The number that decides whether a passive PoE run works is not printed on either box. It is the minimum input voltage of the device at the far end — the point where its regulator gives up — and almost nobody puts it on the label. A 24 V camera that needs at least 21 V fails at 48 m on perfectly good Cat5e. The same camera, if it tolerates 18 V, runs to 96 m. Same injector, same cable, same wattage. Only the tolerance changed.
I have this argument about twice a month. Someone has a 12 V passive injector, a 40 m run, and a camera that draws 9 W. It worked on the bench. Not on the wall. So they buy a bigger injector, then a thicker cable, then they call the camera defective. Passive PoE has no negotiation, so nothing in the system corrects for the voltage the copper eats. You either do the arithmetic before you pull the cable, or you do it afterward with a meter and a bad mood.
Two Numbers on the Box, One That Actually Matters
A passive injector is a DC brick with an RJ45 on it. It puts a fixed voltage on the cable the moment you plug it in — no detection probe, no classification, no current limit worth the name. The voltage is only correct at the injector's own jack.
Every metre between the injector and the device is a resistor in series with your load. On an 802.3af port the PSE sits at 44–57 V and the PD must work down to 37 V, which leaves roughly 7 V of headroom for the cable. On a passive 12 V run you have whatever tolerance the device's regulator happens to have, and that is frequently 1 V or less.
Run the same job at three voltages and the asymmetry shows up immediately. A 12 W load — small PTZ camera, radio, door controller — on a full 100 m channel at 12.5 Ω loop resistance, the allowance the standards use to cover patch cords and a warm riser. Real 24 AWG comes in under that, so treat these as pessimistic:
| Source voltage | Current at 12 W | Loop drop, 100 m | Voltage at the device | Watts lost in cable |
|---|---|---|---|---|
| 12 V passive | 1.00 A | 12.5 V | Never starts | 12.5 W |
| 24 V passive | 0.50 A | 6.25 V | 17.75 V | 3.13 W |
| 48 V passive | 0.25 A | 3.13 V | 44.87 V | 0.78 W |
| 48 V, 802.3at, 24 W | 0.50 A | 6.25 V | 41.75 V | 3.13 W |
The last row is worth a second look. A 24 W device on 48 V at 100 m lands at 41.75 V, just under the 42.5 V input floor that 802.3at sets for the PD. Not a defective switch. A run on the edge, which passes a bench test at 20 m and fails in the field at 95 m.
Four Times the Loss, Half the Distance
The slogan is "higher voltage, lower current, less loss". The exponent is the part that bites.
- Fix the load power. P = 12 W. It does not change with source voltage; the device wants 12 W either way.
- Get the current. I = P ÷ V. At 12 V that is 1.00 A, at 24 V it is 0.50 A, at 48 V it is 0.25 A.
- Get the loss. P_loss = I² × R. With R = 12.5 Ω that is 12.5 W, 3.13 W and 0.78 W.
- Compare. Halving the voltage doubles the current, and doubling current quadruples the loss. Going from 48 V to 12 V multiplies cable loss by sixteen.
- Add temperature. Copper rises about 0.393 % per °C, so 12.5 Ω is closer to 13.7 Ω in a 45 °C riser. Every number above gets about 10 % worse in summer.
Sixteen times. That is why a 12 V passive run feels like it works and then quietly stops: at 100 m the cable would dissipate 12.5 W to deliver 12 W, the device sees nothing, and the injector sees a near short.
Where each voltage actually gives up
Distance is not a property of the standard. It falls out of the device's floor voltage. Work backwards: allowed drop = source − floor, maximum loop resistance = allowed drop ÷ current, distance = resistance ÷ 0.125 Ω per metre.
- 12 V, 12 W, 10 V floor. Allowed drop 2 V at 1 A, so 2 Ω — about 16 m. A 12 V passive camera is a 16-metre technology at that wattage, whatever the "transmission distance: 100 m" line means. That line describes the data link, not the power.
