BRENNEN

Industries / Technology / Telecommunications / Fiber optics

Testing and fault finding

Testing is the work of checking that fiber, connectors and finished links carry light as they should, and of finding where and why they do not.

7

Milestones achieved, from 1976 to 2022.

3

To watch for: announced, not yet confirmed outside.

3

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1976

    Timing light that comes back

    M. K. Barnoski and S. M. Jensen of Hughes Research Laboratories fired laser pulses into a glass fiber and recorded the faint light scattered back as each pulse travelled; from that record, taken at one end only, they could work out the fiber's total loss. This is the idea behind the optical time-domain reflectometer (OTDR), which times the echoes of a light pulse to show loss, splices and breaks along a fiber. In March 1977 S. D. Personick described an optical-fiber time-domain reflectometer in the Bell System Technical Journal.

    Lessona technician at one end of a fiber can see its whole length.

    opg.optica.orgvtda.org

    Credit

    M. K. Barnoski and S. M. Jensen (Hughes Research Laboratories); S. D. Personick

  2. 1990

    Testing from the office

    NTT began using AURORA, a system that runs fiber tests from a terminal at a maintenance base instead of by hand on site. It tests fibers automatically, isolates faults and detects water getting into cables. NTT brought in a cheaper, better version in 1996, linked it to its other operations systems in 2000, and added a lower-cost Optical Testing Module in 2002.

    rd.ntt

    Credit

    NTT

  3. 2007

    Testing without cutting service

    ITU-T approved Recommendation L.66 in May 2007. It sets aside a maintenance wavelength band for test light, with filter rules, so that a fiber already carrying customers' traffic in an access network can be tested without disturbing it.

    itu.int

    Credit

    ITU-T

  4. 2009

    Measuring dirty end faces

    The IEC published IEC 61300-3-35 on 20 November 2009, a method for "visual and automated inspection" of a connector's end face, the polished glass tip where two fibers meet, for scratches, pits and dirt. Test-gear makers name dirt as the leading cause of failures: EXFO, citing NTT Advanced Technology, says 80% of network owners call contamination the root cause, and the connector maker Senko, citing the same source, says connector problems are four of the top five causes.

    Lessonthe most common fault was dirt, not glass or electronics.

    webstore.iec.chexfo.comsenko.com

    Credit

    IEC TC 86/SC 86B

  5. 2010

    One launch for every tester

    In 2010 the TIA published TIA-526-14-B, adopted from IEC 61280-4-1, which defines "encircled flux," a tight rule for how a test source's light fills the core of multimode fiber, the wide-core fiber used in buildings and data centers. Before it, the same link could read up to 1 dB differently depending on the light source, a spread that was tolerated until Ethernet loss budgets at 40 and 100 Gbit/s fell to 1.9 dB on OM3 fiber and 1.5 dB on OM4. Field devices that meet the rule arrived in 2013.

    Lessona loss test is only as repeatable as its light source, so the tester had to be standardized as well as the cable.

    cablinginstall.comcablinginstall.com

    Credit

    IEC; TIA committee TR-42, chaired by Bob Jensen

  6. 2022

    Looking is not enough

    Edition 3.0 of IEC 61300-3-35, published on 16 September 2022, keeps numerical criteria for end-face images but says visual inspection does not replace measuring loss, and that "a connector cannot be rejected by just failing visual inspection." It also narrowed the cladding zone from 115 µm to 110 µm and dropped inspection requirements for the two outer zones.

    Lessonan image of a flaw does not prove the flaw loses light, so the optical measurement decides.

    webstore.iec.ch

    Credit

    IEC TC 86/SC 86B

  7. 2022

    A fault dataset others reuse

    On 17 October 2022 Khouloud Abdelli and co-authors posted a dataset of optical fiber faults on IEEE DataPort: segments of real OTDR traces from a lab setup spanning tens of kilometers, in eight classes from normal to fiber cut and dirty connector, about 16,000 samples per class. A team at the Technical University of Denmark and Politecnico di Milano reused it in 2025, and their classifier reached 92.6% accuracy on the original data.

    Lessonshared data lets others test their methods, but one lab's setup limits what models learn.

    re.public.polimi.it

    Credit

    Abdelli and co-authors; reuse by Healy, Baum and Musumeci

Watch for

  1. AI grading of multifiber connectors

    VIAVI says its INX 760 probe microscope, launched in March 2024, is "the industry's first such device to implement AI-driven image analysis," needed for MT ferrules because glass in their polymer surface can look like fibers or defects. EXFO says its FIP-500 has the "industry's first AI-driven endface analysis" and grades an MPO-12 connector in under 10 seconds. The two "first" claims conflict.

    Achieved whenan independent lab, standards body or large user publishes a blind comparison of either tool's verdicts against expert grading and loss measurements on multifiber connectors.

    blog.viavisolutions.comexfo.comcablinginstall.com

    Claimed by

    VIAVI, September 2025; EXFO, February 2026

  2. Certifying hollow-core fiber links

    Hollow-core fiber scatters light back differently from ordinary fiber, which changes how OTDR tests read. VIAVI says it has the industry's first all-in-one kit for certifying such links over medium and long distances, validated in trials with fiber makers, contractors and "three leading hyperscalers," none of them named.

