Most broadband technologies put electronics between you and the provider: a cabinet with power, cooling, and things that fail. , which stands for Gigabit Passive Optical Network, takes the opposite bet. Between the provider's equipment and your home there is nothing but glass - fibers and small optical splitters that need no electricity at all. That single design decision explains almost everything else about how it works.

Origins: the copper ran out of headroom, not the fibre

Fibre to the home was technically possible for decades before it was affordable. The obstacle was never the glass; it was that giving every subscriber a dedicated fibre back to the exchange meant a port, a laser and a strand per customer, which is a cost that scales linearly with take-up in exactly the way a telephone company cannot fund.

Passive optical networking is the answer to that economics problem, and its defining choice is in the middle word. The splitter between the exchange and the homes has no power, no electronics and nothing to fail: it divides incoming light among its outputs and combines what comes back. One expensive port at the exchange is shared by many homes, and the shared plant needs no cabinet, no rectifier and no maintenance visit.

The lineage is standardised and worth knowing, because installed networks span all of it:

  • APON/BPON (ITU-T G.983, late 1990s) — the first generation, -based, modest speeds, and the proof that passive splitting worked in the field.
  • GPON (ITU-T G.984, 2003 onward) — the one most deployed: 2.488 Gbps downstream, 1.244 Gbps upstream, shared across the tree.
  • EPON and 10G-EPON ( 802.3ah and successors) — the -native branch, dominant in parts of Asia, same physics with a different framing.
  • XG-PON and XGS-PON (ITU-T G.987, G.9807.1) — 10 Gbps, with XGS-PON symmetrical. Crucially, these use different wavelengths from GPON, so an operator can run them over the same fibre and the same splitter and migrate subscriber by subscriber.
  • NG-PON2 and 25/50G-PON — multiple wavelengths per fibre and higher rates, aimed as much at mobile backhaul and business services as at homes.

That wavelength-coexistence property is why the passive plant is such a good investment: the electronics at both ends get replaced across generations, and the glass, the splitters and the trenching — the expensive, disruptive part — stay.

The tree of glass

A GPON deployment is a tree. At the root, in the provider's central office or a street cabinet, sits the - the Optical Line Terminal. From one OLT port, a single fiber runs out into the neighborhood. Along the way it passes through one or more optical splitters: passive components that divide the light among branches the way a prism divides a beam. A 1:2 splitter sends half the light each way; cascade the splits and you reach the common ratios of 1:32, 1:64, or 1:128 subscribers sharing one OLT port.

At each leaf of the tree sits an - the Optical Network Terminal, the box on your wall that turns light back into Ethernet. The standards also call it an ONU, an Optical Network Unit; field technicians use the two names interchangeably.

Nothing between the OLT and the ONT is powered. No cabinet electronics to fail in a heat wave, no batteries to replace, nothing for a storm to reboot. The plant is glass, connectors, and splice trays. This is the "passive" in the name, and it is why the architecture won the fiber-to-the-home era: the expensive, failure-prone parts live only at the two ends.

Downstream: everyone hears everything

Because a splitter divides light without judgment, every ONT on the tree receives everything the OLT transmits. Downstream GPON is a broadcast at 2.488 gigabits per second: your neighbor's frames physically arrive at your ONT, and yours at theirs.

The system deals with this the only way it can - cryptographically. Each subscriber's downstream traffic is encrypted with , the Advanced Encryption Standard, under keys negotiated per ONT, so the frames your ONT cannot decrypt are noise to it. The ONT filters by identifier and decrypts only its own. Privacy on a shared medium is not a courtesy here; it is a protocol requirement.

Upstream: light on a schedule

Upstream is the harder problem. If two ONTs transmitted at once, their light would collide at the splitter - the splitter works in both directions, combining as happily as it divides. So upstream GPON, at 1.244 gigabits per second, is run on TDMA, Time Division Multiple Access: the OLT grants each ONT precise time slots, and each ONT fires its laser in bursts only when told.

For the schedule to work, the OLT must know how far away each ONT is - light from a home 1 km out and a home 19 km out takes measurably different time to arrive. During activation the OLT performs ranging: it measures each ONT's round-trip delay and assigns an equalization delay so that every burst lands in its slot regardless of distance. This is why a new ONT takes a moment to come online, and why the standard caps the differential reach of a tree.

The asymmetry - 2.488 down, 1.244 up - was a deliberate bet on traffic patterns of the 2000s. Its successor XGS-PON, where XGS stands for 10 Gigabit Symmetric, pays that debt: 10 gigabits in both directions, usually over the same glass on different wavelengths, which lets a provider run both generations on one tree during migration.

