True between chips it can vary from 1.0 to 1.2V... this is why I mentioned the calibration. With calibration the error is 1mV (0.1%).

If we don't already own a Digital Multimeter that is certified to +/- 0.1% accuracy,
then how can we calibrate the ATmega 328 ADC to +/- 0.1%, using the Band Gap Reference Voltage ?

What are the details for this precise calibration procedure for the ATmega 328 ADC ?

 

You need a precision meter to calibrate or a know good reference voltage Because the offset is consistent you can put an offset in EEPROM and calibrate it. They do it at the factory on some chips but I’m sure it costs more.

The procedure according to microchip http://ww1.microchip.com/downloads/en/AppNotes/doc8060.pdf

 

I did not make myself clear. I do not expect 0.01% accuracy or anywhere near it; I used precision resistors in my test so that I would know that any error was not the resistor but the meter. I was hoping for something better than a DVM would give me.
As to Kelvin leads, totally unnecessary at my expected level. There are, actually no leads as such--just a ZIF socket into which the resistor leads are plugged.
An internet search of Ardino ohmmeters was enlightening, if not encouraging. One seemingly competent study indicated worthless readings below about 100 ohms and over about 1000 (I've forgotten the exact range.)
Maybe I'll buy another meter to check that I did not just get a dud. Maybe I'll just pitch it.

Assuming the Meter uses the internal 10 Bit ADC,
a) At the upper end, near "Full Scale", of a range, readings could be as good as +/- 1% accuracy
b) At the lower end, near "1/10th scale", of a decade range , may be +/- 2% accuracy, or worse.

The accuracy will vary based upon where there resistance value falls within the specific Decade Auto-Range.

My friend has a $350 "Fluke 159" and it is only +/- 1% for resistance measurement, at best.
NOTE: that it states "at best", it does not state "worst case".

 

You need a precision meter to calibrate or a know good reference voltage Because the offset is consistent you can put an offset in EEPROM and calibrate it. They do it at the factory on some chips but I’m sure it costs more.

The procedure according to microchip http://ww1.microchip.com/downloads/en/AppNotes/doc8060.pdf

Unfortunately, that does not apply to the ATmega 328 that is used in the Meter being discussed in this thread.
Also, it is illogical to tell the TS to buy a $400.00+ 0.1% DVM, to calibrate the $10 Meter being discussed.

A LM4040 0.1% Precision Dual Voltage Source for $7.50 ...
https://www.adafruit.com/product/2200

 

There's no fixing that meter... this is just for reference as the bandgap can be an accurate source and as stated before it's how you utilize it that matters. ADC can impart more errors based on how things are wired... It would be better to have different ranges as no one resistor divider would be accurate for all situations. So just saying don't blame the chip... blame the guys who designed and made this thing.

 

I don't expect the same accuracy out of a $10 meter as I would from a NIST certified meter. Kind of like a $10 watch that only errs by a few minutes a week/month. You get what you pay for. When working with devices that typically have 5-20% tolerances it may be adequate for the job and is probably more accurate than 5-20% error.

 

When working with devices that typically have 5-20% tolerances it may be adequate for the job and is probably more accurate than 5-20% error.

since most of his post pertain to LEDs, how on earth will he ever make sure he has the correct value resistor for them?

 

Tolerances multiply. Let’s say your measuring a component that is rated 5% tolerance. With a meter with a 10% tolerance. The upper range with which you can be confident is (1.05)(1.1) or within 15%... And it’s actually worse. 15.5%.

Not a very good confidence interval. Considering the error could be in both directions, there is a 31% range that the actual value could be.

 

since most of his post pertain to LEDs, how on earth will he ever make sure he has the correct value resistor for them?

LEDs? There is a lot of slop there. Their light is fairly uniform over a large current range. In that case the typical value for Vf and a midrange current value is good enough.

 

Tolerances multiply. Let’s say your measuring a component that is rated 5% tolerance. With a meter with a 10% tolerance. The upper range with which you can be confident is (1.05)(1.1) or within 15%... And it’s actually worse. 15.5%.
Not a very good confidence interval. Considering the error could be in both directions, there is a 31% range that the actual value could be.

I do not agree in assessing the "stack up" of multiple tolerances. They do not multiply, but I think you mean you mean worst case additive. Even in that instance, since each tolerance is probably something like a 95% confidence interval, as you stack them up, it becomes extremely unlikely that each value (say 10 resistor in series) will be at the maximum of its 95% confidence limit [(1-0.95)^10].

Here's a reference on stacking tolerances: https://pdfs.semanticscholar.org/229e/2da19777edf2dee3521334093c8e28273901.pdf

 

LEDs? There is a lot of slop there. Their light is fairly uniform over a large current range. In that case the typical value for Vf and a midrange current value is good enough.

