SFDR and CNR pitfalls in analog RoF: what the datasheet hides on a real fiber run

[Technical note — Fiberaw / F-tone]

Analog RoF SFDR and CNR numbers from a datasheet are usually measured on a short, clean patch. Field fiber changes the story. Checklist we run before arguing with a vendor or a customer about "the link is noisy".

**1. Define the metric before the plot**
- CNR: carrier vs noise in a stated RBW / noise bandwidth — write both.
- SFDR: usually two-tone, spur-free dynamic range to third-order products — state tone spacing and whether you mean OIP3-limited or noise-floor-limited SFDR.
- Mixing CNR language with SFDR language in the same email is how programmes stall.

**2. Datasheet traps**
- SFDR quoted at one optical power / one frequency / one fiber length.
- RIN assumed white; real DFB RIN + reflections + driver noise are not.
- PD / TIA NF quoted without the actual photocurrent you will run.
- No statement of MZM bias point, Vπ, or whether bias-lock was on.

**3. Fiber-run killers (often missed)**
- Connector dirt / APC vs UPC mix → reflections raise effective RIN and IMD.
- Chromatic-dispersion RF fading: nulls look like "CNR collapse at one band edge".
- EDFA ASE that looks fine on an optical power meter but eats CNR.
- PD saturation / clipping: RF still present, SFDR already dead.
- Multi-carrier crest factor: two-tone SFDR ≠ LTE/5G waveform EVM.

**4. Bench sequence that separates causes**
1. Short patch baseline: CNR + two-tone SFDR at nominal Popt.
2. Calibrated optical attenuator sweep: RF gain slope (~2 dB RF / 1 dB optical) and where SFDR peaks.
3. Insert the real fiber (or spool of same length/type); re-measure; note frequency-dependent dips.
4. Only then touch bias, EDFA gain, or RF drive — one knob at a time.

**5. Habit**
File three screenshots with every RoF bring-up: short-patch SFDR, attenuator sweep, real-fiber CNR vs frequency. Without them, every later "noise" ticket becomes archaeology.

Open question: do you gate analog RoF on two-tone SFDR, on modulated EVM/ACLR, or both — and which one fails first on long SMF?

— Fiberaw / F-tone

I gate on both, but I give them different jobs, because they fail for different reasons and a single gate hides that distinction. Two-tone SFDR is my linearity diagnostic: it tells me whether third-order products, bias drift, or photodiode saturation are eating the usable range at a known tone spacing and optical power. Modulated EVM and ACLR are my system gate, because they are what the actual waveform has to live with, including crest factor, filtering, and frequency-selective effects that two tones never see.
On a long single-mode fiber run, the modulated measurement usually fails first in my experience. Chromatic-dispersion fading does not degrade every frequency equally, so a two-tone test at a favourable centre frequency can still look clean while the modulated signal loses carrier-to-noise at a band edge or suffers a frequency-dependent amplitude and phase tilt across its occupied bandwidth. Added loss, amplified spontaneous emission, and reflections also raise the noise floor, which directly costs EVM even when the third-order intercept has not moved much.
That does not make SFDR redundant. When EVM fails, I go back to the short-patch baseline and the attenuator sweep you describe, and I use SFDR to separate a linearity problem from a noise or fading problem. If SFDR is still close to baseline while EVM has degraded on the real fiber, I look at dispersion, connector reflections, and noise first. If SFDR has collapsed as well, I look at bias, drive level, and saturation before touching the fiber. Keeping both measurements, with tone spacing, bandwidth, optical power, and photocurrent written next to each result, is what stops a good short-patch number from being used to excuse a failing field link.

Marking this - our project starts something similar next month, learning from everyone's experience.