Glossary
The optical-transport vocabulary in one place, written for someone who already knows networking. Use the filter to jump to a term. Definitions are concise and operational — enough to read a diagram, an alarm, or a datasheet.
Dark fiber
Fiber you lease or own with no carrier electronics on it — just glass between two points. You provide the optics on both ends, which means you also own the loss budget, the wavelength plan, and the operations. Contrast with a lit service.
Why it matters: "we bought dark fiber" is not a service — the right next question is always "what are we lighting it with, and does the budget close?"
Lit service
A managed connectivity service where the provider supplies and operates the transport electronics; you hand off Ethernet and never touch the optical layer. You trade control and cost-at-scale for simplicity. Contrast with dark fiber, where you own the optics.
Why it matters: the dark-vs-lit choice decides who owns the loss budget, the SLA, and the 2am call.
WDM
Wavelength-division multiplexing — carrying many independent signals on one fiber, each on a different wavelength (color) of light. Each wavelength is its own service and can run a different rate or protocol. It is how one leased pair becomes dozens of circuits.
Why it matters: WDM is the multiplier that makes a single expensive strand economical.
CWDM
Coarse WDM (ITU-T G.694.2) — a small number of widely spaced channels, ~20 nm apart, so lasers can be uncooled and cheap. The wide spacing limits channel count and rules out C-band amplification. Best for short reach and a handful of services.
Typical: up to ~18 channels, 1270–1610 nm; no EDFA. Why it matters: cheapest way to get a few services onto one pair.
DWDM
Dense WDM (ITU-T G.694.1) — many closely spaced channels on the C-band grid at 50 or 100 GHz spacing, enabling dozens-plus of channels and long, amplified reach. The tight spacing demands temperature-stabilized, precisely tuned lasers. The workhorse of high-capacity transport.
Typical: ~40–96 channels in the C-band; EDFA-amplifiable. Why it matters: the default for scale and reach.
Wavelength
A specific color of light (transport lives around 1550 nm, the C-band) used to carry one signal. In WDM, each service rides its own wavelength so they don't interfere. Wavelength (nm) and frequency (THz) are two names for the same channel.
Why it matters: assigning a wavelength is assigning a lane — two services on the same one collide.
Channel
A named slot on the DWDM grid corresponding to a specific wavelength/frequency. Assigning a "channel" means picking which color a service uses and which mux/demux port it lands on. Channels must match ports the passive filters actually pass.
Why it matters: the channel plan register is what stops two services being assigned the same slot.
ITU grid
The standardized set of DWDM frequencies (spacing such as 100 GHz or 50 GHz, anchored at 193.1 THz) that vendors align to so equipment interoperates. It is defined in ITU-T G.694.1 for DWDM and G.694.2 for CWDM. Staying on-grid is what lets a mux from one vendor pass a wave from another.
Why it matters: off-grid lasers drift into a neighbor's channel and break both.
Transponder
A card that converts a client signal (e.g. grey 100GbE) into a specific DWDM wavelength for the line, and back. One client in, one wavelength out — no aggregation. It is the simplest way to put one service on one color.
Why it matters: if you don't need to groom lower-rate clients together, a transponder is the clean choice.
Muxponder
Like a transponder but it multiplexes several lower-rate client signals into one higher-rate wavelength (e.g. up to ten 10G clients into one 100G line). The aggregation happens in the OTN layer as ODU tributaries. One line wave, many client ports.
Why it matters: it fills a wavelength efficiently instead of burning a whole color on a single 10G.
Mux
Multiplexer — the passive device that combines multiple wavelengths onto one fiber at the transmit end. It has no power and adds a fixed insertion loss to every channel. Its counterpart at the far end is the demux.
Typical: ~3–4 dB insertion loss per mux+demux pair. Why it matters: that loss goes straight into every channel's budget.
Demux
Demultiplexer — the passive device that separates the combined wavelengths back into individual signals at the receive end. Physically it is usually the same filter component as a mux, run in reverse. Like the mux, it adds insertion loss and no gain.
Why it matters: mux and demux loss together is often the single biggest lumped element in a metro budget.
OADM
Optical add/drop multiplexer — drops selected wavelengths to a local site and passes the rest straight through, with a fixed channel plan set in hardware. It lets an intermediate site take its own services without terminating every wave. Lower loss than a ROADM but not reconfigurable.
