That's weird, the bigger the cap, the less bandwidth the circuit has, thus it is more stable. Can you please share the waveform of that 50mV p-p? I wonder what the frequency is. If it's 50/60Hz (or multiple of it) it's mains hum. Otherwise it is oscillating, or it's a noise from somewhere.
I attach both the waveform obtained with the two capacitors in series as well as the oscillation present when there is no capacitor whatsoever.
With two capacitors in series, the circuit works perfectly now, no problem whatsoever. It might very well be that the oscillation in the 20mV range is noise of some kind.
The signal with the 2 caps does not look much like noise, though it does not look like an oscillation either. It looks more like EMI, like a radio signal, SMP ripple or some other relatively strong interference. Ocillation would normally be a more repeatable waveform and noise would be more random, except if the circuit is very close to the edge of oscillation.
It definitely is not oscillation. Adding another capacitor in series doesn't change the amplitude or frequency of the oscillation at all. The circuit sits in a shielded box, without any trace of hum. I will follow previous advice and solder a 100nF directly to the board and see if it is a PS problem.
Okey great news a 100nF helped tremendously. The remained noise seems to be pretty much EMI. However, I already can report great results. With the input connected to the turned off power supply, a Fluke87V measures 0.3mV (presumably from the alredy mentioned interference). Preeliminary tests with mV signals reveal great perfomance, even though the offset is noticeable. Note how the output cable goes outside the shielded box. It is interesting how just covering it with sticky piece of aluminum tape (and connecting this "shield" to ground) reduces the signal measured in the oscilloscope by a noticeable degree. If at the end I want to display the signal on an oscilloscope, I think I should just add a low pass filter (which it seems like the Fluke already has incorporated).
Is the box painted on the inside? Its clearly not presenting a problem to you now, but its worth keeping in mind that you can get strange effects due to static charges and things I can't remember the name of, so sometimes coating it with a relatively high resistivity coating like Aquadag or nickel loaded spray paint can give avoid some problems particularity if you're chasing down to fA's or have high voltages involved
Well penfold, it very well might have been the paint (it is indeed painting in the inside), or charge building in my vinyl gloves, anything else.
I think that the op amp is dead. When turned on, it saturates or gets very close to doing so, then after a minute or two it briefly goes down to the correct value and then goes to saturation again. In the next step, I will add the protection resistor, and also a more robust input connector.
It will be a while because I just started the memorizing process of informations for exams known as university, but will report in as soon as everything is set in.
Once again, thanks to everyone. This forum has been extremely welcoming and helpful, unlike most internet places I've visited, and will do my best to contribute to it and make it even better.
Well penfold, it very well might have been the paint (it is indeed painting in the inside), or charge building in my vinyl gloves, anything else.
I wasn't really suggesting it could be what killed it, just maybe in the future if you have any strange inconsistent offsets or strange non-linearities,
Best of luck in your exams
The BAV199 has a reverse leakage of at least 100pA at zero volts
How do you know this? The datasheet (from nexperia) promises much better performance.
The datasheets have quite a large spread between the typical and maximum leakage specs. This is because testing to low leakage takes time and is expensive.
The numbers, especiall for the test limits are different between manufactures, so the number can vary somewhat From the DS I have, I get
Philips BAV199: 3 pA typical and 5 nA max at 75 V (rather high voltage)
Interfest 2N4391 100 pA max at 20 V
Fairchild An6609 process 51: typical current near 15 pA for 20 V
The BAV199 has a low typical leakage, but a rather crude test. The main difference for the way more expensive PAD1-5 is that they are actually tested to lower currents.
So chances are that at some 10 V the leakage of the PAD , BAV199 and 2N4117 JFET gate is similar, just the price is different (some $20, $0.20, $5 ), refelcting the test limits of some 2-5 pA , 5 nA and 10 pA. I would be surprized if the lower cost MMBF4117 are actually tested to 10 pA - more like they accidentially coppied the data from the 2N4117.
Diode leakage may be an interresting thing to measure with the DIY nV meter. Just make sure to keep RF out.
Edit:
For testing the TIA type meter, the offset / bias current is measurend with an open input. The TIA does not like the input shorted and than gives erratic errors from the OPs offset voltage. So the test case is an open input. The capacitor in Fb should be relatively small, maybe some 10 pF (preferrably PS type) or the like.
With CMOS opamps, I found it better to include a decent package size series resistor on the opamp input (inside the feedback loop) and use the on-chip protection diodes for minimum leakage current.
Eg. A 1M 2W resistor will limit the protection diode current to 1mA at 1kV - well within ratings. It works particularly well with opamps that have bootstrapped protection diodes such as the LMC662 or LMP7721 which give typical input bias currents of 3fA and have internal protection diode rating of 5mA (10mA on the LMP7721).
It is a cheaper approach too.