In Volume 2, Number 13, we were discussing the use of a "picket fence" technique for noise isolation. To continue that discussion, I have invited an outside speaker to address our forum on this topic.
W. Michael King, of Costa Mesa California, is an expert in (among other things) EMC design for high-speed digital products. I've certainly gained a lot of useful insights from talking with Michael over the years, and hope you enjoy reading what he has to say on the subject of isolation.
Comments from Michael King
Yes, my history includes a great deal of multi-gigahertz layout work at the circuit board level, usually achieving about 100 to 110 dB of intra-board isolation. (For example, at ~ 2 GHz, in the hub transceiver for a PCS radio board, it is necessary to isolate the 1.6 W transmitter section from the micro-volts sensitivity receiver sections, with separation distances of a few inches.)
To save rattle, and get to the core of the approach concept, I find that (for me) the method that "works" is to "view" the board as a sequence of functional modules that require various amounts of inter-module isolation. The modules, assuming that they signal-connect to each other, can be thought of something like "boxes" that share a common substrate platform (the circuit board). Then, it becomes only a simple matter of devising on-board shielding methods, using shielding techniques essentially, except that the dielectric constant of the board must be factored into the consideration for transfer impedances, rather than "air".
The normal 1/20th wavelength "fence-post" interval that one might consider if the "shield" on an inter-layer basis was between modules "in air" must be factored for not only the dielectric constant effect, but also for flux fringing (as eddys between the fence posts) between the "modules" that can setup induction-field-transfers from "module-to-module".
In general, inter-module spacing (still using the concept that the segments of circuit on the board are modules related to each other) for high isolation is required to be more than 3X the gap distance between the fence posts, the V-planes need to be undercut relative to the ground planes by at least 50 H (though 100 H is preferred and usually represents the 98 percent boundary), and the fence post intervals need to be ~ 1/20 lambda (tangent-to-tangent) for the gaps between the posts. For inter-module, and occasionally intra-module, isolation, the anti-pad clearances in ground planes can be a major source of coupling, depending on proximity, so for high isolation in the gigaplus range, blind vias are often demanded by the applications.
Hope that helps,