Engineering

Test & Measurement — proving the fiber and reading the coherent link

Before you light a modern coherent wavelength you have to prove the glass, and once it is running you have to read the link's health from the right instrument. The tools that worked in the 10G on/off era do not all translate cleanly to dense, flexgrid, coherent systems. This page covers the advanced OTDR, the fiber characterization a coherent span needs, why the classic OSNR measurement breaks, and what the coherent receiver itself tells you.

The mindset shift

In the direct-detect era you measured the signal with an external optical instrument and read a number. In the coherent era the best instrument is often the transponder's own DSP — it reports Q-factor, pre-FEC BER, and even an internal OSNR estimate that an external analyzer physically cannot get anymore. Learn when to trust the external box and when to trust the receiver.

The OTDR, revisited

An OTDR (Optical Time-Domain Reflectometer) fires pulses down a fiber and times the light that reflects and scatters back, building a trace of loss vs distance. It is how you locate splices, connectors, bends and breaks. In the coherent era a few subtleties matter more than they used to.

The "gainer" artifact — why you shoot bidirectionally

An OTDR infers loss at a splice from how much backscattered light it sees before and after that point. But backscatter depends on the fiber's own properties, which differ slightly between two different fibers spliced together. The result: a splice can appear to gain light (a step up in the trace) — which is physically impossible for a passive joint. This false "gainer" is an artifact of measuring from one direction only.

THE GAINER ARTIFACT AT A SPLICE BETWEEN TWO DIFFERENT FIBERS Measured A -> B: Measured B -> A: loss ___ loss ___ | \___ ___/ | | \__ (step UP!?) __/ | | \___ / | +----------------> dist +----------------> dist looks like GAIN looks like extra LOSS Neither single direction is the truth. The real splice loss is the AVERAGE of the two directions: real loss = ( loss_A->B + loss_B->A ) / 2 Bidirectional averaging cancels the backscatter mismatch and removes the false gainer. Illustrative.

Mental model

The apparent gain in one direction is exactly cancelled by apparent extra loss in the other. Averaging the two directions removes the fiber-to-fiber backscatter difference and leaves the true splice loss. This is why acceptance testing of a real span is done bidirectionally — a single-ended trace can under- or over-report every splice.

Dead zones and the resolution-vs-range tradeoff

Right after a strong reflection (a connector) the OTDR's detector is briefly blinded and cannot resolve events — the dead zone. You control it with pulse width, and that is a tradeoff:

Short pulse

Better resolution — small dead zone, can separate two close-together events (e.g. two connectors a few metres apart). But it puts less energy on the fiber, so it cannot reach as far and the far end is noisy.

Long pulse

Better range — more energy travels farther, so you can test a long span. But the dead zone is larger, so close-together events blur together. You trade the ability to resolve nearby events for reach.

There is no single "correct" pulse width; you often shoot multiple ranges — short pulse for the near end and connectors, long pulse for the far end.

Wavelength dependence and in-service monitoring

Bend loss is wavelength-dependent: longer wavelengths leak out of bends more easily. That makes 1625 nm and 1650 nm more sensitive to bends than the 1550 nm signal band. Because those wavelengths sit outside the C-band traffic, an OTDR can monitor the fiber at 1625/1650 nm in service — watching for a developing bend or degradation without touching live traffic. A problem often shows at 1625 nm before it affects the 1550 nm channels.

Characterize the fiber BEFORE lighting a coherent span

A coherent transponder is sensitive to impairments that a 10G link shrugged off. Before turn-up, characterize the span so you know what the optics must tolerate — and so you have a baseline to trend against later.

Pre-coherent fiber characterization checklist

Attenuation / OTDR: total span loss, per-splice and per-connector loss, event map, no unexpected reflections — measured bidirectionally.
Chromatic dispersion (CD, ps/nm): how much the pulse spreads with wavelength across this span — the coherent DSP compensates it, but you must know it is within range.
PMD (ps): polarization-mode dispersion — a statistical spread that the DSP also handles up to a limit; old or stressed fiber can exceed it.
Optical return loss (ORL): total reflected power back toward the source — poor ORL (dirty/damaged connectors) degrades performance.

CD and PMD barely mattered for a slow direct-detect signal but are first-class concerns for coherent — even though the DSP corrects both, it corrects them only up to a stated limit, so you must confirm the span sits inside that limit.

The OSNR measurement problem

OSNR (optical signal-to-noise ratio) is the number that decides whether a wavelength decodes cleanly. Measuring it used to be simple; on a coherent flexgrid system it is not.

The classic out-of-band method (and when it works)

An OSA (Optical Spectrum Analyzer) sweeps the spectrum and shows each channel as a peak. The classic method reads the noise floor in the gaps between channels and interpolates it underneath each signal to estimate OSNR. This works fine when there is visible noise between the channels — i.e. widely spaced channels with clear gaps.

OUT-OF-BAND OSNR ON A WIDELY-SPACED GRID (OSA can do this) power | /\ /\ /\ | / \ / \ / \ <- signal peaks | / \ / \ / \ | / \______/ \______/ \ |._______| |______| |_______. <- noise visible in the GAPS +----------------------------------------> wavelength ^ interpolate this floor UNDER each peak = OSNR

Why it fails for dense / flexgrid / coherent — in-band OSNR

On a dense or flexgrid coherent system the channels are packed tightly and the signal is spectrally wide. There is no clear gap — the noise you care about sits underneath the signal itself. This is in-band OSNR, and an OSA cannot interpolate it because there is nothing to interpolate from.

