Filtering thermistors

[Gibson486]

My co worker and I were doing a design review, and we got to the thermistor circuit. We got into a discussion about the pros and cons of adding an rc filter. Do you guys normally do that? Personally, I have usually just filtered in software if it went straight to a micro and an ADC that had low offsets, but I have added rc filters if I ended up using an op amp (added to the output) before the ADC. What have you guys done?

[Mattjd]

depends how much environmental noise there is (ie shit around the thermristor influencing it, that isn't supposed to be)

[Sal Ammoniac]

What are you trying to filter?

Normally there are two reasons to filter a sampled input signal: to eliminate noise and to filter out signal above the Nyquist frequency of your sampling rate (anti-aliasing filter). For a thermistor the sampling rate is low enough for the second case not to be an issue.

I've never used a filter between an ADC and a thermistor (or any other temperature sensor). Given the low data rates it's trivial to filter in software using simple averaging, an exponential filter, etc.

[daqq]

Well, the environment I used them in was very noisy and they were user removable. So, yeah, I added a pretty slow RC filter - nothing special really. This served both to filter the signal (this was a VERY slow system) and it also added a parallel cap to the ADC, which might be a good idea in some cases. Also, there was ESD/overvoltage protection since the thermistor could be plugged into the system.

[Siwastaja]

Often the reason for the RC filter for a thermistor circuit isn't the noise filtering, but because it's a lot cheaper and smaller solution to provide the ADC with the low enough input impedance than the standard opamp follower circuit. A 0.1uF BOM reuse MLCC is typically fine for this job. This limits the allowable sample rate compared to the standard opamp solution, but with a thermistor, this is not an issue.

Some MCUs have configurable sampling time that can work with input impedances in kilo-ohm range, applicable for many thermistors directly, but this slows down converting other ADC channels as well.

ESD protection and noise filtering come as an extra.

[Kleinstein]

The is also filtering for the other signal direction: the capacitor also tends to reduce EMI coming from the µC. So having at least some filtering is a good idea. 

The thermal signal is usually slow, but this does not directly says on how fast the sampling is and thus which frequency is relevant for anti aliasing. Externally coupled noise / spurs could also cause aliasing. So it might be a good idea to sample at least fast enough to get the usual line related crap and do averaging over multiples of a line period.

P.S. not all thermal signals are slow, there are pretty fast systems if small. The Inkjet print-heads are such an example. They heat up to some 200-300 C in a ms.

[David Hess]

The input filter is also needed to block RF which could cause DC offsets that cannot be removed in software.  This is a common problem in DC signal conditioning which shows up as mysterious DC offsets in the input stage and low frequency noise.

[lukewren]

Normally there are two reasons to filter a sampled input signal: to eliminate noise and to filter out signal above the Nyquist frequency of your sampling rate (anti-aliasing filter). For a thermistor the sampling rate is low enough for the second case not to be an issue.

Low sample rates mean more of your input is above the Nyquist. More noise will be aliased into your band of interest, and filtering after you've sampled won't be able to eliminate this. OTOH, an analog filter can remove noise before it gets aliased down.

However, if your band of interest is still well below your Nyquist, you are still able to remove a lot of the noise after the fact -- but some noise was irreversibly added by sampling. It sounds like this was the case in your projects.

 I also trust you when you say that the systems you've built work fine I guess this was just a slip of the tongue with sample rate vs band of interest