I liked STH sensirion as erik suggested . Pl see in the
datasheet I have got a little confusion on page 2/9 ,it shows a graph
Fig 1 , temperature accuracy of SHT11 Transducer ,and on the left hand
side Repetablity +/- 0.1C .What I understand is if I wanna read 22.65C
then according to its accuracy graph ,I read it with an err of +/-0.6C
for SHT11 series is this *RIGHT* or I'm mistaken what is the precision.

'Repeatibility' means, that if you repeatedly take data from your sensor, it will differ from reading to reading by +-0.1°C. Assume you read 22.65°C. Then, due to specified error, real temperature can lie in the error band of 22.65°C +-0.6°C. E.g. real temperature can be 22.65°C + 0.4°C.
If you now repeat this measurement some time later, real temperature can be 22.65°C + 0.4°C +-0.1°C. Important is, that you can calibrate your sensor in order to correct for '+0.4°C' error term, but you cannot correct for 'repeatibility error'!! So, 'repeatibility error' is a very important parameter of any sensor.
Especially for pressure sensors 'repeatibility error' is a critical parameter. Here, it's major source is 'hysteresis'.

My temp range of interest for fine regulation is *ONLY* 20C to 35C I wanna follow a
comfort level equation for HVAC system for the Bus on wheels can I achieve an
such accuracy of 0.1C with a NTC 5K thermistor can ya please show me some
good schematic I found simple bridge on the net with no conditioning
very *RAW* and one as Bannister suggested with Opamp.

Now, that I know, what you are needing temperature sensor for, namely air temperature control, I can tell you, that +-0.1°C accuracy is not really needed! Why that? If you make a precision temperature measurement in such an environment like car or bus, you will observe, that temperature isn't homogenous at all. Dependent on location where you measure, temperature can vary about several °C! So, all standard air temperature control applications don't use temperature sensors providing +-0.1°C accuracy. This would not make any sense, but only increase costs and design efforts.

I would highly suggest you to use a circuit, which is very widely used for such purposes. You will see it in the following. It's almost identical to Charles' design, but does not use NTC sensor, but KTY 10-6 silicon temperature sensor. Advantage is, that curve is much more linear than standard NTC. A further software linearization is not necessary in most cases. Here is datasheet:
http://www.infineon.com/cmc_uploa...kt_10_.pdf

I want to focuse, that you can also use any other way, suggested in this thread, of course. It's only, what I would choose. It's here:




From datasheet we see, that R20 = 1922Ohm, R25 = 2000Ohm and R40 = 2246Ohm.
Circuit above will then provide following voltages at output: U20 = 1.042V, U25 = 1.444V and U40 = 2.669V. (All data are typicals.)

Common mode input voltage of LM324 is in allowed range (0V...3.5V). Input voltage span is 2.669V - 1.042V = 1.627V. Divided by 20K yields 81mV per °C. So, 8bit ADC with span of 0V...5V will give a resolution of about 4 bit per °C, which is fully acceptable. (Uref of circuit should be Uref of ADC.)

Circuit shows a trim poti. It allows you to run several apllications containing identical software with sensors showing some tolerance. Adjust trim poti, so that at 25°C 1.444V at output is achieved. No individual calibration curve has to be fabricated!

In very noisy environment circuit will profit of some filtering: 1nF (X7R) across 47k resistor and 100nF (X7R) across KTY 10-6.

Kai