Troubleshooting Unstable ADC Reading for PT100 Signal Circuit

[Chandra214]

We have a four wire PT100 signal conditioning circuit, which transpires resistance to voltage.
7mV constitutes one degree celsius. we are using a 12bit ADC to decode the analog input.
When measured from a true RMS multi meter, the ADC input voltages are stable upto millivolt ( 4 1/2 digit multi meter). But upon reading the ADC values in software, we are unable to see the same stability in the voltage. the ADC shows a fluctuation of few millivolts.
Can somebody throw some light on the issue.

Thanks,
Prathap Chandra

 

[MATLABdude]

You may be seeing several (different) effects here...

1) What is the sampling rate of your DMM? Probably lower than that of your ADC.

2) How much bits of instability do you see? (least significant bit, least significant two bits, three?) Are you sampling at an appropriate rate (greater than the ADC settling time, for the given system impedance?) Given the extrema, is this more than a degree or two? Is this acceptable for your application?

With that in mind, you can buy ICs or modules (meant to work with PT100 platinum sensors) that have guaranteed accuracy of 0.1C or better (and either a digital output, or amplified analog output).

 

[Chandra214]

I must admit, Am feeling like a novice now..
The sampling rate of the DMM is 3times/sec, where as ADC sampling rate is in MHz range.

We see around four bits of instability.

how do i calculate an appropriate sampling rate for a given impedance?

right now we are getting acceptable values, but we are trying to achieve accuracy.

 

[f95toli]

With an ADC you will normally have to integrate to get any kind of stability(the multimeter is already doing this internally). I am note sure if you are referring to the input of a commercial DAQ (e.g. from National Instruments) when you write "ADC", if so there should be a setting for this in the drivers.
Otherwise you simply have to do it manually, there are all sorts of "tricks" one can use to e.g. get rid of 50/60 Hz noise etc; but one can usually get away with simply calculating the mean value of a few hundred samples (depending on sampling rate).

Btw, a fluctuation of a few mV is actually reasonably good and don't expect a huge improvement no matter what you do; ADCs are inherently noisy so one should always try to feed them a signal that is as high as possible.
In practice this means that one should always use a pre-amplifier (with a suitable BW) when dealing with low-level signals and get the signal into the 0.1-1V range or so before sending it into the ADC.

 

[Chandra214]

with sampling hundred times and taking its mean is serving the purpose.
we are getting readings with +/- 0.25 degC.
Thanks for your inputs.

 

[Topher925]

Is this for a thermocouple or RTD? Thermocouples are by their nature very noisy, 0.25 deg C is pretty good for any regular thermocouple especially with just a 12-bit ADC.

A multimeter isn't going to tell you much in terms of fluctuation or the quality of the signal. For that you need an oscilloscope. Otherwise you won't know if its the signal or the ADC that's generating the majority of the noise.

 

[Chandra214]

PT100 is a RTD
Measurement using scope showed ambient noise level of 20mV

 

[vk6kro]

Also, check the power supply for the ADC. Some of these effectively use the power supply as the reference voltage for the A to D conversion.

If the supply voltage has quite small hum or noise levels on it, it may be different at different times when the conversion is taking place.