DDC264: INL spec

Hi Jerry,

Maybe let me start from the beginning although I may explain some stuff that you already know...

When you use the DDC to measure a given current, you expect it to give you something very close to the real value but with some error. For instance, if you don't apply any input, you would expect the device to output "zero" (well, actually 4095 in the DDC264) but you may get 4200. The difference between the two is called zero error (as it is the error for zero input). If you inject the FSR (say 100pC if you are in Range 2) you should be getting 2^20-1 code but may get something else. That would be your full-scale error.

If you inject any other value in between you may expect something but will get something else. Etc... In order to give an indication of what that error can be the DS lists several parameters. One way is to sweep the input from zero to FSR and take measurements. Then fit a line to the results and give a description of the line and the error to that line. Of course, there are many criteria to choose that line but regardless, once we choose that line, we put some data to explain how "wrong' that line is respect to the ideal one. For instance, we can give the error on the slope (gain error) and on the crossing with zero (offset error). Then we give the INL (the error, the difference between the real and the line). 

Imagine that no matter what point we measure (what input we put, small or big), the difference between what we measure and the line is within 0.1pC. Then you could say that INL=+/-0.1pC, regardless of the measured magnitude. This is actually the case for many ADCs. In the DDC we choose what is called a best fit line (for the value on the tables, not the graph). It would be too long to explain what best line we pick, but it results on the error actually not being constant, but growing with the measurement. Smaller signals have smaller deviation to the line and bigger signals, have bigger. Hence the INL is given as a factor proportional to the measurement (the reading): 0.025% of reading. But somebody would then think that the error is zero when the signal is zero, which is not the case. Hence the 1ppm.

In the graphs, as mentioned, the line is an endpoint line, so, the error between what you read and the line on the extremes is zero and the difference is biggest somewhere in between. For instance, in figure 1, Range 2 (100pC), if you input 40pC you would get about 400k reading. The graph is showing that actually you may have an error of 20LSBs when you are measuring that value. The error is the difference between what you measure and the end-point line. Nevertheless, remember that the line does not represent zero absolute error. For instance, there may be a 5% gain error... so you would still have to add that to the 20LSBs.

Hope this helps. If not, I recommend you look at http://www.ti.com/lit/an/slaa013/slaa013.pdf or any other book where these concepts would be explained more in detail.

Best regards,

Eduardo

Hi Jerry,

I think there is some confusion on the way to read the INL spec. First term is ±0.025% of Reading. I.e., if the input is for instance 20pC, this error term on whatever the ADC returns is ±0.025% * 20pC. The 2nd term is constant, no matter what the input is. In this case ±1 ppm of FSR. If one is using Range 3, then it is ±1 ppm of 150pC. So, the total typical INL error for 20pC input would be  ±0.025% * 20pC ±1 ppm of 150pC. To this, you would have to add the offset and gain errors...

On figure 1, yes, this is likely generated from one typical device. I.e., other devices will be close to this, but not exactly the same. 

Not sure what you mean for your last sentence... In a general ADC, a typical spec (whatever that is, INL, offset, gain...) is obtained usually as an average of many devices... Certainly in the DDC, the INL for a given device will be different than that from another device  (like for any other spec), but it'll be close to the typical INL, which is what is listed on the DS. It is true that in a generic ADC, usually the INL is a constant value independent of the input to be measured.

Regards,

Edu