OSNR and GOSNR Concatenation Workbench

Per-Section Noise Budget

Each row holds one section. The noise share column is the fraction of receiver noise that the section contributes. In per-section mode every white field is editable, and the checkbox removes a section from the sum without deleting the row.

Cumulative Accumulation Along the Section Chain

OSNR falls monotonically along the chain because every section adds noise to a signal restored to the same power. The gap between the curves is the nonlinear penalty accumulating in parallel with amplified spontaneous emission (ASE).

Table 2: Cumulative OSNR and GOSNR after each section
Section	Cumulative OSNR (dB)	Cumulative GOSNR (dB)	Requirement (dB)

Per-Section Noise Contribution

Reciprocal summation weights sections by noise power, so a section with a 3 dB worse OSNR contributes twice the noise of its neighbors. The bars show that weighting; the overlaid line shows the section OSNR that produced it.

Table 3: Noise contribution share by section
Section	Noise share (%)	Section OSNR (dB)	Section SNR(NLI) (dB)

Launch Power Sweep of ASE, NLI and GOSNR

Raising per-channel launch power lifts the ASE-limited OSNR one decibel per decibel and degrades the nonlinear term two decibels per decibel. GOSNR therefore has a single maximum, and at that maximum the nonlinear noise power equals half the ASE noise power, which places the peak 1.76 dB below the ASE-only OSNR.

Table 4: Launch power sweep, sampled every 2 dB
Offset (dB)	Launch power (dBm)	OSNR (dB)	SNR(NLI) (dB)	GOSNR (dB)

GOSNR Versus Section Count

For identical sections the ASE-limited OSNR follows a 10·log10(N) law exactly. GOSNR follows the same slope while the nonlinear accumulation exponent stays at zero, and falls faster once partial coherence is applied.

Table 5: Scaling with the number of identical sections
Sections	OSNR (dB)	GOSNR (dB)	Loss against one section (dB)

Receiver Noise Composition by Source

The same reciprocal arithmetic partitions receiver noise into its three sources. A design that is 90% ASE-limited responds to amplifier and span changes; one that is 60% nonlinear responds to launch power and modulation choices instead.

Table 6: Receiver noise composition
Noise source	Share of receiver noise (%)	Equivalent SNR (dB)

Takeaway: concatenation is addition of noise, never addition of decibels. Convert every section OSNR to a linear ratio, add the reciprocals, invert, and the total lands below the worst section. Adding the nonlinear and transceiver reciprocals to the same sum turns OSNR into GOSNR without changing the arithmetic.

Calculation Model and Formulas

The calculator implements four equations. The first concatenates sections, the second extends the sum to nonlinear and transceiver noise, the third derives a section OSNR from physical span parameters, and the fourth converts the optical ratio into the electrical signal-to-noise ratio the coherent receiver reports. Every constant below is stated with its evidence class.

Decibel arithmetic. A decibel is a logarithm, so it supports exactly two operations: addition and subtraction. Adding decibels multiplies the ratios behind them, which is why gains and losses along a path accumulate by plain addition — 20 dB of gain followed by 3 dB of loss is 17 dB, exactly. Everything else needs conversion to linear ratios first. Doubling a decibel figure does not double the ratio, it squares it: 20 dB becomes 40 dB, and 100 becomes 10 000. Powers do not add in decibels either — two equal noise contributions sit 3.01 dB above one of them, not 6 dB and not twice the number written on the page. Two shortcuts worth committing to memory, both exact: doubling a power adds 3.01 dB, and N identical sections cost 10·log10(N), so ten sections cost exactly 10.0 dB and twenty cost 13.01 dB. This calculator sums noise powers, which is why every decibel value entered is converted to a linear ratio, summed there, and converted back at the end. A spreadsheet column that adds OSNR values in decibels and labels the result a total is describing a network nobody built.

Reciprocal Summation of Section OSNR

Amplified spontaneous emission from each amplifier adds as optical power. Referring each contribution to the signal power at the same point makes the noise-to-signal ratios additive, and the noise-to-signal ratio is the reciprocal of OSNR. This is the cascaded-element relation standard-specified in ITU-T G.680 for optical network elements in series, and it is the reason a chain of ten identical spans loses exactly 10.0 dB against one span.

