Optical Signal-to-Noise Ratio (OSNR)

The Limit That Takes Over From Loss

On an unamplified link the constraint is straightforward: too little light arrives and the receiver cannot see it. Add optical amplifiers and that stops being true, because gain is cheap. What is not cheap is the amplified spontaneous emission each EDFA adds along with the gain.

OSNR is signal power divided by that noise power, measured in a 0.1 nm reference bandwidth — about 12.5 GHz at 1550 nm — by convention, so that numbers from different instruments and different symbol rates can be compared. On an amplified system it is the number that decides whether the link closes.

The Arithmetic

For one span a good estimate is OSNR = 58 + per-channel launch power in dBm - span loss in dB - amplifier noise figure in dB. Launch +1 dBm per channel into a 20 dB span and recover it with a 4.5 dB amplifier and you start at about 34.5 dB.

For N identical spans, subtract 10 log₁₀(N). Every doubling of the span count costs 3 dB and nothing recovers it, because the signal is amplified and so is all the noise that came before. Ten spans cost 10 dB against one.

What the Modulation Format Demands

The requirement side is set by bits per symbol. Reading the published OpenZR+ modes: 100G DP-QPSK needs about 12.5 dB, 200G DP-QPSK 16 dB, 300G DP-8QAM 21 dB, and 400G DP-16QAM 24 dB. The OIF 400ZR mode, which uses a weaker FEC, asks for 26 dB.

That ladder is the whole reach-versus-capacity trade in one column of numbers. Halving the line rate buys roughly 8 dB, which is nearly three doublings of span count. It is why a coherent system's mode selection is an OSNR budgeting exercise rather than a feature choice, and why ZR and ZR+ differ mainly in what modes they offer.

What Goes Wrong

The first mistake is trying to fix poor OSNR with more launch power. Above the optimum, extra power buys nonlinear penalty — four-wave mixing, cross-phase modulation between neighbouring channels — faster than it buys signal, and the curve turns back on itself.

The second is trusting receive power. A DWDM receiver reading a healthy -8 dBm while the link errors is the classic OSNR failure: the power is real, most of it is noise. Measure with an optical spectrum analyser, and on a densely packed grid remember that filtered noise between channels can make a conventional out-of-band reading optimistic. The power converter handles the dBm arithmetic that goes with it.

Related Terms

EDFA, coherent optics, 400ZR and ZR+, DWDM and FEC.