CWDM vs DWDM: Which Wavelength Technology to Choose?

2026年9月2日 LOG admin

Wavelength Division Multiplexing (WDM) lets multiple data streams travel over one fiber pair by giving each stream its own color of light, and the two mainstream flavors - Coarse WDM (CWDM) and Dense WDM (DWDM) - split the optical spectrum very differently. This article compares CWDM and DWDM on channel spacing, reach, cost and upgrade path, so you can pick the technology that matches the distance and capacity of your link instead of defaulting to whichever is more popular.

The short version: CWDM is cheap, simple and passive, and it works well on short links with modest channel counts. DWDM is denser, amplifier-friendly and scalable into multi-terabit territory, and it costs more per component. The decision usually comes down to three numbers — distance, required channel count, and how much growth you expect in the next five years.

Channel Spacing: The Core Difference

CWDM follows ITU-T G.694.2 with channels spaced 20 nm apart, covering 1270-1610 nm - up to 18 channels total (9 in each direction with BiDi pairs). DWDM follows ITU-T G.694.1 with channels packed 100 GHz (~0.8 nm) or 50 GHz apart inside the C-band (and L-band) around 1550 nm, fitting 40, 88, 96 or even 160 channels onto one fiber.

The naming is literal. "Coarse" refers to the wide 20 nm spacing, which allows low-cost, loosely specified lasers because each channel sits in a wide, forgiving slot. "Dense" refers to narrow spacing that squeezes many wavelengths into a small band, which requires tightly controlled, temperature-stabilised (cooled) lasers and much more precise filters. Everything else about the two technologies follows from that one design difference.

Two practical consequences follow. First, a CWDM mux is a passive device with no power supply and no control plane; a DWDM terminal is often a full system with amplifiers, per-channel monitoring and management. Second, the failure modes differ: CWDM faults are mostly connector and loss problems, while DWDM adds amplifier and OSNR-related failure modes.

Reach and the Water Peak Problem

CWDM - up to about 70 km unamplified; between 40 and 70 km the usable channel count shrinks because the 1370-1430 nm range sits in the fiber water peak where loss reaches ~1.0 dB/km versus ~0.25 dB/km in the 1550 nm region.
DWDM - up to 80 km unamplified and 1000+ km with amplification (EDFA, Raman), because all channels live in the low-loss 1550 nm window where erbium-doped amplifiers work best.

The water peak deserves emphasis because it silently breaks CWDM designs. Legacy multimode and older single-mode fibre attenuates the 1370-1430 nm channels much more heavily than the rest of the plan. A CWDM link designed with 8 channels over 50 km of such fibre can work perfectly for the channels below 1370 nm while the mid-band channels fail outright — and the failure looks like a bad optic, not a fibre property. On modern low-water-peak fibre (often marked ZWP or "low water peak") the effect is small, but you must know which fibre you actually have before you commit to a channel plan.

Cost and Deployment Reality

CWDM wins on cost and simplicity for short links and limited capacity: uncooled lasers, passive mux/demux, no amplifier planning. DWDM costs more per mux and per transceiver but buys capacity headroom - adding channels later only requires new transceivers if the mux was sized correctly. As one CTO put it, the shift is rarely a preference for DWDM over CWDM; it is capacity and distance catching up with you earlier than expected. For links where growth is likely, starting with DWDM avoids a forklift upgrade.

Be careful with the word "cheaper", though. The mux and the optics are cheaper for CWDM, but if you later need amplification you cannot amplify CWDM on the same plan — EDFAs only cover the C-band around 1550 nm. That means a CWDM system that outgrows its reach cannot be extended with an amplifier; it must be replaced. The cheapest option at day one is not always the cheapest over the asset's life, and reach plus capacity growth are what make the difference.

Comparison at a Glance

Factor	CWDM	DWDM
Standard	ITU-T G.694.2	ITU-T G.694.1
Channel spacing	20 nm	100/50 GHz (~0.8/0.4 nm)
Typical channels	Up to 18	40-160
Wavelength region	1270-1610 nm	C-band / L-band (~1530-1625 nm)
Unamplified reach	~70 km (less in water peak)	~80 km
Amplified reach	Not applicable	1000+ km
Laser type	Uncooled, wide tolerance	Cooled, tightly specified
Typical use	Metro/access, short dark fiber	Long-haul, high capacity, DCI

How the Passive Part Works

A WDM link has four elements on each end, in order: transceiver, mux/demux, the fibre span itself, and the far-end demux/mux. The mux is a passive optical filter that combines wavelengths onto one fibre; the demux splits them apart again. Because the mux is reciprocal, the same unit typically performs both functions depending on which port light enters.

Two hardware variants dominate:

Band or filter-based mux — uses thin-film filters or arrayed waveguide gratings. Low loss (typically 1.5-3 dB insertion loss for 8 channels, more for CWDM's wider plan), fully passive, and the standard for both CWDM and small DWDM terminals.
Add/drop or ROADM-based system — allows individual wavelengths to be added or dropped at intermediate sites without breaking the express path. This is the DWDM option that makes multi-site rings practical, and it is where the cost and complexity of DWDM mostly lives.

Insertion loss compounds: a mux adds loss, a demux adds more, connectors add roughly 0.3-0.5 dB each, and splices add ~0.1 dB. Budget them all before choosing optics.

