Q. Do I really need to use dithering?

I have been working on an album and am in the final stage of exporting the tracks to WAV, which will then be burned to CD. All the tracks are in 24-bit audio, but I know that for it to meet the audio CD standard they should be 16-bit.

To increase the volume of a track, I put the Oxford Inflator plug-in on the master out, which tends to work well. The problem is that when I put a dithering plug-in (from Cubase SX3) at the end of the chain (after the Oxford Inflator), the output starts clipping. However, when I just use the Inflator on its own, everything works fine. My question is: do I need to use dithering, or can I just export the master to 16-bit without it?

Via SOS web site

SOS Technical Editor Hugh Robjohns replies: 

You must use dithering because you are reducing the word length from 24 to 16 bits. If you don’t dither, you will end up with unwanted truncation distortions, and although they may not be obvious to everyone during a track, they may well become very obvious during any fade-outs or fade-ins.

I suspect the reason for the clipping you’re getting is because you have set Inflator to raise the peak level of your tracks to the maximum 0dBFS.

Simple triangular dither (which is what I suspect you are using in that Cubase plug-in) adds a low-level broadband noise signal, equivalent to the 16th bit level, to the output of Inflator (it adds noise at about -93dBFS). The result is that where the signal is already hitting 0dBFS from the Inflator process, the added dither noise will be just enough to push it over the top into clipping.

There are two solutions. The first is to adjust Inflator so that it raises the peaks to something a little less than 0dBFS. Something like -0.5dBFS should work better, leaving just enough room for the dither noise without clipping. You may need to experiment a little, but something in the range -0.3 to -1dBFS will cure the problem.

The other solution would be to use a more sophisticated form of dither noise that has been ‘noise shaped’. These dither variations reduce the level of dither noise across the lower half of the audio spectrum (where most of the musical signal energy is) and instead put more dither noise energy up at the higher frequencies, where there tends to be little musical energy and therefore more headroom available to accommodate the dither noise (and where our ears are less sensitive to the noise anyway).

Noise-shaped dither subjectively sounds quieter than simple triangular dither because it takes advantage of the ear’s non-linear frequency response to low-level sounds, although the total dither noise energy remains exactly the same for both forms, and that’s the critical aspect as far as proper dithering is concerned.

There are lots of different noise-shaped dither systems around, some generic and some bespoke commercial forms such as POW-R, Prism’s SNS, Apogee’s UV22 and Sony’s Super Bit-mapping.  

Published in SOS March 2010