Precision RTD measurement - correct for tempco of ref resistor?

[max_torque]

I'm doing a device to read some precision RTD sensors. I've chosen a drive architecture for the 4 wire sensors that uses a simple current source to drive through the sensor, and then to drive through a reference resistor. The voltage across the sensor is fed into the differential inputs of a 20 bit ADC, and the voltage generated by the reference resistor is used as the ADCs reference voltage. In this way, the measurements digitised output is the ratiometric comparison of the resistance of the sensor to that of the "fixed" reference resistor.   This therefore means the precision and drift of the current source are relatively unimportant.

So far, so good (hopefully!)  However, the tempco of that reference resistor is going to dominate things i think, unless i use some ridiculously trick resistor, which i want to avoid. As i have a micro on the device that is serialising the output of the ADC, then i can also use that micro to measure the actual temperature of the reference resistor, and apply a correction factor to the output of the ADC i think?

Does this sound sensible?

The actual temperature of the unit is likely to vary by a maximum of 55 degC from say 5 degC to 60 degC (it's an industrial device, but for indoor / lab use), so for a boggo resistor with a 100 ppm characteristic with a nominal 400 ohms resistance at 25 degC that means a change in resistance of 2.2 ohms, or a change in ADC counts of approx 1400 counts over that range.  Those counts can be easily characterised and added to the ADC output to come up with a corrected output, which is then converted to the appropriate temperature value depending on the actual characteristic of the RTD being used.

[DaJMasta]

Do you need the reference resistor?  If your ADC has a reasonably low drift reference, the RTD's base construction and performance should be pretty good, and the 4 wire measurement basically avoids issues with cable length or quality.

A standard RTD is going to be like 0.5% tolerance or better, a well established tempco that is very linear in the temperature range you're describing (unless it's not the temps you mentioned), and very good stability... so I'm not sure why you'd need a reference, but if you do, it's going to have to be a pretty fancy one.

[max_torque]

If i use a fixed voltage reference for the ADC, then i need to ensure i excite the RTD with a precise current!   By using the reference resistor to also create the reference voltage, i don't need a high quality current source, which is going to be both more complex and more expensive that a single decent resistor.

I quick look shows 0.01% 5 ppm resistors for around the £1 mark (1.3$ US) and i doubt i can make a 5ppm current source for that!

[Dr. Frank]

Hi,
I have built a lot of precision current sources for temperature measurements with PT100 sensors @1mA, or 10uA for special temperature diodes, all going down to cryogenic temperatures, 20K for PT100. I also first used 5ppm/K reference resistors, and trimmed the current to precise nominal currents.. usually 0.01% total accuracy was good enough.. see Lake Shore sources, or do error calculation..PT100 have much higher T.C., so 5ppm/K is fine for 0.01°C I would estimate.
The real problem you'll have are the thermo couple voltages and lead resistances which especially for PT100 will ruin your precision..at 100Ohm and 0.35Ohm/K.
Therefore, you need a good 4 wire connection for the PT100 and the reference resistor, and a switched current source to eliminate these e.mf.
That's the reason, why a switched current source is better.
Frank

[splin]

Correcting for the temperature coefficient of the reference resistor is perfectly reasonable but the TCRs of resistors vary non-linearly with temperature. Metal film resistors in particular can exhibit +ve or -ve TCR at room temperature. Ni-Cr resistors typically have parabolic characteristics where the TCR changes from +ve to -ve anywhere between,  say,  0 and 100C. You can see some examples on page 181 here:

https://core.ac.uk/download/pdf/4147785.pdf

There is also quite good information here, but bear in mind it is pushing Vishay's expensive bulk metal foil products:

http://www.vishaypg.com/doc?63519

Resistors of the same type may have differing TCRs both within a batch and especially between batches. Thus you would need to either:

a)  Calibrate every single resistor over the 5 to 55C range or

b) Thoroughly characterise the TCR curve with temperature of sufficient samples of each batch of resistors. With 5ppm precision resistors however,  the chances are that the resistance variation over your limited temperature range is rather better than the specified 5ppm such that temperature compensation is not worthwhile.

Also bear in mind that the stability of the resistor over time may be a more significant factor than TCR, depending on how long a calibration period you need for your product. Metal film resistors, even the precision varieties costing $1 or more, rarely have any useful time drift specifications beyond a 1000 hour load life spec which typically may be 1000ppm or more. Some precision resistors have annual shelf life stability @ 25C specifications such as the NOMCA resistor network @100ppm/year, but they are few and far between. In reality you would probably see rather less than 200ppm change in the first year reducing to less than 100ppm subsequently, especially if the resistor spends most of its time at room temperature or less.

Problem is that you can't guarantee the time drift without extensive characterisation. If this is a problem, you might be better using a precision wirewound or metal foil resistor but they are costly. Also, all non-hermetically sealed resistors are sensitive to humidity and might exhibit 25ppm shift with seasonal humidity changes if you're lucky or more likely 50ppm or more. Again the data sheets wont help you.

Of course, if this is a one-off project then most of the above is irrelevant...    

[langwadt]

Hi,
I have built a lot of precision current sources for temperature measurements with PT100 sensors @1mA, or 10uA for special temperature diodes, all going down to cryogenic temperatures, 20K for PT100. I also first used 5ppm/K reference resistors, and trimmed the current to precise nominal currents.. usually 0.01% total accuracy was good enough.. see Lake Shore sources, or do error calculation..PT100 have much higher T.C., so 5ppm/K is fine for 0.01°C I would estimate.
The real problem you'll have are the thermo couple voltages and lead resistances which especially for PT100 will ruin your precision..at 100Ohm and 0.35Ohm/K.
Therefore, you need a good 4 wire connection for the PT100 and the reference resistor, and a switched current source to eliminate these e.mf.
That's the reason, why a switched current source is better.
Frank

I've had issues using currents source, because of the high impendance and the two sides not being balanced impedance it gets sensitive to noise

[max_torque]

The real problem you'll have are the thermo couple voltages and lead resistances which especially for PT100 will ruin your precision..at 100Ohm and 0.35Ohm/K.

Therefore, you need a good 4 wire connection for the PT100 and the reference resistor, and a switched current source to eliminate these e.mf.

I'm using 4 wire RTD's so lead resistances shouldn't be an issue.

  When you say "switched" current source, do you mean a bi-polar source /sink, ie using an AC excitation you can generate a positive and negative voltage across the RTD element and take two measurements (or demodulate AC->DC in hardware) in order to cancel any "dc" offset from thermocouple effects?


Luckily for me, my measurement range of temperatures is really pretty small, approx -20 degC to 120 degC, which for a standard PT100 is 92.16 to 146.06 ohms, a range of 53.9 ohms.  At 1mA excitation, that's obviously 53.8 mV for 385 uV/degC, way above typical thermocouple voltages of ~45uV/C.

In my application, i am measuring a relatively high flow water system (~20 l/min) so self heating of the RTDs shouldn't be an issue (because the waters thermal mass dominates massively) so i could push the excitation currents up significantly to further separate the RTD resistance from thermocouple effects?  A 3mA excitation would be 1.3 mW of self heating at 120 degC, compared to the thermal power of the flow at around 120 kW !!

I also could swap to PT1000 RTDs as well if i really need too i guess   




 

[awallin]

I built a pt100 frontend in 2013 for a narrow temperature range around +32C. There's a multisim picture there:
http://www.anderswallin.net/wp-content/uploads/2013/06/pt100_piiri.png
(the ad8554 is single-supply so may work only in simulations, not real life)

For the op-amp I think the final choice was OPA2188. The tempco of the input offset voltage will make a difference, depending on what temperature fluctuations you have in your lab.
The gain-setting resistors were 4x-arrays, tempco +/-10ppm/K matching.
I think the reference resistors were around 10euro each - the best reasonably priced SMD ones I could find. There's two reference resistors to center the operating point around the desired +32C temperature.

