DWDM or CWDM: which ITU-T grid to link two company sites?

Two sites, a single pair of dark fibre available — perhaps leased, perhaps the last spare one in a shared consortium cable — and five services to carry over it: LAN extension, storage replication, telephony, video surveillance, a backup link to the data centre. The technical answer is almost always the same: multiplex several wavelengths onto the same fibre, with a passive multiplexer combining them at one end and a demultiplexer separating them at the other — to the equipment at each end, every channel remains a dedicated point-to-point link. The question the specification has to settle before anyone signs is a different one: CWDM or DWDM? The difference between the two lies entirely in the spacing between one channel and the next — and in what that spacing costs. It isn’t a matter of taste: two ITU-T Recommendations fix it precisely — and a third, surprisingly, leaves the most practical point blank.

CWDM: the ITU-T G.694.2 grid — fewer channels, cheaper lasers.

Recommendation ITU-T G.694.2, approved on 14 December 2003 and never revised since, sets the CWDM grid: 18 channels spaced exactly 20 nm apart, from 1271 nm to 1611 nm. The wide spacing is a deliberate choice: Appendix I explains that it exists to allow uncooled lasers, whose wavelength drifts with temperature — the standard tolerates a total drift of roughly ±6-7 nm — together with wide-passband filters, both considerably cheaper than their DWDM counterparts. On G.652.D fibre, free of the absorption peak around 1383 nm that penalised older fibre, all 18 channels are usable; on an older run, operators in practice stay within the 8 channels of the 1471-1611 nm window.

DWDM: the ITU-T G.694.1 grid — narrow channels, with no ceiling set by the standard.

ITU-T G.694.1 — now in its third edition, approved on 29 October 2020 after those of 2002 and 2012 — anchors the DWDM grid to a reference frequency: 193.1 THz. From there, a formula generates the permitted channels: 193.1 + n × 0.0125 THz for 12.5 GHz spacing, and likewise with a step of 0.025 (25 GHz), 0.05 (50 GHz) or 0.1 THz and multiples thereof (100 GHz and wider), with n an integer. The channel anchored at 193.1000 THz converts directly to 1552.5244 nm; a tighter 50 GHz step shifts the adjacent channel to 193.15 THz, i.e. 1552.1225 nm — a 0.4 nm difference, invisible to the eye but perfectly resolved by a narrow-grid demultiplexer. The third edition also added the flexible grid, with definitions for “frequency slot” and “slot width”, for channels of differing width sharing the same infrastructure. Unlike CWDM, though, the standard sets no maximum number of channels: the example table lists frequencies from 195.9 THz down past 185 THz, but says so explicitly — the endpoints shown are illustrative, not normative. How many channels actually fit on a link depends on the passband of the multiplexer and, where fitted, of the optical amplifier — not on any limit written into G.694.1.

Where the specification finds no number: what ITU-T G.671 does — and doesn’t — say.

For PON splitters, as we’ve written before: clause 6.6 of ITU-T G.671 sets minimum and maximum insertion loss for every split ratio, a precise figure to cite in a tender. For WDM multiplexers, the same Recommendation — now in its ninth edition, approved on 29 November 2025 — holds a surprise. Clause 6.19 distinguishes three families: CWDM (6.19.1), DWDM 1×X (6.19.2) and Wide WDM (6.19.3, with spacing wider still than CWDM). For the first two, almost every parameter — channel insertion loss, adjacent- and non-adjacent-channel isolation, crosstalk, PDL — is marked “ffs” (for further study) or “sba” (subject to bilateral agreement): the standard states the test method — IEC 61300-3-4 and 61300-3-7 for insertion loss, IEC 61300-3-6 for reflectance, IEC 61300-3-2 for PDL — but no pass/fail value. Only for Wide WDM does clause 6.19.3 set a formula: maximum insertion loss 1.5 × log₂(X) dB, where X is the number of channels. For CWDM and DWDM, that figure has to come from the manufacturer’s datasheet — and be verified with the same IEC method the standard already names.

DWDM or CWDM: which to choose.

With those two constraints clear, the choice comes down to a handful of questions, not a price comparison on the individual transceiver. How many independent services need to travel over the same fibre today, and in a few years? Above a handful of channels, the practical ceiling of CWDM’s 18 comes into view quickly. Will the link need optical amplification, or will it stay a passive point-to-point run? DWDM’s cooled lasers and narrow spacing are built for networks that may grow in reach; CWDM’s uncooled lasers remain the cheaper option on passive runs of limited length. And does whoever will certify the link already have the instruments to measure at a narrow grid? An OTDR or power meter calibrated for a CWDM run doesn’t automatically characterise channels 50 or 100 GHz apart.

For two sites consolidating four or five services onto a dark fibre, CWDM is almost always enough, and cheaper. For an interconnection between data halls that needs to carry dozens of channels, or grow in capacity without laying more fibre, DWDM is the only grid that allows it — at the cost of temperature-stabilised lasers and a specification that has to be stricter on loss values, precisely because the standard, here, doesn’t set them for you.

What to actually write in the specification.

The number of channels required today, and the growth margin expected — not just the CWDM or DWDM label.
The exact grid — nominal wavelengths (CWDM, ITU-T G.694.2) or frequency and spacing (DWDM, ITU-T G.694.1) — citing the Recommendation, not a generic “WDM-compliant”.
The maximum channel insertion-loss value declared by the manufacturer, with test method IEC 61300-3-4/3-7: the ITU-T standard does not supply it.
The link to the overall optical budget of the run: the multiplexer’s loss adds to cable, splice and connector losses like any other passive component.

The point.

CWDM and DWDM solve the same problem — too many services, one fibre — with two grids that ITU-T has fixed in opposite ways: rigid and channel-counted for the first, open for the second. Neither one, though, sets on its own how much the multiplexer is allowed to lose: that figure is written by whoever signs the specification. That’s why every site-to-site interconnection we design — two offices or two data halls — starts from the right grid and the loss value demanded of the manufacturer, not from the label on the box.

Do you need to link two sites or two data halls over fibre that’s already in the ground, and aren’t sure whether CWDM is enough? Talk to an engineer: the choice of grid is made together, against the real number of services to carry.