Ekinops Dark Fiber Learning Path
Engineering

Capacity & Design — Filling the Fiber and Growing It

You leased one fiber pair. It can carry a single service today or dozens of services for a decade. This page is about how you fill that glass on purpose — choosing a channel grid, mapping services to wavelengths, leaving room to grow, and proving the whole design actually closes before anyone lights it.

The one question this page answers

"I have a fiber pair between two sites. How do I put services on it now, and how do I keep adding capacity later without ripping it apart?" The answer is a repeatable design method: pick a grid, lay out a channel plan with room to spare, size the amplification for the channels you'll grow into, and run the design-closure checklist that ties every other page on this site together.

Mental model

Think of the fiber's usable spectrum as a parking lot. The grid is how you paint the lines: wide stalls (100 GHz) waste space but are easy; narrow stalls (50 GHz) fit more cars; flexible stalls (flex-grid) let you size each space to the vehicle. A channel plan is the map of which service parks where. Guard bands are the empty stalls you leave between cars so doors don't bang. And you never fill every stall on day one — you leave room for the cars that arrive next year.

Step 1 — Choose a grid

The C-band (around 1550 nm, the low-loss window from the testing page) is divided into channels by an ITU standard. Every wavelength you provision must land on a defined channel — you can't just tune anywhere. The grid you choose sets how many services the fiber can hold and how tightly they pack.

GridStandardSpacingBest for
CWDM (coarse)ITU-T G.694.220 nmA handful of services, short reach, low cost. Wide spacing means no cooling and cheap optics — but few channels and generally no amplification.
DWDM 100 GHzITU-T G.694.1100 GHz (~0.8 nm)The easy DWDM default. Roughly ~48 channels across the C-band. Forgiving of small laser drift; simple filters.
DWDM 50 GHzITU-T G.694.150 GHz (~0.4 nm)Doubles the channel count (~96 in the C-band) for high-density fills. Needs tighter optics and filters.
Flex-gridITU-T G.694.1variable (12.5 GHz slices)Sizes each channel's spectrum to the signal. A wide 400G coherent carrier can claim more spectrum; a narrow signal claims less. The modern choice for mixed-rate, coherent networks.
Key idea — the grid is a commitment

The grid is anchored at 193.1 THz in the C-band (the ITU-T G.694.1 reference frequency), and every channel is defined relative to that anchor. Your mux/demux filters are built for a specific grid — a 100 GHz filter cannot pass a 50 GHz plan. Choosing the grid up front, and buying filters to match, is a decision you live with for the life of the system. Pick the grid for where you're going, not just today's two services.

Step 2 — Map services to channels

A channel plan is a table (and a picture) that assigns every service a specific ITU channel. It is the single source of truth that stops two services landing on the same wavelength — the collision that makes both fail. Write it down in a register; never keep it in your head.

A channel-plan picture

Below is an illustrative 100 GHz plan on one fiber pair. Read it left-to-right as frequency across the C-band. Filled slots are lit services; empty slots are headroom you deliberately left; the shaded edges are guard band.

C-BAND on ONE fiber pair — 100 GHz grid (illustrative, not a real channel map) low freq ◄──────────────────────────────────────────────────────► high freq ~191.x THz 193.1 THz (anchor) ~196.x THz ch: 17 19 21 23 25 27 29 31 33 35 37 39 41 ... ┌────┬────┬────┬────┬────┬────┬────┬────┬────┬────┬────┬────┬────┐ │////│ A │ ·· │ B │ ·· │ C │ ·· │ D │ ·· │ ·· │ ·· │ ·· │////│ └────┴────┴────┴────┴────┴────┴────┴────┴────┴────┴────┴────┴────┘ ^^^^ 100G free 100G free 10G free 100G free free free ^^^^ guard svc svc svc svc guard band (grow into the free channels later) band A = 100G to DC-East B = 100G to DC-West C = 10G customer D = 100G protect "free" = provisioned in the register as HEADROOM — filter port exists, no light yet
ServiceRateITU channel (illustrative)Mux portStatus
A — DC-East100Gch 19P19Lit
B — DC-West100Gch 23P23Lit
C — Customer10Gch 27P27Lit
D — Protect100Gch 31P31Lit
(reserved)ch 33–41P33–P41Headroom

Channel numbers and ports here are teaching placeholders. The real channel-to-port map comes from the exact mux/demux you deploy — verify against the Ekinops docs and the component BOM before you commit a plan.

