I use resistive impedance matching network and then recalculate the resulting impedance to get the "real value" on few frequency points. Not sure this is a correct way - nevertheless the results looks similar to the simulated ones. In my case it is more important, why simulation, where everything seems to be correct (see my first message), shows bad VSWR from the outputs. Nevertheless the results looks similar to the simulated ones.
Here it is more important and interesting why already in simulation, where everything seems to be correct (see my first message, I used s-param file for TC1.5-52 balun), VSWR from the outputs is bad.