Why it matters: cheap, low-loss add/drop when the channel plan won't change.
ROADM
Reconfigurable OADM — an add/drop node where you choose in software which wavelengths drop locally versus express through, without re-patching. It uses wavelength-selective switches (WSS) internally. More flexible than a fixed OADM, at higher cost and insertion loss.
Why it matters: it turns a wavelength move into a config change instead of a truck roll.
Amplifier
A device that boosts the optical signal to overcome span loss, without converting it to electrical (all-optical, all wavelengths at once). It adds power but also adds noise, degrading OSNR a little each time. Enables long spans between regeneration points.
Why it matters: amplification is where you turn when a passive budget won't close.
EDFA
Erbium-doped fiber amplifier — the common optical amplifier for the C-band; boosts every wavelength in the band simultaneously. It is why DWDM scales over long distances. It amplifies power but cannot fix a degraded signal's shape or timing.
Typical: ~15–25 dB gain in the C-band. Why it matters: the standard reach tool for DWDM.
Raman amplifier
An amplifier that pumps the transmission fiber itself so gain is distributed along the span rather than lumped at a node. This gives a better noise figure, extending reach on long or lossy links, often paired with an EDFA. More complex and higher-powered than an EDFA alone.
Why it matters: squeezes extra reach out of spans where an EDFA can't close the OSNR.
OSC
Optical supervisory channel — a separate management wavelength (outside the traffic band) carrying telemetry and control between line-system nodes. It rides the same fiber but is terminated at every node. Because it traverses the whole span, losing it is a strong, fast fiber-fault signal.
Why it matters: OSC-down usually means a fiber event before any traffic alarm clears.
Client side
The port/optic facing the customer equipment (router/switch/SAN) — typically a grey, short-reach signal at a standard rate. It is where the service enters and exits the transport. Its counterpart is the line side.
Why it matters: a client-side fault looks like a customer outage but is inside your gear.
Line side
The port/optic facing the WDM fiber plant — a specific DWDM wavelength tuned for distance, often coherent. It is the transport side of a transponder/muxponder. Its power levels and pre-FEC are what you read to judge the optical path.
Why it matters: line-side Rx and pre-FEC are your first two troubleshooting readings.
OTN
Optical Transport Network (ITU-T G.709) — a digital wrapper that carries client signals with added FEC, end-to-end performance monitoring, and multiplexing. It wraps a client into an ODU, then an OTU for the fiber. It is what lets sub-rate services share a wavelength and be monitored per-service.
Why it matters: OTN is the framing that makes grooming, FEC, and PM possible.
OTU
Optical Transport Unit — the ODU plus FEC and section overhead, framed for one hop across a wavelength. It is the "on the fiber" container. OTU4 (~111.8 Gbit/s) is the one that carries 100GbE.
Typical: OTU2 ≈ 10.7 G, OTU4 ≈ 111.8 G. Why it matters: the OTU rate is what the line optic actually transmits.
ODU
Optical Data Unit — the container carrying the customer payload with end-to-end path monitoring; the service object you groom, switch, and protect. Lower-rate ODUs (e.g. ODU2 for 10G) multiplex into higher-rate ones (ODU4 for 100G) in tributary slots. It is the "service" in the OTN world.
Why it matters: when you provision or protect a service, you're acting on an ODU.
FEC
Forward error correction — extra redundancy carried with the signal so the receiver can fix bit errors without retransmission. It effectively buys you signal-to-noise, letting a "noisy" span deliver clean traffic. Stronger FEC schemes buy more reach.
Typical: standard G.709 GFEC ≈ +6 dB coding gain; advanced FEC gives more. Why it matters: FEC margin is real reach.
BER
Bit error rate — the fraction of bits received in error, e.g. 1e-12. Lower is better; it is the bottom-line measure of signal quality on a span. In OTN you watch it as pre-FEC and post-FEC separately.
Why it matters: post-FEC BER above zero means the customer is taking errors right now.
Pre-FEC errors
Errors measured before FEC correction — errors the FEC is successfully fixing. A rising pre-FEC rate is the early-warning signal of a degrading span, even while traffic is still perfectly clean. It is your best leading indicator.
Why it matters: trend pre-FEC and you fix the span before customers ever notice.
Post-FEC errors
Errors remaining after FEC has done its work. Any post-FEC error means FEC ran out of correction margin — the fault is no longer contained and the customer is now seeing errors. It is a hard problem, not a warning.
Why it matters: post-FEC > 0 is an outage-grade event; pre-FEC is a heads-up.
dBm
Absolute optical power referenced to 1 mW on a log scale, so 0 dBm = 1 mW. It reads negative for typical receive levels (e.g. -18 dBm) and positive for strong transmit levels. Your Rx and Tx power readings are always in dBm.
Why it matters: "is the power OK?" is answered in dBm against the optic's window.
dB
A ratio (relative), not an absolute level. Loss and gain are in dB — every +3 dB is roughly ×2 power and +10 dB is exactly ×10. The difference between two dBm values is a dB value.
Why it matters: keeping dB (a ratio) separate from dBm (a level) is the number-one unit mistake to avoid.
Insertion loss
The power lost by inserting a component (connector, splice, mux, patch) into the path, measured in dB. The sum of every element's insertion loss is the core of a loss budget. It is what you add up to decide whether a span closes.
Typical: connector ~0.3 dB, fusion splice ~0.1 dB, mux+demux ~3.5 dB. Why it matters: the budget is just insertion losses summed.
Return loss
How much light is reflected back toward the source at a connector or junction, in dB — higher is better (less reflection). Reflections destabilize lasers and add noise. APC connectors give much higher return loss than UPC.
Typical: APC ≥ 60 dB, UPC ~50 dB. Why it matters: poor return loss = laser noise and instability.
Reflectance
The reflection at a single event (e.g. one connector), expressed as a negative dB value — more negative means less reflective and better. It is the per-event view an OTDR reports, where return loss is the aggregate. A dirty or damaged connector shows up as a high reflectance spike.
Why it matters: a reflectance spike on an OTDR pinpoints the bad connector.
OTDR
Optical time-domain reflectometer — a tester that fires pulses down a fiber and maps loss and reflection events by distance. It locates cuts, splices, bends, and bad connectors to the meter. Requires launch (and ideally receive) cables to see the near and far ends.
Why it matters: it turns "the span lost 4 dB" into "there's a bad splice at 12.3 km."
Power meter
A tester that reads absolute optical power (dBm) at a point on a fiber. Paired with a matched light source it measures end-to-end insertion loss (an OLTS test). It tells you total loss but not where the loss is.
Why it matters: the fastest go/no-go on whether a span meets its budget.
Light source
A calibrated transmitter at a known wavelength used with a power meter to perform an insertion-loss (dB) measurement across a fiber or patch. Together they form an Optical Loss Test Set (OLTS). Test at the wavelength the service will use, since attenuation is wavelength-dependent.
Why it matters: a 1310 nm test doesn't prove a 1550 nm service.
APC
Angled physical contact connector — the ferrule endface is polished at an 8° angle (usually green) so back-reflection is steered into the cladding, not back down the core. This gives very high return loss. Never mate APC to UPC — the angle mismatch wrecks both the loss and the reflection.
Typical: 8° polish, return loss ≥ 60 dB. Why it matters: APC↔UPC mismating is a classic self-inflicted fault.
UPC
Ultra physical contact connector — a flat/domed polish (often blue) with higher reflectance than APC. It must be mated only to UPC. Common on many client/grey optics where reflection is less critical.
Typical: return loss ~50 dB. Why it matters: mixing UPC into an APC line side degrades return loss badly.
LC
A small form-factor fiber connector with a latch, the dominant connector on transceivers and high-density panels. Its small footprint is why modern optics use it. Comes in UPC and APC polishes — check which the port expects.
Why it matters: the connector you'll handle most; polish (UPC/APC) still matters.
SC
A larger push-pull square fiber connector, common on older equipment and field patch panels. Bigger than LC, so it's fading from high-density gear. Still widely seen in the outside plant.
Why it matters: you'll meet SC at the panel even when the optic is LC — carry the right patch cords.
Patch panel
The passive frame where fibers terminate and cross-connect via patch cords. Every hop through a panel adds a pair of connector insertion losses to the budget. It is also where polarity and labeling errors creep in.
Why it matters: count every panel hop in the budget — they add up fast.
Demarc
Demarcation point — the physical boundary between your responsibility and the customer's (or between two providers). Faults on their side of the demarc are theirs to fix. It is where acceptance testing stops and finger-pointing starts.
Why it matters: knowing the demarc tells you whose problem an outage is.
Fiber pair
Two strands, one for each direction of transmission (Tx on one, Rx on the other). It is the common physical design for a bidirectional service. Crossed strands pass a power test but never bring the link up.
Why it matters: "power present, link down" is very often a swapped pair.
Single-fiber bidirectional
Both directions carried on one strand using a different wavelength each way. It saves a strand — valuable when fiber is scarce or expensive — but must be patched and designed differently from a pair. The two directions use distinct optics.
Why it matters: halves strand count but you can't just swap in a normal pair optic.
Protection switching
Automatically moving a service from a failed working path to a pre-provisioned protect path to ride through a fault. It targets a switch time fast enough that upper layers barely notice. Real protection also requires the two paths to be physically diverse.
Typical: ~50 ms switch target. Why it matters: non-diverse "protection" dies to one backhoe.
Service ID
The unique identifier tying together a customer service across ports, wavelength, and spans. It is the first thing you quote on any ticket or support case. In the NMS it is the object that carries the baseline and PM.
Why it matters: without a service ID, correlating an alarm to a customer is guesswork.
Performance monitoring
Rolling, binned counters (typically 15-minute and 24-hour bins) for power, FEC/BER, and errored seconds. It is the trend data that catches slow degradation before it becomes an outage. Compared against the turn-up baseline, it turns "something feels off" into a number.
Why it matters: PM history is how you catch a dying laser weeks early.
OSNR
Optical signal-to-noise ratio — the ratio of signal power to accumulated amplifier noise, in dB, and the real currency of long-haul reach. Every EDFA adds a little noise, so OSNR only ever drops along a chain of amplifiers. When OSNR falls too low, pre-FEC errors climb even though the power looks fine.
Why it matters: "power's fine but pre-FEC is rising" is usually an OSNR problem, not a loss problem.
Chromatic dispersion (CD)
Different wavelengths travel at slightly different speeds in fiber, so a pulse spreads out over distance and smears into its neighbors. It accumulates with length (ps/nm) and is the classic long-reach impairment. Coherent optics compensate it electronically; older direct-detect systems needed dispersion-compensating modules.
Typical: ~17 ps/nm/km for G.652 at 1550 nm. Why it matters: it limits reach on non-coherent links.
PMD
Polarization mode dispersion — the two polarization states of a signal travel at slightly different speeds, spreading the pulse in a way that varies randomly with temperature and stress. Unlike CD it is not fixed, which makes it harder to compensate. It mainly bites older fiber at high rates and long reach.
Why it matters: a high-PMD leased strand can cap the rate you can run over it.
Coherent optics
Modern line optics that detect both amplitude and phase (and both polarizations), using DSP to compensate chromatic dispersion and PMD electronically. This is why a 100G/400G coherent wave crosses long spans with no external dispersion compensation. It also enables higher-order modulation for more bits per wavelength.
Why it matters: coherent is why 40 km 100G is "just plug it in" and long-haul got simpler.
ZR / ZR+
Standardized 400G coherent pluggable optics (400ZR from the OIF; ZR+ is the higher-power, longer-reach extension) in a QSFP-DD/OSFP form factor. They let a router or switch source a DWDM wave directly, blurring the line between the transport and IP layers (IP-over-DWDM). Reach and features vary by variant, so verify against the datasheet.
Why it matters: ZR/ZR+ can move the transponder function into the router — a real design fork.
WSS
Wavelength-selective switch — the engine inside a ROADM that routes any wavelength from an input to any output port under software control. It is what makes add/drop-vs-express reconfigurable without re-patching. It adds insertion loss but buys flexibility.
Why it matters: the WSS is the part that turns a wavelength move into a config change.
Tributary slot
A time-slot within a higher-rate OTN container (e.g. within an ODU4) that carries one lower-rate tributary such as an ODU2. Mapping a 10G client means assigning it specific tributary slots. When the slots are full, the container is full — no more grooming without a new wavelength.
Why it matters: "is there room for another 10G?" is really "are there free tributary slots?"