Core: Transport technologies

Optical transport - CWDM vs DWDM

Miro Kosut

Miro Kosut

— 7 min read
Share

Requirement / business driver

Carry capacity across distance - campus/metro links, data-centre interconnect (DCI), or fibre-exhaust relief - as cost-effectively as the growth curve allows.

Constraints

Number of wavelengths needed now and later, distance (amplification?), budget, own vs leased fibre, and operational capacity.

Options

  • Grey optics over dark fibre - one wavelength per pair; simplest.
  • CWDM - ~8-18 widely-spaced wavelengths, passive (no amplification), cheap, ~40-80 km.
  • DWDM - dense wavelengths (40-96+), amplified / coherent, long-haul, higher cost.
  • Leased wavelength / managed service - provider owns the fibre and optical layer.

Comparison

optical-transport-matrix

Tradeoffs

CWDM trades channel count and distance for low cost and simplicity; DWDM buys high capacity and long reach at real capex/ops cost; grey-over-dark is fine for a single short link; leased-lambda swaps capex for recurring opex when you have no fibre or no optical-ops team.

Recommendation / justification

Modest metro capacity on a budget -> CWDM; high channel count, long distance, amplification, or growth -> DWDM (coherent for very long-haul); short point-to-point -> grey optics on dark fibre; no fibre / no ops capacity -> leased wavelength. Justify by the capacity roadmap and distance, not just today's link.

What would change this (mid-scenario twist)

  • Capacity outgrows CWDM's channels -> migrate to DWDM.
  • Distance/amplification needed -> DWDM (coherent).
  • No spare fibre -> leased wavelength or DWDM to reuse existing strands.

Validation checks

  • Does the optical power budget close (loss, connectors, distance)?
  • Is there a wavelength plan and protection (diverse path) for critical spans?

IPv6 / dual-stack note

Purely L1 transport - agnostic to v4/v6.

ZR/ZR+ pluggables - DCI without a line system

The modern metro-DCI default is the coherent pluggable: 400/800G ZR/ZR+ optics (DP-16QAM coherent modulation) sit in ordinary switch ports and reach ~120-500 km, muxable through passive DWDM filters to tens of terabits per fiber pair - so a dedicated telecom line system (transponders, ROADMs, amplifier chains) is only required for long-haul distances or multi-point optical topologies. The companion economic rule: trenching and right-of-way dominate DCI cost, not the fiber itself, so builds lay far more pairs than needed - overprovisioned dark fiber is the cheap hedge against future bandwidth demand.

The fibre plant decides before you do

DWDM rides whatever glass is already trenched, and the fibre type constrains the mux plan. G.652 (standard SMF) is the workhorse: fine at 1310 nm, good for DWDM once dispersion is managed. Legacy G.653 dispersion-shifted fibre is hostile to C-band DWDM - its zero-dispersion point sits mid-band, so non-linear effects wreck dense channels (single-channel 1550 was its whole point). G.655 NZDSF keeps a little dispersion on purpose and takes DWDM across C+L. Reduced-water-peak G.652.C opens the water-peak attenuation valley - the variant that makes dense (>8-wavelength) CWDM viable. Band names to reason with: C (1530-1565 nm) is where EDFAs live and DWDM starts, L (1565-1625 nm) is the doubling lever. Survey the plant - fibre type, splice records, real measured loss - before promising a channel plan.

Three linear impairments frame every span: attenuation, chromatic dispersion (CD), and OSNR. The budget is arithmetic: Tx output power - Rx sensitivity = budget, spent on fibre (engineer at ~0.25 dB/km at 1550 nm), splices, patch panels, bends and connector dirt, keeping ~1 dB margin - a 20 dB budget is roughly 80 km of dark fibre, but measure (OTDR or source-and-meter) rather than assume. The trap is multiplexing: mux/demux filters tax the same budget - push 64 channels through a filter pair (~11 dB) and that 20 dB span collapses to ~9 dB, ~36 km - reach more than halves the moment a link goes multi-channel. CD grows linearly with distance and starts biting at 10G+: compensate optically (DCM spools, which burn budget) or let coherent DSP absorb it electronically - one more reason coherent won. In amplified systems the currency becomes OSNR: EDFAs add noise as they amplify, and when amplification and compensation run out the last resort is 3R regeneration (re-amplify, re-shape, re-time) - per-channel, bit-rate- and protocol-specific, the costliest tool on the shelf.

ROADM degrees: broadcast-and-select vs route-and-select

A ROADM node is built from wavelength-selective switch (WSS) stages, and the internal architecture is a real fork. Broadcast-and-select splits every ingress passively to all degrees and lets each egress WSS block what it doesn't want - simple, but splitter loss grows with the degree count. Route-and-select adds a second WSS per degree so nothing is broadcast: lower insertion loss, OSNR preserved, and colorless / omnidirectional add-drop scales. Service-card vocabulary worth owning: a transponder maps one client onto one wavelength, a muxponder aggregates many, and an alien wavelength is someone else's coherent signal carried over your line system - the interop case the OpenZR+/OpenROADM ecosystems exist to standardise.

The coherent-pluggable reach ladder

The pluggable family now covers nearly the whole distance axis, and the ladder is worth knowing by rung. 400G-ZR: metro access, ~120 km, C-FEC. 400G-ZR+: O-FEC and trunk-rate flexibility - 1,400 km at 400G, and downshifting the trunk to 300/200/100G buys further reach and tolerates worse OSNR (an n x 100GE muxponder mode included). Bright ZR+ raises launch power to +1 dBm - less amplification needed on the same span. ULH pluggables stretch to ~3,000 km by spending spectrum: higher baud (up to ~118 GBaud) in wider flex-grid circuits (~137.5 GHz) with probabilistic constellation shaping - flex-grid stops being optional at these widths. The 800G generation (up to ~131 GBaud, oFEC) adds the other lever: C- and L-band tunable variants, doubling a fibre pair. The mechanism under all of it: bits-per-symbol versus OSNR - polarisation-multiplexed 16QAM at 400G demands ~22.5 dB OSNR where QPSK at 100G lives at ~11.5 dB - modulation density is traded directly against reach. Planning anchors per fibre pair: C-band about 64 x 400G or 32 x 800G = ~25.6 Tb/s, roughly ~51 Tb/s with C+L.

The line system is disaggregating too

Routed optical networking removed the transponder shelf; the line system itself is now shrinking - down to a dual-EDFA in a pluggable hosted in ordinary router ports for point-to-point spans up to ~120 km, with compact 1-2RU open line systems above that. Three behaviours of modern OLS design matter to an architect: an ingress amplifier on the line terminal exists because ZR-class sources launch weak (~-10 dBm) - the line system compensates for the pluggable; embedded ASE loads the full spectrum from day one so amplifier gain and per-channel performance do not shift as real channels are added; and C to C+L expansion is hitless on current systems - the capacity double without a maintenance window. One operating system across the IP and optical layers is the quiet operational win the routed-optical pitch rests on.

Spaced repetition

The key practical difference between CWDM and DWDM is ==CWDM = few widely-spaced channels, passive/no amplification, short-metro; DWDM = many dense channels, amplified/coherent, long-haul==.

You pick the optical-transport option primarily on [...] (not just the current single link).

You pick the optical-transport option primarily on the capacity roadmap and distance (not just the current single link).

When you have no spare fibre or no optical-ops team, the natural choice is a [...].

When you have no spare fibre or no optical-ops team, the natural choice is a leased wavelength / managed service.

[...] put 400/800G DCI into ordinary switch ports for ~120-500 km, deferring full telecom line systems to long-haul or multi-point - and since trenching dominates cost, DCI builds overprovision fiber pairs.

ZR/ZR+ coherent pluggables put 400/800G DCI into ordinary switch ports for ~120-500 km, deferring full telecom line systems to long-haul or multi-point - and since trenching dominates cost, DCI builds overprovision fiber pairs.

Which legacy fibre type is hostile to C-band DWDM, and why?

G.653 dispersion-shifted fibre - its zero-dispersion point sits mid-C-band, so non-linear effects wreck dense channels. G.652 SMF is the DWDM workhorse (with dispersion management), G.655 NZDSF handles C+L, and reduced-water-peak G.652.C is what makes dense CWDM viable.

Dark-fibre budget arithmetic for a DWDM span?

Budget = Tx power - Rx sensitivity, spent at ~0.25 dB/km (1550 nm) plus splices, connectors and bends with ~1 dB margin - 20 dB is roughly 80 km. Going multi-channel taxes the same budget with mux/demux filter loss (a 64-channel pair ~11 dB -> ~36 km). Measure with OTDR; don't assume.

Broadcast-and-select vs route-and-select ROADM?

B&S passively splits every ingress to all degrees and blocks at the egress WSS - simple, but splitter loss grows with degree count. R&S adds a second WSS per degree: no broadcast splitter, lower insertion loss, OSNR preserved, and colorless/omnidirectional add-drop scales.

The coherent-pluggable reach ladder: ZR [...], ZR+ [...], ULH [...] - and L-band variants [...].

The coherent-pluggable reach ladder: ZR ~120 km metro, ZR+ 1,400 km at 400G with rate-downshift buying more reach, ULH ~3,000 km via higher baud in wider flex-grid circuits - and L-band variants double the fibre pair.

In amplified DWDM the design currency is [...]; when amplification and dispersion compensation run out, the last resort is [...].

In amplified DWDM the design currency is OSNR - EDFAs add noise as they amplify; when amplification and dispersion compensation run out, the last resort is 3R regeneration: per-channel, rate- and protocol-specific, the costliest option.

Sources

  • Cisco Live BRKOPT-1007 Optical Networking Fundamentals (CL 2026, 57 slides, compromissless full pass): fibre-plant table, impairment triad and budget arithmetic incl. the multi-channel filter tax, ROADM fork, coherent-pluggable reach ladder, line-system behaviours. Platform tables and DCO spec sheets excluded by invariant.
  • Cisco Press, DWDM Network Designs and Engineering Solutions; Optical Network Design and Implementation.

domain: Core · blueprint-ref: Core 1.2 CWDM/DWDM · type: design-decision · status: complete · tags: [core, core/transport, tradeoff/cost, tradeoff/scale]

Read more