Thank you for your answers. Frequencies are 400-2400MHz. It seems the problem stems from my measurements. My cheap VNA does not allow for automatic port extension to substract the measurement cable impact so is running the open/short/load calibration with the measurement cable connected to the VNA port on one side and soldered to the PCB matching network start on the other side the right way to do it ? Thank you !

It's not real clear what you are saying. You have a cable soldered to the impedance you are trying to measure? Then you are calibrating at the cable connector?

In that case you have to de-embed the cable from the measurement. Assuming that you have a well matched cable, you have to know the insertion loss and phase shift of the cable (or S21 of the cable). Divide your measurement by (S21)^2. Note that you are dividing two vectors, so divide the magnitudes and subtract the angles.

So it's important that the cable is well matched and as little loss and phase shift as possible. If the phase shift of the cable is more than 180 degrees your math gets a little more complicated. (The phase shift of the cable goes from zero degrees at zero frequency and rotates clockwise. To de-embed it, you have to rotate counter clockwise).

How do you know S21 of the cable? That's another problem. One way would be to measure a cable that is twice as long, then cut it in half.