ADS1248: PGA temperature effects and calibration questions

Mark Pearson

Part Number: ADS1248
Other Parts Discussed in Thread: DAC8562, LMT85

My customer has the following:

The PGA (per the data sheet) has a tempco associated with supply voltage, data rate and specific gain. For reference see Fig's 17-20: 5V and Fig's 21-24; 3.3V of the data sheet. We set the gain to 16 and 32, use a 3V supply and 2.5V reference. We have a operating temp of -20 to 55 C. Our testing shows the tempco impacting our system performance. Without going into circuit detail, we actually used this ADC to servo a TI DAC output. Works well … at room temp.

ADC calibration: However, our software engineer has read on the e2e forums, that ADC calibration can be painful to constantly tweek the gain cal constants. Question: The curve shown on Fig 23: is that THE gain vs temp curve, or is that a typical curve? If just typical, do we need to do a two point against the PGA, for each ADS1248 to fit and fix this gain skew? Or, do you folks have some recommendations on calibrating out the gain & offset errors over temperature?

I don’t see how I can add a circuit to compensate. This is in a thermocouple measurement circuit. We do know the cold junction temperature. There are also the built in temp sensing diodes. I have checked “e2e” but nothing jumping out at me.

My associated questions:

1) PSRR is 100dB, min. Is this over temperature?

2) I believe one can do a self-cal (either at 25C or at Temp) – as outlined in section 9.4.5.3, page 40 of the datasheet by using

SELFOCAL. yes, there is a manual version by using SYSOCAL, but it sounds like we want to avoid this from a S/W perspective.

Gain error is calibrated with the SYSGCAL (and from what I can tell, this is a self-cal test). Correct?

Thanks!

over 5 years ago

0 Joseph Wu over 5 years ago

TI__Guru* 92715 points

Mark,

For the gain vs temperature curves in the Typical Characteristics section, I would call these typical curves for the gain, not exact. I'll skip to the two questions that you ask at the and then come back to the gain/offset calibration in more detail.

1. The PSRR is listed in as 100dB as a minimum in the Electrical Characteristics table. This should be good for the temperature range. I would note there is some variance with PGA gain. This is noted in the graph in Figure 32.

2. For most systems, we recommend a self offset calibration and no gain calibration. For SYSOCAL, this is most useful when there is some sort of front end outside of the ADC that contributes some large offset. The SYSOCAL would remove this offset. As for the gain calibration, the FSC register stores the gain calibration coefficient. That value is updated with a gain calibration. Generally, the values trimmed into the part will give good gain performance and are trimmed for each PGA gain.

In the ADS1248, the input and reference are capacitively sampled into the ADC (the modulator of this delta-sigma type ADC). The FSC is for finely calibrating the difference between the input capacitors and reference capacitors.

As for the temperature coefficient errors, it's hard for me to give recommendations without knowing the accuracy the customer is looking for. Also, I'd like to compare those numbers with the drift they are seeing in their system. What is the magnitude of the offset drift and gain error drift they get in their system? I would imagine some of the gain drift they see is the drift of the external 2.5V reference voltage. What external reference are they using?

For most thermocouples, the accuracy is going to be worse than that of RTD measurements based on the sensor itself. For most thermocouples, I would have guessed that a 16-bit ADC would be enough for an accurate temperature measurement.

Joseph Wu

0 Mark Pearson over 5 years ago in reply to Joseph Wu

TI__Genius 12705 points

Joseph,
Here are the Customer responses:
"I shall try my best to answer this with circuit basic description. A DAC8562 set’s a simulated thermocouple voltage. An analog diff channel on the ADS1248 reads that. Circuit servos based on the read. There is an op amp between the DAC and ADS to gain (actually divide) the DAC output based on 2.5V reference. At room temp this works extremely well. We observe a fairly linear response due to temp between -20C and 50C. I would say at the extremes, we are out 30uV – 40uV. We have a spec of 5uV. Maybe we are crazy. Especially when I compare to a competitors product.
We have implemented an offset cal tied to reference cal. This did help a great deal between 23C and 50C. However, -20C show the 30uV to 40uV.
No offense, but a competitor 2.5V reference is being used. However, I would expect that the reference being the same for both the DAC and ADS, these would ratiometrically wash? "

For the 16-bit resolution observation: "Normally 16 bits, but wanting those uV of TC connections measurement – the ADS1248 looked like the correct part."
I hope these responses help...

