We have a board with an ADC0848 A/D chip that is giving erratic readings on one particular channel on some boards but maybe not all boards. The chann el that reads erratically is used to measure a temperature value that is us ed to compensate a voltage that is read from another channel. An increasin g value on an A/D input (CH8) that is measuring electric current, results i n increasingly erratic/ unstable readings on the temperature A/D channel (C H7).
Has anyone encountered this kind of problem with the ADC0848 chip?
I have never used exactly this one but I do use a very similar one from TI (I2C). Can't think of many other possibilities than charge transfer between neighbour channels; normally you would have a 2k/100n-ish filter on each analog input, this eliminates the problem. If you do have that and still have the problem.... well, no idea. Sometimes new silicon (new fab process on good silicon, too) just does not work, I had recently an ADR441 reference which drifts like a champion (about 10 times the spec), formerly they were just OK.
Not so long ago I had the undervoltage thing on that same 8 bit
8 input I2C ADC I mentioned earlier. One of the inputs is meant to detect a "HV shutdown" request from a HPGe detector (typically it comes when the detector runs out of liquid nitrogen, the vacuum can get compromised and a HV arcing can kill the input FET of the preamp - also cooled and in the vacuum area so not easily replaceable etc.). Well, there are various standards for that and the netMCA has to cope with all, hence using an ADC input (and an on/off pullup to 5V on the input). One of the detector types pulls the signal to -24V and releases it (open collector style) to signal "no nitrogen" (so if the cable disconnects the HV will shut down). This goes through a 100k to an ADC input; I thought 240uA would be no problem for the clamping diodes on the ADC chip. Turned out almost right - it worked, but all other channels got shifted by 1-2 LSB when the -24 was connected. Had to add an additional 18k in series to the cable so with the 3k pullup to 5V the input voltage would be above GND....
That sounds like a less-severe version of what I've seen. In my case we were over-voltage, and dropping more current into the protection diodes. All of the channels went to maximum.
I _think_ that what happens is that there's some parasitic transistors that get turned on when the protection diodes conduct -- or maybe it's just that some bit of what's supposed to be insulating silicon suddenly has a bunch of carriers on it.
At any rate, our solution was a similar kludge to yours, followed by a firm rule for future products that no circuit should ever be able to pull an ADC input outside of the power supply range, even if you had to add components to make it so.
Control systems, embedded software and circuit design
Must be something like that, clearly the clamping diodes are not meant to conduct during normal operation... It must be about the voltage drop over the diodes (I never measured how much it was, saw this only with the case closed, in fact it was some 10km from me :)) like you suggest.
Yep, now I also keep that in mind. Our series resistor was easy to put, inside the housing of a D-9 cable the customer gets to access the HV shutdown input anyway.