New eevblog multimeter?

[CDaniel]

Fool me once shame on you , fool me twice shame on me 
If the same company is involved that can't test and do a relative simple firmware right , I would never buy again . And there are some hardware issues too , like weak rotary switch

[Marco1971]

I did find another Shelly1PM and a standard Shelly1, I thought I saw them when we rearranged the office the other day.
Attached is the capture from the scope from the PM version since that’s the one I used in the video. Keep in mind that I am using a Hantek CC-65 that is rated for the lowest reading of 20mA although I have seen this go down to 5mA before it fails.
I tested this with looping the wire on the clamp two and three times and had the same waveform without the loop.
I also confirmed this Shelly1PM exhibited the same behavior using the 121GW, as did the standard Shelly1.

I am also including two screen shots.
The one with 11.5mA is just the shelly being powered.
The one showing 20mA (actual is 18.5mA) is just the shelly engaging the relay without any light bulb.

Hello Scott,
many thanks for your oscillograms! Very interesting.
So inside these modules are small and simple SMPUs, which create small square wave current pulses, which have a high Crest factor, like the problem in ACV, with high DC superimposed.

The 121GW, like many other DMMs, calculates the RMS by fast digitizing and directly calculating the sum of the squared values of the samples.
The amplitude of about 55mA_pp of these pulses in the idle case would normally create an overrange, as this is 10 times the 5mA range, which is for some reason not detected by the HY3131, or by the SW.

Dave has built in an additional x10 amplifier to feature 'low burden' on the current ranges, and the HY3131 provides additionally two selectable x10 stages. So I assume, that an undetected analogue signal clipping inside the HY3131 occurs. 55mA_p over 1 Ohm shunt give 55mV_p, times 10 once gives 550mV_p, and then calculated in the 50mA range to give about 12mA_rms. That's clipped to a display of about 5mA in the 5mA range, i.e. factor 2.5 low, i.e. one of the amplifiers, or the A/D is clipping at about 2.2V_p.
This makes sense, as the HY3131 is powered from 3.3V 3.6V, and AGND is at about 1.2V, so the positive margin is 2.4V.

So as the fast A/D should be able to handle high Crest factiors, i.e. high peak amplitudes, there should be a window comparator (in SW?) to detect any pulses at its input higher than 2V, or so, to range up in that case, although the RMS value does not exceed the current range yet. .. I've tried to figure out from the datasheet, from the configuration manual, and from the circuit diagram, how to detect such an overload condition, but could not find out, how to do this properly.

Frank

Hi, 121GW uses a "True RMS to DC" converter i.e an Analog Devices AD8436

https://www.analog.com/media/en/technical-documentation/data-sheets/AD8436.pdf

Fluke 87V uses an Analog Device AD737J

https://www.analog.com/media/en/technical-documentation/data-sheets/AD737.pdf

Only analog computations not in the digital domain.

Marco1971

[EEVblog]

I'm in need of at least 2-3 new meters at my job, and i need an autohold badly ... i make tons of measurements  and the old flukes we have are  getting tiresome to use  ...

My new $150 class meter has AutoHold, might be an option. Shouldn't be too far away, but I've said that before...

[Dr. Frank]

Hi, 121GW uses a "True RMS to DC" converter i.e an Analog Devices AD8436

https://www.analog.com/media/en/technical-documentation/data-sheets/AD8436.pdf

Fluke 87V uses an Analog Device AD737J

https://www.analog.com/media/en/technical-documentation/data-sheets/AD737.pdf

Only analog computations not in the digital domain.

Marco1971

The conversion principle is not relevant in the end, concerning the clipping of the OPAmps.
Anyhow, I also first stumbled across this RMS chip AD8436, but it is AC coupled only, and its input label RMS_IN seems not to be connected anywhere else in the schematic, so maybe this IC is a placeholder only?

I'll have a look, if it's really assembled inside the 121GW.

