Ekinops Dark Fiber Learning Path
Engineering

ROADM & Flexgrid — routing wavelengths without touching a patch cord

So far the wavelengths on a fiber have gone end to end. A real network has junctions — sites where some wavelengths must turn toward a different city while others pass straight through. A ROADM is the device that does that routing in the optical domain, remotely, without anyone unplugging a fiber. This page explains what is inside one, how it is described (CDC, node degree), and how the modern flexible grid lets a fiber carry today's 400G and 800G superchannels.

What a ROADM is, in one sentence

A ROADM (Reconfigurable Optical Add/Drop Multiplexer) lets you choose, by software, which wavelengths drop at this site, which add here, and which pass straight through to another direction — for any wavelength, to any direction, without physically re-cabling. "Reconfigurable" is the whole point: the routing is a command, not a truck roll.

Inside a ROADM — the WSS

The engine of every ROADM is a WSS — Wavelength-Selective Switch. A WSS takes a fiber carrying many wavelengths, separates them internally, and can send each individual wavelength out a different port — and it can attenuate each one independently on the way. That last ability is as important as the switching: it is how a ROADM levels the channels so they all arrive at the next amplifier at the right power.

A WSS ROUTES EACH WAVELENGTH INDEPENDENTLY in (many λ) ─►┌───────────────┐─► out port A (λ1, λ4 …) λ1 λ2 λ3 │ WSS │─► out port B (λ2 …) λ4 λ5 … │ splits by λ, │─► out port C (λ3, λ5 …) │ steers + sets│ │ per-λ level │ └───────────────┘ Any input wavelength can be sent to any output port, at a per-wavelength attenuation the software chooses. Switching AND power leveling happen in the same element.

Express vs add/drop — the two jobs at a node

At any ROADM site, each wavelength does exactly one of two things:

Express (pass-through)

The wavelength arrives from one line direction and leaves toward another line direction without being converted to electrical signal. It stays optical the whole way through the node. Express traffic is cheap — no transponder is consumed at this site — and it is how a wavelength crosses many intermediate cities on a single optical path.

Add / drop

A drop peels a wavelength off the line and sends it to a local transponder to be received here. An add takes a locally generated wavelength and injects it onto a line direction. Add/drop is where traffic enters and leaves the optical layer — it consumes a transponder and an add/drop port.

Node degree — how many directions a node serves

The node degree is simply the number of line directions a ROADM node connects to. A node at the end of a chain is degree-1; a node in the middle of a line is degree-2; a node where three or four fiber routes meet is degree-3 or degree-4. Higher degree means more express paths to manage and, usually, one WSS per direction.

A ROADM node expressing wavelengths through and dropping and adding one wavelength A central ROADM node built on a wavelength-selective switch connects two line directions, West and East, making it a degree-two node. Four wavelengths, lambda 1 to lambda 4, arrive from the West. Lambda 1, lambda 2 and lambda 4 pass straight through as express traffic and leave toward the East. Lambda 3 is dropped down to a local drop port and receive transponder, and a locally generated lambda 3 prime is added from a transmit transponder back onto the line. Add and drop happen at the node's local add/drop bank. ROADM node WSS · per-λ routing + leveling express: λ1 λ2 λ4 → drop λ3 · add λ3′ West (in): λ1 λ2 λ3 λ4 East (out): λ1 λ2 λ4 drop λ3 add λ3′ Drop port Rx transponder Add port Tx transponder Degree-2 node — two line directions (West, East) + local add/drop bank
A ROADM node: λ1 λ2 λ4 express straight through West→East, while λ3 is dropped to a local receive transponder and λ3′ is added from a local transmit transponder. Text version below.

CDC — the three freedoms of a modern add/drop bank

Early ROADMs could switch express traffic freely but had rigid add/drop banks: a given port only worked on one fixed wavelength and one fixed direction, and two services wanting the same wavelength could block each other. CDC removes those three restrictions. The letters stand for three independent freedoms:

C — Colorless

Any wavelength on any add/drop port. A colorless port is not hard-wired to one "color" (wavelength). You can tune the transponder on that port to whatever wavelength the plan needs, by software, without moving fibers. Without colorless, every add/drop port is pre-assigned a wavelength and re-planning means re-patching.

