Information Technology Reference
In-Depth Information
controller
controller
Complexity,
Generating Bit
Target Bit
Video 1
encoding
unit
encoding
unit
buf
VBR
stream
(Ch1)
Video 2
encoding
unit
encoding
unit
Bitrate
buf
VBR
stream
(Ch1+Ch2)
Ch.1
Video 3
Ch.2
encoding
unit
encoding
unit
Ch.3
buf
VBR
stream
(Ch1+Ch2+Ch3)
Ch.4
Video 4
time
encoding
unit
encoding
unit
buf
Muliplex
CBR
stream
(TS1+TS2+TS3+TS4)
Pre-encoders
Main-encoders
Fig. 10
Two-stage spatially parallel encoding in multiplexing mode
increases encoding quality and the latency of the CODEC. However, small delay
decreases them. The value of frame delay is determined by a trade off between the
encoding quality and the delay of the CODEC system. We selected one GOP delay
as a good trade-off point. In Stage 2, the pre- and main-encoders encode all partial
frames. After the
i
-th frame is encoded by the pre-encoders, the controller obtains
complexity
Xp
i
,
k
and generated bits
S
i
,
k
of partial frame (i,k) from the
k
-th pre-
encoder in Stage 3. Then, in Stage 4, the function of “Future X based STEP1()”
calculates the target bits,
TAm
j
,ofthe(
i
N
)-th non-separated frame by taking the
complexity of the next N frames into consideration. In other words, this complexity
information on the next N frame enables the controller to determine a precise value
for the target bits of the main-encoder, and the encoding quality of the main-encoder
will thus be superior to that of the pre-encoder. This is why we allocated an N frame
buffer in front of the main-encoder.
The output VBR ES of the main encoders are multiplexed into one CBR TS
through coordinating the multiplex modules in the encoder LSIs. The output bi-
trates of the main-encoders are changed at the same timing, while the bitrate of the
multiplexed TS is held constant. This bit-allocation process enables the overall pic-
ture quality to be kept almost constant, because it updates the rate-distortion models
of all frame parts by using the coding information from all encoders in every frame.
−
