[sdiy] Ground planes for mixed signal designs - to split or not to split

Eric Brombaugh ebrombaugh1 at cox.net
Mon Dec 15 16:06:42 CET 2008


Seb,

I'm sure you'll get a lot of feedback on this one - it's a topic that 
comes up often and it seems that everyone has an opinion on it, even the 
folks who've never done a layout in their lives. :)  Add to that the 
fact that even the manufacturer's appnotes often disagree on the best 
practices and you've got a real point of contention. Kind of like asking 
which computer language is better.

For the sake of completeness, I'll chime in too. The boards I've been 
working on lately (ARM + FPGAs + 2GHz RF receivers with very high 
sensitivity requirements) only have one ground connection. The ground 
plane is not completely isolated, rather there are cuts which restrict 
the flow of current loops to a specific path. All single-ended interface 
signals between digital & analog cross over at the area where there are 
no cuts - this allows return currents for these signals to flow as 
closely as possible to the outgoing signal. Differential (LVDS) signal 
pairs are allowed to cross anywhere since they provide their own return 
path. This technique appears to work well.

YMMV, OMG, XYZ, BBQ.

Eric

Seb Francis wrote:
> I've been having a discussion off-list about splitting (or not) ground 
> planes for mixed signal designs.  I'd like to open the discussion to the 
> list as this seems to be a tricky area where there is not always a clear 
> optimal solution.
> 
> The specific case I'm looking at is my Digi-Mod design which can be 
> summarised as follows:
> 
> Has precision analog audio & CV circuitry
> Has high frequency digital circuitry
> The 3 main places where these domains meet are:
> - Audio CODEC (which has both an AGND and DGND pin)
> - CV DAC (which has a single GND pin)
> - dsPIC built in ADC (the dsPIC has both AGND and DGND pins)
> 
> My current thinking is as follows:
> 
> Have a separate ground plane for AGND and DGND, with a separate power 
> connector pin, connected back at the PSU (off-board).
> Have the audio CODEC and CV DAC on the boundary between the 2 planes
> Use the AGND plane for both A&D CODEC ground connections and the CV DAC 
> ground.
> The dsPIC needs to be in the digital zone, but with AGND connection 
> routed from the AGND plane.
> 
> I'm using this split ground approach because it seems right to isolate 
> the analog and digital as much as possible, and also other manufacturers 
> such as MOTM use this approach, but I do have concerns with it:
> 
> On the dsPIC the AGND and DGND pins could be potentially quite different 
> as there may be a lot of inductance between the 2 ground planes.  Will 
> this degrade the performance of the dsPIC ADC?
> 
> I've read that routing signals across ground planes can cause problems, 
> but I am having to route the CV input signals across to the dsPIC, and 
> also potentially some digital signals unless I'm very careful to have 
> the CODEC and DAC exactly straddling the 2 ground planes.  And maybe the 
> MIDI UART signals will have to cross zones because of the position where 
> the MIDI headers are.
> 
> Another thing is that my 'analog' 3.3V power is made from filtering the 
> digital 3.3V via an inductor and capacitor (makes sense to do it this 
> way round as the 'analog' 3.3V is very low current so a small, but high 
> inductance SMD inductor can be used).  However the 3.3V regulator is 
> referenced to DGND, whereas the analog 3.3V devices have ground 
> reference AGND.
> 
> So it gets me wondering if splitting the grounds is really the right 
> thing to do.  And what are the best practice alternatives to avoid 
> digital noise in the analog domain without splitting the ground planes?
> 
> Seb
> 
> P.S. The circuit in question:
> http://burnit.co.uk/sdiy/stuff/Digi-Mod.pdf
> 
> 
> 
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at dropmix.xs4all.nl
> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy
> 




More information about the Synth-diy mailing list