Understanding of counts, acuracy, ppm etc

[001]

Hi!

I know what tolerance is not accuracy.

But tell me please why counts  named so strange?
Why 1999 meter is "3.5 digit"?
What is for 3200 meter? "3 x/y digit"? 

So what counts NECESSARY for known accuracy?
For example
1999 display can indicate 1/2000=0.05% Does it mean what I can use 3.5 digit only for meters rated above 0.1% ?

19999 display can indicate 1/20000=0.005% Does it mean what It is useless for 0.1% meters?
(Is billion counts in china handheld meters like UT61E is for marketing purpose only?)

And what ppm stability is ok for 3.5 digit meter? for 4.5 etc?
I.e.: If 3.5 meter teoretically reads 0.05%  so what ppm stability  it must have for REAL 0.05% readings?

Some formulas and/or eqations?

Thanx!

[Gyro]

If you forum search 'counts digits' at the top level of the forum you will find a quite a lot of threads covering this subject - including one of your own from last October.

https://www.eevblog.com/forum/metrology/nebie-question-about-counts-and-acuracy/

[001]

yea

but i cant understand speaking man in video 
I can read only
(For a long time via google translate)

[Gyro]

The other threads that come up in a 'counts digits' 'forum search are very informative though.

[TUMEMBER]

https://static.elektroda.pl/attach/AN82_6405927.png

[Harfner]

The most basic digital multimeters have a resolution of 3 1/2 digits. Here 1/2 is not a fraction, so 3.5 is a misleading way of writing. 3 1/2 is used to express the fact that the most significant digit can only go up to 1. A 3 1/2 multimeter has 2000 counts.
Likewise a 3 3/4 multimeter has 4000 counts and a 3 5/6 multimeter has 6000 counts. A multimeter with 3 000 000 counts has 6 2/3 digits according to this strange convention.
Multimeters generaliy have higher resolution than accuracy. 3 1/2 digit multimeters typically have 0.5% accuracy. If the multimeter shows 1.999 the real value lies between 1.8 and 2.050 the last digit is completely random.

[iMo]

3 1/2 digit multimeters typically have 0.5% accuracy. If the multimeter shows 1.999 the real value lies between 1.8 and 2.050 the last digit is completely random.

1.999 * 0.5 /100 = 0.01

1.999 - 0.01 = 1.989
1.999 + 0.01 = 2.009

[EEVblog]

Oldie

[metrologist]

That's what came to mind. Good ol' memories.

[LapTop006]

Something else we noticed today while I was talking up the new Keithley bench DMMs was they advertise having a 16-bit ADC, which of course isn't nearly enough to record six and a half digits on its own.

I assume this is the slope coversion thing that is common in bench DMMs where they integrate the result over time to get the extended response.

[AlfBaz]

Here's a link to the complete application note containing the conversion graph TUMEMBER posted
www.analog.com/media/en/technical-documentation/application-notes/an82f.pdf

[Cerebus]

Something else we noticed today while I was talking up the new Keithley bench DMMs was they advertise having a 16-bit ADC, which of course isn't nearly enough to record six and a half digits on its own.

They say that it has 16 bit 1 Ms/s ability when you use it as a digitizer. That is not the same thing as saying it has a 16 bit ADC. That's saying when you repetitively sample with a 1us acquisition time it can still deliver 16 bits.

For the cynical, and I know you're out there, it's quite likely that's a delivered 16 bits, not an ENOB of 16. Given that they are claiming 4 1/2 digits (±0.005% of range + 40uV) at 0.0005 NPLC (~10us acquisition time); if that's 4 1/2 digits for 20,000 count then that's an ENOB of 15.29 including the sign bit at 100 kHz.

[Gyro]

Oldie

Sadly, that's the video that the OP said he couldn't follow in his previous thread.

To be fair, his problem is with following English speakers in general rather than your particular version of it!   

[metrologist]

might be easier if one uses the gear settings icon to slow the speed down?

[David Hess]

But tell me please why counts  named so strange?
Why 1999 meter is "3.5 digit"?

As far as I know, marketing decided that a single 1 counted as 1/2 of a digit.  It is a pretty good trade off considering that one extra segment doubles the range from 999 to 1999.

What is for 3200 meter? "3 x/y digit"?

I have never noticed any constancy.  Everything above 1/2 of a digit seems to be 3/4 digits.

So what counts NECESSARY for known accuracy?
For example
1999 display can indicate 1/2000=0.05% Does it mean what I can use 3.5 digit only for meters rated above 0.1% ?

That would keep the precision consistent with the accuracy but in practice most high resolution meters have at least 10 times more resolution than their their accuracy.

19999 display can indicate 1/20000=0.005% Does it mean what It is useless for 0.1% meters?
(Is billion counts in china handheld meters like UT61E is for marketing purpose only?)

There is some marketing influence but a 20,000 count meter allows fewer range changes when making lower accuracy measurements so it is not completely useless.  A meter may not be accurate enough to support its least significant digits but it can still accurately measure a relative change in its least significant digits.

