【Editorial Discussion】

On an RF-over-Fiber (RFoF) link, noise figure and SFDR often get chased by turning up laser power or EDFA gain, then adding RF gain at one end or the other. The first move is frequently wrong, because the dominant noise term changes with optical power, and the two ends of the link do not contribute equally.

1) Write down the noise budget before turning a knob. Relative intensity noise (RIN), shot noise, and photodiode + LNA thermal noise trade places as received optical power rises. Below a certain power, shot or thermal noise still dominates, so more optical power helps NF. Once you are RIN-limited, more optical power mainly buys you compression and fiber nonlinearities, not a quieter RF output.

2) Gain allocation is not symmetric. RF gain before the laser can bury the following noise, but it drives the laser toward clipping and chirp, and it can pull the optical modulation index into distortion. RF gain after the photodiode is easier to add, but it amplifies everything already on the photocurrent, including RIN and shot noise. If the photodiode+LNA NF is already the floor, pre-laser gain is the wrong medicine.

3) Measure SFDR and NF at the same reference plane. A “better” NF with a collapsed two-tone SFDR usually means you moved into laser clipping, amplifier compression, or fiber Brillouin / SBS effects. Record optical modulation index, received power, and the two-tone spacing with every NF number, or the next person cannot reproduce it.

4) Temperature and fiber length change the answer. The same gain plan that looks clean on a bench jumper can fail over a long plant: loss, dispersion, and reflection-induced RIN all move the crossover point. A lab NF figure without the plant loss and connector budget is not a field NF figure.

Open question: in your last RFoF NF problem, did raising optical power actually help, or did you only recover SFDR after cutting pre-laser RF gain? Any case where the link was already RIN-limited and more EDFA gain made compression worse? Please share the received power, OMI, and which term you think was dominant — no customer names needed.