Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
CA 02304917 2001-O1-10
20104-8634D
1
Method and apparatus for digitally processing a high definition
television augmentation signal.
BACKGROUND OF THE INVENTION
The present invention relates to a method and
apparatus for digitally processing the components of an
augmentation signal as part of a multi-channel high definition
television (HDTV) delivery system. Such a system has been
described for example in U.S. Patent 4,694,338. In such a
system, an HDTV source signal, having for example a 16:9 aspect
ratio and a 525 sequentially scanned or 1050, 2:1 interlaced
line structure, is divided into two signals. A conventional
television signal, for example an NTSC encoded signal
receivable on a standard broadcast receiver, is transmitted
over a regular television transmission medium, for example a
standard television channel and an augmentation signal
providing the additional wider aspect ratio and high resolution
information of the source signal is transmitted over a second
transmission medium, for example a second television channel or
part of one. A high definition receiver can then be used to
receive and combine both signals into a high definition
television display.
SUMMARY OF THE INVENTION
The applications referred to hereinabove describe
systems utilizing analog augmentation signal configurations.
For example, systems using a standard 6 MHz wide NTSC channel
and a 3 MHz wide augmentation channel which could be
transmitted on a presently unassigned television channel. The
only available terrestrial channel capacity however, appears to
be the so called "taboo" channels which are currently
restricted
CA 02304917 2000-OS-04
2
for television transmission. if however, the amount of
power needed to transmit the augmentation signal on such
a channel could be effectively reduced, use of these
"taboo" channels could become practical.
The instant application comprises a digital
processing method and apparatus which digitally encodes
the augmentation signal components cornpressing them so
that the information they provide can be transrnitted in
a narrow RF channel having for example a 3 MHz wide
t0 bandwidth, using a digital transmission scheme such as
aPSK (quadraphase shift keying). This permits a significant
reduction of transmitted power on the augmentation
channel, subsequently reducing the amount of interference
to other channels using the same frequency band.
15 Digital transmission of the augmentation signal not only
permits the use of reduced power, but also enables taking
advantage of the various known error detectior~/correction
schemes, so that a better quality signal can be achieved.
A major disadvantage of digital transrnission
20 is, however, the increase in the transmission bandwidth
necessary. The increased bandwidth requires either the
use of a channel with a wider bandwidth or implernerrt:ation
of data compression schemes to reduce the required
bandwidth. The invention comprises an approach for
25 encoding the enhancement signals based on the latter
solution.
Extensive research in tt~e area of image coding
has shown that two-dimensional (2-D) discrete block
cosine transform (DCT) encoding has attractive performance
3o compared to other encoding techniques while being
practical for hardware implementation. however, ttvese
investigation have traditionally dealt with stancf~,r~l
full-band signals. The compression scheme used herein is
a muditic:d version of one described in a paper ent.itlecl
35 "Adaptive Intra/Inter Frame DCT Coding of TV Pictures"
.lone '19f38, C. Hemus (LEP Philips, France) and the
implt;mentation of a two-dimensional DC~ with Vt.(; fur
' CA 02304917 2001-O1-10
20104-8634D
3
full band two-dimensional signals (regular images) was first
introduced by W. Chen and W.K. Pratt in a paper entitled "Scene
Adaptive Coder" (IEEE Transaction on Communications, March
1984). The results of these studies cannot be readily extended
to the encoding of augmentation signals which in part possess
only high frequency content.
Augmentation signal components derived using for
example, an encoder first described in U.S. Patent 4,694,338
(issued on September 15, 1987), and comprising, for example, a
luminance panel signal Yp, two chrominance panel signals (Ip
and Qp), a high frequency luminance signal YH and a line
difference signal LD, are those processed in accordance with
the invention. The instant invention takes these augmentation
signal components, samples them and converts them into for
example, 8 bit per pixel digital signals. Each of these
digital signals is then fed into a separate coder which reduces
the number of bits per pixel required to reconstruct the
original signal components. This compression is achieved by
quantization and removal of redundancy. By quantization, we
mean the elimination of certain information. The compression
scheme is based on the use of DCT (discrete cosine transform)
processing in combination with VLC (variable length coding) and
each augmentation signal component has its own coder, which is
adapted to its unique characteristics.
BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 is a diagram of an encoder described in U.S.
Patent No. 4,694,338.
