Characterizing the entire reference circuit is exactly what I thought I was doing (as it's what matters in the end), but I hadn't really thought about that in detail.
or if the charts I produced are just a curiosity.
Hello,
I believe that you can improve the T.C. behaviour around the heater setpoint significantly with the "zero T.C." zener current.
Of cause higher currents may create a worse long term stability.
On the other side you should get better short therm (less noise with good thermal shielding) and better temperature stability. (lower overall T.C.). Noise should decrease by a factor of 2 if using a factor 4 higher current similar to a array of 4 LM399 with 1 mA each.
A good test would be probably the "tilting experiment" in comparison of 1 mA and "zero T.C." current.
https://www.eevblog.com/forum/projects/lm399-based-10-v-reference/msg360779/#msg360779This test uses the imperfection of the LM399 that the heater is not symmetrical to the zener and the temperature sensing element.
With best regards
Andreas
I think the shift of the top of the TCV curve in the last graph is due to self-heating of the LM399. If you turn on the current and take a voltage reading then turn off the current, and allow some thermal settling time between measurements-- I think these shifts will go away.
I started the 7.5mA and 9.5mA tests by letting the references heat the oven to get a sense for the self-heating, but now that you suggest it, I'm pretty sure I didn't compensate for it nearly enough. Your measuring method sounds like a much better way to eliminate the error. It would be harder to do over the entire range, but so long as the parabola is still true, I would only need to spot check a few points to get a rough sense for the shift. I will endeavor to do that this weekend.
Unfortunately, zener diodes, buried or otherwise, are just plain noisy, very careful processing at the wafer level can go a long way to minimizing wideband noise and even paring down shot noise a bit but they are noisy, just like the gas regulator tubes back in the good ole days. These references also have additional circuitry in parallel with the zeners which add their own bit of noise to the output too. While filtering can decrease overall wideband noise, it has its own set of compromises too. In the case of a voltage reference that is running for relatively long periods of time, a longer time constant in the filter is not so much of an issue.
If you are using the reference as a short term on source, then the time constant becomes much more important. The filters would require PWW (not necessarily tight tolerance) resistors and polypropylene capacitors but even these capacitors will impart small voltage errors if the resistances are of high values. Active filtering with appropriate op amps could possibly provide a good compromise. Multiple summed voltage references (again with PWW resistors) will generally provide the least amount of total noise.
Since the internal circuitry of the LM399(A) shunts extra current around the Zener, there is no way to decrease the awful noise spec by increasing current-- so you may as well run it at the 1.0mA current to get the best long term stability.
That´s a good point, I did not recognize this.
I made 2 measurements with LM399#19 out of my ageing PCB.
One with a 6k8 pull up to 14.1V stabilized from a battery. (around 1 mA)
the other with 6K8 + 1K in parallel (around 8 mA)
All packaged into a cookie tin can and amplified *10000 with a 0.1 ... 10 Hz 4th order bandpass.
So the scaling on the oscilloscope is 1sec/div on time axis and 0.5uV/div in y-axis.
So there is no visible difference between 1 mA and 8 mA.
On the other side I do not understand from where do you get the figure of 10uVpp of noise.
Measured with which bandwidth?
With correct setup of 0.1 ... 10Hz bandwidth I usually measure values from 2.5uVpp to 4uVpp as average with std deviation of 0.2 .. 0.4. Only one LM399 is around 7uVpp.
With best regards
Andreas
Hello,
if I look at the current LM329 datasheet it looks more like 3-4uVpp (Figure 15)
http://www.ti.com/lit/ds/symlink/lm329.pdfHow could you live without noise measurements up to now?
How will you detect the "stinkers" of your references?
I am not aware of a TI appnote for measuring noise of references.
(Although perhaps there is one for low noise voltage regulators using a (lower noise) voltage reference to compensate DC)
Usually only cirquits of OP-Amps are shown for this purpose at TI.
But this is more easy: no DC-Offset and the DUT is part of of the amplifier.
I was inspired of LT AN124
http://cds.linear.com/docs/en/application-note/an124f.pdfBut since I only wanted to measure down to 1uVpp I used standard precision OP-Amp cirquits like LT1037 and LT1012 for the 2 amplification and filter stages.
The result is a noise floor below 200nVpp.
With best regards
Andreas
Yes, that tantalum capacitor was a real special, something on the order of $1,200 each but at the time, it was the best solution to the circuit requirements. I rather doubt very many of those were built.
Every now and then, on eBay, some Chinese manufacturer will put up a large capacitance polypropylene capacitor for a good price. Not too long ago, I remember seeing a 400uF poly for $14 and change. They don't show up all the time but they have specs better than the tantalum capacitor Jim was stuck with. They are not too small but they are not the elephant in the room any more either.
You can also use standard (85 deg) electrolytic capacitors (35V)
if you select for lowest (< 20nA) leakage current @10V after 48 hours forming time.
But you will have to bias the capacitor all the time even when not using the cirquit.
