Solartron 7061 noise

[grizewald]

Dear metrologists, maybe you can give me some insight into something I discovered recently...

I'm a big fan of Solartron's multimeters, what with them being an excellent example of good British engineering and the fact that I hail from the UK, despite the Swedish flag under my user name. I recently found a 7061 meter, complete with all the original measurement cables, on eBay and bought it without a second thought. (Just the three Fischer connectors on the measurement cables are probably worth what I paid for the meter.)

When it arrived, I started doing some testing to compare the 7061 to my superbly stable 7075 meter so that I could get a feel for what kind of condition it was in. One of the first tests I did was to connect the 7061 to my Muirhead Weston cell to get a feel for initial accuracy and see how stable it was. The result of that test was both good and bad. The good part was that the meter read to within 20μV of what the temperature adjusted voltage of the Weston cell should be. The bad part was that I observed a +/- 3μV variation while measuring the cell.

My 7075 just sits there with even the last digit completely stable when measuring the Weston cell, so I was quite surprised to see that a more modern meter was so noisy by comparison.

Looking at the warmed up 7061 with my thermal camera gave me this image:





Most of the heat is coming from the transformer which uses the case as a heat sink.





One big difference between the 7075 and the 7061 is that the 7075 has its reference diode in an oven to keep it and its leads at a constant temperature. In contrast, the reference diode in the 7061 is left naked and raised up from the PCB at the rear right of the analogue board. (D301A in the next picture.)





Looking at the heat distribution, it's more than likely that there are air currents from convection rolling around inside the case. Surely those currents will also mean that the reference diode is subjected to local thermocouple effects due to these air currents creating temperature gradients on the body and legs of the diode? I know that the Zener current is hand trimmed on each meter to coincide with the zero TC point of the diode, but that's not going to help against local temperature gradients is it?

Although I don't have my GPIB interface completed yet so that I can log readings from the 7061, it occurred to me that the 7061 has built in statistics functions that I could use to look at how unstable the meter actually is. I therefore enabled the MaxMin and Statistics programs to get some more scientific results than just observing the meter's readings over an hour or so.

I also decided to use some cotton to shield the reference diode from any eventual draughts and see what effect that had.





I made two runs, one with my little cotton tea cosy and one without. The meter had been on for more than 48 hours when I ran the two tests, one for 1.5 hours and the other for 1.75 hours. During the test, the measurement leads were shorted and I also turned off the filter function to better expose the meter's noise. I left auto-ranging on as the meter is going to stay at the lowest range anyway with shorted inputs.

The results were very interesting!



              With cosy        Without cosy
MAX      3.166496e-7     2.495944e-7
MIN      -3.241002e-7    -5.178153e-7
PP         6.407499e-7      7.674098e-7
MEAN  -1.678598e-8    -1.631235e-7
SD         6.158931e-8     1.010704e-7
VAR       3.793245e-15   1.021523e-14
RMS      6.383583e-8     1.918972e-7


There would seem to be an improvement in noise of a whole order of magnitude with my cotton draught shield.

Far be it from me, as a simple hobbyist, to question the Solartron designers, but surely there's something wrong here? Having the reference diode in the rear right of the case with a roasting hot transformer creating constant air currents inside without any protection from those air currents would seem to be guaranteed to cause noise in the reference voltage.

Am I missing something here? The figures would seem to speak for themselves as to how effective my little cotton pad is. Any input from the knowledgeable and experienced volt-nuts here would be much appreciated!

(edit: I got the column heading swapped and it looked like my cotton pad made things worse rather than better!)

[Kleinstein]

When measuring a short, the reference should have very little effect. Noise from the reference is multiplicative and thus not important near zero.

At the highest gain the limiting factor is usually the input amplifier. From the pictures this looks like a LDR base chopper stabilized amplifier.  If this i still using old style neon tubes, this could be a source of trouble.

The first part would be to check if the variations are more with the input amplifier, the ADC or the reference. All 3 are possible in case of the 1 V reference (I assume a 1-2 V range).