Figure 2 is a diagram of a high definition NTSC
(HDNTSC) encoder described in U.S. Patent No. 4,873,567.
CA 02304917 2000-OS-04
4
Figure 3 is a flow diagram representing a general
description of the digital processing and transmission scheme of
the instant invention.
Figure 4a describes one embodiment of a coder for the
panel information.
Figure 4b describes one embodiment of a decoder for
panel information.
Figure 5 describes the reordering of processed panel
information.
Figure 6 is a graph representing the buffer regulation
procedure.
Figure 7 is a graph describing the criteria for
determining the weights for panel information.
Figure 8 describes one embodiment of an LD encoder.
Figure 9 describes one embodiment of an LD decoder.
Figure 10 describes the reordering of LD information.
Figure 11 describes one embodiment of a YH coder.
Figure 12 describes one embodiment of a YH decoder.
Figure 13 describes the reordering of YH information.
Figure 14 is Table 1 which shows how the number of Mega-
samples/sec is computed for each signal component.
Figure 15 is Table 2 which illustrates the computation
of bit rate for each signal component and the combined signals.
DETAILED DESCRIPTION OF THE INVENTION
Figure 1 is one embodiment of an encoder which can be
used to derive luminance and chrominance signal components-for'
both center and panel portions of a high definition television
CA 02304917 2000-OS-04
4a
source signal. The operation of this encoder ZO and its use in
the derivation of an augmentation signal is fully described in
the '338 patent in which this figure appears~as Fig. 4. As
explained therein, a HDTV source is processed and outputs 18, 20
and 22 of the 3-channel demultiplexer 32 provide respectively, a
luminance (Yc), and two
CA 02304917 2001-O1-10
20104-8634D
chrominance components, for example Qc and Ic, for the center
portion of the HDTV source. This center portion presents
picture information which, as taught by the '338 patent, is
then NTSC encoded and transmitted over one standard television
5 channel. The outputs 12, 14 and 16 of 3-channel demultiplexer
32 provide respectively Yp, Qp and Ip components for the panels
derived from the HDTV source signal.
Figure 2 describes an HDNTSC encoder which utilizes
the encoder 10 in conjunction with bandpass filter 35 to derive
a high frequency luminance component YH from the HDTV source
signal. This encoder is disclosed as Fig. 2a of the 4,873,567
patent which describes a system providing extended horizontal
resolution of luminance and chrominance in a high definition
television system. The YH component 36 is available at the
output of the bandpass filter (BPF) 35. A line difference (LD)
signal component is available at output 38.
The described embodiment of the instant invention
provides for the processing and transmission of the derived
augmentation signal in an approximately 3 MHz wide RF channel.
The augmentation signal consists, for example, of the following
signal components: Yp, Ip, Qp, LD and YH.
A flow diagram of one embodiment of the digital
compression and transmission scheme of the invention is shown
in Figure 3. Each of the signal components is fed through a
separate analog to digital converter (A/D) which uses a
sampling frequency approximately twice the bandwidth of the
respective signal component. In this way the lowest possible
number of samples/sec is obtained before coding. The analog
augmentation
20104-8634D
CA 02304917 2001-O1-10
6
signal components Yp, Qp, Ip, LO and YH are bandlimited
as described in Table 1. Like the applications cited
herein, the augmentation signal of the instant invention
utilizes a line period of approximately 127 Ns end
better known and described in these applications as a
"superline". Fig. 14 shows how the number of Megasamples/
sec. is computed for each of the signal components.
After sampling, each of the digital signal components
is converted into a stream of 8 bit pixels or samples
52. The resulting number of megasamples/sec for the
signal components is 2.54 MS/sec, 0.36 MS/sec, 0.72 MS/
sec, 2.74 MS/sec, and 2.46 MS/sec for Qp, Ip, LD and YH,
respectively. The digital signal components are then
passed through separate coders labeled CoderP, CoderQ,
CoderI, CoderLD, and CoderYH 53. Average resulting
compression rates are about 1.0 bits/pixel for Yp, Ip
and Qp. Rates of about 0.2 bits/pixel fur LD and about
0.3 bits/pixel for YH have also been achieved.
The digital signal components are then multiplexed into
a single bit stream 55. Based on previous computations
of MS/sec and average compression rates, this bit
stream will have bit rate of about 4.91 Megabits/sec.
