Fiber Testing & Loss Budget
Before you light a span, you prove it can carry light. This is the math and the measurements that separate "should work" from "will work."
The test toolkit
Optical power meter (OPM)
Measures absolute received power in dBm. Paired with a light source, it gives you total end-to-end loss of a span.
Light source
A stable, known-power laser at the test wavelength. You inject at one end and read at the other with the power meter.
OTDR
Optical Time-Domain Reflectometer. Sends pulses and times the reflections to map every event along the fiber by distance.
Two ways to test — and when to use each
| Light source + power meter (insertion-loss test) | OTDR | |
|---|---|---|
| Gives you | One accurate total-loss number for the whole span | A distance map: location + loss of every connector, splice, bend, break |
| Best for | Acceptance: "does end-to-end loss meet the budget?" | Diagnosis: "where is the loss / where is the fault?" |
| Accuracy of total loss | Highest (this is the reference method) | Estimate — good for locating events, less exact for absolute end-to-end loss |
| Needs | Access to both ends | Access to one end + launch/receive cables |
Use light source + power meter to accept a span (does it pass the budget?), and OTDR to find where a problem is. They answer different questions — good turn-ups use both.
Tier 1 vs Tier 2 (the industry shorthand)
Structured-cabling standards name these two levels of certification. You'll hear them constantly:
| Tier 1 — OLTS | Tier 2 — OTDR | |
|---|---|---|
| Instrument | Optical Loss Test Set (calibrated source + power meter, often paired) | Optical Time-Domain Reflectometer |
| Produces | One true end-to-end insertion loss in dB (plus length) | A distance-vs-loss trace — every event located and characterized |
| Question answered | "Does the whole span pass the budget?" | "What is each event, and where?" |
| Truth of total loss | Definitive (this is the reference method) | Estimate — good for events, not the final acceptance number |
| Access | Both ends, simultaneously | One end (+ launch/receive fiber) |
The OLTS is the scale that weighs the whole package — one trustworthy number. The OTDR is the X-ray — it shows you what's inside and exactly where. Acceptance is decided by the OLTS number; the OTDR tells you why that number is what it is and where to fix it.
What causes loss
- Fiber attenuation: the glass itself loses light per kilometer. Typical single-mode is about 0.25 dB/km at 1550 nm (varies by wavelength and fiber).
- Connector loss: each mated connector pair loses power — plan roughly 0.3 dB each (dirty/poor ones lose far more).
- Splice loss: fusion splices are low, around 0.1 dB each; mechanical splices are higher.
- Patch-panel loss: every panel port is a connector pair — count them.
- Mux/demux loss: passive WDM filters have real insertion loss, often several dB total (add + drop).
- Macrobends: a fiber bent too tightly leaks light. A cable-tie cinched hard or a sharp corner can add loss and even show as an OTDR event.
Fiber attenuation by wavelength
Attenuation depends heavily on the wavelength. This is why long-haul DWDM lives at 1550 nm — the glass simply loses less light there.
| Wavelength | Typical SMF attenuation | Notes |
|---|---|---|
| 1310 nm | 0.33–0.35 dB/km | Higher loss. Common on grey/client optics and CWDM O-band. |
| 1550 nm (C-band) | 0.19–0.25 dB/km | Lowest loss. DWDM workhorse. Use 0.25 dB/km for conservative budgets. |
| ~1383 nm | high (spike) | Water-peak in legacy fiber; low-water-peak (OS2) fiber suppresses it. |
Budget rule of thumb: use 0.25 dB/km @ 1550 nm unless you have the actual fiber data. It's slightly pessimistic, which is what you want in a margin calculation.
Loss per element (planning values)
| Element | Plan for | Comment |
|---|---|---|
| Connector (mated pair) | 0.3–0.75 dB | Use 0.3 dB for a clean reference-grade pair; poor/dirty ones far exceed 0.75. |
| Fusion splice | 0.05–0.1 dB | The good kind. Permanent, low loss, low reflection. |
| Mechanical splice | 0.1–0.3 dB | Higher loss and reflection than fusion; temporary/field-repair use. |
| Patch-panel pass-through | ≈ one connector pair | Every panel port you cross is a mated pair — count it. |
| Mux/demux (WDM filter) | several dB total | Real insertion loss; take add + drop from the component data sheet. |
Launch and receive cables (OTDR)
An OTDR is "blind" for the first stretch of fiber (the dead zone) and can't properly measure the very first or very last connector. A launch cable (before the span) and a receive cable (after it) push those blind spots outside your span so the OTDR can measure your real first and last connectors.
