A tight bend in a fiber patch cord is a fault you installed yourself. The link comes up, the readings look acceptable, the ticket closes — and then the link flaps under load, or drops after a temperature swing, or fails outright months later when the stressed glass finally gives up. Nothing was dirty, nothing was broken, nothing was swapped. The cord was just folded too hard around a cable manager, or pulled tight around a Velcro strap, or creased where it exits the panel. Bend radius is the least glamorous rule in fiber work and one of the most common causes of faults that look like something else.
What a bend does to the light
Light travels down the core because it reflects off the core-cladding boundary at a shallow angle. Bend the fiber and that angle gets steeper at the outside of the bend; past the critical angle the light stops reflecting and escapes through the cladding instead. That escaped light is macrobending loss, and it shows up as a lower received power reading.
The loss grows fast as the bend gets tighter — halving the radius can add decibels, not tenths — and it is worse at longer wavelengths, so a 1550 nm single-mode link suffers more from the same bend than a 1310 nm one. A bend that costs 0.2 dB today can cost 2 dB after the cable settles, the rack warms up, or someone tugs the cord while working the next panel over.
The practical rule: if the received power is low and everything else checks out, the bend is the suspect, not the optic.
The numbers: 10x installed, 20x under tension
The rule most cable makers publish is simple: minimum bend radius is 10 times the cable outside diameter once installed and under no load, and 20 times the diameter while pulling or under tension. For a standard 3 mm patch cord that is a 30 mm radius installed and 60 mm while pulling.
- 3 mm patch cord: 30 mm installed, 60 mm pulling
- 2 mm patch cord: 20 mm installed, 40 mm pulling
- 900 µm tight-buffered fiber: 9 mm installed, 18 mm pulling
- 3 mm bend-insensitive fiber (G.657.A2): 7.5 mm, per the datasheet
Bend-insensitive fiber — G.657, the kind in most modern patch cords — tolerates tighter bends than the old G.652 rule. That does not mean bend radius stops mattering: it means the limit moved, and the cable datasheet is the authority. If the manufacturer rates the cord at 30 mm, a 15 mm bend is a fault, not a neat dress.
The radius is measured to the inside of the bend, at the tightest point. A cord that looks fine from the front can be folded hard where it disappears behind the manager. The quick check I use on the floor: if the bend is tighter than the curve of my thumb, it is too tight. A thumb is about 30 mm across, which is no coincidence — it matches the 3 mm cord rule, and it is always in my pocket.
Where bends hide in the rack
Bends rarely happen in the open. They hide where the cord changes direction out of sight: pinched behind a horizontal cable manager, folded hard where it exits the patch panel, pulled tight around a Velcro strap that someone cinched down during a tidy-up, or kinked inside a service loop that was coiled too small.
The worst one I see regularly is the cord dressed beautifully across the front of the rack and then folded at ninety degrees behind the manager to reach the port — the visible half looks perfect and the hidden half is creased. Another is the slack loop: a service loop coiled to a 50 mm diameter looks tidy and is a bend fault waiting for a temperature swing. A loop is just a bend that goes all the way round, and its radius still has to clear 10x the diameter.
When you open a panel to work a link, look at the back of the manager, not just the front. The bend is almost always behind something.
Finding a bend fault: power, OTDR, and the two-wavelength test
The diagnosis order is power first, OTDR second, eyes third. Check the received power against the loss budget: if the light level is low but the connectors are clean, the splices test good, and the budget says the link should pass, suspect a bend.
An OTDR trace shows a bend as a step down — a sharp, localised drop that is not at a splice or connector — and the two-wavelength comparison confirms it: a bend loses more at 1550 nm than at 1310 nm, while a splice or a dirty connector loses about the same at both. That wavelength difference is the fingerprint. A splice is flat across wavelengths; a bend is not.
On the floor, once the trace points at a distance, walk that distance along the cord run and look at every turn, strap, and manager edge. Then straighten the suspect bend and watch the received power — if it comes back, the bend was the fault, and the fix is re-dressing the cord with a proper radius, not replacing the optic.
When a bend is permanent damage
A bend straightened quickly usually recovers fully — the light comes back and the link holds. A bend held tight for weeks is different. Glass under constant tension develops microcracks that grow over time, and the fiber can snap months after the bend was installed, or the loss can creep up until the link drops under load.
If you find a hard crease in the jacket — a visible fold line, a flattened section, a white stress mark — treat the cord as damaged, not as fixed, even if the readings recover when you straighten it. Replace the cord, scope the new one, and dress it with a proper radius. The same applies to a service loop that was coiled too small for months: the cord may test fine today and fail in the summer when the rack runs hotter. A patch cord costs a few dollars; a 2 a.m. link-down call costs a shift. Replace the creased cord.
Recording the bend in the ticket
The bend belongs in the ticket, because the next person on the link needs to know the fault was in the dressing, not the hardware — otherwise they will swap the optic and close the same ticket again. The minimum useful line:
- Received power against the budget, and whether it failed
- The OTDR evidence, with both wavelengths
- Where the bend was and how tight it was
- What you did about it — re-dressed, and whether the cord was replaced
- The received power after the fix
Five fields, and the next tech knows the fault was a bend, where it was hiding, and that the cord was replaced rather than just straightened — which is exactly the information that stops the same ticket from being opened twice.
A question for you
What is the tightest bend you have found on a live link — and did the readings recover when you straightened it, or did the cord have to go?
Top comments (0)