A camera on a pole reboots every evening around seven. It comes back within ninety seconds, works all night, works all next day, and reboots again. Swap the port on the switch: fine for three weeks, then the same pattern. Swap the camera: fine for a month, then it starts again. Replace the injector, re-terminate both ends, update firmware, and the fault moves to a different camera on the same run.
I have chased that exact symptom four times in fifteen years, and three of those four ended in a cable. Not a broken cable — a cable that passed a continuity test, passed a cheap link tester, negotiated gigabit, and still could not carry the power the label promised. The label said 23 AWG. It did not say what the 23 AWG was made of.
Two evenings a week for a month is roughly the point where a customer stops believing the problem is theirs.
What "23 AWG" actually guarantees
AWG is a diameter, nothing more. 23 AWG means the conductor is 0.573 mm across. It says nothing about the metal inside, the thickness of any plating, or how many strands the conductor was built from. A copper-clad aluminium conductor can be drawn to exactly 23 AWG and hit the diameter spec on the drawing while carrying two to three times the resistance of copper of the same size.
The standard that actually matters for PoE is not the gauge line on the box, it is the maximum conductor resistance in the regional cable standard. For North American Cat5e and Cat6, that ceiling is set per 100 m at 20 °C. The values below are the limits, not typicals:
| Conductor | Max DC resistance, one conductor, per 100 m at 20 °C |
|---|---|
| 22 AWG solid bare copper | 5.8 Ω |
| 23 AWG solid bare copper | 7.4 Ω |
| 24 AWG solid bare copper | 9.4 Ω |
| Copper-clad aluminium, nominal 23 AWG | no standard figure — measured samples run 1.8 to 3.5× the copper value |
That last row is the whole problem. There is no CCA entry in the cable standard because CCA is not a compliant conductor for the category. Which means a reel can be sold as "Cat6 23 AWG" while the property that decides whether PoE works has never been tested, never been published, and in most cases never been measured by the seller either.
The number that decides everything: loop resistance
PoE does not use one conductor per polarity. It uses two. On 802.3af and 802.3at, power goes out on two pairs: one pair carries the positive side, the other carries the return. Within each pair the two conductors are in parallel, so their resistance halves. The two pairs are in series with the load, so the halves add back up.
The result is convenient: the loop resistance of a PoE run is approximately equal to the resistance of one conductor over the same length. For a full 100 m pull of 23 AWG copper at its limit, that is about 7.4 Ω. For a good 24 AWG patch-heavy run, a little over 9 Ω.
Now put current through it. An 802.3at device drawing 25.5 W at the powered device side pulls roughly 600 mA. Voltage dropped in the cable is current times loop resistance, and the powered device only sees what is left of the PSE output.
The budget is tighter than most people assume. Under 802.3at and 802.3bt, the PSE may output as little as 50 V, and the powered device is only guaranteed to keep running down to 42.5 V. That leaves 7.5 V of total loss budget for the entire cable run — not just the horizontal cable, but every patch cord, every punch-down, every coupler in the path.
| Conductor, 100 m | Loop resistance | Drop at 600 mA | PD sees from a 50 V PSE | Verdict |
|---|---|---|---|---|
| 23 AWG bare copper | 7.4 Ω | 4.4 V | 45.6 V | inside budget |
| 24 AWG bare copper | 9.4 Ω | 5.6 V | 44.4 V | inside, thin margin |
| CCA, 1.8× copper resistance | ~13 Ω | 7.8 V | 42.2 V | at or below the floor |
| CCA, 3× copper resistance | ~22 Ω | 13.2 V | 36.8 V | device browns out |
Read the third row again. A cable only 80% worse than copper eats the entire loss budget before you have accounted for a single patch cord. And that is at 20 °C, at nominal current, on a reel that was probably not the worst one in the shipment.
