Decent benchtop multimeter for beginner?

[tautech]

The reason for there are 3  DMM's is comparison.
This is the idea. TWO meters are stable from ther beginning and ONE not so stable. Not the precission.
SDM3065X is not defect. Is just inferior as design.



P.S. Reference value is 10.000005

And the time scale is ?
us, ms, s ? ? ?

[skander36]

The reason for there are 3  DMM's is comparison.
This is the idea. TWO meters are stable from ther beginning and ONE not so stable. Not the precission.
SDM3065X is not defect. Is just inferior as design.



P.S. Reference value is 10.000005

And the time scale is ?
us, ms, s ? ? ?

I've said before Rob. about 1H45min.

[KungFuJosh]

The reason for there are 3  DMM's is comparison.
This is the idea. TWO meters are stable from ther beginning and ONE not so stable. Not the precission.
SDM3065X is not defect. Is just inferior as design.



P.S. Reference value is 10.000005

Not one of your meters agrees with that reference value.

As has been said, no bench meters are expected to be perfect at power on. If two more expensive meters are more stable at power on, cool. Buy more expensive meters.

One of the reasons the SDM3065X costs less is because the LM399 is not burned in ahead of time. If you want stability faster, put at least a thousand hours on the chip.

Another question I have is regarding the setup. Which meter was tested first? Was the reference warmed up? I imagine your reference also needs warm up time before it's stable. I know mine do.

If your variables aren't consistent, your comparison isn't useful. Give equal warmup time to the reference, and to the meters, and then see what the results are.

[bdunham7]

P.S. Reference value is 10.000005

What is it and how do you know its value to better than 0.5ppm?

[bdunham7]

Not one of your meters agrees with that reference value.

Although the difference is more than I might expect for the Keysight (assuming his reference value is accurate) the Keithley and KS are both within their stated specfications.  The Siglent is not.

As has been said, no bench meters are expected to be perfect at power on. If two more expensive meters are more stable at power on, cool. Buy more expensive meters.

Sure, but being "better" in terms of stability and warmup performance is a much different issue than "meeting specifications" vs "not meeting specifications".  My HP 34401A also has significant power-on drift and is roughly 15-20ppm low at power-up.  But it is still well within specs even though it is 30 years old and has never been adjusted.

One of the reasons the SDM3065X costs less is because the LM399 is not burned in ahead of time. If you want stability faster, put at least a thousand hours on the chip.

That makes them useless or even dangerous (in a liability sort of way) in many professional applications.  See the post by tzaboo earlier.  Drifting out of tolerance before the first calibration interval is an unmitigated disaster in some instances and IMO is completely unacceptable for a professional-grade instrument.  As you say, you get what you pay for (if you're lucky).  However, none of this has anything to do with the particular behavior of this meter and Siglent's calibration practices.  This isn't the first story involving shortcomings on both of those fronts:

https://www.eevblog.com/forum/testgear/siglent-sdm3065x-issues/

Another question I have is regarding the setup. Which meter was tested first?

I'm assuming they were all tested simultaneously in parallel.  If not, then we need some details.

[KungFuJosh]

Although the difference is more than I might expect for the Keysight (assuming his reference value is accurate) the Keithley and KS are both within their stated specfications.  The Siglent is not.

I don't agree, and I won't assume anything, because the stated value of the reference is very likely BS. Not one of the meters agree with it, and how many digits would the meter to certify that value need to be? If we're playing the throw out the last digit rule, then you need an 8.5 digit meter to certify the ref as stated.

If you average the 2 more expensive meters, or choose either of them, and accept that as the ref value, then the Siglent is well within it's accuracy range:

If the reference is 9.9998 then 9.9998 x 0.9985 = 9.9848. Or 9.9999 source voltage would be 9.9849. Both of those would be within threshold, no?

ETA: Let's pretend that an 8.5 digit DMM meter was used to certify the ref, and that the DMM was also traceably calibrated. What conditions were required to meet that value? How long was warmup time for the ref and the meter? What was the temperature and humidity? Were any of those guidelines followed here? We know the answer is no, they were not, so the best we can do is average 2 or 3 of the 6.5 digit meters available to find a reasonable value.

[bdunham7]

If you average the 2 more expensive meters, or choose either of them, and accept that as the ref value, then the Siglent is well within it's accuracy range:

I'm perfectly OK with not assuming the reference is accurate, but even if you opt to believe that the Keysight is exactly accurate--an assumption that would most benefit the Siglent--there's still a ~50ppm discrepancy in the Siglent readings after the 1:45 warmup time.  That's not within spec, is it?

Now if you assume the Keithley and Keysight are reading high near the top of their tolerances and the reference is actually much lower (say 25ppm below 10V) then the Siglent might be in-spec albeit still quite drifty.  However, that's what I think Scott Adams refers to as "pretzel logic" and I think the answer here is very simple:  The meter has excessive power-on drift (perhaps due to a TC or TC compensation issue) and it was calibrated cold.  This is very similar to what happened to KG7AMV in the thread I linked.

Hopefully skander36 will fill us in on the details of the reference and whether these meters were all tested simultaneously.

[skander36]

Josh, you write a lot of personal assumptions (AKA BS).

The point is not he performance but a comparison between them.
The meters was tested in the same time. As KS and KT was able to keep a reasonable reading, so Siglent should stay aside. If not in my opinion is inferior. You may have your personal opinion of course.
-No, none of the meters doesn't show the actual value of the reference, they can't as they are 6.5 digit and this value is from a 3458A used for calibration. Actually the real value is 10.000001 after adjustement. 10.000005 was before (after 1 year and 8 month from fabrication) see picture from adjustement.
The drift will eventually stop after 3 about hours - see attached.

[skander36]

P.S. Reference value is 10.000005

What is it and how do you know its value to better than 0.5ppm?

I  said before, is a PDVS2 mini, calibrated in January this year.

[Kleinstein]

...
One of the reasons the SDM3065X costs less is because the LM399 is not burned in ahead of time. If you want stability faster, put at least a thousand hours on the chip.

The drift directly after turn on and the long term drift are different issues. It is common practice to have a mains powered meter powered up for an extended time (e.g. > 2h , possibly more in a fan-less design) before counting on the accuracy. So the calibration should also be for the later value, not the initial value. The initial drift can also vary quite a bit between units. So that SDM3065 could be an especially bad one - others can be much better. It would need more units to really make a point from it.

The design is different and the way the SDM3065 is build the 20 V range is not the most stable. Chances are the 2 V range can be more stable with the SDM3065 and the 1 V range of the DMM6500 and KS34465 are likely less stable than the 10 V range. A comparison at 1 V may go the other way.
Want the SDM3065 to look good: do the comparison at 15 or 20 V.

The reference with little burn in is a different issue and effect the long term drift. This a downside of the SDM3065 and makes the claimed accuracy a bit dubious for the initial phase and may requite an adjustment with the 1st calibration to really get it accurate. With the lower grade 34460 version from KS (still more expensive than the Sigilent), I am not so sure they get a proper burn in - maybe the rejects from the better meters. It would not need a few 1000 hours, just 1 week could do a lot.

[skander36]

One of the reasons the SDM3065X costs less is because the LM399 is not burned in ahead of time. If you want stability faster, put at least a thousand hours on the chip.

The meter is about 6 years old, and last year Siglent has certified /calibrated as good.

[KungFuJosh]

That's not within spec, is it?

Yes, it's still in range based on the 0.0015% I mentioned above. Which is wrong. Apparently that accuracy applies only to the 2V range (within 24 hour relative to calibration standards).

The meter is about 6 years old, and last year Siglent has certified /calibrated as good.

Skander's meter is 6 years old, and calibrated about a year ago. 20V range: 0.0040% of reading, + 0.0004% of range is the relevant stated accuracy. Even if we pretend it's within 24 hours (which we know is not true) that would be 0.0020 + 0.0003. In either of those cases, it's well within those ranges.

Skander, I'd like to see what it looks like if you let the meter and the standard warm up for 1 hour and then record the data over a long period and see how much drift there is.

I did the same warm up, and I'm recording data now with TestController (I'll follow up in the regular SDM meter thread).

If your meter shows to drift significantly when correctly warmed up and setup, then Siglent should absolutely correct it.

[bdunham7]

I  said before, is a PDVS2 mini, calibrated in January this year.

OK, I must have missed that.  In any case, that is a 5ppm (typical) reference not 0.5ppm, so that needs to be taken into account when analyzing your results.  I don't think it makes much difference to the conclusion though.

[bdunham7]

In either of those cases, it's well within those ranges.

