3. Technical Implementation

3.1 The Wavelength Grid: ITU-T G.694.2

The 18 nominal center wavelengths (nm):

Important: "18 theoretical channels" and "channels usable in practice" are two different things. The gap is caused by the E-band water peak, discussed separately below.

3.2 Why 20 nm: An Arithmetic Exercise in Wavelength Tolerance Budget

This is the root of CWDM's entire cost advantage and deserves detail.

The wavelength of an uncooled DFB laser deviates from its design value for two reasons:

1. Manufacturing spread: wavelength variation between chips from the same batch. Typical upper bound about ±2.3 nm.

2. Temperature drift: chip temperature varies with the environment and the wavelength drifts with it. Across the industrial temperature range of −40 to +85 °C, measured drift is bounded at about ±4.2 nm.

Combined, an uncooled laser needs roughly ±6.5 nm of headroom in the worst case. Working backward from the −40 to +85 °C span (125 °C), the equivalent temperature coefficient is about 0.067 nm/°C.

G.694.2 defines a usable per-channel width of about 13 nm (the 20 nm spacing minus inter-channel guard bands). In other words:

Usable channel width 13 nm

− manufacturing spread ±2.3 nm

− temperature drift ±4.2 nm

= remaining margin ≈ 2 nm

(before modulation chirp and aging)

That budget already uses the 20 nm spacing to its limit. If the spacing were narrowed to DWDM's 0.8 nm (200 GHz), the same device would have to hold drift to the 0.01 nm/°C order - achievable only by adding a TEC to lock the chip temperature.

And so an order-of-magnitude cost gap appears:

That last row is interesting: TFF filters themselves are thermally excellent (under 0.002 nm/°C). The bottleneck was never the filter - it is the laser. CWDM's cost advantage fundamentally consists of moving the wavelength-stability problem from the component side to the system side.

As a side effect of low power: a 16-channel DWDM multiplexer needs a 6U chassis with fan trays and redundant power supplies, while the equivalent-function CWDM multiplexer fits in 1U - or even a single board. Where rack space is leased from a third party, that indirect cost (including air-conditioning load) pushes DWDM's lifecycle cost even higher.

3.3 The E-Band Water Peak: A Band That Exists in the Standard but Is Avoided in Practice

Standard G.652 single-mode fiber has a water peak near 1383 nm caused by hydroxyl (OH⁻) ion absorption. The typical excess loss there is about 0.5 dB/km, and in the worst case 2 dB/km or more.

The consequence is direct: 5 of the 18 CWDM channels fall in the E-band (1371 / 1391 / 1411 / 1431 / 1451 nm) and are unreliable on conventional G.652 fiber. Carriers would not accept the risk that purchased equipment might not work on some or all of their metro fiber, so the first metro CWDM products covered only the O, S, C, and L bands.

This is why 8-channel CWDM systems in practice commonly use one of two plans:

• S+C+L: 1471 / 1491 / 1511 / 1531 / 1551 / 1571 / 1591 / 1611 nm - all low-loss, the most commonly used.

• O-band: 1271 / 1291 / 1311 / 1331 / 1351 / 1371 … - low dispersion, suited to higher bit rates.

The solution comes from the fiber itself. ITU-T G.652.C substantially suppresses the 1383 nm water peak and releases the E-band. Both OFS and Corning supply fiber conforming to that standard (often marketed as zero-water-peak fiber). Where new cable is laid or metro fiber is upgraded, choosing G.652.C/D directly makes more channels usable.

One more useful by-product: early dispersion-shifted fiber (DSF) cannot be used with C-band DWDM due to four-wave mixing (FWM), but it can be used with CWDM. In markets with a large DSF installed base, such as Japan, this avoided tearing up and re-laying entire cable sections.

3.4 Passive Mux/Demux: TFF Dominant, AWG and FBG Supplementary

All wavelength selection in a CWDM system is performed by passive multiplexers/demultiplexers.

Mainstream approach: thin-film filters (TFF). TFF achieves wavelength selection through interference in multiple dielectric thin-film layers. A single-channel TFF is a discrete component; in practice they are combined into 4 / 8 / 16 / 18-channel modules using a cascaded structure.

One inherent property of cascaded structures must be noted: insertion loss accumulates with channel count. This is why, as channel count rises, the per-channel price advantage shrinks.

Other approaches:

• AWG (Arrayed Waveguide Grating). An integrated waveguide device with a clear size advantage at high channel counts, but usually applied to DWDM and high-count scenarios; less common in CWDM.

• FBG (Fiber Bragg Grating) plus circulator. Suited to OADM nodes, capable of single-wavelength add/drop.

• Broadband couplers. Used for wideband splitting such as 1310/1550 nm - essentially "two-wavelength" WDM, not usually called CWDM.

Typical specifications (quantitative reference only; always defer to vendor datasheets):

Unidirectional or bidirectional? CWDM multiplexers can be specified in two ways:

• Dual-fiber unidirectional: one fiber carries all downstream wavelengths, the other carries upstream.

• Single-fiber bidirectional (BiDi): on the same fiber, downstream uses one wavelength group and upstream uses another. The advantages are lower first-in cost (especially for leased-fiber scenarios) and directly halved fiber occupancy when fiber is exhausted.

3.5 Optical Modules: Uncooled DFB Is the Star

Transmitter

• Primary device: directly modulated DFB laser (DML), uncooled design.

• Why DFB: narrow linewidth and high side-mode suppression ratio, giving dispersion performance close to that of directly modulated DWDM lasers. As a result, CWDM lasers can transmit 2.5 Gbit/s over 80 km on G.652 fiber.

• Typical output power: about 1 mW (0 dBm) - not high, but sufficient when paired with an APD receiver.

• Higher-rate scenarios: at 10G and above, or over longer distances, EML (Electro-absorption Modulated Laser) is usually adopted to combat dispersion.

• The VCSEL route: 850 nm / 1310 nm VCSELs have long been in volume production for Gigabit and 10G multimode applications. Long-wavelength (1500–1610 nm) VCSELs have also been developed for single-mode WDM; 80 km transmission at 2.5 Gbit/s has been demonstrated with very small power penalty.

Receiver

• CWDM and DWDM receivers are essentially the same class of device, typically requiring a broader bandwidth covering the full CWDM band to accommodate multiple rates and protocols.

• The front-end detector is wavelength-agnostic PIN or APD; wavelength selectivity is provided entirely by the CWDM filter in front of it.

• PIN: low cost, simple circuitry.

• APD: improves receiver sensitivity by 9–10 dB, at the cost of high-voltage bias and temperature compensation.

A common misconception: a CWDM optical module is not a "generic module customized per wavelength." Each wavelength is a separate part number (e.g., CWDM-SFP-1471, CWDM-SFP-1491 …). Inventory and sparing strategy must be planned along the wavelength dimension - this is the most commonly underestimated cost item in CWDM operations.

3.6 Link Budget and Distance Limits

CWDM's distance limit arises from three superposed physical constraints.

Constraint 1: No amplification (the most fundamental). EDFAs (erbium-doped fiber amplifiers) operate only in the C-band (roughly 1530–1565 nm), whereas CWDM spans more than 340 nm across the O, E, S, C, and L bands. This means in a CWDM system, most wavelengths simply cannot be amplified by an EDFA.

A CWDM link is therefore fundamentally an unrepeatered passive network, with distance directly limited by the power budget - typically around 70–80 km (some datasheets state "up to 80 km"; the real figure depends on channel count and insertion loss).

The only ways to extend distance:

• OEO regeneration (3R regeneration): re-amplify, reshape, retime. Multi-wavelength 3R repeaters enhanced with OEO add/drop become regenerative OADMs (R-OADMs).

• SOA (semiconductor optical amplifier): broadband amplification covering O/E/S/C/L, suited to metro networks with many nodes and lower per-node capacity.

• 1R amplification: amplitude regeneration only, requiring either per-wavelength handling or a broadband amplifier.

Constraint 2: Chromatic dispersion. G.652 fiber has zero dispersion at 1310 nm, so dispersion is significant near 1611 nm. In the OADM bus case published by RBN, 2.488 Gbit/s (OC-48/STM-16) at about 80 km is already approaching the dispersion limit of many directly modulated CWDM lasers, driven mainly by accumulated dispersion at 1611 nm.

Constraint 3: Cumulative loss across cascaded OADMs. The maximum size of a passive CWDM bus network can be calculated directly, in much the same way as tap-loss calculations for the coaxial segment of an HFC network. Consider a 5-node, 60 km CWDM bus (an 8-wavelength Mux/Demux plus four 2-channel OADMs at 15 km intervals):

Assumptions: APD receiver, 1 dB OADM insertion loss (express and add/drop), 4 dB Mux/Demux insertion loss, 0.4 dB/km fiber plus splice loss at 1470 nm and/or 1610 nm.

The result is loss-limited, with node power levels of −10 / −11 / −17 / −18 / −24 / −25 / −31 / −32 dBm end to end.

The conclusion is clear: a CWDM network without amplification or regeneration is intrinsically confined to metro access scenarios with a small number of nodes. Larger scale requires repeaters or amplifiers.

3.7 CWDM vs DWDM: An Engineering Comparison

A conclusion validated repeatedly: CWDM and DWDM are not substitutes but layers. CWDM bridges DWDM's long-haul and core capacity down to the metro, campus, and end user. Where fiber is scarce and budget constrained, the two can even be deployed in hybrid mode (CWDM + DWDM hybrid): CWDM handling multi-protocol demultiplexing at the access layer, DWDM handling long-reach high-capacity transport at the core.

4. Key Components and System Composition

A complete point-to-point CWDM system consists of the following:

[Service Equipment]                          [Service Equipment]

             |                                                           |

[Colored Module]                               [Colored Module]

             |                                                           |

[CWDM MUX] ----- single-fiber / dual-fiber SMF ----- [CWDM DEMUX]

             |                                                           |

(optional OADM intermediate nodes, passive add/drop)

• Colored module: CWDM optical modules customized per wavelength - the "color source" of the whole system.

• CWDM MUX/DEMUX: passive multiplexer/demultiplexer, essentially cascaded passive TFF.

• OADM: optical add/drop multiplexer, passively adding or dropping specific wavelengths at intermediate nodes with no optical-electrical conversion.

• Optical protection: combined with optical switches, enables 1+1 optical channel protection, with millisecond-order switching time and complete protocol transparency.

• Fiber: G.652.C/D (low water peak) is recommended, to release E-band channels.

• Connectors and patch cords: outdoor scenarios commonly use IP65-rated enclosures with ODVA connectors.

CWDM's single greatest operational advantage is this: intermediate nodes can be entirely passive. No power, no fans, no software, no parameters to tune - in outdoor cabinets, street cabinets, and building telecom rooms, the reliability benefit of that property far outweighs any question of technical sophistication.