0 Joseph Wu over 5 years ago in reply to Mark Pearson

TI__Guru* 92715 points

Mark,

I had thought this was a simple thermocouple circuit, but it looks like there are a few more parts to their setup. I'd still like to know what the components are even if they aren't TI parts. It just helps to track down the drift characteristics of the all the components involved. This would also include the op-amp used as the buffer. It would help to have a full schematic to look at where the sources of error come from. If there are other components (series filter resistances, protection diodes that have leakage), there may be other errors to consider.

There will be some ratiometric cancellation if the DAC and the ADC have the same reference, but then the DAC also has it's own gain drift which needs to be calibrated out. I'm not sure if the DAC or ADC has the larger drift both seem similar.

To start, I think it would be best to have some precision measurements of both the ADC input (or DAC output) and the reference over temperature. This will give some idea if the the ADC measurement is really drifting, or if the DAC output has a larger drift.

I would also note that most thermocouples have a fairly large tolerance error. The best thermocouples have an error of ±0.5°C plus a tolerance proportional to temperature. If a precision measurement is necessary, using an RTD is probably a better solution.

Joseph Wu

0 Joseph Wu over 5 years ago in reply to Joseph Wu

TI__Guru* 92715 points

Mark,

I haven't heard from you for a while. Has your customer isolated the drift errors for their system? I think it's important to consider the contributions of drift for all the components in the system and I was waiting for a full circuit and data on what was coming out.

If they've been able to solve their problems great, I'll close this post. If they haven't then you can always post back and re-open this thread.

Joseph Wu

0 Mark Pearson over 5 years ago in reply to Joseph Wu

TI__Genius 12705 points

Thanks for checking in Joseph. It turns out that part of the problem was that their design was violating the PGA common mode range:

"Page 5 states an external Vref should be less than your analog supply minus 1. So let’s say you have 3V analog but stick in a 2.5V external reference. What bad things might happen?

Another question. When measuring this external ref, data sheet states you get Vrefext/4 … APRROXIMATELY! Page 35. What is approximately? If I am using the external reference for other stuff and the ADS1248 hardware is setting bit weight based on the actual reference, what is the impact?

Now we have done external ref and made a measurement using the MUX internals. We see maybe 4 – 5 mV of difference between the ADS1248 measured result and me using a DMM. Are we missing something?"

And my answers to the designer:

Ah, yes – with respect to the External VREF Signal level. Well, you won’t be blowing up the ADS1248, but you will be violating the PGA common mode input range.

I would refer you to page 72, section 10.1.3 of the datasheet

See page 27 for the 9.3.2.1 PGA Common-Mode Voltage Requirements.

For the “…you get Vrefext/4 … APRROXIMATELY” question. Yes, I saw this too, though I never really looked into it before. I will look into it for you now.

Question: We see maybe 4 – 5 mV of difference between the ADS1248 measured result and me using a DMM. Are we missing something?

Answer: I don’t know enough about the situation to say with a high certainty. I would need to know where the precision source is (are we measuring the External VREF – or something else). How long are the DMM leads, their impedance, and the input leakage current? How much noise are they picking up (this foes to the fact that many DMMs are measuring RMS voltages and not true DC. I assume a calibration was done on the ADS1248, yes? If so, then the ADS1248 (input referred Vos) would be less than +/-15uV.

