Re: Digital TV

[Raymond <Reb.barnett@xxxxxxx>]

Rather than give you my understanding of MPEG2, here's why the encoder
matters, straight from the horse's mouth.

Ray.

-----Original Message-----
From: ian.poole@xxxxxxx [mailto:ian.poole@xxxxxxx]
Sent: 23 August 1999 14:11
To: Reb.barnett@xxxxxxx
Subject: Re:MPEG quality

First, the quality of the end picture is TOTALLY DEPENDENT on the
encoder. Crap
encoder gives crap picture. However, the quality sould be independent of
the
decoder, ignoring post processing (filtering, data embedding etcetera)

The MPEG specification states what makes a data stream a valid MPEG-2
bitstream.
It does NOT give explicit instructions on how to construct the bitstream
in the
first place. MPEG-2 is often reffered to as a "tool-kit for
compression", as
opposed to a standard because MPEG-2 is just that - a group of
mathematical
tools for compressing video. The quality of the final picture depends on
which
of these tools are used, and how they are implemented. The decoder,
however, is
told via the bitstream which tools were used in the encoder, and hence
dumbly
follows with the inverse tools. Our company employes 3 phD
mathematicians
full-time to develop algorithms to automatically select the best tool
for each
frame of video.

If a signal is compressed, it does _NOT_ have to be inferior to the
original
uncompressed picture. To achieve any decent level of compression, MPEG-2
is
lossy, but turn up the bit rate and it theoretically becomes loss-less.

This is why the picture is dependent on the encoder:
GOP structure:
MPEG consists of three main types of picture. "I" frames, which
related
to
JPEGS, and are stand alone frames. "P" pictures, which are
referenced
from the
previous "I" frame, and finally "B" pictures which
referenced from both
"I" and
"P" pictures. The pictures can be encoded as frames or fields.
All MPEG
sequences have to contain "I" frames, but after that it is up to
the
encoder to
decide when/where "P" and "B" pictures go. The dumbest
MPEG stream (as
proposed
by Sony) uses JUST I-frames, but has a very poor compression ratio.

Rate Control
So you want a 5MBit per second MPEG-2 output. What proportion of that
5Mbits do
you assign to the 3 types of picture? And how do you control it? (The
MPEG-2
output is buffered using a FIFO - the output is a continuous xMBits, but
the
input can be whatever you want so long as the buffer never fills, nor
does it
ever empty) At 25frames per second, thats an _average_ of 0.2MBits per
frame.
What if you assign 1MBits to a frame with lots of detail? If it takes
longer
than a frame period to transmit the compressed image (ie lots of bits,
and
hopefully lots of quality), this means the next frames have to take less
than a
frame period, thus imposing conditions on the number of bits assigned to
those
following frames, and hence the compression factor and _quality_ for
those
frames.

Motion Vectors
Block matched or phase-correlation? Or some hybrid. Block matched gives
better
video coefficiants, but the motion vectors are pseudo random noise,
which must
be encoded along with the video, resulting in poor compression. Phase
Correlation does not always give such nice video coefficiants, but the
motion
vectors are highly structured, and so compress very well. Also, block
matching
introduces artifacts into areas of plain colour (check out the cartoon
network -
the shimmering in the background is caused by the encoder's use of block
matched
motion vectors)



*******************************************************************
Tel +44 (0)1730 821 188
email ian.poole@xxxxxxx
www.snellwilcox.com
Snell and Wilcox; Durford Mill; Petersfield; Hants;
GU31 5AZ; UK
*******************************************************************



____________________Reply Separator____________________
>The quality of the encoder is not an issue as the encoder is only doing
what
>the MPEG standard is asking.  All, well most, encoders for TV broadcast
are
>MPEG compliant and DVB compliant so  the quality comes down to the
bandwidth
>and only the bandwidth (OK, also the screen you are watching it on and
the
>human eye).

>If a signal, picture is compressed then it has to be inferior as some
of the
>components of the picture have been taken away on the encoding and and
>replace with assumptions, i.e. this bit of sky is blue so this bit must
also
>be blue.


------------------------------------------------------------------------
MyPoints-Free Rewards When You're Online.
Start with up to 150 Points for joining!
http://clickhere.egroups.com/click/805


eGroups.com home: http://www.egroups.com/group/ukha_d
http://www.egroups.com - Simplifying
group communications