Nothing new in this paper, but it's a pretty succinct and short summary of why NOS no-filter DACs aren't such a good idea:
 
http://www.hagtech.com/pdf/not16bit.pdf

Seems very biased to start with and there are no adequate descriptions of how each set of data was generated and where the errors may be. It would also be nice for a 16-bit comparison as there is little to no evidence in the paper that 24-bit sampling is better.
 
Otherwise an interesting short read.

This paper entirely misses the point : oversampling is necessary to avoid the flawed frequency response of the "zero order hold" digital-to-analog conversion, with a treble roll-off that reaches -4 dB at 22050 Hz.
Zero order hold is the technical choice of generating analog "stairs" instead of analog "spikes" from the digital data.
 
Oversampling at twice the original frequency before performing the conversion rejects the roll-off at 44100 Hz instead of 22050 Hz. Oversample 8x and the roll off occurs at 176400 hz, where there is no signal, and where the noise can be muted without affecting the audio bandwidth.
 
The problems that the paper talks about exist, but are off-topic here. They should be dealt with using dithering, not oversampling.