Coherent Optics: Pre-FEC BER, OSNR and DSP Checks

Coherent pluggables changed optical troubleshooting fundamentally. With grey optics you checked receive power and, if it was inside the window, the link worked. With coherent modules the link can show perfect power and still deliver errors, because what matters is signal quality - OSNR, dispersion compensation and the DSP's ability to lock. This article walks the telemetry a coherent module exposes, what each value tells you, and how to map symptoms to causes instead of swapping optics until something works.

The Telemetry That Matters

Read them together. Good Rx power with poor OSNR and a rising pre-FEC BER is an optical signal-to-noise problem; good power and good OSNR with a high EVM points at the module or the DSP; errors that appear only at certain times of day point at thermal expansion and connector condition.

Why Grey-Optic Habits Fail on a Coherent Link

A 10G or 25G grey link is a simple amplitude-modulated signal. Receive power above a sensitivity floor and below a saturation ceiling is essentially the entire health check, and a link that is inside its power window either works or does not. Coherent transmission is a different animal: the receiver digitises the optical field - both quadratures of both polarisations - and a DSP reconstructs the constellation, compensating chromatic dispersion, polarisation-mode dispersion and carrier phase error in the electrical domain.

That architecture gives coherent optics enormous reach and, at the same time, it decouples link health from optical power almost completely. You can have:

• Perfect power and unusable signal: too much noise (OSNR deficit) so the constellation is too diffuse to decode, even though every photodetector sees a healthy level.
• Perfect power and a locked-but-struggling DSP: the receiver converges, but a parameter such as polarisation tracking is near its limit, so error correction runs hot and pre-FEC BER trends upward.
• Perfect power and a link that will not come up: a mismatch in modulation format, baud rate or FEC mode means the DSP is hunting for a constellation that is not there.

The practical rule is that on coherent links, power is a sanity check and signal quality is the actual health metric. Trouble tickets that end "the optics look fine" usually mean only that the power reading was green.

The FEC Limit and What "Too Close" Is

Every coherent link has a pre-FEC BER at which the FEC can no longer correct fast enough - the FEC limit, or forward error correction threshold. Above it, post-FEC BER becomes non-zero and frames are lost. Typical hard-decision thresholds sit around 1e-3 to 2e-3 pre-FEC BER depending on the coding, and the specific number for a module is a datasheet value, not a rule of thumb.

What matters operationally is not the absolute number but the margin to it. If a link runs at 1e-6 pre-FEC BER against a 1e-3 limit, it has three orders of magnitude of headroom and is healthy. If it runs at 5e-4 against the same limit, it is a marginal design that will fail the first time a connector degrades or an amplifier drifts. Trend matters more than the instantaneous value: a pre-FEC BER that doubles every month reaches the limit on a predictable schedule, and that is a maintenance ticket long before it becomes an outage.

OSNR, OSNR Penalty and SNR Are Not the Same Number

These three get used interchangeably in conversation and confused in troubleshooting, so it is worth separating them:

• OSNR is the optical signal-to-noise ratio in the channel, typically referenced to a 0.1 nm measurement bandwidth. It is dominated by amplifier noise - the EDFA noise figure and the number of amplifiers in the path.
• OSNR penalty is the extra effective noise you must subtract for impairments that are not additive noise: fibre non-linearity, filtering, polarisation effects, and crosstalk from neighbouring channels. A link can meet its OSNR budget and still miss its performance target because the penalty is worse than assumed.
• SNR, as reported by a module's DSP, is the electrical signal-to-noise ratio after equalisation - a different domain entirely. It correlates with performance but is not directly comparable to a link-budget OSNR figure.

When a module reports "OSNR" from its own estimator, treat it as a useful relative metric rather than an absolute budget number. Comparing a field measurement with a design budget produced a different way is how teams chase a phantom 3 dB for a week.

DSP Lock, Training Time and What They Reveal

The DSP lock state is the first thing to read on a link that will not come up. Training involves blind equalisation, carrier frequency estimation, carrier phase recovery and polarisation tracking, and each stage can fail independently. A useful field test is to compare training time against a known-good link of the same type and length: a receiver that locks in two seconds on a healthy channel and takes twenty seconds on a suspect one is telling you the channel is noisy or impaired, even before any BER becomes measurable.

Two specific signatures are worth knowing:

• Frequency offset near the edge of the carrier-recovery range. Transmitter and receiver local oscillators are specified with a tolerance; if the offset is large, the DSP is operating near the limit of its search window and lock becomes intermittent, especially as temperature changes. It is a real failure mode and one that swapping one end of a link usually fixes.
• EVM elevated with a clean optical power reading. If modulation quality is poor at the transmitter, the receive side pays for it no matter how clean the fibre is. This points at the module and, occasionally, at the host port driving it.

