HP 34401A hacks and upgrades

[iMo]

While still targeting the HP399 replacement with my ADR1399 (showing lowest noise off all my 399s) I've made the first step - adding a 3.3u foil (wima red brick) in parallel to R442.
So far my ext_ref_10V dropped by aprox 2.5ppm (2h after soldering the capacitor in).

FYI - the MCU_ADC in the 80C196 and the 4053 switch in the ADC are powered from the output of the "U401-B" (AD706) creating 5V from the divider wired to the 399.

+5REF - powers the MCU_ADC (aprox 5mA based on DS)
+5VB - powers the 4053 integrator switch in the main ADC
..

See below the planned changes. Based on my simulation the RMS noise at +REF10 should drop by 43% (0.1-10Hz) and 5.4x (0.1Hz-10kHz). The -10REF should be aprox 10% worse then. That drops thanks the 47k/3.3uF low pass. Another drop will be made by the 1399 itself (..hopefully).

[GigaJoe]

shorting +-15 by 10uf not so much ....   i would short each 0.1 by some polymer caps, at least 100uf , and do per each capacitors everywhere as +-15 everywhere .  ideally separate this opamp by 20-50 ohm resistors ,  if PSRR low ... ( but it would be difficult,  you may look at short C441, no sure how it may behave ... for zener you can try 40uf .... or even 2 polymer caps in series with resistor ... the 47K at input , how much  noise from R ?

[Dr. Frank]

While still targeting the HP399 replacement with my ADR1399 (showing lowest noise off all my 399s) I've made the first step - adding a 3.3u foil (wima red brick) in parallel to R442.
So far my ext_ref_10V dropped by aprox 2.5ppm (2h after soldering the capacitor in).

FYI - the MCU_ADC in the 80C196 and the 4053 switch in the ADC are powered from the output of the "U401-B" (AD706) creating 5V from the divider wired to the 399.

+5REF - powers the MCU_ADC (aprox 5mA based on DS)
+5VB - powers the 4053 integrator switch in the main ADC
..

See below the planned changes. Based on my simulation the RMS noise at +REF10 should drop by 43% (0.1-10Hz) and 5.4x (0.1Hz-10kHz). The -10REF should be aprox 10% worse then. That drops thanks the 47k/3.3uF low pass. Another drop will be made by the 1399 itself (..hopefully).

Good idea to replace the LM399 by the ADR1399. I just want to do the same on my old 34401A.

I can confirm, that the ADR1399 gives a noise advantage over the LM399, at least in my FLUKE 343A calibrator:
https://www.eevblog.com/forum/repair/fluke-343a-cleaning-repair-and-improvements/msg6225429/#msg6225429

Here, noise improved by about 10% only, but I guess that the electronic components from 1969, as well the 584V bias voltage overwhelm the real noise improvement effect by far.

I had replaced the LM399 inside my 34465A by an LTZ1000A and found a 50% noise improvement. As the ADR1399 is in the noise ballpark of the LTZ, I estimate, that you might get a 20..30% improvement for the 34401A.
Please keep us updated about the noise improvement, Thanks a lot.

Frank
 

[Kleinstein]

The 34401 has a rather noisy ADC and there is not much that can be done about this (much of the noise is from the µC internal ADC).
A ADR1399 reference could help a little with longer time noise. It mainly avoids the popcorn type jumps of the LM399 and thus low frequency reference noise. For the short term (e.g. less than 10 seconds) there is little improvement expected. Some extra filtering of the reference can help, but the 47 K 3.3 µF is likely overkill in combination with the ADR1399. With some 160 ms RC one can nearly bridge the AZ loop with 10 PLC (the zero reading makes littel use of the reference voltage). It likely still makes little difference as the ADC is more noisy than the reference at that frequency.

[iMo]

@GigaJoe:

Code: [Select]

Simulation with ADR1399 and AD706 and HP schematics:
RMS noise in nV at +REF10, R=47k, C=3u3

       0.1Hz-10Hz    0.1Hz-10kHz
no RC      326         5451
R only     349         5240
RC         188         982
I've also tried to block +5REF (powers the entire MCU_ADC module, btw.) and the FLASH ADC_MCU input (fed by 1k off the AD711). Soldered directly at the MCU pins, against ANGND (in schematics wired together with VSS).
The FLASH blocked by 82p C0G, it messed up the reading - tens of uVolts randomly jumping.
The 100nF ceramics at +5REF dropped by aprox 1ppm, and increased a little bit the STD.
The 4n7 foil at +5REF with no drop visible, and STD under investigation, next morning we will see..
The lowest STD was with none capacitor at +5REF.
Could it be the U401-B AD706 is somehow unstable with such large capacitive loads??
The load in original schematics is 100ohm/100nF at +5REF and in parallel 46ohm/100nF at 4053.

@Frank: Way back I replaced the HP399 with one of my lowest noise LM399 and the noise decrease was visible. The issue with that LM399 was it popped couple of times per day (nice 5uV jumps).

PS: nVolts in the above table, of course..

[Kleinstein]

The AD706 datasheet shows an example with 1 nF capacitive load: quite some overshoot, but still not oscillating. So the OP-amp is somewhat tolerant to capacitive loading. The extra resistors of 100 ohms and 46 ohms would provide extra isolation of the capacitive load. It is a bit on the low side for isolation, but one also does not want too much resistance.

The relevant frequencies for the reference noise depend on the PLC setting. Likely one would use 10 PLC and longer, which is averaging 10 PLC conversions.
The main frequencies that would contribute are
1) the very low frequency range, below some 1.2 Hz, often well below 1 Hz, here the RC filter is barely effective
2) some aliasing from not really using the reference only half the time (zero conversions):  1.3 - 3.7 Hz   (can become a narrower band around 2.5 Hz for higher PLC), the RC filter has some limited effect
3) around 1/2 modulation frequency , ~ 170 kHz range, filtering is easy. The limited speed of the OP-amps already provides some filtering.

Filtering the  "flash" ADC input is not a real option. The ADC reading is only the fly and thus needs to be relatively fast.
The µC internal ADC is not a flash type, but a conventional, relatively fast for it's time SAR type with 10 bits.

[iMo]

Frankly, this is the first time I've been digging into that AD706 output - there are two +5VB and +5BV voltages in the schematics so a little bit confusing to find the labels easily..
Seeing now the entire MCU_ADC module in the 80C196 AND the 4053 switches in the integrator are powered by that single AD706 opamp via those simple RC filters is something I would call a Tragedy.. 

[Kleinstein]

The HC4053 switches would not need much power (should be some 100 µA). Most of the 80C196 µC is still powered normally. Only the ADC reference is from the AD706. So both are light loads.

[iMo]

..Most of the 80C196 µC is still powered normally. Only the ADC reference is from the AD706..

That would be nice, but it is not the case, I am afraid..
Max 5mA IREF of the ADC module (analog+digital) powered by +5VREF (and the power loss at the AD706 powered off 15V??)..
Imagine the "digital" mess at the +5VREF node (the 0.5V lost at the 100ohm).. 

[iMo]

Below an overnight measurement without a capacitor and with 4n7 foil between pins 12 and 13 of the MCU (+5VREF power and ANGND).
STD made with a sliding window of 100 samples, 100NPLC, 10V ADR1001#1 ref, both with the 3u3 foil at U401-B input, still with the HP399.
Median of that data (simple math approach but otherwise visible first glance):
none cap median: 749nV
with 4n7 foil median: 838nV
Either the AD706 is at the edge of instability or there is a grounding issue, or..

PS: the voltage at the 4n7 (+5VREF against ANGND) while measuring is 4.977V, at the U401B's input is 4.993V, HP399 out is 7.002V, power is +14.706V and -14.390V.

[Kleinstein]

The VREF pin of the µC has already 100 ohms in series for isolation and there are already 100 nF in the plan. So I doubt an extra 4.7 n would make a difference to the OP-amp loading. It is more that the EMI pic up can change.

The curves show quite some spikes, so the noise seems to include random jumps, like reference popcorn noise. So the test for noise should be done with a shorted input. Also noise would better be measured for the 10 PLC mode and thus with less averaging before.

[iMo]

Yep, the spikes are the main reason I want to change the HP399.
Added some voltages above. It seems the MCU_ADC IREF is indeed small.
The additional noise may come from coupling to the 4n7 capacitor off the MCU package, for example.. Anyhow, the traces are long, the 5V from the 399 divider goes to the AD706 placed obove near the HP array (and 4053) and then it goes back down to the MCU, so perhaps 20cm or more. Therefore I mess with the 4n7 there..
Shot from xdevs.com.. Traces not accurate, demo only..

[iMo]

..Also noise would better be measured for the 10 PLC mode and thus with less averaging before.

While still waiting for the low TC resistors delivery, I replaced the HP399 with my lowest noise LM399#2. Thankfully it did not jump over this overnight measurement.
10NPLC against myADR1001, raw data with STD sliding 100samples (the std over last 40secs).
4n7 at +5REF removed, the 3u3 at the 5V divider (U401B input) still there.

