Core and cladding
A fibre is a glass core inside a glass cladding with a slightly lower refractive index, 125 µm across in total. A plastic coating takes it to 250 µm, or 200 µm for some newer fibres.
Fibre optics
Fibre optics, from light to link
Fibre carries almost every link I design beyond the desk: between floors, between racks and between sites. This page explains how light travels in single mode and multimode fibre. It then covers fibre grades, reach, cables, connectors, polarity, transceivers, splicing, loss budgets, testing and safety, and ends with a fault finder.
How fibre works
Light in glass, and why it reaches so far
Fibre carries data as pulses of light, usually infrared, through a glass core thinner than a hair. Everything else on this page, from fibre types to loss budgets, follows from two ideas: how many paths the light can take, and how much of it is lost on the way.
Swipe sideways to see the whole diagram.
Orange packets take different paths through the multimode core and arrive at different times. That spread, called modal dispersion, is what limits multimode reach. The teal packets in single mode share one path, which is why single mode reaches tens of kilometres.
Total internal reflection
Light that meets the cladding at a shallow enough angle reflects back into the core every time, so it follows the fibre round gentle bends. A tight bend lets light escape. That loss is a macrobend.
Modes
A mode is one path light can take. A 50 µm or 62.5 µm core carries many modes, so it is multimode. A 9 µm core carries one, so it is single mode.
Dispersion
A pulse spreads as it travels. In multimode fibre most of the spread comes from modes arriving at different times. In single mode, colours travelling at slightly different speeds cause it (chromatic dispersion).
Attenuation
Glass absorbs and scatters a little light per km. It is measured in dB (decibels): 3 dB is half the light, and 10 dB is a tenth. Connectors and splices add more.
Why fibre
Fibre carries more bandwidth over longer distances than copper, ignores electrical noise and lightning, and needs no earthing between buildings. It is also light and thin, which matters in a full tray.
The wavelengths in use
Swipe sideways to see the whole diagram.
850 nm. The multimode window. Cheap VCSEL (vertical cavity surface emitting laser) optics work here, and almost every short reach data centre link uses it. Loss is high, about 2.5 dB per km in practice, but the links are short.
850 to 953 nm. OM5 fibre is specified across this band, so shortwave WDM (wavelength division multiplexing) and BiDi (bidirectional) optics can put two or four wavelengths on each fibre.
1310 nm, the O band. Standard single mode fibre has almost no chromatic dispersion here, so LR (long reach) optics and most data centre single mode optics use it. Loss is about 0.35 dB per km.
The water peak, 1383 nm. Older fibre lost a lot of light here because of water in the glass. Low water peak fibre (ITU-T G.652.D) removed most of it, which opened the whole band for CWDM (coarse WDM) from 1271 to 1611 nm.
1550 nm, the C band. The lowest loss, about 0.2 dB per km, and the band optical amplifiers work in. Long haul links, DWDM (dense WDM) and coherent long reach optics live here.
PON (passive optical network) wavelengths. GPON sends downstream at 1490 nm and upstream at 1310 nm. XGS-PON uses 1577 nm down and 1270 nm up, so both can share one fibre.
| Wavelength | Fibre | What uses it | Loss |
|---|---|---|---|
| 850 nm | Multimode | VCSEL lasers, the cheapest optics | About 2.5 dB/km in practice; 3.5 dB/km is the long standing TIA (Telecommunications Industry Association) maximum |
| 1300 nm | Multimode | The older LED (light emitting diode) window | About 0.6 dB/km; 1.5 dB/km maximum |
| 1310 nm | Single mode (O band) | Low dispersion: most data centre single mode optics | About 0.35 dB/km (typical) |
| 1490 and 1577 nm | Single mode | PON downstream (GPON, XGS-PON) | About 0.2 to 0.25 dB/km (typical) |
| 1550 nm | Single mode (C band) | Lowest loss; long reach and DWDM | About 0.2 dB/km (typical) |
Loss figures marked typical are what good modern fibre measures. The maximums are the values cabling standards allow, which I use for worst case design. Copper links and the 100 m rule are on the copper cabling page.
Fibre types
OM1 to OS2, and what each one is for
Multimode fibre is graded OM1 to OM5 by its modal bandwidth: how well it keeps a laser pulse together. Single mode is graded by loss. The grade, more than anything else, sets how far a link can reach.
Swipe sideways to see the whole diagram.
OM1. A 62.5 µm core from the LED era. It cannot be joined to 50 µm fibre without a large loss, and 10G reaches only 33 m. Where I find it, I plan to replace it.
