And the results are in...!
Over the last 90 days I have seen my LM299 reference drift about 137uV. Since the HP3458A was not in the correct mode during the first measurement there is a little error. The drift can be between 132.1uV (13.2 ppm) and 142.0uV (14.2 ppm). This is not bad and in the datasheet they specify a typical drift of 20 ppm over 1000 hours. After 180 days there is only 37uV (3.7 ppm) drift since the 90 days measurement.
I also learned that the LT1013 opamp is not suitable as a voltage booster in this kind of circuit. The drift on the 10V reference side is 270uV or 27 ppm over 90 days. This is of course not much...but you know.
During the 90 days the voltage reference has been powered and it still is. Since the LM299 I used was "new" it will be interesting to see how much it will drift in the future. It has now had a 90 day burn-in period which should be enough.
Not long ago I received my LTZ1000A from Linear Technology. So now I have started designing a new and better reference...or so I hope. Follow the development of my LTZ1000A reference here.
After cleaning up the lab I found a bag of LM299. This is a voltage reference...and in the better end of precision references. So I decided to build my own 10V reference. I thought it would be fun to see how precise I could get it. Below you can see the final result.
On the internet I found a lot of information about different ways to approach this idea. I learned that the most precise reference out there is the LTZ1000A from Linear Technology. But since I already had lots of the LM299 I decided to used this one for my project. Maybe in the future I will build something with the LTZ1000A.
Building a precise reference can be difficult. There are lots of things that can introduce noise and instabilities. I tried to follow many of the correct design approaches. I did see lots of people designing their references to an insane degree. I decided to find some place in the middle...maybe leaning to the insane degree.
One thing I did was use precision resistors from Vishay. Again I had some RNC90 resistors in my "junk-box" and a precision trimmer which was a good start. A good Op Amp was needed and I used the LT1013 since this is a good low noise Op Amp. The HP3458A 8.5 digit multimeter from Agilent uses this Op Amp in its reference section. And since this is the best DMM out there its was a no brainer.
Below are some of the build pictures.
First mistake I made (according to some) was building it on standard epoxy FR4. To minimize heat transfer I drilled some holes around the reference itself. Heat travelling in the FR4 is kept low since there is not much material to travel through. The leads to the reference and the leads to the heater is seperated. This way noise from the heat regulation is not producing interference on the reference wires.
I wrapped some Kapton tape around the TO-house to isolate it electrically from the ESD-foam. And at the bottom I covered the leads with Kapton again to isolated from the ESD-foam.
The ESD-foam isolates the LM299 so that temperature changes do not affect the output voltage. But this foam is conductive and is therefore isolated from the circuit with the Kapton tape.
This is the finished PCB and the only thing missing is an enclosure. Again to isolate the circuit from temperature changes I used some pink ESD-foam inside the enclosure.
...more foam pictures...
...and done. I decided to put a small heatsink on the LM2941 regulator. Not that it matters much...but just in case. The regulator is getting 20V and it outputs 15,61V. And the whole circuit draws less than 30mA. This will be less than 150mW of power.
The schematic is pretty simple. About 6.99V comes from the reference which is then amplified to 10.0000V in the precision amplifier circuit. I used two inputs a V(supply) and V(battery) this way I can move the reference from place to place without disconnecting the power. The whole circuit operates on 15.61V which is a regulated voltage delivered from the LM2941. The voltage from the LM2941 is actually very stable. Resistors R8, R9 and R11 are 1% and all the others are RNC90. Trimmer R7 is a Bulk Metal precision trimmer from Vishay.
To reduce noise a 1nF capacitor was added in the feedback loop. By adding this I got rid of "hum" problems when my hand was near the PCB. I could see a variation of several micro volts on the output without the capacitor. Resistor R11 was added as bias compensation so that both input sees the same impedance.
Before calibrating the 10V reference I had the whole thing powered for several days. This way I am sure that it has "burned-in" and the voltages will not change significantly. As can be seen in the pictures the input voltage from the regulator is 15.6148V and the reference voltage is 6.99040V.
The final result is shown here. A nice 9.999966V which is only 33uV from 10V. This is about 3 ppm which is fine by me. :)
But all this precision talk has led to another problem! All my DMM's need calibration.