Accurate 4W measurement using ADC

Have you looked at a dual slope integrating converter (eg the Microchip TC 850 ~ 17 bits)? You get autozero, synchronous noise rejection (60 Hz), inherent monotonicity. Not blazing fast - but temperature measurement never have to be.

Also, if you use a stable current source, you won't have to do the extra measurement and the software correction.

--
Grizzly H.

You might find that being able to reverse the direction of the excitation current is helpful, since that will reverse any thermocouple effects, and you can swap the polarity and average the two readings to eliminate the thermocouple errors.

Be careful where you connect the reference voltage inputs of the different ADCs. Maybe connect them all together, and you might be able to connect them across the reference resistor, to do a ratiometric measurement.

Other things that can be a problem:

1. Input current - many converters draw significant input current which with approx. 100 Ohm RTDs can cause a voltage error a much bigger error than the ADC voltage accuracy and thermocouple effects, and is often voltage-dependent.

1. RF rectification - if you add a buffer op-amp to solve the input bias current problem (or maybe even if you don't), the semiconductor devices can easily rectify any nearby TV stations etc. and you can get a big error that changes depending on time of day, where you stand, whether the cabinet is open, etc.

2. Self heating - there is a compromise between putting enough current through the RTD to get a decently accurate measurement of the voltage across it, and inadvertently changing the resistance due to the power that you are dissipating in the RTD. The compromise might be improved somewhat if you briefly turn on the excitation for a reading and then turn it off again for most of the time, though I don't know what the time constant of thin-film RTDs is.

1. If you are measuring differences between temperatures, e.g. for calorimetry, then you might not care so much about the absolute temperature but be really interested in differences. In that case there can be some benefit from sharing the same reference resistor between multiple channels, so that any unavoidable drift in the reference resistor will only affect the absolute temperature readings, not the differential temperatures (at least to a good approximation, for small differential temperatures).

2. With 4-wire sensors it would be really nice to have a mode where the input wiring can be switched to check that there are no broken wires, like the contact check feature on Keithley sourcemeters. The current and voltage wires should be connected with some reasonable wiring resistance in between but many thermocouple digitisers can't tell if a wire breaks.

1. Leakage check - it would also be nice if the RTD wiring could be temporarily floated to positive and/or negative voltages, to see if there is any leakage current flowing, that would indicate a faulty sensor (e.g. terminals full of water).

One upon a time I had to solve the above problems to improve a commercially bought data acquisition card, involving lots of soldering and voiding warranties.

I should make an accurate RTD digitiser and sell it. Short of buying e.g. a keysight or Fluke laboratory DMM it is difficult to buy such a thing that really works, without the ohmmeter dominating the error budget.

Chris

That's weird. What did the datasheet say about the accuracy? Really, two monolithically matched 24-bit converters running off the same reference should be a lot closer than 4096 LSBs!

Did you run any experiments locally, e.g. shorting the inputs of two converters together (with a nice beefy bypass to prevent their input current spikes talking to each other) and seeing if they read the same?

If they work like that, I'd strongly suspect one or more of the items in Chris's list.

Cheers

Phil HObbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

On a sunny day (Mon, 14 Dec 2015 05:13:48 -0800 (PST)) it happened snipped-for-privacy@yahoo.com wrote in :

Your question makes no sense 100/50e-3 = 2000

2^11 = 2048 An 11 bit ADC should be enough. You must have an other error source.

That assumes the response is linear, which it isn't because I don't use use a CCS, I use a bias resistor because it's cheaper.

That's probably a good A/D, but dual supply voltage and parallel interface. Ick.

The current source is somewhat of a problem because these devices will have many inputs, and multiple precision current sources tend to drive the cost up. An A/D input and precision resistor is cheaper.

Reversing the current does no such thing, the thermal EMF stays the same... Thermal EMFs are canceled when you difference the readings for precisely that reason.

It's an MCP3914, and accuracy is dependent on all sort of things including what the PGA is set to. I opened a ticket with Microchip and apparently wh at I was seeing is considered to be acceptable because it's actual applicat ion is "energy monitoring".

An experiment I did was to measure the voltage right at the diff. inputs wi th a calibrated Agilent meter, measure the reference voltage, and compare t hat to the ADC output. There was clearly an error of a few percent, and wo rse, it wasn't consistent throughout the input range. For the record, I al so use an MCP3901 in another application and it's much more accurate, but o nly has 2 channels.

ature sensor. The resistance being measured is well under 100 ohms, and to tal measurement error cannot exceed 50 milliohms. The sensor is at the end of a cable which has conductors that can vary in resistance, which is the reason for the 4 wire measurement strategy.

accuracy isn't quite where it needs to be. I figured that any ADC error w ould be consistent across channels and so because I'm performing 4W measure ment, the error would cancel out. Apparently since this ADC has completely independent converters the error is not the same across all inputs.

exed and contains only one actual converter. 12 bits may be plenty of reso lution so it's possible that any micro with a 12-bit differential mux ADC m ay work. High sampling speed isn't all that important (10s/s probably ok).

-bit delta-sigma with independent converters to be accurate enough for some thing like this, or is it better to stick with a multiplexed converter? An y other advice? The way I'm doing the 4W measurement is by using a low TCR /high tolerance resistor to supply current to the sensor, then measuring th e voltage across both that resistor, the sensor, and applying math magic in firmware.

.

I did consider most of these, except for the thermocouple effects. That re quires more circuitry and unless I can prove it's really an issue, I'd rath er avoid implementing something like that. I think the issue rests primari ly with the ADC itself.

Right now I'm testing out another ADC used in a different product that is h igher accuracy, but more importantly, only has one internal converter. We' ll see how this turns out.

You can use a delta-sigma, like the AD7793, and use it ratiometrically. Start with a reasonably stable reference voltage. Drop that through a good resistor (Susumu thinfilm maybe) and then through your RTD. Use the drop across the ref resistor as the ADC reference, and then digitize the drop across your RTD. You'll need some other way to detect an open loop.

We have also used multiplexed delta-sigmas for 4-wire RTD measurements, with PPM accuracy, again ratiometric against a good resistor. If the ref resistor drop is not the ADC reference, but measured like any other input, there will be a little time skew between the measurements, which could cause small problems.

At 12-bit level, multiplexing should work fine. A fast SAR ADC can scan the channels quickly and avoid the time skew issue; average mathematically as needed.

We've found the uP on-chip ADCs to be fairly nonlinear, and noisy, so don't use them for precise stuff.

I am surprised that the Microchip part is causing the problem described. Most delta-sigma ADCs are insanely accurate as regards digitizing the ratio of Vin to Vref. Maybe the problem is elsewhere. You need roughly 0.1% accuracy, and should be seeing way better. What's your excitation source? What's the ADC reference? Can you post the schematic?

That's just a dumb vintage technique to avoid computation and is no longer relevant in the modern world...

Yikes, what a pile of crap. Good to know.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

Rather than actually reversing the polarity, consider chopping the power. You should get most of the benefits of AC excitation for a lot less complexity.

I'm with Phil, that I have a hard time believing that a problem that could be solved with a well-calibrated 12-bit ADC can't be touched with an uncalibrated 24-bit ADC. 12 vs. 16 I could barely believe, but not 12 vs. 24 unless the 24-bit part is monumentally crappy.

--
www.wescottdesign.com

Right. Connect two resistors in series, from some stable voltage. Measure the two 4-wire voltage drops, divide.

That should work to PPMs. Something else is wrong.

Analog multiplexers can self-heat a little and change Ron versus time, which can have interesting effects if the RTD and reference voltages are measured sequentially. The reference resistor could self-heat a little too. Any lowpass noise filters can add to the fun. Reference voltage transient loading droop ditto. But none of those is likely to cause the sorts of errors you seem to be seeing.

A few per cent??!!

The Analog Devices delta-sigma ADCs have errors of a few PPM!

There are others, including some specifically intended for this purpose like the Maxim MAX31865 RTD-to-Digital Converter - everything in one package for $3.50 in singles.

An n-way current mirror (MPQ3904 - quad NPN ~$1.50) or a precision current source (LM334

It's not my favorite because it doesn't detect an open loop sensibly, but it is very accurate and eliminates problems caused by measuring the reference resistor and the RTD at different times, which can cause subtle errors. Delta-sigma ADCs are slow, so thermal tails can cause problems.

An RTD system should be ratiometric on one good thinfilm reference resistor, which eliminates all sorts of error contributions. There's nothing old-fashioned about that.

A LTC2420 works with a Pt-100 or Pt-1000 sensor to well below the accuracy limit of the sensor itself. The only component needed is a precision 10k resistor.

--

-TV

It's more "classic" than old-fashioned.

Crystal Semiconductors did sell the CS5520 20-bit bridge transducer A/D converter which supported low frequency, 'reversing DC', AC-bridge excitation, which might have been interesting but Cirrus Logic discontinued it in 2009.

They do appear to be still selling other interesting products.

--
Bill Sloman, Sydney