How to Measure Fiber Attenuation and Bend Loss: Cutback, OTDR and Mandrel Methods

Scope

This article gives procedures for measuring the attenuation coefficient of an optical fiber, in dB/km, and its loss under bending. Attenuation is measured by the cutback method, which is the reference method, and by optical time-domain reflectometry, which needs access to one end only; the same cutback measurement made with a broadband source gives the spectral attenuation curve. Bend loss is measured by wrapping the fiber around mandrels of known radius. Measuring the loss of a finished link or patch cord, with connectors, is covered in Measuring Insertion Loss and Return Loss, and the loss of on-chip waveguides in Cutback Propagation Loss. The expected values for each fiber type are in Types of Optical Fiber.

Cutback method

Light is launched into a long length of fiber and the output power measured; the fiber is then cut a short distance from the launch end, without disturbing the launch, and the power measured again at the new end. Because the launch is untouched, its coupling loss is the same in both measurements and cancels, and the difference is the loss of the length removed:

$$\alpha \;=\; \frac{10\log_{10}\!\left(P_\text{short}/P_\text{long}\right)}{L_\text{long} - L_\text{short}}.$$

Procedure

1. Prepare the launch. Couple a stable source (a laser or LED at the test wavelength) into the fiber under test and fix the input end so that nothing moves for the rest of the measurement.
2. Condition the launch. For single-mode fiber, a loop of small diameter near the input (a single turn around a mandrel of about 30 mm diameter is common) strips cladding light and the LP11 mode, which otherwise reach the short-length measurement but not the long one. For multimode fiber, a mode scrambler or mode filter sets a defined modal distribution; without it the result depends on how the fiber was launched.
3. Measure the long length. Cleave the far end, inspect the cleave, and record the power $P_\text{long}$ on an optical power meter with a large-area detector.
4. Cut back. Cut the fiber about 2 m from the input, beyond the mode filter, without touching the launch. Cleave, inspect and record $P_\text{short}$.
5. Repeat the short-length cleave and measurement two or three times; the spread gives the cleave-and-coupling uncertainty.
6. Compute $\alpha$ from the powers and the length difference, which is taken from the spool's marked length or measured with an OTDR.

Worked example

A 10.000 km spool of standard single-mode fiber at 1550 nm gives $P_\text{long}$ = −3.10 dBm; after cutting back to 2 m, $P_\text{short}$ = −1.12 dBm. The loss difference is 1.98 dB over 9.998 km, so $\alpha$ = 0.198 dB/km, a typical value for G.652 fiber at this wavelength. A repeatability of ±0.02 dB in the short-length reading contributes ±0.002 dB/km here; over a 1 km sample the same reading would contribute ±0.02 dB/km, which is why the sample should be as long as possible.

Spectral attenuation

Replacing the laser with a broadband source (a tungsten-halogen lamp through a monochromator, or a supercontinuum source) and the power meter with an optical spectrum analyzer gives $\alpha(\lambda)$ from the ratio of the two spectra. The curve shows the Rayleigh scattering floor falling as $\lambda^{-4}$, the infrared absorption edge beyond about 1600 nm, and any hydroxyl absorption peak near 1383 nm, which is small in low-water-peak fiber.

OTDR method

An optical time-domain reflectometer sends pulses into one end of the fiber and records the light backscattered from each point along it. The trace, in dB against distance, falls with a slope equal to the fiber's attenuation (the instrument scales for the round trip), so the attenuation over a section is read from the slope between two markers or, better, from a least-squares fit over the section.

1. Connect the fiber through a launch cord long enough to clear the instrument's dead zone.
2. Choose a pulse width short enough to resolve the features of interest and long enough to keep the far end above the noise floor, and set the group index for the fiber type.
3. Average until the trace is smooth over the section to be measured.
4. Fit the slope over a section free of splices and connectors.

A single-ended trace measures the backscatter, which depends on the fiber's mode field diameter and scattering coefficient as well as on its loss. Where fibers of different types are joined, a splice can appear as a gain in one direction and an exaggerated loss in the other. Measuring from both ends and averaging the two values removes this effect, and bidirectional averaging is standard for acceptance testing of spliced links.

Bend loss

Bend loss is measured as the increase in loss when a known length of fiber is wound around a mandrel of known radius. Macrobend loss grows rapidly as the radius falls and as the wavelength rises, because the mode is less tightly confined at longer wavelengths; it is therefore measured at the longest wavelength of use, usually 1550 nm and 1625 nm.

1. Launch into the fiber as for cutback, with the output on a power meter, and record the power with the fiber laid straight or in loose loops of large radius.
2. Wind the specified number of turns around a mandrel of the specified radius, without tension and without crossing turns, and record the power again.
3. The bend loss is the difference, in dB, divided by the number of turns if a per-turn value is wanted.
4. Repeat for several numbers of turns: the loss should grow linearly with the number of turns, and a nonlinear result indicates microbending from crossings or tension, or a reference that drifted.

A source whose power drifts by more than the bend loss being measured makes the result meaningless; for bend-insensitive fiber, whose loss per turn can be a small fraction of a dB, a reference path through a splitter or a source with an output monitor is needed. Microbend loss, from small random deformations, is not measured this way but by winding the fiber under tension onto a drum with a rough surface and comparing with a loose winding.

Common errors

Launch not conditioned. Cladding modes and, in single-mode fiber operated near its cutoff wavelength, the LP11 mode, reach the short-length measurement and inflate the apparent loss difference.

Poor cleaves. A bad cleave at the short end adds loss to $P_\text{short}$ and reduces the measured attenuation; inspect every cleave.

Detector underfilled or overfilled. The power meter must capture the whole output cone of both ends; use a large-area detector or an integrating sphere.

Temperature and handling. Moving the spool between measurements changes microbending; leave it undisturbed.

Single-ended OTDR through mixed fiber. Read splice losses only as bidirectional averages.

References: IEC 60793-1-40, Optical fibres – Attenuation measurement methods; IEC 60793-1-47, Optical fibres – Macrobending loss; D. Derickson (ed.), Fiber Optic Test and Measurement (Prentice Hall, 1998), chapters on attenuation and OTDR.