I think one of us is reading that chart wrong.  What do you see the SDM3065X reading at the 1:45 (end of the chart) point?  How many ppm below 10.00000V is that?

I'm seeing about 9.99933V, which is 670uV or 67ppm below 10V.

[KungFuJosh]

In either of those cases, it's well within those ranges.

I think one of us is reading that chart wrong.  What do you see the SDM3065X reading at the 1:45 (end of the chart) point?  How many ppm below 10.00000V is that?

The incorrectly stated accuracy was +/- 0.0015%. 10.00000000000000000000000000 x 0.9985 = 9.985. At no point in the chart does the measure value drop below 9.9992. It is within that incorrect range.

The correct accuracy for his meter in the 20V range is 0.0040%. That would place the bottom range: 10.00000000000000000000000000000000000000000000000000 x 0.996 = 9.9960. The meter is clearly within the accuracy spec.

[Kleinstein]

After warm up the SDM3065 seems to be some 65 ppm off the nominal 10 V from the reference and some 50 ppm lower than the other 2 meters. Both would be outside the 1 year specs (40 ppm of 10 V + 8 ppm of fixed part). So the reading should be between 9.99952 and 10.00048.  It is not far off, so maybe still in the gray zone where we can not for sure say it is out, but more likely out than in.

Given the time of year, part of the problem could also be the temperature - it may not be close to 23 C. A high temperature could explain a lower reading. An extra 10 K would allow for an extra 50 ppm.

There is a certain problem with the usual calibration practice to only check against the spec limit (e.g. for 1 year). Ideally there should be a warning if the meter just passes the 1 year specs, but not the shorter (e.g. 90 day) specs. This would indicate that the meter was OK for the past, but has a chance to drift outside in the next year or so. So one would either need to include correction factors or do an adjustment to make the meter direct read good values again. For the future it is just not good enough to just pass the 1 year limits, one needs some headroom for future drift and temperature variations.

[zepto]

I'm surprised by the minimal warm-up drift of the other 2 meters, but I do not have any experience with either of those models.

I recently tested some old meters from hp, fluke, advantest, and keithley with possible intent of adjusting. Despite some being 40 years old they all decided a high quality reference was within 40 ppm of 10V, the two oldest suggested it was within <10ppm. Most of them had not been turned on in the last year, I have no idea when they were last calibrated. I bought them used and untested beyond power on. I have never worked on any of them and they are better off for that fact.

I was under the impression that HP's spec sheets for multimeters served as a guarantee and they were liable for damages.

I do not recommend to hobbyists that they buy an old LCR meter like pm3603, hp4261a,esi252 etc when DE-5000 exists. There is not a similar device that makes precision multimeters from the past obsolete. Even older devices that lacked digital interfaces could have their usefulness extended with a $15 mcu dev board and a little bit of effort. The graphing capability of modern meters is inferior to properly processing the data on a computer, and they don't offer anything over older meters if you just want to take a series of 1 plc or 10 plc readings.

Recommending sdm3045x for $400 to a hobbyist in the USA seems malicious.

[skander36]

I see you like it
The source csv file from Test controller (the one and only).

[bdunham7]

The correct accuracy for his meter in the 20V range is 0.0040%. That would place the bottom range: 10.00000000000000000000000000000000000000000000000000 x 0.996 = 9.9960. The meter is clearly within the accuracy spec.

Your math is one or two decimal places off depending on where you look.  100.0000% - .0040% = 99.9960% or 0.999960.  4 nines.  The lower threshold would be 10V x 0.999960 = 9.99960V.  The lowest readings appear to be about 9.99933V, well below the threshold.

[nctnico]

I'm surprised by the minimal warm-up drift of the other 2 meters, but I do not have any experience with either of those models.

The Keysight one Skander36 tested has a temperature coefficient compensation mode. Likely that is enabled. My guess is the Keithley DMM6500 has a similar feature.

[Kleinstein]

The KS meter is a quite high grade part and the actual performance can also vary. The main part to expect is temperature rise (e.g. some 10 K as a crude estimate for meter with a fan) times TC  (2 ppm/K spec limit for the 34465). So one should expect less than some 20 ppm or the grid spacing in the graph. The actual TC is often quite a bit smaller as actually measuring the TC is difficult and thus some design headroom needed.
Some of the initial drift (the first few minutes) are likely missed with the meters, as they take quite some time to boot and than may come added time to actually start the test. The curve looks like it is not that steep to start with and thus some initial part missing - the first few minutes can make quite a difference to the look.
The actual turn on drift can also vary quite a bit between units and there can be some lucky ones with really low drift.

The more relevant parameter may actually be the TC, so the effect of a change in the room temperature.


The $300 price range for a new bench meter is tricky. One can get some 4.5 digits based one a chip set for hand helds for some $200 or even less, but these often miss on 4 wire ohms and high impedance. Due to the low volume the bench versions tend to be more expensive than a comparble hand held version and no accuracy advantage at the lower end.

Bench meters with 4 Wire ohms and high impedance mode start at more like $400, like with the SDM3045, but there are a few alternatives too.
The SDM3045 is a bit 2 sided thing. One the one side it is kind of based on a 5.5 digit design (SDM3055E is essentially the same harware) that is software crippled. One still gets relatively good noise performance. The crippling however causes some issues with the ranges that can be really anoying. So one may consider the not that much more expensive SDM3055 as the proper 5.5 digit version.

Some $300 is the range where one may get a used 6 digit meter in the US - maybe no longer a HP34401, but still some reasonable used bench meter around in the US. One issue with a used DMM can be the PC interface: the old ones often come with GPIB and a GPIB card / adapter add to the costs.

[slavoy]

I see you like it
The source csv file from Test controller (the one and only).

Noticeably outside the specification, but I’m not knowledgeable enough in metrology to calculate the uncertainty of your method. Siglent might just as well be right on the low edge.

[KungFuJosh]

Okaaaay, I did not understand how the accuracy was supposed to be calculated. It's pretty goofy the way it works. It's like there's an extra step just to confuse me. 😉

So the real accuracy is supposed to be:
0.0040 + 0.0004 = 10 x (0.0040 / 100) + 10 x (0.0004 / 100) = +/-0.00044
.0004 + .00004 = .00044

That makes the lower number 9.99996, but that's only assuming that the calibrated reference really matches the calibrated number in his environment, which it does not, according to three 6.5 digit DMMs.

I saw Skander's calibration post, and the Siglent was usually above the nominal value, but now it's under. That's not awesome.

However, since the meter is 6 years old, it's not under warranty anyway. Maybe it would be better to let an independent lab calibrate / adjust the meter?

SiglentNA quoted me $300 just to calibrate. Trescal told me $200, and a couple smaller but still certified/traceable labs quoted me around there, including the data. I would rather an independent lab calibrate it anyway; I would certainly never pay the manufacturer extra for a potentially biases cal. 😉

I would still like to see longer data after everything is properly warmed up.

[bdunham7]

That makes the lower number 9.99996,

Now you have FIVE nines!     When it gets confusing, using the same number of digits as appear on the display can be helpful. 9.99960 is what you are looking for, or 40 counts.  Actually you'd use 9.99952, another 8 counts lower, since there is the additional 4ppm of range and the range is 20V. 

I saw Skander's calibration post, and the Siglent was usually above the nominal value, but now it's under. That's not awesome.

If I'm looking at the same post you are, that measurement was at 18V, not 10 so linearity could be an issue.  That much non-linearity would, IMO, truly be "not awesome" but still a possibility.  I don't think there is an INL spec for the SDM3065X.

You posted a screenshot in another thread showing your SDM3065X with about 2ppm p-p noise over a 30-minute period.  What was it measuring on cold power-up with the reference warmed up?  Do you get anywhere near as much power-up settling as skander36?

[Kleinstein]

It is indeed odd to see high values in the calibration report and low values here. There are a few points that can cause such effects: a rather high TC and different temperatures, an offset error (e.g. zero adjust too soon after trun on or at a much different temperature) or a linearity problem as the calibration is at 18 V  and not at 10 V. Another possible issue can be EMI, e.g. from interference between the meters or some local source.

The full scale for the SDM3065 is 20 V to the 40+4 ppm makes it  400 µV + 80 µV = 480 µV or 48 ppm of the 10 V signal.

The rather high costs for calibration are a problem with many more budget meters. This is not only the Sigilent. Here the DMM6500 may have a slight edge with specs for 2 years, not just 1year. For hobby use one usually not really needs the calibration as an extra expensive piece of paper to tell that is meter was supposedly OK at the test points.