HP/Agilent 1675x logic analyzer card memory up-hack

[MarkL]

In the course of repairing some 1675x cards, I discovered that HP/Agilent is using resistors to set the model ID, and hence the capture memory size on these old logic analyzer cards.  The main clues were:

- All models already have the full complement of memory for their series,
- There was sloppy soldering work (solder splatter, flux residue) on unpopulated resistor pads that were in an obvious grouping (indicating that some had been removed after manufacture),
- The only indication of the model number was a solitary adhesive sticker,
- And finally, HP/Agilent is known to use ID resistors in other products.

I haven't seen this info anywhere else, so I'm posting here.

I've verified the following ID conversions work (with full memory testing):

  16751A --> 16752A
  16754A --> 16756A
  16755A --> 16756A

Specs and other details for the cards discussed in this post:

  http://literature.cdn.keysight.com/litweb/pdf/5968-9661E.pdf

----------

Series: 16750A, 16751A, 16752A
Memory: 34 x Micron 48LC4M16A2-8E
Max model: 16752A with 32MSa

There are nine ID resistor footprints arranged in a column: R735, R65, R66, R68(?), R63, R73, unlabeled, R78, and R82.  They are near the pair of Altera LQFP's near the backplane connector.  (It's possible R735 does something else besides ID, but it's unpopulated and near the others so I'm including it.)  To make a 16752A, all the resistors are populated except R735, R66, R63, and R82.  Resistors are 1k 0805.



There are also "B" versions of each of these cards.  The (bad) photos I can find look like they have the same layout, so I would expect it to work with them also with possibly different resistor labels or combinations (maybe one controls the A vs. B?).

----------

Series: 16753A, 16754A, 16755A, 16756A
Memory: 18 x Micron 46V16M16-75
Max model: 16756A with 64Msa

There are eight ID resistor footprints arranged in a matrix: two rows A and B, and four columns 1, 2, 3, and 4.  They are near the Xilinx FPGA U128, which is near the backplane connector.   To make a 16756A, all resistors are populated except B3 and B4.  Resistors are 1k 0805.



There's also an EEPROM on this card, but it doesn't seem to have anything to do with ID.  It's probably only used for calibration data.

----------

Other Cards (which I don't have so I can't try it)

The 16740A, 16741A, and 16742A from photos look very similar (perhaps identical?) to the 16750A/51A/52A series and with ID resistors in the same PCB area.  Hopefully a 1674x owner can figure this one out since it might need different resistor positions populated.  Please post your results if you try it.

I'll also note that the 1674x and 1675x series cards are identical spec-wise except for memory configurations, occurrence counters, and maximum state speed.  So, I suppose it's possible that a 1674x could be turned into a 1675x by using only ID resistors, but it would require a closer look at the memory configuration on the board and subsequent testing at higher speeds.  Again, something for a 1674x owner to try.


It may also be possible to turn a 16718A (8Msa) into a 16719A (32MSa), but I can't find enough photos to know if the board layouts and memory configuration are the same.  The 16719A at least has a similar board layout to the 16740A/41A/42A and 16750A/51A/52A series, so it probably has ID resistors.


Somewhat related to this topic, I've also heard of a hack to turn a 16533A 250MHz 1Gsa/sec scope card into a 16534A 500MHz 2Gsa/sec scope card via a resistor ID.  If anyone knows the specifics, please feel free to share.  Or if you have a 16533A, post a high-res photo (front and back) and I can probably figure it out from there.

----------

Good luck if you try any of this, and please post your results here if you do!

[gslick]

Interesting information. This is the first time I have seen this. Thanks for sharing.

I'll have to check what types of modules I have. I think have all three of 16750, 16751, and 16752, and at least either a 16718 or 16719, or maybe both. Is the 16717 populated with a different memory configuration? Maybe only a 16741 from that line.

I've never bothered doing a careful examination of differences between the modules myself.

(I also have some of the 16753 - 16756 line that I have never used as I don't have any use for the 90-pin POD connectors).

[MarkL]

Interesting information. This is the first time I have seen this. Thanks for sharing.

No problem; hope it's of some use.

I'll have to check what types of modules I have. I think have all three of 16750, 16751, and 16752, and at least either a 16718 or 16719, or maybe both. Is the 16717 populated with a different memory configuration? Maybe only a 16741 from that line.

I have a 16717 and it's significantly different than the others.  It has 34 x OKI M5416283 (256k x 16).  That comes out to 2Msa (34 * 256k*16 / 68channels), which is the 16717 spec, so there's no additional memory to be enabled.  Same with the 16715.