The values from the short test look similar and if the units are volts, the noise looks OK. There are some details no given for the test, but as far as I see it the ADC and amplifier look kind of OK so far. So it would point to a problem with the reference (or maybe the clock, of input current noise from the amplifier). However the test with a short says nothing about the reference stability.

[grizewald]

When measuring a short, the reference should have very little effect. Noise from the reference is multiplicative and thus not important near zero.

At the highest gain the limiting factor is usually the input amplifier. From the pictures this looks like a LDR base chopper stabilized amplifier.  If this i still using old style neon tubes, this could be a source of trouble.

The 7061 uses the classic ICL7650 chopper amp. You're thinking of the 7075 when it comes to the LDR/LED chopper.

The first part would be to check if the variations are more with the input amplifier, the ADC or the reference. All 3 are possible in case of the 1 V reference (I assume a 1-2 V range).

The values from the short test look similar and if the units are volts, the noise looks OK. There are some details no given for the test, but as far as I see it the ADC and amplifier look kind of OK so far. So it would point to a problem with the reference (or maybe the clock, of input current noise from the amplifier). However the test with a short says nothing about the reference stability.

The units are indeed volts, but my question is really about why the Solartron designers put the reference diode flapping around in the breeze inside a case with obvious thermally induced air currents without giving it some kind of protection against those air currents. I'm sure I'm not smarter than the people who designed the meter, so I'm at a loss to understand why my simple cotton draught protector lowers the mean of the zero volt measurement from -1.631235e-7 to -1.678598e-8?  (I've edited my original post as I had the column headings the wrong way around).

[Kleinstein]

I can understand that the unshielded zener looks odd. Some thermal shielding could help to get a little less noise at the reference. However the zener itself already has some noise, so the thermal effect may not be so bad if the current is adjusted well and thus the TC low.
The zener diodes run at a relatively high current and thus get quite some power. So much thermal insulation alone may not be a good solution.

I would consider it some random effect that padding the zener effects the average offset seen. There normally would not be much effect expected and the actual offset very much depends on the last adjustment / zero cal. So not sure which value is better. The offset value has not much to do with noise / variations seen, when measuring the 1 V ref.  0.1 µV difference in the offset has not much in common with 3 µV_pp variation.

The measured noise with a short looks not too high (some 0.7 µV_pp) for the lowest range. So the amplifier is likely not the culprit.  The ADC noise is relatively easy to test, with a short in the higher range ((10/20V).
One could also check if the noise is more white noise of has significant 1/f parts.

[martinr33]

It isn't quite a whole order of magnitude - more like 30%. It just happened to cross the decade line. 

I would check out those few electrolytics. I have a similar vintage (Datron 1281) machine that got a lot quieter when I replaced the electrolytics.

[grizewald]

I finally got my GPIB interface up and running, so I connected my Weston cell and logged the voltage for 24 hours.

The results (without anything on the diode) show clearly that I was chasing ghosts. I've plotted the graph below to include temperature as the Weston cell is a very good thermometer. I think there are a few jumps in there due to the meter's drift correction every 15 minutes, but the drift correction is so small that it's hardly relevant.

[Kleinstein]

The jumps due to the drift compensation still look relatively large compared to the normal noise. Intuitively I would have expect the jumps to be not larger than the normal noise. So there could be a slight problem.

[grizewald]

Looking at the raw data, the jump that happens at 22:00 goes from 1.0183910V to 1.0183926V, so 1.6μV. Maybe that is excessive. What would you say?

It's also clear from the raw data that the jump comes from a drift compensation as the time delta between the two readings is 28 seconds instead of the 8 seconds that a 7 digit reading with filtering on normally takes.

[grizewald]

Extracting all the lines from the raw data where the difference between two readings is more than 1μV confirms that each one happens when there's a drift compensation as the time delta between the two readings is always around 20 seconds longer than expected.

