Recording - February 2016 - 71

Theoreticians maintain that greater bit
depth only makes for a quieter signal, but
here my guru's distinction between theory
and practice kicks in: most audio engineers will tell you that the difference
between 16- and 24-bit signals is clearly
audible as a difference in clarity and
detail.
Why the discrepancy? Well, it's important to remember that the conversion happens in hardware, and hardware lives in
the real world. We may be hearing the
limitations of that real-world hardware
rather than a theoretical difference. I
don't know, but I do know that studio
engineers, who also have to live in the
real world, almost inevitably record using
24 bits. (What about the 32-bit files
offered as an option in many DAW systems? It turns out these aren't really 32
bits; instead, they're a way of formatting
24-bit numbers that makes them easier
for a computer to work with. The audio
signal is still sliced into 24-bit slices.)
This encoding system, which digitizes
the analog signal into binary-coded
words, is called pulse code modulation,
or PCM. It's used for audio recording
and telephone transmissions around the
world, and at this moment a PCM system is busy transmitting pictures of Pluto
to the planet Earth. The basic principle
of PCM is that (to super-simplify) if the

A/D conversion is done accurately, if
the D/A conversion is done accurately,
and if what's in between (a storage or
transmission system) delivers the right
numbers at the right time, it can honestly
reproduce the original signal waveform.

referring to sample rates as "kHz", and
although it's not literally correct, audio
engineers keep doing it. Why?
The answer is that although it's wrong,
it's not completely wrong. To explain why,
I need to take a rather wide detour.

Time in a bottle
I just mentioned time for, well, the first
time. I've talked so far about how the
audio signal is sliced on the vertical
axis-representing how big the audio
signal is with numbers (the essence of
digital recording). Something equally
crucial for defining how a converter
turns analog waveforms into numbers is
when it measures: the horizontal axis of
our graph. The rate at which the waveform is sampled is called, logically
enough, the sampling rate.
Compact discs use a sampling rate of
44,100 samples per second, while most
video formats sample the audio portion
at 48,000 samples per second, and
"high-density" audio often samples at
96,000, 192,000 or even 384,000
samples per second. Why so high?
Seemingly unrelated: The sample rates
are sometimes described as "44.1 kHz"
(for CD), "96 kHz", etc. This is technically incorrect; we're talking about samples
per second, whereas "Hz" means cycles
per second. But the habit persists of

Connecting the dots
Every now and again a novice in audio
recording comes to me with the earthshattering discovery that digital recording
can't possibly work. Inevitably they show
me a drawing like Figure 2, which depicts
an analog waveform being sampled by
an analog-to-digital converter; the sample
points are again shown as red dots. If you
then take the red dots (samples) and turn
them back into analog audio, you discover that there's an ambiguity; the dots can
be reconstructed in more than one way.
Figure 3 shows one possible waveform
that can be made from the dots.
That waveform looks nothing like the
original. So is my interlocutor right? Is
digital recording impossible? Here's the
surprise: the argument is completely
valid. Its scope, however, is less than universal; the reality is that you can't sample
(and therefore digitize) that waveform
with those sample points. Otherwise the
system will create those ambiguous frequencies which aren't in the original signal; that's called aliasing.

http://www.realtraps.com

Table of Contents for the Digital Edition of Recording - February 2016