Crosstalk Coupling: Single-Ended vs. Differential

The paper begins with four propositions. 1), the effects of crosstalk coupling decrease with trace separation. 2), crosstalk coupled to a differential pair has meaning only for the differential component of the crosstalk on the differential pair, not the common mode component of the crosstalk. A differential component only exists because the outside trace is (perhaps only slightly) further away from the source than is the inside trace. 3), crosstalk caused by a differential pair would be equal and opposite, and therefore cancel, on a victim trace were it not for the (perhaps only slight) separation of the differential traces themselves. Since one trace (of the pair) is (slightly) closer to the victim trace than is the other trace of the differential pair, that trace will couple slightly more strongly and there will be a small differential coupling to a victim trace. 4), differential pair coupling to another differential pair would combine these last two effects and should be quite small. The relationships between these four propositions are quantified and then tested against four PCB structures using the Mentor Graphics Hyperlynx simulation tool. The four structures are: microstrip, deeply embedded microstrip (with a thick dielectric layer above the trace), balanced stripline, and asymmetric stripline. The results of the simulations are as predicted for microstrip configurations and for stripline configurations when the traces are close together relative to the second reference plane. But singleended coupling drops off more quickly than predicted with increased spacing for stripline environments. Differential coupling, however, does not drop off in the same manner for stripline configurations. INTRODUCTION AND BACKGROUND Crosstalk is often a serious consideration in PCB design. It is reasonably well understood that the primary strategies available to board designers for reducing crosstalk between traces are (a) rout sensitive traces close to their underlying reference planes and (b) spread the traces apart1. How "close" and how "far" are policy variables, the responsibilities for which are usually reserved for the circuit or system design engineer. A benefit often ascribed to the use of differential signals and traces is that they are less susceptible to radiated noise (and therefore crosstalk), and that they radiate less noise (and therefore cause less crosstalk) than ordinary single-ended traces2. If this benefit is true (and it is) then the worst PCB crosstalk environment (all other things equal) would be where a single-ended aggressor trace couples into a singleended victim trace, and the best environment would be where a differential aggressor pair couple into a differential victim pair. The case of a differential aggressor pair coupling into a single-ended victim, or of a single ended aggressor trace coupling into a differential victim pair would represent environments somewhere between the other two extremes. Typically, after board stackup decisions are made, the only variable left to control crosstalk is the spacing between traces. Thus, system engineers may give board designers layout rules for spacing. A typical rule may be spacing between traces of 5*H (H being the height above the plane, for example). These types of rules may be derived through simulations, or they may simply be "carry-overs" from previous designs, previous engineers, or even previous companies! When (and if) these layout rules are supplied, they are usually done so without regard to the types of signals (single-ended or differential) being routed. But if the degree of crosstalk is a function of the types of signals being routed, then the layout rules should (presumably) reflect this. The purpose of this paper is to look at the various signal and trace environments and compare them from a crosstalk standpoint. Given a better quantitative understanding of the relative magnitude of the crosstalk noise signals for these different environments, it may be possible to adjust layout rules more intelligently for more efficient use of board real estate.