"NU180" - A U180 drop-in replacement for the 3458A.

[Zondar]

The problem with off-the-shelf arrays is that the good tracking values (e.g. 5 ppm) are only achieved within one package. As soon as you cross a package boundary, you lose that, and the net tracking defaults to the absolute TCR value (e.g. 25 ppm). This application is not well suited for these arrays since most ratios would require crossing package boundaries.

By contrast, foil resistors offer a vastly better 0.2 ppm TCR, which can't be beat by any off-the-shelf arrays that I've seen.

On the other hand, it kind of appears that it's the resistance values (%) that are suspected of causing 114 errors rather than TCR (drifting). Perhaps some resistors, e.g. 16.7k, need to be tweaked in value? But once values are certain, the foil resistors will still have an advantage over the arrays (0.01% vs. 0.1% absolute). The down-side is expense, but 8.5 digits was never expected to be easy or cheap.

[MiDi]

VPG BMF are really good, but their DS specs are - let's call it - marketing driven.
IRL you will have a hard time to find matched resistors that track better than 0.2ppm/K - just have a look at their BMF RN specs.
In contrast the Vishay RNs tracking specs are quite conservative.

[KE5FX]

Dumb question time: these days you wouldn't obsess over the perfect ratio for ramp-up and ramp-down time constants, you'd go for the most stable solution and calibrate the time error out digitally.  Would it make more sense to just re-engineer the whole ADC board using modern parts and techniques?

[Zondar]

KE5FX: Go for it! It would be a massive project, though. I think some here like Kleinstein have been working on similar things.

I do think that the 3458A, being ~40 years old, is past its prime as a design. Maybe it's still technically the king, but if so, only by a thin margin, and they can be very difficult to keep working.

MiDi: I haven't researched VPG's products. The ones I custom ordered are Vishay's FRSM (replaces VSMP), quoting 0.01% and 0.2 ppm. I ordered a full set except for the 37.5 and 16.7's.

[KE5FX]

Hey, I wasn't volunteering.   Just curious as to what drove the decision to attack it at the U180 side, given that it seems nontrivial to unsolder and replace.

There aren't too many connections between the rest of the unit and the ADC board, it appears.  A disassembly of the 80C51 mask ROM would probably be the place to start, in order to see if there's any hope at all.

[wraper]

Hey, I wasn't volunteering.   Just curious as to what drove the decision to attack it at the U180 side, given that it seems nontrivial to unsolder and replace.

Depends on how much are you willing to heat the PCB. Could be done in a few minutes with some preheating and hot air (most likely you would not like to heat precision electronics this much). Low melting point alloy + large soldering iron tip would do it in a few minutes as well. Removing solder from the holes will take a few minutes too. Assembly + replacement of U180 is certainly easier, faster and cheaper than making the whole ADC PCB.

[MiDi]

Dumb question time: these days you wouldn't obsess over the perfect ratio for ramp-up and ramp-down time constants, you'd go for the most stable solution and calibrate the time error out digitally.  Would it make more sense to just re-engineer the whole ADC board using modern parts and techniques?

Modern approach would be to just have the MS-ADC optimized for the longer integration times and use fast ADC-IC for fast acquisition - like fluke.
No time interpolator & fast S&H required anymore.
Spec wise is not much room for improvement, so nearly 40 year old design still holds.

[Zondar]

Just curious as to what drove the decision to attack it at the U180 side...

There were a few reasons: To better understand the frustrating problems attributed to dying U180's and hopefully learn if/how they can be resolved. To start a fun but well-defined project (as opposed to a whole ADC, which could go in many directions). And to help those in the community (including myself) with a misbehaving U180. I'm very pleased that others, more knowledgeable about the 3458A than I am, have contributed and taken up the challenge too!

I've wondered how I would define "success" in this project. An 8.5-digit-worthy replacement is the ultimate goal, naturally. But I would be very happy and satisfied if a proper 7.5 digits was achieved. That would turn my hopeless instrument into one that is genuinely useful while going beyond the more common 6.5 digit DVMs.

