Figure 1 depicts two 100-Ω differential pairs of PCB (printed-circuit-board) traces. A solid-plane layer appears at the bottom of the figure. An upper solid plane exists somewhere above the figure. This plot shows only a subset of the full stripline cross section.
A pattern of thin, colored lines represents the magnetic field resulting from current in the left-most pair. Technically, the lines prescribe contours of constant 2-D magnetic scalar potential. The magnetic field is most intense where the lines fall closest together and least intense on the right side of the diagram where the lines spread far apart. The field intensity proximate to the conductors appears very intense. Pixelation effects in the figure may create what look like moiré interference patterns in the dense region. The moiré patterns are not real. The field patterns proximate to the conductors form a set of concentric curves near the surface of each tiny conductor.
You can estimate the crosstalk picked up by the victim pair, wires A and B, based on the field patterns in this diagram. Your estimate will not be perfect because the mere presence of wires A and B will distort somewhat the high-frequency magnetic-field pattern, but the general principle remains quite useful. Simply count the number of magnetic-field lines that pass between wires A and B to estimate the crosstalk picked up between them.
There are 96 magnetic-field lines between the two conductors on the left side. You can't see those lines, but I know how many there are because I wrote the code that generated this figure. Three of the lines pass through the flux window drawn between the centerlines of victim traces A and B. In this type of diagram, the near-end crosstalk ratio is simply the ratio of 3 to 96—about 3 That's all there is to estimating crosstalk. Just count how many lines, out of the total, pass between the differential traces of interest.
Experts in electric- and magnetic-field calculations may recognize that my procedure takes into account only magnetic-field effects, ignoring capacitive coupling. That's OK. The procedure works because, in a stripline configuration, inductive- and capacitive-coupling effects are nearly always exactly equal. Compute one, and you have the other.
Now, let's change something. Grab traces A and B and move them both to the left by about one-half a trace width. In response to that action, two new magnetic lines of force fall within the victim's flux window. Crosstalk increases to 5, almost double the previous result. Obviously, the spacing between the aggressor pair and the victim pair greatly affects crosstalk.
Next, put the traces back to their original positions and try something different. This time, leave trace A in its original position while you bring B to the left by one-half a trace width. Notice that the line density near B is quite sparse. After you slide B into its new position, the number of flux lines penetrating the window does not change. Three flux lines still penetrate. Crosstalk hardly changes. In a more accurate plot, you would surely observe some small effect, but I think you get the picture: the spacing between the wires of a differential-stripline pair only mildly affects crosstalk.
In this example, the spacing between wires A and B has little effect because the crosstalk effect is not well-balanced. The aggressive field affects wire A much more strongly than it affects wire B. Differential systems can reject noise only if it affects both wires equally. In a nonsymmetric situation, like this, differential-noise cancellation does not occur.
I hope Figure 1 helps you visualize why differential PCB traces do not significantly reduce crosstalk from other traces. To reduce intertrace crosstalk, you must enforce spacing rules between the aggressor and the victim, much as you would with single-ended signals.
Let me wrap up this article with one more figure, this time showing the magnetic-field pattern associated with broadside-coupled pairs.