[opendtv] Re: News: CEA FORECASTS CONSUMER ELECTRONICS REVENUE
- From: Craig Birkmaier <craig@xxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Sun, 14 Jan 2007 07:55:45 -0500
At 5:17 PM -0500 1/13/07, Albert Manfredi wrote:
There's no question that doing this stuff in digital gets you closer
to the Shannon limit, as of today anyway. Although I'm also positive
that better analog techniques *can* be developed.
Sometimes yes, sometimes no. It just depends on what you are trying
to push through the pipe.
The beauty of analog video encoding is that you take the hit on
quality when it is encoded, then the quality is consistent. You will
lose some information in the encoder, but everything will have the
same limits ( the one area that gets hairy is how much vertical
detail you can push through the channel when the source is
interlaced, without creating objectionable flicker artifacts on
scanning displays).
With digital compression you don't need to brick wall filter
everything as was the case for analog - I am being overly simplistic
here, as the spectra for analog encoding rolls off smoothly, not
sharp like a brick wall. But there is a finite limit on what can pass
through the channel and everything looks pretty because it is all
filtered properly.
With digital we can be far more selective about what we can squeeze
through the channel. In theory we can push to the bandwidth limits of
the channel 13.5 MHz for SDTV, ~30 MHz for HDTV. With the right
information content we can hit these limits, if there is high
frequency content in the source. But this quickly goes away if we try
to push too much information through the channel at a given bitrate.
When we start to quantize the image in the digital encoder, THE FIRST
THING that gets thrown out is the high frequency detail
As most HD sources have little more than noise in the spectra above
20 MHz, it is common practice to roll off the response starting at
about 20-22 MHz, with no information above 25 MHz. By comparison,
most NTSC equipment has a smooth roll-off from 4.2 MHz down to about
8 MHz. This roll-off occurs in the source, before it sees the first
digital encoder, so there is NOTHING close to 30 MHz detail in the
spectra.
Then MPEG-2 gets a hold of the source and things go downhill from
there. Virtually all MPEG encoders now have pre-processing to deal
with encoder stress. These pre-processes include noise reduction and
low pass filtering. IF the encoder is stressed, you can just roll off
the frequency response to limit the high frequency detail. This has
been common practice for the DBS operators since day one. DirecTV is
still getting many complaints about the quality of their HD content,
as they pre-filter so they can get the stream down to an average of
about 10 Mbps.
But even with 25 Mbps, you will still see some scenes that will fall
apart - just ask anyone who is shooting with an HDV camcorder.
And then there is the dirty little secret of MPEG encoding. It can
easily create images with very high frequency ERRORS. Take some
nicely filtered text keyed over any average scene. If you look
closely at the pixels in the transition between the text and the
background you will see nice controlled rise times - you will not see
very high contrast pixels next to one another. But when the DCT
transform is quantized, many of those nice smooth transitions are
butchered. It is quite easy to generate an image with a black pixel
next to a white pixel, and the edges of the text will come alive with
quantization errors.
So theory is nice Bert, but in practice, reality can be VERY
different. For digital video, it is not the quantity of pixels that
one can push through the channel that matters. It is the quality of
the pixels that make it through that matters. In most cases, less is
actually more. I can easily demonstrate how you can make BETTER
pictures with source that is downsampled to about 0.5 Mpixels per
frame then encoded properly to preserve the integrity of the samples,
versus encoding the original 2 Mpixel source, only to have it trashed
by excessive quantization.
Regards
Craig
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.
Other related posts: