DWDM line card: replace, repair or re-specify?
Replace a card that shows monotonic pre-FEC BER growth against falling transmit power across 15-minute registers, or repeated OTUk-LOF and SES on clean fiber. Repair fits a failed pluggable line module or converter; re-specification fits a shelf where spares, quotes and capacity all point to a higher rate.
Line card anatomy: client mapping, G.709 framing, line optics
A DWDM line card turns client traffic into a wavelength. Client ports accept Ethernet, SDH/SONET, Fibre Channel or CPRI; the card maps each signal into an OTN container, adds G.709 overhead and forward error correction, and drives a laser onto a slot on the ITU-T G.694.1 frequency grid. Most shelves pass it through an amplifier and a ROADM before it reaches the span.
At 100G the frame arithmetic is explicit: a 100GbE signal arrives at 103.125 Gb/s after 64/66b coding (IEEE 802.3), OPU4 carries it at 104.36 Gb/s, ODU4 frames it at 104.79, and OTU4 transmits at 111.81 Gb/s (ITU-T G.709). Reed-Solomon RS(255,239) spends 16 of every 255 symbols on parity, 6.7% overhead, and corrects up to eight symbol errors per codeword.
Transponders, muxponders and OTN tributary cards all sell as line cards. A transponder carries one client per line port, a muxponder aggregates several clients into one line signal, and an OTN tributary card switches ODUk containers without touching the optical layer, so repair options follow the optics.
A CFP2-DCO module (CFP2-DCO MSA) or a QSFP-DD module sits in a cage, so a laser failure is a module swap; a soldered optical bench goes to a depot for the same fault, and 10G cards with XFP or SFP+ cages stay field-serviceable. The line card and interface card guide covers slot and chassis matching for both layouts.
Reading a pre-FEC BER curve: aging vs fault
Pre-FEC BER is the counter that shows a laser dying before traffic drops. Export it per 15-minute bin with transmit power, receive power, laser bias current and board temperature; G.874 15-minute and 24-hour registers supply the OTN counters, and SFF-8636 on a QSFP28 line optic or OIF-CMIS on a QSFP-DD adds module-level diagnostics. The 100G DWDM transceiver guide lists the module families in those cages.
The shape of the curve says more than its level. A step inside one bin that stays there is an event: a connector touched during maintenance, a splice, a dirty ferrule, an amplifier gain change. A curve that doubles every few weeks while transmit power falls is a laser losing slope efficiency, and cleaning will not recover it. Oscillation that tracks the afternoon temperature peak is a thermal-margin problem, usually airflow or a fan tray. A reset that clears BER for a few hours and lets it climb again points at firmware or a power rail.
Absolute BER measures distance to the FEC limit; it says little about speed of travel. Coherent DSPs on 100G and 200G cards are usually specified with pre-FEC thresholds between 1e-3 and 1e-2 in the vendor datasheet, so 1e-9 leaves six decades of margin and 1e-4 leaves one or two. Remaining life follows the doubling time: days remaining ≈ doubling time × log2(threshold ÷ current BER).
A card that moved from 1e-11 to 1e-9 in 120 days doubled 6.6 times, an 18-day doubling time. From 1e-9 to 1e-3 is 19.9 doublings, roughly 360 days. Treat the figure as an upper bound, re-measure monthly and derate it, because laser aging accelerates near end of life.
| Pattern in 15-minute registers | Likely cause | Test that separates them |
|---|---|---|
| Step inside one bin, stable afterwards | Connector, splice, patch panel, amplifier gain change | Inspect and clean end faces to IEC 61300-3-35, re-measure span loss |
| Monotonic rise, Tx power 2 dB below commissioning | Laser aging | Swap the line optic if pluggable; compare bias current with the commissioning record |
| Rise tracks board temperature, falls at night | Thermal margin, airflow | Check fan tray and inlet temperature; compare the slot with a neighbour |
| High BER in one direction, Rx power low | Receive side or upstream span | Measure power at both ends; check amplifier gain and tilt |
| Clears on reset, returns within hours | Firmware, PSU rail, DSP | Re-seat, check rail voltages, substitute a known-good card |
Corrected-block totals are the FEC working; uncorrectable blocks and SES are the FEC losing. A 24-hour sum matters more than any single 15-minute bin, and the alarm belongs on the trend.
