Anyone have more than one 3457a and compared their DC readings?

[rigrunner]

Just picked up another 3457a and noticed that it seemed a bit more noisier than my existing meter.
Let them warm up for a couple of hours and then fed them both 1V and logged a little data for the 3V and 30V ranges. Both meters are 2703A serial numbered.

Just curious if anyone else has done a comparison.

Edit: 3457a settings were AZERO=ON, MATH=OFF, NRDGS=1, NPLC=10, FIXEDZ=OFF,;NDIG=6

 

[e61_phil]

Hi,

I have only one 3457A but mine is much noiser/unstable than my 3456A (in 10V Range). I've attached a short comparison in with 1V and 10V from my Fluke 343A. Both meters are configured to 10 NPLC.

[rigrunner]

Is the 3456 a cleaner trace because 10v input is at the top of the selected DC range whereas 10V on the 3457 is only 1/3 of the selected range?

I've spent some time this evening cleaning the A2 board and the holes in the input hybrid.
30V range with 1V input as before with log data after cleaning overlaid. It has made quite a difference, but i still have the odd spike here an there.
Much better than it was though

[e61_phil]

I think the noise is more related to this "divide and then amplify again" input in the 30V range. In other meters (3456A, 3458A, 34401A) the 10V input goes directly to the ADC.

Your cleaning results are quite impressive. Do you use IPA for cleaning?

[rigrunner]

Your cleaning results are quite impressive.

I'm so pleased my wife doesn't read this forum 

I put the A2 and Vreference  in a plastic container with 500ml of IPA.  Stood the A2 board on one end and spent about 20-25 minutes washing the IPA over both sides of the board with a new paint brush.  To clean the input hybrid i used string soaked in IPA to pull through the holes.

[e61_phil]

I forgot to save the data with 1V in the 10V Range of the 3456A and 30V Range of the 3457A. The 3456A is even stable in the lower end of a range.

I bought the 3457A due to the 7.5 digit resolution, but it seems quite useless if 6.5 digits aren't stable. I'm interested in short term (many minutes to hours) drift measurements (and I have no 3458A at home )

[rigrunner]

I bought the 3457A due to the 7.5 digit resolution, but it seems quite useless if 6.5 digits aren't stable. I'm interested in short term (many minutes to hours) drift measurements (and I have no 3458A at home )

I've picked up a second 3457a to try and monitor thermal drift of a reference.  The noise isn't helpful

I noticed a mod to the precharge offset adjust (U216) in Dave's teardown that isn't listed in the service manual.
<https://www.flickr.com/photos/eevblog/8488474552/in/album-72157632803534496/>
The remainder of my noise seems to be coming from this area so i'm a little curious about it.

[Kleinstein]

The 3457 is relatively noisy. This is for two reasons. One is that is uses just a simple resistor for current limiting in the protection. The second point is that the best range is for 3 V maximum and this already includes some amplification (thus less stable). So it is a little hard to compare with other meters that usually have a best range of 10 V or 20 V nominal, that uses no amplification before the ADC. So for something like 2.5 V the 3457 is not that bad, but at 7 or 10 V it uses the 30 V range with an divider at the input. So you have to compare the 30 V range to something like a 100 or 200 V range on other meters - these are also often not that great and more noisy than the 10 V range.

It is not that unusual to get the full resolution only with relatively slow reading, and still some noise visible. Many other meters are similar in that respect.

I won't expect the precharge circuit part to be a big problem - at least if it works.

[rigrunner]

I won't expect the precharge circuit part to be a big problem - at least if it works.

I substituted the reference and the hybrid from my less noisy meter in to the noisy one and it made very little difference to the remaining noise peaks i have.
Only thing between the hybrid and the dac that i can think  could introduce noise is U216.   

[Kleinstein]

I remember two tasks where U216 was defective and thus caused trouble.
Normally I would not expect the pre-charge adjustment to cause any significant noise. If there is trouble with the precharge I would expect too much bias current as the main trouble. So the right test would be to measure the voltage with a 10 M or similar resistor at the input.

Also note that when changing the input hybrid, there might be need to do an adjustment of the precharge offset. I am not sure if this is done during power up - it might need to be triggered by hand or an extra calibration step.

To check for the noise source, the interesting ranges are more the 3 V and 30 mV range. The 30 mV range is mainly testing the input amplifier, the 3 V range is more influenced by the ADC and less be the amplifier. If really needed one could test the ADC alone by removing the jumper JM101. Also looking at data with and without autozero and different integration would help - as a side effect you could find comparable data in the thread about the noise comparison thread.
If precharge is a problem mainly the mode with active autozero should be influenced.

[rigrunner]

Had a little time to run quick tests with AZ on/off. Both 30mV scale 10 NPLC with 15MOhm across hi lo terminals.

I'll get more data later this evening.

[e61_phil]

The 3457A software feels like a small 3458A but the meter itself is quite disappointing. With this amount of noise one can average the 6.5 digits and will get the same average values (instead of the 7.5 digit), I think.

[rigrunner]

Also note that when changing the input hybrid, there might be need to do an adjustment of the precharge offset. I am not sure if this is done during power up - it might need to be triggered by hand or an extra calibration step.

It passed self test and autocal with the swapped hybrid. The SM suggests only to adjust A2R113 if HW error and Auxerr=16.

To check for the noise source, the interesting ranges are more the 3 V and 30 mV range. The 30 mV range is mainly testing the input amplifier, the 3 V range is more influenced by the ADC and less be the amplifier. If really needed one could test the ADC alone by removing the jumper JM101. Also looking at data with and without autozero and different integration would help - as a side effect you could find comparable data in the thread about the noise comparison thread.
If precharge is a problem mainly the mode with active autozero should be influenced.

There's a definite negative offset when AZ is on. Readings are quite wild with AZ off. 3V and 30mV ranges behave the same.
I'm going to take another look at the input amplifier section.

Whilst i was waiting for the readings to be collected i took a little closer look at the mod in Dave's 3457a.
R216 511K  changed to  49K9 1%
R217 511K  actually 510K
Rx01 N/A    added 30K1 1%  Pin 9 U216 to +15V
Rx02 N/A    added 200K 1%  Pin 13 U216 to +15V
Rev A, B and C of the A2 board are in the SM i have and there's no notes regarding these changes that i can see.

[rigrunner]

With this amount of noise one can average the 6.5 digits and will get the same average values (instead of the 7.5 digit), I think.

I agree with you again.

[Kleinstein]

The bias currents are very low (e.g. 4 µV at 15 MOhm is about 0.25 pA), both with and without AZ active. This is nearly too good to be true - other meters are specified at something like < 30 pA .

The autocal might be enough to get the settings adjusted for the changes hybrid already. This might also adjust the bias to the very good values. Also the AZ reading look much better than the readings without. So I won't expect trouble with the precharge circuit part.

The readings in the 3 V range are really bad compared to the 30 mV range. This indicates trouble from the ADC and not from the input amplifier. Here especially the non AZ readings shows something that looks like popcorn noise: more or less random steps of considerable size. The AZ mode can compensate for most of this and only short pulses remain, similar to differentiating the non AZ signal.

However finding the origin of trouble in the ADC circuit could be really difficult. My guess would be one of the JFETs (especially Q513, Q514) - but there are quite a few. A first thing would be to check the supplies in this area for ripple.

[rigrunner]

Spent the evening scoping pretty much everything in the A/D circuit. Q515 had a nice crisp signal on the gate and a scratchy/jumpy signal on the drain.
SM lists Q515 as a Siliconix FN2761. I can't find anything datasheet wise for it    I've put a j110 in place for testing purposes and the initial results look good to me. 

The resistor across hi lo measures 16.245Mohm and it is reading around 50 µV at the moment. I'm out of time for warming it up and logging for any reasonable amount of time but a quick log is attached. I'll gather more data later on.

[Kleinstein]

50 µV at 16 M is still a really low input current - should not have changes anyway. Variations could be just a question of temperature. The test for the noise of the meter should be done without the resistor - with the resistor you see the noise of the resistor and the still low input current noise.

Q515 is used as a switch.  The properties and type of this FET should not be that critical. So I don't think is is responsible for the "scatchy" signal at it's drain. I would not be surprised if the DMM would still work without it, or replaced by a 1 K resistor. I am more surprised to see this FET at all.

It's more Q513 that sets this signal. Also leakage of Q511 is a possible candidate. This circuit part is for compensating the offset of Q514. So it might be interesting to look at the DC level at the drain of Q515. A relatively large offset might be better compensated in another way.

[rigrunner]

I've picked up a 2n4291 as a permanent replacement for Q515. Hopefully i'll have time to fit it later today and retest with a short and a resistor for comparison.
I'll make a note of temperature at the time of logging and i'll check Q513/4 again also.

Speaking of meter noise. How much do you think Q111, Q112 and U111 contribute to the meter noise in the 3V range?

[Kleinstein]

U111 should not give any significant noise contribution as it's inside a loop controlled by Q111. Q112 is just bootstrapping so it should not give a significant noise contribution either. Q111 is the main noise source for the input amplifier. Running the differential amplifier at about 0.4 mA total, I would expect some like a 0.1-0.3 µV for a 0.1 -10 Hz band. Estimate of the 1/f noise is difficult and it can be quite different from sample to sample. Q111 will be likely the limiting part for the 30 mV range, but not so much the limiting factor for the 3 V range.

There is a jumper (JM101) to test the ADC by itself.
For the 3 V range I would more expect noise from the ADC itself.

Here Q514 should give nearly as much noise as Q111. If leaky Q511 could also contribute to a noisy signal at the drain of Q515. Some of the resistors inside U511 can also contribute, especially for shorter integration times. Also U516 and maybe U513 can contribute to the noise.  Also power supply ripple / noise can contribute.

The big question is what could go wrong and cause a much higher than normal noise for the ADC. For me candidates are Q514 and supply ripple / noise. A noisy signal at the drain of Q515 points a toward Q513 and maybe leakage from Q511 / C512. So it might be good idea to check this signal (or the other side of R519) again. It should be jumping between the offset of Q514 (e.g. a voltage in the +-20 mV range) and a smaller value (e.g. 1% of the signal before) near zero.

[rigrunner]

Finally found some time to replace the fet and retest.

I didn't see anything worrisome on or around Q511/2/3/4/5. I've put a .1 µF across U507 supply and also between pin 3 and 5 of the ADC to try and take the edge off the digital noise on the 5V rail. 

Results of brief test after power up with short across hi-lo attached.

[Kleinstein]

The data for the 3 V range now look good. It's mainly jumping between 2 value, with a few of a third value.

The data for the 30 mV range without AZ still shows rather large noise - this should be noise from the input amplifier, especially Q111. To me this noise looks higher (about 5 µV_pp)  than normal (would expect about 1 µV_pp due to rather large time window). At least I would expect better performance from decent fets. With a second working meter you could compare in the 30 mV range without AZ and low NPLC setting.

The AZ modes look much better because the AZ mode can to a large part compensate for such popcorn-noise like steps. However noise would be higher too and there are these few spikes as leftover.

[rigrunner]

I'll power up both meters in a little while and compare the two on 30mV range.

(I meant to mention this in my previous post but was in a bit of a rush when i posted it.)
The reason i asked about U111 introducing noise was a thought that crossed my mind when looking at the schematic. The 3457a was designed with parts available in the 80s.  I wondered if it was worth replacing any parts in the amplifier with newer, lower noise parts?

[Kleinstein]

For the ADC it looks like it get to the limits of it's digital resolution - so nothing you can do about this.

The main noise source of the input amplifier (low level DC) should be the dual FET Q111 and maybe the resistor for input protection. JFETs are not much better now than they were in the 1980's - it's only more difficult to get JFETs. The major weakness is having only 3 V (nominal) maximum voltage at the input and no range without amplification at the input - even the 3 V range uses some amplification and thus is not super stable. 

The noise for the 30 mV range looks relatively high so it could be that Q111 is not a good one (e.g. one with high 1/f noise or not likening the current it gets). One might want to check the setting of R113 (DC offset). One could test a little higher current (e.g. resistor in the 20-50 K range in parallel to R114) - sometimes popcorn noise depends on current
If one has to replace Q111 one would likely need to go for a more modern type anyway, as this an old hard to get type - could be a good one though.

The other point is checking the supply: bad caps and thus to much ripple on the supply can look a like noise.

[rigrunner]

There are large offset jumps that make the 30mV noise look bad. If these were removed it would be similar noise to my other meter.

I've not spotted any ripple on the supply rails, just digital noise during readings. I'll experiment with R113 and paralleling R114 and see if it makes any difference.

[rigrunner]

Adjusting RV113 had no effect on spikes. Adding resistor to R114 shifted reading but had no effect on spikes.
Scoping pins 5,6 and 7 on U111 showed jumps on pin 7 but nothing corresponding on pins 5 or 6. After replacing U111 the readings look better. (Bear in mind that there are no shields or lids on the meter at the moment and warm up is probably 20 minutes.)


Edit:

False dawn. Once warmed up it's back to jumping around