Author Topic: Dithering on 8-bit scope?  (Read 3090 times)

0 Members and 1 Guest are viewing this topic.

Offline CirclotronTopic starter

  • Super Contributor
  • ***
  • Posts: 3362
  • Country: au
Dithering on 8-bit scope?
« on: August 19, 2023, 10:27:13 pm »
On a DSO that uses an 8-bit A/D converter, would it be possible to have an option where at slower scan rates you could have the LSB of the A/D dithered so as to gain several extra bits of resolution? And by dithered, I mean add a high frequency triangular signal of exactly 1 LSB peak to peak to the incoming waveform and then measure the duty cycle of the resulting LSB as it jiggles between two adjacent values to see where it lies between those two LSB values.

Obviously this approach wouldn’t work at higher signal frequencies and scan rates because the A/D would be too busy, but at the lower end, particularly at audio frequencies it might have half a chance. So, would it be a practical thing to do? Or do some scopes already do this?
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 17518
  • Country: 00
Re: Dithering on 8-bit scope?
« Reply #1 on: August 20, 2023, 12:28:57 am »
I don't know a 'scope that does it and they're not programmable.

OTOH many 'scopes have a "high res" mode which averages adjacent samples. You can usually set the window size.

There's also an "averaging" mode which overlays adjacent waves on top of each other when a signal is repetitive.

Both these techniques can give a couple more "bits".
 

Offline CirclotronTopic starter

  • Super Contributor
  • ***
  • Posts: 3362
  • Country: au
Re: Dithering on 8-bit scope?
« Reply #2 on: August 20, 2023, 12:59:00 am »
My scope has both hi res and averaging modes. Naturally there is a limit to the resolution of the signal displayed on the screen because of the pixel pitch, but if the individual samples are logged and brought out externally to be examined, in hi res or average mode have they been truncated to 8 bits or do they have greater resolution?
 

Online tautech

  • Super Contributor
  • ***
  • Posts: 29810
  • Country: nz
  • Taupaki Technologies Ltd. Siglent Distributor NZ.
    • Taupaki Technologies Ltd.
Re: Dithering on 8-bit scope?
« Reply #3 on: August 20, 2023, 01:56:46 am »
My scope has both hi res and averaging modes. Naturally there is a limit to the resolution of the signal displayed on the screen because of the pixel pitch, but if the individual samples are logged and brought out externally to be examined, in hi res or average mode have they been truncated to 8 bits or do they have greater resolution?
Not only as oversampling is employed therefore limiting BW before the risk of aliasing.
Also note how mem depth is automatically managed.
Some examples from the 8 bit SDS1104X-E and can do more from other models if/as required.
Avid Rabid Hobbyist.
Some stuff seen @ Siglent HQ cannot be shared.
 

Online tautech

  • Super Contributor
  • ***
  • Posts: 29810
  • Country: nz
  • Taupaki Technologies Ltd. Siglent Distributor NZ.
    • Taupaki Technologies Ltd.
Re: Dithering on 8-bit scope?
« Reply #4 on: August 20, 2023, 04:34:08 am »
On a DSO that uses an 8-bit A/D converter, would it be possible to have an option where at slower scan rates you could have the LSB of the A/D dithered so as to gain several extra bits of resolution? And by dithered, I mean add a high frequency triangular signal of exactly 1 LSB peak to peak to the incoming waveform and then measure the duty cycle of the resulting LSB as it jiggles between two adjacent values to see where it lies between those two LSB values.
There are also other ways to accomplish this with vertical zoom however that is sometimes governed by channel offset limits and requires some mucking about to maintain that part of the waveform you need inspect more closely.

But some modern scopes have an improved vertical zoom feature to keep the waveform part you need inspect on the display where in the following screenshots a 100x zoom has been applied from the 1V/div to 10mV/div which with a 10x probe = 1mV/div.
With these there was a need to keep the trigger level setting on or near the display.
Avid Rabid Hobbyist.
Some stuff seen @ Siglent HQ cannot be shared.
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 17518
  • Country: 00
Re: Dithering on 8-bit scope?
« Reply #5 on: August 20, 2023, 05:30:42 am »
My scope has both hi res and averaging modes. Naturally there is a limit to the resolution of the signal displayed on the screen because of the pixel pitch, but if the individual samples are logged and brought out externally to be examined, in hi res or average mode have they been truncated to 8 bits or do they have greater resolution?

They'll  be truncated.

