Another DIY voltage reference...

[Oldcodger]

Originally Posted by BigClick

Is there anyone local(ish) to Walsall who has a calibrated DVM that I could compare my thirty year old BT Digital to?

I'd be surprised if the BT digital is not as per calibrated DVM. I've got one of these and a recently "acquired" professionally calibrated 4 digit Fluke meter and I was amazed to find that the BT meter was spot on.

[Al (astral highway)]

Originally Posted by mhennessy

Hello...

Hey Mark, what a brilliant idea. I may replicate this and in any case, it brings up some very interesting points indeed. Thanks for posting this

One query I have is on the way you've passed the various multicoloured wires through the top of the veroboard (pic 2). This looks amazingly professional but I'm assuming it also has a functional purpose - is it to strain relieve the wires?

Thanks again!

[Brian R Pateman]

I often do this too. It does provide strain relief but also means that it is easier to route the wires from the board to the switches. Consequently it's more difficult to trap them when putting a small plastic case together.

I'll be interested in Mark's thinking.

[mhennessy]

Originally Posted by astral highway

One query I have is on the way you've passed the various multicoloured wires through the top of the veroboard (pic 2). This looks amazingly professional but I'm assuming it also has a functional purpose - is it to strain relieve the wires?

Thanks for the compliment

Yes, strain relief. Fiddly, but worth it. IIRC, I had to slightly enlarge the holes in the Veroboard as the wire was too thick, but that only took 30 seconds...

All the best,

Mark

[Julesomega]

Returning to this thread now I've had time to read and think about it, I can see this is a useful and well thought out project. The statistical theory of improving accuracy by multiplication is counter-intuitive and would not have occurred to me. The range switching and output protection are impressive augmentations.

Following the "Statistical DVM Reference" webpage shows the latter sophistications progressively being added while the statistical basis is being dropped, and we end up with one reference device being used for the 2.5V o-p, 2 for 5V, and it is not til you select 12.5V that you benefit from the full 0.48% accuracy from the 1% devices, Would it perhaps be more comprehensive to derive all outputs from the 12.5V by using resistive dividers made of series or parallel combinations of equal values?

[Radio Wrangler]

The statistics are fascinating, and the experimental confirmation is good.

There is one factor which can spoil things, though. The statistical distribution has to have a central peak which is accurate, or your result has convergence around this different value. Beware of parts sold as different grades. If all the best parts are selected out and sold more profitably, the remaining parts have a pdf with a hole in the middle. This ceases to be damaging if the twin peaks are symmetrically disposed around the perfect value. Systematic errors can also skew the pdf peak, so all the parts in one batch, or all the parts from one production machine could tend in the same direction.

Lots of theories adding worries, but in the end, Mark's measurements showed he got a real improvement. Neat bit of construction, too.

David

[kalee20]

Originally Posted by mhennessy

I first ran into this when Douglas Self published a pre-amp design back in 1996... The design called for a 50nF capacitor, so he used five 10nF 1% polystyrene types that gave 50nF at 0.48%. I was impressed!

At the same time, the Maplin catalogue had a "box-out" that explained that combining 1% resistors gave you an overall tolerance of 1%. Hmm...

Ignoring switch resistance, Maplin are correct, combining 1% resistors does give an overall tolerance of 1%. (All 1% high - overall value is 1% high. All 1% low, overall value is 1% low. So, 1% is the best that can be guaranteed.)

However, if the resistors are independent of each other (ie from different batches) then it is likely that above-tolerance in some, will combine with below-tolerance in others. So the final value is more likely to be rather better than 1%, even though it cannot be guaranteed.

This probably explains Doug Self's result, that his 10nF capacitors came from a spread of values. Of course, he could have measured each one, combined a few lows with a few highs, and got significantly better than 'worst case.'

The same would apply to voltage references.

