[Solved] 3457a huge warm up time - ADC fault?

[rigrunner]

I've got two 3457a meters, one has always been fine and the other has had a few issues.
I've been running comparisons between the two meters for a while and i've got one final problem that i think is caused by a faulty ADC but i'm not certain.

The problem meter takes over 6 hours to become stable after power on whereas my good meter warms up in around 2 hours or so. The issue seems to stem from the 10V reference moving upwards after it has warmed up.  It doesn't seem to be caused by the LM399 itself or the reference board as swapping the reference boards between meters gives the same results, and replacing the LM399 on the board also gives the same results.  The supply voltages for the reference are stable during the warm up.

There is a resistive divider across pins 21,22 and 23 in the ADC and pin 22 feeds back to the reference board. I think this might be causing reference to shift as the ADC heats up.

I've attached graphs of logged data showing the downward drift of the reference on meter 2., the warm up @ 100 NPLC and a part of the circuit that has the reference and the ADC. (The reference board doesn't have a diagram in the SM)

Before i condemn the ADC is there anything else i should check?

[Kleinstein]

I am not shure about the time scale of the 2 graphs, but the 10 V reference does not look that bad. It could be in part be the resistors in the ADC-hybrid, but there are a few other parts (OP, resistor) on the board too. There is also TP6 for the negative reference. The more important value should be the difference form TP6 to TP7, not the individual values.

Normally a not so good part should only give a stronger drift during warmup, not a longer time constant. If it is really a longer time constant and not only higher amplitude and thus visible over longer time, it could be the fan / filter (if the 3457 has one). A broken of to slow running fan can change the time constant as it takes longer to stabilize the temperature and the temperature will rise higher.

It might be also worth looking at the actual temperatures.

[rigrunner]

I am not shure about the time scale of the 2 graphs, but the 10 V reference does not look that bad. It could be in part be the resistors in the ADC-hybrid, but there are a few other parts (OP, resistor) on the board too.

Each logged point on the graphs is roughly 5 seconds 10V reference graph is approximately 45 minutes of data. A little over 9 hours for the from cold data.

If i overlay the offset adjusted 10V reference data on the from cold for the same time frame i see correlation. (graph attached)

By swapping the reference boards between the two meters i'd expect to see the drift move with the reference board if the cause was anything on the board itself?

There is also TP6 for the negative reference. The more important value should be the difference form TP6 to TP7, not the individual values.

The -10V also drifts. I haven't measured the difference between the two though.

Normally a not so good part should only give a stronger drift during warmup, not a longer time constant. If it is really a longer time constant and not only higher amplitude and thus visible over longer time, it could be the fan / filter (if the 3457 has one). A broken of to slow running fan can change the time constant as it takes longer to stabilize the temperature and the temperature will rise higher.

The 3457a is a closed plastic case, no air holes or fan to contend with. Internal case temperatures seem to be driven by the 3 voltage regulators on the A2 board which run at around 50 degrees C.

It might be also worth looking at the actual temperatures.

I don't have anything that i can log temperatures from automatically, but i'll take some points from the graph and write down the temps for those.

I left the logging running and it seems to become stable at around 12.5 hours.

[rigrunner]

Taped a thermocouple to the top of the ADC. Meter hadn't fully cooled from the previous run.
Room temperature was 22C and i measured 24.4C on the ADC with the covers off which rose to 26.2C when the covers went back on.

Made a note of temperature at some points during another logging session.

[rigrunner]

Here is the voltage measured between TP6 and TP7 from cold. Looks to be pretty much the inverse trace of the meter drift.

[Kleinstein]

Slightly more than 40 C is a reasonable temperature, not too hot, especially for a system without fan. The voltage drift also seems to mainly follow the measured temperature, except for the very fast initial part, which could be the LM399 reference.

Without a fan I don't think there is anything that could alter the time constant. So it's down to a rather high temperature coefficient for the +-10 V reference.  I looks like they path is  7 V reference (likely negative) to the -10 V and than to the +10 V. So it might be interesting to see if the -10 V are drifting to. From the +10 V and +-10 V difference data, it looks like that.

