A fibre optic loss budget is a single sum. You start with the optical power the transmitter launches, subtract every loss between the two ends, and check what arrives against the power the receiver needs in order to work:
Received power (dBm) = launch power (dBm) − total link loss (dB)
Margin (dB) = received power (dBm) − receiver sensitivity (dBm)
Total link loss is the fibre itself, plus every connector, splice and splitter along the way:
Total loss = (length × attenuation per km) + (connectors × loss each) + (splices × loss each) + splitter loss
The received power has to land inside a window. Too little light and the receiver cannot recover the data. Too much light and its front end saturates, which fails just as hard.
-1.0 dBm ┬── receiver maximum input (too much light)
│
│ the received power must
│ land inside this window
-11.7 dBm ┼── received power
│
│ margin 4.1 dB
│
-15.8 dBm ┴── receiver sensitivity (too little light)
Most engineers check the bottom of that window and forget the top. On a short link with a long reach optic, the top is the one that catches you.
Optical power is quoted in dBm, which is power referenced to one milliwatt. A transmitter might launch at −4.7 dBm and a receiver might need at least −15.8 dBm to recover the data. Everything in between is subtraction, and because the units are logarithmic you add the losses in dB rather than multiplying ratios.
Follow the light down an illustrative path and the budget is just a running total:
TX ───────────────────── launch -4.7 dBm
│
[C] mated connector pair -0.5 dB -5.2 dBm
│
═══ fibre, 12 km at 1550 nm -2.6 dB -7.8 dBm
│
[S] fusion splice -0.1 dB -7.9 dBm
│
[X] splitter, 1:8 -10.5 dB -18.4 dBm
│
[C] mated connector pair -0.5 dB -18.9 dBm
│
RX ───────────────────── arrives -18.9 dBm
sensitivity -24.0 dBm
────────────────────────────────
margin 5.1 dB
Notice how the splitter dwarfs everything else. On a passive optical network that one component decides the design, while on a point to point link the fibre and the connectors share the work.
A fibre link matters here because it fails differently from a copper or radio link. There is no long, audible degradation you can watch coming. In practice the transition is narrower than it is abrupt: on a modern system with forward error correction you first pass through a region where the FEC is quietly correcting a rising number of errors, and then, over a small further change in received power, the errors outrun the correction and the link drops. The loss budget tells you how far you are from the edge of that region, which is why the margin figure is far more useful than a bare pass or fail verdict.
Why it matters in the field
On a remote site, a link that fails in six months because the budget had no repair margin is not a maintenance ticket. It is a four hour drive and a day of lost production. Margin is bought at the design stage, where it costs nothing.
Most people mean the power budget when they say loss budget, but a fibre link has to pass two independent tests, and a design that satisfies one can still fail the other.
On multimode the limit is modal dispersion, which is why multimode is specified by an effective modal bandwidth. OM3 is rated at 2000 MHz·km at 850 nm and OM4 at 4700 MHz·km, and that is what sets the 300 m and 400 m reach figures for 10 Gbit/s, rather than any shortage of optical power.
On singlemode the limit is chromatic dispersion, roughly 17 ps per nanometre per kilometre at 1550 nm. G.652 fibre has its dispersion null near 1310 nm, which is exactly why 1310 nm is preferred where dispersion would otherwise bite. At 10 Gbit/s over 80 km at 1550 nm you can have ample power and still fail, and the fix is a dispersion compensating module, not a bigger laser.
The working rule is that at 10 Gbit/s or below, and under roughly 40 km, the power budget is the one that decides your design. Beyond that, on many modern Ethernet links, sufficient power is necessary but no longer sufficient, and you must check both.
Attenuation depends on the fibre type and, strongly, on the wavelength. Longer wavelengths scatter less, which is why 1550 nm reaches further than 1310 nm on the same glass.
| Fibre and wavelength | Typical measured (dB/km) | Design maximum (dB/km) |
|---|---|---|
| Multimode OM3 and OM4 at 850 nm | 2.3 to 3.0 | 3.5 |
| Multimode OM3 and OM4 at 1300 nm | 0.6 to 1.0 | 1.5 |
| Singlemode G.652.D at 1310 nm | 0.32 to 0.36 | 0.4 |
| Singlemode G.652.D at 1550 nm | 0.19 to 0.22 | 0.25 |
| Singlemode G.652.D at 1625 nm | 0.21 to 0.25 | 0.3 |
Budget against the design maximum column, which is deliberately pessimistic. The typical column is what a good installation actually measures. Structured cabling standards allow more headroom again, permitting up to 0.5 dB/km for outside plant singlemode and 1.0 dB/km inside plant, because the standard has to cover cable that has been pulled hard, aged and repaired.
Design tip
Use the attenuation figure from your cable manufacturer’s datasheet in preference to any published table. The numbers above exist to sanity check a design, not to replace a specification.
Every join costs light. On a short link the joins dominate the budget completely, and it is common for a 200 m in-building run to lose more in its four connector pairs than in all of its fibre.
| Element | Typical measured | Design allowance | Standard maximum |
|---|---|---|---|
| Fusion splice | 0.02 to 0.10 dB | 0.10 dB | 0.30 dB |
| Mechanical splice | 0.10 to 0.30 dB | 0.30 dB | 0.30 dB |
| Mated connector pair | 0.20 to 0.50 dB | 0.50 dB | 0.75 dB |
| Field terminated connector | 0.30 to 0.75 dB | 0.75 dB | 0.75 dB |
Three points are worth drawing out of that table.
A mated pair means two connectors joined in an adaptor, so one patch panel port counts as one mated pair and not two. A fusion splice is roughly five times better than a connector, which is why a permanent joint should be fused rather than patched wherever you can reach it. And connector loss is the item most likely to drift over time, because it degrades with dirt, with mating cycles and with careless handling in a way that a fused splice never does.
Common mistake
The most frequent cause of a link that suddenly misses its budget is a dirty end face. A fingerprint on a ferrule can add several dB on its own. Inspect and clean before you measure, every time, and do it before you start blaming the design.
The optical module sets both ends of the budget. Its minimum launch power is where you start, its sensitivity is the line you must stay above, and its maximum input is the ceiling you must stay below. The gap between minimum launch and sensitivity is the loss the link is allowed to have.
| Module | Wavelength | Min launch | Sensitivity | Approx budget | Max input |
|---|---|---|---|---|---|
| 1000BASE-LX | 1310 nm | −9.5 dBm | −19.0 dBm | 9.5 dB | −3 dBm |
| 10GBASE-LR | 1310 nm | −8.2 dBm | −14.4 dBm | 6.2 dB | +0.5 dBm |
| 10GBASE-ER | 1550 nm | −4.7 dBm | −15.8 dBm | 11.1 dB | −1.0 dBm |
| 10GBASE-ZR (vendor type) | 1550 nm | 0 dBm | −24.0 dBm | 24 dB | −7 dBm |
These are the values commonly quoted for these module types. Treat them as a planning aid and confirm against the datasheet for the module you are actually buying, because vendors differ and the datasheet is the only authority that counts.
Two things here catch people out. The workhorse 10 km optic, 10GBASE-LR, has a budget of only about 6.2 dB, which is far less than most engineers assume and is easily eaten by a long run with a couple of patch panels. And ZR is not an IEEE standard type at all. It is a de facto vendor category, so its figures vary between manufacturers more than the standardised types do.
Design tip
Always budget using the minimum transmitter power and the maximum loss of every component. Typical values describe the link on a good day. The worst case combination is the one the link will eventually present you with.
A surface mine needs a 10 Gbit/s link between the comms hut and the processing plant, 12 km apart, on outside plant G.652.D singlemode. The cable arrives on 4 km drums, so there are two mid-span fusion joins. At each end the outside plant cable is fused to a pigtail in a patch panel, and a patch cord runs from that panel to the switch.
That gives four fusion splices in total, being two mid-span joins plus one pigtail splice at each end, and two mated connector pairs, being one patch panel port at each end.
First attempt, 10GBASE-LR at 1310 nm. Using design values throughout:
The LR budget is 6.2 dB, so the link passes with 0.60 dB to spare. That is not a design, it is a coincidence. One future repair splice, one dirty connector or one slightly out of spec drum takes the link down, and nothing is left for ageing. The honest verdict is that LR does not do this job.
Second attempt, 10GBASE-ER at 1550 nm. The longer wavelength cuts the fibre loss by nearly half:
Against the ER budget of 11.1 dB that leaves 7.06 dB of margin, which is generous. But the design is not finished, because the other end of the budget now bites.
Every receiver has a maximum input power as well as a minimum. Give it more light than that and the front end saturates, the eye closes and the link errors or drops. It presents as a fault on a link that has too much power rather than too little, which is not where anyone looks first.
The overload check uses the opposite extreme from the sensitivity check. It takes the maximum launch power against the minimum possible loss, using best case measured components rather than design values:
A 10GBASE-ER module can launch as high as +4.0 dBm, so the best case received power is +4.0 − 2.76 = +1.24 dBm. The ER maximum input is −1.0 dBm. The receiver is being overdriven by more than 2 dB, and the link that passed the power budget with 7 dB to spare will not run.
The fix is an inline optical attenuator. Adding a 3 dB pad and rerunning both checks:
10GBASE-ER, 1550 nm, worst case path
Launch (minimum) -4.70 dBm
│
│ patch cord to panel, mated pair -0.50 dB
│ pigtail fusion splice -0.10 dB
│
│ 12 km G.652.D at 1550 nm -2.64 dB
│ 2 mid-span fusion joins -0.20 dB
│
│ pigtail fusion splice -0.10 dB
│ panel to patch cord, mated pair -0.50 dB
│ inline attenuator -3.00 dB
▼
Arrives -11.74 dBm
Receiver sensitivity -15.80 dBm
────────────────────────────────────────────────
Margin 4.06 dB
That 0.76 dB of overload headroom is uncomfortably tight, and a 4 dB pad would open it up. The catch is that every dB you add to protect the top of the budget comes straight out of the 4.06 dB of margin at the bottom. Balancing those two constraints against each other is the whole craft of the exercise, and it is why a loss budget cannot be reduced to a single number.
On a passive optical network the splitter dominates everything else. Splitting light N ways costs at least 10 × log₁₀(N) dB before you account for any imperfection in the device, and real splitters add an excess loss on top of that ideal figure.
| Split ratio | Ideal loss (dB) | Typical specified maximum (dB) |
|---|---|---|
| 1:2 | 3.0 | 3.6 to 4.0 |
| 1:4 | 6.0 | 7.2 to 7.4 |
| 1:8 | 9.0 | 10.5 to 10.8 |
| 1:16 | 12.0 | 13.5 to 14.2 |
| 1:32 | 15.0 | 16.8 to 17.7 |
| 1:64 | 18.0 | 20.5 to 21.5 |
GPON gives you a defined budget to spend. ITU-T G.984.2 sets Class B+ at 13 to 28 dB and Class C+ at 17 to 32 dB, and those two numbers are the floor and the ceiling of the same window described at the top of this article.
A 1:32 split on a B+ system consumes about 17 dB of a 28 dB budget on the splitter alone. That leaves roughly 11 dB for all the fibre, splices and connectors across both the feeder and the distribution side. On a sparse regional network that is not much, which is why the split ratio and the reach are a single joint decision rather than two independent ones.
A budget with no margin has already failed, because it assumes the link on the day it is commissioned is the link you will have in ten years. It will not be.
Hold margin for four things:
A total system margin of about 3 dB is the widely used working figure and a sensible default. On a long outside plant route with poor access, budget more. Holding an extra 2 dB at design time costs nothing next to a link that cannot survive its second repair.
The budget is a prediction. Two instruments check it, and they answer different questions.
An optical loss test set measures end to end insertion loss with a known source at one end and a power meter at the other. It measures exactly the quantity the budget predicts, so it is the instrument that verifies the budget, and it is the acceptance test.
An OTDR fires a pulse and reads the light scattered back, producing a trace along the length of the fibre. It cannot verify total link loss as authoritatively as an OLTS, but it shows you where the loss is, which the OLTS cannot. It is the fault finding and quality instrument, and it is how you catch the one bad splice at 7 km before you sign the link off.
Use both. The OLTS proves the link meets its budget, and the OTDR proves it does so for the right reasons while giving you a baseline trace to compare against after the first fault.
A fibre optic loss budget is launch power minus total link loss, checked against receiver sensitivity, with margin held back for the repairs and the ageing that will certainly come. Run it with worst case numbers, using minimum launch power and maximum component loss, or you are describing a link you will not get.
