DSP在手机H.264解码应用中的初步分析

haichao

H.264  is the latest video compression standard especially
suitable for wireless network.

In wireless communications, bandwidth is too expensive to waste.
For most of the mobile users, bandwidth for steaming video will
not go beyond 128kbps. Among current video compression
standards, only H.264 can provide high quality at such low bit
rate, because compared with former video coding standards
such as MPEG-4 part 2, it saves up to 40% in bit rate and
provides error resilience features. Thus we adopted H.264
decoding as our target application. However, the improvement
in performance also introduces increase in computational complexity,
which requires more powerful hardware. At the same
time, there are several image and video coding standards currently
used such as JPEG and MPEG-4. Although ASIC design
can meet real-time performance, it is not programmable which
lacks flexibility for heterogeneous media standards. Hence
programmable DSP processor is believed by many people to
be the trend since it provides both real-time performance and
flexibility.

New Features of H.264

1) Entropy Decoding: CAVLC is context adaptive variable
length coding based on 4×4 blocks. It is combination of both
entropy coding and run length coding.
2) Inverse Integer Transform: In H.264, IDCT has been
replaced by inverse integer transform (IIT) which only involves
shift and addition without multiplication. Thus this part is
not as computational intensive as in former standards while it
provides more accuracy without drift caused by approximation
in transform.

3) Quarter Pixel Interpolation and Variable Block Size
Motion Compensation: Most of the existing standards support
the motion estimation accuracy up to 1/2 pixel. In H.264, the
maximum accuracy is 1/4 pixel. In former standards, motion
estimation is based on 16x16 macroblock for luma component
as well as 8x8 block for chroma component. While in H.264,
variable block-size motion compensation with smaller block
sizes is supported, thus picture can be restored with higher
accuracy. However, this largely increases the computational
complexity.

4) De-blocking Filter: De-blocking filter is a new added
part for H.264 in order to minimize the artifacts introduced
mainly by present compression technique such as intra prediction.
Although it does not promote the PSNR greatly, it
can largely improve the subjective appearance.


Performance Table:

Decoding Steps MIPS Cost (QCIF 30fps)
------------------------------------------------
Entropy Decoding (CAVLC) 66
Inverse Quantization 21
Inverse Transform 40
Intra Prediction 30
Inter Prediction (Interpolation) 256
De-blocking Filter 467
Other 30

As show above, three tasks appear to be most computational
complex in H.264 decoding which are CAVLC decoding,
interpolation and de-blocking filter.

lcos

很不错 的。

landis

好象
Decoding Steps MIPS Cost (QCIF 30fps)
------------------------------------------------
Entropy Decoding (CAVLC) 66
Inverse Quantization 21
Inverse Transform 40
Intra Prediction 30
Inter Prediction (Interpolation) 256
De-blocking Filter 467
Other 30
上面的数值相加，需要好几百 MIPS,不知道使用什么DSP