I'm also noticing there's the 16710A (8k), 16711A (32k), and 16712A (128k) which I missed on the datasheet before.  They also have the same specs except memory, so they may be a candidate too.

There may also be opportunities for cards in the 16500 series, but I have zero experience with them.  I think 16900 series card options are software controlled.

(I also have some of the 16753 - 16756 line that I have never used as I don't have any use for the 90-pin POD connectors).

The 90-pin flying lead probe kits are outrageously expensive, which kind-of kills the attractiveness of those cards compared to the others.  I'm looking at building some single-ended probe breakouts (some day) out of E5378A adapters.  They're pretty cheap on ebay (~ $20), but I'll never get the kind of fidelity needed for the speeds of those cards.

[MarkL]

I've never bothered doing a careful examination of differences between the modules myself.

I don't think it would have occurred to me either.  What tipped me off was reading the service manual for the 16755A I was repairing at the time.  The manual covers all four 16753A/54A/55A/56A cards and talked about the use of 256Mbit DDR SRAMs for the acquisition memory.  My lesser card would presumably not need all the memory installed, yet, it didn't have any unpopulated memory footprints.  Hmmm...  Why is that?

[texaspyro]

You can also upgrade a 16533A scope board to a 16534A board with a resistor mod.  Sorry, I can't find my notes on which resistors to change and no longer have a 16533A to compare to a 16534A. I seem to remember all the needed resistors were on the back.

[alm]

I remember adding up the memory installed on a 16750A once and figuring out is was much more than the specified memory depth. Good job on figuring out the resistor combinations, I could never find sufficiently high resolution pictures of the 16751/16752A cards to figure out the correct resistor combinations.

The 1675xA and 1675xB are the only cases I know where cards with different models are supported in a mixed master/slave configuration (so 16750A and 16750B may be connected through ribbon cables as master/slave, unlike say a 16741A and 16750A. So I expect the 1675xA and 1675xB series to be very similar. Not sure what the exact difference is that triggered changing the model. A slight change in specifications?

The 90-pin flying lead probe kits are outrageously expensive, which kind-of kills the attractiveness of those cards compared to the others.  I'm looking at building some single-ended probe breakouts (some day) out of E5378A adapters.  They're pretty cheap on ebay (~ $20), but I'll never get the kind of fidelity needed for the speeds of those cards.

What's the point of using the 16753+/16760 series if you are not using differential inputs or high speed? The factor two increase in memory depth compared to the 16752A is barely worth the hassle. The various adapters to single-ended Mictor or single-ended or differential Samtec are probably the only affordable solution with decent signal integrity. Many later HP logic analyzer boards have Mictor footprints on them, and you can obviously design in whatever connector you want on your own boards. The 90-pin flying lead pods are more complicated than the older 40-pin pods to improve signal integrity (not surprising, given that the 40-pin pods were designed in the late eighties).

[MarkL]

What's the point of using the 16753+/16760 series if you are not using differential inputs or high speed? The factor two increase in memory depth compared to the 16752A is barely worth the hassle.
...

Not much, I agree.  My mainframe came with a pair of 16751A, and a 16754A.  I never really used the 16754A for anything.  I later came across a broken 16755A for dirt cheap which I was able to fix mostly for the fun and challenge of it.

One additional nice feature in the higher speed cards is the eye scan across all inputs, but again not something I've needed or currently have the appropriate probing for.

No idea on the A vs. B versions.  They look the same from all the photos I've seen, and a diff on the A and B service manuals reveals nothing.

[fisafisa]

aaagh!
as I suspected the 16533 can become 16534....
anybody with high res pics of the back of a 16534?

thanks

[texaspyro]

aaagh!
as I suspected the 16533 can become 16534....
anybody with high res pics of the back of a 16534?

thanks

I think these are the guys:

On 16534A: top side, near the Actel chip:  R810=0  R809=1004
On 16533A:  R810=1004 R809=0

I remember swapping two resistors...  on the 16533A they were hand soldered at the factory.

Also, there are at least two versions of the 16534 (A and A2)... one has a two PGA chips, the other has a single PGA chip and a chip that does not have pins... The different chip is labeled U508 and U509.

[MarkL]

aaagh!
as I suspected the 16533 can become 16534....
anybody with high res pics of the back of a 16534?

thanks

I think these are the guys:

On 16534A: top side, near the Actel chip:  R810=0  R809=1004
On 16533A:  R810=1004 R809=0

I remember swapping two resistors...  on the 16533A they were hand soldered at the factory.

Good memory!  That seems to be the recipe!

I took one of my 16534A cards and swapped R809 and R810.  It became a 16533A.

Using a 0dB reference point at 10Mhz, the 16534A had a -3dB point at 623MHz (not bad - I hadn't measured this before).  When it became a 16533A the -3dB point was reduced to 346MHz.  So, even back then it seems HP was doing their artificial limits on BW.

Someone with a card manufactured as a real 16533A should double check this to make sure there's no other changes needed.

Also, there are at least two versions of the 16534 (A and A2)... one has a two PGA chips, the other has a single PGA chip and a chip that does not have pins... The different chip is labeled U508 and U509.

There's at least three versions that I've seen.  U501 and U509 vary.  U508, the trigger comparator, is either there or not.  The comparator function and some driver circuitry for it around the U200 DAC appears to have been absorbed into U509 in some card versions.

[texaspyro]

Good memory!  That seems to be the recipe!

I took one of my 16534A cards and swapped R809 and R810.  It became a 16533A.

Using a 0dB reference point at 10Mhz, the 16534A had a -3dB point at 623MHz (not bad - I hadn't measured this before).  When it became a 16533A the -3dB point was reduced to 346MHz.  So, even back then it seems HP was doing their artificial limits on BW.

Someone with a card manufactured as a real 16533A should double check this to make sure there's no other changes needed.

I originally found those resistors when, many years ago,  I bought a couple of expansion chassis that had 4 x 16533A's in them.  Last night I dug out a defective 16534A that I tried to back-grade to a 16533A to see if that made a difference (didn't) and spotted the swapped resistors.   I originally thought the resistors were by one of the PGA chips...  That was probably a different board I was remembering.

[fisafisa]

Thanks.
I will try the modification.

[DocBen]

Hi there,

I can confirm that a 16740a can be converted to a 16742a by removing one resistor. I used a picture of a 16742a.
I also tried to convert it to a 16752b but that failed, probably because I only have 1.5k resistors and I think the resistors create a voltage divider so I probably had the wrong voltages.
I will try again once I have some 1k resistors.

[DocBen]

Yatta!

So I have finally gotten around to try to get a 16740a to be a 16752b.

On my third attempt I succeed. (keep in mind however this just passes the self tests. I havent made any actual measurements with it so it might fail when doing actual work)

from R735 to R82:  - -RR-RRR- 
does it. (- open, R well, you know)

I used 1K 0.1% Resistors just like they are used on the board

[MarkL]

Wow, that's interesting!  Of the many possibilities for ID upping, I would have ranked the 16740A --> 16752B as having one of the least chances of actually working.

Does your 16740A have the same amount of memory populated as a 16752B?  The 16752A has 34 x Micron 48LC4M16A2-8E (4M x 16).

[gslick]

Does your 16740A have the same amount of memory populated as a 16752B?  The 16752A has 34 x Micron 48LC4M16A2-8E (4M x 16).

I just took a look at the 16741A that I have. It has (34x) MT48LC4M16A2-75 64Mb (1 Meg x 16 x 4 banks) parts. So if my math is right that should be sufficient for 32M samples of 68 channels, even though the 16740A, 16741A, and 16742A are 1M, 4M, and 16M analyzers.

[MarkL]

Does your 16740A have the same amount of memory populated as a 16752B?  The 16752A has 34 x Micron 48LC4M16A2-8E (4M x 16).

I just took a look at the 16741A that I have. It has (34x) MT48LC4M16A2-75 64Mb (1 Meg x 16 x 4 banks) parts. So if my math is right that should be sufficient for 32M samples of 68 channels, even though the 16740A, 16741A, and 16742A are 1M, 4M, and 16M analyzers.

Your math is right: 34chips * 4M*16/chip = 2176M/68ch = 32M/ch

Plus you have the -75 version which is a 133MHz part and I have the -8E part in a 16751A which is 125MHz, so it would be reasonable to assume the upped 1674x cards can also do the full sample rate of 400Msa/s.  Some verification testing from DocBen (or yourself, if you feel like playing with it), would tell for sure.

That must have been some pretty fine segmentation in the market to make Agilent produce 6 different performance levels on essentially the same hardware.  Go sales team!

[gslick]

There is flux residue around the pads of two of the unpopulated resistors on my 16741A module. I wonder if those resistors were populated during manufactured and initial testing in a maximum configuration and then removed when the module was configured to be a 16741A.

Does someone have a photo of a 16742A for comparison?

[DocBen]

Well it makes a lot of sense to have the exact same hardware, because it effectivly eliminates testing and designing of different products and eliminates the need for early product differentiation.
So you can buy in bulk.

I didnt try to solder in all resistors, but it will probably not be recognized by the analyzer (factory test setting?)

Also if you think about: the memory is run of the mill Pc100/PC133 memory not even special temperature range or rad hardend. Even if they built 10000 cards thats not even 500k chips for the whole lifetime of the cards. A PC manufacturer back in the day probably used that per day or per week. Micron probably asked when they would stop sampling and start buying 

I guess thats also the reason why all the other chips come in multiples: it is just cheaper (or you cant even get them in lower quantities).

@gslick: I think that card was originally a different model. might have been a 16750b not sure.
forgot to take a picture of my card when I modded it to be a 16742a but I think R68 belongs at R63 not sure. ebay constantly wants to give me a free iphone so I cant look at the pictures of 16742a there right now, but you can give it try

[DocBen]

I just checked what kind of equipment you need to actually verify specs (from the service guide):
- a pulse generator with 200 MHz (still reasonable) with 2.5 ns pulse width and < 600 ps rise time (and i'm out)
- a digitizing oszilloscope (check) with >= 6 GHz bandwidth, < 58 ps rise time (and i'm out)

so at least an "offical" verification is off the table for me.

I might do some experiments with an FPGA ( upduino when (if) I get it ) as it supposedly has a 275 MHz PLL onboard that can drive an I/O pin. Maybe clock doubling with an XOR gate from potato semi and a clock buffer. That setup could in theory reach 800 MHz but I don't have an oscilloscope that can verify that so 

[DocBen]

the 16960As have a similar feature it seems. There is a field of resistor rows above the main ASIC.

And it is suspiciously different for 4M and 16M cards

Now I only have to unbrick the three I have. Maybe I'll get to that this weekend, maybe next.

Does anybody have the (native) 100M version of the card?
I think there is different resistor value on there. Mine are all blu, dont know the value yet, but on pictures the first resistor of the second row looks different on the 100M card. But maybe thats just the pictures I saw on ebay.

[CatCow]

Glad I saw this thread today... I was working on my 16903A/16910A this morning, after finally obtaining some parts as an experimental repair. Got a good deal on it from eBay, has two 16910A cards - 250MHz state, 256k memory on one and 4M memory on the other. The 4M card had a large number of memory errors. It may not be the memory, but I decided to try replacing those chips - my first attempt at SMT rework. Well there are also clock errors, and now after the "repair" another memory chip is throwing errors.

I noticed that there is a similar pad of jumpers for the board ID to the older models... Two rows of six small blue SMT jumpers. I didn't remove one to measure, got about 4.2k ohms across an installed one. As seen in the attached photo, the 4M 250MHz card has all jumpers in place except B6. I didn't take a photo of the 256k, but it had all jumpers except B5. Given the failed attempt at repair and other more difficult to diagnose/repair problems on the board, I will probably attempt to play with the jumper locations and see what happens. If it works out well, maybe it will be worth trying to hunt down the clock error...

[gslick]

One thing that is different about the 16910A module is that it can be upgraded in the field by the customer by installing a software license. Possibly the licensed sample depth / state speed configuration is stored in the serial eeprom along with the serial number.

If you were adventurous and have multiple 16910A modules to experiment with maybe you could try removing and reading the serial eeproms to compare the contents between multiple modules. Maybe there would be obvious differences for the module serial numbers, any maybe some not quite as obvious differences for licensed sample depth / state speed configuration option differences.

[CatCow]

Well the 16910A looks to be more controlled by software than hardware as you suspected. I did manage to break it enough to say that it was an invalid board(and an FPGA error message that wasn't showing up otherwise), but anything that worked was not recognized as a different spec module. I may spend some time tracing the clock signals and see if there is anything to be done there for trying to make the board operational, otherwise I see a lot of clock fanout chips that may find their way into a GPSDO project(would be just my luck if that was the bad part). Bridging all the pads has brought it back to a point where it will be recognized, but unless I find a cheap board on eBay to go in it, I'll probably just use the single working board. Not exactly limiting for what I might use it for.

[perdrix]

Did anyone who has both 16717A and 16719A boards ever find the time to compare to see it the 16717A can be upgraded too (I'm up for upgrading the ROM ICs if I can find suitable ones).

If necessary I can pull a 16717A from the analyzer and take pix. 

Or are the 16718A/19A different beasties altogether?

Thanks
Dave