When a PoE camera does not power up, the fault is almost never inside the camera. It is a copper fault, a budget or class problem at the switch, a detection failure caused by the wrong device or a damaged front end, or a run beyond the channel limit. Work the list from the switch outward, measure instead of guessing, and nine checks will find the fault in under twenty minutes.
Installers start at the camera because that is the end that is dark, then spend an hour swapping hardware that was never broken. Start at the PSE port instead, and every step after it becomes elimination.
What a PoE port does before it sends power
An 802.3af or 802.3at PSE does not apply 48 V and hope. It runs detection first: a limited voltage between 2.8 V and 10 V is presented across the pairs, and the PSE looks for a signature resistance of about 25 kΩ (accepted window 23.75–26.25 kΩ). Only when that handshake succeeds does the port classify the load with short current pulses, and only then does it raise the port to operating voltage.
That sequence gives the single most useful diagnostic fact in PoE work: a port that never lights has usually failed detection, not power delivery. Detection fails for a short list of reasons — an open pair, a shorted pair, a device with no signature at all, or a PD front end destroyed by an earlier surge. If the LED blinks or turns amber and then drops, detection passed and the problem is downstream: classification, budget, or the load itself.
Memorise the operating windows, because a meter reading either confirms the port or condemns it. For 802.3af and 802.3at the PSE must output 44–57 V DC and the PD must tolerate 37–57 V at its input; for 802.3bt those tighten to 50–57 V and 42.5–57 V. A port reading 48–54 V open-circuit is healthy. A port reading 0 V with a known-good PD attached is a detection or budget failure, not a dead supply.
Steps 1 to 3: prove the port and the copper
Step 1 — read the port, then swap it. Check link and PoE LEDs on the switch. Move the cable to another PoE port on the same switch. If the camera comes up, the original port is out of budget, administratively disabled, or has a failed PSE controller. Note the class and wattage the switch reports for that port; one line of telemetry eliminates half the remaining steps.
Step 2 — confirm the switch can afford the port. A 24-port 802.3at switch with a 370 W supply cannot deliver 30 W on all 24 ports; 370 ÷ 30 leaves twelve ports at full class. Many units refuse to power the thirteenth PD rather than brown out the rack. Sum the allocated class wattages against the switch budget and check for a per-port limit set below the camera's draw. The arithmetic is set out in our guide to calculating a PoE power budget.
Step 3 — measure the pair, do not eyeball the cable. Unplug both ends and read DC loop resistance with the far end shorted. A healthy 100 m Cat5e pair reads about 12.5 Ω; 30 Ω or more means a bad crimp, a corroded IDC or a damaged conductor, and that will drop enough voltage to keep a PD below its undervoltage lockout. On outdoor runs, check the RJ45 for green corrosion at the pins.
Steps 4 to 6: prove the negotiation
Step 4 — check that the class matches the camera. A camera rated 802.3at on a port that only implements 802.3af will either refuse to start or boot and reset the moment the IR illuminator engages. The numbers are the tell: 802.3af guarantees 15.4 W at the PSE and 12.95 W at the PD at up to 350 mA; 802.3at gives 30 W and 25.5 W at up to 600 mA; 802.3bt Type 3 raises that to 60 W and 51 W, and Type 4 to 90 W and 71.3 W. A dome with heater and IR that draws 14 W cold and 26 W at night needs Type 2 or better.
Step 5 — watch for the boot loop. Power for a few seconds, then drop, is a PD restart rather than a dead port. Three causes dominate: insufficient allocated power as above, a splitter or extender in the path that cannot pass the class through, and inrush current tripping the port's overcurrent protection. Isolate path devices one at a time.
Step 6 — verify the device type the port expects. Passive 24 V gear on an active 802.3at port never starts, because it presents no 25 kΩ signature. The reverse mistake is destructive: a passive injector pushing 48–57 V into a 24 V device will kill it. Confirm voltage and pinout first.
Steps 7 to 9: prove the run and the device
Step 7 — measure voltage at the far end, under load. Open-circuit voltage at the camera proves almost nothing; the number that matters is the voltage while the PD draws current. With Cat5e's 12.5 Ω per-pair loop resistance, 600 mA of 802.3at current loses 7.5 V in the cable and dissipates 4.5 W as heat over 100 m. From a 53 V PSE that still lands at 45.5 V at the PD — comfortable. At 250 m the same arithmetic fails completely, which is why the channel limit exists and why PoE extenders re-power a segment instead of stretching one.
Step 8 — check distance and hop count honestly. The 100 m figure is a channel limit covering both patch cords, the patch panel and the horizontal run, not just the ceiling cable. A 90 m run plus two 5 m cords is already at the limit. Count extender hops: each re-powers the signal but adds conversion loss and budget, and passive units add voltage drop with no negotiation at all. Runs that must cross several hundred metres belong on fibre with a media converter at the far end.
Step 9 — test the camera on a bench supply or a short cord. A 2 m factory patch cord into a verified port removes every field variable at once. If the camera powers up there, the fault is in the run you bypassed. If it stays dark, the PD front end is gone — usually a failed signature resistor or a blown TVS diode after a lightning event weeks earlier. That is a replacement, not a repair.
Symptoms, causes and the values that confirm them
These come from the standard, not a vendor sheet.
| Symptom | Most likely cause | Confirming measurement | Fix |
|---|---|---|---|
| PoE LED never lights | Detection failure: open pair, shorted pair, no signature | Pair loop resistance, far end shorted: ~12.5 Ω per 100 m | Re-terminate or replace the run; replace a PD with no signature |
| LED blinks, then goes dark | Insufficient budget or a per-port limit | Allocated class wattage vs switch budget | Move the PD, raise the limit, or add PSE capacity |
| Powers up, resets at night | Class mismatch when IR and heater engage | Reported class vs nameplate; current at full load | Move to an 802.3at or bt port |
| Works on a short cord, not in place | Voltage drop over distance or a bad joint | Voltage at the PD under load; above 37 V for af/at | Shorten the run, re-terminate, or insert an extender |
| No power and no link | Cable fault or wrong port; not a PoE issue | Cable tester; link LED on a non-PoE port | Fix the physical layer, then retest PoE |
| Parameter | 802.3af | 802.3at | 802.3bt Type 3 / 4 |
|---|---|---|---|
| PSE power per port | 15.4 W | 30 W | 60 W / 90 W |
| PD power guaranteed | 12.95 W | 25.5 W | 51 W / 71.3 W |
| Maximum current | 350 mA | 600 mA | 600 mA / 960 mA over four pairs |
| Cable loss at 100 m, Cat5e | 4.4 V at 350 mA | 7.5 V at 600 mA | ~6 V per pair-set at 480 mA |
The cable loss row is where marginal installations die. Cat5e's 12.5 Ω per-pair loop resistance is fine for 802.3af, workable for 802.3at, and starts to bite at 802.3bt, where the 42.5 V PD floor leaves less headroom. For 60 W and 90 W devices, plan Cat6 or Cat6a for margin and heat.
Three findings justify abandoning diagnosis. A PD that fails on a short patch cord into a verified port. A pair above roughly 25 Ω that cannot be re-terminated because the damage is inside a wall. A switch whose budget is fully allocated at the port count you need — no configuration change creates wattage, and the honest fix is a PoE switch with a larger supply or a second switch closer to the load.
Frequently asked questions
Why does my camera work on a short cable but not on the installed run?
Because the run adds resistance, and resistance adds voltage drop in proportion to current. A 100 m Cat5e run at 600 mA loses 7.5 V, so a marginal PSE at 48 V delivers 40.5 V — near the 37 V floor, where tolerant cameras boot and strict ones do not. Measure under load.
Can a camera draw too little power to be detected?
No. Detection depends on the 25 kΩ signature resistor, not on current draw. A 3 W camera is detected as reliably as a 25 W one; if detection fails, the signature is missing or the pairs are damaged.
Does an injector fix a switch that has run out of budget?
Yes, and it is often the cheapest fix for one camera. A midspan injector adds its own PSE, so the camera stops competing for the switch allocation. The cost is another mains outlet and another failure point in the rack.
Why does the camera reboot only at night?
The IR illuminator and heater draw the most power after dark, so a supply adequate at noon is inadequate at 22:00. Compare allocated class against peak draw and move the camera to a higher-class port.
How many extenders can one run carry?
Each hop re-powers the segment but consumes budget and adds conversion loss, and passive units add drop with no negotiation. Two active hops on a well-provisioned 802.3at port is the working ceiling for most cameras; beyond that, fibre is cheaper than the failures. The alternatives are compared in our article on extending PoE beyond 100 metres.
Next step
If steps 1 to 9 point at the switch, the camera count or the distance rather than the camera itself, the fix is a specification problem, not a repair. Send us the port count, the camera models with class ratings, the longest run in metres and the cable category, and our engineers will size the PSE budget and topology. Browse PoE switches and extenders for the current range, or contact our team with your site plan and we will return a bill of materials you can install against.