After my conversation with Bob Dobkin, my take-away was that there is nothing wrong with building a voltage reference around some paralleled LM399A's [or even just one of them really]. The more you parallel, the lower the DC-10Hz noise will be. Because they run so hot, hysteresis is not much of an issue, and after a long burn-in period they are only going to drift 1-2ppm/year [if left on 24/7], which is really not that bad. Additionally, there is an opportunity to build a voltage reference that has no batteries [or at least the batteries don't need to hold the reference circuit up for more than a few hours]-- Bob said that these devices don't drift if they are off. So, you could build a voltage reference that you leave off most of the time, and then only turn it on an hour or so before you are going to use it [or calibrate it]. In this use case, the voltage reference could hold it's calibration within 1ppm of the SI volt for many years, which is important for hobbyists, because calibrations can be rather expensive [especially if they are using a JJA].
My burn-in recommendation for the LM399(A) is fairly radical-- it would involve building a very well insulated oven that was controlled at 150C [+/- 5C]. The LM399(A)'s are placed in this oven for ONE YEAR [and now you can see the need for excellent insulation!] This will provide "artificial aging" that approximates about 67 years of natural aging, and could result in references that drift less than 1ppm/year, even if you left them on all the time-- and much less if you only turn them on when you need them. To "soften" the die-attach [which can cause sudden jumps of 1ppm or so-- like a "pop"], the references are placed in a live circuit [no need for accurate resistors in this circuit], and then that is placed in a freezer, and the power is cycled on/off [one minute on 1 minute off] for 90-days. This way, if there are any bubbles or micro-cracks in the die bonding material, these will be "worked out" over this period. For obvious reasons, it is more economical to use this burn-in procedure on a hundred or more references at a time-- and because it takes so long, one has to have great patience.
Bob Dobkin said that you should parallel at least 6 of the LM399(A)'s, but I think 4 of them would be sufficient, and this is still less cost than an LTZ1000(A) based reference. You get a sqrt(N) reduction of noise, so this would economically reduce the noise by a factor of 2. The next level would be 9 devices for a noise reduction factor of 3, and then 16 of them for a noise reduction factor of 4. 9 devices would exceed the cost of an LTZ1000(A) reference, so 4 LM399(A)'s is about the economic limit for this technique.
Bob also said that the current limiting resistor for the LM399(A)'s Zener should be tied to the stable 10V output. This causes a start-up problem, but in the two threads I think this issue has been solved in different ways-- and all of them should work.
In the LTZ1000 thread, you can see one of my early designs of a PWM circuit for the 7V-to-10V boost circuit. Since then, I have refined the technique and it is very much simplified [using only one 32-bit PWM stage, and one filter]. No critical resistor ratios or absolute resistor values are needed, and thus there will be no resistor-related drift [at least in the boost circuit]. I have some more work to do on the digital side of things, and of course I need to build and test a statistically large enough population of these references-- but once I have done that successfully I can post my findings. Since the boost circuit could apply to either an LM399(A) based or an LTZ1000(A) based reference, I will probably start a thread just for that and link to the two voltage reference threads.
Hello,
Just another updated ageing chart of the CH6 + CH7 LM399 devices.
If you remember this was the "slot or not" PCB of branadic.
The drift is now scaled in ppm.
...
With best regards
Andreas
I went through the whole thread again, but could not find any description, how your monitoring system works, and on which volt reference it is based.
Therefore, I could not draw any conclusion, which drift you really display here, i.e. really the drift of the DUT, or the drift of the monitoring system.
Would you mind explaining your practical setup, and how you are able to determine the absolute drift of the LM399s?
Thanks
Frank
Hello Frank, Ken,
...
ADC#13 is a well aged LTC2400 based device with a AD586LQ voltage reference and a temperature sensor.
T.C. is compensated by a 3rd order correction curve.
..
ADC#13 stability can be found by the following charts of my "daily" automated measurements.
These attached charts are done with offset compensation.
Offset is measured once at beginning of the measurement and subtracted from the following readings.
If I set day 0 at the beginning of LM399 CH6 + CH7 measurements you can see that during same time
ADC13 does not drift more than about +/- 1 ppm against two LTZ1000A references (blue and green)
which is also negligible against the 10ppm of the LM399 CH7 reference.
For the LTZ1000A references I try to get reliable calibrations.
But the only thing that I can say from comparisons to other instruments like Keithley 2000 with calibration history
or Fluke 5520A is that the drift of the LTZ devices is up to 2 ppm/year against those devices.
But I cannot tell wether the calibrator drifts or the LTZ1000A.
I will still need some years to make a final decision.
With best regards
Andreas
Hello Andreas,
although the AD586 is quite mediocre concerning TC and ageing, your setup with TC compensation and comparison (frequent calibration?) against your LTZ1000As gives good stability / uncertainty for your ADC#13, I think.
.....
What do you think about that?
Frank
I am also interested how is switching done between different channels?
What is needed, is a traveling LTZ standard, and a group of volt-nuts willing to do the comparison.
What do you think about that?
Frank
I am just starting to read this "LM399 based 10V Reference" Subject,,,, am on pg 7,, But am wondering what spice model anyone might be using for the zener.. Although I use Proteus,,, I don't see anything in LTSpice on the LM399.... You all have done such a good job describing the device's in's and outs... Never seen so much activity for a 4 terminal device... Any response on this will be most appreciated..
Wonderful then,,,,, Any thoughts on using a the "Portable Calibrator" circuit WITH " Walt Jung's, Analog Devices, Build An Ultra-Low-Noise Voltage Reference... Electronic Design 6/24/93".... Of course substituting both op-amps with the LTC2057... Like everyone in the world, I'm looking for low drift,, low noise... all in a 3 terminal package...aaahhhhhhhh. Any thoughts will be appreciated, thanks...
I did not understand how the flicker or 1/f noise of the reference is measured. I could understand that a chopper amp could reduce the impact of the needed voltage amplification. But the amp could only amplify a voltage difference of two potentials or one potential to a ground reference point. But the reference is the DUT itself have an offset of 7V against the chopper amplifier. So there should be second reference which significant lower or well characterized flicker noise to counter the DUT reference voltage. The difference should also be smaller than the requested gain of the measurement.
So what kind of offset source is used?
I did not understand how the flicker or 1/f noise of the reference is measured. I could understand that a chopper amp could reduce the impact of the needed voltage amplification. But the amp could only amplify a voltage difference of two potentials or one potential to a ground reference point. But the reference is the DUT itself have an offset of 7V against the chopper amplifier. So there should be second reference which significant lower or well characterized flicker noise to counter the DUT reference voltage. The difference should also be smaller than the requested gain of the measurement.
So what kind of offset source is used?
I use a large (3200 uF) electrolytic capacitor selected for low leakage current.
Noise floor is <0.2 uV together with a 1K input impedance and a LT1037 OP-Amp
in the first amplifier stage.
See also AN124 of Linear Technology.
With best regards
Andreas