Sorry to be such a pest, but can you please elaborate on this? Do you mean I should avoid running long signal traces over gaps in the ground plane? If that's the case, isn't that the whole point of adding gaps in the gnd plane, to avoid adding low pass filters to the circuit? Is the concern at that point the lack of shielding that the gnd plane would provide but doesn't anymore?
Right, you're making a bastard plane at this point. It should still have reasonably good conductivity between all points (and especially along the directions of signals routed over it), but it should be somewhat distant from those signals, which is where the holes come in.
Consider if you built the circuit using twisted pairs to carry signals (with ground) between sections (or perhaps twin lead, which at Zo ~ 300 ohms has much lower capacitance than regular ~100 ohm twisted pair). All the grounds tie together at each local ground, and pairs route between localities. (A locale, a subcircuit, would probably be a given amp stage.) Which will naturally lead to some ground loops, but small ones, and controllable by means of collecting together pairs that are running in similar directions. (Or a ladder architecture where power flows right to left, and signal left to right -- as drawn on the schematic -- which is a quite good way to do it. No worry about star grounding, the ladder architecture already guarantees that!)
Effectively, you can expand that a bit by using a pour -- it could also be a "mesh" or grid pour at this point -- giving more conductivity than the pair-routed case: less voltage drop for a given current flow, and so less need to take stock of the voltage drops across ground.
If the above is right, what I was trying to do is to avoid stray capacitance, sacrificing shielding in traces that will either already have significant gain, or won't be amplified all that much (low channel)
Let me know if I got it wrong, please
Yeah, it's not far off, just the cases I mentioned.
It's hard to talk through this without a framework; I can approach this from an RF-port perspective, where a signal is a transmission line over ground, and the low-frequency equivalent circuit arises out of the impedances and delays of those transmission lines. This is very general, and hard to express in terms of a textbook "here's a voltage and here's an [absolute] ground" circuit. (Partly for precisely that reason, that the textbook circuit assumes uniform ground.)
At low frequencies, transmission lines become reactances based on their characteristic lengths and impedances. Which in turn come from the dimensions of the circuit, and the frequencies of interest. At DC, the textbook circuit is indeed the correct case (given you can write down the ideally small, but still nonzero, resistances of ground itself). But that's not a very interesting case. So at AC, you can look at important dimensions, like how close signals are to ground, and their lengths; or the spacing of signals, compared to ground (which doesn't depend on transmission line behavior, very much: avoiding crosstalk or oscillation by sheer distance between signals, or shielding by routing ground inbetween them, works at all frequencies!).
The hazard occurs when, at a frequency of interest, a ground slot (say) is long and wide enough to drop significant voltage, and a trace crosses over that gap -- then the voltage drop across that gap is impressed upon the trace, and interference gets in. In works perfectly symmetrically, where a signal crossing a gap carries a current across that gap, exciting it (allowing interference to get out: say, RFI from a digital circuit).
Probably, you're nowhere near the frequency * length where this even can be an issue, and so it's merely "best practices" -- not anything pertinent. (I mean, I would like to think that's enough of a reason, but obviously, it doesn't have to be.
)
Tim