I was reading the Fibre Channel spec (FC-PH) and I noticed that they spec the skew of cables rather tightly. I don't understand why and I'm hoping you can clue me in....
Imagine a differential serial interface transmitted on a twisted pair cable.
Imagine that the + and - conductors in the cable are of different lengths resulting in a propagation delay difference, aka skew.
What impact does the +/- pair skew have on the received signal's characteristics?
I don't see a change in the received pulse width, since the differential receiver sees an equal skew from + to - and - to + transitions. The skew in a single cable is fixed and doesn't move around (jitter). There might be a change in noise margin, resulting in a form of jitter, in applications using AC coupled receivers since the + signal swing will above and to, and the - signal swing below and to the receiver's reference.
I've come to understand that in certain twisted-pair cables, like IBM type-I shielded twisted pair, the two wires of the pair are rather weakly coupled to each other. That is, there is a LOT of capacitance per unit length between each signal wire and the surrounding shield, and not so much between the two wires. Basically, what you get here is almost like having two independent transmission lines, each registering 75 ohms to ground, rather than a single 150-ohm balanced twisted arrangement.
Try it out: send a signal on one wire, and **nothing** on the other. You'll pick up some crosstalk on the other wire. If the wires had been tightly coupled, you would have gotten a full-sized, but opposite polarity signal at the receiving end of the second wire.
In such an arrangement, you can get enough skew between wires so that the signal received on the + wire, at the end of a long cable, is more than one bit removed from the signal on the - wire. This totally scrambles the signal. As you approach this BER performance drops. This is the effect of what you refer to as a deterioration in the noise margin.
Dr. Howard Johnson