Date:       11 May 90                                           X3T9.2/90-77R0
From:       Kurt Chan, X3T9.2 Principal, Hewlett-Packard
To:         X3T9.2 Membership
Subject:    Single-ended crosstalk measurement proposal

Although I swore to abstain from further cable testing, it wasn't a very 
solemn vow, and it now appears more answers are needed to attempt to correlate 
shielded cable crosstalk to the system tests performed by Sun Microsystems.

Some incentive to reexamine these tests came from the realization that my 
previous system crosstalk test was suboptimal. I chose to measure the time 
required for the cables to propagate a data pattern and match exactly the 
transmitted pattern, knowing that crosstalk would manifest itself in the worst 
case as data mismatches and therefore delay in the comparison. One problem 
with this approach was that the differences between cables were very small: 2-
3 nanoseconds in the presence of a 50-90ns baseline (the propagation delay of 
the cable plus cable reflections, plus comparator/transceiver delay). Thus 
slight differences in cable propagation delay would mask the delay effects of 
crosstalk, making the measurement a very sensitive one.

To more accurately determine the effect of adjacent signal transitions on 
quiescent lines, a different test is needed. 

OBJECTIVES

1. Reproduce as closely as possible the environment under which the cables 
   will actually be    used: open collector drivers and appropriate 
   termination, multiple active pairs acting upon    a quiescent pair.

2. Keep the procedure fast, simple and repeatable by minimizing the number of 
   measurements    and avoiding the use of sophisticated test equipment.

3. Make the test results independent pair position within the cable.  To avoid 
   dismantling    cable samples which have already been terminated in molded 
   assemblies, a symmetrical test    is needed which does not rely on knowing 
   which pairs are adjacent to the pair under test.

4. Make the procedure extensible to other environments (such as differential 
   transceivers, or    P-cables).

TEST SETUP

25 open-collector drivers being driven from the same source will continuously 
output an alternating 0..0-1F..F pattern; the frequency is controlled by an 
external pulse generator. 25 switches enable the drivers onto the cable pairs. 
The near end terminators, cable under test, and far end terminators are all on 
the other side of the switches. One conductor of each pair is grounded along 
with the cable shield. All 25 pairs are terminated; TERMPWR is provided on the 
test fixture.

CROSSTALK MEASUREMENT

With 24 switches closed,  a peak voltage measurement made on the quiescent 
line using a digitizing oscilloscope. The crosstalk effect of all other 24 
lines on the open line will be measured.

The results will certainly be worst-case, since there are only 18 (not 25) 
SCSI signals and only about half of those 18 will change at any one time  
under normal circumstances. Nevertheless, I expect these measurements to 
correlate with parametric measurement methods such as those proposed by Jim 
Fiala of 3M.

Ultimately, I would like to see us be able to provide implementors with a 
recommended termination algorithm for each SCSI pair given the pair impedance, 
relative position in the cable, and crosstalk susceptibility of the pair 
(using whatever empirical or parametric crosstalk measurements we can agree 
upon).

I will arrange to have the original cable samples sent back to me, hopefully 
before the Wichita plenary.