A datasheet says sixteen millimeters. The working channel of the scope, at the worst angle in its travel, is closer to twelve. The fiber ran clean through the first ten procedures and then started losing power, so the natural conclusion is that the fiber was mis-specified — that someone should have bought the tighter-bending grade, or the larger core, or a different vendor.
Before swapping anything, it is worth understanding what that sixteen actually was. It was not a measurement of anything. It was a design allowance: a statement that this fiber may be routed at that radius without the manufacturer accepting responsibility for a specific loss figure. That is not a subtlety in the fine print. It is the difference between a number you can design to and a number you cannot, and it is the reason so many bend-related failures get misdiagnosed as material faults.
There are at least four different quantities circulating under the name “minimum bend radius”, and only one of them describes the thing you are actually bending.
The number on the sheet is an allowance, not a measurement
Take the most-cited pair of numbers in the fiber world: bend-insensitive single-mode grades specified for a 10 mm design radius and a 7.5 mm design radius. Both figures are geometry allowances — they tell you how tightly the fiber is permitted to be routed. Neither comes with a loss value attached to that radius, because no loss value has been published for it.
The optical guarantees live in a completely different table, and they are stated as conditions rather than as a single radius: loss per ten turns at 15 mm, loss for a single turn at 10 mm, loss for a single turn at 7.5 mm — each quoted only at named wavelengths. The tightest radius at which those grades are actually characterised is 15 mm. The 7.5 mm figure is a design limit, and it exists so that a network planner knows what geometry is permitted, not so that a reader can assume a guaranteed penalty at that radius.
This distinction has a practical consequence that shows up in tenders. Specify “7.5 mm bend radius” and you have quoted a geometry allowance while assuming an optical guarantee comes bundled with it. If your design genuinely routes fiber near that limit, the useful action is not to cite the grade — it is to ask for the loss to be measured at that radius and reported, because the datasheet does not contain that number and never did.
The reading rule: every time you meet a bend radius, ask whether it is the radius at which the fiber survives, or the radius at which it stops working, or merely the radius the vendor is willing to permit. Then ask for the conditions.
Two numbers at the same radius are not comparable
Here is the trap that catches experienced buyers. Suppose two grades are both quoted at 15 mm. One is quoted as 0.25 dB and the other as 0.4 dB, and the conclusion drawn is that the first grade is the better bender.
It is a misreading, because the two numbers were not measured the same way. In the published data for these grades, one is characterised at 15 mm over ten turns, and the other at 15 mm over a single turn. Ten turns accumulate loss that one turn does not. Placing those two figures side by side and ranking them is comparing a stack of coils against a single bend.
Bend loss is a function of three things at once: the radius, the number of turns, and the wavelength. Change any one and the number changes, which means a bend specification is really a triple, and a comparison is only meaningful when all three match. Vendors rarely choose the same test condition, so a bend-loss figure without its radius, turn count and wavelength is not a specification at all — it is a fragment.
| Grade class | Test condition | Loss at 1550 nm | Loss at 1625 nm |
| A1 | 15 mm, 10 turns | ≤ 0.25 dB | ≤ 1.0 dB |
| A1 | 10 mm, 1 turn | ≤ 0.75 dB | ≤ 1.5 dB |
| A2 | 15 mm, 10 turns | ≤ 0.03 dB | ≤ 0.10 dB |
| A2 | 10 mm, 1 turn | ≤ 0.10 dB | ≤ 0.20 dB |
| A2 | 7.5 mm, 1 turn | ≤ 0.50 dB | ≤ 1.00 dB |
Read the last column against the one before it. Moving 75 nm — five percent of the wavelength — multiplied the same bend loss by roughly two to five times, depending on the row. Keep that ratio in mind, because it becomes the most important fact in this article for anyone delivering laser energy rather than data.
The four quantities wearing one name
It is worth separating them explicitly, because a specification written against the wrong one will be wrong in a way that only shows up in the field.
| Called | What it actually means | Where it comes from | What it ignores |
| Optical bend radius | Loss per N turns at a stated wavelength | Measured, expressed in dB | Whether the glass survives |
| Mechanical bend radius | Short-term and long-term survival limits | Stress fatigue, derived from the proof test | What happens to your light |
| Assembly bend radius | A multiple of the finished cable’s outer diameter | Installation practice | The fiber type entirely |
| Geometric bend radius | The tightest inside curve in your actual routing | Your drawing, your scope, your tray | All of the above |
Vendors publish the first two. Your device imposes the fourth. The number you should be designing to is the third, and it is almost always the largest of the four — which is the source of the whole problem.
