Type 4 current, contact resistance, unmating under load, and what to write on the purchase order
A PoE++ camera that reboots at 3 a.m. is rarely a “bad Cat6A cable.” IEEE 802.3bt Type 4 puts 90 W on the PSE and 71.3 W at the PD; the missing watts are I²R in the channel. The plug is a few millimetres of that channel, but it is the only millimetres a field crew can assemble wrong, unplug live, and replace with a part that was never specified for 960 mA per pair. The existing camera-selection notes on this site cover fit and weather. This piece covers current, heat, unbalance and arcing — the four things a gold-plating number does not tell you.
Quick answer
- If the PD browns out only on long, bundled runsTreat it as channel resistance and bundle temperature first. TIA TSB-184-A holds bundle-centre rise to about 15 °C; Type 4 on 24 AWG UTP is commonly modelled around a 31-cable bundle before that limit is crossed. The plug is not the heat source.
- If one camera dies and its neighbour on the same switch does notLook at DC resistance unbalance at that termination. TIA/IEEE allow 3 % (or 0.20 Ω) within a pair and 7 % (or 0.20 Ω) pair-to-pair. A blade sitting on insulation is enough to trip those numbers.
- If ports flake after someone “just unplugged it to test”That is unmating under load. IEC 60512-99-002 is the endurance schedule for it. Gold thickness does not move the arc; contact geometry does (wipe zone vs seated zone).
- If the RFQ only says “PoE++ compatible”Reject it. Write Type (3 or 4), current per pair, DCRU limits, unmating-under-load evidence, conductor construction (solid/stranded/3-prong), and a cord AWG rule. “Compatible” is not a clause.
1. What 802.3bt actually puts through eight contacts
IEEE 802.3bt-2018 (Type 3 and Type 4, marketed as PoE++ or 4PPoE) is the first Ethernet power clause that uses all four pairs as a matter of course. The numbers that matter at the connector are current and class, not the “90 W” printed on a switch faceplate.
| Type 1 (af) | Type 2 (at) | Type 3 (bt) | Type 4 (bt) | |
|---|---|---|---|---|
| PSE max | 15.4 W | 30 W | 60 W | 90 W |
| PD available | 12.95 W | 25.5 W | 51 W | 71.3 W |
| PSE voltage | 44–57 V | 50–57 V | 50–57 V | 52–57 V |
| Max current | 350 mA | 600 mA / pair | 600 mA / pair | 960 mA / pair |
| Pairs used | 2 | 2 | 4 | 4 (mandatory) |
| Classes | 1–3 | 1–4 | 1–6 | 1–8 |
The 18.7 W gap between a Type 4 PSE and its PD is cable (and connector) dissipation, not “overhead.” On a 100 m channel that dissipation is spread along the copper. On a 3 m 28 AWG patch cord it is concentrated. On a plug it is concentrated further — but only if contact resistance is high. A clean 8P8C interface at a few milliohms dissipates milliwatts. A blade that landed on insulation dissipates enough to brown the housing and still pass a 1G autotest.
Current is common-mode on each pair: the two conductors of a pair share the pair current. Type 4 at 960 mA per pair is therefore about 480 mA per conductor if the pair is balanced. That is why DC resistance unbalance is a PoE problem before it is a heating problem: the magnetics in the PSE and the PD see the difference, not the average.
Do not buy a PoE++ RJ45 plug because a datasheet says “PoE++.” IEEE does not rate 8P8C plugs. The clause rates the channel. A plug is “PoE++ capable” only in the sense that its contacts, plating, crimp and housing survive the current, the temperature and the unmate-under-load cycles you actually apply.
2. Where the heat is, and where it is not
TIA TSB-184-A (Guidelines for Supporting Power Delivery Over Balanced Twisted-Pair Cabling) is the document that models conductor temperature rise from PoE current. The working rule it established, later reflected in TIA-568.1-D derating practice, is to keep the temperature rise at the centre of a bundle around 15 °C. Rise depends on bundle size, current, number of energised pairs, gauge and construction — not on the brand of plug.
Secondary modelling that follows TSB-184-A typically puts a 24 AWG UTP Type 4 bundle near 31 cables, a 23 AWG UTP bundle near 40, and a 23 AWG shielded bundle higher still, because the screen is a heat path. Treat those counts as order-of-magnitude, not as a permit. The bulletin’s own tables are the ones to quote in a tender.
Two consequences for the connector:
- The plug is not the thermal bottleneck of a horizontal run. A 90 m 23 AWG cable has ohms; a mated 8P8C interface has milliohms. If the jacket is warm in the tray, look at bundle size and ambient, not at gold thickness.
