I am measuring the port extension using active marker method from 10MHz up to 20GHz using Network Analyzer (PNA-X). Why the S11 & S22 shape in the smith chart format is different when I put the active marker to 10GHz and 5GHz?
I wonder how do you measure the port extension. After calibration did you connect the cables to your fixture and measure the port extension? Or you measure port extension on open cable. In first case it might indicate that the transmission lines are not the same length. In second case it might be that the ending of the cable do not much the true characteristic impedance of the cable. That happen with cables that have connectors with particular gender: for example open SMA male connector do not match 50 ohms unless you connect the female (for example cal kit open). This is because distance of the center pin of SMA male connector to outer ground do not match the 50 ohms characteristic impedance. The SMA female do match 50 ohms characteristic impedance as the distance between outer center tube and ground is specific for the particular characteristic impedance. Some time instead of cal kit open standard you can use SMA female to female adapter and it serve similar purpose to check the system as the cal kit female open.
If that not the case it mean that you cable might be damaged, or the cal kit standard are damaged and become intermittent during calibration.
Let me explain my thinking (I developed the feature, so sorry it's not clear):
Normally a fixture will be broadband, and so taking the nominal group delay across the current frequency range is a good estimate of the delay. But if a fixture has frequency dependency (for example if it rolls-off a lot at high frequency) then it won't act like a normal transmission line, and at low frequency (where you are likely interested in the data) the port extension will be wrong. OR: if you have a band limited fixture (maybe due to some biasing structure you are using) then the low frequency data may be corrupted by the bias network (thinking here about a stub-line bias structure). In either case, you don't want to use the full frequency range to compute the electrical delay of the fixture, but rather we added a capability to place a marker at a good spot on the trace and compute the delay from the region near the marker (in fact it executes the same function as marker-to-delay). So, if your delay is different at different frequencies, it means your fixture is not like a transmission line, and a single delay number won't work over whole range to remove the fixture effects. So chose the region where you are most interested in.
Retrieving data ...