Oscilloscopes - What happened to equivalent-time sampling?

[David Hess]

That is a great example.  The probes should have a better response than that but maybe the extra length of the female-to-female BNC adapter explains it.

[borjam]

Thanks for all the reply's!
<snipped>

And of course I could do another experiment using just dots mode, as far as I understand, when using dot's mode, there's no interpolation involved.

Correct, dots are the acquisition sample points and if you worked with just enough to form a continuous trace you could accurately compare results against vector traces and sin(x)/x interpolation.

Beware, not in the Rigol DS1000Z series. Dots are, confusingly, post interpolation as well.

[tggzzz]

Thanks for all the reply's!
<snipped>

And of course I could do another experiment using just dots mode, as far as I understand, when using dot's mode, there's no interpolation involved.

Correct, dots are the acquisition sample points and if you worked with just enough to form a continuous trace you could accurately compare results against vector traces and sin(x)/x interpolation.

Beware, not in the Rigol DS1000Z series. Dots are, confusingly, post interpolation as well.

Really? That would be seriously annoying.

I hate "helpful" instruments that prevent me observing what they are measuring. One example is the fuel gauge in a car.

It used to be that the displayed value varied according to the angle the sender was at relative to the car's body. OK, that showed a different value when I was on a hill, but when I was on the flat it returned to the standard value. I could interpret it and know what it was (and wasn't) telling me.[1]

Now the displayed value is processesd by a computer, and some software implementor or marketeer decided that wasn't good enough. Now, when I park on a hill and return, about 1/8 of a tank of fuel has apparently evaporated - and it doesn't reappear when I'm back on a horizontal road. It does reappear after I've been parked on the flat for more than about an hour.

I can interpret and understand reality. I can't double guess what a software engineer wants me to think reality is.

[1] even when, after I finally got around to adding a fuel gauge to one car, there was no smoothing so it wobbled violently between quarter and half full - and the meter was calibrated in roentgens/hour.

[borjam]

Beware, not in the Rigol DS1000Z series. Dots are, confusingly, post interpolation as well.

Really? That would be seriously annoying.

Really. I though I had got insane when I tried Leo Bodnar's pulse generator.

I own both a Siglent SDS1202X-E and a Rigol DS1074Z. The Siglent did the right thing. The Rigol still added some interpolation in dots mode. I can try again and post some screenshots.

And yes, dots should be samples, pure and simple.

[Wolfgang]

To me teasing a 100MHz scope with 3.5ns risetime with a pulse generator with 100ps or less is a *bit* of an academic exercise. Why do this ?

[dzseki]

To me teasing a 100MHz scope with 3.5ns risetime with a pulse generator with 100ps or less is a *bit* of an academic exercise. Why do this ?

I don't think so, back then everyone had a "different" understanding on scope bandwidth. You may know that some scopes were somewhat underspecified in terms of bandwidth, but this can't be generalized of course, so it is not a bad idea to check what you have.

[MrW0lf]

That is a great example.  The probes should have a better response than that but maybe the extra length of the female-to-female BNC adapter explains it.

Checked: Indeed probe is ok. Direct line probe - FF adapter - pulser is also pretty good. So it must be some effects due other components in setup, since even reconfiguring and connecting probes directly to T's did not change much.

[borjam]

To me teasing a 100MHz scope with 3.5ns risetime with a pulse generator with 100ps or less is a *bit* of an academic exercise. Why do this ?

Maybe a bute or even a brute

Note that I don't own a similar pulse generator, so it was 100 ps or nothing.

Nevertheless I learned something important about the DS1000Z, it interpolates samples in dot mode.

[Wolfgang]

When you compare the gain in precision when you change from a 100MHz scope to a 200MHz to the extra information you gain by ETS its very clear that it *is* academic.

Rule of thumb: When making risetime measurements, the scope risetime should best be an order of magnitude below the DUT risetime (or the bandwidth 10 times as much).
Even this will get you just the first 4 harmonics of a square wave, you can image how square it would look then. Risetimes add in a root sum squares fashion, so instrument and probe risetimes and its uncertainties make measurements close to bandwidth limits very inaccurate. Why do this ?

[borjam]

Even this will get you just the first 4 harmonics of a square wave, you can image how square it would look then. Risetimes add in a root sum squares fashion, so instrument and probe risetimes and its uncertainties make measurements close to bandwidth limits very inaccurate. Why do this ?

The pulse generator has a male BNC connector. It was connected directly to the scope input via a 50 ohm terminator. No probes.

[Wolfgang]

Even without probes (I distrust them anyway for all high speed measurements) the risetime problem stays the same. I suggest that you at least verify your setup with a trusty scope with a large bandwidth (1GHz up), then you can believe whats on the screen.

Of course, having fun when playing around is always OK 

[dzseki]

Rule of thumb: When making risetime measurements, the scope risetime should best be an order of magnitude below the DUT risetime (or the bandwidth 10 times as much).
Even this will get you just the first 4 harmonics of a square wave, you can image how square it would look then. Risetimes add in a root sum squares fashion, so instrument and probe risetimes and its uncertainties make measurements close to bandwidth limits very inaccurate. Why do this ?

But beyond 100MHz source signal you'll have hard time to find an affordable oscilloscope to poke with... So you feel like you want to correct the seen rise time with the known constants (like the scope's self rise time, rise time of probing) while this is not elegant, but is often practical...
Not that I'd use a Rigol DS1000 for high frequency measurements, but I don't think either it is overkill to characterize it's bandwidth. And for that the fastest pulse risetime is the most practical, as always.

[Wolfgang]

No need to buy a multi-GHz beast (see TEA thread), but just try it out on one you have access to. 

[MrW0lf]

You may know that some scopes were somewhat underspecified in terms of bandwidth, but this can't be generalized of course, so it is not a bad idea to check what you have.

Still are. My 2048B has about 140MHz analog (~2.4ns RT) confirmed by various methods. All other properly designed 100MHz scopes should be about the same.

On the subject of "pointlessness" of measurements at bw limit: Exact same setup (incl probes) connected to 500MHz scope. This time left window is 50GSa/s ETS all channels and right window 1.25GSa/s RTS all channels.
Of course scope inputs itself may change picture a little, but what can observe?  Somewhat more small details. But did 100MHz scope running in ETS mode miss important details - no! Just enough info to see ringing and seek the source in physical setup. On 4 channels at same time! Absolutely not possible at 250MSa/s.
Also can notice that even on fairly fast scope running 4x1.25GSa/s averaging RTS trace may throw off timing. Both cursor and auto-measurments are spot on in ETS window (signal is exactly 20MHz) while tiny bit off in RTS one. Same can be said for signal traces. So overall - excessive sample rates rock. Excellent helper for shared ADC scopes.

[tggzzz]

When you compare the gain in precision when you change from a 100MHz scope to a 200MHz to the extra information you gain by ETS its very clear that it *is* academic.

Whether or not it is academic depends on the sampling rate.

Example 1: the first digitising scopes I used professionally were two man lift boat anchors, with a 1GHz bandwidth and 25MS/s (or was if 40MS/s; I forget).

Example 2: I have a sampling scope with a 50ps risetime source and a guaranteed 140ps sggregate risetime. The sampling rate is <<100kS/s.

[David Hess]

To me teasing a 100MHz scope with 3.5ns risetime with a pulse generator with 100ps or less is a *bit* of an academic exercise. Why do this ?

It is routine for transient response testing and calibration of the oscilloscope.

It can also be used to verify the measurement setup including the probes.  I also regularly do this to test RF attenuators within the bandwidth of the oscilloscope since I lack a suitable VNA.

That is a great example.  The probes should have a better response than that but maybe the extra length of the female-to-female BNC adapter explains it.