- The plug is a local hot-spot if the crimp is bad. Heat then sits in 8 mm of phosphor bronze and a polycarbonate housing rated UL 94. That is a different failure: discoloration at the blade windows, a camera that dies after the enclosure lid goes on, a contact that works when cold and opens when the PoE class steps up.
Insertion loss also drifts with temperature. A link that passed at 20 °C can sit on the wrong side of 2 dB after a summer afternoon in a closed bundle. Re-test warm if the complaint is “it worked in March.”
What we will not put a number on. ΔT at the plug body at 960 mA per pair is a first-party measurement. We have not published one on this page. If a supplier quotes “PoE++ 90 W rated” without a current, a duration, a bundle or a thermocouple location, they have not measured it either.
3. Contact resistance vs DC resistance unbalance
Two different quantities get collapsed into “the plug is resistive.” Separate them on the RFQ.
| Quantity | Typical limit cited | What a plug defect does to it |
|---|---|---|
| DC loop resistance, permanent link | ≤21 Ω (TIA); channel often cited ≤25 Ω (IEEE) | Almost nothing, unless a conductor is nicked or a blade sits on insulation. Loop resistance is a cable-length number. |
| DC resistance unbalance within a pair | 3 % of loop, or 0.20 Ω, whichever is greater | One short crimp, one long untwist, one strand missed. This is the plug’s real PoE fingerprint. |
| DC resistance unbalance pair-to-pair | 7 % of the parallel resistance of the two pairs, or 0.20 Ω | Load-bar inversion, mixed solid/stranded in one plug, one pair not bottomed. |
| Mated contact resistance (component) | Often specified ≤20 mΩ initial (IEC 60512-2-1 / EIA-364 millivolt method) | Plating, normal force, contamination. Ask for the change after life, not the initial screenshot. |
IEEE 802.3bt also requires the PSE to tolerate a DC unbalance current of 3 % of the powered pairset current (clause 145.2.10.12). That is an equipment requirement. It is not a licence for sloppy terminations. Transformers saturate; data distorts; Type 3/4 four-pair power can simply stop.
Fluke DSX-class testers measure both within-pair and pair-to-pair DCRU. A green NEXT pass with a red DCRU is a PoE fail waiting for the first Class 8 PD. Put DCRU in the handover, not only the Cat6A autotest.
On incoming inspection of plugs, milliohm contact resistance on terminated samples is the useful number. A workable internal rule used on our own lots: ≤10 mΩ initial on a terminated sample and ≤2 mΩ change after 50 insertion cycles. Tune the numbers; keep the structure (sample count, method, change-from-initial).
4. Unmating under load — the failure that gold thickness does not prevent
A PSE does not drop power because you pulled a plug. The PD classification is still valid; current is still flowing. Separation then draws an arc as the last metal-to-metal point opens. The arc is not a shock hazard at these voltages. It is a plating hazard. Erosion and corrosion start at the spark and creep into the seated contact zone if the geometry does not keep them apart.
IEC 60512-99-001 was written around 55 V DC and 600 mA per contact (Type 2 current). IEC 60512-99-002 is the Type 3/4 schedule: unmating under electrical load, 100 operations, with the later 2025 revision of 99-002 cited in the trade press at 55 V DC and 2 000 mA per conductor and a 20 mΩ maximum resistance change after the sequence. Quote the edition on the PO. Do not accept a Type 2 certificate as Type 4 evidence.
Design, not plating thickness, decides where the pit sits:
- A contact with a distinct wipe / make-first-break-last zone, kept millimetres away from the fully seated patch, lets the arc live in a sacrificial area.
- A flat, short contact puts the pit on the same gold the jack will sit on for the next five years.
Repeated live unplug can weld. Re-inserting a pitted plug then dumps debris into the jack. The field habit that causes this is “I’ll just swap this camera while the switch is up.” Write the rule on the panel: de-energise the port or use a powered-device shutdown before you pull a Type 3/4 plug.
5. Blade geometry, nickel and gold — what actually carries 480 mA
Three mechanical facts get sold as electrical ones.
3-prong vs 2-prong. A 3-prong (three-point) blade is made to catch stranded conductors; a 2-prong (two-point) blade is made to straddle a solid. Using a 2-prong on 26 AWG stranded patch is a classic DCRU factory. Using a 3-prong on a large solid 23 AWG Cat6A can pierce past the copper and leave a weak spring. Match the blade to the conductor construction, then verify with a cut-away. Category printed on the bag is not a blade spec.
