MapYourTech · MapYourBasics Series

Noise Modelling for Optical Communication

The four noise sources that bound every link — ASE, nonlinear interference, shot, and thermal — and the OSNR and GSNR math that turns them into a reach number.

Introduction

Noise sets the ceiling on every optical link. It is the random fluctuation in signal power, phase, and timing that decides how cleanly a receiver recovers bits, and it comes from four physical mechanisms that behave very differently: amplified spontaneous emission that adds with every amplifier, nonlinear interference that grows with the cube of launch power, shot noise from the discrete arrival of photons, and thermal noise in the receiver electronics. Modelling them is what lets an engineer predict reach and capacity before a single fiber is lit. This guide builds each mechanism from its physics, gives the OSNR and GSNR math that combines them, and provides four calculators to run the budget. The thread running through all of it: ASE accumulates, NLI is cubic in power, and the optimal launch power sits where the two cross.

One metric to carry: the Generalized Signal-to-Noise Ratio (GSNR) combines ASE and NLI into a single feasibility number, GSNR = P_ch / (P_ASE + P_NLI). It is the metric modern planning tools compute, because OSNR alone misses the nonlinear half of the problem.

1. Fundamentals and Core Concepts

Where noise comes from

Quantum nature of light: photons arrive at random intervals, producing shot noise at the photodetector — the fundamental quantum floor.

Amplifier physics: spontaneous emission in optical amplifiers is amplified with the signal as ASE noise, the dominant source in amplified multi-span systems.

Thermal agitation: random electron motion in the receiver front end produces Johnson-Nyquist thermal noise, dominant at low received power.

Kerr nonlinearity: intensity-dependent refractive index couples channels and generates nonlinear interference through SPM, XPM, and FWM — growing as the cube of launch power.

Where modelling decides the outcome

Table 1: Scenarios where noise modelling is decisive
Scenario	Impact	Key parameters
Long-haul transmission	ASE accumulates over many amplified spans	OSNR, span loss, amplifier spacing
High-capacity systems	NLI caps launch power and spectral efficiency	Channel power, fiber nonlinearity, spacing
High-order modulation	Dense constellations need high SNR for low BER	Required OSNR, format, FEC overhead
Multiband systems	SRS tilt and band-dependent loss imbalance noise	SRS tilt, band allocation, power management

Takeaway: Over a 2000 km C+L system with around 50 amplified spans, the difference between an accurate noise model and an optimistic one is the difference between planning for 16 Tb/s and stalling at 8 Tb/s — the model directly sets the business case.

2. Mathematical Framework

OSNR

OSNR (dB) = 10·log₁₀(P_signal / (PSD_noise × B_ref))

P_signal is the per-channel power, PSD_noise the noise power spectral density, and B_ref the reference bandwidth (12.5 GHz / 0.1 nm).

ASE power from one amplifier

P_ASE = n_sp · hν · (G − 1) · B_o

n_sp is the spontaneous-emission factor (1.5–2.5 for EDFAs), G the linear gain, B_o the optical bandwidth.

Per-span and multi-span OSNR (engineering form)

OSNR_span = 58 + P_ch − L_span − NF

OSNR_total = OSNR_span − 10·log₁₀(N)

P_ch is per-channel power (dBm), L_span the span loss the amplifier overcomes, NF the noise figure, and +58 the bandwidth constant at 1550 nm over 12.5 GHz. The +58 already carries the photon-energy and bandwidth terms — do not also subtract 10·log₁₀(B_ref), or the result is wrong by about 100 dB.

GSNR with nonlinear interference

1/GSNR = 1/OSNR + 1/SNR_NLI

GSNR = P_signal / (P_ASE + P_NLI)

NLI power spectral density (GN model)

PSD_NLI(f) ≈ (16/27) · γ² · P³ · (L_eff²/L_span) · N · g(f)

γ is the nonlinear coefficient (about 1.2 /W/km for SMF, near 0.001 for hollow-core fiber), P channel power, L_eff the effective length, N the span count. The cubic P³ dependence is why launch power has a hard optimum — the GN model captures SPM, XPM, and FWM as a single additive Gaussian term once dispersion has decorrelated the signal.

