EEVblog #1226 - How To Get Better Accuracy On Your Oscilloscope

[EEVblog]

How to get better waveform measurement accuracy on your oscilloscope.
Two advantages of the oscilloscope vernier control.

[eV1Te]

I think the reason why the Siglent is not showing a standard deviation (or showing 0 pV) is because of the low-noise front end, hence there is effectively no noise in the digitized signal and all measurements are placed in exact same ADC value/code.
You can easily test this by reducing the amplitude of the sine wave to 1/10 and increasing the vertical gain of the scope proportionally. On a higher gain you should have more noise and the standard deviation should work without reading 0 pV if I am correct.

More detailed info here:
https://www.analog.com/en/analog-dialogue/articles/adc-input-noise.html
"Now, consider the case of an ADC that has extremely low input-referred noise, and the histogram shows a single code no matter how many samples are taken. What will digital averaging do for this ADC? This answer is simple—it will do nothing! No matter how many samples are averaged, the answer will be the same. However, as soon as enough noise is added to the input signal, so that there is more than one code in the histogram, the averaging method starts working again. Thus—interestingly—some small amount of noise is good..."


Edit:
One more thing, at 5:30 you have a histogram on the screen and you labeled it as the Vp-p histogram, which is incorrect or confusing. It is the histogram of all measured points over the entire wave, not only the Vp-p values. If you made a histogram of the Vp-p values (the values that the scope uses for Mean, Min, Max, etc) you would have a very narrow peak with only a few bins. On the Siglent only a single bin would be filled since the standard deviation was 0 pV.

[NiHaoMike]

Why would the fine adjustment have any effect on the measurements if most scopes only have a few fixed steps for the AFE?

[eV1Te]

Why would the fine adjustment have any effect on the measurements if most scopes only have a few fixed steps for the AFE?

A scope front end usually has many more fixed steps than the only the coarse gain setting. However in most cases it is probably not infinitely variable as you might believe when you turn the vernier/fine control. Hence when you turn the fine control you will occasionally change the analog gain, but other times it will be just a software scaling. Hence if you just change the scale slightly you will not see any difference, but if you scale it more than say 1 dB you might step over to a new analog range and benefit from it.

Here is an example I found quickly with 1 dB stepping of the gain, plus an additional 1x, 2x, 5x attenuator for coarse attenuation of large signals:
http://www.ti.com/lit/ug/tiduba4/tiduba4.pdf

[Rerouter]

E.g. the siglent 1204x-e has 3 attenuator steps. Followed by a variable gain amplifier with 2 ranges a high gain and a low gain range each with 127 steps.

Even when signlents fine gain is giving you 4.5 to 4.6 mV/div as you turn it. Its actually taking multiple steps. Because convienant round division numbers for us dont always line up and it needs to keep a little extra adc resolution top and bottom to make up for it.

[David Hess]

I would be really cagey about accuracy, resolution, and the vertical vernier adjustment.

1. Many oscilloscopes will happily display extra digits of resolution which have no significance even for integrated measurements.  Good oscilloscopes will prune the number of significant digits.
2. The number of counts is usually *not* by 1.  Watch the least significant digit and see if any numbers are missing.
3. The accuracy is significantly worse than the resolution. (1)
4. There are many ways to implement the vernier adjustment.  Many DSOs implement it on the digital side so quantization noise increases and there is no increase in dynamic range when it is used.  If it is implemented on the analog side, then accuracy is questionable; variable gain amplifiers with a linear control response are difficult.

(1) Unless making a slideback measurement which most DSOs do not support in any significant way.

[Dr. Frank]

That's an inspiring video, generally about A/D converters, resolution, noise and accuracy.

But it's not correct, especially in this case, to identify StD directly with accuracy.

StD is a measure of random noise, or SNR.
Systematic errors also have to be taken into account, when judging 'accuracy'.

So using the vernier, or a lower range, gives more dynamic in the measurement, higher SNR and lower StD.

The systematic errors, especially when using the vernier is a few % , is determined by A/D resolution, by the fixed divider calibration, and by the vernier divider/attenuator precision, latter mechanism not being transparent to the user..

If you can really separate the systematic errors from the noise error, the accuracy concerning these random errors might be something like StD / sqrt(N), where N is the number of samples.