Frank

[Marco1971]

Hi, 121GW uses a "True RMS to DC" converter i.e an Analog Devices AD8436

https://www.analog.com/media/en/technical-documentation/data-sheets/AD8436.pdf

Fluke 87V uses an Analog Device AD737J

https://www.analog.com/media/en/technical-documentation/data-sheets/AD737.pdf

Only analog computations not in the digital domain.

Marco1971

The conversion principle is not relevant in the end, concerning the clipping of the OPAmps.
Anyhow, I also first stumbled across this RMS chip AD8436, but it is AC coupled only, and its input label RMS_IN seems not to be connected anywhere else in the schematic, so maybe this IC is a placeholder only?

I'll have a look, if it's really assembled inside the 121GW.

Frank

The conversion remain in the analog domain not in digital as you stated before.

IC17 on mainboard

https://lygte-info.dk/review/DMMEEVBlog%20121GW%20UK.html

Marco1971

[Dr. Frank]

The conversion remain in the analog domain not in digital as you stated before.

IC17 on mainboard

https://lygte-info.dk/review/DMMEEVBlog%20121GW%20UK.html

Marco1971

You're correct.. I also found this missing label RMS_IN.

Then the explanation of all this mess is easy..
For this low burden current measurements, in the 5mA range, the shunt voltage is amplified by 10 by U8, MAX 4238, then again by 10 by the OP1 of the HY3131, where the clipping occurs. Then the signal goes directly to the RMS converter U17, and its DC output is measured by HY3131.

Therefore, there is no physical option to detect this clipping of OP1, as the AC signal is routed outside the HY3131. 

That's a real HW design flaw.

The ACV problem has different root causes.

Frank

[coromonadalix]

Would an  op with an agc gain applied to it could correct this clipping problem ??

Or simply put the low burden option is a burden in the circuitry  10x gain with another 10x gain is kinda too much ?

[Kleinstein]

I don't think the extra gain should be a problem here - if at all there could be a problem with the reference point used for the amplification of the max4238. However this should normally be correct the other side of the shunt. For the DMM chip it makes no difference wether the signal come directly from the shunt or from the extra amplifier. The max4238 should only give a small output signal like 0.2-0.5 V max, comparable to a normal shunt signal. So this OP should not run into saturation. However the signal may be a little larger than from a normal shunt and leave the common mode range of the input - something that the normal protection would prevent directly at the shunt.

There may be a tiny chance that the x10 from the max4238 and x10 from the DMM chip together may start oscillating under certain condition via a feedback through the virtual ground.

The maximum crest factor (peak values compared to RMS) may be limited by the design / level the RMS chip is driven too. So the design balance may be more towards low noise / low maximum crest factor instead of higher permitted crest factor at the cost of more noise and drift. Especially currents may show a high crest factor. Ideally (e.g. new meter) there would be detection of exceeding the peak levels and switch to the next larger range or a special high crest factor mode.

[dcac]

I measured the maximum output swing of OP1 in hy3131 and it’s almost rail to rail (0V to 3.6V).

And AGND in ACA mode is set to 1.8V. So relative to this it equals a swing from -1.8 to +1.8 volt. And with the 100x amplification (max4238+OP1) that equals -18mV to +18mV maximum over the 1 ohm shunt.

So any current waveform exceeding this range will definitely be clipped by OP1 before it reaches the AD8436 RMS converter.

[CDaniel]

Then this is a basic design fault , the current is allmost never a sine wave but pulses and you should give enough headroom without posibility of clipping 
But there is also an issue with detection of shorter pulses ... it's not switching in the next range 50mA as the true-rms value is greater than 5mA .
So for clipping with 10x op-amp the experiment should be done with true-rms values bellow 5mA to see what crest factor is admisible and if in real life is measuring ok .

[Kleinstein]

Some 18 mV / mA maximum would be rather small maximum crest factor of some 3.5 before clipping starts at the top of the 5 mA range. This is especially poor without a warning.

The RMS/DC converter could in principle support a crest factor of at least 10  (not sure this also works with the rather low supply). So it's a question of how to set the ranges.
A SW fix could be to already step to the 50 mA range at some 2 mA, not just at 5.5 mA, effectively reducing the 5 mA range to 2 mA or 2.5 mA in automatic mode.
A HW fix would be reducing the gain before the RMS chip and have more after.

[CDaniel]

Even if the true-rms converter has poor crest factor capability it's not near as bad as clipping the signal upstream  so removing the part above a crest factor  altogether

[floobydust]

I see software fixes possible, where you switch an A/D to the raw input and do peak detection to look at the crests and  model if the op-amp is near overloaded by pulses. You're not using the (slow) A/D samples for the actual measurement but for the autoranging decision. Aliasing would happen at high frequencies, but for mains frequencies this is doable.
It's hard to tell with the indecipherable HY3131 datasheet that is six years old, she's a headbanger 

[joeqsmith]

but I could push it enough to make my two normal meters disagree by a fair margin.

And what was the correct value? Was the Brymen or the Fluke closest or where they all quite a bit off?

Sorry, I wasn't ignoring you but wasn't sure so I repeated the test, this time using an even lower duty cycle to exasperate the problem.   

The three meters were again placed in series and their total resistance measured.  These were placed in series with a 50 ohm resistor.  I assume the Arb is 50ohms and the cables are zero.  Applied a 60Hz, 0-5Vpeak pulse with a duty cycle of 0.005% or 83us wide.   

I calculate the current to be 6.62988mA.   
121GW; 0.5432 ACmA
Fluke 189;  1.954mA AC+DC
Brymen BM869s;  6.6535mA AC+DC

That's asking a lot out of any meter.  Normally I would be using a scope to make this sort of measurement.  As part of evaluating the 121,  I had made a video showing how fast some of the meters I have are when looking at short pulses like this.   I have a couple of CEM meters that are very fast when using the voltage mode.  I'm not sure how it would do with current.     

*****
Just noticed.... I had cleaned up my Arb software today.   That should read AC+DC Amps.  The AC+DC to the left would be voltage.   Maybe there are other mistakes.  Best to check my math...

[EEVblog]

I just tried this short pulse current measurement. The 121GW is again somewhat range dependent on this. If you fix the range it's much better.
Also compared the "upgraded" design and it reads higher than the others on short pulses for some reason.
Also compared the new $150 class meter.
The Keysight U1282A seems to perform the best, but new $150 class meter beats the BM879 when the pulse gets shorter like, see one photo with 30us, it holds up better.
All the meters including the 121 are about the same above about 200us.
I haven't measured or calculated the "exact" expected value.

[EEVblog]

[maginnovision]

That's on par with what I've seen from my five keysight meters, including the small offset for AC current.

[MiroS]

Dave short pulse case has nothing to do with 121GW problem with bad current measurement.  I think people here are right , this is design HW flaw which probably can be masked in software, by the cost of slowing down 121GW.

Other than that I think that this new one may have a problem as well. At first look instant zero looks like a good result, but I checked  my trusty FK28II , Tektronix DMM,  Brymen , FK289 and all they work the same way for your pulse test. I believe it is so for a reason, so if that new one is not doing as all respected multimeters than I would rather think that this new one has a problem somewhere or  is using a  new kind of technology which is rather at very, very small probablity in my opinion.

In context of new multimer, all this flaws with 121GW and qestion if this is secure multimeter - maybe we have here someone from 'certification company' to share some insides -  is there any chance to recognize problems like we see with 121GW in certification process ?

It got me thinking that certificate by X is not the same as certificate by Y, if so that this will  make me sad. Dave which company will certificate this new multimeter?

[EEVblog]

Dave short pulse case has nothing to do with 121GW problem with bad current measurement.

