Cut open a Cat5e or Cat6 cable and you get eight conductors in four colour-coded pairs, each pair wound around itself like a rope. Most engineers know the twisting is there to fight interference. Far fewer notice the detail that actually makes it work: the four pairs are not twisted at the same rate. Count the turns per centimetre on the blue pair and then the orange pair, and the numbers are deliberately different.
That difference is not a manufacturing tolerance. It is the design.
Why twisting cancels noise at all
The idea is older than Ethernet by nearly a century. Alexander Graham Bell patented twisted pair wiring in 1881, because telephone lines strung alongside telegraph and power lines were picking up everything those neighbours radiated.
The mechanism is differential signalling. A pair carries the same signal twice — once normal, once inverted — and the receiver looks only at the difference between the two conductors. Any interference that hits both wires equally shows up as a common-mode offset on both, and subtracting one from the other makes it disappear.
The catch is the word equally. If one conductor ran permanently closer to the noise source than the other, it would pick up more, the interference would no longer be identical on both, and the subtraction would leave a residue. Twisting fixes exactly that: as the wires wind around each other, each one spends the same total length nearest to any external field. Their exposure is averaged out to be identical, so the noise stays common-mode and cancels.
The part that surprises people: different twist rates
External interference is only half the problem. The other half sits inside the jacket — three other pairs, each of them a small transmitter in its own right, running centimetres away for the whole length of the cable. Signal leaking from one pair into another is crosstalk, and the near-end version (NEXT) is one of the hardest parameters to pass in a cable spec.
Here is the failure mode. If every pair used an identical twist rate, then two adjacent pairs would hold the same relative geometry along the entire run: the blue wire would sit next to the orange wire in the same orientation, turn after turn, metre after metre. Their coupling would be consistent, in phase, and cumulative — the worst case.
Giving each pair its own twist rate (its lay length) means the pairs constantly slide out of alignment with each other. A conductor that is closest to its neighbour at one point is furthest from it a few centimetres later. The coupling that builds up over one stretch is undone over the next, and averaged along the run it collapses toward zero. Manufacturers push it further by choosing lay lengths that are not simple multiples of one another, so the geometry does not repeat on any short cycle.
Cat6 adds a plastic spline down the centre to physically hold the pairs apart; shielded Cat7 and Cat8 wrap each pair in its own foil. But the twist rate variation is the free part of the design, and it does most of the work.
Why this matters when you terminate the cable
Because the twist is the noise protection, untwisting the wires throws it away — and you have to untwist a little to land the conductors in an RJ45 plug or a keystone jack.
The standards therefore cap it: no more than 13 mm (0.5 in) of untwisted conductor at the termination. That limit is not conservative padding. Over-stripping is one of the most common reasons an otherwise healthy link fails NEXT during certification, and it is invisible once the boot is on. The cable is fine, the copper is fine, and the tester still reports a fail.
Similar logic covers the rest of the install: do not exceed the minimum bend radius (it deforms the pair geometry), do not overtighten cable ties, and do not stretch the cable while pulling it. All three change the twist, and all three degrade the same parameter.
The IoT angle
This lands hardest on industrial and building-automation deployments, where the network cable is rarely in a clean environment. Runs share trays and conduit with motor feeds, variable frequency drives, contactors, fluorescent ballasts and welding gear — sources that generate exactly the kind of broadband electrical noise twisted pair exists to reject.
In those installs the twist is not a detail; it is the entire EMI strategy for the copper. It costs nothing, it is already in the cable you bought, and the only way to lose it is by terminating carelessly. When a sensor gateway shows intermittent packet loss that no firmware change fixes, the physical layer is where we look first — and the answer is frequently 30 mm of untwisted conductor inside a plug someone crimped in a hurry.
The same principle scales up. Differential pairs on a PCB follow it too: USB, CAN, RS-485 and Ethernet traces are routed as tightly coupled pairs for the same reason, which is why length matching and consistent spacing are non-negotiable in our PCB and hardware work. Noise immunity is bought with geometry, not with software.
If you are specifying cabling for a connected-device deployment, or debugging a link that passes a continuity test and still drops packets, get in touch — the physical layer is usually the cheapest place to fix reliability.
Top comments (0)