Request an intervention

Operational notes Engineering

Can you run an OTDR on a live fibre without taking the customer down?

6 min read

Light trails from night traffic on a curving carriageway, in black and white, with street lamps lit in the background
Traffic does not stop to be measured: the measurement has to pass alongside it, on another wavelength.

“Test that link, but do not stop the traffic.” The request turns up when a link degrades and nobody will sign off an outage window. It can be done, but not with the instrument and method of acceptance testing: in-service measurement is not an OTDR run more carefully, it is a different architecture, decided when the network is designed. The reference is an ITU-T Recommendation almost no specification cites — and those who cite it usually cite the wrong number.

First things first: L.66 is now called L.313.

The document is “Optical fibre cable maintenance criteria for in-service fibre testing in access networks”, approved on 18 May 2007 by ITU-T Study Group 6 as L.66. The ITU record carries a note: “Former ITU-T L.66 renumbered as ITU-T L.313 on 2016-02-15 without further modification and without being republished”. Identical text, new number since 15 February 2016; the L.313 (05/07) record shows as “In force”. Anyone writing “to ITU-T L.66” in a 2026 specification cites a number renumbered ten years ago: the content is right, the reference is not.

Why you cannot simply fire 1550 nm down a live fibre.

Clause 5 sets two requirements that look obvious and are not: testing without degrading the communication signals, and remaining able to evaluate the fibre even under interference from the communication light. Two opposite problems.

Towards the user, the test pulse travels the same fibre as the signal and lands on the terminal equipment’s photodiode, which is broadband: it does not reject light because it “is not the signal”. Towards the instrument, the communication light passes through the injection coupler into the OTDR detector, adding to a backscatter far weaker than the launched pulse.

The answer is not to lower the test power. Clause 7.1 is blunt: the maintenance wavelength must not be one used for communication signals. If the test falls inside the transmission band, no filter removes it without removing the signal too. The separation is spectral, not a matter of level: lower power only costs dynamic range.

The maintenance band, and why 1650 nm.

Communication bands extend up to the L-band, 1565-1625 nm, with minimum channel spacing defined as 12.5 GHz. Above that sits the U-band, 1625-1675 nm, marked in Figure 1 of the Recommendation as the maintenance band; the test wavelength is 1650 nm, following the allocation in ITU-T L.41. The reason, in clause 7.2, is simple: the U-band is not used for communication signals.

The same clause sets the geometry of the spectrum: with A and C the source edges, B and D those of the cut-off filter, B < A < λtest < C < D must hold. It adds the caveat that weighs most in the field: sideband noise from the source, falling outside the filter’s cut-off band, must be suppressed sufficiently against that cut-off value (Lt dB) — otherwise the filter is not enough. Appendix II, Japanese experience reported for information, quantifies it: a fibre Bragg grating filter of −52.48 dB at the peak wavelength fell to an effective rejection of −20.38 dB, because the source suppressed its sidebands by only some −40/−50 dB.

The filter at the far end: the item almost nobody specifies.

The arrangement in clause 7.3 is spare: at the exchange a coupler injects the test light towards the user’s equipment; at the end of the line a test light cut-off filter of cut-off value Lt dB stops that light reaching the receiver. The value is not a catalogue number: the Recommendation ties it to the system’s S/X ratio and the powers below the coupler, through Lt >> SX − Pcd + Pt.

That filter sits at the customer’s premises. Appendix I places it as close as possible to the end of the optical distribution network, typically in the connector in front of the terminal equipment, with an average insertion loss below 1 dB at the communication wavelength. It is decided when the termination is chosen and the patch lead installed, not afterwards. Anyone who has not installed it cannot measure in service — no instrument makes up for it. Where the filters are must be recorded in the as-built, or at the first fault nobody knows which links are measurable.

Symmetrically, clause 7.4 requires the instrument to tolerate the communication light and a filter before the detector to reject it: the criterion is to keep the fluctuation induced on the trace within 0.2 dB, the residual signal 10 dB below the Rayleigh backscatter.

The PON case.

Appendix III matters most on FTTH. With a splitter in place, an OTDR launched from the exchange cannot pinpoint faults in the branched region between divider and user equipment: backscatter from the branches accumulates and is inseparable in the trace. So the measurement is made from the user side — equivalent to measuring a point-to-point link — and must still be in service, because other branches on the same splitter may be active. What the divider costs in the budget is a separate calculation, already done here.

The alternative the Recommendation declares and does not cover.

This is the route many take without knowing it. In the scope, clause 1 flags an alternative approach — monitoring key parameters of the transmission equipment, such as OLT transmitted power and ONU received power — and states expressly that it is not examined in the Recommendation. It is simpler, often already available with no extra hardware, and has one precise limit: it tells you that something has changed, not where. An alarm threshold, not a fault location. The out-of-band route locates, but it costs: a filter on every termination, spectral requirements on the source, instruments that tolerate the live light. Neither replaces the other: choosing one without knowing it is the real mistake.

What to write into the specification.

  1. Test band stated: U-band, 1650 nm, with an explicit reference to ITU-T L.313 (formerly L.66), clause 7.1.
  2. A cut-off filter at every user termination, as close as possible to the end of the ODN, tested at activation — not “provided for”.
  3. The cut-off value Lt required in dB, derived from the system’s S/X ratio, plus the sideband suppression requirement on the test source.
  4. The injection point: where the coupler sits and from which side the measurement is made, with the PON case handled separately.
  5. The deliverable expected: native trace and values with the measurement parameters, not a pass/fail — true in service as at acceptance, where the pulse width changes what you see.
  6. A record in the as-built of where every filter is installed, with the termination code: without it, measurability is lost at the first reconfiguration.

The point.

If the dedicated band, the filter at the far end and compliant instruments are not there, the choice is not between measuring well and measuring badly: it is between taking the customer down and not measuring. A design decision, taken before installation. That is how we work: measurement, certification and documentation defined together with the design, from backbones to data halls — because a network you cannot measure while live can only be diagnosed by switching it off.

Have a specification to write, or an installed network to make measurable in service? Get in touch: checking what is provided at the terminations costs far less than an outage window.

Sources