The following are two possible grounding schemes for a sensitive instrument that contains high gain analog circuits and potentially noisey digital circuits:
This circuit is connected to an external power supply over a 1M cable. There are 3 separate ground conductors in the cable running back to the star ground of external power supply PCB.
The scheme in the first illustration above has an issue. The digital noise can be significant enough to bleed into the analog circuit. Originally I theorized that with a large bypass capacitor connected close to the ground of the digital circuits, that any noise would be minimal and leave the analog circuit unaffected. Normally this works because the analog ground has it's own conductor. Unfortunately, this is not always true. Specifically, an LED driver (like CAT4016 or TLC5925) can generate very sharp currents. No amount or combination of local bypassing seems to have much of any effect on the voltage spikes generated by these chips. If I remove the LEDs and leave the digital communication there is no noise. But the load of the LEDs creates large 100mV spikes in the power ground wire. If the analog ground is connected locally to the power ground, this noise will bleed into the analog circuit (it is sometimes desirable to connect the output of an amplifier generating larger currents to the power ground instead of the analog groung for example).
In the second illustration above, the digital ground is given it's own dedicated conductor. In this case, the digital noise problem is completely resolved. LED drivers and digital noise are completely isolated. My test circuit has a SNR of more than 100dB and I suspect it's better than that but my analyzer is a cheap codec-in-a-box thing that runs on 5V so it's dynamic range is limited.
However, here is the problem:
To give the digital ground it's own dedicated conductor, I moved the chassis ground to the power ground conductor. So electrical discharge into the enclosure it will cause a significant spike in the power ground. Also, the power ground is potentially a source of noise that might be coupled into the analog circuit through shields and adjacent traces.
So the question is, is sharing the power ground with chassis ground a good idea? The enclosure would actually be grounded trough this conductor. It should only have power bypass capacitors and maybe some larger AC currents. But there shouldn't be any DC.
Note: Only ground wires are shown in the illustration. The cable would of course have other power wires but I'm reasonably curtain that the problem has to do with grounding and as such only the ground networks are shown in the illustrations. So there are numerous other power lines that have been omitted.
So what do you think? Is it ok to use the power ground as chassis ground?