In the metrology folder, we have discussed the noise figures of various 6 1/2 .. 8 1/2 digit DMMs, also using the STD at different time scales, that is the Allan Deviation, to judge the noise performance.. demonstrating that some DMMs really have StD worth 8 digits (~0.02ppm), but accuracy in the end being 100 times worse..

So I propose a Fundamental Friday Video about this stuff (*), could be interesting.

Frank

(*) 'Accuracy and Noise of A/D Converter Systems - Evaluation and Measurements'

[thinkfat]

Are digital oscilloscopes actually calibrated when you use the vertical vernier? Analog scopes were not, you basically only used the fine vertical adjustment for rise/fall time measurements, where you adjusted the waveform to lie exactly on markers on the screen and measured the time between the waveform crossing some dotted lines.

[Rerouter]

Atleast for the siglent it does indeed store the calibration of just about every combination of attenuator and variable amplifier. In every case it is using the channel offset DAC as the root of trust and calibrating only for DC offset. This means when you switch to AC on the very low ranges you may see an offset of a small fraction of a division.

[EEVblog]

But it's not correct, especially in this case, to identify StD directly with accuracy.
StD is a measure of random noise, or SNR.
Systematic errors also have to be taken into account, when judging 'accuracy'.

I wasn't equating StD directly with accuracy. I implied in the video it was a higher SNR, and said in the video you "can think of it as giving a more accurate result" or some such.
Perhaps I should have use more "accurate" terminology 

[rf-loop]

I think the reason why the Siglent is not showing a standard deviation (or showing 0 pV) is because of the low-noise front end, hence there is effectively no noise in the digitized signal and all measurements are placed in exact same ADC value/code.

I have tested it and as can guess it is as you told. 

I repeat with signal around same kind as in Dave's video where he try and laugh / wodering like Liza in wonderland what Siglent display (SD). Even when he see 4.12/4.12/4.12 he still laugh this 0V SD 
(but it is wrong when it inhibit measuring min and max and due to it, also mean, it need also show *** as SD. It is some kind of minibug)

Ok  I drive roughly 4Vpp 1kHz sine and setting scope like Dave for 1.4Mpts memory for reduce samplerate and 1ms/div.

After then I adjust signal generator carefully so that peaks max samples level stay in same ADC step  over every single acquisitions. Min 4.12V Max 4.12V and of course average 4.12V and of course Std 0.
I run it many times because history buffer can keep only last 19 acquisitions with this setup.
After then I zoom deeply in to sine top and/or bottom and check every 19 acquisition. There is not any exception in max value. Both, sinewave top and bottom, give just 2 ADC steps (noise) but all peak values stay in same ADC level and not random one (or more) ADC step  overshoots.
 I did this check many times. Always when signal stay stable and just in "golden" position/level it keep this "miracle". Of course, math is math. I hope simple math use same rules also in Australia (sarc.)
If I adjust signal tiny bit higher or lover then it start generate more random variations in peak value and  it start of course show also SD. I can easy find these "golden" levels when I fine adjust signal generator level.  It works and calculate just right least in this particular check what I did. Even in case it show 0 Std in my test (and I have checked it, if it show anything different then there is error in math (with this data what I  look in my test. Of course I can also adjust signal so that it continuously give other than 0 Std. Only need change signal just tiny bit. And again, it calculate right.)
Then Dave wonder 2 digit resolution. 10V full range and 8bit ADC and electronic EE wonder 2 decimal    
Just like adjusting scopes, first need use brain and after then muscle.   
Btw, some times it can tell even some femto or even less volt SD also in cases when there is bigger difference in max and min but we run it extremely long time and max-min diff is just from some extremely rare single shot. (I leave this puzzle here and look if we can laugh later.).  Who understand this, he do not laugh. But who have perhaps lack of enough real experience and knowledge may roll they eyes and talk something ridiculous. Just need do bit homeworks.

Attached image. No signal. Terminated 50ohm. Reduced mem fore get more acquisitions to history buffer.
Every sequential acquisition every ADC sample just in same ADC step. Now if we add here very low noise signal and level is just so that it match optimally these ADC levels result can be  "amazing" until user understand what is going on.
Because lack of enough noise.

This is common problem also if we try get more resolution using example  averaging. We need noise. For this kind of purpose some advanced instruments need add noise for get better accuracy/resolution. This same may happen in time axis (example in some advanced frequency calculators) and level axis.

