When a potential divider just isnt accurate enough

Hi folks.

I am working on a project at the moment where i am comparing voltages around different areas of a circuit, these are then being fed into a PIC which compares them and decides if they are within parameters.

The pic is programmed and works correctly and the bulk of the circuit is designed, however I am having a problem with one aspect of the circuit, that being the potential divider.

The circuit thats voltages I am comparing runs at 24V which would destroy a PIC if I fed it straight into it, so I was going to use a 10:1 potential divider to step the voltage down, this is where the problem lies: I can compensate for the inaccuracies of the resistors by simply changing a few values within the code, however I cannot compensate for the change in resistance as the resistors heat up. (The voltage changes that I am trying to detect are within the mV range)

I am completely drawing blanks as to how to resolve this problem, other than just having to wait for the circuit to heat up everytime I want to use it, which is an option, but I would prefer to have a more temperature stable system, any ideas are much appreciated.

/Pete

Sounds like you need extreme accuracy!

Several things come to mind:

1. the obvious one, use resistors with low temperature coefficient.
2. drop the voltage with a temperature compensated Zener before the potential divider, then use equal value and types of resistors. If both change in value equally the mid point will be constant.
3. if possible connect two resistors of the same type and values across the reference voltage and measure the mid-point voltage, using the reading to compensate the real one.
4. measure the temperature and compensate in software.
5. add thermistor in series with one of the resistors.
6. Consider other methods of measurement, such as charge or discharge time on a capacitor where counting rather than quantifying is used.
7. if self heating is a problem due to the current through the resistors, increase their value to reduce the current then buffer the voltage with an op-amp before measuring it.


I'm assuming you are using an external precision voltage reference, the internal ones have drift of their own. If you are using the supply as reference check that is rock steady as well, it will make far more difference than the resistance drifting on the readings you see.

Brian.

Thanks Brian, helpful stuff.

Yeah, the accuracy of this project is somewhat extreme lol.

I did think of trying number 2 but could only find zeners with a reverse voltage drop with an accuracy of within 2%, which is too high for this purpose.

Think I might give number 7 a try as it is purely the self heating that is causing the problem, the reason I didn't just increase the value of resistors before was that the maximum input impedance reccomended by Microchip was 2800 Ohms but that wouldn't matter if i buffered it through a opamp, excellent, will give that a go.

Also, I am using an voltage reference IC to supply a constant 3V to the PIC =)

Many thanks.

Pete

To give a rough estimation, 1 % resistors are usually sufficient related to the expectable accuracy of the PIC ADC, which may be around 8 Bit absolute precision. If you need more, you should e.g. consider an external 16 Bit ADC. There are smart devices like TI ADS1100 with I2C interface.

Cheers for the advice, the A/D converter I am using is 10Bits which is deffinately not accurate enough for what i need with my current set up, need to completely re-think my setup, have ballsed up bigtime lol.....

Cheers guys.

Dont need ne more advice on this subject if anyone else is thinking about posting, thanks tho!

Hi,
maybe is only your Uref not enough accurate?
National has over 30 years a tep compensated Zener (at 6.3V) & I remember thats output has only some ppm changes...
Newly has Linear Techn. an 0.25ppm noise,2ppm/Cels drift & +/- 0.025% (max) accuracy reference family, LTC6655, with the typical 1,25/2,0/2.5/3,3/4.03 & 5V versions.
K.

Thanks Karesz.

I am using a linear tech reference voltage at 3V at the moment =)

Good_ fein; what kind (they have lot of in program),
then use resistor networks i.e. 1% verisons, but the tracking of their will be better_ up to 0.01% regularly ...
Otherwise Im sure; you must redesign so that over all will be higher resistor values to find= smaller /lower self heating effects.
As one more: you must have a cover over your circuit for thermal streaming effects "switch of"/eliminate.
K.