Author Topic: Anomalies and/or Bugs in/between the Rigol MSO/DS1000Z, MSO/DS2000A, MSO/DS4000  (Read 30456 times)

0 Members and 1 Guest are viewing this topic.

Offline rf-loop

  • Super Contributor
  • ***
  • Posts: 4064
  • Country: fi
  • Born in Finland with DLL21 in hand
Here are two more images that perhaps show the High Res recalculations more succinctly.

I'm now confused totally.

What you mean with these (and there is also more images in other message).

You have used normal mode and high-res mode.
I can see these images sample rate is 2GSa/s.
It is maximum speed of this scope ADC.

How it can do any High-Res type averaging with this speed?
Stupid question: What exatly is Rigol High-Res mode? Do it real High-Res mode at all or is it some kind of diplay cosmetic  make-up?
« Last Edit: November 23, 2014, 10:09:37 am by rf-loop »
I drive a LEC (low el. consumption) BEV car. Smoke exhaust pipes - go to museum. In Finland quite all electric power is made using nuclear, wind, solar and water.

Wises must compel the mad barbarians to stop their crimes against humanity. Where have the wises gone?
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16547
  • Country: us
  • DavidH
BTW, continuing our previous conversation from the other thread (Hi Res mode and Tektronix - deemed OT by OCs), I found some specifications from Tektronix as to how they map waveform values to the display (at least on the TDS XXXA series):

According to the specs, Normal mode maps 6.8 bits to 460px (i.e. ~4.128 pixels per value - some rather serious upscaling) - while Hi Res mode maps 9.7 bits to 460px (i.e. ~0.55px per value). From these numbers it appears as if you would only see a 1:1 mapping when Hi Res was equivalent to 8.85 bits of resolution, otherwise it would almost always be downscaling (unless I'm completely missing some salient point - which is totally possible  ;D ).

I am not sure what you found but I suspect the numbers, 6.8 bits and 9.7 bits, came from the ENOB (effective number of bits) specifications that Tektronix often includes or used to include.  They represent the dynamic range of the digitizer and have nothing to do with mapping ADC levels to the display.

You can find the same type of ENOB specifications in the datasheets for the ADCs Rigol is using but they represent ideal values; the real numbers are lower do to outside factors like clocking and noise.

Tektronix was pretty paranoid about using integer mappings between the ADC levels and the display.  This is why the 7854 uses 102.4 points per division so its 10 bit ADC maps evenly to 8 and 10 divisions exactly (Its 8 x 10 cm display is 1024 x 1024 points or 325 x 260 dpi so Apple must have invented it.  The extra 2 vertical divisions are overscanned.) and their later vector graphics DSOs with 25 or 50 points per division used overscan for the extra points if a power of 2 display record was used if they were displayed at all.  25 points per division and 8 divisions yields 200 points so the extra 56 level from the 8 bit display record are overscanned.  The same thing happens horizontally with power of 2 record lengths.

The TDS460A series specifications (the TDS500 series are the same) are very specific about using 50 pixels per division with a 8 x 10 graticule producing a 401 x 501 pixel graticule on a 480 x 640 display and mapping 25 8-bit levels per vertical division so 2 pixels per 8-bit level.  8 vertical divisions only yields 200 8-bit levels but Tektronix liked to keep clipping outside of the visible display.  The raster display and graticule provide no overscan like their earlier vector CRT displays.

Quote
Other than that, I have found no indication that the added resolution in the waveform record would be usable in real time. It would only be useful when the DSO was stopped - or for viewing/processing on an external device (which, granted, would still be a nice feature to have).

I am not sure why the documentation would discuss it.

I have used the added resolution supported by the waveform record in real time on the 2440 series and old TDS series DSOs.  That is how these oscilloscopes either scale vertically without changing the input sensitivity or provide volt per division settings higher than the highest input sensitivity.  Averaging or high resolution mode is needed to provide a better than 8 bit waveform record because if only 8 bit resolution is available, the quantization levels become evident.  I do not remember exactly what the TDS series does (I think it forces averaging or high resolution mode depending on the sample rate if you try to set a higher vertical sensitivity than the hardware supports on a live waveform) but the 2440 disallows vertical scaling in real time if averaging is not used.  When stopped it allows it on any saved waveform which is how the 8-bit quantization can be displayed like many of the Rigol photos in this discussion.

The automatic measurements make use of the high resolution waveform record as well while the Rigols apparently make measurements on the display record yielding some odd behavior and limitations.

I do not have a TDS series DSO to play with anymore but what I remember is that high resolution mode did the same thing and works as expected.  If the waveform is scaled in real time without adjusting the vertical sensitivity or when saved, then the full >8 bit waveform record was used to generate the display.
 

Offline electronic_eel

  • Regular Contributor
  • *
  • Posts: 201
Here is another issue I found on my MSO4000 (firmware 00.02.02.SP1, hw 1.3):

Once the probe ratio is set higher (e.g. 100x), the measure functions on the math channel return totally bogus values. I found this while measuring on mains with my 100x probe.

This is the most easy way I found to reproduce this:

- Connect the scope to a function generator outputting a sine signal with e.g. 2Vpp @ 1KHz, use e.g. BNC and not a probe with automatic ratio detection (like the probes supplied with the scope)
- Go into the channel menu, probe and select a 100x ratio
- Activate the math channel with an advanced expression and enter CH1*CH1
- Go to the measure menu and select Math as source
- Add some measure values, e.g. Vmin and Vmax
- Go back to the math menu and scale the math channel to 5.00KU
- Move the position of the math channel on the screen up and down
- Watch the Vmin and Vmax measures for the math channel reporting totally bogus values (e.g. negative), depending on where on the screen the math channel is positioned.

I attached some examples from my scope, just compare the purple Max, Min and Avg values.

I can only observe this with a high probe ratio setting so I guess this is some kind of overflow.

Could somebody check if this problem is also occuring on the 2000A and 1000Z series?

Thanks.
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
it's just calculating how to make that thick line thinner, but nothing is really gained with regard to sampling, you could also make that line thinner in your head with same results, you didn't really gain any more adc vertical resolution.

so it's really cheating, and the waveform shown is just a guess....

if you had a scope with more bits in adc, it would show something else, something more genuine...this is a play with 8bit data....
This is just successive sample averaging producing more effective bits - the formula and method for doing it is well-documented. I'm not sure what you expect to happen when you press Hi Res - would the display grow more pixels? Even with a bigger ADC and higher resolution display, when you zoom in far enough, you see jagged edges.
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
I'm trying to replicate this on the DS1000Z, let me know if i'm doing anything wrong.

As I mentioned in a previous message, this is impossible on a DS1000Z. It doesn't recalculate the data, so you're just enlarging the 8-bit record.
 

Offline eV1Te

  • Regular Contributor
  • *
  • Posts: 186
  • Country: se
  • Your trusted friend in science!
    • richardandersson.net
I'm trying to replicate this on the DS1000Z, let me know if i'm doing anything wrong.

As I mentioned in a previous message, this is impossible on a DS1000Z. It doesn't recalculate the data, so you're just enlarging the 8-bit record.

I liked your demonstration of how the DS2000 recalculates the data. I completely understand that the DS1000z does not do this.

However what I am trying to say is that the DS1000z never displays anything above 8 bits, even while running. For example in the 1 mV and 500 uV range, the scope amplifies the signal in software hence it is possible to see bit steps clearly while running or stopping.

The hires mode or the averaging mode does not change the bit steps at all, and there is no difference when changing vertical after stopping either.

Hence the conclusion I can draw is that the Hires mode on the DS1000z is broken, or it never existed in the first place, and should thus be removed from Rigols advertisement and the specification for the instrument.

If this is incorrect and there is some settings that enables more than 8 bits, please correct me as I have spend hours trying to find such a condition (I would find it useful for low noise measurements).

I now have to resort to exporting the entire 24 MB of data in order to convert it to 140 kB of 12 bit data on my computer (post dithering), which takes significant amount of time to write to a USB stick.
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
I am not sure what you found but I suspect the numbers, 6.8 bits and 9.7 bits, came from the ENOB (effective number of bits) specifications that Tektronix often includes or used to include.  They represent the dynamic range of the digitizer and have nothing to do with mapping ADC levels to the display

It certainly has to do with mapping - the specification is very specific about the divisions - but it's true I had forgotten to take into account that it was effective bits, which makes much more sense in the calculations (with more tidy numbers). So, yes, that means it's 2px per level in 8-bit mode, 9-bits would map 1:1, and bit values above that are downscaled (un-zoomed).

