Detection
The port applies a low test voltage and looks for the 25 kΩ signature resistance every PD presents. No signature, no power, so a printer or a laptop on that port sees data and nothing else.

A camera, an access point or a door station needs a network connection and a supply. PoE carries both over the same Cat5e or Cat6 cable, so the device position needs no socket. This guide covers the standards, the pairs, the cable and the power budget.
PoE feeds a device through the same eight conductors that carry its network traffic.
Power over Ethernet is a family of IEEE standards that put a direct voltage of roughly 48 V on the twisted pairs of an ordinary network cable. The supplying end is the PSE, power sourcing equipment: a PoE switch, or an injector inserted into the link. The fed end is the PD, powered device: a camera, an access point, an IP intercom, a touch panel or a KNX IP interface.
Nothing about the cable changes. The same Cat5e or Cat6 that already runs to the device carries the supply, so that position needs no socket, no adapter and no second low voltage cable, and every device on the switch can ride through a power cut behind one UPS in the rack.
A PoE port carries no voltage until it has proof that a PoE device is on the other end, which is what makes it safe to plug a laptop into it.
The port applies a low test voltage and looks for the 25 kΩ signature resistance every PD presents. No signature, no power, so a printer or a laptop on that port sees data and nothing else.
The device declares how much it will draw, as a class from 0 to 8. The switch reserves that much of its budget instead of guessing, and refuses the port when the budget is already spent.
The port raises the full voltage and keeps watching. Unplug the device and the port falls back to the test voltage within a few hundred milliseconds. Type 2 and above can also renegotiate the power over LLDP once the link is up.
Every standard is backward compatible: a Type 1 device works on a Type 4 port and draws only what it needs. The power is quoted twice because the cable itself consumes part of it.
The original standard, still the right size for small devices.
The volume standard today. Most cameras and access points assume it.
Four pairs under load, which is what lifts the ceiling past 30 W.
The top of the range, mainly for screens and PoE lighting.
| Class | Reserved at the port | Available at the device |
|---|---|---|
| 0, unclassified | 15.4 W | 12.95 W |
| 1 | 4.0 W | 3.84 W |
| 2 | 7.0 W | 6.49 W |
| 3 | 15.4 W | 12.95 W |
| 4 | 30 W | 25.5 W |
| 5 | 45 W | 40 W |
| 6 | 60 W | 51 W |
| 7 | 75 W | 62 W |
| 8 | 90 W | 71.3 W |
A device that does not classify is treated as class 0, so the port reserves the full 15.4 W for it. On a switch with a tight budget that is the difference between eight working ports and four.

A network cable holds four twisted pairs. Which of them carry the supply depends on the standard, and every device has to accept either arrangement.
The supply rides on the data pairs 1/2 and 3/6, superimposed on the signal. On a 10/100 link the other two pairs stay idle.
Used by most PoE switches.
The supply uses pairs 4/5 and 7/8, which on a 10/100 link carry nothing else, so power and data stay physically apart.
Typical of midspan injectors.
All four pairs carry the supply at once. The current per conductor halves and the losses fall, and that is what makes 60 W and 90 W possible.
Required for Type 3 and Type 4.
On a gigabit link all four pairs carry data in any case and the supply is simply superimposed on them. The drawings show the 10/100 case, where the difference between the two alternatives is visible.
PoE is a direct current on the same copper, so what decides the result is the cross section and the heat, not the bandwidth alone.
Enough for Type 1 and Type 2, provided the conductor is solid copper.
The thicker AWG 23 conductor runs cooler and drops less voltage, which is the reason to prefer it over Cat5e for PoE.
Better behaved in a full bundle, and the safe choice where Type 4 runs near the length limit.
Two ways to put power onto the link, and one way to take it back off for a device that has no PoE input.
The supply is built into the port. One device in the rack, one budget to manage, and per port control from the switch interface.
Use whenAnything above two powered devices.
Sits between an ordinary switch and the device and adds the supply from a mains socket. Cheap, and the quick answer when one camera has to join an existing switch.
Use whenOne or two devices, or a switch that cannot be replaced.
The inverse of an injector: it takes the PoE link and hands the device an ordinary network port plus a 12 V or 5 V output.
Use whenA device with no PoE input sits where there is no socket.
A switch has a total budget as well as a per port limit, and it is the total that runs out first. A worked example for a house with cameras and access points:
| Device | Qty | Per device | Total |
|---|---|---|---|
| Outdoor cameras with heater, Type 2 | 6 | 25 W | 150 W |
| Indoor cameras, Type 1 | 8 | 8 W | 64 W |
| Wi-Fi access points, Type 2 | 4 | 20 W | 80 W |
| Video door station, Type 3 | 1 | 45 W | 45 W |
| KNX IP interface and IP intercom, Type 1 | 2 | 6 W | 12 W |
| Connected load | 351 W | ||
351 W of load wants a switch of about 450 to 500 W, not 370 W. The reserve covers the losses in the cable, the camera heaters that only draw in winter, and the device someone adds two years later. Check the per port limit as well: a 500 W switch that gives 30 W per port still cannot feed a 60 W panel.
We size the cabling, the switch and the power budget together with the KNX and lighting side, so the rack is planned once and not three times.