Agilent 16900/16950 Logic Analyzer: Multi Card Modules - what are they for?

[ingowien]

My 16903A contains two 16950 modules. Probably not the best choice for a hobbyist as they are differential and pods are expensive and I'd better had 16750 or 16752 but they were in the machine. If someone has excessive pods for a reasonable price or knows a good source, please let me know.

The service manual shows that these cards can be installed as standalone cards (as they are in my machine) or as multi-card modules if they are connected with some cables that I obviously don't have. As far as I understand the same stacking to multi-card modules is possible for 1695x and 1675x cards as long as the card type is the same except for A/B models. However the manual doesn't really explain what the benefit of multi-card modules is (or I couldn't find it). I assume two things will happen:
- the combined cards will show as one module in the software
- there will be better trigger or time sync capabilities
Is this true? Or will the cards still show as individual slots A/B/C?

Can somebody explain the details to me? I have to admit that I haven't really checked what the limits in envirmonments with more than one card are. If I do store-qualified timing measurements as I'm probing relatively slow systems and differential storage probably provides maximum storage for my task, do I need cards to be in a multi-card module to get (relatively) exact time stamps? Or can I start acquisition on both cards at the same time without grouping them and get the same result? Is the same true for triggers? As far as I understand I can trigger a second card from a first one. Can I only arm the second card for a trigger or can I fire acquisition on the second card immediately from a trigger in the first card?

Or are multi-card modules only required if the trigger conditions (i.e. the sigals to be evaluated) span across more than one card? Or can I trigger on slot A.0 && slot B.0 even if both are single-cards modules?

What is the exact benefit of multi-card modules, what are workarounds and what is just nice to have. And are there disadvantages? I assume you can't trigger the cards individually if in a multi-card module, is this correct?

As I surprise I found that my modules would be upgradable to 64M (though the 16M or 32M in half channel mode isn't bad). I assume that's no longer possible as the analyzer is out of service and wouldn't be affordable. Or do you know if some easy modification has become known as for many other measurement devices? Do all 16950 cards have the 64M on board or are there different versions?

best regards
Ingo

[DaJMasta]

I don't have my experience using it, but my unit has two pattern generator boards grouped as a module.  They appear as a single entry in the UI of the analyzer, and they effectively act as a single module with twice the number of outputs.

My assumption is that synchronization between modules requires them to be joined - they can be run at the same time without it, but, for example, could not use the same state clock unless externally attached to both, but there may be software benefits as well.  A multi card module, since it's treated as one element, is probably easier to assign all of the inputs to a single bus for capture or analysis or export or whatnot.  It could be that you can assign elements of the bus between multiple modules that aren't connected, but again since the synchronization may be required, it may not be possible.


The big disadvantage with the 16950 cards are that they use 90 pin pods, which only were around for a generation or so and are much more expensive to find on the used market.  I believe they are differential until the connector, effectively, and then are single ended from there (though pairs of these can still operate in differential modes, I think it's 34 single ended or 17 differential from the pod itself on a single 90 pin connector).  That said, the 16950 cards are also much faster than the 16700 series cards, with a 600MHz base state analysis speed.  There are some 169xx series cards that offer even faster, up to 2GHz state measurements, but the cards beyond the 16950s have a minimum speed of 20MHz or so, so at least with the 16950s you can go really far down for legacy work.

I had just figured they populated the right number of RAM chips for the configuration, but it could be that they're software or hardware unlockable.  I do know that if you make a multi-card module by connecting cards internally, it will operate at the memory capacity of the lowest card.

[MarkL]

If you're looking to buy 16752 cards, you might be interested in this thread to convert 1674x and 1675x cards to 16752:

  https://www.eevblog.com/forum/testgear/hpagilent-1675x-logic-analyzer-card-memory-up-hack/msg1307240/#msg1307240

I don't think there are any other mechanisms for a 16903A to determine the 1675x card type, so it should work on that chassis too.  The 16950 modules capture depth is license controlled, as you've discovered, and I haven't heard of anyone "upping" them.

(Hint: Don't buy old 167xx cards "As-is".  They have aging issues.  Make sure you buy them as returnable/refundable if they don't work.)


In terms of the the behavior of multi-card sets, I can only speak for the 16700 series with 1675x series cards.

When you combine the cards, they operate as a single card with one being the master of the others in the set.  The trigger configuration (such as looking for specific words) can span all the inputs of the cards in the set.  You cannot trigger the individual cards in a set separately.

Here's one non-obvious thing.  When operating in state or transitional mode, you can choose to halve the capture memory or set aside a pod pair for time tags.  So, if you want to keep max capture memory, you need to lose one side of one card.  For this reason, I usually install the cards in groups of 2 or more to preserve the capture depth and still have decent number of input channels.  That's one reason for combining cards.