PS: with shorted inputs at 10V range the std=1.7uV (over 1000 samples).
So at the same level as the measurement above.. The ADR1399 will not help much it seems..

[Kleinstein]

The 120 nV noise for the shorted input case looks like it is for auto ranging and thus in the 100 mV range. That would be noise mainly from the amplifier.
For check of the ADC and comparison for the 10 V range the more interesting number would be for the manual 10 V range.

The 10 PLC noise of some 1.8 µV looks relatively good for a 34401 and this is already with a 10 V input. With a shorted input one could expect a little less (though not that much).

I looked again at the calculated noise for the 10 PLC case. The µC internal ADC part is still a significant part, but there is also the OP27 current noise as a large contribution at 10 PLC. The current noise specs are not that detailed (only 1 values for 2 inputs with partial correlation). So there is quite some uncertainty in how noisy the OP27 actually is.
There are now alternatives with a slightly lower noise (for some 15 K source impedance at 2.5 Hz) : e.g. OPA205, OPA141 ?. Besides changing the OP27 one would also need to adjust the divider for the different speed (maybe also replace the rather slow AD711 in the integrator).

[iMo]

.. corrected above.. it went to 100mV range.. 

The shorted inputs std I see now is 1.707uV calc over 1000 samples at 10NPLC at the manually set 10V range. So the ADR1399 will not help much, unless we hack the ADC somehow..
Below raw data and std over sliding 100samples.

PS: added a longer STD - interesting fluctuations.. Mean at aprox 1.6uV..

[Kleinstein]

Replacing the LM399 with a ADR1399 would help with the typical popcorn noise of the LM399. The sample without the visible popcorn noise is the exception and one can still not be sure if it would eventually jump (I have one that jmps about once an hour and yours may be even slower). With the ADC noise with a short the better reference could not help - there a little filtering for the reference is more effective  (still a small effect).

[iMo]

..the LM399#2 (I have in there now) jumps from perhaps 1 to 3x per day (based on three longer measurements, a typical 5uV sharp step up, and then after X hours down, or vice versa).

No idea how the ADR1399#1 jumps, it got perhaps 2000+ hours burnin, but I never did long term measurments with it. Its noise I saw with my noise meter was the half of the LM399#2. The second ADR1399#2 (worse noise to #1) died, perhaps a static or what while I messed with it (it started to fluctuate significantly like mV).

The old HP399 does not "jump in a typical sharp 5uV step", but its random walk is large compared to the LM399#2 (I see "almost" none).

[iMo]

..
I looked again at the calculated noise for the 10 PLC case. The µC internal ADC part is still a significant part, but there is also the OP27 current noise as a large contribution at 10 PLC. The current noise specs are not that detailed (only 1 values for 2 inputs with partial correlation). So there is quite some uncertainty in how noisy the OP27 actually is.
There are now alternatives with a slightly lower noise (for some 15 K source impedance at 2.5 Hz) : e.g. OPA205, OPA141 ?. Besides changing the OP27 one would also need to adjust the divider for the different speed (maybe also replace the rather slow AD711 in the integrator).

I have here some OPA140, OPA145.. The issue I see with such a replacement is whether the built in "service-guide adjustment" algorithm will incorporate such changes. In a thread in past a perhaps insider had revealed some math used in the meter, with some crazy second and third order coefficients, etc. So the question is whether a simple service-guide adjustment "off the front panel" is enough to incorporate the opamps repalcements.
If yes, I would be prone to replace them.. 

[Kleinstein]

I would not expect that there would be any changes be needed to special calibration parameters.

A 3rd order correction term could be there to compensate for a thermal effect of the input resistor. With 100 K at the input one may very well get away without it. So I don't expect this to be an issue.

A U² part could come from the R_on nonlinearity of the HC4053. Again with 100 K this would not be very much (maybe 1 ppm FS, likely less)
      This may be a part corrected from the positive 10 V and negative 10 V adjustment step.

An abs (U) part or different gain for the positive and negative side. This may happen from interactions from the switches to the clock. Ideally this should not be much. This may be a part corrected from the positive 10 V and negative 10 V adjustment step. However one could have only one parameter abs(U) or U² from this 2 scale factors.

There is a adjustment factor for the µC internal ADC and integrator gain. This is DNL relatated and may well be the special service mode.
There should be one more factor to take into account how much of the last reference step still counts to  before the auxiliary ADC reading and how much is effecting only the next conversion. This parameter would mainly effect the noise, as the last (and first) ref step is somewhat random. It would make sense to measure this parameter together with the µC internal ADC gain.

I see very little effect of the integrator on these parameters. So likely no change needed. If at all it could be last parameter of how to split the last step that could be slightly effected from a change in the settling. Ideally the internal ADC reading should be at a time where the integrator is settled (e.g. just before switching). To get the last step to be 100% effective, the sampling may however overlap with the switching and thus react to settling.

With a replacement of the OP27 one would also have to adjust the divider after the OP-amp because of the different GBW:
e.g. for an OPA140 or OPA141:  a larger value for R420, roughly 12 K because of a higher GBW (11 MHz typ)
e.g. for an OPA145:  a smaller value for R420, roughly 5.6 K because of a lower GBW (5.5 MHz typ)
With the resistor adjusted according to the GBW, the settling should not change much. It would be changing the AD711 to something faster that could change (improve) the settling.

[iMo]

The posts of alan.bain on the math and cal constants inside an 34401A:

https://www.eevblog.com/forum/testgear/hp-agilent-34401a-hidden-menu/msg4644106/#msg4644106

https://www.eevblog.com/forum/testgear/hp-agilent-34401a-hidden-menu/msg5130093/#msg5130093

[GigaJoe]

I assume anything recalculated under calibration process.. otherwise ,   under manuf. process chips need a bin selection process, it doesn't make sense.

[iMo]

FYI - latest overnight maesurement with my lowest noise LM399#2 in 34401A, raw data, 10PLC etc. as above. A pity..

[iMo]

I assume anything recalculated under calibration process.. otherwise ,   under manuf. process chips need a bin selection process, it doesn't make sense.

Based on the math above there are second and third order coeficients for the curvature compensation of the ADC. I somehow doubt those come from the simple adjustment procedure described in the service manual (like for the 10V DC range - enter the calibration voltage close to 10V and press Enter). I also somehow doubt these coeficients are kept constant during X years of the DMM production (because of the spread of the parts parameters actually used, for example)..

PS: ..my current understanding (perhaps wrong one, btw) is that every DMM (or its motherboard) undergoes a special adjustment procedure during the fab process and gets its own set of "coeficients". Afterwards, the user may adjust "some of them only" during that simple service-manual-adjustment-procedure. This user-adj-proc is intended for simple "linear-like ax+b" adjustments only, like a reference_drift/ref_replacement (ie 399, LTZ1000, etc.) or 10k cal resistor drift/replacement, etc..
So, when somebody replaces some components in the ADC and then shows at the end 10.000000V on the display - the story may not be over..

[Kleinstein]

AFAIK the normal adjustment procedure has seprate +10 V and -10 V steps that can be adjusted. As in the cal constants there is only 1 gain, the parameter to correct for a difference is via the U² correction term. So that linearity correction parameter is very likely adjusted there. The service manual also states that this step is corrent ADC linearity.

The µC internal ADC is likely also adjusted with the normal calibration: the math looks like the multiply the coarse steps with the number. This would mean, that after adjusting that number one would have to recalculate the "normal" calibration gain factors. To avoid rounding errors it would make sense to at least redo the normal adjustment after the scale factor for the µC internal ADC is changed. It could still be part of a special calibration step, that should suggset redoing gain calibration after this.

The optional 500 V step for the 1000V range is very likely doing the 3rd order correction for that range. This because heating of the divider is a major contribution for this range.

The open question is the 3rd order term for the other ranges. This would need accurate test voltages like 5V and 10 V.
This is likely some factory or special service calibration step, as it is demanding on the testpoints to be better than the intrinsic ADC accuracy.
Because of the LM399 refrence noise the test / adjustment should do several repeats. Even with my much lower ADC it takes quite some time.

[iMo]

So perhaps next week I would have following components at hand:
47k 0603 10ppm/C  - variant A for the RC filter
8k2 0603 10ppm/C  - variant B for the RC filter
820 1206 10ppm/C  - for the I_1399 (aprox 3.6mA)
5.1  1206 for the 1399 snubber
wima foil red bricks 3u3, 1u, 100n
ceramic 1u/25V 0805 for the snubber..

I think the setup will work for both ADR1399 and LM399. Now what values to choose for the RC filter ?

The 0603 sizes because of the rather limited space on the pcb (will be soldered over an already cut trace near the AD706 U400-A).

Btw there is an error in my schemtics and info above - the blocking capacitor at the +5REF is 10nF in the HP/Agilent schematics (so 100ohm/10nF), not 100nF as I mentioned above. The blocking cap at the 4053 Vcc is 100nF in the HP schematics (so 46ohm/100nF). Was discussed re AD706 U401-B as the power source for the MCU_ADC and 4053.