Nickel underplate. Gold is porous at the thicknesses used on 8P8C contacts. Nickel is the diffusion barrier and the hardness layer. A “50µ" gold” contact with a thin or missing nickel barrier fails in mixed gas and salt spray long before a 15µ" contact with a proper barrier. For the plating argument itself, use the site’s gold plating guide (3µ", 15µ", 50µ") — this article will not repeat it. For PoE the question is contact resistance after environment, not the gold number on the carton.
Normal force and seating depth. Type 4 does not need a special “high-current” 8P8C body. It needs the blade to sit on copper, the housing to hold the jack’s contact in the designed patch, and the latch to keep that geometry through thermal cycles. A plug that seats 0.5 mm proud has already lost normal force. PoE then finishes the job.
6. 26 AWG and 28 AWG cords are a different problem
Horizontal cable is 22–24 AWG. Patch cords are not. TIA allows 22–28 AWG in cords, with extra rules for 28 AWG: channel length of the 28 AWG portion capped at 15 m, insertion-loss derating per Annex G, and TSB-184-A-1 for bundle temperature of 26/28 AWG cords.
A slim 28 AWG cord on a Type 4 port is a heat and voltage-drop concentrator sitting in the one place with the least airflow — the patch field. It also needs a plug window sized for 28 AWG insulation, which a “Cat6A 23 AWG” pass-through will not provide. Two failures then stack: high cord resistance, and a crimp that never bit the conductor.
Practical split used on jobs that survive handover:
- Type 4 horizontal: 23 AWG (Cat6A), bundle planned to TSB-184-A.
- Type 4 patch: 24 AWG, short, not bundled into a 48-cord loom if you can avoid it.
- 28 AWG: data-only, or Type 1/2, ventilated, and short. Not a Type 4 camera trunk.
Gauge selection for the plug itself is covered in how to choose an RJ45 plug by wire gauge. The PoE-specific addition is: the same plug family is not valid across 23 AWG solid horizontal and 28 AWG stranded patch.
7. Purchase language that can be inspected
| Clause | Write this | Evidence |
|---|---|---|
| Power class | IEEE 802.3bt Type 3 and/or Type 4; 600 or 960 mA per pair stated | Not a logo. A test current, duration and thermocouple location, or an explicit “not type-tested; channel is the rated object.” |
| Unmating under load | IEC 60512-99-002 (edition dated), 100 cycles, resistance change limit | Test report with voltage, current per conductor, cycle count, ΔR. |
| Contact resistance | IEC 60512-2-1 millivolt method; initial and after life | Change-from-initial, not a single “<20 mΩ” line. |
| DCRU on terminated samples | Within-pair 3 %/0.20 Ω; pair-to-pair 7 %/0.20 Ω on your cable P/N | DSX (or equivalent) native file on 10 terminated samples per lot. |
| Blade | 2-prong / 3-prong / dual; solid and/or stranded; AWG and insulation OD | Cross-section photo of a crimp on your cable. |
| Housing | UL 94 class, operating temperature, latch retention after cycles | Material declaration; no “90 W plastic” claims. |
| Cord rule | Forbidden gauges for Type 4 (e.g. 28 AWG not permitted on Class 7/8) | Written in the job spec, not left to the installer. |
For camera-side fit, weather and enclosure practice, keep using how to choose RJ45 plugs for PoE IP camera installations. That page is the installation companion. This page is the current-and-heat companion. They should point at each other; they should not copy each other.
8. Field checklist
| # | Check | Accept when |
|---|---|---|
| 1 | PSE type vs PD class | Type 4 port for Class 7/8 PDs; no “90 W” sticker as the only evidence |
| 2 | Plug blade vs conductor | 3-prong on stranded, 2-prong or specified dual on solid; cut-away agrees |
| 3 | All eight blades on copper | No insulation in the window; pass-through tips trimmed flush |
| 4 | Jacket in the crimp ramp | No rotation by hand |
| 5 | DCRU on the finished link | Within-pair ≤3 % or 0.20 Ω; pair-to-pair ≤7 % or 0.20 Ω |
| 6 | Loop resistance | Permanent link ≤21 Ω; matches length and gauge |
| 7 | Bundle size at Type 4 current | Planned to TSB-184-A; 15 °C rise not treated as optional |
| 8 | Patch cord AWG | No 28 AWG on Type 4 trunks; 28 AWG length ≤15 m if used at all |
| 9 | Live unplug rule | Port powered down before unmate on Type 3/4 |
| 10 | Re-test warm if the site is hot | Worst margin still ≥2 dB with the bundle at operating temperature |
Frequently asked questions
Tap a question to expand it.
