Optimizing the RF side of a radiosonde receiving station can be a bit tricky, but here's some thoughts on this:
In an ideal world, our receive systems would only be limited by thermal noise, and so we'd be targeting the best noise figure we can on our receive chain. Noise figure is directly related to receive sensitivity, so the lower our noise figure the better our sensitivity.
In the real world, we are often limited by other factors, such as:
- Intermodulation Distortion (IMD) issues: Other strong signals (which could be inside or outside your receiver's passband) mixing and resulting in unwanted IMD products in the frequency range we are interested in. This may set an upper limit on your receiver gain.
- Local Noise Floor problems: A higher-than-thermal noise floor due to interference (this could be broadband, narrowband, or anything in-between) created by electronic devices in the vicinity or your receive antenna. Pretty much anything with a switching power supply produces RF noise to some degree, as do ethernet cables.
SDR Only
If you were using only a SDR, then the best noise figure is achieved at essentially maximum gain. According to measurements conducted by SDR Makerspace the best noise figure of a RTLSDR is about 6 dB, which occurs at absolutely full gain. (The Airspy R2 gives similar performance).
Unfortunately at this gain you'll often be limited by other factors like receiver intermodulation distortion (IMD), so it's sometimes best to back the gain off a bit from maximum. If you use AGC the SDR will generally wind the gain up pretty close to maximum. In many cases this is 'good enough', so often there isn't much point in doing much more optimization unless there's obvious IMD issues occurring (lots of spurs in your spectrum plot).
SDR + Preamplifier
For the case where you have a preamp (and ideally a filter!) near your antenna, the system noise figure (again... in an ideal world where we are limited by thermal noise) is almost entirely set by the noise figure of that preamplifier. In the case of an Uputronics filtered preamp, that's about 1 dB.
There are of course contributions to the noise figure from other devices in the receive chain, but realistically unless you are using up all the gain (~21 dB for the Uputronics unit) of that preamplifier in subsequent losses (be it cable loss, or poor noise figure devices) then you are probably not going to degrade the overall system noise figure too badly.
So how do we pick a suitable gain setting in this situation?
You could choose to leave the SDR in AGC mode, but the extra signal levels hitting the SDR due to the preamp may result in worse IMD issues.
I've found 2 ways you can tweak the gain settings. Both of them require viewing the SDR output 'live' with SDR software like GQRX, SDR++, or otherwise. This might mean plugging the SDR into some other computer to do this, or running rtl_tcp on the machine the SDR is attached to and connecting to it via a network.
Empirical, based on Radiosonde signal SNR
The first way is to watch a radiosonde signal in-flight and adjust the gain for maximum SNR. If you increase the gain and the SNR doesn't go up, then you've probably hit your system's performance limit and any additional gain increase is just going to result in limited dynamic range and possible IMD issues.
Empirical, based on noise floor increase
If you don't have a radiosonde signal available, then you can get a rough idea of a suitable gain setting by initially setting the gain to 0, then watching your noise floor as you increase the gain. For maybe the first 15 db or so of gain settings the noise floor is unlikely to change much (indicating your receiver is limited by its internal noise figure), but as you increase the gain you will eventually see the noise floor start to rise. It's around this gain setting that your receive system has now become limited by the noise figure of the preamp (or possibly by other external noise), and any additional SDR gain won't help.