How do I troubleshoot a high noise figure measurement that does not match my cascade analysis?

When a measured noise figure is higher than the cascade analysis prediction (Friis formula), the discrepancy is caused by contributions that were not included in the analysis or by measurement errors. The most common overlooked noise contributions are: input cable and connector losses before the LNA (any loss before the first amplifier adds directly to the system noise figure: a 0.5 dB cable loss adds approximately 0.5 dB to the noise figure), LNA noise figure under actual bias conditions (the datasheet NF is measured at optimal bias; actual bias may differ, especially if voltage drops across bias resistors are not accounted for), LNA input mismatch (mismatch between the source impedance and the LNA's optimal noise impedance increases the noise figure; the minimum noise figure is achieved only at the optimal source impedance), image noise in mixers (a mixer folds noise from the image frequency into the IF band, adding 3 dB of noise if no image-reject filter is present), and passive losses between stages (each loss between amplifier stages contributes additional noise). Common measurement errors include: incorrect noise source ENR (Excess Noise Ratio) calibration data, cold source temperature assumption errors (the actual physical temperature may not be exactly 290K), noise source mismatch with the DUT input, and instrument uncertainty (noise figure analyzers have their own uncertainty of 0.1-0.5 dB).