How to obtain a accurate analog reference voltage, with little drift?

[hmeijdam]

The electric fence consumes between 170-260 milliAmpere at 12 Volt, depending if vegetation touches the fence.

The extra load of a Attiny1624 I estimate at 5 milliAmpere. If I use a LDO it will still be 5mA at 12V, if I use a switching regulator it should be less. Once I have built my prototype I will measure consumption and decide if it's worth putting a sleep function in.

I like your idea of driving the reference from a IO pin, but I would connect the 2.5V ref to the AREF pin and be done with just one conversion.

The ATTiny1624 input resistance of the ADC is also 10K (datasheet chapter 33.16).

I am ok with programming all by setting up registers directly, though that usually takes me a lot of time. I find the modern ATtiny registers quite a bit more complex than the old ones.

Luckily @DrAzzy put a lot of effort in supporting and documenting the ADC for megaTinyCore, so I am studying that as we speak.

I will ask the ADC for a 16 bit resolution with the analogReadEnh() function. that should do 1024 samples in one go and drop the least significant bit. At the risk that I am just measuring noise with the last bits. I would be very happy already if I get a stable 12 bits resolution.

[johnerrington]

@hmeijdam

The ATTiny1624 input resistance of the ADC is also 10K (datasheet chapter 33.16).

You need to interpret that in the light of 30.3.3.5.2; its NOT the input resistance, its an internal resistance that doesnt affect the voltage applied at the input. The ACTUAL input resistance (ie Vin/Iin) is VERY high. I've shown a 10M resistor doesnt change the measurement significantly (ie within the limits of precision of the converter).

A high input resistance in the external circuit will increase the settling time - but since you are measuring from the same input terminal each time that isnt relevant.

"After switching input or reference, the ADC requires time to settle."

Yes I dont see why you cant apply the reference voltage directly to VREFA

[apf1979]

I'd add a temperature sensor too. Take a measurement at room temperature then one after the setup has been cold acclimated and use map() or generate a function to compensate. Take the guesswork out of it.

[hmeijdam]

@johnerrington That's a good one. I thought I red it somewhere about the MOhm impedance, but did not find it in the datasheet.
I was thinking about a 20K-10K divider, that would take 0.4mA from the battery. I can double those values to get to 0.2mA.
Plenty of time to add a delay() before a ADC conversion to get the reference and ADC settled upfront.

@apf1979 The Attiny1624 has a temperature sensor onboard, even with calibration data in the sigrow. Datasheet chapter 30.3.3.7. Never worked with it.

[stitech]

@johnerrington

A high input resistance in the external circuit will increase the settling time

That is about the time it takes for the external circuit to charge the sample/hold capacitor.
But there’s also leakage currents that may affect the voltage at the input if the resistance of the external circuit is high.

[johnerrington]

@stitech

But there’s also leakage currents that may affect the voltage at the input if the resistance of the external circuit is high.

True. All I can say is that I tested this on a Nano and the 10M did not change the ADC reading by more than 0.5LSB (using oversampling)
I dont expect much difference on other MCU's but of course some may have higher native resolution ADC's.
I've repeated the test with a 300M resistor, and 16384 oversampling, then taken the raw summed ADC value and divided by 16384. The "error" is less than 1 adc digit.
Results:

A0:R=5k6	A1: R=300M
740.41	740.24
Since 1 LSB = 5/1024 = 5mV the leakage current is only a few pA.

[wvmarle]

The input resistance of the pins are somewhere in the 1-10GΩ order of magnitude. Even a 10M source resistance is two orders of magnitude smaller, so will have little to no effect on this.

Not that it such a high source resistance is a good idea

[stitech]

The datasheet gives a typical leakage current of <0.05μA. A 10kΩ resistance then gives 0.5mV deviation. Worst case scenario: 17 bit precision, internal reference 1.024 V, 1 LSB = 7.8μV, the deviation being 64LSB.
To put this in perspective: The datasheet gives a Total unadjusted error (int. ref. 4.096V, resolution 12bit) 30LSB max. (Which is better than 1%.)
Of course you are allowed to calculate your own case, to obtain the precision you want/need.

