Trimming my reference voltage: pot divider, or trim the chip itself?

[technix]

In my DMM project I have settled down on using REF03 (DIP-8 THT) or ADR03 (SO-8 SMT) precision 2.5V reference chip for the voltage reference, divided with a resistive divider down to 1V for ICL7135. Now how should I trim this?

The two possibilities are:

1) Connect the voltage output to a 10-turn pot and trim the voltage there to 1V, or
2) Connect the voltage output to a fixed 1K/1.5K 1% metal film resistor divider and use the TRIM input of the chip to trim the voltage to precisely 1V.

Which one should I go?

[Alex Nikitin]

In my DMM project I have settled down on using REF03 (DIP-8 THT) or ADR03 (SO-8 SMT) precision 2.5V reference chip for the voltage reference, divided with a resistive divider down to 1V for ICL7135. Now how should I trim this?

The two possibilities are:

1) Connect the voltage output to a 10-turn pot and trim the voltage there to 1V, or
2) Connect the voltage output to a fixed 1K/1.5K 1% metal film resistor divider and use the TRIM input of the chip to trim the voltage to precisely 1V.

Which one should I go?

Let's do a simple calculation. For this chip 1 count = 50ppm FS. Ideally you would prefer the reference drift below 1 count in a reasonable temperature range, say +/- 5C, making it <10ppm/C drift. And that is the total drift, of the voltage reference itself and the resistive divider after it. A multiturn cermet pot has a tempco ~100ppm/C. Even top of the range Bulk Foil pots are specified at 10-15ppm/C. I would not use a single pot as a divider in this application. It would make sense to use a couple of low tempco (below 10ppm/C) 0.1% resistors (say, 1K and 1.5K) and a larger value trimmer (say 20K) with 470K resistor from the trimmer viper pin to the reference input, which would allow you to trim the reference in about (+0.3% -0.2%) range with less than 1ppm/C temperature drift from the adjustment circuit.

Cheers

Alex

[technix]

In my DMM project I have settled down on using REF03 (DIP-8 THT) or ADR03 (SO-8 SMT) precision 2.5V reference chip for the voltage reference, divided with a resistive divider down to 1V for ICL7135. Now how should I trim this?

The two possibilities are:

1) Connect the voltage output to a 10-turn pot and trim the voltage there to 1V, or
2) Connect the voltage output to a fixed 1K/1.5K 1% metal film resistor divider and use the TRIM input of the chip to trim the voltage to precisely 1V.

Which one should I go?

Let's do a simple calculation. For this chip 1 count = 50ppm FS. Ideally you would prefer the reference drift below 1 count in a reasonable temperature range, say +/- 5C, making it <10ppm/C drift. And that is the total drift, of the voltage reference itself and the resistive divider after it. A multiturn cermet pot has a tempco ~100ppm/C. Even top of the range Bulk Foil pots are specified at 10-15ppm/C. I would not use a single pot as a divider in this application. It would make sense to use a couple of low tempco (below 10ppm/C) 0.1% resistors (say, 1K and 1.5K) and a larger value trimmer (say 20K) with 470K resistor from the trimmer viper pin to the reference input, which would allow you to trim the reference in about (+0.3% -0.2%) range with less than 1ppm/C temperature drift from the adjustment circuit.

Cheers

Alex

I took the trimming circuit from REF03 datasheet (unmodified) and specified my usual 1% metal film through hole resistors there. I think the relative tempco won't be that much as those are made from the same material if I chose those from batches made by the same manufacturer. The fixed divider is 15k/10k, the same 1% metal film through-hole job.

[Gyro]

I think the application circuit might be a comment on your choice of reference rather than needed accuracy. The Ref03 is specified as 10ppm/'C typ, 50ppm/'C max.

Assuming that you have a typical one then according to Alex's calculation then that is your entire drift budget right there, before you even start adding in the TC of the divider resistors.

Unfortunately there is no way to guarantee that standard metal film resistors, even from the same manufacturer will drift the same way, even between individual resistors from the same batch. That's why their specified +/-50ppm/'C and they don't provide a tracking spec. Low TC 0.1% resistors aren't very expensive, why compromise this early on.

As Alex says, you need to keep your trimming range to an absolute minimum, trimmers don't do well at all for TC.

The venerable old Nat Semi (Now TI) AN184 'References for A/D Converters' is an excellent resource for understanding error budgets and compromises, eg using higher spec reference and lower spec resistors and vice-versa. Maybe you could try a low cost buried zener reference like the REF102C...

http://www.ti.com/lit/an/snva510b/snva510b.pdf

[technix]

I think the application circuit might be a comment on your choice of reference rather than needed accuracy. The Ref03 is specified as 10ppm/'C typ, 50ppm/'C max.

Assuming that you have a typical one then according to Alex's calculation then that is your entire drift budget right there, before you even start adding in the TC of the divider resistors.

Unfortunately there is no way to guarantee that standard metal film resistors, even from the same manufacturer will drift the same way, even between individual resistors from the same batch. That's why their specified +/-50ppm/'C and they don't provide a tracking spec. Low TC 0.1% resistors aren't very expensive, why compromise this early on.

As Alex says, you need to keep your trimming range to an absolute minimum, trimmers don't do well at all for TC.

The venerable old Nat Semi (Now TI) AN184 'References for A/D Converters' is an excellent resource for understanding error budgets and compromises, eg using higher spec reference and lower spec resistors and vice-versa. Maybe you could try a low cost buried zener reference like the REF102C...

http://www.ti.com/lit/an/snva510b/snva510b.pdf

I don't have a separate 15V rail to do this and I may have trouble get a fourth SMPS into my power module (there is already three of them: +5V +/-10mV 100mA analog positive with MC34063, -5V +/-10mV 100mA analog negative with MC34063 and +5V +/-150mV 500mA digital power with LM2596, all sourcing from a 9V battery) so I am pretty much stuck with REF03. I don't think MC1403 or TL431 (which can't even be trimmed) will cut it. This build have one special limit that it is THT-only (or ADR03 will work)

All the reference chips still made that supports this level of drift parameters are either SMT-only or go directly to LM399 (which I do have a few) but 399 will throw my battery life out of window outright even if I squeezed in the required +15V power supply.

[Gyro]

In that case you're going to have to tighten up on your resistor specs (have you checked the app note yet?) - have you considered what you're going to use  for your input dividers yet? You're going to need close tolerance low TC for those too.

I understand you wanting to stay through-hole but for your reference but with your limited supply rail too you are restricting yourself quite badly - can't you even allow yourself one of those little SOIC8 to DIP converter boards, that would open up lots of other options (eg the LM4140 to pick one at random). The long term drift of the REF03 is quite poor too.

[Alex Nikitin]

This build have one special limit that it is THT-only (or ADR03 will work)

Have a look at LT1019*CN8-2.5 and LT1460*CN8-2.5, LT1460GCZ-2.5 .

Cheers

Alex

[technix]

In my DMM project I have settled down on using REF03 (DIP-8 THT) or ADR03 (SO-8 SMT) precision 2.5V reference chip for the voltage reference, divided with a resistive divider down to 1V for ICL7135. Now how should I trim this?

The two possibilities are:

1) Connect the voltage output to a 10-turn pot and trim the voltage there to 1V, or
2) Connect the voltage output to a fixed 1K/1.5K 1% metal film resistor divider and use the TRIM input of the chip to trim the voltage to precisely 1V.

Which one should I go?

Let's do a simple calculation. For this chip 1 count = 50ppm FS. Ideally you would prefer the reference drift below 1 count in a reasonable temperature range, say +/- 5C, making it <10ppm/C drift. And that is the total drift, of the voltage reference itself and the resistive divider after it. A multiturn cermet pot has a tempco ~100ppm/C. Even top of the range Bulk Foil pots are specified at 10-15ppm/C. I would not use a single pot as a divider in this application. It would make sense to use a couple of low tempco (below 10ppm/C) 0.1% resistors (say, 1K and 1.5K) and a larger value trimmer (say 20K) with 470K resistor from the trimmer viper pin to the reference input, which would allow you to trim the reference in about (+0.3% -0.2%) range with less than 1ppm/C temperature drift from the adjustment circuit.

Cheers

Alex

I can tolerate drift up to 2 LSB. So given the structure of REF03 (or ADR03,) to minimize drift I should use matched 1k/1.5k 25ppm resistors? (Don't go 10ppm - no seller stock them here and those are very, very, very expensive - a few US dollars each, orders start with quantity of 100 expensive.) Can I use my stock 1% job for the trimming parts?

[Gyro]

Don't go 10ppm - no seller stock them here and those are very, very, very expensive - a few US dollars each, orders start with quantity of 100 expensive.)

Really? I'm genuinely surprised that you can't get them economically in your country! You learn something new every day.

A suggestion that might just work: You could always put a pair of your 1% ones in close thermal coupling and actually measure the drift while you heat / cool them. With luck you might be able to match a pair that track each other well. You may need to repeat several times, and be careful soldering them afterwards so that you don't shift their drift characteristics or value too much. Better than nothing.

[Kleinstein]

With the switched mode converter, I would consider something like a forward or resonant converter (e.g. Royer) to produce all voltages together from one transformer. Noise is often much lower than hard switched converters, especially flyback.
Depending on the input stage one may want something like +-12 to  +-15 V anyway.
When battery operated power consumption should be much lower, more like  < 100 mW for the hole circuit.

Still a 2.5 V or 1.2x V reference might be ok. There are several to choose from. With the reference TH is also prefered, as SMD versions tend to react stronger to mechanical stress and the thin case may react faster to humidity. Often the TC is considerably better in the 10-30 C range - so one might get away with a tested lower grade, possibly even LM285 or similar. Still long term drift may be a problem.

Usually the external deviver is better than trimming at the ref chip itself. It is also much more flexible so you can change the reference if needed. The two resistors for coarse division should be low TC (e.g. < 25 ppm/K) - it gets a little less critical if you start from 1.2 V. The third resistor to connect to the trimmer is less critical, as the trimmer should not adjust much more than 1 %.  One might even consider including a digital pot for trimming - this way the µC could to do fine tuning depending on the range selected and thus do software calibration even with a display oriented ADC.

[Alex Nikitin]

A suggestion that might just work: You could always put a pair of your 1% ones in close thermal coupling and actually measure the drift while you heat / cool them. With luck you might be able to match a pair that track each other well. You may need to repeat several times, and be careful soldering them afterwards so that you don't shift their drift characteristics or value too much. Better than nothing.

A better suggestion would be to use five exactly the same resistors from one batch, three in parallel on the top and two in parallel at the bottom part of the divider. The actual value does not matter in a reasonable range (from 1K to 47K or about, ideally from 3.3K to 10K), just select the best available type.

Cheers

Alex

[dom0]

[  ] Use trimpots
[x ] don't trim at all
[x ] calibrate in software instead

Because a software trim will beat an analog gain trim easily in repeatability, stability, and ease of adjustment (speed and precision). I'm not sure why anyone would want to build one of these trimmer monsters of multimeters like they did in the early 70s. I mean, you do know, that you will need the same, expensive, scheme of adjustment for each range in both offset and gain?

Well you can do it in digital hardware, too (at least if you're a masochist).

[technix]

With the switched mode converter, I would consider something like a forward or resonant converter (e.g. Royer) to produce all voltages together from one transformer. Noise is often much lower than hard switched converters, especially flyback.
Depending on the input stage one may want something like +-12 to  +-15 V anyway.
When battery operated power consumption should be much lower, more like  < 100 mW for the hole circuit.

Still a 2.5 V or 1.2x V reference might be ok. There are several to choose from. With the reference TH is also prefered, as SMD versions tend to react stronger to mechanical stress and the thin case may react faster to humidity. Often the TC is considerably better in the 10-30 C range - so one might get away with a tested lower grade, possibly even LM285 or similar. Still long term drift may be a problem.

Usually the external deviver is better than trimming at the ref chip itself. It is also much more flexible so you can change the reference if needed. The two resistors for coarse division should be low TC (e.g. < 25 ppm/K) - it gets a little less critical if you start from 1.2 V. The third resistor to connect to the trimmer is less critical, as the trimmer should not adjust much more than 1 %.  One might even consider including a digital pot for trimming - this way the µC could to do fine tuning depending on the range selected and thus do software calibration even with a display oriented ADC.

The SMPS have no need for isolation (already battery powered) so I am using classic non-isolating converters.

[technix]

[  ] Use trimpots
[x ] don't trim at all
[x ] calibrate in software instead

Because a software trim will beat an analog gain trim easily in repeatability, stability, and ease of adjustment (speed and precision). I'm not sure why anyone would want to build one of these trimmer monsters of multimeters like they did in the early 70s. I mean, you do know, that you will need the same, expensive, scheme of adjustment for each range in both offset and gain?

Well you can do it in digital hardware, too (at least if you're a masochist).

Well the current version have no microcontroller in it. All it have that remotely resembles software is a few GALs.

[Kleinstein]

Not using a µC is a poor decission: the DMM would need quite a large number of trimmers (one per range). Software calibration is much easier and the µc could also do things like auto-ranging and display controll. So no more outdated GAL needed. Even if you stick to the old ICL7135 software calibration is possible with one digital pot to replace something like 10 analog ones. It gets even easier with a binary ADC.   There are even µC with integrated 24 Bit ADC.

[technix]

Not using a µC is a poor decission: the DMM would need quite a large number of trimmers (one per range). Software calibration is much easier and the µc could also do things like auto-ranging and display controll. So no more outdated GAL needed. Even if you stick to the old ICL7135 software calibration is possible with one digital pot to replace something like 10 analog ones. It gets even easier with a binary ADC.   There are even µC with integrated 24 Bit ADC.

Well if that is the case I might as well go directly to MCP3911/MCU system, and upgrade the reference to ADR02B and have all trimming done in software. I am still going to keep the AD736 RMS to DC converters so I can keep the main MCU simple. The MCU part I am yet to decide - ATmega32U4 with built-in USB interface, ATmega328P with CH340G USB interface, or CH563 with both USB and Ethernet.