Without a schematics for the reference board, it's a little hard to see how the -7 to -10 V conversion is done. I don't see a way the divider with two 30 K resistors in ADC hybrid could do the job (maybe if there is a mistake in the plan, having 30 K and 70 K ? - than this could make some sense). So it might be worth to check the voltage at pin 22 of the adc hybrid, going to the ref board: with two 30 K resistors this should be about -5 V, with 70K/30 K it would be -7 V.

The ADC hybrid is used to do the -10 to +10 V inversion. With so much drift (150 µV), I would not expect the external OP U516 to be the culprit.

Anyway with a higher than normal TK of the resistors in the ADC hybrid, there is not that much one can do about it. For the 100K/100K pair one could in theory replace it with an external resistor pair - though not a very simple modification. For the 7V/10 V conversion it's more difficult to get a 70K/30K matched couple  that is better than the original one. It's a question whether it is worth the effort.

As a crude workaround one could even go for an added temperature stabilization. Either a heater, or maybe a small fan.

[rigrunner]

it might be interesting to see if the -10 V are drifting to. From the +10 V and +-10 V difference data, it looks like that.

I'll log the -10V at TP6 shortly.

Without a schematics for the reference board, it's a little hard to see how the -7 to -10 V conversion is done. I don't see a way the divider with two 30 K resistors in ADC hybrid could do the job (maybe if there is a mistake in the plan, having 30 K and 70 K ? - than this could make some sense). So it might be worth to check the voltage at pin 22 of the adc hybrid, going to the ref board: with two 30 K resistors this should be about -5 V, with 70K/30 K it would be -7 V.

Attached is what i read the reference board being.
ADC pin 22 is at -7V.

The ADC hybrid is used to do the -10 to +10 V inversion. With so much drift (150 µV), I would not expect the external OP U516 to be the culprit.

You are correct. Heating U516 made no appreciable difference to the voltage across TP6-TP7.

Anyway with a higher than normal TK of the resistors in the ADC hybrid, there is not that much one can do about it. For the 100K/100K pair one could in theory replace it with an external resistor pair - though not a very simple modification. For the 7V/10 V conversion it's more difficult to get a 70K/30K matched couple  that is better than the original one. It's a question whether it is worth the effort.

I'd like to try and resolve it.

As a crude workaround one could even go for an added temperature stabilization. Either a heater, or maybe a small fan.

Moving one of the voltage regulators to across the ADC would do it 

[rigrunner]

The -10V at TP6 looks as expected.
Does this point to U516 and/or the 100K divider on ADC pins 18,19 and 20 being the cause of the drift?

[rigrunner]

The answer is no. Replacing the 100K divider with a pair of resistors yields the same trace.

[Kleinstein]

The last curve looks slightly better than the earlier ones. The shape will be similar in most cases anyway, as this is the temperature profile; the main part to optimize is the amount of change.

The curve shows about 7 ppm from the maximum till 1000 seconds. Given an estimated temperature rise of 15 K, this is only a TC of 0,5  ppm / K. This is not such a bad figure.  So it might be that the replacement resistors are not that much better than original 100 K pair.

There are also resistors in the input amplifier, to set the input amplification - the 3457 is special in the way that there is no range without amplification at the input stage. So there is no especially good 10 V range without amplification in the input stage and thus less gain drift. The 30 V range is already using the input divider  - so compare it to the 100 V range of other 6 digit meters.

[rigrunner]

The curve shows about 7 ppm from the maximum till 1000 seconds. Given an estimated temperature rise of 15 K, this is only a TC of 0,5  ppm / K. This is not such a bad figure.  So it might be that the replacement resistors are not that much better than original 100 K pair.

It's the warm up time that is bugging me.
Just as a test i've replaced the 70/30 divider with a pair of external resistors . There is a monstrous negative drift showing at the moment.

There are also resistors in the input amplifier, to set the input amplification - the 3457 is special in the way that there is no range without amplification at the input stage. So there is no especially good 10 V range without amplification in the input stage and thus less gain drift. The 30 V range is already using the input divider  - so compare it to the 100 V range of other 6 digit meters.

I've already exchanged R111,R112 in the input amplifier to rule those out. And i've swapped input hybrids to rule out the switched resistors. Different fets for Q111 and Q112 have little effect on the curve. 

To best compare the drift across the 30mV, 3V and 30V ranges should i keep the input at say 30mV and change ranges or should i set the input to 30mV, 3V and 11V respectively (EDC only goes up to 11V).

[rigrunner]

After seeing the replacement divider making no difference or making things worse i left the ADC and went back to the input amplifier.

Too many hours of logging later i found that most of the drift was caused by Q113 and CR113/114. I still have the drift from the ADC divider<>Reference board but that is very slight once the longer term drift is removed.

Attached is the latest logged data against the earlier 20 hour run.

 

[Kleinstein]

Q113 and CR113/CR114 should not give a significant contribution to drift, unless they are really broken. They are just part of a current mirror in the input amplifier. If at all the input amplifier might give some offset drift - but this is usually taken care of by the auto zero function. The drift of the input amplifier is also mainly set be Q111 and only very small contributions from U111, R111, R112 and other parts.  So Q113 must be near dead before causing trouble.

The 30 V range of the 3457 is not really good, as the input amplifier only sees a 1:100 divided signal and thus is rather sensitive to drift of the input amplifier. So it can be important to use AZ in the 30 V range. The best range of the 3457 is the 3 V range.

Drift of the reference and the 7 to 10 V step is different from offset drift: this gives a change in the scale factor, similar to possibly drift of the amplification setting resistors inside U101.

I don't see an easy way to improve on the stability at the ADC / ref section. My best guess would be a temperature controlled fan - this could speed up the time it takes to get a stable temperature, and a slightly lower temperature also tends to be positive. However it will only work well in a limited temperature range (e.g. 18 C to 28 C, lower with heater). Also a fan can cause extra trouble, like dust collection and local temperature fluctuations from turbulence.

[rigrunner]

Q113 and CR113/CR114 should not give a significant contribution to drift, unless they are really broken. They are just part of a current mirror in the input amplifier. If at all the input amplifier might give some offset drift - but this is usually taken care of by the auto zero function. The drift of the input amplifier is also mainly set be Q111 and only very small contributions from U111, R111, R112 and other parts.  So Q113 must be near dead before causing trouble.

Cooling the transistor and diodes whilst running caused the drift to reset. I replaced Q113 with a BC328 and the diodes with BAV21 restarted and set it logging again. The data looked stable after 90 minutes, so i left it running.

The 30 V range of the 3457 is not really good, as the input amplifier only sees a 1:100 divided signal and thus is rather sensitive to drift of the input amplifier. So it can be important to use AZ in the 30 V range. The best range of the 3457 is the 3 V range.

AZ is on for all the data i posted in this thread. I'll run some more tests with different NPLC and AZ off at some point.
The only reason i've been concentrating on the 30V range is that i want to monitor some modifications i've made to my EDC and the reference is around 7V.

Drift of the reference and the 7 to 10 V step is different from offset drift: this gives a change in the scale factor, similar to possibly drift of the amplification setting resistors inside U101.

I don't see an easy way to improve on the stability at the ADC / ref section. My best guess would be a temperature controlled fan - this could speed up the time it takes to get a stable temperature, and a slightly lower temperature also tends to be positive. However it will only work well in a limited temperature range (e.g. 18 C to 28 C, lower with heater). Also a fan can cause extra trouble, like dust collection and local temperature fluctuations from turbulence.

My attempts with the dividers showed that i'm wasting my time in that area. The reference<> ADC drift is going to have to stay.

I'm more that happy with the results i've had so far

Thank you Kleinstein for your help with this and my previous q111 noise problem