GPIB plugging problems - SOLVED

[EC8010]

Should be easy enough? Yet I've had trouble with the screws. So I looked up "GPIB connector" on Wikipedia to see what thread is used, and reference #21 took me to a Hewlett-Packard document that states on page 28 that:

Metric threads (ISO M3.9 x 0.6 type) are specified. Metric fasteners are typically coloured black. Some existing cables use English threads (6-32UNK). They are coloured silver. DO NOT ATTEMPT TO MATE ENGLISH AND METRIC FASTENERS, as damage to hardware may result.

Ah ha! Now I know why I had trouble. Some of my cables have black fasteners, some silver. But it gets worse, far worse. Firstly, there is no such thing as M3.9 x 0.6, and M3.9 would be much bigger than any GPIB screw I've seen. I think HP meant M3.5 x 0.6 which is commonly used for electrical stuff. Next, there is no such thing as 6-32UNK. But there is such a thing as #6-32 UNC. It's an American thread, by the way. The #6 refers to diameter, the 32 its pitch in TPI. A #6-32 screw is too small for the GPIB screws, and the pitch is too coarse. But #5-40 seems to match GPIB screws both for diameter and pitch. And guess what, 40TPI corresponds to a metric pitch of 0.635mm, and a #5 screw is 1/8" diameter, or 3.175mm. So M3.5 x 0.6 is sufficiently similar to #5-40 UNC that you could force one into the other and cause damage.

I ran an M3.5 x 0.6 tap into the heads of my (black) screws and varying amounts of swarf came out. They should have been M3.5, but I suspect they were #5-40 UNC (described elsewhere as an obsolete thread, but apparently alive and kicking). I've ordered a #5-40 UNC tap and die set.

To sum up: Don't believe that HP document. Any GPIB screw you see is probably #5-40, even if it's black. But it might be M3.5 x 0.6. No wonder we have trouble!

[tooki]

Should be easy enough? Yet I've had trouble with the screws. So I looked up "GPIB connector" on Wikipedia to see what thread is used, and reference #21 took me to a Hewlett-Packard document that states on page 28 that:

Metric threads (ISO M3.9 x 0.6 type) are specified. Metric fasteners are typically coloured black. Some existing cables use English threads (6-32UNK). They are coloured silver. DO NOT ATTEMPT TO MATE ENGLISH AND METRIC FASTENERS, as damage to hardware may result.

I think you mean reference #20, not #21. But it’s much smarter to just post a link to the reference itself, since wiki articles change over time.

http://bitsavers.org/pdf/hp/hpib/TutorialDescrOfHPIB.pdf

And here’s a different edition of the same document, with grayscale images, etc:
https://www.hpmemoryproject.org/an/pdf/hp-ib_tutorial_1980.pdf

Ah ha! Now I know why I had trouble. Some of my cables have black fasteners, some silver. But it gets worse, far worse. Firstly, there is no such thing as M3.9 x 0.6, and M3.9 would be much bigger than any GPIB screw I've seen. I think HP meant M3.5 x 0.6 which is commonly used for electrical stuff.

There seems to be wide agreement that the screw is M3.5x0.6, which does suggest that M3.9x0.6 is a typo. (Probably confused with M3.9 self-tapping screws.)

FYI, M3.5 is no longer widely used, and in fact is very hard to find now. They’re only encountered in old stuff. Fastener manufacturers have largely removed them from their stock item catalogs.

Next, there is no such thing as 6-32UNK. But there is such a thing as #6-32 UNC.

Ehhhh, the statement I bolded is incorrect.

Just a little info about the American thread types:

The primary standard is called UN (unified national thread). UNC (UN coarse) and UNF (UN fine) are tables that define particular thread sizes, but because the thread pitch is given in the thread size, a given size can only fit into one of those tables. So a #6-32 UN thread is always UNC, for example, because the #6 UNF thread is #6-40. The thread standard itself is simply “UN”.

