Why the grid is defined in frequency

A 50 GHz channel is 0.401 nm wide at 1550 and 0.286 at 1310. That is not a rounding difference.

The anchor

The DWDM grid is built from a single reference frequency — 193.1 THz — plus a whole number of channel spacings, in both directions. Everything else follows from that.

Wavelengths are derived, not defined. 193.1 THz happens to be 1552.524 nm, which is why that oddly specific figure turns up everywhere in optical documentation.

Why frequency and not wavelength

Because frequency is what the hardware controls. A laser is stabilised to a frequency by a temperature control loop; a filter passes a frequency band. Wavelength is a derived quantity, related by λ = c/f — which is a reciprocal, not a proportion.

The consequence catches people out regularly. On a 50 GHz grid, adjacent channels differ by:

• 0.286 nm near 1310 nm
• 0.390 nm at 1530 nm
• 0.401 nm at 1550 nm
• 0.409 nm at 1565 nm

So anyone who builds a channel list by repeatedly adding 0.4 nm produces numbers that are approximately right in the middle of the C band and progressively wrong towards either end. Across a full band the accumulated error is more than a channel.

The width in nanometres is Δλ = λ²·Δf/c. The λ² is the whole story: the same frequency span occupies more wavelength as wavelength increases.

CWDM is the other way round, for a reason

CWDM channels genuinely are spaced in wavelength — 20 nm apart, starting at 1271 nm — and the reason is economic rather than physical.

A DWDM laser has to be held within a few gigahertz of its channel, which needs a thermoelectric cooler and a control loop to keep it there as ambient temperature changes. A CWDM channel is roughly a hundred times wider, so an uncooled laser that drifts several nanometres over its temperature range still sits comfortably inside its channel. Remove the cooler and the control electronics and the optics become dramatically cheaper.

The trade is capacity and reach: eighteen channels rather than ninety-six, and no optical amplification, because the channels are spread far outside the narrow band an erbium-doped amplifier covers.

The water peak

Four CWDM channels — roughly 1371 to 1431 nm — sit in a region where residual hydroxyl ions from manufacturing absorb strongly. On older fibre those channels can be several times lossier than their neighbours, which makes them effectively unusable.

Low-water-peak fibre eliminates it, and most fibre installed this century is. But a great deal of older cable is still in the ground, and a plan that assumes all eighteen channels are available on a route of unknown vintage is a plan that will need revisiting after the first OTDR trace.

The channel clash that passes a naive check

Two services on the same channel is easy to catch: sort the list and look for duplicates.

The one that gets missed is overlapping spectrum on different channel numbers. Put a 100 GHz carrier on channel 0 of a 50 GHz grid and it occupies the spectrum channel 1 was meant to use. The channel numbers are different, so a check that compares numbers passes it, and the collision only appears as unexplained errors on one of the two services.

This is ordinary rather than exotic. Mixed line rates on one fibre — 10G alongside 100G or 400G — means mixed channel widths, and the moment widths differ, channel numbers stop being a sufficient description of what occupies what.

The only reliable check compares actual occupied spectrum: centre frequency plus and minus half the carrier width, for every service, against every other.

Flex-grid, briefly

Newer systems abandon fixed channel spacing altogether and allocate spectrum in 12.5 GHz slices, so a carrier takes as many slices as it needs. A 400G carrier might occupy 75 GHz where a 10G one takes 37.5.

That makes the overlap question the only question — there are no channel numbers to compare, just occupied spectrum and gaps. It also introduces fragmentation: after enough churn, the free spectrum is scattered in pieces too small to fit a wide carrier, exactly as a disk fragments.

Frequently asked questions

Why is the DWDM grid defined in frequency?

Because that is what the hardware controls — lasers are stabilised to a frequency and filters pass a frequency band. Wavelength is derived, and the relationship is reciprocal, so channels evenly spaced in frequency are unevenly spaced in wavelength.

How wide is a 50 GHz channel in nanometres?

About 0.401 nm at 1550, but only 0.286 at 1310 and 0.409 at 1565. The width scales with the square of the wavelength, so treating it as a constant across a band accumulates more than a channel of error.

What is 193.1 THz?

The anchor the grid is built from — every channel is that frequency plus a whole number of spacings. Its wavelength is 1552.524 nm, which is why that figure appears throughout optical documentation.

Why is CWDM spaced in wavelength?

Because the point of CWDM is cost. A 20 nm channel is wide enough for an uncooled laser that drifts with ambient temperature, removing the thermoelectric cooler and control loop. The trade is eighteen channels instead of ninety-six and no optical amplification.

What is the water peak?

A region near 1383 nm where residual hydroxyl ions absorb strongly. Four CWDM channels sit in it, and on older fibre they can be several times lossier than their neighbours. Low-water-peak fibre removes it, but plenty of older cable is still in service.

How can two different channel numbers clash?

If the carriers are wider than the grid spacing. A 100 GHz carrier on channel 0 of a 50 GHz grid occupies channel 1's spectrum — the numbers differ, so a check comparing numbers passes it. Mixed line rates on one fibre make this common.

What is flex-grid?

Allocating spectrum in 12.5 GHz slices rather than fixed channels, so a carrier takes as many slices as it needs. It makes overlap the only question, and it introduces fragmentation — free spectrum scattered in pieces too small for a wide carrier.