NTC interface IC?

[bitbanger]

Hi Folks -

I'm working on a quick one-off board which needs to measure two localized (within 8-10") surface temps with modest accuracy (+/-0.5°C) over a range of 20-125°C. I've decided on NTCs for a variety of reasons (primarily due to availability of industrial probes with required mounting provisions, the accuracy spec., simplicity).

I have used a cold junction compensation IC before for thermocouples, but I was surprised to find a lack of NTC interface IC's. The one or two I did find were outrageously expensive ($15+/ea). Is there something I'm overlooking? Trying to avoid linearization via external passives/algorithm/calibration. Lazy, perhaps,  but I don't want to spend a whole lot of time on this and don't mind spending a few bucks on an IC to help in this regard.

Thanks in advance!

[ajb]

With a thermocouple, if you know that it's a k (or whatever) type, you immediately know the transfer function.  With an NTC you need to at least know it's nominal resistance and it's B value (there are soother values you'd want to use for really high precision, but for low precision measurements they're usually negligible).  There are enough available combinations of nominal resistance and B values that it's rather hard to make an IC that can accommodate all of them.

The hardware can be as simple as a voltage divider, although if you want to maximize your ADC resolution it's often helpful if you can bias the top and bottom of the divider above and below your ADC input range.  You could also use a current source and/or an opamp for gain and offset if you want to get fancy.  As long as you have a bit of resolution to spare on your ADC you don't need to worry about linearizing anything in hardware.  Software-wise, the only algorithm you need is a single formula, or even a lookup table you could precompute in Excel.

[NW27]

Maxim have a nice 8 pin k type interface that uses i2c to communicate to the micro. The linearity correct is already built in.

[rstofer]

Absolutely, Maxim have the MAX6682.

You said $15/ea is "outrageously expensive". If you live in a "rich" country that's probably worth about 1 hour of your time!

9 minutes at my current 'retirement' rate.  It would be more if I had to pay myself benefits.  I think I need a raise...

[ebclr]

http://www.lpi.usra.edu/lunar/ALSEP/pdf/LinearThermistorsUseAsTempSensors%20-%20ATM-1108.pdf

[nctnico]

I'd use PT1000 sensors. Connect to 3.3V from 2.2k Ohm to an ADC input and you'll have enough accuracy (the ADC reference also needs to be from the same 3.3V).

[ebclr]

0.5 C with NTC will be possible only with several points of calibration and linearization on a one to one sample basys, and with no long term stability, I guess milk does not come from rocks