5 dollar voltmeter with 5 digit (0.1mv) resolution

[Shock]

Sperb can you link the one you purchased please? If possible take a photo of your pcb as well. Just so I and others can get or check we have the correct one.

[b_force]

Sperb can you link the one you purchased please? If possible take a photo of your pcb as well. Just so I and others can get or check we have the correct one.

Found some schematic here at least (plus some further explanation)
https://eddy-em.livejournal.com/73375.html

[floobydust]

Careful, there's two different styles of these 5-digit panel meters. I look at the pictures to figure out which is being sold but even then, pic is 4-wire and I get 3-wire. If a vendor does not show a pic of the board, I would not buy.

SOT23-6 A/D (MCP3421) and some modules have only three wires for power and ground and V input, others have two connectors etc.
SOIC-16 A/D it's a chinese dual-slope part and I forgot where I posted the part number.

[magic]

Here.

Are those with SOIC16 ADC any good?

[Kleinstein]

The ADCs in SO16 case are likely OK too. The MCP3421 is not that special: just small, cheap and easy to use.
The other likely Chinese ADC version may be a bit cheaper but probably similar quality. It may even have an advantage with better 50 Hz supression. The MCP3421 has good 60 Hz suppression, but no good 50 Hz supression (one can improve on this by averaging over 3 consecutive readings).
Knowing the ADC and HW configuration would in theory give the option to use the Chinese HW and just make a custom SW version.

I somewhat doubt that the 16 pin ADC is a dual slope - they went largely out of favor because they need a relatitvely bulky capacitor (usually some 100 nF...500 nF with low loss). The ICL7106 even needs 2 more large capacitors. SD ADC are made with all parts on chip and are mean-while quite cheap. Also modern handheld DMMs use SD ADCs and usually no more dual slope. 

[b_force]

The amount of bits doesn't say that much to begin with.

I have seen ADC's with an higher amount of bits, but with an higher amount of error a well.
Resulting basically in a ADC with much lower bits. (since the rest of the bits vanish into those errors)

Also a proper voltage reference is always needed of course.

This can be somewhat stretched when being manually calibrated.

I just calculated the total relative error estimation (with those wonderful partial derivatives).
These devices will just give me enough accuracy.

I was just hoping there was some of the shelf solution.
Maybe I can write those Ebay/Ali shops a message if they can reprogram something for me.

I am also happy to make my own PCB, it's pretty straightforward, but those MCP3421 are out of stock (like most other things atm )

[floobydust]

I found it, SOIC-16 "22 bit" A/D marked "BX5815" and nobody knows where it comes from in the gutters of Shenzhen. It looks like a TL431 is the reference.
At first I thought a knockoff of TC500 dual-slope.
Or the IC is possibly a remarked Titan Micro delta-sigma 24-bit TM7707 or 16-bit TM7705. They appear to be "enhanced" knockoffs of Analog Devices 16-bit delta-sigma AD7705.
I wouldn't doubt it's 16 bits, as the display at 99,999 count almost 17 bits worth.

more in this thread https://www.eevblog.com/forum/testgear/are-sub-$10-5-digit-ammeters-worth-buying/

[magic]

If TL431 is the reference then they likely aren't as good as MCP3421, although the latter's TC specification is only "typical"

That being said, mine shows 5.000 when measuring a MAX6250 5V ±1mV reference chip.
Now it makes me wonder if those meters also cheat and round to full volts. Anyone tried?

[floobydust]

I tried a couple I have (MCP3421 A/D model) and it's kind of hilarious. There is no evidence of rounding to favorite values, and they are within about 2 counts verses a 34401a across the span of 0-25V, and VCC changes do not affect readings. But not so great firmware.

Both only show 4 digits i.e. 9.999V then the 5th comes on past that i.e. to show 10.000V
One module shows 5 digits up to ~4.3V but it freaks out:

If it is already in 5-digit display mode, it will move the decimal point at around 4.3900V i.e. increasing input it displays 4.3899V then goes to 4.390V (it does flicker 31.895V as it moves the DP to the right).
If it is already in 4-digit display mode, it will move the decimal point at around 4.3100V i.e. decreasing input it displays 4.310V then goes to 4.3089V
Of course a cold-boot does aggravate all this as well. In slow motion video, cold boot with input steady 4.3087V it displays: 0.000/0.430/0.5037/3.4920/4.3083V and settles there.

