TRL (through-reflect-line) calibration is a complex subject. TRL calibration and de-embedding are two separate mechanisms for the removal of fixture errors within a measurement. In fact there are at least five mechanisms that may be appropriate, depending upon the fixture, its errors, the measurement and the accuracy required:

1. Reference plane shift or rotation – Removes the effect of delay in the fixture feed lines, but not their loss, mismatch loss or flatness errors. Applies for transmission (S21, S12) and reflection (S11, S22)
measurements.

2. Normalization – Removes the effects of loss and flatness of the feedlines for transmission measurements.

3. Time-domain gating – Assuming that time-domain translation is available for the accurate time location of fixture errors, an additional time-domain gating function can be used to gate them out and then convert back to the frequency or s-parameter domain. This is effective for reflection measurements and removes fixture mismatch and delay, but not fixture loss.

4. De-embedding – Uses S-parameter data files for each of the test port feedlines. These are mathematically backed out of (de-embedded from) all transmission and reflection measurements, wholly removing their errors (at least to the level of accuracy in those de-embed files).
5. TRL calibration – While we can readily and accurately define and apply SOL (short-open-load) calibration standards to a connectorized test port, we can't so readily define SOL standards to apply at the end face of any given fixture feedline. If we could, we would be able to define a calibration reference plane right at the DUT terminals and remove all errors to the feed side of that reference plane. TRL calibration instead uses lengths of transmission line that have identical structure to the feed lines, often several lengths (to achieve a broadband calibration) and with various terminations; a practical, if inconveniently time-consuming, process and one that cannot fully address fixture mismatch errors.

The PicoVNA supports mechanisms (1), (2), (1)+(2) and (4).

Despite having the time-domain function available, we do not have the gating function that would support (3). Nor do we support (5) in its many complex forms that in practice become necessary.

Our User's Guide has some good background and of course instruction in the use of reference plane shift,
normalization and de-embedding. This will help you to decide which method is most appropriate to your application.

If you have to apply de-embedding to your measurement, there are two options for the derivation of the de-embed sparameter files. Most commonly a simulation model of the feedlines would be used. You might also have a measurement option. If the feedline and launch points are geometrically identical, for instance, you might measure Port 1 and 2 feeds as a through pair and attribute half of the transmission parameters and an average of the mismatch parameters to each. The choice depends on the details of your fixture, your available fixture test pieces and your application.