UNJ, UNR, and UNK, on the other hand, are not tables of UN thread sizes, but separate standards with slightly different thread geometry to improve performance. UNR specifies that the external thread must have a rounded root, and UNK was just UNR with extra measurements made mandatory. UNJ specifies modified internal and external thread geometry for the highest performance. UNK, apparently, was “obsoleted in 1979” according to the first source below, which has a comprehensive overview:

https://www.ring-plug-thread-gages.com/ti-UNR-vs-UN.htm (this website has a few other interesting pages covering minutiae of the UN thread family and seemingly every other thread family in existence)

https://www.ultrasonic-resonators.org/misc/references/articles/Kanter__'Taking_the_Mystery_Out_of_Screw_Thread_Forms'.pdf

https://jlmetrology.com/screw.html



Anyhow, this means that a male #6-32 screw in the UNK standard would technically be, for example, a 6-32UNKC, and it would mate with a #6-32UNC or #6-32UNJC female thread.

While UNK screws are understandably impossible to find, since they’ve been obsolete for 45 years, UNR screws appear to be readily available, including #6-32UNRC.

It could be a typo in HP’s documentation and that they intended UNC, but chances are it’s not, and they were simply specifying the very best variant at the time, in which case they actually should have written UNKC or UNRC.


Other HP documentation, like this one http://www.bitsavers.org/pdf/hp/98xx/98034-90000_hpibIntf_May79.pdf describe the old units as using “National Coarse” threads, which would be UNC, and thus UN. But that’d be the same fundamental dimensions as UNK.

It's an American thread, by the way. The #6 refers to diameter, the 32 its pitch in TPI. A #6-32 screw is too small for the GPIB screws, and the pitch is too coarse. But #5-40 seems to match GPIB screws both for diameter and pitch. And guess what, 40TPI corresponds to a metric pitch of 0.635mm, and a #5 screw is 1/8" diameter, or 3.175mm. So M3.5 x 0.6 is sufficiently similar to #5-40 UNC that you could force one into the other and cause damage.

I ran an M3.5 x 0.6 tap into the heads of my (black) screws and varying amounts of swarf came out. They should have been M3.5, but I suspect they were #5-40 UNC (described elsewhere as an obsolete thread, but apparently alive and kicking). I've ordered a #5-40 UNC tap and die set.

To sum up: Don't believe that HP document. Any GPIB screw you see is probably #5-40, even if it's black. But it might be M3.5 x 0.6. No wonder we have trouble!

The mounting side of the GPIB connector screw is not specified, but apparently in many, it was 6-32. (See e.g. here.) So it’s not inconceivable that early connectors really did use #6-32 for the mating-side screw, too.

It is highly unlikely to be a #5-40 thread, as the odd-numbered screw sizes (like odd-numbered AWG wire sizes) are exceedingly rare in the wild. It’s much more likely to be M3.5 like the world in general agrees they used, or #6-32 (regardless of whether UNC, UNKC, or UNRC).

But it will be interesting to see what you find once you get your #5-40 tap and die set!

[tooki]

Ok, debate over: here’s an old HP manual stating that the metric versions are M3.5x0.6 and the old versions #6-32 UNC: https://archive.org/details/manualsbase-id-548540/page/n207/mode/2up on page A-6 (PDF page 10)

[edpalmer42]

Part of the problem is that GPIB (General Purpose Interface Bus) started out as HPIB (Hewlett Packard Interface Bus).  It's not surprising that HPIB would specify American threads.  Later, the bus was standardized by the IEEE and things were converted to metric.

Ed

[MarkL]

To quote the GPIB standard, IEEE 488.1-2003:

  ISO metric thread M3.5 x 0.6 or equivalent Optimum Metric Fastener System (OMFS) thread 3.5PO.6

[EC8010]

Thank you to everybody who responded. OK, so the consensus is that the thread ought to be M3.5 x 0.6. Which, as a number of people have pointed out, is very hard to buy. Fastener distributors rarely stock it, nor even electronics distributors. Bizarrely, it is the thread for securing UK domestic mains sockets and light switches in their pattresses. My Prologix GPIB-USB adapter has long enough threads to check properly and they are definitely M3.5.

Specification consensus is also that old versions of GPIB connectors used 6-32 UNC. Which makes sense because HP is an American manufacturer and you'd expect them to use an American thread. So I measured again and the 6-32 screws I have had an outside diameter of 3.07mm, whereas a #5 screw is specified as being 3.175mm, which is somewhat nearer to a measured M3.5 diameter of 3.37mm. I can't help feel that a gorilla would be needed to force an M3.5 screw into a #6-32 UNC stud. Further, the thread pitch is entirely wrong. A #6-32 UNC has a pitch of 32 teeth per inch (TPI), whereas the 0.6mm pitch of an M3.5 screw is 42.33r TPI. The two threads are nowhere near being the same. But a #5-40 UNC (40 TPI) is much closer to M3.5 x 0.6 and could perhaps be forced without noticing when groping awkwardly round the back of an instrument.

And thank you to tooki for that thread reference; never heard of UNK before. But I note that even in the reference it's only a footnote!

But my point is not so much about published standards as practicalities. I assumed that the published standard of M3.5 x 0.6mm was correct, but when I ran an M3.5 plug tap into the heads of the fasteners of my newly manufactured GPIB cables, more swarf than expected came out. (I do a lot of mechanical engineering, hence having a set of M3.5 taps to hand.) I expect a little swarf when cleaning a thread, but this was a little too much. So I don't believe the fasteners in newly manufactured GPIB cables were M3.5 x 0.6. And that's the important point.

I will note that after running the M3.5 x 0.6mm tap into the studs, the new cables plugged up happily.

[tooki]

Could it just be that your M3.5 tap and die set is for a different tolerance class than the M3.5 in the connector? That might explain the swarf.

As for the #5-40: your reasoning is sound, but nonetheless almost certainly wrong, in that there is no reason to believe that these connectors were ever made with #5-40 threads. Your main objection seems to be that “a gorilla” would be needed to cross-thread M3.5 and #6-32. It could be, but probably isn’t as much as expected, especially with cheap, soft screws. But I also don’t doubt for one second that people would do this even if it took significant force! :p

[Hydron]

Were the gpib cables which produced much swarf the cheap eBay ones? Because my experience with them suggests that the female threads are total shit, and might just be that far out of spec.

[dietert1]

When i made some USB-GPIB interfaces about 15 years ago i found somebody who made those stubs M3.5 to M3 and as i didn't know that HP paper, they were made in steel with zinc coating.
Even our oldest HP devices (e.g. a HP 59309A clock) all have those black stubs and i never had a problem fitting the cables. For automation running reliably for a month or so it's necessary to tighten the screws.

Regards, Dieter

[EC8010]

Were the GPIB cables which produced much swarf the cheap eBay ones?

They were indeed. I must admit I had not considered the possibility that the thread was intended to be M3.5 but badly made. I suppose I should have. 

Whatever the problem, I shall run a plug tap into all my female threads. I'm not entirely certain that the male thread is M3.5 (it's quite difficult to positively identify an unknown thread), but it does allow me to gently tighten stacked connectors without risking something seizing. I'm already putting a spanner on the nut on the instrument so that I can feel exactly what I'm doing and avoid the DB9 problem of disturbing the instrument nut.

All this mechanical stuff has come about because I have discovered that GPIB connectors need to be done up reasonably (not gorilla) tightly to work reliably. I have to clamber up on the bench and bash my head against the ceiling to get at GPIB. I don't mind doing it once, but I expect it (once plugged correctly) to work and continue working so that any problems are down to me not sending the right commands.

Another interesting thing about GPIB... In the manual for the GPIB adapter for Tek TDS3000 series oscilloscopes, it says that more than two thirds of the devices on the bus must be powered. Until yesterday, I had eight devices on the bus and have had no trouble just switching on instrument on and talking to it. I now have ten, but haven't yet tested it, and the Tek adapter card hasn't arrived yet. I'm guessing that the early busses might have been genuine TTL but that later are CMOS and this affects electrical loading.

[guenthert]

[..]
Another interesting thing about GPIB... In the manual for the GPIB adapter for Tek TDS3000 series oscilloscopes, it says that more than two thirds of the devices on the bus must be powered. Until yesterday, I had eight devices on the bus and have had no trouble just switching on instrument on and talking to it. I now have ten, but haven't yet tested it, and the Tek adapter card hasn't arrived yet. I'm guessing that the early busses might have been genuine TTL but that later are CMOS and this affects electrical loading.

   Yes, the standard is from a time when instruments were still steam powered.  Not so ancient instruments load the bus considerably less (only way why something like the AR488 can be used with multiple instruments at all), even though they are all supposed to terminate the bus (thanks to the requirements to allow for "star" connections).

[MarkL]

...
Another interesting thing about GPIB... In the manual for the GPIB adapter for Tek TDS3000 series oscilloscopes, it says that more than two thirds of the devices on the bus must be powered. Until yesterday, I had eight devices on the bus and have had no trouble just switching on instrument on and talking to it. I now have ten, but haven't yet tested it, and the Tek adapter card hasn't arrived yet. I'm guessing that the early busses might have been genuine TTL but that later are CMOS and this affects electrical loading.

I have 12 devices on my GPIB bus, including a TDS3054 with the TDS3GM (GPIB/RS232)  module.  There was never any problem talking to the TDS3054 with all the other devices off using a National Instruments PCI-GPIB card.

IEEE 488.1-2003 has this to say about it:

8.2.4 Devices powered off and on
A system will operate without adversely affecting normal data transfer with at least two-thirds of the devices powered on. A system will operate correctly with any number of devices powered off, provided all of those devices powered off do not degrade the specified high state condition, that is, that the voltage on each signal line with all its output drivers passive false should exceed +2.5 V with respect to the logic ground at each device.

Unless special precautions are taken (that is, use of special driver circuits beyond the scope of this standard), powering a device to on while the system is running may cause faulty operation.

[dietert1]

I have been using a HP 59401A for GPIB debugging. It brings out GPIB signals on its back and i used that to check GPIB timing and signal levels with a scope. But with power off it pulls down the bus, so it can cause problems in a small setup with just a controller and one more GPIB device.
GPIB driver chips do exist that don't pull down the bus with power off.

Regards, Dieter

[EC8010]

I was horrified when I saw diagrams of permissible networks that included stars and spurs. I cut my teeth on analogue video where correct termination of transmission lines was essential, so my GPIB bus starts with the Prologix and loops through all the instruments to end at the final instrument where there is a GPIB terminator.

That IEE 488.1 2003 comment is very useful. If things should misbehave, I'll hang a 'scope on the bus.

I looked up HP 59401A on Fleabay. None in the UK, but a number in the US of A ranging from a bit pricy to outrageous. Nice bit of kit, though. Too rich for me.

[pdenisowski]

All this mechanical stuff has come about because I have discovered that GPIB connectors need to be done up reasonably (not gorilla) tightly to work reliably.

I think part of the issue is the sheer bulk of the GPIB connector:  it's easy to think (and it may look as if) it's pushed in all the way when it's not.  And given the size, stiffness, and orientation of the cable, it's often hard to get a nice snug fit without screwing it down.

Another interesting thing about GPIB... In the manual for the GPIB adapter for Tek TDS3000 series oscilloscopes, it says that more than two thirds of the devices on the bus must be powered. Until yesterday, I had eight devices on the bus and have had no trouble just switching on instrument on and talking to it. I now have ten, but haven't yet tested it, and the Tek adapter card hasn't arrived yet. I'm guessing that the early busses might have been genuine TTL but that later are CMOS and this affects electrical loading.

I recently did a YouTube video about GPIB (or "grandpa's interface bus" ... lol) and I debated whether to include the "2/3 powered on" part.  I left it out of the final version for exactly the reason you mentioned: it's not quite as stringent of a "requirement" as it used to be.

[EC8010]

I have been giving the M3.5 issue more thought. It's not exactly hard to cut an M3.5 thread, and it has been pointed out that although a #5-40 thread is quite close to M3.5, it's a highly unlikely thread. The simplest explanation is that the cheap Fleabay cables have correctly cut M3.5 threads but that the subsequent chemical blackening is shoddy (probably done quickly with poorly controlled chemical strength), leading to undersize internal threads and oversize external.

[MarkL]

You might want to stick with genuine HP GPIB cables.  I learned a similar lesson early on with cheap cables.

While it doesn't solve any issues you might be seeing on the equipment side, at least you can have reliable cables.  They are built like a tank.  Available in varying lengths, part numbers 10833A thru 10833D.  Most are available for around USD$20 or less, used.

[unseenninja]

I don't use GPIB cables. I just plug an AR-488, built on the back of the GPIB plug into each device and then plug them all into a USB hub. I've not found a case yet where I actually need to do a parallel poll with multiple devices. I just adapt the bash scripts that control my test to integrate multiple devices into it.

I'm more than happy not to have multiple kilos of uncooperative cable strung between all my devices.

[dietert1]

Yes, USB plugs don't have any screws. RJ45 ethernet plugs are as easy to use as USB, but more reliable and secure.

[EC8010]

Thanks to everybody who contributed to this thread. I'm fairly sure that my problems were caused by shoddy chemical blackening of the screws, which I have solved by running a plug tap to the head of the GPIB screw. By the way, you will find that in order to make taps, they often have a cone that prevents the plug tap from cutting to the bottom of the hole, so I grind those cones off. Whilst grinding, you need to periodically dip the tap in water to avoid heat build up that would draw the temper of the tap.

Obviously, I could have solved the problem by buying better cables. Used GPIB cables may well be $20 in the US of A but postage across the pond is extortionate. For the cost saving, I'm happy to fix poor threads.

[tooki]

By the way, you will find that in order to make taps, they often have a cone that prevents the plug tap from cutting to the bottom of the hole, so I grind those cones off.

The “cone” (I assume you mean the taper at the point) is not a manufacturing artifact, it’s there deliberately on hand taps to guide the tap, because without it, it’d be dangerously easy to tap at an angle.

Machine taps, being intended to be used on drill presses, lathes, etc. are designed with the expectation that they will be mounted flawlessly coaxial to the hole, so they a) dispense with the taper and so can usually cut nearly to the base of the hole, and b) cut with one tool all at once, rather than 3 sequential ones like in hand taps.

[EC8010]

I mean the very short cone with a typical 90 degree included angle seen on smaller taps, not the much gentler taper that leads into the thread. No way is the 90 degree cone going to assist in getting the tap square. I notice that my larger (hand) taps do not have the cone.

I fully understand your point about the need for taps to go in squarely and without wobble as they do so (even more important). I have made a sprung centre for guiding taps in the lathe and a couple of tapping fixtures. Unsurprisingly, threads cut using the fixtures are much better than those cut by holding a tap wrench by hand and hoping it's square. The first version started life as a hand-cranked pillar drill (bought at a car boot sale for £15) and is used for taps up to M6, gripped in ER11 collets. The larger version (on the same stand) takes ER20 collets and can do quite large threads but was made when M10 threads had to be cut at the edge of Paxolin with only 0.2mm clearance between cut thread and edge. Sixteen threads were cut, none with breakthrough, and the job went easily. Obviously, it took some time to make the tools, but they are used a lot and have easily recovered their investment in time.