X3T9.2/90-100 MEMORANDUM -- 26 Jun 1990 TO: John Lohmeyer, Larry Lamers, and all Cable Types FROM: Bill Spence, TI SUBJECT: SCSI Cables Observations and Proposals As a result of Kurt Chan's and Jim Fiala's inputs at the last Cable session, and as a result of a few experiments of my own and additional conversations with cable engineers, I feel that a lot of insight into the cable matter has been achieved. Herebelow is an attempt to summarize these insights and draw some conclusions. 1. Twisted pairs do indeed couple to each other, and the coupling is probably more inductive than capacitive at the impedance levels where we operate. But this coupling is somewhat erratic. Even between two pair twisted in the same direction with the same pitch, pressed into close proximity, the degree of coupling can be changed simply by adjusting the relative longi- tudinal positions, or, equivalently, by rotating one pair relative to the other. 2. The amount of coupling may not be exactly controllable, but it can be reduced and limited to some maximum level by various proprietary design schemes--varying pitch, direction of twist, lay in the cable, buffering (spacing by dielectric fillers or layers), etc. 3. Kurt Chan's crosstalk measurement proposal (90-77 and 90-93) seems to meet our needs very well. It is relatively quick and easy to perform. Although taking a reading on one conductor with all other ungrounded conductors being driven may seem unreal, Jim Fiala's testing suggests that the most of the signal picked up comes from nearby pairs anyway. 4. A more important way to view Kurt's results may be in terms of worst-case coupling, rather than average coupling. A well designed cable will have no pairs closely coupled. This can be examined in his "scattergrams" at the end of 90-93. He set them up, I believe, to show how close any signal comes to the possbile transition voltage of a receiver. Set up dif- ferently, they might illustrate the point more sharply, but even so the results can be seen. In this concept, what is desired is that the lowest point for any cable be as high as possible. Viewed this way, the cable identified only as Astro 28/30AWG and the Madison 4099 come out best, with lowest points of 2.28 v--still a far cry from flat ribbon cable at 2.48 v, but offering significantly greater safety margin than some of the others. As Kurt emphasized, however, in all cables, how the pairs are assigned appears to be the critical element. Rather than proceed immediately to a decision about what to put in the Stan- dard, I suggest that we try to draw up guidelines for choosing a good SCSI cable. If we can accomplish that, we probably can extract the essence that properly would be included in the Standard. Here are some points I think might provide guidance. 5. We should try to stick to performance criteria and avoid as much as prac- tical going into construction details. This leaves room for inventive- ness to come up with better solutions in the future. Granted that many users will be dependent on the vendor to say what performance criteria his cable meets; it is not uncommon to be dependent on vendor specs. 6. We should consider working out guidelines both for cables and for cable assemblies. We shouldn't try to nail the best cable-pair/connector pin scheme; we should leave room for inventiveness to come up with superior schemes. (But I have no quarrel with our recognizing in some fashion any good ideas we come upon. I am excited by the prospect of surrounding the REQ and ACK pairs with a layer of relatively inactive pairs, leaving the data and other active pairs in the outer layer--preferably with a buffer film of low dielectric constant between the outer layer and the shield.) 7. We should try to work out guidelines for how both cables and cable assemblies should perform when applied to single-ended service and to differential service. There is an obvious advantage to a SCSI cable or cable assembly which will perform well in both types of service, but we should leave an opening for someone to offer a superior cable for single- ended service only. It may be that the guidelines for differential service can be minimal, if the experience of those using differential so indicates. 8. My proposed starting point for such guidelines is what is in the Standard now, soft as it is. I would leave the impedance requirement soft, I would add a soft crosstalk limit (specifying it individually by function in the cable assemblies), probably establish an upper limit on line loss, and possibly something on ratio of max to min propagation velocity. Needless to say, all of this is open to improvement. It may fall short of ensuring that all SCSI cables work in all circumstances while maintaining minimal pricing (we ain't going to get that), but it would provide a good framework in which a cable or cable assembly buyer could evaluate and compare market offerings.