Another installment in the Datron DMM weird issues series.
Happy with having two of these 1062 units seemingly repaired - revived to the point where they power up without errors and give fair readings in all modes - something was still not right with either of them. The readings wobble at almost-regular intervals, in a roughly recurring pattern. This happens even with the filter on.
Neither unit has been calibrated and I expect readings to be off, but wobbling/jumping of multiple lower digits isn't on my list of expected behaviour for a precision DMM. This is my first experience using these things so I could not make direct comparisons except between the two units I had in front of me - and they both do the same thing.
I confirmed the problem using a precision 10k resistor in ohms mode and a precision DC source configured to 50.000mV. Both cases wobble by about the same relative magnitude. This suggested an issue somewhere on the analog board or analog-digital interface, which is known to contain ...interesting design choices.
So I set about testing the opto-isolators used in the digital-analog pathway as these have been reported as a recurring issue. Some trickery has been used to perform logic using opto-isolators with known CTR differences to produce something like gate delays for correct combining of signals. Other threads (and pages) cover this in more detail. The optos are known to degrade, changing their timings and messing up signals.
I tried to characterize the ones most involved in this process (M1 .. M4,M5,M6) and came to a few conclusions. Those marked white are 'fast', or high CTR. Those marked red are slow, or low CTR. In at least one case resistor values are changed to set timings outwith the CTR alone. Fun stuff. Basically if you plug in a new device without measuring it first, you might just make things worse.
All the new devices I had ordered were just slightly better than 'slow' - ok for slots marked red, but not suited for use in the slots marked white. Fortunately the originals were still good enough and all produced similar measurements.While debugging, I tried to use the original optos, arranged as they were when the unit arrived. Two of the original optos were faulty/dead so I used the replacements for those slots - the rest remained original.
After mucking about endlessly with different patterns of optos, I concluded that:
- its easy to mess things up by using the parts with the wrong spec. typically the unit will report bad readings or just show an error.
- they were not involved in the problem I was debugging, having no effect on the jumping values
- the optos were good enough to produce the correct signals for the null-detection events described in the service manual and nothing I did with the optos improved on that situation. it either continued to do the exactly the same, or didn't work properly at all.
So I decided to give up on that exercise and leave the optos alone. I'll measure them more properly another day but for the unit under test at least, they're ok.
I then hooked up my scope to monitor the A2D integration circuit...
...to view the waveforms described in the service manaul:
The scope showed the expected signal when a voltage is available at the input:
However the waveform was not stable! The spacing between events in the integration cycle was varying erratically.
I indicated this variation in red on the diagram above.Here are two images superimposed, capturing the integrator exposed to a constant, precision input voltage:
Anyway the erratic movement of the integration period and the jumping of the lower digits on the display readout seemed to be related, both apparently cyclic and similar period range. It's also pretty certain that a change in integration period is going to change the measurement.
Something funny going on with the timings driving all this stuff.
I checked voltages around the board looking for ripple and other effects. Nothing apparent.
I did the usual hand-wavey routine of replacing caps near sensitive areas and playing with decoupling capacitor sizes etc. Tried different caps at the 40v DC convertor etc. etc.. None of this really made a dent in the problem. So having got bored with that I went back to debugging it properly.
Using the service manual & schematic, I followed various signals back from the integration circuit to the Ferranti chip that handles the A2D logic and found the wobble present on most of those signals. This was interesting. I followed this as far back as the digital board.
At this point I started looking at the 6800 CPU clock and found that to be erratic as well...
The manual has a very small paragraph on 'line locking' which describes a the rejection of mains 50/60hz from measurements by phase locking with the mains transformer. That got my attention.
The schematic for the line locking circuit shows that it is primarily a digital circuit, with a small analog portion which tweaks the master clock. Sticking probes on various parts of this circuit presents a confusing picture of clocks masking other clocks, counters, dividers and so on. Not very straightforward.
Without bothering to figure out the details of how that mess compares the line and reference clocks, it is already clear the analog part takes output from that comparison and uses it to drive a kind of variable capacitor, implemented with varicap diodes and some other sensitive stuff, which in turn sets the frequency offset for the clock, which in turn drives the 6800, the Ferranti thing, the A2D period and everything downstream from there.
To confirm I was in the right area, I stuck a capacitor across the 10M resistor across the varicap diode. This effectively removes the variable cap and stops the line locking.
The master clock stops wobbling. The integrator stops wobbling.
At this point I notice the adjustable/trimmer coil on the board, and in the line locking schematic. I usually never touch those without a really good reason, and this seemed like a really good reason. I turned this perhaps 120' clockwise and watch the 6800 clock change from jumpy through a slow ramp, to a dead stop. A bit more adjustment finds the ramp will reverse direction. So I find the midpoint between these two extremes and put the lid + shield back on the DMM.
A few measurements later, the jumpy digits have gone. Readings are rock solid in all modes, all ranges.
I won't be surprised to find this is all documented properly in the appropriate part of the calibration guide. Which I haven't even started reading properly. But I guess solving it that way would have been less fun...