I am using an Agilent 34401A. When I measure the external reference, heck it is dead nuts. That is, I measure 1.8V (or 2.5V when using it). Standard DMM probe leads – maybe 3 feet or less. I am struggling with my ability to measure the 1.8 but when we ask the hardware it reports, let’s say 1.795V. Using that for all calculations when the bit weight is actually based on the 1.8 V would cause problems, yes?

Here is the example. Using an LMT85 for our cold junction. I measure the voltage with the DMM. Ask the ADS1248 to report that voltage it is off by maybe 7 to 8 mV. This is using the external reference. Then we use the internal reference. Heck, the ADS1248 reports the correct LMT85 voltage.

From the attached schematic, if I supply a reference into REFP1 and REFN1 as shown, let’s say 1.8V, I would expect the hardware to report 1.8V. Am I doing something wrong? When I review the RTD circuit example in the data sheet, basically same gig, or so I think.

Question: is it OK to connect REFN1 to circuit common (ground). Example?

Answer: Yes – connect the reference GND to REF1N and to VSS (if VSS in not at a different potential). Sure, for an example – see the EVM schematic.

Joseph - if you could comment on the question" When measuring this external ref, data sheet states you get Vrefext/4 … APRROXIMATELY! Page 35. What is approximately?" I would appreciate that.

0 Joseph Wu over 5 years ago in reply to Mark Pearson

TI__Guru* 92715 points

Mark,

Just to be clear, the external reference sampling should be set so that VREFP is below analog supply voltage minus 1V. The consequence is that you might get some bad readings. The source of this error is because of the method of sampling for the reference into the ADC. If the reference is a higher voltage, the sampling might be such that some of the reference sampling charge is lost as it is sampled and discharged into the ADC. This could cause a large gain error/non linearity. This particular specification doens't have to do with the PGA input range, but rather problems in reference sampling.

The VREF/4 measurement in the system monitor is a pretty coarse measurement. Recently I was asked about the accuracies for this measurement, and you can find a response here:

e2e.ti.com/.../708493

In short, I'd guess that the accuracy is ~10% based on old final test data. This coarse measurement is designed to be used to check to see if the system is still functional. For example, if you're measuring an RTD and the RTD burns out. If you're using a ratiometric measurement, you can check to see if any current is still going to the reference resistor. If your using a 3-wire measurement, and there should be 2mA going into a 820Ω reference resistor, you could expect to see 1.64V. However, if your system monitor showed that the reference was 0.0V, you'd know that the RTD burned out and is now high impedance. Otherwise, you might think that the RTD temperature was so high that the measurement is out of range.

In reading the rest of the text, my guess for their error is that the external reference is off from the expected value. Just as any ADC, the input voltage is sampled and compared to the reference and the output data reports a value that is proportional to that measurement. If they are using the internal reference, and the reference is perfectly accurate, then the output data for a 1.8V measurement should be (1.8*2^23)/2.048 = 7372800d or 708000d. However if the reference is off, then the ADC will be off. If you measure the same voltage and the measurement low by 1%, then it's likely that the reference is high by 1%. To check, the customer could just measure the external reference voltage with the 34401A (and they could compare it to the VREFOUT voltage too). If they have an external VREF, what are they using?

I did look at the schematic that you sent via email. I just had a couple of quick comments.

- R27 shouldn't be populated because the AINN would be at ground and that's not a valid input point.

- D15 works as a good DC level shift to lift AINN off the ground point.

- VBATTST is connected to REFP0, but it does no good if REFN0 isn't connected.

- There are a few TVS diodes (D9, D11, D12, D13) normally you would think of them as not part of the circuit unless theres a HV event, but the current leakage might be significant. Because they use high value resistors, this might cause problems with thermocouple biasing. I'd check to see that the thermocouple is still near the midpoint between supplies.

I think that should cover it. If I missed something or you have other questions, just post back.

Joseph Wu

Data converters

Data converters forum

ADS1248: PGA temperature effects and calibration questions