[volt-nuts] Traveling Standards - Measuring Protocol

[WarrenS]

Andreas

Nice data
The name 'noise' of a high precision reference can mean many things.
The band width limited noise between 0.1Hz and 10 Hz, what is sometimes 
measured, and what I think is displayed on your scope picture,
is not the 'Problem noise' for sub PPM references.
The problem noise is the much lower frequency, random happening voltage 
steps, that can occur from several times a hour to several times a minute, 
and each last for a random length of time, from seconds to minutes to 
forever.
That type of noise can not easily be filtered out with a simple long time 
constant RC or averaging.
I'd guess that "Pop Corn noise" is what is happening in your LM399 (#1) that 
is making it "Less Stable".
To measure the presence of that kind of noise, an hour plus long graph 
showing the high resolution voltage, using about a 10 sec filter, works 
good.
The best way to reduce that kind of noise is first by selection and them by 
paralleling several low noise references so that the output is the Average 
of all the devices.

ws

***************
***************
[volt-nuts] Traveling Standards - Measuring Protocol
Andreas Jahn Andreas_-_Jahn at t-online.de

Hello Warren,

I'm sorry that it got not very clear with the noise.
Most of the noise shown in my diagram is not from the LM399 but from my ADC.

When measuring the LM399 #2 alone I have around 1ppm of low frequency noise
( around 6,5uVpp)
Attached a typical measurement with the oscilloscope and a 4th order
bandpass-amplifier.
1 ppm (pp) noise is not too bad compared to the 0.6ppm of typical buried
zener references like MAX6350.

My other LM399 (#1) has only around 4 uVpp noise.
But this reference is far less stable over time than LM399 #2.
(although I have heard that with zeners the stability can
be derrived from change of noise before and after pre-ageing).

Of course noise is dependant on the zener current. So with the
1mA of the LM399 you will never reach the low noise values
of a 7.5mA 1N82x or the 5mA LTZ1000.

With best regards

Andreas

***************

----- Original Message ----- 
From: "WarrenS" <warrensjmail-one at yahoo.com>
To: <volt-nuts at febo.com>
Sent: Thursday, September 08, 2011 4:21 PM
Subject: Re: [volt-nuts] Traveling Standards - Measuring Protocol

>
> A good thing to include in the "Measuring Protocol"
> The type of filtering used  and a 24 hr plot would be a great thing to
> show.
>
> In the LM399 data reported below, Its noise starts at 10-30 uv PP  (3 PPM)
> Once per minute averaged readings gave 2 uv PP (0.3 PPM)
> 21 min moving average data is below 0.5 uv PP  (under 0.1 PP)
> The PP variation over a couple hours time where the temperature is not
> changing is about 0.5 PPM
>
> What I've seen is that the LM399 has good TC over a wide temperature but
> they
> can be pretty noisy PP short and medium term when measured at the sub PPM
> level.
>
> The total PP noise on the plot I posted of the selected 1N825#2, running
> at its 'Zero-TC' current
> shows about 0.1 PPM variation over the couple hours where the temperature
> is not changing.
> The plot is once per second samples taken from with a 15 sec analog RC
> filter time constant.
> http://www.febo.com/pipermail/volt-nuts/attachments/20110904/15d2a3ab/attachment-0001.jpg
>
> All 1N825 are not the same. Starting with a selected low noise one,
> it must then be aged and run in at 50 ma for a long time before it is
> fully stabilize.
>
> ws
>
> ***********************
>> From: "Andreas Jahn" <Andreas_-_Jahn at t-online.de>
>> To: "Discussion of precise voltage measurement" <volt-nuts at febo.com>
>>
>>Hello,
>>
>>the red curve is one measurement with an integration time of 1 minute.
>>(about 350 measurement values of LTC2400 ADC with 10-30uVpp
>>noise averaged giving around 2 uVpp for 1 minute integration time).
>>
>>the light blue trace is the sliding average over 21 minutes of
>>the red curve.10 minutes before, current minute and 10 minutes after.
>>(so noise is reduced down to below 0.5uVpp).
>>
>>Total measuremen time (x-Axis) is around 940 minutes.
>>
>>with best regards
>>Andreas
>