Where the decibels go

Every optical link lives on a power budget: the transmitter launches so many dB of light, the receiver needs so many to hear, and everything in between spends the difference. GPON class B+ optics, the common case, budget about 28 dB end to end.

Splitters are the big spenders. Every 1:2 division costs about 3 dB - half the light, by definition, plus a little loss - so a 1:64 tree spends roughly 18 dB on splitting alone. The fiber itself takes about 0.35 dB per kilometer at GPON's wavelengths, each connector 0.3 dB or so, each fusion splice a few hundredths. Add it up and a 1:64 split at 20 km arrives near the edge of the budget - which is exactly the design point of the standard.

This arithmetic is the technician's daily reality. A dirty connector that eats 2 dB, a drop cable bent past its radius, one bad mechanical splice: any of these can push a marginal leaf over the line, and the failure looks like mystery packet loss rather than darkness. The - the Optical Time-Domain Reflectometer - exists to turn that mystery into a distance: fire a pulse, read the reflections, and the trace shows how many meters out the budget is leaking.

Architecture, and the numbers that shape a design

Split ratio is the central trade. A 1:32 split shares one port's capacity among 32 homes; 1:64 halves the per-home share and halves the cost per subscriber. Splits are also frequently cascaded — a 1:4 in a street cabinet feeding 1:16 units at the poles — which changes where the loss falls but not the total.

The loss budget is the design. Each split costs light: roughly 3.5 dB per doubling, so about 17-18 dB for 1:32 and 20-21 dB for 1:64, before adding fibre attenuation, splices and connectors. The standards define optical classes with a total budget, and a design is simply the arithmetic of staying inside it. This is why a distant subscriber on a 1:64 split fails while a nearby one on the same tree is fine, and why the answer is often a different split ratio rather than a different .

Reach is commonly up to 20 km, and the practical limit is usually the loss budget rather than the distance.

The terminology, since every document assumes it. The OLT (optical line terminal) is the exchange-side equipment; the ONT or ONU (optical network terminal or unit) is the device in the home; the ODN (optical distribution network) is the passive plant between them.

Ranging and equalisation. Because homes sit at different distances, their upstream bursts would collide even with perfect scheduling. The OLT measures each ONT's round-trip delay and hands it an equalisation offset, so every burst arrives in its slot regardless of distance. This is also why a rogue ONT transmitting outside its slot degrades an entire tree, and why finding it is a physical-layer hunt rather than a configuration review.

Security follows from the topology. Downstream frames reach every home on the tree, so encryption is not optional decoration: GPON encrypts downstream traffic per subscriber, with the key established during activation. Anyone reasoning about a shared medium should recognise the argument from wireless — the medium is shared, so confidentiality has to be added rather than assumed.

Interoperability and dependencies

Interoperability is the sore point. The standards specify the optics and the framing, but the management interface for the home device historically was not, so an OLT and an ONT from different manufacturers frequently do not fully interoperate. OMCI, the management channel defined in the standards, is what makes multi-vendor deployment possible, and how completely a vendor implements it is a purchasing question with a long tail — an operator locked to one ONT supplier discovers it at the worst moment, during a supply shortage.

Beyond the optics, the dependencies that actually cause outages are the ones around it: power at the home (a fibre network survives a mains cut only for as long as the ONT's battery, which is why voice-over-fibre changed emergency-calling assumptions), the subscriber-management platform that authorises an ONT onto the tree, address assignment and carrier-grade translation behind it, and the physical records: a passive network has no way to tell you which port on which splitter serves which house, so the documentation is the topology.

The vendor landscape, by category

  • Large telecom equipment vendors — Huawei, Nokia, ZTE, Ericsson. Full OLT platforms at scale, deep integration with their own management stacks, and the geopolitical constraints that now shape which of them an operator may deploy.
  • Access specialists — Calix, Adtran, DZS and similar, strong with regional and community operators, typically with a management platform aimed at smaller engineering teams.
  • Merchant-silicon and white-box — Broadcom silicon underneath much of the market, with open management stacks and disaggregated OLTs appealing to operators who want to escape vendor-tied ONTs.
  • ONT suppliers, frequently distinct from the OLT vendor, which is exactly where the OMCI interoperability question bites.

The buying question that matters is not line rate, since the standards fix that. It is which ONTs the OLT will accept and manage, because that decides whether the operator has one supplier or a market.

What to remember

One fiber, split passively, shared by schedule. Downstream is an encrypted broadcast; upstream is a timed choir where the OLT conducts. The whole design trades electronics in the field for arithmetic in decibels - and when GPON misbehaves, the answer is almost always in that arithmetic.