Siglent SDG1032X sine distortion at 1 kHz ?

[rf-loop]

My previous image.      




Bullshit warning aka fakenews warning!

I do not believe these all harmonics and/or these harmonics levels are true out from SDG1032X/62X output! 

Real spectrum analyzer image is very different!  Later about this. Now other things keep busy.

[Martin72]

Interesting that it could be "fake".
If you were to add up the thd from your picture, they would probably be in the specification.
In an ideal world, the harmonics would fall continuously after the fundamental wave.
That they do not do so here has several reasons - but that everything you see could then be "fake"...
I'm looking forward to seeing what you pull out of your hat later.

[2N3055]

I don't have SDG1000X/2000X...
This is for SDG6000X

[rf-loop]

Interesting that it could be "fake".
If you were to add up the thd from your picture, they would probably be in the specification.
In an ideal world, the harmonics would fall continuously after the fundamental wave.
That they do not do so here has several reasons - but that everything you see could then be "fake"...
I'm looking forward to seeing what you pull out of your hat later.

Naturally "all" is not fake but...  I do not like oscilloscope generate so much these "fake" harmonics.
So if someone ask/hope that Siglent implement automatic THD measurement.... please do not. Garbage in leads always to garbage out and this garbage we do not need, even if salesmen can add one row to features list..

Due to some reasons I use here more low signal level. Signal level is -20dBm.  (it is also attenuated because it is not designed for under 9kHz so I believe mixer level is not problem)

Then other thing. SSA specs start 9Khz. Using SSA trace B,  I have done some kind of "level correction" curve. Yes this is not now rocket science...  only just for fun and make SDS FFT bit questionable for this purpose. Look these "harmonics" in SDS image specially after 7th . What generate these all peaks. It is sure they do not come from generator - least not even close this level. 

This requires further research. It also requires different methods and tools.


Note: there need correct levels specially under 3kHz. B trace input level was -21dBm (swept with MaxHold). A trace input level -20dBm. Fixed 1kHz -20dBm


CH1  input level -20dBm. Fixed 1kHz Sine -20dBm fom SDG1032/62X
(-20dBm  is 63.3mVpp)

[Calvin]

Hi,

just a number of quickn´dirty measurements of my SDG2042X with my QuantAsylum QA401 24Bit audio analyzer.
Yellow traces are from the analyzer, red traces from the SDG2042X.
Frequencies: 1kHz and 100Hz, Amplitude: 0dBV (note the analyzer´s ADC has its THD-minimum at -16dBV input level)
It shows that the analyzer´s own generator creates a cleaner Signal and at lower noise due to much lower bandwidth.

regards
Calvin

[Martin72]

So if someone ask/hope that Siglent implement automatic THD measurement.... please do not. Garbage in leads always to garbage out and this garbage we do not need

If you (or others) can figure out why the scope is outputting "wrong" harmonics (only in amplitude?), then the THD feature for the FFT function would be one of the most useful features ever.
I know far too little about FFT, but my little layman's understanding finds it remarkable that the higher harmonics look almost mirrored.
It would be interesting to see what our HDO6034A "makes" of the signal with its SA function.
Unfortunately I am on holiday...

[rf-loop]

So if someone ask/hope that Siglent implement automatic THD measurement.... please do not. Garbage in leads always to garbage out and this garbage we do not need

If you (or others) can figure out why the scope is outputting "wrong" harmonics (only in amplitude?), then the THD feature for the FFT function would be one of the most useful features ever.
I know far too little about FFT, but my little layman's understanding finds it remarkable that the higher harmonics look almost mirrored.
It would be interesting to see what our HDO6034A "makes" of the signal with its SA function.
Unfortunately I am on holiday...

Example in my last image FFT full span is 25MHz.
Only over 25MHz signals are mirrored. Of course if there is (ref my image) 49.992 MHz signal it is then displayed as 8kHz and same if there is 50.008MHz signal and so on but using SA I believe these are produced mainly inside SDS unfortunately. With SA I can not find anything what explain these when I look SDG output. Because explanation is not there in SDG signal,  then, imho, one possible guilty is oscilloscope.


Do you think SDG generate these when it is set for 1kHz sine. With spectrum there can see one bit higher spur at 16.4MHz but even it is still inside FFT span and its level is around -85dBm and other spurs are more weak least up to 200MHz.

