When to Apply Dithering?
There are a number of choices of dithering to pick from, and
dithering is actually appearing in far more places than ever
before, making it even easier to misuse and abuse this
critical stage of sound production, reduction, and
reproduction. If uncertain about what you're doing with
dithering, the best thing to remember is simply this:
Do not apply dithering to anything unless you're at the
VERY LAST STAGE of production, as you're
transferring your material to the format you will be
listening to it in. In most situations, you only want to
apply dithering once, and the time to do it is just as you're going to the CD that will be the Master CD for
duplication during Mastering.
A good rule of thumb is to apply dithering only when you
absolutely MUST reduce the bit-rate of the material you're
working on. An ideal scenario when working in a DAW like Pro
Tools (or any others) is to work in 24-bit and to mix down
to 24-bit as well. There is no need to go to 16-bit until
you want to make a CD to listen to your music, but that CD
should be made from 24-bit masters that every DAW is capable
of producing. There are far more choices of dithering and
far more flavors of dithering than most are aware of, and
the point of the following discussion is to help point out
why dithering can often be the most critical stage of
production and mastering. Improper dithering can vastly
change the sound of your material, and improperly or
repeatedly applied, can introduce clipping, noise,
artifacts, and all kinds of unpleasantness to mixes you
worked so hard making perfect.
What is Dithering?
Noise-shaping is a method of improving the signal-to-noise
ratio of dither, resulting in an SNR more like the higher
resolution source. It can't be perfectly done - if we could
fit 24 pounds into a 16-pound bag we wouldn't need the
bigger bag.
Noise-shaping curves are tailored to the limits of human
hearing. For 16-bit output, the object is to keep the total
power of the curve at 16 bits, but to lower the noise in the
areas where the human ear is most sensitive. In critical
bands it is possible to lower the noise to 20-bit levels.
In Figure 1, an FFT illustrates the dilemma at three
ninth-order curves: Meridian's shape D (blue curve), POW-R
dither type 3 (yellow) and Waves Ultra dither type 1 (green,
and from its new L2 Ultramaximizer). Notice how each
manufacturer follows the same general shape, but each one
chooses fine variations on that shape. Noise masks signal in
the same frequency band. Thus, in the frequency ranges where
the dither noise is reduced, more of the source music and
ambience is revealed (unmasked).
Different Kinds of Dithering
The best thing to remember is this:POW-R is short for
Psychoacoustically Optimized Wordlength Reduction. POW-R
dither is owned and licensed by the POW-R consortium LLC, a
group formed by four independent audio companies Millennia
Media, Lake DSP, Weiss and Z-Systems. The goal was to
produce the most sonically transparent dithering algorithm
possible. The process the consortium came up with is not
just dither with noise-shaping, but a patent-pending
approach shrouded in secrecy.
POW-R is similar to other wordlength reduction schemes in
its use of a random noise signal as part of the
signal/quantization error decorrelation process. How this
noise signal is generated and what its spectral
characteristics and statistical properties are,
differentiates POW-R.
POW-R dither comes in three flavors intended to fit
different kinds of music. The POW-R consortium will license
the algorithm noncompetitively to its own members and to
additional licensees, including notable DAW, processor and
converter manufacturers.
Dithering in Use
That's the theory. My skepticism turned to wonder - my first
reaction when I hooked up the POW-R box was "is this thing
on?" I thought I was listening to the source! I rechecked my
connections, inspected my scope and switched around until I
was sure there was no wrong connection. Apparently the POW-R
dither avoids major soundstage shrinkage and tonality
changes associated with other dithers.
There are three types of POW-R dither. Previous experience
has shown me that rock'n'roll music frequently benefits from
a gentle noise curve or a UV-22 approach; I think these
curves cover up some of the evils in the source and tend to
be less edgy sounding on brashy material. Classical music
tends to benefit from the ninth-order curves, like Meridian
shape D, which has greater depth and width yet has a bit of
a sheen to it that makes it unsuitable for bright music.
Because each dither affects the tonality of the sound, I
integrate it into the mastering process.
Ideally, a dither should be tonally neutral, with the 16-bit
final sounding like the 24-bit master. In the FFT of Figure
2, note that POW-R type 1 (blue) is a near-Nyquist dither
second-order curve, which resembles UV-22 in its shape. Type
2 (green) is a slightly steeper fifth-order curve, which
resembles Meridian type E. And type 3 (red) is the severe
ninth-order curve, like Meridian shape D or the new Waves
Ultra curve.
As a serious dither connoisseur, I expected to hear
meaningful differences between the POW-R curves. But again,
I was wrong. Using a delicate 24-bit classical piano solo
recording, I could not reliably distinguish the three 16-bit
curves from each other or from the 24-bit source! This has
never happened before. I usually notice some degradation and
a meaningful difference between any box's curve shapes.
I tried an excellent recording of a country rock group,
which has a lot more high-frequency information than the
classical piano. This recording came in on 1/2" 30 IPS
analog tape and I had previously mastered it using various
digital processors and Meridian D dither. Starting with
POW-R type 3 (the sharpest curve), I was amazed at the
clarity, depth and lack of congestion in the music.
With this music, I was finally able to hear the smallest
difference between the 24-bit source and the 16-bit dithered
version; it's hard to describe, the 16-bit being a bit more
closed in but a respectable facsimile of the original in all
respects.
I then tried POW-R type 1 and heard a bit more degradation
compared to type 3, but nothing to write home about. I
certainly can't use POW-R for tone control; it may soften
the sound, but more subtly than its predecessors.
Basically, after a few tests, I decided that the way to use
POW-R dither is to stick with the best-sounding curve; POW-R
Type 3. This is not the dither to use if you need to cover
up some evils in your source.
POW-R neither disguises nor enhances any part of the
spectrum.
We can also conclude that the effect is music-dependent;
complex stereophonic music appears to expose dither problems
more than simple. As a control, I compared the POW-R against
my reference dither, Meridian's shape D. I had originally
chosen shape D over other dithers because it revealed more
of the source and added a little bite, which was desirable
with this music. So I was concerned that POW-R's tonal
neutrality might produce a more recessed sound. I found that
POW-R reveals so much more of the source that I no longer
needed the Meridian's artificial edge - I was obviously
using the edge to make up for the its lack of resolution.
Instead, the transparency of the original source carried the
day.
I really started noticing the congestion of the Meridian
algorithm, yet the Meridian's resolution is slightly better
than other dithers. It's all a matter of degree, but it
didn't take long for me to conclude that the POW-R dither is
better than anything else out there. Negligibly different
from the source, it's an incredible achievement.
Summary
Putting this in perspective, many people are insensitive to
the virtues of the different dithering systems. What order
of magnitude of differences are we talking about? If you are
inclined to music that has depth, space, clarity and purity
of tone, you will recognize the qualities provided by a
superior dither. A high-resolution monitor system is useful
to distinguish the sound of subtly different dithers from
each other and from the source.
Examples of White Noise Dithered various ways
Below, you’re not looking at anything other than Noise
Floors of the music in relation to frequency. In other
words, using the POW-R3 as an example, the noise floor,
after dithering, sits at about-140db across most of the
sound spectrum, but drops to -165db around 3K, and
sharply increases to about 105db from 12K to 20K. Since
our ears are not too sensitive to noise in lower
frequencies, very sensitive around 2K – 5K, and and less
sensitive at higher ones, the noise curves are created
to “hide” the noise created from reducing wordlengths
from 24-bit to 16-bit in places our ears don’t hear
(WOW!).
|