With an additional overhead of 20~ added for error
protection (see below) the bitrate is 5.89 Mb/sec.
Computation of bit rate for each of the signal components
and the combined signal is illustrated by Fig. 15.
The: column of Fig. 15 marked "Coding Rate" indicates the
operating rate of the system for each of the signal
components. This constraint is irnposed on each codes.
Conventional error protection/compensatiun
methods can be used 56 to process the bit stream pr:iur
to modulation and transmission in order to mitigate
the effects of noise in the transmission channel me:jium.
The hit stream can be transmitted in a 3 MHz wide RF
channel using for example a 2 bit/sec./Hz QPSK trans-
mission scheme 58.
CA 02304917 2001-O1-10
20104-8634D
7
At the receiving end, the QPSK bit stream is
received and demodulated 60, and any error protection/
compensation method used is compensated for 62. The
demodulated bit stream is then demultiplexed into
individual compressed digital augmentation signal
components Yp, pp, Ip, LD and YH 64. These compressed
signals are then decoded by passing each signal through
its own decoder 66. The recovered digital augmentation
signal components can now be converted back into analog
form and fed into a 2 channel decoder 68 which is fully
described in for example, the U.S. Patent No. 4,873,567.
The. block diagram of an embodiment of a coder
for panel information, i.e. Yp, Ip, or fjp, is shown in
Fig. 4a. The panels from two consecutive fields are
combined to produce one panel frame. More specifically,
the even numbered lines of even numbered fields are
intCl'leavad with odd numbered lines of odd numbered
fields. The resulting combination is regarded as one
frame of panels and focus the input 100 of the panel
encoder 53a. The frame is partitioned into separate
blocks of equal size N, where N - 16 and two-dimensional
(2-D) discrete cosine transform (DCT) processing is
applied separately and independently on each block
using transform means 110.
The coder 53a operates on a black by block
basis and performs the same series of operations on
each block. Its operation therefore, will be described
only for one block represented by x(i,j). Here, we wi:l1
assume that the blocks in one Frame are indexed from
left to right and top to bottom (i.e., similar to a
raster scan format) and the indexing is continued in d
similar fashion for subsequent frames. AFter tha 2-Cl Gi:~
transformation, the do coefficient, X(0,0), which
represants the average gray level or luminance of the
block, is separately quantized using for example a 9-bit
quantizer 112 (a simple float to integer rounding
opCration), ttie output of which is than provided to
20104-8634D CA 02304917 2001-01-l0
8
multiplexer 115. The rest of the coefficients X(u,v),
where fu,v - 0,1,...,N, N - 1; (u, v) ~ (0,0)}
are treated separately as described below. The activity
of each block, which is indicative of amount of detail
(for example edge content) of the block, is then
calculated and classified using block classification
means 114 according to the formula:
A(m,n) - max I X (u,v)~
(u,v)~(0,0)
u,v - j0,1,...,N - 1~
where A(m,n) represents the activity of the block (m, n).
The block is then assigned to one of K - 4 activity
classes, using a fixed set of decision levels
0 = ro < rl = 10 < ?'2 _ 25 < Z'3 = 50 < 2"4 _ 255.
That is, block (m,n) is assigned to class k iF r
k-1
< A~m,n) < 2'k. The assigned class of the block requires
two bits (1od24) of inFormation to be uniqnel.y identi-
fiable. The assigned class along with the normalization
factor, (described helow) are then used to select a sat
of weights suitable for that class and normalization
factor. Selection of the appropriate weights takes place
using weighting means 116. The weights associated
with all coefficients indexed by n - (u+v)/2, where
o - (),1,...2(N-1) are calculated according to the formula:
wi(n) _ wi(N - 1) +Iwi(N - 1 ) - 1, (n _ N ,. 1)/
(kN),
where i indicates the block number or the operating st,;te.
Calculation of these weights is illustrated by the graph
of Figure 7. As is clear form the previous expression,
the weight assigned to each coefficient depends on the
calculated value of wi(n - 1 ), referred to as the
central coefficient or the normalization factor which
is in turn evaluated using normalization means 124
acc:ucwi i nd to formula
wi(N) - f(l.~i-1(N), bs).
Eierc:, f(.) is a piece-wise linear function having a
hufF~:n status (bs) as shown in the graph of Figure Ei.
