Dr Frank, that’s a very nice, very high performance design! I like the star connection between the 10 ohm and the LM399. Thanks for sharing.
Yes, that's right. All six ranges are strictly laid out for Kelvin connections. On top of the shunt resistors, the double pole relays form drive and sense connections, on their bottom, the supply current is fed from one side, and the sense side, including the LM399 ground point, arrives from the opposite side.
I wonder if the 100k and 899k HP resistor are picking up some of the magnetic field from the 50Hz transformer. Would a cap directly across the resistors help with the low frequency noise? A 0.1ufd cap has a 31k ohms impedance at 50 Hz. That may help attenuate any 50Hz picked up by the large wire wound resistor.
No, it's not magnetic pick-up, as an additional magnetic shield, or a different orientation of the transformer did not make a difference.
I may give the 0.1µF capacitor a try, but the disturbances might enter or influence the circuit everywhere else.. obviously the whole ground is riding on this 50Hz hum, and the 470nF capacitor short-cuts that.
This noise is also present in the other ranges, on the same order of magnitude, but due to the higher absolute currents, are of no consequence there.
It would add a delay to the control loop response.
Why? The circuit is stable under all conditions, I see no oscillations, only this noise with 50Hz repetition rate.
If I put the 3458A in series, it will add spikes, which might drive the 34465A into overload, again.
Therefore, much of the noise seems to be created outside my circuit.
I did not realize the 34465A measured the low currents that way. Are there other meters that use an amplified low resistance (1k ohm) shunt?
There are not so many other DMMs on the market, which resolve 1pA, or have a 1µA range.
Usually, that's the reign of electrometers, which operate a bit different, and feature double shielding by means of a triax cable.
Anyway, I also wondered why KS designed it that way. Other ranges are not consequently designed for low burden, and a chopper OpAmp might be more expensive than an additional quality shunt resistor.
In the past, HP justified the lack of more current ranges on the 34401A by the additional price of such shunt resistors.
The 34465a uses a 1k ohm res shunt. The HP3458A uses about 45k. With a 1ua current the 34465 develops a 1mV dc drop across its shunt. For 1 ppm display resolution the shunt voltage needs to be measured with 1nV resolution. Ok, that should be a Keysight problem right? But that also adds a very difficult requirement for the calibrators. It means your output voltage needs to be stable to 1nV. That could be difficult. Am I thinking about this correctly?
Everything is much easier with a normal 45k ohm shunt in the meter (like the HP3458A). The voltage developed is 0.045V and 1ppm of noise is 45nV. That’s not easy but it is 45 times easier.
Well, the problem is somewhat different. The OpAmps only need to amplify a 1nV level to 100nV, which the A/D then is able to resolve properly. So, there's also no output voltage which has to be stable to 1nV.
The problem, you pinpointed here, are un-controllable / fluctuating thermocouple voltages on the order of 1nV across the 1kOhm shunt, to achieve 1pA resolution. Using higher shunt values, implying higher burden voltages, reduces this problem.. You're right, I prefer the 3458A for these µA ranges, although the 34465A works surprisingly well, anyhow.
Maybe a 100k ohm resistor in series between the calibrator and the 34465A would eliminate the noise. Now the output voltage stability requirements for the calibrator would become 100 times easier and more practical.
Nope, I don't think, that this will improve the noise.
The circuit intentionally operates at a high 1V level across the shunts, and that's the relevant
voltage level you have to take into account.
The chopper has absolutely no problem to stably regulate 1V to < 1ppm.
In the end, the stability / noise for time scales > 20msec is extremely good, just compare that to specifications of other current sources from Time Electronics (model 1024), Lake Shore, and Fluke.
In the next design anyhow, I will even use a 100mV reference voltage for the higher current ranges, as here, the 100mA range is very non-linear (~20ppm), due to self-heating effects.. an output of 100mA requires at least one minute to stabilize.
Frank