Peter,

I agree about a separate ADC as we two have mentioned before. I started out my company with a design that was based on a very good ADC. As I mentioned, the accumulating errors in dealing with ADCs led me to remove ADCs altogether the following year to achieve a level of accuracy my competition couldn't accomplish.

Granted most implementations of ADCs don't need to deal with accuracy to sell. If the customers don't know the difference, it can be wasted effort and wasted profit. But if anyone can accomplish a product, there is generally not much profit in it. :)

ADCs in a microprocessors are the ultimate application of this ignorance principle. You can minimize circuit board layout, sometimes cost, and if the customer doesn't care... GREAT!

The best you can do with an ADC is select an external chip, layout the board so it has an isolated analog power and ground plane from the digital planes, use Excel spreadsheets to properly pick the right bit resolution and update speed for the intended accuracy, add all the normal analog bus decoupling caps, choose ratiometric ADCs if you're advanced and figure out all the ramifications of that, maybe add a thermal measurement sensor to allow compensation due to thermal induced error and then calculate your total system error rate.

Few people seem to realize that you have so many modules compounding the error in processing ADC results:

SENSOR or Signal:
()Error of encoding to voltage signal.
()Error modified by thermal range.
()Error modified by power noise.
()Error modified by signal transmission.
ADC Stage:
()Error of decoding voltage signal.
()Error due to bit resolution.
()Error due to ADC cycle time.
()Error modified by thermal range.
()Error modified by power noise.
MICRO Stage:
()Error due to update rate.
()Error due to bit clipping for algorithm.
()Error due to ALGORITHM selected.
()Error due to computation cycle time.
()Error due to binary to decimal conversion.
()Error due to display/report rounding.

Of course most people don't understand how to combine these error rates for a total system expected error.

I work all this out in an Excel Spreadsheet BEFORE I select chips. I've had subcontractor design engineers recommend ADCs that were completely unusable for the expressed purpose. Apparently, they don't understand the WHOLE PICTURE of these chips that require expertise in both analog and digital implications.

After calculating the error expectations on numerous systems I tend to smirk when I see a numeric display with a value more than one decimal point. Sure an engineer can easily believe he's gone through all the proper steps to post several digits of resolution but its only value may be in expressing RELATIVE readings and not accuracy as they expect.

As a businessman/engineer, I've won't share my solutions because those secrets keep my business popular. I take great pride in the knowledge that one of my primary competitors spent 18 months trying to accomplish our early ADC capability and aborted the whole project. And we've advanced even beyond that. :)

So, if there is any lesson to the general reader of this forum absent my secrets, its that your choices regarding how many digits of display resolution or how many bits of resolution you transmit to other modules, is usually just a marketing decision to attract customers beliefs, not the true accuracy of your products capability.

aka j