How an OTDR "sees" the fiber
An OTDR fires pulses of light down the fiber and listens to what comes back. Two things return: a faint, continuous Rayleigh backscatter from the glass itself (this is the sloping baseline that reveals per-km attenuation), and sharp Fresnel reflections wherever the light hits a glass-to-air boundary (connectors, mechanical splices, breaks, the far end).
| Event on the trace | Looks like | Usually is… |
|---|---|---|
| Reflective event | A spike up, then a drop | Connector, mechanical splice, or an open/broken end. |
| Non-reflective event | A step down, no spike | Fusion splice or a macrobend. |
| Sloped baseline | Gentle downhill line | Normal fiber attenuation (dB/km). |
| End reflection | Big spike, then noise floor | The far end (or a break, if it's not where you expect). |
A splice between two fibers of slightly different backscatter can read as a gain ("gainer") from one direction and extra loss from the other — neither is the true value. Shoot the span from both ends and average the two readings to get the real splice loss. A one-direction splice number is not trustworthy on its own.
Chasing phantom "events" that repeat at even distance intervals. Strong reflections (often from a UPC or an open connector) bounce back and forth and create ghost echoes further down the trace. They aren't real fiber events. Using APC connectors, launch/receive fiber, and the right pulse width suppresses them.
The loss-budget formula
A loss budget is just an addition problem. Add every source of loss along the path, add an engineering margin for the future, and compare the total against the optical system's allowed budget.
Total Expected Loss = fiber distance loss + connector loss + splice loss + patch-panel loss + mux/demux loss + engineering margin
Worked example — a 40 km path
A single-mode span, 40 km long, running through DWDM mux/demux at each end. Here is the budget, line by line.
| Loss source | Calculation | Loss (dB) |
|---|---|---|
| Fiber distance loss | 40 km × 0.25 dB/km | 10.0 |
| Connector loss | 4 connectors × 0.3 dB | 1.2 |
| Splice loss | 8 splices × 0.1 dB | 0.8 |
| Mux/demux loss | combined add + drop | 3.5 |
| Engineering margin | headroom for aging/repairs | 3.0 |
| Total Expected Loss | 18.5 | |
If the optical system's budget is 24 dB, then:
Remaining margin = 24 − 18.5 = 5.5 dB. Positive margin means the link is viable, with 5.5 dB of headroom for future splices, connector aging, and repairs. If the number had come out negative, you'd need amplification, fewer/better connections, or a shorter/cleaner path.
Worked example 2 — a 10 km metro span
A short single-mode metro run, 10 km, two patch panels each end, a couple of field splices, no WDM filters (grey optics). Wavelength 1550 nm.
| Loss source | Calculation | Loss (dB) |
|---|---|---|
| Fiber distance loss | 10 km × 0.25 dB/km | 2.5 |
| Connector loss | 6 connector pairs × 0.3 dB | 1.8 |
| Splice loss | 4 fusion splices × 0.1 dB | 0.4 |
| Engineering margin | headroom | 3.0 |
| Total Expected Loss | 6.7 | |
Against a modest 12 dB grey-optic budget: 12 − 6.7 = 5.3 dB margin. Comfortable. Notice that on a short span the connectors, not the fiber, dominate the loss (1.8 dB of connectors vs 2.5 dB of glass) — which is why connector cleanliness matters even on "easy" short links.
Worked example 3 — an 80 km DWDM span
A long single-mode DWDM span, 80 km, mux/demux at each end, several splices from the outside-plant build. 1550 nm.
| Loss source | Calculation | Loss (dB) |
|---|---|---|
| Fiber distance loss | 80 km × 0.25 dB/km | 20.0 |
| Connector loss | 4 connector pairs × 0.3 dB | 1.2 |
| Splice loss | 12 fusion splices × 0.1 dB | 1.2 |
| Mux/demux loss | combined add + drop | 3.5 |
| Engineering margin | headroom | 3.0 |
| Total Expected Loss | 28.9 | |
Against a 24 dB passive budget: 24 − 28.9 = −4.9 dB. Negative — the link fails as a passive design. Your options: add an EDFA / optical amplifier, use higher-power/higher-sensitivity optics (a bigger system budget), reduce loss (fewer/cleaner connections), or accept a shorter reach. This is the distance where amplification stops being optional.
When loss is no longer the limit
On short/moderate spans, loss is the wall. Push distance and bit-rate up and other impairments become the real limiter — you can have plenty of power and still not close the link:
- Chromatic dispersion (CD): different wavelengths travel at slightly different speeds, smearing the pulse. Measured in ps/(nm·km) — it accumulates with distance and bites harder at higher bit-rates.
- Polarization-mode dispersion (PMD): the two polarization states arrive at slightly different times; a fiber-quality issue that matters on long, high-rate spans.
- OSNR (optical signal-to-noise ratio): each amplifier adds noise. On amplified/long systems the receiver often runs out of OSNR before it runs out of power — so the budget becomes an OSNR budget, not just a loss budget.
Coherent optics + forward error correction (FEC) rewrite the reach story. Coherent receivers compensate large amounts of CD/PMD electronically, and FEC buys back several dB of effective sensitivity. A modern 100G/400G coherent wavelength can cross spans that would be impossible for a simple direct-detect optic at the same loss — verify the specific reach against the transponder's data sheet (and against current Ekinops documentation for Ekinops-specific optics).
Reading a power level at turn-up
Once the link is lit, you read received power (dBm) at the transceiver and compare it to the optic's receiver window — the range between its sensitivity (minimum it can decode) and its overload (maximum before it saturates). You're aiming to land comfortably inside that window, not just above the floor.
Too hot (near/over overload)
Reading is higher than expected (e.g. −2 dBm on an optic that overloads at −3). The receiver saturates and throws errors. Fix with an inline optical attenuator of the right value, or check for a missing expected loss (did you skip a mux stage?).
Too cold (near/under sensitivity)
Reading is lower than the budget predicted. Suspect a dirty connector, an extra/bad splice, a bend, or a wrong-wavelength optic. Compare against your loss budget — if you're 3 dB low, something is eating ~half your power.
Celebrating "link up" without reading the power. A link can come up while sitting 1 dB from the sensitivity floor — it works today and drops the first cold night or after one connector ages. Record the actual Rx power and confirm it has margin on both sides of the window.
Budget checker (fill in your own span)
Enter your span's numbers below. The total and remaining margin update as you type. It's the same addition as the worked examples — the calculator just does the arithmetic. (Self-contained; nothing leaves your browser.)
| Loss source | Your inputs | Loss (dB) |
|---|---|---|
| Fiber distance loss | km × dB/km | — |
| Connector loss | pairs × dB | — |
| Splice loss | splices × dB | — |
| Patch-panel loss | ports × dB | — |
| Mux/demux loss | dB (add + drop) | — |
| Engineering margin | dB (typically 2–3) | — |
| Total Expected Loss | — | |
| System budget dB − Total = remaining margin | — | |
| — | ||
Defaults reproduce the 40 km worked example (18.5 dB total, 5.5 dB margin). Change any field to model your own span.
Designing to zero margin. A span that "just passes" at turn-up will fail later — connectors age, repairs add splices, and temperature shifts loss. Always carry engineering margin (commonly 2–3 dB) and treat a link with no margin as not viable.
- No launch (or receive) fiber: the first and last connectors sit in the dead zone and go unmeasured — the two most likely bad connectors on the whole span are the ones you can't see. Always use launch and receive fiber.
- Unidirectional splice reading: trusting a one-direction OTDR splice number. Backscatter differences produce "gainers" and inflated losses. Average both directions for the true value.
- Dirty reference cords: setting your OLTS reference with contaminated test jumpers. Every later measurement is then offset by that dirt — or worse, you contaminate the ports you test. Inspect and clean reference cords before referencing, and re-verify the reference periodically.
- Wrong test wavelength: testing at 1310 nm and budgeting at 1550 nm (or vice versa). Loss and bend-sensitivity differ by wavelength — test at the wavelength(s) the service will actually run.
1. Your budget says 12 dB but the OTDR shows a single 2.5 dB event at 8 km. What is it likely to be?
2. Total expected loss is 22 dB and the system budget is 24 dB. Ship it?
3. Which test do you reach for to locate a fault, not just measure it?
Try it: read an OTDR trace
Pick a scenario and read the trace the way you would in the field — reflective spikes at connectors, small non-reflective steps at splices, and either a clean end reflection or a drop to the noise floor for a cut. The interpretation names the event and the fix.