Then run it a second time at the other extreme, because a receiver given too much light fails as hard as one given too little, and on a short link with a long reach optic that is the failure you will actually meet. Hold about 3 dB of margin, fuse every joint you can reach rather than patching it, clean the end faces before you measure, and remember that above 10 Gbit/s and 40 km the power budget is only the first of two tests.
Do that and the link becomes a calculation rather than a hope, which matters most on exactly the sites where a failure means a four hour drive.
What is a fibre optic loss budget? It is the calculation that compares the optical power a transmitter launches against the power the receiver needs, after subtracting every loss along the link. If the power arriving is above the receiver sensitivity by a healthy margin, typically 3 dB, the link will work.
How do you calculate optical link loss? Add the loss of the fibre, being the length in kilometres multiplied by the attenuation in dB/km, to the loss of every connector pair, splice and splitter in the path. For example, 12 km of singlemode at 0.22 dB/km with four fusion splices at 0.1 dB and two connector pairs at 0.5 dB comes to 4.04 dB in total.
How much margin should a fibre link have? About 3 dB is the common working target. It covers future repair splices, connector ageing, temperature variation and test equipment tolerance. Long or hard to reach routes justify more.
Can a fibre receiver get too much light? Yes. Every receiver has a maximum input power as well as a minimum sensitivity, and exceeding it saturates the receiver and causes errors. This is common when a long reach optic such as 10GBASE-ER or ZR is used on a short link, and the fix is an inline optical attenuator.
Why is 1550 nm used for long links instead of 1310 nm? Because attenuation is lower. Singlemode fibre loses roughly 0.20 dB/km at 1550 nm against roughly 0.35 dB/km at 1310 nm, so the same glass reaches considerably further. The trade off is that chromatic dispersion is higher at 1550 nm, which matters on long, high speed links.
How much loss does a splitter add? At least 10 × log₁₀(N) dB, so about 15 dB for a 1:32 split, plus the excess loss of the device itself, which brings a typical 1:32 splitter to about 17 dB. On a GPON Class B+ budget of 28 dB, that leaves roughly 11 dB for everything else.
What is the difference between an OLTS and an OTDR? An optical loss test set measures the total end to end loss and is what you use to verify the budget. An OTDR shows where along the fibre the loss occurs and is what you use to find faults and check splice quality.
A single column of subtractions in a spreadsheet answers one narrow question, which is whether one path, at one wavelength, passes at one set of assumptions. A real design has to hold several of these in step at once: both ends of the budget, two wavelengths, the worst path on a split rather than the average one, the dispersion check, and the link as it will be after the next cable strike adds two splices to it. Change one input and six sums have to be reworked, which is where the errors creep in.
Build it in noIM₃
The fibre optic system designer builds the full link, runs the power budget at both the sensitivity and the overload limits, and tracks the margin as you change the route. The connector and splice loss calculator totals the joint losses for a given path. For passive optical networks, the PON optical budget calculator and the PON split ratio planner work out what a given split ratio leaves you for reach.
Component figures vary between manufacturers, so design against the datasheet for the parts you are specifying, and check the current, in force version of any standard before relying on it. The worked example is our own.
Related Insights
Regional Australia is facing a growing spectrum and connectivity challenge driven by sparse infrastructure, satellite dependence, and rising demand for wireless services. Here is what is really happening.
Apr 1, 2026
Noise figure is how many decibels of noise a stage adds on top of a perfect receiver, and in a chain those figures do not add. The Friis cascade formula divides every later stage by the gain ahead of it, which is why the first stage sets the system noise figure and why a lossy feeder run in front of the amplifier is the most expensive mistake on the tower. This guide covers noise figure, noise factor and noise temperature, the cascade formula, why passive loss equals noise figure decibel for decibel, a worked four stage chain that gains 2.9 dB purely by moving the LNA, how much gain is enough, the dynamic range you pay for it, Y factor measurement, and when chasing a lower noise figure buys you nothing at all.
Jul 13, 2026
Intermodulation is the interference created when two or more signals mix in a non-linear device and produce new signals at the sums and differences of their frequencies. The third-order products at 2f1 minus f2 and 2f2 minus f1 fall closest to the carriers and are the usual troublemakers. This guide explains active and passive intermodulation, where the products land, the third-order intercept point IP3 and the 3 to 1 rule, a worked multi carrier example, and the filtering, isolation and frequency planning that keep IM off a site.
Jul 11, 2026