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

[pdenisowski]

I'll post images from actual measurements if anyone here is interested.

I'm interested   Always amazed at the different ways people find to use XY mode

[nctnico]

Several posts in this topic display the FFT harmonics from the output signal of a "function generator." Typically the ratio of fundamental/harmonics shown is no better than -50dB for function generators. ...

Thank you so much for giving us valuable insight in the obscure field of audio engineering.  We learn something new every day!

In case you did not understand Tautechs response: This thread originally was (and somehow still is) about the distortion in AWGs (Arbitrary Waveform Generators), which have been shown to provide decent distortion levels down to -80 dBc.

In case you did not read the posts within this thread: unsurprisingly, it has been found that the linearity of the frontend of an average general purpose DSO cannot keep up with the modern AWGs, hence we get much worse distortion figures than what the generators are actually capable of.

After doing some testing (with a Tektronix AFG31000 ) the question is more like: how useful is long FFT where it comes to showing valid results or not. I did some testing with the RTM3004 which is limited to 128kpoints. As a result the FFT noise floor is about -80dB and shows no distortion products. On my GW Instek with 1Mpts FFT (resulting in a lower noise floor) I can see the front-end distortion products just fine.

[gf]

After doing some testing (with a Tektronix AFG31000 ) the question is more like: how useful is long FFT where it comes to showing valid results or not. I did some testing with the RTM3004 which is limited to 128kpoints. As a result the FFT noise floor is about -80dB and shows no distortion products. On my GW Instek with 1Mpts FFT (resulting in a lower noise floor) I can see the front-end distortion products just fine.

Even if you consider anything below (say) -50dBc as potentially invalid, a long FFT is still useful if you need to separate closely spaced frequencies.

[2N3055]

Several posts in this topic display the FFT harmonics from the output signal of a "function generator." Typically the ratio of fundamental/harmonics shown is no better than -50dB for function generators. ...

Thank you so much for giving us valuable insight in the obscure field of audio engineering.  We learn something new every day!

In case you did not understand Tautechs response: This thread originally was (and somehow still is) about the distortion in AWGs (Arbitrary Waveform Generators), which have been shown to provide decent distortion levels down to -80 dBc.

In case you did not read the posts within this thread: unsurprisingly, it has been found that the linearity of the frontend of an average general purpose DSO cannot keep up with the modern AWGs, hence we get much worse distortion figures than what the generators are actually capable of.

After doing some testing (with a Tektronix AFG31000 ) the question is more like: how useful is long FFT where it comes to showing valid results or not. I did some testing with the RTM3004 which is limited to 128kpoints. As a result the FFT noise floor is about -80dB and shows no distortion products. On my GW Instek with 1Mpts FFT (resulting in a lower noise floor) I can see the front-end distortion products just fine.

-80dB noise floor? You mean dynamic range?
At what sensitivity and signal level?

On SDS2000X HD noise floor is approaching -140dBm..

On following capture dynamic range aproaches 100 dB (-5dBV minus -103dBV)
Same number of points..
No wonder nonlinearities are seen...
If dynamic range were 80 dB they also would not be seen..

[JeremyC]

Back to topic, comparison...

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

SDG2122X:



SDG1062X:

I’m new to the forum, however I watch it since ~2019.
In my opinion many people don’t try to understand FFT backgrounds and they do misinterpreting the results. Somebody should create new topic like “FFT good practices / primer”, independent of the equipment vendor. FFT can be confusing… FFT windows may sound confusing, almost each vendor has own interpretation of FFT, etc… But all are base on the same math formulas.
For beginners:
  - Check your DSO ADC bits, and if:
   - it’s 8 bit, ignore everything below 48dB in respect to dbV should be ignored.
   - it’s 10 bit, ignore  everything below 60dB in respect to dbV should be ignored.
   - it’s 12 bit, ignore  everything below 72dB in respect to dbV should be ignored.
   - it’s 14 bit, ignore  everything below 84dB in respect to dbV should be ignored.
   - it’s 16 bit, ignore  everything below 96dB in respect to dbV should be ignored.
… and so on.

