Measuring Distortions with the Scope:What you see is not what you really have..

[Martin72]

Hi,

Note that you are trying to characterize a 16 bit generator with a 12 bit oscillscope.

I realise that, but the scope should not show anything that is actually not there, and that is the current situation.
And I don't think it's because you get a 16 bit signal with a 12 bit scope(the very most user got 8bit ones).
Approaches have already been shown as to what the problem could be, and that worries me at least.
Not that you generally have to doubt everything that the FFT function provides or that you first have to make an increased effort instead of just roughly adjusting it and that's it.

A very interesting and important topic, it should be separated from this thread, otherwise you won't find it again so quickly.
Regarding the original topic here:
And then I wonder what else you can compare between the models that doesn't need to be doubted because you did it with a scope?

[rf-loop]

One of the findings one could derive from this would be of a fatalistic nature, namely that one can forget the FFT function with oscilloscopes.
Is this really the case, or do you have to try a little harder to get a more credible result?
And if so, in which direction one would have to go for it....

Not at all, in the same way that the frequency measurement function is not useless just because it does not have the same resolution and accuracy as a standalone frequency counter. An oscillscope is not a replacement for an audio analyzer (neither is a spectrum analyzer btw).

Every instrument creates some distortion of its own (not only in the ADC but more importantly the front-end as well). It will depend on the frequency and amplitude of the input signal and the vertical setting as well. And unfortunately, it is usually poorly characterised in oscilloscopes (the SDS2000X HD datasheet only says SFDR >= 45dBc which is not very helpful). Once you get close to those distortion limits, the error bars increase. The distortion of the generator and the receiver do not necessarily add in a straightforward way either. In some cases they may cancel partially and a noisier input signal may show less distortion because of dithering effects.

The SDG2000X has pretty clean output in the audio range (significantly better than the datasheet specs at most settings). Almost all oscilloscopes will struggle there, with some exceptions like the remarkable but specialized PicoScope 4262. Note that you are trying to characterize a 16 bit generator with a 12 bit oscillscope. Now that doesn't tell you anything definite about linearity at all but IMHO it should give you pause and make you proceed with careful consideration and healthy scepticism.

What can you do to improve your measurements? First, know your instrument and its limitations. You can try to measure an oscillator that is known to have distortion levels significantly better than your oscilloscope as a reference. That also allows you to find the settings that give you the best results. A quick sanity check is to add some in-line attenuation (on a spectrum analyzer, change the input attenuator setting). If the relative levels of the harmonics change, you will know that the distortion is from the receiver.

You can also add an external notch filter to attenuate the fundamental. This will enable you to measure down to very low distortion levels.

Well said!

Also example I have used several kind of spectrum analyzer over tens of years. Most (all) of them display "fake" signals.  Non input related spurs and then input related. Look generator harmonics with SA and wonder what harmonics are generated or affected due to SA own generations... and so on. Simple: There is not ideal instruments on this Tellus, not in history, not now and never in future. Every instrument display sum of errors mixed with some kind of truth what we do not know.
With FFT display human eye see these so... omg...so big false peaks. User need just understand and just know his instrument limits and lies and not believe everything.

[Martin72]

OK, so we have the philosophical approach already.
Do not trust what you see, use your imagination.
Btw, this is what it should look like to me, a 1Khz square wave, steadily decreasing harmonics.

[rf-loop]

OK, so we have the philosophical approach already.
Do not trust what you see, use your imagination.
Btw, this is what it should look like to me, a 1Khz square wave, steadily decreasing harmonics.

Yep, roughly so.

You can now zoom in and look just 577th harmonic and check 576th and 578th... then measure these and tell "truth" and think why these levels are just right or how they are... (not seriously...  )

[2N3055]

Hi,

Note that you are trying to characterize a 16 bit generator with a 12 bit oscillscope.

I realise that, but the scope should not show anything that is actually not there, and that is the current situation.

Well scope doesn't show anything that is not there in normal time domain mode.....

FFT is math mode... In math mode there are no limits on result.
You can make math channel where you multiply a signal by 10E6 and you will get weird stuff.

Like others said, with SA you need to characterise for spurs to know if signal peaks are real.

