Noise Isolation

I am currently involved in a project where large isolation is required between areas on a PCB (e.g., >80 dB from 100 kHz to 1 GHz). In order to obtain this isolation, we consider applying the technique of "Picket Fence" (Via Fence) for increasing isolation between conductors on a PCB.

I cannot find any references (papers, books, etc.) on that issue, nor is it mentioned in your book. According to your opinion, is this technique effective? How much isolation can be obtained between conductors embedded between ground planes, or between ground and power planes, when they are "isolated" by a "picket/via fence"?

I would be grateful if you could refer me to any references or data on this issue or enlighten me with some comments of your favor. I have gone through the "classics" (your book included) and could find nothing there, and I need some references for the work we are involved in...?

Thanks for your interest in High-Speed Digital Design.

I'm not aware of any good, rigorous research on this topic. One thing I can tell you is that, for parallel lines, the worst case value of the coefficient of coupling K (i.e., the crosstalk coefficient) is approximately:

K = 1 / (1 + (D/H)^^2 )

Where:

  • D is the separation between the centerlines of the traces, and
  • H is the effective height above the nearest reference plane.

[This is the near-end crosstalk coefficient, for fast-edged signals. Far-end crosstalk is less, see discussion starting p. 204 in my book. Also, if two-times the propagation delay of the parallel traces is less than the risetime of the signal, then the crosstalk reduces further, proportional to the ratio of those two quantities.]

Let's do an example. Say the traces are 0.005" above the reference plane.

Now say we separate them by 0.500" (one-half inch). The D/H ratio is 100:1.

When we use (D/H) squared in the formula above, we get a cross-coupling ratio of 1 part in 10,000, independent of signal rise time (it works all the way into the GHz range). This is 80 dB isolation. Just moving 1/2 inch apart, over a solid plane, is an extremely effective measure.

Keep in mind that the total crosstalk onto a victim line is the sum of crosstalk from each of several nearby aggressors, and the crosstalk waveforms can aggregate appreciably. If you are planning a number of parallel traces (such as 128-bit high-speed digital bus) then you will want to take that into account.

One classic mechanism for reducing parallel-line crosstalk further is to encapsulate the parallel signals between two plane layers (with the same voltage on both planes) and then to install a row of vias between the parallel signals, separating them. I'd put the vias on about 1/4" centers, and then measure the crosstalk on a sample bare board. If you do this experiment, be sure the measure the crosstalk with the vias installed, and then DRILL THEM OUT and do the experiment again, so you know precisely what was the beneficial effect of the vias. Also, when measuring crosstalk, always terminate the source and destination ends of each line with impedances representing what you will really have in the real system. Usually this means shorting the source end of the victim trace to ground (as if it were being driven by a low- impedance source). Use a pulse generator to drive the aggressive trace, matching the pulse rise-time, peak voltage, and peak current of the real source.

Best regards,
Dr. Howard Johnson