De-embedding S-parameters

Updated June 29, 2009

De-embedding is an extremely important topic. One of these days we'll develop some better content for this page. Be patient! If you have expertise in this area, please contribute!

While we are on the subject, here's something that bugs us. If you have data that is not de-embedded, it is stupid to say it is "un-de-embedded". Please just say "embedded"!

De-embedding is the act of taking data that is measured in a test fixture and removing the effects of the fixture so that the data is accurate to reference planes that are more useful, using vector measurements of known standards. For example, you can de-embed a FET measurement so that the reference planes are the gate and drain busbars; these data can be used in an amplifier design or to create an equivalant ciruit model.

If your component is matched to fifty ohms, you don't always need to properly de-embed it to understand its performance. In most cases you only need to remove attenuation of input and output traces from the test fixture. This is a scalar version of de-embedding, and you can do this in Excel or even with a calculator, so long as you have data on fixture losse (over frequency) and can apportion it to input and output sides. Sometimes the fixture is symmetrical, and you can split its loss in half to make corrections.

There are many de-embedding schemes, here are just a few. The acronyms refer to the standards that are measured in order to create the "error boxes".

LRM: load-reflect-match

SOLT: short-open-load-through