Common mistake — the un-recorded channel

Provisioning a wavelength and never writing it in the channel-plan register. Six months later someone "picks a free channel," tunes onto a channel that is already lit, and takes down a live customer. The register — not memory, not a sticky note — is what makes the plan safe. Every add updates it; every design review checks it.

Step 3 — Guard bands and expansion headroom

Guard bands

A guard band is spectrum you leave empty between or beside signals so they don't interfere. Lasers drift a little with temperature and age; filters aren't perfectly sharp; wider coherent carriers spill slightly into their neighbors. The guard band absorbs all of that. On a fixed grid the spacing itself is the guard; on flex-grid you explicitly allocate guard slices around each carrier.

Expansion headroom

Headroom is spectrum (and shelf slots, and amplifier gain) you deliberately leave unused so growth is an add, not a rebuild. A fiber filled to the last channel on day one has nowhere to go — the next service means a second fiber pair or a forklift upgrade. Leaving free channels, spare mux ports, and amplifier margin is what turns "we're full" into "provision the next one."

Key idea — design for the second and third service

The cheapest capacity is the capacity you planned for. A mux with spare ports, an amplifier sized for the eventual channel count, and a channel plan with reserved slots let you grow by adding a card and a patch instead of returning to site with new passive gear. Headroom is not waste — it is the difference between a five-minute provisioning task and a maintenance window.

Step 4 — Plan amplifier gain for the channel count

If your span needs amplification (see the amplification & safety page), the amplifier must be sized for the number of channels you will eventually run, not just today's one or two. An optical amplifier shares its output power across all the channels passing through it.

Mental model — power is shared

An amplifier has a total output power budget, and every channel takes a slice. Add more channels and each one's share drops unless total power rises. That is why you plan gain and per-channel power for the full fill: an amplifier tuned perfectly for two channels can starve each channel's power (and its OSNR) when you grow to twenty. Per-channel power must stay balanced within the amplifier's limits across the whole plan.

  • Size for the target channel count. Decide the eventual fill (e.g. "up to 40 channels") and confirm the amplifier can hold per-channel power and OSNR at that fill, not just at launch.
  • Keep channels balanced. A big power difference between channels causes problems; the goal is a flat, balanced per-channel power across the band, within the amplifier's specified range.
  • Watch OSNR, not just power. Each amplifier adds noise. On amplified spans the worst-case channel often runs out of OSNR before it runs out of power — the design must close on OSNR for that worst channel (see link engineering).
  • Verify against real specs. Gain range, output power, noise figure, and supported channel counts come from the amplifier's data sheet and the Ekinops BOM — never from a course approximation.

Step 5 — The upgrade path: three ways to grow

When "the fiber is getting full," you have three levers, roughly in order of how disruptive they are. Good design reaches for them in this order.

1 · Add channels

Light another wavelength in a free slot of the existing plan. If you left headroom and the amplifier was sized for it, this is a card, a patch, and a register update. The least disruptive growth — and the reason you left headroom in the first place.

2 · Raise per-channel rate

Carry more bits on the wavelengths you already have — e.g. move a 100G service to 200G or 400G on the same channel. This grows capacity without consuming new spectrum, though a higher-rate carrier may need more spectrum (flex-grid) and has tighter OSNR/dispersion requirements.

3 · Go coherent / denser modulation

Modern coherent optics pack more bits per hertz (higher spectral efficiency) and tolerate dispersion electronically. Moving to coherent (or a denser modulation format) lifts the ceiling on both channel rate and reach — at the cost of tighter OSNR requirements. This is how one fiber pair reaches multi-terabit totals.

Illustrative capacity math

How much can one fiber pair hold? It is just channels × per-channel rate. The numbers below are illustrative to show the shape of the growth — verify real reach and rate against the optic data sheets and Ekinops docs.

FillChannelsPer-channel rateFiber-pair total
50 GHz grid, 100G each96100G≈ 9.6 Tb/s
50 GHz grid, 400G each96400G≈ 38.4 Tb/s

Illustrative only. The jump from 9.6 Tb/s to 38.4 Tb/s on the same fiber comes entirely from lever #2/#3 — raising the per-channel rate with coherent optics, no new glass. Spectral efficiency (bits carried per hertz of spectrum) is the underlying measure: coherent formats carry more bits/Hz, which is why the same 96 channels hold four times the traffic.

Step 6 — The design-closure checklist

A design is not "done" because the loss budget passes. Closure means every impairment is accounted for at once. This is the checklist that ties the whole site together — each gate links to the page that teaches it. If any single gate fails, the design is not ready; it will bite you at turn-up or on the first fault.

