Kurt, when you measure these cables, do you compare the twists in the colored cables? If all the wires are twisted the same would that cause interference and lead to poor test results? If so, would changing the twists result in better test results?
We don't compare the twists when measuring them, but when the cable was manufactured, the twist rates were set relative to one another to minimize crosstalk. And yes, you're right about the twist-rate issue; here's how it works, more or less.
Common-mode noise rejection in a twisted pair cable works if we assume that the source of the interfering noise hits the cable symmetrically--that is, if there's +10 microvolts on the "plus" member of the pair, there's likewise +10 microvolts on the "minus" member. If the receiving circuit is properly balanced, the differential amplifier on the receiving end will not "see" either of these because they arrive in time with, and at the same amplitude as, one another.
For external noise sources--e.g., a noisy doorbell buzzer located a foot away from the cable--this assumption of symmetry works well. The problem, when you get to crosstalk, is that there is a pattern to the signals in the other pairs. If, worst-case scenario, the twist rates in all of the pairs are identical, the noise is extremely asymmetrical, because the same wires will repeatedly approach, and separate from, one another again and again along the length of the cable. So, if the "plus" conductor of pair 1 is repeatedly close to the "plus" conductor of pair 2, and is repeatedly (relatively, of course) far from the "minus" conductor of pair 2, this will mean that the noise which comes from the "plus" conductor of pair 1 is highly asymmetrical in pair 2. At the same time, the corresponding profile of the "minus" conductor of pair 1 relative to pair 2 will be equally asymmetrical, and in opposite direction. Now that we've completely invalidated the "symmetrical noise" principle for this hypothetical cable, common mode noise rejection works very, very poorly and crosstalk is really high.
If you've got four pairs in a bundle, too, you have to worry about six different crosstalk relationships--so you can't just arrange the twists so that pairs 1 and 2 have minimized crosstalk. You've got to find four different twist rates that make the relationship between the pairs such that the crosstalk from each pair hits each other pair in a relatively symmetrical fashion, so that common mode noise rejection can do its work.
My suspicion is that if one were to look at some of the "Cat 6" cables we tested, you'd find that the twist rates are not well-chosen, and that this is why some of them are really lousy on crosstalk.
To sum up: you're very right that the twist rates matter; you're right that having the same twist rates would strongly affect the test rates; and yes, if that were the case, changing the twist rates would improve things. Finding the optimal twist rates, by the way, turns out to be a fantastically difficult mathematics exercise--once upon a time, Belden actually rented a Cray supercomputer to solve that problem. Of course, these days computing power isn't quite so difficult to find as it was back then.
Another problem, incidentally, on crosstalk is that common-mode noise rejection in a cable gets tougher as frequencies go higher, simply because no pair is ever completely free from intrapair (not to be confused with interpair) skew--that is, the electrical length of the two conductors isn't identical. This means that even when noise IS completely symmetrical relative to the cable, it may not cancel as well as one expects because it arrives sooner on one conductor, and finishes later on the other. When we're dealing with analog mike cable, this is not an issue because we're talking about picoseconds here--but in high-speed data, picoseconds matter.
Kurt
