Crosstalk Calculator — near-end and far-end coupling between PCB traces

The crosstalk calculator above estimates how much of a switching signal leaks onto the trace running beside it. It gives you two separate answers, because crosstalk is two separate phenomena with different physics, different waveforms and different fixes. Near-end crosstalk appears at the end of the victim trace closest to the aggressor's driver and lasts for twice the flight time of the coupled section. Far-end crosstalk appears at the opposite end as a single narrow spike whose width is the edge rate. Confusing the two is the most common reason a crosstalk problem resists being fixed.

Arb Digital publishes free engineering calculators that name their assumptions instead of burying them. This page uses the published first-order coupling rule of thumb, prints the intermediate quantities so you can see where the number came from, and states clearly that a two-dimensional field solver run against your actual stack-up is the answer you would sign off a design with. The tool is for deciding whether a spacing is roughly sane before you commit to a routing plan, and for understanding which knob moves which number.

What This Crosstalk Calculator Does

Two traces running parallel over a reference plane share mutual capacitance and mutual inductance. When the aggressor's voltage changes, the mutual capacitance injects a current into the victim proportional to the rate of change of voltage, and the mutual inductance injects a voltage proportional to the rate of change of current. Those two contributions travel in opposite directions along the victim. Backward, towards the near end, they add. Forward, towards the far end, they subtract.

That subtraction is why microstrip and stripline behave so differently. In a stripline the dielectric is homogeneous, the two contributions match, and they cancel almost exactly in the forward direction. A microstrip has laminate below and air above, the cancellation is incomplete, and far-end crosstalk survives — often larger than the near-end figure on a long, fast bus.

The hero number is the near-end crosstalk voltage. The grid gives the spacing ratio, the coupled length at which near-end coupling stops growing, the isolation in decibels, and the far-end peak.

How to Use It

1. Enter the real edge rate. A 100 MHz clock from a driver with a 150 ps edge behaves like a 2 GHz signal for crosstalk purposes. The clock frequency is almost irrelevant; the edge is everything.
2. Measure spacing edge to edge, and height to the nearest plane. The ratio s/h is the term that dominates the answer. If your stack-up has the plane two layers down because a layer was reassigned, use that larger height — it makes coupling far worse.
3. Count only the genuinely parallel length. Sections where the traces diverge contribute almost nothing. Long parallel runs in a connector breakout or under a BGA usually contribute most of the total.
4. Set K_max for your geometry. Leave it near 0.55 for a microstrip. Reduce it for a stripline, where the second reference plane pulls field lines away from the neighbour and cuts coupling substantially.
5. Put a real K_fe in. Take the far-end coefficient from a field solver or a measurement on your own stack-up; the default is a placeholder.

The Formula: How Crosstalk Is Estimated

The near-end coupling coefficient uses the widely published microstrip approximation K_b = K_max ÷ (1 + (s/h)²), where s is the edge-to-edge spacing and h is the height above the reference plane. It is an empirical fit, not a derivation, and it is the reason the tool exposes K_max as an input rather than hiding a constant.

Backward crosstalk does not grow without limit. Energy coupled at a point travels back towards the near end while the aggressor's edge continues forward, so a near-end observer sees a plateau lasting twice the one-way delay of the coupled section. The coupling reaches its full value once the coupled length exceeds the saturation length L_sat = t_r × v ÷ 2, where v is the propagation velocity. Below that, the coupling scales linearly with length, so K = K_b × L ÷ L_sat.

Propagation velocity comes from the effective permittivity: v = c ÷ √ε_eff, using the exact NIST CODATA value for the speed of light in vacuum of 299,792,458 m/s. Far-end crosstalk is computed as V_fext = V × K_fe × L ÷ t_r, which captures the two behaviours that matter: it grows with coupled length and it grows as the edge gets faster.

Work the defaults through by hand. Spacing 0.5 mm over a 0.2 mm height gives s/h = 2.5, so 1 + 2.5² = 7.25 and K_b = 0.55 ÷ 7.25 = 0.0759. With ε_eff = 3.2 the velocity is 299,792,458 ÷ 1.7889 = 1.6759 × 10⁸ m/s, so a 200 ps edge gives L_sat = 200 × 10⁻¹² × 1.6759 × 10⁸ ÷ 2 = 16.76 mm. The 50 mm coupled length is well past that, so the coupling is saturated and the near-end voltage is 0.0759 × 3.3 = 0.250 V, an isolation of 20 log₁₀(0.0759) = −22.4 dB. The far-end peak is 3.3 × 2 ps/cm × 5 cm ÷ 200 ps = 0.165 V.

Why Saturation Changes How You Read the Answer

The saturation length is the single most useful number on the page and the one most often skipped. Below it, halving the coupled length halves the near-end crosstalk. Above it, halving the coupled length does nothing at all to the amplitude — it only shortens how long the plateau lasts. Teams routinely spend a respin shortening a parallel run that was already three times the saturation length, and are baffled when the measured coupling does not move.

Far-end crosstalk behaves in the opposite way. It has no saturation. It keeps growing linearly with coupled length for as long as the traces stay parallel, which is why on a long backplane or a wide DDR bus the far-end spike is usually the failure and the near-end plateau is a curiosity. If your problem is at the far end, shortening the run genuinely helps; if it is at the near end and you are past saturation, only spacing, a lower height or a stripline layer will help.

Where the Rule of Thumb Stops Being Trustworthy

The 1/(1 + (s/h)²) form is a fit to weakly coupled microstrip pairs over a solid plane, and it degrades outside that. Below roughly half a dielectric height of spacing the real coupling is stronger than the fit predicts, and the pair stops behaving as two independent lines at all — it is a differential structure whose even and odd modes have different impedances and velocities. At very wide spacing the formula predicts a small number that is swamped in practice by coupling through shared return paths, connectors and package pins.

The model also assumes an unbroken reference plane. A split or a slot under the coupled section forces return current to detour, raises the effective height dramatically and can increase coupling by an order of magnitude. The Missouri University of Science and Technology Electromagnetic Compatibility Laboratory publishes research on exactly these signal-integrity and return-path mechanisms, and it is worth reading before assuming a plane is solid because the layer stack says so.

What Actually Reduces Crosstalk, in Order of Effectiveness

Bringing the trace closer to its reference plane is the strongest lever available, because it appears in the denominator as h and it shrinks both the coupling and the loop area at once. A 0.1 mm dielectric with the same 0.5 mm spacing gives s/h = 5 and a coupling coefficient about four times lower. It costs nothing in board area; it costs a stack-up decision.

Spacing is next, and it is the one most designers reach for. The familiar 3×h design rule exists because it puts s/h at 3 and drops the coefficient to about a tenth of K_max, which is usually enough. Going beyond 4×h buys very little for the area it consumes.

Slowing the edge helps the far end proportionally and helps the near end only until you drop below saturation. Moving a critical net to a stripline layer removes most of the far-end coupling outright. Guard traces disappoint most often: an ungrounded one does nothing, and one grounded only at its ends can resonate. Stitch it to the plane with vias at intervals well under a quarter wavelength, or spend the area on spacing instead.

The underlying field behaviour is standard electromagnetics. MIT OpenCourseWare's 6.013 Electromagnetics and Applications course materials cover the guided-wave theory that the coupled-line equations are built on, if you want the derivation rather than the rule of thumb.

How This Sits Next to the Other Signal Tools

This page answers "how much noise lands on the neighbour". It does not give the characteristic impedance of either line — that is the job of the cable impedance calculator, which handles the single-line geometry. Once you have an impedance, the RLC impedance calculator and the reactance calculator cover the terminations, while the attenuation calculator covers loss along the line rather than coupling across to a neighbour.

Common Mistakes to Avoid

• Using clock frequency instead of edge rate — a slow bus driven by a fast part couples like a fast bus, and the frequency on the schematic tells you nothing about the edge.
• Measuring height to the wrong plane — if the nearest solid reference is two dielectrics away because a layer was reassigned to signals, h doubles and coupling rises sharply.
• Shortening a run that is already saturated — past the saturation length, near-end amplitude does not care how long the coupled section is.
• Assuming far-end crosstalk is small because a stripline reference said so — that cancellation only holds for a homogeneous dielectric, and any microstrip breaks it.
• Adding an unstitched guard trace — a floating or end-only-grounded guard provides no benefit and can resonate, while the same board area spent on spacing would have worked.

Related Free Tools From Arb Digital

Start with the cable impedance calculator for the single-line geometry, then bring the result here for coupling. The attenuation calculator covers loss along the line, the decibel calculator converts ratios into isolation figures, and the wavelength calculator sets your via-stitching intervals. For the circuit end of the problem, the RLC impedance calculator, the reactance calculator and the voltage divider calculator handle terminations and level shifting. Everything Arb Digital publishes is on the free online tools hub.

Frequently Asked Questions

This tool is provided for educational and preliminary design use. It applies a published first-order approximation to weakly coupled traces over a solid reference plane, and does not model plane splits, multiple aggressors, connectors, packages, dielectric loss or mode conversion. Verify any design with a field solver and measurement.