Hello DiligentMinds.com,
wow that is a lot of very good and interesting Information. Thank you very much, for sharing your knowledge. That gives me indeed some new ideas.
...consider using an LT1051 instead of the LT1013 ... you could use a quad chopper (LT1053A)
I will check the LTC1051 and LTC1053
The two 70K resistors should be ultra-high quality metal-film type-- accuracy is not that important (they could be 0.1%), but they should have as low a temperature coefficient of resistance [TCR] that you can find-- probably 5ppm.
The 120-ohm, 12K/1K-divider, and 10V-output-divider resistors all need to be a metal-foil resistors from Vishay Precision Group-- their newest "Z1-foil" resistors are the best, and will have the best long-term stability. These will be about $10 each, if purchased from an authorized distributor. If you can afford it, the hermetically sealed type will have even better long term stability-- and will reduce variability due to humidity and barometric pressure changes.
I tried to find the best possible resistors and ordered Vishay Z-Foil (the hermetically sealed VHP101 and the lower cost Z201). They have very long lead time (probably 5 more weeks for me to wait). I wonder where your 10$ price point comes from, because I paid up to more than 50$ for one, so please let me know.
The "LTZ1000(A)" does *NOT* need to be "stress relieved".
That is good news, because I still have no idea how to do that with diptrace (but I am still interested to know if/how it is doable)
although it should be covered with something (both top and bottom), because any air flow (as in convection currents) will cause tremendous low-frequency noise (due to the thermal EMF's between the Kovar package leads and the copper PC board). You can use anything for the cover-- you're just trying to stop air-flow-- not make an insulated box.
I have only thought of a top cover so far, but now will think about how to do a bottom one too
Your schematic doesn't appear to have any way to adjust the 10V output. If I were doing this, I would make the 10V-divider a 3-resistor type (with 3 resistors in series). The bottom resistor would always be 5K (for 5V) and the next two would total up to 5K, but their values would be governed by the actual voltage of the 7V reference output (after 300-hour burn-in). When you order metal-foil resistors, you can specify the exact value you need, and they will supply you with that value (at the tolerance that you pay for). The 10V-divider resistor tolerances should be (at least) 0.01%. The reason for the 3-resistor divider, is that now you have a 5V node that can be used to adjust the 10V output. At this 5V node, you would tie 1M-ohms to 5M-ohms of resistance (depending on the sensitivity you need) made up of one or more 5ppm metal-film resistors. The other end of this large resistor would be tied to either a potentiometer (the Vishay Precision Group 1285G series) or (even better) an unbuffered 16-bit or 18-bit DAC (TI DAC8871 or Analog Devices AD5781). You would tie the Vref-H directly to 10V-out at the banana jack, and Vref-L directly to ground at the banana jack-- do not use buffers (as their ap-notes say to do) on either the reference inputs or on the output voltage. Without the buffers, the INL spec will suffer some, but the repeatability and time+temperature drift of the output will be ~0.05ppm.
About the trimming of the 10V, I have considered some options, but not decided yet which to choose. That is why the schematic still shows the AN86 suggestion with fixed Vishay VHD200 ratio set. But I have not orderd this and will probably go for either:
- similar to your suggestion with a precision trimming circuit as you decribed (also already ordered together with the other parts the Z-Foil Trimmers 1260 and 1285)
- with KVD as shown in AN86 (see att.), for this I could use my Fluke 720A and/or my ESI Dekapot
In the drawing are X= solder copper juctions, how should/can this be done? Just leave a gaps and bridge it with solder or with a via and fill it with solder?
Your DAC suggestion is probably to complicated for me.
And I do not know yet if I even maybe leave it as simple as possible and just go for a chopper buffered nominal 7.2 Volt Output and do anything else externally (if I ever want/need it).
Another thing you could do is place a (fc=100mHz) low pass active filter between the reference output and the 10V amplifier-- something using a very low-noise amp (like the LT1112), and a DC-agnostic topology for the filter (like here):
http://www.millertechinc.com/pdf_files/MTI%20TN094%20Zero%20DC%20Offset%20LPF%20and%20the%20D%20Element.htm
Another thing you might consider (especially if you are using an LT1053 quad chopper), is to use 2 of the chopper amps to "discipline" a low-noise amplifier (like the LT1097 for example)-- this will lower the DC error and 1/f noise of the LT1097, while at the same time, you will enjoy the low voltage noise.
I will check that
And finally, if you can afford it, I would do 2 additional things-- first, think about putting the whole circuit in an oven. In this case, you would use the lower-cost LTZ1000 (not the 'A' version), and control the outer oven's heater with the LTZ1000's temperature sensing transistor. In this configuration, the LTZ's internal heater is not used. Set the oven temperature to keep the LTZ die at ~45-deg-C, for a reasonable compromise between a good environmental temperature range, and low long-term drift-- (higher temperatures cause more drift over time). The other thing I would do (if you are really serious) is go with the hermetic resistors (as I mentioned above) and also use a hermetic packaged LT1053. (The other amps and the buffer don't matter-- their drift is controlled by the LT1053, and the LT1112's drift will not affect the DC value of the reference if you use the DC-agnostic filter I showed you in the link).
If you can only afford one hermetic resistor, let it be the 120-ohm Zener current control resistor (or whatever value you use for this)-- as this resistor is the most critical in the circuit. The other resistors (the temperature control divider and the 10V divider) will tend to drift together, and we are only interested in their ratio (mostly if you discount the error caused by the output trim circuit if the 10V divider drifts).
If you do all of the above, you would probably get around 0.75ppm (7.5uV) of drift per year, which is *VERY* good. But if you REALLY want to get crazy, forget about the hermetic resistors and LT1053 (just use the epoxy versions) and place the whole circuit on top of a double-stack Peltier device-- all inside of a hermetic package (yes, they make them that big). Inside of the hermetic package, you would insulate the circuit with silicone foam rubber (which can handle the high heat used to solder the cover on). The hermetic package would then be fastened to a heat sink that would be fan-cooled. So now, instead of controlling the LTZ's die temperature to 45-deg-C, you are controlling it to 10-deg-C (or even lower if you filled the package with nitrogen or argon before sealing). Doing this would get the long-term drift down to less than 0.25ppm (2.5uV) per year-- which is phenomenal, and would far exceed anything currently available at any price (except for a Josephson Junction Array).
How do I know all of this? I have been studying this problem for a very lonnnnnng time...
I have already bought several LTZ1000ACH, LT1013ACN8 and Z-Foil resistors (and have some Burster) to make two or maybe 4 boards.
The hermetic packaged LTC1051/LTC1053 I have not seen/found yet. Do you know where to get it?
About your really get crazy suggestion, I would like to talk about that in more Detail.
Thanks again for your great post.
bye
quarks
Home
Help
Search
Login
Register
