Optical Communication Products · Shenzhen, China
Tel: +86 0755 23766314  ·  marketing@fotfiberoptic.com
Request a Quote

Fiber Optic & Networking Glossary

70 terms, defined properly. Not one-line dictionary stubs — each entry explains the mechanism, says where you actually meet it, and links to the products and calculators that go with it. Written and maintained by the FOT engineering team.

Modules, Packaging & Form Factors

How the optics are built and housed — from pluggable cages to co-packaged engines.

Co-Packaged OpticsCPOThe optical engine is mounted on the same substrate as the switch ASIC, so the high-speed electrical path shrinks from centimetres to millimetres.Distributed Feedback LaserDFBA single-wavelength semiconductor laser with a grating built into the cavity — the workhorse source for single-mode optics.Electro-absorption Modulated LaserEMLA DFB laser with an external modulator integrated on the same chip — narrower spectrum, longer reach.eXtra-dense Pluggable OpticsXPOA liquid-cooled multi-terabit pluggable class that reaches co-packaged densities without giving up field replaceability.Linear Pluggable OpticsLPOA pluggable transceiver with the DSP removed — the host SerDes drives the optics directly through a linear analogue path.Linear Receive OpticsLROA half-way house between DSP modules and LPO: the DSP stays on the transmit side, the receive path goes linear.Multi-Source AgreementMSAAn industry agreement that fixes a module's mechanical, electrical and management interfaces so parts from different vendors interchange.Near-Packaged OpticsNPOOptical engines placed close to the switch ASIC on the same board, but not on the ASIC package itself.Optical Module DSPDSPThe chip inside a transceiver that equalises, retimes and regenerates the signal in both directions.OSFPAn eight-lane pluggable form factor for 400G, 800G and 1.6T, designed from the start with a larger thermal envelope than QSFP-DD.OSFP-XDA sixteen-lane extension of OSFP, aimed at 1.6T and 3.2T without waiting for the next SerDes generation.QSFP-DDQuad Small Form-factor Pluggable Double Density — eight lanes in a body that still accepts legacy QSFP modules.QSFP112A four-lane QSFP running at 100G per lane — 400G in the familiar QSFP footprint.Silicon PhotonicsSiPhBuilding optical circuits — waveguides, modulators, splitters, detectors — in silicon using semiconductor fabrication.Vertical-Cavity Surface-Emitting LaserVCSELA cheap, low-power 850 nm laser that emits from the wafer surface — the source behind every multimode SR module.

Fibre, Connectors & Cabling

The passive plant: glass grades, ferrules, polarity and the loss they contribute.

APC vs UPC PolishAPC / UPCWhether a ferrule end face is flat or angled at eight degrees — and why the two must never be mated.Base-8 vs Base-12 vs Base-16 MPObase-8Whether your MPO plant is built in groups of 8, 12 or 16 fibres — and how much of it modern optics can actually use.Effective Modal BandwidthEMBThe figure of merit for multimode fibre — how far a laser-launched pulse can travel before modal dispersion closes the eye.End-Face InspectionIEC 61300-3-35Looking at a ferrule under magnification and grading it against a pass/fail standard before you mate it.Fiber PigtailpigtailA short length of fibre with a factory-terminated connector on one end and bare fibre on the other, for splicing into the plant.G.657 Bend-Insensitive FiberG.657Single-mode fibre engineered to tolerate tight bends without leaking light — what patch cords and FTTH drops are made of.Insertion LossILHow much optical power a component removes from the link — measured in dB, and the currency every link budget is spent in.LC UnibootunibootA duplex LC patch cord that uses one round cable instead of two zipcord halves — roughly half the volume in the cable manager.MPO / MTP ConnectorMPO / MTPA multi-fibre connector carrying 8, 12, 16, 24 or more fibres in one ferrule — the interface parallel optics runs on.MPO PolarityType A/B/CThe wiring scheme that ensures every transmit fibre reaches the far end's receive — and the single most common cause of dead parallel links.OM3 Multimode FiberOM3Laser-optimised 50/125 µm multimode fibre specified at 2000 MHz·km effective modal bandwidth — the first grade built for VCSELs.OM4 Multimode FiberOM450/125 µm multimode at 4700 MHz·km — the default multimode grade for data centres.OM5 Wideband Multimode FiberOM5Multimode fibre specified across 850–953 nm so several wavelengths can share one fibre pair — not simply 'better OM4'.OS2 / G.652.D Single-Mode FiberOS2The standard single-mode fibre — nine-micron core, zero water peak, usable from 1260 to 1625 nm.Return LossRLHow much light a connection reflects back toward the source — quoted as a positive number you want to be large.Very Small Form Factor ConnectorsVSFFThe MDC, SN and CS connector families — duplex connectors small enough to put three or four in the space an LC duplex occupies.

WDM, Coherent & Transport

Putting many wavelengths down one fibre, and pushing them a long way.

400ZR and ZR+400ZRCoherent pluggables that put DWDM transport straight into a router or switch port, removing the separate transponder.Chromatic DispersionCDDifferent wavelengths travelling at different speeds in glass, smearing pulses in time as they propagate.Coarse Wavelength Division MultiplexingCWDMEighteen channels on 20 nm spacing from 1271 to 1611 nm — wide enough that the lasers need no cooling.Coherent OpticscoherentDetection that recovers amplitude, phase and polarisation — allowing far more bits per symbol and DSP compensation of fibre impairments.DCO and ACODCO / ACOWhether the coherent DSP lives inside the pluggable module (DCO) or on the host line card (ACO).Dense Wavelength Division MultiplexingDWDMDozens of channels on a 100, 50, 25 or 12.5 GHz grid anchored at 193.1 THz — the capacity workhorse of long-haul.Erbium-Doped Fiber AmplifierEDFAA length of erbium-doped fibre that amplifies every C-band wavelength at once, optically, without conversion to electrical.LAN-WDMFour tightly spaced O-band wavelengths around 1295–1310 nm, used by LR4 and ER4 transceivers.Medium Wavelength Division MultiplexingMWDMA twelve-channel plan that doubles the low CWDM channels by shifting each by ±3.5 nm — created for 5G fronthaul.Optical Add-Drop MultiplexerOADMA passive filter that removes selected wavelengths from a fibre and inserts others, passing the rest through untouched.Optical Signal-to-Noise RatioOSNRSignal power against amplifier noise in a reference bandwidth — the number that decides whether an amplified link closes.Polarization Mode DispersionPMDThe two polarisation states of a signal travelling at slightly different speeds — a statistical, time-varying impairment.Short Wavelength Division MultiplexingSWDMFour VCSEL wavelengths between 850 and 940 nm on one multimode fibre pair — quadrupling capacity without more fibre.

Data Centre & AI Networking

Fabrics, congestion control and the topologies that GPU clusters forced into existence.

Explicit Congestion NotificationECNMarking packets to signal congestion instead of dropping them — the early-warning half of lossless Ethernet.GPUDirect RDMAGPUDirectLetting a network adapter read and write GPU memory directly, without staging through host memory.InfiniBandIBA purpose-built cluster interconnect that is lossless by design, with credit-based flow control instead of drop-and-retry.Leaf-Spine TopologyClosA two-tier non-blocking fabric where every leaf connects to every spine — the standard data-centre architecture.Lossless EthernetlosslessAn Ethernet fabric configured so that congestion causes delay rather than dropped packets — a prerequisite for RoCE.NVLinkA proprietary GPU-to-GPU interconnect with far higher bandwidth and lower latency than any network — the scale-up fabric inside a node.Optical Circuit SwitchingOCSSwitching light physically — steering a whole fibre path with mirrors rather than forwarding packets electronically.Priority Flow ControlPFCA link-level pause mechanism that stops traffic on one priority class instead of dropping it — and the source of most RoCE pain.Rail-Optimized Topologyrail-optimisedA GPU cluster wiring scheme in which every GPU of the same index across all servers connects to the same dedicated switch.RDMA over Converged EthernetRoCERemote direct memory access carried over ordinary Ethernet — one machine writes into another's memory without involving either CPU.Scale-Up, Scale-Out and Scale-Acrossscale-upThe three tiers of AI infrastructure: inside the node, across the cluster, and between data centres.Ultra Ethernet ConsortiumUECAn industry effort to redesign Ethernet's transport for AI and HPC — keeping the Ethernet ecosystem, replacing the parts RDMA struggles with.

Questions about your project?

Send the specification and an engineer answers it — no obligation, and no sales sequence. If the answer is that you do not need what we sell, that is the answer you get.

Usually answered within one working day.