Turn it around and the same arithmetic tells you how far a conductor can actually go. Working backwards from a 7.5 V drop budget at 600 mA gives a maximum loop resistance of 12.5 Ω:
| Conductor | Resistance per 100 m | Max run at 25.5 W (7.5 V budget) |
|---|---|---|
| 23 AWG bare copper | 7.4 Ω | about 168 m |
| 24 AWG bare copper | 9.4 Ω | about 133 m |
| CCA at 1.8× copper | ~13 Ω | about 96 m |
| CCA at 3× copper | ~22 Ω | about 57 m |
The copper figures exceed 100 m on purpose. Real installations carry margin, which is precisely why a compliant cable keeps working when someone adds a 30 m patch run nobody documented. CCA at three times copper resistance falls off a cliff at 57 m — and 57 m is a perfectly ordinary distance between a mid-span cabinet and a pole.
Why the failure looks like everything except the cable
Here is what makes this expensive to diagnose. PoE discovery does not test your cable. The PSE applies a low voltage, looks for the 25 kΩ signature resistor that identifies a valid powered device, optionally performs class hardware negotiation, and then applies full voltage. All of that happens at almost no current, so a high-resistance conductor passes every step without complaint.
The failure only appears under load, and it appears as a voltage problem rather than a cable problem. The powered device browns out, its own supervisor resets it, and it comes back up drawing less current — which is why it often recovers on its own and works until the next peak. On a camera with IR illuminators, the peak arrives at dusk. On an outdoor unit with a heater, it arrives when the temperature drops. On an access point, it arrives when the office fills up.
Four field symptoms that should send you to the cable, not the electronics:
1. The fault follows the run, not the device. Move the camera to another port on the same cable and it still fails. Move it to a different cable and it behaves. That is a cable signature, and no amount of firmware will fix it.
2. It is time-of-day dependent. Anything that correlates with darkness, cold, or occupancy is correlating with current draw.
3. Link stays up while power drops. Data needs a fraction of the current that PoE needs. A cable that is marginal for power is often perfectly fine for a 1 Gbps link, so the switch reports a healthy link and the device still reboots.
4. A cheap tester says the run is good. Most sub-$200 testers measure continuity and wiremap. Almost none measure loop resistance under load, which is the only number that matters here.
Heat makes it worse, and cabinets are hot
Copper resistance rises with temperature, at roughly 0.4% per degree Celsius. That sounds small until you apply it. A run measured at 20 °C on the bench gains about 12% resistance at 50 °C. A sealed wall cabinet running a loaded PoE switch in a Chinese summer sits well above that, and the cable inside it is at cabinet temperature, not ambient.
Apply the correction to the numbers above and a 24 AWG copper run that had 5.6 V of drop now has 6.3 V. Still inside budget. Apply it to CCA at 1.8× and the drop goes from 7.8 V to 8.7 V — now clearly outside, on a cable that measured borderline in air conditioning. This is the mechanism behind the "it worked all winter and died in July" report, and it is why I never accept a bench measurement as proof of a site.
If a PoE cabinet is going to run above 45 °C for weeks at a time, the honest answer is not better cable alone. It is a cabinet with airflow, and an industrial-rated switch that is specified to derate its PoE output rather than shut down.
Testing a reel before you pull it
You cannot inspect your way to an answer here. Plating thickness is invisible, and the seller's datasheet is a copy of somebody else's datasheet. Measure it, or insist on a measurement:
1. Cut a two-metre sample from the reel you will actually install. Not from the sample box, not from the previous shipment. Reels are bought in lots and lots differ.
2. Measure one conductor with a four-wire method — a milliohm meter, or a bench supply pushing a known 1 A and a meter reading the drop across the sample. Two-wire multimeter readings on a two-metre length are dominated by lead resistance and will mislead you.
3. Convert to Ω per 100 m and compare against the table above. Allow a small amount for twist: stranded and twisted conductors are physically longer than the cable they sit in, typically a few percent.
4. Measure at a second temperature if the cable will live in a hot space, or apply the 0.4% per degree correction yourself and check the result against the 7.5 V budget.
5. Cut the conductor lengthwise and look at the cross-section. A copper-clad conductor shows a silver core under a thin copper skin. It is the fastest confirmation in the field and it costs one destroyed sample.
6. Keep the sample and the measurement. When a run fails in eighteen months, the reel that produced it is long gone and you will want the number.
