Port Extension and De-Embedding for Fixtures and PCB Traces
Port extension removes only the phase of an ideal lossless line, using θ = 360 × l / λ. De-embedding removes the full S-parameters of the fixture, including loss, and TRL on the board avoids the fixture model altogether.
For practitioners
Where the reference plane sits
A VNA calibration places the measurement at a reference plane. Normally this is the cable end of the calibrated connector. Everything the calibration standards touched is corrected. Everything beyond that point is still part of your result.
A device on a board is not at that plane. Between the calibrated connector and the device pins you have a board connector, a launch and a PCB trace. The VNA measures all of it together with the device. The result is a cascade, not the device alone.
Treat the fixture as a two-port on each side of the device. Each has its own S-parameters, and each adds three effects:
- Delay. The trace adds phase to S21 and rotates S11 and S22.
- Loss. Conductor and dielectric loss add to the insertion loss.
- Mismatch. The connector, launch and trace impedance step add their own reflections, which interact with the device and create ripple.
The job is to move the reference plane from the connector to the device. Two software methods do this, and a third avoids the problem by calibrating at the device.
The trace and its electrical length
A microstrip line is a conductor over a ground plane with a dielectric between them. Its width and the dielectric height set the impedance. The geometry also sets the wavelength on the line, which is shorter than in air.
For a trace of length l and wavelength λ on that line, the electrical length is θ = 360 × l / λ degrees. The trace shifts the phase of S21 by that angle. A quarter-wavelength trace gives 90°. A full wavelength gives 360°.
The effect grows with frequency, because λ gets shorter while l stays fixed. A trace that is invisible at low frequency can rotate a phase plot by many turns at the top of your band. Magnitude of a matched, lossless trace stays flat, which is why phase errors are easy to miss in a magnitude-only plot.
Port extension: phase only
Port extension tells the VNA to remove a known electrical delay from a port. The model is an ideal lossless line. It applies a phase correction and nothing else.
For a transmission measurement through one extended section of electrical length θ, the correction is:
For a reflection measurement the wave travels the section twice, so the rotation is doubled:
Port extension is simple. You enter a delay or a length, and you can often set it by watching the S11 phase of an open at the device and adjusting until the trace is flat. It needs no fixture file.
Its limit is the model. A lossy fixture adds insertion loss that port extension alone cannot remove. It also cannot remove mismatch at the connector or launch. It is accurate only when the fixture is short, low loss and well matched, so that phase is the dominant error.
De-embedding: remove the whole fixture
De-embedding mathematically removes the S-parameters of the fixture, using a measured or modelled fixture description. Because it uses complex S-parameters, it removes loss, delay and reflections together.
The usual method converts S-parameters to transfer (T) parameters, because T-parameters multiply in a cascade. If the measured chain is the input fixture A, the device, and the output fixture B, then:
The fixture description can come from two places:
- A measurement. You measure a fixture standard, for example a through line built like the fixture, or an open and short at the device plane, and derive A and B from it.
- A model. You build A and B from a simulation or an equation model of the line, launch and connector.
The accuracy of de-embedding is the accuracy of the fixture description. If the model is wrong, the error is built into the result. A line that does not match the real board in width, dielectric or launch will leave a residual in both loss and phase. Inverting a poor model can also amplify noise.
To check a model's line impedance, the Microstrip Impedance and Width Calculator gives the width for a target impedance on FR-4 or Rogers.
Calibrating on the board with TRL
TRL calibration made on the board itself places the reference plane at the device without a separate fixture model. The thru, reflect and line standards are built on the same board, with the same launch and the same substrate, so the fixture is part of the calibration.
The cost is board area and design effort. You need standards that fit the layout, and a line length that suits the frequency span. The benefit is that no fixture file is needed and the plane lands at the device.
Comparing the three
| Method | Corrects | Needs | Accuracy |
|---|---|---|---|
| Port extension | Phase only | A delay or length | Good for short, low-loss, matched fixtures |
| De-embedding | Loss, phase, mismatch | A measured or modelled fixture | Limited by fixture description |
| TRL on the board | Whole fixture | Standards on the board | Highest, with more build effort |
Choose by what dominates the error. If phase is the only problem, port extension is enough. If the fixture has loss or reflections, use de-embedding. If you can build standards on the board and need the best result, calibrate there.
Worked example
Assume a 15 mm microstrip trace and a wavelength on the line of 60 mm at the test frequency. These are assumed values for the example.
θ = 360 × 15 / 60 = 90°
The trace shifts S21 by 90°. Port extension of 90° on that port restores the phase of S21. For S11 at the same port, the rotation is 2 × 90° = 180°, so the open at the device would read 180° away from its true phase without correction.
Now assume the trace also has 0.4 dB of insertion loss, again an assumed value. The corrected S21 still reads 0.4 dB low, because port extension does not touch magnitude. Only de-embedding, or a calibration on the board, removes that 0.4 dB.
In plain words · for students and beginners
Port extension and de-embedding in plain words
The idea in an everyday picture
Imagine you want to know how long your own walk takes, but you must start from the car park, not your front door. The path from the car park to your door adds time and effort that are not yours. To get your own number, you subtract that path.
The car park is the connector where the VNA was calibrated. Your front door is the device. The path between them is the fixture and trace.
The picture breaks down in one place. A path adds time, but a real trace adds time and also tiredness: loss. Subtracting only the time leaves the tiredness behind.
The same example in simple steps
- The trace is 15 mm long. We assume a wavelength on it of 60 mm.
- Divide: 15 / 60 = 0.25, a quarter of a wavelength.
- Multiply by 360: 0.25 × 360 = 90°. The trace twists the phase of the signal by 90°.
- Port extension removes that 90°. The phase now looks as if the device were at the connector end of nothing, right at the device.
- The trace also lost 0.4 dB in our example. The size of the signal is still 0.4 dB too small, because port extension does not touch size.
- De-embedding removes both the twist and the loss, because it removes the whole fixture.
The curve above shows that the width of a trace sets its impedance. A fixture trace with the wrong width is not a plain delay. It is a small mismatch, and that is one more reason a delay-only fix can fall short.
Mistakes beginners make
- Using port extension and expecting the loss to vanish. It only shifts phase.
- Trusting a fixture model that does not match the real board. The error from the wrong model goes straight into the result.
- Forgetting that reflection is doubled. The signal goes along the trace and back, so S11 rotates by twice the angle that S21 does.
- Reading only the magnitude plot. A big phase error can hide while the magnitude looks fine.
Remember
- The calibration puts the reference plane at the connector, not at your device.
- Port extension fixes phase only. De-embedding fixes loss, phase and mismatch.
- Calibrating with TRL on the board puts the plane at the device without a fixture model.
Related pages on RF View
Frequently asked questions
What is the difference between port extension and de-embedding?
Port extension removes the delay of an ideal lossless line, so it corrects phase only. De-embedding removes the complete S-parameters of the fixture, so it can also remove loss and mismatch.
Can port extension remove the insertion loss of a PCB trace?
No. A lossy fixture adds insertion loss, and port extension shifts phase only. You need de-embedding, or a calibration made on the board, to remove that loss.
How do I calculate the phase shift of a microstrip trace?
Use θ = 360 × l / λ degrees, where l is the trace length and λ is the wavelength on the line at the measurement frequency. A trace of one quarter wavelength shifts S21 by 90°.
When is TRL on the board better than de-embedding?
TRL made on the board itself places the reference plane at the device without a separate fixture model. It is the better choice when you can build the thru, reflect and line standards on the same board.