Formula 1 — OSNR concatenation

  1 / OSNRtotal  =  1 / OSNR1  +  1 / OSNR2  +  ...  +  1 / OSNRN

Where (all OSNR values in linear power ratio, 0.1 nm reference bandwidth):
  OSNRi   = linear OSNR of section i          [ratio]
  N        = number of sections in the path    [count]

Decibel conversion, applied before and after the sum, never during it:
  OSNR[linear] = 10 ^ ( OSNR[dB] / 10 )
  OSNR[dB]     = 10 x log10( OSNR[linear] )

Identical sections collapse to:
  OSNRtotal[dB] = OSNRsection[dB] - 10 x log10( N )

Extension to GOSNR

A coherent receiver does not distinguish ASE from nonlinear interference: both appear as additive Gaussian noise on the recovered constellation. The Gaussian Noise model exploits that equivalence, which is why the nonlinear term enters the same reciprocal sum. The transceiver term captures the back-to-back implementation noise of the module and is included when the module specification quotes it separately.

Formula 2 — generalized OSNR

  1 / GOSNR  =  1 / OSNRASE  +  1 / SNRNLI  +  1 / SNRTRX

Where:
  OSNRASE  = concatenated ASE-limited OSNR of Formula 1   [ratio]
  SNRNLI   = signal to nonlinear interference ratio        [ratio]
  SNRTRX   = transceiver back-to-back SNR                   [ratio]

Nonlinear accumulation over N sections:
  1 / SNRNLI,total = ( SUM 1 / SNRNLI,i ) x N ^ eps
  eps = 0        incoherent accumulation
  eps = 0.03 to 0.10   partially coherent, low dispersion-length product

Launch power scaling of the nonlinear term (cubic NLI power law):
  SNRNLI[dB]( P ) = SNRNLI[dB]( Pref ) - 2 x ( P - Pref )

Design rule. GOSNR peaks where the nonlinear noise power reaches half the ASE noise power. At that point GOSNR sits 10·log10(3/2) = 1.76 dB below the ASE-only OSNR, a theoretical-limit result of the Gaussian Noise model that holds independently of fiber type, span count and modulation format. Any design further than about 1 dB from that offset is trading margin for nothing.

Section OSNR from Span Parameters

When no measured value exists, each section OSNR follows from the power reaching the amplifier and the amplifier noise figure. The constant 58.0 dB is derived from physical constants: Planck constant h, optical frequency 193.4 THz and the 12.48 GHz reference bandwidth that corresponds to 0.1 nm at 1550 nm, giving 10·log10(h·ν·ν_ref) = −58.0 dBm.

Formula 3 — span-model section OSNR

  Lspan  = alpha x Lkm + Lextra
  OSNRi[dB] = Pch - Lspan - NF + 58.0

Where:
  Pch     = per-channel launch power into the span   [dBm]
  Lspan   = total section loss                       [dB]
  alpha  = fiber attenuation coefficient            [dB/km]
  NF     = amplifier noise figure                   [dB]
  58.0   = -10 x log10( h x v x vr ) with vr = 12.48 GHz   [dB]

Example, 80 km G.652.D span at 0.20 dB/km, 0 dBm per channel, 5.0 dB NF:
  Lspan = 0.20 x 80 = 16.0 dB
  OSNRi = 0.0 - 16.0 - 5.0 + 58.0 = 37.0 dB

Reference Bandwidth Conversion

OSNR is quoted in a 0.1 nm optical reference bandwidth; the coherent receiver reports an electrical SNR over its own symbol rate and across two polarizations. The conversion below is the standard dual-polarization relation and is what makes a 20.5 dB module requirement comparable with a 26 dB link result at 64 GBd.

Formula 4 — optical to electrical signal-to-noise ratio

  SNR[dB] = OSNR[dB] - 10 x log10( Rs / Bref ) + 10 x log10( 2 )

Where:
  Rs     = symbol rate                       [GBd]
  Bref   = 12.5 GHz reference bandwidth (0.1 nm at 1550 nm)
  10 x log10( 2 ) = 3.01 dB, dual-polarization term

Example at 64 GBd:
  SNR = OSNR - 10 x log10( 64 / 12.5 ) + 3.01 = OSNR - 4.08 dB

Figure 1: Noise contributions from each section collect on a common bus and enter one reciprocal sum. The same sum carries the nonlinear and transceiver terms, which is what converts OSNR into GOSNR without changing the arithmetic.