VDelta 1.1e-06
2020-02-16 16:29:43,24.17,40.53,983.69,+1.0183882
2020-02-16 16:30:11,24.17,40.53,983.69,+1.0183871

VDelta 1.6e-06
2020-02-16 21:59:54,24.00,40.58,983.98,+1.0183910
2020-02-16 22:00:22,23.99,40.61,984.00,+1.0183926

VDelta 1.3e-06
2020-02-16 23:44:58,23.92,39.94,983.66,+1.0183939
2020-02-16 23:45:25,23.92,39.94,983.66,+1.0183952

VDelta 1.6e-06
2020-02-17 02:45:04,23.82,38.76,982.96,+1.0183990
2020-02-17 02:45:32,23.82,38.76,982.96,+1.0184006

VDelta 1.4e-06
2020-02-17 06:00:05,23.76,38.15,984.77,+1.0184018
2020-02-17 06:00:32,23.76,38.15,984.77,+1.0184032

VDelta 1.2e-06
2020-02-17 07:15:07,24.00,38.99,985.61,+1.0183979
2020-02-17 07:15:35,24.00,38.99,985.61,+1.0183967

VDelta 1.2e-06
2020-02-17 12:15:16,23.89,37.10,988.15,+1.0183956
2020-02-17 12:15:44,23.89,37.13,988.18,+1.0183944

VDelta 1.1e-06
2020-02-17 13:30:11,23.86,37.15,988.32,+1.0183945
2020-02-17 13:30:39,23.87,37.13,988.18,+1.0183956

I'm sure I remember reading something about a bug with Solartron 70x1 meters and drift compensation, but I can't for the life of me find the thread.

[Kleinstein]

1.6 µV is quite some jump for 7 digit readings (0.1 µV resolution ). As it takes some 28 seconds instead of 8 seconds it would be an extra 20 seconds to do the drift correction, so probably 2 readings + some waiting. I would expect lower noise than 1 µV, but maybe I am wrong with that and hope for too much. Compared to modern meters the old solartrons are kind of noisy.

Besides the noise, there can be a second contribution from resistor / leakage drift at the ADC. So if the ADC really drifts that much the compensation is right with the jumps. However the curve shows times (e.g. 0:00 to 4:00) where the temperature goes steady in one direction and still jumps up and down. If it would be real drift I would expect the jumps to be one direction mainly.

The more normal noise test is with a short at the input, as this takes out the source (a weston cell may show some noise and react to current noise from the DMM) and the DMM reference.

There was some trouble with the zero adjustment due to not waiting long enough for dielectric absorption to cam down. This was with the 7071/7081 however. There could be a similar problem here - but hard to do anything about this.

p.s. Having 8 seconds between readings is suspicious: could it be that the meter is still running in a kind of 60 Hz mode ?

[grizewald]

p.s. Having 8 seconds between readings is suspicious: could it be that the meter is still running in a kind of 60 Hz mode ?

No, it's correct. The manual states 2 seconds for a 7 digit voltage measurement, multiplied by 4 when filtering is enabled.

You're right to draw attention to the direction of the jumps. The one at around 02:45 is particularly odd where there's a downward drift of around 0.8μV before the sudden jump up of 1.6μV from where it slowly drifts down again to about where it started, despite the fact that the temperature is dropping and the Weston cell's voltage should be increasing.

I will run some more tests with shorted inputs on the 10V range and see if that gives any more clues as to what might be going on here.

Thanks for your time and thoughts!

[grizewald]

I thought it was interesting that the manual contains the following text about drift correction:

DRIFT CORRECTION
The instrument automatically compensates for internal drift every
15 minutes but does not interrupt a GPIB input or a keyboard trigger.
Under remote control, drift correction may be turned on and off.
For example, in tracking measurements lasting longer than 15 minutes,
drift should be turned off for continuity of results.

I am currently running a log with shorted inputs on 10V range with filter on and drift correction on. I will let that run for a few hours and then repeat with drift correction off. Results later!

I also double checked the mains frequency switches and can confirm that they are set for 50Hz.