[mawyatt]

Then other thing. SSA specs start 9Khz. Using SSA trace B,  I have done some kind of "level correction" curve. Yes this is not now rocket science...  only just for fun and make SDS FFT bit questionable for this purpose. Look these "harmonics" in SDS image specially after 7th . What generate these all peaks. It is sure they do not come from generator - least not even close this level. 

This requires further research. It also requires different methods and tools.

We must remember that the DSO FFT is created from data gathered from the ADC, be it 8, 10 or 12 bit core. Whereas the SA is from a swept type measurement. Recall most SA that use the swept mode also feature a log amp which has a large DR, more so than conventional linear amps, and possible input scaling after the passive LPF or BPF in the signal processing chain.

Also, the FFT is not benefiting from the dynamic scaling like we see with the FRA/Bode modes of these DSOs, it just the preamp and ADC at a fixed scale factor, and thus limited by such since the "log" dB scale is just a computation from the ADC raw data and not "preprocessed" before ADC conversion like the Log Amps & Scaling in a conventional SA.

I'm not surprised the FFT shows additional "artifacts" & various different harmonic levels over the conventional SA with it's swept capability.

Agree, the THD may not be an attractive added feature for signals with low THD, likely limited to something on the order of the core ADC and input amplifier linearity, so maybe 40~50dB, and thus not very attractive for any low level distortion characterization.

Like the use of the Trace B correction display to revel the error when the SA is operated below the lower frequency spec limit 

Edit: Should also add that these signals in question here originate from a 14 or 16 bit core DAC, whereas the DSO has a 8, 10 or 12 bit core ADC. In signal processing terms this is bass-ackwards (old engineering term to highlight a backwards condition  ), the measurement should have a higher effective resolution/precision than the signal of interest if one expects the results to be meaningfull regarding the "quality" of the original signal.

Best

[Martin72]

Like the use of the Trace B correction display to revel the error when the SA is operated below the lower frequency spec limit 

You could also choose a frequency that is in the range, 10khz for example.

[mawyatt]

Hi,

just a number of quickn´dirty measurements of my SDG2042X with my QuantAsylum QA401 24Bit audio analyzer.
Yellow traces are from the analyzer, red traces from the SDG2042X.
Frequencies: 1kHz and 100Hz, Amplitude: 0dBV (note the analyzer´s ADC has its THD-minimum at -16dBV input level)
It shows that the analyzer´s own generator creates a cleaner Signal and at lower noise due to much lower bandwidth.

regards
Calvin

Did some similar tests over here at 10KHz with Pico Scope which has a 16 bit ADC.

https://www.eevblog.com/forum/testgear/comparison-between-siglent-sdg1000x-and-2000x/25/

Best,

[electr_peter]

I do not believe these all harmonics and/or these harmonics levels are true out from SDG1032X/62X output! 

You are right to be suspicious. Harmonics are an effect (mostly) of digitization in ADC which creates additional spurs due to "sharpness" of digitization levels and low internal noise in ADC. Additional dither noise before ADC helps to smooth out these spurs and achieve better SFDR.
This is an issue with a scope, not with AWG.

Sources (to name a few):
ADC Input Noise: The Good, The Bad, and The Ugly. Is No Noise Good Noise?
Dithering in Analog-to-digital Conversion
AN-804 Improving A/D Converter Performance Using Dither

[Martin72]

FFT, two different vertical settings:

https://www.eevblog.com/forum/testgear/comparison-between-siglent-sdg1000x-and-2000x/msg4604809/#msg4604809

[electr_peter]

Can you try this measuring setup:

• SDG1032X 1 kHz signal with added controllable random noise level which has RMS of around 1/3-1/2 of LSB (LSB w.r.t. to SDS scope)
• SDS2000X input with FFT

With zero noise FFT looks like a comb (which is BS). With added noise noise floor should rise, but most higher harmonics should disappear. Noise level should be controllable to see the effect. If dither hypothesis is correct, dither noise should help to improve FFT view.
Problem without dither is that averaging does not fully work in time domain, which results in "sharp" corners on a waveform, which are shown as a comb on FFT. If signal is 0.8LSB, ADC will always show 1LSB and average will be 1LSB. With dither, ADC sometimes will show 0LSB and average as correct 0.8LSB.