- Use the vernier and adjust that your wave to covers 90 - 95% of the p-2-p signal on the screen.
- Don’t use any kind of averaging, ERES, highres, etc.
- Use AC coupling (most scopes will be fine with frequencies > 30Hz in AC coupling, check yours).
- Use plenty of samples… it will depend on your scope.
- Ignore myths...

As good start I suggest to check THD + Scope published by W2AEW about 3 years ago:



If anybody is in high-end audio, you should ignore oscilloscope and use good quality ADC/DAC combined with software like Audio Rightmark, ARTA, or REW suite. ...Or buy specialized equipment, however it may come pricey…

[mawyatt]

To evaluate the DSOs ability to properly represent the THD of a waveform, rather than use a waveform of unknown low THD (sine-wave for example), use a waveform with known THD such as a square-wave or triangle-wave. Altho these waveforms may have an uncertainty in creation from the generator like the sine-wave they have a THD that can be computed theoretically and not ideally zero, so the results from a DSO FFT evaluation can be compared to the theoretical THD of the applied waveform.

Because these waveforms naturally possess higher levels of controlled THD compared to a uncontrolled abet low THD sine-wave, they can be measured by the DSO with the lower resolution ADCs & performance Input amplifiers/attenuators within their span of limited performance.

Here's a few THD for common waveforms derived from this IEEE paper:

https://www.researchgate.net/publication/260672713_Analytic_Method_for_the_Computation_of_the_Total_Harmonic_Distortion_by_the_Cauchy_Method_of_Residues

Square-wave       THD = Sqrt[(pi^2)/8 -1] ~ 48.3%
Triangle-wave      THD = Sqrt[(pi^4)/96 -1] ~ 12.1%
Sawtooth-wave    THD = Sqrt[((pi^2)/6 -1} ~ 80.3%

Pulse-wave          THD = Sqrt[((D(1-D)*(pi^2))/2(sin^2(pi*D))) -1] which varies from 48.3% to 191% as D (Duty Cycle) varies from 0.5 to 0.9, or 0.1 as THD function is symmetrical about D = 0.5 (Square-wave).

We are unable to verify our DSO FFT against these theoretical waveform THD readings at this time, so others may want to utilize this information and report results. Hopefully we'll be able to perform these measurements ourselves and report the results soon.

Of course this introduces another "uncertainty" with the waveform "quality", some of the modern AWGs likely are good sources at lower frequencies (as shown with sine-waves), and we do have an accurate by design custom low frequency squarewave source with precise duty cycle and amplitude available.

Anyway, hopes this helps with the FFT evaluations regarding THD.

As always, YMMV.

Best,

[_Wim_]

- Don’t use any kind of averaging, ERES, highres, etc.
....
As good start I suggest to check THD + Scope published by W2AEW about 3 years ago:

I think you need to re-watch the referenced video by W2AEW (#65: Basics of using FFT on an oscilloscope), because it perfectly makes sense to use HI-RES or ERES to increase the effective number of bits. This way an 8-bit scope can have a larger effective number of bits at a much lower bandwidth.

A good comparison between ERES and HI-RES can be found here: https://teledynelecroy.com/doc/differences-between-eres-and-hires

[markone]

- Don’t use any kind of averaging, ERES, highres, etc.
....
As good start I suggest to check THD + Scope published by W2AEW about 3 years ago:

I think you need to re-watch the referenced video by W2AEW (#65: Basics of using FFT on an oscilloscope), because it perfectly makes sense to use HI-RES or ERES to increase the effective number of bits. This way an 8-bit scope can have a larger effective number of bits at a much lower bandwidth.

A good comparison between ERES and HI-RES can be found here: https://teledynelecroy.com/doc/differences-between-eres-and-hires

The point is that any math process that extend dynamic range with ADC sampling add artifacts and alter BW, so what JeremyC is saying makes perfect sense if you want to be sure  to no catch phantom FFT peaks or miss real ones with BW cut.

I personally agree with that, while I often use ERES equivalent functions in YT mode I have no interest to see FFT fake deep dynamic ranges in 8-12 bits scopes, I find all those small false frequency components quite distracting and when I see 100-120dB FFT screenshot taken from ordinary scopes I make a smile.

[elecdonia]

After doing some testing (with a Tektronix AFG31000 ) the question is more like: how useful is long FFT where it comes to showing valid results or not. I did some testing with the RTM3004 which is limited to 128kpoints. As a result the FFT noise floor is about -80dB and shows no distortion products. On my GW Instek with 1Mpts FFT (resulting in a lower noise floor) I can see the front-end distortion products just fine.

Even if you consider anything below (say) -50dBc as potentially invalid, a long FFT is still useful if you need to separate closely spaced frequencies.

This earlier post describes FFT measurement of the AWG signal from a Siglent which worked out very well, clearly displaying harmonic amplitudes -100dB below the fundamental amplitude. OP lists most of the measurement parameters.
     https://www.eevblog.com/forum/testgear/comparison-between-siglent-sdg1000x-and-2000x/msg4615288/#msg4615288

In my opinion the use of 24-bit A-D to analyze the signal from the AWG is significant. Until recently 24-bit A-D converters were nearly unobtainable at any price. But today they are mass-produced and affordable. My thinking is that having 24-bit resolution for the analysis system is valuable.

[_Wim_]

The point is that any math process that extend dynamic range with ADC sampling add artifacts and alter BW, so what JeremyC is saying makes perfect sense if you want to be sure  to no catch phantom FFT peaks or miss real ones with BW cut.

I personally agree with that, while I often use ERES equivalent functions in YT mode I have no interest to see FFT fake deep dynamic ranges in 8-12 bits scopes, I find all those small false frequency components quite distracting and when I see 100-120dB FFT screenshot taken from ordinary scopes I make a smile.

I agree that the very low noise floor shown is artifact, but if you take into account the effective number of bits (including the bits created by enhancement from HIRES or ERES), than the actual dynamic range can be calculated (20*log(2^enob)) and is in my opinion "real".

Many high resolutions ADCs use oversampling to achieve their high resolution, and for me oversampling and HIRES/ERES are essentially the same. Bandwidth reduction and bit-enhancement can be seen from the attached tables (copied from the Lecroy explanation)

So for a 10 bit scope a 3 bit enhancement is certainly possible for these low frequency ranges in the posts above, giving around 78db of dynamic range. Anything below that must indeed be taking with a grain of salt, but I just wanted to point out that an 8-bit scope has more that 48db of dynamic range...

[mawyatt]

To evaluate the DSOs ability to properly represent the THD of a waveform, rather than use a waveform of unknown low THD (sine-wave for example), use a waveform with known THD such as a square-wave or triangle-wave. Altho these waveforms may have an uncertainty in creation from the generator like the sine-wave they have a THD that can be computed theoretically and not ideally zero, so the results from a DSO FFT evaluation can be compared to the theoretical THD of the applied waveform.

Because these waveforms naturally possess higher levels of controlled THD compared to a uncontrolled abet low THD sine-wave, they can be measured by the DSO with the lower resolution ADCs & performance Input amplifiers/attenuators within their span of limited performance.

Here's a few THD for common waveforms derived from this IEEE paper:

https://www.researchgate.net/publication/260672713_Analytic_Method_for_the_Computation_of_the_Total_Harmonic_Distortion_by_the_Cauchy_Method_of_Residues

Square-wave       THD = Sqrt[(pi^2)/8 -1] ~ 48.3%
Triangle-wave      THD = Sqrt[(pi^4)/96 -1] ~ 12.1%
Sawtooth-wave    THD = Sqrt[((pi^2)/6 -1} ~ 80.3%

Pulse-wave          THD = Sqrt[((D(1-D)*(pi^2))/2(sin^2(pi*D))) -1] which varies from 48.3% to 191% as D (Duty Cycle) varies from 0.5 to 0.9, or 0.1 as THD function is symmetrical about D = 0.5 (Square-wave).