But... if you're bringing them out externally you could do the dithering there.  :)
 

Offline johansen

  • Super Contributor
  • ***
  • Posts: 1181
Re: Dithering on 8-bit scope?
« Reply #6 on: August 20, 2023, 05:36:55 am »
The hp3582 did all of that 50 years ago. And explained in the manual how it works.

There doesnt appear to be any money in building something better.

Granted the hp3582 ran a 4004 processor and a 12 bit adc which was oversampled to get 14 or more....but Your modern scope has enough processing power to take that hp 3582 to mars and back...but there is no money in programming itt to take full advantage of its hardware.

My 6000 count dmm produces more accurate rms values of clean sine waves than does my hacked rigol 1054.
« Last Edit: August 20, 2023, 05:51:04 am by johansen »
 
The following users thanked this post: Someone

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 15151
  • Country: de
Re: Dithering on 8-bit scope?
« Reply #7 on: August 20, 2023, 07:18:38 am »
With many faster scopes there is enough noise in the system that no dithering is needed, just averaging / oversampling. Many scopes allow this to get at least a little more resolution or at least get true 8 Bit or what the screen allows.
There are often still limitations to the memory organzation and the theoretical higher resolution may not be used reduced back down to 8 Bit or so. Especially older scope may skip on the extra resolution because of limited memory.
 

Online tautech

  • Super Contributor
  • ***
  • Posts: 29810
  • Country: nz
  • Taupaki Technologies Ltd. Siglent Distributor NZ.
    • Taupaki Technologies Ltd.
Re: Dithering on 8-bit scope?
« Reply #8 on: August 20, 2023, 08:39:29 am »
With many faster scopes there is enough noise in the system that no dithering is needed, just averaging / oversampling. Many scopes allow this to get at least a little more resolution or at least get true 8 Bit or what the screen allows.
There are often still limitations to the memory organization and the theoretical higher resolution may not be used reduced back down to 8 Bit or so. Especially older scope may skip on the extra resolution because of limited memory.
Yes Acquisition modes (HRES AVG) work like that however we also have Lecroy to thank for seeing these modes for what they really are, maths formulae where they do not suffer the sampling/acquisition mode penalties HRES and AVG do when they are not treated as maths functions.
While this screenshot is from a 12 bit scope, the sampling and mem depth are unaffected when Math ERES and AVG are engaged.
Avid Rabid Hobbyist.
Some stuff seen @ Siglent HQ cannot be shared.
 

Offline Performa01

  • Super Contributor
  • ***
  • Posts: 1729
  • Country: at
Re: Dithering on 8-bit scope?
« Reply #9 on: August 20, 2023, 11:31:35 am »
On a DSO that uses an 8-bit A/D converter, would it be possible to have an option where at slower scan rates you could have the LSB of the A/D dithered so as to gain several extra bits of resolution? And by dithered, I mean add a high frequency triangular signal of exactly 1 LSB peak to peak to the incoming waveform and then measure the duty cycle of the resulting LSB as it jiggles between two adjacent values to see where it lies between those two LSB values.

Obviously this approach wouldn’t work at higher signal frequencies and scan rates because the A/D would be too busy, but at the lower end, particularly at audio frequencies it might have half a chance. So, would it be a practical thing to do? Or do some scopes already do this?
Basically, there's no scope that "measures the duty cycle of the LSB", yet on the other hand any serious DSO should be able to do this - as long as they have deep measurements (use the entire record without decimation) and provide measurement statistics, that is. Since even "serious" DSOs like Keysight Megazoom (DSOX, MSOX) series do not meet all these requirements, one has to be careful here.

With an appropriate dither, the LSB will toggle at a ratio according to the intermediate value of the sampled voltage. Collecting a few hundred measurements in the statistics and looking at the mean value usually gives a reliable and fairly accurate (on scopes with deep measurements) result (taking the duty cycle of the LSB into account). Basically, this could already be what you're looking for. It's also not limited to low frequencies, but you need a stable signal over a certain time interval (depending on the frequency) to gather all the data.

If you're not after measurements but simply the graphical representation on the screen, then HiRes, ERES or Average Acquisition modes and/or math functions can do what you want. Many midrange scopes provide a resolution enhancement up to 16 bits based on oversampling. This reduces the input bandwidth though. You can get this function even in the top entry level Siglent SDS2000X Plus, where there is a 10 bit Acquisition mode that limits the bandwidth to 100 MHz and halves the sample memory (because it now has to be 16 bits wide), but has no drawbacks otherwise. And it is almost indistinguishable from the "real thing", except that the noise, hence also dynamic range (ENOB) is only improved by a single bit, as always, when math is used to increase the resolution instead of real hardware. But then again, even real hardware 12 bit scopes rarely ever reach more than 9 bits ENOB at best...

You might need an additional dither on very low noise scopes like the SDS2000X Plus. At least at lower sensitivities, say >10 mV/div. At high sensitivities (down to 500 µV/div) there is enough noise from the frontend to serve as a dither. The other cases are managable as well. Usually the measurements with passive high impedance probes capture enough environmental noise - even low levels of mains hum or signals from nearby radio stations or other transmitters will do. If it is a lab standard 50 ohms test setup, we can always add a dither signal via a resistive power combiner. In any case, it has never been a problem for me in practice as the linked demonstrations show.

My scope has both hi res and averaging modes. Naturally there is a limit to the resolution of the signal displayed on the screen because of the pixel pitch, but if the individual samples are logged and brought out externally to be examined, in hi res or average mode have they been truncated to 8 bits or do they have greater resolution?
The screen of a modern DSO could at least display 9 bits vertically, and with vertical zoom you can inspect even higher resolution captures quite closely. On most advanced scopes like the Siglent SDS2000X Plus, the full sample memory will be exported. That means you get 16 bit data as soon as you're working with a resolution >8 bits, 8 bit data otherwise.

Of course we can easily get 16 bit resolution even with the moderate SDS2000X Plus, if we use the 10 bit mode combined with ERES 3.0 (Extended Resolution 3.0 bits). ERES 3.0 provides 3 bits of increase in ENOB and 6 bits of resolution enhancement. This way, we get 16 bits, can store it and export it.

Of course, the ERES operation further limits the bandwidth of the scope. In case of ERES 3.0 and full channel mode with 1 GSa/s sample rate, it would be only 7 MHz - but you get 4 x 16 bits...

Examples
Here is a demonstration of the 10 bit acquisition mode:

https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg2786952/#msg2786952

This is a demnstration of ERES and Average (3 consecutive posts):

https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg2840282/#msg2840282

And finally the demonstration of 10 bit acquisition mode combined with ERES 3.0:

https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg2857298/#msg2857298



« Last Edit: August 20, 2023, 01:22:45 pm by Performa01 »
 
The following users thanked this post: Someone, Circlotron

Offline Someone

  • Super Contributor
  • ***
  • Posts: 5155
  • Country: au
    • send complaints here
Re: Dithering on 8-bit scope?
« Reply #10 on: August 20, 2023, 11:08:25 pm »
My scope has both hi res and averaging modes. Naturally there is a limit to the resolution of the signal displayed on the screen because of the pixel pitch, but if the individual samples are logged and brought out externally to be examined, in hi res or average mode have they been truncated to 8 bits or do they have greater resolution?
They'll  be truncated.
Examples?

Tek/HPAK both add more bits in their high res and averaging (acquisition) modes, plainly visible when zoomed in. Tek provide an illustrative example in:
"Tools to Boost Oscilloscope Measurement Resolution to More than 11 Bits"
page 17
https://download.tek.com/document/Tools%20to%20Boost%20Oscilloscope%20%20Measurement%20Resolution%20to%20%20More%20than%2011%20Bit_48W_27802_4.pdf

Where the 8bit quantisation is directly/side-by-side compared to the higher depth averaging mode.
 
The following users thanked this post: Performa01, egonotto, 2N3055

Online 2N3055

  • Super Contributor
  • ***
  • Posts: 7464
  • Country: hr
Re: Dithering on 8-bit scope?
« Reply #11 on: August 21, 2023, 10:55:51 am »
My 6000 count dmm produces more accurate rms values of clean sine waves than does my hacked rigol 1054.

That is only because DS1000Z has particularly bad implementation of RMS from data decimated for screen..

Even Keysight MSOX3000T (that works on decimated data too, but much larger set of 32k points) does a very decent job for a 8 bitter.
Any of scopes that do measurements on full data (like Siglents and LeCroy or some new Rigols, to mention few) will easily achieve better result for RMS than average 6000 digit meter.
Not to mention they will do that for waveforms with huge crest factor (non-sinusoidal) and for BW no normal meter will have..
"Just hard work is not enough - it must be applied sensibly."
Dr. Richard W. Hamming
 
The following users thanked this post: Performa01, Someone

Offline Berni

  • Super Contributor
  • ***
  • Posts: 5050
  • Country: si
Re: Dithering on 8-bit scope?
« Reply #12 on: August 21, 2023, 11:34:24 am »
There is no need to introduce a dithering waveform because scope front end and ADC combined tend to be already noisy enough due to being so high bandwidth. So the noise is a sort of free dithering waveform.

