DWDM 50GHz vs 100GHz Channel Spacing on the ITU Grid

Every DWDM deployment starts with a channel plan, and every channel plan comes from ITU-T G.694.1. Choosing 50 GHz instead of 100 GHz spacing doubles the number of channels you can fit in the C-band — but it also halves the guard band between neighbours, so it demands tighter laser tolerance, better filtering and higher OSNR discipline. Understanding the grid framework is what lets you answer the practical question: what spacing does this link and these optics actually support?

The Fixed Grid: One Frequency, Several Spacings

G.694.1 anchors the grid at 193.1 THz (about 1552 nm) and defines nominal central frequencies by simple arithmetic:

The historical pattern is worth remembering: the sequence grew by sub-dividing the original 100 GHz grid in halves. That is why 50 GHz channels line up with 100 GHz channels on every second slot, and why a 100 GHz channel plan can migrate to 50 GHz by filling the gaps rather than by rebuilding the spectrum.

From Frequency to Wavelength: Reading the Channel Plan

Planners speak in terahertz, installers speak in nanometres, and the two are related by nothing more than λ = c / f. Every frequency on the grid maps to exactly one wavelength, which is why a channel plan should always be printed with both columns — a frequency label survives a change of fibre and optics, a wavelength label does not.

λ(nm) = 299792.458 / f(THz)

f (THz)     λ (nm)      role in a C-band plan
196.1       1528.77     blue edge of the conventional C-band
195.1       1536.61     upper C-band, often used for expansion
194.1       1544.53     100 GHz channel number n = 10
193.1       1552.52     grid anchor, n = 0
192.1       1560.61     100 GHz channel number n = -10
191.3       1567.13     red edge (~1565-1567 nm, keep-out for some amps)
191.4       1566.31     lowest 100 GHz slot inside the conventional band

The conventional C-band runs from roughly 1530 nm to 1565 nm, which is about 191.56 THz to 195.94 THz — a little under 4.4 THz of usable spectrum. That single number is what every channel-spacing decision is really about: how many lambdas can you carve out of 4.4 THz, and what do you have to pay in optics quality to fill it densely.

Capacity per Band: The Real Trade-Off

Count channels in the conventional C-band (roughly 1530–1565 nm, about 4.4 THz of usable spectrum):

• 100 GHz: ~44 channels — the metro workhorse, easiest to filter, cheapest mux/demux, tolerant of wavelength drift.
• 50 GHz: ~88 channels — the long-haul and 100G+ standard; doubles capacity but demands stable, well-controlled transmitters.
• 25 GHz: ~176 channels — historically used to pack many 10G lambdas; largely displaced by coherent 100G/200G at 50 GHz.

Realities behind the numbers: at 50 GHz the adjacent-channel crosstalk risk grows, so transceiver wavelength stability and mux/demux passband shape matter much more; at 100 GHz you gain passband flatness but pay in channel count. Filter cascade penalties accumulate as you traverse more ROADMs, which is why 100 GHz remains common in metro rings with many add/drop nodes.

What Tight Spacing Actually Costs in Optics

Halving the spacing halves the guard band, and the guard band is the budget that absorbs every source of frequency error in the link. For a fixed-grid 100 GHz system the transmitter is typically held to ±2.5 GHz; a 50 GHz system tightens that to about ±1.5 GHz, and a 25 GHz plan tighter still. Those numbers are not administrative — they have to cover laser aging, temperature drift, and the fact that the receiving filter is not a perfect rectangle.

The one that bites in practice is the third and fourth rows. A narrow passband is good for OSNR because it rejects amplified spontaneous emission from the EDFA chain, but if it is too narrow relative to the signal's occupied bandwidth it clips the spectrum, and clipping looks exactly like a dirty signal at the receiver — degraded pre-FEC BER with no obvious cause. Every ROADM adds another filter to the cascade, so a 50 GHz plan with six ROADM hops has a much thinner margin than the same plan over two hops.

Baud Rate, Modulation and Why Tight Grids Force Coherent

Occupied bandwidth scales with baud rate, not with line rate. A 10G NRZ signal occupies roughly 10–15 GHz and fits comfortably in either grid. A 100G coherent signal at ~30–35 GBaud with PM-QPSK occupies 37–45 GHz, which fits a 50 GHz slot with little room to spare. Push to 200G at ~60–70 GBaud, and 50 GHz becomes tight enough that engineers reach for probabilistic constellation shaping, Nyquist filtering, or a wider slot from the flexible grid.

• 10G/25G per lambda: the signal is far narrower than the slot — spacing is limited by filter technology, not by the signal.
• 100G/200G coherent: the signal fills the slot; OSNR and filtering both matter, and C-band 50 GHz is the natural home.
• 400G/800G per lambda: the signal needs more than 50 GHz of spectrum, so fixed-grid plans give way to explicit slot widths.

The Flexible Grid and Super-Channels

Edition 2 of G.694.1 adds a flexible (elastic) grid: instead of a fixed grid, a channel is defined by a slot width, with 12.5 GHz granularity, and the frequency by 193.1 + n × 6.25 GHz. This is how modern systems carry 400G and 800G: rather than forcing a high-baud signal into a fixed 50 GHz slot, they allocate multiples of 12.5 GHz — for example 75 GHz for a 400G channel — and can place several smaller channels in the leftover spectrum.