We are unable to verify our DSO FFT against these theoretical waveform THD readings at this time, so others may want to utilize this information and report results. Hopefully we'll be able to perform these measurements ourselves and report the results soon.

Of course this introduces another "uncertainty" with the waveform "quality", some of the modern AWGs likely are good sources at lower frequencies (as shown with sine-waves), and we do have an accurate by design custom low frequency squarewave source with precise duty cycle and amplitude available.

Anyway, hopes this helps with the FFT evaluations regarding THD.

As always, YMMV.

Best,

Got a chance for a quick measurement. This is with SDS2000X+ in 10-bit mode using SDG2042X for waveforms.

Triangle-wave at 5VPP, THD 12.0% (12.09%) using first 19 harmonics (Theoretical 12.1%), using 50 harmonics 12.10%
Square-wave at 5VPP, THD 46.5% (46.64%) using first 29 harmonics (Theoretical 48.3%), using 50 harmonics 47.42%, 100 harmonics 48.14%
Sawtooth-wave at 5VPP 100 harmonics 80.05% (Theoretical 80.3%)
Pulse-wave 25% Duty Cycle at 5VPP, 100 harmonics, THD 91.99% (Theoretical 92.23%)

We could see the results converging towards a slightly higher THD as we added more harmonics in the measurements, which is to be expected as the higher order harmonics are being truncated (not included) which yields a lower overall THD rendering. We also confirmed the Square-wave result with the mentioned precision custom source. The squarewave error is more susceptible to harmonic truncation because the harmonics fall of as 1/f, whereas the triangle wave falls off as 1/f^2.

Anyway, this gives us a little more confidence in the DSO FFT ability to render a reasonable THD result for waveforms with higher THD, curious what others find utilizing these waveforms.

Best,

Edit: Added results from (PicoScope 4262) with 16-bit core ADC.

[mawyatt]

Many high resolutions ADCs use oversampling to achieve their high resolution, and for me oversampling and HIRES/ERES are essentially the same. Bandwidth reduction and bit-enhancement can be seen from the attached tables (copied from the Lecroy explanation)

So for a 10 bit scope a 3 bit enhancement is certainly possible for these low frequency ranges in the posts above, giving around 78db of dynamic range. Anything below that must indeed be taking with a grain of salt, but I just wanted to point out that an 8-bit scope has more that 48db of dynamic range...

Having been involved in ADC chip developments in the past, it's funny when folks think an ADC chip is the single conversion type one often refers to as a "Flash" single step conversion, when they really are an oversampled multi-step conversion type. The prior mentioned 24-bit resolution types are massively oversampled Delta-Sigma types, which are very CMOS friendly and yield nicely to CMOS advanced technologies, and why they are so reasonably priced (small chip size). However, no free lunch as they tend to operate at lower frequencies, but with CMOS process advancements are moving up in frequency and soon maybe into the lower RF bands. Hopefully we'll soon see some new CMOS friendly ADC architectures appear and trade off massive oversampling with waveform quantization in both amplitude and time, and support post Analog to Digital conversion Anti-Aliasing filtering dictated by the actual input waveform metrics!!

Best,

[rf-loop]

Some may say that when the ADC is 8 or 10 or 12 bit then this and that dynamic range and below this and that level FFT displayed things are more like shit.
So or so. But in practical example this happens in real life with real instruments and not with student books.

Oscilloscope
SDS2504X HD
Generator SDG1000X

Input CH1, 50ohm AC, Full BW, 50mV/div (with this setup 4dBm signal is roughly full screen (thewre is stioll small overlap after screen 8div before ADC F.S.
Acquisition mode Normal,
FFT also normal and window Flat Top.
FFT span 300Hz (30Hz/div) and center 100020Hz

There are two signals. Marker 1 signal is 99930Hz from SDG CH1 and Marker 2 signal is 100070Hz from SDG CH2 but there is ~30dB attenuator before splitter.

SDG Ch2 signal is amplitude modulated using modulation depth 0.2% and modulating frequency 50Hz so sidebands are 60dBc

Now think that marker 1 signal is in practice maximum level (near ADC full scale. Screen height 400mVpp responds -3.98 dBm




In image (77), there can see modulation sidebands well and even levels are measured quite well. These mod. Peaks are around -90dB below marker 1 signal (maximum level) and -60dBc (carrier is 30dB below maximum level. )

As can also see there is still nice distance to noise floor.




Image (78), all same but FFT trace average 8 what lightly cleans random noise to more close its average.

[nctnico]

- Don’t use any kind of averaging, ERES, highres, etc.
....
As good start I suggest to check THD + Scope published by W2AEW about 3 years ago:

I think you need to re-watch the referenced video by W2AEW (#65: Basics of using FFT on an oscilloscope), because it perfectly makes sense to use HI-RES or ERES to increase the effective number of bits. This way an 8-bit scope can have a larger effective number of bits at a much lower bandwidth.

A good comparison between ERES and HI-RES can be found here: https://teledynelecroy.com/doc/differences-between-eres-and-hires

The point is that any math process that extend dynamic range with ADC sampling add artifacts and alter BW, so what JeremyC is saying makes perfect sense if you want to be sure  to no catch phantom FFT peaks or miss real ones with BW cut.

I personally agree with that, while I often use ERES equivalent functions in YT mode I have no interest to see FFT fake deep dynamic ranges in 8-12 bits scopes, I find all those small false frequency components quite distracting and when I see 100-120dB FFT screenshot taken from ordinary scopes I make a smile.

Agreed. For starters there has to be enough white / Gaussion noise for eres / high resolution to work well. Eres and high-res are nice tools to make a signal look cleaner but if you zoom in far enough, you can see all kinds of weird distortions. I recall doing some testing on my good old Tektronix TDS744A. Turning the 20MHz bandwidth limit on, made the high-res produce all kinds of fantasy signals. The same goes for FFT; you need to be really carefull with interpreting what you are seeing. Deep FFT is useful to look at closely spaced frequencies (as others have mentioned as well) but the extra dynamic range might not be there at all. Or put differently: the reliability / confidence level of the measurement results that are below the dynamic range of the ADC, is very low.

[_Wim_]

Agreed. For starters there has to be enough white / Gaussion noise for eres / high resolution to work well. Eres and high-res are nice tools to make a signal look cleaner but if you zoom in far enough, you can see all kinds of weird distortions. I recall doing some testing on my good old Tektronix TDS744A. Turning the 20MHz bandwidth limit on, made the high-res produce all kinds of fantasy signals. The same goes for FFT; you need to be really carefull with interpreting what you are seeing. Deep FFT is useful to look at closely spaced frequencies (as others have mentioned as well) but the extra dynamic range might not be there at all. Or put differently: the reliability of the measurement results that are below the dynamic range of the ADC, is very low.

I agree some fantasy signals can be shown, but this is in my (limited) experience always below the effective dynamic range (so not the dynamic range visualized, but the once calculated using the enhanced number of bits).

Is there in your opinion a difference between oversampling (https://www.analog.com/media/en/technical-documentation/tech-articles/Increase-Dynamic-Range-with-SAR-ADCs-Using-Oversampling.pdf) and HIRES? For me it seems this is exactly the same, but I could be missing something. 

Maybe with my 12-bit pico I am closer the the actual noise floor and never had insufficient gaussian noise? I do not use the FFT on my 8-bit scope (DS1054Z) because it is a pain, but with the 12-bit Pico (and Analog discovery) I always have had quite trustworthy results by using averaging and/or ERES. 

