When it is needed
It is not always possible to calibrate directly at the device plane. A transistor soldered to a board, a component in a fixture, or a chip with access traces make it necessary to measure through lines, connectors and transitions that are not part of the device. De-embedding removes that effect after the measurement.
How it works
A fixture between the calibrated plane and the device is a two-port network. If its behavior is known, the total measured network can be seen as the cascade of the input network, the device and the output network. To work with cascades, S-parameters are converted to transfer parameters, in which the cascade is a product of matrices. The measurement is multiplied by the inverses of the fixture matrices, and the result is the device alone.
Where the fixture characterization comes from
- Measurement of auxiliary structures. A thru, an open structure, a short, or two fixture halves joined together, which make it possible to deduce the effect of each side.
- TRL calibration in the fixture. Calibration is performed directly in the device medium.
- Electromagnetic simulation or circuit model. The fixture is calculated from its geometry.
- Port extension. A simple case that compensates only for an electrical delay, without loss or mismatch.
Assumptions and limits
- The fixture must be linear, passive and reproducible, and the auxiliary structures must represent the actual fixture well.
- If the fixture has coupling between its halves or radiation, the cascade model is no longer accurate.
- Model errors propagate: in a high-loss fixture, measurement noise is amplified by the inversion.
What to check first
Confirm that the result is physical: a passive device must not show gain, and a known line must appear as such. If the result oscillates or shows impossible values, this usually indicates an incorrect fixture model or unrepresentative reference structures.