That was my point. Why use high precision resistors when that's what you're doing?

 

I do not agree in assessing the "stack up" of multiple tolerances. They do not multiply, but I think you mean you mean worst case additive. Even in that instance, since each tolerance is probably something like a 95% confidence interval, as you stack them up, it becomes extremely unlikely that each value (say 10 resistor in series) will be at the maximum of its 95% confidence limit [(1-0.95)^10].

Here's a reference on stacking tolerances: https://pdfs.semanticscholar.org/229e/2da19777edf2dee3521334093c8e28273901.pdf

Wonderfully detailed analysis of stacked tolerances. And you’re right, adding the tolerances provides a good estimate.

In my example, we’re stacking a 5% tolerance with a 10% tolerance, which added gives a 15% tolerance. Damn close to my calculated 15.5% value. Mathematically, if a component has a tolerance of 5%, then given v as the purported value, the actual value could be as much as 1.05v. Call it v’. Now consider our measurement could be ±10%, the measurement could be as much as 1.1v’, where v’ is the actual value. Substituting, the measurements could be as much as (1.05)(1.1)v, or 15.5%. The same logic can be applied for the minimum value. This demonstrates that the measurements do multiply.

I’m not going to do it, but mathematically I could prove that adding the tolerances is a reasonable estimate of the actual tolerance. But mathematically, they do multiply.

 

Wonderfully detailed analysis of stacked tolerances. And you’re right, adding the tolerances provides a good estimate.

In my example, we’re stacking a 5% tolerance with a 10% tolerance, which added gives a 15% tolerance. Damn close to my calculated 15.5% value. Mathematically, if a component has a tolerance of 5%, then given v as the purported value, the actual value could be as much as 1.05v. Call it v’. Now consider our measurement could be ±10%, the measurement could be as much as 1.1v’, where v’ is the actual value. Substituting, the measurements could be as much as (1.05)(1.1)v, or 15.5%. The same logic can be applied for the minimum value. This demonstrates that the measurements do multiply.

I’m not going to do it, but mathematically I could prove that adding the tolerances is a reasonable estimate of the actual tolerance. But mathematically, they do multiply.

WOW. None of my posts mention LEDs or multiple tolerances. When you get a new meter of any kind I'll bet that one of the first things you do is measure a known item to check the meter. You probably use a quality item. That's what I did.
Today I watched a bunch of youtube posts. It surprised me that most of the testers never mentioned checking the accuracy. It went like "There, 100.6 ohms." without saying what the resistor is marked. If it was marked 100 ohms his tester worked a lot better than mine. If it was marked 120 ohms . . .
On the other hand, there are a lot of entries about the circuit and much respect to the people who developed it. There is a long set of entries about improving the firmware--worse than Greek to me, but an indication that the meter can be accurate.
It would be nice to have a meter that allowed one to sort resistors by actual value, not marked value. Apparently this isn't it.

 

If you want a good meter, scrounge around for a Harbor Freight coupon that gets you a free DMM when you purchase something. Well worth the money.

I've been in this field for close to 50 years and the only time that I needed accurate measurements of resistors was when I was building shunt resistors to put across a refaced panel meter. Even then I didn't need 0.01% accuracy. Since you say that you are a novice, I'll bet that you don't need it either.

 

since most of his post pertain to LEDs

Huh???

meter with a 10% tolerance

I'd throw the meter away. What I said was that meter is probably MORE accurate than the device tolerance. If I want a highly accurate meter I will pay more that $10 to get one. My watch analogy is that losing a minute or two a week is annoying, but is actually only 2/10000 or 0.02% error unless I missed decimal place..

 

Huh???


I'd throw the meter away.

You do know the the meter with a 10% tolerance wasn’t referring to a real meter? It was an arbitrary meter in an example only.

 

It was an arbitrary meter in an example only.

I'd still throw it away!

 

I am going to add something productive to this thread.
Here is discussion of your type of LCR meter: https://www.eevblog.com/forum/testgear/$20-lcr-esr-transistor-checker-project/
It goes back to 2013.
It is 229 pages.
It shows the hardware, firmware, price evolution of this product.

 

This is pretty standard semi-pro/hobbyist LCR meter: https://www.deree.com.tw/de-5000-lcr-meter.html
The best resistor accuracy is 0.2% at specific ranges. Worst is 2%, again in a specific range, and they tell you to calibrate the meter before making measurements in this range.
I own one. It is sold from 80 to 150 USD.
How you expected 0.1% accuracy from 10 USD toy is beyond me.

 

I second the DE-5000. Got one a couple of years ago when I needed something better than a guess at capacitance and inductance for some filter circuits and oscillator tuning. My workhorse before that was a Wavetek LCR55. The DE-5000 could sort low-value ceramic caps ( < 50pF) and get them in the right order, but I still use the Wavetek.