If a supplier cannot produce a conductor resistance figure in Ω/km from a test report, that is the answer. Suppliers of compliant cable have that number on file because their buyers ask for it every week.
While the reel is open, check the connectors going onto it. A field-terminated plug with the wrong contact plating adds its own resistance where current density is highest. The PoE accessories I keep on the bench all specify gold plating in writing.
What to write into the purchase order
The gauge line on the box is not a specification, so do not accept it as one. Three clauses do the work:
Conductor material, explicitly. "Solid bare copper conductors, no copper-clad aluminium, no copper-clad steel." Written into the PO, this converts a later argument into a documented non-conformance.
Maximum conductor resistance, by number. State the ceiling in Ω per 100 m at 20 °C for the gauge you are buying, and require a per-lot test report. Anything at or under the category limit is acceptable; anything above is rejected regardless of what the drum says.
PoE intent, stated up front. Tell the supplier the cable will carry 802.3at or 802.3bt power, at what distance, in what environment. A supplier who understands the application will quote a different reel than one who is answering a generic Cat6 enquiry, and the price difference is smaller than one callout.
Where CCA is still the right answer
I am not going to pretend CCA has no place. It is lighter, cheaper, and for a patch cord carrying data over three metres at 5 V it is entirely reasonable. It is also fine for non-PoE data runs inside a rack.
The line I draw is at power. If a conductor is carrying PoE current over distance, CCA is the wrong material, and the failure mode it produces — intermittent, load-dependent, invisible to every diagnostic the switch offers — is the most expensive kind to find after handover. On a run carrying 802.3bt at 60 W or 90 W, the current roughly doubles, the loss goes up with the square of it, and the margin that CCA was already eating disappears entirely.
If the distance is the problem rather than the copper, the answer is not a cheaper conductor. It is a PoE extender that re-powers the run mid-span, or a fibre link with a local injector, both of which I have written about separately. If the conductor itself is the problem, buy copper and stop thinking about it.
One last note on switch selection, because it interacts with this. A switch that reports per-port voltage and current under load turns a two-evening cable hunt into a five-minute read. Cheap unmanaged units report neither, which is part of why marginal cabling survives so long on those networks — nobody can see it failing. The PoE switch range I work from lists per-port telemetry on the models where it matters.
Questions I get asked every week
Can I tell CCA from copper by looking at the cable?
By the jacket, no. By the cross-section, yes — bare copper is copper-coloured all the way through, CCA shows a silver-white aluminium core under a thin copper layer. It is also noticeably lighter per box, but weight is a poor test because jacket and filler materials vary.
My link tester passes the run. Isn't that enough?
No. Continuity and wiremap confirm the conductors are connected, not that they can deliver 25 W at the far end. A run with three times the resistance of copper passes both tests comfortably and still browns out a camera at dusk.
Does CCA cause data errors too?
Sometimes, at longer distances and higher speeds, because the same resistance that drops voltage also attenuates the signal. But the typical complaint is power-related reboots, because PoE current is far larger than the current involved in signalling. A CCA run that is stable for data can still be unusable for power.
How much does compliant copper actually cost more?
On a 305 m box, the gap between CCA and bare copper is smaller than the gap between two grades of jacket compound. Against one after-hours callout plus a re-pull, the arithmetic is not close.
Will a higher PoE class fix the voltage drop?
No, it makes it worse. Moving to 802.3bt raises the current, and loss rises with the square of the current. A cable that was marginal at 25.5 W is not rescued by a 60 W port; it is finished faster.
What is the practical distance limit for PoE over compliant copper?
The 100 m channel limit stands, and compliant 23 AWG cable has enough margin inside it that real installations with a few undocumented patches still work. Beyond 100 m the answer is an extender or fibre, which is a different article and a different budget line.
Fifteen years in, the pattern has not changed. When a PoE fault is intermittent, load-dependent, and follows the run rather than the device, the cheapest thing to check is the thing nobody checks: what the conductor is made of. I keep a two-metre cut sample from every reel I sign off, labelled with the lot number and the measured resistance, and it costs one metre of cable.