DIY through-hole multimeter project

[technix]

I have brought this project up before, and I am rounding it up here. This is a design challenge - do build a multimeter that performs good enough for a maker, using only through hole components in 80s fashion.

Here is the specs I am shooting for:

* 4.5 digit, autoranging, true RMS, 4-wire resistance measurement
* Double-layer board, single-side load, through hole only.
* Two-board construct (one main board, one display board)
* Bench multimeter with built-in mains power supply for 240V (replaceable transformer for 120V operation)
* Little to no microcontroller involvement. If MCU is required for some reason, use up to two AT89C2051s (pot-fest)
* 7-segment LED display
* CMOS

The core ADC chip I am using is ICL7135 @ 100kHz.

Here is the ranges:

* Volts: 20mV, 200mV, 2V, 20V, 200V, 600V/2kV (10M or 10k input impedence)
* Amps: 200uA, 2mA, 20mA, 200mA, 2A, 10A (shared jack)
* Ohms: 20R, 200R, 2k, 20k, 200k, 2M

All those ranges covers 6 orders of magnitude, which corresponds to the 6 output pins of the autoranging PLD. Additional modes selectable through the knob:

* Diode drop (12V burden voltage so even those high-voltage LEDs will work, uses the 1mA constant current source)
* Capacitor ESR (0.2V drive voltage - lower than Schottky diode and some Ge diode voltage drop - how to implement it?)

[technix]

Display board is fairly simple as it hosts only the LED display and driver chips. The challenge here is to route all things correctly, matching the size of the case.

[technix]

Here is the reference circuit based on LM399. It emits a buffered 1.0000V reference voltage for the ADC. Resistors R6 and R7 are low tempco precision resistors (hence the different symbol)

I have a split rails supply, so the LM399 is heated by the negative half of the power supply while the reference is constructed using the positive half. The TL431 circuit creates a constant current regulator for LM399. Tweak R1 until there is about 3.3V across R6, this gives about 10mA drive current into LM399. Resistors R4-R7 divides the reference voltage to 1V that is intended for ICL7135. Tweak R4 until you get exactly 1.0000V at pin 6 of IC6. Optionally before tweaking R4, tweak R8 to clear out the null point.

[technix]

The ADC requires clocking. It is done using a crystal oscillator and divider based on CD4060.

Here the CD4060 generates the 100kHz clock signal required by the ADC by dividing 12.8MHz crystal oscillator down. I chose 100kHz for its ability to reject both 50Hz and 60Hz noise. The 555 provides a few seconds of delay when the unit is switched on, allowing the LM399 to warm up. (I was thinking about using 300kHz but that frequency was warned against. 100MHz is a bit harder to achieve with CD4060 crystal oscillator though.)

[technix]

The front end is still up to design considerations. I am thinking about using relay-switched frontend.

[technix]

RMS converter is simple enough, based on AD736.

[technix]

The power supply here is based on a linear design. The input connectors corresponds to the output wires of the transformer I am targeting.

The four linear regulators generates the +/- 12V and +/- 5V rails for the analog section. The single switch-mode regulator generates the 5V rail used by the digital section.

[David Hess]

I have considered doing the same thing but I would go with a simple microcontroller based design and delta-sigma instrumentation converter and push the performance.  One thing which would be really nice is being able to compute and display the standard deviation of DC measurements to measure low frequency noise.

I would also consider designing it as an electrometer with a relay switched 10M shunt for general use.

* Little to no microcontroller involvement. If MCU is required for some reason, use up to two AT89C2051s (pot-fest)

The core ADC chip I am using is ICL7135 @ 300Hz.

An ICL7135 at 300Hz what?  300 conversions per second?  300kHz clock for 2 power line cycle integrations?

The power supply here is based on a linear design. I would prefer a switch-mode one though, if I can find a negative-voltage version of LM2596.

A switching regulator has the potential to add a lot of noise.  I might use one for the LED supply (and float it) though because the current is so much higher than the rest of the circuits.

The front end is still up to design considerations. I am thinking about using relay-switched frontend.

Even old digital multimeters often used JFETs and MOSFETs for switching.  Good relays are awfully expensive.