Where the GN model holds: it assumes the WDM signal is statistically Gaussian, which is true once chromatic dispersion has decorrelated the channels — typically within the first one or two spans of an uncompensated coherent link. It degrades for low-dispersion fiber and very low baud rates, where the Gaussian assumption breaks.

3. Types of Noise

Amplified spontaneous emission (ASE)

Broadband, additive, signal-independent, and dominant in amplified multi-span systems. Noise figure runs 4–6 dB for C-band EDFAs and 6–8 dB for L-band. OSNR drops 3 dB per doubling of spans.

Nonlinear interference (NLI)

From the Kerr effect (n = n₀ + n₂|E|²): self-phase modulation broadens the signal spectrum, cross-phase modulation couples channels, and four-wave mixing generates new tones. NLI power scales as P³, so it sets the maximum usable launch power.

Shot noise

i_shot² = 2qI_photoB_e. The quantum detection floor, more significant in direct-detection systems.

Thermal noise

i_thermal² = 4k_BTB_e/R. Dominant in receiver electronics at low optical power; a low-noise front end or optical pre-amplifier suppresses it.

Table 2: Noise mechanisms compared
Noise type	Spectrum	Power dependence	Mitigation
ASE	Broadband, uniform	Independent of signal	Low-NF amplifiers, shorter spans
NLI	Signal-dependent	∝ P³	Lower launch power, larger A_eff fiber
Shot	White	∝ √P	Higher optical power
Thermal	White	Independent	Low-noise receivers, pre-amplification
SRS tilt	Power tilt across bands	∝ P × Δf	Power pre-emphasis, ASE idlers

Fiber dependence

Standard single-mode fiber: γ ≈ 1.2–1.3 /W/km; NLI dominant above roughly 0 dBm/channel; ASE and NLI comparable in long-haul.

Hollow-core fiber: γ about 1000× lower, so NLI is negligible and ASE plus splice loss become the limits; it can run at much higher launch power, shifting the whole budget.

4. Effects and Impacts

Table 3: OSNR to performance (QPSK reference)
OSNR (dB)	Pre-FEC BER	Status	Action
> 20	< 10^-12	Excellent	Normal operation
15–20	10^-9–10^-12	Good	Monitor
10–15	10^-6–10^-3	Marginal	FEC required
< 10	> 10^-3	Poor	Intervention

Capacity follows the modulation the OSNR supports: at 50 GHz spacing, QPSK (15 dB) gives 100 Gb/s, 16-QAM (21 dB) gives 200 Gb/s, and 64-QAM (28 dB) gives 300 Gb/s — so a 6 dB OSNR loss can cut per-channel capacity by two thirds. Reach follows the 10·log₁₀(N) law: a 1 dB lower noise figure extends reach by roughly 200–300 km in a typical long-haul system.

Table 4: Mitigation by noise type
Impairment	Primary mitigation	Secondary	Typical gain
ASE	Low-NF amplifiers	Distributed Raman	2–3 dB
NLI	Reduced launch power	Digital back-propagation	1–4 dB
Shot	Higher optical power	Balanced detection	3–6 dB
Thermal	Low-noise TIA	Optical pre-amplifier	5–10 dB

5. Interactive Calculators

Four calculators on one consistent model. Simulator 2 is the one to sit with: hold the fiber fixed and sweep launch power, and watch GSNR rise on ASE, peak, then fall as NLI takes over — the optimal launch power is that peak, and on hollow-core fiber it sits far to the right.

6. Techniques and Solutions

Low-NF EDFAs

Gain-flattening filters and optimised erbium length hold NF to 4–5 dB in C-band; each 1 dB of NF is 1 dB of per-span OSNR, so this is the cheapest lever on ASE.

