2x Thru De-Embedding: IEEE 370 Guide for RF PCBs

What Is 2x Thru De-Embedding?

2x thru de-embedding is a technique that removes the electrical effects of test fixtures, connectors, and launches from vector network analyzer (VNA) S-parameter measurements by characterizing a single symmetric through-line standard whose total length equals twice the fixture path. The method splits the measured 2x thru response at its midpoint, extracts each fixture half as an error network, and then mathematically peels those networks away from the device-under-test (DUT) measurement. The result is the true, fixture-free performance of the DUT—insertion loss, return loss, and phase—without requiring the multiple standards that TRL or SOLT calibrations demand.

The approach was formalized in the IEEE 370-2020 standard, which defines the 2x thru de-embedding structure, the bisection algorithm, and the quality metrics (causality, passivity, reciprocity) that a valid de-embedded result must satisfy. If you are measuring anything from a 50-ohm microstrip evaluation coupon to a 28-Gbps channel, 2x thru de-embedding is now the industry baseline.

Why Fixture Effects Make 2x Thru De-Embedding Essential

Every measurement path between your VNA port and the DUT adds loss, phase shift, impedance discontinuities, and mismatch ripple. A typical SMA-to-microstrip launch can introduce 0.15–0.4 dB of loss per connector at 10 GHz, plus return-loss degradation below 15 dB if the transition is not optimized. When the DUT itself is a short trace whose total insertion loss might be only 0.3 dB, the fixture contribution can exceed the quantity you are trying to measure.

Without 2x thru de-embedding, you risk over-stating dielectric and conductor loss, mischaracterizing impedance, and making wrong material or stackup decisions. This is a common trap covered in detail in our guide to insertion loss measurement pitfalls.

How the 2x Thru De-Embedding Method Works, Step by Step

Step 1: Fabricate the 2x-Thru Standard

Build a through-line that includes both fixture halves connected back-to-back with no DUT in between. The structure must be geometrically symmetric about its midpoint. Typical lengths range from 2 inches (≈ 50 mm) to 6 inches depending on the fixture launch depth. The same board material, stackup, surface finish, and connector type used for the DUT board must be replicated exactly. Guidance on building repeatable fixtures is in our RF test fixture design guide.

Step 2: Measure the 2x-Thru with a Calibrated VNA

Perform a standard SOLT or eCal calibration at the coaxial reference plane (the cable ends), then measure the 2x-thru structure’s full 2-port S-parameters. Save the Touchstone (.s2p) file. Frequency resolution of 1–10 MHz per point up to your maximum frequency of interest is typical; finer steps improve time-domain gating accuracy.

Step 3: Bisect and Extract Error Adapters

The bisection algorithm splits the 2x-thru S-parameter matrix at the electrical midpoint. Each half becomes a two-port error adapter (often called a “fixture model”). The math converts S-parameters to T-parameters (transfer matrices), takes the matrix square root (accounting for symmetry), and yields T_fixture for each side. Several commercial tools (Keysight ADS, Ansys Circuit, AtaiTec ISD) and the open-source scikit-rf Python library implement IEEE 370 bisection.

Step 4: De-embed the DUT

Measure the DUT in the same fixtures. Convert the raw DUT S-parameters to T-parameters, then multiply by the inverse of each fixture’s T-matrix:

T_DUT = T_fixtureL⁻¹ · T_measured · T_fixtureR⁻¹

Convert back to S-parameters. The result is the DUT’s intrinsic response, referenced to the fixture–DUT boundary.

Step 5: Validate Quality Metrics

IEEE 370 specifies three checks: the de-embedded result must be causal (impulse response is zero before t = 0), passive (no |S₂₁| > 0 dB), and reciprocal (S₁₂ ≈ S₂₁). If any check fails, suspect asymmetry in the 2x-thru fabrication, connector torque variation, or an incorrect reference-plane definition.

2x Thru De-Embedding vs TRL vs SOLT: When to Use Which

Choose 2x thru de-embedding when you need fast, low-cost fixture removal for PCB-level measurements up to ~50 GHz and can fabricate a symmetric through-line. Choose TRL when you need the highest accuracy at millimeter-wave frequencies or on-wafer probing. Choose SOLT when you are working entirely in a coaxial environment and the reference plane is at the connector.

Fixture Planning: Do You Need a Dedicated Test Coupon?

Before committing board space, estimate whether a dedicated ICT or de-embedding fixture coupon is justified for your measurement campaign. The estimator below helps you weigh DUT count, frequency range, and connector type to decide.

If the tool suggests a dedicated fixture, allocate at least one 2x-thru coupon and one DUT coupon per stackup variant on your panel. For high-mix programs, placing the 2x-thru in the panel border saves active board area while keeping the same copper and dielectric environment.

Copper Roughness: The Hidden Error After 2x Thru De-Embedding

De-embedding removes fixture loss but does not fix errors in your DUT’s own loss model. Above 5 GHz, copper roughness becomes a dominant contributor to conductor loss—often adding 30–50 % on top of smooth-copper predictions. Standard electrodeposited (ED) copper with an Rz of 6–10 µm behaves very differently from RTF (reversed-treat foil, Rz ≈ 2–3 µm) or HVLP (Rz < 2 µm) grades. If your de-embedded insertion loss still looks higher than simulation, roughness is the first suspect.

Use the tool below to estimate the additional loss from your foil’s roughness profile at a given frequency. Enter the RMS roughness (Rq) or peak-to-valley roughness (Rz), trace geometry, and frequency.

For example, a 50-ohm microstrip on RO4350B versus RO4003C with standard 1-oz ED foil (Rq ≈ 1.8 µm) at 20 GHz may show roughly 0.08 dB/inch of additional roughness loss beyond the smooth-conductor baseline. Switching to HVLP foil (Rq ≈ 0.6 µm) cuts that adder by more than half. After 2x thru de-embedding, this kind of detail is exactly what you can finally see and act on. Our page on reducing RF PCB loss ranks roughness among the top three levers available to designers.

Common Pitfalls and How to Avoid Them

• Asymmetric 2x-thru: Even 1–2 mil (25–50 µm) trace-width variation between halves introduces error. Use the same Gerber mirrored, not a separate drawing.
• Connector torque inconsistency: SMA and 2.92 mm connectors should be torqued to the manufacturer’s spec (typically 8 in-lb for SMA). A 2 in-lb deviation can shift return loss by 3–5 dB above 20 GHz.
• Ignoring the reflect standard: IEEE 370 optionally uses a 1x-reflect (open or short at the midpoint) to improve accuracy. For measurements above 30 GHz, adding this standard is strongly recommended.
• Frequency-dependent Dk drift: If the 2x-thru and DUT boards come from different panels or press cycles, slight Dk variation shifts the electrical midpoint. Fabricate both on the same panel whenever possible.
• Applying coaxial-cal de-embedding twice: If you already moved the reference plane to the connector with an eCal, do not also apply a fixture model that includes the connector. Double de-embedding produces non-physical gain artifacts.

Software Tools That Support IEEE 370 2x Thru De-Embedding

Several commercial and open-source packages implement the full IEEE 370 flow:

• Keysight PathWave ADS – built-in IEEE 370 de-embedding wizard.
• Ansys Circuit (HFSS) – automatic fixture extraction from 2x-thru .s2p.
• AtaiTec ISD – lightweight, dedicated to de-embedding.
• scikit-rf (Python) – free, scriptable, supports IEEE 370 bisection and quality checks.

For verifying that a laminate supplier’s published Df actually matches your boards, the de-embedded DUT data feeds directly into the procedure described in our article on verifying a supplier’s loss claim.

FAQ

Can I use 2x thru de-embedding on differential pairs?

Yes. Measure the 2x-thru as a 4-port (or 2-port with mode conversion), then apply mixed-mode S-parameter bisection. IEEE 370 explicitly covers differential 2x-thru structures.

What is the maximum frequency for 2x thru de-embedding?

The method itself has no inherent frequency limit. Practical limits come from the connector (2.92 mm to 40 GHz, 1.85 mm to 67 GHz, 1.0 mm to 110 GHz) and the quality of the fixture symmetry. Above 50 GHz, adding a 1x-reflect standard significantly improves results.

Does the 2x-thru line need to be exactly 50 ohms?

It should match the nominal impedance of the DUT’s transmission line as closely as possible. A few ohms of offset is tolerable because the algorithm captures that mismatch, but large deviations reduce bisection accuracy.

How long should the 2x-thru line be?

Long enough that the two fixture launches are electrically separated in the time domain—typically at least 1 ns of one-way delay (≈ 6 inches on FR-4, ≈ 5 inches on RO4003C). Shorter lines risk overlapping reflections that corrupt the bisection.