- 24 V, 12 W, 18 V floor. Allowed drop 6 V at 0.5 A, so 12 Ω — about 96 m. This is the good case, and why 24 V gear survives on real sites.
- 24 V, 12 W, 21 V floor. Allowed drop 3 V, so 6 Ω — about 48 m. Same hardware, half the reach, because the regulator inside is stricter.
- 24 V, 24 W, 18 V floor. Current 1 A, allowed drop 6 V, so 6 Ω — again about 48 m.
One caveat I have watched people miss for years: a camera is not a resistor. Its input is a switching regulator, so as the voltage sags it pulls more current to hold its output, which increases the drop, which increases the current. Near the point where cable resistance approaches the load's equivalent resistance, the run falls off a cliff instead of degrading. It boots at 70 m, boots at 80 m, refuses at 84 m. You crossed a threshold.
Pinout, Polarity, and the Gigabit Trap
Four things have to match: voltage, which pairs carry it, which conductor is positive, and whether the magnetics can pass a gigabit link.
On 10/100 only two pairs carry data — 1/2 and 3/6 — so 4/5 and 7/8 sit idle, and the obvious cheap design pushes DC down the spare pairs. That is Mode B: 4/5 positive, 7/8 negative. Most passive injectors do it because it needs no transformers at all.
On 1000BASE-T all four pairs carry data. No spare pairs left. Power has to be phantom-fed onto the same conductors through centre-tapped magnetics, and that needs real transformers or the link is corrupted.
Put a 10/100 spare-pair injector in a gigabit path and the failure is not what you expect. The camera powers up. Video arrives. Everything looks healthy, except the link trains at 100 Mbps or flaps between 1000 and 100, and the NVR records at half the bitrate you configured. I have seen a site where a "gigabit" camera ran at 100 Mbps for eight months and the integrator blamed the NVR firmware. The injector cost eleven dollars and had no magnetics on 4/5 and 7/8.
Polarity is the other quiet one. Mode B is 4/5 positive by convention, not by law. Devices with a bridge rectifier do not care; devices without one care enormously. Reverse it and the unit either refuses to boot or lets the smoke out of a reverse-protected diode. Measure with a meter and a breakout adapter, not probes on bare RJ45 contacts, where a slip is a short.
Distance hardware inherits the same question, so ask it before you order. Extenders and media converters both pass power through their own magnetics, and not all of them pass it on the pairs you assumed.
A 48 V Passive Injector Is Not an 802.3af Port
The label says 48 V. An 802.3af port says 48 V too. They are not the same product.
An IEEE PSE does three things before sending meaningful power: it applies 2.7–10.1 V and looks for a 25 kΩ signature, it classifies the device to reserve wattage, and it keeps watching so it can cut power when the device disappears or faults. A passive injector does none of that. It connects the brick to the pins. Plug a laptop into a passive 48 V injector and the NIC's centre taps get 48 V. Whether that ends in a dead NIC depends on the PHY's isolation, which you will not find in a datasheet.
The reverse direction is murkier, and this is where I see confident wrong answers. Feed an 802.3af device from a passive 48 V injector and it may well boot — the front end has a bridge rectifier and an undervoltage lockout, and 48 V clears that lockout comfortably. It may also refuse, because the hot-swap controller is waiting for a detection sequence that never comes. Both behaviours ship in real hardware, and the box will not tell you which one you have.
Even when it boots, you have given up the accounting. No class negotiation means the port does not know what it reserved. No LLDP means the device cannot ask for more when its heater kicks in. No per-port current monitoring means a failing device looks fine until it does not. On anything bigger than a handful of ports that accounting is why you bought a managed PoE switch. One radio on a mast can live without it. A 24-port closet cannot.
12 V gear is the one I would retire first: 12 V 1 A bricks with no headroom, a 5.5 mm barrel spliced onto a pigtail, and a 15 m reality sold as a 100 m solution.
