A tool for calculating the "crosstalk" that appears between two traces running side by side on a printed circuit board. Adjust the trace spacing, coupled length and signal rise time to see the near-end (NEXT) and far-end (FEXT) crosstalk picked up by the neighbouring trace in real time, and design noise-resistant high-speed routing.
Parameters
Trace spacing s
mm
Edge-to-edge distance between adjacent traces
Trace width w
mm
Parallel coupled length
mm
Length of the section where the two traces run parallel
Signal rise time
ps
Time for the aggressor edge to transition from 0 to 100%
Dielectric (substrate) height h
mm
Thickness of the insulating layer between trace and ground plane
Results
—
Near-end crosstalk NEXT (%)
—
Far-end crosstalk FEXT (%)
—
NEXT level (dB)
—
Saturation length (mm)
—
Coupling ratio Cm/Ct
—
3W-rule verdict
—
Coupled-lines cross-section — crosstalk animation
The top is the aggressor (offending) trace, the bottom is the victim (affected) trace. As the aggressor pulse travels, noise couples across and crosstalk pulses appear at the victim's near end (left) and far end (right).
Backward (near-end) crosstalk coefficient k_b and forward (far-end) crosstalk coefficient k_f. C_m/C_t is the mutual-to-self capacitance ratio, L_m/L_t the mutual-to-self inductance ratio. In microstrip the inductive coupling is slightly stronger, so k_f is negative.
Saturation length L_sat and the propagation speed v on the board. t_r is the signal rise time, c the speed of light and ε_eff the effective dielectric constant (about 3).
Near-end and far-end crosstalk for a coupled length ℓ. Near-end crosstalk grows with length but saturates at L_sat, while far-end crosstalk does not saturate and keeps growing in proportion to length.
What is crosstalk between coupled lines?
🙋
"Crosstalk" means a signal leaks onto the neighbouring trace, right? But the traces aren't even connected — how can it leak?
🎓
Exactly — "leaking without being connected" is the whole point of crosstalk. When two board traces run side by side, the moment a fast-switching signal flows on one trace (the aggressor, the offending side), its electric and magnetic fields reach the neighbouring trace (the victim, the affected side) too. The electric field gives capacitive coupling, the magnetic field gives inductive coupling, and through those two a small noise voltage rides onto the victim. The tighter you pack the traces, the stronger this coupling.
🙋
I see. But the results card on the left splits it into "near-end" and "far-end". It's the same crosstalk — why two kinds?
🎓
Good question. The noise born from the coupling travels both "forward" and "backward" along the victim trace. The part that travels backward toward the aggressor's source is near-end crosstalk (NEXT); the part that travels forward with the signal and reaches the victim's far end is far-end crosstalk (FEXT). They differ not only in direction but in behaviour. NEXT plateaus (saturates) once the coupled length is long enough, but FEXT just keeps growing the longer you make the run.
🙋
Wait, NEXT saturates? When I make the "parallel coupled length" longer on the left, the NEXT number stops changing past a point. Why is that?
🎓
That's the interesting part of NEXT. Backward noise is generated continuously while the aggressor's edge sweeps the coupled section, and it piles up toward the source. But once the coupled length exceeds the "saturation length", new noise just replaces old noise and the amplitude stays constant. That saturation length is set by the signal's rise time — the faster the edge, the shorter it is. So if you want to lower a saturated NEXT, the only way is to widen the trace spacing, not shorten the run.
🙋
Then how do I reduce FEXT? If it doesn't saturate, a long trace must get dangerous fast.
🎓
Right — FEXT bites harder the longer the parallel run. So the first defence is to "keep the parallel section short". The second is not to make the edge faster than necessary, since FEXT grows with edge speed. In microstrip routing the capacitive and inductive coupling don't fully cancel, so FEXT never reaches zero. A classic field fix is to route the trace as stripline (an inner layer): both couplings line up and FEXT almost vanishes.
🙋
I get that spacing matters. I keep hearing about the "3W rule" — what is that?
🎓
The 3W rule is a design rule of thumb: "keep the edge-to-edge spacing to a neighbouring trace at least three trace widths." Spacing the traces three widths apart drops the mutual coupling sharply and is said to suppress roughly 70% or more of the crosstalk. For example, sensitive signals like clocks and resets always get a full 3W, while data lines in tight areas might be allowed 2W — you prioritise. Narrow the spacing in the tool on the left and you'll see the 3W verdict flip to "too close" while NEXT jumps up.
Frequently Asked Questions
Crosstalk is the noise that couples from the aggressor trace onto a victim trace, and it splits into two parts by direction of travel. Near-end crosstalk NEXT travels backward toward the aggressor's source; it builds up as the coupled length increases but saturates at a constant value once the run exceeds the saturation length set by the signal's rise time. Far-end crosstalk FEXT travels forward with the signal toward the victim's far end; it does not saturate, it keeps growing in proportion to the coupled length, and it is larger for faster (shorter rise-time) edges.
The 3W rule is a rule of thumb that keeps the edge-to-edge spacing between adjacent traces at least three trace widths. Crosstalk rises sharply as spacing narrows, so spacing the traces three line-widths apart drops the mutual coupling to a roughly negligible level and suppresses on the order of 70% of the crosstalk. It matters most for noise-sensitive signals such as clocks and resets, and next to differential pairs. When board area is tight, prioritise: give sensitive signals a full 3W and let the rest run at 2W or more.
The defences are simple geometry. First, increase the spacing between traces (the 3W rule). Second, keep the parallel run (coupled length) as short as possible. Third, place a grounded trace or plane between sensitive signals (guard trace). Fourth, route critical pairs on different layers. In addition, do not make the signal edges faster than necessary, and terminate properly to suppress ringing. Vary the spacing and coupled length in this tool to see intuitively how NEXT and FEXT respond.
Near-end crosstalk travels backward, so the noise generated while the aggressor's edge sweeps the coupled section piles up toward the source; once the coupled length exceeds the saturation length, new noise simply replaces old noise and the amplitude becomes constant. Far-end crosstalk, by contrast, travels in the same direction and at the same speed as the signal, so all the noise generated along the coupled section keeps accumulating onto the very same pulse. FEXT therefore grows in proportion to the coupled length and never plateaus. It also matters that in microstrip the capacitive and inductive coupling do not fully cancel, so FEXT does not reach zero.
Real-World Applications
High-speed digital board routing: For high-speed signals with rise times of tens to hundreds of picoseconds — DDR memory, PCI Express, HDMI, SerDes links — crosstalk directly eats into the timing margin and eye opening. Designers secure trace spacing with the 3W rule, manage the length of parallel data buses and place guard traces on both sides of sensitive clocks. Quickly estimating NEXT/FEXT from spacing and coupled length, as this tool does, lets you narrow down a routing strategy before running detailed electromagnetic field simulation.
Connectors, cables and backplanes: Crosstalk is not just an on-board problem. The same physics applies to connector pin layouts, adjacent wires in a ribbon cable and parallel traces on a backplane. In connectors you insert ground pins between signal pins to separate aggressor and victim; in cables a "ground-signal-ground" arrangement that assigns a ground wire per signal suppresses crosstalk.
Differential pairs and high-speed interfaces: In differential signalling such as USB, Ethernet and MIPI, the two wires inside a pair are deliberately coupled strongly, but "pair-to-pair crosstalk" with the neighbouring pair is noise. The standard practice is to make the pair-to-pair spacing much larger than the intra-pair spacing and to keep parallel runs between pairs short. Heavy crosstalk leaves skew and jitter that the common-mode rejection of differential signalling alone cannot clean up.
Signal-integrity verification and debugging: On a prototype board, faults like "only one signal misbehaves" or "a glitch appears on a line that runs near the clock" are often caused by crosstalk. Use a simple model like this tool to estimate the coupled length and spacing of a candidate aggressor and victim, then decide whether to revise the layout, split layers or rework the termination. Detailed 3D electromagnetic analysis and TDR measurement are most efficient once that estimate is in hand.
Common Misconceptions and Pitfalls
A common misconception is "crosstalk drops linearly as you move traces apart". In reality the coupling falls steeply with spacing (roughly inversely with the square of the distance). Doubling the spacing cuts the coupling to nearly a quarter, yet moving traces that are already more than 3W apart even further apart yields only a tiny improvement. The "Crosstalk vs trace spacing" chart in this tool clearly shows the steep rise in the narrow region and the flatness in the wide region. Area is finite, so the right approach is to allocate spacing to sensitive signals first.
Next, the belief that "crosstalk always increases the longer the coupled length". This holds for far-end crosstalk but not for near-end crosstalk. NEXT plateaus once the coupled length exceeds the saturation length, and running longer in parallel beyond that point does not increase the amplitude. Conversely, on a trace that is already saturated, "shortening the run a little" barely helps near-end crosstalk — you need to widen the spacing or split layers. Confuse which fix works for NEXT versus FEXT and you can rework endlessly with no improvement.
Finally, assuming that "crosstalk is the same in microstrip and stripline". In a surface microstrip the signal travels partly through air and partly through the board dielectric, so the balance of capacitive and inductive coupling is broken and far-end crosstalk does not reach zero. An inner-layer stripline, on the other hand, is surrounded by a uniform dielectric, so the two couplings line up and far-end crosstalk in theory almost vanishes. Routing a fast, important signal on an inner layer is a standard fix that exploits this property. Keep in mind that the choice of routing layer is itself a crosstalk countermeasure.
How to Use
Enter spacing between traces in mm (typical range 0.1–2.0 for PCB designs)
Set trace width in mm; narrower traces increase coupling coefficient
Input trace length in mm; crosstalk peaks at saturation length then stabilizes
Define rise time in ps; faster edges generate stronger capacitive and inductive coupling
Click Calculate to obtain NEXT, FEXT, coupling ratio, and 3W-rule compliance
Worked Example
Two 0.15 mm copper traces on FR-4 with 0.2 mm spacing, 50 mm length, 100 ps rise time. Simulator returns NEXT = 8.3%, FEXT = 2.1%, NEXT level = −20.6 dB, saturation length = 35 mm, coupling ratio = 0.31. The 3W-rule (spacing ≥ 3× width = 0.45 mm) fails, indicating potential reflections and crosstalk in high-speed digital designs above 1 GHz.
Practical Notes
Increase spacing to 0.5 mm minimum on 8-layer boards carrying differential pairs; NEXT drops below 5% when spacing exceeds 3× trace width
Saturation length depends on dielectric constant and coupling strength; crosstalk does not improve beyond this point regardless of additional length
Rise times faster than 50 ps on tightly coupled traces (≤0.15 mm spacing) demand microstrip or stripline geometry to suppress far-end noise
Use symmetrical trace routing and guard traces between signal pairs to reduce capacitive coupling by 40–60%