Low Frequency noise on LTZ1000A and tempco of unheated LTZ
Andreas,
I can confirm, that the unstabilized LTZ1000 has a TC of around 50ppm/K.
I did not try to use the 200Ohm resistor for tempco trimming, as this
has other disadvantages.
I think, that the tempco depends on the absolute value of the Zener
diode, and therefore a higher or lower total ref voltage may give lower TC.
But this would be too expensive to selct...
I also found a sensitivity / unstability due to grounding problems,
especially the Temp.-stabilization circuit seems to be volatile.
When I made a 24h measurement, I have seen many glitches, jumps of about
-5ppm which last for 10-30s, and then returning to the original value.
Perhaps those glitches are caused by the MUX switch of the DVM (34401A
or 3458A).
If I ground the metal case of the unit, I see a permanent change of the
output in the same order of magnitude.
I will test your hint implementing the 200nF cap, and will see, if this
described sensitivity will be reduced.
Frank
Andreas,
Hello Frank,
I can confirm, that the unstabilized LTZ1000 has a TC of around 50ppm/K.
Ok on the one side i'm relieved to hear that. On the other it makes me
hardly to believe that a tempco of 0.05ppm/K can be reached just by heating
the reference. In this case the temperature regulator should have a
stability of 0.001K???
I did not try to use the 200Ohm resistor for tempco trimming, as this has
other disadvantages.
I think, that the tempco depends on the absolute value of the Zener diode,
and therefore a higher or lower total ref voltage may give lower TC.
But this would be too expensive to selct...
Yes at least the total output voltage will increase by about 800mV, since
the
Zener voltage will flow through this resistor too. On the other side the
tempco
of the external resistor and the heated reference have to be adjusted too.
But anyway, if I have the room in my next layout I will at least put one
resistor as option.
I also found a sensitivity / unstability due to grounding problems,
especially the Temp.-stabilization circuit seems to be volatile.
Yes thats exactly what I'm observing here: Pin 6 of the LTZ1000(A) is the
most
sensitive pin in the whole cirquit. Any induced voltage will give a large
drift
over a relative large time. Since the the error voltage is stored as
temperature change
within the LTZ1000(A).
When I made a 24h measurement, I have seen many glitches, jumps of
about -5ppm which last for 10-30s, and then returning to the original
value.
On my noise measurements I had around up to +/- 15uV which give around
4-5ppm peak-peak
But I did not have a 24h measurement.Only relative short times looking on
the oscilloscope.
Perhaps those glitches are caused by the MUX switch of the DVM (34401A or
3458A).
In my case I do not believe that it has something to do with the ADC. I have
seen some pattern
with repeating rate all 0,8 seconds or 2,8 seconds then switching off for a
time and coming back again.
So for my side it looks like a pulse package heated device like a laser
printer or something like that
causing trouble on the mains line. Perhaps comming in to my cirquit by the
oscilloscope.
If I ground the metal case of the unit, I see a permanent change of the
output in the same order of magnitude.
Thats one reason why I have a battery on my LTZ1000. Now in battery mode
after all cirquit changes
a single earth ground in the cirquit influences the output voltage below my
noise floor of 1uV.
On the other side: when using a switchmode power supply to charge the
battery and using earth ground
on the negative reference output the voltage shifts up to 2mV !!!
I will test your hint implementing the 200nF cap, and will see, if this
described sensitivity will be reduced.
Perhaps I did not write clearly enough what I have done: (I think I should
make some photos).
I have one 100nF between Pin 6 and Pin 7 of the LTZ1000A (necessary).
The other 100nF is between Pin 4 and Pin 7 in my cirquit.
But be carefuly: this capacitor may lead to heavy oscillations in zener
current
without further changes in the cirquitry. I have additionally a resistor of
10K
between Pin 4 of the LTZ and Pin 2 of the LT1013. And a 100nF between
Pin 1 and Pin 2 of the LT1013. Both parts are not in the original cirquit of
the datasheet. Further against your cirquit I have a FET in series with a
220R
resistor instead of your bipolar transistor. So the maximum current through
the zener will be around 14 mA in case of oscillations. With the bipolar
output transistor the zener will be probably destroyed.
The 10K and the 100nF over Pin 1+2 of the LTZ I have needed agaist
large capacitive loads on the output.
In my case I have a 100nF between Ref + and Ref - within the cirquit.
So when in doubt do not use the capacitor between Pin 4 and 7 in your case.
Im not shure if this capacitor will really help.
Both capacitors together reduced the voltage shift of capacitive influence
on the cirquit from 8-10uV to about 4 uV (a factor of 2 .. 2.5).
The heater output voltage on Pin 1 of the LTZ does no longer change
when connecting the switchmode charger. The low frequency
voltage fluctuations in the noise measurement were still there.
In the meantime I did further changes to the cirquitry.
