Archived Forum

Do I understand you correctly, and do you think the method should read and return the internal voltage reference itself, i.e. 1.1 V ?

If so, I'm afraid you're wrong. The code changes the ADC's reference to the internal band-gap reference, then uses that to measure the voltage of the ADC's normal reference voltage AV_cc. It is that voltage that you are reading as 4752 mV.

The band-gap reference is stable to a high degree but the initial value can lie anywhere within the band 1.0 - 1.2 V, thus your 4.752 V is a valid value that is within tolerance for the measurement of a 5 V supply to AV_cc.

Thanks for clarifying Robert.

Perhaps you could help all of us  understand how this code works - either give us a detailed description or an in depth reference so that we too can understand the nature of voltage referencing according to bandgap. From what I have read on forums there are not too many who understand the registers and how they work - a good tutorial is needed, perhaps you can help.

Also I say again that this site needs a "notify" check box under each post.

Thanks Robert.

Do I understand you correctly, and do you think the method should read and return the internal voltage reference itself, i.e. 1.1 V ?   .....   The code changes the ADC's reference to the internal band-gap reference, then uses that to measure the voltage of the ADC's normal reference voltage AVcc.

I don't think that's quite right.  If you selected the 1.1V bandgap reference how would you be able to measure AVcc?  Everything above ~1.1V would clip at 1023.

What the code actually does is measures the 1.1V against AVcc, and then "back calculates" AVcc from what it sees.

platypus the best reference for all those register accesses is always the Atmel datasheet for your part.  Check out the "Input Channel Selections" table in the A/D section.  On a 2560 for example, MUX5:0 selects which input pin gets fed to the A/D.  But if you set it to 011110 then it doesn't select any input pin, but rather selects the 1.1V bandgap.

That's a big ask. It is actually all in the data sheet, but until you've fought that battle a few times, it can be more than a little mysterious. With Glyn & Trystan's permission, I'll put it on my "to do" list. No promises for when, though.

For now what everyone needs to know is readVcc( ) returns the value of AV_cc in mV (nominally 3300 or 5000) using the internal band-gap reference. The uncertainty in that measurement could be ±9%. The repeatability and stability of that measurement is in the order of a few hundred parts per million.

I should have added to my reply that there are two sources of error when you measure power: variations in component values (resistors, transformers) and variation in this reference voltage. The purpose of calibration is to correct for errors from both these sources.

It has been said that the mains-powered processor should not use readVcc( ) because the value of the internal 1.1 V reference is known with less certainty ( ±9%) than the 3.3 V or the 5 V from the regulator i.c. However, the internal reference is more stable than the regulator in the face of external influences. readVcc( ) is of course essential when operating from batteries.

Perhaps you could help all of us  understand how this code works - either give us a detailed description

I'll have a crack.  The code is necessarily cluttered by all the AVR variants it supports,  so start by focusing on your part of interest.  In this example I'll assume a 2560 because I have the datasheet handy.   And then examine it line by line:

ADMUX = _BV(REFS0) | _BV(MUX4) | _BV(MUX3) | _BV(MUX2) | _BV(MUX1);
ADCSRB &= ~_BV(MUX5);  

The above is basically saying:

REFS1:0 = 01  and MUX5:0 = 011110

(MUX5 happens to live in a different register from MUX4:0 which is why it's done separately).

Looking at the attached tables from the 2560 datasheet, you'll see that:

REFS1:0 = 01  means select "AVCC with external capacitor at AREF pin" as the reference voltage.

MUX5:0 = 011110 means measure the 1.1V Vbg

ADCSRA |= _BV(ADSC);

The above starts a conversion running.

while (bit_is_set(ADCSRA,ADSC));

The above waits for the conversion to finish.

result = ADCL;
result |= ADCH<<8;

The above fetches the 10-bit result.  So at this stage result now holds the result of measuring the 1.1V Vbg using AVCC as the reference voltage.   In other words:

result = Vbg / AVCC * 1024

So for instance, if AVCC is 5V and the 1.1V bandgap is spot on at 1.1V, then you would expect result to now hold 1.1 / 5 * 1024  =  225.

The premise of the routine is that if you know one of either AVCC or Vbg, and you have the above result then you can easily deduce the other.

If you trust AVCC, and you want to know what Vbg is for this part, then you can rearrange the above formula to:

Vbg = result * AVCC / 1024.

If you trust Vbg, and you want to know what AVCC is, then you can rearrange it to:

AVCC  = Vbg * 1024 / result

             = 1.1 * 1024 / result         (since you trust Vbg to be 1.1)

             =  1126.4 / result              (in volts)

             = 1126400 / result            (in mV)

which gives rise to the final line of code:

result = 1126400L / result;