Datron 1061A 6.5 digit Multimeter repair
or just a case of CTR degradation
Love at first sight
Some time ago, I stumbled upon that beauty:

(click on image for larger view)
Beeing fascinated by the warm orange glow display, I just couldn't resist. The advertised picture showed the
display in real good and uniform shape, the self test result was a trusty "PASS". So I drove up to the sellers shop,
did a quick test, looked fine. After having done that, I bought it, stuffed into the trunk compartment and had a
nice weekend holiday in Duesseldorf. By the way, thumbs up for that really nice and helpful guy (on that day, he
kindly gave me some help to fix my TDS580 scope I bought from his shop).
Some bonus points: The offer included a full (paper) copy of the service manual, including all schematics and parts lists.
I took the challenge and made a scanned pdf (my very first try to scan such a manual, not the best result, still room left for
improvement). Get it here: Datron 1061A calibration and servicing handbook
Calibration required?
Having noticed a small but notably outside the specs deviation of the reading (reading is about 150ppm too high in comparison to 30ppm specified for the 10V range),
I started some investigation on how to calibrate this unit. Owning the manual, it's a rather simple job in case you've access to a
suitable calibrator. Since I don't have one of these beasts in my home lab, I refrained from this thought.
Resolution step of 10 digits? WTF
So I did some other experiments and investigation, resulting in the discovery of a rather unexpected behaviour of the multimeter:
When measuring a positive voltage and varying this voltage in very small steps of one digit, the display resolution was about 10 digits.
The reading jumped in amounts of 10 digits, not 1 digit as one would expect.
Doing the same thing with the same voltage applied in reverse (measuring negative voltage), the result was as expected. The reading
jumps in steps of one digit.
So something must go subtly wrong here.
Read and understand the manual
Reading the manual, I found out they're using an interesting strategy for the A/D conversion with one detail:
The 4-slope integrating converter uses a slightly different control scheme and different reference voltages for positive and
negative inputs.
Bingo! Thats the point to start troubleshooting, something must go wrong with positive input voltages while negative voltages work
as expected.
 
Hunting down the failure
Open the manual and go to section 3.2.3 to read about the operation priciples of the A/D conversion.
 
So I hooked up the oscilloscope to various test points around the reference multiplexer, the integrator and the null detectors
(yes, there's two of them). Datron did a really good job in providing easily accessible test points:
There's the complete integration cylce on the scope:

yellow: TP3 (2nd Null Detector Output)
cyan: TP5 (Integrator Input)
magenta: TP4 (Integrator Output)
The "Bias" and "Ref2" periods are very short, one can't see them in the integrator output, the "Bias" period is visible as the narrow positive spike in cyan.
Found the integrator slope looking like the timing diagram (fig. 3.13) at TP4, but missing the final Null at
TP3 while measuring positive voltages. Checking the waveform at TP9 revealed a subtle difference between measuring positive and negative
inputs:
Measuring negative voltage: The analog multiplexer (M35) switches at the first Null from Ref1 to Ref2 within a few microseconds, the integrator runs for some 10 us on Ref2 upon final null detection.
Measuring positive voltage: The analog multiplexer (M35) takes some 10us too long to switch from Ref1 to Ref2, resulting in overdriving the
first Null Detector and no second null detection
Watching the converter working while varying the input voltage in steps of approximately one digit reveals: The missing last digit for
positive inputs correlates exactly with the missing second null detection. Gotcha!
See the missing second null detection:

yellow: TP3 (2nd Null Detector Output)
cyan: TP5 (Integrator Input)
magenta: TP4 (Integrator Output)
The first (left hand side) picture shows the missing pulse, while the second (right hand side) shows the correct ending of the integration cycle,
having two pulses at the null detector output. Clearly visible is the longer "Ref2" period at the cyan trace.
Who's to blame now
Having discovered the root of the failure, I now to follow the way of the null detection signal across the isolation barrier through some
discrete digital logic back to the control inputs of the multiplexer. Watching the "a" input (fig. 3.14) and the null detection signal
clearly showed: There must be some unintended delay along this path, interestingly for the falling edge only (of the TP3 signal).
Watching the input and ouput signals of the involved opto-isolators (M1 and M4) quickly showed M1 having a way too slow transition from
"off" to "on" state, causing a greater than 10us delay of the "a" multiplexer input. For correct operation, a delay of less than 3...4 us
is required here. The output waveform of M1 looked like your typical slowish exponential RC charging waveform, not like a digital signal edge
as one would expect from this kind of photocoupler.
Here's the good / bad comparison:

yellow: TP3 (2nd Null Detector Output)
cyan: TP5 (Integrator Input)
green: Collector of Q3 on the digital board, amplified null detector signal past the photocoupler
On the left the output of the bad coupler, right side shows the same signal using the new coupler.
The green trace was acquired at the collector of Q3 here:
 
So the conclusion was: this photocoupler has a serious case of CTR degradation from its original specs, causing its output to creep slowly
from "off" to "on". Replacing the part solved instantly the issue. All integration and null detection waveforms looked just as expected.

