[Coco] Multipak redesign/replacement

[Gene Heskett]

On Tuesday 24 February 2015 21:02:12 RETRO Innovations wrote:
> On 2/24/2015 12:19 AM, Al Hartman wrote:
> > If you check the May 1985 Issue of Rainbow Magazine, page 222 has a
> > review of the Spectrum Projects Multi-Pak Extender Cable.
> >
> > I can't believe Bob Rosen could make something in 1985 that can't be
> > made today. I have one in a box somewhere, and it worked well.
> >
> > 24 inches long, and shielded cable.
> >
> > -[ Al ]-
>
> I can't speak to his product offering, but I am not prepared to create
> something that causes people issues.  I appreciate that certain lengths
> or specific brands of flat cable might do OK on 1.77MHz signals, but
> such cable is not designed for such speeds and should not be used for
> such speeds.
>
> Jim

Au contaire;

Transmission line class 101 is in session:

Such cable is 100% useful at mind boggling speeds.  All we have to do is 
follow best practices for a scsi interface.  That means the cable is driven 
at the end sourcing the signal by a driver whose source impedance matches the 
cable when that cable is looked at as a transmission line, AND that it be 
terminated at the other end with a load impedance that matches this 
transmisson lines working impedance.

For the typical flat ribbon cable, the interwire impedance is approximately 
120 ohms according to the cable pedlars.

So how is this done in the real world for a scsi-II cable, which the specs as 
published say can be run as far as 39 meters?  The simplest way to visualize 
it is with an "active termination" chip on both ends, which is powered by the 
5 volt buss and gives a very low impedance point at its output, holding that 
output at 3.3 volts, at both ends of the cable, which is then connected to 
the individual wires of the cable with a 120 ohm resistor.  This is done at 
both ends of the cable, and only at the ends.  So a signal can be injected at 
any point on that 39 meters of cable, with a 5 nanosecond fall time by an 
open collector transistor stage pulling  that point of the cable to ground.

This "falling edge" then propagates from where ever it originates, to the ends 
of the cable, where it sees this 120 ohm termination resistor.  The resistor 
absorbs the signal by std ohms law theory, and since the cable impedance and 
the terminating resistor are a close match, very little signal is reflected 
from the ends of the cable which could add or subtract from the signal at any 
point on the cable.  If they do not match, then there will be a reflection 
which will bounce back and forth on the cable until it is all absorbed.

You can hook the open collector driver transistor to the cable at any 
arbitrary point, and it will work flawlessly at scsi-II speeds, possibly as 
high as 80 megabytes a second flowing on this cable with no errors.

The secret of the no error operation is the closeness of the termination vs 
the cables impedance.

When it was first brought forth for use as a storage media interface, the bean 
counters didn't like the active termination because it costs money, so only 
the top end drives and interfaces entertained it.  And you paid a premium for 
such card s and drives, typically another 50 to 100 dollars.

What saw the production lines & was stuffed into the boxes to us used a 
resistor pack, usually setup to handle 7 lines per pack, hence 3 packs were 
used to cover the active lines of the scsi-II interface.  These resistor 
packs had a ground pin, and a pin intended to be connected to the 5 volt bus, 
with a 220 ohm resistor in series from the 5 volt connection, with a 330 ohm 
resistor tied to the ground pin, so the junction of the two resistors was 
then held at 3.0 volts.  It took 3 of them and the power draw to power them 
was, and I need to do some math here since we have the equivalent of 21 each 
550 ohm resistors in parallel, each such pair across the power rails then 
drew 5/550, *21, *5=0.954545454545 per pack. Or nearly 3 watts of power as 
heat at both  ends.

Termination wise that resistor combo looked like (using the classic formula 
for paralleled resistors, was a 132 ohm resistor, not a perfect match, but 
close enough that it did work at reasonable cable lengths.

Then it was found that in order to prevent everything on the buss from being 
powered by the 5 volt supply with the highest voltage, even with the device 
on the far end turned off, that there needs to be a very low drop diode to 
feed this 5 volts to the terminators, while preventing that same 5 volts from 
re-entering a device that was powered down, so the designers then specified a 
high current shotkey diode, with a voltage drop in the forward direction that 
was in the .125 volt range.  But between the designers and the production 
line, some damned bean counter looked at the cost of that shotkey diode, and 
bought a 10 cent std SI power diodes for production to stick in those holes 
in the board.

Bam, thats .75 volts of effective voltage drop for the SI diode, so the logic 
one resting voltage, designed for 3 volts, is now only about 2.65 volts, 
assuming the psu is good.  If its starting to sag in its old age, and only 
making 4.90 volts the bus itself is down to 2.5 and change.  This is TTL 
logic levels, where a solid logic one must be at least 2.4 volts, no 
guarantees are made betwen about .8 volts and 2.4 volts, its officially 
declared indeterminate.  Check any chip book.

So in the real world occupied by us, at the mercy of those bean counters who 
don't know squat about semiconductors, scsi-II got a reputation for being a 
cast iron bitch that only worked when it wanted to.  But it wasn't the bus 
design, it was the bean counters that made it that way.  I long since lost 
track of the number of scsi cards in amiga's that died with bus errors, but 
that between my replacing the 10 cent Si diodes with $2 shotkeys, and a tired 
psu with a good one ,aking a 5 volt buss at 5.05 or better, and it became 
bulletproof.  No more errors until the drive died.

All because its logic one noise margins of about .6 volts had been given back 
to it.

So what I am saying is that it can be done, and done very dependably, if its 
done right.  The IDE cable had an 18" max length, and it was so sloppily 
terminated that I have also cured several PC's making data mistakes by the 
simple expedient of cutting off the end connectors and cutting off whatever 
amount of cable I could that still allowed it to reach the drives and 
reinstalling the connectors.

Because the master/slave jumper also turns the terminations off when in the 
slave position, it is imperative that the drive called master is the drive on 
the far end of the cable, and the slave is then attached somewhere in the 
middle.

Class dismissed, says great grandpa Gene.

Cheers, Gene Heskett
-- 
"There are four boxes to be used in defense of liberty:
 soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
Genes Web page <http://geneslinuxbox.net:6309/gene>