N.F. vs R.L.
You have noticed a valid problem. At this low value of NF errors are large. Also, approximations used in deriving calcuations are no longer valid.
At these low values it is usual to switch to measuring and calulating noise temperatures since they add.
Another thing is if the NF is specified at some specific source impedance and what is the RL at that line/source impedance.
it is a math problem.
how can it be a system with mismath loss (0.5) at the begin with n.f. 0.3 dB ?
One situation would be where a quarter wave line is between your circuit and the generator. The line is used for impeance transformation. The 0.5 dB of signal reflected from your circuit back to the generator is reflected at the generator end and has a second chance of getting into your cirucit.
Another similar case is with a half wave line. It can be any impedance and the source and load will "see" each other at their actual impedance.
Follow the link below, it can help you!
http://www.mwrf.com/Globals/PlanetEE/Content/9595.pdf
it can be a single transistor