[Lagerfeldt]

Once you record in 24 bits in your DAW, it will usually operate internally at 32 bit float or maybe 48 bit fixed (ProTools TDM) when processing.

When you bounce your mix for mastering, it should be as a 24 bit file with a peak reference at around -6 to -3 dBFS (the digital ceiling).

During the mastering process the file is kept at the highest possible resolution digitally and when treated via analog gear, captured at 24 bits again. During the final bounce in mastering dithering is applied while going from 24 bits (or higher) and down to 16 bits, which is the required bit rate for an audio cd.

Dithering is a process that adds a type of arbitrary low noise signal (sometimes shaped via psycho acoustics to make it less obvious to the ear) to enhance the perception of dynamics. This enables a 16 bit audio file to sound closer to the original file, maybe approaching 20 bits of resolution in an actual 16 bit file.

A very popular type of dithering is the POW-r#3 (or #1), which is heavily noise shaped.

Once dithered, the file should never be alterede in any way, no eq, no compression, no gain changes or fades. All this should be performed pre-dithering.

Multiple dithering of 16 bit files is not a good idea as the dithering algorithm then can become very audible. So don't record and process files in a multitracking session using dither, especially not if you're going to do any further processing to the files.