Matthew,

> The LM20 and a PIC18F2xK20 are going to be powered 
> from a 3.3V power supply.
> The PIC chip has VREF- and VREF+ inputs to the ADC. I 
> believe that if I can
> set VREF- at 0.9V and VREF+ at 2.5V, I can greatly 
> increase the resolution
> of the ADC over the input range that the LM20 will 
> deliver.

Can't do that, it violates the deltaVRef spec 
(parameter A20 in data sheet).

>
> How do I approach building a voltage divider to 
> provide me with these
> voltage references? How do I select the best 
> order-of-magnitude for the
> resistors? What other considerations do I need to pay 
> attention to?
>

In general, pick resistors that are a relatively high 
order of magnitude to minimize power consumption.  Make 
sure that the input current into the Vref pin (0-150uA) 
is swamped out by the current in the external resistive 
divider.  See parameter A50 (IRef).  Don't forget, that 
if you use resistive dividers that there will be a 
relatively large error in the VRefs. You will need to 
take into account tolerance of the resitors (1% ?) as 
well as tolerance of Vdd (5% ?).  For better accuracy, 
you can use high tolerance voltage references ($$). A 
simpler, more cost effective approach is to calibrate 
out the errors after the unit is built (but be careful 
of VDD variation over input voltage and output 
loading).

> (Now... I am aware that this may be a bit extreme. 
> The PIC has a 10bit ADC,
> and even if I leave VREF- and GND and VREF+ at 3.3V, 
> I can probably get +/-
> 1 degree C resolution, but this still makes for a 
> nice topic of discussion,
> IMHO, and I'm bound to learn something new from the 
> discussion.)
>

Probably not if you use resistive dividers from Vdd and 
use a typical voltage regulator on Vdd.

We also didn't talk about system noise and filtering.

Ken. 

-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist