What PoE does before it puts 48 V on the cable
Plenty of things that are not PoE devices plug into an RJ45 port. That is why a PSE does not just put 48 V out: it checks who is there with a few volts, asks how much power they need, and only then switches on.
The other end might not be a PoE device
A PoE camera, an ordinary PC, an old hub — all of them fit the same RJ45 port. Put 48 V on the pins the moment something is plugged in and you destroy the magnetics or the PHY of anything that was not expecting it.
So PoE puts two checks in front of the power. Detection asks whether this is a PoE device at all; classification asks how much power it wants. Both are done at voltages far below 48 V.
Where USB PD starts at 5 V and negotiates by messages, PoE starts by measuring the electrical character of the other end at even lower voltages. It is measurement, not protocol.
Detection: two points, and read the slope
The PSE applies a voltage somewhere in 2.7–10.1 V and measures the current, limited to 5 mA. A PD is built to look like 25 kΩ across that range; the standard accepts 23.75–26.25 kΩ.
The complication is the diodes in the path. A PD takes power either way round, so there is a bridge rectifier, which means two diodes in series are always in the way. The current does not pass through the origin.
Measure one point and divide, and the offset shows up as extra resistance. At 4 V the current is (4 − 1.4) / 25k = 104 µA, so V / I = 38.5 kΩ. A perfectly good 25 kΩ PD gets rejected.
So the PSE measures two points and takes the difference. At 4 V and 8 V the currents are 104 µA and 264 µA, and ΔV / ΔI = 4 V / 160 µA = 25.0 kΩ. The diode offset cancels in the subtraction.
That is what "stepping the voltage up gradually" actually is here. The steps are not only caution — you cannot get a slope from one point.
Classification: one step up, then read the current
Once detection passes, the PSE raises the voltage to 15.5–20.5 V. The PD draws a current that encodes how much power it wants, and the PSE reads that current to decide how much to reserve.
| Class | PD current | Reserved at the PSE | Available at the PD |
|---|---|---|---|
| 0 | 0–4 mA | 15.4 W | 12.95 W |
| 1 | 9–12 mA | 4.0 W | 3.84 W |
| 2 | 17–20 mA | 7.0 W | 6.49 W |
| 3 | 26–30 mA | 15.4 W | 12.95 W |
| 4 | 36–44 mA | 30 W | 25.5 W |
Class 4 (Type 2 / PoE+) and above need either two-event classification — the same step repeated — or an LLDP negotiation after power is up. 802.3bt (Type 3 / 4) adds classes 5 to 8 and uses all four pairs, up to 71.3 W at the PD.
Classification is a reservation, not a limit. Whether the PD actually uses that power is another matter, but the PSE can no longer offer it to another port. A 48-port switch that runs out of power budget is usually the sum of these reservations.
From switch-on to switch-off
With the class settled, the PSE finally brings up the full voltage. Not instantly: it limits current so the inrush into the PD's input capacitance stays within spec, and the PD controller has its own inrush limiter.
Once power is up, the PSE keeps checking that the PD is still there. That is the MPS (Maintain Power Signature), and the PD has to keep drawing at least 10 mA.
This is where real hardware trips up. If the PD sleeps and drops below 10 mA, the PSE decides it has been unplugged and removes power. "It powers off when it goes into low-power mode" is almost always this. The fix is to keep 10 mA flowing through a dummy load while asleep, or to draw the current in pulses.
The PD minimum voltage comes from the cable drop
The standard gives different minimum voltages at each end: Type 1 is 44 V at the PSE and 37 V at the PD; Type 2 is 50 V and 42.5 V. Those are not numbers to memorise — they fall out of the cable drop.
Type 1: 350 mA maximum, channel resistance capped at 20 Ω. 44 V − 0.35 A × 20 Ω = 37 V.
Type 2: 600 mA maximum, and with four pairs the channel cap is 12.5 Ω. 50 V − 0.6 A × 12.5 Ω = 42.5 V.
Both land exactly on the specified figure. The PD minimum is simply what is left at its terminals with the worst-case cable carrying the maximum current.
So what governs a PoE design is the usual voltage drop. 24 AWG Cat5e is about 9.4 Ω per 100 m per conductor; over two pairs there and back, a 100 m run comes to roughly 9–10 Ω.
Things that catch you out
- Do not hang capacitance on the 25 kΩ signature. The PSE takes its two points quickly. Too much capacitance on that node and the current has not settled, so the resistance reads wrong.
- Classification is a reservation, not a ceiling. Some PSEs will trip on overcurrent if you draw beyond your class. Declare the class that matches what you actually consume.
- Sleep versus MPS. Low-power design and PoE fight each other directly. Either budget the 10 mA (about 0.5 W) away, or draw the current intermittently.
- Cable length costs power. 100 m of Cat5e is around 10 Ω, which is 6–7 V at maximum current. A lower input voltage means more current for the same power, which increases the drop again.
- "Passive PoE" exists and is not this. Some equipment simply puts 48 V on the pins with no detection or classification. A standard PD will work, but plugging in a non-PoE device will destroy it.
Frequently asked questions
Why two measurement points? Would one not do?
My PoE device powers off when it sleeps.
Why is the PD minimum 37 V in one case and 42.5 V in another?
Is it safer to declare a higher class?
Standards and references
- IEEE 802.3 (Clause 33 / 145) — Detection, classification and MPS requirements, and the PSE / PD voltage and current ranges
- IEEE 802.3af / 802.3at / 802.3bt — Power levels and class definitions for Types 1 to 4
- TIA-568 / ISO-IEC 11801 — Cat5e / Cat6 conductor resistance, used for the drop estimates