RoF link-budget field checklist: optical dB, RF dB (square-law), and the five numbers that catch a bad patch cord

[Technical note — Fiberaw / F-tone]

A RoF "link budget" that only lists optical dBm will surprise you on RF day-one. Photodetection is square-law in the linear region: **~2 dB RF for every 1 dB optical** until saturation or AGC breaks the slope. Field checklist:

**Five numbers to write down (before blaming the laser)**
1. Optical power into the PD (dBm) — meter after the last connector, not at the TX faceplate alone.
2. RF power out of the PD/TIA at a known CW tone (dBm).
3. Expected RF change if you pad 1.0 dB optical (should be ~2.0 dB RF if still linear).
4. Noise floor / CNR in a stated RBW at that same tone.
5. Two-tone IMD or a quick SFDR snapshot at the same Popt.

**Patch-cord / connector red flags**
- Optical power "OK" after a re-mate, RF down ~2–4 dB → classic dirty/mismatched connector (square-law bite).
- RF OK, CNR down → look for ASE, RIN/reflections, or RBW/attenuator settings — not just loss.
- Frequency-selective RF dip with flat optical power → dispersion fading or standing-wave / bias issue, not a simple pad.

**Quick linear-region test (10 minutes)**
Sweep a calibrated VOA in 0.5 dB optical steps over ±2 dB around nominal. Plot RF dB vs optical dB. Slope ≈ 2 → linear. Slope collapses → PD/MZM compression or AGC. Slope >> 2 or chaotic → measurement setup or intermittent connector.

**Do not**
- Treat optical margin and RF margin as 1:1.
- Sign off on optical power alone after a cable change.
- Compare CNR across different RBWs without converting.

Discussion: what optical-to-RF slope do you actually measure on your PD at working photocurrent — and at what Popt does the slope break?

— Fiberaw / F-tone

This checklist captures the trap well, because the square-law step is easy to forget once optical power looks comfortable. At a working photocurrent in the linear region I expect the slope you describe: about 2 dB of RF change for every 1 dB of optical change, measured with a calibrated attenuator sweep rather than inferred from a single power reading. The key is to plot RF power against both optical power and photocurrent, so a change in responsivity or bias is not hidden behind a tidy optical number.
Where I see the slope break first is usually not the laser at all. As photocurrent rises, the photodiode can move toward saturation, the following amplifier can compress, or an automatic gain stage can flatten the response, and each of those shows up as a slope falling below 2. A slope above 2, or one that jumps between steps, is a different warning: it often points to a drifting connector, a bias shift, or a measurement setup that changed between steps rather than to a genuine device improvement.
Two habits make the ten-minute test more trustworthy. First, repeat the sweep after re-mating the patch cord, because a dirty or slightly misaligned connector can cost 1 dB optically and therefore about 2 dB in RF, exactly as your red-flag list notes. Second, record the noise and linearity numbers at the same photocurrent as the slope, not at a more convenient level. A link that is linear at low photocurrent can still fail on carrier-to-noise or intermodulation at the working level, and a link that looks noisy at one resolution bandwidth may simply need its bandwidth stated before it is compared with anything else.

Question: how sensitive is this to temperature? Any thermal test data?