X digits of DIY voltmeter for Y dollars?

[ogden]

It is fine for great minds to search for holy grail of ultrastable voltage reference, but it could be helpful for (evblog) community if you come to earth and brainstorm affordable, yet precise autoranging voltmeter project. Perhaps uVolt?

Challenge is simple: most digits/bits per dollar.

Good price/performance reference is 7$ LCR/transistor meter. Is it possible for 20$ of components to do 5&1/2 - digit autoranging millivoltmeter giving max ~60VDC @ 20 +/-10 oC? If not - then how much it would cost for proper 5&1/2 digits? Could we agree on circuit? It could heavily use software calibration/compensation approach with NTC's all over the circuit & maybe other tricks to increase performance for a cheap.

Your thoughts?

[TiN]

You asking wrong side questions. Define how accurate, and how sensitive you need meter to be, what ranges should it have, and how stable? Digits on their own don't mean that much without these questions cleared.

Reason why you see people here using long scale meters is mostly because of the stability (temporal and environmental) and accuracy (linearity, uncertainty), not the digit themselves.

Millivoltmeter, that is 1mV range meter with 5.5digit resolution for example need very careful low noise front end design, with great attention to details and EMF issues. Just pure copper connectors suitable for this range may cost more than 40$ a set of two. Commercial meters that have 1mV range and provide 5.5digits all cost multiple thousands $$.

So try to think of what you are after. And whatever you spend on making long scale meter, even just a prototype, it will be more expensive than going on used market and buying good 6.5 or 7.5-capable meter, trust me on this one . You will need it anyway for your design verification and testing.

[Svgeesus]

A noble goal, but the goal should probably be more precisely stated and generally agreed before any design work. You already stated this this is a voltmeter only, which is good.

Some people for example would not touch, and would advise others against, any meter that doesn't have CAT ratings because their minimum requirement is safely handling 400V and 10A (but conversely, they don't really see the need for more than 3.5 digits). They would steer a design towards spark gaps, barriers and fuses. I suspect that is not the group you had in mind (but you did mention 60V which seems high and would constrain the design).

Others would want to be measuring precise voltages or small voltages (you mentioned ?Voltage as a possible name) but would be fine if the maximum input voltage was 15 or even 5V.

Who is this device aimed at, and can we assume they already have an EEV-style handheld meter? In that case it is adding extra precision (they can read 100s of mV on a 5V range but want to read 10s or single digit mV reliably). Is it an add-on  to an existing meter (like ?Current), or does it have it's own display and controls, or does it just provide data over USB to a laptop or RPi?

I can empathize with the latter group, until recently my best/only meter was a UT-71C which was just not good enough to even check, for example, a claimed 2mV initial accuracy on a 5V voltage reference, let alone try to trim it out. (I'm now a happy owner of a Keysight 34465A).

Actually that could be a third target group for yet another definition of ?Voltage (not at the price point you mentioned): people with a reasonably good 5.5 or 6.5 digit meter, but that is at their spending limit; four times that cost for a good 7.5 or 8.5 digit meter is unattainable. A differential voltage probe (with maybe an LM399A and a DAC, plus a stable, low drift differential amp) to extend the range of their existing meter (measuring the afore-mentioned few mV to 10s of ?V on top of up to 10V offset).

[Kalvin]

This ADS1262, 32-bit ADC Eval kit from TI seems to provide a starting point for an inexpensive voltage measurement platform:

https://xdevs.com/review/ti_ads1262_p1/
https://xdevs.com/review/ti_ads1262_p2/

The kit is pretty expensive $199 form TI, but the chip itself costs around $10.

http://www.ti.com/tool/ADS1262EVM-PDK
http://www.ti.com/product/ADS1262

According to the article, with the on-chip voltage reference one can build a 5 1/2 digit instrument downto 1 mV input level. With a stable external reference one could achieve a somewhat better performance. The design needs to be constructed very carefully as the design is very sensitive to the temperature effects.

