How do I estimate the noise contribution of a printed circuit board trace at millimeter wave frequencies?

Estimating the noise contribution of a PCB trace at millimeter-wave frequencies (30-100+ GHz) is critical because the trace loss increases significantly at these frequencies and every lossy element before the LNA adds directly to the system noise temperature. A lossy passive element at physical temperature T with loss L (in linear, L > 1) contributes a noise temperature of T_noise = T x (1 - 1/L) to the system. At millimeter-wave frequencies, PCB trace losses include: conductor loss (increases as sqrt(f) due to skin effect; at 77 GHz, skin depth in copper is only 0.24 micrometers, and surface roughness of standard PCB copper (1-3 um RMS) significantly exceeds the skin depth, increasing the effective resistance by 50-200% beyond the smooth-conductor approximation), dielectric loss (proportional to frequency and the substrate's loss tangent; at 77 GHz, standard FR4 has tan_d approximately 0.02, giving approximately 3 dB/cm loss, making it unusable; Rogers RO3003 has tan_d approximately 0.001, giving approximately 0.3 dB/cm; Megtron 7 has tan_d approximately 0.002 at 0.5 dB/cm), and radiation loss (microstrip lines radiate at discontinuities and bends at mmW, especially on thick substrates). For a 10 mm microstrip trace at 77 GHz on Rogers RO3003 with 0.5 dB total loss (L = 1.12), the noise contribution at 290 K is T_noise = 290 x (1 - 1/1.12) = 31 K. This is significant and can degrade the system noise figure by 0.4 dB if placed before the LNA.