Yes, an old coax run will carry PoE, and the converter pairs that do it are honest little devices. What is not honest is the number on the box. Those wattage figures are 100-metre figures. Ask the same pair of boxes for 300 metres and the delivered power does not fall in proportion to length — it falls off a cliff, and not a gentle one.
I watched it happen on a retrofit last spring: a building wired in 2003 for analogue cameras, all RG59, all home-run to one rack, and the plan was to keep the copper and change the ends. Six of eight cameras came up clean. The two on the longest runs, around 280 and 340 m, came up fine at noon and dropped at 18:40 three days running. Nothing in the NVR but "video lost". By the time I got there the installer had already replaced both cameras. The new ones did exactly the same thing.
The retrofit only looks free
The pitch for coax PoE is financial, and a good one. Re-pulling eight runs through a finished ceiling in an occupied building costs an order of magnitude more than eight converter pairs, and you lose a week. I would make the same call.
But there is a hidden term in that comparison: you are not keeping a cable, you are keeping a resistance. The old cable was fine for 1 V video at a few milliamps. It was never asked to pass 300 mA of DC, and its shield was never part of a power circuit.
I work at a PoE equipment factory in Shenzhen, on the test side, so most of what I know about long runs comes from days at a bench, not from sales literature. There are two 305 m drums of RG59 and RG6 under it, and they have settled more arguments than any datasheet.
What is actually inside a passive coax pair
These boxes are not Ethernet, whatever the marketing implies. The pair takes a 100BASE-TX signal, converts it to a proprietary PHY that survives 75 Ω impedance, pushes it down the coax, and converts back at the far end. Power rides the same centre conductor with the shield as the return. A small DC-DC board in the far unit rebuilds 48 V for the camera and re-runs detection so the camera sees a valid PoE port.
Three consequences follow. The link is point-to-point: one coax, one camera, no daisy chain. Where the old DVR had eight cameras on eight cables you need eight ports on a PoE switch, not two, and that surprises people at quoting time. The data rate is not a standard either — it is whatever the PHY pair manages, in practice 100 Mbps to roughly 300 m and 10 Mbps beyond, and your switch will show a 100M link without ever mentioning that the far end is running at 10M.
The third one is the camera's power class. It is not the camera's any more. The converter advertises a class and enforces its own limit. If it presents Class 3 while your PTZ wants 25.5 W, the port will cheerfully power a device that browns out the first time the motors move. Most common mis-configuration I see on these jobs.
Constant power makes the drop a cliff, not a slope
Here is the part that took me two afternoons to accept. A camera is not a resistor. It is a constant-power load: it wants its 8 W whether you give it 50 V or 40 V. So as the cable steals voltage, the camera draws more current to hold its wattage, which steals more voltage. Positive feedback like that has a fold in it.
Put numbers on it. The source sits at 53 V, the loop has resistance R, the far converter is 88 percent efficient. Its input current is I = P / (0.88 × V) and the voltage it sees is V = 53 − I × R. Substitute:
0.88 V² − 46.6 V + P·R = 0
which only has a solution while P·R ≤ 618. Past that point there is no operating voltage at all. The converter does not gently deliver less — it collapses, resets, tries again, and the camera boots every forty seconds. The ceiling, in watts, is:
P_max = 0.88 × 53² / (4R)
Loop resistance for RG59 — 11 Ω per 100 m on the 20 AWG centre, about 4 Ω on the braid, so roughly 15 Ω per 100 m at 20 °C — and the ceiling falls fast: 41 W at 100 m, 13.7 W at 300 m, 6.9 W at 600 m. Those are not my opinions. They are the same three numbers the converter vendors print in their own tables, 21 W, 11 W, 7 W, sitting politely under the ceiling with margin. My model reproduces their published table within a few percent, which is why I trust it.
Now go back to the building. At 300 m the ceiling is 13.7 W at 20 °C. That roof void was 50 °C at six in the evening, and copper resistance climbs 0.393 percent per degree, so the loop was nearer 50 Ω and the ceiling nearer 12.3 W. Add the 2.5 W the IR LEDs draw when the cut filter swings over, and the load crosses the ceiling at exactly the hour the camera switches to night mode. That is the 18:40 reboot. Nothing was broken. The design had no margin and the sun went down.
Sizing a coax run, in the order I actually do it
- Measure the loop, don't trust the label. Disconnect both ends, short centre to shield at the far end, read ohms across the pair at the near end. Subtract lead resistance first by touching the probes together. New RG59 lands near 15 Ω per 100 m; over 20 Ω on the same length means corrosion, a bad F connector, or copper-clad braid.
- Estimate the load honestly. Nameplate watts plus 15 percent for IR and heater, then divide by 0.88 for the converter. A 12 W camera becomes about 15.7 W the coax must actually carry.
- Compute the ceiling with the formula above, using R at your hottest expected cable temperature rather than at 20 °C.
- Keep the load under 60 percent of it. Below 60 percent you have behaviour you can predict on a cold night and a hot afternoon; above 80 percent you have a camera that works until it doesn't, in a way nobody can reproduce on demand.
- Check the data side separately. One 4K stream at 8 Mbps fits inside a 10 Mbps link; two streams plus on-camera analytics will not. If the site needs the second stream, you are capped at 300 m regardless of the power maths.
- Only then decide the cable. RG6 buys about 40 percent more distance than RG59 for the same watts, RG11 more again, a twisted-pair extender another 100 m per hop, and fiber with a media converter ends the argument at any distance you can afford to trench.
The same maths, across media
Everything below assumes 53 V, an 88 percent efficient converter, copper at 20 °C. "Usable load" is what I would design to, not the theoretical ceiling.
| Run | DC loop resistance | Ceiling at 53 V | Usable PoE load | Data rate |
|---|---|---|---|---|
| Cat5e, 100 m | 9.4 Ω | 66 W | 25.5 W (600 mA port limit) | 1 Gbps |
| RG6, 100 m | 10.5 Ω | 59 W | 25.5 W | 100 Mbps |
| RG59, 100 m | 15 Ω | 41 W | 21 W | 100 Mbps |
| RG6, 300 m | 31.5 Ω | 19.6 W | 15 W | 100 Mbps |
| RG59, 300 m | 45 Ω | 13.7 W | 11 W | 100 Mbps |
| RG59, 600 m | 90 Ω | 6.9 W | 7 W | 10 Mbps |
| Single-mode fiber + converter | not applicable | — | set by the endpoint PSU | 1 Gbps, kilometres |
The row that should annoy you is the first one. Coax is worse than Cat5e per metre for power, because one centre conductor does the work that four conductors share in a twisted pair. At 100 m the old coax has already given up 25 W of ceiling to a cable you could buy for less. What it gives you is the 300 m, not the efficiency. That is the whole trade, and it is why the 100-metre limit on twisted pair is a different problem with a different answer.
The shield is a power conductor now
For twenty years that braid carried nothing but a reference potential. Now it carries the full return current of a PoE load, through a chain of BNC shells and F connectors that somebody crimped in a hurry in 2004.
I have measured 6 Ω of extra loop resistance across a corroded F-to-BNC adapter on a rooftop. On a 60 m cable that is 40 percent more, from a part the size of a fingernail, invisible until you put a meter on it. When I find one I stop measuring and start replacing connectors, because the corrosion that raised the resistance will raise it further next summer. The same contact-resistance arithmetic applies to every joint on the run.
The second issue is not about ohms. The pair bonds the shield to chassis at both ends, because that is what BNC shells do. If the two ends sit on different earth potentials — two buildings, a sub-panel with a loose neutral, a generator transfer switch — current flows through the shield. It is small, continuous, and it makes the DC return noisy. I have seen a coax camera drop its link every time the lift next door started, and the cause was the shield tying two earths together, not the motor.
So: bond at one end, let the converter's isolation do its job at the other, and if the run leaves the building, fit a coax protector rated for the DC now going through the centre pin. Ordinary video surge protectors are DC-blocking. I have killed a pair by forgetting that.
What I check before signing one off
Six things. The first four take under an hour on a ladder.
- Loop resistance on every run, written down. Not pass or fail — a number. On an eight-camera retrofit the numbers tell you which two runs will be trouble next summer.
- Delivered voltage at the camera end, under full load, IR on. If it is within 2 V of the converter's minimum input, the run has no margin and I say so in writing.
- Which class the converter advertises, against what the camera needs. A mismatch here produces a fault that looks like a camera fault for weeks.
- Negotiated data rate at the far end, not at the switch. Ten-megabit links are common and nobody notices until the customer adds a second stream.
- A 20-minute thermal soak with load on, then loop resistance again. Copper at 50 °C is 12 percent worse than at 20 °C, and that 12 percent is usually the difference between a design and a hope.
- One deliberate power cycle of the whole rack, all cameras cold-starting together. Coax converters come up slower than the switch, and I have seen a rack where four of eight never finished booting because the port timed out first.
Questions I get asked about coax PoE
Can I feed the converter from a passive 48 V injector?
Sometimes, but check the input range first. Most expect a real 802.3at handshake or a passive 48–53 V feed with current behind it, and a 48 V brick that sags to 46 V under load is already spending margin you were counting on. It also bypasses the port's current limiting, which is the only protection those little DC-DC boards have.
Does the camera see a genuine 802.3at port?
No. The far-end converter terminates detection and re-runs it locally, so the camera sees a valid signature, but the class it is granted is the converter's choice. Anything beyond that class — heaters, pan motors, a second radio — comes out of the converter's headroom, not the switch's.
Can I split one coax run to four cameras?
Not with a passive pair. It is point-to-point by design, and four cameras on one centre conductor means four times the current through the same 15 Ω per 100 m. An EOC switch does it with four coax ports and its own budget — a different product, not a splitter.
Will 10 Mbps carry a 4K camera?
One stream, yes. A 4K camera at 8 Mbps CBR fits comfortably. Two streams, or on-camera analytics pushing an event feed, will not, and the failure is ugly: the main stream stays up while the sub-stream drops frames, so the NVR records and the phone app does not.
Is coax with decades of 12 V on it still usable?
Often, and the loop measurement tells you in ten minutes. What kills these runs is not copper, it is connectors and any water that got into the jacket. I have used 25-year-old RG59 that read 16 Ω per 100 m and worked fine, and condemned 8-year-old cable reading 40 Ω because every F connector on the run was crimped onto wet braid.
Where I stop recommending it
My rule of thumb after a few dozen of these: coax PoE is right when the load is under 15 W, the run is under 300 m, one stream is enough, and the alternative is opening a finished ceiling. That covers most analogue-to-IP retrofits, door stations, and a fair number of access points. It is the wrong answer for PTZ cameras with heaters, for anything needing a second stream, and for any run where the customer will add a device next year without calling.
The thing I got wrong early on was treating the cable as the constant and the load as the variable. It is the other way round. The load is fixed by the camera somebody already bought; the cable is what is still open to change. Measure the loop before ordering converters, put the number in the quote, and the 18:40 reboot never happens.
If you have loop readings off a long RG59 estate — especially the ones with a decade of water in the connectors — I would like to see them. My table is built from new cable and my own drums, and field numbers age better than bench numbers.