Rails? What are those? You mean common mode limits?
Umm,
umm,
umm...
You're contradicting yourself: you imply there are no common mode limits*. Then you say "needs to set some reasonable CM voltage".
(*Or imply there are no such things as signals beyond the rail.)
So, this is precisely the issue I have been addressing. :)
For signals entirely within a PCB (and probably between "high speed" connectors of adequate shielding and grounding), the noise is small, and the susceptibility is small. As long as you aren't accidentally routing pairs beneath power switching pours, it'll all stay nice and quiet.
The exact scenario is putting LVDS into cables. That's fundamentally the problem, and why ALL signaling standards are either heavily shielded (USB, HDMI, etc.) or transformer coupled (Ethernet, others?).
Hmm, suppose it's worth noting, the slower, auto/industrial standards (RS-485, CAN, etc.) have receivers with beyond-the-rail input range (usually achieved with divider resistors -- probably too slow an approach for high speed signals, so you never see it done there). They still have limited range, but not being confined to a miserable 3.3V rail certainly helps against your basic IEC 61000-4-3 and -6 tests.
They're usually drawn as N-ch pairs, or something like that, so it's kind of surprising they don't crap out at < 0.8V or whatever. Maybe they do the R2R comparator trick of having complementary diff pairs, wiring their outputs together. This should be indicated in the equivalent circuit, but... you know TI, if you get an equivalent at all, you should feel lucky.
They only guarantee offset to +/-100mV of course, but typical is much better. What'd you measure, more like 10mV?
Tim