PCB crosstalk: near and far-end coupling, Johnson's rule, the 3W rule, and a worked example
Two traces that run side by side share a field, and a fast edge on one appears as a pulse on the other. How much depends almost entirely on one ratio, the centre-to-centre spacing divided by the height over the ground plane, and on whether the run is long enough for the coupling to saturate. This calculator takes the traces' width, gap and height, the coupled length and the aggressor's edge, and gives the coupling coefficient, the near-end pulse in volts and decibels, a rough far-end figure for microstrips, the spacing that keeps the crosstalk under a target and what the 3W rule would give, with the pair drawn from above, edge and pulse, and in cross-section with the field that leaks across.
How to use the crosstalk calculator
- Type the trace width, the gap between the traces and their height over the plane, and pick microstrip (outer layer, one plane) or stripline (buried) and the substrate in the menu row. The coupling coefficient k appears at once.
- Add the coupled length, the aggressor's rise time (1n, 500p…) and its swing for the near-end pulse in volts and the rough far-end figure.
- Type an allowed crosstalk in percent for the gap that keeps under it; the Gap table tab lists the coupling for gaps from half a width to ten, each with a Use button.
What crosstalk is
A trace over a plane carries its signal in the field between the copper and the plane, and that field does not stop at the trace's edge: some of it fringes sideways. A neighbouring trace inside that fringe sees part of the aggressor's voltage through the capacitance between them and part of its current through the mutual inductance. Both couplings act only while the aggressor is changing, so what appears on the victim is a pulse at each edge, not a copy of the signal. The pulse that travels back to the victim's driver end is the near-end (or backward) crosstalk; the one that travels along with the aggressor's edge to the far end is the far-end (forward) crosstalk. Near-end crosstalk has the aggressor's polarity, lasts twice the coupled run's delay and has a maximum set by the geometry; far-end crosstalk on a microstrip is a spike of the opposite polarity about one rise time long that grows with the run.
Johnson's rule
k = 1 / (1 + (D/h)²)
Howard Johnson's rule of thumb from High-Speed Digital Design: the saturated near-end crosstalk is this share of the aggressor's swing, with D the centre-to-centre spacing (gap plus width) and h the height over the plane. It is a fit to field-solver results for a terminated victim and is usually within 20 %. The lesson is in its shape. Twice the height apart (D/h = 2) gives 20 %; four heights, 6 %; ten heights, 1 %. The spacing only matters relative to the height, so the fastest way to quieten a board is to bring the plane closer, not to spread the traces.
Saturation and the far end
The near-end pulse is the sum of the coupling along the whole run arriving back at the near end over twice the run's delay. If the edge is slower than that, the contributions overlap only partly and the pulse is k × 2Td/tr; once twice the delay reaches the rise time the pulse is k and stays there, growing only in length. On FR-4 a 1 ns edge saturates at about 85 mm of coupled microstrip (72 mm of stripline), a 100 ps edge at under 9 mm. The far end is different: on a microstrip the odd and even modes travel at slightly different speeds (part of the field is in air), so a forward pulse builds along the run, roughly k × Td/tr of the swing until it too saturates. In a stripline both modes travel at the same speed and the forward contributions cancel; a buried pair has almost no far-end crosstalk, which is one reason fast buses go on inner layers.
What reduces it
In order of effect: bring the plane closer (k falls with the square of D/h), widen the spacing (the 3W rule, centres three widths apart, is a habit that works because most traces sit one to two widths above their plane), shorten the coupled run, slow the edge with a series resistor at the driver, and move the pair to a stripline layer. A guard trace between them helps only if it is stitched to ground with vias every fraction of a rise time's length; floating, it can make things worse. And keep the plane whole: a slot under the traces forces both return currents through the same gap and couples them through the shared return, which no spacing fixes.
Your pair, step by step
- Geometry: D = 0.20 mm + 0.20 mm = 0.40 mm, h = 0.20 mm, so D/h = 2.00 and k = 1 / (1 + 2.00²) = 20%.
- Saturation: 100.0 mm delays 590 ps; twice that against 1 ns → saturated. Saturates past 84.8 mm.
- Near end: 20% × 3.3 V = 660 mV (-14 dB) for 1.18 ns. Far end ≈ 389 mV.
- For 5%: D = 0.20 mm √(1/0.05 − 1) = 0.87 mm, a gap of 0.67 mm; the 3W rule (gap 0.40 mm) gives 10%.
Worked example: 0.2 mm traces 0.2 mm apart on a 0.2 mm prepreg
Two 0.2 mm traces with a 0.2 mm gap on an outer layer 0.2 mm above the plane: D = 0.4 mm, D/h = 2, so k = 1 / (1 + 4) = 20%. Over 100 mm on FR-4 the run delays 590 ps, twice that exceeds a 1 ns edge (the run saturates past 84.8 mm), so the near end is saturated: a 3.3 V swing puts 660 mV on the victim for 1.18 ns, enough to false-trigger a clock input; at 50 mm the run would be unsaturated and the pulse 389 mV. With the 3W rule's 0.4 mm gap, D/h = 3 and k falls to 10%, 330 mV; for 5 % the centres must be h √19 = 0.87 mm apart, a gap of 0.67 mm. Moving the plane to 0.1 mm instead, with the original 0.2 mm gap, gives D/h = 4 and 5.88% without touching the routing.
Questions
How much crosstalk is acceptable?
A common budget is 5 % of the swing for data lines, and 2–3 % for clocks, resets, strobes and analogue inputs, which are the lines a glitch can hurt. Several aggressors add up, so a bus beside a clock needs the stricter figure per neighbour.
Does the 3W rule always work?
It works when the traces are one to two widths above their plane, which covers most multilayer stacks: centres three widths apart is then D/h of 1.5 to 3 and k of 10 to 30 %, a 30 % reduction or more over touching traces. On a two-layer board with 1.6 mm to the plane it does almost nothing, as this calculator shows.
Why is the far-end figure only rough?
Far-end crosstalk depends on the difference between the odd and even-mode speeds, which depends on how much of the field is in air; it needs the coupled-line parameters a field solver gives. The estimate here is the right order for a microstrip and zero for a stripline.
Is crosstalk set by the clock frequency?
No, by the edge rate. A 1 MHz signal with 1 ns edges crosstalks exactly like a 100 MHz one with the same edges, just less often. That is why slowing the edges with a series resistor is such an effective fix.