Edge-Coupled Asymmetric (Offset) Stripline Calculator
Calculate differential and common-mode impedance for an edge-coupled pair sitting off-centre between two reference planes — the usual case in an asymmetric stack-up.
Cohn coupled stripline × offset factor fitted to 2-D MoMAPI slug edge-coupled-internal-asymmetric
Formula
S. B. Cohn, IRE Trans. MTT-3 (1955) for the centred case; offset correction fitted to the committed solver in src/lib/pcb/__tests__/solver.
| Symbol | Quantity | Your value |
|---|---|---|
| Zdiff | Differential impedance, 2·Z_odd (Ω) | 97.7 Ω |
| W | Trace width (mm) | 130 µm |
| S | Edge-to-edge spacing (mm) | 200 µm |
| h₁ | Dielectric below the traces (mm) | 200 µm |
| h₂ | Dielectric above the traces (mm) | 500 µm |
| κ | Offset correction, identically 1 at h₁ = h₂ | — |
Assumptions
6 items
- 01One homogeneous dielectric between two solid planes, εeff = εr, quasi-static and lossless; the frequency sets only λg = c/(f·√εr).
- 02Each mode’s capacitance is Cohn’s exact centred coupling ratio applied to the offset single line, C_mode = C_single,offset × (C_mode,centred/C_single,centred) × κ, all at zero thickness, with κ fitted to a 2-D method-of-moments solver and exactly 1 when h₁ = h₂.
- 03The fit covers W/b 0.05–0.8, S/b 0.05–1.0 and a nearer gap of 0.125–0.5 of b = h₁ + h₂, worst case 1.85 % on a mode impedance (1.9 % published); outside it the result is flagged as extrapolated and Within Validated Range reads 0.
- 04Copper thickness is added once, after the offset scaling: each edge’s increment from the thick offset single line and, in the odd mode, the sidewall across the gap, with the centred pair’s fitted gap terms, so the two pairs agree exactly at h₁ = h₂; the trace is a rectangle.
- 05Zdiff = 2·Zodd and Zcommon = Zeven/2; the offset ratio is |h₁ − h₂|/(h₁ + h₂), and a height that is not positive is refused.
- 06Synthesis solves the trace width or the spacing numerically within its range for the target Zdiff; the material picker sets εr to the laminate’s design Dk.
Worked example
100 Ω pair off-centre, 0.15 mm below and 0.35 mm above
An LVDS pair of 0.1 mm traces sits 0.15 mm above one plane and 0.35 mm below the other in FR-4 (εr 4.4), with 17.5 µm copper. Solve the edge-to-edge spacing for Zdiff = 100 Ω.
| Step | Expression | Result |
|---|---|---|
| Spacing for 100 Ω (solved) | Zdiff(S) = 100 Ω; W = 0.1 mm, h₁ = 0.15 mm, h₂ = 0.35 mm | 0.1528 mm |
| One trace alone | Z₀(W; h₁ = 0.15 mm, h₂ = 0.35 mm) | 60.5 Ω |
| Odd-mode capacitance | C_odd,centred × (C_single,offset / C_single,centred) × (1 − κ_odd) + 2·ΔC_edge + φ·2t/S | 3.592 |
| Differential impedance | 2 × 376.73 / (√4.4 × 3.592) | 100 Ω |
| Common-mode impedance | 70.8 / 2 | 35.4 Ω |
How it works
Most differential pairs are not centred between their reference planes. A pair on an inner signal layer usually has a thin prepreg to one plane and a thicker core to the other, and that asymmetry moves the impedance by more than the tolerance the fabricator is holding.
The offset case has no exact closed form, so this calculator is built as an exact base with a correction on top. The base is the exact centred coupling ratio applied to the offset single line — both of which are already known quantities. What that decomposition misses is a residual correction, and that correction was fitted to a two-dimensional method-of-moments field solver committed alongside the calculator.
The important property of the correction is that it vanishes identically when the two dielectric heights are equal. That means the reduction to the exact symmetric case holds by construction rather than by luck: set h₁ = h₂ and this calculator returns the symmetric calculator's answer to the last digit, not merely to within a tolerance.
Offsetting the pair moves it closer to one plane, which raises the capacitance to that plane and lowers every mode impedance. The effect is not subtle. Moving a pair from centred to a quarter of the plane spacing typically drops the differential impedance by 10 to 20%, which is why treating an offset pair as centred is a common way for a controlled-impedance board to come back out of specification.
The model's residual against the solver is 1.85% worst case on a mode impedance and 0.44% rms, over trace widths of 0.05 to 0.8 of the plane spacing, spacings of 0.05 to 1.0, and a nearer gap between 0.125 and 0.5 of the plane spacing. Outside that range the result is still returned, but it is flagged as extrapolated.
Practical tips
- Run the centred case first as a reference, then apply the real offset. The difference between the two is the number to take to a stack-up review.
- If the offset is large and the impedance target is tight, ask whether the pair can move to a more balanced layer before adjusting width and spacing to compensate.
- Check the extrapolation flag before trusting a number. Very thin outer gaps in particular fall outside the fitted range.
- An offset pair is also asymmetric in its coupling to the two planes, so plane-referencing discontinuities matter more; keep a continuous reference under the whole route.
- Where the two heights are nearly equal, use the symmetric calculator — its answer is exact rather than fitted, so there is no residual at all.
- Compare the odd-mode impedance against the single-trace value to see how much of the result is coupling and how much is placement.
Common mistakes
- Averaging the two dielectric heights and using the symmetric calculator. The relationship is not linear in the offset, so the average gives the wrong answer in the direction that flatters the design.
- Taking the heights from the drawing rather than from the fabricator's actual stack-up. Prepreg thickness after lamination is not the thickness on the datasheet, and it is the built value that sets the impedance.
- Ignoring which plane is nearer. The model is symmetric under swapping the two heights, so only the offset matters — but confusing the two while reading a stack-up produces the wrong offset entirely.
- Treating the extrapolation flag as cosmetic. Outside the validated range the model has no stated accuracy, and a fit outside the geometry it was fitted over is not a calculation, it is a guess.
- Forgetting that the copper thickness sits inside the plane-to-plane distance. The dielectric heights are measured from the copper faces, not from the layer centreline.
Frequently asked questions
5 questions
When does a pair end up off-centre?
Almost always, on any stack-up that is not deliberately symmetric. Inner signal layers typically have a thin prepreg to one plane and a thicker core to the other, and the resulting offset is normal rather than exceptional.
How much does the offset actually change Z_diff?
Moving a pair from centred to a quarter of the plane spacing typically drops the differential impedance by 10 to 20%. That is several times the fabricator's impedance tolerance, so it cannot be absorbed.
Why does the calculator tell me the result is extrapolated?
Because the offset correction is fitted, and a fit outside the geometry it was fitted over has no stated accuracy. Inside the published range the worst case is 1.85%; outside it the number is still returned but flagged, so you know to check it against a field solve.
Is the answer the same if I swap the two heights?
Yes, exactly. Only the magnitude of the offset matters, not which plane is nearer, and the model is symmetric under the swap by construction.
Does this handle a dual-stripline stack-up?
It handles the impedance of the pair against its two planes. A dual-stripline arrangement adds coupling to the second signal layer, which is a separate question — use the dual stripline calculator for that interaction.
Shop Components
As an Amazon Associate we earn from qualifying purchases.