Ekinops Dark Fiber Learning Path
Engineering

Link Engineering — will this path close, and how far?

A loss budget answers "does enough power arrive?" But power is only half the story. A signal can arrive strong and still be unreadable because it is buried in noise. This page is about the second budget — the OSNR budget — and how it decides whether a wavelength will actually carry error-free traffic, and for how many spans.

The two budgets, side by side

The loss (power) budget asks: after all the fiber and connector loss, is the received power above the receiver's minimum? The OSNR budget asks: after all the amplifiers have added their noise, is the signal-to-noise ratio still above what this modulation and FEC need to decode cleanly? A link must pass both. A short link usually dies on power; a long amplified link usually dies on OSNR.

What OSNR actually measures

OSNR is the Optical Signal-to-Noise Ratio — the ratio, in decibels, of the wanted signal power to the background optical noise power, measured in a standard 0.1 nm reference bandwidth. Higher is better. The reference bandwidth matters: everyone measures noise over the same 0.1 nm window so numbers are comparable between instruments and vendors.

Mental model

Think of OSNR as how far the signal "sticks up" above the noise floor. Every amplifier you pass through raises the noise floor a little, so the signal sticks up a little less. Loss lowers the whole signal; amplifiers restore the height of the signal but also raise the floor. Once the floor climbs close enough to the signal, the receiver can no longer tell ones from zeros reliably — and no amount of extra amplification helps, because it lifts the floor too.

Where the noise comes from — ASE

The dominant noise source on an amplified line system is ASE — Amplified Spontaneous Emission. Every optical amplifier (an EDFA, for example) spontaneously emits a little broadband light of its own, amplifies it alongside your signal, and hands the sum to the next span. That added noise never goes away — the next amplifier amplifies your signal and the accumulated noise, then adds its own on top. Noise accumulates; it is not reset by amplification. Only a full 3R regenerator (a decode-and-retransmit) resets OSNR to like-new.

ASE ACCUMULATION across a chain of identical amplified spans Tx ─►[BST]══span══[EDFA]══span══[EDFA]══span══[EDFA]══► Rx │ │ │ │ noise floor rises a step at EACH amplifier and is carried forward: OSNR: high ───────► lower ─────► lower ─────► lower ─────► floor (span 1) (span 2) (span 4) (span 8) signal │▁▁▁█▁▁▁ │▁▁▂█▂▁▁ │▁▂▃█▃▂▁ │▂▃▄█▄▃▂ the floor over │ │ │ │ keeps rising noise └noise floor └ a bit up └ higher └ near signal → errors Amplifiers restore POWER every span, but the accumulated ASE noise rides along. Reach ends when OSNR falls to the receiver's requirement.
Rule of thumb — doubling spans costs ~3 dB of OSNR

For a chain of identical amplified spans, the accumulated OSNR falls by 10·log₁₀(N) for N spans. Since 10·log₁₀(2) ≈ 3 dB, every time you double the number of spans you lose about 3 dB of OSNR. Go from 1 span to 2 → −3 dB; 2 to 4 → another −3 dB; 4 to 8 → another −3 dB. This is why reach does not scale linearly: the tenth span hurts far less than the second one, but the total keeps eroding.

Estimating OSNR per amplifier

A widely used engineering rule of thumb for the OSNR contributed by a single amplified span is:

OSNR(dB) ≈ 58 + Pin(dBm) − NF(dB) − span_loss(dB) 58 a constant folding in the 0.1 nm reference bandwidth and fundamental photon/quantum noise terms (C-band) Pin per-channel power LAUNCHED into the span (dBm) NF amplifier noise figure (dB) — how much noise it adds (~4–6 dB EDFA) span_loss the fiber+connector loss of that span (dB) For N identical spans, subtract 10·log10(N): OSNR_total(dB) ≈ [ 58 + Pin − NF − span_loss ] − 10·log10(N)

Read it as a lever board. Launch a little hotter (Pin up) → better OSNR — but only up to the point where fiber nonlinearity bites (covered on the Amplification page). Lower-noise amplifiers (NF down) → better OSNR. Shorter or lower-loss spans (span_loss down) → better OSNR. More spans (N up) → worse OSNR. Every design decision moves one of these terms.

Treat these numbers as illustrative

The constant 58, the doubling rule, and the reach figures below are engineering approximations for teaching. Real designs use the vendor's planning tool with the exact amplifier, fiber, and channel data. Use these to understand which way each lever moves — not to sign off a live circuit. Always verify against Ekinops docs / BOM and the actual line-system planner.