A typical delivery fiber product page, ours included, follows the same convention: power, wavelength range, core options, connector and length — properties of the fiber. Nothing on it describes the routing, because a routing is not a property of a fiber.
Your assembly’s number is a multiple of its diameter
The installation trade solved this long ago by refusing to talk about the fiber at all. The field rule is stated in terms of the cable’s outer diameter: twenty times the diameter while the cable is under tension, ten times the diameter once it is permanently in place. The reasoning behind it is geometric rather than optical. When you bend a finished assembly, you are not bending the fiber; you are bending a bundle of jacket, strength members and buffer, and the fiber inside is dragged along a path whose radius is set by the outermost layer, not the glass.
Run that arithmetic on a delivery fiber and the picture changes completely:
| What you are bending | Governing number | Minimum radius |
| Bare fiber, long-term limit | Mechanical, from the fiber datasheet | about 16 mm |
| Same fiber in a 3 mm jacketed assembly, fixed routing | 10 × assembly OD | about 30 mm |
| Same assembly, flexing during a procedure | 20 × assembly OD | about 60 mm |
Same fiber, three numbers, a factor of nearly four between the most generous and the most demanding. And note the direction of the error: the fiber datasheet’s figure is smaller than the assembly’s, so a designer reading only the fiber specification will happily plan a routing that violates the finished assembly by a factor of two. When it fails, the symptom will look like a bad batch of fiber.
One honest caveat. The ten-times and twenty-times factors come from telecom and utility installation practice, where the goal is protecting a cable that may sit in a duct for decades. A medical delivery assembly is a different animal, and the equivalent number for it should come from the assembly’s own testing rather than from a convention. But the direction of the arithmetic does not change: the number that governs your routing belongs to the assembly, and the fiber’s number is only a floor underneath it.
Two details that people get wrong when they measure. Radius is measured to the inside of the bend, not the outer curve, which matters when the thing being bent is a thick jacketed cable and the two numbers differ by the cable diameter. And a storage loop is naturally described by its diameter, which is twice the radius — so a coil that satisfies a ten-times-diameter radius rule is a coil whose diameter is twenty times the cable diameter. Half of all bend violations in the field are measurement errors in one of these two directions.
The wavelength decides how much of that radius you can afford
Now the part that matters most for laser delivery and is almost never mentioned in bend discussions, because the bend literature was written by the telecom industry for wavelengths around 1.3 and 1.55 µm.
Bend loss grows steeply as wavelength increases. The mechanism is straightforward: a longer wavelength is held less tightly by the core, so its field spreads further into the cladding, and a given curvature throws more of it away. In the table above, a five percent wavelength change produced a two-to-five-fold loss increase. Extend that trend across the range a laser delivery fiber actually covers and the consequences get serious.
Our own bare fiber is specified from 400 nm to 2200 nm — one all-silica product line spanning a wavelength ratio of more than five to one. A bend that is entirely comfortable at visible wavelengths is a different problem entirely in the near infrared, and the same physical fiber is being sold to both users. The practical conclusions are worth stating plainly:
- There is no single minimum bend radius for a fiber. There is a minimum bend radius for a fiber at a wavelength, a radius and a turn count.
- An infrared delivery fiber has less bend headroom than a visible one. A holmium system at 2100 nm sits at the least forgiving end of the range; a photodynamic or other visible-wavelength application sits at the most forgiving end, using the same glass.
- A bend-loss figure borrowed from a 1550 nm telecom datasheet understates your problem by a wide margin. It is measured far below your operating wavelength, at a radius and turn count that are unlikely to match your routing.
The levers you actually have are NA and core diameter, not carefulness
When the routing is tight and the number does not fit, “handle it more carefully” is not a design fix. There are three real dials, and two of them are on the datasheet.
