Crosstalk Calculator

How much of an edge on one trace appears on the one beside it: near-end and far-end crosstalk as a fraction and in decibels, the length beyond which near-end crosstalk stops growing, and the two rules that fall out of the algebra — that moving the plane closer beats moving the traces apart, and that far-end crosstalk cancels in a stripline and does not in a microstrip.

Last updated 24 Sep 2026

Near-end and far-end crosstalk, and the two rules behind them

Geometry + edge rate → coupled voltage

Which geometry?The distinction decides whether far-end crosstalk exists at all. In a uniform dielectric the even and odd modes travel at exactly the same speed, the inductive and capacitive coupling cancel at the far end, and far-end crosstalk from this mechanism is zero. A microstrip has part of its field in air, the odd mode has more of it there than the even mode does, the two speeds differ, and the cancellation is incomplete.

Trace width (mm)Used for the modal permittivity split, which needs the width-to-height ratio. The single-line impedance is a separate input below, because that is what the microstrip and stripline calculators on this site are for.

Edge-to-edge gap between the two traces (mm)The gap, not the centre-to-centre pitch. Everything on this page depends on it through the ratio of gap to dielectric height.

Dielectric height to the reference plane (mm)For a microstrip, the height above the plane. For a stripline, the plane-to-plane thickness. This is the other half of the ratio that decides the coupling, and it is the input that makes the design rule: crosstalk depends on the gap measured in dielectric heights, so halving the height is worth as much as doubling the gap — and it costs nothing in board area.

Relative permittivity εᵣOnly used for the modal split that produces far-end crosstalk, and it has no effect at all in the homogeneous case where that split is zero.

Where does the coupling coefficient come from?The geometric estimate uses the same IPC-2141 edge-coupled fits that the differential pair impedance calculator on this site uses, so the two pages agree. If you have a field solver, or you have already run that calculator, the modal impedances are the better input — they carry the copper thickness and the exact stack-up, which the fit does not.

Even-mode impedance (Ω)One trace’s impedance when both are driven the same way. Larger than the single-line impedance.

Odd-mode impedance (Ω)One trace’s impedance when they are driven oppositely — half the differential impedance. Smaller than the single-line impedance. The single-line value is the geometric mean of the two.

Single-line characteristic impedance (Ω)From your stack-up. Used only for the per-unit-length inductance and capacitance reported below; the crosstalk coefficients themselves do not depend on it.

Propagation delay (ps/mm)From the propagation delay calculator on this site, or from √ε_eff ÷ c. About 6 ps/mm for an FR-4 microstrip and about 7 ps/mm for an FR-4 stripline — 152 and 178 ps per inch. It sets the coupled length’s electrical length and therefore the saturation point.

Coupled length — the parallel run (mm)The length over which the two traces run parallel and close. Only the part that is actually coupled counts; a pair that diverges has stopped coupling.

Rise time of the aggressor, 10 to 90% (ps)The edge rate, not the clock period. This is the input that has changed most over the last twenty years and it is why crosstalk problems appear on boards that used to work: the same layout with a faster driver has the same near-end crosstalk but far more far-end crosstalk, because far-end crosstalk is inversely proportional to the rise time.

Aggressor step amplitude ON THE LINE (V)The size of the edge travelling down the aggressor, which is not the same as the driver’s open-circuit swing. A source-terminated driver launches half its swing onto the line, so a 3.3 V driver with a series resistor puts 1.65 V on the trace. Getting this wrong is a clean factor of two.

Victim’s near-end terminationThe backward wave arrives at the victim’s near end and does whatever that end does with it. A matched termination absorbs it; an open reflects it in phase and doubles the voltage there; a hard low-impedance source shorts it out, which is why a driver at the near end sees very little and the receiver at the same end sees twice as much.

Victim’s far-end terminationSame argument at the other end, for the forward wave. Most real high-speed receivers are closer to open than to matched, which is the pessimistic case and often the right one to design to.

The two lines, with the coupling drawn as what it is: a mutual capacitance between them, and a pair of coupled inductors carrying the mutual inductance — both computed from your own geometry. The aggressor's driver is at the left, so the BACKWARD wave appears at the victim's left-hand end and the FORWARD wave at its right-hand end, which is the end most receivers are at. That is the pairing people reverse. The mutual capacitance turns amber above 5% near-end crosstalk and red above 10%.