Which alarm patterns point at the card, and which at the fiber plant?
G.798 defines OTN defects per layer, and the layer tells you where to look. LOS and LOF sit at the section level: no light, or light without frame alignment. OTUk-BDI is the far end reporting a receive defect, which places the fault upstream of the far end, in your transmit direction. OTUk-TIM means the trail trace identifier does not match the expected string, a provisioning error that usually follows a card swap. PM-BIP-8 errors with no LOS point at a degrading span or a marginal receiver.
Alarm history over 30 days beats the current alarm list. One LOS burst at 03:14 with nothing since is a fiber event. LOS every few hours at the same time of day is thermal or a loose connector. Repeated LOF on a span whose loss tests normal points at the card’s receiver or DSP.
| Alarm or counter | Layer, per G.798 | Usual meaning |
|---|---|---|
| LOS | OTUk/OTS | No incoming light: fiber cut, disconnected patchcord, failed amplifier |
| LOF | OTUk | Light present, frame alignment lost: severe BER or DSP fault |
| OTUk-BDI | OTUk | Far end has a receive defect: look upstream of the far end |
| OTUk-TIM | OTUk | Trace mismatch: provisioning after a swap |
| ODUk-OCI | ODUk | Open connection: cross-connection missing or not provisioned |
| PM-BIP-8 errors, SES | ODUk PM (G.826) | Errors that passed framing: degrading span or receiver |
| 10 consecutive SES | G.827 availability | The trail counts as unavailable |
A G.873.1 1+1 ODUk protection group can hide a failing card for weeks: it switches in tens of milliseconds, with 50 ms the usual objective, and leaves the failed card in service. Evidence sits in the protection-switch count and the BDI history; pull both before declaring a card healthy.
Three decision paths and their thresholds
Policy rows assume protected services at a 99.99% objective; the G.826 and G.827 rows are fixed by standard.
| Criterion | Repair | Replace like-for-like | Re-specify |
|---|---|---|---|
| Pre-FEC BER trend | Single step, stable for 30 days | Monotonic rise with Tx power falling | Healthy BER, line rate below 25% of shelf capability |
| FEC and SES | Zero uncorrectable blocks in 24 h | Any SES; 10 consecutive SES (G.827) | n/a |
| Alarm history, 30 days | OTUk-TIM only, after a swap | Repeated LOF or LOS on clean fiber | Repeated vendor EOL notices |
| Optics | Pluggable module failed | Soldered optical assembly or DSP failed | Pluggable 400ZR covers the link |
| Age and support | In support, under 5 years | EOL, over 7 years | EOL and capacity constrained |
| Spares | Spare on shelf, lead time under 2 weeks | No spare, lead time over 8 weeks | Spares scarce, fiber pair full |
| Repair quote | Under 30% of replacement | Over 50% of replacement | n/a |
| Capacity | Port count covers 24 months | Port count covers 24 months | Demand exceeds shelf by 2x |
When does repair beat replacement on cost?
Repair wins when the failed item is a field-replaceable sub-assembly. CFP2-DCO and QSFP-DD line optics, client SFP and XFP cages, fan trays and power modules come out with a screwdriver; the card stays in the shelf and the outage is a module swap. Depot repair wins when the fault sits in a socketed converter or connector and the quote is a small fraction of replacement. It loses when the DSP, FPGA or soldered optical bench is involved, because the bill approaches the price of a used card and the residual life is unknown.
Let r be the repair charge, R the replacement price and p the probability that the repaired card fails again inside your planning horizon. Repair is cheaper when p × R < R − r. At r = 0.3R, replacement wins once p exceeds 0.7; at r = 0.5R, it wins above 0.5. A shop that cannot name the sub-assembly it replaced gives you no basis for estimating p.