Checked: Indeed probe is ok. Direct line probe - FF adapter - pulser is also pretty good. So it must be some effects due other components in setup, since even reconfiguring and connecting probes directly to T's did not change much.

I have used basically the same test setup before but with female BNC-T adapters so the x10 probe plugged into the 50 ohm transmission line without the extra length of the BNC gender changer and the probe response was practically identical to the oscilloscope's calibrated input response.

For higher frequencies where the extra length of the probe tip adapter and BNC-T would cause problems, they used to make special in-line probe tip adapters which were like a BNC-T except the probe plugged directly into it.  These were useful up to at least 1 GHz if an active probe was used.

When you compare the gain in precision when you change from a 100MHz scope to a 200MHz to the extra information you gain by ETS its very clear that it *is* academic.

A higher bandwidth oscilloscope will of course provide a more accurate measurement at higher frequencies but that by itself is not sufficient if the sample rate is not high enough even with averaging.  ETS provides an order of magnitude increase in sample rate where real time sampling is not required.

Tektronix has been providing a lot of application notes talking about the advantages of real time sample rates with averaging for making accurate rise and fall time measurements but they leave out that this does not provide the accurate transient response measurements that ETS provides through much higher sample rates.  I think this is just a case of marketing what you have and concealing its disadvantages.

[rf-loop]

Beware, not in the Rigol DS1000Z series. Dots are, confusingly, post interpolation as well.

Really? That would be seriously annoying.

Really. I though I had got insane when I tried Leo Bodnar's pulse generator.

I own both a Siglent SDS1202X-E and a Rigol DS1074Z. The Siglent did the right thing. The Rigol still added some interpolation in dots mode. I can try again and post some screenshots.

And yes, dots should be samples, pure and simple.

Here is some findings from year 2014 about Rigol Sin(x)/x mess.  DS1074Z.
I have not anymore Rigol so I can not repeat tests using current FW.

In this thread I show what kind of mistake Rigol do with with Sin(x)/x interpolation... oh and not only this alone.

One main problem is that Rigol destroy real sample data what Siglent do not.

there is enough test images / data what show this error so that this time I can say they do wrong, totally wrong. And still Rigol engineer do not mind or even understand what goes wrong with method they use.


More fun that this nearly same debate was here in forum long long time ago when Rigol DS1000E ancient model was just new. I show this same error! Also at this time EEVblog Rigol fan club did not understand it at all. It feels nearly like  "it must be Right because Rigol did it"  Only rolling eyes and "so what".. who cares. Why repair errors if product sell well. (omg)
This time I try contact Rigol Beijing. Not any single answer. Like talking to walls at this time.

One most clear error is that they do not interpolate trace line via true sample points! THis is so bad error that I can ask what are they smoking in Rigol or do they have been never in any basic school for even know what is Sin(x)/x interpolation. They break-violate first rule. They do not respect true sample points and produce Sinc via true sample points. (exept if it is after years finally repaired in some new FW what I have not seen irl how it works.

There is also displayed how it need right so that do not violate this one most important rule. Interpolate via true data points.

Also here is Mechatrommer one comment from 2012
I can not find very old thread where it was originally.

[David Hess]

One most clear error is that they do not interpolate trace line via true sample points! THis is so bad error that I can ask what are they smoking in Rigol or do they have been never in any basic school for even know what is Sin(x)/x interpolation. They break-violate first rule. They do not respect true sample points and produce Sinc via true sample points. (exept if it is after years finally repaired in some new FW what I have not seen irl how it works.

I suspect the reason for this is linked with Rigol's use of an 8-bit display record which also results in averaging only producing an 8-bit result.  If the display record is only 8-bits, then maybe all of the processing is also limited to 8 bits so their sin(x)/x interpolation algorithms suffers from excessive truncation errors.

This design decision could be an optimization for greater speed at the expense of accuracy.