I think it comes down to comparing a controlled impedance line to one which
is out of control. A wire without it's reference called an antenna at some
frequency. The G-S-G truly brings the wire back under control a buys allot
of performance.
---------------------- Forwarded by Brad D Knott/WLGORE on 09/19/97 04:04
PM ---------------------------
[email protected] on 09/19/97 03:22:31 PM
To: [email protected]
cc: (bcc: Brad D Knott/WLGORE)
Subject: [SI-LIST] : Single-ended SCSI vs. differential SCSI
I understand the many advantages of differential signaling over twisted
pairs -- no need to worry about ground loops or ground bounce (as much),
much less crosstalk, far less suseptibility to induced noise and hence
better EMI performance. Where I work, we use LVDS (200mV swing) to ship
digital camera data about twelve feet at 30MHz. It works quite well (not
that this implies we really know what we're doing or that we employ any
SI gurus :-) ).
Anyway, I'm curious as to how well single-ended SCSI works, particularly
in comparison to true differential SCSI and also the "fake" 25 wire
version of SCSI used by Apple. In the 25 wire version, there's only one
ground and no twisted pairs -- clearly making it difficult to have a
large amount of signal integrity as your clock rate goes up. But how
about the typical single-ended version found in most PCs? Every other
wire is a ground wire, and the ribbon cables are generally not twisted
pairs. How much of an advantage is this over the single ground wire
approach? This is what I figure:
-- The extra ground wire spaces the conductors twice as far apart, giving
you a crosstalk reduction of about four.
-- There seems to be little chance that the current you send down one
wire will come back on its associated ground wire (I'm assuming that at
the receiver end all the grounds are connected together). This creates
varying impedances that probably aren't very friendly to the data edges.
Perhaps the premise about "little chance..." is mitigated by the fact
that the lowest inductance return paths usually will be the nearest
neighbor ground wires and there _will_ be a large amount of roughly
equalized ground currents in each return wire?
-- The lack of twists would make the system less immune to induced noise
and produce more EMI.
It seems to me that this isn't really all that much better than, say, one
ground wire with widely spaced connectors. I suspect I'm missing
something here and it's actually somewhat better than I'm thinking.
Comments?
Interestingly, "wide" SCSI usually uses .025" spaced ribbon cables -- I
wonder how its performance over 34 conductor [one ground] .050" spaced
ribbon cable would be? Awful given the very large (and uneven)
inductance return currents would face?
Finally, let me say that I'm very impressed with the technical knowledge
that the people on this list possess. You guys who can send a gigabit
per second down a single twisted pair thousands of feet -- that's damned
impressive, but the general public is probably never going to recognize
that fact!
---Joel Kolstad
[email protected]