the DHO900 series is supposed to provide. The U/I is quite fast compared to Rigol's legacy instruments, the FFT is lightning-fast and provides quite decent performance (IMO), and the over-all appearance makes it a decent, modern instument to "carry around" in the lab / workshop.
What didn't seem to get better during the ten days that I had to "play" with the DHO914S is the "baked electronics smell" that it emanates and that may actually be quite distracting when evaluating a new electronics design, not knowing where the "burning" smell comes from. In a home lab, maybe even in a living area, I'ld consider theis smell an absolute no-go. Rigol should definitely address their PCB manufacturing/flux cleaning processes to eliminate this problem.
"My" DHO900 is back in its box and will get returned these days. It's been an interesting experience but also one that cost me way too much time for an unfinished product that in it's current configuration (IMO) is not worth spending the time and the money on. At least as long as none of use is getting paid by Rigol to do their work!