But - returning to capacitors - not necessarily to them. I was told by a capacitor manufacturer's rep (polyester film types), that they had improved the accuracy of their process capability to the point where they could be confident of values within 2%. So, for their 100nF 10% capacitors, they set their machines to make 93nF capacitors (which would turn out between 91 and 95nF), label them 100nF 10%, and sell them saving 7% on the materials cost of a true 100nF. Everything was in spec, but for the chap who bought a boxful hoping to be able to measure them and fine one bang on 100nF, he'd be sorely disappointed!

[jez_145]

Yes, I agree that Maplin were essentially correct.

In his book "The Design of Active Crossovers", Doug Self discusses improving accuracy with multiple components:Gaussian Distribution in Section 12.2 (p342). He then goes on to say (p344) "You may object to putting four 1% resistors in series means that the worse case errors can be four times as great. This is obviously true - if all the components are 1% low, or 1% high, the total error will be 4%."

I really don't believe he meant this and suspect he was just checking to see if we were still awake !

Jerry

[mhennessy]

As has probably already been covered originally, if the components in the group have a Gaussian distribution, then the improvement in tolerance is root-N. Taking the Self RIAA preamp example, five 1% capacitors in parallel will have a combined tolerance of 0.1 divided by root-5, which is where the 0.48% comes from. BTW, this was not a measured "result", as implied in #27 - if you read the write-up of the pre-amp, the maths is explained fully. Self discusses the same thing in many of his works. Agree that the quote cited in #28 is out of character - I wonder if that's been clarified in the second edition?

It's quite reasonable to expect component like resistors to come off the production line with a decent Gaussian distribution - this is well documented in many text books - and that's why I take issue with Maplin's comments, which were clearly written by someone who didn't know about this because they simply made a glib statement with no hint that they knew the reality might be a little bit different. Yes, of course you could argue that they are correct in the extreme case, but in the real world, that rarely happens. As I say, it's the lack of knowledge that's the problem.

The joy of this is the ability to end up with something better than you can measure. Suppose you can measure resistors with 0.5% accuracy - not unrealistic for an affordable DMM - then ten pre-selected 10k resistors in parallel can give you a 1k resistor with a 0.16% tolerance. You can pay a lot of money for resistors with that sort of tolerance

Originally Posted by Julesomega

Following the "Statistical DVM Reference" webpage shows the latter sophistications progressively being added while the statistical basis is being dropped

I'll be honest: I'm confused by this comment. The very first diagram shows the 5 references and selector switch, and everything builds on that - the only places where there's no selector switch shown on a diagram is where it is omitted for clarity. If there are places where that is not immediately obvious, please let me know and I'll amend the text accordingly. One can see from the final schematic that the core idea is present throughout.

Originally Posted by Julesomega

and we end up with one reference device being used for the 2.5V o-p, 2 for 5V, and it is not til you select 12.5V that you benefit from the full 0.48% accuracy from the 1% devices

Yes, of course it's only "statistical" when using anything other than 2.5V. But as I selected the voltage sources, that's not an issue. You wouldn't want to put this into mass-production, but as a project for a time-rich hobbyist, that's fine.

Originally Posted by Julesomega

Would it perhaps be more comprehensive to derive all outputs from the 12.5V by using resistive dividers made of series or parallel combinations of equal values?

First of all you've got to get a whole load of precision resistors. Or select them and hope their value doesn't change when you solder them.

(You'll note the trimmer for the divide-by-10 option. I really didn't want to include that at first, so spent ages pre-selecting fixed resistors to do that job. The results were disappointing because they did indeed change their value during soldering.)

Having done that, you've got to buffer the output because there's no way you'd get a sufficiently low source impedance while retaining reasonable battery life. So now we're into the murky world of op-amp specs, plus don't forget the current it draws. Not impossible, of course, but it takes away from the KISS principle of the project.

Another drawback: battery life. You'll note that I'm only supplying current to the number of references required because when there isn't enough battery voltage to cope with 12.5V plus the dropout voltage of the LM317, there's still plenty of life in the cells. If you go with 12.5V followed by a potential divider, then you'll be throwing away perfectly good batteries for no good reason.

However, if the idea appeals, I'd consider using 10V and a high quality 10 turn potentiometer with a dial. In fact, if you haven't already seen it, take a look at my Time Electronics 1030 Microcal write-up - you could get a lot of ideas from that. The LM10 is a handy little IC.

https://www.markhennessy.co.uk/time_...1030_microcal/

OK, that moves away from the "statistical" element of the project, but it's important to realise that the whole thing was just a bit of fun. The idea came to me, I did some initial experimentation, and decided it was worth putting together and doing a bit of a write-up that might inspire others. That's all. It has proved to be very handy, and it's one of my more frequently used bits of gear. For example, I use it to test LEDs, as it's a 2.5mA current source.

I hope that helps,

Mark

[Julesomega]

I appreciate the KISS principle behind your design: my point is that you dropped the statistical benefit of multiple V-refs for no compelling reason. The most accurate reference is at the top of the chain. You divide this down with fixed resistors and there is no increase in battery loading. The division accuracy would benefit from combining resistors. When I make mine I will see if combinations of values will give a close 10:1 ratio.

You were able to select a) V-refs and b) resistors, to bystep the statistical approach. I may not be so fortunate, that's why I will go back to (your) first principles I will also add fine adjustment with a pot at the 12.5V point so that /all ranges can be set precisely with the one preset when I am next passing an obliging Calibration House.

[mhennessy]

Originally Posted by Julesomega

The most accurate reference is at the top of the chain. You divide this down with fixed resistors and there is no increase in battery loading.

Honestly, I've think you've missed the points I made about the potential divider. Obviously, I don't know your expertise level, so forgive me if any of this is stating the obvious, but from what you're saying it's clear that I do need to add a bit more detail.

Let's not worry about accurately selecting the resistors for your potential divider for the time being.

Instead, what value are you going to choose for these resistors? Or more pertinently, what's the total value of your resistor chain?

Let's say 10k. And to keep the numbers simple, let's say your input voltage is 10V rather than 12.5V.

This of course means you've got 1mA flowing in these resistors. Not a big deal, perhaps...

On the other side of the coin, what is the source impedance of your voltage source?

Obviously that will vary with the output voltage you select, and reaches a maximum at 5V, where it will be 2.5k.

This is a big problem for something that is supposed to be a voltage source.

I'm assuming you've read my deliberations around the protection part of the circuit, where I calculated that a 820 ohm resistor (in conjunction with that chunky Zener) would do the job of protecting the LM4040s? But I wasn't happy with such a high output impedance, and discussed how that would introduce errors - eventually ending up with the simple JFET solution...

Clearly, a source impedance of 2,500 ohms is considerably worse than 820, or indeed the 135 that I ended up with. Of course, it's your project, so you must run the numbers and decide what you can accept, but if you invest heavily in making a very accurate voltage source and throw away that accuracy in a potential divider that results in different output voltages depending on what meter you connect to it, then that is not a well balanced design - if doing that, you might as well save money/design effort on the voltage source. Indeed, if this was a commercial project, you would be obliged to do that (or do something else to balance the design).

OK, so to improve the performance of the potential divider, let's divide the resistances by a factor of 10. This gives a total of 1k, and a maximum source impedance of 250 ohms. This is much better, but hopefully you can now see the problem with current consumption that I mentioned: your divider is now drawing 10mA. Not good for something that runs on PP3s.

As I mentioned, the only way to resolve this conflict is to buffer the output of the potential divider. Not impossible, but definitely moves us out of KISS territory.

Originally Posted by Julesomega

I appreciate the KISS principle behind your design: my point is that you dropped the statistical benefit of multiple V-refs for no compelling reason.

If you've understood everything above, you now have your compelling reason!

What I've done ensures a consistently low output impedance at all settings of the voltage control, without having to use an op-amp to buffer a potential divider. That is excellent, though I prefer your choice of word. Compelling indeed

Originally Posted by Julesomega

You were able to select a) V-refs and b) resistors, to bystep the statistical approach. I may not be so fortunate, that's why I will go back to (your) first principles I will also add fine adjustment with a pot at the 12.5V point so that /all ranges can be set precisely with the one preset when I am next passing an obliging Calibration House.

I take your point about selection, but look at the context - I was only aiming for 0.1% initially, which is what the A version of the LM4040 does by default. The fact I was able to select to get much better than that is nice, but effectively "outside the scope of the project". My article includes showing the results of random sampling to show what can be achieved by someone without the means to select them, and the results are pretty good on the whole.

Because the LM4040 is not a high-precision voltage source, I would expect some drift with time. I'm quite up-front about this in the article - there are plenty of voltage references that are higher precision (and higher cost!) than the LM4040, and they should have much better long-term accuracy. But the point is simple - I had a few dozen LM4040s sitting around...

BTW, I didn't select resistors. I didn't need to

A word of caution: if you feel you must include trimmers, do remember that the resistive elements are usually not as good as fixed metal film resistors, so ensure that they exercise the absolute minimum influence over the final result.

I mention this when discussing the divide by 10 section - I put a 50 ohm preset in series with a 1,100 ohm resistor, which gives a plus/minus 2% adjustment range. So if the preset resistor changes by 10% over the years, say from 25 ohms (if it's in the mechanical centre) to 27.5 ohms, the total change seen in the lower limb of the potential divider is from 1125 to 1127.5 ohms - about a quarter of a percent... Imagine if I'd used a 2k preset there, with no resistor. It actually takes a lot of effort to arrive at the optimum amount of range for a pre-set; you need to cover the range caused by the known tolerances plus a bit for safety, but no more. Some test gear - like the old Thurlby PL-series power supplies - are nearly impossible to set up because their presets have far, far too much range.

Again, I apologise if any of this is stating the obvious - I don't know your level of expertise and experience, and can only respond based on what you've written. But I honestly hope this helps. Good luck with your design!

All the best,

Mark

[Radio1950]

Here's my Meter Checker made last year to give me confidence, if I get an unusual measurement.

Made around an AD584 Voltage Reference and Dale 0.1% resistors, all sourced from Mouser, and one extrapolated accuracy resistor.
Used with a 20 volt 5 amp DC supply input, or lower current requirement if the 1.0 amp range is not needed.

It gives-

2.5, 5.0, 10.0 volts
1.0A, 0.1A, 0.01A, 1mA, 50uA, provision for custom, current into 0.0 ohms with 10.0 volts selected. It is proportionately lower if 5.0 or 2.5 volts is selected.
200K, 10K, 1K, 100R, 10R, provision for custom, ohms.

If used to check meters on current scale, you have to allow for meter resistance, and make a mathematical estimate of error.
But it will give a very close result.

Designed for 1.0 % checks, and I probably get 2.0% results or better, which is fine for me. It is only a checker, not a calibrator.

The AD584 has a double tiered darlington type boost circuit to give the 1.0 amp output at 10.0 volts, and all fed from a current boosted 7812 regulated 12 volts. It is good to about 1.5 amp at 10 volts before any noticeable voltage sag. I wanted a simple circuit without feedback current control.

Now, back to my reading about Wurzburg.
.

[David G4EBT]

This topic comes round from time to time (every Leap Year?) and clearly interests quite a few forum members, myself included, but given that we're hobbyists, in my case for the most part restoring vintage equipment which used 10% or even 20% tolerance components, designed to operate on mains power supplies with a voltage input commonly ranging from 220V - 260V, 40 - 100Hz, this striving for accuracy (other than for academic rather than practical reasons), certainly takes us into the 'Volts Nut' arena referred to by Mark in the original thread. See:

https://lists.febo.com/mailman/listi...lists.febo.com

Sure, in a laboratory or sophisticated precision electronic equipment design, construction, testing and repair,or simply for academic interest, but on an upturned chassis of a 1950s domestic radio? I don't think so.

The original thread ran from Jan 2016 - May 2016, and had 112 posts.

In post 5 of that thread, I mentioned that as a compulsive home-brewer, I'd built the 10V Voltage Reference which appeared in Everyday Practical Electronics in June 2011, based on the AD588AQ IC. That IC (.06% accuracy) must have been much cheaper back then as it's now £37.50 and I wouldn't give it a second glance. It was useful and reassuring to check the analogue and digital meters at that time (and since), the most expensive of which was a Toolzone, EL060, which cost me a tenner, the cheapest of which were Maplin 'two for a fiver'. They were all either spot on 10V or no more than 2% or so out.

