Siglent SDS2000X Plus

[gf]

If the interpolated sample rate at 5ns/div is per se already 200GSa/s, then 20x explicit oversamling leads to 4000Gsa/s. This sounds actually plausible. But at the end it only changes the FFT resolution, while the time window (in seconds) is still the same, since sample rate and #points increase by the same factor. I was rather thinking of using timebases >= 50ns/div (where it does not interpolate automatically), and interpolate/upsample explicitly from 2GSa/s to 40 GSa/s, and check how many points we get.

And yes, since the FFT spectrum does not fit with the time domain data, there might be a bug.

[Martin72]

Apart from this and in general, having a formula editor is a real great feature...

Good night guys.
Martin

[2N3055]

I can't use the SCPI command: LIC : DEL to remove the hacked Bandwidth license, maybe who knows how to remove  a bandwidth license？

A bit of advice, there is a separate hack topic, best to keep related questions there:

https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-hack/msg3082536/#msg3082536

I'm sure some of the experts there will know..

[cesare]

I can't find any sort of low pass filter in the maths functions.

I'm playing around with a 1 bit DAC implementation on an FPGA board, and don't want to go to the trouble of actually building the filter section (for now). Is there some other maths function that can get me close? I was thinking ERES might do enough, but is there a better way?

[Martin72]

Could you try adjust it to be somewhere at 200-400GS/s so bin width is not half scope's bandwidth.

With the lowest upsample coefficient (x2) it drops down to 400GS/s, 1024pts.
Effect: Same as before.

Martin

[gf]

I checked all your screenshots again, and it seems that the chosen FFT window size was between 50% and 100% of the screen width (in samples).
Seems that it selects a power of two which falls into this range. Don't know of course, whether this is a rule, or just by accident.
An exception was 5ns/div, where the window size was even less than 1/10 of the screen width.
Sorry, was a mistake. I'll correct in a new message.

EDIT: Interpolate(...) does not change that, and does not reduce the frequency bin size either, since sample rate increases by the same factor as the number of points, and bin size is sample rate divided by number of points.
[ In order to interpolate in the frequency domain, one would rather need to zero pad the window in the time domain to a larger size (2x, 4x,...), before doing the FFT.
Actually, in conjunction with zero padding, the window size would not even need to be a power of two, but only the padded size needs to be a power of two. ]

[rf-loop]

Seems that it selects a power of two which falls into this range. Don't know of course, whether this is a rule, or just by accident.

Not accident. As told in user manual.
"When the number of points in the time domain, N, is less than 2 Mpts, the FFT takes the number which is an integer power of 2 closest to N."
Starting from 2^²¹ down to 2^²

[Orange]

I can't use the SCPI command: LIC : DEL to remove the hacked Bandwidth license, maybe who knows how to remove  a bandwidth license？

Try LIC:DEL (without the spaces around the : character)

[Martin72]

Someone else with fast pulse gen and SDS2000X+ should try..
Maybe there is a bug lurking somewhere in there..

Nobody else except me ?
I think we should make it clear if this is a bug or not...

[gf]

I think we should make it clear if this is a bug or not...

The inflexible window sizes seem to be rather a limitation.
Still, at 20ns/div, you should see a descending spectrum, if you center the impulse the in the FFT window.
Btw, did you also look at the full span (0...25GHz at 20ns/div and 10ns/div)? Maybe the scale is wrong?
Did you rule out Average(...) as potential source of the problem? It was also part of the expression.

[tautech]

Guys, where on earth is the recent post with tips to remove the mainboard from 2kX Plus ?
Particularly need the tricks for the 3 hidden screws under the front panel vinyl overlay.
TIA

[tautech]

Guys, where on earth is the recent post with tips to remove the mainboard from 2kX Plus ?
Particularly need the tricks for the 3 hidden screws under the front panel vinyl overlay.
TIA

These little suckers.....
Proceeding carefully with a heat gun....wish me luck.

[tautech]

Guys, where on earth is the recent post with tips to remove the mainboard from 2kX Plus ?
Particularly need the tricks for the 3 hidden screws under the front panel vinyl overlay.
TIA

These little suckers.....
Proceeding carefully with a heat gun....wish me luck.

All apart now and several pics taken of the off sides of the PCB's you don't see in any teardowns. 
Will post them in the next day or so FYI along with some further info should anyone need to fix a blown 50 Ohm input.
  time in NZ.

[mawyatt]

Guys, where on earth is the recent post with tips to remove the mainboard from 2kX Plus ?
Particularly need the tricks for the 3 hidden screws under the front panel vinyl overlay.
TIA

These little suckers.....
Proceeding carefully with a heat gun....wish me luck.

May recall we had to remove the cover on our new SDS2102X Plus due a drink being spilled into the air vents, painful story in more ways than one 

This required removing the vinyl overlay to gain access to those screws, and we buggered up the overlay. Should have tried the heat gun approach 

Anyway Jason at Siglent US kindly sent a replacement overlay. The scope was cleaned, then reassembled and looks like new

Good luck with your efforts!!

Best,

[tautech]

Guys, where on earth is the recent post with tips to remove the mainboard from 2kX Plus ?
Particularly need the tricks for the 3 hidden screws under the front panel vinyl overlay.
TIA

These little suckers.....
Proceeding carefully with a heat gun....wish me luck.

May recall we had to remove the cover on our new SDS2102X Plus due a drink being spilled into the air vents, painful story in more ways than one 

This required removing the vinyl overlay to gain access to those screws, and we buggered up the overlay. Should have tried the heat gun approach 

Anyway Jason at Siglent US kindly sent a replacement overlay. The scope was cleaned, then reassembled and looks like new

Good luck with your efforts!!

Best,

I well recall Mike and found one of your old posts about the event. Yep I buggered the overlay too and should've used a hot air station for better heat control. 
Luckily our customer almost passes us to and from work so we'll get another overlay in a stock order for him.

[Martin72]

@gf:

No averaging, interpolating to get more points.
Dx = 4

[gf]

The 25 GHz span was actually meant for 50 GSa/s (i.e. for 10ns/div or 20ns/div without extra interpolation), in order to see the full FFT spectrum which is calculated.
At 250 GSa/s, the full 0...Nyqist span were rather 0...125 GHz.

The extra interpolation does not help here anyway. It gives more points, but OTOH also increases sample rate, so it still does not increase the frequency resolution at the end, since the points are distributed over a larger span.

EDIT:

The spectrum is still strange. There is some roll-off now, but at too high frequency.

I'm also unsure whether the edge (including pre- and overshoot) is completely inside the -5ns...+5ns window from the center of the screen. Maybe you  can shift the edge a little bit to the right?
At 20ns/div, the FFT window were -10ns...+10ns, so you had more tolerance for edge placement, and the frequency resolution of the FFT is 2x better as well.

[gf]

Not accident. As told in user manual.
"When the number of points in the time domain, N, is less than 2 Mpts, the FFT takes the number which is an integer power of 2 closest to N."

It is obvious that it needs to round down to the next power of two, since it can only discard, but not "invent" samples.
But what is the "number of points in the time domain", from which the power-of-two FFT size is then derived?

I actually made a mistake when I wrote "FFT window size was between 50% and 100% of the screen width". Checked again, and noticed the mistake.
It only applies to the screenshots with >= 50ns/div, but not for faster time bases.

Below is an updated table, extracted from Martin72's screenshots. FFT window sizes range from 0.512 to 8.192 screen divisions.

Why is the FFT window size so much smaller than the screen width at <= 20ns/div? This can no longer be explined by rounding down the screen width (in samples) to the next smaller power of two.
The interpolated samples for the full screen width need to exist anyway, otherwise the trace could not be displayed. So why not use as many of them as possible?

For instance: 5ns/div reports 100 (raw) samples @2GSa/s (which is the width of the screen).
With 100x interpolation (-> 200GSa/s) this gives about 10000 samples, and a power-of-two FFT size of 8192 were possible. But it chooses only 512.


2000ns/div: FFT window: 16384 ns, or 8.192 div (2GSa/s)
200ns/div: FFT window: 1024 ns, or 5.12 div (2GSa/s)
100ns/div: FFT window: 512 ns, or 5.12 div (2GSa/s)
50ns/div: FFT window: 256 ns, or 5.12 div (2GSa/s)
20ns/div: FFT window: 20.48 ns, or 1.024 div (50GSa/s interpolated)
10ns/div: FFT window: 10.24 ns, or 1.024 div (50GSa/s interpolated)
5ns/div: FFT window: 2.56 ns, or 0.512 div (200GSa/s interpolated)
5ns/div: FFT window: 2.56 ns, or 0.512 div (400GSa/s interpolated)

[rf-loop]

When I look this @Martin72 image....  (and this is nothing personal at all... just one image for example )



I have "serious problem" when I look this image (or same kind of others rel. this sub subject).

Problem is that I do not know if I need cry or laugh. :/

Of course, it can also be expressed by saying: mathematically produced, perhaps correctly calculated, but if you think it gives a picture of reality, it’s mostly garbage.
If this is used seriously for something I will cry.   
If it is just directed for laboratory workshop joke, I can laugh.  25GHz... 

Maybe my signature gives a clue as to the root cause of my problem - or then not.

[Martin72]

When you have cried/laughed enough, maybe you can help and make things clearer in the case why the sds2k+ can't do that, what the HD version can 

[gf]

When I look this @Martin72 image....  (and this is nothing personal at all... just one image for example )
I have "serious problem" when I look this image (or same kind of others rel. this sub subject).

Problem is that I do not know if I need cry or laugh. :/

Of course, it can also be expressed by saying: mathematically produced, perhaps correctly calculated, but if you think it gives a picture of reality, it’s mostly garbage.
If this is used seriously for something I will cry.   
If it is just directed for laboratory workshop joke, I can laugh.  25GHz... 

Maybe my signature gives a clue as to the root cause of my problem - or then not.

The large span is for diagnosis.
Nyquist of the interpolated samples is of course that high (i.e. 25GHz @50GSa/s, or 125GHz @250GSa/s), and the FFT spectrum goes up to Nyquist.
However, if the spectrum of the raw samples is 0...1GHz, then the interpolation is not supposed to introduce any frequencies beyond 1GHz. Consequently 1...25GHz (or 1...125) GHz should be empty.
Aim of the large span was to check that the spectrum is really empty (zero) beyond 1GHz. And as we see it is not. It rather extends up to ~6 GHz before dopping to the noise floor
This is not plausible. I think there is a bug in the calculation.

[rf-loop]

When I look this @Martin72 image....  (and this is nothing personal at all... just one image for example )
I have "serious problem" when I look this image (or same kind of others rel. this sub subject).

Problem is that I do not know if I need cry or laugh. :/

Of course, it can also be expressed by saying: mathematically produced, perhaps correctly calculated, but if you think it gives a picture of reality, it’s mostly garbage.
If this is used seriously for something I will cry.   
If it is just directed for laboratory workshop joke, I can laugh.  25GHz... 

Maybe my signature gives a clue as to the root cause of my problem - or then not.

The large span is for diagnosis.

For diagnosis about what...

The purpose of an oscilloscope is to diagnose a more or less known signal.
What that FFT looks like over (real raw samples)fNyquist part doesn't say anything about the signal being diagnosed with the oscilloscope. But of course, it is always good to remember that whatever occurs on the oscilloscope screen is a mere sum of errors from reality already. And as here, quite detached from any reality. If we think reality is a signal in an oscilloscope input.

I think displaying such "nonsense" data on an oscilloscope screen is pretty much the same as meaningless speech in many words. Perhaps that screen could be filled numbers that have no meaning whatsoever for the purpose for which the tool is used for work.

Who and how this serves some real purpose.

Only some kind of semireality is in ADC raw true samples.
When oscilloscope -3dB BW is around 500MHz ands ADC samplerate is 2GSa/s ands then there is some attenuated signal up to fNyquist and some amount even over...
Then we interpolate (produce fake samples between true raw samples)... my opinion is that in this case we only add shit to nice soup. Then perhaps some noob are watching and wondering... oh well it can display 2GHz... oh more... it can display 10GHz... omg. At least that shit portion of display should be marked with some color that tells it to be shit at least relative to the incoming signal. (all over ~0.9 - 1GHz in this case)  Even when it is mathematically true that with interpolated samples it "looks like" samplerate is higher and then FFT range is more (always 0 to fNyq.).

The basic question should always be: What does this say about the signal being tested. This is purpose of these tools (exept if some just want play fun - what is also sometimes fun).
The purpose of this tool is to measure - analyze and report the signal under test. In this example image, what this FFT tell about signal under test if we look this 0 - 25GHz  full span. Neasly nothing.  Is it really useful to "fool" users with this kind of entertainment on screen.

How ever, all this is or can be somehow useful - and nice and also it may be, way or other,  educational. Know your equipment... and so on.

Gentlemen's, go on...

[Martin72]

Bravo...*Hands clapping*
And now after this you can perhaps tell the fools why the sds2k+ can´t reproduce the signalform like the two chinese market only scopes before.
Is it because of a bug ( last time a FFT bug was fixed), is it because it can´t do it or is it because I´m to too dumb for the right adjustments.
With the last two I can live with.

[2N3055]

As I said before dV/dt is not a symbolic math function, but a numeric algorithm estimating from discrete data samples.
Key words here are sample density at the very fast edges ( and tendency of derivative to go to infinity when confused ) and  fact that it uses 4,8 or more points to calculate slope change for the derivative (depending on the setting). That alone means it is basically using low pass filtered data for the derivative calculation, effectively. I'm not privy to exact formula implementation so can't tell exactly. Will investigate.

I'm still looking into exact explanation how it works, as it seems to work a bit differently between 2000X+ and 6000H12 and 2000XHD. Which is normal, because there are differences in acquisition modes, resolution etc..

If there is any sub optimal behaviour here in any of mentioned scopes, it will be reported.
In meantime, let's move on. A separate topic can be opened if this ends up being worthy of separate writeup..

No need for adversarial tone. We are all friendlies here...

Best,

[MathWizard]

So these SDS2000X-plus are about 2yo now right? Is there any difference between the oldest and newest products ?

This year or next, I'm planning to get the low BW model and hack it. And that will probably be about the best scope I'd ever need, or afford for hobby use.

Is there anything newer and similar in spec's/price close to release ?