OM2. The first 50 µm grade. It still carries 1G for 550 m, but its low modal bandwidth limits 10G to 82 m.
OM3. Laser optimised 50 µm fibre, made for the VCSEL lasers in short reach optics. It carries 10G for 300 m and 100G for 70 m.
OM4. A higher modal bandwidth version of OM3, with more reach at every speed: 400 m at 10G and 100 m at 100G. Most jackets are aqua, and some makers in Europe use violet.
OM5. Specified from 850 to 953 nm, so BiDi and shortwave WDM optics can run several wavelengths per fibre. At a single 850 nm wavelength it behaves like OM4.
OS2. Single mode fibre with a maximum loss of 0.4 dB per km. Reach is set by the optics, from 500 m to 80 km and beyond. Jackets are yellow.
| Type | Core | Modal bandwidth or loss | Jacket (TIA-598) | Where it fits |
|---|---|---|---|---|
| OM1 | 62.5 µm | 200 MHz·km | Orange | Legacy only |
| OM2 | 50 µm | 500 MHz·km | Orange | Legacy 1G |
| OM3 | 50 µm | 2000 MHz·km | Aqua | 10G to 400G short reach |
| OM4 | 50 µm | 4700 MHz·km | Aqua (violet from some makers) | 10G to 400G short reach, longer |
| OM5 | 50 µm | 4700 at 850 nm, 2470 at 953 nm | Lime green | Several wavelengths per fibre |
| OS1a | 9 µm | Not mode limited; 1.0 dB/km maximum | Yellow | Indoor single mode |
| OS2 | 9 µm | Not mode limited; 0.4 dB/km maximum | Yellow | Indoor, outdoor and long distance |
How I choose
For new work I fit OS2 for anything between buildings, between floors over about 100 m, and for most new links above 100G. Inside a data centre row, OM4 still gives the cheapest short reach optics.
OM classes and OS classes
ISO/IEC 11801 defines the OM (optical multimode) and OS (optical single mode) classes. OS1a, from the 2017 edition, is the indoor single mode class at 1.0 dB per km, and OS2 is the low loss class at 0.4 dB per km.
Bend insensitive fibre
ITU-T G.657 fibre tolerates tighter bends than standard G.652.D. A G.657.A1 fibre is rated down to 10 mm radius, A2 to 7.5 mm and B3 to 5 mm, which suits tight trunking and home installs.
The print beats the colour
Jacket colours are a convention, not a guarantee. TIA allows other colours when the jacket print states the fibre type, so I read the print and the drawings before I patch.
Reach
How far each optic goes on each fibre
Every optic is specified for a reach on each fibre grade. Faster multimode optics reach less, because the pulses are shorter and spread matters more. Single mode reach is set by the optic, from 500 m to tens of kilometres.
| Optic | OM1 | OM2 | OM3 | OM4 | OM5 | OS2 |
|---|---|---|---|---|---|---|
| 1000BASE-SX | 275 m | 550 m | 1000 m* | 1000 m* | 1000 m* | No |
| 1000BASE-LX | No | No | No | No | No | 5 km |
| 10GBASE-SR | 33 m | 82 m | 300 m | 400 m | 400 m | No |
| 10GBASE-LR | No | No | No | No | No | 10 km |
| 25GBASE-SR | No | No | 70 m | 100 m | 100 m | No |
| 40GBASE-SR4 | No | No | 100 m | 150 m | 150 m | No |
| 40G BiDi | No | No | 100 m | 150 m | 200 m | No |
| 100GBASE-SR4 | No | No | 70 m | 100 m | 100 m | No |
| 100G BiDi or SWDM4 | No | No | 70 m | 100 m | 150 m | No |
| 100GBASE-DR | No | No | No | No | No | 500 m |
| 100GBASE-FR1 | No | No | No | No | No | 2 km |
| 100GBASE-LR4 | No | No | No | No | No | 10 km |
| 400GBASE-SR8 | No | No | 70 m | 100 m | 100 m | No |
| 400GBASE-SR4.2 | No | No | 70 m | 100 m | 150 m | No |
| 400GBASE-DR4 | No | No | No | No | No | 500 m |
| 400GBASE-FR4 | No | No | No | No | No | 2 km |
| 400GBASE-LR4-6 | No | No | No | No | No | 6 km |
Reach is the IEEE 802.3 or MSA (multi source agreement) figure for the standard optic. * 1000BASE-SX on OM3 and above: the standard specifies OM1 and OM2 only, and vendors quote about 1 km. Extended reach optics from vendors go further, at a price. 1000BASE-LX is specified to 5 km, and many 1000BASE-LX optics are rated to 10 km.
