Measurements, Testing, and Calibration Network Analysis Informational

How do I de-embed fixture and cable effects from my S-parameter measurements?

Q: When should I use TRL instead of de-embedding?

TRL calibration is preferred when: (1) You can fabricate the calibration standards on the same substrate as the DUT (PCB or MMIC wafer). (2) You need the highest possible accuracy (TRL does not rely on lumped standards like SOLT, which have imperfect models at high frequencies). (3) You are characterizing many DUTs on the same substrate (calibrate once, measure many). De-embedding (e.g., 2x-Thru) is preferred when: (1) You cannot fabricate calibration standards on the DUT substrate. (2) You are measuring packaged devices in a test fixture. (3) You want a quick method with minimal additional standards (just one 2x-Thru measurement).

Q: How do I validate that de-embedding worked correctly?

Measure a known reference device using the same fixture and de-embedding procedure. Compare the de-embedded result to: (1) The manufacturer datasheet specifications. (2) A measurement of the same device in a different fixture (independent validation). (3) Simulation results (if a validated model exists). Common validation devices: a precision attenuator (known S21 and S11), a short length of transmission line (known impedance and loss), or a calibration kit verification standard. Also check for physical reasonableness: (a) A passive device should have |S21| ≤ 1 (0 dB) at all frequencies. (b) The S-parameters should be smooth (no sudden jumps or spikes). (c) Reciprocity should hold: S21 ≈ S12 for a passive device. Violations of these checks indicate de-embedding errors.

Q: Can I de-embed using port extension only?

Port extension adds a lossless delay, which is equivalent to physically moving the reference plane by a known distance. It corrects only for the phase shift of the fixture. Port extension is adequate when: the fixture is short ( 25 dB). For lossy or mismatched fixtures: port extension introduces calibration errors because it does not account for fixture loss or internal reflections. Use full S-parameter de-embedding or TRL calibration instead. A quick check: perform port extension and then measure a SHORT at the DUT reference plane. If the return loss is > 0.1 dB or the phase is not exactly 180° ± 2°: the fixture has loss or mismatch that port extension cannot correct.

De-embedding is the mathematical removal of fixture, cable, and adapter effects from a VNA measurement to reveal the true DUT S-parameters. The concept: the measured S-parameters include the DUT plus the S-parameters of the test fixtures on both sides. De-embedding separates these: [S_measured] = [S_fixture1] × [S_DUT] × [S_fixture2]. Solving for [S_DUT]: [S_DUT] = [S_fixture1]^-1 × [S_measured] × [S_fixture2]^-1. Methods: (1) S-parameter de-embedding: if the fixture S-parameters are known (from manufacturer data or separate measurement), the VNA can mathematically remove them. Most modern VNAs have a built-in de-embedding function. (2) Port extension (simple de-embedding): adds a lossless electrical delay to shift the reference plane. Corrects for the phase shift of a known-length transmission line but does not correct for fixture loss or mismatch. Suitable only for low-loss, well-matched fixtures. (3) 2x-Thru de-embedding: measure a "thru" standard consisting of two back-to-back copies of the fixture (fixture1-fixture2 without a DUT). The VNA mathematically splits this measurement in half to extract each fixture S-parameters. Advantage: does not require a separate fixture model. Only requires one additional measurement. Widely used for PCB test fixtures and wafer probes. (4) TRL calibration: calibrate directly at the DUT reference plane using Thru-Reflect-Line standards fabricated in the same medium as the DUT (e.g., on the same PCB or wafer). Eliminates fixture effects entirely at the calibration step. Highest accuracy for planar circuits.

Category: Measurements, Testing, and Calibration

Updated: April 2026

Product Tie-In: VNAs, Calibration Kits, Cables

De-embedding and Fixture Removal

Fixture de-embedding is essential when the DUT cannot be directly connected to the VNA ports, which is the case for: PCB-mounted components, on-wafer devices (MMICs), packaged components in test fixtures, and non-coaxial devices requiring transitions.

T-parameter Conversion for De-embedding

De-embedding requires matrix operations. S-parameters are not directly cascadable (you cannot simply chain S-matrices for series-connected networks). Instead: convert S-parameters to T-parameters (transfer matrix), which cascade by multiplication. The conversion: T11 = -(S11×S22 - S12×S21)/S21. T12 = S11/S21. T21 = -S22/S21. T22 = 1/S21. Cascading: [T_total] = [T_fixture1] × [T_DUT] × [T_fixture2]. De-embedding: [T_DUT] = [T_fixture1]^-1 × [T_total] × [T_fixture2]^-1. Then convert back to S-parameters. The process requires that the fixture T-matrices are known (measured or modeled).

