LM399 based 10 V reference

[Andreas]

I have further doubts that the dielectric absorption set the lower limit above the reference.

From the pairs you can calculate the indidividual noise spectra.

Dielectric absorption (DA) is only a problem for measuring leakage current.
On AC-Signals a 0.1% error of the noise voltage due to DA should be no problem.

Leakage current itself contributes to the noise floor.
Of course it might be a good idea to let the input capacitor always under power to minimize leakage.

I doubt that one can calculate individual noise from pairs.
I'd rather average >9 references to get a lower noise voltage (sqrt(number)) and measure the D.U.T. against the average.

with best regards

Andreas

[rf-design]

I doubt that one can calculate individual noise from pairs.

Indeed you can Click :

www.intersil.com/data/an/an177.pdf

The calculation is not difficult!

The measurement is a pairwise sample DUT noise density over frequency. Because the assumption is that the samples are statistical independend the noise power at sa specific frequency is simply the addition of the noise power of the DUT pair.

If you have n=4 samples you can make n*(n-1)/2=6 pairwise measurements. I will call the reference sample spectral noise power at a specific frequency simply a,b,c,d.

What you measure with the pairs is the sum of the noise powers

a+b
a+c
a+d
b+c
b+d
c+d

You get the noise power of the first sample reference by the following expression:

a=(1/3)*(((a+b)+(a+c)+(a+d))-(1/2)*((b+c)+(b+d)+(c+d)))

a=(1/3)*((3*a+b+c+d)-(1/2)*(2*b+2*c+2*d))

a=(1/3)*((3*a+b+c+d)-(b+c+d))

a=(1/3)*(3*a)

a=a

For the general case n the expression is:

a=1/(n-1)*(sumwith(a)-1/(n-2)*sumnot(a))



The application note from LT does not give a calculation example. It is only stated that you have to use a much better reference as a DUT partner.

[Andreas]

What you measure with the pairs is the sum of the noise powers

Hello,

what do you want to measure?
wideband noise > 10 Hz or
1/f noise 0.1 .. 10 Hz or
drift (below 0.1 Hz).

For wideband noise you are right: the power (effective value) adds.
The voltage adds only geometrically. (square root of the squared sum).

For 1/f noise you usually can only measure the peak-peak voltage value.
And if you really have done such measurements you now that
from measurement to measurement you have a lot of variation which
makes it practically not possible to measure differences.

Except when one of the 2 references has much lower noise.
-> my suggestion to average at least 9 references as
    "low noise reference" to have less than 10% error for the DUT.

With best regards

Andreas

[rf-design]

What you measure with the pairs is the sum of the noise powers

Hello,

what do you want to measure?
wideband noise > 10 Hz or
1/f noise 0.1 .. 10 Hz or
drift (below 0.1 Hz).

For wideband noise you are right: the power (effective value) adds.
The voltage adds only geometrically. (square root of the squared sum).

For 1/f noise you usually can only measure the peak-peak voltage value.
And if you really have done such measurements you now that
from measurement to measurement you have a lot of variation which
makes it practically not possible to measure differences.

Except when one of the 2 references has much lower noise.
-> my suggestion to average at least 9 references as
    "low noise reference" to have less than 10% error for the DUT.

With best regards

Andreas

Andreas,

my thought about the reference noise problem was to avoid in general a requirement to have better reference to characterize or measure other references. I know that similar principles are also used for phase-noise measurements at levels where the best economical sources are less than 10dB away. As the amplifiers could be build with 1/f noise and drift much less than the expected reference noise and drift a differential measurement seems to be an easy solution.

Because for 10 references you have 45 pairs you could not measure all pairs at the same time the result will be dependend on the statistical nature of the drift and noise process. So it is assumed that the noise spectrum, which is hole picture of short drift and 1/f noise, is stationary. So otherwise the calculation of the last post is not valid.

The measurement sample could be every 100ms over 2^14=16384 samples. So a pair measurement need less than 30min. I know that it is difficult to interpret a noise spectrum extending down to 306uHz. Otherwise if the noise spectrum follow a 1/f rule the noise power integrated over a frequency range with a fixed factor of the upper limit to the lower limit will give an equal noise power. So there is no further gain in noise reduction to measure longer. If the spectrum follow (1/f)^k with k<1 than gain is small. Finally the 1/f or short time stabilty set for most systems a lower limit. And I know that it is difficult to differentiate that from an againg which typical goes only in one direction.

