[#3300] Wavetek 7000, the hidden gemstone.

[chekhov]

Today little accident likely revealed the whole story of failure of this particular unit.
Attempt to connect chassis to 10V LO output lead to immediate 'reboot' of the unit  .
Rather quickly the culprit was found, it was a battery pack:

One of bolts that keep this battery pack shorts single cell to chassis, depending on how it was screwed and addition al unit movements.
Thus, initially burned 1 ohm resistor was likely the consequence of doing the same what I did, but for a longer period of time. Could you imagine, this little mechanical issue lead to whole this story  .

[TheSteve]

Wow, a pretty terrible assembly job of that battery pack, nice job finding the fault.

[branadic]

Just thinking loud...

When I've build the reference presented here, I was first measuring the initial and residual t.c. with the oven set to the z.t.c. point of the zener and found it to be at +0.175 ppm/K. The reference was buffered only.
I've then compensated this t.c. and ended up at -0.0105 ppm/K after this step.
Afterwards I've installed the boost to 10 V and found the t.c. to be -0.195 ppm/K, that was then trimmed within the boost stage.

So question is, could the residual positive t.c. of the reference have compensated for the negative t.c. of the boost stage? Seems like I need more investigation to answer that question, so I've already removed the compensation on that board.

-branadic-

[Kleinstein]

It is well possible that they only cared about the TC of the final result and choose the parts for the boost stage to compensate for the small TC of the zener part.  The compensation could be intentional (select the parts before soldering) or just luck (some units work well, other may get a rework or selected trim parts).

[dietert1]

As far as i remember we have seen schematics that included a nice measurement of zener temperature from the base-emitter voltage of LTZ1000 Q1 and then a resistor array used with jumpers to set TC in a binary fashion, with very fine LSB. No luck involved!
By the way my LTFLUs include a very similar temperature measurement circuit, but i roughly compensated TC and then used the residual nonlinear TC to adjust the TEC oven for the zero TC temperature. In the LTFLU case the 10 V boost is included in the same oven.

Regards, Dieter

[aronake]

Does any of all knowledgeable people here know if there is any disadvantage of reducing what is called R2 in the LTZ1000 data sheet to 30K as Datron/Wavetek have down to 30K? Analog may just not have thought of the benefit of a much lower R2, but for the ADR1000 datasheet they just lowered it to 61.9K, and it ought to be well known to them with the benefit of lower R2 by the time they wrote the ADR1000 data sheet. And for ADR1000, there is even more benefit than in LTZ1000 as no need for the extra around 10 ohm resistor to bring the non heated TC to zero. Their unheated schematic in the LTZ1000 data sheet show they were aware of the benefits of the 10 ohm compensator and the lower R2.

Amazing reverse engineering here by Chekhov of the schematics!

And the "secret" final TC adjustment mentioned in some places mentioned in relation to this design, what is that?

[Kleinstein]

Lowering R2 gives less noise from the transistor - usually more the higher frequency part, not so much the 1/f noise.
At 100 µA a small transistor (e.g. bc847) should be at some 4 nV/sqrt(Hz) of white noise, which is way less than the LTZ1000 noise. So the transistor noise should not be that relevant.

However it also comes with a slightly higher TC of the ref. part if not temperature stabilzed. The difference is not much: some 22 mV higher BE voltage, which results in 70 µV/K less for the temperture effect, or 10 ppm/K more for the overall TC (e.g. 60 ppm/K instead of 50 ppm/K). This makes the temperature stability and setpoint divider a little more relevant.

The adr1000 has a different TC to start with, here more current could even lower the TC.

[aronake]

Lowering R2 gives less noise from the transistor - usually more the higher frequency part, not so much the 1/f noise.
At 100 µA a small transistor (e.g. bc847) should be at some 4 nV/sqrt(Hz) of white noise, which is way less than the LTZ1000 noise. So the transistor noise should not be that relevant.

However it also comes with a slightly higher TC of the ref. part if not temperature stabilzed. The difference is not much: some 22 mV higher BE voltage, which results in 70 µV/K less for the temperture effect, or 10 ppm/K more for the overall TC (e.g. 60 ppm/K instead of 50 ppm/K). This makes the temperature stability and setpoint divider a little more relevant.

The adr1000 has a different TC to start with, here more current could even lower the TC.

Thanks for this! My pretty basic tests with ADR1000 seems to indicate 0 TC unheated at around 25K ohm for R2 though. So benefit would be lower noice and lower TC. One abvious disadvantage thugh is slightly more current draw, but that ought to be compensated by the heater drwaing less current. So still very unclear why data sheet say 69K, but there seems to be benefit in most areas by setting this lower.

[chekhov]

You may try different combinations of R2 and R1, can also decrease current back. Maybe question of time spent running thermal sweeps with different combinations in order to select desired temp set point.
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5507800/#msg5507800
Unfortunately that all does not help with LTZ1000, and upper Rz may add much more long-term instability and its own TC as well.

[aronake]

You may try different combinations of R2 and R1, can also decrease current back. Maybe question of time spent running thermal sweeps with different combinations in order to select desired temp set point.
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5507800/#msg5507800
Unfortunately that all does not help with LTZ1000, and upper Rz may add much more long-term instability and its own TC as well.

I am working on this now.

I use an IET PRS300 for R2.
https://www.ietlabs.com/prs-300-programmable-decade-resistor.html

So I can automate change of R2 as temperature goes up and down. Then trying 70, 80, 90 and 100 ohm for R1 which I change on the board. Using an ADR1000.

I still dont see any benefit in such high R2 as 70K as per the data sheet.

[aronake]

I have found a VERY strong reason for not using too low R2. R2 is allready the resistor that have most impact on output voltage. Lowering this from 69K as per the data sheet to 24K where typically unheated TC is zero, dependence on the resistor increase more than 10 times! At 69K it is still very possible to make a very low heated TC, but lowering R2 to 24K would make this even better, but as R2 age with time and change resistance value this would have a much bigger inpact on outpur voltage so a much worse long term stability. And long term stability generally being much more important than 0 TC.

What i did was to keep ADR1000 datasheet circuit at 65 degrees, then change R2 to some different values and measured voltage output.

[chekhov]

Yep, that's true, if I'm not mistaken, around 20k R2 influence factor is smth like 55:1, far from what datasheet shows. But this is where 0 TC is located for me.
I think that more relevant info for ADR1000 thread.

[Kleinstein]

The increasing effect of R2 drift is indeed a reason not to lower the resistor. It is however not as bad as the numbers in the table suggest. The relevant part is the relevent change in the resistor, not the absolute change. It still gets roughly 3 times worse at 24 K.