It's driven by a Keygilent scanning multimeter (34970A). A DAC-output at +5V drives the Howland-source to produce +500uA through the pt100s, then the voltages are measured, then we switch to -500uA, then measure again. Everything is done very slowly, after setting the current there's a minute of waiting for everything to settle, then a slow measurement, then at -500uA again a minute of settling, etc. Taking the average of both current directions we collect one data point per 5 minutes. The frontend has an overall gain of 1V/K. (so the DMM +/-10V range is +/-10K in temperature)
Attached an image which shows typical performance over one month. The noise is at 1mK level - haven't done a proper analysis of systematic errors, might be at the 10+ mK level.

One problem is that the op-amp inputs tend to rectify RF/AC pickup on the measurement wires. So if you simultaneously run RF-sources (1-100MHz) in the lab then switching these on/off can cause DC-offset in the measured temperatures. We improved this a bit by looping the pt100-cables multiple loops around large ferrite-rings.

Would be interesting to look at Lake Shore or other schematics - are they available?

[max_torque]

wow, that ^^^ is certainly going to town in the name of accuracy!  I don't need to be anything like as good as that!


Question for ADC experts:  What is the difference between adding gain in front of the ADC, vs simply lowering the reference voltage?

I guess gain can be added with better components (lower noise) whereas the noise of the ADC is more likely to be fixed by the choice of ADC used.

For example, if i use a MCP3550   ( http://ww1.microchip.com/downloads/en/devicedoc/21950c.pdf ) which can accept a Vref down to 100 mV, what is the actual effect of using such a "Low" reference voltage?

if the noise of the adc is fixed (2.5 uV rms according to the data sheet) , with a 100mV reference, that noise is equivalent to anything greater than about 15 bits?

[Dr. Frank]

I'm using 4 wire RTD's so lead resistances shouldn't be an issue.

  When you say "switched" current source, do you mean a bi-polar source /sink, ie using an AC excitation you can generate a positive and negative voltage across the RTD element and take two measurements (or demodulate AC->DC in hardware) in order to cancel any "dc" offset from thermocouple effects?


Luckily for me, my measurement range of temperatures is really pretty small, approx -20 degC to 120 degC, which for a standard PT100 is 92.16 to 146.06 ohms, a range of 53.9 ohms.  At 1mA excitation, that's obviously 53.8 mV for 385 uV/degC, way above typical thermocouple voltages of ~45uV/C.

In my application, i am measuring a relatively high flow water system (~20 l/min) so self heating of the RTDs shouldn't be an issue (because the waters thermal mass dominates massively) so i could push the excitation currents up significantly to further separate the RTD resistance from thermocouple effects?  A 3mA excitation would be 1.3 mW of self heating at 120 degC, compared to the thermal power of the flow at around 120 kW !!

I also could swap to PT1000 RTDs as well if i really need too i guess   

I used an on/off switching, because it's more easy but +/- switching is equally good, maybe more complicated to apply.
It depends on the required resolution and accuracy, whether the emf has to be compensated, or not. Simply do the error calculus, or measure once the emf ,to get an idea how your system behaves.. 45uV seems quite high..
I would not increase the current over the standard 1mA, and reduce that for a PT1000 sensor.

[Kleinstein]

wow, that ^^^ is certainly going to town in the name of accuracy!  I don't need to be anything like as good as that!


Question for ADC experts:  What is the difference between adding gain in front of the ADC, vs simply lowering the reference voltage?

I guess gain can be added with better components (lower noise) whereas the noise of the ADC is more likely to be fixed by the choice of ADC used.

For example, if i use a MCP3550   ( http://ww1.microchip.com/downloads/en/devicedoc/21950c.pdf ) which can accept a Vref down to 100 mV, what is the actual effect of using such a "Low" reference voltage?

if the noise of the adc is fixed (2.5 uV rms according to the data sheet) , with a 100mV reference, that noise is equivalent to anything greater than about 15 bits?

How far on can go with lowering the reference voltage, depends on the ADC. The main disadvantage is the noise. There may not be full specs for a rather low reference voltage: something like bias, drift etc. may be tested only for more normal reference. It can still be OK, but may need extra testing if for more than simple home use. The MCP3550 allows for a very simple circuit, but is not very good at noise. Other SD ADC often offer an additional internal gain option, that often does improve the noise somewhat. The internal gain is also usually very stable, as it is done by multiple sampling and not real voltage gain through resistors.

External amplification can be lower noise, but it also needs to be stable and thus may be relatively expensive resistors (or networks) to set the gain. The uncertainty in the gain adds to the errors. The idea of directly using a high resolution ADC is avoiding those gain setting resistors. In the modern times a 24 Bit ADC may be use as a lower cost replacement for 2 precision resistors  .

For the simple circuit, using the reference resistor directly for the ADC reference, it can be a little tricky to turn off the current. Reversing the current could work with CMOS switches for reversal of the current leads. However this will add some leakage currents and cause transients on switching. The reference input may need some extra capacitance for the ADC to work really linear. This may extend the transients in time - so switching would need to be rather slow.

[max_torque]

I'm looking mainly for repeatability and stability of measurement, rather than absolute accuracy.

I need to determine a temperature between around -10 and 110 degC to an absolute accuracy of 0.05 degC  (0.1 degC would be acceptable)  I expect the device to be regularly calibrated (multipoint cal) probably checked every 3 months and calibrated if it falls outside the spec. Cal points will probably be 0 degC and 100 degC, with perhaps a 50 degC check in the middle.

The MCP3553 can sample up to 60 samples per second at 20.6 bits, and my process has a very long time constant, in the order of 10's of seconds per degree (because it's a large mass of water) so i will be able to oversample and remove AC noise in software.  As such, the DC offsets are probably more of a concern, and the variability of those offsets (with time) i guess more important than there basic magnitude (because the calibration will remove fixed offsets)


Using a reference resistor to generate the voltage reference for the ADC directly is good for a low parts count, and doesn't introduce any additional significant offsets into that task, however that resistor will need to be carefully sized.  My RTD is going to change between roughly 90 and 140 ohms, so to use the biggest "span" of the ADC possible the reference resistor needs to be reasonably close to the upper value. However, as the ref resistor gets smaller, the Vref of the ADC falls, the internal ADC noise increases ie more bits are "lost" to each noise quanta:  EG

200 ohm ref resistor at 1mA = 200mV Vref at 20.6 bits, each bit is 0.126 uV, so the claimed 2.5uV of noise is 4.3 bits and my ADC range used is 75% of the available range

400 ohm ref resistor at 1mA = 400mV Vref at 20.6 bits, each bit is 0.252 uV, so the claimed 2.5 uV of noise is 3.3 bits and my ADC range used is 37.5 % of the available range

In reality, i suspect the answer will be the answer so to speak, and that the effects that dominate due to layout etc won't be changed much by minor tuning of the component values?

[Kleinstein]

From the data-sheet it looks like it does not make a large difference weather a 200 or 400 Ohms resistor is used as the reference. The main limiting point is ADC noise and offsets. It looks like the noise is not getting better with a very low reference (it may make a difference from 5 to 3 V).
The points to observe may be the importance of copper traces and the heating effect. Another point could be capacitance at the reference and maybe stray RF.

[max_torque]

I'm going to knock together a quick test pcb and see how it performs i think! too many unknowns to work out at the moment!

Has anyone got any recommendation for precision resistors i can use as reference resistors for system checking & calibration?  100 ohms (0 degC) and around 140 ohms (~100degC)