Eddy-current sensors are not just for prox. Kaman specializes in custom displacement measuring systems. We can run at frequencies as high a 120khz (well, we probably can go faster, but that's the fastest shipped to date) and get resolutions down to 2 nm. Not at the same time, at least not yet, we're working on that. We have one system which is a custom PCB & sensor design where the resolution requirement is >3 nm @ 1khz and that's a number we can ship from production without tweaking. These systems range in the 1.7 to 2.8 nm range, with a 10 Volt max output, which translates to a noise measurement of 113 to 187 MICROVOLTS PEAK TO PEAK. I don't agree with companies that quote great resolutions and use RMS measurements. Think about what you're doing here! As a sensor user, I'd want to know the uncertainty of my measurement and you're not going to get that from RMS. I spent a lot of years dealing with video noise in Medical Imaging Equipment and there, RMS noise makes sense, because the eye detects noise as RMS. The eye (and the brain that's hopefully behind it) does averaging. Now go and measure something and you'll see that the p-p noise is full range of disturbance of your numbers. The RMS noise does not, and for a very important reason....RMS doesn't notice peaks. If you have some lousy op-amps in a circuit with a lot of shot noise, your RMS measurement will look the same as a good system, but won't perform as well. And, you have the inevitable marketeer that likes to play with the crest factor to 'make things look better'. I've always used 6.6 as the factor between RMS and p-p for random noise. I have seen some use factors as high as 9 and 10. To me this makes the RMS measurement useless.
Anyways, that's my take on the situation.
Getting great resolution at high frequencies is not easy. Filtering to get great resolution is easy, but of course you lose speed and there's the resulting phase shift to deal with.