Fixed-grid mindset        Flexgrid mindset
------------------------  -------------------------------------------
channel = grid slot       channel = slot width m x 12.5 GHz
plan is uniform 50 GHz    plan mixes 37.5 / 50 / 75 / 100 GHz widths
guard band is implicit    guard band is an explicit unallocated n slot
WSS switches 50 GHz       WSS switches 6.25 GHz granularity

Practical consequence: a fixed 50 GHz plan is a special case of flexgrid (m = 4), which is why a flexgrid-capable ROADM can host both legacy and new channels — but only if your channel plan is documented in a form the network planning tool understands.

Two properties of the flexible grid matter more than the arithmetic. First, the slot edges move: two channels are allowed to sit at 6.25 GHz granularity positions rather than on a uniform ladder, so the "guard band" is no longer an artefact of the grid but a number you choose and must document. Second, the network must be able to describe a channel as a range of slots. That description is what the WSS configuration, the ROADM control plane and the monitoring system all agree on; when it is wrong, the failure mode is not a dead channel but an intermittently degraded one, which is far harder to find.

Building a Channel Plan: A Worked Example

Suppose you have a four-span C-band link inside a metro ring, three ROADM nodes, and a mix of 100G and 400G demand. A defensible plan looks like this:

1. Fix the usable band. 1530–1565 nm is roughly 191.56–195.94 THz; the EDFA gain flatness window is narrower, so start from what the amplifiers actually specify (often 191.6–196.0 THz on modern C-band EDFAs).
2. Subtract the red-edge keep-out. Below about 191.4 THz (above 1566 nm) some amplifiers lose gain flatness, so the lowest 100 GHz slots are often left empty.
3. Lay out 100 GHz slots first: 191.4, 191.5, 191.6 … 195.9 THz. That is roughly 46 slots, of which maybe 40 are inside the flat window.
4. Decide what the 400G lambdas need. If they require 75 GHz of slot width, place them as flexgrid slots and leave their guard bands as unallocated spectrum.
5. Fill the remaining spectrum with 50 GHz slots only where the transceivers, WSS and monitored OSNR margin all support it.
6. Write the plan down: for each channel, record frequency, wavelength, slot width, direction, and the port it occupies at every ROADM. A "lost channel" alarm is only useful if it maps to a frequency.

Choosing a Spacing: A Decision Table

Migrating an Existing 100 GHz Plan to 50 GHz

The most common real-world task is not designing a grid from scratch but doubling the capacity of one that already exists. The sequence that works is: audit first, then interleave, then re-equalise.

• Audit every optic. A 100 GHz-only transceiver cannot be moved to an odd 50 GHz slot. Replacing optics is usually the bulk of the project cost, so it comes first — not after the plan is drawn.
• Check the filters end to end. Fixed-grid mux/demux units built for 100 GHz will happily pass a 50 GHz neighbour and destroy it. Each filter in the path must be either replaced or confirmed to have 50 GHz-class passbands.
• Confirm the ROADMs. A WSS with 50 GHz granularity can express 50 GHz channels; a flexgrid WSS adds the finer granularity needed for future super-channels.
• Interleave, do not re-plan. Because 50 GHz channels fall on every second 100 GHz slot plus the gaps between them, an existing plan can be extended by filling the gaps. Existing services keep their frequencies, which means no re-tuning, no hit, no re-qualification.
• Re-equalise after the change. Adding channels changes per-channel power and the amplifier operating point; power equalisation must be redone at every ROADM before the OSNR numbers mean anything.

Validation Checklist Before You Turn Up the Link

1. Confirm every transceiver's supported grid and channel numbers (a 50 GHz-only optic cannot be placed on an odd 100 GHz slot).
2. Check the mux/demux and ROADM passbands support the chosen spacing end to end — a flexgrid node with fixed-grid filters breaks super-channels.
3. Verify laser frequency stability and the resulting OSNR margin; tight spacing leaves less room for drift.
4. Record the channel plan in the monitoring system so a "lost channel" alert maps to a frequency, not just a port.
5. Leave documented gaps (guard bands) for future expansion instead of filling the band with today's demand.
6. Measure pre-FEC BER and OSNR per channel after equalisation, not just at the amplifier output — a plan that looks healthy in the EDFA can be marginal at the receiver.

Common Mistakes

• Treating the grid as continuous spectrum. The slots between nominal frequencies are not free space; a laser 10 GHz off its nominal frequency is off-plan and will eventually collide with a neighbour.
• Assuming a 50 GHz system is a 100 GHz system with more channels. It is a different optical budget: tighter filters, tighter lasers, less OSNR margin.
• Counting channels from the band edges. The usable count comes from the amplifier and filter passband, not from 1530–1565 nm.
• Documenting wavelengths but not frequencies. Maintenance crews replace optics by channel number; frequency is what the grid is defined in.
• Filling the band to 100%. Guard bands and spare slots are what let you add a service next quarter without re-planning the link.

相关阅读：ROADM 功率均衡与 APC、相干光模块：Pre-FEC BER、OSNR 与 DSP 检查、800G ZR/ZR+ 相干可插拔光模块指南 以及 DWDM 系统组成：Mux、EDFA 与 Demux。

原文链接：ITU-T G.694.1 - Spectral grids for WDM applications