MIME-Version: 1.0
In-Reply-To: <54F98855.4060903@optonline.net>
References: <CAF3uCZ=G0FztqBtpfvssYCz9WrUORX8+J45xmH9+KutaX6c+gw AT mail DOT gmail DOT com>
	<201503051621 DOT t25GL09H018380 AT envy DOT delorie DOT com>
	<54F98855 DOT 4060903 AT optonline DOT net>
Date: Fri, 6 Mar 2015 22:34:18 +0530
Message-ID: <CALT8Ef78+6GhFaJzCymGLaAZBVbR-KcZgqt=Qn=KnkvOZw+uMw@mail.gmail.com>
Subject: Re: [geda-user] [OT] Temperature sensor and control recommendation
From: Shashank Chintalagiri <shashank DOT chintalagiri AT gmail DOT com>
To: geda-user AT delorie DOT com
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Reply-To: geda-user AT delorie DOT com
Errors-To: nobody AT delorie DOT com
Precedence: bulk

Getting precision out of a temperature sensor is fairly
straightforward. Oversampling and decimation works just fine, like
everyone's suggested. RTDs excited by a garden variety opamp current
source measured with SARs or even most SD ADCs are usually noisy
enough to give you the necessary entropy to do this. Thermocouples are
a little _too_ noisy to give you that sort of precision convincingly,
though you can always produce a readout with that many digits. If you
update it slowly enough, it'll look like it isn't actually as noisy as
it is.

On a practical note, though, getting that sort of _accuracy_ is a
whole different sort of problem. You need to solve problems such as
thermal resistance / coupling to whatever you're trying to measure the
temperature of, related lags in measurement, and sensor calibration.
Standard tables will only go so far, if you're really trying to hit
0.1C accuracy. Similar offsets and gain errors in any conditioning
circuits you may have are relatively simpler to deal with, though you
should keep in mind what the ambient temperature of the measuring
circuit is going to be - usually that tends to cause a long-term drift
and / or wander which is almost impossible to get rid of in software.

On Fri, Mar 6, 2015 at 4:28 PM, gene glick <geneglick AT optonline DOT net> wrote:
> On 03/05/2015 11:21 AM, DJ Delorie wrote:
>>
>>
>>> I'm facing the need to control the temperature of a small sample in a
>>> chamber between room temperature and about 450°C with a precision of
>>> 0.3°C or better.
>>
>>
>> I use a thermocouple to monitor my woodstove, but I don't care so much
>> about precision.  I use a DS2760 thermocouple kit from Parallax for
>> it, and a high-temp thermocouple probe from Omega.com.
>>
>> For monitoring my geothermal system, I used RTDs and an MCU's ADC to
>> measure them.  I got extra precision by doing each measurement 64
>> times and averaging, and the tech who calibrated my geothermal system
>> says they're spot-on.
>>
>> I use the same averaging trick on my thermostats to get 0.1F readings
>> on a 1C-rated sensor.  If your sensor isn't noisy enough to use this
>> trick, you can always add noise - you're basically building a 1-bit
>> ADC.
>>
>>
> we use this trick at work as well - converts a 12-bit ADC to 16-bit
> precision. I think the relationship requires you need 2^n samples for each
> bit of additional resolution. This is off the top of my head, but is
> probably close. So if I remember correctly, we had to sample an extra 16X to
> get the additional 4-bit precision. It works really well. But yes, it relies
> on noise to work and assumes the measurement is relatively stable over the
> sampling period.  I believe the technical term for this trick is decimation.
>
> gene
>
>


-- 

Chintalagiri Shashank
Indian Institute of Technology, Kanpur

http://blog.chintal.in