[Color Computer][Coco] Tandy Hard Disk Controller

[Gene Heskett]

On Saturday 04 March 2006 23:25, Mark Marlette wrote:
>Gene,
>
>The 4n1 worked, but it didn't even follow the electrical spec for
>SCSI. I'm sure if you would have loaded the bus with all devices it
>would have failed. Not the case with the TC^3.

I do now that it failed when a cdrom was also attached and set for the 
next address up.  I haven't tried that with your tc^3 yet though, I 
need to salvage another connector to install, & find the round tuit to 
get it done.
 
>Glazing over??? Come on Gene I wanted you to teach me something I
>didn't already know about SCSI. :)

Doubtfull if I could have taught you anything about scsi, even a decade 
ago.  The one single thing I've long since learned is that very few 
engineers building this stuff, really, truely, understand scsi's 
termination.  The biggest single reason scsi has a reputation for 
needing ritual sacrifices, vestal virgins prefered to make the bus 
actually work I'll discuss.

The first huge engineering, or bean counter mistake, I think sometimes 
the engineers get over-ridden in their parts procurements by bean 
counters that don't always understand the nuances of the difference in 
something as simple as a diode.  The engineer usually knows about the 
problem, and specs a schotkey for the bus isolation diode in an attempt 
to maintain a decent noise margin in the resting logic 1 voltage.  With 
that, the usual 220-330 resistor network used per line gives maybe 2.85 
volts.  Thats arrived at via assuming the psu is 5 volts, the schotkey 
drop is .15 to .2 volts, leaving a 4.8 to 4.85 voltage for the 
termpacks.  So the logic line on the buss is sitting at an open 
collector voltage of 2.88 volts, and the bus has a noise margin of 440 
millivolts above a guaranteed logic 1.  With decent cabling, the buss 
will work error free even if it does ring a little.  And it will.

Now sub a small si diode for that, which may well drop that 4.8 volts to 
4.3 or less as the si diode will drop .75 volts at that current level 
typically. (4.3/550)*330 now equals 2.58 volts, only a 180 millivolt 
noise margin.  Then due to wireing losses and old age in the psu, its 
now down to 4.88 volts on the 5 volt line at the motherboard connector, 
and by the time it gets thru the motherboard to the scsi card & the top 
of that diode, its maybe 4.6 volts.  Subtract the diodes nominal drop 
of .7 volts and you have (3.9/550)*330=2.38 volts and no noise margin.  
The bus rings maybe .7 volts on a signal edge, and there's not a ttl 
gate in the world that won't see that and make a mistake.  Not even 
active terms on the other end will fix that.  And people wind up 
sacrificing goats, chickens (I used to use KFC's, but after they made 
me sick a few times I got smart), whatever, trying to find the magic 
incantation that makes it work.

I don't recall how many scsi busses (7 or 8 at least) I've converted 
from flakey to dead solid with nothing more than a small schotkey power 
diode replacing the cheapo si diode that the card shipped with back 
when I was doing some amiga work.  And one run of an expensive card 
actually had the termpacks installed backwards so the resting voltage 
was only about 1.9.  Soldered in, it was easier to cut and jump the pcb 
to fix it than to extract 3 termpacks and turn them around.  I called 
the maker and he wasn't the least bit embarrased.  And that was the 
last of those cards we ever bought.

The bus 'ringing' is just another word for VSWR, and those 220-330 
resistor packs are just the nearest handy std value that attempts to 
match the impedance of a ribbon cable, but misses the mark by at least 
10 ohms most of the time for flat cables, & I haven't the foggiest how 
far off it is for round cables, I shudder to contemplate it.  And I've 
had much better luck with std ribbon cable than with rounded stuff.  At 
least one can mathematicly calculate the flat ribbons impedance with a 
10% slop figure and come up with something in the 118-124 ohm range if 
I did it right.  The termpacks 220 and 330 in parallel give you around 
132 if the tolerances are nominal, so you can see there is an error, 
leading to some ringing.

Active terms are another horse entirely, generally consisting of a bias 
supply, and a power op-amp that supplies about 3 volts at a very low 
impedance.  All the data lines are then fanned out from this fixed 
supply, usually by trimmed onchip r's of about 132 ohms per line, so 
both the logic 1 voltage they supply, and the terminating impedance are 
a lot closer to ideal.  The development of this technology was driven 
by the portable business, where every watt you burn is that much less 
battery life.  Termpack resistors draw .818 watts of power full time, 
the active terms draw it only when the lines are carrying data, and 
then only in proportion to the logic 0's on the buss at that instant.

>All the flavors of SCSI, the spec is something to get lost in for
>sure. Know it all? Nope, just enough for the CoCo. :)
>
>Mark
>Cloud-9

And thats my contribution to the knowledge pool vis-a-vis the care and 
feeding of a scsi bus.

To a lot of folks its voodoo, to someone familiar with transmission 
lines, which both the scsi and the atapi/ide busses are, the 
explanation, and the fixes to make it work, makes sense.

Relatively unk is that in the case of ide/atapi drives, the one jumpered 
as master also turns the cable terms on, and that means it should be on 
the end of the cable, with the slave drive in the middle.  Its not 
always happy if the master is on the center connector, and the slave, 
or nothing, is on the end of the cable.  Thats asking for trouble with 
a big kick me sign taped on your back.

[...]

-- 
Cheers, Gene
People having trouble with vz bouncing email to me should add the word
'online' between the 'verizon', and the dot which bypasses vz's
stupid bounce rules.  I do use spamassassin too. :-)
Yahoo.com and AOL/TW attorneys please note, additions to the above
message by Gene Heskett are:
Copyright 2006 by Maurice Eugene Heskett, all rights reserved.