The total differential skew at points x(t) and y(t) in Figure 1 equals the delay in the top path minus the delay in the bottom path, tskew=t1+t3+t5–(t2+t4+t6). Rewriting this equation to group common terms produces tskew=(t1–t2)+(t3–t4)+(t5–t6), a number dependent on both the device under test and the matching of the test cables or probes.
Many types of equipment allow you to automatically calibrate out, or deskew, the effect of test cables. To deskew your setup, first connect the two cables directly from b to c and from x to y, shunting around the device under test. Should delay t1+t5 exceed delay t2+t6, your equipment automatically inserts a compensating delay in the measuring circuit. This procedure doesn't eliminate skew; it just inserts a permanent compensating delay, Δt=t1+t5–(t2+t6), in series with pathway t6. Assuming the two pathways within the device under test operate in an independent, uncoupled manner, further measurements of skew with the device under test in place should always return the correct (deskewed) answer.
What happens if you crisscross the connections at points c and y? Then the signal x(t) comes from the inverted side of the generator, passing through delays t2, t4, the crisscrossed connection at c–y, and finally t5. A new measurement of skew now gives you tcrossed=(t2–t1)+(t4–t3)+(t5–t6). One-half of the sum (tskew+tcrossed) extracts the skew following point c–y, which includes the last cable section plus any artificial skew correction embedded within the receiver. One-half the difference (tskew–tcrossed) extracts the skew preceding point c. By crossing the connections at several points and retaking the data each time, you can determine the amount of skew within each individual section of the setup.
Next, while the connections at c–y remain crisscrossed, make a second crisscross at b and x. The average of this new measurement, tdoublecross, with the original, tskew, isolates just the skew in the test setup, (t1–t2)+(t5–t6). Half of (tskew–tdoublecross) extracts just the skew within the device under test. Crisscrossing is the basic technique for investigating the accumulation of skew in highly cascaded systems.
A highly coupled device under test, such as a transformer, turns up some serious flaws in the deskewing procedure. Assume in Figure 1 that t2, t5, and t6 equal zero, but t1 doesn't. Place an amount of compensating skew (Δt=t1) in series with t6 to automatically balance out (deskew) t1. With the cables directly connected from b to c and x to y, the equipment reads zero skew. Now insert a perfect, zero-skew transformer into the device-under-test position. A perfectly balanced transformer passes differential signals un-changed but blocks common-mode signals. The signal at point c–y therefore by definition has zero common-mode content and zero differential skew (Reference 1). In this case, even though the transformer is perfect, an automatically deskewed equipment setup displays a skew equal to the negative of its internal compensation, (–Δt). That's not right, illustrating perfectly my point that you can't depend on automatic deskewing when measuring skew in tightly coupled differential systems.
In a tightly coupled system, you must separately match both transmitting and receiving cables for zero skew before you start measuring. You know your setup is right when you can crisscross the instrument cables at any point without altering the results.
 Johnson, Howard, "Common-mode analysis of skew," EDN, Jan 22, 2004