It looks as though this is a very noisy environment. It also looks to me as though the precision you believe you need is excessive. Further, you haven't mentioned how much noise thee is in your logic system. Have you explored the power-gnd noise in your power supply? Have you examined your voltage reference?

You probably can take 12- or even 16-bit samples, but how many meaningful bits of valid information can you reasonably anticipate? Keep in mind that whatever induced noise finds its way into your circuitry will probably also find its way into your test equipment.

This is not an easy measurement situation.

RE