Analog ground question

[amd]

Hi everyone,
I have an analog grounding question. It is asked a lot but it often seems more opinion than facts. In my application I have a 24 bit ADC and I need to sample a low voltage signal (-10 to 10mV). I made a shielding cabinet around the ADC and then thought about how to ground the thing. I have a 3.3V low noise LDO for the analog stuff and a DC-DC Converter for the 1.8V digital section which operates from 1 to 2MHz switching frequency. So I want to make sure that the analog section is not affected by switching noise (there is bluetooth on board too, also GPS which has a high dynamic range of current consumption all operating from the 1.8V rail).

So here are a couple of things I thought about and do not know if they make sense or not:
1) To route out the analog inputs I used the bottom layer of the 4 layer board and made a track above it (called it guard, it should be a shield effectively) and connected this track only at one point. This shielding track is currently twice as wide as the analog input track. Is this nonsense or useful? (it is semi-transparent on the image)

2) Worst part: Grounding. I first made a common ground not separating analog ground and digital ground. Then I thought I should separate them since digital ground might have switching noise and a lot of people say it is absolutely essential to do so. On the other hand James Bryant of Analog devices has this article that states that AGND and DGND should be joined right at the chip level. (http://www.analog.com/static/imported-files/rarely_asked_questions/RAQ_groundingADCs.pdf)

So I wanted to ask for some inputs on this issue...whats your take on this? I've attached a screenshot of the analog section

Daniel

[f5r5e5d]

why the analog inputs going different ways? probably need differential at that level, should be routed together for thermal balance of Thermocouple Effect as well as the usual reasons

digital interface and power might work better closer together too - enter through the same "gap" in the stiches

[amd]

why the analog inputs going different ways? probably need differential at that level, should be routed together for thermal balance of Thermocouple Effect as well as the usual reasons

digital interface and power might work better closer together too - enter through the same "gap" in the stiches

The inputs go off to electrodes which are physically placed at a certailn distance. The board sits inbetween those two electrodes.

Regarding the digital power: Before I had the common ground plane which I consider going back to, wouldn't it be sufficient to have the ground below the digital data lines to have a current return path close to the lines? There would be a 1.8V layer on layer 3, I would have thought that the decoupling caps for the digital supply can go another way.

Daniel

[f5r5e5d]

are you buffering and/or amplifying the electrode V? or does the ADC input have gain? true differential? what BW? (I assume some delta-sigma few Hz BW?)

how do the electrodes look over frequency - internal to the electronic packaging? shielded cable? benign or RF noisy environment?


on the digital side I'm thinking in terms of moat/bridge type plane dissection for the analog-digital ground

I expect some of the digital signals completing their loops thru the digital V+ since bypass C are never perfect - so routing V+ along side the signal trace cuts loop area