Sorcery with AD595

A few month ago I was playing with thermocouples trying to make an Arduino-based temperature sensor for reprap (3d printer) extruder hot end. Another project was a temperature sensor for laboratory application - temperature logger that required accuracy at least 0.1C in the range of 10 to 200C.
I spent a few months with no success, I cursed AD595 - huge amount of electric noise was collected by a thermocouple that provided a temperature jump of 100C when I turned on stepper motors that was unacceptable at all. My laboratory unit was totally miserable - my temperature reading was stable within +/-3C that was far too high.

I read AD595 (AD594) datasheet, a paper by Analog Devices with a promising title: "Taking the Uncertainty out of thermocouple measurement..." - you can easily find this pdf in the internet. I read this stuff, but the information was rather misleading (not mentioning a few mistakes in the schematics in the latter file). For the last 3 weeks I spend 2-3 hours every Friday evening trying to solve the problem. I switched to high-resolution ADC (MCP3426) to increase my resolution, but electrical noises were still immense.
Fortunately, I found a very good page about thermocouples by National Instruments - it looks that these guys do know a lot about thermocouples.

The solution was simple - you MUST NOT follow the recommendation of the datasheet file connecting one lead of the thermocouple straight to the ground! Instead, you must include a (pull-down) resistor, 10k-100k, between one thermocouple pin (pin 1 or pin 14, it does not matter which pin) and the ground. This greatly decreased the electric-noise-induced current picked by the thermocouple, improving your accuracy to about +/-0.1C.

 

Further improvement can be achieved using a low pass filter after the AD595 (pins 8 and 9). It is easy - you should find impedance of your ADC (see datasheets) that is an order of 10 MOhm, calculate a capacitor needed to attenuate any noises with the frequency of an order of 1Hz - use the calculator. For my ADC with the impedance of 25MOhm, I used 0.47uF tantalum capacitor that improved my accuracy to at least +/- 0.02C. A brief note, different types of capacitors do differ. For low pass filters, use ceramic, polypropylene or tantalum capacitors only. Do not use electrolytic or paper-based because the latter have either high leak current of significant inductance (try to google for more details elsewhere).

PS Do not try to insert a low pass filter into a thermocouple circuit - it creates more problems. Just use one after the AD595.