[0culus]

I would be really cagey about accuracy, resolution, and the vertical vernier adjustment.

1. Many oscilloscopes will happily display extra digits of resolution which have no significance even for integrated measurements.  Good oscilloscopes will prune the number of significant digits.
2. The number of counts is usually *not* by 1.  Watch the least significant digit and see if any numbers are missing.
3. The accuracy is significantly worse than the resolution. (1)
4. There are many ways to implement the vernier adjustment.  Many DSOs implement it on the digital side so quantization noise increases and there is no increase in dynamic range when it is used.  If it is implemented on the analog side, then accuracy is questionable; variable gain amplifiers with a linear control response are difficult.

(1) Unless making a slideback measurement which most DSOs do not support in any significant way.

[Emphasis mine]

In this case, it's no different than the vernier on an analog scope right? All three analog scopes I own (Tek 2465B, 7904A, HP 1727A) have some way of indicating that the vertical measurement is not calibrated when the vernier is out of it's normal position.

[rf-loop]

I would be really cagey about accuracy, resolution, and the vertical vernier adjustment.

1. Many oscilloscopes will happily display extra digits of resolution which have no significance even for integrated measurements.  Good oscilloscopes will prune the number of significant digits.
2. The number of counts is usually *not* by 1.  Watch the least significant digit and see if any numbers are missing.
3. The accuracy is significantly worse than the resolution. (1)
4. There are many ways to implement the vernier adjustment.  Many DSOs implement it on the digital side so quantization noise increases and there is no increase in dynamic range when it is used.  If it is implemented on the analog side, then accuracy is questionable; variable gain amplifiers with a linear control response are difficult.

(1) Unless making a slideback measurement which most DSOs do not support in any significant way.

[Emphasis mine]

In this case, it's no different than the vernier on an analog scope right? All three analog scopes I own (Tek 2465B, 7904A, HP 1727A) have some way of indicating that the vertical measurement is not calibrated when the vernier is out of it's normal position.

Example in Siglent.   There is three analog voltage bands selected by relays.
Steps (rough and fine) inside V bands are selected with digitally controlled VGA method before ADC part of system is AD8370) Hardware relay selected voltage bands are:

Band I:
500uV/div to 118mV/div rough steps (500uV (10uV finestep), 1 (20uV finestep) , 2 (60uV fine step) , 5 (100uV finestep), 10 (200uV finestep), 20 (600uV fs), 50 (1mV fs), 100mV (2mV fs up to 200mV/div)

Band II:
120mV/div to 1.18V/div rough steps (200 (6mV fs), 500mV (10mV fs), 1V (20mV fs up to 200mV/div)

Band III: 1.20V/div to 10.0V/div rough steps (2 (60mV fs), 5 (100mV fs up to 10V/div), 10V)

What ever step, rough or fine. Same VGA system handle it inside voltage band.
I do not know what steps are included in SelfCal routine, but selfcal takes quite long time and as we know thewre is lot of brute force processing power in system. So perhaps something more than just some single sigle steps in every V Band. In ~600 seconds it can do quite lot.

Between these rough steps there are fine steps.  Rough steps (1-2-5) are some points in total fine steps table. (Not at all as analog scope vernier what do not have any steps) So talking about analog scope vernier what is uncalibrated (least I have not seen old analog scopes with calibrated vernier even when it is technically possible)  due to its nature and modern digital scope fine steps very different principle is more like utterly useless mess. There is no analog vernier at all in example Siglent (and as in many other modern DPO's). Part of level control is done with digitally controlled variable gain amplifier. Example as in Siglent SDS1000X-E  (and something similar principle in many other brands also) it is Analog Devices AD8370. With its pros and cons!!

VGA (Variable Gain Amplifier)  Analog Devices AD8370

As far as I know there is not digital side (after ADC) fine steps "vernier" at all in this Siglent model what exist in this video, nothing like example Keysight 1mV/div derived from 4mV/div digital zoom shit, how about fine steps there...

[David Hess]

In this case, it's no different than the vernier on an analog scope right? All three analog scopes I own (Tek 2465B, 7904A, HP 1727A) have some way of indicating that the vertical measurement is not calibrated when the vernier is out of it's normal position.