And what ppm stability is ok for 3.5 digit meter? for 4.5 etc?
I.e.: If 3.5 meter teoretically reads 0.05%  so what ppm stability  it must have for REAL 0.05% readings?

Stability over time and temperature is included in the accuracy specification.

[Kalvin]

Sadly, that's the video that the OP said he couldn't follow in his previous thread.

To be fair, his problem is with following English speakers in general rather than your particular version of it!   

One can enable subtitles and/or slow playback speed down. The subtitles may be sometimes a bit goofy, but the short view I had subtitles enabled looked pretty good to me. Just keep in mind that the automatic voice recognition is not flawless and may even produce some major errors. These automatic subtitles do not seem to have any punctuation whatsoever, but they may still be helpful for improved comprehension.

[maxwell3e10]

The most basic digital multimeters have a resolution of 3 1/2 digits. Here 1/2 is not a fraction, so 3.5 is a misleading way of writing. 3 1/2 is used to express the fact that the most significant digit can only go up to 1. A 3 1/2 multimeter has 2000 counts.
Likewise a 3 3/4 multimeter has 4000 counts and a 3 5/6 multimeter has 6000 counts. A multimeter with 3 000 000 counts has 6 2/3 digits according to this strange convention.
Multimeters generaliy have higher resolution than accuracy. 3 1/2 digit multimeters typically have 0.5% accuracy. If the multimeter shows 1.999 the real value lies between 1.8 and 2.050 the last digit is completely random.

This funny definition always bothered me. It seems that a 6.5  digit meter should have 3 333 333 counts. My guess is that the manufactures first rounded it down to 3000000 and then to 2000000 and still called a 6.5 digit meter, then coming up with a  contrived way to justify it. 

[Kleinstein]

The half digit convention likely goes back to old days rules on rounding. So even before having an ADC in mind there was a concept of counting a leading 1 as half a digit.  With many old DMMs having a +-1999 counts range it made perfect sense to call this 3.5 digits. There is no real need to fuss with 1/3 digits.

It gets a little odd if some meter that can give +-1200000 counts call this 5.5 digits - with 1100 counts one could argue the second digit would be a 1/2 digit  only too.

Anyway with modern DMMs the counts / display resolution is not that important anymore, especially at higher resolution. While at 3.5 digits the last one might still be accurate, in the 6 digit an higher range it is common to have errors that are way higher than the last digit. In addition the computer output usually gives even finer steps, though noise may limit the usefulness. 

Today the pure resolution (meant as quantization steps) of the ADC is not an important price factor anymore - the relevant parts are more the stability of the reference and dividers / gain stages. So to grade DMMs, the count's are not that important, it is more the TC,  long term reference drift and linearity that makes the difference. For a handheld add safety.

[serg-el]

https://radiokot.ru/forum/viewtopic.php?f=10&t=139114
Мне больше нравится вариант АК в дробном выражении.

Отсюда: http://pro-radio.ru/measure/13498-4/
"4-1/4 - шкала 12000
Числитель – наибольшая цифра в не полном разряде, 1-ца в данном случае. Знаменатель – число градаций не полного разряда. Градации не полного разряда такие, четыре:
09ххх, 10ххх, 11ххх, 12000.

Если шкала 15000, то градации не полного разряда такие:
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15000.
Разрядность 4-1/7.

Для 1800:
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15ххх, 16ххх, 17ххх, 18000.
Получается 10, значит, для шкалы 18000: 4-1/10, а для 19000: 4-1/11. Так?

Что-то тут не так, проверим на нормальных шкалах:
3-1/2
0999, 1999 (две градации);
0999,1999, 2999, 3999, 4999, 5999, 6999, 7999, 8999 (9 градаций), запись 3-8/9. На нормальной шкале все нормально и понятно.
--

А что если старший разряд может иметь максимум 2, а полная шкала 22000. Как в таком случае записать?
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15ххх, 16ххх, 17ххх, 18ххх, 19ххх, 20ххх, 21ххх, 22000.
4-2/14"


В этом варианте записи всё выглядит логично, и доступно расшифровывается."

[Cerebus]

https://radiokot.ru/forum/viewtopic.php?f=10&t=139114
Мне больше нравится вариант АК в дробном выражении.

Отсюда: http://pro-radio.ru/measure/13498-4/
"4-1/4 - шкала 12000
Числитель – наибольшая цифра в не полном разряде, 1-ца в данном случае. Знаменатель – число градаций не полного разряда. Градации не полного разряда такие, четыре:
09ххх, 10ххх, 11ххх, 12000.

Если шкала 15000, то градации не полного разряда такие:
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15000.
Разрядность 4-1/7.

Для 1800:
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15ххх, 16ххх, 17ххх, 18000.
Получается 10, значит, для шкалы 18000: 4-1/10, а для 19000: 4-1/11. Так?

Что-то тут не так, проверим на нормальных шкалах:
3-1/2
0999, 1999 (две градации);
0999,1999, 2999, 3999, 4999, 5999, 6999, 7999, 8999 (9 градаций), запись 3-8/9. На нормальной шкале все нормально и понятно.
--

А что если старший разряд может иметь максимум 2, а полная шкала 22000. Как в таком случае записать?
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15ххх, 16ххх, 17ххх, 18ххх, 19ххх, 20ххх, 21ххх, 22000.
4-2/14"


В этом варианте записи всё выглядит логично, и доступно расшифровывается."

English please.
For all most of us know you might be exhorting us to revolt on the 25th October. 

[001]

https://radiokot.ru/forum/viewtopic.php?f=10&t=139114
Мне больше нравится вариант АК в дробном выражении.

Отсюда: http://pro-radio.ru/measure/13498-4/
"4-1/4 - шкала 12000
Числитель – наибольшая цифра в не полном разряде, 1-ца в данном случае. Знаменатель – число градаций не полного разряда. Градации не полного разряда такие, четыре:
09ххх, 10ххх, 11ххх, 12000.

Если шкала 15000, то градации не полного разряда такие:
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15000.
Разрядность 4-1/7.

Для 1800:
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15ххх, 16ххх, 17ххх, 18000.
Получается 10, значит, для шкалы 18000: 4-1/10, а для 19000: 4-1/11. Так?

Что-то тут не так, проверим на нормальных шкалах:
3-1/2
0999, 1999 (две градации);
0999,1999, 2999, 3999, 4999, 5999, 6999, 7999, 8999 (9 градаций), запись 3-8/9. На нормальной шкале все нормально и понятно.
--

А что если старший разряд может иметь максимум 2, а полная шкала 22000. Как в таком случае записать?
09ххх, 10ххх, 11ххх, 12ххх, 13ххх, 14ххх, 15ххх, 16ххх, 17ххх, 18ххх, 19ххх, 20ххх, 21ххх, 22000.
4-2/14"


В этом варианте записи всё выглядит логично, и доступно расшифровывается."

Hi
It is hard to translate it via google
lets speak english becouse its more "google-friendly" for translate
Now I see some sort of heresy instead info
Is it Yours own opinion or google translate joke?

[MadTux]

For all math illiterate persons:
ln(DVM Count)/ln(10) => number of digits
ln(DVM Count)/ln(2)  => appr. bits of ADC

ppm, accuracy, drift etc. has absoulutely nothing to do with that.
A 10ppm accuracy,  6 1/2 digit DVM in 3 1/2 digit mode still has the same error and accuacy, thereby all digits are valid, if the DVM is in good conditon.
A cheap shit chinese/broken/uncalibrated DVM might also have 6 1/2 digits, but might be useless on the last 3 digits because of broken condition, chinese quality, noise or lack of calibration.

Hope that helps

[Cerebus]

For all math illiterate persons:
ln(DVM Count)/ln(10) => number of digits
ln(DVM Count)/ln(2)  => appr. bits of ADC

ppm, accuracy, drift etc. has absoulutely nothing to do with that.
A 10ppm accuracy,  6 1/2 digit DVM in 3 1/2 digit mode still has the same error and accuacy, thereby all digits are valid, if the DVM is in good conditon.
A cheap shit chinese/broken/uncalibrated DVM might also have 6 1/2 digits, but might be useless on the last 3 digits because of broken condition, chinese quality, noise or lack of calibration.

Hope that helps

For those illiterate in the design of long scale DVMs: No.

The lower 'digit' modes in long scale DVMs use smaller aperture windows, fewer (or shorter) cycles of the core ADC (almost always a double/multislope integrating ADC) and so do not have the same errors or accuracy as full scale 6.5 digit or higher modes. The whole point of having lower count modes on an instrument is not to throw away resolution for no good purpose but to trade off accuracy for faster readings at lower resolution. Accuracy and resolution are inherently linked as it is quite obvious that you cannot have accuracy better than the resolution of your measurement, worse yes, but not better. Assuming no other error sources, the measurement accuracy of an instrument is limited by measurement resolution to ±¹/₂ lsb.

To put some concrete numbers on it: The HP 3456A has 90 day specified accuracy on its 10V DC range of ±(0.0025% of reading + 2 counts) for a 6 digit reading and ±(0.06% + 2 counts) for a 4 digit reading. Six digit readings are taken with an aperture time of 10 nplc^* or more, 4 digit readings with an aperture time of 0.01 nplc.

^*number of power line cycles

[MadTux]

Sorry, accuracy of course decreases if less digits are displayed, but the error of measurement is the same (with same NPLC), no matter how many digits are displayed.

I.e. I can set the HP-3456A into 4 digit mode (4, store, N- digits) while setting the NPLC to 100 (100, store, NPLC). Then I get all the low error of 100NPLC with only 4 digits display, last 2 digits are simply discarded.

Same works on pair of Keithley 2010/2015. Both meters fully agree in 4 and 5 digit mode with 10NPLC, while they usually have differences on the last 2 digits in 6/7 digit mode. Without the extra digits, the eror simply isn't displayed.