General > General Technical Chat
The Rigol DS1052E
<< < (61/166) > >>
dimlow:

--- Quote from: rossmoffett on March 18, 2010, 05:11:50 am ---Dave has taken a break, he posted a note in the "Announcements" forum saying he's going walkabout for a while.  I'm sure the deluge of e-mails and consistent output of quality video blogs has been taxing his time for a while now!

--- End quote ---

Yea, but that was on the 9th he has posted a new video blog on the 16th so i assumed he was back.
Mark_O:

--- Quote from: rossmoffett on March 18, 2010, 05:11:50 am ---You're saying it's 20% slower maximum, but that's actually the minimum.  The maximum would be 600 MS/s slower with both scopes in single channel operation, and 300 MS/s with both scopes in dual channel operation.  That's 50% and 50% maximum difference.  The minimum difference would be 20% with the DS1052E in dual-channel operation and the C model in single channel operation.
--- End quote ---

Are you talking about using LongMemory, or not?  It's hard to tell.  Personally, I use ShortMemory so infrequently, I hardly remember it's there.   :)  In ShortMemory mode it's 1000M/500MSa, vs. 400M/200MSa, as you say.  In LongMemory mode, it's 500M/250MSa, vs. 400M/200MSa.  That's what I was referring to.  My apologies if I wasn't clear about that.

It's possible you still think the 1000E series will do 500MSa in dual-channel mode, even with LongMemory.  That would be understandable, because that's what Rigol published in their Feb'09 manual for those scopes.  However, they corrected themselves in the Jun'09 edition, and show the correct 250MSa max-rate in Long dual-channel mode.  Try it yourself, and see where it maxes out in dual-channel LongMemory mode.


--- Quote ---I do say this makes a huge difference in performance, I operated them side by side to test my modifications and the noise in the C model made the readings far worse than even my 50 MHz unmodified channel.
--- End quote ---

Yes!, I was very interested in your personal evaluation of the two different models, because I have no E-series unit here to compare my older C-series with.  I wasn't calling your opinion into question at all... just asking because the spec differences are smaller than they initially appear to be.  Thanks for sharing your personal comparisons.

- Mark
alm:

--- Quote from: Mark_O on March 18, 2010, 04:01:35 am ---I'm surprised to hear that Dave made that claim.  Especially since it's not true.  You don't need 10-times oversampling to be able to accurately reconstruct a waveshape.  Assuming a gaussian-distribution (the normal case), you can exactly reconstruct any arbitrary waveform, using sin(x)/x with as little as 2.5x oversampling.  However, to accomplish that requires very good filters (i.e., expensive), and with the quality of filtering on most consumer-grade instruments, you actually need ~4x oversampling to achieve the same results.  (To get technical, it varies, depending on the instrument, from 3-5x... but 4x is a good rule-of-thumb.)

But 10x is really overkill.  Thus even the 400MSa on the earlier 100MHz 1000C-series was perfectly fine.  Based on your claim, they'd only be good to 40 MHz (and the 100 MHz-version of the 1000E would top out at 50 MHz).
--- End quote ---

One issue with applying the Nyquist-Shannon sampling theory to DSO's is that the theory talks about the total bandwidth of the signal. The bandwidth of an oscilloscope is the -3dB point. The bandwidth in the sampling theory is the highest frequency component. This would require a hard cut-off above 100MHz, as opposed to the Gaussian roll-off that's normal for oscilloscopes. Some of the high-end oscilloscopes, which do oversample less than 10x because they can't make the ADC's and memory fast enough, have a much sharper input filter (there's an Agilent appnote about this). Because your signal may have higher frequency components, you can't treat it like a simple Nyquist reconstruction, so more samples are useful.
Mark_O:
Ross,

I just went back and reviewed your original comment...
> We have 400 MS/s 100 MHz model C Rigol scopes at school... <

And your more recent:
> I do say this makes a huge difference in performance, I operated them side by side to test my modifications and the noise in the C model made the readings far worse than even my 50 MHz unmodified channel. <

You do realize that the residual noise displayed on the 100MHz model-C will always be higher than on your unmodified 50MHz model-E, simply due to the bandwidth differential?  All other things being equal, wider BW scopes will always exhibit higher apparent noise levels.  That's the nature of the beast, and what we have BW limiting, digital filtering options, averaging, and ET sampling for.  To reduce the contribution of noise, and see the signals underneath.

If the E-series do have less residual noise (which is possible, though the noise-samples Dave posted for his 1052E didn't look significantly different than what I was used to on my 1102CD), it's not a result of higher sampling rates.

Agilent did a paper on noise in digital scopes a while back, and it was quite illuminating.  Even very expensive scopes can exhibit surprisingly high levels of relative noise, under some circumstances as much as 50% of one vertical division on a peak-to-peak basis!  Turn on infinite Persistence, and things can get pretty ugly.  ;)

- Mark
Mark_O:

--- Quote from: alm on March 18, 2010, 10:06:09 am ---One issue with applying the Nyquist-Shannon sampling theory to DSO's is that the theory talks about the total bandwidth of the signal. The bandwidth of an oscilloscope is the -3dB point. The bandwidth in the sampling theory is the highest frequency component. This would require a hard cut-off above 100MHz, as opposed to the Gaussian roll-off that's normal for oscilloscopes. Some of the high-end oscilloscopes, which do oversample less than 10x because they can't make the ADC's and memory fast enough, have a much sharper input filter (there's an Agilent appnote about this). Because your signal may have higher frequency components, you can't treat it like a simple Nyquist reconstruction, so more samples are useful.

--- End quote ---

Alm, yes, that's correct.  I think the root cause for the confusion was a misinterpretation of what Dave "said".  I was talking about oversampling on the highest-frequency component of a signal.  What Dave, and Simon, and Ross were referring to was evidently the need to sample 10 times per-cycle.  And Dave's right on that... you need a decent number of samples, even with sin(x)/x reconstruction, to be able to capture the higher-order harmonics of a lower-frequency waveform.  No argument there.

- Mark
Navigation
Message Index
Next page
Previous page
There was an error while thanking
Thanking...

Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod