The link certifies clean on Friday and goes intermittent by Wednesday. Nothing was moved, nothing was cut, and the tester still passes it on a cool morning.
Fiber is unforgiving in a specific way. Copper degrades gradually and tells you — the error counters climb, the link trains down. Fiber either works or it does not, and the failures cluster in four places: wrong fiber type, wrong polish, a bend nobody measured, an end face nobody looked at.
Single-Mode or Multimode: Decide By Distance
The physical difference is simple. Multimode fiber has a large core, almost always 50 or 62.5 microns, which lets light travel in many paths at once. Single-mode shrinks the core to about 9 microns so only one path exists — no modal dispersion to manage, which is why its bandwidth is described as virtually infinite.
Multimode runs with LED sources at 850 and 1300 nm, and lasers at 850 and 1310 nm. Single-mode runs lasers at 1310 and 1550 nm, where loss is lower.
The attenuation difference is not small. Multimode loses 3 dB/km at 850 nm versus 1 dB/km at 1300 nm for the older grades — half the light gone in one kilometer at the short wavelength, a fifth of it at the long one. Single-mode sits at 0.4 dB/km at 1310 nm and 0.25 dB/km at 1550 nm. The physics behind that gap is scattering, which scales with the inverse fourth power of wavelength. Double the wavelength and you cut scattering loss sixteen-fold.
So the decision rule is distance and rate, not price per meter. Multimode is cheaper to terminate and cheaper to drive — short-wavelength optics cost less than long-wavelength ones. Single-mode wins as soon as the run is long, or when you want the installation to still be adequate at rates that have not been invented yet.
One warning that belongs in every design review
You cannot mix fiber types. Connecting single-mode to multimode can produce 20 dB of loss — that is 99% of the power, which is not a degraded link, it is no link. Even stepping between 62.5/125 and 50/125 multimode can cost 3 dB or more, over half the signal.
This matters most in retrofits, where the existing backbone is an older grade and somebody assumes a patch cord will bridge the difference.
The Grades, and What They Buy You
The OM and OS designations are TIA types for the same underlying glass, sorted by performance.
Grade | Core/cladding | Attenuation @850/1300 nm | Bandwidth @850/1300 nm |
|---|
OM1 | 62.5/125 µm | 3 / 1 dB/km | 160–200 / 500 MHz·km |
|---|
OM2 | 50/125 µm | 3 / 1 dB/km | 500 / 500 MHz·km |
|---|
OM3 | 50/125 µm | 2.5 / 0.8 dB/km | 1500 / 500 MHz·km |
|---|
OM4 | 50/125 µm | 2.5 / 0.8 dB/km | 3500 / 500 MHz·km |
|---|
OM5 | 50/125 µm | 2.5 / 0.8 dB/km | 3500 / 500 MHz·km |
|---|
OS1 (single-mode) | 9/125 µm | 0.4 / 0.25 dB/km | — |
|---|
OS2 (single-mode) | 9/125 µm | 0.4 / 0.25 dB/km | — |
|---|
Read the bandwidth column, because that is what you are buying. OM1 and OM2 were built for LED-era LANs. OM3 and OM4 are laser-optimized for 850 nm VCSELs, and OM4’s higher bandwidth is what lets it carry a given rate further. OM5 is wideband multimode, built for wavelength division multiplexing in the 850–950 nm range rather than for a single higher rate.
On the single-mode side, OS1 and OS2 share the same attenuation figures here; the distinction is in the TIA designations and where the cable is meant to be used, with OS1 aimed at premises cabling and OS2 at outdoor plant. Neither has a bandwidth figure worth quoting, which is rather the point.
Two practical notes. OM5 does not buy extra distance on a single wavelength — if the design does not use short-wavelength division multiplexing, you have paid for capability you are not using. And OM4 is the sensible floor for a new multimode backbone. Specifying OM3 today is choosing to re-pull sooner.
Connectors and Polish: The Color Code Exists For a Reason
The connector family is smaller than people expect. LC is the small-form-factor connector that dominates patch panels and high-density work. SC is the square push-pull connector still common in carrier and FTTx work. ST is the older bayonet twist-lock. MPO and MTP are multi-fiber connectors carrying 8, 12 or more fibers at once, used where trunk cables have to land in a single position.
The part that causes real damage is the end-face polish, and it comes in three types. PC is the original physical-contact polish. UPC is a superior flat polish with tighter tolerances. APC is an angled polish, ground at an angle so that reflected light leaves the core instead of travelling back toward the source. Connector color tells you which you are holding: APC is green, UPC is typically blue.
APC exists because some applications cannot tolerate back reflections — analog video transport and higher-rate PON systems are the usual examples. The trade is that an angled end face cannot make physical contact with a flat one.
So do not mate APC to UPC. The two geometries cannot seat against each other, and forcing the connection puts the end faces at risk. This is the single best argument for the color convention: green to green, blue to blue, and if the colors disagree, stop.
Bend Radius: The Rule, and the Two Ways People Get It Wrong
The standard guidance is straightforward: minimum bend radius under tension during pulling is 20 times the cable diameter. Once installed and no longer under tension, the long-term minimum is 10 times the cable diameter. Some cables are specified at 15 times rather than 20, so the manufacturer’s data sheet governs.
The first mistake is confusing radius with diameter, and it is expensive. If a 13 mm cable needs a 20 × bend radius, that is 260 mm of radius. A pulley sized at 260 mm of diameter does not meet it — the pulley needs a 520 mm diameter, because diameter is twice the radius. The same error applies to slack storage: with a 10 × installed radius, storage loops must be no smaller than 20 times the cable diameter.
The second mistake is assuming a bend that violates the spec will look wrong. It usually does not. Damage shows up as broken fibers, fibers carrying higher loss under stress, and structural damage in the cable — and none of that announces itself as a kink. It shows up later, as an intermittent link. It also voids the cable warranty, since improper installation typically sits in the exclusions.
Where bends are unavoidable — patch cords at a crowded panel, small-count cables routed around tight enclosures — bend-insensitive fiber is the answer. It adds a low-index glass trench around the core that reflects light which would otherwise escape back into the core. It exists in both single-mode and multimode, and is worth specifying for patch cords.
Contamination: The Number One Cause
If you take one operational point from this article, take this one. Contaminated connections remain the number one cause of fiber-related problems and test failures. That is not a vendor slogan — it is the headline finding in Fluke Networks’ guidance on fiber inspection, and it holds across data centers, campus backbones and enterprise networks.
The mechanism is unglamorous. A single particle of dust at the core interface blocks light and scatters what it does not block. The connector still mates, still clicks, still looks fine in a photograph. The loss shows up as a number on a tester, or six months later as an unstable link.
Two rules follow.
Inspect every end face, always — including brand-new patch cords out of the bag. Factory dust caps are not a cleanliness guarantee, and a connector that has been sitting open in a pouch has been collecting whatever is in the air.
Clean, then inspect again. The act of cleaning can itself introduce contamination, so every end face should be inspected after every cleaning attempt. A pass before cleaning and a fail after it is more common than people expect, and it is the reason one-click cleaners and a scope belong in the same pouch.
Our own fiber optic cable range and HDMI active optical cables are supplied with end-face condition in mind, and our test and inspection process covers how assemblies are verified before they leave. The certifications we hold are listed separately, and quality management sets out how that is kept consistent across production runs.
What to Test Before You Sign Off
Attenuation and reflection are the two things a fiber link has to control, and they are measured separately.
Insertion loss is the light you lost getting from one end to the other — the sum of fiber attenuation, splices, connectors and any bends. Return loss is the light that came back at you, which matters more at higher rates and in analog systems.
A basic loss test set gives you insertion loss end to end. Return loss needs a meter that measures reflectance, and an OTDR is what you reach for when a link fails and you need to know where. On an MPO trunk, an OTDR is close to essential — it is the only practical way to find which of the twelve fibers is bad.
Then there is the check no tester replaces: a scope on every connector. Loss testing can pass a dirty connector on a short link and fail the same connector on a long one, because you had margin to spare. End-face inspection is the only way to see the problem that will still be there when the margin is gone.
Frequently Asked Questions
Can I use multimode for a 400-meter backbone?
At 850 nm over OM4, a 10G signal reaches well past that. The question is what rate the link has to carry in ten years, and whether you want the upgrade path to depend on the optics at both ends.
Is single-mode always the "better" choice?
It carries further with less loss and has effectively unlimited bandwidth. It also needs more expensive optics. On a 60-meter in-building run, that trade rarely pays.
My connector is green and the panel is blue. Can I use an adapter?
An adapter changes the connector type, not the polish. Mating APC to UPC is a geometry problem, not a gender problem, so a green-to-blue adapter does not solve it. Use the correct patch cord.
Does a clean end face stay clean?
No. Every mating and unmating is an opportunity to transfer contamination, and an exposed connector collects dust. Inspect before every mate, and re-inspect anything that has been sitting open.
The Short Version
Pick single-mode or multimode from distance and rate, and never mix the two — the loss is measured in tens of decibels, not tenths. Buy OM4 or better if you are installing multimode at all.
Match polish to polish, and treat the color code as a safety rule. Keep every bend inside 20 × cable diameter under tension and 10 × after installation, checking whether the number on the drawing is a radius or a diameter. And inspect every end face, every time, after every cleaning attempt — because contamination, not bandwidth, is what takes fiber links down.
If you are specifying a fiber programme and want to confirm what we can build, terminate and test against it, send us the details — grade, connector type, polish and the environment the cable is going into — and we will tell you what fits.
Sources
- Fiber Optic Association — Optical Fiber, including core and cladding sizes, multimode and single-mode wavelengths, the attenuation and bandwidth tables for OM1–OM5 and OS1/OS2, the caution against mixing fiber types, and bend-insensitive fiber — https://www.thefoa.org/tech/ref/basic/fiber.html
- Fiber Optic Association — Fiber Optic Cable Bend Radius or Diameter, including the 20 × under-tension and 10 × installed guidance, the radius versus diameter trap, pulley sizing and the worked 13 mm example — https://www.thefoa.org/tech/ref/install/bend_radius.html
- Fluke Networks — The Golden Rule of Fiber Inspection: Clean and Inspect Again, including the finding that contaminated connections remain the number one cause of fiber-related problems and test failures — https://www.flukenetworks.com/blog/cabling-chronicles/golden-rule-fiber-inspection-keeps-times
- Fluke Networks — White Paper: Fiber Contamination, Cleaning and Inspection, including the point that cleaning can itself introduce contamination and every end face should be inspected after every cleaning attempt — https://www.flukenetworks.com/edocs/wp-fiber-cleaning-and-inspection-white-paper
- trueCABLE — PC vs UPC vs APC Fiber Connectors: What Is the Difference, including the polish types and their reflectance behavior — https://www.truecable.com/blogs/cable-academy/pc-vs-upc-vs-apc-fiber-connectors-what-is-the-difference
- trueCABLE — Bend-Insensitive Fiber: What Is It?, including minimum bend radius figures for bend-insensitive single-mode — https://www.truecable.com/blogs/cable-academy/bend-insensitive-fiber-what-is-it
