Hallo Ashish,

the key question of this grounding issue is always: What is happening on digital ground?
If in the moment of conversion digital noise on digital ground plane is negligible, then the separation between both ground planes theoretically is not needed. But who knows, what is happening on digital ground plane? If you don't put digital circuitry into sleep or idle mode (if this is possible at all), then it's rather unprobable that digital ground noise is negligible. Then, it's better to keep digital and analog ground planes totally separated and to connect them at only one point. By this any intermixing of digital and analog ground return currents is widely eliminated. Provided, that there are no other effects, like charge injection via stray capacitance or magnetic coupling of digital signal lines to analog circuitry.
Where you make this single connection between both ground planes depends on your application: If you have only one ADC or DAC, then you can make this connection right at the chip. But if you use several ADCs and/or DACs, then it's better to make this single connection at power supplies. This is especially valid if needed precision is high, means if 14bit or 16bit converters are used.
Please have also a look at this datasheet

http://www.analog.com/UploadedF...D660_a.pdf

and read the chapter 'grounding'.

So, it's not wise to connect analog and digital ground planes under each chip again and again, as you mentioned this. This will cause digital ground return currents also to flow across analog ground planes, which definitely must be avoided.
This might only work in very certain cases: When 8bit converters are used, with no need of high precision, while digital circuitry is quiet in the moment of sampling and converting, etc.

Another word about digital ground return currents:
These annoying currents, which make the ground potential vary between separated points and introduce ground noise, come in two flavors:

1. Digital ground return currents as consequence of 'communication' between digital chips. Here no filtering is possible. Relevantly eliminating ground noise by inserting low pass filters into digital signal path would corrupt digital signal integrity, which will cause a desaster.
Such currents must be hindered from running across analog grounds. If they are flowing during sampling and converting phase, all the precision of converter will be lost.

2. Digital ground return currents resulting from supply current changes of digital chip. Here filtering can be extremely helpful! You can filter out this ground noise to such a low amount, that the resulting ground return currents might be allowed to flow across analog ground planes, especially if precision of converters isn't very high.
This would allow you to use one common supply voltage for both analog and digital circuitry and make it available for the digital circuitry by intensive low pass filtering. This would also allow you to ease the design of single point grounding methodes, as this single grounding point can be made nearly totally free of digital noise then.


Kai