A repair report should name the replaced part number, list before-and-after test results and state the warranty term. A report that says “repaired and tested” without a part number is a red flag. Board-level repairs return with default firmware, so budget an hour for commissioning and a trace check; a mismatched expected TTI produces OTUk-TIM that reads like a fresh fault.
How much does each path cost over five years?
Five-year cost is dominated by what buyers leave out of the quote: spare holding, downtime exposure and the second failure. A like-for-like replacement sets the index at 1.00; other values are policy inputs to swap for your own quotes.
| Cost line | Repair | Replace like-for-like | Re-specify to 400G |
|---|---|---|---|
| Card or repair charge | 0.30 | 1.00 | 2.60 |
| Freight, both ways | 0.03 | 0.02 | 0.05 |
| Engineering and commissioning | 0.05 | 0.03 | 0.30 |
| Spare holding, 5 years at 20% a year | 1.00 | 1.00 | 2.60 |
| Expected repeat failure (p × replacement) | 0.35 | 0.10 | 0.10 |
| Downtime exposure, protected service | 0.05 | 0.05 | 0.05 |
| Power and rack space, 5 years | 0.15 | 0.15 | 0.10 |
| Total | 1.93 | 2.35 | 5.80 |
At these inputs repair totals 1.93 against 2.35 for replacement and 5.80 for re-specification, so re-specification carries more than twice the five-year cost before any capacity benefit is counted.
The ranking flips when the service is unprotected, because repair turnaround becomes outage time. A card with a 200,000-hour MTBF (Telcordia SR-332 style prediction) and an eight-week repair cycle gives 200,000 ÷ 201,344 = 99.33% availability, about 58 hours down a year; with a one-hour swap the same card reaches 99.9995%, about 2.6 minutes.
When is re-specifying the line rate the cheaper answer?
One C-band pair makes the comparison concrete: 4.4 THz of usable width between 1530 and 1565 nm, 88 slots at 50 GHz carrying 100G for 8.8 Tb/s, and 58 slots at 400G for 23.2 Tb/s, a 2.6× capacity gain on the same glass.
| Line rate | Slot width | bit/s/Hz | Capacity on 4.4 THz of C-band |
|---|---|---|---|
| 100G | 50 GHz | 2.0 | ~8.8 Tb/s |
| 200G | 50 GHz | 4.0 | ~17.6 Tb/s |
| 400G | 75 GHz | 5.3 | ~23.2 Tb/s |
| 400G | 100 GHz | 4.0 | ~17.6 Tb/s |
| 800G | 100 GHz | 8.0 | ~35.2 Tb/s |
On a 16-slot shelf, 400G per slot gives 6.4 Tb/s and 100G per slot gives 1.6 Tb/s. Serving 3.2 Tb/s takes two cards at 400G and eight at 100G; the difference shows up in power, rack units and spare count as much as in capex. The ROADM side needs checking too: 400G at 75 GHz requires a flex-grid or 75 GHz-capable wavelength-selective switch plan, and a fixed 50 GHz line system forces 100 GHz slots, giving back part of the gain. The CWDM vs DWDM metro guide covers the outside-plant constraints.
Where a router-to-router link runs under 80 km, an OIF 400ZR QSFP-DD removes the transponder card entirely and the router port becomes the line port. Conditions are strict: the 80 km objective in the 400ZR implementation agreement, router support for the module’s power and cooling, and a free grid slot. Links that need ODUk cross-connection, tandem connection monitoring or protection switching keep a card. A 10G card can give way to a 100G card in the same slot only where the backplane, cooling budget, shelf manager, ROADM grid and NMS model library all support the higher rate; vendors publish a compatibility matrix per shelf.
A shelf with three years of life left and 30% utilisation does not justify re-specification; firmware, NMS templates, training and a second spares pool carry over as cost.
