AD converter, try 1

Just recently I ran for the first time in a situation where I actually needed pretty high-precision AD converter for a specific measurement. Previously I had only used 10- or 12-bit converters, and even those were generally situated in less than optimal location on the boards (read: in middle of digital circuitry), and those designs weren't... great, so to say, so I was kinda nervous about this thing.

But nevertheless, it needed to be done, and so I took every single tidbit of information I had read from anywhere and applied it to this one board. Ground planes? Check. Low-noise power supply and reference? Check. Separate power supply for digital logic? Check. Truckload of ground vias? Check. And so on and so on. In the end I ended up with 4-layer board had scarcely more stuff on it than power supplies, AD converter chip itself , connectors and some passives. I might've gotten away with just two layers, but why push it too much.

And what I got out of it?

20 bits. 20 bits of clear, noise-free signal. This uses 3,3v reference, so this means about 3,15 microvolts per bit. And of course further averaging could bring noise down even further, but in this case, this result is a great success, but for my application this was easily enough. And board didn't even need second design round, it worked perfectly the very first time.

For a first attempt, I really think this was pretty nice result.