7.74%Example

two 0.35 mm microstrips 0.25 mm apart over a 0.2 mm dielectric of εᵣ 4.3, running parallel for 50 mm, with a 1.0 V step of 200 ps rise time on the aggressor and both victim ends terminated

Two coefficients, and what each one depends on

k = (Z_even − Z_odd) ÷ (Z_even + Z_odd) ≈ ½(L_m/L + C_m/C)
V_NEXT = (k ÷ 2) · V_line · min(1, 2T_D ÷ t_r)
V_FEXT = (V_line ÷ 2) · min(1, |T_even − T_odd| ÷ t_r), T_even − T_odd = T_D(L_m/L − C_m/C)
saturation length = t_r ÷ (2 · t_pd)

V line: the amplitude of the step travelling on the aggressor, not the driver’s open-circuit swing. A source-terminated driver launches half its swing
the sum vs the difference: near-end crosstalk depends on the SUM of the inductive and capacitive coupling and far-end crosstalk on their DIFFERENCE. That difference is zero in a uniform dielectric, which is why a stripline has no far-end crosstalk
min(1, …): near-end crosstalk saturates once the round trip of the coupled section exceeds the rise time; after that a longer run makes the pulse last longer rather than grow
g = s/h: coupling depends on the gap measured in dielectric heights. Halving the height does as much as doubling the gap and costs no board area

Worked example

two 0.35 mm microstrips 0.25 mm apart over a 0.2 mm dielectric of εᵣ 4.3, running parallel for 50 mm, with a 1.0 V step of 200 ps rise time on the aggressor and both victim ends terminated
The gap is 1.250 dielectric heights. The IPC-2141 edge-coupled fit gives a coupling factor of 0.85543, so the coupling coefficient k = (1 − f²) ÷ (1 + f²) = 0.1549 and the near-end coefficient is half of that, 0.0774
The coupled section is 300 ps one way, so its round trip is 600 ps against a 200 ps edge — saturated. Near-end crosstalk is therefore the full 0.0774 × 1.0 V = 77.45 mV, which is 7.74% or -22.22 dB below the aggressor, and it lasts 600 ps
Far-end crosstalk needs the modal split. Kirschning and Jansen give an even-mode effective permittivity of 3.4549 and an odd-mode one of 2.9714 against 3.2388 for the line on its own, so the two modes arrive 22.50 ps apart over this length
That is 11.2% of the rise time, so the far-end pulse is half the step times that: 56.25 mV, 5.62% or -25.00 dB down, and of opposite polarity to the driving edge. On a stripline it would be exactly zero, because the two modes would arrive together
The two design numbers. Near-end crosstalk stops growing beyond 16.67 mm of parallel run, so shortening a 50 mm run will not help until it is below that. And doubling the gap to 0.50 mm takes the near-end crosstalk to 22.25 mV, while leaving the gap alone and bringing the plane to 0.10 mm instead reaches 22.25 mV — the same gap in half the dielectric height, for no board area at all

Near-end and far-end crosstalk behave differently in every respect

The row that decides stack-ups is the second one. Far-end crosstalk vanishing in a homogeneous dielectric is a much stronger argument for putting a critical bus on an inner layer than anything about shielding, and it is exact rather than approximate — as long as the dielectric really is uniform, which glass-weave texture and resin-rich regions spoil to a degree.
	Near-end (backward)	Far-end (forward)
Depends on	the SUM of the inductive and capacitive coupling	their DIFFERENCE
In a uniform dielectric	unchanged — it is there in a stripline, a coax and a twisted pair	zero, because the two modes travel at the same speed
With coupled length	grows until the round trip reaches the rise time, then saturates	grows linearly, with no saturation until the modal delay difference reaches the rise time
With rise time	independent of it once saturated	inversely proportional — a faster driver makes it worse in direct proportion
Polarity	same as the aggressor’s edge	opposite, for a microstrip where the inductive coupling ratio exceeds the capacitive one
Duration	twice the one-way delay of the coupled section	about one rise time
Which end it appears at	the end the aggressor’s driver is at	the end the aggressor’s receiver is at
Who usually cares	a receiver sharing the near end with the aggressor’s driver	the receiver at the far end, which is most receivers