Whole thing stems from the fact that input noise is really low. On my Keysight 3104T noise floor is so high you cannot see spurs at -60dBm

[rf-loop]

Hi,

Note that you are trying to characterize a 16 bit generator with a 12 bit oscillscope.

I realise that, but the scope should not show anything that is actually not there, and that is the current situation.

FFT is math mode... In math mode there are no limits on result.
You can make math channel where you multiply a signal by 10E6 and you will get weird stuff.

Like others said, with SA you need to characterise for spurs to know if signal peaks are real.

Whole thing stems from the fact that input noise is really low. On my Keysight 3104T noise floor is so high you cannot see spurs at -60dBm

In this regard, I just previously suggested using paper or a piece of fabric to make curtains. Then just close a curtain in front that covers the things that shouldn't be taken so seriously. Of course, one could also make the curtains in another way... like Keysight, for example. This way you can conveniently cover it up and no one will make a fuss about "strange" signal spikes somewhere deep...
Maybe someone should suggest that... you could easily implement a noise generator there to satisfy the users, because then the lies will be removed --- oops --- hidden.

[Detlev]

Hello everyone,

for comparison I made an FFT of my SDG6022x with the RTB2004 (RTB2K-COM4). The values of the 6022 are very similar to those of the 2122X



Many greetings
Detlev

[Martin72]

Like the comment function of the RTB...

By the way, I remembered that I still have something on the shelf with which I can measure THD.
My Neutrik A1....
With this puppy it should be possible to get "more trustful" values.

[mawyatt]

Hello everyone,

for comparison I made an FFT of my SDG6022x with the RTB2004 (RTB2K-COM4). The values of the 6022 are very similar to those of the 2122X

(Attachment Link)

Many greetings
Detlev

Thanks for the plots!!

If you check the earlier posts we provided, it's likely that the DSO is contributing more of the higher frequency harmonics than either AWG. This is because these AWGs provide pretty good signal fidelity from 16 bit DACs and thus more so than the typical general purpose bench top DSO can cleanly resolve.

Best,

[Performa01]

My bad – In my initial comments I simply forgot that the linearity of the average scope frontend (including SDS2000X HD) is not up to the task of characterizing the harmonic distortion of an SDG1000X or 2000X.

One of the findings one could derive from this would be of a fatalistic nature, namely that one can forget the FFT function with oscilloscopes.
Is this really the case, or do you have to try a little harder to get a more credible result?
And if so, in which direction one would have to go for it....

If you look up the datasheet of PGAs that can be used for a scope like this, such as LMH6518, you will find the harmonic distortions specified somewhere between -44 and -50 dBc. Considering this, the SDS2000X HD still does a fairly decent job.

The FFT is still useful for many tasks, including distortion measurements within the usual range of interest, which hardly ever exceeds -60 dBc except for high end audio. There is a reason, why specialized audio analyzers exist.

You just have to take into account that the SFDR of a general purpose scope frontend might be only about 60 dB in practice – many (if not most) of the more affordable old SA boat anchors haven't been much better btw.

For instance, there is no restriction for single tone narrowband measurements, like phase noise or modulation spectra, where especially the SDS2000X HD will give you an exceptional dynamic range that clearly exceeds the ~72 dB that you could expect from the 12 bits.

The THD of a general purpose oscilloscope frontend is usually not specified in the datasheet. On the other hand, the Picoscope 4262 has a guaranteed linearity of 16 bits – but that is certainly not a general purpose instrument, with its 16 bit converters and limited bandwidth of just 5 MHz. Yet this is the way to go if you need to measure down to -96 dBc and don't want to be restricted to the audio frequency range like with soundcards.

[mawyatt]

Why we've often requested & performed the classic Two Tone IMD to evaluate DSO linearity. Also why we have the PicoScope 4262, altho don't care much for instruments that require a PC to function (and prefer Macs anyway), and because of this the 4262 rarely sees use!!

Since these modern DSOs are venturing outside the old oscilloscope measurement realm, we are somewhat witnessing the effects of an instrument designed to display a waveform result of maybe 40~50dB on screen range. Whereas a proper Spectrum Analyzer is designed to display a 100~120dB range, big difference with intended usage and on screen Dynamic Range between these instruments.

