Introduction

A DWDM planning tool that declares a lightpath feasible has evaluated roughly forty equations in a fixed order, and a planner who knows the order can tell within a minute which of them is the one that failed. The chain starts with a span loss and a launch power, converts them into an amplified spontaneous emission (ASE) noise floor through a single reference constant, adds a nonlinear interference term that scales with the cube of that same launch power, subtracts four penalty functions for chromatic dispersion, polarization mode dispersion, polarization dependent loss and nonlinear accumulation, and maps whatever margin survives onto a Q-factor and a pre-forward-error-correction bit error rate. Post-FEC error-free operation is a threshold test on the last number in that chain.

What makes the chain worth learning as a chain rather than as a set of independent formulas is that the same physical quantity appears in it three times with three different exponents. Per-channel launch power enters the ASE term linearly, so every decibel of extra power buys a decibel of optical signal-to-noise ratio (OSNR). It enters the nonlinear interference term cubically, so every decibel of extra power costs three decibels of nonlinear noise. And it enters the stimulated Raman scattering (SRS) tilt term through the total band power, so in a C+L system extra power in one band reshapes the spectrum of the other. Those three exponents fix the shape of every optimization decision in the rest of this reference, and they are also why extra launch power is worth taking only up to the point where the nonlinear term reaches half the ASE term.

This reference sets out each equation with its variable definitions and units, a worked numerical value, and a live sandbox where the equation has a clear input-to-output relationship worth experimenting with. The worked values are not illustrative: the single-span, cascade and submarine cases are carried end to end through the same chain a production planning tool applies, every intermediate value is reproducible from the stated inputs, and the Q-factor chain closes on its own reported Q-factor, Q-factor margin and pre-FEC bit error rate to three decimal places. Readers who want the noise accounting from first principles before the engineering forms should start with the OSNR fundamentals primer; readers who want the closed-form nonlinear theory behind the engineering approximations used here should read the Gaussian Noise model treatment.

1. Reference Frame and the −58 dBm Constant

Optical signal-to-noise ratio is a ratio of two powers measured in different ways, and every disagreement between two tools that both report "OSNR" traces back to the reference bandwidth. The signal term is the total power of one channel. The noise term is the ASE power inside a fixed spectral window centred on that channel, conventionally 0.1 nm.

At 1550 nm the 0.1 nm window converts to an optical frequency span through the standard wavelength-to-frequency relation, and this conversion is the reason the number 12.5 GHz appears throughout coherent link engineering.

The ASE power spectral density at the output of an ideal amplifier is set by quantum mechanics, not by design choice, and evaluating it over the reference bandwidth produces a single constant that collapses all the physical constants into one number.

Because it is a quantum floor rather than an engineering parameter, the constant does not change with equipment or with route. It shifts marginally with band: evaluated at the L-band centre near 188.5 THz instead of 193.4 THz, the constant moves by 10·log₁₀(188.5/193.4) = −0.11 dB, which production tools absorb rather than track. What does change is the third quantity in Eq. 3 when a receiver, not a spectrum analyzer, is the measuring instrument. A coherent receiver integrates noise over its own matched filter, whose noise bandwidth equals the symbol rate, so converting a spectrum-analyzer OSNR into the electrical signal-to-noise ratio a transceiver datasheet is written against requires an explicit bandwidth ratio.

The practical size of that correction matters. A 138 GBd channel — the symbol rate of the 800 Gb/s line interfaces used in the worked cases later in this reference — carries a conversion term of 10·log₁₀(138/12.5) = 10.4 dB. A receiver needing 12.9 dB of electrical SNR therefore needs 23.3 dB of measured OSNR, and that is exactly the threshold value the planning tool applies. Confusing the two domains produces a 10 dB design error, which is the single most expensive arithmetic mistake available in this discipline.

2. Span Loss, Wavelength-Dependent Loss and Channel Attenuation

Span loss is the input to everything downstream, and it is not one number per span but one number per channel. Silica attenuation falls across the C-band and rises again into the L-band, so the blue edge of the C-band arrives with measurably less power than the red edge after a long span. Planning tools model this as a linear tilt about the band centre rather than a full spectral curve.

The corresponding L-band form replaces the ITU channel index with absolute frequency and normalises to the 4.8 THz L-band width, so the tilt term becomes FiberTilt · (f_j − f_mid,L) / (4.8 · L) with f_mid,L = 188.475 THz. Channel power and the ASE noise accompanying it are then attenuated identically, which is why a passive span changes power levels without changing OSNR.