[nctnico]

Agreed. For starters there has to be enough white / Gaussion noise for eres / high resolution to work well. Eres and high-res are nice tools to make a signal look cleaner but if you zoom in far enough, you can see all kinds of weird distortions. I recall doing some testing on my good old Tektronix TDS744A. Turning the 20MHz bandwidth limit on, made the high-res produce all kinds of fantasy signals. The same goes for FFT; you need to be really carefull with interpreting what you are seeing. Deep FFT is useful to look at closely spaced frequencies (as others have mentioned as well) but the extra dynamic range might not be there at all. Or put differently: the reliability of the measurement results that are below the dynamic range of the ADC, is very low.

I agree some fantasy signals can be shown, but this is in my (limited) experience always below the effective dynamic range (so not the dynamic range visualized, but the once calculated using the enhanced number of bits).

Is there in your opinion a difference between oversampling (https://www.analog.com/media/en/technical-documentation/tech-articles/Increase-Dynamic-Range-with-SAR-ADCs-Using-Oversampling.pdf) and HIRES? For me it seems this is exactly the same, but I could be missing something. 

It is exactly the same.

[JeremyC]

I think you need to re-watch the referenced video by W2AEW (#65: Basics of using FFT on an oscilloscope), because it perfectly makes sense to use HI-RES or ERES to increase the effective number of bits. This way an 8-bit scope can have a larger effective number of bits at a much lower bandwidth.

I watched this video many times and I read many FFT related publications from the “A” vendors in the last few years.
Sorry, I should mention in my posts that I disagree with W2AEW about the averaging/hires modes with FFT. In my opinion W2AEW is excellent (if not the best) information source and I’m glad he’s sharing his knowledge and experience with the public. But in this topic I would disagree with his opinion about averaging combined with FFT, sorry.

[nctnico]

Think of it this way: if you can make extra information appear magically from an 8 bit ADC, then why aren't we all using scopes with 1 bit ADCs? Even cheaper to make!

[mawyatt]

Think of it this way: if you can make extra information appear magically from an 8 bit ADC, then why aren't we all using scopes with 1 bit ADCs? Even cheaper to make!

There is no "magic" in these concepts and in fact many of the slower Delta Sigma types are simply 1 bit core ADC types (1 bit comparator) with massive Oversampling and Multi-Order Modulators followed by high order Decimation Filters and easily achieve well beyond 20 ENOB, these are the common 24 bit Delta Sigma ADC chips that only cost a few $!!

Many of the higher speed ADC chips employ techniques similar to the 1 bit core Delta Sigma ADC, except they utilize more than 1 bit (usually 3 ~ 4) in the "Comparator" and DAC "Feedback" path to speed the overall conversions up, and digitize the Signal - Feedback difference instead of integrating such as the DS ADC do (if they utilize integration then some of the benefits of Modulator induced "Noise Pushing" can be employed, but this slows things down).

As was mentioned by another earlier post, the HIRES and ERES are effectively doing something similar to what some of the core silicon ADC chips are doing to enhance resolution, and as shown is always a tradeoff between resolution and speed. This is where the mentioned new type ADC architecture may come into play which makes amplitude quantizations but at non-uniform sample intervals which are dictated by the input signal, so simultaneous amplitude and time quantizations...but this is another topic well beyond the scope being discussed here.

So there are many ways/techniques to approach the ADC, some are completely within the CMOS chip, some are external software enhanced, some external hardware enhanced, and some both hardware/software enhanced. They all are trading off speed/BW for resolution, but none rely on "magic" to achieve their respective results.

What you may have interpreted as "magic" in rf-loops notes with the lower level dual unequal tones, one of which is lightly AM modulated, is the effect of the larger tone acting as a "dither" for the lower ADC bits to resolve the smaller tone with AM modulation. This was (maybe still is) a classic technique to improve the apparent resolution/linearity of a non-linear device with a non-responsive zone, called "dead zone". The ADC is non-linear across each LSB transition because it has a "dead zone" where the bit doesn't respond to the input signal, where added noise (maybe internal) or additional signal will modulate the LSB transition, effectively "dithering" across the LSB dead zone and subsequent filtering can reduce the dead zone effects and resolve into fractional LSBs revealing input signal fine details below the LSB as shown.