Tho it is true that not many scopes have a good implementation of a high resolution mode. Most will just run the scope as normal and then slap filtering over it in software. Most of the Keysight scopes also change the filtering as you zoom in/out in time, so i never know how downsampled of a signal i am looking at.

But when it is implemented properly it is a very nice feature. The only scope i personally used with it is a Agilent MSO9000. That scope (and all of the modern Keysight Infiniium series aka PC windows based) has a high resolution mode where you can select 9bit 10bit 11bit 12bit and this places downsampling filter directly after the ADC, so the only side effect you see is having your max available samplerate reduced. You get full memory, no extra slowdown etc... The ADC just appears like it is spitting out samples with more bits more slowly. Since these scopes have very high sample rates (this one does 20GS/s) you still get plenty of bandwidth(i think it can do 500MHz on 12bit), while the performance looks remarkably close to a real 12bit ADC scope. Because of this you can just leave it in high res mode most of the time and enjoy the sweet low noise. More scopes should do this!

This feature also goes well with FFT since the results of that are also cleaner. However don't expect magic, it will still pick up a spurious tone here and there since this is not a spectrum analyzer (the lower noise just makes the spurious tones more apparent). Also a lot of scopes have not so great FFT implementations anyway.
 
The following users thanked this post: edavid

Offline switchabl

  • Frequent Contributor
  • **
  • Posts: 445
  • Country: de
Re: Dithering on 8-bit scope?
« Reply #13 on: August 21, 2023, 04:43:41 pm »
Yes, the lower-end Keysight scopes don't really allow fine-grained control over acqusition and filtering (the automatic settings are pretty well optimized though). Depending on the timebase setting, you still get up to ~3 extra bits in high resolution mode.

Normal mode (single shot at 5V/div, zoomed to 200mV/div):


High Resolution:


Filtering seems to happen in two stages (up to some limit):
- acquisition: if the memory is too small at full sample rate, average during down-sampling
- display: if there are more samples than pixels (depending on zoom level), average again

As you can see, even at 5V/div, there is no need to add extra noise. High resolution/ERES can be quite useful in some cases but it is not a replacement for native 12bit scopes. It does nothing to improve linearity and spurious tones and it costs a lot of bandwidth. As a rule of thumb, to gain 1 bit/6dB in SNR, you need to down-sample by a factor of 4.
 
The following users thanked this post: edavid, Circlotron

Offline Someone

  • Super Contributor
  • ***
  • Posts: 5155
  • Country: au
    • send complaints here
Re: Dithering on 8-bit scope?
« Reply #14 on: August 22, 2023, 02:25:42 am »
High resolution/ERES can be quite useful in some cases but it is not a replacement for native 12bit scopes. It does nothing to improve linearity and spurious tones and it costs a lot of bandwidth. As a rule of thumb, to gain 1 bit/6dB in SNR, you need to down-sample by a factor of 4.
in some cases it won't even consume bandwidth, as the ADC sample rate could be high enough to grab a few bits for "free" (Keysight 50x oversampling that was described as "excessive" by a particular opinionated poster).
 

Offline EggertEnjoyer123

  • Frequent Contributor
  • **
  • Posts: 335
  • Country: us
Re: Dithering on 8-bit scope?
« Reply #15 on: August 22, 2023, 03:37:15 am »
My Lecroy Waverunner oscilloscope can get up to 3 bits through averaging (at a significant cost to bandwidth).

Interestingly the FFT isn't different (at least at frequencies below the cutoff). This is expected because the extra bits are obtained from filtering (so all the lower frequencies are still let through). So if you are looking for a signal at a low frequency that's obscured by the noise... well you're out of luck.

Edit: Oops I forgot to turn off the bandwidth limit
« Last Edit: August 22, 2023, 03:48:17 am by EggertEnjoyer123 »
 
The following users thanked this post: Someone, egonotto

Offline Berni

  • Super Contributor
  • ***
  • Posts: 5050
  • Country: si
Re: Dithering on 8-bit scope?
« Reply #16 on: August 22, 2023, 11:52:10 pm »
Did a quick test and it looks to reduce noise across the board on the Infiniium scopes. I fed it a tiny -40dBm sinewave to make an example of a case where you are not far from the noise floor.

It looks lower all over, even removes some of the spurs (tho this is more likely due to the ADC running faster)




Here is an example of squeezing the most dynamic range out of it. it is being fed a +6dBm sine wave while the noise floor is down near -130dBm. Tho most of this comes from doing a huge FFT window (as you can see it captured 15 seconds of waveform, and yeah you can get a cup of coffee while you wait for it to calculate). This scope has a very good FFT implementation. Tho the harmonics are likely mostly coming from the scopes front end, no amount of averaging will remove non linearity errors.