Quote
The TDS460A series specifications (the TDS500 series are the same) are very specific about using 50 pixels per division with a 8 x 10 graticule producing a 401 x 501 pixel graticule on a 480 x 640 display and mapping 25 8-bit levels per vertical division so 2 pixels per 8-bit level.  8 vertical divisions only yields 200 8-bit levels but Tektronix liked to keep clipping outside of the visible display.

The DS2000 uses the same mapping.

Quote
I am not sure why the documentation would discuss it.

Well, ok then - I'll leave it there... and just assume those expensive DSOs can do it. :)

Quote
...while the Rigols apparently make measurements on the display record...

You keep mentioning this as if it's an anomaly in low-cost scopes. Instead, could you perhaps come up with some other low-cost brands which don't do this?  ;)  Agilent X-series? Nope, they do it too.

You have to keep in mind - Agilent was Rigol's mentor. Much of the behavior and feature set of Rigol's DSOs, for good or ill, is learned or copied from Agilent.
« Last Edit: November 24, 2014, 08:13:29 am by marmad »
 

Offline i4004

  • Regular Contributor
  • *
  • Posts: 71
  • Country: hr


The hires mode or the averaging mode does not change the bit steps at all, and there is no difference when changing vertical after stopping either.

Hence the conclusion I can draw is that the Hires mode on the DS1000z is broken, or it never existed in the first place, and should thus be removed from Rigols advertisement and the specification for the instrument.


even 2k's "hires" mode is nothing usefull really, while on 1000z it doesn't even exist.
how to test it?
do you see any meaningful change when you turn it on? if not, then it doesn't exist.

but again, even on 2k this is just a gimmick....just like the perfectly still line on grounded input...that doesn't exist, why would you do it? for whom?
this thread should be called "rigol gimmicks/tricks".

edit/ok, it does something on 2k, not totally useless, looking at lastest pix from marmad.

This is just successive sample averaging producing more effective bits - the formula and method for doing it is well-documented. I'm not sure what you expect to happen when you press Hi Res - would the display grow more pixels? Even with a bigger ADC and higher resolution display, when you zoom in far enough, you see jagged edges.

well i would expect that in 21st century we've moved away from number of bits we first saw in 1972. for starters.
http://en.wikipedia.org/wiki/8-bit#Notable_8-bit_CPUs

8bit adcs are even older
http://www.analog.com/library/analogdialogue/archives/39-06/chapter%201%20data%20converter%20history%20f.pdf
page 23 onwards.

offcourse more bits have their limits too (12bits requires 8k display, i didn't kid much..heh), but they do exponentially grow, as you know...1024 levels of 10bit would be good enough.
that is 5x  more zoom. oh yeah, what is "ultra zoom" on rigol?
hehe...why do i even ask?  :scared:
is it by any chance just the horizontal zoom?  >:(
did you ever hear anybody call "zoom" enlarging just one axis of the image?
meh....it's more advertising than technology....simillar to the way new flat tvs are marketed, with whole lots of interesting tech terms that in the end bring nothing or next to nothing.
there were 200hz panels, now 400hz....   :blah:

btw. is anybody making 10bit adcs that output whole 1k resolution if not via its display then via pc interface?

but...hehm..now i know why these things are relatively cheap...

i'm familiar with these concepts from video processing....your pdf mentions both sample and capture averaging...i would call first spatial averaging, and latter temporal averaging....
i presume rigol2k has the former, and i know the latter would be more usefull to average the noise out...i'm also familiar with dithering of images...

but either way it's just a trick to make a waveform look a bit better in a similar way that adding noise to video makes lores video easier on the eyes, but that still has nothing to do with hires video... no extra data was introduced....you just played a bit with existing data to reduce noise....
and video nr filters are always a compromise, you lose some and gain some...it's same here...
talking about video again: i would prefer to not use noise reduction at all because when you do it poorly it's worse than not doing it at all...and you can't really do it properly.
and i see same goes for 'hires' concept in scopes: why don't they just call it noise reduction?
or "virtual resolution improvement"....or both....there really is no need to lie about it.....
it's not hires.

the bottom point is that you can play with software on small amount of bits, OR do it properly and don't play at all and just improve adc ie employ 10bit adcs. map every level to every pixel, at least via pc interface...for scope's own display just do a bicubic resize of hires image...it'll be allright.
waaaaay better than 200 freakin vertical pixels/levels.

to that extent what's the price difference between 10 and 8bit adcs that we're discussing here?

btw. i don't thin it makes any sense to calculate effective number of bits the way you do it, to end up with, for example, "6.8 bits and 9.7 bits" numbers, because bits are bits, they don't have decimal point. because the whole thing is kinda like those "ghetto blaster" "boom boxes" which would state ridiculous 'pmpo' power outputs: it's  just a gimick in attempt to sell something.
it's not really there and it won't really help the measuring.

dave said it well:
https://www.eevblog.com/forum/blog/eevblog-683-rigol-ds1000z-ds2000-oscilloscope-jitter-problems/msg554787/?topicseen#msg554787
although i don't agree that these days 8bit adc is all we should expect, even for such a low price!
« Last Edit: November 23, 2014, 07:27:05 pm by i4004 »
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
However what I am trying to say is that the DS1000z never displays anything above 8 bits, even while running. For example in the 1 mV and 500 uV range, the scope amplifies the signal in software hence it is possible to see bit steps clearly while running or stopping.

The hires mode or the averaging mode does not change the bit steps at all, and there is no difference when changing vertical after stopping either.

Hence the conclusion I can draw is that the Hires mode on the DS1000z is broken, or it never existed in the first place, and should thus be removed from Rigols advertisement and the specification for the instrument.

If this is incorrect and there is some settings that enables more than 8 bits, please correct me as I have spend hours trying to find such a condition (I would find it useful for low noise measurements).
I've written and posted a lot of information about this in other threads, but I thought I would try to gather together some of it here for clarity:

There at least three big operational differences that I've noticed in the implementation of High Res on the DS1000Z as compared to the DS2000:

1) High Res can not be applied to a captured waveform (stopped DSO) on the DS1000Z; you can only use it when running.

2) High Res on the DS1000Z is destructive - i.e. the original samples are unavailable. On the DS2000, you can turn High Res back OFF after capturing a waveform; on the DS1000Z, once the DSO is stopped, you can't recover the original waveform to the display - although I haven't tested saving sample memory to USB or PC, so perhaps the original samples ave available that way.

3) Most importantly: on the DS1000Z, High Res mode appears as if it might be implemented after display decimation - or at least, in some kind of strange combination with it - resulting in some aliasing and other artifacts. The DS2000 follows the mathematical model of successive sample averaging quite well, while the DS1000Z seems to be doing things poorly in comparison.

The DS1000Z is doing some averaging (as shown in the images below), but it's not clear to me exactly when/how well it works for each possible memory depth setting (I haven't had time for in-depth tests yet). The 12k and 120k setting seem to do some averaging in the us/div range, while the 1.2M and higher settings seem to do almost nothing unless in the ms/div and slower ranges (perhaps higher memory depths overwhelm the ability of the engine to keep up unless at slower sample rates). There also seems to be a problem with higher frequencies leaking past the stopband sometimes, as shown in the image of the sweep roll-off @ 12k below.

The simplest way to get a picture of what the DSO is doing while in High Res is to test the bandwidth. As I've mentioned many time, successive sample averaging (boxcar filtering) is a low pass filter with a sharp roll-off to a null, then subsequent peaks and nulls through the stopband (as shown in this image):




To illustrate the effects of this, I used a repeated sine-sweep pattern - first to my DS2000 in Normal mode:




When I turn on High Res (@ 1.4M), you see the sweep BW roll-off, just as you should:




When I switch the memory depth to 14k, the roll-off gets sharper because of the lower sample rate (thus lower bandwidth):




Here is the same signal to the MSO1000Z:




When I turn on High Res (@ 12M), there is very little change:




When I switch the memory depth to 120k, there is a roll-off, but fairly gentle (perhaps every 4 samples averaged):




When I switch the memory depth to 12k, there is a sharper roll-off, but then a rise again - almost like a notch filter (clearly incorrect behavior):





And here is a sine wave FFT - in Normal, then High Res (DS2000, then DS1000Z):










« Last Edit: November 23, 2014, 07:34:51 pm by marmad »
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
but again, even on 2k this is just a gimmick....just like the perfectly still line on grounded input...that doesn't exist, why would you do it? for whom?
this thread should be called "rigol gimmicks/tricks".

Yes, you've stated your opinion quite clearly a number of times now - regardless of the fact that Rigol didn't invent any of these "gimmicks/tricks".

But I'm not quite clear what your dog in this race is: did you buy one of these scopes without reading the specs and expect a 12-bit ADC? Or just here to rant?  ;D
 

Offline i4004

  • Regular Contributor
  • *
  • Posts: 71
  • Country: hr
but if they copied the tricks did they at least pay copyright fees?  ;D

Quote
But I'm not quite clear what your dog in this race is

nah, just slightly surprised we're still in 8bit time....
i still like my 256 level(maybe) 200 line ( :P) scope!  :-DD  ;)

and my dog is that rigol should come clear about the stuff it does...i mean "up to 12bits"?
come on...and you have simillar specs for both 1 and 2k z scopes....

1k spec sheet says
Quote
High Resolution 12 bit (max)
and 2k
Quote
High Resolution 12 bits of resolution when ?5 ?s/div @ 1 GSa/s (or ?10 ?s/div @ 500 MSa/s).
« Last Edit: November 23, 2014, 07:44:24 pm by i4004 »
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Here is another issue I found on my MSO4000 (firmware 00.02.02.SP1, hw 1.3):

Could somebody check if this problem is also occuring on the 2000A and 1000Z series?

Thanks.

@electronic_eel: Hi! I confirmed that the bug DOES exist in the DS2000A, but it DOES NOT exist on the MSO1000Z (you can actually see the code recalculate and replace the incorrect values after moving the offset). So they've fixed it in their latest model - now we've just got to get them to port the fix to the DS2000A/DS4000.
 

Offline electronic_eel

  • Regular Contributor
  • *
  • Posts: 201
@electronic_eel: Hi! I confirmed that the bug DOES exist in the DS2000A, but it DOES NOT exist on the MSO1000Z
Thanks for checking this, I thought nobody's gonna read / check this after all this bit resolution / high res talk   :-*

(you can actually see the code recalculate and replace the incorrect values after moving the offset). So they've fixed it in their latest model - now we've just got to get them to port the fix to the DS2000A/DS4000.
So on the 1000Z the scope displays the bogus values first and after a second or similar the values get overwritten with correct ones?

If that is the case I'm not sure if it is really an intentional fix or they just restructured the measure code a bit and this is just a byproduct. I guess the statistics still get completely garbled up once the bogus values show up for a short time? If yes then I'd still count this a bug...
 

Offline eV1Te

  • Regular Contributor
  • *
  • Posts: 186
  • Country: se
  • Your trusted friend in science!
    • richardandersson.net

I've written and posted a lot of information about this in other threads, but I thought I would try to gather together some of it here for clarity:

There at least three big operational differences that I've noticed in the implementation of High Res on the DS1000Z as compared to the DS2000:

1) High Res can not be applied to a captured waveform (stopped DSO) on the DS1000Z; you can only use it when running.

2) High Res on the DS1000Z is destructive - i.e. the original samples are unavailable. On the DS2000, you can turn High Res back OFF after capturing a waveform; on the DS1000Z, once the DSO is stopped, you can't recover the original waveform to the display - although I haven't tested saving sample memory to USB or PC, so perhaps the original samples ave available that way.

3) Most importantly: on the DS1000Z, High Res mode appears as if it might be implemented after display decimation - or at least, in some kind of strange combination with it - resulting in some aliasing and other artifacts. The DS2000 follows the mathematical model of successive sample averaging quite well, while the DS1000Z seems to be doing things poorly in comparison.


That is a nice selection of measurements, the BW definitely shows that it does some boxcar averaging.

So am I right in assuming that the reason why we both see a reduction in BW and also no additional bits of resolution, is probably due to the fact that the DS1000z boxcar averaging is done all with 8bit integer math? That could explain all the observations we have made, and it also sounds reasonable since it has less processing power than the DS2000.
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Thanks for checking this, I thought nobody's gonna read / check this after all this bit resolution / high res talk   :-*

No problem at all. I'd read your post, but just didn't get around to running the tests until I responded to some other posts  ;)

So on the 1000Z the scope displays the bogus values first and after a second or similar the values get overwritten with correct ones?

More like a fraction (0.25) of a second.

Quote
If that is the case I'm not sure if it is really an intentional fix or they just restructured the measure code a bit and this is just a byproduct. I guess the statistics still get completely garbled up once the bogus values show up for a short time? If yes then I'd still count this a bug...

I would say that someone at Rigol noticed that changing the offset screwed up the Math measurements, and they tacked on a routine to update them correctly.

 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
So am I right in assuming that the reason why we both see a reduction in BW and also no additional bits of resolution, is probably due to the fact that the DS1000z boxcar averaging is done all with 8bit integer math?

That's possible - although the more tests I do, the more I start to become convinced that the High Res feature is really not implemented correctly on the DS1000Z - either by fault or design. Often the effect is almost exactly the same as normal Average, when it really shouldn't be.

Take a look at the following 3 images; I've sent a 1MHz sine wave to both scopes (BTW, it doesn't matter that the DS2000 and DS1000Z are using different timebases - what matters is that the sample rates are as similar as possible):

Given a sample rate of 20MSa/s on the DS2000 with 256-sample averaging, the -3db bandwidth should be 34kHz - so a 1MHz signal should be virtually invisible (which it is). Compare that to the DS1000Z with a similar sample rate (25MSa/s) and 1MHz input signal. The resulting High Res display is almost exactly the same as you get using Average - which is exhibiting aliasing.
« Last Edit: November 23, 2014, 11:09:29 pm by marmad »
 

Offline TMM

  • Frequent Contributor
  • **
  • Posts: 471
  • Country: au
That's possible - although the more tests I do, the more I start to become convinced that the High Res feature is really not implemented correctly on the DS1000Z - either by fault or design. Often the effect is almost exactly the same as normal Average, when it really shouldn't be.
You've finally seen the light eh?
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
That's possible - although the more tests I do, the more I start to become convinced that the High Res feature is really not implemented correctly on the DS1000Z - either by fault or design. Often the effect is almost exactly the same as normal Average, when it really shouldn't be.
You've finally seen the light eh?
I've been posting there was a problem with High Res on the DS1000Z series for about a month now.

But as I showed with the sweep images above, the DSO is clearly doing some boxcar averaging (which I believe you asked me post actual documentation to prove before - so there you go :) ). I actually think the problem might be that it's doing some display decimation before the averaging (at least with large sample sizes) - which would be a terrible idea, but would save time and processing power. But in any case, the only results I've been able to see so far were either correct - but with very small averaging sizes (perhaps 4 - 16 samples) or else large averaging sizes with bad artifacts (like aliasing).
« Last Edit: November 24, 2014, 04:28:05 am by marmad »
 

Offline TMM

  • Frequent Contributor
  • **
  • Posts: 471
  • Country: au
Another thing to note is that on the 2K, when you enable high-res the intensity grading is disabled?

On the 1K the intensity grading runs even in high res. I wonder if that could explain why the data seems to still be 8bit (they round the high res data back to 8bit in order for the intensity grading to work?). It doesn't explain why the high freq rolloff is different to the 2K though.
« Last Edit: November 24, 2014, 07:15:12 am by TMM »
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
Another thing to note is that on the 2K, when you enable high-res the intensity grading is disabled?

Rigol coded the intensity-grading on the DS2000 in a fairly strict, non-adaptive way - which can sometimes be good and sometimes bad. Since the grading is based on both complete waveform update hits and overlapping sample hits (e.g. 1.4M samples to 700 horizontal display px), when High Res is turned on, all of a sudden large groups of samples are averaged to single points - i.e. single hits - so the intensity grading is reduced immediately. You can get some grading (depending on the memory depth and timebase), but it's much reduced from Normal mode.
« Last Edit: November 24, 2014, 08:20:23 am by marmad »
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16547
  • Country: us
  • DavidH
nah, just slightly surprised we're still in 8bit time....

There is a good reason for this and it is also why analog oscilloscopes support a maximum sensitivity for a given bandwidth; the total noise from the vertical input amplifier over a wide bandwidth is considerable so in practice greater than about a 2mV/div or 5mV/div sensitivity on a 100 MHz oscilloscope is not useful.

If you want greater useful sensitivity, then the input bandwidth has to be limited and some analog oscilloscopes do this automatically.  DSOs can provide greater sensitivity and resolution through averaging or high resolution mode which both remove noise.  A higher resolution ADC is not needed and does nothing in this case.

Another problem with higher resolution is that fast settling time to 8 bits is difficult enough to achieve.  If 12 bit resolution is used, it does not reveal anything new on fast signals because they will not settle any faster.  Improvements can be made in this area but they are difficult.

12 bit and higher resolution oscilloscopes do exist, even the Tektronix 7854 is 10 bits and has a 10 bit display to support that, but they are still limited by noise and settling time so the extra resolution is not useful in a general way.  Sampling oscilloscopes often have much higher than 8 bit resolution.
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16547
  • Country: us
  • DavidH
EEVblog ate my original reply with a redirect so what is below is abbreviated.

The TDS460A series specifications (the TDS500 series are the same) are very specific about using 50 pixels per division with a 8 x 10 graticule producing a 401 x 501 pixel graticule on a 480 x 640 display and mapping 25 8-bit levels per vertical division so 2 pixels per 8-bit level.  8 vertical divisions only yields 200 8-bit levels but Tektronix liked to keep clipping outside of the visible display.

The DS2000 uses the same mapping.

An integer mapping prevents or minimizes aliasing from the display.  The TDS documentation discusses how those DSOs apply interpolation when this is not the case.  Many Tektronix models have a mode where the actual sample points are highlighted to distinguish them from interpolated points.

Quote
Quote
I am not sure why the documentation would discuss it.

Well, ok then - I'll leave it there... and just assume those expensive DSOs can do it. :)

The DSOs I have certainly do but they date from a time when all DSOs were expensive.

While testing I found something interesting on the Tektronix 2232; it reports the amount of memory each saved waveform takes.  1k record length waveforms take 1kbyte and 4k record length waveforms take 4kbytes which is consistent with 8 bit acquisition records and 8 bit waveform records.  When averaging is used, the waveform records double in size taking 2kbytes or 8kbytes.

Quote
Quote
...while the Rigols apparently make measurements on the display record...

You keep mentioning this as if it's an anomaly in low-cost scopes. Instead, could you perhaps come up with some other low-cost brands which don't do this?  ;)  Agilent X-series? Nope, they do it too.

The Rigols were the first oscilloscopes I considered when looking for a new DSO and I expected better.  Making measurements (and protocol decoding) directly from the display is something I would expect in a toy no matter who makes it.

There is a caveat with the above however.  If DPO type operation produces the display record during acquisition which is the case with older DPOs, then there is no choice if measurements are made because the display record is the only thing available.  That explains limitations on oscilloscopes where DPO operation is used and may explain some of the limitations on the Rigol Ultravision oscilloscopes where DPO operation is fixed.

Quote
You have to keep in mind - Agilent was Rigol's mentor. Much of the behavior and feature set of Rigol's DSOs, for good or ill, is learned or copied from Agilent.

Current DSOs from Agilent appear to be better than the ones from Tektronix but I would carefully evaluate either for this type of problem.
 

Offline Teneyes

  • Frequent Contributor
  • **
  • Posts: 498
  • Country: ca
I started new thread just for "Rigol DSO Trigger position Offset w/ AC coupled trggering , DS2000, MSO1000"
and I will collect confirmations of which models have this bug and return results to this thread later.
New Thread Here
« Last Edit: November 26, 2014, 05:50:22 pm by Teneyes »
IiIiIiIiIi  --  curiosity killed the cat but, satisfaction brought it back
 

Offline i4004

  • Regular Contributor
  • *
  • Posts: 71
  • Country: hr
btw. anybody compared rigol's 'video' tigger to other dso's simillar triggering items?
on pal or ntsc...
would you expect perfectly still waveform even wothout stopping the scope?
 

Offline marmadTopic starter

  • Super Contributor
  • ***
  • Posts: 2979
  • Country: aq
    • DaysAlive
The Rigols were the first oscilloscopes I considered when looking for a new DSO and I expected better.  Making measurements (and protocol decoding) directly from the display is something I would expect in a toy no matter who makes it.

It's SOP below a certain price point, so I guess it's just a market full of toys. Too bad for you; you'll likely never be able to buy a low-cost DSO.

Personally, I wouldn't count on an oscilloscope for making precise measurements (a habit I learned from trying to count graticule tick marks on analog scopes) - just as I wouldn't count on an MSO for complex logic analyzing - feels rather like using a screwdriver to hammer a nail.
« Last Edit: November 25, 2014, 01:17:48 am by marmad »
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf