Trying to display bandwith via Math on Siglent SDS2k+/2kHD/800X HD

[Martin72]

Hi all,

As mentioned (and suggested by 2N3055) in the Siglent SDS2000X plus thread, I´m starting a dedicated one for this "problem".
Actually in questionmarks because it´s not really clear if this is a problem sitting in front of the scope...
What was happen ?
In the siglent thread the question came up if you can´t display the bandwith (frequency response) by using the math functions.
You can:

https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg4156720/#msg4156720

I´ve tried to reproduce it on my sds2504x+ but it failed somehow:

https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg4157086/#msg4157086

Then 2N3055 took another scope from siglent, it works.
And so on, and so on, see the following posts there.
The expression used in the first time by 2N3055 is FFT(d(Channel)/dx).

At work I checked this with three different scopes from lecroy.
-Waverunner 9054, 500Mhz, 20GSa/s
-HDO6034A, 350Mhz, 10GSa/s
-Wavesurfer WS3024Z, 200Mhz, 4GSa/s

On all the models you can´t do very much on the FFT adjustments(like memorypoints, markers and so on the siglent got).
Max. resolution was 10ns/div, above (5ns and more) the ws3024z didn´t display the function proper(No problem for the Waverunner)
Today I used the same settings at home on my siglent for comparison all the scopes directly.
Tomorrow I´ll follow the questions/suggestions from gf:

- FFT(d(Intrp(C1))/dt)
- 20x interpolation
- dx as short as possible
- Flat Top window function
- span 0...1GHz
- 2µs/div (as before) might be a reasonable starting point.

https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg4171216/#msg4171216

Goal of this new thread is to find out, if this is a bug to fix or if this siglent model is simply not capable to do this.
Or it´s me who is not capable to do this... 
The last is no problem, I´m always willing to learn.
EDIT : Title changed.

[Martin72]

Pics from the "big" scopes..

[rf-loop]

I do not have same kind of signal available now.
But because these are just partially for fun, I use SDS1104X-E oscilloscope Trig out for get edge what is still possible to measure with this scope, so it is suitable signal for this scope when it works 2GSa/s real samplerate. (if use 1GSa/s this signal rise is too fast). This all is not meaningful at all in this case.

But then this what was @gf ask (originally in other thred)

- FFT(d(Intrp(C1))/dt)
- 20x interpolation
- dx as short as possible
- Flat Top window function
...
...

When I look your images I can not find Interpolation.


Here I have used SDS2504X HD

- FFT(d(Intrp(C1))/dt)
- 20x interpolation
- dx 4 (as short as possible)
- FFT Flat Top window function

1st image (1)a  is 0-2GHz span (iterpolated samplerate 40GSa/s)
2nd image (2)a is full 0-20GHz span (iterpolated samplerate 40GSa/s)
Full span based to real samplerate is of course 0-1GHz

[Martin72]

Hi,

When I look your images I can not find Interpolation.

It´s because I didn´t it so far - Until now...

- FFT(d(Intrp(C1))/dt)
- 20x interpolation
- dx as short as possible
- Flat Top window function
- span 0...1GHz
- 2µs/div (as before) might be a reasonable starting point.

Here it is, dx is 4 and can´t be lowered.
I´ve saved all timebases between 2µs and 2ns, here a selection.

[gf]

What was the exact math expression here? I just see that it starts with "FFT(Intrp(..."

I think I aked for: FFT(d(Intrp(Cx))/dt).
I.e. interpolate first, then d/dx, and finally FFT.

[Martin72]

FFT(d(Intrp(Cx))/dt).

Damned..

See settings pic...
Will repeat it later

[rf-loop]

Just a small margin note (Settings.png)

Naturally if select Menu Style [Embedded], it does not cover the signal area.

Sometimes when take document images, it can be useful to keep some desired settings menu visible, without loosing any part of signal. (in some cases signal rendering for display may be better with Menu Style [Floating])

[tautech]

Just a small margin note (Settings.png)

Naturally if select Menu Style [Embedded], it does not cover the signal area.

Sometimes when take document images, it can be useful to keep some desired settings menu visible, without loosing any part of signal. (in some cases signal rendering for display may be better with Menu Style [Floating])

This only works if the user has fully explored Display settings and set Hide Menu = OFF .....not some setting of 10s that Siglent think best for newbies to this scope....where's my menu gone ? 

[Martin72]

Thank you,
It was my hope that the real siglent-knowers join this thread after they were very calm in the original one and indeed, the solution was the menu style..

Back to topic..

@gf:
Now the corrected ones and you´ll see, nothing hast changed in relation to the expression before.

[gf]

Not good. Seems like dx=4 were interpreted as dx=80, when d/dx is applied to 20x interpolated data. At timebase >= 50ns/div, this makes this d/dx implementation more or less useless for frequencies above say 100MHz ¹⁾
[ I'd rather expected that dx=4 means 4 samples of the actual sample rate of the given data, i.e. 4 samples @40GSa/s when the data fed to d/dx is 20x interpolated. ]

¹⁾ There is no clear border, of course, but it depends on the tolerated error.
The diagram below show the amplitude error (in the frequency domain) of a 2-point finite difference approximation of the 1st derivative, for delta x = 2ns.

[Martin72]

In the manual, dx can be selected between 2 and 20.
In real, it starts with 4, higher you can only get when using the keypad - the knob does nothing.
And: When you choose "max" on the keypad, dx has the value 256..
Tomorrow I´ll look at the adjustments the lecory scopes offers on d/dx and FFT.
I expect nearly "nothing", because "somehow" it works as expected without doing anything special.

[gf]

... This makes this d/dx implementation more or less useless for frequencies above say 100MHz ¹⁾

More exactly I should add "... for timebase >= 50ns/div" to this statement.

At <= 20ns/div the flat spectrum is not plausible either, but this seems to be a different issue.

[Martin72]

Hi,

What else can we do to manifest that this is a bug ?

[tv84]

What else can we do to manifest that this is a bug ?

A placard?? 

[Martin72]

Later...

[gf]

At work I checked this with three different scopes from lecroy.
-Waverunner 9054, 500Mhz, 20GSa/s
-HDO6034A, 350Mhz, 10GSa/s
-Wavesurfer WS3024Z, 200Mhz, 4GSa/s

The spectrum in the WS... image also looks a bit strange in the 900-1000 MHz region. I wouldn't expect it to ascend again.
Since you just display FFT(C1), there is no d/dx in the play which could garble the spectrum. I suggest to check how it looks up to 2GHz (half sample rate).



And the time domain trace shows some ripple as well (~10 cycles / div => ~1GHz), which is too regular to be random noise.

[gf]

What else can we do to manifest that this is a bug ?

One key aspect seems to be the definition of a "point".

The range of “dx”in the d/dtmenu is2~20. The measurement unitsarepointand thecorresponding time difference range is 0.02~0.20div. “div” indicates the number of the pixel points that each division has and is 100 for the SDS2000X Plus. If dx = 10 points, the time difference is 0.01*10 = 0.1div Then the differential operator calculates the average slope in10 points of the selected sourceand dx represents the time difference between the 10 sampled points.

On the one hand it talks about pixel points, on the other hand it talks about sampled points (which I would associate with samples). So what is really a "point"

At 1us/div, 4 "pixel points" were 40ns, and at 2us/div, 4 "pixel points" were 80ns (according to the definition in the handbook: 1 div = 100 pixel points).
But that's not what we see. The spectrum with the two big lobes tells us that the actual dx was 2ns, at all timebases >= 50ns/div. And 2ns are 4 samples @2GSa/s.

So the implementation obviously assumes that dx is in samples, and the stuff in the handbook regarding div and pixel points seems to be wrong an just leads to confusion.

Nevertheless, "number of samples" is still an incomplete time specification if the sample rate of these samples is not specifed as well.
For your experiments at > 50ns/div, the scope obviously assumed the raw sample rate of 2GSa/s, even if the d/dt operator is applied to interpolated samples @40GSa/s.
While this could be considered correct, if it were unambiguously documented, I still find it a significant limitation, since It hiders you to do what I intended to do.
[ Similiar to a camera in "auto" mode, which does whatever it thinks -- just not what the photographer wants it to do ]

For <= 20ns things still need more investigation. Could you please capture a single shot of say 4,000,000 raw samples (@2GSa/s) of this signal (I guess 200µs/div gives you that many points?), save them to an ASCII file, and post a link to the file? I would like to get a feeling for what can be extracted from these data, and what to expect.

[Martin72]

Now I got a appx 125MB csv. file on the stick..

What your post with the WS concerns, I want to repeat this thing, because a day before it was different(FFT, last pic from the first post here):

(see pic below)

Can´t believe it´s only because the samplerate was 2GSa/s at this time (nice "trap" : Measure-paras were active...).

[Martin72]

https://drive.google.com/file/d/1vUSKhswd9ki8PlOd07ZWp-cu3aet98cf/view?usp=sharing


csv-file

[gf]

Thank you. I'll take a look. It can take a while until I find some free time...

[Performa01]

The “stuff in the handbook” is most likely a (copy&paste) leftover from prehistoric manuals dating back to long forgotten times, when math and measurements were performed on screen data.

The “Dx” parameter in the SDS2000X Plus math refers to sample points and we actually need to take the sample rate into account whenever we want to convert it to an absolute time.
However, this is a deep memory scope; a sensible setting for the max. record length is 10/20 MPts, which results in a constant sample rate up to 1 ms/div, i.e. a record length of 10 ms. Heck, if the need should be, this could even be extended to 100 ms by selecting 100/200 MPts in the Acquire menu. I very much doubt that many users will have the need for the derivative function on slow waveforms <10 Hz and then at frequently changing horizontal time bases.

The little but more advanced brother SDS2000X HD already offers a constant sample rate setting just like the SDS6000. This can solve any problems and situations, where a function depends on the sample rate.

Regarding the attempts to get the frequency response from a fast risetime pulse, there are a number of reasons, why it doesn’t work as expected on the SDS2000X Plus right now.

1.   The scale is wrong. The FFT-sample rate is increased by two orders of magnitude whenever the argument is a derivative. This bug seems to be exclusive to the SDS2000X Plus.
2.   The input field of the formula editor is limited, so we just cannot add the average function there. But this is mandatory to get a nice and clean frequency response graph. On the other Siglent scopes we can use the average acquisition mode, so the lengthy expression “average()” does not eat up precious formula space.

A longer formula string has already been requested. I’ve also reported the FFT bug today.

[Performa01]

Here’s a demonstration what the SDS2504X HD looks like with a moderate ~500ps risetime pulse coming from the SDG7102A. 32x averaging before up-sampling.

[gf]

The “Dx” parameter in the SDS2000X Plus math refers to sample points and we actually need to take the sample rate into account whenever we want to convert it to an absolute time.

At 20ns/div and 10ns/div, FFT(Cx) shows a sample rate of 50GSa/s, and at 5ns/div it shows a sample rate of 200GSa/s.
This is obviously the interpolated sample rate, and it seems that interpolation joins in automatically at <= 20ns/div.

What is dx=4 now supposed to mean at (say) 10ns/div? Still 2ns (4 / 2 GSa/s), or 80ps (4 / 50 GSa/s)?

EDIT: Looking at the image in your previous message it must be (significantly) less than 2ns, otherwise the spectrum would look different. But is it 80ps then?

[rf-loop]

Here’s a demonstration what the SDS2504X HD looks like with a moderate ~500ps risetime pulse coming from the SDG7102A. 32x averaging before up-sampling.

Perhaps marker 1 horizontal position is better in ≥ 1/2 Δf 
(also looks weird your marker 2 vertical position looks like below trace level)

[gf]

Related to the frequency response calculation I also wonder:

• Is it possible to add/subtract/multiply/divide freqency domain traces (results of FFT) with math functions?
• If yes, in which space is the operation carried out? Complex, linear magnitude, log magnitude (dB)?
• Is it possible to load an (externally and synthetically generated) trace from e.g. USB stick, and use it as argument to math functions?