Best,

[Performa01]

No doubt that the real deal – a true higher resolution ADC – is the superior solution. Yet there is nothing wrong with a proper ERES/HiRes implementation as long as the user thinks a little. And that should not be too much asked, since there are so many situations in T&M where we cannot trust blindingly and just copy some values from an instrument to the lab protocol without a second thought.

For instance we might run into troubles when we try to measure a -60 dBm signal using FFT when the corresponding input channel is set to 100 mV/div (800 mVpp = +2 dBm full scale). This is clearly outside the genuine 8 bit dynamic range.

SDS2354X Plus_LVL_10MHz_100mV_-60dBm_8bit

The measured amplitude is -57.5 dBm, hence 2.5 dB off – just remember, we try to measure a 632 µVpp signal at an 800 mVpp full scale input range. In a low noise DSO, the so called "process gain" doesn't work that well, fair enough.

But then again, not many will attempt to measure such a weak signal by using a particular insensitive input range of the scope – we would rather choose 1mV/div instead:

SDS2354X Plus_LVL_10MHz_1mV_-60dBm_8bit

Unsurprisingly, the measured amplitude is correct now; -59.9 dBm means that the error is just 0.1 dB.

For a fair comparison, now that we used a 40 dB more sensitive input range, we should also try to measure a 40 dB lower signal, i.e. -100 dBm instead of -60 dBm:

SDS2354X Plus_LVL_10MHz_1mV_-100dBm_8bit

All of a sudden even the signals outside the genuine dynamic range are measured accurately: -99.9 dBm is only 0.1 dB off again. This is because at a high sensitivity of 1 mV/div we have sufficient inherent noise to make the resolution enhancement inherent to a longer FFT fully work.

What if there are any stronger signals present, so we cannot select a higher sensitivity without overdriving the scope input? All the better, then even the lower sensitivities will work, because now we don't need noise as dither anymore, but have those stronger signals instead.


Of course, a 12 bit SDS2000X HD doesn't have any troubles measuring -60 dBm even in the insensitive 100 mV/div (+2 dBm FS) range. After all, this is still within its genuine dynamic range of ~72 dB.

SDS2504X HD_LVL_10MHz_100mV_-60dBm_Normal

The last screenshot demonstrates how we can still measure -100 dBm (6.3 µVpp, far outside the genuine dynamic range) at 100 mV/div (800 mVpp) with only 0.74 dB error:

SDS2504X HD_LVL_10MHz_100mV_-100dBm_Normal

[_Wim_]

Sorry, I should mention in my posts that I disagree with W2AEW about the averaging/hires modes with FFT. In my opinion W2AEW is excellent (if not the best) information source and I’m glad he’s sharing his knowledge and experience with the public. But in this topic I would disagree with his opinion about averaging combined with FFT, sorry.

No need to say sorry. As seen from the replies above, it is clearly something were different opinions exist. I agree about W2AEW, excellent channel, have watched all of his video's, even the HAM ones as a non-HAM.

[_Wim_]

No doubt that the real deal – a true higher resolution ADC – is the superior solution. Yet there is nothing wrong with a proper ERES/HiRes implementation as long as the user thinks a little. And that should not be too much asked, since there are so many situations in T&M where we cannot trust blindingly and just copy some values from an instrument to the lab protocol without a second thought.

For instance we might run into troubles when we try to measure a -60 dBm signal using FFT when the corresponding input channel is set to 100 mV/div (800 mVpp = +2 dBm full scale). This is clearly outside the genuine 8 bit dynamic range.