In post 11 of the original thread, Jeremy (Pamphonica) pointed to a ready made voltage reference available on ebay described as: 'AD584 4 Channel 2.5V/7.5V/5V/10V High Precision Voltage Reference Module'. It's still available at £14.04 post free - less than half the current price of the AD588AQ IC that I used in the EPE 10V Voltage Reference I built.

https://www.ebay.co.uk/c/581937500

Other than the interest and enjoyment of home-brewing, such an off-the-shelf Voltage Reference does blow the cost/benefit argument out of the water. But that said, if cost/benefit was the only criteria, we wouldn't be restoring old radios, the costs of which often far exceeds the monetary value of the radio. (Think DAC90A). All hobbies cost money - that's the price we pay for the enjoyment we derive.

Of course, all that a Voltage Reference will do is to give an indication of accuracy of a meter on the DC Volts ranges, but if we want to check the accuracy of the resistance ranges, we can of course do that with close tolerance resistors. Even 0.1% accuracy resistors are available cheaply:

https://uk.rs-online.com/web/c/?sra=...1%25+resistors

If we have a variable voltage power supply (I do) having checked that my meter voltage range is accurate, I can set the output of the power supply to exactly 10 Volts, then with the meter on the mA range, can connect say a 100 Ohm resistor across the power supply with the meter in series. From Ohms Law, I can expect to see 10/100 - 0.1A (100mA). If I substitute the 100 Ohm resistor for 1,000 Ohms, I can expect the meter to read 10/1000 - .01A (10uA), and so on.

Really, as a hobbyist, for all practical purposes, that's good enough for me.

I'm an old guy restoring radios in a garden shed - I'm not putting rockets into space!

The most alarming thing to me about the pics below has nothing to do with measuring Voltage, but the inability of my brain to accurately calculate elapsed time since past events. It's surreal, and more than a little scary, that eleven years has elapsed since I built that, but it seems like last week.

[mhennessy]

Sure, if time and cost are your primary concerns, then just pick up a pre-built module. To support people wishing to do that, I did a bunch of reviews of AD584-based units a while back: https://www.markhennessy.co.uk/ad584_references/

But obviously time and cost were not why I built my original project. It was an idea I had, prototyped, and decided to build. I had a lovely time doing it, and ended up with something which is considerably better than those Chinese modules. Yes, it might be better than David needs, but this forum covers a lot more than just 1950s domestic radios. And we're not all hobbyists

I have a Chinese AD588 module sat here. It is encased, complete with a rechargeable (via USB) battery, and cost less than a tenner if memory serves. I never did get around to documenting it - partly because it doesn't really "bring anything to the party" over and above the AD584 reviews I'd already done, and partly because I found it was no longer available when I did think about starting it. I see that the ICs are readily available for just a few pounds on eBay, but obviously you're taking a bit of a chance there - they could easily be fake or rejects. But you'd only buy an AD588 today is to repair something that used one IMHO - there are cheaper/better references out there.

Regarding the 10% capacitors made to 2% tolerance (not accuracy as it is often called), that happens. Just like petrol, the old British Standard said "at least a gallon per indicated gallon" now it's 1 litre +/- 1%, so we all get 990ml of petrol for a litre. There is an E symbol for quantity which means the mean quantity, doesn't apply to petrol.

Back on topic, resistors as opposed to capacitors don't need any more material to be accurate so they probably better now.

[Julesomega]

@Mark: I started off congratulating you on the elegant and KISS design of your little calibrator project but I can't help thinking you're trying to put me off copying it now! As I said before, I like the idea of a unit whose accuracy can be assured without recourse to traceable calibration in a lab. You introduced me to the appealing concept of combining several components of one tolerance to achieve a higher tolerance. I shall be happy enough if the accuracy is slightly compromised when trying to calibrate Avominors as I want it primarily for DMMs and maybe oscilloscopes, minimum input resistance 1MΩ. I'd ideally like a combination of the features of yours and the Time Electronics 1030.

Taking a lengthy look for some reference ICs I found some packets of low-spec devices for 2.45, 2.50 and 5.0V before I found a bag of LM399AH 7V references. Although these are only the 5% type (the better selected versions are 2%) I reckon that I can connect 10 in series to get better than 1% overall. The datasheet is on datasheetarchive where you will see that these devices are ovenised for temperature stability which will be particularly useful for us working in the back room or garden shed - the warm-up time to settle within .05% is only 3 sec.

Looking at AN-184: References for A/D Converters shows these were intended for accurate data conversion, but the author shows how the device can be used to give a buffered output (for Avos?) with a temperature coefficient of 3 or even 1ppm/°C. I attach his suggested circuit diagram here.

While we are in isolation I had hoped there'd be plenty of time to start projects like this, but then there's SWMB Reasured Then Obeyed so it could be some time yet! Please allow me to send you a couple to play around with, I'm sure you will appreciate the spec's. There are plenty to be had on eBay, but the device markings always look very different to my NOS 1980s pieces, and we know what that usually means

[mhennessy]

Hi Julian,

That's a kind offer, but I've already got a couple of them here. They're nice chips. One thing to watch about putting them in series: the heater supply is connected via an intrinsic diode to the zener, so you have to pay attention to where the heater supplies sit relative to the zener - to be completely safe you'd basically need 10 floating supplies, though I'm sure that could be simplified somewhat.

Alternatively, why not put them in parallel? The statistical trick should work just as well.

Bob Pease writes a bit about this here:

https://www.electronicdesign.com/arc...y-stuff-anyhow

Having summed the output, a (good!) op-amp could be used to get the voltage up to something sensible like 10V, and then you could implement your potential divider idea (and optionally, perhaps a second op-amp could buffer the output?). The op-amps mentioned in your link are promising, but from a quick look they appear to be obsolete now. I'm sure there's something else that's suitable and available. For example, the LT1001 mentioned in the LM399 datasheet is available in DIP from RS for about £4 - I've not checked the datasheet yet. The diagram towards the end of the LM399 datasheet ("Portable Calibrator") is a good starting point for this. Note the way the reference zener is supplied with the bulk of the operating current from the output (via the 5k resistor) - a good trick to improve stability in light of variations of the supply voltage.

The only downside of this approach might be the initial calibration, but you'd be welcome to visit here after the lockdown and put it on my Keithley 2015 if you can't find anyone nearer. Once done, it should be pretty stable long-term. Certainly better than the LM4040.

Anyway, that was the vague plan for mine, but other things took over. But now you've got me thinking...

[Julesomega]

Thanks for pointing out about the heater supplies, I hadn't spotted that. As for the op-amp type, I'll have to sort through the op-amp box next

When SWMB has been placated I'll make a start on this, and try to arrange to take her to visit her old friends in the Cotswolds so that I can slip away to take you up on your kind offer.

[Oldcodger]

Originally Posted by mhennessy

Thanks, Andy

The secret of Veroboard is simple: 0.1" graph paper

I'm not sure how easy it is to find these days, but there are websites that can create it for you. I'd guess that a laser print would be better than inkjet, as you'll do a lot of rubbing out as part of the process - you'll see what I mean when you see this example: https://www.vintage-radio.net/forum/...1&postcount=52

If you haven't see it before, that thread is full of useful tips about Veroboard...

For smaller projects that don't warrant a PCB, I use one of the PCB programs. Express is my tool of the moment. First step, make a master. I use a 9 hole wide by 25 hole long x .1" matrix and draw my tracks.
I make them circa 0.1 wide with 0.04 between them, so spacing between the holes is circa 0.06.
Add in track holes and it's sorted.
that is the underside but I also make a tp view to help locating the wire links.

It might take a few tries, but you've always got the master and the last best guess. For track cuts, I use a circle ,track width in a different colour. For links I use another colour. Components, I measure and