One trace switches and the one beside it moves

Two conductors running side by side over a reference plane are coupled by two mechanisms at once. The electric field between them is a mutual capacitance, which injects a current into the victim proportional to the aggressor’s dV/dt. The magnetic field around the aggressor links the victim’s loop, which induces a voltage proportional to dI/dt. The two produce waves travelling in opposite directions along the victim, and that is the whole reason near-end and far-end crosstalk are different quantities rather than the same quantity measured at two places.

The backward wave adds and the forward wave subtracts. Towards the aggressor’s driver, the capacitive and inductive contributions have the same sign and add: near-end crosstalk depends on the SUM of the two coupling ratios and is (k/2) times the aggressor’s step, where k is the familiar coupling coefficient (Z_even − Z_odd) ÷ (Z_even + Z_odd). Towards the aggressor’s receiver they have opposite signs and subtract: far-end crosstalk depends on their DIFFERENCE. In a uniform dielectric that difference is exactly zero, because the inductive and capacitive coupling ratios are then equal — which is the same statement as saying the even and odd modes travel at the same speed. A stripline, a coax and a twisted pair in a uniform jacket have no far-end crosstalk from this mechanism at all. A microstrip does, because part of its field is in air, the odd mode has more of its field there than the even mode does, and the cancellation is incomplete.

Near-end crosstalk saturates and far-end does not. The backward wave from a point on the line takes the round trip to get back to the near end, so contributions from the whole coupled length arrive spread over twice its one-way delay. Once that spread exceeds the rise time, the near-end pulse has reached its full height and a longer parallel run only makes it last longer. The saturation length is t_r ÷ 2t_pd, which for a 200 ps edge on FR-4 microstrip is about 17 mm — so almost every parallel run on a real board is saturated, and shortening it is not the fix people expect it to be. Far-end crosstalk, by contrast, accumulates along the whole length and is inversely proportional to the rise time, which is why a layout that worked with a 2 ns driver fails with a 200 ps one.

The design rule that falls out. Coupling depends on the gap measured in dielectric heights, not in millimetres. That single fact carries the whole of the practical advice. Doubling the gap helps; halving the height to the reference plane helps by the same amount, costs no board area at all, and improves the impedance control and the return path at the same time. On a modern stack-up the dielectric under an outer layer is often 0.1 mm or less precisely for this reason. The familiar three-widths rule is this rule in disguise, and it is only right for the stack-up it was derived on: the number that matters is the ratio of gap to height, and a 3 W rule on a thick dielectric is much weaker than the same rule on a thin one. The chart plots both trades side by side so you can see which buys more on your own stack-up.

What this page assumes. It is the weak-coupling model: first-order in the coupling, both lines close to matched, two conductors only. It is checked in this batch against a finite-difference time-domain solve of the two-conductor telegrapher equations with full inductance and capacitance matrices — near-end agrees to better than 1% and far-end to better than 2% at the coupling levels here — and both degrade as the coupling rises. It does not model a third aggressor, and crosstalk from several neighbours adds; it does not model reflections beyond the single termination factor at each end; and it does not model loss, which reduces far-end crosstalk on a long lossy line. The single-line geometry it builds on is owned by the microstrip impedance calculator, the stripline calculator, the differential pair calculator — whose coupling coefficient is exactly the k used here — and the propagation delay calculator.

Frequently asked questions

Is the 3 W rule right?

It is right for the stack-up it was derived on and misleading everywhere else. Coupling depends on the gap measured in dielectric heights, so the same three-trace-width spacing gives quite different crosstalk over a 0.1 mm dielectric and over a 0.5 mm one. Work in gap-over-height and the rule becomes a number you can actually check.

Why does moving the ground plane closer help more than moving the traces apart?

Because the coupling depends on the gap divided by the height, so both act on the same ratio — but halving the height costs no board area, doubles nothing, and improves the return path and the impedance control at the same time, while doubling the gap costs you the space. The chart on this page plots the two trades against each other for your own geometry.