2x-Thru Method

The 2x-Thru (also called AFR: automatic fixture removal) is the most practical de-embedding method for PCB fixtures: (1) Define the fixture: the test fixture is a PCB launch (connector to trace transition + some trace length). The DUT is in the middle. (2) Fabricate a 2x-Thru standard: a PCB with two fixture halves connected back-to-back (no DUT, just two launches connected by a short trace). (3) Measure the 2x-Thru on the VNA. (4) Apply the de-embedding algorithm: the software bisects the 2x-Thru measurement at the midpoint, extracting S_fixture1 and S_fixture2 (assuming the fixture is symmetric and reciprocal). (5) Measure the DUT-in-fixture. Apply de-embedding: S_DUT = de-embed(S_measured, S_fixture1, S_fixture2). Assumptions: the fixtures are identical (or symmetrical) and the fixture response is separable from the DUT. These assumptions hold well for PCB test coupons where the fixtures are fabricated on the same panel as the DUT. Tools: Keysight ADS AFR, Anritsu fixture de-embedding, Simbeor de-embedding, and open-source Python scikit-rf library.

Sources of De-embedding Error

(1) Fixture model accuracy: if the fixture S-parameters are incorrect, the de-embedded DUT data will have systematic errors. Validate the fixture model by measuring a known reference DUT and comparing the de-embedded result to the known value. (2) Symmetry assumption: the 2x-Thru method assumes both fixture halves are identical. Manufacturing variations (connector soldering, trace width tolerance) break this assumption. Effect: typically < ±0.1 dB and < ±2° phase error for well-controlled PCB fabrication. (3) Fixture-DUT interaction: the fixture model is extracted without the DUT present. If the DUT significantly changes the impedance at the fixture-DUT interface (e.g., a highly mismatched device), the fixture behavior may change (higher-order mode excitation, resonance shifts). This is a limitation of all linearized de-embedding methods. (4) Reference plane uncertainty: the mathematical "cut" between the fixture and DUT must be at a well-defined physical location. Uncertainty in the cut location introduces phase error proportional to the electrical length uncertainty.

De-embedding Equations

T-params: T11 = -(S11·S22 - S12·S21)/S21
Cascade: T_total = T_fix1 × T_DUT × T_fix2
De-embed: T_DUT = T_fix1⁻¹ × T_meas × T_fix2⁻¹
Port Extension: φ_shift = 2π·f·delay
2x-Thru: S_fix = bisect(S_thru_thru)

Common Questions

Frequently Asked Questions

When should I use TRL instead of de-embedding?

TRL calibration is preferred when: (1) You can fabricate the calibration standards on the same substrate as the DUT (PCB or MMIC wafer). (2) You need the highest possible accuracy (TRL does not rely on lumped standards like SOLT, which have imperfect models at high frequencies). (3) You are characterizing many DUTs on the same substrate (calibrate once, measure many). De-embedding (e.g., 2x-Thru) is preferred when: (1) You cannot fabricate calibration standards on the DUT substrate. (2) You are measuring packaged devices in a test fixture. (3) You want a quick method with minimal additional standards (just one 2x-Thru measurement).

How do I validate that de-embedding worked correctly?

Measure a known reference device using the same fixture and de-embedding procedure. Compare the de-embedded result to: (1) The manufacturer datasheet specifications. (2) A measurement of the same device in a different fixture (independent validation). (3) Simulation results (if a validated model exists). Common validation devices: a precision attenuator (known S21 and S11), a short length of transmission line (known impedance and loss), or a calibration kit verification standard. Also check for physical reasonableness: (a) A passive device should have |S21| ≤ 1 (0 dB) at all frequencies. (b) The S-parameters should be smooth (no sudden jumps or spikes). (c) Reciprocity should hold: S21 ≈ S12 for a passive device. Violations of these checks indicate de-embedding errors.

Can I de-embed using port extension only?

Port extension adds a lossless delay, which is equivalent to physically moving the reference plane by a known distance. It corrects only for the phase shift of the fixture. Port extension is adequate when: the fixture is short (< lambda/4), the fixture loss is negligible (< 0.1 dB), and the fixture is well-matched (RL > 25 dB). For lossy or mismatched fixtures: port extension introduces calibration errors because it does not account for fixture loss or internal reflections. Use full S-parameter de-embedding or TRL calibration instead. A quick check: perform port extension and then measure a SHORT at the DUT reference plane. If the return loss is > 0.1 dB or the phase is not exactly 180° ± 2°: the fixture has loss or mismatch that port extension cannot correct.

Keep Reading