With this in mind we are seeing the modern DSO becoming a more versatile instrument than a Time Domain waveform display device (O-Scope), and the OEMS are addressing such by providing higher resolution core ADCs, Digital Logic Analyzer capabilities and such.

As Performao1 indicated the DSO input PGA, along with the core ADC are the likely culprits limiting linearity, and for the intended usage (Time Domain Linear Display) provide very respectable performance, however when pressed into the Frequency Domain (Log Display) that begins to show limitations. We must remember that the DSO FFT displays are just a mathematical representation of the Time Domain captured raw data from the ADC and thus limited by the mentioned PGA and ADC channel linearity and noise.

As we see better performing lower cost ADCs and PGAs ICs come available, I'm sure we'll see better FFT performance like the Pico Scope 4262 at much higher frequency capable stand alone bench DSOs, and at affordable prices. Maybe we'll see some new type ADC architectures emerging from the research channels, we know of one (Non-Uniform Sampling), but that's another topic as we've already diverted from the OP topic.

Best,

[Detlev]

Hello everyone,

I have here a comparison of three FFT analyses. The SGD6022 is always set identically. Once I measured without an attenuator and then one with 10dB and 20dB attenuator.

I would expect the spectral distribution to be identical for all three FFT measurements since the AWG was not changed. However, we see three different FFTs.

From my point of view, this underlines that a DSO is only conditionally suitable for these low levels of harmonics.

Many greetings
Detlev

[mawyatt]

Why we've often requested & performed the classic Two Tone IMD to evaluate DSO linearity......

Here's using the SDG2000X AWG with the Wave Combine feature producing the Two Tone IMD mentioned above. AWG Ch1 is set to 0dBm at 99.5KHz, Ch2 0dBm at 100.5KHz.

The SDS2000X+ was set to 50 ohm input Z and the FFT performed, see PNG 153.

Then the special purpose PicoScope 4262 was used with an external 50 ohm terminator added, see Pico_AWG4.

Note the added "artifacts" added by the SDS relative to the PS, and also note the difference in the IMD products as shown.

This clearly shows that a typical DSO is not good enough to show the waveform quality of a quality AWG (get similar results with SDG6000X). A special purpose DSO is required to revel the signal quality of these 16 bit DAC based AWGs, even when in this case we were using the Wave Combine feature of the SDG2000X (and 6000) which likely adds a little non-linearity to the resultant waveform. Does anyone know if this combining is done in the digital domain (pre DAC) rather than at the analog post DAC output?

Would expect the HD version of the SDS2000X+ with it's core 12 bit ADC to have a better result, and interested to "See" how well the new Rigol 12 bit ADC DSO behaves (we asked awhile ago for a Two Tone IMD, but no now has offered yet).

Edit: Added a result from our Spectrum Analyzer with a 10dB PAD in front (almost always keep a PAD on front end, saves replacing SA front ends), see PNG18. The SA amplitude hasn't been calibrated, but the relative IMD readings are good. Note the lack of artifacts and the level of the IMD Products. It's interesting that the PicoScope shows even better IMD results than the SA, a tribute to the PicoScope and of course the SDG2000X+ AWG signal source!!

Best,

[Martin72]

Maybe I should give this thread a new name and start a new comparison thread..
Serious, this topic about the limits of a scope is too important IMO..

[mawyatt]

Well it is showing just how good these AWGs are 

Agree tho, the DSO FFT issue certainly belongs in it's own thread.

Best

[baldurn]

Edit: Added a result from our Spectrum Analyzer with a 10dB PAD in front (almost always keep a PAD on front end, saves replacing SA front ends), see PNG18. The SA amplitude hasn't been calibrated, but the relative IMD readings are good. Note the lack of artifacts and the level of the IMD Products. It's interesting that the PicoScope shows even better IMD results than the SA, a tribute to the PicoScope and of course the SDG2000X+ AWG signal source!!

How do we know that the extra IMD products shown on the Pico are real?

[mawyatt]

Edit: Added a result from our Spectrum Analyzer with a 10dB PAD in front (almost always keep a PAD on front end, saves replacing SA front ends), see PNG18. The SA amplitude hasn't been calibrated, but the relative IMD readings are good. Note the lack of artifacts and the level of the IMD Products. It's interesting that the PicoScope shows even better IMD results than the SA, a tribute to the PicoScope and of course the SDG2000X+ AWG signal source!!

How do we know that the extra IMD products shown on the Pico are real?

We don't, but they are lowest (~ -100dBc ref to 0dBm signal peaks, noise floor ~ -115dBm)) of the group of tests, so would baseline them as the resultant of the PS and AWG. Please note the SA doesn't show some of these artifacts, but the PS shows some that are below the SA noise floor ( ~ -103dBm, signal peaks is ~ -11dBm due to input PAD). Since we can't "assume" the PS contributed these, truth would be of course both contributed but we can't separate either, so safe assumption it's the AWG if one is looking for the signal source "quality".

Of course if we only had the general purpose DSO results, then this would "paint" an entirely different picture of the AWG, since without another "good" measurement instrument we couldn't defer/question the DSO results. Please remember in engineering measurements, one is always looking for a source significantly better than the measuring instrument, this helps to "verify" that the measuring instrument is good for the task. Also applies to evaluating a DUT, where one wants the "input signal" to be significantly better than DUT, so the end resultant measurement is mostly due to the DUT effects and not the poor quality input signal (or measuring instrument).

Best,

[baldurn]

IMD is very common in receivers so I could easily see all three instruments generating those IMDs internally. Maybe generate and measure each signal independently before combining them externally using a passive power combiner. If that shows the same IMD products it should be the measuring instrument doing it.

I would try it but I do not have a power combiner available.

[mawyatt]

Agree, the resistive combiner should produce the best overall IMD from the signal various sources. However, these AWGs under discussion have a built-in combiner, whether it's digital or analog don't know, but produces a Two Tone Signal without any additional components or cables, just one cable, and produces very respectable results (better than we expected) as shown by the various previous plots.


As always YMMV,

Best

[blackdog]

Hi,

I did some measurements on the function generators available here in my lab.

I have limited myself to the audio region at 2V RMS output at a 50 Ohm load.
These THD measurements were done with Sine signals, if time is available, I will show from the generators also the square wave representation, but I will do that in a separate post.
All equipment used had been on for at least 1 hour before I started measuring.

These types of measurements take a fair amount of time and I have to do this in between my work. :-)

The equipment used:
Hameg /R&S:   HMF2525
Siglent:            SDG1032
Siglent:            SDG2042
Rigol:               DG4162

The scope used for the THD pictures is a Hameg /R&S: HMO3004
And the THD meter is an Audio Precision: Portable One Plus
The spectrum picturers are made with a Siglent SDS2104 Plus

First the SDG1032X
14Bit Generator
This is the display of the Audio precision Analyser and it shows the measured THD at a bandwidth of 400Hz to 80KHZ, rounded off the THD is 0.07%.



.
This scope picture shows the measured 10KHz Sinus in the color yellow and the green trace is the residue of the Analyzer output.
Do not look at the size of the green trace, this is because the analyzer output is auto scale.
I only show the green trace to get an idea of what the THD looks like.


.
The following picture shows what the FFT looks like done with the Siglent scope.
Keep in mind that the table always varies a bit, especially at the somewhat higher harmonics.
Keep in mind the table value of -+1dB.



********************************************************************************************************************************
********************************************************************************************************************************

Now the Siglent SDG2042X
16Bit Generator
This is the display of the Audio Precision Analyser and it shows the measured THD at a bandwidth of 400Hz to 80KHZ, rounded off the THD is 0.007%.


.
This scope picture shows the measured 10KHz Sinus in the color yellow and the green trace is the residue of the Analyzer output.
Do not look at the size of the green trace, this is because the analyzer output is auto scale.
I only show the green trace to get an idea of what the THD looks like.


.
The following picture shows what the FFT looks like done with the Siglent scope.
Keep in mind that the table always varies a bit, especially at the somewhat higher harmonics.
Keep in mind the table value of -+1dB.


********************************************************************************************************************************
********************************************************************************************************************************

Now the Hameg HMF2525
14Bit Generator

This is the display of the Audio Precision Analyser and it shows the measured THD at a bandwidth of 400Hz to 80KHZ, rounded off the THD is 0.023%.


.
This scope picture shows the measured 10KHz Sinus in the color yellow and the green trace is the residue of the Analyzer output.
Do not look at the size of the green trace, this is because the analyzer output is auto scale.
I only show the green trace to get an idea of what the THD looks like.


.
The following picture shows what the FFT looks like done with the Siglent scope.
Keep in mind that the table always varies a bit, especially at the somewhat higher harmonics.
Keep in mind the table value of -+1dB.



********************************************************************************************************************************
********************************************************************************************************************************

Now the Rigol DG4162
14Bit Generator

This is the display of the Audio Precision Analyser and it shows the measured THD at a bandwidth of 400Hz to 80KHZ, rounded off the THD is 0.011%.


.
This scope picture shows the measured 10KHz Sinus in the color yellow and the green trace is the residue of the Analyzer output.
Do not look at the size of the green trace, this is because the analyzer output is auto scale.
I only show the green trace to get an idea of what the THD looks like.


.
The following picture shows what the FFT looks like done with the Siglent scope.
Keep in mind that the table always varies a bit, especially at the somewhat higher harmonics.
Keep in mind the table value of -+1dB.



********************************************************************************************************************************
********************************************************************************************************************************

Now the Audio Precision: Portable One Plus - Generator
Analog Generator

This is the display of the Audio Precision Analyser and it shows the measured THD at a bandwidth of 400Hz to 80KHZ, rounded off the THD is 0.0008%.
THD with my PC and a verry good Soundcart is about 0.0002.


.
This scope picture shows the measured 10KHz Sinus in the color yellow and the green trace is the residue of the Analyzer output.
Do not look at the size of the green trace, this is because the analyzer output is auto scale.
I only show the green trace to get an idea of what the THD looks like, this is almost all noise...


.
The following picture shows what the FFT looks like done with the Siglent scope.
Keep in mind that the table always varies a bit, especially at the somewhat higher harmonics.
Keep in mind the table value of -+1dB.



.
And what does the last picture, which is the FFT from de AP generator on the Siglent Scoop show us now....
The Siglent scope I used, does not have sufficient capabilities to make good FFT measurements below say 50dB.

The slightly newer 12Bit Siglent series scoops will do 15 to 20dB better I think, the SDS2104X is a basic 8Bit scoop with 10Bit oversampling.

Later I will show some pictures of the square wave the generators can make.

Kind regards,
Bram

[Martin72]

As I said, I should rename the thread..
But nice work, tomorrow when the XLR plugs arrived I´ll follow with the neutrik A1.
And the scope FFT pictures are nearly same looking...

[tautech]

As I said, I should rename the thread..

You can, just go back to the OP and do it there with a Modify.
The topic URL remains the same and we will all see new posts flags and/or get notifications if we are setup for them.

[Martin72]

I know that, but I have no idea yet what the new title might be.
Currently, two threads are "hijacked" by the FFT thing..

[rf-loop]

I know that, but I have no idea yet what the new title might be.
Currently, two threads are "hijacked" by the FFT thing..

Least one reason is here:

Back to topic, comparison...

FFT of SDG1062X and 2122X, sinewave 1khz, appx 0dBm @50Ohm:
[  images ]

I think it was very important to handle this FFT thing because this kind of instruments FFT is wrong tool for evaluating these said generators signal quality by measuring, for example, THD and comparing them.
The result is almost garbage.
Which has now been proven in this context as well.

[Performa01]

Agree, the resistive combiner should produce the best overall IMD from the signal various sources. However, these AWGs under discussion have a built-in combiner, whether it's digital or analog don't know, but produces a Two Tone Signal without any additional components or cables, just one cable, and produces very respectable results (better than we expected) as shown by the various previous plots.

The question about the combining method has been dealt with multiple times already, see for example reply #159 here:

https://www.eevblog.com/forum/testgear/two-tone-test-with-scope-and-sa/msg4250476/#msg4250476

I am amazed about your stellar results. Maybe Siglent have been able to achieve some improvement for this function since the last time I've checked it...