    Achieved whena named hyperscaler, fiber maker or independent lab publishes link results measured with it.

    viavisolutions.com

    Claimed by

    VIAVI, January 2026

  3. Three-minute subsea fault finding

    EXFO's USO-1618 is an OTDR built for undersea cables that have amplifiers along their length. EXFO says it tests each fiber pair in three minutes, against up to six hours with conventional methods, and locates a fault to within about 5 m.

    Achieved whena cable owner or repair contractor reports a real fault located with it and confirms the position at repair.

    subtelforum.com

    Claimed by

    EXFO, September 2026

Open

  1. 01

    Real end-face images, openly graded

    An open set of real microscope images of connector end faces, single-fiber (LC and SC, flat and angled polish) and MT-ferrule MPO-12, each with a defect mask, defect type, size in µm, zone, and a pass or fail under the 2022 rules, released with a baseline grader.

    The brief
    Credit · unclaimed
  2. 02

    Bench OTDR traces with answers

    A public benchmark of full OTDR traces from bench-built links with known faults (spools, connectors, splices, tight bends around a mandrel, reflectors), kept as the original.sor files, taken on at least three makers' OTDRs in both directions and at several pulse widths, with true positions, losses and reflectances from known lengths and power-meter readings.

    The brief
    Credit · unclaimed
  3. 03

    Open labels for OTDR events

    A tool, either an addition to OpenOTDR or a new one built on otdrs, that loads.sor files from any maker, lets a person mark and edit events, exports the labels in a documented open format, and writes the edited file back.

    The brief
    Credit · unclaimed

Open 01 · the brief

Real end-face images, openly graded

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open set of real microscope images of connector end faces, single-fiber (LC and SC, flat and angled polish) and MT-ferrule MPO-12, each with a defect mask, defect type, size in µm, zone, and a pass or fail under the 2022 rules, released with a baseline grader.

Why it is still open

The open projects found are partial. FiberSight, an open IEC 61300-3-35 grader, reports results "on a synthetic dataset" that "should be re-validated on real microscope images," and still checks the two outer zones that the 2022 edition no longer requires. FEB-Bench has 535 images labeled only for boundaries, and MCF-Endface 216 splicer images labeled only for core positions. The AI graders from VIAVI and EXFO are closed.

github.comgithub.comhuggingface.co

Built with what exists

A probe microscope that exports images, with VIAVI INX 760 or EXFO FIP-500 verdicts as a reference; practice patch cords and MPO cords; cleaning tools; CVAT or Roboflow for labeling; OpenCV and PyTorch with a YOLO-family segmentation model; FiberSight's code as a start. To run the grader on a device the size of a scope, see the Note "An open compiler".

  1. First step

    Image 50 unpowered cords before and after cleaning with one probe, save the raw images and the probe's verdicts, and publish a 100-image pilot with a datasheet.

  2. Done looks like

    At least 1,000 real images from at least two microscope models, at least 200 of them of MT ferrules, each labeled by two people with their agreement reported and paired with a commercial analyzer's verdict and a measured insertion loss, released under CC BY 4.0, with a baseline that agrees with the reference pass or fail on at least 95% of held-out images and every disagreement listed.

Open 02 · the brief

Bench OTDR traces with answers

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public benchmark of full OTDR traces from bench-built links with known faults (spools, connectors, splices, tight bends around a mandrel, reflectors), kept as the original.sor files, taken on at least three makers' OTDRs in both directions and at several pulse widths, with true positions, losses and reflectances from known lengths and power-meter readings.

Why it is still open

A 2025 paper says "the limited availability of labelled failure data in optical fiber systems has been a significant barrier," citing operators' privacy concerns, and its authors had to make synthetic data for rare classes. The main open real dataset holds trace segments from one lab, many GitHub projects train on synthetic traces, and the.sor file format "is unfortunately not open," with its specification sold for $750.

re.public.polimi.itgithub.com

Built with what exists

Rented or university OTDRs; spools of unlit G.652 fiber with launch and receive cords; a light source and power meter; the open-source parsers pyOTDR and otdrs; PyTorch. No live networks, and each instrument's laser-safety label followed.

  1. First step

    Build one three-spool link with two connectors, one fusion splice and one mandrel bend, shoot it both ways at three pulse widths, and publish the.sor files, the otdrs output and the build sheet.

  2. Done looks like

    At least 500 traces from at least 10 link layouts and 3 makers, labels in a documented JSON format, a build sheet for each layout so anyone can rebuild it, and a baseline event detector with its F1 score reported against the truth and against each OTDR's built-in analysis.

Open 03 · the brief

Open labels for OTDR events

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A tool, either an addition to OpenOTDR or a new one built on otdrs, that loads.sor files from any maker, lets a person mark and edit events, exports the labels in a documented open format, and writes the edited file back.

Why it is still open

OpenOTDR's roadmap still lists "Label features in trace data" and "Export SOR & SOD Files" as not done, and its code was last updated in August 2024. Makers' own analysis software exists, but no open tool with an open label format was found.

github.comgithub.com

Built with what exists

otdrs, whose README says the files it writes "open without issue in several professional applications"; pyOTDR; a Plotly or web interface.

  1. First step

    Write the label schema (event type, position in m, loss in dB, reflectance in dB, and where the truth came from) and a script that converts the key-events block in otdrs output into it.

  2. Done looks like

    It opens sample.sor files from at least three makers, labels survive export and re-import unchanged, edited files open in at least one maker's viewer, and it is used to label the bench benchmark above.