BR
Reiner

[blackdog]

Hi,

Can i also play? 

This is a portable reference i am designing, is uses a LM399.
I've done my best to keep the noise low and the current to the heater as low as possible.
Therefore, I use three 9V batteries, it gives me a lower heater current and longer live of the batteries.
Only 2 euros each, for alkaline here in the Netherlands.

The corner of the filter is about 1 a 1,5 Hz, this is low enough to filter out most of the noise.
If you are calibrating a digital multimeter, choose the longest integration time, 1 to 2 seconds.
There are also elements for the protection of the circuit on the output circuit.

I now own a calibrated Agilent 3458a, later i wil show a short term plot of the portable reference.

Shoot at it!

Kind regarts,
Blackdog

[MK]

Hi Blackdog, with a rai to rail output of a cmos amp and with a FET for Q1, it should self-start without the 180K R1 resistor and the output should  be slightly more stable too.

regards.

[wiss]

You should definitely remove R1 (me without an hard empirical proof).

[blackdog]

Hi,

I know that R1 should be removed, but i need to do some test.
Today, i received the last components, and from other references i had build using the same schematic setup
R1 was not necessary, we will see wats come out of it...
Thanks for the remarks.

A little test print, with the LM399, LM368 10V reference and a 2K Rhopoint 8E16 who is putting the current in de reference
drift within 2PPM with long cable's to a Tektronix DMM4050 Multimeter, not bad :-)

Kind regarts,
Blackdog

[Mickle T.]

LM399H with the external oven
- Noise (0.1-10 Hz) ~ 2.6 uV p-p;
- TempCo ~ 0.25 ppm/C;
- T(oven) = 52 C;
- Long-term stability is unknown.

[Andreas]

LM399H with the external oven
- TempCo ~ 0.25 ppm/C;

a) Is the tempco including transformation to 10V?
b) what is the tempco of the zener alone (unheated). (Did you select for tempco?)

With best regards

Andreas

[Mickle T.]

1. There is no transformation to 10 V.
2. Zener have a broken heater and have a huge TC ~ 70 ppm/C.

[ManateeMafia]

Mickle,

This is the first time I have seen that voltmeter in your posts. Is this another one of your neat projects?

[Mickle T.]

Yes it is. This is a simple homemade 8.5-digits DVM/voltage comparator.

[Andreas]

Hello Mickle,

I think it´s time for a teardown.
Isnt it?

Which ADC
which Reference
which processor.

With best regards

Andreas

[wiss]

Yup!!! 

[Vgkid]

I agree with the above statements.

[Mickle T.]

Which ADC
which Reference
which processor.

ADC - Mark-space type with convergence accelerating signal.
Reference - external 7-14 V.
Inguard CPU - Atmega8.
Outguard CPU - AMD Geode 300 MHz.
Some photos is here: http://bbs.38hot.net/forum.php?mod=viewthread&tid=102351&extra=page%3D1&page=3

[Andreas]

Hello Mickle,

nice project.
What do you mean with "mark-space" type?
I guess it is some kind of first order sigma delta modulator?

On the pictures the reading changes only by 1 count (0.1uV).
Which would be better than on a HP3458a.
But how much do the values really change?
Do you have a standard deviation figure for the readings?

With best regards

Andreas

[wiss]

Mickle, could you plot the DVM measuring it`s own ref for half an hour or so? (and post here)
Youtube video would be ok if no computer interface 

[babysitter]

My mother today emailed me that she considers the LM399 thermography movie by Andreas and branadic "art"

[Mickle T.]

Hello Andreas,
"Mark-space ADC" is a term proposed by Solartron (Pat. US3942172A, US4340883 e.t.c.). I have used the structure of the ADC, the average between Solartron DMMs and Valhalla 2720GS with some improvements. Noise is much more than 0.1 uV 
This project is suspended now. I will continue to deal with them, since I can afford to buy an LTZ and make the internal reference.

Mickle.

[wiss]

That looks very much like my 7075! Total noise about +- 2 uV at 14 V range, no averaging.

[MK]

Well done Mickle, that  is quite an acchievment. I went to the 38hot link you gave, unfortunately no images shown for me in the UK? I am impressed that you can read and write chinese as well.

[quarks]

MickleT, great work as always. Please keep sharing.
Many thanks
quarks

[macfly]

Hi Mickle,

I agree with the other knowledge-hungry guys: please show us your great work in detail.   

Best regards,

macfly