D — Directionless

Any add/drop port to any line direction. A directionless port can send its wavelength out the North, East, or South line by software choice. Without directionless, a transponder is tied to one direction and rerouting to a different fiber route means physically re-cabling it.

C — Contentionless

No internal same-wavelength blocking. A contentionless bank can add/drop two uses of the same wavelength toward two different directions at the same time without one blocking the other internally. Without it, the second service on a duplicate wavelength is blocked even though the spectrum on the two fibers is free.

The add/drop structure that makes CDC work — an MCS

Colorless-directionless add/drop is often built with an MCS — Multicast Switch. An MCS is a broadcast-and-select fabric: it fans every line direction out to every add/drop port, so any transponder can reach any direction and tune to any wavelength. Adding the "contentionless" property means arranging the MCS (and its tunable filters/transponders) so duplicate wavelengths toward different directions never share the one blocking path.

A CDC ADD/DROP BANK (multicast-switch style, conceptual) Line directions: [N-WSS] [E-WSS] [S-WSS] │ │ │ └────┬────┴────┬────┘ │ MCS │ broadcast every ┌──────┴─────────┴──────┐ direction to │ fans all directions │ every port │ out to every port │ └─┬───┬───┬───┬───┬───┬──┘ add/drop ports ► T T T T T T (tunable transponders) │ Colorless : each T tunes to ANY wavelength (not fixed color) Directionless: each T reaches ANY line direction via the MCS Contentionless: two T's on the SAME λ can go to DIFFERENT directions without internal blocking

A contention example — why the last "C" matters

Suppose two customers both need wavelength λ20, one routed North and one routed South. The spectrum is physically free — λ20 is unused on both the North and South fibers. On a contentionless bank both services light up fine. On a bank that is colorless and directionless but not contentionless, the two λ20 uses collide on a shared internal path and the second one is blocked even though the fiber has room. Contentionless is what turns "the spectrum is free but I can't provision it" into "just provision it."

CONTENTION: same wavelength, two directions Request: λ20 → North AND λ20 → South Fiber: North fiber λ20 = FREE, South fiber λ20 = FREE NOT contentionless: Contentionless: λ20→North ✔ (took the λ20→North ✔ only λ20 path) λ20→South ✔ (independent path) λ20→South X BLOCKED (internal clash, not a fiber clash) Contentionless removes the internal duplicate-λ blocking so the only limit is real spectrum on the fiber. (Illustrative example.)

Per-channel power equalization

Because the WSS can attenuate each wavelength independently, a ROADM does not just route — it equalizes. Channels arrive at a node having each been through different numbers of amplifiers and different fiber, so they arrive at different powers. The node sets each channel's attenuation so they all leave at the target per-channel power, giving the downstream amplifiers a flat, predictable input. This ties directly to the nonlinear and OSNR budgets: a channel that runs hot risks nonlinearity; one that runs cold loses OSNR. Leveling keeps every channel in its window.

Operational read

When a single channel through a multi-node path shows a power or OSNR problem that the end sites cannot explain, an intermediate ROADM's equalization is a place to look — a mis-leveled channel gets progressively worse span by span. Per-channel leveling is a routine, expected function of the node, not an exception.

Fixed grid vs flexible grid

For decades WDM used a fixed grid: wavelengths sat on evenly spaced slots, commonly 50 GHz or 100 GHz apart, each slot the same width. That works while every channel fits in a fixed-width slot. Modern high-rate carriers — 400G, 800G, and superchannels that gang several carriers together — do not fit neatly in one fixed slot; some need more than 50 GHz, and forcing them onto 100 GHz slots wastes spectrum.

The flexible grid, standardized in ITU-T G.694.1, replaces fixed slots with a fine granularity of 12.5 GHz slices. Instead of assigning a channel to a fixed slot, you allocate it a media channel — a contiguous group of slices just wide enough for that carrier, plus its guard. A narrow carrier takes a few slices; a wide superchannel takes many. The fiber's spectrum is carved to fit the traffic, not the other way around.

FIXED GRID vs FLEXIBLE GRID (G.694.1, 12.5 GHz slices) FIXED 50 GHz — every channel gets the same width slot: │ 50 │ 50 │ 50 │ 50 │ 50 │ 50 │ ← a wide 400G/800G carrier does │ λ1 │ λ2 │ λ3 │ λ4 │ λ5 │ λ6 │ not fit; a narrow one wastes space FLEXIBLE GRID — spectrum sliced at 12.5 GHz; media channels sized to fit: ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ← 12.5 GHz slices (fine ruler) [ 100G ][ superchannel ][ 200G ] (4 sl) (many slices, e.g. (few several ganged slices) carriers = 1 media channel) A "media channel" = a contiguous block of 12.5 GHz slices sized to the carrier + guard. Wide carriers get wide blocks; narrow ones stay tight. (Slice widths per G.694.1; carrier widths illustrative.)
Why flexgrid exists

Higher bit rates use wider carriers, and packing them onto a coarse fixed grid either does not fit or wastes spectrum. Flexgrid lets each carrier claim exactly the spectrum it needs at 12.5 GHz resolution, which is what makes efficient 400G/800G and superchannel transport possible on the same fiber that once carried fixed 50/100 GHz channels. A flexgrid-capable WSS can define these arbitrary-width media channels; a fixed-grid WSS cannot.

Regen placement in a meshed ROADM network

Express pass-through keeps a wavelength optical across many nodes — but only as far as its optical reach allows. When a path is longer than any single wavelength can go error-free, you must regenerate: drop the signal to a transponder at an intermediate ROADM, clean it up, and re-transmit it as a fresh wavelength for the next optical segment. Choosing where to place that regen in a meshed network is a design decision: pick a node that is roughly within reach of both the source and destination, has a spare transponder pair, and ideally sits where several long paths could share the regen site.

REGEN PLACEMENT WHEN REACH RUNS OUT A ──express──express──express──✗ (reach exceeded before D)──► D Insert a regen at C, within reach of both ends: A ──express──express──►[ C: DROP → clean → ADD ]──express──► D │ fresh wavelength, OSNR reset, dispersion reset (a full 3R regeneration — see Protection & OTN) Rule of thumb: regenerate at a node that both segments can reach with margin, that has spare transponders, and that many long paths can share. (Reach limits are per the planning tool — illustrative here.)

Software control — a note

Everything on this page is set by software: which wavelength drops where, each channel's attenuation, each media channel's width. Larger networks coordinate that centrally with an SDN controller, and control-plane signaling such as GMPLS can compute and set up optical paths across many nodes automatically. Conceptually, the ROADMs provide the reconfigurable hardware and the controller decides how to use it. The exact controller, and which features a given Ekinops ROADM supports, must be checked against Ekinops documentation and the platform BOM.

Field checklist — working with ROADM nodes

1. What is the single element inside a ROADM that both routes each wavelength and levels its power, and what does its name stand for?

The WSS — Wavelength-Selective Switch. It separates the wavelengths, steers each one to a chosen output port, and applies a per-wavelength attenuation, so routing and per-channel equalization happen in the same element.

2. Spell out C-D-C and give one thing each freedom removes.

Colorless — any wavelength on any add/drop port (removes ports being hard-wired to one wavelength). Directionless — any port to any line direction (removes transponders being tied to one direction). Contentionless — no internal same-wavelength blocking (removes duplicate-λ collisions when the fiber spectrum is actually free).

3. Two customers both need λ20 — one North, one South — and both fibers have λ20 free. On a colorless+directionless-but-not-contentionless bank, what happens, and why?

The second λ20 service is blocked. Even though the spectrum is physically free on both fibers, the two uses of the same wavelength collide on a shared internal path. Only a contentionless bank gives them independent paths, so the only real limit becomes spectrum on the fiber.

4. Why can a 400G/800G superchannel be awkward on a fixed 50/100 GHz grid, and how does flexgrid solve it?

Wide, high-rate carriers may not fit one fixed slot, and forcing them onto coarse slots wastes spectrum. Flexgrid (ITU-T G.694.1) slices the spectrum at 12.5 GHz and assigns each carrier a "media channel" — a contiguous block of slices sized exactly to that carrier plus guard — so wide carriers get wide blocks and narrow ones stay tight.

5. A wavelength's path is longer than its optical reach across a meshed ROADM network. What do you do, and what makes a good site for it?

Place a regenerator: drop the signal at an intermediate node, clean it (3R), and re-transmit a fresh wavelength for the next segment. A good regen site is within reach of both segments with margin, has spare transponders, and ideally sits where several long paths can share it.