Numerical aperture. NA is usually introduced as an acceptance-cone specification, which is what it means at the input, but its more useful role in a delivery fiber is as a bend-tolerance specification. A higher NA means a larger refractive-index step between core and cladding, which confines the light more tightly against curvature. Ray-tracing calculations make the size of the effect concrete: to reach the same fraction of transmitted power, a fiber at NA 0.20 needs a bend radius of the order of 6000 µm, while a fiber at NA 0.50 manages with roughly 1000 µm. NA 0.22 corresponds to a full acceptance cone of about 25.4 degrees, and the higher-NA variant of the same fiber accepts a visibly wider one.
That is why the two NA options on the bare fiber specification are not an interchangeable detail. Choosing the higher-NA fiber is choosing bend tolerance. The price is on the output side: a wider acceptance cone is a wider emission cone, and a beam that diverges faster is harder to work with at the far end. In a geometry where the beam has to cross a gap and strike a target — as in a side-firing fiber, where the beam is deflected up to 90 degrees from the fiber axis and the tip is a rigid 1.3 mm glass body — that divergence is part of the clinical behavior, not an abstraction. Bend tolerance bought at the input can be paid for at the output.
Core diameter. Larger cores are more forgiving of the connector and less forgiving of the bend. A 940 µm core may require a bend radius several times that of a 200 µm core in the same glass family, which is the trade documented on any high-power fiber datasheet. So the two levers push against each other: NA up buys bend tolerance, core diameter up buys alignment tolerance and costs bend tolerance.
The routing itself. Which leaves the third dial, and it is free. A delivery fiber’s tightest radius is rarely a physical constraint of the device; it is a consequence of where someone put the coil, how the cable was dressed in the tray, and whether the strain relief was allowed to do its job.
What bends for real: macro and micro
Everything above concerns macrobend loss — a visible curve that radiates light. It is computable, which is why it dominates the literature and the datasheets. The loss that ruins real assemblies is often microbend: tiny, invisible deformations caused by a cable tie pulled too tight, a clamp edge, an uneven surface, an adhesive bead, or thermal contraction of a jacket pressing on the glass. The mechanism is different — axial irregularity couples the guided light into radiating modes — and it scales with the number of modes in the fiber, which means a large-core multimode delivery fiber is more exposed to it than a single-mode telecom fiber.
This is why “the jacket looks fine” is not evidence of anything. A microfracture or a microbend from a clamp does not announce itself, the assembly can pass a certification test on the bench, and the degradation appears over a device’s service life. It is also why the internal construction of a delivery assembly — buffer, jacket, strain relief, how the cable is restrained inside a handpiece — is part of its bend specification and not merely packaging. In a lab setup, the cable is whatever holds the fiber. In a device, the cable decides what the fiber is allowed to do.
Auditing your own routing
The minimum bend radius your fiber actually needs can be derived in about ten minutes, and it is not a number you look up.
- Find the tightest curve, not the nominal one. The storage coil, the tray, the fold in a sterile pack, the worst configuration of a scope’s angulation rather than its resting position, the strain relief where the cable enters a handpiece. The tightest curve is almost never in the drawing you were given.
- Measure the inside of the curve with a known-diameter object you already own — a marker barrel, a coin, a mandrel — and write the number down in radius, not diameter.
- Classify static or dynamic. Anything that moves during a procedure is dynamic. Dynamic is the twenty-times case, and it is also the case where fatigue accumulates over the life of the device.
- Apply the assembly factor, then add margin for the person who repacks the tray at the end of the day, because that is when the tightest bend of the week gets made.
- Test the assembly as routed, not as delivered. This is the step that most often gets skipped: assemblies are typically characterised straight on a bench, with a clean launch and an undisturbed fiber. That measurement certifies a routing you will never use. Measure transmitted power with the cable dressed as it will be in the device, at the wavelength you will use, and compare against the straight-bench figure. The difference is your real bend budget, and it will tell you whether the problem is the fiber, the assembly or the person who coiled it.
The number you derive
The minimum bend radius your fiber actually needs is not printed anywhere. It is the tightest inside radius in your routing, in the static or dynamic case, at your operating wavelength, multiplied by the factor your finished assembly imposes, with margin on top. The fiber datasheet’s long-term figure is the floor under that calculation, not the answer to it.
And when the answer does not fit inside your device — which is the situation that produces most of these conversations — the available moves are to change the NA, to change the core diameter, or to change the routing. Three dials, two of them on the datasheet, one of them free. Being careful is not one of the three.