Spare-parts position and repair viability
Every repair path assumes a spare on the shelf. Without one, turnaround is outage, and the availability arithmetic above rules out a 99.99% objective. The minimum position for a protected service is one spare per shelf type, powered, licensed and configured, not one per card population; for unprotected single-card services, hold a spare per site or publish the MTTR you will accept. The rule breaks when a shelf carries two card generations: one spare then covers neither population, and the position needs one per generation.
The secondary market changes the answer for end-of-life cards. Stock lists carry Huawei OSN-series boards such as the TN54THA-002 and TN13M40V02, Nokia 1830 PSS shelves and Huawei OSN 8800 T32 subracks, so a card the OEM no longer sells can still arrive within days; the equipment catalogue lists condition and part number for each unit. A card bought this way carries no OEM support entitlement, so the supplier’s test record and warranty take its place.
Firmware and licence state need checking before purchase, and the load should match the shelf’s running firmware; a spare on an older load can be rejected by the shelf manager or come up degraded, and a trace mismatch produces OTUk-TIM after the swap. A spare that sits unpowered for two years carries an unknown load and an unknown laser: power it up once a quarter, read the DDM values and compare them with the commissioning baseline. A module whose bias current has crept 15% above the recorded value during storage will fail early in its first year of service.
Pre-pull test sequence
Fiber comes first, because that is where the cheap failures are. Inspect and clean both end faces to the acceptance zones of IEC 61300-3-35, then measure span loss against the commissioning record; a connector that adds 3 dB looks exactly like a receiver going deaf.
Shelf isolation comes next: loop the client signal at the client port to prove the client side, and loop the line side through an attenuator sized for the receiver’s range to prove the transmit chain. A PRBS (ITU-T O.150) through the OTN path, read at the far end, gives the error count. A card that passes a 24-hour run with zero errors at nominal power is not the fault, whatever the BER counter recorded during the event. The step that gets skipped is the attenuator: a bare loopback can overload the receiver, and the test then passes for the wrong reason.
If the line optic is pluggable, swap it with a known-good module before opening a repair ticket. The module swap clears a large share of card-level tickets at module cost. Keep the removed module and its DDM log; bias current and temperature history are the evidence a repair vendor will ask for.
Frequently asked questions
Can a DWDM line card be repaired at component level?
Yes, if the part is accessible. Power converters, connectors, pluggable optics and fan trays are routine depot work. DSP, FPGA and soldered optical assemblies usually are not: the parts are not sold separately and rework risk exceeds the value of a used card. Ask for the replaced part number before approving the quote. A converter swap is worth doing; a 1,000-ball FPGA reflow rarely is.
Does a rising FEC-corrected error count mean the card is failing?
No. Corrected errors are the FEC doing its job. The counters that matter are uncorrectable blocks, ES and SES; G.826 defines SES as a second containing at least 30% errored blocks. A card whose corrected count climbs while uncorrectable stays at zero still has margin. Once SES appears, frames are being dropped.
How long should a DWDM line card last?
Vendor MTBF predictions for telecom line cards are usually quoted between 100,000 and 300,000 hours (Telcordia SR-332 methods), 11 to 34 years at a 40°C base. Lasers age faster than boards: a line optic that has run eight years at high case temperature can fail well inside that figure, and the pre-FEC BER trend predicts it better than the purchase date.
Is a refurbished DWDM line card safe for production?
A tested refurbished card with a warranty is normal practice for an end-of-life shelf. Insist on test records at the line rate you will run, the optical module’s bias current, and a return window covering a full commissioning cycle. Our refurbished equipment inspection checklist lists the checks to ask for.
When is re-specifying to a different line rate cheaper than repairing?
When the shelf still has spare capacity at a different rate. Moving a marginal 10G lambda to a lower-rate service, or consolidating two half-full lambdas onto one 100G card, avoids the repair bill and the spare-part search at the same time. The deciding number is the cost of the free slot versus the cost of the repair, not the age of the card.
Need a specific part?
Send the part number and quantity — we confirm condition, availability and price before you commit.