Why is there no far-end crosstalk in a stripline?

Because far-end crosstalk depends on the difference between the inductive and capacitive coupling ratios, and in a uniform dielectric those two are equal. Equivalently: the even and odd modes travel at exactly the same speed, so the two halves of the disturbance arrive together and cancel. A microstrip has part of its field in air, the odd mode has more of it there, the speeds differ and the cancellation is incomplete. In practice a stripline’s far-end crosstalk is small rather than exactly zero, because glass weave and resin-rich regions make the dielectric less than perfectly uniform.

Making the parallel run shorter did not help. Why?

Because near-end crosstalk saturates. Once the round trip of the coupled section exceeds the rise time, the near-end pulse is at full height and a shorter run only makes it briefer. The saturation length is on this page — for a fast edge on FR-4 it is millimetres, so most real runs are already saturated. Far-end crosstalk does keep falling with length, so if the receiver is at the far end, shortening the run does help.

Which crosstalk should I worry about?

Usually the one at the end where the receiver is, which is the far end, and on an outer layer that is the one that grows with every faster driver. Near-end matters when a receiver shares the near end with the aggressor’s driver — a bidirectional bus, or a connector where several signals turn round together.

Do I use the driver’s swing or the voltage on the line?

The voltage on the line, and the two differ by a factor of two for a source-terminated driver: a series resistor equal to the line impedance launches half the swing and the full amplitude only appears after the reflection at the open far end. Using the driver’s swing when the line carries half of it doubles every number on this page.

Related calculators

References

Johnson HW, Graham M. High-Speed Digital Design: A Handbook of Black Magic. Prentice Hall, 1993. Chapter 5 on transmission lines and chapter 6 on crosstalk: the backward and forward coupling coefficients, the saturation of near-end crosstalk at a coupled length of one half the rise-time distance, the gap-over-height dependence behind the familiar spacing rules, and the cancellation of far-end crosstalk in a homogeneous dielectric. The model on this page is the weak-coupling approximation developed there and it is stated as such.
IPC. IPC-2141A, Design guide for high-speed controlled impedance circuit boards. The edge-coupled fits used for the geometric estimate — Z_diff = 2Z₀(1 − 0.48 e^(−0.96 s/h)) for microstrip and Z_diff = 2Z₀(1 − 0.347 e^(−2.9 s/b)) for stripline — which are the same fits this site’s differential pair calculator uses, so the two pages agree by construction. Stated for roughly 0.3 ≤ s/h ≤ 4; the page warns when you leave that range.
Kirschning M, Jansen RH. Accurate wide-range design equations for the frequency-dependent characteristic of parallel coupled microstrip lines. IEEE Transactions on Microwave Theory and Techniques, 1984. The static even- and odd-mode effective permittivities used here for the modal delay difference that produces far-end crosstalk. Implemented with Hammerstad and Jensen’s single-line effective permittivity so that the two modes converge to it exactly as the gap widens — checked in this page’s build script, where the far-end term goes to zero to within 10⁻⁶ at large gaps for every width and permittivity tried.
Hammerstad E, Jensen Ø. Accurate models for microstrip computer-aided design. IEEE MTT-S International Microwave Symposium Digest, 1980. The single-line effective permittivity, ε_eff = (εᵣ+1)/2 + (εᵣ−1)/2·(1+10/u)^(−a(u)b(εᵣ)), with a(u) = 1 + (1/49)ln[(u⁴+(u/52)²)/(u⁴+0.432)] + (1/18.7)ln[1+(u/18.1)³] and b(εᵣ) = 0.564[(εᵣ−0.9)/(εᵣ+3)]^0.053, quoted as accurate to better than 0.2%. Checked here against the Hammerstad 1975 form this site’s microstrip page uses: the two agree to better than 0.6% over w/h from 0.3 to 6.
Paul CR. Analysis of Multiconductor Transmission Lines, 2nd ed. Wiley, 2008. The multiconductor telegrapher equations with full L and C matrices, which is what this batch’s independent check solves in the time domain to verify the closed forms above.