A traditional noise parameter measurement setup, it includes a vector network analyzer (VNA) and a separate noise figure analyzer. For s-parameter measurements, the tuner is set to 50 ohms, and the two RF switches connect the device under test (DUT) to the VNA. For noise measurements, the switches connect the noise source to the DUT input and the noise receiver to the DUT output. An optional load tuner (not shown) is sometimes used when the DUT is highly reflective, to reduce sensitivity to error.
The tuner is pre-characterized at every frequency independently. This means that there is a unique set of tuner positions for each frequency, ensuring a good spread of source impedance points at every frequency. The tuner can be characterized separately, or as part of an in-situ system calibration. The advantage of doing it separately is that the same tuner file can be used for a long time, and then a hybrid in-situ calibration can quickly get the remaining s-parameter blocks.
The in-situ system calibration normally uses two VNA calibrations: a 2-port calibration at the DUT reference planes, and a 1-port s22 calibration at the noise source reference plane. By subtracting error terms, the 2-port x-parameters from the noise source to the DUT can then be determined. If the optional load tuner is used, then a 1-port s11 calibration at the noise receiver reference plane is also used to determine the 2-port s-parameters from the DUT to the noise receiver.
A hybrid in-situ calibration uses tuner data that is already characterized. The same VNA calibrations are still used to determine the 2-port s-parameters from the noise source to the DUT plane, which are then de-embedded to remove the tuner s-parameters. The result will be s-parameter blocks that include everything except the tuner.
After the system TRL calibration, the traditional noise receiver calibration and DUT noise parameter measurement are both done one frequency at a time. This is because the noise parameter extraction involves complex math that is sensitive to small errors, and the pattern of source impedance points is important to get well-conditioned data. Measuring one frequency at a time solves this by allowing the impedance pattern to be selected in an optimal manner for each frequency.