Distributed Raman

A counter-propagating pump roughly 100 nm below the signal gives distributed gain and an effective NF near 0 to −2 dB — 3–5 dB OSNR and meaningful reach extension, at the cost of pump management and power.

Digital back-propagation

Numerically inverting the nonlinear Schrödinger equation recovers 1–4 dB of nonlinear penalty, but the FFT load per span is heavy — reserved for high-value long-haul links.

Forward error correction

Hard-decision RS gives 6–7 dB net coding gain at 7% overhead; soft-decision LDPC reaches 11–12 dB at 20%. FEC is the strongest lever for relaxing the OSNR requirement.

Coherent detection

Capturing amplitude and phase across both polarizations enables DSP compensation of linear impairments and gives 3–6 dB over direct detection.

Table 5: Techniques compared
Technique	Target noise	Gain	Cost
Low-NF EDFA	ASE	1–2 dB/span	Medium
Raman	ASE	3–5 dB	High
Digital back-propagation	NLI	1–4 dB	DSP load
Advanced FEC	All	11–14 dB NCG	Medium
Coherent RX	All	3–6 dB	High

7. Design Methodology

Define capacity, distance, and target pre-FEC BER; set the required OSNR for the format; build the amplifier chain; optimise launch power against NLI; then verify the GSNR budget with margin. The required OSNR is the format threshold plus a 2–3 dB implementation margin.

Practical Example — 1600 km, 80×100G DP-QPSK.
20 spans of 80 km, fiber 0.2 dB/km → span loss 16 dB. Amplifier total output +17 dBm across 80 channels → per-channel +17 − 10·log₁₀(80) = −2 dBm. NF 5 dB.
Per-span OSNR = 58 + (−2) − 16 − 5 = 35 dB.
Receiver OSNR = 35 − 10·log₁₀(20) = 35 − 13 = 22 dB.
Against a 13.5 dB QPSK target (with margin) that is 8.5 dB of headroom — the link closes comfortably. The headroom is also what you would spend to move to 16-QAM (about 21 dB required); there a low-NF amplifier or Raman buys the extra few dB.

Per-channel, not total: the single most common budgeting error here is feeding total amplifier output into the per-channel OSNR formula. For 80 channels at +17 dBm total, the per-channel figure is −2 dBm — 19 dB lower. Use per-channel power, and never subtract a separate bandwidth term on top of the +58 constant.

Pitfalls to avoid

Pitfall: ignoring implementation margin. Fix: add 2–3 dB over the theoretical threshold.

Pitfall: linearising NLI. Fix: use the GN model or measurement, not a fixed dB penalty.

Pitfall: one model for all bands. Fix: model S, C, and L separately — loss and SRS differ.

Pitfall: designing for fresh fiber. Fix: budget 5–10 years of component aging.

Pitfall: under-counting splice loss. Fix: 0.5–1 dB for hollow-core splices, 0.1–0.3 dB for SMF.

8. Applications and Case Studies

Practical Example — 2000 km backbone, 2.5 to 12 Tb/s.
Twenty-five 80 km spans on legacy 6 dB NF EDFAs gave a marginal 11 dB OSNR. Swapping to 4.5 dB NF amplifiers added 1.5 dB, adding C+L doubled the spectrum, hybrid Raman on the worst spans added 2.5 dB, and 16-QAM with 11 dB-NCG soft-decision FEC closed the loop — OSNR reached 15 dB, capacity rose to 12 Tb/s, and cost per bit fell about 75%. C+L expansion beat denser C-band packing on cost.

Practical Example — 200 km hollow-core DCI.
On hollow-core fiber the latency dropped about a third (light travels faster in air) and γ fell roughly 1000×, so NLI vanished as a limit. Running unrepeated at +12 dBm launch (versus 0 dBm on SMF) with coherent receivers and optical pre-amplification, the link carried DP-64QAM at 32 Tb/s with a 22 dB receiver-limited OSNR and no inline amplifiers — ASE accumulation eliminated, splice quality now the dominant concern.

Practical Example — 6000 km submarine, 25-year life.
120 repeaters at 50 km spacing on hybrid Raman+EDFA (effective NF 3.5 dB), low-loss fiber → span loss about 10 dB, per-channel near −1 dBm. Per-span OSNR = 58 − 1 − 10 − 3.5 = 43.5 dB; over 120 spans, 43.5 − 10·log₁₀(120) = 43.5 − 20.8 = 22.7 dB at deployment, falling to about 22 dB end-of-life after aging margin — enough for DP-16QAM. Short low-loss spans are what make the budget close: fewer dB lost per hop means a high per-span OSNR survives the 20.8 dB cascade penalty.

Table 6: Symptom-to-cause guide
Symptom	Likely cause	Fix
OSNR drift over time	Amplifier aging	Check pump power, replace pumps
High BER on some channels	Gain ripple	Adjust GFF or per-channel equalisation
Power-dependent BER floor	Excessive NLI	Reduce launch power 2–3 dB
Band-edge channel failure	Low OSNR or residual CD	Adjust pre-emphasis or DSP
All channels degraded equally	ASE buildup	Lower amplifier NF or add Raman

Table 7: Quick-reference values
Parameter	Typical range	Best practice
EDFA NF (C-band)	4–6 dB	≤ 5 dB
EDFA NF (L-band)	5–8 dB	≤ 6 dB
Launch power (SMF)	−2 to +3 dBm	0 to +1 dBm/ch
Span loss	15–20 dB	16–18 dB
Receiver OSNR	12–25 dB	> 15 dB

Main Points

1. Four mechanisms bound a link: ASE, NLI, shot, and thermal noise — each scales differently.

2. ASE dominates amplified multi-span systems and accumulates as 10·log₁₀(N).

3. NLI scales as P³, so launch power has a hard optimum where GSNR peaks.

4. GSNR — not OSNR alone — is the feasibility metric, since it carries the nonlinear half.

5. The engineering OSNR form is 58 + P_ch − L_span − NF − 10·log₁₀(N); use per-channel power.

6. The +58 constant already carries the bandwidth term — never subtract 10·log(B_ref) again.

7. Advanced FEC (11–14 dB NCG) is the strongest lever for relaxing the OSNR requirement.

8. Hollow-core fiber moves the limit from NLI to ASE and splice loss, allowing far higher launch power.

9. Short low-loss spans give a high per-span OSNR that survives the cascade penalty — the submarine trick.

10. Carry 2–3 dB implementation margin plus aging; validate the model against measurement.

References

ITU-T, Optical monitoring for dense wavelength division multiplexing systems (G.697), ITU-T Study Group 15.
P. Poggiolini et al., The GN-Model of Fiber Non-Linear Propagation and its Applications, Journal of Lightwave Technology.
ITU-T, Forward error correction for high bit-rate DWDM submarine systems (G.975.1), ITU-T Study Group 15.

Sanjay Yadav, "Optical Network Communications: An Engineer's Perspective" — Bridge the Gap Between Theory and Practice in Optical Networking.

Developed by MapYourTech Team

For educational purposes in Optical Networking Communications Technologies

Note: This guide is based on industry standards, best practices, and real-world implementation experiences. Specific implementations may vary based on equipment vendors, network topology, and regulatory requirements. Always consult with qualified network engineers and follow vendor documentation for actual deployments.

Feedback Welcome: If you have any suggestions, corrections, or improvements to propose, please feel free to write to us at [email protected]

Start with DWDM Fundamentals

871+ Technical Articles

Engineering Knowledge Base

Noise Modelling for Optical Communication

Noise Modelling for Optical Communication

Introduction