Practical Example — four-section link with one degraded amplifier

Three sections run 80 km at 0.20 dB/km with 5.0 dB noise figure amplifiers, giving 37.0 dB each at 0 dBm per channel. The third section carries an extra 6 dB of loss after a cable repair, so its OSNR is 31.0 dB. The linear reciprocals are 2.00×10⁻⁴ for each good section and 7.94×10⁻⁴ for the degraded one; the sum is 1.394×10⁻³, and the total ASE-limited OSNR is 28.6 dB. Four healthy sections would have given 31.0 dB. One section 6 dB down therefore costs 2.4 dB end to end, and it alone contributes 57% of the ASE noise at the receiver — which is why re-splicing that section returns more margin than replacing all three healthy amplifiers with 4.0 dB units.

Takeaway: the noise share column identifies the section worth fixing. Sections below about 10% share return less than 0.5 dB even if their noise is removed entirely, and effort spent on them buys nothing measurable at the receiver.

Reference Values and Application Notes

Default values in the calculator describe a C-band terrestrial line system on G.652.D fiber. The tables below give the ranges a design engineer can substitute, with the evidence class of each range stated in the caption. Module thresholds vary between vendors and firmware releases, so the transponder specification always overrides the table.

Required OSNR by Modulation Format

Values below are receiver thresholds. The formula chain behind them — symbol rate, FEC coding gain, implementation penalty and the OSNR to SNR conversion — is set out in the coherent optical transmission formula reference.

Table 7: Typical required OSNR at the receiver, 0.1 nm reference bandwidth (vendor-specified ranges, confirm against the module datasheet)
Line rate and format	Symbol rate (GBd)	Required OSNR (dB)	Typical application
100G DP-QPSK	32	12–15	Long haul and subsea
200G DP-16QAM	32–35	17–20	Regional and long haul
400G DP-16QAM	60–65	20–23	Regional and metro core
400G DP-QPSK	120–130	18–21	Long haul at high symbol rate
800G DP-64QAM	90–96	26–29	Data centre interconnect

Typical Per-Section Parameter Ranges

Table 8: Section parameter ranges used by the calculator defaults (field-measured and vendor-specified, as marked)
Parameter	Typical range	Evidence class	Effect on total
G.652.D attenuation at 1550 nm	0.18–0.22 dB/km	Field-measured	1 dB of loss costs 1 dB of section OSNR
G.654.E attenuation at 1550 nm	0.15–0.17 dB/km	Vendor-specified	Lower loss raises every section equally
C-band EDFA noise figure	4.5–6.5 dB	Vendor-specified	1 dB of noise figure costs 1 dB of section OSNR
Terrestrial span length	60–100 km	Design practice	Shorter spans raise OSNR and add amplifiers
Repeatered subsea span length	40–60 km	Design practice	Keeps section OSNR high over 6,000–13,000 km
Section SNR(NLI) at 0 dBm per channel	27–30 dB	Gaussian Noise model	Sets the launch power optimum
Transceiver back-to-back SNR	24–28 dB	Vendor-specified	Caps GOSNR regardless of line quality

Application Notes

Three usage points recur in design reviews. First, section boundaries are a choice: a section can be one amplified span, one optical multiplex section between reconfigurable optical add-drop multiplexers (ROADMs), or a complete vendor domain quoted as a single OSNR figure. The arithmetic does not care, as long as every OSNR value refers to the same reference bandwidth and every section appears exactly once. Second, per-channel power drives every result, and using composite power instead inflates the answer by 10·log10 of the channel count. Third, the nonlinear term belongs to the fiber spans only: a passive add/drop element contributes ASE-equivalent loss but no nonlinear interference, so its SNR(NLI) field is left blank rather than set to a large number.

Engineering note. A measured OSNR taken with an optical spectrum analyzer on a coherent signal reads high, because in-band noise sits under the signal spectrum where interpolation cannot see it. Integral or polarization-nulling methods return the value that matches this calculation. Feeding an interpolated analyzer reading into the concatenation produces an end-to-end figure that field measurement will not reproduce.

Takeaway: the calculator answers three design questions in one pass — what the receiver sees, which section produced the noise, and how far the launch power sits from its optimum. The third answer is often worth more than the first, because it costs nothing but a power adjustment.

References

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

Carry These Numbers Further

Optical Network Designer Take the section budget into a full topology. Built-in optical component library, span and node modelling, end-to-end design on the browser. Open the designer Optical Toolkit for iOS The quick utilities and reference data on your phone: unit conversions, budget checks and specification lookups for the field and the lab. View on the App Store