Is there such a thing as a “PoE++ RJ45 plug”?
Not as an IEEE-rated component. 802.3bt rates the channel and the PSE/PD. A plug can be shown to survive Type 4 current, DCRU limits on a named cable, and IEC 60512-99-002 unmating. If the only evidence is the words “PoE++ compatible,” you are buying a label.
Does 50µ" gold make a plug safe for 90 W?
No. Gold thickness is a corrosion-and-cycle property. Type 4 current is carried by copper area, normal force and a nickel barrier. Unmating-under-load damage is a geometry problem. Use thicker gold where the cord will be unplugged often or sit in a corrosive room; do not use it as a power rating.
Why does a camera work on a 3 m test lead and fail on the 90 m run?
Voltage drop and heat. Type 4 leaves 18.7 W in the channel at full load. On 3 m that is negligible; on 90 m of 24 AWG in a bundle it is not. Confirm loop resistance against length, then bundle size against TSB-184-A, then DCRU at the two terminations. In that order.
Can I use pass-through plugs on PoE++?
Yes if the window fits the insulation OD, the blades match solid or stranded, and the tips are trimmed flush so they cannot short or corrode into the jack. Pass-through does not add current capacity and does not remove DCRU risk. A 23 AWG Cat6A solid in a window cut for 24 AWG is still a miss, PoE or not.
Should I test DCRU on every link or on a sample?
On Type 4 / Class 7–8 jobs, every permanent link. The test is seconds on a DSX-class certifier and it is the only field number that sees a one-blade miss. On Type 1/2 camera jobs, a statistical sample plus any link that already looks marginal on loop resistance is usually enough.
Is shielded cable required for PoE++?
Not by IEEE. Shielding is an EMI and heat-spreading choice. TSB-184-A modelling generally allows larger bundles on shielded 23 AWG than on UTP of the same gauge, because the screen conducts heat. If you do use shielded cable, the plug must complete the screen — that mechanism is in the companion article on 10G shielded links, not here.
What current should a supplier put on a plug drawing?
A component current at a stated temperature, or a reference to IEC 60512-5 (current-temperature derating), plus the unmating-under-load schedule. “90 W” is not a current. 90 W at 52 V is about 1.7 A for the whole channel, shared across four pairs. The drawing should say milliamps per contact, not watts per port.
Send the cable P/N, not a wattage
We crimp shielded and UTP Cat6A/Cat8 plugs against the cable you actually buy: blade type, insulation OD, jacket OD, and a DCRU file on the terminated sample. If you need Type 4 patch cords, say so — the 28 AWG body is a different window.
Related: PoE IP camera plug selection · Gold plating 3µ" / 15µ" / 50µ" · Plug by wire gauge · Shielded plug and 10G link failure
Sources and standards referenced
- IEEE Std 802.3bt-2018 / Clause 145 — Type 3 (60 W / 51 W) and Type 4 (90 W / 71.3 W), 4-pair power, 600 mA and 960 mA per pair, classes 5–8, pairset unbalance 3 % of pairset current (145.2.10.12).
- IEEE Std 802.3af / 802.3at — Type 1 15.4/12.95 W at 350 mA; Type 2 30/25.5 W at 600 mA per pair.
- TIA TSB-184-A and TSB-184-A-1 — power delivery over balanced twisted pair; ~15 °C bundle-centre rise; 26/28 AWG cord addendum. TIA-568.1-D — conductor temperature derating for bundled PoE.
- ANSI/TIA-568.2-D / 568.2-E — permanent-link DC loop ≤21 Ω; DCRU 3 %/0.20 Ω within pair, 7 %/0.20 Ω pair-to-pair. IEEE channel loop often cited ≤25 Ω.
- IEC 60512-2-1 (contact resistance, millivolt method); IEC 60512-5 (current-temperature derating); IEC 60512-99-001 / 99-002 (unmating under electrical load). 99-002 Ed. 2.1 (2025-04) is cited in the trade press at 55 V DC, 2 000 mA per conductor, 100 cycles, 20 mΩ ΔR — confirm the edition in your copy before a tender.
- TIA cord rules for 28 AWG: 15 m maximum in the channel; insertion-loss derating per Annex G.
Figures on this page are standard limits and secondary published models, not thermocouple results from our line. Bundle-size counts (~31 / ~40) are order-of-magnitude figures from TSB-184-A-based modelling in the trade press; use the bulletin’s tables in a contract. For a named plug on a named cable we will send DCRU files and a crimp cross-section. We will not invent a wattage rating for an 8P8C body.
WeChat ID: