External I/O (Smartstream, etc.).

[matthiaspaul]

(09-16-2015 05:36 AM)cyrille de brébisson Wrote:  The basic issue is the precision that you need to have on the analog part in order to differentiate between the probes. [...]
They are identified by resistors. They are around 75 different probes.

From a today's engineering point of view, I would call this a design fault - either a left-over from how someone would have approached this more than a decade ago, or an outgrow from aggressive (and misled) cost optimization, as for just a few cents more in parts one could use f.e. a serial EEPROM or other chip to identify probes via a digital protocol (through a (1- or) 2-wire interface). I take it that your data logger was interfacing an existing external design which you could not influence.

(The resistor approach is still okay to distinguish between perhaps a dozen different "probes", but IMO not for such a high number of probes. For comparison, Sony has introduced a new camera hotshoe named the Multi Interface Shoe in 2012. It uses three (analog) ID pins, of which at least one is connected to a small number of different pull-up or pull-down resistors in the devices and by which the camera can detect the attached type of device and select the corresponding interface. Devices are flashes, video lights, cabled and wireless microphones, GPS receivers, monitors, various adapters etc. Perhaps not the smartest approach for a future-oriented interface in 2012 either, but it is obvious that the number of possible types of devices is quite limited by the nature of the application "camera hotshoe", so it still works.)

Quote:Since the sensing circuitry is all 3.3 or lower voltages (but input are in the 5V range), you need to divide the inputs before digitizing them.
So, the analog and digitizing needs to be accurate to within 1/75=1.3%. This is VERY hard to achieve...

While it is possible to achieve this accuracy with the low resolution 8 to 12 bit ADCs built into microcontrollers, from own experience I know that it can be difficult and requires a good ground and power plane design and possibly a carefully decoupled analog reference input (and 1% or better resistors, of course).

But since the value of a probe's resistor does not change (putting temperature etc. aside for a moment), you can extend the time it takes to measure the resistance considerably (from milliseconds to perhaps half a second) and apply some filtering in the analog domain (f.e. capacitors) or in post-processing (averaging multiple measurements in a sliding window approach). Also, one way to greatly improve results is not (only) relying on the ad-hoc approach of using a resistive voltage divider, but to use an active driver circuit into the ADC input. This will also help eliminate the influence of the ADC itself. Alternatively, you might "translate" the resistor value into another property which can be measured with higher accuracy without relying on the internal ADCs, f.e. let the resistor change the frequency of a (voltage- and temperature-compensated) multivibrator (or compare against a reference multivibrator), and then use one of the timers to count impulses within a given timeframe.

Greetings,

Matthias