>About maximum level:
The maximum level of a digital signal is governed by the highest peak in the file. Simple normalization finds the
highest peak, then raises the entire signal so that this peak is at the maximum value. However, many of these peaks
may be of very short duration and can usually be reduced in level by several dBs with minimal audible side effects.
Those familiar with digital editing systems may even have proved this for themselves by ‘redrawing’ some troublesome
peaks by hand. By transparently controlling these peaks, the entire level of the file can be raised several more
dB than by simple normalization resulting in a higher average signal level.
>About maximum resolution:
Any digital signal processing that alters the original digital data (mixing, gain changes, EQ, dynamic processing,
etc.) generally increases the number of bits required to represent the signal. Conventional truncation results in a loss
of signal-resolution each time the signal is processed. The human ear uses this low level information to construct a
mental image of the stereo soundstage, so any compromise in this area manifests itself as a loss of spaciousness and
transparency.
Even when processing 16-bit signals, it is normal to process with at least 24 bits resolution, or more (as in the L1,
which is now double-precision). However, as soon as the resolution is pulled back down to 16 bits by rounding or
truncation (by removing the bottom 8 bits), the resulting rounding error produces an audible distortion at low
signal levels, and a permanent loss of digital resolution that can never be recovered. If the audio signal is repeatedly
processed and truncated back to 16 bits, the losses accumulate, causing a significant loss of fidelity, most evident as
a loss of the tonal subtleties of low-level sounds within a mix. The human ear uses such low level information to
construct a mental image of the stereo soundstage, so any compromise in this area manifests itself as a loss of
spaciousness and transparency.
The solution is to properly dither and noise-shape a signal each time the wordlength is increased then reduced
(such as nearly every digital signal process will require).
>Why use them, and what are they?:
Proper dithering is simply this: before the requantization (reduction of the wordlength), a precisely controlled
amount of noise (termed ‘dither’) is added to the signal. This can convert the low-level nonlinear distortion caused
by truncation into a simple steady hiss, thereby removing all traces of low-level non-linearity, but at the expense of
a very slightly increased background noise. Obviously, increased noise levels are not ideal in high quality audio
applications, but fortunately, the perceived level of this dither noise can be greatly reduced by ‘shaping’ the noise in
such a way that it falls into an area of the audio spectrum where the human ear is least sensitive.
The main point of maximum resolution is simple: to ‘capture’ the best possible quality into a shorter wordlength
(smaller bitdepth) from a longer wordlength (higher resolution).
>An example of resolution improvement:
Note: Using an L1 Limiter here from WAVES
To hear what truncation (an extremely severe form of quantization error) sounds like at the 1-bit level (present in
every signal, no matter the bitdepth) and how WAVES IDR and noise-shaping works, try this simple (and exaggerated)
example:
1 -Using a 16 bit, 44.1kHz soundfile, select a desired region. Select the +L1 plug-in within your application
and set your system so you can monitor the L1 output.
2 - Set Dither Type to non, and Shaping to non. Set Quantize to 8 bit for easier auditioning of
quantization distortion.
3 - Listen to the output of the L1 (preview or realtime, depending on your host application).
4 -To more easily hear the quantization error, reduce the the input level at least 30dB, until you hear the
sound becoming distorted in a 'fuzzy' way. If necessary (depending on the input level), pull the Out
Ceiling level down also, to about -15dB. Leave the Threshold at 0.0. (You will have to raise your monitor
levels quite a bit to hear the distortion noise - so do NOT click Bypass until you turn your monitor
level down!) This quantization distortion you are hearing is present in varying degrees in ALL digital
signals at the 1-bit level.
5 - Click Shaping twice. It changes to Normal. You will hear the music cleanly (the nonlinear distortions
are gone), and that a steady hiss in the background is now present. Of course, since you have reduced the
input level, this hiss will be much more prominent than in actual work and is exaggerated just like the
distortion. You can cycle throughout the Shaping options to listen to the quantization error-removing
effects of IDR’s different noise-shaping filters.
6 - For this example, leave Dither set to non. Audition the noise-shaping settings to familiarize yourself with
the distortion-removing properties of IDR. For 16-bit work, the flexibility of the IDR implementation of
L1 gives you the choice of dither and noise-shaping. Please read the following sections on type1, type2,
and Noise-shaping.