[grizewald]

This morning I turned on the 7061 and let it warm up properly from about 09:00 to 12:30. Then I shorted the inputs and captured data in two sessions. Both sessions had filtering off and drift compensation off. The first session was with the meter on the 10V range (which only gives 6 digits after the decimal) and the second session was on the 1V range for the full 7 digits after the decimal.

I have to say that I'm mystified by the results.







The log from the 10V range shows a maximum variation of close on 12μV peak to peak, which is horrible. The 1V range shows only around 1.2μV.

If I'm using the 1V range then I should be seeing the amplifier noise together with the reference noise, the 10V range should only be reference noise. I'm confused!

[Kleinstein]

The noise seen with a short in the 10 V range is mainly from the ADC. the reference would be only a very small contribution, if at all.

The 1 V range has some analog amplification x 10. So noise of the ADC is effectively reduced by a factor of 10. So the 1.2µV seen in the 1 V range absolutely make sense. As it looks this is still to a large part noise from the ADC and not yet much from the amplifier.
Looks like it needs the 100 mV or even 10 mV range to really see the amplifier noise.


Chances are that the slow variations seen are actually drift from the ADC, at least for the 10 V case.
These variation look not that small. So the drift compensation could be actually chasing true drift / low frequency noise. The meter internal temperature may be different from the Weston cell.

I looks a little like some of the "noise" may actually be a very low frequency AC part. This makes me think, that the synchronization with the grid may not work. A not so well working clock PLL is a known problem with some of the Solartron meters.

Getting 12 µV peak to peak noise in the 10 V range is not that bad. This is something like 2 µV_RMS . This is not that bad for such an old meter. More resolution would need a little more filtering or longer integration. This can be more effective than just the white noise case, as this effectively extends the integration time and reduces the quantization noise.

[grizewald]

I'm having a little difficulty understanding this. The ADC is comparing the input voltage to the reference voltage and producing an output which is the difference between the two. Correct?

If the input is almost zero and the meter is not amplifying the input (the 10V case), then the comparison is only involving the input voltage and the reference voltage. If the input is being amplified by 10 before comparison (the 1V case), how does that reduce the ADC noise by 10?

...

Thinking about this while I'm writing it, if the input signal is almost zero and we amplify if by 10, we're increasing the noise on the input by 10. So we end up comparing an input voltage with ten times more noise than the unamplified version with the reference voltage, therefore making the reference noise ten times less significant. Right?

I've read about the problems some of the Solartron meters have with locking the PLL to the mains frequency and I'll try to look at how well the PLL is working on my meter.

Thanks again for giving me the benefit of your experience!

[Kleinstein]

The ADC is measuring the ratio of the external voltage to the reference.  So external voltage divided by ref. voltage.  Near zero the actual size of the reference voltage does not matter and thus reference noise (at least the usual low frequency one) does not matter. At zero reading the ADC would still get near zero result even of the reference voltage would double or is cut in half.
It is additive noise of the ADC itself that matters, not noise of the reference (that would be multiplicative).
Noise of the reference can get relevant if a significant voltage (e.g. more than 1/2 the full scale) is measured. The reference noise will show up with as   "reference noise" * "measured voltage" / "nominal reference voltage", so no reference noise at zero measured voltage.

The displayed voltage is for the input signal. So if there is a gain of 10 before the ADC the result from the ADC is divided by 10 to get the voltage reading.  So noise of the ADC would show up lower by that factor of 10.

[grizewald]

Thanks. Now I understand.

I'm only just starting out in my journey down the volt nut rabbit hole, and it's not always obvious how things relate. Your patience is much appreciated!

[grizewald]

So, I'm still looking at the noise levels on the 7061. Plotting a half hour with shorted inputs on the 10V range with filtering and drift correction turned off gives me this:



Pretty much the 12μV noise previously observed.

As speculated on earlier in this thread, most of the noise would seem to be coming from the ADC. The 7061 uses the original Intersil ICL7650 chopper amp, so I ordered a new ICL7650S, which is the "improved" version of the original and also ordered an LTC1052 which is a direct replacement. The chopper amp is socketed, so testing the alternatives is easy.

