you have multiple samples within each 350MHz time period. Averaging them won't make the bandwidth less then 350Mhz, it simply reduces the noise inherent in the ADC process.
Hires is not averaging.
It's a calculation done on adjacent samples. It might be the mathematical average, it might be a FIR filter, we don't know.
We do know it's done in an FPGA with steps of 2x, 4x, 8x, 16x so I'm leaning towards "averaging".
Averaging is useful only for repetitive, low phase noise (low jitter) signals.
That usage of the word "averaging" isn't what was meant here, and you know it.
So you could say that 1 bit improvement would be transparent , like you say. But not more
That's the worst case scenario.
For 2 bit improvement BW is 0,0625 (1/16) of original Nyquist (8GS sample rate, 4GHz Nyquist effective 250MHz).
2 bit improvement is roughly what MSO5000 need to show similar RMS of noise to RTB2000 or SDS2000X+.
So with 1 channel enabled you can get as good as a Siglent (16x oversample, 500Mhz NYquist). Even with two channels enabled you can still get very close (8x oversample, 500Mhz Nyquist).
Or I could just say your are right. It doesn't matter, and we should all just stop using any scopes with more than 20 MHz BW because it is all just some unwanted interference we don't care about anyways.
Or we can be grownups and admit that most signals in daily life simply aren't so small that we need to do any of this. This makes for a nice bullet point on the sales brochure but it's mostly an edge case.
Hires can be implemented only one way. As a filter, of whatever implementation you chose. But result is same, lowpass filtering of current sample buffer.
If it is not implemented that way then it is not Hires, but something else. Simple. So we know exactly how it's made.
And it creates BW limiting.
Also averaging is averaging. That is also term that is well defined in this context. In this context it means repetitive triggers and combining (averaging) multiple buffers on top of each other. It does not create BW limiting and can be used to extract signals that are autocorrelated and in a fixed timing to trigger from noise. Signals have to be stable and repetitive.
I calculated whole table to show how rapidly BW drops and how it drops to very low bandwidth if you try to use Hires to actually have some usefulness from it. From it you took one data point and proclaimed victory.. Numbers are if you run at fastest sample rate. If you go with slower timebases and scope decides to drop sample rate BW will drop too.
Problem with Hires that is annoying that
BW changes all the time in seemingly unpredictable ways. Every time you change timebase sample rate might change. You enable or disable a channel it changes. You change how much Hires, it changes. You change memory size it might change. You end up with a scope that has "random" BW. What it does is that signal on the screen changes all the time. Looking at I/O pin on Arduino, signal risetime will vary from 1ns to 5 ns depending on Hires. You go slower timebase, it will drop slower than that.
That is confusing as hell.
Expectation from the scope is that BW won't change with you twiddling knobs and signal won't look different at different setting of timebase.
Hires is a
special acquisition mode. Like peak detect mode it is there for special occasions.
Yes we could behave like adults. And admit that you are wrong. You are pushing that worse specification of a specific device are not important to all and everybody, because you are of the opinion that it doesn't matter because everybody out there is doing same thing as you, apparently.
Signals in real life are absolutely small and large and all in between. All together on a same scope screen.
What are you saying, that we don't need vertical sensitivities better than 100mV/DIV, and scopes faster than 50Mhz?
I said many times, if you are doing only digital level signals, on low speed digital, MSO5000 is going to do good job for you. In it's price class it is a decent device for that kind of work. It has few quirks, but it will do that work.
If you want to look at analog domain properly, on mV level signals, any number of devices (including cheaper ones like Micsigs, SDS1104X-E, hires Picos etc) are going to be better for that. Tool for the job.
Stretching that to a statement that it makes it same as a devices with much better specs is what is wrong in your efforts. These workarounds might work for you, for the stuff you do. Good for you and enjoy. That doesn't make it universal truth. Just workarounds that happen to be good enough for you.