2000 May 03
26
Philips Semiconductors
Product specification
Encoder for MPEG2 image recording
(EMPIRE)
SAA6750H
Table 10
Digital vertical timing (NTSC)
LINE NUMBER
F
V
H (EAV)
H (SAV)
1 to 3
4 to 19
20 to 263
264 and 265
266 to 282
283 to 525
1
0
0
0
1
1
1
1
0
1
1
0
1
1
1
1
1
1
0
0
0
0
0
0
7.3.3.8
Video timing reference codes (ITU-T 656)
There are two timing reference signals, one at the
beginning of each video data block (start of active video,
SAV) and one at the end of each video data block (end of
active video, EAV).
Each timing reference signal consists of a four word
sequence in the following format: FF 00 00 XY (values are
expressed in hexadecimal notation). The first three words
are a fixed preamble. The forth word XY contains
information defining field 2 identification, the state of field
blanking, and the state of line blanking. The assignment of
bits within the timing reference signal is shown in Table 11.
Table 11
Video timing reference codes
Notes
1.
2.
3.
4.
F = 0 during field 1; F = 1 during field 2.
V = 1 during field blanking; V = 0 elsewhere.
H = 0 in SAV; H = 1 in EAV.
Protection bits are ignored by SAA6750H data
processing.
7.4
Macroblock processor
7.4.1
G
ENERAL
The MacroBlock Processor (MBP) performs the
compression of macroblocks. It fetches its input data from
the external DRAM memory where this was stored by the
video front-end and formatter. The data processing is
macroblock related. The processing start information and
the global scheduling is provided by the global controller
module.
The functionality of the MBP is controlled by the
Application Specific Instruction-set Processor (ASIP).
The ASIP does also perform some computing of data
needed by the MBP. The compressed data is fed to the
packer module.
7.4.2
F
UNCTIONAL DESCRIPTION
7.4.2.1
General
The MBP performs source coding on macroblock level.
It contains several items: motion estimation; motion
compensation,noisereductionandframefieldconversion;
Discrete and Inverse Discrete Cosine Transformations
(DCT and IDCT), quantization and inverse quantization;
motion decompensation and frame-field conversion;
zigzag scanning; DC trend removal (residue); Run-Length
Encoding (RLE) and Variable-Length Encoding (VLE).
7.4.2.2
Motion estimation
The motion estimator considers frame based motion.
Furthermore, the frame distance is one frame and,
consequently, can only be used for P frames.
The motion estimation is based on the recursive block
matching algorithm. Per macroblock the ASIP must feed
the motion estimator with five candidate vectors.
Depending on a control word, the last two vectors can be
relative to the computed vector of the previous macroblock
or can be absolute. The vectors are compared by the
Minimum Absolute Difference (MAD) of the estimated
macroblock in the previous frame and the current
macroblock. The vector that leads to the smallest MAD is
selected. The fifth vector gets a penalty and can be used
as random vector candidate. The two coordinates of the
selected vector and the corresponding MAD value are
returned to the ASIP.
7.4.2.3
Noise filtering
The availability of the motion estimator makes motion
compensated adaptive temporal filtering possible.
The functioning of this filter can be programmed by two
parameters. These parameters are provided by the ASIP.
The noise reduction may only be activated if control
bit INTRA is set to logic 0 (see Table 22).
1
F
(1)
V
(2)
H
(3)
P
3(4)
P
2(4)
P
1(4)
P
0(4)