Review, teardown and potential improvement of SRS DC205 Voltage source

[maxwell3e10]

This is a detailed review, noise measurements and teardown of Stanford Research Systems DC205  voltage source. The source can generate DC voltages on three ranges +/-1V, +/-10V and +/-100V with 6 digit resolution and 4-wire sensing. The current is limited to 20-50 mA. It is based on LM399 voltage reference and has 1 ppm short term stability and linearity. Overall the voltage source works quite well and may be the best commercial DC voltage source short of a calibrator. However, it would be nice to improve it even further by switching to an LTZ1000 reference as well as changing a few chips. I have also explored using the 4-wire sensing capability to increase its output current and to convert it to a current source.

This picture is of course carefully timed during initial warm-up, but the absolute accuracy of the 3 sources I checked is quite good. The voltages are measured by a HP3458 recently calibrated to TiN voltage standard from USA cal club.
Unit 1: 1V: +2.5 ppm, 10V:-3.3ppm, 100V: 0.5 ppm
Unit 2: 10V: -1.6ppm, -10V: -1.1ppm
Unit 3: 10V: +0.9ppm

The linearity is about +/- 1 ppm. The DC source uses a lower grade AD5791A and applies a correction. Since the digital resolution is 1 ppm, there are sharp steps of 1 ppm in the residuals.


The noise spectrum of the output is quite good, but clearly can be improved at low frequency with a better voltage reference and op-amps. One can see they contribute about equally to the drift because low-frequency noise decreases when the output voltage is set to 3V, but it doesn't go down further for lower voltages or for 1V output range.

The broadband noise of the voltage source is pretty good at 50nV/Hz^{1/2}, comparable to LTZ1000 reference.

Below are the pictures of the insides, with some of the chips labeled.
It uses a number of ADA4522 zero-drift amplifiers and LT5400 resistor networks for voltage scaling. However, there are also OPA192 FET input amplifiers which maybe responsible for some of 1/f noise. Perhaps replacing them with zero-drift OPA189 would help. But I have not yet tried to trace the circuit and would welcome people's observations and suggestions.

[Kleinstein]

Besides the OP (OPA192) the ADC itself can also contribute to low frequency noise. The low frequency noise level for the AD5791should still be a bit better than what is seen.

There are plenty of AZ OP, so it is not clear that there are some non AZ OPs in the critical path and contribute much to the low frequency noise.
The 100 V range would need some relatively high value divider and this may not be suiteable to work with a low noise AZ OP. A direct replacement of the OPs is not that easy.

The INL curve looks quite good - so the LM399 reference currently inside seems to be a good one, well better than average.

[Andreas]

What I would try first:

shielding the (long) legs of the LM399 against air drafts.
(if possible also on the other side of the PCB).

BTW: how is the influence of tilting the device on the output voltage?
https://www.eevblog.com/forum/metrology/project-pimp-a-keithley-2000/msg1106829/#msg1106829

with best regards

Andreas

[maxwell3e10]

Thanks for the suggestions. The sensitivity to tilting the voltage source is not very large, the biggest effect is when it's resting on the left side, the change in voltage is 1-1.5 ppm. Since the power supplies are on the left and there is no fan, it's probably not surprising. I will try thermal insulation when I find a piece of pipe insulation of the right diameter to fit over LM399.

But after starting real-world use of the voltage source, I quickly saw a limitation. Its output impedance is only low for low frequency, for frequencies above 1 kHz the output impedance rises to about 20 Ohm. I think the problem is that voltage sense feedback has relatively low bandwidth. I can see why SRS would want to limit the bandwidth to ensure good stability under a variety of conditions. But in case there is high-frequency noise injected into the output, it's nice to have a wider bandwidth feedback. One could also add a capacitor at the output, but it appears non is built-in.