1 · Loss budget closes (with margin)

Total expected loss summed and compared to the system budget, with real engineering margin left over (commonly > 3 dB). Both directions checked. → Testing & Loss Budget

2 · OSNR closes for the worst-case channel

On amplified/long spans, confirm the noisiest channel still has enough optical signal-to-noise ratio at full fill — not just the best channel. → Link Engineering

3 · CD & PMD within tolerance for the rate

Accumulated chromatic dispersion and PMD are inside what the optic tolerates at its bit-rate — higher rates are far less forgiving. → Dispersion

4 · Per-channel power balanced within amp limits

Channels are power-balanced across the band and every channel sits inside the amplifier's specified range at the target fill. → Amplification & Safety

5 · Protection path physically diverse

Any protected service rides a genuinely diverse path — separate ducts, handholes, and building entrances — and the protect path's own budget closes. → Protection & OTN

6 · Baselines captured

Turn-up records local/remote Tx/Rx (dBm), pre-/post-FEC, service ID, wavelength, and slot/port, stored in the NMS and acceptance record so later degradation is diagnosable. → Celestis NMS

Common mistake — closing on loss alone

Declaring victory because "the budget passes." A span can pass the loss budget and still fail on OSNR, on dispersion at a higher rate, or because the protect path shares a duct with the working path. Closure is the AND of all six gates, checked together, at the fill you intend to grow into — not the loss gate alone at today's single channel.

A worked design walk-through

Tie it together on a small example. Two data centers, one leased fiber pair, growing over three years.

Light one 100G wavelength on a 100 GHz grid. Choose a grid you can grow into (100 GHz now, room to move to 50 GHz filters later if needed). Buy a mux/demux with spare ports. Run the six-gate closure: loss budget closes with margin; at 100G over a short span, CD/PMD are comfortable and the span may be passive. Record λ1 in the register. Capture baselines.

Two new services land in reserved slots — no new passive gear, because you left headroom. Re-run closure with all three channels: does per-channel power still balance? If the span now needs an amplifier, is it sized for the eventual fill, not just three channels? Does OSNR still close for the worst channel? Update the register with λ2 and λ3.

Rather than lease a second pair, raise the per-channel rate: move the two 100G services to 400G coherent on the same channels (lever #2/#3). Capacity roughly quadruples on those waves with no new fiber. But 400G is far tighter on OSNR and dispersion — re-run gates 2 and 3 hard, and confirm the amplifier still holds per-channel power. This is exactly why you sized the amplifier and left OSNR margin back in Year 1.

Field checklist — before committing a capacity plan

1. You have a 100 GHz mux/demux installed and someone asks to move to a 96-channel 50 GHz plan next quarter. What has to change?

The passive filters. A 100 GHz filter cannot pass a 50 GHz channel plan — the mux/demux must be replaced (or was chosen as flex/50 GHz-capable up front). This is why the grid is a long-lived commitment you pick for where you're going.

2. An amplifier was tuned perfectly for two channels. You grow to twenty. What can go wrong even though each channel's optic is fine?

The amplifier's total output power is shared across all channels. Twenty channels each get a smaller slice than two did, so per-channel power (and OSNR) can drop below spec. Size the amplifier and per-channel power for the full intended fill, not the initial two.

3. Your loss budget closes with 6 dB of margin. Is the design done?

No. Loss is only gate 1 of 6. You still have to confirm OSNR closes for the worst channel, CD/PMD are within tolerance for the rate, per-channel power is balanced, protection is physically diverse, and baselines are captured. Closure is the AND of all six.

4. The fiber is nearly full and you need more capacity but can't lease a second pair. What's the least-disruptive lever?

Raise the per-channel rate on wavelengths you already run (e.g. 100G → 400G coherent). It multiplies capacity on the same channels with no new glass — but re-check OSNR and dispersion, because higher rates are far less forgiving.

5. Illustratively, what's the fiber-pair total for 96 channels at 400G each, and where did the growth over 96×100G come from?

≈ 38.4 Tb/s (96 × 400G), versus ≈ 9.6 Tb/s for 96 × 100G. The 4× gain came from raising the per-channel rate via coherent optics / higher spectral efficiency — same 96 channels, same fiber, more bits per hertz. (Illustrative figures.)
Take it to the labs

The Design Labs walk this exact method hands-on: a two-site 100G design, then adding services, wavelengths, a third site, protection, and operating it all. Do the labs on paper first, then hold every deliverable against the six-gate closure checklist above.