[technix]

I have considered doing the same thing but I would go with a simple microcontroller based design and delta-sigma instrumentation converter and push the performance.  One thing which would be really nice is being able to compute and display the standard deviation of DC measurements to measure low frequency noise.

I would also consider designing it as an electrometer with a relay switched 10M shunt for general use.

* Little to no microcontroller involvement. If MCU is required for some reason, use up to two AT89C2051s (pot-fest)

The core ADC chip I am using is ICL7135 @ 300Hz.

An ICL7135 at 300Hz what?  300 conversions per second?  300kHz clock for 2 power line cycle integrations?

300Hz at its clock input pin.

The power supply here is based on a linear design. I would prefer a switch-mode one though, if I can find a negative-voltage version of LM2596.

A switching regulator has the potential to add a lot of noise.  I might use one for the LED supply (and float it) though because the current is so much higher than the rest of the circuits.

Well then, 7812/7912 and 7805/7905 can still work.

The front end is still up to design considerations. I am thinking about using relay-switched frontend.

Even old digital multimeters often used JFETs and MOSFETs for switching.  Good relays are awfully expensive.

I will look into that. I just have to imagine what kind of heatsinking I am subscribing into.

[David Hess]

300Hz at its clock input pin.

I think you mean 300 kHz then.

Even old digital multimeters often used JFETs and MOSFETs for switching.  Good relays are awfully expensive.

I will look into that. I just have to imagine what kind of heatsinking I am subscribing into.

There should not be any even if you switch high current ranges.

Check out Dave's discussion about designing a better multimeter which deals with switching the current ranges.

[Kleinstein]

Today I would also prefer an sigma delta converter based design - but this is a little hard to get if you want through hole only. There are mainly to choices for an DMM:
1) use a display oriented ADC like the ICL7135, so that every ADC step directly corresponds to one display digit. Adjustment is usually done by quite a lot of pots, like one for every range, but no µC needed.

2) use an ADC with higher resolution and do adjustment and compensation for divider errors in software. This is how modern DMMs are usually build, as there are no or very few trim pots needed. However there are not many high resolution ADCs available in DIP form.

For switching, usually CMOS switches are easy to use as long as the voltage is low (e.g. +-15 V) and inside the supply range. Relais might have trouble with thermal EMF and are bulky. If feasible I would avoid them for low voltage signals, though not at all costs.

For the input there is also the choice of classical form with one basic voltage range (like 200 mV) and than a multi tap divider for all the other range. This is how old analog meters and cheap old  DMMs were build. The other option is a variable amplification or attenuation behind the initial amplification for ranges of maybe 100 mV/1 V / 10 V and than only one input divider of about 100:1 for all the higher voltage ranges. This allows lower noise and high input impedance (e.g. > 10 GOhms) for the low ranges without using the divider, but it usually needs a higher supply (like +-15 V.. +-20V). So this is mainly an option for mains powered meters.

A relatively easy version would be a relay switched 100:1 divider followed by an CMOS switched amplifier for something like times 10 / 1 / 0.1 .

[David Hess]

Today I would also prefer an sigma delta converter based design - but this is a little hard to get if you want through hole only.

For expediency, I settle for small outline, shrink small outline package, or mini small outline parts where absolutely required.  Or they can be mounted on a DIP adapter.

[bitseeker]

technix, this is a cool project. Going to follow along. I agree with David to stick with linear supplies. You're not going to be drawing much power here, so better to minimize the noise.

David Hess, interested to see your version as well and what more can be accomplished at the DIY level when not limited to THT.

[David Hess]

David Hess, interested to see your version as well and what more can be accomplished at the DIY level when not limited to THT.

The delta-sigma ADC could be the only surface mount part but logic or a microcontroller is required to interpret its output.

I have many standard capability bench DMMs so if I went to design something from scratch, I would want to include capabilities that I miss like an electrometer input to medium voltages (at least +/-20V), low current measurement, and automated low frequency noise measurement which is just a firmware function if you have a microcontroller.

And at least these three capabilities would be pretty easy to add to a more conventional design.

Other all through hole designs which may be more fun perhaps but a lot more work include the multi-slope run-up integrating converter like the old Siliconix LD series (I do not think anybody makes an integrated chip set for this type anymore), the charge balancing voltage-to-frequency converter like that commonly implemented with the LTC1043, and the ratio counting design which a lot of early multimeters used.  The ratio counting design is interesting because a universal counter with A/B ratio mode could be used as the display during development.

[Kleinstein]

I would not go for a voltage to frequency converter design anymore. The variable frequency causes trouble with nonlinearity due to charge injection. So accuracy is limited. If you really want to build your own ADC, one could do an integrating converter (e.g. similar the HP34401, which is also used in some supplies) based on an µC - even one in an DIP package. Still using a ready made SD ADC is way easier and some come in an still easy to solder SO8 package.

A ICL7135 based solution might be still fun, as it can work without an µC. But don't expect the highest performance. It also gets less attractive with more ranges, as individual trimmers are needed. So I would try to keep such an design simple as it will not get ultimate performance (good INL, low noise, flexible reading rates) anyway. I would even skip auto-ranging on this, at least for the beginning.

[technix]

Today I would also prefer an sigma delta converter based design - but this is a little hard to get if you want through hole only. There are mainly to choices for an DMM:
1) use a display oriented ADC like the ICL7135, so that every ADC step directly corresponds to one display digit. Adjustment is usually done by quite a lot of pots, like one for every range, but no µC needed.

2) use an ADC with higher resolution and do adjustment and compensation for divider errors in software. This is how modern DMMs are usually build, as there are no or very few trim pots needed. However there are not many high resolution ADCs available in DIP form.

For switching, usually CMOS switches are easy to use as long as the voltage is low (e.g. +-15 V) and inside the supply range. Relais might have trouble with thermal EMF and are bulky. If feasible I would avoid them for low voltage signals, though not at all costs.

For the input there is also the choice of classical form with one basic voltage range (like 200 mV) and than a multi tap divider for all the other range. This is how old analog meters and cheap old  DMMs were build. The other option is a variable amplification or attenuation behind the initial amplification for ranges of maybe 100 mV/1 V / 10 V and than only one input divider of about 100:1 for all the higher voltage ranges. This allows lower noise and high input impedance (e.g. > 10 GOhms) for the low ranges without using the divider, but it usually needs a higher supply (like +-15 V.. +-20V). So this is mainly an option for mains powered meters.

A relatively easy version would be a relay switched 100:1 divider followed by an CMOS switched amplifier for something like times 10 / 1 / 0.1 .

About that CMOS switched amplifier, is CD4051 series a good switching element?

About controlling the board, I am thinking preferring GAL16V8 or 74HC over AT89C2051 (the only allowed MCU in this design) Is it doable?

[technix]

I would not go for a voltage to frequency converter design anymore. The variable frequency causes trouble with nonlinearity due to charge injection. So accuracy is limited. If you really want to build your own ADC, one could do an integrating converter (e.g. similar the HP34401, which is also used in some supplies) based on an µC - even one in an DIP package. Still using a ready made SD ADC is way easier and some come in an still easy to solder SO8 package.

A ICL7135 based solution might be still fun, as it can work without an µC. But don't expect the highest performance. It also gets less attractive with more ranges, as individual trimmers are needed. So I would try to keep such an design simple as it will not get ultimate performance (good INL, low noise, flexible reading rates) anyway. I would even skip auto-ranging on this, at least for the beginning.

I know that this is doable and even AT89C2051 have the required hardware for it (a single analog comparator) but as I have said this project requires as little MCU involvement as possible.

[Kleinstein]

cd4051 (more likely CD 4052/3) switches are cheap, but limited to something like a +-6 .. 8 V supply (depending on the letters at the end / exact version).  Otherwise they can be good enough.
There are a few switches like ADG409 that also work with higher voltage up to about +-20V, e.g. if you want high impedance even up to +-20 V or so.

For Autoranging I would prefer an simple µC, though also just 74HC... can work.
AFAIK GAL16V...  run pretty hot and are rather outdated - maybe even hard to get. So I would avoid them, even if it takes a few more 74HC chips.

[David Hess]

I would not go for a voltage to frequency converter design anymore. The variable frequency causes trouble with nonlinearity due to charge injection. So accuracy is limited.

The charge balancing types are as good as a dual-slope integrating converter like a ICL7135 though and can be better.  They do not have autozero built in of course so that might have to be added which makes them noncompetitive economically; ICL7135 type converters are dirt cheap.

If you really want to build your own ADC, one could do an integrating converter (e.g. similar the HP34401, which is also used in some supplies) based on an µC - even one in an DIP package.

I do not know about the HP34401 but its fully documented predecessors are very similar to the multi-slope run-up LD series made by Siliconix.  I do not know that I like them all that much but I have to admit that they are low noise and I guess that is the way to go if you want better than integrated delta-sigma performance.  The LD series only went to 4.5 digits and competed with the ubiquitous dual-slope converters.  I am not sure why they were discontinued.

[bitseeker]

I have many standard capability bench DMMs so if I went to design something from scratch, I would want to include capabilities that I miss like an electrometer input to medium voltages (at least +/-20V), low current measurement, and automated low frequency noise measurement which is just a firmware function if you have a microcontroller.

And at least these three capabilities would be pretty easy to add to a more conventional design.

Hmm, could add some (all?) of those to an existing meter a la uCurrent style external module?

Other all through hole designs which may be more fun perhaps but a lot more work include the multi-slope run-up integrating converter like the old Siliconix LD series (I do not think anybody makes an integrated chip set for this type anymore), the charge balancing voltage-to-frequency converter like that commonly implemented with the LTC1043, and the ratio counting design which a lot of early multimeters used.  The ratio counting design is interesting because a universal counter with A/B ratio mode could be used as the display during development.

Yeah, those sound much more serious. I have yet to get into things like the multi-slope converter in the 34401A.

[bitseeker]

Shouldn't be too hard to adapt to use a micro-USB or barrel jack power supply. Definitely a good idea for those who don't want to mess with mains power.

[Monkeh]

* Bench multimeter with built-in mains power supply for 240V (replaceable transformer for 120V operation)

Don't be silly, just use a dual primary transformer.

[technix]

Can you consider an optional build without mains wiring? Maybe a USB powerbank or USB supply. Or even an external 12V AC plugpack if 5V DC is not practical.

I think less is more in this style of project.

This makes the power supply subsystem a bit difficult to design. And I have a specific spare mains-rated transformer I want to use.

[David Hess]

Can you consider an optional build without mains wiring? Maybe a USB powerbank or USB supply. Or even an external 12V AC plugpack if 5V DC is not practical.

I think less is more in this style of project.

Doing this adds a serious complication.  You do *not* want any possibility of sharing a common ground between the power inputs and the signal inputs.  While any given USB power supply may or may not be galvanically isolated including from earth ground, a USB port on a computer or hub shares a common ground with other peripheral ports.  Bench voltmeters inputs are galvanically isolated from their power input and earth ground for a very good reason.

So if you want to use USB power, you need to also design in an inverter or other scheme for galvanic isolation.

[David Hess]

I have many standard capability bench DMMs so if I went to design something from scratch, I would want to include capabilities that I miss like an electrometer input to medium voltages (at least +/-20V), low current measurement, and automated low frequency noise measurement which is just a firmware function if you have a microcontroller.

And at least these three capabilities would be pretty easy to add to a more conventional design.

Hmm, could add some (all?) of those to an existing meter a la uCurrent style external module?

Except for low frequency noise measurement they could and even that could be added if there is a data output but it would be easier to include them in a multimeter design from the start.

Other all through hole designs which may be more fun perhaps but a lot more work include the multi-slope run-up integrating converter like the old Siliconix LD series (I do not think anybody makes an integrated chip set for this type anymore), the charge balancing voltage-to-frequency converter like that commonly implemented with the LTC1043, and the ratio counting design which a lot of early multimeters used.  The ratio counting design is interesting because a universal counter with A/B ratio mode could be used as the display during development.

Yeah, those sound much more serious. I have yet to get into things like the multi-slope converter in the 34401A.

Ah, so the selection of the ICL7135 was not so much for tradition as simplicity.