Sonifex NICA X, Справочное руководство

Страница: 61 / 64

NICA X Reference Manual
53
A
PPENDIX
C T
HE APT
-X100 A
UDIO
D
ATA
C
OMPRESSION
S
YSTEM
C.1 - Introduction
Linear PCM coding of audio has now become virtually synonymous with ultimate sonic quality. Certainly, it has
raised both professional and public expectations of the performance of audible media. For most of its applications
16 bit linear PCM produces fine results, but to take advantage of it in many situations is not without its problems.
In most of these, the overriding limiting factor is the enormous amount of bandwidth it occupies, with the general
result that it is too costly or complex to be practically implemented. For example, using a linear 16 bit PCM
system sampling at 32kHz, the basic binary bit rate will be 512kbit/s per channel.
The bandwidth problem of PCM is particularly acute in transmission environments, where the transmission costs
are proportional to the channel capacity. As a result, a number of alternative music coding schemes have been
developed which require significantly lower operating bandwidth for these circuits. Such schemes, employing near
instantaneous or delta modulation techniques, operate at bit rates of around 256-400kbit/s per channel, and
include error protection overheads. However, for many potential low capacity digital audio applications, for
example ISDN, these levels of compression are still inadequate.
The excessive data rates of these existing music compression schemes have been due primarily to their adherence
to relatively simple digital companding techniques. These systems exploit little of the natural redundancies
associated with the sound signals of interest, unlike their speech coding counterparts. The situation has remained,
partly because of the higher sampling overheads involved, and until recently, the absence of high speed Digital
Signal Processing (DSP) hardware.
Sub-band ADPCM can be described as a "medium-complexity" scheme which appears particularly suited to high
quality audio coding and which does exploit the considerable natural redundancies of audio. A high coding
efficiency is ensured in this system as it not only incorporates the benefits of digital companding, but also takes
advantage of time and spectral redundancies by using linear prediction and sub-band coding.
This Chapter presents the concepts involved in sub-band ADPCM, explaining the nature of the signal redundancies
which are exploited to provide a transparent 16 to 4 bit compression.
C.2 - Linear PCM Digital Audio Coding
The process by which audio is sampled and coded with a binary bit stream has the advantage of producing an
accurately repeatable quality. Noise and speed variations introduced by storage and transmission media can be
eliminated, as can multiple generation degradation, and performance criteria such as frequency response and
distortion can be better controlled and therefore virtually guaranteed. However, the problems created by the very
much higher frequencies required to code audio digitally, coupled with the necessity to maintain the original
integrity of the code throughout any processing, inhibit and complicate its use in many applications.
The bit rate of PCM digital audio is defined by the sample frequency and the word length. The analogue signal
must be sampled at least twice per cycle in order to code the highest frequencies accurately. In practice it is also
necessary to filter the high frequencies to avoid anti-aliasing distortion. By coding in this way an analogue
waveform is systematically dissected into many component parts, each one of which is accurately labelled so that
it can be exactly reconstructed again, regardless of whether or not the components are audible to the human ear.
In many cases it will always be desirable to do this, however there are a great many instances where it is a
considerable extravagance if not a positive disadvantage.
C.3 - Characteristics Of Audio
An audio signal can be defined as an analogue waveform containing frequency components to which the human
ear responds. To be considered musical the signal must also be pleasing, and this is a factor on which some very
low bit rate coding schemes base greater emphasis than others, but all rely on it to some extent.
The characteristics of the response of the human ear are fundamental to the elimination of coding redundant data.
Loud sounds mask quiet sounds of similar frequency, for example, and sensitivity is biased to low frequency
sounds. Musical notes are made up of fundamental frequencies and a series of harmonics and, because of the
tendency of the ear to lower frequencies, there are few signals that are considered musical with a fundamental
frequency above about 4kHz. In fact music and audio signals in general exhibit a diversity of redundant
characteristics of which very low bit rate coders are designed to take advantage.
The foremost requirement of hi-fi music coding is the maintenance of a high coding transparency. This implies
that the quality, bandwidth and distortion/noise levels of both original and coded music should not be subjectively
different. In theory a process relying on the inherent redundancy in music to maintain signal quality might not
prove satisfactory for non-redundant signals. Fortunately, most signals of this class already incorporate specific
perceptual redundancies to compensate for this (e.g. noisy signals which will invariably mask coding error). This is
also true for transient signals, which are exceptionally tolerant because of, amongst other things, temporal
masking (the response of the ear to short, sharp rises in level).