Required OSNR vs actual OSNR — and margin

Every modulation format and FEC combination needs a certain minimum OSNR to decode error-free. Slower, simpler signals need less; faster, denser ones need more. Illustrative required-OSNR values:

Signal (illustrative)Roughly needs OSNRWhy
10G OOK (on/off keying, direct-detect)~15–20 dBSimple format, older FEC — needs a clean signal
100G DP-QPSK with SD-FEC (coherent)~12–15 dBStrong soft-decision FEC tolerates far more noise
Higher-order coherent (16QAM, 64QAM)Progressively moreDenser constellations pack more bits but sit closer together — less noise tolerance

The link is engineered so that actual OSNR ≥ required OSNR + margins. You do not design to the exact edge — you stack margin on top:

OSNR MARGIN STACK (design so the actual sits comfortably above) actual OSNR ┃██████████████████████████████ e.g. 20.0 dB (what you have) ┃ aging margin ┃▒▒▒▒ ~1–2 dB (fiber/amp age, dirt, ┃ repairs, drift over life) design margin┃▒▒▒▒▒▒ ~2–3 dB (planning uncertainty, ┃ temperature, tolerances) required ┃████████████████ e.g. 14.0 dB (format+FEC floor) OSNR floor ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ System OSNR margin = actual − required = 6 dB, of which ~4–5 dB is deliberately reserved for aging + design. If actual dips to required, BER climbs and the link starts erroring. Never plan to zero margin.

Q-factor and BER — the quality-to-errors bridge

OSNR predicts noise; what you ultimately care about is the error rate. The bridge between them is the Q-factor. Q measures how cleanly the "1" level separates from the "0" level relative to their noise. It is quoted either as a linear number or in decibels: Q(dB) = 20·log₁₀(Q_linear). Higher Q → fewer errors.

Q maps to BER (Bit Error Ratio) through the complementary error function:

BER ≈ ½ · erfc( Q_linear / √2 ) erfc() is the "complementary error function" — it just answers "what fraction of a bell-curve of noise spills past the decision threshold?" You do not compute it by hand; you read the mapping: higher Q → noise rarely crosses the threshold → lower BER lower Q → noise often crosses the threshold → higher BER
Q (linear)Q (dB)Approx BERMeaning
~6.0~15.6 dB~1×10⁻¹²Classic pre-FEC target for an uncoded system — "error free" by legacy convention
~5.3~14.5 dB~1×10⁻⁹Marginal for uncoded; needs FEC
~4.7~13.4 dB~1×10⁻⁷Would be a bad link uncoded — but fine as a pre-FEC input to strong FEC
lowerlower~2×10⁻² (pre-FEC)SD-FEC can still recover this to error-free post-FEC

Q ≈ 15.6 dB ↔ BER ≈ 1×10⁻¹² is the number to memorize: it is the classic "error-free" pre-FEC target for older uncoded systems.

Pre-FEC vs post-FEC — the number that matters

Mental model

FEC (Forward Error Correction) adds redundant bits so the receiver can repair a certain number of errors without asking for a retransmit. So there are two BERs: the pre-FEC BER (errors arriving off the fiber, before correction) and the post-FEC BER (errors that survive after correction — these are what actually reach the customer). A healthy modern link can run a terrible pre-FEC BER and still deliver zero post-FEC errors. FEC is doing exactly its job.

Operational rule

Pre-FEC BER is your early-warning gauge; post-FEC BER is the alarm. Watch pre-FEC trend upward over days and you can see a link aging toward its FEC limit before customers feel anything. The moment post-FEC errors appear, the link has run out of correction headroom and traffic is being hit — that is customer-affecting and demands action now.

How FEC buys OSNR and reach

Strong FEC lets a link tolerate far more noise before post-FEC errors appear. That tolerance is called coding gain — effectively "free" OSNR. Soft-Decision FEC (SD-FEC) delivers roughly 11 dB+ of coding gain and can run at a pre-FEC BER as high as ~2×10⁻² (2% of bits arriving wrong!) yet output error-free traffic. Older Hard-Decision FEC (HD-FEC) gives roughly ~6 dB. That difference in coding gain converts directly into either more spans or a denser modulation format.

WHAT FEC BUYS YOU Required OSNR to hit error-free: Uncoded (Q=15.6 dB target) ┃██████████████████████ needs the most OSNR HD-FEC (~6 dB gain) ┃██████████████ ~6 dB relief SD-FEC (~11 dB+ gain) ┃████████ ~11 dB+ relief → far more reach Every dB of coding gain lowers the required OSNR by ~1 dB. And since doubling spans costs ~3 dB, ~11 dB of SD-FEC gain buys roughly 2^(11/3) ≈ 12x the number of spans versus uncoded — the single biggest reason modern coherent links go so far.