20104-8634D
CA 02304917 2001-O1-10
9
The buffer status is described as the ratio of number
of bits in the buffer 12 to the buffer size.
All the transform coefficients, except for the
do coefficient, are divided by the weights calculated
as above 118 and then quantized using a simple float
to integer rounding operation in quantizer 120.
The output of the quantizer 120, X(u,v), can then be
reordered (scanned) 122 using the pseudo zig-zag scanning
pattern illustrated in Fig. 5 in order to provide better
match for thecaariable-length coder (VLC) 128.
The VLC 128 consists of several tables, each
designed for a specific normalization value ~ri(N - 1),
and activity class, k. The normalized values provide a
better match to input statistics than previously provided
by single coders such as those described by the prior
art. The appropriate table, selected based on the
normalization value and class assignment, is then used
to encode X(u,v) to the ULC 128. The output of VLC 128
is fed to buffer 126 for further preparation (channel
coding and modulation) for transmission. The status
(bs) of the buffer 126 is then updated and the norrnali-
zation value for the next block, i + 1, is calculated
using normalization means 124. The output of buffer 126
is combined with the output of quantizer 112 and the
respective block class provided by block classificat:iao
means 114, in multiplexer 115. The multiplexed output
is then modulated on to a carrier in modulator 117 arid
transmitted over part (i.e. 3 MHz) of a television
channel 119.
The role of the decoder 53b, in the absence of
transmission errors, is to recover the digital signal
components input to the coder 53a, allowing for some
degradation due to quantization depending on the
oE~arating rate of the system.
Figure 4h describes one embodiment of a decoder
for panel information, i.e. Yp, Ip or lip. The information
mo r:harrnel 11J is demodulated by demoduldtur 1~ 1 arosJ
CA 02304917 2000-OS-04
the do coefficient X(0,0) is separated from the rest
of the coefficients and the respective assigned blor:k
class in demultiplexer 125 and provided to inverse DCT
transformation means 123. The rest of the demultiplexed
5 coefficients are input to buffer 130. The output of the
buffer 130 is fed to the variable-length decoder (V(..D)
132 while the status of the buffer 130 is used to
calculate, using normalization means 134, the normalization
value used in the coder 53a. The respective assigned
10 block class is provided from demultiplexer 125 to
weighting means 13b. The normalization value from 134
and the assigned block class 125a from demultiplexer
125 are input to and thereby enable the VLD 132 to
select the table used in the VLC 128 and therefore
to recover the value of the coefficients, X(u,v).
The resulting coefficients are then processed 132a to
compensate for any recording process used during encoding.
Weighting means 13b derives the weight factors used in
encoder 53a for the coefficients (using the block class
and normalization factors) and these factors are then
combined in multiplier 140 with the processed coefficients
from Vt_D 132. Along with the do coefficients, X(0,0),
coefficients X(u,v) are inverse transformed in inverse
DCT means 123 to provide the output of the decoder
~2(i ~J)
The encoding and decoding procedures for l.0
components are similar to those for the panel cornponeots
as shown in block diagrams Figs. 8 and 9. Because of
peculiar nature of the LD component, there is no weighting
of the DCT coefficients. That is, all the coefficients
are normalized using the same normalization value.
Iv addition, the (0,0) coefficient in DCT domain is
treated the same as the rest of coefficients.
1=urt:lmrmore, the recording method using for l.D is
~trr~5,;r~ Lo better suit the statistical prot~erties of
this signal and is shown in Fig. 10. Test; results
strowed t:trat the value of the normalization factor is
CA 02304917 2000-OS-04
11
usually small and therefore only one codebook is used
in the VLC and the normalization value is taken as
a real number.
Block diagrams representing the operation of
embodiments of an encoder and decoder for YH are shown
in Figs. 11 and 12. The operation of the encoder for YH
is very similar to that for LD except that the reordering
process, shown in Fig. 13, is selected to better suit
the properties of the YH component. In the above described
t0 embodiments, the means for transforming, normalizing
weighting and variable length encoding/decoding the
signal components can be accomplished using hardware
or a computer program. Persons skilled in the art
will be able to provide the means shown and use thorn in
15 accordance with the instant invention.
The foregoing disclosure and description
of the invention is illustrative and explanatory thereof
an~i various changes in the details of the embodiments
shown may be made within the scope of the appended
20 claims without departing from the spirit of the invention.