[Kleinstein]

My ADC experiments go towards a complete alternative DMM, so far DC only and also not much faster than 1 KSPS (SW currently 1 PLC only).
It allows testing with only reduce functions and can add extra functions (like 20 V range) where easy.
I did think about an A3 replacement, but it is still quite some effort. My ADC version is more on the simple side (only 1 x LV4053 switch so far). Good noise but still a bit higher INL than the 3458.
Noise wise there is still room for some 10% improvement. A step in that direction is higher voltage and higher resitors for the ref. currents (e.g. 15 V and 100 K instead of 12 V and 80 K). This would also make it easier to use standard resistor arrays.


The rundown convergence error is likely not due to poor TC matching, but more some issues with the absolute values and maybe contact issues.
The smallest slope current should be at some -1.2 µA and thus still well larger than switch leakage. The run-down part does not operate the switches in U180, so charge injection, switching delays or similar should not cause the error message.
The higher switch capacitance could increase the overshoot and this way a bit slower run-down.

For the initial testing and understanding the problem it could help to know how much the small currents are off. So this would be measurements of the Integrator waveform. This could also give a hint on at what level the error in the current the error message comes up. For a test one could have a trimmer for some added offset (e.g. some 10 M to a +-12 V adjustable voltage). This way one should be able to to trim to a working state and see the limits.

I would consider something like 2 times the noise and maybe 0.5 ppm INL a success. It may not need that much extra tweeks to get there.

[Zondar]

It may not need that much extra tweeks to get there.

Please be as explicit as possible about what tweaks or experiments you would propose, such as specifically which nodes to probe, which specific resistors to adjust, etc.

At the moment, you seem to suspect the 16.7 resistor(s), and wanghar found that adjusting one or both of these seemed to make a big difference.

I made a little box containing a coarse and fine-adjust <=1 MOhm variable resistance that could be put in parallel with any of the resistors. Series would typically require some surgery, which I'd do with some reluctance if indications for that are good. (I'm not willing to hack my A3 board more than the socket I already put in, though.)

Update: Interesting! I lowered the resistance of the 16.7k resistors (one at a time) with my little box, and that allowed running DCV without errors.

The one associated with pin 19 had to be reduced almost by half. The one associated with pin 21 had to be reduced by a third. Both of these affected the digitized value.

I tried several more with similar results. The magnitude of the changes required are too large to make sense as a solution, but maybe there's a clue in this.

[wanghar]

Since the ACAL DCV was performed yesterday, the #2 board has been continuously measuring voltage for 16 hours with normal operation and no errors. During this period, the indoor temperature fluctuated by approximately 3 degrees, and the voltage Vpp was around 65μV. The overall temperature coefficient is 2.3 ppm/°C, which is acceptable considering the 10 ppm resistors used. Of course, this also includes the influence of the voltage source, though its impact is much smaller. The average standard deviation during this period was 0.079 ppm. Under the same testing conditions, the original U180 on this board had an average standard deviation of about 0.05 ppm. It is approximately 60% higher.

The ACAL ALL has just been completed and passed in one go also without any errors. I will unload the #2 board and see how to make #1 board working better.

[Kleinstein]

It may not need that much extra tweeks to get there.

Please be as explicit as possible about what tweaks or experiments you would propose, such as specifically which nodes to probe, which specific resistors to adjust, etc.

A first test would be to measure the integrator output during run-down with a DSO. This would be TP112 or if too noisy alternatively TP140 (slope amplifier output). For a trigger and to see which run-down step on the the signals at the HC14 (e.g. S16 or pin19 or pin21 of U180) would good.
The slopes would give an idea of the current from the ladder (and 604 K) resistor. Here especially the last 2 rundown steps would be of interest.

If the 16.x K at pins 19 /21 need so much change, there seem to be a bigger issue than just normal resistor tolerances. Are the -12 V and 5 V ref levels OK ?  The gain from -12 to 5 V sets the overall current from the ladder and would act like changing the 604 K (R154).
Another point to check is if some  37 K and 16.X K resistors are mixed up.

When much lower values for the 16.x K are needed to make the ADC work this means too positive a current. For a more negative current one could add resistance in parallel to R154, at the U180 socket this would be pins 14 and 24. Chances are one would need a relatively large value, like 1-5 M.

[iMo]

..During this period, the indoor temperature fluctuated by approximately 3 degrees, and the voltage Vpp was around 65μV. The overall temperature coefficient is 2.3 ppm/°C, which is acceptable considering the 10 ppm resistors used..

After soldering all the resistors and switches in onto the pcb one has to wait for some time (like X weeks ??) to settle down the stuff, perhaps to anneal somehow the pcb too, imho..

[wanghar]

Today's update on board #1.  Actually I'm seeing the same issue as described by Zondar. It throws a 114 error after running for about ten seconds, but then it works normally when switched to the ACV range to measure AC voltage. Based on previous analysis, the problem is likely some deviation in the resistance matching on the ladder network. Just like with board #2, I tried adding a 200-ohm resistor in series with the two parallel 33K resistors, but this time it didn't work. The problem remains, with no improvement at all. I didn't have a suitable trimmer resistor on hand, so I tried 100 ohms and 300 ohms, but neither worked.

I ran a simulation and found that the influence of these resistors on the S1 current varies. If all resistances change by 0.1%, R154 on the A3 board has the greatest impact on the S1 current (-1.77%), followed by the connected 37.5K at PIN23 (1.26%), then the middle 37.5K (1.01%), and after that the two 16.7K resistors. The simulation also showed that the +5ref (applied to the resistor network via the 74HC14) significantly affects the S1 current. A 0.1% change in voltage causes a -1.67% change in S1 current.

Since the measured -12ref was at -12.16V, I tried increasing the +5ref voltage to balance the negative current. I first paralleled a 4M resistor across the 24K resistor, raising the +5ref from 5.07V to 5.12V, but it still threw the 114 error. I further reduced the parallel resistor value, all the way down to 1M. At this point, the +5ref reached 5.16V, and it finally stopped throwing the 114 error.

Board #1 has now passed the ACAL ALL test and its continuous testing performance is basically consistent with board #2.

[Zondar]

A quick report:

Following wanghar's experience, I put my 1.1 MOhm trimmer box in parallel with the 24k resistor on board 2, adjusted it a bit, and that allowed DCV to run much better. Great investigative work wanghar!

I don't have a GPIB setup, so I can't produce nice charts like his. But I'll want to try my first board again, and will report back with anything more that I've learned.

[chuckb]

Exciting development work on this project!
When simulating with the 74HC14 keep in mind it has about 50 ohms of output resistance. This resistance has a +4000 ppm/C (silicon) temp co. The resistance varies with bus voltage and may be different for a logic high output verses a low output.
The 74AC14 has roughly half the output resistance so it's temp co influence will be half.
Good luck!

[Kleinstein]

The HC14 resistance part is known. It adds to the 16.7 K resistors. These resistors have a limited effect on the small slopes and the small slopes only contribute relatively little to the result. Taking the finest step as starting point it is only some LSB 256 steps from the small step part. So these resistors would only need to be accurate to some 0.4% and the 16.7 K are only roughtly 1/3 of this. So 1% tolearce with the 16.7 K would be roughtly acceptable. With this the 50 ohms even with a high TC are not that relevant. It is more that the math calls for 16.666 K including the switch resistance. So the ideal values is more like 16.616 K and not 16.7 K. However the difference is still just in the tolerance window, though still not ideal.

[Zondar]

I think the 114 errors will need more work to understand and correct. After some more experimentation, I haven't found the adjustment that wanghar came up with to be fully satisfactory, at least with my board #2. Yes, it helped, but not well enough, as it still fails after 10-15 minutes or so. Furthermore, an ACAL DCV works for the 10V range, but fails with several errors immediately after it switches ranges.

Meanwhile, I received the precision resistors, but I don't have a board on hand that can accept the 2512 ones. I'll need to put one into fabrication, which unfortunately takes 3 weeks or so. I need to switch to a faster board fabricator!

I'm thinking about putting some or all but the 2512's (50 and 80k's) in an existing board as an interim measure.

Edit: I measure the 4x 80k resistors. All four were better than 0.003%

[Kleinstein]

I think it is still too early to use the good resistors. Even the resistors used now should be good enough to make the ADC work. It would be mainly the gain TC that is expected to be not so good. It looks like there is some extra current to the integrator and chances are this extra current may well be the part that is not stable. Not sure what this is. It could be leakage from flux and maybe control signals to the switches that excedd the supply - even if still in the absolute maximum specs excessive current there can cause substrate currents and this way extra leakage.

The next point would be more looking at the integrator output to see what condition causes the error 114. It could be both too negative or too positive a current from the ladder. It could also be excessive ringing of the integrator (e.g. due to more capacitance)

[wanghar]

After adjusting the +5Vref by paralleling an additional resistor across the 24k resistor, my #1 board has demonstrated excellent stability (except during power-on stage which still requires external aid). The longest test so far was a continuous 24-hour run with no errors, including switching between different DCV ranges, all without any issues. After the +5Vref voltage adjustment, the standard deviation for the 10V measurement has further decreased to 0.066ppm. Increasing the +5Vref voltage might not actually be solving the -S1 step current issue, but more likely addressing the problem at the +S4 stage. I tried observing the run-down waveform with a DSO, but not success due to the  limitation of  machine capability  on my hand, the +S4 and -S1 stages can no longer be effectively identified—their difference were swamped by noise.

Regarding the failure to pass hardware self-test at power-on, both #1 and #2 boards have this issue. However, once started with auxiliary assistance, they enter a very stable state. In a previous post, someone suggested wearing gloves to touch and press the specific area. I tried that and it didn't work. Finally, I found that during the startup phase, an additional resistance is needed between pin 6 of the ADG788 (U11) and GND. This resistance must be removed immediately after passing the self-test, otherwise, it will trigger an 114 error. I've tried adding the PIN_KEEPER option and Power_reset option in the ATF1502 code, but none were effective. I suspect this is related to the IO initialization of the ATF1502, but for now this issue remains unsolved.

[Kleinstein]

At which integration time is the 0.066 ppm noise ?  It would be OK noise for 1 PLC, but an issue when at 10 PLC.
One may get a fist check on the INL (especially issues with the slope ratio) from a histogram in 10 V, 1 PLC non AZ mode with a short at the input. The drift would also give a hint to resistor stability, as one major source of drift would be the +- ref. resistor ratio drift.

If noise at the integrator output is an issue, one could look at the slope amplifier output. This would add some reasonable low noise gain for the final steps.
Getting a really good reading can be tricky - after all it is where the 3458 ADC sees it's noise limits.

[wanghar]

This average standard deviation of 0.066ppm is just for comparison purposes - the original U180 on this same board measured an average STD of about 0.05ppm. This reading includes noise from both the voltage source and the whole machine.

[wanghar]

A high level of noise makes it trickier to assess INL. Probably the most straightforward way is to do a side-by-side comparison with a second 3458A taking readings together.

[Zondar]

Yes, it might be too early. What are people's opinions about potential performance degradation when moving precision surface-mount resistors between boards? Anyway, I'm probably going to order a second set.

Wanghar, I'm confused about your description of the start-up problem. Which IC is "U11"? I don't see that one. And pin 6 is GND, right? That doesn't make sense in the context of your problem. Anyway, my board doesn't have a start-up issue.

There are almost no meaningful differences between your board and mine. Different values were used for the 6k/4.14 resistors but in essentially the same ratio, and you slightly adjusted the 24k resistor. That's it as far as I can tell. I suppose it's possible that the tune of our machines differs somewhat? For example, how important is it that the +12, -12 and 5V references are exact?

I've mostly focused on U180, but it's past time for me to study more about the circuitry and operation of the ADC as a whole. What is "AZ" mode, by the way?

Edit: This morning, Board #2 (with an adjustment to the 24k resistor) is not working as well as yesterday. It's producing 114 errors that won't go away, or at best only a few digitizations are performed before stopping. This is similar to what Board #1 was doing (after initially working great for the first few hours).

Measured on NU180 to its ground. True to U180 I did not connect pins 4, 8 and 64, or  10 and 14 on the board (looks like wanghar did):

+12 ref = +11.98446 @ Pin 4, +11.98445 @ Pin 8, +11.98448 @ Pin 64
-12 ref = -11.98670 @ Pin 10, -11.98671 @ Pin 14
+5 ref = 4.99501 (without 24k resistor mod).

With the 24k resistor mod, +5 was reading ~5.15V, which I think is too much for this mod to be "right."

[wanghar]

Yes, in my circuit, I only replaced the ADG734 with an ADG788. Another minor difference is that I added voltage modulation, which shouldn't have any significant impact. But I'm not sure why there's a startup issue. I suspect the ADG788 and ADG734 are actually the same die, just in different packages.

AZ stands for Auto Zero. It means each measurement cycle consists of one measurement of the voltage being tested and one measurement of the zero point (GND), then the difference is calculated. This helps reduce some noise. Actually, the 3458A's default mode on startup is AZ.

The component designations in my circuit are slightly different—please refer to the attachment.