RFoF link design: the three numbers that actually set your dynamic range

Started by fiberben · 2026-11-17 14:38 UTC · 18 views · 2 replies
F Member
Most RFoF links I get asked to debug are not broken. They were designed around the only number everybody quotes in a datasheet headline: gain. Gain tells you where the link sits, not how much signal it can carry before it lies to you. Three numbers decide whether the link survives contact with a real antenna: equivalent input noise (EIN), spurious-free dynamic range (SFDR), and the compression point that sets where SFDR stops being meaningful. Everything below is written for the common case of a 1310 nm DFB with direct modulation, or a 1550 nm external modulator with an EDFA in front of a PIN detector.

1) Optical loss is not RF loss: it is twice RF loss

An analog optical link is square-law at the detector. Optical power arriving at the photodiode is proportional to the square of the modulated optical field, so a 3 dB drop in received optical power costs about 6 dB of RF power. People routinely budget a 10 km span at 3.5 dB (0.35 dB/km at 1310 nm), add 0.6 dB for two connector pairs, call it 4.1 dB, and then are surprised when the measured RF gain is 8 dB low. If your spreadsheet is in optical dB, convert before it meets an RF cascade calculation: RF_delta = 2 x optical_delta for small signals. Do this before you add the post-detector amplifier, not after. Amplifier gain cannot recover signal-to-noise, it only re-scales the noise you already have.

2) EIN, not noise figure, is the number that survives a long fiber run

The detector plus TIA sets a thermal floor, the laser RIN and shot noise ride on top of it, and the fiber adds nothing but loss. For a well-behaved link the limiting terms are RIN and the received optical power. A DFB with -150 dB/Hz RIN at 10 mW and a link with 6 dB optical loss gives an EIN around -125 to -130 dBm/Hz in a 50 ohm system; the same transmitter at 1 mW sits roughly 10 dB worse. This is why "the module works on a 2 km lab spool" and "the module is marginal on a 15 km tower run" are the same module: nothing in the RF chain changed, only the optical loss term.

Practical rule: put the EIN on the link budget at the input, treat the whole optical span as an attenuator with 2x electrical slope, then cascade the post-detector LNA. If the post-amp has 2 dB NF and 20 dB gain, the system NF after the link is dominated by whatever the link presents, not by the amplifier. Choosing a lower-noise amplifier is usually a waste of money; buying 6 dB of optical power or 6 dB less fiber loss is not.

3) SFDR: where the link actually earns its keep

SFDR is a two-tone measurement, and it is quoted in dB·Hz^(2/3). For an externally modulated link biased at quadrature, values around 105-112 dB·Hz^(2/3) are normal for a single octave, limited by the modulator's third-order behaviour. Directly modulated DFBs are worse, typically 95-105 dB·Hz^(2/3) depending on bias and frequency, because relaxation oscillation and chirp add both distortion and phase-to-intensity conversion over the fiber. If your application is a receive-only antenna remoting link feeding a spectrum monitor, this hardly matters. If you are upconverting a multichannel QAM or OFDM signal with high PAPR, it matters more than gain.

The trap is bias drift. An MZM parked at quadrature with a simple bias tee will wander a few degrees over a 0-70 C swing, and the second-order term grows fast as it leaves quadrature. A link that measured 110 at 25 C can read 90 when the cabinet is hot. If the datasheet does not state a bias control scheme, ask. The answer tells you whether the module is for a lab bench or a tower.

4) A bench sequence that predicts field failures

Four measurements, about an hour, on any candidate module:

a. Laser off, RF in: the noise floor of the detector and post-amp chain. If this is not at least 10 dB below your expected EIN, your measurement rig is lying to you.

b. Gain vs frequency at three optical lengths: 0 km, 10 km, 20 km. The slope should follow 2x the optical loss, no more. A steeper slope usually means connector contamination or a failing launch condition, not physics.

c. 1 dB compression sweep at the highest expected input, at 70 C as well as 25 C. Compression often degrades first with temperature because the laser slope efficiency drops.

d. Two-tone at 1 MHz spacing, then at 10 MHz spacing. If the IM3 changes materially with spacing, you are seeing chromatic-dispersion-induced distortion, and at 1550 nm that means you need to shorten the span or add dispersion compensation.

5) Selection shortlist by reach

Under 2 km: direct-modulated 1310 nm is the cheap and honest answer; link budget is easy, distortion is manageable.
2-20 km: 1310 nm direct modulation still works for narrowband signals, but for wideband or high-SFDR work move to 1550 nm external modulation, and add an EDFA only when the loss budget genuinely needs it. An EDFA buys optical power and adds its own noise figure, roughly 5-6 dB; it does not improve SFDR.
Beyond 20 km at 1550 nm: dispersion becomes the design constraint, so plan for a compensated or dispersion-shifted path, or accept a narrower modulation bandwidth.

6) One note on sourcing

Disclosure: I work with F-tone Networks, who make RFoF modules and industrial optical transceivers. That does not change the arithmetic above, and I would rather people run the four measurements on our parts and on anyone else's than trust a headline gain figure. If you want the module-level numbers to compare against your own budget, the RF module line is at f-tone.com and the RFoF category index there lists bandwidth, wavelength and connector options side by side.

If you have a link that measures fine on the bench and misbehaves in the field, post the gain, EIN and the optical length and I will try to say which of the three numbers is actually moving.
L Admin
Thank you, fiberben — putting EIN, SFDR and the compression point next to gain, and treating optical loss as twice the RF loss in the cascade, is the habit we want on every "the module worked on a 2 km spool and failed on the tower" ticket.

Editorial keepers (soft promo = 0):
1. Laser-off RF noise floor at least 10 dB below the expected EIN before you trust a field number. A post-detector amplifier does not recover SNR; it only rescales the noise the link already presented.
2. Gain versus 0 / 10 / 20 km should follow 2× the optical loss. A steeper slope is usually launch or connector condition, not a new physics term. Archive received optical power (or photocurrent) on the same line as EIN and SFDR.
3. Two-tone at 1 MHz and at 10 MHz spacing: if IM3 moves with spacing, treat dispersion (especially at 1550 nm) before a modulator ECO. Bias-control scheme belongs on the SFDR traveler — a quadrature wander over temperature is not a laser lot.

Open: on your last bench-good / field-bad analog link, which of the three numbers actually moved — EIN with optical loss, SFDR with bias or tone spacing, or the compression point — and do you refuse to sign off a headline gain without them?

— rfforum 编辑部 / Editorial Team
H Member
Have you tried looking at it from the signal integrity angle? That's how we found our root cause.
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.