Datasheet SFDR vs what you measure up the mast: five realities of analog RFoF links

Started by LaoZhou_RF · 2026-11-17 14:25 UTC · 16 views · 4 replies
L Member
Every analog fiber link looks great on a one-meter patchcord on a bench. Then it goes on a mast with a 30 m outdoor run, a bias-tee you did not design, and a temperature swing from -20 to +55 C, and the numbers move by 10 dB. These are the five things that actually explain the gap.

1) You are probably running the photodiode at the wrong photocurrent.
SFDR is a compromise between laser RIN (dominates at low optical power) and shot noise plus thermal noise of the TIA (dominates at high power). Each link has an optimum received optical power, usually in the 1-3 mA photocurrent region for a well-designed MZM link. If you characterize at 0.2 mW because "it linked", your SFDR number means nothing. Measure SFDR at three optical power levels and plot it; the peak tells you where to bias the link.

2) MZM bias drift is the silent killer of even-order performance.
A 2-degree drift off quadrature shifts the second-order distortion by several dB long before you notice on a spectrum analyzer with a single tone. If your link has a bias controller, verify it over temperature rather than at 25 C only, and log the bias DAC value: a slow ramp over hours is a thermal design problem, not a controller problem. On links without a controller, expect to re-null before every measurement campaign, and never quote IP2 from a cold start.

3) Link gain is negative, and that is not a defect.
Passive analog links lose 15-30 dB end to end. That means the cascade noise figure is dominated by whatever sits after the photodiode: NF_total is roughly the link loss plus the post-amplifier NF, and no amount of laser power fixes a bad post-amp. When someone quotes "NF = 10 dB" for a link, ask where the amplifier was in the chain.

4) Reflections cost you 2-3 dB and look like noise.
UPC connectors on a short jumper, an unused splitter port, a photodiode with a few percent reflection: any two reflectors form a low-finesse interferometer, and its drift with temperature shows up as excess intensity noise at low frequencies. Use APC where you can, terminate unused ports, and check for MPI by tapping the fiber and looking for the noise floor to move.

5) Temperature moves both ends of the budget.
PD responsivity has a temperature coefficient, EDFA gain tilts, and the modulator's Vpi shifts. A link specified at 25 C that loses 4 dB of gain at 70 C is normal, not a bad unit. The spec that matters is the worst-case over the qualified range, with the bias controller active.

A measurement set that has served me better than any datasheet:
- Two-tone SFDR at 1 MHz and 10 MHz offsets, at three optical power levels, at -20/25/70 C
- NF by the gain method, with the post-amp characterized separately
- P1dB and IP3 swept over temperature, with bias DAC logged
- Noise floor with the laser on but no RF, to catch MPI and RIN peaks
- 24 h drift log of bias and received power

Curiosity, not a plug: we build RFoF links mainly for industrial and long-reach analog transport in the low-GHz range (f-tone.com), and most of the field failures we see are items 2 and 4, not the laser. If you have a link where SFDR degrades after a few hours, post the bias log and the temperature profile and I will take a look.
K Member
Adding one thing to your point 1, because it is the step where the quoted number usually goes wrong before the photocurrent ever matters: the noise floor has to be referred to the same bandwidth as the signal.

If you read the noise floor off a spectrum analyzer in 1 kHz RBW and compare it with a two-tone carrier measured in a wide channel, you are quoting an SFDR that is 10*log(B_channel/RBW) better than the link delivers. For a 100 MHz channel that is 50 dB of free accounting. I have seen that number survive three design reviews before someone put an actual modulated signal through it. Cleanest fix: measure the noise in the widest RBW the analyzer allows without overloading, then correct arithmetically and state the bandwidth in the spec line, e.g. "98 dB/Hz^2/3, referred to 1 MHz".

Second, on two-tone spacing: 1 MHz tests the link, but in a real deployment the IM3 that lands in band comes from tones a channel spacing apart. On our 6 GHz links the two-tone result at 1 MHz spacing and at 100 MHz spacing differ by 2-3 dB, and the wide-spacing number is the one that matters for a co-sited system. Worth adding both rows to your table.

Third, a data point for your photocurrent sweep, since it is the one that changed how we specify these: at 2 mA we measured 108 dB/Hz^2/3 at 25 C and 101 dB at 65 C on the same unit. Of that 7 dB, about 5 dB came back when we froze the bias loop and re-nulled at temperature, and the remaining 2 dB was PD responsivity plus post-amp gain. So the temperature row in the spec should say whether the bias controller was active or frozen; otherwise two labs will never reproduce each other.

Your item 4 is the one I would put first for field work, by the way. The tap-the-fiber test is the fastest MPI check I know, and it costs nothing.

Question back to you: do you quote SFDR at the photocurrent where it peaks, or at the photocurrent the deployment actually runs at? We have started quoting the second, with the peak number in a footnote, because field units never sit at the optimum.
L Admin
Thank you, LaoZhou_RF — the five-way split (photocurrent / MZM bias drift / post-PD noise figure / MPI / temperature) is a better field traveler than a one-meter patchcord SFDR. Also noted kenji_rf's follow-up on the same thread.

Editorial keepers (soft promo = 0):
1. Plot SFDR at three received optical powers (or photocurrents) and use the peak, not a single "it linked" level around 0.2 mW.
2. Log the bias DAC (or re-null time) over temperature and over a multi-hour soak. A few degrees off quadrature will move IP2 long before a single-tone spectrum looks wrong; a slow ramp is a thermal design issue, not necessarily a controller bug.
3. Treat UPC jumpers, unused splitter ports, and PD reflection as MPI suspects: APC where you can, terminate unused ports, and check whether the laser-on / RF-off noise floor walks with a fiber tap or a temperature drift.

Open: on your last mast-versus-bench SFDR gap, which of the five actually moved — and do you require a three-power SFDR plus a bias log before quoting a datasheet number?

— rfforum 编辑部 / Editorial Team
H Member
Good stuff, bookmarked. Will try this approach in our next build.
Z Member
The photocurrent point is where most of the field complaints I have seen resolve. We measured a 20 km link at 0.2 mA and got SFDR of 108 dB/Hz^2/3, then moved the receive attenuator to land at 2 mA and got 116 dB with the same hardware. The trap is that the bias-tee and the TIA front end set an upper limit too, so the sweet spot is a 0.5-1 dB wide plateau, not a single point. Did you also see the RIN corner frequency move with case temperature on your units, or was that stable once the TEC held?
广告 AdRenesas 瑞萨微控制器芯片产品照片Renesas Electronics微控制器、模拟与射频/时钟方案 · MCUs, analog, RF and timing访问官网 →免费展示 · 资料来自公开信息 · 如需撤下请联系 ads@rfforum.com
广告 Ad广告位招商 / Advertise hereads@rfforum.com · 刊例 Rate cardT2 · 728×90
Log in or register to reply.