LTZ1000 daughter board

[Grandchuck]

Having completed two of Dr. Frank's LTZ1000 reference boards (many thanks to him and to cellularmitosis for their contributions). I have compared them to the LTZ1000 daughterboard associated with the AD5791 evaluation module.  There is more to be done (data collection) but so far Frank's design yields a standard deviation a little over 1 microvolt and the daughterboard a bit over 4 microvolts.  It is noteworthy that the FX reference standard deviation ala TiN measures quite a bit less than one microvolt.  I have yet to get my hands on TiN's gem but my turn on the USA cal club list is approaching.

One big unknown is my equipment which is a vintage 2011 3458A DMM that has never been recalibrated.  Back in 2011, the DC volt gain with a 10 volt input had a measurement uncertainty of 3.64 E-5.

On to a question.  The attached partial schematic shows R40 and R39 values of 68k ohms and they happen to be surface mount devices (I know nothing else about them).  I am wondering if it would be a good idea to replace those with better devices.  The LTZ1000 data sheet cites a 0.2 ppm or a 0.3 ppm change for a 0.01% change in resistance with those two parts.

[RandallMcRee]

. . .

On to a question.  The attached partial schematic shows R40 and R39 values of 68k ohms and they happen to be surface mount devices (I know nothing else about them).  I am wondering if it would be a good idea to replace those with better devices.  The LTZ1000 data sheet cites a 0.2 ppm or a 0.3 ppm change for a 0.01% change in resistance with those two parts.

There are some very good resistors available in SMD. So, there is no guarantee that replacing them will be a good thing!
Personally, I would try to evaluate the module, first, and only make changes where indicated. Don't fix what is not broken.

[chuckb]

There is more to be done (data collection) but so far Frank's design yields a standard deviation a little over 1 microvolt and the daughterboard a bit over 4 microvolts. 

If your concern is the 4 microvolts of output noise then there are two areas you may want to investigate.

The 1N270 diode is strange. It is a Radio Frequency Germanium diode. An RF demodulator is not what you want in the middle of your precision circuit. Also the dc voltage drop of the heater germanium diode may be to small under high ambient temperature conditions (low heater current) for correct operation of the LTZ1000.

The chip operating temperature is set to 95 deg C by the 15k resistor (R41). This, by itself, will not cause extra noise but it will greatly increase the possibility for noise caused by convection air currents over the PCB Copper and Kovar leads of the LTZ1000. This thermocouple has a 35uV / deg C reaction. It is easy to get a 0.1 deg C temperature difference if the LTZ1000 leads and pcb connections are not sealed and protected from air currents. The LTZ1000 operating temperature can also be reduce by changing R41 from 15k (95 deg C) to 13k (60 deg C) or lower. This can be quickly tested by paralleling a resistor across the existing 15k Vishay Z201 resistor ($18).

Good Luck!
 

[Dr. Frank]

Deeply back in this thread, there is a table of measured 'attenuation' factors for the different resistors. Any change of one of the 70k is attenuated by 250, the other by 1000, so forget about the datasheet. I think, R2 and R3 designators are referenced to latter schematic.
Also keep in mind, that hp used thin film only for these resistors.
So one of these should be better, less timely and T.C. drift .

I also recommend to lower the 15k to 12..13k for best stability.
Don't forget: Noise is not everything .
(... Physics is everything!)

Frank

[Grandchuck]

The standard deviation is now under one microvolt!  Thanks to all of you!

Changed from 15k ohm to 13k ohm as suggested and replaced the diodes with 1N4148s.

R39 and R40 remain the same.  As pointed out, their effect is not pronounced and the originals are of good quality.