I was wondering how they avoid being swamped by analog noise in the lower ranges with such a high dynamic range.

I suppose they use different gains per DAC and map that to the respective dynamic ranges of the f32 signal?