I have a high precision voltage reference and I need to build a symmetrical voltage reference from it. As the goal is to get a few ppm tracking from the reference, it is very difficult to create a way to invert the voltage using a OpAmp as an inverter as it requires precision resistors and tuning.
So I tried the LTC1043 solution that can be found in several datasheets and app notes, especially a Jim Williams one in which the precision is said to be around +/1ppm (or +/-2ppm in the datasheet).
It is a very simple and straightforward technique using a sample and hold circuit reversion the ground location between input and output.
The problem is that I couldn't achieve something in this area of precision. For a 5V input (and +/8V power supply), the error is about 2.5-3mV ! Which is 500 to 600ppm ! There is no load on the output, I'm using polypropylene capacitors (1uF) and 10n capacitor for the internal oscillator. Everything is shielded. Power supplies are very clean and linear types.
So I tried to simulated with LTSpice and to my surprise, the results are no far from what I get, even though the error is less severe (about 150uV). Simulating a non inverting S/H does give perfect results, but as soon as I invert the output, there is a serious signal degradation.
Measuring voltage at the sampling capacitor gives half the error and the other half is on the hold one. It seems that there are some leakage that discharge the capacitor that does not exist when using the no inverting way.
Whatever I try, I get exactly the same results. Have I missed something ?
If there is no way to use this method to invert a voltage reference, what would be the best one ?
Thanks in advance,