I just wanted to repeat the test Joe did because I thought it would be interesting.
But isn't the problem video in question about some sort of pulsed current?

[MiroS]

Dave short pulse case has nothing to do with 121GW problem with bad current measurement.

I just wanted to repeat the test Joe did because I thought it would be interesting.
But isn't the problem video in question about some sort of pulsed current?

I do not think so. I think people here are right regarding  HW design flaw. That has nothing to do with pulse width, but pulse high.
I am at opinion that BM869 works OK, and instant zero in this new multimeter  may indicate the other problem. I edited my previous post with this and other remarks/questions.

[EEVblog]

In context of new multimer, all this flaws with 121GW and qestion if this is secure multimeter - maybe we have here someone from 'certification company' to share some insides -  is there any chance to recognize problems like we see with 121GW in certification process ?
It got me thinking that certificate by X is not the same as certificate by Y, if so that this will  make me sad. Dave which company will certificate this new multimeter?

The only "Certification" in multimeters is 61010 et.al, and that is related to safety, not really to performance aspects. And even then when you start talking about this kind of pulse stuff, it's almost infinite. You could have 0-99% duty cycle, pulse widths from naf all to whatever, against every frequency across the whole bandwidth across all signal levels across all the ranges. The permutations are insane, and that's just for pulse current that we are talking here let alone all the other functions.
No such magic certification exists.

[EEVblog]

I am at opinion that BM869 works OK, and instant zero in this new multimeter  may indicate the other problem.

Why would a meter displaying zero current instantly when you remove said current "indicate a problem"?
If anything, a meter that continues to display current when there is no current would be the problem, surely? Just saying"well, that's how they all work" is not much of an argument.

[MiroS]

I am at opinion that BM869 works OK, and instant zero in this new multimeter  may indicate the other problem.

Why would a meter displaying zero current instantly when you remove said current "indicate a problem"?
If anything, a meter that continues to display current when there is no current would be the problem, surely?

I think if you measure RMS than if voltage is 100V at time t0, and zero at t1 than RMS over AC period will be not zero right? Otherwise I would say instant zero is instantaneous value, not RMS.
That may be something like that.  To be sure I would need to refresh in my head metrology lectures and do some math.  The other reason may be that AC in grids is never very clean, so maybe safer is to average AC for loger time that to show instantaneous value? or filtering is doing causing that  delay ? Anyway instant zero may indicate a problem somewhere in my opinion.

[EEVblog]

I am at opinion that BM869 works OK, and instant zero in this new multimeter  may indicate the other problem.

Why would a meter displaying zero current instantly when you remove said current "indicate a problem"?
If anything, a meter that continues to display current when there is no current would be the problem, surely?

I think if you measure RMS than if voltage is 100V at time t0, and zero at t1 than RMS over AC period will be not zero right? Otherwise I would say instant zero is instantaneous value, not RMS.
That may be something like that.  To be sure I would need to refresh in my head metrology lectures and do some math.  The other reason may be that AC in grids is never very clean, so maybe safer is to average AC for loger time that to show instantaneous value? or filtering is doing causing that  delay ? Anyway instant zero may indicate a problem somewhere in my opinion.

It's not actually "instant", meters typically sample 3-5 times per second, so say a 200ms period.
If you want to do software averaging over several sampling periods, that's fine, it's typically a feature called "smoothing" on Fluke's for example. Also AVG mode can be used.
But if you remove the current then why wouldn't you want your meter to read zero within the next sample? Let's say within 2 samples because it could have been removed mid-sample.

[coromonadalix]

If you do some of theses test on a few bench meters with faster sampling rate  what will happen ?   will they catch the pulse(s) better ?

The norm with many meter brand is 3-5 times a sec, never saw an hand held meter with faster sample rates, and my 28-29 gossen are at 2 - 2.5 a sec ... wich i find slooooooooow as hell.

Sadly i dont have a sig generator to do some of theses tests