[rf-loop]

Can you try this measuring setup:

• SDG1032X 1 kHz signal with added controllable random noise level which has RMS of around 1/3-1/2 of LSB (LSB w.r.t. to SDS scope)
• SDS2000X input with FFT

With zero noise FFT looks like a comb (which is BS). With added noise noise floor should rise, but most higher harmonics should disappear. Noise level should be controllable to see the effect. If dither hypothesis is correct, dither noise should help to improve FFT view.
Problem without dither is that averaging does not fully work in time domain, which results in "sharp" corners on a waveform, which are shown as a comb on FFT. If signal is 0.8LSB, ADC will always show 1LSB and average will be 1LSB. With dither, ADC sometimes will show 0LSB and average as correct 0.8LSB.

There is not zero noise in ADC input. There is "lot" of wide band random noise from different parts in front end.
As can see in this image below what include noise before ADC and ADC itself.

Input CH1 no signal (just open BNC, AC 50ohm)
100mV/div. Zoomed in 2mV/div for display noise level. (every horizontal line is one ADC step)
Display have 30s persistence.

Then there is FFT.  (FFT split window mode and then time domain zoomed in (time domain zoom displayed overlayed FFT display area) zoomed in horizontally and vertically zoomed in enough for count ADC steps (in this display area either the time domain scale or the FFT scale is visible, not both)
Upper FFT cursor is 2dBm level. It is same as oscilloscope vertical displayed full scale (if go more details ADC full scale is tiny bit more but nonsense here)
FFT (using these used settings) noise floor top is around -99dB down from full scale. (2dBm to -97dBm)

When look Ch1 noise peak level it is roughly 4.8mVpp  what looks like 24 steps. (if it is just 0.2mV/step then full scale is 819mV (display vertical 800mV+ bit overlap ))

[electr_peter]

Noise is definitely non zero. Does 20MHz limit has an effect on this noise?
This leaves the question what is going on with the FFT comb. DNL error or something funky inside SDS?

[rf-loop]

Does 20MHz limit has an effect on this noise?

Yes, naturally. Not directly just only reduced by BW ratio, there is so many different noise sources and noise is not pure straight white noise coming to then BW filtering.
Example just I measured (very roughly and same 100mV/div and all just as in previous image) full BW 600uVacrms (stdev) and 20MHz BW  450uVacrms for example. Also we know this noise is not flat white noise.

Here nice info: Intersil (Renesas)/ "Understanding Noise in the Signal Chain"

https://www.renesas.com/us/en/document/ppt/understanding-noise-signal-chain

[electr_peter]

Alternative title for this topic and related comparison-between-siglent-sdg1000x-and-2000x:
"Siglent SDG*000X low noise output highlights limitations of Siglent SDS2000X HD scope". Oh, the irony

[2N3055]

Alternative title for this topic and related comparison-between-siglent-sdg1000x-and-2000x:
"Siglent SDG*000X low noise output highlights limitations of Siglent SDS2000X HD scope". Oh, the irony

You are so funny...  You make no sense, but funny....

So it's a limitation that oscilloscope has spurs 20 db below it's native dynamic range... Who knew it works that way...

[electr_peter]

Alternative title for this topic and related comparison-between-siglent-sdg1000x-and-2000x:
"Siglent SDG*000X low noise output highlights limitations of Siglent SDS2000X HD scope". Oh, the irony

You are so funny...  You make no sense, but funny....

So it's a limitation that oscilloscope has spurs 20 db below it's native dynamic range... Who knew it works that way...

Humor is different for everyone.
To be serious, it is clear from specs and experiments that SDG has very low noise and SDS HD is great 12-bit low noise scope. FFT stuff is just a reminder to be careful with readings near noise floor. Just the fact that this nuance gets attention signifies bright future for scopes with 10-12 bit ADCs and low noise AFEs.

[electr_peter]

Simulations showed that ADC dither is not an issue (noise is high enough to hide quantization effect), but small non-linearities in AFE could cause FFT to show "comb". Assuming non-linear behavior is stable over time, it can be estimated by downloading raw time domain samples and comparing with "ideal" input signal. Not sure if at such low distortion level it is possible to do a digital non-linear correction after the fact or whether it is worth the effort, this requires more modeling. Of course, correction would be scope and input setting specific and would require very pure input signal.

FFT view at different input levels and SDS HD scope settings could get small insight at which AFE part causes distortion, maybe there is a sweet spot somewhere.