The Day I Fried a Bridge With the Wrong White Box
Both injectors were white. Both said 48 V. One was 802.3af, one was passive, and I grabbed the wrong one from the same bin at the end of a long day. The bridge came up, ran forty minutes, then died in a way that smelled like the inside of a power supply. I spent the next morning telling the customer it was firmware before I opened the case and saw the burn mark on the DC input. My mistake. Since then no two injectors in my kit live in the same colour, and every one gets voltage, current and pairs written on it in marker.
The other thing I got wrong for years was trusting the distance claim. Now I run this before quoting any passive job:
- Find the input range on the device label, not just the nominal voltage. If it only says "24 V", assume a 21 V floor until proven otherwise.
- Read the brick's output current. Available power is V × A, then subtract 20 % for cable loss on runs over 40 m. A brick that only just covers the load will brown out on inrush.
- Get the loop resistance. 24 AWG is 9.4 Ω per 100 m per conductor, and with two conductors per polarity in parallel the loop lands near 9.4 Ω per 100 m.
- Compute allowed drop, divide by current, divide by 0.125 Ω/m. That is the real maximum distance — compare it to the run you have, with 10 % margin.
- Confirm pair assignment and polarity, then confirm a gigabit rating if the device needs gigabit.
- Power one device, measure at the far end under load, write the number down. It becomes the baseline next time someone says the camera "just stopped".
- If the arithmetic does not clear, do not buy thicker cable first. Move to 48 V, put a converter at the device, or shorten the run. Cable is the most expensive of those three.
Related, and cheaper to read than to learn: the resistance figures by cable category are in my Cat5e/Cat6/Cat6a comparison. Short version — the category gap is far smaller than the voltage gap. Cat5e to Cat6a buys a few percent of resistance and a lot of stiffness. On a marginal run only one of those is worth the money.
FAQ
Can I run a 24 V passive device from a 48 V PoE switch?
Not directly. An 802.3af or 802.3at port will not detect the device, so it usually withholds power — the safe outcome. If it does energise the port, you have put 48 V into a 24 V input. Use a PoE splitter with a 24 V output at the device end: the clean bridge from an active plant to legacy passive gear.
Will a 24 V passive injector damage a 48 V 802.3af camera?
Probably not damage, but it will not work. The camera's detection circuit sees 24 V instead of the 2.7–10.1 V probe, gets no valid signature, and never closes its hot-swap path. You end up with an unpowered camera and a perfectly healthy link, which is a confusing pair of symptoms.
Why does my passive PoE camera work on the bench and fail on the wall?
Because the bench cable was 2 m and the wall cable is 60 m. At 12 V and 1 A, 60 m of 24 AWG eats 5.6 V before the device sees anything. Measure at the camera end under load, not open circuit, and the answer is usually obvious.
Is 48 V passive the same as 802.3af?
No. Same nominal voltage, entirely different behaviour. 802.3af requires detection, classification and continuous fault monitoring. A passive 48 V injector has none of the three. The matching voltage is a coincidence of design, not a compatibility guarantee.
Does a passive injector hurt the data rate?
Only if it was built for 10/100 and you are running gigabit. A spare-pair injector with no magnetics on 4/5 and 7/8 corrupts 1000BASE-T and the link falls back to 100 Mbps. A properly built gigabit passive injector phantom-feeds through centre taps and never touches the signal. Look for a 10/100/1000 rating instead of assuming.
If you take one thing from this: passive PoE is a resistance problem wearing a voltage costume. The standard did not fail you, the arithmetic did, and the arithmetic is knowable before anyone climbs a ladder. The two numbers worth getting in writing are the device's minimum input voltage and the run's loop resistance. Everything else follows.
If you are working through a mixed plant right now — a 24 V radio beside 48 V cameras, a 12 V controller on a spare pair — I am curious which part broke first, because it is never the part anyone suspected. The cable gets blamed, then the device, then the firmware, while the injector with the wrong pair assignment sits in the rack doing the damage. If you have a run with numbers on it, put them in the comments and I will check the arithmetic against mine. That is how I learned most of what is here.