That is right and I have never seen any use that method (paraphase amplifier or Gilbert cell with cross connected outputs) while displaying a calibrated value but it is not impossible to do.  Nobody bothered because it simply was not needed; the vertical vernier was intended for waveform comparison or rise and fall time measurement using the graticule and neither require calibration.

I think HP might have done it on their very early automated oscilloscopes, maybe the HP 1980?  The Tektronix 2465 series implemented a calibrated *horizontal* vernier but that is significantly easier because the sweep is proportional to an accurately controlled current.

What ever step, rough or fine. Same VGA system handle it inside voltage band.
I do not know what steps are included in SelfCal routine, but selfcal takes quite long time and as we know thewre is lot of brute force processing power in system. So perhaps something more than just some single sigle steps in every V Band. In ~600 seconds it can do quite lot.

There are all kinds of ways it can be implemented in the analog or digital domain with analog or digital control.

As far as I know there is not digital side (after ADC) fine steps "vernier" at all in this Siglent model what exist in this video, nothing like example Keysight 1mV/div derived from 4mV/div digital zoom shit, how about fine steps there...

I have run across this occasionally in low end DSOs or at least something which produced the same result.

The ubiquitous HMCAD1511 ADC has both analog side and digital side gain controls which apparently produce up to 0.01% gain resolution.  The later act on the raw ADC resolution so within limits, do not result in missing codes.  What is missing are any specifications for gain accuracy and I really wonder how it varies with temperature.

I am not saying that it is impossible to produce good accuracy with a calibrated vernier.  I just would not take it for granted when no specifications are given.

[rf-loop]

What ever step, rough or fine. Same VGA system handle it inside voltage band.
I do not know what steps are included in SelfCal routine, but selfcal takes quite long time and as we know thewre is lot of brute force processing power in system. So perhaps something more than just some single sigle steps in every V Band. In ~600 seconds it can do quite lot.

There are all kinds of ways it can be implemented in the analog or digital domain with analog or digital control.

As far as I know there is not digital side (after ADC) fine steps "vernier" at all in this Siglent model what exist in this video, nothing like example Keysight 1mV/div derived from 4mV/div digital zoom shit, how about fine steps there...

I have run across this occasionally in low end DSOs or at least something which produced the same result.

The ubiquitous HMCAD1511 ADC has both analog side and digital side gain controls which apparently produce up to 0.01% gain resolution.  The later act on the raw ADC resolution so within limits, do not result in missing codes.  What is missing are any specifications for gain accuracy and I really wonder how it varies with temperature.

I am not saying that it is impossible to produce good accuracy with a calibrated vernier.  I just would not take it for granted when no specifications are given.

<i do not really understand where is your blind spot for this "calibtrated or uncalibrated vernier".  Example Siglent SDS1000X-E models, what one exist in this Dave's video have.  In these models there are not at all uncalibrated "vernier" like in your analog scopes. There is 3 voltage bands what have different circuit and these are selected using relays. Due to component variations etc these bands need calibrate. Also there are some things what need adjust in repair/service calibration manually.
But then, there is these main steps as 1-2-5 sequence. And between these also fine steps.  All these steps, 1-2-5 steps and fine steps are included in same specification about vertical.  It do not matter if user use 1.1 - 2.2 - 5.5 steps or what ever available steps.  How we even can have separatew specs for fine steps and rough steps. They are nothing else but UI programmer have selected these 1 - 2 - 5 steps from huge table of all possible V/div steps.  880mV/div is with just same accuracy as 76mV/div or 100mV/div or 500mV/div or 4.82V/div. No matter if you use fine or main steps. But vernier like in analog scope there is not at all.  Analog scope analog potentiometer vernier is normally out of cal. Yes, this can also do as calibrated but it need good precision potentiometer, dial and adjustments for calibrate it to inside given specifications. I have not seen.
Modern digital scope, expensive or even good quality low end have fine steps and rough steps.
IN Siglent all steps, fine and rough are based to same circuits. As told previously band relays/circuits and digitally step adjustable VGA AD8370 + HMCAD1511.

Every single step, fine or rough (both are selected from same "table") is included to this specification and fine steps are not excluded. I can not see any reason or even possible to give separate specs for fine steps and to these 1-2-5 steps but I hope Siglent  edit data sheet (1-2-5 sequence) so that there read (1-2-5 sequence and fine steps) or something similar: