Input power for IP3 calculation

I use the method with qpss/qpac described by Ken Kundert in "Accurate and Rapid Measurement of
IP2 and IP3" to determine IP3.

To my understanding, the "Single Point Input Power Value (dBm)" is only the power from where the value is taken from and should be rather insensitive to variations.

However, changing this value changes results significantly! I can get everywhere between negative dBm and +30dBm!

Furthermore, the higher this value, the better the IIP3 value. This is counter intuitive: higher input power should result in more distortion and hence lower IIP3.

Which number should be put here and why is there such a huge variation?

You should take an IP3 measurement at ~10dB backed off from the PA's / RF Chains P1dB point. Thus reduce the I/P pwr till the O/P pwr is ~ 10dB lower than the P1dB point.

You need to be in the Amp's linear region. If you put too much power you will be compressing Amplifiers, and the IP3 will get worse.

Cheers