DENSE / FLEXGRID COHERENT (out-of-band OSA method FAILS) power | ____ ____ ____ | / \ / \ / \ <- wide signals, packed tight |/ \ / \ / \ | \/ \/ \ <- NO gap: channels touch | noise is UNDER the signal, not between channels +----------------------------------------> wavelength there is no clear floor to read -> OSA interpolation is invalid

Common mistake

Trusting an OSA's out-of-band OSNR number on a dense coherent line. With no gap between channels, the interpolated "noise floor" is really the neighbouring signals' skirts, and the reading is optimistic and meaningless. In-band OSNR needs either a polarization-nulling instrument method or — far more practical in the field — the coherent receiver's own reported OSNR/Q and pre-FEC BER.

Read the link from the DSP

The coherent receiver's DSP is measuring the signal continuously and exposes the health metrics that actually matter:

Q-factor

How cleanly the receiver separates the constellation points from the noise. Higher Q = healthier link. Quoted linear or in dB. A convenient single-number health gauge.

Pre-FEC BER (trend it!)

The error rate before correction. It is not an alarm by itself — strong FEC is meant to run at a high pre-FEC BER. What matters is the trend: a slowly rising pre-FEC BER is margin erosion, an early warning that the link is aging toward its FEC limit.

Operational rule

Pre-FEC BER is the early-warning gauge; post-FEC errors are the alarm. Baseline pre-FEC BER at turn-up and watch it trend. Post-FEC errors mean correction has run out of headroom and traffic is being hit — customer-affecting, act now. The DSP's internal OSNR/Q estimate is the practical substitute for an external OSNR measurement you can no longer make in-band.

OCM — the ROADM's built-in eyes

A ROADM includes an OCM (Optical Channel Monitor): a built-in spectrum monitor that reports per-channel power (and, depending on design, an OSNR estimate) for every wavelength passing through the node. It lets the ROADM balance channel powers automatically and gives operators per-channel visibility without wheeling an OSA to site. Treat OCM data as a routine health feed — a channel drifting in power shows up there first.

The coherent-span acceptance record

When you accept a coherent span, capture a baseline of every meaningful metric so future readings can be compared against a known-good state. A representative acceptance record:

Metric	Instrument / source	Why it is recorded
OSNR / Q-factor	Coherent receiver DSP (not out-of-band OSA on dense lines)	Core signal-health baseline to trend against
Pre-FEC BER	Coherent receiver DSP	Early-warning gauge; only meaningful if you have the turn-up baseline
Post-FEC BER	Coherent receiver DSP	Must be error-free at acceptance
Chromatic dispersion (CD, ps/nm)	Dispersion test set / DSP report	Confirms the span is within the transponder's CD tolerance
PMD (ps)	PMD analyzer / DSP report	Confirms the span is within PMD tolerance
Per-port / per-channel power	Power meter / ROADM OCM	Confirms launch and receive powers are in the engineered window
OTDR / OLTS results	OTDR (bidirectional) + optical loss test set	Splice/connector map and end-to-end insertion loss baseline

Record these as the span's as-built baseline. The single most valuable troubleshooting asset months later is a clean set of turn-up numbers to compare a degraded link against.

Common mistakes to avoid

Reading out-of-band OSNR from an OSA on a dense/flexgrid coherent line — the number is invalid without a real gap between channels.
Accepting single-direction OTDR splice losses — the gainer artifact means you must average both directions.
Skipping CD/PMD/ORL characterization because "the DSP handles it" — the DSP handles it only up to a limit you must verify.
Not recording a turn-up baseline — with nothing to trend against, pre-FEC BER numbers are far less useful later.
Using one OTDR pulse width for everything — near-end resolution and far-end range need different pulses.

1. An OTDR trace shows a splice that appears to GAIN light. What is happening, and how do you get the true loss?

It is the "gainer" artifact — a backscatter mismatch between two different fibers spliced together, seen from one direction only. Shoot the span in both directions and average: real loss = (loss A→B + loss B→A) / 2. A passive splice can never actually add power.

2. Why can't you read OSNR the classic out-of-band way on a dense flexgrid coherent system?

The out-of-band method interpolates the noise floor from the gaps between channels. On a dense/flexgrid coherent line the channels are packed with no clear gap and the noise sits under the signal (in-band OSNR). There is nothing to interpolate, so the OSA reading is invalid. Use a polarization-nulling method, or in practice the coherent receiver's own reported OSNR/Q and pre-FEC BER.

3. Pre-FEC BER on a link is 2×10⁻³ and steady, post-FEC is error-free. Is this an alarm?

No. Strong FEC is designed to run at a high pre-FEC BER and still deliver error-free post-FEC traffic. Pre-FEC BER is an early-warning gauge — what matters is the trend (rising = margin erosion). The alarm is any post-FEC errors. Here post-FEC is clean, so the link is healthy.

4. Why test the fiber at 1625 nm when the traffic is at 1550 nm?

Bend loss is wavelength-dependent and 1625/1650 nm is more bend-sensitive, so it reveals a developing bend earlier. Because it sits outside the C-band traffic, it can be used for in-service monitoring — watching the fiber without disturbing live channels.

5. Which four fiber parameters should you characterize before lighting a coherent span, and why CD and PMD specifically?

Attenuation/OTDR, chromatic dispersion (CD, ps/nm), PMD (ps), and optical return loss (ORL). CD and PMD matter because a coherent DSP compensates both, but only up to a stated limit — you must confirm the span sits within that limit before turn-up.

6. What does a ROADM's OCM give you, and why is it useful day-to-day?

An OCM (optical channel monitor) reports per-channel power (and often an OSNR estimate) for every wavelength through the node. It lets the ROADM balance powers automatically and gives operators per-channel visibility without bringing an OSA to site — a channel drifting in power shows up there first.