Otherwise you will have excess noise for the first 24-48 hours when measuring.
With best regards
Andreas
I used 3 each tantalum capacitors in parallel, they were 68uF/16V, they couple the input from the device under test to a 25 ohm 0.1% 25 ppm resistor to the inverting input.
Cutoff frequency is 31.2 Hz, but not 0.1 Hz.
I've dublicated the
1/f noise amplifier by Andreas.Attached is a picture off my board. For its input cap (1.000µF||2.200µF) I've measured electrolytic caps with same value (1.000µF and 2.200µF) f different vendors and temperature type to find the one with lowest leackage current. Best cap I found were 85°C types by Yageo. The combination of 1.000µF/25 || 2.200µF/35V for the low noise amp showed a leackage of <5nA after 24h on a 9V block. 105°C types were decades worse.
The amp is somewhat hand sensitve through its aluminium case and needs to be placed with the DUT in a shielded cookie box.
I've also build the
10Hz - 100kHz amp that is shown in AN83 in different versions (60dB for oscilloscope applicaton, 80dB version combined with a STM32 board and its 12bit ADC).
I've used Oscon Caps (330µF, 6.3V) as input cap, but found that SMD tantal caps (330µF, 10V) do work as well. For higher voltages I've got 3 pretty expensive
FFB54D0117KJC 110µF, 75Vdc polyester caps that can be paralleled.
Both amps are pretty neat.
I used 3 each tantalum capacitors in parallel, they were 68uF/16V, they couple the input from the device under test to a 25 ohm 0.1% 25 ppm resistor to the inverting input.
Cutoff frequency is 31.2 Hz, but not 0.1 Hz.
Hello,
That also explains the "low noise" of the Fluke 515A which is specced with up to 0.01% rms of range.
I would not connect such a low impedant capacitive source to a sensitive unbuffered LTZ1000 reference.
The last time I did this I got a hysteresis shift on the output voltage.
With a LM399 (1 Ohms differential impedance) you will get 4% less amplification since the 25 Ohms and 1 Ohms are in series.
With best regards
Andreas
With best regards
Andreas
Attached is a picture off my board.
Hello branadic,
I would remove the socket of the first input stage to further reduce the noise.
And its always a good idea to do some thermal isolation (e.g. a piece of cotton or cloth)
with best regards
Andreas
I would remove the socket of the first input stage to further reduce the noise.
Thanks, could be an option, I will keep that in mind and try that maybe some day.
And its always a good idea to do some thermal isolation (e.g. a piece of cotton or cloth)
The board is assembled in a standard aluminium case together with the batteries and some cotton, but this is not shown here.
I think the shift of the top of the TCV curve in the last graph is due to self-heating of the LM399. If you turn on the current and take a voltage reading then turn off the current, and allow some thermal settling time between measurements-- I think these shifts will go away.
Confirmed. I took spot measurements to locate the peak of the parabola, and it appears to be right around where it was at 1mA. I'm going to remove the errant graphs lest they confuse someone who doesn't read the whole thread
Simulation: before and after
Hello,
although the discussion gets now rather off topic:
The LT1028 has very good voltage noise specs but rather bad current noise.
Above 200-300 Ohms input resistance the current noise dominates on a LT1028.
So just flipping the resistors will not help in this case for a good flicker noise amplifier.
With best regards
Andreas
These two notes from Linear Tech should help with op amp selection, one is older, the other an updated version.
although the discussion gets now rather off topic:
With best regards
Andreas
EEVblog Electronics Community Forum
*
A Free & Open Forum For Electronics Enthusiasts & Professionals
I have removed my postings from this thread.
although the discussion gets now rather off topic:
I have removed my postings from this thread.
It´s your decision.
I mentioned this to suggest to you to spin off a thread with a appropriate title.
In my opinion nobody will seek for a flicker noise amplifier within a LM399 thread.
And also I think that a flicker noise amplifier is more complex than it looks on the first glance.
So it would be worth a own thread.
with best regards
Andreas
I intend to build 2 or 3 lm399 10 V references, what are the suggestions for the number of ppm span for a trimpot in the chain. I expect to pre-age them then select of pad the volt setting resistors and to have some low value resistors of various values in the chain with 1-2 ohms with a 100 ohm pot in parallel for setting to final value periodically. If i make the tim range too small I will have to cut or solder too often, if I make the trim range too large hysteresis in setting the value will limit the stability? what are the thoughts about a best span to choose?
Hello,
after proper pre-aging of the LM399 the further drift is in the range of about 1-2ppm/year for the LM399.
So the trimming range is mainly dependant on the drift of your voltage divider from 7 to 10 V.
I would use a trimming scheme similar to the "Standard Cell Replacement" cirquit on page 5 of the datasheet.
http://www.linear.com/docs/3317With best regards
Andreas
If you need more stability in the trim range, check out the trimming methods discussed in an42.
I decided to use the fluke 732A timming technique in the end.