(It's now more like a capacitor grave yard ;-)
I measured the output voltage with a Oscilloscope on the heater (pin 1 of
LTZ) and
recognized a relative high frequent noise with up to 15mVpp in AC coupling.
Since I have learned that 2mV Heater voltage change will result in 4.1uV
change
in output voltage I decided to calm down the heater regulator.
When adapting the LT1013 pins with a steel needle I found out that Pins 1,2
and 6
are sensitive.
So the resulting measures are:
A 10nF capacitor between Pin 6 + 7 of the LT1013 (necessary).
After this the heater voltage noise is below 2mV.
For the sensitive pins I added further 100nF capacitors:
One from Pin 5 on Pin 6 on the LT1013.
(I had preferred connecting between Pin 6 and Pin 4 of the LT1013 but this
leads
to heavy oscillations on the heater output voltage).
The other from Pin 2 to Pin 3 on the LT1013. But this change will probably
only
work with all the other changes aroud the current regulator with FET.
So I would not recommend this capacitor in your case.
With all these further changes it seems that the low frequency noise
is reduced from +/- 15uV to around +/- 3uV.
Most of it will be due to the 10nF capacitor.
I had still some 1-2uV relative fast spikes on the output voltage which
never could be from the reference cirquit.
On the other side I recogized that the 0.1Hz .. 10Hz amplifier had
only a high-pass filter on the input coupling directly on the OP-Amp
pin. With a additional 4K7 + 1uF Low pass filter (33 Hz) in front of
the first high pass these fast spikes disappeared.
What is left is a 4uV sensitivity to the switchmode charger
and about +/-3uV ( = 6uV or 1 ppm) low frequency noise in total.
I will have to observe the low frequency noise. Maybe it will
be lesser on week ends than within the week.
In the next layout I will have to pay more attention to the routing
of the power supply lines with regard to the output lines.
With best regards
Andreas
Ok on the one side i'm relieved to hear that. On the other it makes me
hardly to believe that a tempco of 0.05ppm/K can be reached just by heating
the reference. In this case the temperature regulator should have a
stability of 0.001K???
Obviously, that's exactly the case.
In the Datron 4910 manual, p. 1-4, "principles of operation", it's described that the chips temperature is controlled to within 1mK.
The spec is 0.05ppm/K. Although both values do not have to do directly with each other, that seems to be the description of a 50ppm/K TC of the unheated element.
Well, do not add too many capacitance, otherwise the temp regulation may get stuck, heating the element to over 100°C, which may cause drift.
regards Frank
Well, do not add too many capacitance, otherwise the temp regulation may
get stuck,
heating the element to over 100°C, which may cause drift.
regards Frank
Hello Frank,
10nF * 10K or 100nF * around 10K will give a time constant of 0.1 .. 1ms.
The thermal time constant that I have observed is in the range of around
100 ms. So the distance should be enough as long as the output does not
oscillate.
On the other side the unbuffered output is very critical. By accidently
loading the output by an unpowered ADC the LTZ-cirquit has drawn about 75mA
(against 20mA in steady state). Shorting/loading the output of the
unbuffered LTZ will set the heater setpoint temperature to a very large
value. I observed that after this treatment the output was shifted by -11uV
(-1.5 ppm). The shifted value lasted as long as I did several power-cycles
with off-time > 2min to cool down the reference.
After 3 or 4 powercycles the old value was restored. So there seems to be
some thermal hysteresis for large temperature excursions. Unfortunately the
hysteresis is reached even by connecting of a 1uF capacitor at the output of
the reference.
Do you have a buffering cirquit on your devices? Im looking for a cirquit
which does not add to thermal drift and noise. And is not sensitive to
EMC-noise of course.
with best regards
Andreas
Update:
On my noise measurements I had around up to +/- 15uV which give around
4-5ppm peak-peak
But I did not have a 24h measurement.Only relative short times looking on
the oscilloscope.
Perhaps those glitches are caused by the MUX switch of the DVM (34401A or
3458A).
In my case I do not believe that it has something to do with the ADC. I
have seen some pattern
with repeating rate all 0,8 seconds or 2,8 seconds then switching off for
a time and coming back again.
So for my side it looks like a pulse package heated device like a laser
printer or something like that
causing trouble on the mains line. Perhaps comming in to my cirquit by the
oscilloscope.
The low frequency noise was not generated by the LTZ1000A in my case
but came from the measurement amplifier.
For the 0.1Hz lower edge Frequency I had a 20uF foil capacitor
- a 150k pull down resistor. Together with the current noise of
the operational amplifier (LT1013) some kind of shot noise came up.
The fatal thing was that this low frequency noise was only generated
when the LTZ1000 was connected.
Never with other (more noisy) references,
and never when the input of the amplifier was shorted.
Now I'm using a much more low impedant design (3200uF + 1k) togeter
with a LT1037 low noise amplifier to avoid the current noise.
The measurement results are now close to the datasheet value of the LTZ.
best regards
Andreas