If you don't combine the cards, you can trigger/arm one card from another by using the IMB (intermodule bus).  But the trigger setup and data capture is completely separate, which can be inconvenient depending on the width of your buses.  You can't have an "and" trigger across separate cards on the 16700, and I think the same for the 16900 according to my reading the manual.

According to the specs, at least for the 16700 series, the time correlation when using the IMB is <=2.0ns if you're doing separate triggering.  If you're working on old/slow systems, it's probably not a factor.

[ingowien]

Thank you for the clarification. Using only 1/half of the modules to double memory was something I was planning to do, too. Yes, 2ns are not an issue for many situations, e.g. when evaluating some hardware where you need to analyze/decode the CPU bus but only basically need to check when some other signals start/stop/change or show a given behaviour. In such situations it's just a matter of having enough channels, not ns accuracy.

Thanks for pointing to the 16752 article, I've seen this some time ago. And I've already considered getting a 16750.
Regarding the aging issues: Do you know if these detected in the self test of are there special checks to be applied?

[MarkL]

Thank you for the clarification. Using only 1/half of the modules to double memory was something I was planning to do, too.
...

To be clear, I wasn't referring to half-channel mode.  If, for example, you wanted to correlate transition times across two separate cards that are capturing in state mode, you'll need to use time tags.  Turning on time tags has an impact on either memory depth or number of usable input pods (your choice).  You might want to read up on time tags.

Thanks for pointing to the 16752 article, I've seen this some time ago. And I've already considered getting a 16750.
Regarding the aging issues: Do you know if these detected in the self test of are there special checks to be applied?

All the problems I've encountered with aging, namely the corrosion issue, have been detected in self-test, and in some of the scope cards with self-calibration.

I purchased one set of cards that I found were obviously yanked out by their ribbon cables, and it severed the center conductor in some of the signal wires.  So, I'd recommend also testing each signal and clock input to the tip of the flying lead (or whatever probe you're using) to make sure you can see it on the analyzer.

And speaking of corrosion, I'd also recommend removing the plastic runners from the bottom of all the cards and cleaning the bottoms thoroughly with a brush and IPA.  I've had several cards go bad in my possession because I didn't do this.  The adhesive and foam tape can get very hard and brittle over time.  A heat gun will soften the adhesive and make it easier to get the runners off and/or remove any remaining adhesive.  Just make sure you get it all off.

[ingowien]

... I usually install the cards in groups of 2 or more to preserve the capture depth and still have decent number of input channels.

Now I understand, thanks for pointing it out. The 16752 manual states that "Time tags are stored in either a dedicated tag chip, if a spare pod exists, or are interleaved..."
Is it a pod or a pod pair? Or does it depend on the exact configuration or card? Or is the 16752 just 2 pods? I'm sure I've read this but the documentation is spread across so many files...
The 16950 consists of 4 pods, 16 bits each. The manual references pod pairs there:
"state options, 600MHz mode: A warning will appear stating that one Pod pair is required for timetag storage."
"Reserve Pods 1 and 2 for timetag storage, and assign Pods 3 and 4 to My 16950A."

So obviously if a pod pair (32 bits, that would make sense) is required for time tags, than it's more efficient to use 32 bits from 2x64 channels for time tags leaving 3x32 channels per memory word than in a single module configuration where only 1x32 channels would remain. This is indeed a huge point for stacks of 2 cards.

Concerning the multi-cards modules: It seems to be hard to get those 40 pin or 50 pin cables, there's one pair offered for $75 + $45 shipping, doesn't seem like a good deal to me. Do you think it would make sense to get some, just in case they might be required in some special situation? Or do you know of any inexpensive sources for such parts? As for any kind of test gear, the accessories almost always get lost. I'm sure my analyzer had an accessory bag attached but it was removed some time. And often people don't know which cables or adapters are expensive and which are the cheap ones that can be dropped. It might be easier to get a 32M/64M 1675x with twice the storage than 2 of these 50pin connector cables for my 16950.

And speaking of corrosion, I'd also recommend removing the plastic runners from the bottom of all the cards and cleaning the bottoms thoroughly with a brush and IPA.

Thanks a lot for the tip with the runners, I've never heard of something like this. Should they then be put on again after cleaning or not? I assume they are for mechanical protection, correct?

[MarkL]

...
Is it a pod or a pod pair? Or does it depend on the exact configuration or card? Or is the 16752 just 2 pods? I'm sure I've read this but the documentation is spread across so many files...

It's a pod pair on the 1675x, so it's 2x (16ch + clk).  Basically half the card.

The 16950 consists of 4 pods, 16 bits each. The manual references pod pairs there:
"state options, 600MHz mode: A warning will appear stating that one Pod pair is required for timetag storage."
"Reserve Pods 1 and 2 for timetag storage, and assign Pods 3 and 4 to My 16950A."

Sounds like the same deal for the 16950 and 1675x.

So obviously if a pod pair (32 bits, that would make sense) is required for time tags, than it's more efficient to use 32 bits from 2x64 channels for time tags leaving 3x32 channels per memory word than in a single module configuration where only 1x32 channels would remain. This is indeed a huge point for stacks of 2 cards.

Exactly.

Concerning the multi-cards modules: It seems to be hard to get those 40 pin or 50 pin cables, there's one pair offered for $75 + $45 shipping, doesn't seem like a good deal to me. Do you think it would make sense to get some, just in case they might be required in some special situation? Or do you know of any inexpensive sources for such parts? As for any kind of test gear, the accessories almost always get lost. I'm sure my analyzer had an accessory bag attached but it was removed some time. And often people don't know which cables or adapters are expensive and which are the cheap ones that can be dropped. It might be easier to get a 32M/64M 1675x with twice the storage than 2 of these 50pin connector cables for my 16950.

The 1675x cables are 40 pins (part #16715-61601), and I wouldn't spend that much for just the cable.  Often you can get the cables with the card, or if you just want the cables sometimes you can find a dead card (16715/6/7/8/9, 16740/1/2, 16750/1/2) with cables for cheap.

And almost all the time the flying leads are sold separately.  E5383A, 54838-61608, or 01650-61608 are all 40-pin compatible.

I usually end up fishing around on ebay.  Those with "Best Offer" can sometimes result in some good deals.  Some sellers post a high price, but they also know these old logic analyzers and accessories don't sell well and they'll take a low bid just to unload it.  Especially dead cards.  There's no schematics or repair information so they have little value.

The 16950A is a 90-pin differential cable, and you're going to end up paying a fortune for flying lead probes and other 90-pin probing solutions.  It's nice being able to do differential and higher speeds, but if you don't need it I wouldn't invest in it.

And speaking of corrosion, I'd also recommend removing the plastic runners from the bottom of all the cards and cleaning the bottoms thoroughly with a brush and IPA.

Thanks a lot for the tip with the runners, I've never heard of something like this. Should they then be put on again after cleaning or not? I assume they are for mechanical protection, correct?

The corrosion is a very common problem if you search around for it.  Some people even claim a conspiracy of "planned obsolescence' it's so pervasive.

You're right it's for mechanical protection when sliding the cards in and out.  I haven't put the runners back on my cards and I'll probably be sorry that I didn't.  But I don't move around cards very often and I'd figure I'd take my chances with being careful over applying another potentially destructive adhesive.

One way to be nice to the cards is to put them all in and pull them all out as a group.  It's easier to control the sliding against each other since the boards flex and fan out a little when they're sticking out.

[ingowien]

May I ask some more questions concerning the corrosion issues?
I'm not quite comfortable with the idea to partially disassemble these boards, yet. I definitely trust your experience but I'm not that confident I wouldn't damage more than this could help or prevent from happening. I'll definitely follow your suggestion with some heat and use some mild plastic tools and nothing else. And no force that could suddenly slip and cause damage.
I now found your other contributions and the pictures explaining the problem, thanks a lot.

So your assumption was that it could be the glue, correct? It seems they used some kind of thin double sided foam tape to mount the runners. Would you recommend to put some of these back on, maybe with hot glue? Or could it be the spacers itself or moisture beneath that causes corrosion and it would be best to remove them completely? I think I have to get used to the idea that not all HP/Agilent decisions were the best. I might be tempted to add a full-size sheet of thin FR4 or something similar but I think this could lead to RAM overheating. On the other hand it's probably overkill if one is careful.

Did you see if Agilent changed the runners or the way they have been mounted at a later time?

I extracted my 16750, it's 16750-66504 but I didn't find a date code. First I tought I'd see some signs of corrosion and I tried to remove the railer on the lower right side with the least traces under it.(next to the backplane connector). It seems that there's no corrosion, it's just slight discolorization of what seems to be a relatively solid layer of glue between the board and the foam. The board looks good and the tin on the vias doesn't show any sign of corrosion. But I'm not familiar with this type of corrosion, I only know that from battery acid as I had in my prehistoric 8673B where the solder mask was eaten away and the tin had this really ugly green to white discolorization. The only thing that withstood the acid were the gold plated traces.

