[volt-nuts] Low Frequency noise on LTZ1000A and tempco of unheated LTZ

[Andreas Jahn]

Hello,

I have built my first LTZ1000A reference but there are still some issues 
which affect stability.

When measuring the 0,1 Hz ... 10Hz noise with a battery powered 4th order 
bandpass amplifier (10000x) and a ordinary analog oscilloscope:
In principle I can see the 1,2uVpp noise stated in the datasheet 
as 1,5 to 2uVpp randomly distributed noise on the oscilloscope.
(the amplifier has 0,55uVpp according to datasheet).

But there are large excursions of typical 4-6uV and up to about 15uV
with a risetime of 1-3 seconds as low frequency noise overlaid to 
the random noise.
I have already done better thermal shielding but the low frequency
noise is still there.

On the other side I see on the oscilloscope that the cirquit from the
LTZ1000 datasheet is very sensitive to parasitic capacities.
When nearing with the hand to the metal case with the LTZ cirquit
I can change the offset voltage of the noise measurement by the
distance of my hand.

When connecting a switchmode power supply to charge the batteries
of the LTZ reference I could see a voltage change of around -8uV
(towards lower voltages) measured with a 24bit ADC.
I could also see that the LTZ heater voltage changed by 2mV when 
connecting the charger. Since I did not expect this I made further
investigations.

I switched the heater off and got a voltage change of about -20mV!!!
My nominal setpoint (12K5 + 1K) is about 50.5 degrees celsius.
I added a 22K in parallel to the 1K resistor to change the setpoint
temperature to 61.3 degrees. So I got a change of +3.7 mV
which calculates to a tempco of +48ppm/K or 345uV/K for the unheated zener.

First I thougth that this tempco is due to a defective part. but
when calculating from the datasheet that a 100ppm resistor divider change
will give a 1ppm output voltage shift I would calculate as follows:
Nominal setpoint around 514 mV. Change by 100ppm calculates to 51uV.
51uV / 2mV/K gives a 0.026K change for 1ppm output voltage shift.
Thus from the datasheet I can calculate around 39ppm/K for the unheated zener.

On the other side the 2mV heater voltage shift that I observed give
around 30uW (with around 15mA heater current). With the 400K/W
of the LTZ1000A this gives a temperature shift of 0.012K
and a voltage shift due to temperature of around 4.1uV.

So 4uV of the 8uV of the EMC of the switchmode supply are thermal
effect by disturbing the temperature measurement transistor of the LTZ.
I have now put  2*100nF capacitors directly to the 2 base pins of the LTZ
against ground. So now I have no heater voltage shift and only about -4uV
output voltage shift when connecting the switchmode powersupply.
Unfortunately my output voltage changed by +30uV after connecting the capacitors.
Later I have seen that in the Datron 4910 reference cirquit (a reverse engeenered
schematic that someone has posted) the base of
the temperature transistor is filtered with 2*100nF.

But still I have the large low frequency fluctuations.

@Frank: possibly part of the voltage changes from your tilting experiment might be
due to capacitive changes of the I/O-Lines or power supply lines of the LTZ.

Questions:
- can anyone confirm the large tempco (40-50ppm/K) of the unheated LTZ1000 zener.

- has anyone experience how the sensitivity to EMC can be further reduced
  (target is <1uV shift with my switchmode charger).

- does anyone have the original Datron 4910 reference cirquit or repair handbook
  and can post the cirquit. (perhaps the answer is there).

- What is the best way to improve the low frequency noise.
  Is it really necessary to use a slotted pcb to isolate the LTZ or is it
  sufficient to use a isolating cap as it is used on the HP3458A reference board.

- should the legs of the LTZ be as short as possible or is it better to leave them 
  longer and shield them by some foam.

- to get a better behaviour of the low frequency noise my idea would be to use the
  datasheet cirquit on page 6 of the datasheet with the 200Ohm trim pot.
  http://cds.linear.com/docs/Datasheet/1000afc.pdf
  when having near 0ppm with the unheated zener the heated version should have
  lesser low frequency noise.
  Has anyone seen any reference cirquit in this way?

- or where is the "sweet spot" on the LTZ where the -VBE and the +Zener tempco
  cancel really out?
  I have already tried to change the 70K from the ref amp into 50K but this is the
  wrong direction (from 48 to 52 ppm/K). The 120 Ohms resistor seems to have
  nearly no influence on tempco when changing to 107 Ohms by adding 
  a 1K resistor in parallel.

With best regards

Andreas