Probably someone here in the forum has tested this or some other 24-bit or 32-bit ADC, and can provide much better information and insight into the possibilities and potential problems.

[ogden]

Define how accurate, and how sensitive you need meter to be, what ranges should it have, and how stable? Digits on their own don't mean that much without these questions cleared.

I mentioned 5&1/2 digit, max ~60VDC @ 20 +/-10 oC. In my opinion this more or less includes stability specs. If not enough specs - then please  elaborate. The lower voltage range - the better, but not that important. Let's say resulting device shall be able to measure 1.0000 volts w/o significant error. Everything else is bonus. What I am aiming at - max performance for given amount of $, for b3ginners who at the moment have just 3.something digit multimeter and want more. uVolt most likely was overstretch.

Millivoltmeter, that is 1mV range meter with 5.5digit resolution for example need very careful low noise front end design

Oh.. Then obviously 1mV range shall be disqualified. As I said in my offering to discussion - "$" is paramount to "V".

So try to think of what you are after.

I am after ridiculously cheap voltmeter with surprisingly good (for a buck) performance

And whatever you spend on making long scale meter, even just a prototype, it will be more expensive than going on used market and buying good 6.5 or 7.5-capable meter, trust me on this one .

Trust me, I already have bench multimeter so what I am aiming at is not personal benefit.

So you say that it's impossible to do collaboration here and somehow develop said voltmeter so it costs for individual let's say <= 20$ (or <= 40$ whatever) in circuit components BOM?

[ogden]

Some people for example would not touch, and would advise others against, any meter that doesn't have CAT ratings because their minimum requirement is safely handling 400V and 10A (but conversely, they don't really see the need for more than 3.5 digits). They would steer a design towards spark gaps, barriers and fuses. I suspect that is not the group you had in mind (but you did mention 60V which seems high and would constrain the design).

Nothing that requires CAT ratings definitely. If stated 60V is too much, then we derate requirements to 30V or so. Target is audience using said 7$ LCR/transistor meter, device obviously same form factor - bare PCB with widely available uC & display on it. Case is up-to user.

Is it an add-on  to an existing meter (like ?Current), or does it have it's own display and controls, or does it just provide data over USB to a laptop or RPi?

Stand-alone device, having own display and some pushbutton(s) if necessary, powered preferably by 2xAA batteries or 4xAA if that matters.

people with a reasonably good 5.5 or 6.5 digit meter, but that is at their spending limit; four times that cost for a good 7.5 or 8.5 digit meter is unattainable.

Target group would be people who 1) want to build something 2) have nothing better than some handheld 3.something digits multimeter. After build they supposedly shall either obtain some transfer standard or visit lab/friend with instrument having enough precision to do calibration.

[guenthert]

Doesn't this thread fit better in <Projects> rather than <Metrology>?  For $7 we're hardly getting a used copper lug. 

[ogden]

This ADS1262, 32-bit ADC Eval kit from TI seems to provide a starting point for an inexpensive voltage measurement platform:

Thanx! 32 bits is mind boggling I stated 5-1/2 digits for a purpose so 20 bit ADC could be in the list of possible choices as well, not to mention wide selection of 24 bit ADC's. My main concern is stable, still low-cost reference and obviously decade divider resistors as well.

I would be happy to hear from voltunts guru that ovenized/T-compensated TL431 together with 100ppm/oC decade resistor set on small daughterboard is enough for 5.5 digits at 10..30 oC temperature range. Most likely I am dreaming, right?

[ogden]

Doesn't this thread fit better in <Projects> rather than <Metrology>?  For $7 we're hardly getting a used copper lug. 

I said 20$, not 7$. Next time please read whole post, ok? How much do you bet for 5.5 digits? Preferably with some justification.

p.s. Those who live in <projects> have no clue about subject, so don't be negative right at the beginning pleez.

[zhtoor]

upto 5 1/2 digit dmm (6 1/2 maybe?) chipset $10-$20, see attached file and :-

https://www.digikey.com/products/en/integrated-circuits-ics/data-acquisition-analog-to-digital-converters-adc/700?k=ald500&k=&pkeyword=ald500&pv153=11&FV=ffe002bc&mnonly=0&ColumnSort=0&page=1&quantity=0&ptm=0&fid=0&pageSize=25

regards.

-zia

[floobydust]

There's a difference between resolution and accuracy. Look at the Accuracy Translator nomogram here: https://www.eevblog.com/forum/metrology/7-5digit-diy-voltmeter/msg1391177/#msg1391177

Ali/eBay has 5-digit voltmeters for about $5. "Measurement accuracy: (0.3% + 2words) {means the table worst case maximum error of less than 10mv normal within error 2mv}". Whatever that means. It's just a single chip MCU.
Is this good enough? /s

Because with a $$ budget, the PCB, voltage reference IC, low noise power supply, low thermal EMF connections, precision divider resistors, display and other circuitry cost money beyond $20.

You put forth a gauntlet but people on the forums prefer to design high performance, not bottom price.  Bottom price already exists.

[ramon]

Challenge is simple: most digits/bits per dollar.
[...]
Your thoughts?

Second hand equipment.

US $20 ?  I am the proud owner if this wonderful pocket meter :  http://www.tenmars.com/YF-220A.html   I paid US $19.24 (NT $560). Hard to beat. Also have one LTC2400 module at US$ 24

However, if you want to make a cheap DIY voltmeter kit then you could start with a blank sheet defining price, specifications, and component selection (input resistor divider, Protection, ADC, MCU, Display, battery/power circuit, ... Case?) and maybe someone will jump in to help.

[Svgeesus]

Doesn't this thread fit better in <Projects> rather than <Metrology>?  For $7 we're hardly getting a used copper lug. 

As a challenge to voltnuts for "what can your exertise build for minimum cost" it seems on target to me. The right people are here.

[Kleinstein]

With modern chips building a good DIY meter is not that difficult any more. However the prices for resistors has not gone down so much and the once cheap 1N82x refs are now rather expensive. So one may very well end up spending half the money on resistors. Building only a voltmeter as opposed to a DMM can save quite a bit, as no shunts are needed.  A tricky part it calibration, as the lower grade meters might need calibration in all ranges - this might need some higher quality gear. The type of design depends on the requirements or cost limits. Possible solution would be:

At the low end something like an MCP3421 ADC with internal ref and some divider, amplifier, just like the Chinese modules for $3-5. This would be not all the way to 5 digits, but close. Add a µC with PC interface and there is already some use to it.

The next class would be build around a better SD ADC like LTC2410 or similar and an external reference. This might already reach 6 digit noise level, but with limited linearity. Cost depends a lot on the accuracy and ranges. Low voltage ranges are not that difficult anymore, if one does not need super low bias currents. The difficult part would be more like the higher ranges with a divider at the input. The down side with the integrated ADCs is that they only work for some 2-5 V. So one would likely get high impedance only for the low voltage ranges and everything above 2-5 V would be 10 M or similar and with lower accuracy.

For even higher resolution - the way to go would likely be a self made ADC and thus quite a bit of higher effort. I am currently testing such an ADC design. The semiconductors could be surprisingly cheap (e.g. $10 range), but it might take a few PCB iterations to make it work well and for good performance it might need some $15-$100 in resistors for the ADC and a similar amount for the input amplifier. The usual reference in the 6 digit range is the LM399 at bout $10. Even if the linearity turns out to be good enough to do the range transfer with the ADC - it still need verification.
   
AC ranges are a separate issue. One normally would not get that much resolution here anyway. In the budget range it works surprisingly good to do RMS in software with an µC - even if only a 12 bit ADC.

[MasterT]

I almost finished my 2 channels uVoltmeter, components list:

1. Voltage refs - MAX6164, 5 ppm/C,  4$ (CAD).
2. ADC 24-bits - NAU7802,  3$
3. Microcontroler, arduino UNO 4$, (for lowering cost, maybe AtTiny or something below 2$).
4 LCD - Nokia 5110, 2$

What left, is voltage divider. Nothing fancy, 2-4  resistors array,  1-5% initial accuracy and 50-100 ppm/C not so important for dividers, thermal tracking in between resistors is.
 Works well on 4V range, close to 18-bits noise free or 5 1/2 digits. The problem is 40-60V range ( divider by 10-15), internal PGA of the ADC doesn't help much. If divider is constant, than I have noise and non-linearity at the low end. Using opto relay in the front end, would easily double projects cost. Still thinking, if I need micro-volts resolution above 5V, probably not.   

[ogden]

You put forth a gauntlet but people on the forums prefer to design high performance, not bottom price.

That's sad actually - that today engineers often are forgetting what innovative and excellent design means. They just search best components around, throw it all together using traditional approach/circuit - to get target performance ignoring that by using their heads, most likely similar performance can be achieved using less money, thou most likely using more design time. Good example where engineers were using their heads: EDC 4601 AC Voltage Standard (EEVblog #709 episode), check service manual page 24. There's 1ppm (!) reference using two zeners in elegant compensation circuit. Something like this is what I am looking here for

I started this discussion in hope that some can pick-up challenge of achieving as much as possible using as low cost as possible (as 7$ LCR/component meter do). Those who say "just buy used multimeter" missed essence completely. It's not my build, I just came to brainstorm. After all I own decent enough bench multimeter already.

Reference most likely is one component which is as it is. Possible choices are LM399 (obviously), then REF5050 (3ppm) and mentioned MAX6164 (5ppm). PCB can be made compatible to accommodate different references - layout for default ref and  pins to connect daughterboard. Those who don't mind some bigger thermal drift, can put on that board reference they have in their bin.

Next comes ADC. I would like to discuss multi- or dual-slope converter here. I feel/hope that this is doable for lower cost than some decent 20/24 -bit ADCs.

Regarding resistors - what you say about putting some 50 or even 100ppm thin film resistors on daughter board so they are thermally coupled as much as possible? On this daughter board I offer to put NTC as well (to measure temperature of said board). - Those who seek for as much performance as possible - they can do two point calibration, at two different temperatures. This could possibly allow to cancel some of resistor TC drift-induced error. What you say?

Those who ask me to state not only resolution (5 1/2 digits), but accuracy as well - honestly I don't know. As good as possible for BOM cost which does not let you break a bank. I would like to see what we can come-up with.  I don't want to state unrealistic requirements and because of that ruin all the discussion. 0.005% at TCAL +/-1oC? Doable? If this needs 50$ worth of resistor arrays, then what can be achievable when we spend just 10$ for resistors that gives let's say 4 ranges (50V, 5V, 500mV and 50mV)? Daughter board PCB can be made in dimensions compatible to some well-known brand decade divider - so again those who seek for best performance, can open their wallets and get it.

Obviously 20$ total BOM is just start of discussion. I would put BOM pain threshold at let's say 50$, but then most likely it will be not so desirable for masses anymore. Who knows..

[ogden]

For even higher resolution - the way to go would likely be a self made ADC and thus quite a bit of higher effort. I am currently testing such an ADC design.

Max 21bits is all what's needed here. Could you possibly share more about your work? Is it dual slope converter? Do you use uC timer? If it is possible to beat LTC2410 price - I am extremely interested to hear from you as much as you are willing to share.

[MasterT]

Next comes ADC. I would like to discuss multi- or dual-slope converter here. I feel/hope that this is doable for lower cost than some decent 20/24 -bit ADCs.

Regarding resistors - what you say about putting some 50 or even 100ppm thin film resistors on daughter board so they are thermally coupled as much as possible? On this daughter board I offer to put NTC as well (to measure temperature of said board). - Those who seek for as much performance as possible - they can do two point calibration, at two different temperatures. This could possibly allow to cancel some of resistor TC drift-induced error. What you say?

IMHO, sigma-delta 24-bits adc has at least one advantage compare to dual slope, it's practically immune to EMI, RF and noise, including electrical 50/60 Hz. 
Resistors array I mention, sometimes they call it resistor networks,
https://wiki.analog.com/university/courses/electronics/electronics-lab-resistors
could help to null out relatively fast temperature drift, on a seconds scale. Than, every minute or let say  30 sec, microcontroller can switch input to reference for re-calibration slow drift. Spending a few bucks on solid state  opto-relay, may save a lot on precise low TCR resistors, PGA and anything else you 'd need otherwise to get 1-10 microvolts right.
 Since voltmeters usually  don't have to monitor inputs w/o interruption, spending 100 milliseconds on re-calibration  could goes unnoticeable for end user.

[David Hess]

To my mind the problem is input buffering to provide high input impedance and range switching.  High input impedance goes along with input protection so it is desirable even if not required.

So what is the highest range and what is the highest resolution?  20 volts with 5-1/2 digits is only 100uV resolution which I expect is not enough.

I would be inclined to drop the typical high input impedance divider for range switching and replace it with a high compliance high input impedance buffer followed by a low impedance divider making a low cost high resolution electrometer style voltmeter.  This also allows a selectable constant 10 megohm or gigohm+ input resistance but it also requires a bipolar high voltage supply.

[Kleinstein]

For even higher resolution - the way to go would likely be a self made ADC and thus quite a bit of higher effort. I am currently testing such an ADC design.

Max 21bits is all what's needed here. Could you possibly share more about your work? Is it dual slope converter? Do you use uC timer? If it is possible to beat LTC2410 price - I am extremely interested to hear from you as much as you are willing to share.

It is a simple µC controlled multi-slope converter, with much of the timing controlled by run time (thus an ASM program), but it also uses a timer at one place. Rundown is using one slow slope and the µC internal ADC for the final part.
I am using a µC (Atmel AVR) and like the HP34401 the cheap HC4053 switches.  A low cost version would use something like  AVR(e.g. Mega48), 74HC4053, Ne5534, LM393, MCP6001, 2 x OP07, 1 x TL081 and 1  better OP like OPA137 or OP177. It still needs at least one (ideally 5 or 6) good resistor to get a good linearity - self heating limits linearity on resistors with a higher TC. The same limitation also applies to a divider and gain stage. So the minimum would be likely something like thin film resistors with < 20 ppm/K, possibly 5 ppm/K. So the BOM costs might be slightly lower than the LTC2410, but not that much and stability would not be that good with cheap resistors. I am still not sure how good linearity is, this would need a test with better resistors and a full linearity test is difficult.

One the positive side the SD ADC chips are small, low power, easy to use and may work without an extra zero phase. The choice of ADC also effects the input part.  A difference is that such an integrating ADC could work with a +-10 V or similar (one is rather free in choosing) range and could thus directly measure a 7 V reference like the LM399. The LTC2410 or similar would need something like a super stable divider or other tricks to use a 7 V reference as a long term reference. So a chip based ADC would likely use a 3-5 V or similar reference.

[Kalvin]

@Kleinstein Would you see that a cheap FPGA like $8 ICE40 could bring some benefits for the design as the FPGA can quite easily clocked at 100 MHz thus obtaining better time resolution compared to AVR?  And could the FPGA be used to implement the overall ADC state machine, and even help with the residue voltage measurement https://en.wikipedia.org/wiki/Integrating_ADC#Residue_ADC so that one could use an Atmega328 (10-bit ADC) even without time-critical ASM programming?

[branadic]

Could you possibly share more about your work? Is it dual slope converter? Do you use uC timer? If it is possible to beat LTC2410 price - I am extremely interested to hear from you as much as you are willing to share.

https://www.eevblog.com/forum/projects/multislope-design/

If I had a wish, I would like to have a 0-10V ADC with the highest resolution, accuracy and linearity possible (reference LTZ1000?) and a multichannel multiplexer to measure voltage references (up to 10V) fully automatic. No need for full DMM or DVM capabilities as this is what ready to buy instruments can do.

-branadic-