How to read a Temeraturesensor the most precise way?

[97hilfel]

Hello  everybody,
Im pretty new here in the forum,
currently Im working on a project and I want to measure the temperature of a NTC 100K sensor the most precise way,
so Im thinking about a 12 bit ADC and a Voltage reference source plus a constat 0.5mA current source to get a linear sensor value.
The problem is that I dont know any ICs that create a constant current nor any for a voltagereference. I got a 12 bit ADC already integrated in my µcontroller (Atmel ATSAMD20J18). About the problem of heatig up the temperatuer sensor when I let a current flow trough it constantly I taught I just use a verry low drain-source resistence n-channel MOSFET.
I hope I get a few awnsers
Felix

PS excuse me for my bad english

[tszaboo]

With a REF200 you can make 400uA, and it is quite accurate. Otherwise, with any* series voltage reference, you can force a certain voltage across a resistor. An then the current should go through your NTC.
But I think you over-think the problem. The NTC initial error will be quite high anyway, so a radiometric approach is usually enough for an NTC. If you want to have higher accuracy, you need an RTD, like a PT100.

[Kleinstein]

You should use a ratiometric reading, so derive the current from the ref. voltage used by the ADC and a resistor. In the simple case this just a resistor and the NTC as a divider.
At 100 K you likely need some kind of buffer and can not directly connect to the ADC. So you might as well have a simple bridge amplifier with a single OP. Usually there is no need for very high resolution of the ADC, as the NTC changes its value quite a lot.

[97hilfel]

The PT100 is not adapt for my temperatuer range, I need to measure up to 300°C so Im trying to get at lest the best out of my ADC because I cannot adapt the voltage range because I probably need to measure from (in the worst case) -20°C up to 300°C.
Thanks by the way for your replys

[mstoer]

A PT100 (platinum RTD) is good up to just over 900C. I've used them at 800C even myself. Why won't it work for your temperature range?

[Alexei.Polkhanov]

Try Analog Devices ADuC845/848 Micro instead of Atmel - it has 24 bit ADC inside as well as built-in current sources for designs just like that.

Look at page 91 of datasheet: http://www.keil.com/dd/docs/datashts/adi/aduc845_47_48_ds.pdf

They have 4-wire connection example as well.

[AndyC_772]

What's the point, if the NTC has an absolute tolerance of 5% anyway?

A thermistor isn't a precision device. It's a cheap, simple, easy to use device, which gives a reasonable approximation to temperature using simple electronics. Great for controlling a cooling fan or detecting if something has overheated, but no use at all if you need to know absolute temperature with any real degree of accuracy.

You can improve accuracy by calibrating it against a known reference, which may be a workable solution for a one-off, but if this is a production item, you're better off with a platinum resistance device.

Neither will give a response that's linear with temperature, but both have resistance-temperature characteristics that are known, and which your microcontroller can compensate for mathematically. In the case of a PT sensor, that characteristic is actually accurate and well defined, and very consistent between manufacturers. For an NTC thermistor, not so.

[mikerj]

Hello  everybody,
Im pretty new here in the forum,
currently Im working on a project and I want to measure the temperature of a NTC 100K sensor the most precise way,
so Im thinking about a 12 bit ADC and a Voltage reference source plus a constat 0.5mA current source to get a linear sensor value

Using a constant current source with an NTC thermistor will make the output LESS linear than using a fixed resistor to form a potential divider.  This can give resolution problems at the hot end of your measurement range, and overall worse dynamic range.
A constant current source also means that that thermistor self heating can increase measurement errors at low temperatures when the resistance is higher.  In general, don't bother using a constant current source to bias a thermistor.

Note that thermistors tend to have fairly wide tolerance on both the 25C resistance value, and the "Beta" value.  Using the nominal RT function to linearise the output will not give you the best accuracy; you will need to characterise your thermistor to get the best possible accuracy.  Also note that extended high temperature operation can cause thermistors to drift in value.

[Kleinstein]

For such a large temperature range a NTC is not a good solution, because the resistor changes too much to be well measured at the high and low end with the same circuit. Also stability at 300 C might be an issue. There are a few accurate NTCs available, but the typical cheap ones are are not really accurate.

So I would also tend towards a PT100 or maybe PT1000. 

[97hilfel]

Thanks so far for all replys,
I will include some ideas and look at other ideas if they are suted for my application.

[Neilm]

There are a load of I2C compatible chips that measure the temperature. If you need lots of accuracy, you could use an ADT7420 +/-0.25 degC over a large temperature range