Since my GTR is getting close to done, I need to get my Analog-to-CAN boards done. I added some auto-calibration routines (since the resistors in the sampling filter are only 1% accurate) to get the channels equal to each other. Endpoints are pretty close, but a tad off in the center. With the input at 5.001V, the channels all get 5.001Vs in the average, but a few have more noise than the others.. and the maximum noise is +- 12-14mV. That noise is almost certainly because the 5.001V input is from the power rail. If you look at the zero input (all 8 channels grounded) most channels are +/- 1mV, which is not bad considering the board layout and the very inexpensive 12 bit ADC.

Right now I sampling at 400hz and decimating to 100hz, but I’ll up that to sample at 1000hz and decimate to 100hz. Doing 8 channels at 100hz over a 1 mbit CAN bus is pretty easy, as that amounts to 200 messages/sec, or about 3% bus load.

Since this is for automotive sensors like pressure sensors and temperature sensors, 1mv noise is sufficient. With a 100psig sensor in the typical 0.5-4.5V output range, 1mv represents .025psi, smaller than the physical variations I would be measuring.