| Direct
Stream Transfer |
 |
|
 |
|
 |
Super
Audio CD |
|
|
|
|
|
|
|
Direct Stream
Transfer (DST)
|
As
seen with the
hybrid disc structure,
the 4.7GB layer of the SACD disc can hold two
complete, 74-minute versions of the music :
2-channel high-quality (DSD) stereo and a 6-channel
high-quality (DSD) surround sound. Such an amount of
data can be stored on a single layer thanks to a new
losless coding method developed by Philips and
called Direct Stream Transfer (DST).
|
Indeed,
the DSD sigma-delta modulation presents a high
oversampling rate, resulting in a raw audio
data-capacity which is typically 4 times as high as
that needed for current CD signals.
|
Data
reduction is consequently necessary to store all
high-density channels on a single layer. In general,
there are two different types of bit-rate reduction
methods : on the one hand, lossy data reduction
methods choose parts of the original signal which
can be ignored, and remove these parts from the
final data stream to be stored in digital format.
These lossy methods, often based on psychoacoustic
models, include MPEG-1 and MPEG-2 for video signals,
and Dolby Digital (AC3) and DTS for audio signals.
|
On
the other hand, lossless coding methods have been
primarily developed for computer applications. They
reduce the amount of data while preserving the
original data integrity, bit for bit. Although the
sigma-delta modulated DSD signal inherently has a
very noisy structure, it turned out that lossless
coding can be successfully applied and consequently
the required capacity for storage or transmission
can be reduced. Direct Stream Transfer is a very
sophisticated coding method involving data framing,
adaptive prediction and entropy encoding stages.
|
In
order to enable random access of a losslessly coded
stream, this lossless coding scheme operates indeed
on a frame basis. Coding parameters such as
prediction coefficients are optimised once per
frame.
|
Entropy
encoding stages take advantage of the non-uniform
distribution of symbols appearance probability. In
other words, the shortest codes are assigned to the
most frequent symbols, while the longest codes are
assigned to the least frequent symbols. The
substitution table is built during the encoding
phase and has to be transmitted with the final
encoded data. In the case of DST, the entropy coding
is adaptive, i.e. it depends on statistics related
to the audio data to be compressed.
|
In
the case of text (ASCII) documents, this type of
encoding method is very efficient, since some
letters (ex : e) have a higher appearance
probability than others (ex : x). It has also very
good results with PCM signals because low amplitude
values are more recurrent (gaussian probability).
However, basic entropy coding is less efficient with
DSD signals, since amplitude is related to pulses
and not any more to absolute amplitude values. As a
consequence, a new adaptive prediction method has
been developed.
|
Indeed,
the lowest achievable coding gain of a frame is
important in determining system parameters such as
buffer size and processing power. It is therefore
desirable to have the lowest achievable gain as high
as possible. It was found that by allowing adaptive prediction
filter structures, the worst-case
lossless coding gain of any frame could typically be
improved. An improvement in the worst-case coding
gain at the same time decreases the gain variations
and consequently, the constant coding gain at the
output of the buffer increases.
|
Thus,
Direct Stream Transfer can achieve a 50% reduction
in bit rate, with zero loss in data integrity. This
is very impressive since halving the data required
means doubling the storage capacity.
|
|
|
|
|
|
|
 |
