A call for Bin-Files as for new Siglent HD-Models

[hpw]

I am very interested to get a BIN file of full possible sample size as may 100M or 200M or even more as like to see the 1/f noise or spurs.

As from all new 800 HD, 1000 HD, 2000 HD, 3000HD, 6000 H12 & 7000A (HD) models

May with the following settings:

- select ONLY one Channels as Ch1 or Ch2

- Zero Offset the input and probe scale to 1

- Input shorted as as for a) 50E or b) 1M by 50E resistor or opened connections

- May the ERES to consider, currently to look at full BW of the DSO

- Lowest Y division as 1mV or 500uV

- As on my SDS2k + (10 bits), requires to set the time divisions may in the 2..5ms range,
  anyway the time division icon will display current sample size

- save it as BIN file to a USB stick (as large data may not saved to local file space)

- Copy the BIN file to your PC and zip/rar it please as 100M sampled data zipped gets about 10MBytes

and please PM me for a link to download the file using a file share service & additional text about the used input setup (50E/1M, opened ERES & etc) .... 

Hp

[rf-loop]

I am very interested to get a BIN file of full possible sample size as may 100M or 200M or even more as like to see the 1/f noise or spurs.

As from all new 800 HD, 1000 HD, 2000 HD, 6000 HD & 7000A HD models

May with the following settings:

- select ONLY one Channels as Ch1 or Ch2

- Zero Offset the input and probe scale to 1

- Input shorted as as for a) 50E or b) 1M by 50E resistor or opened connections

- May the ERES to consider, currently to look at full BW of the DSO

- Lowest Y division as 1mV or 500uV

- As on my SDS2k + (10 bits), requires to set the time divisions may in the 2..5ms range,
  anyway the time division icon will display current sample size

- save it as BIN file to a USB stick (as large data may not saved to local file space)

- Copy the BIN file to your PC and zip/rar it please as 100M sampled data zipped gets about 10MBytes

and please PM me for a link to download the file using a file share service & additional text about the used input setup (50E/1M, opened ERES & etc) .... 

Hp

What is SDS6000 HD. Outside China here is SDS6000A  and it is 8 bit.
Then here outside China we have now SDS800X HD, SDS1000X HD, SDS2000X HD, SDS3000X HD and SDS7000A  and these all are 12 bit models.
It is good practice to use right model names for avoid confusions and for help also sarch engines.

Because you did not ask SDS3000X HD, here it is. (~460M .zip)  Data start at 0x1000

Inside .zip there is SDS3104X_HD_Binary_C4_1.bin  (it include one single shot trace data: 800000000 bytes data 400Mpoints trace + 4096 byte header)

500uv/div
Input 50ohm (no signal)
10ms/div
memory 400Mpts
Sampling 4Gsa/s
Full BW

[ebastler]

Could you please clarify the conditions you are looking for? Thanks!

- Input shorted as as for a) 50E or b) 1M by 50E resistor or opened connections

So which one is it? Shorted, 50 Ohms, or open?

- May the ERES to consider, currently to look at full BW of the DSO

I understand that as "Currently I want to look at the full bandwidth of the scope. Maybe we can look at ERES later." So no ERES for now, right?

- As on my SDS2k + (10 bits), requires to set the time divisions may in the 2..5ms range,
  anyway the time division icon will display current sample size

Hang on, those 10 bits are in ERES mode, aren't they? So do you want ERES or not?

[hpw]

As rf-loop at first:

500uv/div
Input 50ohm (no signal)
10ms/div
memory 400Mpts
Sampling 4Gsa/s
Full BW

[egonotto]

Hello,

please the data with 1 V/div also.
Most modern oscilloscopes are very good at 1 mV/div. But at 1 V/div, opinions are divided and there are still big differences.

And if possible in the 1 MOhm and 50 Ohm branch.

Best regards
egonotto

[hpw]

I am very interested to get a BIN file of full possible sample size as may 100M or 200M or even more as like to see the 1/f noise or spurs.

As from all new 800 HD, 1000 HD, 2000 HD, 6000 HD & 7000A HD models

May with the following settings:

- select ONLY one Channels as Ch1 or Ch2

- Zero Offset the input and probe scale to 1

- Input shorted as as for a) 50E or b) 1M by 50E resistor or opened connections

- May the ERES to consider, currently to look at full BW of the DSO

- Lowest Y division as 1mV or 500uV

- As on my SDS2k + (10 bits), requires to set the time divisions may in the 2..5ms range,
  anyway the time division icon will display current sample size

- save it as BIN file to a USB stick (as large data may not saved to local file space)

- Copy the BIN file to your PC and zip/rar it please as 100M sampled data zipped gets about 10MBytes

and please PM me for a link to download the file using a file share service & additional text about the used input setup (50E/1M, opened ERES & etc) .... 

Hp

What is SDS6000 HD. Outside China here is SDS6000A  and it is 8 bit.
Then here outside China we have now SDS800X HD, SDS1000X HD, SDS2000X HD, SDS3000X HD and SDS7000A  and these all are 12 bit models.
It is good practice to use right model names for avoid confusions and for help also sarch engines.

Because you did not ask SDS3000X HD, here it is. (~460M .zip)  Data start at 0x1000

Inside .zip there is SDS3104X_HD_Binary_C4_1.bin  (it include one single shot trace data: 800000000 bytes data 400Mpoints trace + 4096 byte header)

500uv/div
Input 50ohm (no signal)
10ms/div
memory 400Mpts
Sampling 4Gsa/s
Full BW

Look attached, very noisy, as may the input requires some terminations....

Removed rtHz as something went false 

Just my 2 cents

Hp

[Performa01]

A few facts to consider:

At the highest sensitivities like 500 µV/div, there is no advantage having a 12-bit DSO – the ENOB will be less than 8 bits anyway because of the frontend noise, so no significant difference to an 8-bit DSO.

It is different for the 10-bit mode in a SDS2000X Plus, because this is DSP filtering and limits the bandwidth to 125 MHz. Because of this we will see an advantage even at the high sensitivities.

We cannot just zip random noise and hope to achieve a compression factor of 20. My experiments rather hint on a compression factor of 2.

There is a file for the SDS824X HD

Channel 4, 500 µV/div, 50 ohms terminated;
5 ms/div, 100 Mpts, 2 GSa/s;
Full bandwidth = 254 MHz;

Unsurprisingly, its size is about 90 MB:

https://mega.nz/file/zSJn2KAS#l8JMLXh0_uEakIIrVbC-suonbcblvjp59ObsYzixDYg

[Performa01]

Look attached, very noisy, as may the input requires some terminations....

What we see is a noise floor around -140 dB with lots of spurs. The diagram says the y-axis would be dBFS.
Have you ever thought wat this actually means? FS = full scale = 500 µV * 8 = 4 mVpp = 1.4 mVrms.
140 dB below that would be 140 pVrms – spoken out: picovolts!

The same goes for the 2nd plot which claims to show the noise density in nV/√Hz. Do you realize that a passive 50 ohm resistor alone exhibits already 0.9 nV/√Hz, hence numbers at or below that are utter BS?

And then you talk about very noisy…

[rf-loop]

Look attached, very noisy, as may the input requires some terminations....

What we see is a noise floor around -140 dB with lots of spurs. The diagram says the y-axis would be dBFS.
Have you ever thought wat this actually means? FS = full sale = 500 µV * 8 = 4 mVpp = 1.4 mVrms.
140 dB below that would be 140 pVrms – spoken out: picovolts!

The same goes for the 2nd plot which claims to show the noise density in nV/√Hz. Do you realize that a passive 50 ohm resistor alone exhibits already 0.9 nV/√Hz, hence numbers at or below that are utter BS?

And then you talk about very noisy…

Applause!

[hpw]

As now in this Siglent BIN task:

1Veff is as 3.0103dBFS ... currently no calibration was used just the samples corrected & amplified by given BIN data.
Nothing more, so the rthz needs to be checked again.

Also keep in mind that rtHz scaling is for NOISE only valid, as peaks gets false values.

hp

[hpw]

A few facts to consider:

At the highest sensitivities like 500 µV/div, there is no advantage having a 12-bit DSO – the ENOB will be less than 8 bits anyway because of the frontend noise, so no significant difference to an 8-bit DSO.

It is different for the 10-bit mode in a SDS2000X Plus, because this is DSP filtering and limits the bandwidth to 125 MHz. Because of this we will see an advantage even at the high sensitivities.

We cannot just zip random noise and hope to achieve a compression factor of 20. My experiments rather hint on a compression factor of 2.

There is a file for the SDS824X HD

Channel 4, 500 µV/div, 50 ohms terminated;
5 ms/div, 100 Mpts, 2 GSa/s;
Full bandwidth = 254 MHz;

Unsurprisingly, its size is about 90 MB:

https://mega.nz/file/zSJn2KAS#l8JMLXh0_uEakIIrVbC-suonbcblvjp59ObsYzixDYg

The Herbie Man plot still with dBFS.

[Performa01]

Okay, let’s see:

A 64 Mpts FFT should result in a frequency step df = 953.67 Hz.

Assuming a proper Flattop window, the RBW (resolution bandwidth) would be about 3.6 kHz.

I don’t know where the peak at about 60 MHz comes from, but let’s assume the real noise floor is shown above 100 MHz, hence about -144 dB “FS”. Since I already know the real noise density is some 2.4 nV/√Hz at higher frequencies (which is pretty good), I can calculate the voltage level there: 2.4 nV * √3600 = 2.4 nV * 60 = 144 nVrms. This happens to be -142.85 dBVrms.

At a setting of 500 µV/div, no one can deny that full scale is inevitably 1.4 mVrms at a grid that is 8 divisions vertically. The noise level above 100 MHz would then be about -80 dBFS.

If the 144 dBFS were true, the reference level would have to be 2.28Vrms = 6.45 Vpp, equivalent to a vertical gain of 806 mV/div…

[rf-loop]

And when we talk about noise, BW is important.
Here SDS3104X HD (same what was used for this 400Mpts noise binary). Note that this measurement is not perfect. Signal is from old (in its time, HP "state of art" generator, today, boat anchor)
I do not know its real flatness. But I know that it is quite sure in its specs. Also cable, around 80cm Suhner M17/84
6MHz - 2GHz sweep (and it was long time manual sweep, thank god this instrument have very good optical encoder with ball bearings... perhaps I need take one step motor for this sweep... it have also automatic hopping sweep but it some times produce high glitches )
But so or so...this is however, sufficient in relation to this matter.

[egonotto]

Hello,

I have a question for rf-loop.

In the SDS3000X HD manual on page 63 it says: "Under the full Memory Depth, SDS3000X HD can still operate at full speed sampling at timebases <= 10 ms/div which provides high resolution captures for long time duration events". Is that correct? If so, 400 Mpts should be available when using CH1 and CH3.
And how large is the segmented memory?

Best regards
egonotto

[rf-loop]

Hello,

I have a question for rf-loop.

1) In the SDS3000X HD manual on page 63 it says: "Under the full Memory Depth, SDS3000X HD can still operate at full speed sampling at timebases <= 10 ms/div which provides high resolution captures for long time duration events". Is that correct?

2) If so, 400 Mpts should be available when using CH1 and CH3.


3) And how large is the segmented memory?

Best regards
egonotto

1. Yes.
2 No.  1Ch 400M, 2Ch 200M for both   3 or 4 Ch 100M for all.

3. It is bit more complex to explain. Because it depends... how many channels and one segment sample points.
(I have not yet made full table with all or most important combinations but perhaps (this is observed) up to total 650Mpts 652.760Mpts 652.400 Mpts when 2 channels (only if 4GSa/s) or 4 channels on but not know if there is still combination what give more. But yes, total max is more than just 400Mps)
ETA:corrections

[hpw]

Well, as it gets. On Audio the sample data comes as dBFS as calibration comes after.

Now on BIN the data comes as in Volts, so calibration has been done before. So different situation(s).

The given pictures are from my upcoming Pro Version. As things changed a lot and internal used calibrations is different.
Mainly changed things for PN measurements (1/f) and new ENF as Enhanced Noise Floor/Filter as getting the noise level even 12dB down & 3D view.

As the incoming data is already in Volts normalized, I have to think about as different concept.

The noise scales gets any down as higher FFT size used. Noise as a rule of dump, without any applied FFT Windows.

Here attached my updated rtHz plots.

Hp

[Performa01]

Sorry, but I'm giving up at this point.

dBFS is universal and not related to audio. And everyone should know what it actually means.
dBU would be used in audio pretty exclusively, but irrelevant here.
FFT without the correct (Flattop) window function is totally worthless.

The noise density plots are incorrect once again - apart from the fact that they cannot show the 1/f noise as long as there is a linear frequency axis.

I've already measured the noise density long time ago, (Reply #1, 5th screenshot), so we can know what the noise density of the SDS824X HD is: 3.0 nV/√Hz at 300 kHz and 2.4 nV/√Hz at 10 MHz.

https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293744/#msg5293744

[hpw]

Sorry, but I'm giving up at this point.

dBFS is universal and not related to audio. And everyone should know what it actually means.
dBU would be used in audio pretty exclusively, but irrelevant here.
FFT without the correct (Flattop) window function is totally worthless.

The noise density plots are incorrect once again - apart from the fact that they cannot show the 1/f noise as long as there is a linear frequency axis.

I've already measured the noise density long time ago, (Reply #1, 5th screenshot), so we can know what the noise density of the SDS824X HD is: 3.0 nV/√Hz at 300 kHz and 2.4 nV/√Hz at 10 MHz.

https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293744/#msg5293744

The normal deal is dBFS as it goes to DAC and comes from a ADC. The digital world do not tell about any scale.

And one more: FFT windows is used to filter out signals, for noise no windows used.

About your:

Here you can see the noise characteristic up to 25 MHz at a RBW of ~90 Hz. In order to prevent aliasing taking effect, the noise measurement has been done of a 20 MHz bandwidth limited input channel. Since this bandwidth limiter is only first order, a digital 30 MHz lowpass filter has been added.

So the given BIN files is without any filtering. Just to fiddle out your calculations and false mentioned scaling.
While had this one once claimed as noise gets any false scaling.

hp

[2N3055]

Hello,

I have a question for rf-loop.

1) In the SDS3000X HD manual on page 63 it says: "Under the full Memory Depth, SDS3000X HD can still operate at full speed sampling at timebases <= 10 ms/div which provides high resolution captures for long time duration events". Is that correct?

2) If so, 400 Mpts should be available when using CH1 and CH3.


3) And how large is the segmented memory?

Best regards
egonotto

1. Yes.
2 No.  1Ch 400M, 2Ch 200M for both   3 or 4 Ch 100M for all.

3. It is bit more complex to explain. Because it depends... how many channels and one segment sample points.
(I have not yet made full table with all or most important combinations but perhaps (this is observed) up to total 650Mpts 652.760Mpts when 2 channels (only if 4GSa/s) or 4 channels on but not know if there is still combination what give more. But yes, total max is more than just 400Mps)
ETA:corrections

I verify 652.760 MPts total (163,19 MPts per CH) with 10kpts length.
It is 159 Mpts with 1Mpts length, per channel.
At very short and very long you get a bit less because of fragmentation...

[hpw]

Sorry, but I'm giving up at this point.

dBFS is universal and not related to audio. And everyone should know what it actually means.
dBU would be used in audio pretty exclusively, but irrelevant here.
FFT without the correct (Flattop) window function is totally worthless.

The noise density plots are incorrect once again - apart from the fact that they cannot show the 1/f noise as long as there is a linear frequency axis.

I've already measured the noise density long time ago, (Reply #1, 5th screenshot), so we can know what the noise density of the SDS824X HD is: 3.0 nV/√Hz at 300 kHz and 2.4 nV/√Hz at 10 MHz.

https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5293744/#msg5293744

Please simple provide the BIN data of your F4 measurements.

While like to go into the deep issue as to see where the false scaling happens.

While you used picture data or given table data as already normalized by ENBW to RBW* what is not the real FFT RBW as

50MHz / 2M samples as 23.81Hz...

This means for me, as this RBW is already multiplied by the FlatTop Window factor of about 3.8.

In other words if you multiply your values by 2, you will come simple in the regions of my values even you use a filtered samples.

So again, please provide this BIN file ... attached my by log scales.

[gf]

...
While you used picture data or given table data as already normalized by ENBW to RBW* what is not the real FFT RBW as

50MHz / 2M samples as 23.81Hz...

This means for me, as this RBW is already multiplied by the FlatTop Window factor of about 3.8.
...

What matters for the noise density calculation is neither the FFT bin spacing, nor the -3dB bandwidth (RBW), but the ENBW.
The ENBW of Matlab's flattopwin is about 3.7703 x FFT bin spacing (or IOW, 3.7703 * sample_rate / num_fft_points).
If you use a different variant of the flattop window, then the factor may be different.

Code: [Select]

>> x = flattopwin(100000,"periodic");
>> x /= sum(x);
>> X = fft(x);
>> sum(X.*conj(X))
ans =  3.7703

Or even simper:

Code: [Select]

>> x = flattopwin(100000,"periodic");
>> mean(x.*x)/mean(x)^2
ans =  3.7703


[Performa01]

A new measurement with attached data file:

https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5373974/#msg5373974

[hpw]

A new measurement with attached data file:

https://www.eevblog.com/forum/testgear/sds800x-hd-review-demonstration-thread/msg5373974/#msg5373974

OK, thank you so far  for the given BIN files.

After using & investigation the given BIN files:

- Important, to use the equal FFT size as 2M, to consider

- My given spectrum's uses peak values

- To get into the given rtHz figures, AVERAGING is the critical part as > 32 times to use

- This requires a high amount of sample given on the BIN file as 2M * 32 = 64M minimum as 100M samples as best size

- Various FFT Window types as: None, Hann, Flatop and any better once (higher orders than 3) .... must deliver equal rthz figures

- Hann = Hanning is a common mistake, as false used NAME, while simple Hanning do not exist at all and was misused for a long time 


Currently large BIN files of the SDS1000 HD, SDS6000A H12, SDS7000A are not posted yet.

The main goal was to get BIN files and hidden to get the performance of those new 12 bit DSO's.

Hp

[gf]

Moved my answer from the Siglent SDS800X HD Review & Demonstration Thread - I think it fits better here.

@Noise Density 2: Any chance to save 100x AVG by 2M as one 200M BIN file size to save?

Of course, I could just capture a long record so that you can play with it – but that I’ve done already and since the total sample memory in the SDS824X HD is just 100 Mpts, I can’t provide anything longer than that.

@hpw: IMO, this should be sufficient. With Welch's method with 80% overlap and 2M window size (FFT size), you end up in averaging 234 spectra anyway. This results in a lower noise floor variance than averaging 100 independently captured 2M traces. [ Btw, ~80% overlap is a good value in conjunction with a flattop window. More overlap does not lead to a significant improvement of the noise floor variance, although even more spectra will be averaged. ]

Don't know if there exists a ready-to-use .bin file loader for Matlab/Octave.
Here is an Octave example with random samples:

Code: [Select]

% create 100M samples of white random noise, 80uV RMS
sample_rate = 2e9
samples = 80e-6 * randn(1,100e6);

% analyze
win = flattopwin(2*1024*1024,"periodic")';
[PSD,f] = pwelch(samples,win,0.8,[],sample_rate);

% plot
semilogy(f,sqrt(PSD))
grid on
ylim([1e-10 1e-4])
xlabel("Frequency [Hz]")
ylabel("V/sqrt(Hz)")


[gf]

I tried a spectral analysis of SDS824X_HD_Binary_C4_1.bin (as posted by Performa01, 100M points, 2GSa/s, 500µV/div) with Welch's method (2M window size, 80% overlap), and it results in a very similar shape as the SDS824X_HD_ND_2GSa_1mV screenshot (although calculated from a different data set).

[ PS: The plot is just calculated from the raw sample values, so the absolute vertical scaling is not yet correct.
I was too lazy yet to find out how to find the scaling factor from raw sample values to Volts in .bin file header. ]

EDIT: Does anybody know the "code_per_div" value for the SDS800X?
The value 25 from the following paper obviously applies only to 8-bit samples.
https://siglentna.com/wp-content/uploads/dlm_uploads/2021/09/How-to-Extract-Data-from-the-Binary-File.pdf
Is it 256*25=6400 for 16-bit samples? (just a guess)

EDIT: Added a plot with scaling factor code_per_div=480*16.
Instead of the average PSD I have plotted the 95% confidence interval to get a feeling for the uncertainty.