OSNR Calculation for Coherent 400ZR Links: Step-by-Step Examples
How to compute OSNR for 400ZR coherent DCI links: the 0.1 nm reference, the per-span formula with the 58 dB constant, span stacking with 10·log10(N), and two worked examples (single 80 km span and a 3-span amplified route) against realistic required-OSNR figures.
What OSNR measures — and the 0.1 nm reference
OSNR is the ratio of channel power to accumulated ASE noise power, evaluated in a fixed reference bandwidth:
OSNR = P_ch / P_ASE(0.1 nm)
For an EDFA with gain G and noise figure NF, the ASE power in 0.1 nm at 1550 nm is:
P_ASE(dBm) = NF(dB) − 58 dBm + G(dB)
(The −58 dBm constant is hν·B for B = 0.1 nm ≈ 12.5 GHz at 1550 nm.) With a line amplifier gain matching span loss L, each span contributes:
OSNR_span = 58 + P_ch − NF − L_span (all in dB / dBm, ref 0.1 nm)
Stacking spans: the 10·log₁₀(N) penalty
Noise adds in linear units, so for N equal spans:
OSNR_total = OSNR_span − 10·log₁₀(N)
For unequal spans, sum the linear noise contributions: 1/OSNR_total = Σᵢ 1/OSNRᵢ (linear). The equal-span shortcut is accurate enough for planning when spans are within a few dB of each other.
Worked example 1 — single 80 km span (the 400ZR sweet spot)
| Parameter | Value |
|---|---|
| Span length | 80 km |
| Fiber attenuation | 0.22 dB/km → L = 17.6 dB |
| Connector/patch allowance | +0.9 dB → L = 18.5 dB |
| Line EDFA gain | 18.5 dB (gain = loss) |
| EDFA noise figure | 5.5 dB |
| Channel power P_ch | 0 dBm |
OSNR = 58 + 0 − 5.5 − 18.5 = 34.0 dB
Against a 400ZR requirement of ~26 dB (26 dB required + ~1.5 dB implementation penalty), margin is ≈ 6.5 dB — comfortable. This is why unamplified-or-single-amp 80 km DCI routes are the textbook 400ZR deployment.
Worked example 2 — three 80 km spans with inline amplifiers
| Parameter | Value |
|---|---|
| Spans | 3 × 80 km, each L = 20 dB (incl. connectors) |
| Inline EDFA NF | 5.5 dB |
| Channel power P_ch | 0 dBm |
Per span: OSNR_span = 58 + 0 − 5.5 − 20 = 32.5 dB
Total: OSNR_total = 32.5 − 10·log₁₀(3) = 32.5 − 4.77 = 27.7 dB
Requirement: 26 dB + 1.5 dB implementation penalty = 27.5 dB → margin 0.2 dB. That is a coin flip in production. Options, in rough order of cost:
| Option | OSNR gain | Notes |
|---|---|---|
| Raise P_ch +2 dB | +2 dB | watch nonlinear threshold (16QAM is sensitive) |
| Lower-loss spans (better splices, fewer connectors) | +0.5–1.5 dB | cheapest if the plant allows |
| Distributed Raman assist | +3–5 dB | adds cost/complexity per span |
| Drop to 2 spans (re-route) | +1.76 dB | N-penalty drops from 4.77 to 3.01 dB |
Required OSNR: planning figures by format
| Format | Modulation / FEC | Typical required OSNR @ 0.1 nm |
|---|---|---|
| 400ZR | DP-16QAM + oFEC | ≈ 25–27 dB |
| 400ZR+ (80–120 km class) | DP-16QAM + stronger oFEC | ≈ 24–26 dB |
| 100G IM-DD (duplex) | PAM4, different regime | not comparable (use direct-error budgets) |
Notes that keep the math honest:
- Implementation penalty: transceiver-specific, typically 1–2 dB at 400ZR baud rates; vendor datasheets sometimes quote gOSNR which already includes it.
- 0.1 nm vs GHz: some instruments report noise in GHz; 0.1 nm ≈ 12.5 GHz at 1550 nm — convert before comparing.
- Nonlinear effects (SPM/XPM) do not appear in linear OSNR math. At 16QAM and typical DCI powers, they consume part of your margin invisibly — one more reason for the 3 dB floor.
Step-by-step checklist for your own link
- Measure or estimate span loss per section (fiber km × 0.22 dB/km + connectors + patch panels). DDM on existing optics gives a reality check.
- Pick amplifier NF from the datasheet (EDFA line amps: 5–6 dB typical).
- Fix P_ch from the transceiver spec and nonlinearity headroom (0 dBm is a sane default for 400ZR).
- Compute OSNR_span = 58 + P_ch − NF − L, then subtract 10·log₁₀(N).
- Compare against required OSNR + implementation penalty + 3 dB margin.
- After turn-up, verify with pre-FEC BER / DDM on both ends and reconcile against the prediction — the deltas teach you your plant's true NF.
Bottom line
- The whole calculation reduces to one line per span: 58 + P_ch − NF − L, minus 10·log₁₀(N) for span stacking.
- Single-span 80 km 400ZR works with margin to spare; three spans is where margin starts disappearing, and every option to recover it costs money or physics.
- Keep 3 dB of margin, use gOSNR figures when a vendor gives them, and validate against live pre-FEC BER rather than trusting the spreadsheet alone.
Compiled from OIF 400ZR implementation agreement concepts, standard EDFA noise theory, and common DCI engineering practice; original synthesis, no reproduction.
FAQ
- What OSNR does 400ZR actually need?
- For DP-16QAM with open-amplification FEC (oFEC) at the 400ZR pre-FEC threshold, planning figures cluster around 25–27 dB in 0.1 nm. Always add the transponder implementation penalty (1–2 dB) and keep 3 dB or more of engineering margin.
- Why is OSNR referenced to 0.1 nm?
- ASE noise power depends on measurement bandwidth, so a fixed optical reference bandwidth (0.1 nm ≈ 12.5 GHz at 1550 nm) makes numbers comparable across instruments and vendors. The ASE in that band at 1550 nm works out to about −58 dBm per dB of gain with NF = 0 dB.
- What is the difference between OSNR and gOSNR?
- OSNR is measured in the line (typically with the channel off or via nulling). gOSNR (generalized OSNR) is the value back-calculated at the receiver that includes linear and nonlinear impairments the transmitter/receiver actually experiences — it is the number that predicts pre-FEC BER.
- How much margin should I keep on a 400ZR link?
- Industry practice is 2–3 dB of system margin on top of required OSNR plus implementation penalty, to cover connector drift, EDFA gain ripple, aging, and measurement uncertainty.