Try it: will this link reach?
Fits. 10GBASE-SR on OM4 is specified to 400 m, so a 250 m link has 150 m to spare. Check the loss budget too: reach assumes the standard allowance for connectors.
- Standard reach
- 400 m
- Your link
- 250 m
- Headroom
- 150 m
Cables
What is inside a fibre cable
The fibre is the same in every cable. What changes is how it is protected: from pulling, temperature, water, crushing and fire. The construction decides where a cable can go and how fast it can be terminated.
Swipe sideways to see the whole diagram.
Tight buffered. Easy to work with and to terminate, but the glass shares the strain when the cable is pulled or gets cold. I use it inside buildings.
Loose tube. The tubes protect the fibres from strain and temperature. Each tube and each fibre follows the colour code, so tube 2, fibre 3 is orange tube, green fibre.
Ribbon. The densest construction. Mass fusion splicing joins a whole ribbon in one go, which saves days on a large build.
Armoured. Heavier and stiffer, with a larger bend radius. The metal armour needs earthing and bonding where it enters a building.
| Construction | What it is | Where I use it |
|---|---|---|
| Tight buffered | 900 µm buffer on each fibre | Inside buildings, risers, patch cords |
| Loose tube | Fibres in gel filled or dry tubes | Outdoors, ducts, between buildings |
| Ribbon | 12 fibres bonded side by side | High fibre counts and MPO (multi fibre push on) trunks |
| Armoured | Steel tape or corrugated steel | Direct burial and rodent risk |
| Blown fibre | Fibre units blown into microducts later | Campuses that will grow |
The fibre colour code
- 1 Blue
- 2 Orange
- 3 Green
- 4 Brown
- 5 Slate
- 6 White
- 7 Red
- 8 Black
- 9 Yellow
- 10 Violet
- 11 Rose
- 12 Aqua
TIA-598 sets the order for fibres in a tube and for tubes in a cable. After twelve, the sequence repeats with a stripe or ring mark. I record tube and fibre numbers on every splice and patch schedule, never just a colour.
Indoor fire performance
Indoor cable in the UK and EU is rated under the CPR (Construction Products Regulation). LSZH (low smoke zero halogen) sheaths give off less smoke and no acid gas. The ratings are on the copper cabling page.
Outdoor cable indoors
Outdoor polyethylene sheaths are not made for long indoor runs. I change to an indoor or universal cable at a joint near the building entry, as the fire strategy requires.
Bend radius
I keep to the maker's bend radius: typically 10 times the cable diameter when it is not under load, and 20 times while it is being pulled. Bend insensitive fibre helps, but the cable still has limits.
Fibre counts
Cables come in 12, 24, 48, 96, 144 fibres and more. I size for today's need plus spare fibres, because pulling a second cable later costs far more than the extra fibres.
Connectors
LC, SC, MPO and the rest, and how they meet
A connector holds the fibre in a ceramic ferrule and presses two ferrules end to end, so the cores line up within a micron or so. Most losses and most faults in a fibre link happen at connectors. The LC (Lucent connector) is the one I handle most.
Swipe sideways to see the whole diagram.
LC. The small form factor connector on almost every pluggable optic. A duplex LC carries transmit and receive, and the clip often lets me swap the pair if the polarity is wrong.
SC. Larger and easy to handle. Still common in older patch panels, and in fibre to the premises, where the APC (angled physical contact) version is the norm.
ST. A bayonet connector from older multimode networks. I still meet it on plant floors and in older riser panels.
FC. A threaded connector that holds firm under vibration. Mostly on test gear and older single mode equipment now.
MPO. Many fibres in one rectangular ferrule, for parallel optics and for trunks that break out to LC. Pinned and unpinned ends must pair correctly, and the key sets polarity.
SN and MDC. VSFF (very small form factor) connectors put a duplex pair in a body about a third the width of a duplex LC. Several fit in one QSFP-DD port, so a 400G optic can break out without a cassette.
UPC and APC. UPC (ultra physical contact) end faces reflect a little light straight back. APC end faces are angled at 8 degrees, so the reflection leaves the core. APC is green; UPC single mode is blue.
| Connector | Ferrule | Coupling | Where I see it |
|---|---|---|---|
| LC (Lucent connector) | 1.25 mm | Push and latch | Most duplex pluggable optics, patch panels |
| SC (subscriber connector) | 2.5 mm | Push pull | Older panels, PON and FTTP (fibre to the premises) |
| ST (straight tip) | 2.5 mm | Bayonet | Legacy multimode, industrial |
| FC (ferrule connector) | 2.5 mm | Threaded | Test equipment, older single mode |
| MPO and MTP | 12, 16 or 24 fibres per row | Push pull, pinned or unpinned | Parallel optics and trunks |
| SN, MDC and CS | 1.25 mm, duplex | Push pull | 400G and 800G breakout, very dense panels |
Connector and adapter colours
| Colour | Meaning |
|---|---|
| Beige | Multimode 62.5 µm (OM1) |
| Black | Multimode 50 µm (OM2) |
| Aqua | Multimode OM3 and OM4 |
| Lime green | Multimode OM5 |
| Blue | Single mode UPC |
| Green | Single mode APC |
Return loss matters more on single mode
Light reflected back into a single mode laser upsets it, and reflections ruin analogue video and PON. That is why service provider and FTTP work uses APC, and why I never mix UPC and APC in one mated pair.
Pinned and unpinned MPO
Two MPO connectors only mate if one has guide pins and the other does not. Transceivers are usually pinned, so the patch cords to them are unpinned. I record pin state on every MPO schedule.
SN and MDC are not the same
Both put two 1.25 mm ferrules in a narrow body, and they look alike. They do not mate with each other, so I standardise on one per site.
Field or factory
Factory polished, pre terminated cables give the lowest loss and the fastest install. Field terminations, by splicing a factory made pigtail, suit cables whose length is not known in advance.
Polarity
Making sure transmit meets receive
Every fibre link sends on one fibre and receives on another. Polarity is the plan that makes sure each transmitter reaches the right receiver, through every cord, panel and trunk. Most "no light" faults on new installs are polarity.
Swipe sideways to see the whole diagram.
Light leaves switch A on its transmit fibre and must arrive on switch B's receive fibre, and the same the other way. Every duplex link is two one way paths.
In the TIA A to B scheme, a duplex patch cord connects position A at one end to position B at the other. So each cord crosses the pair once.
The installed cabling between panels is also built A to B, so it crosses once more. Panels and adapters keep the order they are given.
Cord, link, cord: three crossings, an odd number. Transmit lands on receive at both ends and the link comes up.
Swap one cord for an A to A cord and there are two crossings. Transmit meets transmit, both receivers see no light, and the ports stay down.
Try it: does transmit meet receive?
The link comes up. Three crossings, an odd number, so transmit lands on receive at both ends.
- Crossings
- 3
- Switch A transmit lands on
- Switch B receive
MPO trunks and polarity methods
Swipe sideways to see the whole diagram.
Method A. A Type A trunk keeps the order: fibre 1 to fibre 1. One end is key up and the other key down. The pair is flipped by using an A to B cord at one end and an A to A cord at the other, so two kinds of cord are needed.
Method B. A Type B trunk reverses the order: fibre 1 to fibre 12. Both ends are key up, which suits parallel optics such as SR4 and DR4. A to B cords work at both ends. Key up to key up does not suit angled APC MPO connectors.
Method C. A Type C trunk flips each pair inside the cable, so A to B cords work at both ends for duplex links. It does not suit parallel optics, so I avoid it for new builds. TIA-568.3-E adds universal methods, U1 and U2, built on Type B trunks.
The rule I check
For a duplex path, count every crossing from one transceiver to the other. An odd number works. An even number puts transmit on transmit, and both ports stay dark.
Parallel optics
An SR4 or DR4 optic transmits on fibres 1 to 4 and receives on 9 to 12. A Type B path, key up at both ends, maps 1 to 12 and lands every transmit lane on a receive lane.
One method per site
Mixing methods is how polarity faults start. I set one method in the design, record it on every schedule, and order cassettes and cords to match.
Fix it at the cord
When a duplex link has the wrong polarity, I swap the pair at one patch cord, which many LC clips allow, rather than changing the panel. Then I update the record.
Transceivers
The optic at each end, from SFP to OSFP
A transceiver, or optic, turns the switch's electrical lanes into light and back. Its form factor sets the speed, its optical type sets the fibre and reach, and both ends must match.
Swipe sideways to see the whole diagram.
Orange packets are the transmit path from the switch ASIC (application specific integrated circuit), teal the receive path, and purple the DOM (digital optical monitoring) readings the switch polls. The retimer or DSP (digital signal processor) cleans up the electrical signal. Short reach optics use cheap VCSEL lasers at 850 nm; single mode optics use DFB (distributed feedback) or EML (electro absorption modulated) lasers.
| Form factor | Electrical lanes | Common speeds | Optical connector |
|---|---|---|---|
| SFP (small form factor pluggable) | 1 lane | 1G | Duplex LC |
| SFP+ | 1 lane | 10G | Duplex LC |
| SFP28 | 1 lane | 25G | Duplex LC |
| SFP56 | 1 lane, PAM4 (four level pulse amplitude modulation) | 50G | Duplex LC |
| QSFP+ (quad SFP) | 4 × 10G | 40G | Duplex LC or MPO-12 |
| QSFP28 | 4 × 25G | 100G | Duplex LC or MPO-12 |
| QSFP56 | 4 × 50G, PAM4 | 200G | Duplex LC or MPO-12 |
| QSFP-DD (double density) | 8 × 50G or 8 × 100G, PAM4 | 400G, 800G | Duplex LC, MPO, SN or MDC |
| OSFP (octal SFP) | 8 × 50G or 8 × 100G, PAM4 | 400G, 800G | As QSFP-DD; larger, with room for more cooling |
PAM4 (four level pulse amplitude modulation) sends two bits per symbol, where NRZ (non return to zero) sends one. It doubles the data rate per lane, and it is why 50G and 100G lanes need forward error correction. DAC (direct attach copper), AOC (active optical cable) and breakout cables are on the data centre cabling page.
Reading the name of an optic
| Code | Meaning | What it tells me |
|---|---|---|
| SR | Short reach | Multimode at 850 nm: tens to hundreds of metres |
| DR | Data centre reach | Single mode, 500 m, often parallel (DR4 uses four fibre pairs) |
| FR | Far reach | Single mode, 2 km, usually four wavelengths on one pair (FR4) |
| LR | Long reach | Single mode, 10 km (400GBASE-LR4-6 is 6 km) |
| ER | Extended reach | Single mode, 40 km |
| ZR | The 80 km class | Single mode, coherent optics for data centre interconnect |
| BiDi | Bidirectional | Two wavelengths on each fibre, so 40G or 100G runs on one duplex pair |
| SWDM4, CWDM4 | Wavelength multiplexing | Four wavelengths on one duplex pair: SWDM (shortwave WDM) on multimode, CWDM near 1310 nm on single mode |
The number after the code counts lanes: four fibre pairs, or four wavelengths. A suffix such as .2 means two wavelengths per fibre, as in 400GBASE-SR4.2.
What the DOM tells me
| Reading | What it measures | What I look for |
|---|---|---|
| Temperature | Inside the module | High values point to poor airflow or a failing module |
| Supply voltage | From the host | Rarely the problem, but worth a look |
| Laser bias current | Drive current to the laser | Rising over months means the laser is ageing |
| Transmit power | Light leaving the module, in dBm | Low means a failing laser |
| Receive power | Light arriving, in dBm | Low means loss on the path: dirt, bends or distance. Too high on short single mode runs can overload the receiver |
Coding and compatibility
Switches read the module's EEPROM (electrically erasable memory) and some refuse optics they do not recognise. I buy optics coded for the platform, and check the vendor's compatibility matrix for the exact switch and software release.
Both ends must match
An SR optic only talks to an SR optic on the same fibre type. A 100GBASE-LR4 will not talk to a 100GBASE-FR1, even though both are single mode. I put the optic type on every link record.
Heat and power
A 400G or 800G optic can draw over 10 W (typical), so a switch full of them needs the airflow its design assumes. Blanking plates in empty ports matter.
Wavelength division
CWDM (coarse wavelength division multiplexing) spaces wavelengths 20 nm apart from 1271 to 1611 nm. DWDM (dense WDM) packs them much closer in the C band. Both put many links on one fibre pair.
Splicing
Joining fibre so the light barely notices
Fibre is joined by splicing, either fused into one piece of glass or held end to end in a sleeve. Splices join cable lengths, fit connectors in the field, and repair breaks.
Swipe sideways to see the whole diagram.
I strip the 250 µm coating from both ends, leaving bare 125 µm glass. A thermal or mechanical stripper keeps the glass unscratched.
I wipe the bare glass with lint free tissue and pure alcohol, once and in one direction. Any dirt now ends up in the joint.
A precision cleaver scores and snaps the glass to leave a flat, square end. A poor cleave angle is the main cause of a high loss splice.
The splicer moves both fibres into line. Core alignment splicers use cameras to line up the cores themselves, which matters most on single mode.
An electric arc melts the two ends together into one piece of glass. The whole fuse takes a few seconds.
The splicer estimates the loss from its images. A good fusion splice is typically a few hundredths of a dB. I test the whole link afterwards, because the estimate is not a measurement.
A heat shrink sleeve with a steel rod protects the joint. It goes into a splice tray, in order, with tube and fibre numbers recorded.
| Method | How it works | Loss | Where I use it |
|---|---|---|---|
| Fusion splice | Glass melted into one piece | Typically under 0.05 dB (indicative) | Joining cables, pigtails, repairs |
| Mechanical splice | Ends held in a gel filled sleeve | A few tenths of a dB (indicative) | Quick repairs and some field kits |
| Factory pre terminated | Connectors fitted and tested in the factory | Lowest connector loss, with a test report | Data centre trunks and known lengths |
| Spliced pigtail | Factory connector on a short tail, spliced on | Factory connector loss plus one splice | Field termination of any cable |
Mass fusion
Ribbon splicers join twelve fibres in one arc. On a 144 fibre joint that is twelve splices instead of 144, which is why high count builds use ribbon cable.
Splice trays and enclosures
Every splice sits in a tray in a closure or a wall box, in colour code order, with spare fibre coiled at a safe radius. Outdoor joints use sealed closures.
Pre terminated where I can
Inside data centres I specify factory terminated trunks with test reports. Installation is faster, cleaner and the loss is known before the cable arrives.
The 0.3 dB ceiling
TIA allows up to 0.3 dB for a splice, and I use that figure for worst case design. A fusion splice measuring anywhere near it gets redone.
Loss budgets
Adding up every dB before I order cable
A link works when the light that arrives is strong enough for the receiver. A loss budget adds the loss of the fibre, every mated connector and every splice, and checks the total against what the optics allow.
Swipe sideways to see the whole diagram.
The optic sets the budget. A 100GBASE-SR4 channel on OM4 allows 1.9 dB of loss over up to 100 m. Every connector and metre of fibre must fit inside it.
Ninety metres of multimode at the standard maximum of 3.5 dB per km costs 0.32 dB. On short links the fibre itself is the small part.
Two mated pairs, one at each patch panel, at the TIA maximum of 0.75 dB each, add 1.5 dB. The total is 1.82 dB: inside the budget with almost nothing to spare.
A cross connect adds two more mated pairs. At 0.75 dB each, the link needs 3.32 dB and fails, even though it is still under 100 m.
Low loss MPO connectors rated at 0.35 dB per pair bring the same cross connected link to 1.72 dB. This is why I specify low loss parts for 40G and faster.
Try it: loss budget calculator
Tight. The link fits with only 0.08 dB to spare. A dirty connector or one more patch would push it over. Clean, test, and consider low loss parts.
- Fibre
- 0.32 dB
- Connectors
- 1.50 dB
- Splices
- 0.00 dB
- Total loss
- 1.82 dB
- Budget
- 1.90 dB
- Margin
- 0.08 dB
The numbers I design with
| Element | Standard maximum | Typical modern parts |
|---|---|---|
| Multimode fibre at 850 nm | 3.0 to 3.5 dB/km, depending on the standard and edition | About 2.5 dB/km |
| Multimode fibre at 1300 nm | 1.5 dB/km | About 0.6 dB/km |
| Single mode, indoor (OS1a) | 1.0 dB/km | About 0.35 to 0.5 dB/km |
| Single mode, OS2 | 0.4 dB/km | About 0.35 dB/km at 1310 nm, 0.2 at 1550 nm |
| Mated connector pair | 0.75 dB | About 0.2 to 0.35 dB for low loss LC and MPO |
| Splice | 0.3 dB | Under 0.05 dB for a good fusion splice |
Loss budget and power budget
The loss budget is what the cabling loses. The power budget is what the optics can tolerate: transmit power minus receiver sensitivity, less a penalty allowance. IEEE turns that into a maximum channel insertion loss for each optic.
Which connectors count
The connectors plugged into the optics at each end are part of the optic's test, so I count the mated pairs in between: panels, cassettes and cross connects.
Margin
I aim to leave at least 0.5 dB (typical) of margin on short reach multimode links and more on long single mode ones, for ageing, repairs and the dirt nobody admits to.
Too little loss
On short single mode links with long reach optics, the light can be too strong and overload the receiver. The data sheet gives a maximum receive power; an attenuator fixes it.
Testing
Clean it, test it, and keep the results
Dirt is the commonest cause of fibre faults, and testing is how I prove a link before anyone relies on it. The rule on every site I run is simple: inspect before you connect.
Swipe sideways to see the whole diagram.
I look at every end face with a camera based inspection scope before it is connected, on both sides of the mated pair. Here there is dirt across the core, so it fails.
A one click cleaner or a cleaning cassette wipes the end face. Dry cleaning removes most loose dust, and it is where I always start.
One spot is left on the cladding. Never assume a clean worked. The scope decides, not the cleaner.
For oil or stubborn residue, I use a cleaning fluid made for fibre and finish dry straight away. Alcohol that is left to dry leaves its own film.
The core and cladding are clear. The one remaining speck sits on the contact zone and has survived several cleans, so it counts as embedded and acceptable. Now, and only now, I connect.
| Test | Tool | When I use it | What it proves |
|---|---|---|---|
| Inspection | Camera inspection scope | Every end face, before every connection | Pass or fail per IEC 61300-3-35 |
| Tier 1 (basic) | OLTS (optical loss test set): light source and power meter | Every link at handover | Loss against the limit, plus length and polarity |
| Tier 2 (extended) | OTDR (optical time domain reflectometer) | Backbones, outside plant, any link that fails Tier 1 | Where each loss is: connectors, splices, bends, breaks |
| Fault finding | VFL (visual fault locator), red light at 650 nm | Short runs, patch cords, panels | Breaks and tight bends glow red |
Setting the reference
An OLTS is referenced before testing, so the result includes the right connectors. TIA describes one, two and three jumper methods; the one jumper method is the one I ask for, as it includes both end connections.
Encircled flux
For multimode, how the light is launched changes the result. Test sets that meet the encircled flux launch condition give repeatable results, and I ask for it in the test specification.
Both wavelengths
I test multimode at 850 and 1300 nm and single mode at 1310 and 1550 nm. The difference between wavelengths says a lot about bends.
Results are a deliverable
Every link's test results, in the tester's native file, go into the handover pack with the drawings and the schedule. They are the baseline for every fault later.
Reading an OTDR trace
Swipe sideways to see the whole diagram.
The OTDR sends pulses and measures the light scattered back. Its own connector blinds it for a few metres, the dead zone, so a launch reel of known fibre goes first.
A mated connector shows as a spike, from its reflection, followed by a step down, which is its loss. Thanks to the launch reel, the first panel can be measured.
A fusion splice shows as a small step with no spike, because it reflects almost nothing. Each event gets a distance, so the splice tray can be found.
A tight bend is a step that grows at longer wavelengths. Testing at 1310 and 1550 nm and comparing the two tells a bend from a bad splice.
A receive reel after the far panel lets the OTDR measure the last connector too. Without it, the far end would be lost in the end reflection.
The end of the fibre gives a large reflection, then the trace falls to noise. The distance to it confirms the length on the drawings.
OTDR launch reels are usually hundreds of metres long. A rule of thumb from EXFO is the pulse width in nanoseconds divided by ten, in metres, plus 20%: about 600 m for a 5,000 ns pulse.
Safety
Working safely with light and glass
Fibre is safe to handle, but it carries invisible laser light and it leaves glass splinters. These are the habits I insist on wherever fibre is worked.
Invisible light
Most fibre systems use infrared light that the eye cannot see, so a live fibre looks dark. I treat every fibre as live until a power meter says otherwise.
Never look into a fibre
I never look into a fibre end or an optic port, with or without a magnifier. A camera based inspection scope shows the end face on a screen instead.
Laser classes
IEC 60825-1 sets laser classes and IEC 60825-2 covers fibre systems. Short reach optics are usually Class 1, eye safe in normal use. Red fault locators are often Class 2 or 2M, and the M means no magnifiers.
Glass shards
Cleaved fibre ends are tiny glass splinters that pierce skin and do not show on an X ray. They go straight into a marked sharps container. I work on a dark mat, with no food or drink nearby.
Chemicals
Cleaning fluids, gel from loose tube cables and isopropyl alcohol need ventilation and gloves. I check the safety data sheets on site.
Outside plant
Work in manholes, ducts and on poles has its own risks: confined spaces, traffic and height. Those jobs follow the site rules and the right permits.
PON and outside plant
Fibre beyond the building: PON, dark fibre and ducts
Outside the data centre, fibre reaches homes and businesses through passive optical networks, and links sites over leased dark fibre. A little knowledge of both helps when a branch or campus depends on them.
Swipe sideways to see the whole diagram.
Downstream, the OLT (optical line terminal) broadcasts to every ONT (optical network terminal), and each keeps only its own encrypted traffic. Upstream, each ONT transmits in time slots the OLT grants, so the orange packets arrive one after another.
| Technology | Standard | Downstream | Upstream | Down wavelength | Up wavelength |
|---|---|---|---|---|---|
| GPON | ITU-T G.984 | About 2.5 Gb/s | About 1.25 Gb/s | 1490 nm | 1310 nm |
| XGS-PON | ITU-T G.9807.1 | About 10 Gb/s | About 10 Gb/s | 1577 nm | 1270 nm |
Split ratios of 1:32 or 1:64 are common (typical); XGS-PON allows up to 1:128. Because the wavelengths differ, GPON and XGS-PON can share the same fibre and splitters during an upgrade.
Where I meet PON
More and more UK business and home broadband arrives over FTTP (fibre to the premises). For a branch, I care about the service it delivers and its handoff, usually Ethernet, rather than the PON behind it.
Dark fibre
Dark fibre is leased glass with no service on it: I light it with my own optics. It suits data centre interconnects and campuses, and its loss and length decide the optics.
Outside plant
Between buildings, cable runs in ducts, chambers and joints. I design for spare ducts, diverse routes for resilient links, and sealed, labelled joint closures.
Diverse routes
Two fibres in the same duct are one digger away from failing together. For resilient sites I ask carriers for route maps and separate building entries. More on this is on the WAN (wide area network) page.
Troubleshooting
Finding a fibre fault from end to end
Fibre faults come down to a short list: dirt, polarity, the wrong fibre or connector, bends, breaks, and optics that do not match. Choose a symptom to see where it usually sits and what I do first.
Swipe sideways to see the whole diagram.
Dirty connector. The commonest fault of all. I inspect both ends of the mated pair, clean, inspect again and compare receive power with the baseline from handover.
Polarity. Transmit is meeting transmit. If neither end sees any light and both optics show transmit power, I count the crossings and swap the pair at one cord.
Mixed fibre types. A 62.5 µm cord in a 50 µm link, or a multimode cord on single mode, loses several dB. I read the jacket print, not the colour.
UPC mated to APC. The angled face meets a flat one, leaving an air gap. Loss is high and the end faces can be damaged, so I inspect both after correcting it.
A macrobend. A cable trapped by a door, a tight tie or an overfilled tray. The OTDR gives the distance; a VFL shows the glow on short runs.
Receiver overload. A long reach optic on a few metres of fibre can deliver more light than the receiver accepts. A fixed attenuator, sized from the data sheet, fixes it.
Mismatched optics. LR at one end and FR at the other, different wavelengths, or a different FEC (forward error correction) setting. I compare the optic type and settings at both ends.
A break. The OTDR end event gives the distance to the break, which I match to the route drawings. Then the repair: a splice in a new closure, and a fresh test.
The commands I start with
| Platform | Command | What it shows |
|---|---|---|
| Cisco IOS XE | show interfaces transceiver detail | Temperature, voltage, bias, transmit and receive power, with thresholds |
| Cisco NX-OS | show interface ethernet1/1 transceiver details | The same, per lane on multi lane optics |
| Arista EOS | show interfaces transceiver | Optical power and temperature for every port |
| Juniper Junos | show interfaces diagnostics optics et-0/0/0 | Power per lane, with alarm and warning flags |
| Linux | ethtool -m eth0 | The module's identity and DOM readings |
I compare receive power with the baseline recorded at handover. A drop of a few dB on a link that used to be fine is almost always a connector. My general method is on the troubleshooting page, and the terms are in the glossary.
Projects
Where I use it
My home lab runs on copper, so my fibre work comes from data centre, campus and estate projects. Fibre decisions sit inside most of them: which grade, which optics, how much spare, and how it is tested before handover.
A data centre from an empty shell
Delivered the £25M Network Rail data centre build from an empty shell, covering structured cabling, building management systems and mechanical and electrical integration.
AI ready data centre infrastructure
Designed AI ready data centre infrastructure for high density compute clusters, covering network, power, cooling and structured cabling, with the rack layouts and deployment documentation the build teams worked from.
Cabling for high performance workloads
Prepared data centre infrastructure for high performance workloads: rack allocation, structured cabling pathways and deployment coordination.
A Tier 3 data centre
Tier 3 data centre and managed infrastructure for over 300 customer sites on Cisco ACI (Application Centric Infrastructure) and Nexus.
A refresh across Europe, the Middle East and Africa
Audit, high and low level design and implementation of LAN (local area network), WAN, SD-WAN and data centre infrastructure, replacing end of life equipment with minimal impact on global operations.
Fibre and copper cabling
I have trained in fibre and copper structured cabling, alongside power and electrical systems, which is the grounding behind the cabling decisions on this page.
The physical layer behind every link: copper is on the copper cabling page, and trunks, breakouts and switch optics are on the data centre cabling page. See all credentials
Explore
More areas I design and deliver
Every page follows the same idea: how it works, the choices that matter, and where I have used it. The foundations pages explain the basics the others build on.