First off, I installed the ICL7650S and took a ~30 minute sample to give:



It's a small improvement, with p-p noise reduced to around 9μV p-p worst case and as low as 5μV at times. I also captured a much longer time period which has some strange results which I'll show later.

Then I switched to the LTC1052 and made a similar ~30 minute plot:



Worst case noise here is about 8μV p-p and as low as 4μV at times, so another small improvement. I also captured a longer time period log as well.

Looking at the longer log for the ICL7650S, it has some interesting features:



There is an overall negative drift over the approximately 8 hours that the test ran from +8μV to -23μV for a total span of 31μV. There are some inexplicable 30μV bursts of noise starting at 07:00 which I've not included in the overall drift. I don't remember hearing anyone in my apartment block using power tools around that time, but then again, I was still asleep.

The longer term plot for the LTC1052 looks like this:



In this case there's a drift from +14μV to -7μV for a total span of 21μV.

The overall noise for both seems to be around the same at about 8μV, but the LTC1052 seems considerably more stable. I might be able to win 30% less noise and a boost in stability by using the LTC1052, but I suspect that this change will affect the meter's calibration, so I'm currently looking at re-measuring some standards and comparing the results with the readings I took just after I had the 7061 calibrated.
 

[grizewald]

Switching the chopper amp seems to have had negligible effect on the meter's calibration.

I have a newly built KX reference and with the meter's original chopper, right after I had it calibrated, the KX read between 7.134501V and 7.134508V. (I didn't note how many readings that spread comes from)

Now, with the LTC1052, I'm seeing 7.134509V to 7.134516V over 85 readings for an 8μV difference to the previous readings.

That's well within the meter's specified 24 hour calibration stability of +/- 5ppm of reading + 6 counts on the 10V range in 7.5 digit mode, so I think I'll keep the LTC1052 fitted.

The next question is where might the rest of the noise be coming from? Could the 7.5V zeners which clamp the + and - supply voltages to the chopper from the +15V and -15V rails be a problem? Or maybe the LM301 which follows the output of the chopper amp? (That's not socketed, but it wouldn't be a problem to add one.)

[Andreas]

Hello,

those "Bursts" look familiar to me.
I also had those together with my LTC2400 based ADCs.
Obviously some interference of the chopper frequency (LTC1050 in my case) together with LTC1043 switching frequency and/or LTC2400 sampling.
In my case the whole is temperature dependant (reproducible).

With a different OP-Amp (LTC2057) the noise amplitude is less.

with best regards

Andreas

[grizewald]

Interesting. That's quite a "pop"!

Looking at my plot when the noise occurred, I can see that the temperature was very stable for the hour leading up to the sudden 30μV jumps. The readings (2 seconds apart) are hovering around the -12μV mark when suddenly I see:

-0.000016
-0.000014
-0.000023
-0.000031
-0.000020
-0.000000
-0.000024
-0.000012
-0.000015

This repeats several times.

Maybe it's just a feature of the "improved" ICL7650. I'll keep an eye out for the same behaviour with the LTC1052 that I have in there now.

[Kleinstein]

The LM301 may contribute to the noise. A possible upgrade would be something like LF356, AD711 or OPA134. There is no real need to go all the way to a OPA140. With a JFET OP one would likely also want to increase the value of R209 - this resistor can also contribute to the noise and a higher resistance value (e.g. 50K) can help.

However it looks like one is already at the point where the quantization gets visible. So there is a limit to possible improvements.

One odd point is that I don't see supply decoupling at the AZ OP.

[grizewald]

There doesn't seem to be any decoupling on the rails for the chopper amp on this part of the circuit diagram or elsewhere.
Maybe it's something to do with the floating nature of the analogue circuitry. 

[iMo]

Solder a 100nF/25V ceramic between the Vcc and Vss pins of the chopper then..

[MiDi]

There doesn't seem to be any decoupling on the rails for the chopper amp on this part of the circuit diagram or elsewhere.