Happy End
Having replaced the photocoupler (which was a jellybean 6N136, having some of a similar kind - TLP759 in the junk box), I checked the
reading against my good old trusty HP34401 and the recently acquired Keithley 2015: The reading is inside the expectable limit for this
particular test setup.

Thats a 44ppm difference, quite a good match for a DMM just repaired and of unknown last calibration date.
Not yet :-(
Two days later, I did some linearity checking (measuring voltages of 10V, 1V, 0.1V, 0.01V, 0.001V, both polarities, all in the 10V range), that showed a
constant offset of 21 digits, except for 0V which reads zero. Appears to be some mis-counting of the converter circuit, maybe caused
by photocoupler propagation delay time. If one looks at the scope traces of the M1 input and output signals, the second null detection
pulse looks quite noisy, maybe there's something wrong here.So there's some investigation left to do now.
Another one bites the dust
First, I had a closer look at the noisy null detector waveform (yellow trace), measured at TP3:

Interestingly, that waveform at the second transition from +8V to -8V looked very similar each time the scope triggered, not just random noise.
Looking at the second null detector schematic:

The fast comparator (M15, an LM311) doesn't have any hysteresis build around it and is a quite fast one. Opposite to my opinion of how to build a
fast acting comparator, C3 provides some negative feedback around the comparator. My guess is: this kind of feedback increases linearity of the
circuit around zero input, for the expense of these small spikes near the trigger point. The photocoupler does a quite efficient filtering of these
spikes as they are too short to propagate. So I considered this waveform as intended by the engineers and moved on.
There are some more 6N136 couplers: M4, M5, M6:

These couplers isolate the multiplexer control signals on their way from the digital board to the analog board. Since the integration results of
the A/D converter not only depends on the reference voltage but also on the timing of the multiplexer switching the various voltages to integrate,
the propagation delays of these couplers have influence on the result. According to the schematics and parts lists, these are specially selected
parts (marked with white or red dots). I didn't find the selection criteria, but basically there are two of them: current transfer ratio and
propagation delay. Both parameters (among others) vary over production lots and I suspect them having influence on the accuracy of the A/D converter
circuit here.
So I started to measure the propagation delay of these three photocouplers:

There's a difference of about 200ns between M4 and M5 (both beeing binned "white"). So I suspected M5 to be too slow - but take a closer
look at the schematic: They have different resistor values for the LED current and the photodiode bias, deliberately making M5 slower.
Swapping M4 and M5 confirmed that, so I consider them OK.
Look at the third scope waveform: This is M6, having a very slowish fall time at its output (TP10). Looking at the resistor values, one would expect
a waveform rather similar to M5. Swapping M6 and M5 confirmed M6 beeing slow. But I didn't know what Datron's selection criteria for "RED" were or even mattered at all, anyway:
Neither M1 nor M6 didn't have a color spot at all. Beeing somewhat confused now, I put M6 into M1's location (the one that I found degraded before) to
discover M6 beeing even more degraded than M1 was.
So I replaced M6 and M1 again, using non-selected but similar looking new 6N136s. The propagation delay waveform at M6 looked now plausible (like M5), and M1 also was
operating as expected (including the noisy waveform). Checking the 10V range linearity, there was still more deviation than specified, but changed to slightly lower deviation.
Autocal to the rescue
This multimeter is called "Autocal" by Datron, which means: Most of the calibration is done in software. Nothing unusal today, but I believe quite revolutionary the days back then.
Following the manuals post-repair instructions, I adjusted R20 (null comparator offset), did the "LIN" autocal routine, then adjusted R23. The result: All linearity checks are
now within the specification. Now a complete re-calibration is required for the instrument. For now, I did this in some rudimental way for the DC ranges, the results are promising,
still leaving me to check and re-calibrate if necessary the Ohms and AC ranges.
Bag of tricks
Finding the faults inside the A/D-converter and analyzing the circuitry was a quite interesting thing. The engineers used a lot of tricks and compensation
techniques to squeeze the most out of common off-the-shelf semiconductor components. In comparison to comparators and amplifiers that are available today, these components
are rather lousy (which is a correct technical term according to Bob Pease). I'm a bit concerned about the many photocouplers found in this design, because they are
known to drift in their parameters. Especially the propagation delay issues associated to the 6N136 could have been solved more elegantly by using inductive coupling techniques,
as seen in other instruments, like the famous Fluke 5100B.
There's an interesting repair report (written in German language) at amplifier.cd Datron 1061 repair,
having a similar experience with one of the linear photocouplers used in the multimeter. CTR degradation of photocouplers is a known
failure mode, if you ever made off-line switching power supplies you might have experienced dying feedback photocouplers causing the
control loop to run wild and finally the supply to fail with a bang. Luckily the long term stability of photocouplers is greatly improved today and
smps control loops usually take care of this failure mode, preventing at least the bang.
Back to Wunderkis.de (German language only)