Do you know if this issue was addressed at some time or are all 16750 affected? Unfortunatley I couldn't find something like a date code.

But you're probably right, it will be better to get all of them removed before chemistry hits the board on some unlucky day.

Another question: I always hate to remove those cards from a system. They are so hard to get out and I don't trust those plastic levers. They look like the levers HP used on their old modules that almost always break when you just look at them. Can the 16800/16900 levers really be used to get a card out or is it better to use a different lever or mechanism to get them out? I'm sure you shouldn't use the 16750 connectors though they look solid and the 16950 doesn't even have exposed connectors. So is there a recommended method putting the lowest possible strain on all the components?

[MarkL]

So your assumption was that it could be the glue, correct?

Yes.  The corrosion issues are always around or under the runners.  I suppose it could also be the foam material itself, but I'm assuming it's the glue since it's in direct contact with the board and that's where it's the worst.

It seems they used some kind of thin double sided foam tape to mount the runners. Would you recommend to put some of these back on, maybe with hot glue? Or could it be the spacers itself or moisture beneath that causes corrosion and it would be best to remove them completely?

Hot glue may not be a bad idea.  Maybe a few dots of it per runner; I wouldn't slather it on.  It's entirely up to you if you want to put them back on.  It would certainly be mechanically safer for the board.

Did you see if Agilent changed the runners or the way they have been mounted at a later time?

On the cards I have, they are (were) all mounted the same way with foam tape.  But they are old cards, dating from early 2000s, that will run in a 16700 series chassis.  I don't have any recent ones to compare.

Do you know if this issue was addressed at some time or are all 16750 affected? Unfortunatley I couldn't find something like a date code.

You can look at some of the chips to figure out the manufacture year.  Many of them will have a date code on them.

I don't know if the problem has been addressed in later versions of the 16750.  I can only say of the cards I have, all of them have shown some level of corrosion: 16715/7, 16740, 16750/1/2, 16753/4/5, and 16534, in a total of about 25 cards.  And of those 25, 10 are corroded beyond repair.  I tried to repair most of them.

Actually, I do have one card, a 16720 pattern generator that is spotless and works perfectly, and guess what - it has no runners on the bottom.  (There are no components on the bottom.)

Another question: I always hate to remove those cards from a system. They are so hard to get out and I don't trust those plastic levers.
...

I don't know since I don't have a 16800/16900 chassis.  Maybe someone else can answer.

The 16700 series uses knurled thumb screws which are rather annoying.  It sounds like they tried to fix that with ejection levers.  I think HP/Agilent could have done a better job with the mechanical design on these analyzers.

[ingowien]

I decided to take them all off and you're right, there is corrosion. It's very light and only on maybe 10 vias that aren't so shiny as they should be. Solder mask seems to be intact and I hope the traces are, too. The card didn't have issues before and I hope it won't have when I re-install it. The date codes I found are from 2000 so it seems to be as old as your cards. Maybe I'm just lucky or the analyzer was in a dry or cold area or just wasn't turned on that long. But I'm sure the corrosion would get worse and you've saved my analyzer, thanks a lot.

The runners are made of some clear plastic. As you don't have to be careful once they are off, l it's not that hard to get the foam tape off. The condition was varying a lot, some tapes still kind of soft where others were really hard. I think it shouldn't be a problem to clean them though IPA isn't really the best cleaner for that kind of stuff. Is that kind of plastic probably isn't the problem I'll probably put them back on once it's clear that the card runs with 32M.

I took the two 16950 out, those were really stuck in the chassis. Had to use two screwdrivers, one left one right to gently lever them out. Once you get used to the feeling when it moves for the first time it's OK but yes, definitely not the best mechanical solution they created. I definitely wouldn't dare to use the plastic levers to get them out. The runners are smaller on that boards but similar.

I think the thumb screws are the same. Unfortunately the're too short to push the card out and on one side two of the thread inlays in the chassis are gone, so they are pretty useless. Especially useless for disassembly.

So again thanks a lot for all your knowledge and tips, this helped a lot in understanding the analyzer and I think it definitely has saved it from corrosion death.

[MarkL]

If you find a good solvent to get the adhesive off the runners without attacking the plastic, please let us know!

I have a bag of them now and if I ever decide to put them back on I'll need to clean them off first.

[philpem]

Heating the runners up to about 100C-150C with a hot-air solder station then scraping the foam tape off worked for me. Followed by IPA afterwards to remove the glue.

For the PCB, 50/50 isopropyl+acetone mix should get the crud off from where the tape was. Check with a magnifier/microscope to see if any of the pads need cleanup or repair.