I have seen that on a couple of designs and asked me everytime if there is any disadvantage of local decoupling?

[Kleinstein]

More decoupling capacitors could in theory cause RF resonances between capacitors if the conditions are bad. In this case a few ohms in series with the supply would add damping and essentially avoid resonances. In this case the zener diodes should have enough resistance.
With some chips the lack of local decoupling can slow down switching. Sometimes this helps, but this is an exception.
Adding decoupling is at least one of the few mods to try.

[grizewald]

I will certainly contemplate putting a cap between the two rails right on top of the chopper amp, but for the moment, I'm looking at the PLL and oscillator board.

I had performed the very simple check from the manual (measure voltage at test point, if it's 3V +/- 0.2V you're good) but hadn't looked at the inputs to the PLL chip to see if they were actually locked. So last night I got the Rigol on the test points and discovered this:



(ch1 is MAINS_IN, ch2 is COMP_IN and ch3 is the clock control)

It's not exactly what you'd call locked and regardless of the control signal trying to pull the COMP_IN signal back into phase with the mains, the phase difference was fairly constant, which is more than I can say for the frequency with never settled on anything.

So I twiddled the adjustment on the clock board until the PLL managed to lock the two signals, but the result was less than great. It would correct the phase, but the response was horribly slow. If the COMP_IN signal drifted before the MAINS_IN signal, it would produce a negative signal on the clock control to push it back. If it drifted after, it would produce a positive signal to pull it back again. If the phases were locked, the clock control would remain at the mid point. Sounds good, and that's what it's supposed to do, except I'd expect it to take much less than several seconds to re-lock the phases and I'd also expect the lock to stay locked after maybe a minute at most, but no. It would drift forward and backwards all the time and the clock frequency would never stabilise.

So I've now taken the clock board out of the meter and have it running on the bench with the clock control input fixed at 3V. After a few minutes to warm up and stabilise, it's holding stable. Except the frequency is 52.346 MHz instead of 49.152 MHz. So I'll adjust it to 49.152 MHz, put it back in the meter and see if the PLL can now manage to phase lock the clock to the mains frequency and keep it there. If not, I guess it's time to start tracking down an MV2115 varactor diode.

I looked at the circuit diagram of the 49.152 MHz crystal clock option and was very surprised to find no input for the PLL control signal. A crystal is going to give a more stable frequency, but surely it still needs to be phase locked to the mains for proper AC interference rejection?

[Kleinstein]

The mains locking with the PLL is a 2 sided thing. It can improve the mains hum suppression, but it can also add noise, as the mains frequency is not perfectly stable over a short time. The response of the PLL has to be slow - this is intentional, as the frequency during one conversion should not change much. Some deviation in the phase is the smaller problem. The slow response is not easy to get with analog means.

Most modern meters tend to use no mains PLL, but just a suitable crystal clock and than for some modes do a synchronization of the start of the conversion.

[grizewald]

Well, I managed to get the frequency of the oscillator board as close to 49.152 MHz as I could.



Adjusting the frequency really accurately is next to impossible as you just need to breathe on the adjustment slug to make the frequency change by 50 KHz. After re-installing the board in the meter, I'm happy to see that the PLL now locks with a maximum deviation either side of 300ns, which is well within the +/- 3% specification.

The only negative is that it made absolutely no difference to the amount of noise!

I'm starting to think I might be chasing an impossible goal here. I saw that e61_phil's massive 7081 noise thread ran out into the sand with no resolution. Maybe my expectations are unrealistic, but it seems to me that having a 7.5 digit meter where the last 1.5 digits are effectively a random number is kind of pointless. Maybe that's why Solartron added the program feature, so that one can run a series of readings into the stats program and get it to output the mean instead of even looking at the filtered 7.5 digit reading.

[Kleinstein]

The solartron meters have there limitations. They seem to be nice with quite good linearity, but are also slow and limited resolution when doing fast (e.g. 5 sps) conversions.

