A while ago I posted this at the UAD forum, I thought it may be useful here....
Originally Posted by scopeAhaa, very interesting topic!Originally Posted by J-Mac!Originally Posted by jdnuisanceThe question is this, if dither is "adding" low level noise, and you have your limiter set to 0dbfs then aren't you going to clip your signal with the dithering?, or does dither somehow not raise the overall level? JD
JD, I usually set mine to 0dbfs. Dithering to my ears seems to kill low level noise. I use the Waves L1. But I'll be soon purchasing the Precision Limiter (UA) and try to put it through it's paces. If it's what everyone says it is, I will more than likely find my new baby for mastering.
dithering *IS* the process of adding low leve noise, or hiss.
Of course necessary when you are bouncing from a higher bit depth to a lower one.
Without dithering - you are simply rounding, or truncating the bits.
This will create both harmonic distortion and intermodulation distortion.
If your DAW only uses say 16 bits internal resolution then any form of gain change ITB whether it be normalization or running a signal through a plugin with any kind of filter, or simply summing on a buss will require a dither (since multiplying 2 16 bit values will give you a 32 bit value)
Luckily most DAWs use either 24 bit or 32 bit floating point making dithering for such things not necessary due to the increased dynamic range.
If we take the example of bouncing out to a 16 bit file from your DAW, its done by adding noise with a level lower than the LSB before then of course rounding to 16 bits. The effect this has is simply to spead the effect of all the tiny length errors caused by the rounding. It is spread across the audio spectrum as white noise. The noise is added simply be adding randomly either a 0 or 1 to the LSB.
As well as having different properties as to how well they reduce the distortion, the different dithering algorithms also change the shape of the noise so it manifests in areas of the spectrum that are less objectionable to the human ear.
Theoretically aswell, dithering should allow you to hear audio that you wouldnt hear without the dither (i.e. those signals that fall below the level of the LSB). The added noise will increase the level of some of those signals over the level of the LSB, and thus cause a bit transition.
Fortunately our ears and brains work in such a way that we can easily discern the difference between noise and another signal and almost filter out the noise - analagous to listening in to a conversation in a noisy room.
Now as to the original question.
The process of randomly adding 0 or 1 to the LSB of a 24bit signal before rounding down to 16 will produce a peak noise level of around -90dBFS, and an average noise level of around -93dBFS.
If in the prior stage you are limiting to 0dBFS
Because by definition we are adding noise to a signal, then yes, its my understanding that theoretically we can hit the peaks when adding dither.