Potential conflict between internal and external ADC reference voltage

[jonathanperret]

Problem

While reviewing an update to the AYAB shield, it appeared the original AYAB schematic connected the 5V pin from the Arduino (coming from USB), known as VCC, to the Arduino's AREF pin.

Looking at the schematics of previous and current versions of the shield, and the schematic of the EMSL Interface board, it seems this design has been copied in all existing versions of the AYAB hardware.

However, when driving the AREF pin with an external voltage like this, both the Arduino documentation and the ATmega328P datasheet warn that the ADC should not be used without first being configured to disconnect the internal voltage reference from the AREF pin:

If you’re using an external reference on the AREF pin, you must set the analog reference to EXTERNAL before calling analogRead().

Because the AREF pin is also an output pin that exposes the internal voltage reference when it is used, enabling the 1.1V internal reference while supplying 5V to the pin could damage the microcontroller.

But the AYAB firmware does not call analogReference at any point, so it is in violation of the specification of the microcontroller and could in theory damage it.

How to fix this

This is clearly an oversight in the AYAB hardware design; however, since the existing hardware cannot be globally replaced, changing the hardware design now would not affect the existing AYAB installations. Instead, we can modify the AYAB firmware to call analogReference(EXTERNAL) upon initialization. Testing on actual hardware is required however, in case calling analogReference(EXTERNAL) has an unforeseen effect on the values read from the ADC.

Why this might not be a real problem

First off, as far as I can tell no issue has ever been reported with the AYAB hardware that could be attributed to this problem.

In practice, what the Arduino libraries do by default is to use the ADC in a mode that uses the AVCC as a voltage reference (REFSn = 1). The AVCC pin is connected to VCC on the Arduino board as recommended in the ATmega328P datasheet. So, according to the datasheet, this would mean the AVCC pin is internally connected to the AREF pin in that mode:

Looking at this diagram, it seems relatively harmless to short AVCC and AREF externally as long as the 1.1V reference is never selected, which the AYAB firmware never does?

[jpcornil-git]

Correct, as long as you don't select the internal reference, shorting AVCC with AREF is safe but better to connect only a capacitor to AREF in that mode to filter out AVCC noise (but not critical here)

[jonathanperret]

Correct, as long as you don't select the internal reference, shorting AVCC with AREF is safe but better to connect only a capacitor to AREF in that mode to filter out AVCC noise (but not critical here)

The UNO board already has a capacitor between AREF and ground for that purpose, and another one between AVCC and ground, so it looks like we’re covered on that front.

[jpcornil-git]

Best is to leave AREF floating; as illustrated below, the capacitor on AREF together with the MOSFET "on" resistance act as an RC filter reducing noise of the ADC reference/conversion result.

When you short AVCC to AREF as well, then decoupling is much worse (impedance of AVCC source much lower than internal/small mosfet)

Again, not an issue for this application but better to fix whenever possible; lower noise as explained above but also no risk to damage the UNO if you turn on the internal référence by accident.

[jonathanperret]

Best is to leave AREF floating; as illustrated below, the capacitor on AREF together with the MOSFET "on" resistance act as an RC filter reducing noise of the ADC reference/conversion result.

Again, not an issue for this application but better to fix whenever possible; lower noise as explained above but also no risk to damage the UNO if you turn on the internal référence by accident.

Thanks for the detailed explanation.

So I guess we should disconnect AREF on the next version of the shield. The problem if we do that is that we can no longer unconditionally call analogReference(EXTERNAL) in the firmware as the documentation says we should when supplying AREF, because it wouldn’t work on a fixed shield with floating AREF.

I guess we can just fix the shield and leave the firmware as it is, since it seems to have been working fine so far.

[jpcornil-git]

This is what I would do, i.e. leave FW as-is and fix the shield moving foward.

I wouldn't use analogReference(EXTERNAL) unless you have to, e.g. AREF level is different than AVCC/default or you want to add additional noise filtering between AVCC and AREF

[jonathanperret]

I just did a quick test on an Arduino Uno R3 to confirm that the analog readings were identical with or without an AREF-VCC short which confirms we can remove the short from the shield.

[jonathanperret]

The AREF connection was cut for future hardware in AllYarnsAreBeautiful/ayab-hardware#48 so I think there's no longer a need to worry about this. Bad things could happen on existing hardware if the firmware ever enabled the 1.1V internal reference but I can't see a reason that anyone would ever be tempted to do that.