For the clock, there is the option to use a fixed crystal clock and accept that the integration time is not exactly a multiple of mains periods.
The PLL part is not easy, though there is no absolute need to have an exact in phase lock. Depending on how the PLL is made there is even an advantage if there is some shift, as this avoids some coupling.

[grizewald]

Surprisingly, I can actually buy a VCXO with an output frequency of 49.152 MHz from stock at mouser. That might be a nice upgrade, delivering both the accuracy of a crystal oscillator and the ability to phase lock it with the mains frequency.

It would need a 3.3V power rail and some level translation, but that's not difficult. I might add one to my next order as they're only €20.

[Kleinstein]

The pull range of an VCXO may not be sufficient for the variations of the mains frequency.  It may work much of the time, but could get unlooked at some times.

[serg-el]

Pay attention to this topic.

https://radiokot.ru/forum/viewtopic.php?f=10&t=129145&hilit=solartron
http://www.ko4bb.com/getsimple/index.php?id=download&file=06_Misc_Test_Equipment/Solartron/Solartron_7081/Solartron_7081_Documentation_pack_-_Mickle.T.zip

https://www.eevblog.com/forum/repair/repair-of-solartron-7081-sn718/

[grizewald]

The pull range of an VCXO may not be sufficient for the variations of the mains frequency.  It may work much of the time, but could get unlooked at some times.

The pull range of the VCXO I found is +/- 100ppm or +/- 4915 KHz. What with all the divisions down from 49.152 MHz to get the 50Hz that the PLL compares to the mains frequency, I'm having a little trouble working out if that's enough! I suspect it isn't.

[Kleinstein]

The mains frequency is under favorable condition much of the time within some +-0.01 Hz (+-200 ppm), but not all the time. Worst case can be up to some 0.1 - 0.15 Hz.
The 100 ppm adjustment range for the VCXO would directly translate to +-50 ppm for mains.
I would consider some 500 ppm pull range as a minimum to get a reasonable good mains lock at least most of the time.

[grizewald]

500ppm sounds right and I guess that's why they went for the MC1648 VCO and why the crystal oscillator option isn't a VCXO; having had a good look around, a VCXO with 500ppm pull range doesn't exist!

[branadic]

I'm quite curious if you grizwald still have the LTC1052 in the meter without any issues. Have  you changed the zener diodes? I saw the



the other day and for some reason after replacing the ICL7650 by an LTC1052 his meter released the magic smoke. Seems like some problem of the rails at or of the LTC1052.
I wonder, is it related to his specific board revision 706105O5Y of the analog board only that this happened or is this a general problem replacing the old chopper amp?

-branadic-

[Kleinstein]

May my guess would be something like an latchup event at the LTC1052 - though I don't know why it should happen.  In the video one of the 15 V supplies was a bit high. With only 7.5 V Zeners the OP would get a nominal 15 V, but with only 2 mA supply current the 7.5 V zeners may have less than 7.5 V drop and thus maybe a supply at the upper limit (18 V limit). The LTC1052 has the specs for +-5 V supply and in the circuit the input and ouput signal is all close to zero. So a +-5 V supply (e.g. 10 V zeners) would be the safer way to run the OP.  The LTC1052 may also be not too happy to work without local decoupling, so it may oscillate and draw a much higher supply current that could increase the heat to the diodes, which may respond with to much supply to the OP while very hot.

A point to check may be the regulator at the supply - maybe oscillation or a poor solderjoint to give spikes with more than +-15 V.

The LTC1052 and ICL7650 are very similar - the main difference is in the noise specs. Even the supply current is very similar.
The LTC1052 is a bit slower, but this should not be a big deal: In the circuit the speed is anyway set by the local FB cap. So the OPs speed would not matter much.

[MiDi]

We discussed that already in the ltz1000 thread, where Dr. Frank and branadics version had same issues.
Only the current limiting of the lt1763 saved it from self-destruction after latch-up.
On power rails with higher current limit it is likely the LTC1052 gets damaged.

[branadic]

Played a bit with Solatron7061 this weekend and monitored some voltages. I spotted large hum in the readings, so I looked into the oscillator of the PLL and bingo, it was way off. So I brought it back into the range where the PLL was able to lock and readings are much better now. Figure 1 and 2 (100 samples) were taken with filter turned off. Figure 3 shows the effect with filter turned on, reducing the repitition rate in 7 digit mode to about 7.4 s.

-branadic-

[grizewald]

I'm quite curious if you grizwald still have the LTC1052 in the meter without any issues. Have  you changed the zener diodes? I saw the

Sorry for the late reply, I haven't been on the forum for ages! (Pressure of work, life etc.)

I had the LTC1052 in the meter for about a month and although it did result in about 30% less noise, it changed the meter's calibration.
Seeing as I'd actually paid to have the meter calibrated, sacrificing my calibration for 30% less noise seemed like a poor deal, so I put the original ICL7650 back in.

Defpom's problem looks like an unlucky infant failure to me, maybe precipitated by the slightly high 15V rails as Kleinstein suggests.

[grizewald]

Played a bit with Solatron7061 this weekend and monitored some voltages. I spotted large hum in the readings, so I looked into the oscillator of the PLL and bingo, it was way off. So I brought it back into the range where the PLL was able to lock and readings are much better now. Figure 1 and 2 (100 samples) were taken with filter turned off. Figure 3 shows the effect with filter turned on, reducing the repitition rate in 7 digit mode to about 7.4 s.

-branadic-

It would be very interesting to see what you get if you repeat the burst mode capture that both Dek and I tried in this thread: https://www.eevblog.com/forum/repair/solartron7061/msg3047176/#msg3047176

[branadic]

Why should I use the burst mode instead of taking continuous readings?

-branad-

[grizewald]

To see if you get the same 50Hz noise as we did when the chopper is running at 1.5KHz.  As you can see in the linked thread, the discussion kind of ran out into the sand, but even so, another data point would be interesting.

[TheDefpom]

I had a poke  around my 7061 (original version with PCB3 and 5) yesterday and scoped the oscillator and PLL etc, basically the same results everyone else is seeing, a PLL that never really seems to lock properly, just constantly adjusting its control voltage and the frequency jumping around by huge amounts.

As I have played with PLL's and VCO's in the past in my designs for CB radio conversion boards, I played around with it a bit and even tried tweaking the tantalum capacitor values to see if I could improve the phase detectors control voltage filter characteristics, but with no real luck, I could influence it and maybe improve it very slightly, but nothing fixed the response and control accuracy.

One thing I did do was replace the Philips 47uF 10V electrolytic capacitor at C422, this capacitor was completely dead (I have found that the same type of Philips capacitors in the Datron multimeters are also always dead, so I always have them as a suspect), anyway, replacing that capacitor appears to have improved the noise on the display slightly, it doesn't appear to be  jumping around quite as much as it was, so that might be something worth checking.

I was hoping to dig right into this PLL/VCO system and try to find a resolution to its instability, I will try again in the near future as I am sure it should be possible to get it locked properly, I have some other ideas for things I could try but after spending a few hours on it yesterday I have put the meter back together and moved onto my other projects.

[Kleinstein]

With increasing numbers of SMPS and PV systems in the grid the mains frequency may show more jitter / variations than in the old times when the supply came mainly from large classical generators. So the PLL may have more difficult conditions than at design time.

To get reduche the effect of wavefrom distorions / higher frequency noise on the main waveform, one could consider a little filtering for the mains signal pic off.

For the PLL it may be worth looking at the VCO control voltage an look if there are some dominant frequencies. Ideally this would not be the case, but a poorly tuned loop filter can show some amplifications / tendency for ringing.

The alternative to a PLL is to have a stable crstal clock and accept a lettle less mains hum suppresion.

[branadic]

When I adjusted the VCO on my unit I used a Keysight 53230A in trend chart mode, this way I was able to observe the PLL trying to lock by "sweeping" the frequency. So the frequency I adjusted the VCO to is not neccessarily 49.152MHz, but close enough to get the PLL to lock.

-branadic-