Reading the Module

show interface Ethernet1/1 transceiver detail
show interface Ethernet1/1 transceiver dom          ! optical power, temperature, bias
show interface Ethernet1/1 counters detailed
show interface Ethernet1/1 fec                      ! pre-FEC and post-FEC counters

# Linux host side
ethtool -m enp1s0f0
ethtool -S enp1s0f0 | grep -Ei 'fec|ber|snr|osnr'

Add temperature and laser bias current to the routine. A bias current that creeps upward over months with a stable output power is a laser ageing; it is one of the few genuinely predictive signals available from a pluggable, and it justifies a planned replacement rather than a panicked one.

Symptom to Cause Mapping

• Link never comes up: DSP cannot converge, wrong channel/speed or modulation on one end, or a physical fibre break. Check DSP training state before blaming power.
• Intermittent flapping: OSNR margin too tight, contamination in a connector, or thermal cycling.
• Link up with correctable errors climbing: fibre ageing, amplifier drift, increasing connector loss - the classic slow degradation that a link-up SNMP trap will never tell you about.
• Link up with uncorrectable errors: severe OSNR deficit, non-linear penalty at high launch power, or DSP compensation exceeded. Clean connectors and verify per-channel power before replacing hardware.
• Errors only during the warmest hours: thermal expansion changing connector mating, or a module running above its specified operating temperature.

Why OSNR Is Not Always the Number You Think

Classic OSNR measurement (IEC 61280-2-9, the interpolation method) measures noise at the midpoints between the channel and its neighbours. That method breaks for polarisation-multiplexed coherent signals because the signal cannot be extinguished by a polarisation beam splitter, which is why in-band OSNR requires a polarisation-aware measurement or the module's own estimate. When comparing a field measurement with the design budget, make sure both were produced the same way - otherwise you will chase a phantom 3 dB.

Thermal Effects, Connectors and Slow Degradation

Optical degradation is rarely sudden. The failure modes that fill tickets creep:

• Connector contamination. A fingerprint or a dust particle scatters light, producing back-reflection and insertion loss that grows as the particle migrates. It is the single most common physical cause of a link that was fine last quarter and is marginal now.
• Amplifier ageing. An EDFA's output can droop slowly, or its noise figure can rise, quietly eating OSNR margin on a link that has not changed.
• Thermal expansion. Cables, patch panels and module cages all move with temperature, which is why "errors only in the afternoon" is a recognisable signature rather than a coincidence.
• Bias-current drift. Laser ageing under constant-power control shows up as rising bias current long before it shows up as BER.

None of these is visible from a single measurement. They are visible from a baseline and a trend, which is the whole reason the next section matters.

A Practical Turn-Up Order

1. Record pre-FEC BER, OSNR and EVM at turn-up as the baseline for the link.
2. Clean and inspect every connector before measuring anything - it resolves a surprising share of degradation tickets.
3. Verify per-channel power at each amplifier and equalise at the ROADM before judging OSNR.
4. Monitor the trend, not the instantaneous value: a pre-FEC BER that doubles over a month is a ticket long before it reaches the FEC limit.
5. Store the baseline where the on-call engineer can reach it, next to the link's design budget and the FEC limit for that module type.

FAQ

Should I replace a module at the first sign of pre-FEC BER? No. Pre-FEC BER is a quality indicator with enormous range; the question is always margin to the FEC limit and direction of travel. Replacing hardware on a single measurement wastes optics and hides the real cause. Replace when the trend is clearly downward and the physical checks - connectors, per-channel power, bias current - all come back clean.

Why does my module report a different OSNR than the test set? Because they measure different things. The test set measures in-band noise in the optical domain; the module estimates signal quality after equalisation. Use each for its own purpose and never compare them as if they were the same quantity.

Can a coherent link show zero post-FEC errors and still be broken? Yes, in a sense: a link can be error-free and still be running with almost no margin, so the next minor degradation produces an outage with no warning. That is exactly the case a pre-FEC BER trend is there to catch.

Related reading: Optical transceiver power: dBm RX/TX thresholds, DWDM system components: mux, EDFA, demux, ROADM power equalization and APC, DWDM 50GHz vs 100GHz channel spacing, Optical power budget calculation and SFP optical transceiver troubleshooting checklist.

原文链接：AscentOptics: Coherent optical module troubleshooting guide