R
Member
CATV-style analog fiber links are sold with three big numbers (gain, NF, SFDR), but real deployments clip or floor out in ways the datasheet doesn't explain. Here's a working model for where the margin hides.
The three-line picture
An intensity-modulated direct-detection (IM-DD) link has:
- Signal line: output RF power ∝ (R·P_in·m·G)², m = modulation index;
- CSO/CTB / 3rd-order distortion: rises 2× (2nd-order) or 3× (3rd-order) per dB of signal — from laser chirp + fiber dispersion interacting under IM, and from RIN... careful: chirp-dispersion produces 2nd-order even-order distortion that is the killer for NLC channels;
- Noise floor: RIN-dominated (relative intensity noise, dB/√Hz), then receiver thermal, then shot term at high power.
Small-signal SFDR = (2/3)·(OIP3 − noise floor) — so a link with −150 dBm/√Hz RIN floor over 10 MHz noise BW (−80... let's be careful with numbers in your own links) gives you tens of dB of room; the practical ceiling is the 2nd-order limit: CSO beats land right on carriers unless you choose frequency plans to avoid them — in DVB/CATS plants this is why you never just "fill all channels".
What engineers actually hit first
1. Chirp–dispersion cancellation/null: at certain RF frequencies × fiber lengths, the even-order distortion nulls and the IM transfer dips 10-15 dB (classic 20 km G.652 null at ~3–4 GHz/1550). Your 6 GHz front-haul can be inside the null; move the RF band or add DSB/SSB cleverness.
2. RIN vs bias point: relaxing extinction-ratio-like operation (more DC bias, less index) buys RIN margin instantly until clipping. Check the laser's RIN vs I curve, not the headline.
3. MPIN thermal + responsivity drift at −40..+85 over the year: a 6 MHz "shift" of the link gain looks like interference but is just the PD's old age.
4. Reflections from connector faces (bad APC polish!) → multiple-path interference → ripple in the OIP3 itself, not just in gain.
A sanity workflow for new links
- Step 1: measure RIN on the bench at planned bias, note −dB/√Hz; define noise BW (per carrier);
- Step 2: sweep RF at fixed m and find the 2nd-order products vs channel plan; set max operating level 10 dB below the first beating CSO product crossing signal;
- Step 3: fiber-length sweep for chirp nulls before committing frequency plans per route;
- Step 4: temp-cycle the whole chain, log gain vs T, design link budget from the worst corner not the 25 °C one.
RoF people of the forum: are you measuring OIP3 vs fiber length routinely, or trusting vendor curves? Curious where reality departs most.
The three-line picture
An intensity-modulated direct-detection (IM-DD) link has:
- Signal line: output RF power ∝ (R·P_in·m·G)², m = modulation index;
- CSO/CTB / 3rd-order distortion: rises 2× (2nd-order) or 3× (3rd-order) per dB of signal — from laser chirp + fiber dispersion interacting under IM, and from RIN... careful: chirp-dispersion produces 2nd-order even-order distortion that is the killer for NLC channels;
- Noise floor: RIN-dominated (relative intensity noise, dB/√Hz), then receiver thermal, then shot term at high power.
Small-signal SFDR = (2/3)·(OIP3 − noise floor) — so a link with −150 dBm/√Hz RIN floor over 10 MHz noise BW (−80... let's be careful with numbers in your own links) gives you tens of dB of room; the practical ceiling is the 2nd-order limit: CSO beats land right on carriers unless you choose frequency plans to avoid them — in DVB/CATS plants this is why you never just "fill all channels".
What engineers actually hit first
1. Chirp–dispersion cancellation/null: at certain RF frequencies × fiber lengths, the even-order distortion nulls and the IM transfer dips 10-15 dB (classic 20 km G.652 null at ~3–4 GHz/1550). Your 6 GHz front-haul can be inside the null; move the RF band or add DSB/SSB cleverness.
2. RIN vs bias point: relaxing extinction-ratio-like operation (more DC bias, less index) buys RIN margin instantly until clipping. Check the laser's RIN vs I curve, not the headline.
3. MPIN thermal + responsivity drift at −40..+85 over the year: a 6 MHz "shift" of the link gain looks like interference but is just the PD's old age.
4. Reflections from connector faces (bad APC polish!) → multiple-path interference → ripple in the OIP3 itself, not just in gain.
A sanity workflow for new links
- Step 1: measure RIN on the bench at planned bias, note −dB/√Hz; define noise BW (per carrier);
- Step 2: sweep RF at fixed m and find the 2nd-order products vs channel plan; set max operating level 10 dB below the first beating CSO product crossing signal;
- Step 3: fiber-length sweep for chirp nulls before committing frequency plans per route;
- Step 4: temp-cycle the whole chain, log gain vs T, design link budget from the worst corner not the 25 °C one.
RoF people of the forum: are you measuring OIP3 vs fiber length routinely, or trusting vendor curves? Curious where reality departs most.
