ADS1248 integral nonlinearity

Hi Markus,

Regarding the INL, 15ppm is worse case while the typical is much less.  In most cases you will see the typical response for the ADS1248.  When looking at the ADC, you should be looking at the total error.  Gain error usually dominates, especially as you get closer to full scale.  Also 19 bits noise free might be a bit optimistic over the range of the input.  There is a lot of potential for external noise that will supersede the noise tables in the datasheet that are based on the absolute best noise of the converter using shorted inputs.  Even the ability to get these numbers is quite difficult in practice and requires careful PCB layout.

The RTD itself will have linearity error that can be quite significant over the range you are measuring.  As there are a number of variables to consider, a simple calculation can often pull you beyond the true total error, like in the case of using maximum error versus typical.  You can start with a factor of INL error for the ADC but it changes over the input range of the device.  The best approach is to take measurements over specific conditions and build a correction table using actual conditions as you will need to linearize both the ADC and the RTD.

Regarding ESD, it is 2kV HBM and 500V CDM for the device.  Quite often environmental conditions will pick up transients that are quite large.  Additional protection may be necessary that will also include EMI/RFI filters.  This filtering can be RC based and/or ferrites.  For sustained transient events you may need TVS/Schottky diodes.  These can be leaky and another possible source of error.  The amount of protection needed will depend on the environment that the device will be operating in.

Another area that is not often considered is the ratiometric measurement itself.  In theory the effects of noise and drift of the measurement can be reduced, but if done incorrectly the measurement can actually become worse.  One of my colleagues has written this application note that is worth reviewing.

http://www.ti.com/analog/docs/litabsmultiplefilelist.tsp?literatureNumber=sbaa201&docCategoryId=1&familyId=2019

Best regards,

Bob B

Hi Bob,

thank you for your answer and the application note.

Using  a proper PCB layout, balanced EMI filters on input path / reference path and correction tables, what would be a realistic, accessible number of "noise free" bits regarding this application?

best regards

Markus

Hi Markus,

This is a really difficult question to answer as it is dependent on many factors.  Of course the starting place is with the ADC conversion itself.  You will never get better than that, however the noise tests and results given in the datasheet are with shorted inputs.  The level of noise typically increases as the input signal increases, and worsens close to full-scale.  You can get an idea of the noise from Table 1 and 2 in the datasheet.  The best case scenario of 20sps at a gain of 32 is about 17.5 bits noise free.

With a ratiometric measurement you can theoretically cancel some of the noise.  This measurement is assumed to have the same noise/phase relationship of the input signal as it is at the reference.  This is better in theory than in practice as a very diligent effort needs to take place so that the relationship of the input signal to the reference is maintained.  Another way to put this is it is much easier to make it worse than it is to make it better.  Am I suggesting not to make a ratiometric measurement?  No, that is not what I'm suggesting.  I just wish to point out one of the reasons why it is difficult to say how many noise free bits you will get.  Component choice with respect to quality/grade will also make a difference.

If you are willing to lower the data rate, you will improve the noise performance.  If you are willing to average, you can improve the noise performance.

Usually there are other errors associated with the overall measurement.  This will be based on the class of RTD used and the amount of linearization effort to correct the measurement accuracy.  Often times noise is not the dominating factor, but it can be if the degree of measurement resolution and range of temperature being measured forces you to use lower gains that increase the resolution size closer to the level of noise. 

Best regards,

Bob B

Hi Bob,

thanks for the detailed answer. I hope to meet the requirements in my application. I'll do my best :-)

best regards

Markus