[stitech]

The objective is a warning when the battery voltage gets low. Tentative guess: 11.5V, give or take 0.1V.
You might use a divider 27k/10k and internal reference 4.096 V.

The alternative is not much better:
Choose a divider 56k/10k and external reference. Total unadjusted error 10LSB = 0.3%
Add the accuracy of the TL431: 0.5% and the resistors: 0.1%, giving a total of <1%
Total required components: 3 resistors. a small capacitor for noise suppression and a TL431.

[hmeijdam]

I have just finished rigging up my prototype.
I chose the voltage divider (lower right corner picture) 47K - 10K. So I get less than 2.5V and can use the TL431 (top right corner) or the 2.5V internal VREF with the same divider. Curious how they compare, probably no difference

I took a TL750L as 5V regulator, because it has the lowest quiescent current (1mA) of the LDO's I have.

Top center is the small MOSFET that will turn on the power to the GSM/GPRS shield, which will connect via the bunch of pins in the top left corner.

4 pins left is an I2C breakout, so that I can hook up a LCD for debugging and other info.

The blue connector is for connecting the 12V battery.

Now I need to start writing software.

And if everything works, I may need to redesign everything into a compact SMD print, including a smaller GSM/GPRS module. And cast it in epoxy or plastidip. It will have to survive outside in a field box. This prototype and GSM/GPRS shield won't last long there.

Oh. and I use a ATTiny1626, so I can save my smaller ones if size is important.

[stitech]

@hmeijdam

no difference

Components usually perform better than their specs, especially on the workbench (nice room temperature, while the specs cover -40 to +125°C).

[hmeijdam]

I am pretty happy with the results of the voltage measuring.

With below test sketch I measure the voltage at my voltage divider with a resolution of 0.1 mV. I let it sample 10.000 times and then print the min, max and difference between min and max.
Each analogreadEnh() takes 1024 samples and provides a 17 bit resolution, of which I discard the lsb to get a 16 bit number. So in fact for this test it did 10.240.000 ADC conversions.

With the internal 2.5V reference I typically get around 1.6 mV min-max difference and with the external TL431 0.8 mV. Both with a stable temperature.

#include <LCD_I2C.h>
LCD_I2C lcd(0x27, 16, 2);
uint16_t V_min = 0xffff;
uint16_t V_max = 0;
uint16_t counter;

void setup() {
  pinModeFast(PIN_PA4, OUTPUT);
  digitalWriteFast(PIN_PA4, HIGH);
  analogReference(EXTERNAL);
  //analogReference(INTERNAL2V5);
  lcd.begin();
  lcd.backlight();
  lcd.setCursor(0, 0);
  lcd.print("Battery Monitor");
  lcd.setCursor(0, 1);
  delay(2000);
  lcd.clear();
}

void loop() {
  while (counter < 10000){
  uint32_t vddmeasure = analogReadEnh(PIN_PA2, 17) >> 1; // Oversample to 16 bits
  uint16_t millivolts = map(vddmeasure, 0, 0xffff, 0, 25000);
  if (millivolts < V_min ) V_min = millivolts;
  if (millivolts > V_max ) V_max = millivolts;
  lcd.setCursor(0, 0);
  lcd.print(millivolts);
  lcd.print(" ");
  lcd.print(counter);
  lcd.setCursor(0, 1);
  lcd.print(V_min);
  lcd.setCursor(6, 1);
  lcd.print(V_max);
  lcd.setCursor(12, 1);
  lcd.print(V_max - V_min);
  counter++;
  }
}

For the 12V battery this would mean a resolution of 4 mV

[edit about a year later]

The SMS battery monitor is working like a charm this winter.