How do I design a multi-octave LNA with flat noise figure across the entire bandwidth?

Designing a multi-octave LNA (covering 3:1 bandwidth or more, such as 2-18 GHz or 0.5-8 GHz) with flat noise figure across the entire bandwidth is challenging because the transistor's NF_min, Gamma_opt, and available gain all vary significantly over such a wide frequency range (NF_min typically increases 1-3 dB per octave, gain decreases 6 dB per octave). The design approaches are: resistive feedback topology (shunt feedback from drain to gate provides broadband input match and flat gain through negative feedback; the feedback resistor value is chosen to give the desired gain-bandwidth trade-off, typically R_f = 150-500 ohms; noise figure is 1.5-4 dB, increasing at higher frequencies), distributed amplifier topology (a traveling-wave architecture using multiple FET stages connected by artificial transmission lines; inherently wideband from near DC to f_max/2; noise figure of 2-4 dB, set by the first section's noise plus the loss of the input artificial transmission line), balanced amplifier (two identical amplifiers combined with wideband Lange couplers or Wilkinson dividers; provides excellent input/output match over multi-octave bandwidth; noise figure equals single amplifier NF plus coupler loss, approximately 0.3-0.5 dB penalty), and noise equalization (adding a frequency-dependent lossy network at the output that attenuates the lower frequencies where gain is highest, flattening both gain and effective noise figure across the band; this works because at lower frequencies the higher gain overcomes the added loss). The best multi-octave LNAs in GaAs or InP HEMT technology achieve 1.5-3 dB NF with +/- 0.5 dB flatness over 2-20 GHz.