New eevblog multimeter?

[Jay_Diddy_B]

Hi group,
This 'tail' that is being observed is part of the RMS conversion process. The M is for mean (or average). The signal has to be averaged over time, this implies a low pass response.

The tail isn't unique to the current ranges. I took a Fluke 85 III and attached a wire to pin 6 of the AD737 RMS converter.

I put the meter in AC volts and fixed the range at 4V

I applied a burst 1000 cycles of 1kHz sinewave to the input.

Here are the waveforms:



The cyan trace is the input
The blue trace is the output on pin 6 of the AD737
The time base is 400ms/div.



If I look at the Analog devices datasheet:



I find graphs like this one. It basically says the accuracy improves if you do more averaging.
More averaging is a slower transient response.

There are similar graphs on the AD8436 datasheet:





They even have this one, this is similar to the measurement I made:

[ Specified attachment is not available ]

A long 'tail' might be a good thing ...

Regards,
Jay_Diddy_B

[coromonadalix]

ok  you need a good hw and a good fw equilibrium, even if sometimes it misse its objectives, too fast or too slow

[MiroS]

We would be more clear on this getting chipset name in this new multimeter. 

Dave is it possibile to share with us chipset used in this new multimeter?

[EEVblog]

We would be more clear on this getting chipset name in this new multimeter. 
Dave is it possibile to share with us chipset used in this new multimeter?

For the $150 meter, I do not know, it's custom branded.
The other "new" meter is the same HY3131 used in the 121GW.

[EEVblog]

Hi group,
This 'tail' that is being observed is part of the RMS conversion process. The M is for mean (or average). The signal has to be averaged over time, this implies a low pass response.

Yes, that is to be expected, but 10+ seconds?

[joeqsmith]

I had written Brymen about the how their filters worked in the 235 and assumed their response would have been added to the manual.  I'm sure that post is still around.

I tried running the CEM DT-9939.  I thought this meter may hand the short current pulses much better than the Bymen  based on how well it does with short voltage pulses but it does not appear to be the case.   It's a little better than the 189.  It's about as slow as the Keysight shown in Dave's video when removing the current.

[EEVblog]

I had written Brymen about the how their filters worked in the 235 and assumed their response would have been added to the manual.  I'm sure that post is still around.

I've asked Brymen again, so will see what they say.

[joeqsmith]

I had written Brymen about the how their filters worked in the 235 and assumed their response would have been added to the manual.  I'm sure that post is still around.

I've asked Brymen again, so will see what they say.

From 2017

Someone had asked me about the settling time for the Brymen meters.  Basically they were describing a condition where they were trimming a device and using their Brymen to monitor the adjustment.  Because the adjustment was being done very slowly, it appeared their meter went into some sort of filter mode that causes it to have a very long settling time.   I attempted to replicate this and found that both my BM869s and BM235 will do as they describe.  I have linked a short video that shows how I was able to reproduce it.

I assume this is a normal behavior for at  least these two meters.  I could not find any mention of this additional filtering in the manual.  Could you please explain what is going on?  Thanks




BM235 & BM869s firmware are with following reading smooth designs.
 
BM869s
After getting NEW measuring reading, check “NEW reading” and “the AVERAGE of the last 8 readings”.
If ABS ( “NEW reading” -  “the AVERAGE of the last 8 readings”) > 8 counts, then display “NEW reading” directly.
If ABS ( “NEW reading” -  “the AVERAGE of the last 8 readings”) <= 8 counts, then display “the AVERAGE of the last 8 readings” instead.
 
BM235
After getting NEW measuring reading, check “NEW reading” and “the AVERAGE of the last 16 readings”.
If ABS ( “NEW reading” -  “the AVERAGE of the last 16 readings”) > 16 counts, then display “NEW reading” directly.
If ABS ( “NEW reading” -  “the AVERAGE of the last 16 readings”) <= 16 counts, then display “the AVERAGE of the last 16 readings” instead.
 
They are the causes.

[coromonadalix]

We need  at least 8 readings and up to have a good averaging ??   time to add an oversampling  measurement loll

or cram an 34401a or keithley 200x  or any other good bench dmm in a hand held device  loll

[Kleinstein]

The RMS chips need some time for averaging. With the analog chips this is usually not synchronized with the AC waveform and an infinite time pulse response filter. With the usual RMS chips one has to find a balance between reasonable fast response and added errors, especially at lower frequencies (e.g. 50 Hz and below).
AFAIK the RMS chips are somewhat different in the way they do the averaging and with some solutions the time constant depends on the amplitude - this can give slow settling going down.

Another possible problem to cause slow settling is if they use an electrolytic capacitor with high dielectric absorption. The filter cap often needs to be quite large (e.g. > 1 µF) so that electrolytic caps are not unusual, but there are different types to choose.

There are a few different options on how to implement the RMS conversion. Especially the higher end meters may use some kind of digital sampling instead of classical analog RMS. This way one could have a different type of filter, especially an FIR type filter or signal synchronous measurement.
So it is possible to have fast response and an accurate reading.

Another odd point of most analog RMS chips is that the bandwidth depends on the amplitude. With low amplitude they tend to get slower. At the upper end there may be a slew rate limit that can also cause an error with a square wave signal.

With electronic equipment it is not unusual to have a more pulse like wave form and one not allays known this upfront. So a good DMM should work with such waveforms. It is not just SMPS, but already rectifier and filter cap.

[MiroS]

[quote author=EEVblog link=topic=250310.msg3193994#msg3193994 date=1597749867
For the $150 meter, I do not know, it's custom branded.
The other "new" meter is the same HY3131 used in the 121GW.
[/quote]

Thanks Dave.

I guess that this new is SANWA, if so than this will be Brymen chipset, it will be very interesting to compare SANWA and Brymen.
I may be wrong of course, but who else than Fluke and Brymen  have its own chipset those days

[floobydust]

ACV and DCV subroutine from another manufacturer using the same Hycon chip. I've been laughing about it for days 
Those skilled in the art can suggest a better approach to 0.0

Code: [Select]

/*** DMM_DGetAvgValue
**
** Parameters:
**      int cbSamples     - The number of values to be used for the average value       
**      uint8_t *pbErr    - Pointer to the error parameter, the error can be set to:
**          ERRVAL_SUCCESS              0       // success
**          ERRVAL_DMM_VALIDDATATIMEOUT 0xFA    // valid data DMM timeout
**          ERRVAL_DMM_IDXCONFIG        0xFC    // error, wrong current scale index
**
** Return Value:
** double
**          the DMM value, or
**          NAN (not a number) value if errors were detected
** Description:
**      This function computes an average value corresponding to the DMM value 
**      returned by DMM_DGetValue, for the specified number of samples.
**      The function uses Arithmetic mean average value method for all but AC scales,
**      and RMS (Quadratic mean) Average value method for for AC scales.
**      If there is no valid current scale selected, the error is set to ERRVAL_DMM_IDXCONFIG.
**      If there is no valid value retrieved within a specific timeout period, the error is set to ERRVAL_DMM_VALIDDATATIMEOUT.
**      It returns INFINITY when measured values are outside the expected convertor range.
**      When no error is detected, the error is set to ERRVAL_SUCCESS.
**      The error is copied on the byte pointed by pbErr, if pbErr is not null.
**      When errors are detected, the function returns NAN.
 **           
*/
double DMM_DGetAvgValue(int cbSamples, uint8_t *pbErr)
{
    uint8_t fValid = 1;
    double dValAvg = 0.0, dVal;
    int i;
    int idxScale = DMM_GetCurrentScale();
    uint8_t bErr  = DMM_ERR_CheckIdxCalib(idxScale);   
    uint8_t fAC;
    if(bErr == ERRVAL_SUCCESS)
    {
        fAC = DMM_FACScale(idxScale);
        if(fAC)
        {
            // use RMS (Quadratic mean) Average value for AC
            for(i = 0; (i < cbSamples) && fValid; i++)
            {
                dVal = DMM_DGetValue(&bErr);
                fValid = (bErr == ERRVAL_SUCCESS) && (dVal != INFINITY) && (dVal != -INFINITY )&& !DMM_IsNotANumber(dVal);
                if(fValid)
                {
                    dValAvg += pow(dVal, 2);
                }
            }
            if(fValid && cbSamples)
            {
                dValAvg /= cbSamples;               
                dValAvg = sqrt(dValAvg);
            }           
        }
        else
        {
            // use normal (Arithmetic mean) Average value for other than AC.
            for(i = 0; (i < cbSamples) && fValid; i++)
            {
                dVal = DMM_DGetValue(&bErr);
                fValid = (bErr == ERRVAL_SUCCESS) && (dVal != INFINITY) && (dVal != -INFINITY )&& !DMM_IsNotANumber(dVal);
                if(fValid)
                {
                    dValAvg += dVal;
                }
            }
            if(fValid && cbSamples)
            {
                dValAvg /= cbSamples;               
            }
        }
    }
    if(bErr != ERRVAL_SUCCESS)
    {
        dValAvg = NAN;
    }
    if(pbErr)
    {
        *pbErr = bErr;
    }
    return dValAvg;
}


[MiroS]

I spent a few minutes with sheet of paperand easy tests, it looks that instant zero may bring a problem -  limiting factor is low frequency AC range, precision , but the most annoing is screen fluctuation.

I seems like cheaper multimeters may give instant  zero, like Fluke 17B+, that one is instantly showing 'zero', but  at some cost
Result is rock solid for BM869s and badly fluctuationg for 17B+ for low frequency . I do not have cheap 'multimeters' for further tests.

BM869s is  at lower side 5Hz, 17B+ - 40 Hz, so that new one showed by Dave may have probably higher low AC frequency in spec.
Tests conducted staring 1Hz and and up.

Dave can you check spec for low AC frequency for voltage/current mesurement for this new multimeter  / do some tests?

[Kleinstein]

The older AD736 RMS converter has the odd point that settling slows down when the reading is low, so it is inherently slow going back to zero.  The newer AD8436 used in the GW121 does not have this and may be faster without compromising so much on the low frequency of higher crest factor performance. It could be just the choice of RMS converter that makes the difference.
It is usually a balance between response speed and accuracy - for a handheld meter I would often prefer fast response over a little more error at lower frequency. The higher end bench meter can even have the option to chose between 2 modes.

In theory the software could compensate some of the slow end of the settling.

[floobydust]

Post #66 bug is a mains-frequency 1/2 wave rectified waveform. The DMM has plenty of time to notice it needs to range up.
Looking at usec pulses is intellectually titillating but have you heard of the oscilloscope?
It's no wonder this stuff will never get fixed as we move on to the next shiny object.

[EEVblog]

Post #66 bug is a mains-frequency 1/2 wave rectified waveform. The DMM has plenty of time to notice it needs to range up.
Looking at usec pulses is intellectually titillating but have you heard of the oscilloscope?
It's no wonder this stuff will never get fixed as we move on to the next shiny object.

This is not a half wave rectified mains waveform:

[EEVblog]

Dave can you check spec for low AC frequency for voltage/current mesurement for this new multimeter  / do some tests?

New meter is 3dB down at 2.5Hz. BM869 is only a smidge better, but it goes crazy jumping around at anything under 3Hz.
New $150 meter doesn't like anything below 10Hz.

[floobydust]

Post #66 bug is a mains-frequency 1/2 wave rectified waveform. The DMM has plenty of time to notice it needs to range up.
Looking at usec pulses is intellectually titillating but have you heard of the oscilloscope?
It's no wonder this stuff will never get fixed as we move on to the next shiny object.

This is not a half wave rectified mains waveform:

5ms/div, looks like half-wave rectifier current to me. Fullwave is 2x f_mains or 100Hz/120Hz.

[EEVblog]

5ms/div, looks like half-wave rectifier current to me. Fullwave is 2x f_mains or 100Hz/120Hz.

Yes, but it's not a half wave rectified waveshape.

[Wytnucls]

Half wave crest factor is equal to 2. This wave form has a crest factor of 4~5, which is beyond the capability of most multimeters.

[Kleinstein]

The pulses look like half wave rectification with a filter cap. So about what to expect for a very cheap low power SMPS input. So the current waveform in not that unusual anymore and a good multimeter should be Ok with this. I would not mind a few percent of error, but a grossly wrong reading (e.g. 5 mA instead of 18 mA) is quite a bit far off and more like a bug in the GW121.
A relatively small filter cap would cause some error on high crest factor signals, bit this should be only a few percent at 50 Hz. For a current signal a crest factor of 5 is not that unusual and the RMS/DC coverter would in principle allow this (e.g. 5 % error at a crest factor of 10 - though it depends on the details).

From what I have read here, a 100 fold amplification and thus 500 mV (RMS) full scale after the amplifier seems to be too much to allow high crest factor signals. No sure if SW could fix/reduce this by doing automatic range switching already earlier (e.g. at 2 mA / 20 mA). I would prefer a trustworthy reading over extra counts in the AC mode.

[CDaniel]

Of course crest factor can't be an excuse if in the video an Aneng meter is showing correctly ... what are we talking about ? 
As I said in many many cases the current is in pulses , that's how a diode bridge rectifier + filter cap works . This is absolutly  basic

[Kleinstein]

It looks like most handled DMMs have specs with a crest factor of < 3.  I have found something as bad as <1.5 when near the full scale, for some Fluke meters (e.g. 117), though not clear if this applies to voltage readings only or also to current readings. With current readings a higher crest factor is very common - so it would be nice of the meter would work there too. The AD8436 chip allows for a crest factor of 10 with slightly higher error - so in principle it would be possible to build for a higher crest factor.

 The cheaper meters with 2000 counts and thus a 200 mV range may have it easier to allow a higher crest factor (peak values) than meters with a 500 or 600 mV range to get a few more counts. If limited to a rather low crest factor (probably around 1.5 for the GW121 near the end of range) one should at least have a warning when exceeding this limit. One usually does not know the crest factor up front, so some tolerance to a higher crest factor can be an important factor.

[Dr. Frank]

Folks,

this weakness of the 121GW has nothing to do with Crest factor in first instance. This instrument  simply has no proper overload detection for its (internal) OpAmps, at about +/- 1.6V, or so. Therefore, signal clipping occurs, which falsifies of course the RMS calculation.

Overrange / uprange only occurs, when the measured (clipped) RMS value exceeds the chosen range.

This obviously happens in ACI and ACV.

This overload detection requires a fast sampling of the amplified signal, sort of peak-window detection, which is not implemented. Even in ACV it's not implemented, what makes me wonder. The fast internal A/D could serve this purpose.

Better DMMs feature a separate fast window comparator for that.

So you have to live with this weakness and may anticipate manual upranging, when you expect 'difficult', non-sinusoidal signals.

Frank     

[Jay_Diddy_B]

Hi,

To build on what Dr. Frank said.

I scanned the 121GW manual to see what it had to say about crest factor.

The manual echoes the Analog Devices specifications for the AD8436 as having CF of 10.

I am not sure if the crest factor is preserved in the design, because it would require a lot of headroom, without clipping, in the amplifier stage in front of the AD8436.


Crest factor is the peak / RMS

The Fluke 289 DMM has a crest factor of 3 at full scale and increases to 5 at mid-scale. This make sense because there is more headroom at mid-scale.

You can find tables on the internet that illustrate various crest factors.

A 5% duty-cycle square wave has a crest factor approximately equal to 4.5

A 1% duty-cycle square wave has a crest factor of 10

Most waveforms that you measure will be lower.

Regards,
Jay_Diddy_B