I need to make a thermostated enclosure that is stable to < 1 millidegree over timescales of minutes to hours. Likely it will be inside an outer chamber to slow the changes to external variation. A key to this is not only a low noise temperature sensor solution, but one that is stable over time. For this particular application a drift over time is fine as long as the timescale is on the order of half a day or longer as I can correct the output of the actual sensor with repeated calibration measurements (the main sensor is affected by temperature). I've read many references where temperature sensing is done with thermistors or RTD in a bridge and amplified. The bridge is set to null output at the desired temperature of the enclosure, so high gains are possible (very limited dynamic range). The enclosure is heated to higher than ambient so only a heating signal is provided.
I do not require accuracy or calibration (5 or even 10 degree absolute error is not critical), so this should simplify things a bit. I do need a sensor with high resolution. I can't find any reference to the output noise on the ad22103, and would such a sensor be appropriate? I can average for many seconds to take a reading (30 even). Superficially, if the output is 28 mV/degree, then noise needs to be lower than 28 uV. It seems I could be able to digitize the signal directly with at least an 18 bit converter (on 3.3V).
Is this sensor feasible for this application? It certainly offers nice ease of use and a high level output.