REVIEW - Siglent SDS2000 series - A Comparison of Features with Rigol DS2000

[marmad]

I too, do not have the expected result.
We will ask Siglent if High Res is fully functional.

Yes, it's just successive sample averaging - and the by-product is basically a low-pass comb filter.

[Galaxyrise]

From the manual:

High Resolution
This is generally used when the sample rate of the digital converter is higher than the storage rate of the acquisition memory.

Since 2GSamp/s at 200uS/div * 14 div = memory depth (5.6Mpt), maybe it doesn't high res?

[marmad]

Since 2GSamp/s at 200uS/div * 14 div = memory depth (5.6Mpt), maybe it doesn't high res?

The 200us image was just an example. It doesn't do successive sample averaging (high res) at ANY timebase - or any memory setting; I've tried all of them.

[marmad]

Well, since I've finished my first video, I've had much more time to run various tests on the Siglent. And one thing I have to say is that I get more and more impressed with the sampling/processing engine of the Siglent. Not only is it very fast, it actually TRULY does 256 levels of intensity grading (when wfrm/s are high enough, as noted below) - unlike the Rigol which, as I've pointed out before in the past, only seems to do 64 level grading (even though they advertise it as 256). The intensity grading of the Siglent is often very nice - so I'm making a new short video to try to demonstrate how good it can look when running. As far as I know, it's the only DSO doing real 256 level grading that costs less than several thousand bucks.

The only problems with the SDS2000's engine at the moment are it's drop-off in speeds at the faster timebases (1ns, 2ns, etc), and perhaps the need for a slightly more elegant (or switchable) method of dealing with large sample lengths with low wfrm/s - which is what I wanted to clarify here before posting the second video; i.e. the way that the intensity grading differs from the intensity grading that the Rigol is doing:

The Rigol seems to do intensity grading BEFORE downsampling (at least at slower timebases with large memory depths), the Siglent does it after. That means, on the Rigol, the sample memory length can affect the intensity grading - on the Siglent it does not. The Siglent grading is based purely on wfrm/s, with a ratio of something similar to the following table (this is not precise since I haven't tested the exact boundaries - just an indication):

>= 1024 wfrms/s   = 256 levels / 128 color temp.
1023 - 512 wfrm/s = 192 levels / 64 color temp.
512 - 256 wfrm/s  = 32 levels / 32 color temp.
255 - 128 wfrm/s  = 16 levels / 16 color temp.
127 - 64 wfrm/s   = 8 levels / 8 color temp.
<= 63 wfrm/s      = 4 levels / 4 color temp.

Each method has it's advantages at slower timebases (Rigol = more grading information with larger sample sizes but less wfrm/s / Siglent = more wfrm/s) so I'm not proposing one way is necessarily better than the other - it's just interesting to note the difference. Of course, the best would be to have the ability to use EITHER method as desired.

First, two examples of the same waveform at 50ns - with the difference in grading levels noted - show the Siglent with more than 3x more levels:






Second:






And here, demonstrating the difference made by grading before downsampling at 2ms - note that with memory length set to 140k on the Rigol, the waveform image is similar to the 28M length on the Siglent (but not quite as detailed), but when the Rigol is set to 14M, the grading changes to reflect the downsampled data from the larger memory depth waveform:








Siglent intensity-grading demonstration video to follow in the next day or so....

[tautech]

@tautech and rf-loop:  Doing some more testing with the Siglent, and I can't seem to see any evidence that High Res mode is working. Can either of you confirm with Siglent if it's supposed to be a functional feature yet?

I too, do not have the expected result.
We will ask Siglent if High Res is fully functional.

Well it seems we have a High Res button that is waiting for the next firmware release. 

And before you ask when that might be, we should ponder on other fixes that could be included in the next firmware package.
Lets not redesign Siglent's SDS2000, but get the features that are "not quite right", to work as we expect.
We don't want to slow the UI to a crawl with a mass of extra functionality, but I hear you all re Segmented waveform navigation, it is required.
marmad, nctnico, don, can we summarize the required fixes?

[nctnico]

Bugs:
- MSO signals on wrong timescale with decode enabled. IMHO digital signals should be visible with decode enabled.
- 'Peak detect' is not 'peak detect'. but envelope mode
- Signals display wrong (out of phase) when turning timescale or horizontal position knob quickly in 'stop' mode.
- When averaging is enabled the averaged signals are no longer visible in 'stop' mode (*).
- High resolution mode does nothing
- Finite persistance time setting does not seem to change persistance period.
- MSO signals are out of sync with analog channels when zooming in. Sometimes too early sometimes behind.
- Cursors (X / vertical) not working in MSO mode without analog channels active
- Decode doesn't work in single trigger mode
- Serial decoding doesn't always decode (shows empty red balloon) if the timing of the serial data is slightly off.

Improvements:
- Use partial FFT with higher frequency resolution when FFT is zoomed in
- Use averaging on FFT result (on my TDS744A I usually use 10 to 20 averages)
- Allow digital signals for decode
- Make 'intensity adjust' button less sensitive to rotation when selecting something from a menu list
- Make clear that decode and measurement statistics cannot be enabled simultaneously
- Allow decode and measurements to be used simultaneously.

(*) Yesterday I was doing some measurements with the cursors on a very noisy unstable signal. To clean the signal up I enabled averaging. But when I pressed the 'run/stop' button to stop the acquisition the noisy signal is displayed instead of the averaged signal.

edit: added finite persistance bug.
edit2: removed envelope mode
edit3: Added MSO signals out of sync with analog channels
edit4: Cursors not working in MSO mode
edit5: Decode doesn't work in single trigger mode, serial decode bug

[marmad]

- 'Peak detect' is not 'peak detect' but envelope mode

I'm not sure why you think the DSO is doing Envelope mode (which is just a combination of separate Peak Detect acquisitions). Any DSO I've used that's had that mode allowed you to set the number of Peak Detect acquisitions to combine (just like with Average).

Instead, it seems to me that the problem is that (at some/all longer memory lengths and/or slower timebases) the Siglent doesn't do Peak Detect on the sample memory before down-sampling, thus making it possible to miss peaks that existed. And it seems that this might be a general problem with the Siglent in a few areas; i.e. in service of speed (fast throughput) they are doing certain (or all?) things post-decimation.

In my previous post, I pointed out that the Siglent does intensity grading post-decimation, and one could argue that this might be the best way to do it (at least for speed) - but doing Acquisition modes post-decimation can cause problems (or at least unusual behavior). This is perhaps why High Res mode has not been implemented yet; i.e. it's not possible to do it post-decimation.

Even the Average mode seems to be affected - look at the two attached images using identical settings on an AM waveform @ 10us/div, 1200 wfrm/s and Average mode (32).

[WVL_KsZeN]

this one is really strange. When you're looking at noise, how come some traces are brighter (over the complete width) then others? This looks more like some sort of persistence (last captured trace being the brightest)? How can this be?

[marmad]

this one is really strange. When you're looking at noise, how come some traces are brighter (over the complete width) then others? This looks more like some sort of persistence (last captured trace being the brightest)? How can this be?

There are different ways to implement digital phosphor (intensity grading), but usually this entails some type of 'persistence'; i.e. emulating chemical phosphorescence by controlling the rate of decay of the overlaid rasterized waveforms.

[don]

I've used an sds2074 and also thought the grading looked more like a persistence mode and also didn't see anywhere near 256 levels. If have a tek tds754d  which only has 16 levels but I saw more brightness variation on that scope for whatever reason.  Read somewhere in this thread that Siglent grading levels varies with timebase for some odd reason (its just building a histogram, why does it vary?). The timebase variation  has waveform capture rate implications makes me think they  are not handling grading correctly. You should keep track of waveform hits at each pixel and vary the intesisty based on the relative number of hits.

The other odd thing is their grading does not have any compression capability. Meaning if you have a 14Mpts record shown on the display, the entire signal will be the same brightness. They should be compressing the data such that one pixel reflects the slice of time the pixel represents so you can visualize a large record.  Seems they just rail it at min / max.

[marmad]

I've used an sds2074 and also thought the grading looked more like a persistence mode and also didn't see anywhere near 256 levels. If have a tek tds754d  which only has 16 levels but I saw more brightness variation on that scope for whatever reason.  Read somewhere in this thread that Siglent grading levels varies with timebase for some odd reason (its just building a histogram, why does it vary?).

This is discussed in detail 7 posts above this one. The Siglent certainly does do 256 levels (given a minimum number of wfrm/s) - I've analyzed many images from the scope. But, as mentioned in the above post, the levels are tied to the waveform update rate and drop dramatically below ~430 wfrm/s.

You should keep track of waveform hits at each pixel and vary the intensity based on the relative number of hits.

It's slightly more complicated than that. Your Tek TDS754D does 'decay' (persistence) of the pixels hit by the waveforms as well.

The other odd thing is their grading does not have any compression capability. Meaning if you have a 14Mpts record shown on the display, the entire signal will be the same brightness. They should be compressing the data such that one pixel reflects the slice of time the pixel represents so you can visualize a large record.

Yes, they are doing intensity grading post-decimation.

[nctnico]

- 'Peak detect' is not 'peak detect' but envelope mode

I'm not sure why you think the DSO is doing Envelope mode (which is just a combination of separate Peak Detect acquisitions). Any DSO I've used that's had that mode allowed you to set the number of Peak Detect acquisitions to combine (just like with Average).

Instead, it seems to me that the problem is that (at some/all longer memory lengths and/or slower timebases) the Siglent doesn't do Peak Detect on the sample memory before down-sampling, thus making it possible to miss peaks that existed.

Again: there is a very distinct difference between envelope mode and peak detect:
- Peak detect lets the acquisition hardware (ADC) run at full speed, do peak detect and decimate the result from the peak detectors.
- Envelope mode lets the acquisition hardware run at a slower speed or do decimation first and then apply peak detection to the decimated result.

Both modes have their advantages:
- Envelope mode shows less HF noise; basically because the lower sample rate acts as a filter but you could get spurious hits.
- Peak-detect catches any glitch as long as it's frequency is below the bandwidth of the oscilloscope.

The SDS2000 only does envelope mode so Siglent should rename it to envelope mode or implement peak detect properly.

[marmad]

Again: there is a very distinct difference between envelope mode and peak detect:
- Peak detect lets the acquisition hardware (ADC) run at full speed, do peak detect and decimate the result from the peak detectors.
- Envelope mode lets the acquisition hardware run at a slower speed or do decimation first and then apply peak detection to the decimated result.

Where are you getting this particular definition of Envelope mode from? I've never heard of it. Can you post a source?

Here is Tektronix's definition of Envelope mode:

[nctnico]

It seems most seem to agree with Tektronix regarding with the definition of envelope mode but I've seen other definitions as well. For argument's sake: the way Siglent does peak-detect is wrong (and useless) so they shouldn't call it peak detect.

This also offers a short overview of various acquisition modes:
http://www.lanl.gov/conferences/biw10/preprints/MOVNB01_preprint.pdf

[marmad]

For argument's sake: the way Siglent does peak-detect is wrong (and useless) so they shouldn't call it peak detect.

Yes, sending it 10ns pulses, I can see anomalies at timebases >= 5ms/div - or when the sample rate drops to <= 400MSa/s - but not at higher timebases/sample rates.

[rf-loop]

- Peak detect lets the acquisition hardware (ADC) run at full speed,

All modes let ADC run full speed.

But it is clear that there is something wrong in peak detect mode in Siglent this model.

Hi-Lo detector need be in full speed ADC stream of course is in use. But of course it need only if displayed samples rate is below ADC maximum.

[don]

This is discussed in detail 7 posts above this one. The Siglent certainly does do 256 levels (given a minimum number of wfrm/s) - I've analyzed many images from the scope. But, as mentioned in the above post, the levels are tied to the waveform update rate and drop dramatically below ~430 wfrm/s.

Yes, and this is not expected.  Intensity grading should not depend on wfrm/s.  Its displaying accumulated data and the wfrm/s dependency should be how long it took to acquire. As data is acquired, grading is dynamically adjusted but levels do not depend in any way on wfrm/s.  Theoretically you could have data coming in slower and grading should look the same as faster wfrms/s long as number of captured events are the same .

It's slightly more complicated than that. Your Tek TDS754D does 'decay' (persistence) of the pixels hit by the waveforms as well.

By decay, I assume you just mean the way tek accumulates and constantly adjust grading based on new max /min hits. You can select shallow or deep accumulator (32bit or 64). But persistence is not part of tek intensity grading.  All data is kept and displayed with intensity grading.

The other odd thing is their grading does not have any compression capability. Meaning if you have a 14Mpts record shown on the display, the entire signal will be the same brightness. They should be compressing the data such that one pixel reflects the slice of time the pixel represents so you can visualize a large record.

Yes, they are doing intensity grading post-decimation.

Whatever they are doing unfortunatley means no intensity grading on single event captures.

Overall its a pretty basic/partial implementation of intensity grading .  given price not sure what to expect though. Lowest cost agilent or tek probably what to compare to. Rigol is a little goofy as well.  At least for compression of long captures.

[nctnico]

IMHO intensity grading is sort of related to persistance with a finite (short time). I'm not quite sure if 1000 levels of intensity grading make any difference than 10 levels of intensity grading when looking at a signal. Your eyes can only process a limited amount of information anyway. More is just nice for the marketing department so they can print a large number in the specification.

 From what I understand there are 2 kinds of intensity grading:
- one where the most recent signal is the brightest
- one where the brightest places on the screen are where the signal spends most of it's time.

It seems the SDS2000 is doing a combination of both. And I think I have found another bug: I see no difference between the finite persistance set to 1 second or 30 seconds (and anything in between). I'll add the bug to my list.

[marmad]

By decay, I assume you just mean the way tek accumulates and constantly adjust grading based on new max /min hits. You can select shallow or deep accumulator (32bit or 64). But persistence is not part of tek intensity grading.  All data is kept and displayed with intensity grading.

By decay I mean precisely what Tektronix means when they describe their implementation of Digital Phosphor, as in the TDS3000B User Manual:

"A Digital Phosphor Oscilloscope can clearly display intensity modulation in your signals. The oscilloscope automatically overlays subsequent acquisitions and then decays them to simulate the writing and decay of the phosphor in an analog oscilloscope CRT."

This sounds to me exactly like persistence - but of course, at a much faster rate than normal 'persistence' settings on DSOs.

Overall its a pretty basic/partial implementation of intensity grading .  given price not sure what to expect though. Lowest cost agilent or tek probably what to compare to. Rigol is a little goofy as well.  At least for compression of long captures.

I'm not really sure there is one method of doing intensity grading that Tektronix, Agilent, etc. all agree is the definitive implementation. Siglent is using a method that is fast, but contains less information than Rigol's implementation at lower timebases with larger sample depths - it's that simple.

[marmad]

IMHO intensity grading is sort of related to persistance with a finite (short time). I'm not quite sure if 1000 levels of intensity grading make any difference than 10 levels of intensity grading when looking at a signal. Your eyes can only process a limited amount of information anyway. More is just nice for the marketing department so the can print a large number in the specification.

Well, I don't know what your eyes are like, but mine can tell the difference between 10 and 100 levels quite readily. Between 64 and 256 is much more tricky - then it really depends on the signal you're looking at and the overall brightness of the display - it's possible some times, but not always.

From what I understand there are 2 kinds of intensity grading:
- one where the most recent signal is the brightest
- one where the brightest places on the screen are where the signal spends most of it's time.

As well as using / not using a data-density histogram binned from the vertical values occurring in the full waveform memory - which the Rigol DS2000 does - but the Siglent SDS2000 does not.

[nctnico]

IMHO intensity grading is sort of related to persistance with a finite (short time). I'm not quite sure if 1000 levels of intensity grading make any difference than 10 levels of intensity grading when looking at a signal. Your eyes can only process a limited amount of information anyway. More is just nice for the marketing department so the can print a large number in the specification.

Well, I don't know what your eyes are like, but mine can tell the difference between 10 and 100 levels quite readily.

I know but your brain also needs to process that visual information into a meaningful conclusion (is the signal as expected or not; what needs further adjusting).

[marmad]

Added a small video to the first post in this thread looking more closely at the intensity grading on the Siglent SDS2000.

[marmad]

marmad, nctnico, don, can we summarize the required fixes?

IMO, the most immediate necessary additions/changes for the DSO portion (aside from obvious bug fixes) would be:

1) Get the main Acquisition modes working properly (i.e. High Res mode and Peak Detect BEFORE downsampling).
2) Add some functions to make Sequence mode usable as a tool: adjustable speeds for playback, 'Overlay All', etc.
3) Remove the code which disables the selected item for the Universal knob when the Trigger knob is turned (or bumped). This would immediately stop the very annoying behavior of 'losing' what you're adjusting when turning the Universal knob by accidentally nudging the too-close Trigger Level knob (which happens to me regularly).

On a more fundamental level, I don't like Siglent's implementation of slower timebases in Y-T mode (at least, when combined with the way the DSO behaves while waiting for triggers). What this means is that when you're working at a slow timebase, the DSO draws nothing on the display until the entire acquisition is finished (regardless of trigger position - e.g. 7 seconds at 500ms/div) - during which, changing certain settings (e.g. trigger position, channel scale, etc) keeps resetting the trigger (so, of course, re-starting the acquisition), but doesn't change the display of the last captured waveform on the screen.

So you can be adjusting settings for a long time at slower timebases - with no visual feedback of either the last acquired waveform moving onscreen - or seeing a newly acquired waveform appearing past the trigger point. It makes working at slower timebases in Y-T mode quite frustrating.

Later:
1) I think Siglent needs to think of a more elegant solution for doing intensity grading at lower wfrm/s.
2) Add measurements for Math waveforms.

[nctnico]

I added some other bugs to my previous posting as well:
https://www.eevblog.com/forum/testgear/review-siglent-sds2304-a-comparison-of-features-with-rigol-ds2000-series/msg503859/#msg503859

[Hydrawerk]

Well, remember that this is the first digital phosphor scope of this kind ever made by Siglent. It is not like that beloved Rigol DS2000 released after DS6000 and DS4000.