franziski 
I have a LM335 that I would use with an ATmega168 to measure an indoor temperature, in the range 250K..313K (-23°C...+40°C) so the voltage swing is 0.25V..0.313V.
Was wondering if I can attach directly the sensor to ADC input of ATmega168 or I have to use a conditioning stage (to bring the voltage swing to 0..5V)
The resolution needed is 1°C
Thansk in advance
clawson 
If you don't then you will only have just short of 13 steps out of a potential 1024.
5V / 1024 = 0.00488 V/division 0.313V - 0.25V = range of 0.063V 0.063V / 0.00488 = 12.9 distinct readings
franziski If you don't then you will only have just short of 13 steps out of a potential 1024.5V / 1024 = 0.00488 V/division 0.313V - 0.25V = range of 0.063V 0.063V / 0.00488 = 12.9 distinct readings
Your post is a confirm for me. I saw a LM335 used without any conditioning stage and was wondering how it was possible.
franziski My mistake...range of LM335 (10mV/K) is 0,6V [2,53V..3,13V] so the distinct readings would be:
0,6 / 0.00488 = 122.95
So we would have a resolution of 60K/122.95, about 0,48K [0,48 °C] that is ok for my needs (better or equal than 0,5°C) so the conditioning stage should not be necessary.
Can you give me confirm of above calc?
Thanks again.
[cliff: looks right to me]
theusch 
Now, I realize the "engineering is the art of making what you need from what you have". But you certainly could make it easier on yourself.
First,
indoor temperature, in the range 250K..313K (-23°C...+40°C)
Next, a simple device selection of, say, an LM50 at the same price would enable good sub-1-degree precision. 'Cause I've done it. ;)
For your range, that would be roughly 300mV-900mV. It is perfect to be measured with the internal 1.1V bandgap reference. You get roughly 10 ADC counts for each degree.
But, you say, the 1.1V is +/- 10%. No real problem. When you start up, you do a conversion on the internal 1.1V nominal using the "clean" "precision" AVcc as the reference. From there the actual transfer function can be calculated.
When/if temperatures get >~50C then switch to the AVcc as the reference. Some loss of precision but quite a high range.
theusch A properly-biased thermistor can give you a curve that is close enough to linear over the area of interest that it can be used as such. +/- 1 degree is no problem. I'd have to look at my charts to see if fractional degree is practical without a lookup table. As it is a monotonic curve with only a little "belly" the actual accuracy doesn't really matter in a temperature-control application.
Lee
franziski @theusch
In our modhest school-lab are available only some LM35 and LM335 so we have to use these ones.
-20°C seems too low for an lower indoor temperature as you noticed...so we could use a LM35 (10mV/°C) instead which guarantees a 0.25 °C accuracy at room temperature and narrow the range to a more reasonable 0°C..40°C
So doing, we have 81.96 distinct readings (0.4/0.00488) with a resolution of 0.488 °C (40/81.96)
Let me know what do you think about the above approach.
Thanks again for your help.
diy.jack franziski theusch
CANbus.
Lee
franziski Quote:
CANbus.
??? that is a "1-wire" interface device.Lee
OMG I have to relax a bit :shock:
Anyway, LM35 seems simpler to implement.