FFT shows something non-linear from AFE because noise is so low. IMO that's better than  to drown (hide) distortion in high noise floor.

[2N3055]

Simulations showed that ADC dither is not an issue (noise is high enough to hide quantization effect), but small non-linearities in AFE could cause FFT to show "comb". Assuming non-linear behavior is stable over time, it can be estimated by downloading raw time domain samples and comparing with "ideal" input signal. Not sure if at such low distortion level it is possible to do a digital non-linear correction after the fact or whether it is worth the effort, this requires more modeling. Of course, correction would be scope and input setting specific and would require very pure input signal.

FFT view at different input levels and SDS HD scope settings could get small insight at which AFE part causes distortion, maybe there is a sweet spot somewhere.

FFT shows something non-linear from AFE because noise is so low. IMO that's better than  to drown (hide) distortion in high noise floor.

Simulations were not necessarily needed to gain that knowledge. We already knew there is enough noise for dithering.
Nothing wrong with verifying. It is nice when theory and practice agree.
Nonlinearities introduce harmonic and intermodulation distortions, phase delays and deform BW flatness,that is how it works, correct.
Digital non-linear correction is possible in theory but not practical because of sheer volume of real-time data and needed models.
As you go through ranges, various levels of attenuation and PGA gain is used in combination. Any of those combinations and different drive levels would create different distortions.
There are many sweet spots in attenuation ranges, especially if you go from coarse to fine gain settings.

I absolutely agree with last statement (this is what I said before): only reason why people see these spurs is because noise floor is good.

[electr_peter]

Digital non-linear correction is possible in theory but not practical because of sheer volume of real-time data and needed models.
As you go through ranges, various levels of attenuation and PGA gain is used in combination. Any of those combinations and different drive levels would create different distortions.

Checked possibility of digital non-linearity correction, IMO it's not worth the effort.
1) Correction requires non-trivial amount of math operations, proper calibration even more so. 2) Actual distortion is small, but it is non-linear and variable/unstable. Your could be chasing ghosts in AFE during calibration with questionable stability. 3) Adjustment effect is small and almost invisible in time domain for most cases, but it will have some impact on FFT at low signal levels.
Too much effort/cost for small improvement in case of general purpose DSO, better to focus design effort on a more linear AFE instead.

[electr_peter]

So it's a limitation that oscilloscope has spurs 20 db below it's native dynamic range... Who knew it works that way...

In case of FFT spectrum view, number of points in FFT calculations has an effect on noise floor. For \$N = 12\$ bit ADC and \$L = 4096\$ point FFT, FFT noise floor is combination of signal to noise ratio (\$SNR\$) and FFT processing gain
$$FFT\hspace{1ex}noise\hspace{1ex}floor = SNR + processing\hspace{1ex}gain = 107 dB$$
where
$$SNR = 6.02 \cdot N + 1.76 = 74 dB$$
and
$$processing\hspace{1ex}gain = 10 \cdot log_{10}{(L/2)} = 33 dB$$

With bigger number of data points for FFT noise floor is lower. This is related to SA where narrower \$RBW\$ results in lower noise floor. Alternative view is to think about relationship between power spectral density (\$V_{rms}/\sqrt{Hz}\$) and power spectrum (\$dB\$).

Reference: Analog Devices, Taking the Mystery out of the Infamous Formula, "SNR = 6.02N + 1.76dB," and Why You Should Care

[RoV]

Made a test on my SDG2042X (ok, 2122X  ).
Used an E-MU 0202 connected to my PC, acquiring ~20 s at 48000 Hz sampling, 24 bit resolution. Generator at 1 kHz sine, 1 Vpp output. Acquisition gain set to have about 80% level on sin peaks.
Processed with Welch method, using a Blackman window of length 4800 to minimize sidelobes: result in figure, with fundamental normalized at 0 dB.
Highest harmonic is 2nd, at -96.3 dB. Then 3rd is at -105.5, 4th at -105.2, 5th at -105.9.

[RoV]

Test repeated with a lower (~half) gain setting on the E-MU 0202, because I was suspicious about the 2nd harmonic.
Now the 2nd is at -103.5 dB, the 3rd as before. Worst is the 10th at -101.9 dB.
THD, computed up to the 10th harmonic, is -97.5 dBc, or 0.0013%.