Advantest R6581 8.5 digit DMM mini teardown/repair

[dietert1]

As proposed by Kleinstein i adjusted the preregulator replacing the 6K8 adjustment resistor by 1K8 and the first mains transformer now gets 209 Vac, still in 220 V configuration. All low voltage regulators have their headroom.
I implemented many other mods to the meter received from Ismsanmar, including Pomona copper binding posts at the front and a TFT display. Also the reference board was wired with soldered copper wires replacing board connector. Still want to recap the low voltage supplies and mod the input MUX.
In order to test that R6581T i connected it to our ADR1399 and a Keithley 3706. The log at 6 second intervals (100 NPLC with Autozero + 0.76 seconds for R6581T internal temperature measurement) exhibits a point to point standard deviation of about 3.5E-8. In other words: about 250 nVrms noise at this timing (roughly 0.05 Hz).
Internal temperature of the R6581T is about 39 °C at 20 °C ambient. Ambient temperature coefficient is 0.06 ppm/K for the R6581T, yet there are additional deviations depending on speed of temperature change. About 2 uV shift after turning on heaters with a 2 °C ambient temperature rise within 30 minutes.
Image shows one small R6581T malfunction observed during two days of logging, reason unknown (K3706 noise larger than deviation).

Regards, Dieter

[dietert1]

Meanwhile i replaced the 4x 2K2 input protection resistor by a depletion mosfet pair like shown here:
https://www.eevblog.com/forum/metrology/ultra-low-noise-short-term-stable-references/msg5699907/#msg5699907. At small currents it has about 200 Ohm resistance and it should resist about 600 V.

Observations:
- Up shift of the 10 V measurement of our ADR1399 reference by about 3 uV.
- Instrument 10 V range TC is now 0.41 ppm/K
- Stability after numerical TC correction improved a lot. No more drifts depending on ambient temperature change speed.
- R6581T internal temperature readings are less noisy

According to Mickle T.s schematics the modded circuit is not involved with reading the LM35 internal temperature sensor, so there is some mystery. Diagram shows period before and after mod. The mod plus some hours of R6581T warm-up are hidden at the step near reading No. 6000.
To be tested: Null meter noise of the R6581T after mod.

Regards, Dieter

Edit: The diagram doesn't show ambient temperature (about 21 °C) but R6581T internal temperature.

[Kleinstein]

It could make sense to still have some resistance / PTC in series with the depl. FETs. Under ideal conditions the FETs work well to limit the current, but if they fail, they fail short. So some backup in the protection would be a good idea. ESD spikes may well exceed 600 V and there extra resistors with good peak ratings could help.

One the downside the FET version may be more prone to thermal EMF from thermal gradients. In the slightly different version with enhancement FETs and PV OK to turn them on, I have seen a few µV of thermal drift. Depending on the thermal environent this could also get worse.
Just simple resistors are also not all that perfect in that respect - some may show quite some thermal EMF in a temperature gradient. Ideally the 4 resistors in series should somewhat compensate.

The expeted effect could be a change in the offset drift, but hardly and change in the gain or gain TC.

The current limit part can also effect the charge spike from the AZ switching. In some cases this can be quite some current and the protection resistors can be part of what limits the current.  I don't think the 4x2.2 K would be that relevant for the meters noise - if at all in the 100 mV range (the amps parts uses a different path anyway. The zero path seems to have some 8 K for symmetry. 

[dietert1]

So, in order to complete this mod one should replace R021 8K2 used for AutoZero by a 200 Ohm resistor and the 1500 V gas discharge tubes E002 and E003 by the 500 V version. Anyway the R6581T front panel indicates "300 V max".
For the LM35 temperature sensor one could insert a RC low pass filter that protects its output circuit from the MUX switching spikes. TI specify a 2 Ohm max output resistance up to 1 mA load without telling the bandwidth.

Regards, Dieter

[Kleinstein]

I would use more than 200 ohms with the FET protection, more like 1 K to get a lower peak current. There is anyway quite some resistance from the JFET switches. AFAIK the JFETs in the custom switch hybrid are similar to 2N4117 or 8  and thus quite high resistance (e.g. 1-2 K range each). So there is quite some resistance in the input paths anyway. So no real need to go much lower in the input resistance.

Lower voltage gas discharge tubes would make sense.

[dietert1]

When thinking about the LM35 temperature sensor noise, i remembered that in our other R6581T meters i modded it from 15 V supply to 5 V supply in order to reduce self heating. After implementing the mod, temperature noise increased, as the mod implies lifting the LM35 supply pin (1 of 3 pins) and that caused a higher thermal resistance to the PCB. Then i used a short piece of shrink wrap tube to protect the sensor from a possible air flow and it really reduced noise to almost zero. The resolution of the ":SENS:ITEM?" readout is 0,01 °C and the standard deviation of point to point differences is 0.0068 °C now.
Seems like some air flow from the fan enters the guarded section of the meter!

Regards, Dieter

[Kleinstein]

One can get temperture variations not only from a fan blowing air, but also from oscillations / turbulance from convection flow without a fan.
It can help to avoid extra hot spots (like voltage regulators with relatively small heat sink).

[bsw_m]

I discovered a problem with the voltage reference module connector completely by accident.
The connector soldering looked perfect. The problem was discovered after desoldering the connector and attempting to re-solder it. I desoldered the connector to verify the GND connections on the module.
I am also attaching the corrected module diagram.

[leighcorrigall]

I also came to a similar conclusion:

https://www.eevblog.com/forum/metrology/advantest-r6581-service-manual/msg6077931/#msg6077931

[bsw_m]

During the final stage of connector restoration, blowing it with compressed air—the coating was found to be peeling in other areas. Everyone will have to draw their own conclusions about the problem.

[bsw_m]

Another schematic correction for reference module. Attention to R10. Looks like there's another bug in this device.

[dietert1]

That missing resistor with wrong wiring is a proposal of the LTZ1000 datasheet for TC fine tuning. I think the correct wiring is from the heater variable output voltage to the 120R resistor R1.
In practice it is quite some work to find the correct value for that resistor. 390K or 400K appears to be an average value. The best value (for TC = 0) depends on the thermal resistance of the LTZ1000 mount, e.g. on PCB slots, PVC covers, location of heater drive transistor.

Regards, Dieter

[bsw_m]

I started investigating the reference voltage after measuring its TC, which turned out to be high. This revealed problems with the connector and an error in the connection of resistor R10, which is used to fine-tune the reference voltage's TC.

[dietert1]

Be aware that your LTZ1000 TC is negative (about -0.25 ppm/K) and that most LTZ1000 require compensation of a positive residual TC. The proposed circuit with R10 will certainly decrease TC, as the heater voltage and power will decrease with ambient temperature. There is an alternative linear TC compensation circuit  that many other designs use. It has an about 15 Ohm resistor in series with the LTZ1000 zener and reduces TC as well, as it amplifies Ube of the compensation transistor.

So there may be something wrong in your DUT. Maybe the heater reference voltage divider.

Regards, Dieter

[branadic]

bsw_m, the result of your t.c. measurement looks rather odd and uncommon. I supose that's before you have solved the connector issue?

-branadic-

[bsw_m]

The soldering of the connector could not have affected the measurement, since the voltmeter connection for monitoring the output module was made in place of the not installed C5.

I also provide a new TC graph, after adjusting the connection point of resistor R10 (along with the old graph). TC better but not enough.

[branadic]

Even this t.c. graph looks odd. It would be good, if you'd provide more information on your temperature profile and run a full sweep, not just a sweep from room-temperature upwards, but down to room temperature again and to temperatures below room temperature as well.

-branadic-

[bsw_m]

Even this t.c. graph looks odd.
-branadic-

T.C. graph may be looks odd on start since I don't have access to the thermostat right now, heating occurs naturally after the fan in the device being tested is turned off.

[dietert1]

Can you describe your setup?
Do you use a second meter and/or an external reference?
What is the source of the temperature parameter?

[d-el]

Please tell me what the problem could be. I connect a reference resistor using Kelvin probes, enter external calibration mode, and set the value to 10.00035kΩ.
After the calibration is completed, I press the Home button and see a measured value of 10.00175kΩ. If I enable OHM COMP, it shows 10.001014kΩ.
What could be causing this issue?

I also calibrated the voltage by applying 10.0000331V, and after calibration it reads exactly 10.0000331V.

[Mickle T.]

The external resistance calibration procedure's microcode is relatively complex and involves mathematical processing of eight measurements. However, it seems to me that the search for the source of the problem should begin with the cause of the large discrepancy in measurement results with OCOMP mode enabled. A difference of 75 ppm indicates the presence of a 750 µV zero offset in the circuit on the 10 V range, which is completely impossible for a properly functioning multimeter.

[d-el]

I figured it out. The problem was the huge reactive impedance of my reference standard.
I connected a C5-61 resistor and performed the calibration successfully. The values entered during calibration now match the measured values.
With compensation enabled and disabled, the difference is only 0.0000007 kΩ.

The photo shows measurements of an unsuitable resistance standard and measurements of the C5-61.