And as for just how it looks like, here is the sine wave zoomed in. This is with bandwidth wide open to 500MHz (max it can do in 12bit high res). As you can see the waveform is razor sharp, but still has some fuzzynes around it (since it is not just a 1 pixel wide line like you would get with averaging)


« Last Edit: August 22, 2023, 11:55:05 pm by Berni »
 
The following users thanked this post: egonotto

Offline Someone

  • Super Contributor
  • ***
  • Posts: 5155
  • Country: au
    • send complaints here
Re: Dithering on 8-bit scope?
« Reply #17 on: August 23, 2023, 12:43:07 am »
Did a quick test and it looks to reduce noise across the board on the Infiniium scopes.
Note the acquisition sample rate is lower than the peak ADC sample rate, so there is "excess" sample rate in that case to reduce the noise. The reverse explanation is the "normal" mode is using decimation from the ADC to the acquisition memory and throwing away data.

High resolution/ERES can be quite useful in some cases but it is not a replacement for native 12bit scopes. It does nothing to improve linearity and spurious tones and it costs a lot of bandwidth. As a rule of thumb, to gain 1 bit/6dB in SNR, you need to down-sample by a factor of 4.
in some cases it won't even consume bandwidth, as the ADC sample rate could be high enough to grab a few bits for "free" (Keysight 50x oversampling that was described as "excessive" by a particular opinionated poster).
 

Offline Berni

  • Super Contributor
  • ***
  • Posts: 5050
  • Country: si
Re: Dithering on 8-bit scope?
« Reply #18 on: August 23, 2023, 08:01:00 am »
Yes good point, In this case high res mode was actually working with more samples.

I tried it again on a 10MHz sinewave while forcing it use as high of a sample rate as possible. Indeed it leaves the FFT noise floor untouched, just cutting it off above 500MHz.

However these was still something to gain from it. The FFT slows down a lot when it has to compute captures this big (10s of Mpts). So while around the 10 MHz sine wave the FFT looks identical, the scope could compute it in about 0.5 s using highres mode, but it takes an agonizing 10 s to update in normal mode(since it has way more samples to number crunch). So that is a huge speed up for free. (or i could use the extra speed to do an even longer FFT before it gets too slow to use)

Okay but what about more normal scopes? Like the ol classic MSO6000 does 2GSps and has 8Mpts of memory, well the memory is filled up in only 0.004 seconds. If you use 2 channels and continuous run mode that is only 1ms of acquisition time. So as soon as you are using more than about 100us/div time scale the scope is forced to throw away ADC samples. So you quickly again get into a situation where the ADC is too fast to use all samples. With FFT it is even worse since most scopes are usually limited in the number of FFT points to something much smaller than the size of the sample memory. Another example is the good ol 20MHz BW limit, while that is turned on you could high res downsample a 1GS/s ADC down to 100MS/s and still have plenty sample rate to cover the bandwidth.

So there is a lot of use cases where you do have excess sample rate to work with even on entry level scopes. Yet it is very rare to see such a downsampling feature in its proper implementation.
 
The following users thanked this post: Someone

Offline Someone

  • Super Contributor
  • ***
  • Posts: 5155
  • Country: au
    • send complaints here
Re: Dithering on 8-bit scope?
« Reply #19 on: August 24, 2023, 12:58:05 am »
So there is a lot of use cases where you do have excess sample rate to work with even on entry level scopes. Yet it is very rare to see such a downsampling feature in its proper implementation.
Working mostly with audio/MHz stuff, hi-res acquisition mode is on 90% + of the time when I'm using a Megazoom scope, averaging mode 2nd, normal mode 3rd! There are some quirks/non-ideal side effects but they outweigh the noise reduction and resolution benefits for my use cases.
 

Online David Hess

  • Super Contributor
  • ***
  • Posts: 17427
  • Country: us
  • DavidH
Re: Dithering on 8-bit scope?
« Reply #20 on: August 24, 2023, 09:32:05 am »
And by dithered, I mean add a high frequency triangular signal of exactly 1 LSB peak to peak to the incoming waveform and then measure the duty cycle of the resulting LSB as it jiggles between two adjacent values to see where it lies between those two LSB values.

The noise level at the input of the ADC is usually high enough by itself that no additional dithering signal is necessary.

Some old DSOs, like the Tektronix 2232 series, have a low enough noise that the quantization noise of the ADC shows up in the output, but this is not the common case.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf