Poor Man's Reference

[zhtoor]

I have learned that the Weston cell was until not too many years ago considered an excellent reference.  I know there were other "not as good" reference cell constructions before the Weston one became dominant.  That also made me wonder if even taking it further down the "Poor Man's reference" line if there might not be some easy and environmentally suitable inexpensive electrochemical cells that could be constructed and yet which could be interesting for a hobby reference project even if a couple of orders of magnitude worse performing.

well, i have been looking for high purity cadmium rods, (mercury, sulfuric acid etc. is available in pakistan).

regards and thanks for your appreciation.

[Marco]

Adjusting a JFET to its zero temperature coefficient point explains why several older DMMs were able to use them as simple buffers.  These DMMs have the advantage of no charge injection unlike a modern chopping or autozero design.

The best of which used auto-calibration. Auto-calibration is just really slow chopping ...

[zhtoor]

i am actually a little surprised there hasn't been more said in the past about "make it yourself" standards / references since with all the components out there, even some low cost ones, it seems like one could find lots of "good combinations" that work very well despite being simply made.  Even if one had to select parts and tweak the circuit, that is ok for a personal experiment or hobby project.

i am surprised that you are surprised. if we look around us, are the women cooking what they used to? we used to cook delicious bread at
home, make all kinds of fantastic edibles at home including jams, breads cookies etc., we used to do a lot of tinkering at home, almost
every household had a workshop. but alas, we have now become a "consumer" society (both eastern and western included), we are told
to consume brand "x" over brand "y" (a la, ADR431 over LM399). our older generation was a lot more active than us, they used to work
hard, and we consume.

sorry for the rant, but i felt that you may understand.

regards.

[Marco]

Anyway it does sound like there is something to be said for the somewhat cheap discrete implementation for experimental purposes though under $10 gets you something generally comparable and guaranteed "off the shelf" from the integrated version though one other disadvantage of the particular IC version is that it isn't packaged hermetically that I know of. 

Dunno about XFET, but Intersil's FGA relies on an EEPROM type process with junction diodes, which will include a passivation layer. That's hermetic, more prone to stress than a nice roomy can though.

[Andreas]

It all depends on your requirements. No bandgap reference I know of can give you 10ppm/year stability in a hermetic package and the LF noise below 1ppm p-p.

Hello Alex,

the LTC6655 in LS8-package is very close to this.
And I bet that also there will be a good stability at least after a run in time.
Unfortunately I have no own data for long term for this device.

With best regards

Andreas

[zhtoor]

could'nt have said it better evb149.

so count me in, viva la resistance !

regards and may your tribe increase !

[Kleinstein]

If you want to build a DIY votlage reference, I would not worry so much about the TC. The really difficult part is long time stability. For the TC in a DIY reference circuit there often is a way to tune the TC close to zero (e.g. by adjusting the current to a zener type reference). This usually only works good for a small temperature range - but than a coarse temperature stabilization (e.g. a small oven) can do a lot. If you have time for a DIY circuit adjusting an operating point can be used, that is less practical in a monolithic chip.

For long time stability, a critical point might be the drift of resistors. To get a small effect there, a rather nonlinear I/U curve is needed, like a low differential resistance for a zener. It does not really help if you can build a circuit from cheap semiconductor, like of the shelf transistors, but need super stable resistors to make is stable. Even with the LTZ1000 circuit, it takes quite good resistors - and this is a circuit that is already really good in being not so sensitive to the resistors.

[David Hess]

Adjusting a JFET to its zero temperature coefficient point explains why several older DMMs were able to use them as simple buffers.  These DMMs have the advantage of no charge injection unlike a modern chopping or autozero design.

The best of which used auto-calibration. Auto-calibration is just really slow chopping ...

Just like drift is just really low frequency noise.

I keep these old JFET input multimeters around as a sanity check for when I suspect charge injection from one of my more modern multimeters is causing problems.

[zhtoor]

Even with the LTZ1000 circuit, it takes quite good resistors - and this is a circuit that is already really good in being not so sensitive to the resistors.

i agree with you completely.
what i gather from learned people like yourself, MisterDiodes and Mr. Edwin, the three cardinal pillars of references are:-

1. the *resistor* and their ratios, and their stability.
2. stability of the central stability element (zener etc.).
3. tight environmental constraints (like temperature, humidity, mechanical stress etc.).

regards.

[Kleinstein]

The circuit / central reference element determines how stable the resistors need to be:

With the JFET+resistor, JVR type reference, a drift in the resistor translates nearly 1:1 (maybe a factor of 0.5) toward the reference voltage.

The bandgap type circuits are also quite sensitive to resistor / current matching. The initially shown circuit the a Si transistor and a Ge diode might in this respect be actually be relatively good, as uses the same current with parts in series - however Ge transistors are even known to show significant aging, as diffusion in Ge is rather fast. So I won't expect a Ge diode to be super stable.

With a good zener or the LTZ1000 circuit it can be about a factor of 100 lower than the resistor drift.

[Alex Nikitin]

The circuit / central reference element determines how stable the resistors need to be:

With the JFET+resistor, JVR type reference, a drift in the resistor translates nearly 1:1 (maybe a factor of 0.5) toward the reference voltage.

That is not so. The resistor drift in the JVR circuit is attenuated considerably, depending on the JFET used. With the 2N4391, the attenuation factor should be at least 20, so a 20ppm shift in the resistor value will create only 1ppm shift in the output voltage. If the JFET and the resistor are thermally coupled, it should be even possible to compensate for the resistor TC in a certain range of temperatures by adjusting the JFET current.

Cheers

Alex

[Marco]

With the JFET+resistor, JVR type reference, a drift in the resistor translates nearly 1:1 (maybe a factor of 0.5) toward the reference voltage.

Only because he amplifies it to 10V, just the source voltage alone will attenuate the resistor drift by the inverse of transconductance.

[Vtile]

Can someone open a bit of the charge injection mentioned many times. How it is seen and in what magnitudes. Is it a transient kind of effect.

I think I will build a "poor mans reference" some day, but I haven't decided if I go with Conrads 10V opa design or BJT sourced zener like in that Voltage source/reference (1n8xx) Dave have. Well my best digital meter shows full 3,5 digits so not much on it but keeping up old 7th hand equipment would be easier with bunch of precision resistors and voltage ref and my 2nd old FET nA/uV meter need some love.
I have also been wondering how to amplify the ref voltage to 100V or 200V even with some long term and temp stability... Somekind of OPA controlled transistor driven voltage divider.

[zhtoor]

both conrad's design and dave's are based on temp. compensated zeners, 1N8xx are especially good at stability if operated right, if you can find them!.

regards.

[David Hess]

both conrad's design and dave's are based on temp. compensated zeners, 1N8xx are especially good at stability if operated right, if you can find them!.

They are very available but cost $10 or more.  For that price, you can buy an awful lot of performance in an easier to use integrated reference.

[David Hess]

Can someone open a bit of the charge injection mentioned many times. How it is seen and in what magnitudes. Is it a transient kind of effect.

The FET switches used for automatic zeroing and chopping inject charge at the input of the multimeter.  This can either disrupt circuit operation or if it is synchronous with noise coming from the circuit, the intermodulation will produce a DC offset or low frequency tone.  It shows up as inexplicable noise, drift, or DC offset.

I remember it showing up in precision circuits using the OP-07 and various chopper stabilized amplifiers with levels of 10s to 100s of microvolts of offset.  In those cases the easy solution was to add a 10uF aluminum electrolytic capacitor from the output of the operational amplifier to ground.  The large value combined with the finite output resistance of the operational amplifier does not oscillate because the frequency response falls off at such a low frequency but the capacitor also provides a low AC output impedance blocking the charge injection.  Something similar is often recommended now for driving the input of sampling and delta-sigma ADCs for the same reason.

[zhtoor]

update:

i came across an interesting paper discussing 1nv measurements against a JJA using a transfer standard :-

http://ieeexplore.ieee.org/abstract/document/109957/

the transfer standard is discussed in :-

http://iopscience.iop.org/article/10.1088/0022-3735/5/10/026/meta

both files attached.

it looks like alex nikitin is definitely on to something..... see attached circuit and simulate it using ltspice.

regards and comments required.

[Vtile]

Oh, I so like this metrology grade temperature chamber used. Also fits to title of this thread.

A   dozen   n-channel junction FETS, type 2N5457, were checked for temperature coefficient of current by observing the change produced by holding  them between finger and thumb, which raises  their temperature by about 10°C. 

[Marco]

What's also interesting is that a temperature chamber with discrete components was able to get that kind of stability.

Guess there isn't a huge need to get everything monolithic (as long as you don't mind it being a little larger).

[zhtoor]

has anybody tried a supercap to filter out noise across a zener diode like 1N829A?
or maybe think of a rechargeable battery as a *really* supercap?
the noise perfomance could be interesting.

regards.

[zhtoor]

followup:-

please see attached diagram. the batteries are connected in a way that they are trickle charged to
fully charged state of (2.3*3 = 6.9V for lead acid cells), ie; the nominal reference voltage of
LM329/LM399.

here the intent is to use the battery as a *really* supercap, and i think we all know the noise
performance of lead acid batteries.

regards and comments required.

[Kleinstein]

The LM329 / LM399 has a really low impedance. So even with a large "capacitor" there will not be that much filtering. Also, if for some reason (e.g. temperature rise or mechanical disturbance) the voltage in the cells increases, there can be quite some extra current to the reference - possibly too much. So it would need some kind of resistance / current limit.

Lead Acid cells have a significant self discharge. So it takes a noticeable current to keep a constant voltage. There also is a significant temperature dependence of the voltage and leakage current. So for a stable output it would need a really stable temperature.

[zhtoor]

The LM329 / LM399 has a really low impedance. So even with a large "capacitor" there will not be that much filtering. Also, if for some reason (e.g. temperature rise or mechanical disturbance) the voltage in the cells increases, there can be quite some extra current to the reference - possibly too much. So it would need some kind of resistance / current limit.

Lead Acid cells have a significant self discharge. So it takes a noticeable current to keep a constant voltage. There also is a significant temperature dependence of the voltage and leakage current. So for a stable output it would need a really stable temperature.

agreed, but the equivalent "capacitance" of a battery would be *huge*, for example, 1AH battery
would be approx. Q=it = 1*3600s = CV => C = 521F (!) at 6.9V (please check my calcs, or am i out of my mind?)
so a 1 ohm dynamic impedance translates into RC = 521 seconds. a lower cutoff freq 0.0003Hz.

also supercaps can be used, especially maxwell ones, which have been tested for noise performance.

yes, if we have carefully controlled conditions, the noise floor could be difficult to detect.

maybe doing this to ltz1000 zener or 1N829A would be better?

regards.

[zhtoor]

hello all,

help required.

i am trying to simulate the reference part of LM399 (LM329) using ltspice, but running into trouble,
maybe i am using the wrong transistors, (i am using THAT3XX models), or can't adjust the resistors marked for factory adjustment,
or maybe using 6.2V zener instead of 6.3V is a problem. (maybe use 2DW232 model here?, could be interesting)

the voltages and currents work out approx. ok but i should see a hump for a temperature range 0-100degC around 90deg, or any
hump at all, which i cant see.

i have also tried to simulate for a hump using the tesla supplied schematic diagram, (i do see a hump in the bypass transistor collector current
which is only in the tesla schematic as T18).

attached is the schematic / ltspice file.

comments required from learned members of this forum.

[Kleinstein]

The details of the performance depends on the details of the models for both the transistors and the zener. Especially the buried zener might be a little different from the simple models.

Anyway The shunt regulator type circuit in the LM399 is more a convenient replacement for a normal zener, but it also has drawbacks from thermal effects. So it likely is not a practical solution for a circuit build from separate parts. The LTZ1000 circuit, with an external OP is more realistic to build from separate parts. Here it only needs the zener and one transistor to be thermally coupled and in addition the power dissipation of that part is essentially constant. OPs are cheap - so there is no need to do all the gain / loop part by hand.