Hallo Nagarajan,

my reply was addressed to a certain type of 'engineers'. People, which do not make 'any compromises', which do make the 'full step' and think that they get 24bit resolution when they buy 24bit resolution. But when you see their circuits, you will realize very quickly, that finally they will not even have 10bit accuracy, because they do NOT use these techniques, which are neccessary to KEEP this 24bit resolution.
I have studied physics, so I know very well about these very precise devices. I have seen even more precise things. But the truth is, that this extreme precision is not for free. And with my little estimation about loss of accuracy, which can be observed when even using stuff which is known to be very precise, I wanted to demonstrate, that there is much more needed, than simply to use most precise standard components. Even the drift and inaccuracy of only two high precision metal film resistors is enough to totally erode resolution of 24bit ADC. So, in any case lots of additional measures are needed to keep the high resolution.

And how diificult it is to make this job, you tell with your own example: People use 24bit ADC, but they offer 'only' 21bit accuracy. Do you see what I mean? Even these high skilled people loose 3bit! And they are even using temperature compensation and are limiting temperature range. What, do you think, will these people have had pain to achieve this precision?

Please make the following test: Switch-on your balance. Put-on a mass of well known weight. Notate reading. Remove weight, and put-on weight which is much heavier. Remove it and put-on your reference weight, again. Is the reading now identical?
Let the balance untouched for some hours. Make the same game again. Is now the reading identical? What about the reading in two weeks, half a year?
If in all cases reading is identical, showing an deviation of less than +-0.001g, then you have convinced me!

Another point: Specifications about accuracy is often manipulated by statistics. Some manufacturers guarantee a certain accuracy, let's say 1mg, like in your example. But what they guarantee is NOT, that there's no device which shows a bigger error, but something totally different: They guarantee that standard deviation is 1mg. But then, there is a propability of 32% that your actual device will show a bigger error. Correctly spoken, there's no error limit at all. You can end up with an error of 1mg, 0.1mg, 10mg, or whatever you can imagine.
So, even if they guarantee something to you, you don't have any guarantee...

Bye,
Kai