ABOVE BOARD
Fall 1997

PARALLELISM and CROSSTALK II

In normal applications, since the materials are homogeneous, it works out that the inductive and capacitive coupling effects are nearly equal. Thus, they tend to cancel forward crosstalk and the immediate back-ward cross-talk reflection tends to cancel the shorter forward crosstalk pulse. Thus, forward crosstalk is normally not an issue unless we are talking about very long lines (in which case the effects are very hard to predict). Backward crosstalk may or may not be a problem depending on the impedance of the gates on the coupled line, the termination techniques used and the specific circuit characteristics of the signals on the coupled line. If the designer has made good decisions on the layout of the board, it is possible that backward cross-talk will not be a problem even if design rules regarding parallel lengths are violated.

But signal lines coupling into clock lines (a potential for noise problems on clock lines) is of more concern. Clock lines tend to be more power-fully driven and better terminated than other lines, so there is some margin of safety.

Some designers believe a guard band around traces will help to "shield out" crosstalk problems. Note that a guard band increases the separation between parallel traces by at least twice. Since coupling is an extremely linear function of spacing in microstrip structures and approximately a square function of spacing in stripline structures, the separation alone accounts for a two to fourfold reduction in crosstalk! Thus, the effects of a guard band are accounted for in separation distance, rather than electrical shielding, and the space might as well be used for another signal which the designers know from their knowledge of the circuit cannot cause a crosstalk problem. (This is not to say that a guard band should not be used).

In summary, forward crosstalk is normally not an issue. Backward crosstalk may or may not be an issue. To minimize the effects of backward cross-talk on a particular line consider the following:

- Limit the length of the parallel coupling

- Separate the lines

- Make sure that you know the real noise margin of the components you are working with (and that you are not using arbitrary "rule of thumb" guidelines that may not be accurate).

- Place lines with higher noise margins adjacent to or between the "driven" lines of interest.

- Pay attention to line termination procedures at both the source and load ends (if the source termination equals the intrinsic impedance of the line, there will be no backwards crosstalk reflection).

It is easy to focus on crosstalk because CAD systems can provide "reports" based on arbitrary formulas related to parallel trace lengths. Experience shows that the negative effects caused by stubs, impedance mismatches, reflections at stubs, antenna effects and voltage transients caused by fast rise time switching through inductive trace lengths are all far more significant and cause far more problems than does crosstalk. Often these factors are overlooked due to a lack of understanding.

UltraCad Design in Bellevue Wash., specialize in designing high speed boards and can be reached at (206)450 9708.