CWDM Channel Plan in Detail

The G.694.2 grid runs from 1271 nm to 1611 nm in 20 nm steps, giving 18 nominal channels: 1271, 1291, 1311, 1331, 1351, 1371, 1391, 1411, 1431, 1451, 1471, 1491, 1511, 1531, 1551, 1571, 1591 and 1611 nm. In practice, 1361 and 1391 are frequently skipped, and older plans often start at 1470 nm to stay clear of the water peak entirely. A common "8-channel" CWDM deployment uses the upper band (1471-1611 nm) because those channels are both low-loss and amplifier-compatible, which incidentally makes a later DWDM migration more plausible.

DWDM Channel Plan: 50 GHz vs 100 GHz

The G.694.1 grid is defined in frequency, not wavelength, anchored at 193.1 THz. A 100 GHz grid places 44 channels in the conventional C-band; a 50 GHz grid doubles that. The trade-off is filter tolerance: narrower spacing demands more precise lasers and filters and slightly higher insertion loss, and it makes upgrade paths between 50 and 100 GHz grids more awkward than it first appears. Choose the spacing at design time and stay consistent — mixing grids on one system causes interleaving problems that are painful to unwind.

Design Rules: Choosing Between Them

Count the services you need today and in three years. If the answer is 8 or fewer and the span is under about 40 km, CWDM is almost always the right answer.
Measure the actual span length and loss, not the map distance. Fibre route distance, splice count and connector count determine the real budget.
Check whether amplification will ever be needed. If the link is long, or might grow, or must traverse a site where regeneration is impossible, CWDM cannot be extended and DWDM should be considered now.
Confirm the fibre type. Low-water-peak fibre unlocks the mid-band CWDM channels; legacy fibre does not.
Decide on protection. A protected path consumes a second wavelength per service, which can double the required channel count — the fastest way to exhaust a CWDM plan.
Choose optics that match the mux. A DWDM transceiver cannot be muxed by a CWDM filter, and vice versa; the channel grid must be consistent end to end.

Power Budget and Overhead

The working inequality is simple: transmit power − mux loss − fibre loss − connector and splice loss − demux loss − margins ≥ receiver sensitivity. Keep at least 2-3 dB of margin for connector work, fibre ageing and measurement error, and more if the route includes patch panels that will be opened during maintenance. For DWDM, add amplifier gain and noise figure to the calculation: amplification raises power, but it also adds noise, so OSNR — not raw power — is often the limiting factor on long amplified spans.

Migrating from CWDM to DWDM

Migration is more practical than it sounds if the original plan was built thoughtfully. The cleanest route is to have deployed CWDM in the upper band (1471-1611 nm) and to have bought a mux platform that accepts DWDM filters alongside the CWDM ones. In that case, adding DWDM capacity means adding a DWDM filter set and DWDM transceivers to the same fibre, keeping the existing CWDM services running. If the original deployment used the lower band or a fixed sealed mux, the migration usually means replacing the mux at both ends and re-terminating — plan for an outage window.

Commissioning and Verification

# Per-channel power and wavelength check at each end
show interfaces optical 0/1/0/4

# DWDM terminal: per-channel power and OSNR
show controller optics 0/0/0/0

# Physical verification with a power meter, before service turn-up
# Measure each channel individually at the far-end demux output port

Verify each channel's received power against the budgeted value, channel by channel — aggregate power can look correct while a single channel is 5 dB low.
Confirm measured wavelength matches the assigned grid slot; a slightly-off laser still passes traffic but steals margin and can interfere with neighbours on a dense grid.
Record a baseline trace or spectrum for every span, so future degradation has something to be compared against.
Test with the final patch cords and the final connectors in place; measurements taken during commissioning with temporary leads routinely differ from production values.

Common Mistakes and Troubleshooting

Wavelength on the wrong port. The classic CWDM fault: a transceiver installed in the wrong mux port produces a completely dark channel with no alarms anywhere else. Check the port-to-wavelength map first.
Dirty connectors. A single contaminated connector costs several dB and degrades every channel on the fibre. Clean before you debug.
Budgeting from map distance. Route distance is almost always longer than the map line and includes splices nobody documented.
Ignoring the water peak. Mid-band channels failing on legacy fibre is a design error, not a hardware fault.
Assuming CWDM can be amplified. It cannot on the same plan; EDFAs do not cover the CWDM band.
Mixing 50 GHz and 100 GHz components. Causes adjacent-channel interference that appears intermittently and is hard to attribute.

FAQ

Can CWDM and DWDM share one fibre? In principle they occupy different bands and can coexist, but in practice the filters, EDFA band and management model clash; it is rarely worth the complexity. Use one technology per fibre and split services by fibre instead.

Which is better for data centre interconnect? For short DCI with high capacity per link, DWDM (often coherent pluggables) wins on channel count and reach. For a campus or metro ring with a handful of services, CWDM is far cheaper.

How many CWDM channels can I really use? Eighteen exist on the grid but real deployments rarely exceed eight to twelve, limited by fibre water peak, available port count and the optics you can actually buy.

Do I need cooled lasers? For DWDM, yes — wavelength stability matters at 0.4-0.8 nm spacing. For CWDM, uncooled lasers are normal and are a large part of the cost advantage.