Worked reach estimate (illustrative)

Let's walk a full estimate. Given values (all illustrative — not a real Ekinops spec):

  • Per-span loss: 22 dB (about 80 km of standard fiber plus connectors)
  • Amplifier noise figure NF: 5 dB
  • Per-channel launch power Pin: +1 dBm
  • Required OSNR for the transponder (100G DP-QPSK, SD-FEC): 14 dB
  • Reserved margin (design + aging): 4 dB → so we must keep actual OSNR ≥ 18 dB
STEP 1 — single-span OSNR (the per-amp rule) OSNR_1span = 58 + Pin − NF − span_loss = 58 + (+1) − 5 − 22 = 32 dB ← one span, before span-count penalty STEP 2 — apply the N-span penalty: OSNR_N = 32 − 10·log10(N) N 10·log10(N) OSNR_N vs 18 dB floor? 1 0.0 dB 32.0 dB PASS (14 dB margin) 2 3.0 dB 29.0 dB PASS 4 6.0 dB 26.0 dB PASS 8 9.0 dB 23.0 dB PASS 16 12.0 dB 20.0 dB PASS (2 dB above floor) 32 15.1 dB 16.9 dB FAIL ← dropped below the 18 dB floor STEP 3 — find the limit. Between 16 and 32 spans OSNR crosses 18 dB. Set 32 − 10·log10(N) = 18 → 10·log10(N) = 14 → N = 10^1.4 ≈ 25 spans. RESULT: ~25 identical 80 km spans (~2000 km) before OSNR hits the engineered floor. Past that you need a 3R regenerator, Raman gain, a lower-order/stronger-FEC format, or fewer/lower-loss spans. (Illustrative.)

Notice how the answer moves: drop NF from 5 to 4 dB and every OSNR figure improves 1 dB, pushing the limit further. Add Raman to knock ~4 dB off the effective noise and the same happens. Cut span loss to 20 dB and you gain 2 dB per span. Each lever from the rule-of-thumb formula shows up directly in the reach.

When OSNR — not power — is the wall

Common mistake

Adding another amplifier to "fix" a link that is erroring. If the received power is already adequate and the pre-FEC BER is climbing, the limiter is OSNR (or dispersion), not power. Another EDFA raises the signal and the noise floor together — you get a louder bad signal and risk driving the receiver into overload. When OSNR is the wall, the fixes are: a 3R regenerator (resets OSNR), Raman (adds effective OSNR), a stronger-FEC / lower-order transponder, or fewer / lower-loss spans.

Field checklist — engineering an OSNR budget

1. A link's received power is well within spec, but pre-FEC BER is rising over weeks and occasional post-FEC errors now appear. Add an amplifier?

No. Power is fine, so the limiter is OSNR (or dispersion), not power. Another amplifier lifts the noise floor along with the signal. Fixes: a 3R regenerator, Raman gain, a stronger-FEC/lower-order transponder, or fewer/lower-loss spans. Post-FEC errors mean you have run out of correction headroom — this is customer-affecting.

2. You currently run 4 identical spans at 24 dB OSNR. Roughly what OSNR would 8 identical spans give, and 16?

Doubling spans costs ~3 dB. 4→8 spans ≈ 21 dB; 8→16 spans ≈ 18 dB. (Formally, −10·log10(N) from the single-span value.)

3. What Q(dB) corresponds to the classic pre-FEC "error-free" target, and what BER is that?

Q ≈ 15.6 dB ↔ BER ≈ 1×10⁻¹². It is the legacy uncoded target. With strong SD-FEC, a system deliberately runs at a much lower Q / higher pre-FEC BER (up to ~2×10⁻²) yet still delivers error-free post-FEC traffic.

4. Given Pin = 0 dBm, NF = 5 dB, span loss = 23 dB, estimate the single-span OSNR, and the OSNR after 4 such spans.

Single span: 58 + 0 − 5 − 23 = 30 dB. After 4 spans: 30 − 10·log10(4) = 30 − 6 = 24 dB. (Illustrative rule of thumb.)

5. Why is SD-FEC worth roughly twice the coding gain of HD-FEC in practical terms?

SD-FEC (~11 dB+) vs HD-FEC (~6 dB) coding gain. Each extra dB lowers required OSNR by ~1 dB, and since doubling spans costs ~3 dB, the ~5 dB advantage of SD-FEC translates to well over double the reachable span count — the main reason coherent links go so far.

Try it: OSNR simulator

Sweep the span count, per-span loss, amplifier NF and launch power and watch the OSNR margin move. Doubling the spans really does cost ~3 dB. Pick the required OSNR that matches your optic and FEC.