What about testing a 100MHz DS1104Z against another 100MHz DS1104Z ? Shouldn't it be considered also in these "scientific" analyses?
I would be nice to test lots of them but not very practical. Few people with the needed test equipment have access to more than one or two and true 100MHz DS1104Zs are understandably rare.
Wasn't hard. First was a charged capacitor and fast switch. I do not recall what I was doing on the second, but probably something similar with different triggering. What should I be looking for?
Overload in this case depends on volts/second so if it is occurring, it also amplitude dependent. Higher sensitivities are less likely to display it because the volts/second of the edge is lower to stay within the display.
So doing that test at 5 volts/division like you did should show it but there is also the complication that the high impedance divider before the high impedance buffer is being used. That divider may be limiting input bandwidth preventing the problem. The example I showed is at 400 millivolts/division which I assume is actually 200 or 500 millivolts per division as far as the various attenuators. Worst case will be the lowest sensitivity where the high input impedance divider is not used. The table of bandwidth measurements not long ago indicated that the high impedance divider had a major effect on input bandwidth.
I do not see any slew rate related overload in your examples.
Hmm.. I had made an assumption there myself, that of the different bandwidth models actually having different (analog) bandwidth and/or rise times (in the front end path). Which, afaik, requires, for example (might have more to it), hardware low pass filter being enabled/disabled/adjusted in the hardware (via software control on the basis of the model), thus my idea of "hardware used differently". (Put in other words: both models having the filters there = "same hardware"; one disabling a piece of it = "used differently".)
Dave's reverse engineered schematic shows two digital control signals which switch a pair of analog bandwidth filters (they switch some shunt capacitors and it is a common arrangement in oscilloscopes for this function) to provide 20, 50, 70, and 100 MHz bandwidths. These numbers can be calculated from the component values shown in his schematic. The thing he has labeled as "bandwidth" is something else which is not entirely clear.
Look for cause why scope which clearly does not have actual 200MHz analog frontend reports such rise time 0.35/1.65ns=0.212GHz
Plotting amplitude response graph to sinusoidal signal sweep might give some clues.
The other post I mentioned reported the results of such a test and the bandwidth varied significantly with vertical sensitivity which is unusual; usually oscilloscopes are designed to have constant bandwidth. Sometimes bandwidth decreases at the highest vertical sensitivity settings because gain stages are switched instead of attenuation stages but it still remains constant with different signal amplitudes. It used to be a marketable feature if bandwidth was constant but buyers usually assume this is the case when it may not be.
What was not tested is if the bandwidth varied with amplitude indicating non-linearity and a full power bandwidth limit.
The single input divider relay of the DS1000Z switches between 330 and 335mV/div and this makes a big difference: at 335mV selection, the input amplifier "sees" much less signal and the VGA integrated in the A/D converter is adjusted at high gain. I kept that as a reference trace while I changed input sensitivity to 330mV, resulting in the input amp being driven with a much larger signal. The trace shapes and rise times differ by more than 1ns, but both appear to be well within the 100MHz range.
The shape is interesting in your second example. The first shows a Gaussian response but the second shows a first order response.
So the response was Gaussian with the high impedance attenuator and high ADC gain but first order without the high impedance attenuator and low ADC gain.
Your first example seems to show a 5 nanosecond ledge immediately after the edge and it looks like it is in the reference recording of the same signal. I wonder if that is something or not. It is too bad that Leo Bodnar's pulser cannot generate a higher output amplitude.
We know the design is the same or at least I assume it is. We do not know if Rigol grades or selects them.
Lots of people have tried to find a difference over the years, all have failed.
Most users lack either the equipment or experience to find this sort of difference.
(And there's absolutely no reason to think they're using borderline components that might go one way or the other during manufacturing).
Do you mean other than using 2N3904s in 100MHz transconductance amplifiers? That by itself is suspicious.
Rigol did not include those emitter equalization stages with variable tail current without reason. The variable tail current is suspicious as hell because a low tail current would lead to exactly this sort of problem.
Rigol also did not include collector series loads for thermal balancing but with that weird cascode arrangement, maybe they were not needed. Nobody has tested for thermal balance but it is an even more obscure issue and I doubt anybody is using a Rigol oscilloscope where it would matter. Thermal balance affects settling time in the 10s to 1000s of microseconds and is the major limitation for settling time in precision amplifiers.
A hacked 100MHz DS1054Z should be compared to a 100MHz DS1104Z.
It's been done several times over the years and no difference found.
If they are identical which I regard as a likely because for reasons of economy (I doubt Rigol is testing these oscilloscopes and only using the best ones for the higher bandwidth model), then that makes the DS1104Z just as flawed. I only suggested that they are grading them to give them the benefit of my doubt.
An oscilloscope is primarily a visualization tool, with some capability to make low precision measurements.
I at least expect the visualization of the same edge at different vertical sensitivity settings to be consistent. Changes in pulse shape with amplitude and position are actually a very good test for overload.
If the transition time measurements are changing over the range reported for the same edge, then that is even worse.
The individual man hours spent discussing this subject are worth more (in cash) than the money needed to buy a better scope!!
That is what I did by not buying a Rigol. Back when I evaluated them just before the DS1000Z series became available, I concluded they were best to avoid and various reports on the DS1000Z series have convinced me that I made the right decision.
But my curiosity in oscilloscope design and performance is enough that I want to know what is going on. The lack of documentation makes for a challenging puzzle.
I am not totally sure what the complaint is in this post, but I noticed that it was disclaimed as an old test. I think the complaint is the amplitude change when sin(x)/x is off on dots mode. So I made some animations of a current oscilloscope with the 2nd to latest firmware.
Amplitude should not be changing at different time/division settings, sample rates, or interpolation settings whether aliasing is present or not unless insufficient sample points are captured which is not the case here. Somewhere the vertical histogram of the signal is being corrupted although Rigol may not be using it for measurements which does not improve the situation.
I am not really surprised by this behavior in a DSO which makes measurements on the display record instead of the processing record. It is a major design compromise more suitable for a toy.