An IP camera is a small computer with a lens, and its network link looks trivial until the estate grows: a camera on a gate 600 metres away, a cluster of poles across a logistics yard, a perimeter that measures itself in kilometres. Copper Ethernet's 100-metre channel was never sized for this geography. Fiber is how camera estates are actually built at site scale — not as an exotic upgrade, but as the standard spine that copper clusters hang from. This article lays out the three architectures, because 'camera over fiber' names all of them and they are not interchangeable.
Why fiber, stated plainly
Four properties drive the choice. Distance: a fiber span is measured in kilometres, so the control room can be where it belongs rather than where the cable plant permits. Immunity: outdoor copper runs collect induced surges and ground-potential differences between buildings; glass carries none of it, which matters for electronics bolted to poles. Capacity headroom: the same strand serves today's 4K camera and its next two replacements. And topology: fiber spines let cabinets be placed by operational logic (power, access, security) rather than by the 100-metre rule.
Architecture one: fiber to the cabinet, copper to the camera
The workhorse pattern. A fiber spine runs from the control room to a field cabinet; in the cabinet, a switch or media converters land the fiber, and cameras within copper reach (poles and walls around the cabinet) connect by ordinary Ethernet with PoE. Media converters serve the single-link cabinets — including industrial PoE media converter models that power the local camera while converting its uplink. This pattern concentrates electronics in maintainable points and keeps the fiber plant simple. Most estates should start here.
Architecture two: fiber to the camera position
Some positions justify fiber all the way: the very long spur, the electrically hostile route, the camera whose replacement generations will only get hungrier. A media converter pair (or an SFP in a suitable device at each end) carries that single camera's link over its own strand. The discipline that makes this affordable is standardisation — one converter model, one module type, one spare on the shelf for every such position in the estate.
Architecture three: combination extenders for mixed positions
Real positions are mixed: a pole carrying IP cameras and an SDI PTZ, a cabinet serving PoE devices and a legacy feed. The catalogue's most instructive product for this reality is a combination extender documented with eight Gigabit PoE channels plus six forward and two backward 3G-SDI channels over one fiber unit — the mixed position served by one device on one fiber. When a position inventory shows IP and SDI sharing poles, this architecture replaces a cabinet full of pairwise converters.
The power question fiber never answers
Fiber carries no power, full stop. Every architecture above contains a power design wearing a disguise: PoE from the cabinet device (which needs local mains or DC plant), injectors fed from site power at the cluster, or extender chains that pass PoE along copper hops. Designs that draw the fiber beautifully and the power not at all fail in the field. The professional habit: every cluster on the drawing gets two annotations — its fiber route and its power source — before any model is chosen.
- Spine: fiber type, strand count (include spares), route and cabinet landings.
- Cluster: camera count, per-camera rate (today and next generation), PoE class per camera.
- Power: source per cluster, worst-case draw (heaters!), battery/UPS expectations.
- Mixed media: SDI or analog HD positions listed beside the IP count, not discovered later.
Built this way, the estate's fiber becomes boring in the best sense: a plant that accepts camera refreshes as a device swap rather than a rewiring project. The cameras will change every few years. The glass, specified once with headroom, will not care.
Strand discipline: the cheapest future-proofing there is
One practical doctrine ties the three architectures together: always pull more strands than the design needs. Fiber cable's cost is dominated by installation, not glass; the difference between a 4-strand and a 12-strand pull is trivial against the cost of returning to the route. Spare strands are the estate's future: the camera position that becomes a gate controller, the cabinet that gains a second service, the link that needs a diverse path after a dig-in. Every architecture in this article has been saved, somewhere, by a spare strand somebody almost value-engineered away. Specify the spares on the drawing, terminate and test them at installation, and record them — an unrecorded spare is a rumour, not an asset.
Bandwidth arithmetic for people who hate surprises
Camera estates fail quietly on arithmetic more often than on hardware. A modern 4K camera's stream is not its resolution divided by optimism: at working quality settings, budget a range per camera, multiply by the cluster count, add the analytics and recording overheads, and compare against the spine link's honest rate with headroom for the next camera generation. The fiber itself is never the constraint — glass outruns every camera roadmap — but the devices landing it (converter rates, switch ports, the recorder's ingest) are chosen once and lived with. The design habit that prevents the year-two crunch: size the landing electronics one generation above the cameras being installed today, and let the glass be glass. Cameras are consumables in this architecture; the plant is the asset, and arithmetic is how the asset stays ahead of the consumables.
Recording is a network load too
Camera designs habitually size the live-viewing path and forget the recording path, which never sleeps. Every camera's stream travels to its recorder continuously, so the spine carries the sum of all streams at all hours — plus live viewing, plus analytics, plus the export traffic when an incident is investigated (which is, of course, when the estate can least afford contention). The fiber plant shrugs at all of it; the landing electronics must be sized for it. When the cluster inventory is drawn, add the recording destination to the drawing as a first-class node: which cabinet it lands in, which link carries the aggregate, and what that aggregate becomes after the next camera refresh. Estates designed this way treat recording growth as arithmetic. Estates designed around live viewing alone meet it as weather.
One selection habit follows from all three architectures: standardise the landing devices ruthlessly. One converter family for the single-link positions, one injector discipline for the cabinets, one module type per duty. Estates are maintained at night, in weather, by whoever is on call; the kindest thing a design can do for that person is to make every position's hardware identical to the last position's. Variety is a cost that recurs forever, charged against a benefit that was usually a few percent on a purchase order. The drawing should name the standard devices beside the architectures — that is where standardisation stops being a preference and becomes a fact of the estate.