What I dislike about these modules is the averaging, if you input 5.0000V and add 50mVpp sine the numbers are constantly changing, at 10-30Hz they're wandering too much to make sense of. The A/D slowest is 3.75sps for 18-bits.

[Kleinstein]

I have not tried one of those modules, but the MCP 3421 in a DIY circuit. I did not see any problems and with averaging of 3 readings 50 Hz supression is OK.
The claimed 18 bit resolution looks about right - not like many 24 bit SD ADCs where this is only the number of bits provided for the data format. So 5 digits is a bit of a stretch - not a stable last digit, but also not wandering all over the place (e.g. RMS noise similar to last digit).

That an AC background makes the signal wandering is normal, if not at the notch frequency. With some additional averaging the SW could get a more stable, but also slower result. With 60 Hz hum the MCP3421 should have reasonable hum suppression, similar to what you get from a handheld DMM.

The factory trim of the scale factor looks quite good, though there is some doubt they actually get 0.05% accuracy after soldering.
With the additional dividider at the voltmeter input, there is anyway likely a cal factor in the software. It is not that much effort to get the adjustment right.

[b_force]

The factory trim of the scale factor looks quite good, though there is some doubt they actually get 0.05% accuracy after soldering.
With the additional dividider at the voltmeter input, there is anyway likely a cal factor in the software. It is not that much effort to get the adjustment right.

Yes, so with a 0.05% accurate voltage reference, the max theoretical error can only be as good as 0.05%.
On a 10 volt signal, that means 10.000±0.005V.
With a 30V signal that means 30.000±0.015V -> 30.00±0.02V so much so for those 5 digits
Fully assuming that any additional voltage divider won't attribute (which is mathematically and physically impossible)

This is however without any other additional issues, like gain error, match error and DC offset (as well as drift).
With external components you will makes this error a lot worse very quickly.

According the datasheet, the gain error is 0.05% typical, but can be 0.35% maximum.
Point is, we don't know.

I am a little confused what they mean with the PGA error, since there is only one PGA in there?
Usually this is just (quite literally) just an amplifier stage, which it could mean the gain error in de positive vs negative input.
If that's true, this can cause a lot more issues with a typical 0.1% error.

in 18 bit mode it only does 3.75 samples per second, so you won't be even able to measure 50/60Hz hum.
Any low-pass kind of behavior will automatically mean an averaging filter to begin with.
Only in 12 bits mode you will be able to really measure this 50/60Hz hum.

In the end it's always a little odd, having so many bits with a reference that's only 0.05% accurate.
Meaning that it's fine to measure relative differences very well, but absolute differences you still need an external voltage reference and some calibration inside the microcontroller.
Or in simple words, you can detect a difference in signal very well, but the value of this difference a little less.
(the main reason why most professional multimeters/bench meters are not more accurate as they are)

The contribution of an external voltage divider can be calculated as well.

[Kleinstein]

The simple SD ADC linke the MCP3421 essentially does an integration over certain time and this time determines how much of 50/60 Hz hum comes through with the result. You don't have to be able to really measure the 50/60 Hz with this, one actually prefers not to see them and have a blid spot at exactly that frequency. Nominally a  blind spot is at 60 Hz (for 14 bit and more), but there are huge tolerances: 14 bit mode is between 44 and 82 SPS, with 60 SPS nominally. So it depends on luck where the hum suppression works well. Nominally 60 Hz, but some units may have the sweet spot even below 50 Hz.

The accuracy of the reference and scale factor limit the accuracy one can get without an extra calibration / external calibration factor. It can be corrected with an external factor. When using the extra divider one is essentially forced to use such an extra factor and it is well possible the Chinese could as well use the right factor instead of a fake. With that extra factor the initial accuracy of the reference does not matter.
The HP34401 DMM and most other 6 digit DMMs use a LM399 ref. with 2% accuracy.

It is also quite common to have more resolution than accuracy. This especially comes up with higher resolution.  A typical 6 digit DMM has 40 ppm uncertainty on there best range (except frequency), so that is 40 LSB steps and other ranges can get quite a bit worse. It would be strange the other way around. With low resolution meters one is used to get the last digit essentially stable (1 LSB peak to peak). With higher resolution the more common case is an RMS noise of about 1 LSB steps and thus the last digit with about 6 steps peak to peak noise. That is the common way to define the resolution, as the higher grade ADCs tend to give out more bits and the limit is from the noise and often not the quantization.

The PGA part in many SD ADCs is not actuall a true amplifier, but more like a change in the sampling frequency. Essentially all the single chip SD ADC wort with switched capacitors and sampling the input at a rather high rate ( in the MHz range). This transfers charge to an integrator for the conversion. What they call PGA gain is actually only using a higher frequency (and in some cases maybe 2 parallel channels) for the input and this way operating the ADC with gain.  The neat feature of this is that the "gain" is very stable and accurate. I am not sure the display units would use that gain settings other than for the current mode.
One can see an effect of the this more frequent sampling: the input differential impedance is specifies as 2250 Kohms /gain.

[b_force]

It can be corrected with an external factor. When using the extra divider one is essentially forced to use such an extra factor and it is well possible the Chinese could as well use the right factor instead of a fake. With that extra factor the initial accuracy of the reference does not matter.
The HP34401 DMM and most other 6 digit DMMs use a LM399 ref. with 2% accuracy.

Yes, that's called external calibration, which gives an absolute result that is only as good as its reference.

In this case they don't use the LM399 for the absolute voltage reference, because that would mean your error/accuracy in the end will be only as good as 2%.
After they calibrated the device, they use the LM399 for a reference for drift over time and/or temperature.
Drift over time (long term stability) is often not specified for most ADC's and seen the teeny tiny page of this specific ADC probably not so good.
Let alone having some kind of active heating element (like the LM399 has)

So a smart person already learned something from this.
As long as you have a good and stable reference yourself and you can calibrate your devices, one can make cheap gear pretty accurate, as long as you calibrate them on a (very) regular basis!
This is also standard procedure in some research, to calibrate your instruments before you're doing the actual measurements.


btw, the effort to get the calibration right isn't difficult on paper.
In practice this is a pretty labor intensive job, since you have to check every device by hand.
Since wages in Asia are so low, 99.8% of those Chinese meters are all just calibrated by hand with some bench meter or so.

In the end it's all just statistics, there is a chance that your 10 euro dinky meter will be better than a 100k bench meter.
Like I said, you just don't know and the chances will likely be very low!

Anyway, I think we are going off-topic at this point?

 

[Kleinstein]

With the extra divider from relatively cheap resistors they essentiall have to calibrate / adjust the meters. Even with a very accurate reference they should check the accuracy anyway and doing the adjustment does not have to be much more complicated and time consuming. The procedure would be to power the circuit up in a test fixture and let it measure 0 voltage and an externally provided reference voltage (e.g. 10 V or 20 V, depending on the meter) and than send a special command to the meter to tell it to do the adjustment. So much of the adjustment would be internal to the small meter.
An accurate adjustment may take a little more time for averaging or waiting for the circuit to warm up and a more frequent check of the reference.
Ideally there would than be a check if the display reading is actually OK as it should be.


Adjustment would be time consuming the old style with a trimmer. So even in China they tend to prefer software calibration over mechanical trimmers.

[sperb]

As requested, I included some additional information on my post with the alternative firmware for the meter.

here: https://www.eevblog.com/forum/projects/$5-voltmeter-with-5-digit-(0-1mv)-resolution/msg3730789/#msg3730789

Please let me know if that helps.

Regards.

[floobydust]

I didn't see the firmware file(s), where are they located?
I think the STM8 is used for these panel meters because its the lowest cost, small MCU with EEPROM for the calibration. I have not tried soldering in the pushbutton switch to see what it does.

[sperb]

floobydust

Check my previous post (link above).

It is on the attachments.

Regards.

[floobydust]

Check again, there's no software in the attachments of post #141, just the two pictures 

[sperb]

Are you sure?

[floobydust]

AH now I see it, pic filenames I thought are on top. Sorry.
Do you know how much codespace is used, I think the parts have 8KB of flash.