SDS2354X Plus_LVL_10MHz_100mV_-60dBm_8bit

The measured amplitude is -57.5 dBm, hence 2.5 dB off – just remember, we try to measure a 632 µVpp signal at an 800 mVpp full scale input range. In a low noise DSO, the so called "process gain" doesn't work that well, fair enough.

But then again, not many will attempt to measure such a weak signal by using a particular insensitive input range of the scope – we would rather choose 1mV/div instead:

SDS2354X Plus_LVL_10MHz_1mV_-60dBm_8bit

Unsurprisingly, the measured amplitude is correct now; -59.9 dBm means that the error is just 0.1 dB.

For a fair comparison, now that we used a 40 dB more sensitive input range, we should also try to measure a 40 dB lower signal, i.e. -100 dBm instead of -60 dBm:

SDS2354X Plus_LVL_10MHz_1mV_-100dBm_8bit

All of a sudden even the signals outside the genuine dynamic range are measured accurately: -99.9 dBm is only 0.1 dB off again. This is because at a high sensitivity of 1 mV/div we have sufficient inherent noise to make the resolution enhancement inherent to a longer FFT fully work.

What if there are any stronger signals present, so we cannot select a higher sensitivity without overdriving the scope input? All the better, then even the lower sensitivities will work, because now we don't need noise as dither anymore, but have those stronger signals instead.


Of course, a 12 bit SDS2000X HD doesn't have any troubles measuring -60 dBm even in the insensitive 100 mV/div (+2 dBm FS) range. After all, this is still within its genuine dynamic range of ~72 dB.

SDS2504X HD_LVL_10MHz_100mV_-60dBm_Normal

The last screenshot demonstrates how we can still measure -100 dBm (6.3 µVpp, far outside the genuine dynamic range) at 100 mV/div (800 mVpp) with only 0.74 dB error:

SDS2504X HD_LVL_10MHz_100mV_-100dBm_Normal

Thanks for this. Quite impressed by the performance of the SDS2354X!

@nctnico, do you now of a signal that can be generated easily with an AWG where artifacts would appear within the effective dynamic range. Would be nice to do a side by side comparison, and also useful to see what to watch out for...

[tautech]

Thanks for this. Quite impressed by the performance of the SDS2354X!

Do note Performa01 used 2 DSO's, SDS2354X Plus and SDS2354X HD. 

[_Wim_]

Do note Performa01 used 2 DSO's, SDS2354X Plus and SDS2354X HD. 

Ah, missed that one. Thanks for pointing that out! Still, remains very good results with the HD. Would be a excellent replacement for my pico5000...

[gf]

I think you need to re-watch the referenced video by W2AEW (#65: Basics of using FFT on an oscilloscope), because it perfectly makes sense to use HI-RES or ERES to increase the effective number of bits. This way an 8-bit scope can have a larger effective number of bits at a much lower bandwidth.

I watched this video many times and I read many FFT related publications from the “A” vendors in the last few years.
Sorry, I should mention in my posts that I disagree with W2AEW about the averaging/hires modes with FFT. In my opinion W2AEW is excellent (if not the best) information source and I’m glad he’s sharing his knowledge and experience with the public. But in this topic I would disagree with his opinion about averaging combined with FFT, sorry.

HiRes/Eres provides additional processing gain for a subsequent FFT (assuming that the FFT size is already maximum and cannot be increased any more) if the particular HiRes/Eres implementation does decimate the data after filtering, so that the FFT is applied to the lower sample rate.

OTOH, if the HiRes/Eres implementation only applies a lowpass filter (w/o decimation) then it does not increase processing gain of a subseqent FFT, but only distorts the frequency response. No need to filter, the FFT acts as filter bank anyway.

And even if the HiRes implementation does decimate, the next question is whether its filter is an appropriate anti-aliasing filter for the decimation. The traditional boxcar averaging filter certainly isn't -- not at all. For the high dynamic range we are after, we would need an anti-aliasing filter with very high stopband attenuation in order that aliases folded into the 1st Nyquist region drown in the FFT noise floor.