EXTENDED LINEAR PREDICTION TOOLS FOR LOSSLESS AUDIO CODING
Takehiro Moriya *, Dai Tracy Yang ** and Tilman Liebchen ***
* NTT Cyber Space Labs., Tokyo, Japan
** University of Southern California, Los Angeles, USA
*** Technical University of Berlin, Berlin, Germany
ABSTRACT
Two extension tools for enhancing the compression performance
of prediction-based lossless audio coding are proposed. One is
progressive-order prediction of the starting samples at the random
access points, where the information of previous samples is not
available. The first sample is coded as is, the second is predicted
by first-order prediction, the third is predicted by second-order pre-
diction, and so on. This can be efficiently carried out with PAR-
COR (PARtial autoCORrelation) coefficients. The second tool is
inter-channel joint coding. Both predictive coefficients and predic-
tion error signals are efficiently coded by inter-channel differential
or three-tap adaptive prediction. These new prediction tools lead
to a steady reduction in bit rate when random access is activated
and the inter-channel correlation is strong.
1. INTRODUCTION
For the archiving and broadband transmission of music signals,
lossless reconstruction is becoming a more important feature than
high efficiency in compression by means of perceptual coding as
defined in MPEG standards such as MP3 or AAC. Although DVD-
audio and Super CD Audio [1, 2] include proprietary lossless com-
pression schemes, there is a demand for an open and general com-
pression scheme among content-holders and broadcasters. In re-
sponse to this demand, a new lossless coding scheme has been
considered as an extension to the MPEG-4 Audio standard [3, 4].
In the course of the standardization process, a time-domain
compression scheme based on linear predictive coding (LPC) has
been defined as a reference model. This model is proposed by the
Technical University of Berlin and the decoding process is shown
in Fig. 1 [5]. For every frame, the optimum LPC coefficients
are calculated and the associated PARCOR coefficients [6, 7] are
quantized in an arcsine-transformed domain. The prediction error
signal is derived by the quantized predictive coefficients and coded
with a Rice code. For stereo signals, simple inter-channel coding
is applied, where either the L-channel or R-channel together with
the difference between the R- and L-channels are coded.
This paper proposes two extension tools for prediction-based
lossless coding. One is progressive-order prediction to improve
performance in the compression of starting samples at random-
access points. The other is inter-channel joint coding for both pre-
dictive coefficients and prediction error signals. Both tools are
described in detail, and the results of performance evaluation are
given.
2. PROGRESSIVE ORDER PREDICTION
2.1. Random access
Samples of an audio signal usually have strong correlation in the
time domain. Auto-regressive linear prediction is well-known as
one of the most powerful and simple tools for reducing the ampli-
tudes of error signals, enabling reductions of bit rate [2, 8]. How-
ever, in the editing and playback of compressed signals, the ability
to start from a random access point is desirable. We thus have to
reconstruct perfect signals without using any of the previous sig-
nal information. Ensuring this property for auto-regressive linear
prediction leads to a significant loss of compression performance,
since prediction must be shut off at the accessible points. Until
now, the first p samples, where p is the prediction order, are kept
unchanged and required separate coding due to a large amplitude.
2.2. Progressive prediction
For starting samples in the random access frames, progressive-
order prediction is useful as a way of making full use of the avail-
able samples and thus reducing prediction error as much as possi-
ble. While it is of course impossible to predict the first sample, the
second sample is predictable by first-order prediction only from
the previous sample. The prediction error at the (q +1)-th sample
is derivable by q-th order prediction in general.
For this progressive-order prediction, PARCOR coefficients
are convenient, since each coefficient is independent from the pre-
diction order p, while normal auto-regressive LPC coefficients
need to be calculated for every prediction order q upto p. The asso-
ciated lattice filter is shown in Fig. 2, where k
q
represents the q-th
PARCOR coefficient. An example procedure of PARCOR-based
progressive-order prediction is shown in Fig. 3. It is understood
output
(L-ch)
+
entropy
decoder
output
(R-ch)
LPC
+
entropy
decoder
LPC
+
prediction
error
side
information
prediction
error
side
information
Fig. 1. Decoding process of reference predictive coding system
with simple inter-channel prediction.