Fibre to the Home: How PON Networks Deliver Gigabit Internet

The fibre running to a modern house almost never runs to a switch port of its own. It runs to a splitter in a cabinet, where one laser at the exchange is shared between thirty or sixty of your neighbours. That architecture — the passive optical network — is why fibre got cheap enough to deploy at scale, and it is also why the number on your bill is written “up to”.

The shape of a PON

A passive optical network has three parts. At the operator end sits the optical line terminal (OLT), a chassis in an exchange or street cabinet with one port per PON. At your end sits the optical network terminal (ONT), the small box the technician screws to your wall — sometimes integrated into the router, sometimes a separate unit with an Ethernet handoff. Between them is glass and, critically, a passive splitter.

“Passive” is the whole trick. The splitter is an unpowered optical component that divides incoming light among several output fibres and combines light coming the other way. It needs no electricity, no cooling and no cabinet with a battery, which is what makes the outside plant cheap to build and cheap to keep running. The price is optical: every split divides the power. A 1:32 splitter costs you roughly 15 dB of loss before the signal has travelled anywhere, which is why split ratio, reach and optical budget are one problem rather than three.

Cisco’s GPON documentation describes deployments using 16, 32 or 64-way splits, a maximum fibre distance of 20 km and a maximum logical reach of 60 km. Those numbers move with the generation and the optics class, but the trade-off never changes: more homes per laser, less light each.

Downstream is a broadcast. Upstream is a queue.

The two directions of a PON work on completely different principles, and almost every confusing thing about fibre performance comes from that asymmetry.

Downstream, the OLT transmits one continuous stream that reaches every ONT on the tree. Each frame carries an identifier — in GPON, a GEM port ID — and each ONT discards everything not addressed to it. Your ONT is physically receiving your neighbours’ traffic and throwing it away. This is exactly why the ITU-T G.984 series mandates encryption downstream; GPON uses AES-128 so that discarding is not the only thing standing between your neighbour and your packets.

Upstream is the hard direction. Every ONT on the tree shares one fibre back to the OLT, so if two of them transmitted at once their bursts would collide somewhere in the splitter and both would be lost. PON solves this with time-division multiple access: the OLT grants each ONT specific microsecond-scale windows in which it may fire its laser, and the ONT is silent the rest of the time.

Ranging, and why distance matters

Handing out time slots only works if the OLT knows how long light takes to arrive from each ONT — and the ONTs on a tree are at different distances. So every ONT goes through ranging when it registers: the OLT measures round-trip delay and assigns an equalisation delay, an artificial offset that makes every ONT behave as though it sits at the same distance. Without it, a customer 1 km from the cabinet and one 18 km away would trample each other’s slots.

Slot allocation itself is dynamic. The OLT runs a dynamic bandwidth allocation algorithm that reads how much each ONT has queued and shifts grants accordingly, typically several thousand times a second. This is the real mechanism behind “shared” fibre: not a fixed division, but a scheduler.

The generations, and what actually changed

PON generations are distinguished mostly by line rate and wavelength plan. The wavelengths matter more than they look, because a new generation that uses new wavelengths can run over the same glass and the same splitters as the old one — the operator changes the OLT card and the ONT, not the fibre in the ground.

System	Standard	Line rate down / up	Wavelength down / up
GPON	ITU-T G.984 series (first edition 2003)	2.48832 / 1.24416 Gbit/s	1480–1500 nm / 1290–1330 nm
XG-PON1	ITU-T G.987 (2010)	9.95328 / 2.48832 Gbit/s	1577 nm / 1270 nm
XGS-PON	ITU-T G.9807.1 (2016; current edition 2023)	9.95328 Gbit/s symmetric	1577 nm / 1270 nm
25GS-PON	Industry multi-source agreement, not an ITU-T Recommendation	25 Gbit/s symmetric	Vendor-specified
50G-PON	ITU-T G.9804.1/.2/.3 (2019–2021)	50 Gbit/s per channel	1340–1344 nm / band options

G.9807.1, the XGS-PON Recommendation, was first approved in June 2016, revised in February 2023 and amended again in May 2025. It is the workhorse behind most multi-gigabit residential fibre products sold today, because it is the first widely deployed PON that is symmetric — the same roughly 10 Gbit/s in each direction.

The next step is genuinely interesting engineering. ITU’s own account of the Higher Speed PON work notes that 50G-PON is the first PON system to lean on digital signal processing, using 25 Gbit/s-class optics and recovering the rest in the digital domain — deliberately moving cost from optics to silicon, which gets cheaper on Moore’s-Law timescales. G.9804.1 set the requirements in 2019; G.9804.2 and G.9804.3 followed in September 2021. The 25 Gbit/s systems some operators deploy sit outside the ITU line entirely, defined by an industry multi-source agreement rather than a Recommendation.

Why fibre beats cable upstream

Cable’s problem is not the coaxial cable. It is the spectrum plan. A hybrid fibre-coax plant historically allocated only the low end of the spectrum — a few tens of megahertz — to upstream traffic, because that was the leftover space in a system designed to push television downhill. Widening it means re-engineering amplifiers and taps across an entire neighbourhood.

The gap is visible in the specifications. CableLabs’ own material puts DOCSIS 3.1, published in 2013, at up to around 10 Gbps downstream but only 1–2 Gbps upstream, and positions DOCSIS 4.0 as the fix, raising upstream to as much as 6 Gbps. That is a real engineering answer, but it requires plant upgrades. XGS-PON is symmetric by construction: the same laser rate applies in both directions and the only question is how the scheduler divides it.

For a household, the practical consequence is upload. Video calls, cloud backup, security cameras, working from a home office and anything that pushes files rather than pulls them all live in the direction cable was designed to starve.

What “up to” really means here

On a PON, the ceiling is a property of the tree, not of your line. If you buy a 1 Gbps service on an XGS-PON with a 1:32 split, you are one of thirty-two customers sharing roughly 10 Gbit/s. Statistical multiplexing usually makes that fine — households do not peak simultaneously — but “usually” is doing real work in that sentence, and it is the reason evening performance can differ from 3 a.m. performance on a technology with no shared copper anywhere in it.

Some other things that will not show up in the marketing:

• Your ONT’s handoff can be the bottleneck. A 2.5 Gbps service delivered over a gigabit Ethernet port gives you a gigabit. Check the port speed on both the ONT and the router.
• Symmetry is a product decision, not a physics one. An operator running XGS-PON can still sell you 1 Gbps down and 100 Mbps up. The fibre is symmetric; the rate limiter in the OLT is not.
• Upgrades may not need a truck. Because wavelength plans were designed to coexist, an operator can light XGS-PON over the same splitters already carrying GPON and migrate customers ONT by ONT.
• Latency is where fibre quietly wins. No DSL interleaving and no cable request-grant cycle at scale: idle latency on FTTH is typically a few milliseconds to the first hop.

The practical version

When you compare fibre offers, ask which PON generation the operator runs — GPON or XGS-PON — and what the split ratio is on your street. Ask what the ONT’s Ethernet port speed is, since that number, not the PON rate, caps what you can actually take. Then look hardest at the upload figure: that is where fibre’s structural advantage over cable lives, and it is the number an operator can quietly hold back even on hardware fully capable of symmetry.