You are right, the center of current distribution running in the ground plane will be directly underneath the signal trace. But, as inductance of ground plane isn't infinitely low you will have some nonnegligble voltage drop across the path of ground return current in the ground plane of tens of millivolt, depending on rise time of square wave signal. This voltage drop does not allow, that the ground return current is totally located under the signal trace, but will spread over the whole ground plane! Think about how easy it is for the current to flow in the ground plane compared to the current in the signal trace, because ground plane inductance is so much lower than that of signal trace. It's the inductance of signal trace which dominates the inductance of resulting transmission line, and there's no reason, why the current distribution should be totally located under the signal trace.
Another point, which indeed leads to relevant voltage drops in the ground plane resulting from ground return currents is, that with these thoughts allways the points are overlooked, where the ground return current enters the ground plane. Here the inductance of ground plane is much much greater than elsewhere. Alone these ground return current interfaces to ground plane are responsible for relevant voltage drops, independently of actual signal trace length.

Ground return currents will flow in some distance to digital signal traces, relevantly decreased of course, but relevant enough to make trouble with mixed analog/digital applications when sharing the same and unsplitted ground plane!

Please have also a look at the following link

http://www.hottconsultants.com/pdf_files...eynote.pdf

especially pages 11 to 15.

Another argument, why the ground return current cannot be totally located underneath the signal trace is the design rule, not to route digital signal traces close to the edge of PCB, in order not to disturb 'transmission line physics'.

Finally, not to forget that power supply decoupling caps will also contribute to relevant voltage drops on ground plane. These voltage drops are such relevant, that Philips in an application note about the proper use of their video amps recommends the use of separate ground traces for the decoupling caps, being isolated from the ground plane! They accept worse power supply decoupling performance, only to achieve a significant ground noise reduction in their ground plane!!

I have never seen digital PCBs showing less than 1mV ground noise. In most cases, especially when faster digital chips are involved, typical ground noise is in the range of about 10mV, or even more. Such levels will only allow 8bit ADCs/DACs to be used when sharing the same ground plane.

You are right, running most digital applications without ground plane will cause amzingly high ground noise. Using a ground plane will tremendously decrease this ground noise. But unfortunately this is not enough to allow that mixed analog/digital applications share the same ground plane, unless only 8bit accuracy is needed. May be that in some situations, by using certain tricks like invoking idle mode of microcontroller during conversion, stopping any activitiy on other digital circuitry and using slow digital chips a 12bit accuracy application will work with unsplitted ground plane. But this will not really convince me to reject the use of splitted ground planes for my projects.

Kai